doi
stringlengths 28
28
| title
stringlengths 19
311
| abstract
stringlengths 217
5.08k
| plain language summary
stringlengths 115
4.83k
| article
stringlengths 3.87k
161k
|
|---|---|---|---|---|
10.1371/journal.ppat.1002561 | Visualization and Measurement of ATP Levels in Living Cells Replicating Hepatitis C Virus Genome RNA | Adenosine 5′-triphosphate (ATP) is the primary energy currency of all living organisms and participates in a variety of cellular processes. Although ATP requirements during viral lifecycles have been examined in a number of studies, a method by which ATP production can be monitored in real-time, and by which ATP can be quantified in individual cells and subcellular compartments, is lacking, thereby hindering studies aimed at elucidating the precise mechanisms by which viral replication energized by ATP is controlled. In this study, we investigated the fluctuation and distribution of ATP in cells during RNA replication of the hepatitis C virus (HCV), a member of the Flaviviridae family. We demonstrated that cells involved in viral RNA replication actively consumed ATP, thereby reducing cytoplasmic ATP levels. Subsequently, a method to measure ATP levels at putative subcellular sites of HCV RNA replication in living cells was developed by introducing a recently-established Förster resonance energy transfer (FRET)-based ATP indicator, called ATeam, into the NS5A coding region of the HCV replicon. Using this method, we were able to observe the formation of ATP-enriched dot-like structures, which co-localize with non-structural viral proteins, within the cytoplasm of HCV-replicating cells but not in non-replicating cells. The obtained FRET signals allowed us to estimate ATP concentrations within HCV replicating cells as ∼5 mM at possible replicating sites and ∼1 mM at peripheral sites that did not appear to be involved in HCV replication. In contrast, cytoplasmic ATP levels in non-replicating Huh-7 cells were estimated as ∼2 mM. To our knowledge, this is the first study to demonstrate changes in ATP concentration within cells during replication of the HCV genome and increased ATP levels at distinct sites within replicating cells. ATeam may be a powerful tool for the study of energy metabolism during replication of the viral genome.
| ATP is the major energy currency of living cells. Replication of the virus genome is a physiological mechanism that is known to require energy for operations such as the synthesis of DNA or RNA and their unwinding. However, it has been difficult to comprehend how the ATP level is regulated inside single living cells where the virus replicates, since average ATP values in cell extracts have only been estimated using existing methods for ATP measurement. ATeam, which was established in 2009, is a genetically-encoded Förster resonance energy transfer (FRET)-based indicator for ATP that is composed of a small bacterial protein that specifically binds ATP sandwiched between two fluorescent proteins. In this study, by applying ATeam to the subgenomic replicon system, we have developed a method to monitor ATP at putative subcellular sites of RNA replication of the hepatitis C virus (HCV), a major human pathogen associated with liver disease, in living cells. We show here, for the first time, changes in ATP concentrations at distinct sites within cells undergoing HCV RNA replication. ATeam might open the door to understanding how regulation of ATP can affect the lifecycles of pathogens.
| Adenosine 5′-triphosphate (ATP) is the major energy currency of cells and is involved in a variety of cellular processes, including the virus life cycle, in which ATP-dependent reactions essential for virus multiplication are catalyzed by viral-encoded enzymes or complexes consisting of viral and host-cell proteins [1]. However, the lack of a real-time monitoring system for ATP has hindered studies aimed at elucidating the mechanisms by which cellular processes are controlled through ATP. A method for measuring ATP levels in individual living cells has recently been developed using a genetically-encoded FRET-based indicator for ATP, called ATeam, which employs the epsilon subunit of a bacterial F0F1-ATPase [2]. The epsilon subunit has several theoretical advantages for use as an ATP indicator; i) small size (14 kDa), ii) high specific binding to ATP, iii) ATP binding induces a global conformational change and iv) ATP hydrolysis does not occur following binding [3]–[5]. The affinity of ATeam for ATP can be adjusted by changing various amino acid residues in the ATP-binding domain within the subunit. ATeam has enabled researchers to examine the subcellular compartmentation of ATP as well as time-dependent changes in cellular ATP levels under various physiological conditions. For example, the ATeam-based method has been used to demonstrate that ATP levels within the mitochondrial matrix are lower than those in the cytoplasm and the nucleus [2].
Hepatitis C virus (HCV) infects 2–3% of the world population and is a major cause of chronic hepatitis, liver cirrhosis and hepatocellular carcinoma [6]–[8]. HCV possesses a positive-strand RNA genome and belongs to the family Flaviviridae. A precursor polyprotein of ∼3000 amino acids is post- or co-translationally processed by both viral and host proteases into at least ten viral products. The nonstructural (NS) proteins NS3, NS4A, NS4B, NS5A and NS5B are necessary and sufficient for autonomous HCV RNA replication. These proteins form a membrane-associated replication complex (RC), in which NS5B is the RNA-dependent RNA polymerase (RdRp) responsible for copying the RNA genome of the virus during replication [9], [10]. NS3, in addition to its protease activity, functions as a viral helicase capable of separating duplex RNA and DNA in reactions fuelled by ATP hydrolysis [11], [12]. Consistent with other positive-strand RNA viruses, replication of HCV genomic RNA is believed to occur in membrane-bound vesicles. NS3-NS5B proteins, together with several host-cell proteins, form a membrane-associated RC. The HCV RC is localized to distinct dot-like structures within the cytoplasm of HCV replicating cells and can be detected in detergent-resistant membrane structures [13].
In this study, we first used capillary electrophoresis-time-of-flight mass spectrometry (CE-TOF MS) and the original ATeam method to determine ATP levels in cells infected with HCV or replicating HCV RNA. Using these methods, together with an ATP consumption assay, we demonstrated that ATP is actively consumed in cells in which viral RNA replicates, leading to a reduction in cytoplasmic ATP compared to parental cells. To further understand the fluctuation and distribution of ATP in HCV replicating cells, we developed a system to monitor ATP at putative subcellular sites of HCV RNA replication in single living cells by applying ATeam technology to the subgenomic replicon system. Our results show that, in viral RNA-replicating cells, ATP levels are elevated at distinct dot-like structures that may play a supportive role in HCV RNA replication, while cytoplasmic levels of ATP decrease.
As a first approach, the concentration of adenosine nucleotides within HCV-infected and non-infected cells was quantified by CE-TOF MS analysis. ATP levels were approximately 7- and 50-fold higher, respectively, than the levels of ADP and AMP in non-infected Huh-7 cells (Figure 1A). At 9 days post-infection with HCV particles produced from a wild-type JFH-1 isolate [14], the intracellular levels of ATP, ADP and AMP were significantly (52–59%) lower than those in naïve Huh-7 cells (Figure 1A). ATP/ADP and ATP/AMP ratios were comparable among HCV-infected and non-infected cells (Figure 1B). A similar result was obtained using JFH-1/4-5 cells that harbor a HCV subgenomic replicon (SGR) RNA derived from the JFH-1 isolate [15]; the intracellular ATP level of JFH-1/4-5 cells was lower than that of parental Huh-7 cells (Figure S1). These findings are basically consistent with a recent report that phosphorylation-mediated activation of AMP-activated protein kinase is inhibited in cells undergoing HCV genome replication, and that ATP/ADP ratios are similar among cells that do and do not demonstrate HCV replication [16], [17].
To visualize ATP levels in living cells undergoing HCV genomic replication, one of the ATeam indicators, AT1.03YEMK, which has a high affinity for ATP, was introduced into HCV replicon cells carrying SGR RNA or into parental Huh-7 cells and was imaged using confocal fluorescence microscopy. Consistent with previous observations in HeLa cells [2], this ATP indicator was distributed throughout the cytoplasm. FRET signals (Venus/CFP fluorescence emission ratios), which reflect ATP levels in living cells, were calculated from the fluorescent images of CFP and Venus, a variant of YFP that is resistant to intracellular pH [18], within the cytoplasm of individual cells. Each independent measurement was plotted as indicated in Figure 2. Uniform Venus/CFP ratios were observed in Huh-7 cells. These ratios were reduced dramatically following combined treatment with 2-deoxyglucose (2DG) and Oligomycin A (OliA), which inhibit glycolysis and the oxidative phosphorylation of ADP to ATP, respectively [2]. When AT1.03YEMK was expressed in the HCV replicon-harboring cells JFH-1/4-1, JFH-1/4-5 (genotype 2a) and NK5.1/0-9 (genotype 1b) [15], Venus/CFP ratios were significantly lower than those seen in parental Huh-7 cells. This result is consistent with the mass spectrometry results shown in Figures 1A and S1. Venus/CFP ratios were more variable in the replicon-carrying cells compared to Huh-7 cells. It is possible that ATP levels in the replicon cells correlate with viral replication levels, which may vary among the cells tested.
It has been reported that ATP is involved in different steps in the course of HCV replication such as in the initiation of RNA synthesis by NS5B RdRp [9]. NS3 unwinds RNA in an ATP-dependent manner and may be involved in viral replication [11], [19], [20]. NS4A has been shown to enhance the ability of the NS3 helicase to bind RNA in the presence of ATP [21]. In addition, ATP is generally used as a material in RNA synthesis. Together with the above results (Figures 1 and 2), one may hypothesize that active consumption of ATP in cells where HCV RNA replicates efficiently results in lower levels of cytoplasmic ATP compared to cells in the absence of the viral RNA. To study the influence of HCV RNA replication on the consumption of ATP in cells, we used permeabilized HCV replicon cells [13], [22].
Following the addition of ATP to permeabilized cells, reduced ATP levels were detected using a luciferase-based assay (see Materials and Methods for details). Fifteen minutes after the addition of ATP, ATP levels in permeabilized replicon-carrying cells (JFH-1/4-1, JFH-1/4-5 and NK5.1/0-9) were reduced by 82–95%, and this reduction was greater than that observed in control Huh-7 cells (47%)(Figure 3). When the replication of HCV RNA was inhibited by pre-treatment of the cells with the cytidine analogue inhibitor of HCV NS5B polymerase, PSI-6130 [23], [24], for 3 days, the reduction in ATP levels in the replicon cells was comparable to that of Huh-7 cells. A decrease in ATP reduction in the replicon cells was observed even following a 15-min treatment with the inhibitor. An effect of inhibition of viral replication on cytoplasmic ATP levels in replicon cells was also observed by ATeam-based analysis of Venus/CFP ratios following inhibition of replication by IFN-alpha (Figure S2). These results suggest that ATP is actively consumed during viral replication in HCV replicon cells, leading to decreased levels of ATP in the cytoplasm.
Moradpour et al. have established functional HCV replicons that have either an epitope tag or the coding sequence for a green fluorescent protein (GFP) inserted in frame close to the C-terminus of NS5A, which they used to demonstrate incorporation of the NS5A-GFP fusion protein into the viral RC [25]. To further investigate intracellular changes in ATP during HCV replication, we generated HCV JFH-1-based subgenomic replicons harboring an ATeam insertion in the 3′ region of NS5A (SGR-ATeam), as well as plasmids expressing NS5A-ATeam fusion proteins (NS5A-ATeam)(Figures 4A and 4C).
We first tested whether NS5A-ATeam fusion proteins can be used to monitor ATP levels over a range of concentrations in living cells. The Venus/CFP ratios in individual cells expressing NS5A fused either with AT1.03YEMK (Kd = 1.2 mM at 37°C [2]) or with a relatively lower affinity version, AT1.03 (Kd = 3.3 mM at 37°C [2]) were measured. As shown in Figure 4B, differences in the Venus/CFP ratios of NS5A- AT1.03YEMK and NS5A-AT1.03 were similar to those of AT1.03YEMK and AT1.03, although average ratios were lower for NS5A- AT1.03YEMK and NS5A-AT1.03 compared to AT1.03YEMK and AT1.03. In the presence of 2DG and OliA, Venus/CFP ratios of NS5A-AT1.03YEMK were markedly reduced to levels that were comparable to those of AT1.03RK, an inactive mutant with R122K/R126K substitutions [2]. These results demonstrate that NS5A-ATeams can function as ATP indicators, although their dynamic ranges of Venus/CFP ratios are slightly smaller than those of the original, non-fused ATeams.
We next investigated whether the SGR-ATeam could initiate and sustain transient replication of HCV RNA in cells. A RNA polymerase I (Pol I)-derived plasmid, which carries SGR/luc-AT1.03 containing a luciferase reporter gene ([26]; Figure 4C), or its replication-defective mutant were transfected into Huh-7 cells and levels of viral replication were determined by measuring luciferase activity at various time intervals over a five day period (Figure 4D). Although replication of SGR/luc-AT1.03 was delayed compared with parental SGR/luc, the luciferase activity expressed from SGR/luc-AT1.03 rose to approximately a thousand-fold higher than that expressed from SGR/luc-GND-AT1.03 at five days post-transfection. It appears that SGR-AT1.03, which does not carry the luciferase gene, replicated more efficiently than SGR/luc-AT1.03, as determined by Western blotting of the HCV NS5B protein within cells four days post-transfection (Figure 4E). As indicated in Figure 4F, an abundant protein of the same size as that expected for the NS5A-ATeam fusion protein was observed in cells expressing either NS5A-AT1.03 or SGR-AT1.03, indicating that the NS5A-ATeam fusion protein is stable and is not cleaved during HCV replication. Thus, we concluded that the modified replicon constructs in which the ATeam is incorporated into the NS5A region are functional and remain capable of efficient transient replication of HCV RNA.
This SGR-ATeam system that was established to analyze cellular ATP levels was used in living HCV RNA-replicating cells in which membrane-associated RCs are formed through the interaction of viral proteins, including NS5A, and cellular proteins. We compared the subcellular distribution of fluorescent signals expressed from NS5A-ATeams and SGR-ATeams using emission-scanning confocal fluorescence microscopy with a Zeiss META detector. NS5A-AT1.03 and NS5A-AT1.03YEMK were diffusely distributed throughout the cytoplasm (Figure 5A; upper panels). Venus/CFP ratios of NS5A-ATeam constructs were almost constant throughout the cytoplasm (Figure 5A; lower). As expected, Venus/CFP ratios in cells expressing NS5A-AT1.03YEMK were markedly higher than those of NS5A-AT1.03 (Figure 5A; lower). In contrast, cells replicating SGR-AT1.03 and SGR-AT1.03YEMK showed foci of brightly fluorescent dot-like structures in the cytoplasm (Figure 5B; upper panels). Interestingly, some of these fluorescent foci had an apparently higher Venus/CFP ratio than the surrounding cytoplasmic region (Figure 5B; middle and lower panels). Although the number of high Venus/CFP ratios was not consistent between the cells, this phenotype was observed in most of the cells that were replicating SGR-AT1.03 (Figure S3). Such high focal Venus/CFP ratios were not detected in cells replicating SGR-AT1.03RK or in SGR-AT1.03YEMK -replicating cells treated with 2DG and OliA. Thus, foci with a high Venus/CFP ratio apparently represent the presence of high ATP levels at distinct sites in cells replicating HCV RNA. In addition, when a replication-defective polyprotein that extended from NS3 through to the NS5B protein, including NS5A-AT1.03, was expressed, no high Venus/CFP ratio was seen in the cells in spite of the fact that NS5A-AT1.03 was detected in dot-like structures throughout the cytoplasm (Figure S4). These results strongly suggest that the high Venus/CFP ratios observed using the SGR-ATeam system are associated with the replication of HCV RNA.
To investigate whether the high Venus/CFP ratios of the dot-like structures detected in cells replicating SGR-ATeam are located at the HCV RC, FRET images of SGR-AT1.03-replicating cells were analyzed, followed immunofluorescence analysis of cells fixed and stained with either anti-NS5A or anti-NS3 antibodies (Figure 5C). Confocal fluorescence microscopy at high magnification demonstrated that the high Venus/CFP ratios that were identified in foci of various sizes were co-localized with NS5A and NS3 that were possibly membrane-bound within the cytoplasm of the viral replicating cells. Some of the NS3- or NS5A-labeled proteins that were identified by immunofluorescence were not associated with high Venus/CFP ratios. These results are consistent with previous reports, which demonstrated that only some of the expressed HCV NS proteins contribute to viral RNA synthesis [27]. To further investigate the relationship between the cellular sites at which there was a high Venus/CFP ratio and HCV RNA replication, double-stranded RNA (dsRNA) was visualized by staining with a specific anti-dsRNA antibody after FRET imaging (Figure 5C). This staining indicated that dsRNA-containing dot-like structures co-localized with structures that displayed high Venus/CFP ratios. Therefore, it is most likely that the dot-like structures with high Venus/CFP ratios that were detected using the SGR-ATeam system reflect the sites of HCV RNA replication or HCV RCs.
Several studies have shown that mitochondria, which play a central role in ATP metabolism, localize to areas near the membranous web, the likely site of HCV RNA replication [28]. We thus compared the subcellular localization of the fluorescence signals detected in cells expressing SGR-ATeam with that of mitochondria that were visualized by staining with Mitotracker. Foci with high Venus/CFP ratios did not colocalize with, but were localized adjacent to mitochondria in cells that were replicating SGR-AT1.03 (Figure S5). This finding might reflect the fact that ATP can be directly supplied from mitochondria to the sites of viral RNA replication in cells.
Based on the above observations, FRET signals detected within cells expressing SGR-ATeam or NS5A-ATeam can be classified as either signals from distinct dot-like structures, which represent putative subcellular sites of HCV RNA replication, or as signals that are diffuse throughout the cytoplasm. The Venus/CFP emission ratio in individual cells into which NS5A-AT1.03, NS5A-AT1.03YEMK, SGR-AT1.03, SGR-AT1.03YEMK or SGR-AT1.03RK was introduced was determined (Figure 6A). Fluorescent signals corresponding to cytoplasmic ATP were identified by subtracting signals at putative sites of viral RNA replication from signals from the cytoplasmic area as a whole. Cytoplasmic Venus/CFP ratios within cells replicating SGR-AT1.03 and SGR-AT1.03YEMK were lower than those in cells expressing NS5A-AT1.03 and NS5A-AT1.03YEMK, respectively. Therefore, cytoplasmic ATP levels within HCV RNA-replicating cells were lower than in non-replicating cells. This result is consistent with the findings shown in Figure 1A. The average Venus/CFP ratios at potential sites of viral RNA replication were greater than the corresponding cytoplasmic levels in cells replicating SGR-AT1.03 or SGR-AT1.03YEMK. As expected, a significant decrease in Venus/CFP ratios was observed in cells treated with 2DG and OliA.
We next quantified ATP levels within individual cells replicating HCV RNA based on the Venus/CFP ratios obtained. To generate standard curves for this calculation, permeabilized cells expressing NS5A-AT1.03 or NS5A-AT1.03YEMK were prepared by digitonin treatment, followed by the addition of defined concentrations of ATP and subsequent FRET analysis [29], [30]. As shown in Figure 6B, under these experimental conditions, baseline Venus/CFP ratios of approximately 0.1 were detected in the absence of exogenous ATP, and Venus/CFP ratios were observed to increase linearly with increasing ATP concentration. The standard curves thus obtained can be used to estimate the ATP concentrations of unknown samples in which a particular ATeam containing an ATP probe at the C terminus of HCV NS5A, such as NS5A-ATeam or SGR-ATeam, have been introduced. Based on the fluorescent signal obtained in cells replicating SGR-ATeam, as well as in cells expressing NS5A-ATeam, the ATP concentration at putative sites of HCV RNA replication was estimated to be ∼5 mM in the experiments shown in Figures 5A and 5B (average value of putative replication sites; 4.8 mM). After subtraction of the ATP that was localized at the HCV replication sites, the ATP concentration of HCV-replicating SGR cells (∼1 mM) was found to be approximately half that observed in parental non-replicating cells (∼2 mM)(average values in SGR and parental cells; 0.8 mM and 2.2 mM, respectively). To our knowledge, this is the first experiment in which ATP levels were estimated inside living cells during viral genome replication.
Figures 5 and 6A demonstrate changes in ATP concentrations at distinct sites in cells undergoing HCV RNA replication. Finally, we determined the effect of the PSI-6130 inhibitor of HCV replication on the change in subcellular ATP concentration in cells following introduction of SGR-AT1.03, SGR-AT1.03RK or NS5A-AT1.03 (Figure 6C). In general, nucleoside analogue inhibitors of viral replication prevent RNA/DNA synthesis by chain termination immediately after addition to infected cells [23]. Indeed, as shown in Figure 3, a decrease in ATP consumption was detected even following a PSI-6130 treatment period as short as 15 min of permeabilized HCV replicon cells. We therefore analyzed and estimated ATP levels in cells in the presence of PSI-6130 for 10 min and 2 h. ATP concentrations at putative sites of viral RNA replication, as well as cytoplasmic ATP levels, were higher in SGR-AT1.03-replicating cells in the presence of 0.1–5 µM PSI-6130 for 10 min compared to the same cells without inhibitor treatment or to NS5A-AT1.03-expressing cells. A dose-dependent PSI-6130-induced increase in ATP levels at the putative replication sites was observed under the condition used. By treatment with PSI-6130 for 2 h, the ATP levels at putative replication sites were significantly lower than those without inhibitor treatment in SGR-AT1.03-replicating cells. The cytoplasmic ATP levels were similar with or without 2-h treatment (Figure 6C). In HCV SGR-ATeam cells treated with PSI-6130 for 3 days, HCV RNA replication was dramatically inhibited by greater than 90% with no observed cytotoxicity (Figure S6) and, as expected, little or no high Venus/CFP signal was detected anywhere in the cells (data not shown). We adapted the ATeam system to monitor ATP in HCV RNA replicating cells and found increased ATP levels at the putative subcellular sites of the viral replication. Findings obtained from experiments using the viral polymerase inhibitor strongly suggest that changes in ATP concentrations at the distinct sites observed depend on the viral RNA replication.
This paper is the first to demonstrate changes in ATP within cells during viral genome replication. ATP requirements during the virus lifecycle have been studied for years. Several key steps during the viral life cycle, such as genome synthesis, require high-energy phosphoryl groups. For instance, it has been shown that ATP is required for the formation of a preinitiation complex for de novo RNA synthesis by RdRp of flaviviruses [31]. Transcriptional initiation and RNA replication by influenza virus RdRp are functional in an ATP-dependent fashion [32], [33]. An ATP requirement of viral helicase activities has also been reported [34]. Furthermore, it has been demonstrated that ATP is involved in the assembly and/or release of viral structural proteins possibly via interaction with ATP-dependent chaperones [35], [36]. However, it has been controversial as to whether ATP can be concentrated in particular subcellular compartment(s) in infected cells during viral replication. One of the underlying reasons for this controversy may be that a method by which cellular ATP levels can be determined, apart from examination of ATP levels in cellular extracts in the steady-state, has been lacking [37]. Recently Imamura et al. established FRET-based indicators, known as ATeams, for ATP quantification, and have shown that the use of ATeams enables the monitoring of ATP levels in real-time in different cellular compartments within individual cells [2].
In this study, in order to visualize and monitor ATP levels in living cells during replication of the viral genome, we first introduced the original ATeam-expressing plasmids into cells and found that cytoplasmic ATP levels in cells undergoing HCV genotype 1b and 2a RNA replication were lower than those in cured or parental cell lines (Figures 2 and S2). These results agree with the results of CE-TOF MS analysis (Figure 1) and the ATP consumption assay (Figure 3). It is therefore likely that ATP is actively consumed in cells during viral RNA replication, resulting in reduced levels of ATP in the cytoplasm. Furthermore, NS5A-ATeam fusion constructs, in which the ATeam gene was introduced into the C-terminal end of the NS5A coding region, and SGR-ATeam constructs containing a HCV JFH-1-derived subgenomic replicon within the NS5A-ATeam fused sequence as described above, were engineered (Figure 4). The results obtained using several ATeam fusion constructs with different affinities for ATP indicated that NS5A-ATeam fusion constructs can be used as FRET-based ATP indicators, and that the ATeam-tagged HCV replicons are capable of transient replication of viral RNA (Figure 4). It is interesting that our experiment using a SGR-ATeam construct provides evidence for the formation of ATP-enriched foci within cells that support HCV RNA replication (Figures 5 and 6). FRET-signal detection followed by indirect immunofluorescence allowed us to visualize co-localization of viral proteins as well as dsRNA at sites of ATP accumulation in cells (Figure 5), suggesting that these membrane-associated ATP-enriched foci likely represent sites of HCV RNA replication in transient replication assays.
Attempting to precisely quantify ATP within individual cells or particular intracellular compartments is a very challenging process. The luciferin-luciferase reaction has been utilized to monitor cellular ATP levels by measuring the released photon count during catalysis of bioluminescent oxidation by firefly luciferase. A previous study based on the luciferin-luciferase assay estimated basal cytoplasmic ATP levels at ∼1.3 mM, which increased to ∼5 mM during apoptotic cell death [38]. However, the results obtained were likely influenced by cellular levels of luciferase and other assay components, as well as by the pH of the cells. In this study, we describe quantification of ATP in human hepatoma Huh-7 cells undergoing HCV RNA replication using SGR-ATeam technology. Although ATP requirements during the lifecycles of various viruses have been studied for years, the use of ATeam technology enabled us, for the first time, to evaluate ATP concentrations at sites of viral replication within living cells. We here demonstrate that ATP concentrations at these putative subcellular sites of HCV RNA replication approach ∼5 mM (Figure 6). This ATP level is as high as that observed during apoptotic processes such as caspase activation and DNA fragmentation, even though the latter ATP level was determined using a different assay system [38]. Considering that these apoptotic events were not observed at basal ATP levels [38], replication of the viral genome likely also requires high concentrations of cellular ATP. It should be noted that, in contrast to the fluorescent reporter system traditionally used to calculate the ATP/ADP ratio [39], the bacterial epsilon subunit used in ATeam is highly specific for ATP, but not for other nucleotides such as ADP, CTP, GTP or UTP [2], [3]. In evaluating the effect of the HCV polymerase inhibitor on changes in the subcellular ATP concentration in cells replicating SGR-ATeam, an increase in ATP concentration was observed both at putative replication sites and in the cytoplasm of SGR-AT1.03-replicating cells in the presence of PSI-6130 for 10 min (Figure 6C). By contrast, 2-h treatment with the inhibitor resulted in reduction of ATP levels at putative replication sites in the replicon cells. Although the result of the experiment with 10-min treatment may be somewhat unexpected, it might possibly be explained by the following hypothesis. PSI-6130 began to inhibit viral RNA synthesis, leading to a decrease in ATP consumption. Since a mechanism for ATP transport mediated by host cell and/or viral factor(s) is still active during this time period, the ATP level at the replication sites should be increased compared to that during active replication. Higher levels of metabolic intermediates for glyconeogenesis as well as for glycolysis in HCV-infected cells compared to non-infected cells as determined via metabolome analysis (data not shown) may also be implicated in the increased ATP levels at the initial stage of inhibition of HCV replication. It is likely that active consumption of ATP caused by HCV replication and ATP transportation into the replication sites would lead to reduction of cytoplasmic ATP level. Such a change in ATP balance may result in induction of ATP generation and increase in certain metabolic intermediates related to glucose metabolism. These metabolome responses are supposed to maintain in short-term (10 min) treatment with PSI-6130. Thus, inhibition of HCV RNA replication by PSI-6130 under the conditions used may lead to increase in the cytoplasmic ATP level. It is likely that these metabolome responses were not observed after the longer-term (2 h) treatment presumably because the viral replication was inhibited by the inhibitor for a sufficient period of time. Further study is required to address the molecular mechanism underlying change in ATP balance caused by HCV replication and the viral inhibitors.
The mechanism by which ATP accumulates at potential sites of HCV RNA replication remains unclear. We have previously demonstrated that creatine kinase B (CKB), which is an ATP-generating enzyme and maintains cellular energy stores, accumulates in the HCV RC-rich fraction of viral replicating cells [22]. Our earlier results suggest that CKB can be directed to the HCV RC via its interaction with the HCV NS4A protein and thereby functions as a positive regulator for the viral replicase by providing ATP [22]. One may hypothesize that recruitment of the ATP generating machinery into the membrane-associated site, through its interaction with viral proteins comprising the RC, is at least in part linked with elevated concentrations of ATP at a particular site. Through our preliminary study, however, subcellular ATP distribution was not changed significantly in replicon cells where HCV RNA replication was reduced ∼50% by siRNA-mediated knockdown of the CKB gene (data not shown). Another possibility may be implication of communication between mitochondria and membrane-enclosed structures of HCV RC in ATP transport through membrane-to-membrane contact. As indicated in Figure S5, putative sites of the viral RNA replication with high Venus/CFP ratios were mainly localized proximal to mitochondria. Studies are ongoing to understand the mechanism(s) underlying this phenomenon, as well as to determine if changes in ATP levels at intracellular sites supporting replication might also be observed for other RNA or DNA viruses.
In summary, we have used a FRET-based ATP indicator called ATeam to monitor ATP levels in living cells where viral RNA replicates by designing HCV replicons harboring wild-type or mutated ATeam probes inserted into the C-terminal domain of NS5A. We evaluated changes in ATP levels during HCV RNA replication and demonstrated elevated ATP levels at putative sites of replication following detection of FRET signals, which appeared as dot-like foci within the cytoplasm. The ATeam system may become a powerful tool in microbiology research by enabling determination of subcellular ATP localization in living cells infected or associated with microbes, as well as investigation of the regulation of ATP-dependent processes during the lifecycle of various pathogens.
PSI-6130 (β-D-2′-Deoxy-2′-fluoro-2′-C-methylcytidine) and recombinant human IFN-alpha2b were obtained from Pharmasset Inc. (Princeton, NJ) [23], [24] and Schering-Plough (Kenilworth, NJ), respectively. OliA and 2DG were purchased from Sigma-Aldrich (St. Louis, MO). ATP used in this study was complexed with equimolar concentrations of magnesium chloride before use in the experiments.
The construction of the ATeam plasmids pRSET-AT1.03, pRSET-AT1.03YEMK and pRSET-AT1.03R122K/R126K, which express wild-type ATeam (AT1.03), as well as a high-affinity mutant (AT1.03YEMK) and a non-binding mutant (AT1.03RK), has been previously described [2]. pHH/SGR-Luc (also termed SGR/luc) contains cDNA of a subgenomic replicon of HCV JFH-1 isolate (genotype 2a; [14]) with firefly luciferase flanked by the Pol I promoter and the Pol I terminator, yielding efficient RNA replication upon DNA transfection [26]. pHH/SGR-Luc/GND (also termed SGR/luc-GND), in which a point mutation of the GDD motif of the NS5B was introduced in order to abolish RNA-dependent RNA polymerase activity, was used as a negative control. pHH/SGR (also termed SGR) was created by deleting the luciferase gene in pHH/SGR-Luc. To generate a series of SGR-ATeam plasmids, wild-type or mutant ATeam genes were inserted into pHH/SGR-Luc or pHH/SGR at the Xho I site of NS5A (between amino acids 418 and 419) [25]. The ATeam genes were also inserted into the same site of pCAGNS5A, which contains the NS5A gene of JFH-1 downstream of the CAG promoter and hemagglutinin (HA) tag [26], yielding NS5A-ATeam plasmids. To generate a plasmid expressing NS3-NS5B-AT1.03 under the control of the CAG promoter, a DNA fragment containing the coding region of NS3/NS4A/NS4B/NS5A-AT1.03/NS5B of SGR/luc-ATeam was inserted into the pCAGGS vector [40]. Exact cloning strategies are available upon request.
Human hepatoma Huh-7 cells were propagated in Dulbecco's modified Eagle's medium (DMEM) supplemented with 10% fetal calf serum (FCS) as well as minimal essential medium non-essential amino acid (MEM NEAA)(Invitrogen, Carlsbad, CA) in the presence of 100 units/ml of penicillin and 100 µg/ml of streptomycin. The Huh-7-derived cell lines JFH-1/4-1 and JFH-1/4-5, which support replication of SGR RNA of HCV JFH-1 (genotype 2a) and NK5.1/0-9, which carries the SGR RNA of Con1 NK5.1 (genotype 1b), were cultured and maintained under previously described conditions [15]. DNA transfection was performed using a TransIT-LT1 transfection reagent (Takara, Shiga, Japan) in accordance with the manufacturer's instructions.
Huh-7 cells were mock-infected or infected with HCVcc derived from a wild-type JFH-1 isolate at a multiplicity of infection of 1. When most cells had become virus positive, as confirmed by immunofluorescence, with no observable cell damage at 9 days post-infection, equal amounts of cells with and without HCV infection were scraped with MeOH including 10 µM of an internal standard after washing twice with 5% mannitol solution. Replicon cells (JFH-1/4-5) that were cultured in the absence of G418 for 2 days were harvested and prepared as above. The extracts were mixed with chloroform and water, followed by centrifugation at 2,300× g for 5 min at 4°C, The upper aqueous layer was centrifugally filtered through a 5-kDa cutoff filter to remove proteins. The filtrate was lyophilized and dissolved in water, then subjected to CE-TOF MS analysis. CE-TOF MS experiments were performed using an Agilent CE-TOF MS system (Agilent Technologies, Waldbronn, Germany) as described previously [41].
The ATP consumption assay using permeabilized replicon cells was carried out as previously described [13], [22] with slight modifications, so that it was unnecessary to add either exogenous phosphocreatine or creatine phosphokinase to minimize ATP reproduction in cells. Cells (2×106) cultured in the presence or absence of PSI-6130 for 72 h were treated with 5 µg Actinomycin D/ml, followed by trypsinization and 3 washes with cold buffer B (20 mM HEPES-KOH [pH 7.7], 110 mM potassium acetate, 2 mM magnesium acetate, 1 mM EGTA, and 2 mM dithiothreitol). The cells were permeabilized by incubation with buffer B containing 50 µg/ml digitonin for 5 min on ice and the reaction was stopped by washing 3 times with cold buffer B. The permeabilized cells (1×105) were resuspended with 100 µl buffer B containing 5 µM ATP, GTP, CTP, and UTP, 20 µM MgCl2, and 5 µg/ml Actinomycin D. After incubation at 27°C for 15 min, samples were centrifuged, and 20 µl of the supernatant was then mixed with 5 µl of 5× passive lysis buffer (Promega, Madison, WI). The ATP level was determined using a CellTiter-Glo Luminescent cell viability assay system (Promega). All assays were performed at least in triplicate.
Plasmids carrying the ATP indicators were transfected at 48 h (ATeam and NS5A-ATeam) or 4 days (SGR-ATeam) before imaging of the cells. One day before imaging, the cells were seeded onto 30-mm glass-bottomed dishes (AGC Techno Glass, Chiba, Japan) at about 60% confluency. For imaging, the cells were maintained in phenol red-free DMEM containing 20 mM HEPES-KOH [pH 7.7], 10% FCS and MEM NEAA.
Two kinds of confocal microscopies were used to perform the FRET analysis in this study as follows. Since the ways of acquisition of each spectrum were quite different between the two microscopies, differences in the values of the Venus/CFP ratios in different experiments were observed. In Figures 2, 4B and S2, cells were imaged using a confocal inverted microscope FV1000 (Olympus, Tokyo, Japan) equipped with an oil-immersion 60× Olympus UPlanSApo objective (NA = 1.35). Cells were maintained on the microscope at 37°C with a stage-top incubation system (Tokai Hit, Shizuoka, Japan). Cells were excited by a 405-nm laser diode, and CFP and Venus were detected at 480–500 nm and 515–615 nm wavelength ranges, respectively. In the analysis shown in Figures 5, 6, S3, S4 and S5, FRET images were obtained using a Zeiss LSM510 Meta confocal microscope with an oil-immersion 63× Zeiss Plan-APOCHROMAT objective (NA = 1.4)(Carl Zeiss, Jena, Germany). Cells were maintained on the microscope at 37°C with a continuous supply of a 95% air and 5% CO2 mixture using a XL-3 incubator (Carl Zeiss). Cells were excited by a 405-nm blue diode laser, and emission spectra of 433–604 nm wavelength range were obtained using an equipped scanning module (META detector) [42], [43]. Images were computationally processed by a linear unmixing algorithm using the reference spectrum of CFP and Venus, which were obtained from individual fluorescence-expressing cells. All image analyses were performed using MetaMorph (Molecular Devices, Sunnyvale, CA). Fluorescence intensities of cytoplasmic areas in NS5A-ATeam transfected cells were calculated by subtraction of the signal intensities of the nucleus from the signal intensities of the whole cell, which was standardized by the area of the corresponding cytoplasmic region. Fluorescence intensities of cytoplasmic areas and at dot-like structures corresponding to the putative viral replicating sites in SGR-ATeam-transfected cells were measured and calculated as follows. All pixels above CFP intensity levels of 100–200 were selected. The positions of dot-like structures were then determined by examining areas greater than 0.5×10−12 square meters and the intensity of each dot was measured. The fluorescence intensity of the cytoplasmic area, excluding that of the putative viral replicating sites in each cell, was calculated by subtraction of the signal intensities of the nucleus and the dot-like structures, as determined above, from the signal intensity of the whole cell, which was standardized by the area of the corresponding cytoplasmic region. Each Venus/CFP emission ratio was calculated by dividing pixel-by-pixel a Venus image with a CFP image.
To investigate the relationship between Venus/CFP ratios and ATP concentrations in cells, calibration procedures were performed according to previous reports [29], [30]. Huh-7 cells were transfected with NS5A-AT1.03 or NS5A-AT1.03YEMK. Forty-eight hours later, the cells were permeabilized by incubation with buffer B containing 50 µg/ml digitonin for 5 min at room temperature. The reaction was stopped by washing 3 times with buffer B, followed by the addition of known concentrations of ATP in warmed medium for imaging. FRET analysis, with calibration of the signal intensity in the cytoplasm of each cell, was performed as described above. Plots were fitted with Hill equations with a fixed Hill coefficient of 2; R = (Rmax−Rmin)×[ATP]2/([ATP]2+Kd2)+Rmin, where Rmax and Rmin are the maximum and minimum fluorescence ratios, respectively and Kd is the apparent dissociation constant.
To analyze the effect of an inhibitor against HCV NS5B polymerase, the medium for the cells replicating SGR-ATeam was changed to medium containing various concentrations of PSI-6130. After 10-min incubation at 37°C under a continuous supply of 95% air and 5% CO2, fluorescence intensities of cytoplasmic areas and at dot-like structures were determined as described above. Medium containing 0.01% DMSO was used as a negative control.
To visualize mitochondria, MitoTracker Red CMXRos (Molecular Probes, Eugene, OR) was added to the culture medium to a final concentration of 100 nM, incubated for 15 min at 37°C and the cells were then washed twice with phosphate buffered saline (PBS) before FRET analysis of living cells. Images were computationally processed as described above. The reference spectrum of MitoTracker Red CMXRos was obtained from stained parental, non-transfected, Huh-7 cells.
Cells expressing SGR-ATeam were cultured in 30-mm glass-bottomed dishes with an address grid on the coverslip (AGC Techno Glass). After FRET analysis of living cells as described above, the cells were fixed with 4% paraformaldehyde at room temperature for 30 min. After washing with PBS, the cells were permeabilized with PBS containing 0.3% Triton X-100 and individually stained with a rabbit polyclonal antibody against NS3 [44], an anti-NS5A antibody [45], or a mouse monoclonal antibody against dsRNA antibody (Biocenter Ltd., Szirak, Hungary) [46]. The fluorescent secondary antibody used was Alexa Fluor 555-conjugated anti-rabbit- or anti-mouse IgG (Invitrogen). The cells were imaged using a Zeiss LSM510 Meta confocal microscope with an oil-immersion 63× Zeiss Plan-APOCHROMAT objective (NA = 1.4). For dual-color imaging, the ATeam signal was excited with the 488-nm laser line of an argon laser and Alexa Fluor 555 was excited with a 543-nm HeNe laser under MultiTrack mode. Emission filters with a 505- to 530-nm band-pass and 560-nm-long pass filter were used.
Huh-7 cells transfected with SGR/luc or SGR/luc-ATeam were harvested at different time points after transfection (Figure 4D) or at 3 days after treatment with PSI-6130 (Figure S6) and lysed in passive lysis buffer (Promega). To monitor HCV RNA replication, the luciferase activity in cells was determined using a Luciferase Assay system (Promega). All assays were performed at least in triplicate.
Cell viability was assessed using the Cell Proliferation Kit II (Roche, Indianapolis, IN) according to the manufacturer's instructions. The kit measures mitochondrial dehydrogenase activity, which is used as a marker of viable cells, using a colorimetric sodium3′-[1(-phenylaminocarbonyl)-3,4-tetrazolium]-bis(4-methoxy-6-nitro)benzene sulfonic acid hydrate (MTT) assay.
HCV RNA copies in the replicon cells with or without PSI-6130 treatment were determined using the real-time detection reverse transcription polymerase chain reaction (RTD-PCR) described previously [47] with the ABI Prisom 7700 sequence detector system (Applied Biosystems Japan, Tokyo, Japan).
The proteins were transferred onto a polyvinylidene difluoride membrane (Immobilon; Millipore, Bedford, MA) after separation by SDS-PAGE. After blocking, the membranes were probed with a rabbit polyclonal anti-NS5A antibody [44], a rabbit polyclonal anti-NS5B antibody (Chemicon, Temecula, CA), or a mouse polyclonal anti-beta-actin antibody (Sigma-Aldrich), followed by incubation with a peroxidase-conjugated secondary antibody and visualization with an ECL Plus Western blotting detection system (GE Healthcare, Buckinghamshire, UK).
|
10.1371/journal.ppat.1002984 | Tim-3-Expressing CD4+ and CD8+ T Cells in Human Tuberculosis (TB) Exhibit Polarized Effector Memory Phenotypes and Stronger Anti-TB Effector Functions | T-cell immune responses modulated by T-cell immunoglobulin and mucin domain-containing molecule 3 (Tim-3) during Mycobacterium tuberculosis (Mtb) infection in humans remain poorly understood. Here, we found that active TB patients exhibited increases in numbers of Tim-3-expressing CD4+ and CD8+ T cells, which preferentially displayed polarized effector memory phenotypes. Consistent with effector phenotypes, Tim-3+CD4+ and Tim-3+CD8+ T-cell subsets showed greater effector functions for producing Th1/Th22 cytokines and CTL effector molecules than Tim-3− counterparts, and Tim-3-expressing T cells more apparently limited intracellular Mtb replication in macrophages. The increased effector functions for Tim-3-expressing T cells consisted with cellular activation signaling as Tim-3+CD4+ and Tim-3+CD8+ T-cell subsets expressed much higher levels of phosphorylated signaling molecules p38, stat3, stat5, and Erk1/2 than Tim-3- controls. Mechanistic experiments showed that siRNA silencing of Tim-3 or soluble Tim-3 treatment interfering with membrane Tim-3-ligand interaction reduced de novo production of IFN-γ and TNF-α by Tim-3-expressing T cells. Furthermore, stimulation of Tim-3 signaling pathways by antibody cross-linking of membrane Tim-3 augmented effector function of IFN-γ production by CD4+ and CD8+ T cells, suggesting that Tim-3 signaling helped to drive stronger effector functions in active TB patients. This study therefore uncovered a previously unknown mechanism for T-cell immune responses regulated by Tim-3, and findings may have implications for potential immune intervention in TB.
| Tuberculosis (TB), an infectious disease caused by Mycobacterium tuberculosis (Mtb) infection, remains a leading cause of morbidity and mortality worldwide. While CD4+ and CD8+ T-cell effector functions producing Th1 or cytotoxic cytokines are required to mount anti-mycobacterial immunity, insufficiency or failure to mount anti-mycobacterial effector functions by CD4+ and CD8+ T cells may lead to impaired immunity against TB. Therefore, it is important to elucidate functional characteristics and regulatory pathways for Mtb-specific CD4+ and CD8+ T cells during immune responses to Mtb infection. It was postulated that T-cell immunoglobulin and mucin domain-containing molecule 3 (Tim-3) might represent a T-cell exhaustion marker, and expression of Tim-3 on T cells may be linked to progressive loss of secretion of cytokines. Thus, Tim-3 expression on T cells might correlate with T-cell dysfunction and disease pathogenic events. However, T-cell immune responses modulated by Tim-3 in human TB disease remain poorly understood. Here, we found that up-regulation of Tim-3 expression in active human TB disease allows CD4+ and CD8+ T cells to mount stronger, but not impaired, anti-mycobacterium effector functions. This study therefore uncovers a previously unknown mechanism for T-cell immune responses regulated by Tim-3, and has an important implication for TB diagnostics and therapy.
| Tuberculosis (TB), an infectious disease caused by Mycobacterium tuberculosis (Mtb) infection, remains a leading cause of morbidity and mortality worldwide [1]. CD4+ and CD8+ T cells may be important for host immune resistance to TB in humans [2], [3], [4], [5]. In mouse models of Mtb infection, IFN-γ and TNF-α produced by CD4+ and CD8+ T cells have been shown to be critical for immune control of Mtb infection [2], [4], [5]. In addition, CD8+ T cells may contribute to anti-Mtb immunity through releasing bactericidal molecule granulysin or cytotoxic molecules perforin and granzymes killing of Mtb-infected target cells [2], [4], [5], [6]. It is likely that CD4+ and CD8+ T-cell effector functions producing Th1 or cytotoxic cytokines are required to mount anti-mycobacterial immunity [2], [4], [5]. In this context, insufficiency or failure to mount anti-mycobacterial effector functions by CD4+ and CD8+ T cells may lead to impaired immunity against TB [2]. Therefore, it is important to elucidate functional characteristics and regulatory pathways for Mtb-specific CD4+ and CD8+ T cells during immune responses to Mtb infection.
T-cell immunoglobulin and mucin domain-containing molecule 3 (Tim-3) is a membrane protein initially identified as a negative regulator of Th1 immunity in mice [7], [8], [9]. It was postulated that Tim-3, like other members of T-cell inhibitory molecules such as programmed death 1 (PD-1) [10], [11], [12] and co-stimulatory receptor cytotoxic T-lymphocyte antigen-4 (CTLA-4) [13], might represent a T-cell exhaustion marker [12], [14], [15], [16], [17], [18], [19], [20]. A number of studies have suggested that abundant expression of Tim-3 on T cells may be linked to progressive loss of secretion of cytokines such as IL-2, TNF-α and IFN-γ in viral infections [15], [16], [19], [21], [22] or tumors [23], [24]. Thus, Tim-3 expression on T cells might correlate with T-cell dysfunction and disease pathogenic events. We have recently shown that Mtb infection can induce significant up-regulation of Tim-3 expression in macaques [25], suggesting that Tim-3 might be involved in host immune responses during Mtb infection in primates. However, it is not known whether Tim-3 expression or Tim-3 pathway plays a role in modulating T-cell immune responses during Mtb infection in macaques and humans. Elucidating how Tim-3 regulates anti-Mtb effector functions of CD4+ and CD8+ T cells in human TB will help to understand TB immunopathogenesis and have some implications for immune intervention in TB.
Given that Mtb-specific CD4+ and CD8+ T-cell responses are important for anti-mycobacterial immunity, and that Mtb infection drives up-regulation of Tim-3 capable of regulating T-cell effector functions, we hypothesize that Tim-3-expressing CD4+ and CD8+ T-cell subpopulations may play a role in modulating host immune responses during Mtb infection in humans. In the current study, we examined the expression of Tim-3 in TB patients in the context of functional characteristics of Mtb-specific CD4+ and CD8+ T cells. Surprisingly, we found that Tim-3-expressing CD4+ and CD8+ T cells in active TB patients exhibit polarized effector memory phenotypes and stronger but not impaired anti-mycobacterium effector functions. Our findings therefore may implicate a new paradigm for T-cell immune responses regulated by Tim-3 expression in human TB.
We previously demonstrated that Mtb infection induced up-regulation of Tim-3 expression in nonhuman primates [25]. As an initial step to characterize potential roles of Tim-3 expression in human TB, we performed ex vivo polychromatic flow cytometric analysis in 9 healthy controls (HCs), 30 subjects with latent TB infection (LTBI), and 30 untreated active TB patients (Clinical characteristics of the enrolled subjects with LTBI or active TB disease were shown in Supporting information, Table S1). Active TB was confirmed based on assessment of clinical syndromes, chest radiography, and acid-fast bacilli (AFB) staining in sputum, culture isolation of Mtb and PCR detection of Mtb genes [26]. Peripheral blood mononuclear cells (PBMC) isolated from healthy controls (HCs) or subjects with LTBI or active TB disease were stimulated ex vivo first with pooled Mtb Antigen 85-b (Ag85-b)/6 kDa early secretory antigenic target (ESAT-6) peptides (Mtb peptide pool) and then stained for Tim-3 and Gal-9, the only known ligand of Tim-3, or stained directly for Tim-3 or Gal-9 without peptide stimulation. In agreement with TB-driven up-regulation of Tim-3 and Gal-9 in Mtb-infected macaques [25], we found up-regulation of Tim-3 and Gal-9 and significant increases in numbers of Tim-3 or Gal-9-expressing CD4+ and CD8+ T cells in active TB patients when compared with HCs(Figure 1 and Supporting Information, Figure S1). Interestingly, Mtb peptide stimulation drove further increases in numbers of Tim-3 or Gal-9-expressing CD4+ and CD8+ T cells (Figure 1 and Supporting Information, Figure S1). The mean percentages of Tim-3-expressing CD4+ (or CD8+) T cells in PBMC of active TB disease and LTBI after Mtb peptide stimulation were increased approximately 9.3% and 3.0% (or 8.9% and 2.8%) more than those without Mtb peptide stimulation, respectively (Figure 1).These results suggested that active TB up-regulated Tim-3 and drove increases in numbers of Tim-3+CD4+ and CD8+ T cells.
To characterize phenotypic and functional profiles of Tim-3-expressing CD4+ and CD8+ T cells, we first examined whether Tim-3 was predominantly expressed in effector memory or central memory/naïve T-cell subsets in active TB. Blood CD4+ or CD8+ T cells from untreated active TB patients (n = 9) were co-stained for Tim-3 and naïve/memory markers in presence or absence of ex vivo antigenic stimulation by Mtb peptide pool, and then analyzed by polychromatic flow cytometry. Based on differential expression of CD45RA and CCR7 [27], [28], [29], 4 distinct T cell populations were classified as naïve T cells (Tnaive, CD45RA+CCR7+), central memory T cells (TCM, CD45RA−CCR7+), effector memory T cells (TEM, CD45RA−CCR7−), and RA+TEM (TEMRA, also known as terminally differentiated; CD45RA+CCR7−) cells. We found that most of Tim-3-expressing CD4+ and CD8+ T cells freshly isolated from TB patients or stimulated ex vivo with Mtb peptide pool displayed CD45RA−CCR7− TEM phenotype, but not CD45RA+CCR7+ Tnaïve or CD45RA−CCR7+ TCM phenotype(Supporting Information, Figure S2), suggesting a polarization of effector memory phenotype for Tim-3-expressing CD4+ and CD8+ T cells in active TB patients.
Consistently, Tim-3-expressing CD4+ and CD8+ T cells expressed low levels of CD27 (Supporting Information, Figure S3A, B, C) and CD62L (Supporting Information, Figure S3D, E, F). In contrast to a lack of CCR7, CD62L, and CD27 expression, however, we found that Tim-3-expressing CD4+ and CD8+ T cells expressed high levels of CD127, another effector memory surrogate marker (Supporting Information, Figure S3G ,H, I). Furthermore, we determined whether Tim-3-expressing CD4+ and CD8+ T cells expressed CD27−CD45RA− effector memory phenotype by co-staining Tim-3, CD27 and CD45RA on CD3+CD4+ and CD3+CD8+ T cells, respectively. We found that Tim-3-expressing CD4+ and CD8+ T cells preferentially displayed CD27−CD45RA− effector memory phenotype during active TB disease (Supporting Information, Figure S4A, B, and C). Interestingly, we found that most of Tim-3-expressing CD4+ and CD8+ T cells during LTBI displayed CD27−CD45RA−effector memory phenotypes (Supporting Information, Figure S4 D, E, and F and data not shown). These results therefore demonstrated that Tim-3-expressing CD4+ and CD8+ T cells in active TB disease or LTBI preferentially displayed effector memory phenotypes.
Since Mtb-specific CD4+ and CD8+ T cells displaying effector memory phenotypes might be able to exert strong anti-mycobacterial effector functions [30], [31], we examined whether Mtb-driven Tim-3-expressing CD4+ and CD8+ T cells exhibited potent effector functions of cytokine production. We used two approaches to determine the relationship between Tim-3 expression and cytokine responses of CD4+ and CD8+ T cells: (i) PBMC from untreated active TB patients (n = 9) were stimulated ex vivo with Mtb peptides pool, and then stained for Tim-3 and anti-Mtb effector cytokines including IFN-γ, TNF-α, IL-2, and IL-22 and analyzed by polychromatic flow cytometry. This allowed us to evaluate correlation between TB-driven Tim-3 expression and Ag-stimulated cytokine responses of CD4+ and CD8+ T cells. (ii) PBMC from the same TB patients (n = 9) were directly stained for Tim-3 and the above cytokines without peptide stimulation as we recently described [32] to examine correlation between Tim-3 expression and an ability of Tim-3-expressing T cells to de novo produce cytokines. The specificity and utility of the direct intracellular cytokine staining approach has been validated during Mtb infection of macaques and humans as well as in the control settings [32], [33], [34].
Interestingly, in the absence of Mtb peptide stimulation, approximately 18–21% of Tim-3HighCD4+ T-cell subset from active TB patients were able to produce IFN-γ(Figure 2A, B), TNF-α (Supporting Information, Figure S5A, B), IL-2(Supporting Information, Figure S5C, D), and IL-22(Supporting Information, Figure S5E, F). In the presence of Mtb peptide stimulation, 30–34% of Tim-3HighCD4+ T-cell subset could produce those cytokines(Figure 2A,B, Figure S5C,D,E,F). In contrast, Tim-3LowCD4+ T-cell subset did not produce appreciable levels of those cytokines regardless of peptide stimulation(Figure 2A,2B, and Supporting Information, Figure S5A,3B, Figure S5C,D, Figure S5E,F). Similarly, high percentages of Tim-3HighCD8+ T-cell subset, but not Tim-3LowCD8+ T-cell subset, from these subjects with active TB disease were able to produce IFN-γ+(Figure 2A,C), TNF-α+(Supporting Information, Figure S6A, B), IL-2+(Supporting Information, Figure S6C, D) in the presence or absence of ex vivo stimulation with Mtb peptide pool. Interestingly, only ≤2.2% and ≤3.5% cells in Tim-3HighCD4+(or CD8+) T cells from subjects with LTBI were able to produce cytokines in the absence and presence of Mtb peptide, respectively (Figure 2D,E,F, Figure S5, and Figure S6), implicating that TB inflammation or disease course contributed to large increase in Tim-3High T effector cells actively producing cytokines. Notably, Mtb peptide stimulation led to ∼10%(∼12%) and ∼1%(∼1.2%) more increases in cytokine-producing cells in Tim-3HighCD4+(or CD8+) T-cell subset in active TB and LTBI, respectively, when compared to the culture without peptide stimulation (Figure 2, Figure S5, Figure S6), These results suggested that the at least some of Tim-3+ T-cell effector cells were Mtb-specific. Thus, these results demonstrated that consistent with effector phenotypes, Tim-3High CD4+ and CD8+ T-cell subsets exhibited greater effector functions for producing TB-driven IFN-γ, TNF-α, IL-2 and IL-22 cytokines than their Tim-3Low counterparts.
To further characterize the Tim-3 expression and T-cell effector function, Tim-3High T-cell subsets were assessed for the ability to produce cytotoxic effector molecuels perforin and granzyme B in comparisons with their Tim-3Low counterparts. Two above approaches were similarly used in active TB patients (n = 9) and individuals with LTBI (n = 9). We found that much higher percentages of Tim-3High CD4+ and CD8+ T-cell subsets produced perforin (Figure 3) and granzyme B (Supporting Information, Figure S7) than their Tim-3Low controls regardless of presence or absence of Mtb peptide stimulation in either active TB disease or LTBI. Similarly, we observed that, compared with those without peptide stimulation, Tim-3HighCD4+ and Tim-3HighCD8+T-cell subsets showed stronger T-cell effector function of perforin and granzyme B production upon Mtb peptide stimulation (Figure 3 and Supporting information, Figure S7). To evaluate CTL-related degranulation capacity, Tim-3High T-cell subsets were assessed for CD107a expression on cell surface, as CD107a, a lysosome-associated membrane glycoprotein, is expressed on cell surface following release of the cytotoxic granule contents, and is usually considered a hallmark for the degranulation capacity of CTL [20]. When compared with Tim-3Low controls, Tim-3High subsets of CD4+ and CD8+ T cells from individuals with either active TB disease (n = 9) expressed much higher levels of surface CD107a(Supporting Information, Figure S8). Similar higher levels of CD107 expression in Tim-3HighCD4+ and Tim-3HighCD8+ T-cell subsets were also observed in PBMC from individuals with LTBI (data not shown). Thus, Tim-3HighCD4+ and Tim-3HighCD8+ T-cell subsets in active TB disease and LTBI demonstrated greater effector functions of production of Mtb-driven cytotoxic molecules and degranulation than their Tim-3Low counterparts.
To further explore the relationship between Tim-3 expression and effector functions of Tim-3-expressing T cells, we made use of siRNA targeting Tim-3 to knockdown the expression of Tim-3, and evaluated the effects of Tim-3 silencing on production of IFN-γ and TNF-α. siRNA targeting Tim-3(si-Tim-3), but not nontargeting siRNA (si-Control), could significantly knockdown Tim-3, as determined by RT-PCR (Figure 4A) and flow cytometry (Figure 4B, C, F, G). More importantly, silencing of Tim-3 by siRNA in CD4+ and CD8+ T cells led to significant decreases in production of IFN-γ and TNF-α (Figure 4B, D, E, F, H, I), when compared to the controls. These results suggested that Tim-3 expression in active TB was linked to potent IFN-γ and TNF-α responses of CD4+ and CD8+ T cells.
On the other hand, we used Tim-3 ligand competition approach [19] to inhibit Tim-3 signaling pathways and examine the role of Tim-3 signaling in effector function of IFN-γ and TNF-α production by CD4+ and CD8+ T cells. PBMC from active TB patients (n = 9) were cultured in the presence of low concentration of soluble Tim-3 [termed s-Tim-3, i.e. Tim-3-Ig [7]] under the conditions with or without Mtb peptide pool stimulation. Interestingly, addition of low concentration of s-Tim-3 (2 µg/ml) for interfering with membrane Tim-3-Tim-3 ligand interaction significantly reduced the ability of Tim-3High T-cell subsets to produce IFN-γ and TNF-α cytokines (Figure 5). These results suggested that soluble Tim-3 interfering with membrane Tim-3-ligand interaction could impact Tim-3 signaling pathways leading to decreases in effector function of IFN-γ and TNF-α production by Tim-3High CD4+ and CD8+ T cells. However, cytokine-producing T cells in the culture of s-Tim-3 plus Mtb peptide were significantly higher than those of s-Tim-3 alone, suggesting again that some of cytokine-producing Tim-3+CD4+ or CD8+ T cells might be specific for Mtb antigens. The data also supported the notion that Tim-3 expression in active TB helped to drive stronger anti-microbial effector functions of CD4+ and CD8+ T cells.
We then examined whether enhancing of Tim-3 signaling could augment effector function of IFN-γ production by CD4+ and CD8+ T cells. Because anti-Tim-3 mAb could cross-link membrane Tim-3 and enhance Tim-3 signaling [35], [36], we used this approach to enhance Tim-3 signaling. PBMC from active TB patients (n = 9) were incubated with anti-Tim-3 mAb or isotype control Ab in presence or absence of Mtb peptide pool, and then assessed for potential effects of enhancing Tim-3 signal on IFN-γ production by CD4+ and CD8+ T cells. Interestingly, addition of anti-Tim-3 mAb, but not control Ab, to PBMC cultures resulted in significant increases in effector function of IFN-γ production by Tim-3High CD4+ and CD8+ T-cell subsets (Supporting Information, Figure S9). Similar enhancement of T-cell effector functions by cross-linking Tim-3 on T-cell surface was also observed in PBMC of HCs and subjects with LTBI (data not shown). These results provided additional support for the hypothesis that Tim-3 signaling in CD4+ and CD8+ T cells helped to promote effector function of Th1 cytokine production.
While Tim-3 expression could regulate immune status of MΦs [35], our results implicated that Tim-3 expression enhanced T-cell effector functions for producing anti-microbial cytokines and cytotoxic molecules. Because it has also been suggested that IL-1β might contribute to the Tim-3-mediated inhibition of intracellular Mtb growth [37], we then asked whether Tim-3-expressing T cells could limit intracellular Mtb replication in MΦs, and whether IL-1β played a role in Tim-3+ T cell-induced anti-Mtb effector function. To address these questions, Tim-3-expressing CD3+ T cells (Tim-3+CD3+ T cells) were isolated from PBMCs of active TB patients(n = 9) using magnetic beads techniques, as we described [34], and co-cultured with Mtb-infected autologous MΦs in presence or absence of anti-IL-1β Ab or isotype control Ab. The co-cultured cells were then lysed, and lysate was measured for Mtb CFU counts on plates as we previously described [34]. Tim-3− CD3+ T cells served as control. Interestingly, Tim-3-expressing T cells more apparently limited intracellular Mtb growth than Tim-3− CD3+ T cells (Figure 6). Furthermore, treatment with anti-IL-1β Ab, but not isotype control Ab, could reverse the Tim-3+CD3+ T cell-mediated inhibition of intracellular Mtb growth, suggesting that IL-1β might contribute to the limitation of intracellular Mtb growth mediated by Tim-3+ CD3+ T cells. Thus, Tim-3-expressing T cells in active TB patients appeared to mount stronger anti-Mtb effector function limiting intracellular Mtb replication in cultured MΦs.
Given that Tim-3High CD4+ and CD8+ T-cell subsets exhibited greater effector functions of producing Th1 cytokines/cytotoxic molecules and limiting intracellular Mtb growth, we sought to determine a potential signaling mechanism underlying the Tim-3-associated enhancements. Since appropriate activation signaling is usually required for efficient T-cell effector functions in responses to Mtb infection [38], we hypothesized that greater T effector functions of Tim-3High CD4+ and CD8+ T-cell subsets during Mtb infection might be driven by stronger intracellular signaling and activation. To test this hypothesis, we measured the phosphorylation of signaling molecules p38, stat3, stat5, and Erk1/2 in Tim-3High CD4+ and CD8+ T-cell subsets from active TB patients in comparisons with Tim-3Low control subsets. Thus, PBMC from 9 active TB patients were directly stained without ex vivo Mtb peptide stimulation or stimulated ex vivo with Mtb peptide pool, and the phosphorylated (termed P- for simplicity) p38, stat3, stat5, and Erk1/2 were immunologically stained and quantitated by flow cytometry. We found that expression levels of P-p38, P-stat3, P-stat5, and P-Erk1/2 in Tim-3High CD4+ and CD8+ T-cell subsets were much higher than those in Tim-3Low control subsets in cultures with or without Mtb peptide antigen stimulation (Figure 7). The finding that unstimulated Tim-3+, but not Tim-3−, T cells had higher levels of phosphorilated signal molecules was consistent with the ability of Tim-3+ T cells to de novo produce cytokines(Figure 2,3, Figure S5,S6) and to exert anti-Mtb effector function(Figure 6). Furthermore, expression levels of P-p38, P-stat5, and P-Erk1/2 were much higher than that of P-stat3 within Tim-3High CD4+ and CD8+ T-cell subsets (Figure 7). Thus, Tim-3High CD4+ and CD8+ T-cell subsets in active TB patients expressed higher levels of phosphorylated signaling molecules, suggesting that Tim-3-associated increases in T effector functions may involve activation signaling molecules p38, stat5, and Erk1/2 rather than stat3.
In the current study, we have made several interesting observations regarding Mtb infection-induced increases in Tim-3-expressing CD4+ and CD8+ T-cell subsets, Tim-3-related broad effector functions for producing Th1/Th22/cytotoxic cytokines and limiting intracellular Mtb growth, and potential mechanisms underlying Tim-3 signaling-driven enhancements of effector functions. These findings are considered novel as there are no reports, to our knowledge, for in-depth studies of Tim-3-driven T-cell immune responses in active human TB.
Interestingly, active TB patients exhibit up-regulation of Tim-3 expression and increases in Tim-3-expressing CD4+ and CD8+ T cells, and Tim-3-expressing T cells predominantly displayed a polarized effector memory phenotype (lack of expression of CCR7, CD62L, or CD27) [31]. Consistently, Tim-3-expressing Mtb-specific CD4+ and CD8+T cells in active TB patients also express high levels of another effector memory surrogate marker CD127 (IL-7 Receptor α) [39]. It is likely that effector memory phenotypes (CD127+ but CCR7−, CD62L− and CD27−) not only favor Tim-3-expressing T cells for mounting effector functions of anti-microbial cytokine production and cytotoxicity but also facilitate these effector T cells trafficking to Mtb infection sites or inflamed lung tissues [31], [39], [40]. It is noteworthy that the predominant effector memory phenotypes of Mtb-driven Tim-3-expressing CD4+ and CD8+ T cells in active TB patients differ from virus-induced central memory phenotypic Tim-3-expressing CD8+ T cells in HCV- or HIV-1-infected humans [16], [19], [20]. It has been reported that Tim-3-expressing CD8+ T cells during HCV or HIV-1 infection display either dominant CCR7+ central memory or CCR7+ and/or CD27+ phenotype profiles with no or low expression of effector surrogate marker CD127 [16], [19], [20]. The discrepancy of Tim-3-expressing phenotypes between Mtb-infected and HCV-infected or HIV-1-infected patients might result from the natures of pathogens/infections and distinct immune responses to virus versus mycobacteria. A dominant central memory and a lack of preferential effector phenotypes for Tim-3-expressing T cells in HCV/HIV-1-infected patients may help to explain why the virus-driven Tim-3-expressing CD8+ T cells do not adequately produce effector cytokines in response to viral peptide stimulation in vitro [16], [19], [20]. In contrast, the predominant effector memory phenotypes of Tim-3-expressing CD4+ and CD8+ T cells in active TB patients are indeed consistent with enhanced effector functions for these Tim-3HighCD4+ and Tim-3HighCD8+ T-cell subsets.
The ability of Tim-3+ T cells to spontaneously produce cytokines in active TB appears to consist with our recent observation that active Mtb infection, but not SIV/SHIV infection or control setting, allows for intracellular cytokine staining (ICS) detection of cytokine production by T cells without the need for ex vivo stimulation with Mtb antigens [32], [33], [34]. Such an ability to de novo produce cytokines by T cells from Mtb-infected humans or macaques might be due to the fact that a number of T cells have differentiated into highly activated Tim-3+ effector cells capable of producing cytokines in response to active TB-driven immune activation and inflammation. This notion is also supported by the finding that latent Mtb infection did not induce large numbers of T cells that spontaneously produce cytokines in cultures, as only ≤2.2% and ≤3.5% cells in Tim-3+ T cells of LTBI subjects were able to produce cytokines after culture with medium and Mtb peptide, respectively (Figure 2, Figure 3). The high levels of Tim-3+/cytokine+ T cells in active TB might be attributed partially to the BCG vaccination background in active TB patients, as all these patients had BCG vaccination history.
The present study uncovers a surprising finding that Tim-3High CD4+ and Tim-3High CD8+ T-cell subsets exhibited greater effector functions for producing Th1/Th22 cytokines and CTL effector molecules. These enhanced effector functions of Tim-3-expressing T cells in TB are not totally unexpected as anti-microbial effector functions of T cells are usually linked to effector or effector memory phenotypes of these T cells [30], [31], and such enhanced effector functions appear to reflect the immunological features of effector memory phenotypes for Tim-3High CD4+ and CD8+ T-cell subsets. Furthermore, enhanced effector functions of Tim-3-expressing T cells are also supported by the results from our mechanistic studies of cellular activation/signaling molecules as Tim-3High CD4+ and CD8+ T-cell subsets in active TB patients expressed higher levels of phosphorylated signaling molecules p38, stat3, stat5, and Erk1/2. Despite the fact that the downstream molecular activation events for Tim-3 regulation of T cells are largely unknown [41], the up-regulated expression of phosphorylated signaling molecules in Tim-3High T-cell subsets may help to explain stronger effector functions in Tim-3-expressing CD4+ and CD8+ T cells in active TB. The notion that stronger effector functions are driven by greater signaling in Tim-3-expressing T cells is also consistent with a recent finding that Tim-3 may augment T-cell signaling after a short-term stimulation [41]. This connection is also supported by the data from HIV-1-infected humans since impaired phosphorylation of above intracellular signaling molecules correlates with reduced effector functions of Tim-3-expressing CD8+ T cells in HIV infection [19]. A recent study reported that Tim-3-expressing CD8+ T cells in TB patients produced lower levels of IFN-γ than healthy controls [18]. However, subtle IFN-γ responses were detected in both TB and control groups due to the use of ESAT-6 protein, instead of peptide pool, for in vitro stimulation [18]. It is important to note that CD8+ T cells respond poorly to whole protein and that recombinant Mtb ESAT-6 protein actually inhibits T-cell production of IFN-γ [42]. Future studies using MHC I/peptide and MHC II/peptide tetramers will provide a better system in which to elucidate phenotypes and effector functions of Ag-specific Tim-3+CD4+ and Tim-3+CD8+ T cells.
Our results suggest that Tim-3-expressing CD4+ and CD8+ T-cell subsets possess much broader repertoire of effector functions than what was previously described. Earlier studies implicated that Tim-3 might be an exclusive cell surface marker for Th1 cells [9], and it remains unknown whether Tim-3-expressing CD4+ T cells in human TB could differentiate into Th22 and Th17 subsets capable of mounting immune responses to Mtb infection [32], [34]. The current study demonstrates that Tim-3-expressing CD4+ T cells not only can produce Th1 cytokines(IFN-γ, IL-2), but also produce appreciable amounts of IL-22, IL-4 (data not shown), and IL-17A (data not shown), suggesting that Tim-3High CD4+ T-cell subset in active TB are capable to differentiate into Th1, Th2, Th22/Th17 cells. Interestingly, Tim-3High CD8+ T-cell subset in active TB also exhibits broad effector functions producing the above cytokines and cytotoxic molecules. Furthermore, human Tim-3-expressing T cells can function as anti-Mtb effector cells limiting intracellular Mtb growth. Our results are consistent with a recent observation that mouse Tim-3-Gal-9 interaction can lead to inhibition of Mtb replication in macrophages [37]. The broad effector repertoires of Tim-3-expressing CD4+ and CD8+ T cells might be advantageous from the standpoints of host immune responses to Mtb infection.
The findings from our mechanistic experiments suggest that Tim-3 signaling pathways help to enhance effector functions of producing Th1, Th22 cytokines and CTL molecules. Particularly, we show that Tim-3 silencing by siRNA Tim-3 leads to reduced de novo production of IFN-γ and TNF-α by Tim-3-expressing T cells, and that soluble Tim-3 treatment interfering with membrane Tim-3-ligand interaction can also decrease Tim-3-driven activation and effector functions of cytokine production. On the other hand, we demonstrate that stimulation of Tim-3 signaling pathway by Ab cross-linking of membrane Tim-3 can enhance effector function of IFN-γ production by CD4+ and CD8+ T cells. These findings appear to be inconsistent with what were reported in HCV- and HIV-1-infected patients [16], [19]. It has been implicated that expression of Tim-3 on CD8+ T cells may be linked to progressive loss of secretion of Th1 cytokines such as IL-2, TNF-α and IFN-γ in HCV and HIV-1 infections [15], [16], [19], [21], [22]. Nevertheless, these Tim-3-associated negative effects can be explained at least partially by the phenotypic features and impaired activation signaling of Tim-3-expresssing CD8+ T cells in those virus-infected persons. Tim-3-expressing CD8+ T cells in HCV- and HIV-1-infected patients predominantly express central memory phenotypes, rather than effector and effector memory phenotypes [16], [19]; HIV-1 infection leads to impaired Stat5, Erk1/2, and p38 signaling in Tim-3-expressing CD8+ T cells [19]. On the contrary, active TB drives predominant effector memory phenotypes and stronger cellular activation signaling. It is likely that HCV or HIV-1 infection preferentially induces Tim-3-associated central memory or non-effector phenotypes with depression or low levels of cellular signaling, whereas Mtb infection can drive effector memory Tim-3-expressing T cells with enhanced Tim-3 signaling pathways for stronger effector functions. It is noteworthy that studies done to date have only identified Gal-9 as Tim-3 ligand [17], and precise Tim-3-induced signaling pathways remain incompletely understood [17]. From these points of views, we cannot exclude the possibility that viral and Mtb infections would engage in independent co-activation signals in T cells and induce potential different Tim-3 ligands in infected target cells.
It is currently not known whether increased numbers of Tim-3High CD4+ and CD8+ T effector cells in active TB patients are detrimental or beneficial in active Mtb infection. Given that such increases are seen in the setting of active TB, Tim-3-expressing T cells might act as over-reactive effector cells and contribute to TB inflammation and pathologic lesions. This notion is supported by the finding that healthy subjects with LTBI exhibited much lower levels of Tim-3+CD4+ and Tim-3+CD8+ T effector cells producing cytokines. Over production of IFN-γ and TNF-α by Tim-3-expressing T cells may indeed elevate degree of inflammation or damages in active TB, although mouse IFN-γ and TNF-α are important for controlling Mtb infection [30], [43], [44], [45], [46], [47], [48], [49]. On the other hand, increases in Tim-3-expressing CD4+ and CD8+ T cells might result from host responses to high Mtb burden due to postprimary TB or reactivation TB. These responses, although unable to control TB, might develop as disorganized host defense or otherwise reflect protective potential if immune responses to TB can be well coordinated or if pathogenic events leading to high levels of Mtb burden can be intervened. This scenario appears to be supported by the data from the current study since Tim-3-expressing T cells can produce anti-Mtb cytokines IFN-γ/TNF-α/IL-22, and function as effector cells limiting intracellular Mtb growth in macrophages.
Thus, the current study demonstrate that Tim-3-expressing CD4+ and CD8+ T cells in active TB patients exhibit polarized effector memory phenotypes and stronger, but not impaired, anti-mycobacterium effector functions. Our findings therefore may suggest a new paradigm for T-cell immune responses regulated by Tim-3 expression in human TB, and have implications for potential immune intervention in TB.
The active Mtb infection in patients was confirmed based on clinical symptoms, chest radiography, and sputum stain for acid-fast bacilli (AFB), culture and PCR for Mtb, which were done in Shenzhen Third People's Hospital. Subsequently, patients confirmed with Mtb infection received individualized regimens with rifampicine and isoniazide plus either streptomycin or ethambutol. After initiation of TB treatment, patients were evaluated again clinically and bacteriologically to determine the effectiveness of the therapy and the transition of disease. Healthy controls (HCs) are a cohort of individuals negative for tuberculin skin test (TST) with no bacteriological and clinical evidence of TB disease. Subjects with LTBI are a cohort of individuals strongly positive for TST with no bacteriological and clinical evidence of active TB disease. All samples of Mtb-infected individuals or healthy controls were collected with informed written consent according to protocols approved by the Internal Review and the Ethics Boards of Shenzhen Third People's Hospital and Zhongshan School of Medicine of Sun Yat-sen University.
Abs against the following molecules were used: CD3-FITC, CD3-PE, CD3-APC, or CD3-PE/Cy7 (Clone OKT3, ebioscience); CD4-APC, CD4-PE, CD4-PE/Cy7, or CD4-Biotin (Clone RPA-T4, BD); CD8-APC, CD8-PE, CD8-PE/Cy7, or CD8-Biotin (Clone RPA-T8, BD); Tim-3-PE, Tim-3-Alexa fluor488 (Clone 344823, R&D); CCR7(CD197)-FITC (Clone 3D12, ebioscience), CD27-APC (Clone O323, ebioscience), CD45RA-PE/Cy7 (Clone HI100, ebioscience), CD127-FITC (Clone eBioRDR5, ebioscience), CD62L-PE (Clone DREG56, ebioscience); TNF-α-APC, TNF-α-FITC(Clone Mab11, ebioscience); IFN-γ-FITC, IFN-γ-APC (Clone 4s.b3, ebioscience); Granzyme B-FTIC (Clone GB11, BD), perforin-FITC (Clone deltaG9, ebioscience); IL-2-FITC (Clone MQ1-17, BD), IL-4-APC (Clone 8D4-8, ebioscience), IL-17a(Clone eBio64DEC17, ebioscience), IL-22-PE (Clone 22URTI, ebioscience), PE-anti-Galectin-9(Clone 9M1-3,biolegend), anti-IL-1β (Clone AS10, BD). Streptavidin-Phycoerythrin-Texas Red (BD) was used to conjugate CD4-biotin or CD8-biotin antibody. Recombinant human Tim-3 Fc Chimera (i.e.Tim-3-Ig) and the purified Ab for Tim-3 (Clone 344823) were both from R&D. Overlapping Mtb Ag85-b/ESAT-6 pooled peptides(15 a.a. overlapped by 12 spanning entire Ag85 or ESAT6 protein were synthesized and used as we previously described [34]. The purpose choosing combined peptides for ICS was to maximize detection of Mtb-specific T effector cells and to optimally work with a limited amount of blood volume collected from individual subjects.
PBMC were isolated from whole blood by Ficoll (GE health) density gradient centrifugation, as we described previously [32]. Intracellular cytokine staining (ICS) was done as we previously described [32]. We used two approaches for ICS: (i) PBMC from untreated active TB patients were stimulated ex vivo with Mtb peptides pool, and then stained for Tim-3 and anti-Mtb effector cytokines including IFN-γ, TNF-α, IL-2, and IL-22 and analyzed by polychromatic flow cytometry.. (ii) PBMC from the same TB patients were directly stained for the above cytokines without peptide stimulation as we recently described [32], [34]. The specificity and utility of the direct intracellular cytokine staining approach has been validated during Mtb infection of macaques and humans as well as in the control settings [32], [33], [34]. Briefly, PBMC were cultured with or without re-stimulation with Mtb Ag85-b/ESAT-6 peptide pools for 6 hours in presence of brefeldin A (5 µg/ml; BD) in the final 3 hours of culture. Cells were then fixed, permeabilized and washed with the Perm/Wash buffer (BD). For PBMC without in vitro antigenic re-stimulation, cells were fixed, permeabilized and washed with the BD Perm/Wash buffer. After permeabilization, cells were stained using fluochrome-conjugated mAbs or isotype control Abs. Data were acquired on Beckman Coulter Cytomics FC500 (Beckman) and analyzed with CXP (Beckman) software.
PBMC derived from TB patients were incubated with 2 µg/ml soluble form of Tim-3 molecules (human Tim-3 Fc Chimera, purchased from R&D) in presence or absence of Mtb Ag85-b/ESAT-6 pooled peptides for 6 days. Production of IFN-γ and TNF-α by Tim-3-expressing T cells were then analyzed using ICS protocol and flow cytomery.
PBMC from subjects with active TB patients were incubated with 10 µg/ml of purified mouse anti-human mAb against Tim-3 (purchased from R&D) or isotype control IgG (10 µg/ml) in presence or absence of Mtb Ag85-b/ESAT-6 pooled peptides for 6 days. The effects of Tim-3 mAb stimulation on the production of IFN-γ and TNF-α by Tim-3-expressing T cells were then analyzed using ICS protocol and flow cytometry.
PBMC from active TB patients were stimulated with Mtb Ag85-b/ESAT-6 pooled peptides for 30 mins. After fixation and subsequent washing, cells were permeabilized with Perm/Wash buffer (BD). Cells were washed and stained with isotype control Ab or fluochrome-conjugated phosphospecific Abs: P-stat3 (pY705)-PE (Clone 4/P-STAT3, BD), P-erk1/2(pT202/pY204)-Alexa Fluor488 (Clone 20A, BD), P-p38 (pT180/pY182)-Alexa fluor 647 (Clone 36, BD), or P-stat5 (pY694)-FITC (Clone 47, BD).
PBMC derived from TB patients were transiently transfected with 20 nM siRNA targeting Tim-3 (si-Tim-3) or 20 nM nontargeting siRNA (si-control) using Lipofectamine 2000 (Invitrogen). siRNA targeting Tim-3 (si-Tim-3) and nontargeting siRNA (si-control) are commercially available from Ribobio (Guangzhou, China). Knockdown efficiency was analyzed by real-time PCR or flow cytometry 48 hours after transfection. At 2 days after transfection, PBMC were cultured in presence or absence of Mtb Ag85-b/ESAT-6 peptide pools for 6 days, and cytokine production was analyzed by ICS and flow cytometry.
Cell isolation was done as we described previously [34]. Briefly, PBMC were isolated from the blood of active TB patients, and monocytes were obtained by adherence purification on plastic plates. The plates were washed after 2 hours of adherence, and monocytes were detached by cold 2% FBS/PBS. The non-adherent cell fraction containing T cells was stained with anti-Tim-3-PE (ebioscience), followed by anti-PE magnetic beads (Miltenyi Biotec). The stained live cells were then loaded to the purification column following instructions from the manufacturer. The passing fraction was collected as Tim-3− (negative) cells that did not bear Tim-3; Tim-3+ T cells held by anti-PE magnetic beads were then released by releasing buffer (Miltenyi Biotec). The isolated Tim-3+ T cells were stained again with anti-CD3 FITC (BD), followed by anti-FITC magnetic microbeads for secondary purification. The purity of isolated Tim-3+CD3+ T cells and Tim-3−CD3+ T cells is over 95% ([34], data not shown).
This was done similarly like what we recently described [34]. Briefly, autologous monocytes (5×104/well) were cultured in round-bottom 96-well plates with 10% FBS-RPMI 1640 medium in presence of human rIL-4 (BD) and GM-CSF (Sigma-Aldrich) for 8 days. Supernatants were then removed, and Mtb (H37Ra) inoculum was added at a MOI = 1. After overnight infection at 37°C, supernatants were aspirated and each well was washed extensively to remove extracellular Mtb. Enriched Tim-3+CD3+ T cells (5×105/well) or Tim-3−CD3+ T cells (5×105/well) were incubated with Mtb-infected macrophages (MΦs) in presence or absence of anti-IL-1β Ab (10 µg/ml) or IgG (10 µg/ml). After culturing in 5% CO2 at 37°C for 4 days, wells were aspirated, and lysis buffer (0.067% SDS in Middlebrook 7H9) was added to each well. Plates were incubated at 37°C, followed by neutralization of SDS with PBS with 20% BSA. Lysates from each well were pooled, and two 10-fold serial dilutions of lysate in 7H9 medium were made. Aliquots of each dilution of lysate and supernatant were plated onto Middlebrook 7H10 agar and incubated until colonies were large enough to be counted.
Statistical significance was determined with Student t-test (difference between two groups or conditions), and a p value <0.05 in all cases was considered statistically significant (95% confidence interval), as we described previously [32]. Analysis was performed using Prism 5.0 software (GraphPad Software, Inc.).
|
10.1371/journal.pbio.2004880 | METTL3-mediated m6A modification is required for cerebellar development | N6-methyladenosine (m6A) RNA methylation is the most abundant modification on mRNAs and plays important roles in various biological processes. The formation of m6A is catalyzed by a methyltransferase complex including methyltransferase-like 3 (METTL3) as a key factor. However, the in vivo functions of METTL3 and m6A modification in mammalian development remain unclear. Here, we show that specific inactivation of Mettl3 in mouse nervous system causes severe developmental defects in the brain. Mettl3 conditional knockout (cKO) mice manifest cerebellar hypoplasia caused by drastically enhanced apoptosis of newborn cerebellar granule cells (CGCs) in the external granular layer (EGL). METTL3 depletion–induced loss of m6A modification causes extended RNA half-lives and aberrant splicing events, consequently leading to dysregulation of transcriptome-wide gene expression and premature CGC death. Our findings reveal a critical role of METTL3-mediated m6A in regulating the development of mammalian cerebellum.
| N6-methyladenosine (m6A) is an abundant modification in mRNA molecules and regulates mRNA metabolism and various biological processes, such as cell fate control, early embryonic development, sex determination, and diseases like diabetes and obesity. Adenosine methylation is regulated by a large methyltransferase complex and by demethylases, as well as by other binding proteins. METTL3 is one of the core subunits of the methyltransferase complex catalyzing m6A formation. However, the role of METTL3-mediated m6A in mammalian brain development remains unclear mainly because of the lack of specific spatiotemporal knockout animal models, as conventional METTL3 knockout in mice leads to early embryonic death. In this study, we specifically inactivated METTL3 in the developing mouse brain. We detected a drastic depletion of m6A accompanied by severe developmental defects in the cerebellum of these mice. Further analysis established that METTL3-mediated m6A participates in cerebellar development by controlling mRNA stability of genes related to cerebellar development and apoptosis and by regulating alternative splicing of pre-mRNAs of synapse-associated genes.
| Multiple layers of epigenetic modifications play essential roles in neuronal development and brain function in mammals through highly coordinated epigenetic regulatory mechanisms [1–5], such as DNA methylation and demethylation [6–8], histone modifications [9, 10], and noncoding RNAs [11–13]. Additionally, RNA modifications have recently been recognized as a new layer of epigenetic regulation, among which N6-methyladenosine (m6A) is the one being most extensively studied. m6A formation is catalyzed by a methyltransferase complex including methyltransferase-like 3 (METTL3), methyltransferase-like 14 (METTL14), and Wilms’ tumor 1-associating protein (WTAP), among which METTL3 functions as the catalytic subunit [14, 15]. As the most abundant and reversible modification on mRNAs, m6A has been proven to play key roles in regulating RNA stability [16] and RNA splicing [17], as well as mRNA translation efficiency [18–22].
Many essential biological processes are known to be regulated by m6A, including cell fate determination [23, 24] and embryonic development [24, 25, 26]. In neuronal systems, m6A has been shown to be a dynamic modification and increases in adulthood [27], suggesting a potential function in neural plasticity and brain function. Previous studies have shown that conventional knockout of Mettl3 in mice leads to early embryonic lethality [24], and specific depletion of m6A in the nervous system by conditional knockout of Mettl14 results in extended cell cycle and altered neurogenesis in embryonic mouse cortex, through regulating the decay of neurogenesis-related transcripts [28]. Furthermore, loss of m6A eraser, the fat mass and obesity-associated protein (FTO), leads to impaired adult neurogenesis and cognition ability [29]. FTO is enriched in axons, and specifically silencing axonal FTO inhibits axon elongation by suppressing growth associated protein 43 (GAP-43) [30]. Nerve lesion induces the elevation of m6A level, and loss of METTL14 or m6A binding protein YTH N6-methyladenosine RNA binding protein 1 (YTHDF1) inhibits axonal regeneration [31]. All these studies indicate the important roles of m6A signaling in neuronal development and neurogenesis.
The cerebellum offers an ideal model to study neurogenesis, as it contains a limited number of neuronal categories, and different types of cerebellar neurons are generated in a timely order, precisely scheduled by progenitors. The developing cerebellum consists of the external granular layer (EGL), molecular layer (ML), Purkinje cell layer (PCL), and internal granular layer (IGL) [32, 33]. It has been shown that cerebellar granule cell progenitors (GCPs) proliferate in EGL, then migrate through ML to IGL to differentiate into granule cells, generating parallel fibers and forming synaptic connections with Purkinje cells [33]. The EGL, as the exclusive source for cerebellar granule cells (CGCs) [32, 34], gradually becomes thinner and disappears around 21 days postpartum (P21) in mice [33]. Although many studies have indicated the function of epigenetic modifications in cerebellar development, little is known about the roles of m6A signaling in the development of the cerebellum and its associated behavioral phenotype. In the present work, we specifically inactivated Mettl3 in the developing mouse brain using the Nestin-Cre mediated Mettl3 conditional knockout model. Depletion of METTL3 in the brain caused severe developmental defects in both the cortical and cerebellar regions. Further analysis demonstrated drastic apoptosis of newborn CGCs, which is partially responsible for the severe cerebellar hypoplasia and a series of ataxia-like movement disorders. Furthermore, loss of m6A by inactivation of Mettl3 led to extended half-lives of mRNAs from cerebellar development- and apoptosis-associated genes, as well as aberrant mRNA splicing events on synapse-associated genes, which consequently induced inappropriate cell differentiation and cell death, indicating that METTL3-mediated m6A serves as a key regulator in modulating the development of the central nervous system at the posttranscriptional level in mammals.
Consistent with a previous report [35], we detected ubiquitous occupation of METTL3 in various brain regions by immunostaining (S1A Fig), indicating that METTL3 might be involved in the development of mammalian brains. To investigate the functions of METTL3-mediated m6A in brain development, we introduced two loxP sites into the Mettl3 genomic region flanking exons 2–4, a region encoding the methyltransferase catalytic domain in B6D2F1 mice by CRISPR/Cas9 system-assisted homologous recombination (S1B Fig). The resulting Mettl3flox/flox mice were crossed with Nestin-Cre transgenic mice to generate Mettl3 conditional knockout mice (genotype Mettl3flox/flox;Nestin-Cre, named as cKO) (S1B and S1C Fig). Genotyping results confirmed the specific deletion of Mettl3 exons 2–4 in the brain tissue of newborn cKO mice, whereas the Mettl3 coding sequence in their tail tissue remains intact (S1C–S1G Fig). Immunohistochemical staining (S1A Fig) and western blot (S1H Fig) further confirmed that the cKO mice failed to produce METTL3 protein and its isoforms in the whole brain, whereas in other nonneural tissues, such as liver, METTL3 protein remained unperturbed. Littermates of the cKO mice with the Mettl3flox/flox or Mettl3flox/+ genotypes were used as the controls (Ctrl).
The cKO pups had a significantly decreased body weight growth rate as compared to that of the Ctrl (Fig 1A) and were only approximately half the size of their Ctrl littermates by P14 (S2A Fig). The majority of cKO pups died before P20 without artificial feeding (Fig 1B). In addition, the cKO mice demonstrated balance disorders and altered gaits after P7 that became severe at around P14 (S1 Movie). When lifted by the tail, the cKO mice displayed dramatic tremors with tightly curled legs and forward bended head (Fig 1C and S2 and S3 Movies), demonstrating typical features of cerebellar ataxia [36]. We then analyzed the behavior of the mice in an open field test; the cKO mice only generated intermittent and slow movements with much reduced speed and moving distance, and were seldom capable of moving out of the central region (Fig 1D and 1E and S2B–S2D Fig). The smaller body size, premature death, and ataxia-like movement disorders indicated severe developmental defects in the cKO mouse brain. We then performed magnetic resonance imaging (MRI) analysis and identified smaller brain sizes and enlarged ventricles in the brains of Mettl3 cKO mice (S2E Fig), which is consistent with the results in another m6A depletion model via Nestin-Cre mediated conditional knockout of Mettl14 [28]. Furthermore, in addition to the cortex, we also observed shrunk and unstructured cerebellums in the Mettl3 cKO mice (Fig 1F–1H). It is tempting to speculate that the developmental defects in the cerebellums caused by depletion of METTL3 might contribute to the movement disorders of the cKO pups. Ultra high-pressure liquid chromatography tandem-mass spectrometry (UHPLC-MS/MS) analysis showed that the m6A modification on mRNAs of cKO mouse brain tissues were nearly wiped out (S2F Fig), which indicated that the brain developmental defects might be caused by m6A depletion.
A previous report has demonstrated that the m6A RNA methylation level in mouse cerebellum is generally higher than that in other brain regions [35]. Consistently, our western blot showed relatively higher expressions of METTL3 and METTL14 in the cerebellum than those in the cerebral cortex of wild-type mouse (S3A Fig). UHPLC-MS/MS analysis also detected twice the amount of m6A on mRNA from the cerebellum in comparison to that from other brain regions of the wild-type mouse (S3B Fig). We then focused on the effect of METTL3 depletion on the development of the cerebellum. We first systematically investigated the expression pattern of METTL3 in developing mouse cerebellum by immunohistochemical staining, and we found evident and ubiquitous immunostaining signals in the EGL, PCL, ML, and IGL of the cerebellum from embryonic day 16.5 (E16.5) to P14 (S4A–S4C Fig). Interestingly, the METTL3 expression level in the proliferating GCPs within the outer EGL was shown to be relatively lower than that of the newborn CGCs within the inner EGL at P7 (S4B Fig). We confirmed that METTL3 was depleted in the cKO cerebellums as early as E16.5, when no substantial structural defect was seen in the cerebellar region of the cKO mouse brains (S4A–S4C Fig). Macroscopic histological examination showed shrunken size, loss of weight, and remarkable failure in foliation of the cKO cerebellums (Fig 2A–2C). Further immunohistological analysis revealed drastic loss of CGCs in the IGL layer of the cKO cerebellums (Fig 2D and 2E and S4D Fig). In spite of the fact that the number of Purkinje cells in the cKO cerebellums remained almost the same as that in the Ctrl mice, they failed to maintain the well-organized single-cell layer (Fig 2D and 2F and S4D Fig). In addition, Purkinje cells in cKO cerebellums had much shorter dendrites and exhibited reduced calbindin 1 (CABL1, also known as D-28K) staining signal (Fig 2D, 2G and 2H and S4D Fig), indicating defective Purkinje cell maturation in the cKO cerebellums. Glial fibrillary acid protein (GFAP) staining also showed that the Bergmann glia lost their scaffold organization pattern in the cKO cerebellums (Fig 2D and S4D Fig). Based on the above, we concluded that depletion of METTL3 caused severe cerebellar hypoplasia in mice, indicating that m6A played important roles in regulating not only the development of the cortex [28] but also the cerebellar region of the brain.
We further assessed the proliferation potential of GCPs using an in vivo bromodeoxyuridine (BrdU) labeling assay. The cerebellar tissues were dissected from P7 mice 2 h post–BrdU injection and costained with BrdU/antigen identified by monoclonal antibody Ki 67 (Ki67)/DAPI. The numbers of Ki67+ cells exhibited no difference between the Ctrl and cKO samples, indicating that GCPs maintain the ability to proliferate in cKO EGLs (Fig 3A and 3E), although with a slightly reduced rate (S5A and S5B Fig). However, both TUNEL and Cleaved Caspase-3 immunostaining revealed a significantly increased cell apoptosis rate in cKO cerebellum, especially in the EGL region (Fig 3B, 3C, 3F and 3G). Histological examination also detected a large number of karyorrhexis and karyopyknosis, the characteristic nuclear forms of apoptotic cells, in the EGL of cKO cerebellums (S5C Fig). Consistent with the enhanced apoptotic signals, the BrdU+/Ki67− cells, representing postmitotic newborn granule neurons, were drastically reduced in cKO cerebellums at 48 h post–BrdU injection (Fig 3D and 3H). These data demonstrated that the severe depletion of CGCs in cKO cerebellums mainly resulted from the abnormal postmitotic apoptosis of newborn granule cells.
To further investigate the underlying mechanisms for the cerebellar developmental defects in Mettl3 cKO mice, we performed RNA sequencing (RNA-seq) and m6A sequencing (m6A-seq) using mRNAs extracted from the cerebellums of Ctrl and cKO mice at both P7 and P14. Over 10,000 and 15,000 differential m6A peaks were identified in the cerebellar mRNAs of P7 and P14 Ctrl mice, respectively, whereas in Mettl3 cKO mice, only a few hundred (background level) differential m6A peaks were detected (S1, S2 and S3 Tables). Consistent with previous reports [16, 37], the m6A peaks in cerebellar RNAs were also enriched in the regions with RRACH motif (Fig 4A and S6A Fig) and tended to occur near stop codons and within 3′UTRs of mRNAs (Fig 4B and S6B Fig). Among the 15,614 expressed genes in the cerebellum of the P7 Ctrl mice, transcripts of 6,838 genes contain m6A peaks (Fig 4C). Because the presence of m6A peaks could facilitate mRNA degradation [16], we focused on the 696 genes with both m6A peak loss and elevated expression in the cKO cerebellums (Fig 4C), as this group of transcripts was likely stabilized after m6A depletion. Gene ontology analysis revealed that most of these genes were enriched in neural development-associated biological processes and the apoptotic signaling pathway (Fig 4D and S7A and S7B Fig). Similar results were also obtained from the m6A-seq and RNA-seq data at P14 (S6A–S6D and S7C and S7D Figs).
Quantitative reverse transcription-polymerase chain reaction (qRT-PCR) analysis confirmed the up-regulation of genes with m6A loss (S8A and S8C Fig) in cKO cerebellums, including essential factors for cerebellar development such as Atoh1, Cxcr4, Notch2 and its ligand Jag1 [32, 38–41], as well as neuronal apoptosis such as Dapk1, Fadd, and Ngfr [42–44] (Fig 5A and S6E Fig). We further measured the degradation rates of these mRNAs in neural stem cell (NSC) lines established from the Ctrl and cKO neonatal mice (Fig 5B) using the actinomycin-D mediated transcription inhibition assay. We detected prolonged mRNA half-lives of these development- and apoptosis-associated genes in the cKO NSCs (Fig 5C). Because there were no reported apoptosis phenotypes in the developing cortex [28], we compared the expression levels of the apoptosis-promoting genes that showed extended half-lives with the depletion of m6A in both the cerebellar region and the cortex of wild-type mice by quantitative polymerase chain reaction (qPCR), and the results displayed higher expression levels of these genes in the cerebellar region than in the cortex (Fig 5D). Furthermore, in vitro cultured CGCs isolated from P7 mice showed elevated expression levels of these genes with the depletion of METTL3 (Fig 5E). All lines of evidence indicated that these apoptosis-promoting genes functioned predominantly in the cerebellum CGCs rather than in the cortex. Our data demonstrated that the depletion of m6A resulted in a dysregulated gene expression profile during the development of the cerebellum, especially for genes related to cerebellar development and cell apoptosis.
Consistent with a previous report that m6A depletion tends to produce exon-excluded transcripts [17], we also found that the majority of exon-excluded transcripts in cKO cerebellums (as compared with the Ctrl) were derived from m6A-containing transcripts in the Ctrl at both P7 and P14 (Fig 6A and S9A Fig). Interestingly, genes with exon-excluded transcripts in cKO were enriched in synapse-associated pathways, including transmission across chemical synapses and protein–protein interactions at synapses (Fig 6B and S9B Fig). More specifically, we found that genes associated with the two major glutamate receptors, the N-methyl-D-aspartic acid (NMDA) receptors and the α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid (AMPA) receptors, were highly enriched in the pathway analysis (Fig 6B). NMDA receptors were reported to be essential for the maturation and survival of CGCs [44, 45]. We confirmed that the m6A-modified exons (S8B and S8C Fig) in these synapse-associated genes, such as Grin1, Atp2b3, Grm1, and Lrp8, were excluded in cKO cerebellums by RT-PCR (Fig 6C and S9C Fig). The MiniGene assay confirmed the exclusion of exon 21 (C1 region) of Grin1 (S10A Fig) and exon 19 of Lrp8 (S10B Fig) with the depletion of m6A by either Mettl3 knockdown or mutation of the m6A modification sites (S10C and S10D Fig), indicating the regulatory role of m6A modification in RNA splicing and function of the ion channel associated genes.
The NMDA receptor is a heterodimer of glutamate ionotropic receptor NMDA type subunit 1 (GRIN1) and glutamate ionotropic receptor NMDA type subunit 2 (GRIN2) and is permeable to Ca2+ when activated by glutamate. The Grin1 gene contains three alternatively spliced exons (ASEs), namely N1, C1, and C2, and generates at least eight splicing variant isoforms [46]. The C1 region has been shown to be a regulatory domain for calcium influx and is reported to be affinitive to calmodulin (CaM) [47] and inhibit the NMDA receptor in a Ca2+-dependent way [48]. To further investigate the effect of the altered splicing of Grin1 induced by depletion of m6A on the survival of CGCs, we isolated and cultured CGCs from both Ctrl and cKO mouse cerebellums. We overexpressed Grin1001 (C1 excluded variant) or Grin1011 (C1 included variant) in the cKO CGCs by electroporation. We found that cKO CGCs had a much higher intracellular calcium concentration than that of the Ctrl ones, whereas overexpression of Grin1011, but not mCherry or Grin1001, reduced the cellular calcium concentration in the cKO CGCs (Fig 6D and 6E). Consistently, Cell Counting Kit-8 (CCK8) assay detected an increased survival rate of the cKO CGCs with Grin1011 overexpression (Fig 6F). Our findings established that the altered Grin1 splicing induced by m6A depletion resulted in a substantial increase in intracellular calcium concentration and consequently the apoptosis of the CGCs.
It is well established that epigenetic machineries play essential roles in controlling the timing and linage commitment of NSCs during neurogenesis [49, 50]. Epigenetic mechanisms such as DNA methylation and histone modifications regulate brain development and plasticity at the transcriptional level [4, 51]. Additionally, noncoding RNAs and RNA modifications also appeared to add complexities at the epitranscriptomic level in regulating neuronal development and brain function [26, 52–57], and multiple layers of epigenetic modification regulations have been shown to be required for neurogenesis in the developing cerebellum [58–60]. DNA demethylation is highly active during early cerebellum development, and inhibition of DNA demethylation suppresses the circuit formation of developing granular cells [60]. GCPs and glial cells express high levels of histone deacetylase 1 (HDAC1), indicating the potential role of histone modification in neurogenesis of the cerebellum [61]. The microRNAs (miRNAs) have been reported to regulate components of Hedgehog-Patched signaling in normal or transformed GCPs [62]. In addition, alternative transcription combined with alternative splicing gives rise to transcriptomic diversity during cerebellar development [59].
Recent study has reported that specific depletion of m6A in the mouse brain by conditional knockout of Mettl14 causes postponed neurogenesis in the cortex [28]. In our study, we found remarkable cerebellar developmental defects in Mettl3 conditional knockout mouse model, indicating the indispensable roles of METTL14- and METTL3-mediated m6A signaling in neuronal development. In addition, we recognized a severe cerebellar hypoplasia phenotype in Mettl3 cKO mice. Depletion of m6A caused apoptosis of the newborn granule cells, which then resulted in altered Purkinje cell and Bergmann glia architecture. Our findings demonstrated, for the first time, the regulatory function of m6A modification in cerebellar development.
The development of the mouse cerebellum is quite distinct from that of the cerebral region of the brain. Neurogenesis dominantly occurs in the embryonic stage in the cortex, while in the cerebellar region, this process mainly occurs in the postnatal 2 wk [32]. A whole transcriptome-wide m6A methylation analysis depicted a spatial-specific methylation profile of mouse brain and revealed that m6A methylation had a higher level in the cerebellum than that in the cerebral cortex [35], indicating pivotal roles of m6A in this brain region. The development of the mouse cerebellum is a highly organized process involving a cooperated set of genetic and epigenetic regulations. By high throughput RNA-seq and m6A-seq, we revealed that key developmental genes like Atoh1 and Cxcr4 were abnormally up-regulated because of the extended mRNA half-lives induced by m6A depletion in the cKO mice.
During the development of CGCs, apoptosis occurs occasionally as a normal process [63], whereas in the cKO mice, in addition to developmental associated genes, we also detected up-regulation of apoptosis-associated genes like Dapk1 and Fadd, with extended half-lives of their mRNAs. Hence the m6A modification on mRNAs may play an indispensable role in balancing cell survival and apoptosis by regulating mRNA stability of development- and apoptosis-associated genes.
Comparison of the mouse brain m6A-seq data with the mouse synaptic proteome revealed that 76.8% of mouse postsynaptic genes and 30% of presynaptic genes were detected with m6A RNA methylation [35]. In addition to the resistance of decay in the developmental- and apoptosis-associated genes during the cell fate conversion from GCPs to CGCs, we also recognized a set of synapse-related genes that are highly enriched in the abnormally spliced genes when m6A is depleted in the cerebellum. Grin1 had been reported to be associated with CGC survival and contribute to the survival-promoting effect of NMDA receptors in cultured CGCs [64]. In this study, we demonstrated that the altered splicing of Grin1 induced by m6A depletion resulted in excessive calcium influx and finally caused the apoptosis of the CGCs. The regulatory function of METTL3-mediated m6A in alternative splicing of synapse-associated genes implied its pivotal role in neuronal survival, connection, and maturation.
Although it has been reported that cytoplasmic translocation of METTL3 alone could promote protein translation independent of its m6A activity [65], our current study demonstrated that m6A depletion induced by Mettl3 inactivation accounts for both the elevated RNA stability of development- and apoptosis-associated genes (such as Atoh1 and Dapk1) and the altered splicing of synapse-associated genes (such as Grin1), which consequently led to the severe cerebellum development defects manifested by massive apoptosis and impaired differentiation of CGCs in the knockout mouse model. Thus, it is highly unlikely that the functional role of Mettl3 in the mouse cerebellum development is m6A independent.
Contrary to a previous report showing that METTL3 is widely expressed in various regions of the mammalian brain [35], we detected relatively weak METTL3 signals in the proliferating GCPs in the outer EGL of the developing mouse cerebellum by Immunohistochemistry (IHC) (S4B Fig), which might be a likely reason to explain why the proliferation of the GCPs was not substantially affected by depletion of METTL3. However, we detected dramatically elevated expression of METTL3 when GCPs differentiated into the newborn CGCs in the inner layer of EGL (S4B Fig). We highly speculate that enhanced expression of METTL3 is a pivotal event during the differentiation of CGCs.
Taken together, we conclude that METTL3 is an essential regulator of cerebellar neurogenesis and development, and the transcriptome-wide gene expression dysregulation induced by m6A depletion in the absence of METTL3 finally results in CGCs apoptosis and cerebellum hypoplasia (Fig 6G).
The study was approved by the Research Ethic Committee in the Institute of Zoology, Chinese Academy of Sciences in Beijing, China (Approval number: IOZ20150076).
Mouse strains used in this study were B6D2F1 (C57BL/6×DBA2), C57BL/6, and DBA2. B6D2F1 and DBA2 mice were purchased from Beijing Vital River Laboratory Animal Center. C57BL/6 Nestin-Cre mice were purchased from the Jackson Laboratory. All mice were housed under specific pathogen-free (SPF)-grade conditions in the animal facilities of the Institute of Zoology, Chinese Academy of Sciences.
The T7-Cas9 plasmid and the T7-sgRNA backbone used in this study were from Qi Zhou’s lab [66]. T7-sgRNA backbone harbored a T7 promoter followed by two reverse orientated Bbsl restriction sites, which was followed by a small guide RNA scaffold. The T7-sgRNA plasmid construction was performed as previously described [66]. The annealed sgRNA oligonucleotides synthesized from Beijing Genomics Institute (BGI) have two different overhangs that are complementary with the corresponding sticky ends of Bbsl-digested T7-gRNA vector to ensure directional cloning. Oligonucleotides used for CRISPR-Cas9 system-assisted homologous recombination through embryo injection were synthesized in Integrated DNA Technologies. All oligonucleotides were listed in S4 Table.
The sgRNAs and Cas9 mRNA were transcribed by the HiScribe T7 In Vitro Transcription Kit (New England Biolabs) using endonuclease linearized DNA templates according to the manufacturer’s instructions. Cas9 mRNA was capped with the m7G (5′) ppp (5′) G RNA Cap Structure Analog (New England Biolabs) by T7 RNA transcriptase. RNAs were dissolved in DEPC-H2O (New England Biolabs) and used for intracytoplasmic RNA microinjection following previously reported procedures [66]. Briefly, one-cell-stage embryos were collected from female B6D2F1 mice at 0.5 day post-coitum (dpc). Each embryo was microinjected with 25 ng/μL L-sgRNA, 25 ng/μL R-sgRNA, 100 ng/μL Cas9 mRNA, 50 ng/μL L-oligo DNA, and 50 ng/μL R-oligo DNA into the cytoplasm. Injected embryos were subsequently implanted into the oviducts of pseudopregnant B6D2F1 female mice. Full-term pups were obtained by natural labor at 19.5 dpc.
All pups were genotyped and two Mettl3flox/+ founder mice were obtained. The viripotent Mettl3flox/+ founder mice were self-bred or mated with Nestin-Cre mice to generate Mettl3flox/flox and Mettl3flox/+; Nestin-Cre mice, respectively. The Mettl3flox/flox and Mettl3flox/+; Nestin-Cre mice were interbred to generate Mettl3flox/+; Nestin-Cre, Mettl3flox/flox, Mettl3flox/+, and Mettl3flox/flox; Nestin-Cre mice.
Genomic DNA was extracted using the Mouse Direct PCR Kit (Bimake). Briefly, mouse tissues were mixed with 100 μL Buffer L and 2 μL Protease Plus, and incubated at 55 °C for 30 min, then 100 °C for 5 min according to the manufacturer’s instructions.
First, all mouse pups were genotyped to identify the insertion of loxP sites in Mettl3 using tail DNA extract. The length of the PCR product was 222 bp/182 bp (loxP/wt) in Mettl3 intron 1 with primers F1/R1 and 335 bp/295 bp (loxP/wt) in Mettl3 intron 4 with primers F2/R2. Pups with loxP inserted both in Mettl3 intron 1 and intron 4 in the same allele of Mettl3 were identified as Mettl3flox/+. Pups with loxP inserted both in Mettl3 intron 1 and intron 4 in both alleles of Mettl3 were identified as Mettl3flox/flox. The PCR product for Nestin-Cre was 410 bp with primers Nestin-Cre-F/Nestin-Cre-R. Primers F1/R2 were used to further confirm Mettl3 deletion in the brain, producing PCR product with 318 bp/2,554 bp (cKO/wt) in length. All sequences of primers for genotyping are listed in S5 Table.
Mouse brain tissues were harvested after cervical dislocation and were homogenized either by grinding into homogenous powder using mortar and pestle cooled in liquid-nitrogen bath or by mechanical homogenizer using beating beads. Pulverized tissue samples were lysed in RIPA buffer (0.5% NP-40, 50 mM Tris-Cl, pH 8, 150 mM NaCl, 1 mM EDTA) supplemented with proteinase and phosphatase inhibitors. The lysate was centrifuged at 12,000 g at 4 °C for 20 min to remove cell debris. Pierce Coomassie Plus (Bradford) assay kit (Thermo) was used to determine the protein concentration. The fraction (50–100 μg) was separated by 10% SDS-PAGE and analyzed by immunoblotting with corresponding antibodies, anti-Mettl3 (1:1,000, Abcam, ab195352), anti-Mettl14 (1:1,000, Atlas Antibodies, HPA038002), and anti-γ-ACTIN (1:3,000, Santa Cruz, SC65638).
Tail suspension test was performed as described by Pierre-Olivier Frappart [36]. Mice at P14 were hung by the tail. Postures and movements of the mice were observed.
The assay was performed as previously described [67] using P14 Ctrl and cKO mice. The apparatus was a square-shaped arena (750 × 750 mm2, length × width), which was divided into two concentric squares, the central field (325 × 325 mm2 arena) and the peripheral field, and was illuminated evenly at 15 lux. Animals were placed in the center of the arena, and their locomotion was recorded by a camera hung over the arena. A 10-min-long video was analyzed by the TopScan software. Locations of the animals in the area were recorded at the speed of one image per s. The frequency crossing the border, latency in the central field, total moving distance, and velocity of mice recorded in the apparatus were analyzed.
All mice for MRI were anesthetized with 10% chloralic hydras (Sigma) by intraperitoneal injection. Mice were fixed on the scanning coil with a self-made scaffold. Body temperature was maintained by circulating water through warming pads. MRI was conducted using a Bruker 7.0 T MRI equipment (Bruker Medical Systems, CLinScan) and was performed at P14 for both Ctrl and cKO mice. Both the sagittal section and transverse section were scanned. For image acquisition, rapid acquisition relaxation enhanced (RARE) sequence was used. The MRI parameters included T2W repetition time = 2,500 ms, T2W echo time = 60 ms, field of view = 12 × 12 mm, matrix = 240 × 240 pixels, resolution = 50 × 50 μm, and slice thickness = 0.6 mm. MRI images acquired were processed by ImageJ (https://imagej.nih.gov/ij/) to calculate the area of the whole brain section or the cerebellar region. To be brief, the outline of the whole brain or the cerebellum in the image was drawn out as region of interest (ROI); the number of pixels contained within the ROI was calculated by the software. As the resolution is 50 × 50 μm, each pixel represents an area of 2.5 × 10−3 mm2. Three repeats were conducted for each calculation.
In order to examine the proliferation activity of GCPs in the cerebellar EGL, we injected P7 mice with BrdU (Sigma) intraperitoneally at a dose of 50 μg BrdU/g body weight [36]. Mice were humanely killed 2 h later for immunofluorescence staining. To investigate the generation and migration of CGCs, BrdU-injected mice were kept for another 2 d until they were humanely killed at P9 for staining.
To compare the gross appearance and histological organization of the cerebellums of both Ctrl and cKO mice, mice at P14 were humanely killed by breaking the neck. The whole brain or cerebellum was dissected, weighed, or fixed with 4% paraformaldehyde and photographed.
Dissected brains were cut into sagittal blocks and fixed with 4% paraformaldehyde followed by dehydration (70%, 80%, 90%, 100% ethanol) and paraffin embedding. Sections 3 μm in thickness were cut from the paraffin-embedded tissue blocks with a Leica slicing machine (Leica Biosystems) and mounted on poly-D-lysine coated glass slices (Zhong Shan Golding Bridge Biotechnology). Slices were heated at 65 °C for 2 h and then immersed in xylene to remove paraffin. After a series of rehydration processes (100%, 100%, 90%, 80%, 70% ethanol), slices were stained with hematoxylin and eosin (HE) using standard methods and imaged with a Leica Aperio VERSA 8 microscope (Leica Biosystems).
For immunohistochemical staining, rehydrated sections or coverslips were incubated with primary antibodies overnight. Primary antibodies used in this study were for Mettl3 (ab195352, 1:250, Abcam), neuronal nuclei (NeuN, MAB377, 1:500, Millipore), calbindin D-28K (C9848, 1:2,000, Sigma), and glial fibrillary acid protein (GFAP, Z033429, 1:1,000, Dako). Appropriate Horseradish peroxidase conjugated secondary antibodies (anti-Mouse IgG and anti-Rabbit IgG, Vector laboratories) were used according to manufacturer’s recommendations. A 3, 3′-diaminobenzidine kit (DAB, Vector laboratories) was used for color developing. Images were obtained using standard methods and imaged with a Leica Aperio VERSA 8 microscope (Leica Biosystems). Immunohistochemical images were analyzed with Image-Pro Plus (Media Cybernetics). At least three repeats were conducted for each calculation.
Immunofluorescence staining was performed as previously described, with a few modifications [68]. Briefly, rehydrated sections or coverslips were incubated with primary antibodies overnight. Primary antibodies used in this study were for BrdU (C8434, 1:500, Sigma), Ki67 (ab15580, 1:500, Abcam), Cleaved Caspase-3 (9661s, 1:200, Cell Signaling Technology), Phospho-Histone H3 (3377s, 1:500, Ser10, Cell Signaling Technology), Nestin (MAB353, 1:1,000, Millipore,), Sox2 (SC-17320, 1:1,000, Santa Cruz), and Mettl3 (ab195352, 1:1,000, Abcam). Appropriate fluorophore (FITC, Cy3, and Cy5) conjugated secondary antibodies (Jackson ImmunoResearch, 1:1,000) were used according to the manufacturer’s recommendations. Sections were counterstained with DAPI (Invitrogen) and mounted on slices with Dako Fluorescence Mounting Medium (Dako). Fluorescent images were obtained using a Carl Zeiss LSM 780 confocal system. For quantification of Ki67 and Phospho-Histone H3 positive cell ratio, we used three pairs of mice at P7 and quantified three sections of the cerebellum vermis for each Ctrl and cKO mouse, respectively. We lined out the EGL region manually, and the total numbers of Ki67+-, Phospho-Histone H3- and DAPI-positive cells in the EGL region were analyzed with Columbus Image Analyzing System (Perkin Elmer). For the quantification in the assays of long-term BrdU labelling, TUNEL, and Cleaved-Caspase 3 staining, the positive cells in the whole section of the cerebellum vermis were quantified. Three pairs of mice and three sections for each of Ctrl and cKO mouse were used for quantification.
The TUNEL assay was performed using the DeadEnd Fluorometric TUNEL System (Promega) according to the manufacturer’s instruction. Briefly, sections were permeabilized by proteinase K and labeled with rTdT reaction mix for 1 h at 37 °C; the reaction was stopped by 2 × SSC. Sections were counterstained by PI (Invitrogen) and mounted on slices with Dako Fluorescence Mounting Medium (Dako). Images were obtained using a Carl Zeiss LSM 780 confocal system. Fluorescent images were analyzed with Columbus Image Analyzing System (Perkin Elmer). At least three repeats were conducted for each calculation.
Total RNA from mouse cerebellums was extracted using TRIzol reagent (Invitrogen). mRNAs were purified with Dynabeads mRNA Purification Kit (Thermo) according to the manufacturer’s manual, followed by DNase I (M6101, Promega) treatment to remove genomic DNA. RNA concentration was measured using Nanodrop 1000 (Thermo). For quality control, only samples with OD 260/280 nm ratio more than 2 and 260/230 nm values in the range of 2.0–2.2 were used for subsequent experiments. The integrity of RNAs was checked by RNA gel electrophoresis, and only total RNAs having a 28S band twice as bright as the 18S band were used for further study. Two sets of samples were collected as biological replicates.
A total of 400 ng RNA was mixed with 0.1 U Nuclease P1 (Sigma) and 1.0 U calf intestinal phosphatase (New England Biolabs) in the final reaction volume of 50 μL, adjusted with water and incubated at 37 °C overnight. The digested RNA solutions were filtered by ultrafiltration tubes (MW cutoff: 3 kDa, Pall, Port Washington), then subjected to UHPLC-MS/MS analysis for detection of m6A. The UHPLC-MS/MS analysis was performed with an Agilent 1290 UHPLC system coupled with a G6410B triple quadrupole mass spectrometer (Agilent Technologies). A CAPCELL PAK C18 column (100 × 2.1 mm I.D., 3 μm particle size, SHISEIDO) was used for UHPLC separation of mononucleotides. UHPLC separation parameters were used as follows: 0–6.5 min, 5.0% B; 6.6–11.0 min, 20.0% B; 11.1–16.0 min, 100% B; 16.1–24 min, 5.0% B. Solvent A was an aqueous solution of 0.1% formic acid, and solvent B was 100% methanol. The mass spectrometer was operated in the positive ion mode. A multiple reaction monitoring (MRM) mode was adopted, using m/z 282→150 for m6A (collision energy, 15 eV) and m/z 268→136 for A (10 eV). The injection volume for each sample was 5 μL, and the amount of m6A and A was calibrated by standard curves. Nitrogen was used for nebulizing and desolvation gas of MS detection. The nebulization gas was set at 40 psi, the flow rate of desolvation gas was 9 L/min, and the source temperature was set at 300 °C. Capillary voltage was set at 3,500 V. High purity nitrogen (99.999%) was used as collision gas. Each sample was analyzed at least three times.
Methylated RNA immunoprecipitation (MeRIP) was performed as previously described [69]. Briefly, purified mRNA was randomly fragmented to size around 100 nucleotides using Ambion RNA fragmentation reagents and then subjected to immunoprecipitation (IP) with anti-m6A antibody (202003, Synaptic Systems) and protein-A magnetic beads (88845, Pierce) in MeRIP buffer (150 mM NaCl, 10 mM Tris-HCl, pH 7.4, 0.1% NP-40) supplemented with RNase inhibitor. m6A-containing mRNA fragments were eluted with m6A in MeRIP buffer and were purified using TRIzol reagent. For MeRIP-seq, two sets of samples were collected for duplicated biological repeats. For MeRIP-qRT-PCR, the same procedures were carried out except that the purified mRNA was fragmented into about 200 nucleotides; three biological repeats were conducted.
Poly (A) RNA from 1 mg total RNA or purified mRNA and purified m6A-containing fragments were used to generate the cDNA libraries, respectively, according to TruSeq RNA Sample Prep Kit protocol. All samples were sequenced by Illumina HiSeq-3000 with paired-end 101-bp read length. About 100 million and 150 million 2*101 paired-end reads in cKO and Ctrl mice were obtained, respectively.
Raw RNA-seq reads were mapped to the mouse reference genome (mm10) using HISAT2 (Version 2.0.4) with parameters “—novel-splicesite-infile” [70]. StringTie (Version 1.3.1c) was used to construct full-length transcript assembly and estimate transcript abundance with default parameters [71]. Transcripts were filtered by TACO (Version 0.5.1) [72] with default parameters. Differentially expressed genes between the Ctrl and cKO mice at P7 and P14 were identified using edgeR (Version 3.10.0) [73] with fold change ≥ 1.5 and adjusted p-value ≤ 0.1 (Benjamini–Hochberg multiple testing correction) as the thresholds.
To identify decreased or increased m6A peaks in Ctrl and cKO m6A-seq against RNA-seq, two replicates for each group were used for differential peak analysis. The sequences were normalized according to total library size prior to peak calling (see S6 Table for normalization factors). A peak calling prioritization pipeline (PePr, version 1.1.18) was used to identify differential binding sites of m6A caused by Mettl3 conditional knockout [74]. A 200-bp sliding window approach and intergroup normalization were performed for m6A peak calling after removing PCR duplication, and a p-value threshold of 1e-5 (Wald test) was reported. The m6A peaks were annotated by bedtools (version 2.25.0) [75]. The enriched sequence motifs among m6A peaks were identified by HOMER (version 4.7) [76] with motif lengths of 5 nt, 6 nt, 7 nt, and 8 nt, respectively.
Gene Ontology biological processes and Reactome gene sets enrichment analysis were carried out using Metascape with p-value < 0.01 (Banjamini–Hochberg multiple testing correction), with the expressed genes (FPKM > 0.1) in Ctrl mouse cerebellums as the background [77]. For the enriched terms, Kappa scores were used as the similarity metric when performing hierarchical clustering, and then subtrees with similarity > 0.3 were considered as a cluster [77]. Enriched terms were selected according to the Metascape cluster results.
To analyze alternative splicing, the PSI method provided by Schafer and colleagues was used [78]. Briefly, annotations on an exon sequence were created through DEXseq (Version 1.20.2) [79], based on assembled transcripts. The PSI ratios for all exon parts were estimated by modifying the code provided by Schafer and colleagues [77]. PSI comparison supported by at least 10 inclusion and exclusion reads was considered for further analysis. Exons with ΔPSI ≥ 10% in Ctrl mice compared to cKO mice were considered as ASEs. ASEs overlapped with m6A-modified genes in Ctrl were further analyzed to uncover the target exons and their splicing pattern affected by METTL3.
For establishment of the neural stem cell (NSC) lines, neonatal mouse brains were cut into small pieces by ophthalmic scissors and digested with 0.25% Typsin-EDTA (Gibco) for 15 min in 37 °C, 5% CO2. The dissociated cell suspension was plated on Matrigel (BD Biosciences) coated 6-well plates. The basal culture medium contained Neurobasal Medium (Gibco) and DMEM/F12 Medium (Gibco) at a ratio of 1:1, plus 0.5% N-2 Supplement (Gibco), 1% B-27 Serum-Free Supplement (Gibco), 2 mM GlutaMAX-I (Gibco), bFGF (20 ng/mL, R & D systems), and EGF (20 ng/mL, R & D systems). For immunofluorescent staining, NSCs were plated on coverslips coated with poly-L-ornithine (Sigma) and Matrigel (BD Biosciences), successively. HeLa cells were cultured in 37 °C, 5% CO2 in DMEM Medium (Gibco) plus 10% FBS (Gibco), 2 mM GlutaMAX-I (Gibco), and 0.01 mM beta-mercapto-thanol (beta-ME, Gibco).
CGCs were isolated from P7–P8 mouse pups as previously described [80]. To be brief, cerebellums were dissected and washed by HBSS (Gibco) and dissociated with 0.25% Typsin-EDTA (Gibco). Cells were suspended and subjected to Percoll (GE Healthcare) gradient (30% and 65% Percoll) separation. The astroglia and other heavier cells were separated by pre-plating on a poly-D-lysine (Sigma) coated dish for 20 min in a 5% CO2, 37 °C incubator. Purified CGCs were plated on poly-D-lysine coated plates and cultured in Neurobasal Medium (Gibco) containing 2% B-27 Serum-Free Supplement (Gibco), 2 mM GlutaMAX-I (Gibco), and 15 μM glutamate (Gibco) in a 5% CO2, 37 °C incubator. After 20 h, 10 μM Cytosine arabinoside (Ara-C, Sigma) was added to arrest the proliferation of nonneuronal cells.
The plasmids Pmax-mCherry, Pmax-Grin1001-T2A-mCherry, or Pmax-Grin1011-T2A-mCherry were constructed using the Pmax-GFP backbone (Lonza). Approximately 5 × 106 mouse CGCs were resuspended in 100 μL P3 Solution from the Lonza Nucleofector P3 Primary Cell 4D X Kit (Lonza). Additionally, 10 μg of the appropriate plasmid was added to the cell suspension. The cell suspension was then placed in the cuvette and electroporated using a Lonza Nucleofector 4D device (Lonza) under the presupposed program CD150. Cells were then resuspended and cultured in CGC culture medium.
Electroporated CGCs were plated on a 20 mm confocal dish (Nest) precoated with poly-D-lysine. Culture medium of 3 DIV neurons was replaced by 2 mL of Locke’s solution containing 154 mM NaCl, 5 mM KCl, 3.6 mM NaHCO3, 5 mM HEPES, 1.5 mM CaCl2, 1.2 mM MgCl2, and 5.6 mM glucose, pH 7.4, and incubated for 10 min at room temperature. Cells were subsequently incubated for 20 min in 2 mL of Locke’s solution containing 10 pM Fluo-4 (Thermo) and then washed with 2 mL of Locke’s solution for 10 min before starting the experiment. All measurements were made at room temperature (25 °C). Fluo-4 (green) and mCherry (indicating the expression of appropriate plasmid, red) fluorescence were imaged with an inverted Perkin-Elmer microscope. Images were digitized in an image processor connected to a computer equipped with VOLOCITY (version 6.0) software. The formula [Ca2+]i = KdFluo-4 × [F − Fmin/ Fmax − F] [81] was used to calculate the intracellular calcium concentration of mouse CGCs. Fmax and Fmin were obtained by perfusing cells with a salt solution containing 10 mM CaCl2, and subsequently with a Ca2+-free salt solution containing 10 mM EGTA (Calcium Calibration Buffer Kit, Thermo).
Electroporated CGCs were seeded on 48 multi-wells (Corning) coated with poly-D-lysine. At DIV 1, DIV 3, and DIV 5, cells were incubated with CGC culture medium containing 1 × CCK8 (Solarbio) for 4 h in a 5% CO2, 37 °C incubator. The absorbance was measured using a scanning multi-well spectrophotometer (ELISA reader) at a wavelength of 450 nm. The survival rates of CGCs were calculated as OD450DIV3/OD450DIV1 × 100% or OD450DIV5/OD450DIV1 × 100%.
NSCs were plated on 6-well plates, 5 × 105 per well, for 2 d, then cells were treated with actinomycin-D (10 μg/mL, Sigma) and collected for qRT-PCR analysis 3 h, 6 h, 9 h, or 12 h later. The half-life was calculated according to the following equation: ln (Ci/C0) = −kti, where k is degradation rate, Ci is the mRNA value at time i, and ti is the time interval in hours. Three repeats were conducted for each calculation.
In the qRT-PCR experiments, 1 μg of total RNA treated with DNase I was reversely transcribed into cDNA by the Reverse Transcription System (Promega). SYBR Green PCR Master Mix (Toyobo) was used in all qRT-PCR experiments. The relative fold expression changes of genes were calculated using the 2-ΔΔCt method, with Actb as internal control. Semiquantitative RT-PCR products were amplified for 30 cycles. Primers used in these experiments are listed in S7 Table.
For miniGene-based splicing analysis of exon 19 in Lrp8 and exon 21 in Grin1, a fragment containing either wild-type (miniGene) or mutated m6A motif (miniGene-M) was inserted into the pSpliceExpress reporter vector (Addgene). For Lrp8, two m6A motifs (GAACC) were mutated to GATCC. For Grin1, one m6A motif (AGACA) was mutated to AGTCA. MiniGene or miniGene-M was transfected into HeLa cells with control siRNA (siControl) or siRNAs for knockdown of Mettl3 (siMettl3) using Lipofectamine 3000 (Invitrogen), respectively. Forty-eight hours after transfection, total RNA was isolated and reverse transcription was carried out to produce the cDNA. Then, cDNA was used as template for alternative splicing change detection through semiquantitative RT-PCR. AlphaView was used to analyze the density of bands on agar gels. Ratios of exon exclusion were quantified based on the density of bands. All the primers and siRNAs used for this assay are listed in S8 Table.
All statistical analyses (unless stated otherwise) were performed using the R package for Statistical Computing. For experimental data quantification, Student t test was applied using GraphPad Prism 6 software, and the error bar was shown based on SEM (unless stated otherwise). P value < 0.05 was considered statistically significant.
|
10.1371/journal.pgen.1002990 | A Novel Human-Infection-Derived Bacterium Provides Insights into the Evolutionary Origins of Mutualistic Insect–Bacterial Symbioses | Despite extensive study, little is known about the origins of the mutualistic bacterial endosymbionts that inhabit approximately 10% of the world's insects. In this study, we characterized a novel opportunistic human pathogen, designated “strain HS,” and found that it is a close relative of the insect endosymbiont Sodalis glossinidius. Our results indicate that ancestral relatives of strain HS have served as progenitors for the independent descent of Sodalis-allied endosymbionts found in several insect hosts. Comparative analyses indicate that the gene inventories of the insect endosymbionts were independently derived from a common ancestral template through a combination of irreversible degenerative changes. Our results provide compelling support for the notion that mutualists evolve from pathogenic progenitors. They also elucidate the role of degenerative evolutionary processes in shaping the gene inventories of symbiotic bacteria at a very early stage in these mutualistic associations.
| Many insects harbor symbiotic bacteria that perform diverse functions within their hosts. However, the origins of these associations have been difficult to define. In this study we isolate a novel bacterium from a human infection and show that this bacterium is a close relative of the Sodalis-allied clade of insect symbionts. Comparative genomic analyses reveal that this organism maintains many genes that have been inactivated and lost independently in derived insect symbionts as a result of rapid genome degeneration. Our work also shows that recently derived Sodalis-allied symbionts maintain a significant population of “cryptic” pseudogenes that are assumed to have no beneficial function in the symbiosis but have not yet accumulated mutations that disrupt their translation. Taken together, our results show that genome degeneration proceeds rapidly following the onset of symbiosis. They also highlight the potential for diverse insect taxa to acquire closely related insect symbionts as a consequence of vectoring bacterial pathogens to plants and animals.
| Obligate host-associated bacteria often have reduced genome sizes in comparison to related bacteria that are known to engage in free-living or opportunistic lifestyles [1]. This is exemplified by inspection of the genome sequences of mutualistic, maternally transmitted, bacterial endosymbionts of insects, many of which have been maintained in their insect hosts for long periods of evolutionary time [2]. Often these obligate endosymbionts maintain only a small fraction of the gene inventory that is found in related free-living counterparts [3]–[5], indicating that the obligate host-associated lifestyle facilitates genome degeneration and size reduction. At a simple level, the process of genome degeneration in obligate endosymbionts can be viewed as a streamlining of the gene inventory to yield a minimal gene set that is compatible with the symbiotic lifestyle. Genes that have no adaptive benefit are inactivated and deleted as a consequence of mutations that accumulate under relaxed selection at an increased rate in the asexual symbiotic lifestyle as a result of frequent population bottlenecks occurring during symbiont transmission [6].
Although we now have a detailed understanding of the mechanisms and evolutionary trajectory of genome degeneration in ancient obligate insect symbionts, the fundamental question of how these mutualistic associations arise remains to be answered. Studies focusing on insect-bacterial symbioses of recent origin show that closely related bacterial endosymbionts are often found in distantly related insect hosts [7], [8]. This could be explained by the interspecific transmission of symbionts, mediated by parasitic wasps and mites that facilitate the transfer of symbionts between distantly related hosts [9], [10]. Horizontal symbiont transmission could also be mediated by intraspecific mating, as demonstrated in the pea aphid [11]. Another possibility is that symbionts could be acquired de novo from an environmental source.
Symbiont acquisition, at least initially, requires the symbiont to overcome or evade the insect immune response. Given that many insects are known to possess a potent immune system that repels invading microorganisms [12], it has been assumed that mutualistic symbionts arise from pathogenic progenitors that have evolved specialized molecular mechanisms to facilitate evasion of the immune response and invasion of insect tissues [2]. In support of this notion, it has been shown that the genomes of recently acquired mutualistic insect endosymbionts maintain genes similar to virulence factors and toxins that are found in related plant and animal pathogens [13]–[18].
In the current study we describe the discovery of a novel human-infective bacterium, designated “strain HS”, isolated from a patient who sustained a hand wound following impalement with a tree branch. Phylogenetic analyses show that strain HS is a member of the Sodalis-allied clade of insect endosymbionts. Comparative analyses of the genome sequences of strain HS and related insect symbionts suggest that close relatives of strain HS gave rise to mutualistic associates in a wide range of insect hosts.
A 71-year-old male presented to his primary care physician for a routine physical examination three days after sustaining a puncture wound to the right hand. The patient fell and was impaled between the thumb and forefinger by a ∼1 cm diameter branch while removing branches from a dead crab apple tree. Upon presentation the patient denied fever or other constitutional symptoms and had a mild peripheral blood monocytosis (11.8%; reference range = 1.7–9.3%). A palpable cyst was noted in the right hand at the sight of impalement. Warm compresses were applied and cephalexin was prescribed at a dose of 500 mg four times daily for 10 days. The patient was evaluated again three days later due to continuing wound pain. The cyst was drained by aspiration and serosanguineous fluid was submitted for Gram stain and bacterial culture. The Gram stain showed scattered white blood cells, but no bacteria were visualized. A follow-up visit seven days later revealed the presence of an abscess, although the patient was afebrile and without local lymphadenopathy. The abscess was again drained by aspiration and the patient was advised to consult an orthopedic surgeon for evaluation. Subsequent surgery, approximately six weeks later, removed several foreign bodies from the wound and the patient recovered on a second course of cephalexin without incident. Two days after the original cyst aspiration, small numbers of gram negative rods resembling enteric bacteria were isolated on MacConkey agar at 35°C and 5% CO2. Colonies were wet, mucoid, variable in size, and slowly fermented lactose. The isolate could not be definitively identified by a manual phenotypic method (RapID ONE, Remel, Lenexa KS) and was misidentified as Escherichia coli at 98% confidence by an automated system (Phoenix, BD Diagnostics, Sparks, MD).
Phylogenetic analysis of 16S rRNA placed strain HS in a well-supported clade comprising Sodalis-allied insect endosymbionts sharing >97% sequence identity in their 16S rRNA sequences (Figure 1), which is a commonly used threshold for species-level conservation among bacteria [19]. Aside from strain HS, the closest non-insect associated relative of this clade is Biostraticola tofi, which was isolated from a biofilm on a tufa deposit in a hard water rivulet [20]. However, B. tofi shares only 96.5% sequence identity in 16S rRNA with its closest insect associated relative (S. glossinidius), while strain HS shares >99% sequence identity with the primary endosymbionts of the grain weevils Sitophilus oryzae and S. zeamais and with recently discovered endosymbionts from the chestnut weevil, Curculio sikkimensis and the stinkbug, Cantao occelatus [21]–[23]. Analysis of a protein-coding gene, groEL, corroborated these findings, confirming that strain HS is a close relative of the grain weevils, chestnut weevil and stinkbug endosymbionts (Figure 1).
To compare the genome sequences of strain HS and related Sodalis-allied endosymbionts, we aligned a draft sequence assembly of strain HS, comprising a total of 5.15 Mb of DNA in 271 contigs, with the complete genome sequences of the tsetse fly secondary endosymbiont, S. glossinidius (4.3 Mb) [24], [25], and the recently completed sequence of Sitophilus oryzae primary endosymbiont (SOPE; 4.5 Mb). The resulting alignments (Figure 2) reveal a remarkable level of conservation in gene content and organization between strain HS, S. glossinidius and SOPE. To determine if this high level of conservation is simply a consequence of the close evolutionary relationship between these bacteria, we also constructed a whole genome sequence alignment between strain HS and Dickeya dadantii, which represents the next most closely related free-living bacterium whose whole genome sequence is available (Figure S1). This alignment shows that strain HS and D. dadantii are substantially more divergent in terms of their gene inventories, consistent with the notion that they occupy distinct ecological niches. Considering the alignments between strain HS, S. glossinidius and SOPE, it is notable that while the genome sequences of strain HS and S. glossinidius display an increased level of co-linearity, the relationship between strain HS and SOPE is predicted to be closer based on the fact that they share a higher level of genome-wide sequence identity (Figure 2). The genome sequences of strain HS and S. glossinidius demonstrate a typical pattern of polarized nucleotide composition in each replichore (G+C skew, Figure 2), whereas the SOPE genome has numerous perturbations in G+C skew that must result from recent chromosome rearrangements. These rearrangements likely arose as a consequence of intragenomic recombination events between repetitive insertion sequence (IS)-elements, which are highly abundant in the SOPE genome (Figure S2), and have been documented as a causative agent of deletogenic rearrangements in other bacteria [26]–[28].
Although the gene inventories of strain HS, S. glossinidius and SOPE share many genes in common, as expected given their close evolutionary relationship, each bacterium also maintains a fraction of unique genes. In strain HS we identified a total of 1.9 Mb of DNA encoding genes not found in either S. glossinidius or SOPE that are classified in a wide range of functional categories (Figure 3). This indicates that strain HS has many unique genetic and biochemical properties, and is consistent with the observation that strain HS, unlike the fastidious and microaerophilic S. glossinidius [14], grows under atmospheric conditions on minimal media. In addition, strain HS maintains a number of unique genes sharing high levels of sequence identity with virulence factors found in both animal and plant pathogens, including an Hrp-type effector protein that is characteristically utilized by plant pathogenic bacteria [29] (Table S1). This may be indicative of the ability of strain HS to sustain infection in plant tissues. In comparison with strain HS, the unique fractions of the S. glossinidius and SOPE chromosomes are composed almost exclusively of components of mobile genetic elements, including integrated prophage islands and IS-elements. Following excision of these mobile genetic elements in silico prior to alignment, the resulting genome sequences of S. glossinidius (3.21 Mb) and SOPE (3.15 Mb) represent near-perfect subsets of the strain HS genome (Figure 2), indicating that S. glossinidius and SOPE are abridged derivatives of a strain HS-like ancestor.
To further understand genetic differences between strain HS, S. glossinidius and SOPE, we analyzed three genomic regions containing relatively high densities of pseudogenes in both S. glossinidius and SOPE (Figure 4). The most notable finding to arise from this comparison is the absence of pseudogenes in the three genomic regions of strain HS. Furthermore, our comparative analysis shows that S. glossinidius and SOPE each have a unique complement of pseudogenes. Indeed, even for orthologous genes that have been inactivated in both S. glossinidius and SOPE, mutations leading to gene inactivation in each insect symbiont genome are distinct, indicating that gene inactivation and loss took place independently in S. glossinidius and SOPE, mostly as a consequence of small frameshifting indels. However, it should also be noted that the reductions observed in the gene inventories of S. glossinidius and SOPE are very similar at the level of functional categories, indicating that the insect-associated lifestyle imposes similar constraints on the retention of genes encoding core functions such as replication, transcription, translation and energy generation (Figure 3). In order to determine the number of pseudogenes throughout the genome of strain HS, we performed a manual annotation and careful inspection of the complete draft strain HS sequence assembly. Out of a total of 4,002 intact candidate ORFs identified in the draft annotation (Table S1), only 48 (including phage and IS elements) were found to be translationally frameshifted or truncated by more than 10% of the size of their most closely related orthologs in the GenBank database (Table 1). This finding stands in stark contrast to the gene inventories of both S. glossinidius and SOPE, in which pseudogenes represent a substantial fraction of their total genomic coding capacity (Figure 2) [24], [25]. Thus, for both S. glossinidius and SOPE, the predominant evolutionary trajectory following obligate insect association involved the inactivation and/or loss of a substantial component of the ancestral (strain HS-like) gene inventory.
The close evolutionary relationships between strain HS, S. glossinidius and SOPE indicate that the respective insect symbioses are recent in origin. This raises the possibility that a subset of selectively neutral genes in the S. glossinidius and SOPE genomes have not yet accumulated mutations that lead to disruption of their open reading frames. Such “cryptic” pseudogenes are assumed to have no adaptive benefit in the symbiosis and are expected to accumulate nonsense and/or frameshifting mutations in the future [30]. To determine if the genomes of S. glossinidius and SOPE maintain cryptic pseudogenes, we compared the average size of all strain HS genes with the average sizes of strain HS orthologs that are classified either as intact, absent (lost via large deletion) or pseudogenes (visibly disrupted) in the S. glossinidius and SOPE genomes (Figure 5). First, it is important to note that the average size of the absent strain HS orthologs in S. glossinidius and SOPE is not significantly different from the average size of all strain HS ORFs, indicating that large deletion events are not significantly biased with respect to size. However, in both S. glossinidius and SOPE, genes in the pseudogene class were found to have a larger average size in comparison to all strain HS orthologs. Similarly, genes in the intact class were found to have a smaller size in comparison to all strain HS orthologs. This can be explained by the fact that larger genes have an increased likelihood of accumulating at least one disrupting mutation in a given time frame. Based on the same logic, we can infer that the intact gene class contains a subset of smaller, cryptic pseudogenes that have not yet had sufficient time to accumulate any nonsense or frameshifting mutations. Furthermore, since the difference between the average size of intact and disrupted genes is significantly larger in SOPE (192 bases) in comparison to S. glossinidius (77 bases), it follows that SOPE likely maintain a larger number of cryptic pseudogenes than S. glossinidius.
In a previous study, the numbers of cryptic pseudogenes in the recently derived aphid symbiont, Serratia symbiotica, were estimated by extrapolation from a Poisson distribution of disrupting mutations found in existing pseudogenes [30]. The expectation of a Poisson distribution is based on the assumption that the switch to an insect-associated lifestyle leads to the synchronous relaxation of selection on genes no longer required for persistence in an insect host [30]. In the case of both SOPE and S. glossinidius, plots of the densities of disrupting mutations in pseudogenes indicate that the data is overdispersed relative to a Poisson distribution (Figure 6). This effect is exacerbated when current ORF sizes are used for the calculation of mutation densities. This results from the fact that large deletions erase any evidence of previous disrupting mutations. In order to estimate the numbers of cryptic pseudogenes in SOPE and S. glossinidius, we used a Monte Carlo simulation in which a randomly selected class of candidate pseudogenes, selected from all strain HS genes, was permitted to accumulate random disrupting mutations over time, in accordance with ORF size. In this simulation, both pseudogene counts and size differences between the strain HS orthologs of intact and disrupted S. glossinidius and SOPE genes were recorded at regular intervals. The simulation was repeated with an increasing number of neutral genes until the size difference and pseudogene count matched the empirically determined values shown in Figure 5 and Table 1. For S. glossinidius and SOPE, matches were obtained when the predicted numbers of genes evolving under relaxed selection reached 1,470 and 1,530, respectively (Figure 7). Thus, although S. glossinidius and SOPE are predicted to have almost the same numbers of genes evolving under relaxed selection, the degeneration of pseudogenes is at a more advanced stage in S. glossinidius, and SOPE has a larger proportion of neutral genes that have not yet acquired any obvious disrupting changes. Assuming that the relaxation of selection was imposed synchronously at the onset of obligate insect-association, these results suggest that the SOPE-weevil symbiosis originated more recently than the S. glossinidius-tsetse fly symbiosis. This is further supported by a comparison of the estimates of corrected mutation density derived from the simulation (Figure 7). While SOPE is estimated to maintain only 2 disrupting mutations/kb of pseudogenes, S. glossinidius is estimated to maintain more than twice that density of disrupting substitutions (4.39 disrupting mutations/kb). On a related note, we were unable to utilize dN/dS ratios to identify cryptic pseudogenes in SOPE or S. glossinidius. This is likely due to the fact that stochastic variation resulting from differences in expression level, codon bias and other factors greatly exceeds any signal resulting from a recent relaxation of selection.
The transition to obligate insect-association is also known to catalyze base composition bias and accelerated polypeptide sequence evolution on the part of the symbiont [31]. The results outlined in Table 1 show that the genomic GC-contents of S. glossinidius and SOPE are lower than that of strain HS. However, to avoid any bias arising from the differential gene content of these organisms, we also performed comparative analyses focusing solely on orthologous sequences. This facilitated the comparison of 1,355 intact genes and 1,376 pseudogenes shared between strain HS and S. glossinidius, and 1,414 intact genes and 1,194 pseudogenes shared between strain HS and SOPE. Although the symbioses in the current study are anticipated to be relatively recent in origin, comparisons focusing on these shared sequences also show that both S. glossinidius and SOPE have reduced GC-contents relative to strain HS (Figure 8). This effect is most notable at 4-fold degenerate (GC4) sites in S. glossinidius, which demonstrate the highest levels of sequence divergence and AT-bias in comparison to orthologs from strain HS. Assuming that the onset of AT-bias is coincident with the origin of symbiosis, this further supports the notion that the symbiosis involving S. glossinidius is more ancient in origin. It is also notable that the number of substitutions at the 2nd codon position sites of pseudogenes (dGC2, Figure 8) is elevated by approximately the same extent (relative to intact genes) in S. glossinidius and SOPE. This implies that pseudogenes have been evolving under relaxed selection for approximately the same proportion of time since each symbiont diverged from strain HS. However, given that sequence divergence at silent sites (GC4) is greater between strain HS and S. glossinidius, this again invokes the interpretation that pseudogenes arose earlier in the S. glossinidius line of descent. It is also interesting to note that the level of divergence at GC2 sites (dGC2, Figure 8) relative to GC4 sites (dGC4, Figure 8) is greater in SOPE than in S. glossinidius. This can be explained by the fact that the pairwise comparison between strain HS and SOPE is expected to capture an increased proportion of mutations that are fixed in the insect-associated phase of life in which selection on polypeptide evolution is anticipated to be more relaxed.
Considering only those mutations that have led to gene inactivation, we found that the relative ratios of truncating (large) indels, frameshifting (small) indels and nonsense mutations are similar in SOPE and S. glossinidius (Table 2). Inspection of the data reveals that small frameshifting deletions constitute the most abundant class of mutations leading to gene inactivation. However, it should be noted that the effects of large deletions are, for obvious reasons, not captured in our analyses. Another important point is that IS-element insertions appear to have contributed relatively little to the overall spectrum of mutations leading to gene inactivation in SOPE, representing only 10% of the total count. Indeed, the majority of IS-elements in SOPE are located either in intergenic regions or, more commonly, clustered inside other IS-elements. One potential explanation is that IS-element insertions in genic sequences might be more deleterious towards processes of transcription and/or translation in the cell, such that pseudogenes with IS-element insertions are preferentially deleted relative to pseudogenes with nonsense point mutations or small indels. However, it is conspicuous that clustering of IS-elements has also been reported for mobile DNA elements found in eukaryotes, including the MITE elements found in plants [32] and mosquitoes [33], and the Alu and L1 elements found in the human genome [34]. The relative paucity of IS-elements in genic DNA is surprising given the fact that the SOPE genome has such a large number of pseudogenes that provide neutral space for IS-element colonization. However, the inability of IS-elements to occupy this territory can be rationalized as a consequence of an inherited adaptive bias that facilitates the avoidance of genic insertion. This makes sense when considering the perspective of an IS-element residing in a free-living bacterium that has relatively few dispensable genes. It also explains the propensity for IS-elements to insert themselves into the sequences of other IS-elements, because the safety of this approach has already been validated by natural selection. Clearly, in the case of SOPE, when the opportunity arose for expansion into novel territory (i.e. neutralized genic sequences), IS-elements were largely unable to overcome these basic evolutionary directives.
Phylogenetic analysis of strain HS indicates that it shares a close relationship with the Sodalis-allied endosymbionts that are found in a wide range of insect hosts, including tsetse flies, weevils, lice and stinkbugs. In terms of 16S rRNA sequence identity, strain HS is most closely related to endosymbionts found in the chestnut weevil, Curculio sikkimensis and the stinkbug, Cantao occelatus. Interestingly, only limited numbers of these insects maintain Sodalis-allied endosymbionts in their natural environment [21]–[23], suggesting that they do not maintain persistent (maternally-transmitted) infections. Furthermore, it is notable that the sequences from strain HS, C. sikkimensis and C. occelatus are localized on very short branches in our phylogenetic trees, indicating that these particular lineages are evolving slowly in comparison to other Sodalis-allied endosymbionts. This low rate of molecular sequence evolution, along with the observation that the strain HS genome shows no sign of the characteristic degenerative changes that are known to accompany the transition to the obligate host-associated lifestyle, leads us to propose that strain HS represents an environmental progenitor of the Sodalis-allied clade of insect endosymbionts.
Closely related members of the Sodalis-allied clade of insect endosymbionts have now been identified in a wide range of distantly related insect taxa, including some that are known to feed exclusively on plants and others that are known to feed exclusively on animals [8]. Although strain HS was isolated from the wound of a human host, it is difficult to assess the extent of its pathogenic capabilities, due to the fact that antibiotic treatment commenced three days prior to microscopic examination and culturing. In addition, the available evidence indicates that the original source of the infection was a branch from a dead crab apple tree. This implies that strain HS was present either on the bark or in the woody tissue of this tree, possibly acting as a pathogen or saprophyte. Furthermore, it is interesting to note that C. sikkimensis and C. occelatus, whose symbionts are most closely related to strain HS, are both known to feed on trees [35], [36]. In addition, some wood and bark-inhabiting longhorn beetles, including Tetropium castaneum (Figure 1) have recently been found to maintain Sodalis-allied endosymbionts [37]. Moreover, the ability of strain HS to persist in both plant and animal tissues is compatible with the observation that diverse representatives of both herbivorous and carnivorous insects have acquired Sodalis-allied symbionts.
In a comparative sense, relationships involving the Sodalis-allied endosymbionts are considered to be relatively recent in origin. Indeed, evidence of host-symbiont co-speciation only exists in the case of grain weevils, Sitophilus spp., which were estimated to have co-evolved with their Sodalis-allied endosymbionts for a period of around 20 MY, following the replacement of a more ancient lineage of endosymbionts in these insects [38], [39]. The notion of a recent origin of the Sodalis-allied endosymbionts is further supported by the fact that the whole genome sequence of S. glossinidius is substantially larger than that of long-established mutualistic insect endosymbionts, and is close to the size of related free-living bacteria [24]. However, the S. glossinidius genome does have an unusually low coding capacity resulting from the presence of a large number of pseudogenes [24], [25]. This suggests that S. glossinidius is at an intermediate stage in the process of genome degeneration, in which many protein coding genes have been inactivated by indels and nonsense mutations but have not yet been deleted from the genome. In the current study we show that the genome of the grain weevil symbiont, SOPE, is at a similar stage of degeneration as evidenced by the presence of a comparable number of pseudogenes and a large number of repetitive insertion sequence elements.
In a comparative sense, it is interesting to note that SOPE and strain HS share a substantially higher level of sequence similarity, genome-wide, in comparison to S. glossinidius and strain HS (Figure 2). In the context of the progenitor hypothesis, the disparity in the relationship between strain HS, SOPE and S. glossinidius can be explained by the idea that there may be a substantial level of diversity among free-living relatives of the Sodalis-allied symbionts in the environment, and that we simply happened to characterize a representative that is more closely related to the ancestral progenitor of SOPE. While this is likely to be true to some extent, the close relationship between strain HS and SOPE can also be explained by the notion that the SOPE-grain weevil symbiosis has a more recent origin than the S. glossinidius-tsetse symbiosis. Our results provide several compelling lines of evidence in support of this idea. Most significantly, we found that the pseudogenes of S. glossinidius contain a higher average density of disrupting mutations relative to their counterparts in SOPE. This suggests that the pseudogenes of S. glossinidius have been evolving under relaxed selection for a longer period of time, consistent with the hypothesis of a more ancient origin of host association catalyzing the neutralization of these genes. In addition, the genome of SOPE is predicted to have a larger proportion of “cryptic” pseudogenes; genes evolving neutrally that have not yet had sufficient time to accumulate nonsense or frameshifting mutations that disrupt their translation. Finally, it is notable that the GC4 sites of S. glossinidius have a higher AT-content than those of strain HS and SOPE (Figure 8). Assuming that the AT-bias at GC4 sites accumulates in a clock-like manner following the onset of the symbiosis, this again supports a more ancient origin for the symbiosis involving S. glossinidius.
In the current study, a comparative analysis of the genome sequences of strain HS, SOPE and S. glossinidius has provided an unprecedentedly detailed view of the nascent stages of genome degeneration in symbiosis. Taken together, our results indicate that irreversible degenerative changes, including gene inactivation and loss, in addition to base composition bias, commence rapidly following the onset of an obligate relationship. Indeed, the close relationship observed between strain HS and SOPE illustrates the potency of the degenerative evolutionary process at an early stage in the evolution of a symbiotic interaction. This is exemplified by the fact that SOPE is predicted to have lost 55% of its ancestral gene inventory (34% via gene loss and 21% via gene inactivation) in a period of time sufficient to incur a substitution frequency of only 4.3% at the highly variable GC4 sites of intact protein coding genes (Figure 8). Although estimates of genome wide synonymous clock rates vary by several orders of magnitude in bacteria [40], an estimate of μs = 2.2×10−7, derived recently for another insect endosymbiont, Buchnera aphidicola [41], places the divergence of strain HS and SOPE at only c. 28,000 years, which is much more recent than previous estimates obtained for the origin of the SOPE symbiosis [38], [39].
While the broad distribution of recently derived endosymbionts in phylogenetically distant insect hosts has previously been attributed to interspecific symbiont transfer events [10], [11], the results outlined in the current study indicate that diverse insect species can also acquire novel symbionts through the domestication of bacteria that reside in their local environment. In the case of S. glossinidius and SOPE, our comparative analyses support the notion that these symbionts were acquired independently, as evidenced by the presence of distinct mutations in shared pseudogenes. This also implies that symbionts rapidly become specialized towards a given host, likely restricting their abilities to switch hosts. Although the current study highlights the first description of a close free-living relative of the Sodalis-allied symbionts, it should be noted that environmental microbial diversity is vastly undersampled [42]. Thus, it is conceivable that close relatives of extant insect endosymbionts, such as strain HS, are widespread in nature and provide ongoing opportunities for a wide range of insect hosts to domesticate new symbiotic associates. Furthermore, since many insects serve as vectors for plant and animal pathogens [43], it is conceivable that mutualistic associations arise as a consequence of the domestication of vectored pathogens. This hypothesis is compelling because such pathogens are not expected to negatively impact the fitness of their insect vectors [44] and under those circumstances the transition to a mutualistic lifestyle could be achieved without any need to attenuate virulence towards the insect host.
Strain HS was isolated on MacConkey agar at 35°C and 5% CO2. 16S rRNA and groEL sequences were amplified from strain HS using universal primers. Following cloning of PCR products, eight clones were sequenced from each gene and consensus sequences were used in phylogenetic analyses. Sequence alignments were generated for 16S rRNA and groEL using MUSCLE [45]. PhyML [46] was then used to construct phylogenetic trees using the HKY85 [47] model of sequence evolution with 25 random starting trees and 100 bootstrap replicates.
Synchronous cultures of Sitophilus oryzae and Sitophilus zeamais were reared on organic soft white wheat grains and corn kernels respectively, and maintained at 25°C with 70% relative humidity. Bacteriomes (containing the bacterial endosymbionts SOPE and SZPE) were isolated from 5th instar S. oryzae and S. zeamais larvae by dissection and homogenized at a sub-cellular level to release bacteria from host bacteriocyte cells; bacterial cells were then separated from host cells via centrifugation (2,000×g, 5 min). Total genomic DNA was then isolated from bacteria using the Qiagen DNeasy Blood & Tissue Kit (Qiagen, Valencia, CA).
Six mg of genomic DNA was hydrodynamically sheared in 5 mM Tris, 1 mM EDTA, 100 mM NaCl (pH 8) buffer to a mean fragment size of 10 kb. The sample was washed and concentrated by ultrafiltration in a Centricon-100 (Millipore, Billerica, MA) and eluted in 250 µl of 2 mM Tris (pH 8). The fragments were end-repaired by treatment with T4 DNA polymerase (New England Biolab, Beverly, MA) to generate blunt ends. The DNA was then extracted with phenol/chloroform, ethanol precipitated, and 5′ phosphorylated with T4 polynucleotide kinase (NEB). Ten mM of double-stranded, biotinylated oligonucleotide adaptors were blunt-end ligated onto the sheared genomic fragments at room temperature for 25 h using 10,000 cohesive end units of high concentration T4 DNA ligase (NEB). Unligated adaptors were removed by ultrafiltration in a Centricon-100. The adaptored fragments were bound to streptavidin-coated magnetic beads (Invitrogen), and after binding and washing, the adaptored genomic fragments were eluted in 10 mM TE (pH 8). Fragments in the 9.5–11.5 kb size range were gel purified after separation on a 0.7% 1× TAE agarose gel, and the purified DNA was electroeluted from the agarose and desalted by ultrafiltration in a Centricon-100.
pWD42 vector (GenBank: AF129072.1) was linearized by digestion with BamHI (NEB) at 37°C for 4 h, extracted with phenol/chloroform, ethanol precipitated and resuspended in 100 ml of 2 mM Tris (pH 8.0). Ten picomoles of double-stranded, biotinylated oligo adaptors were ligated onto the BamHI-digested vector at 25°C for 16 hrs using 4,000 units of T4 DNA ligase (NEB). Unligated adaptors were removed by ultrafiltration in a Centricon-100. The adaptored vector was bound to streptavidin-coated magnetic beads and the non-biotinylated adaptored vector was eluted in 10 mM TE (pH 8). One hundred ng each of adaptored vector and genomic DNA were annealed without ligase in 10 ml of T4 DNA ligase buffer (NEB) at 25°C for one hour. Two ml aliquots of the annealed vector/insert were transformed into 100 ml of XL-10 chemically competent E. coli cells (Agilent Technologies, Santa Lara, CA) and plated on LB agar plates containing 20 µg/ml ampicillin. A total of 23,808 bacterial colonies were picked into 96-well microtiter dishes containing 600 ml of terrific broth (TB)+20 µg/ml ampicillin and grown at 30°C for 16 h. Fifty ml aliquots were removed from the library cultures, mixed with 50 ml of 14% DMSO, and archived at −80°C. The 200 ml cultures were diluted 1∶4 in TB amp and runaway plasmid replication was induced at 42°C for 2.25 h. Plasmid DNA was purified by alkaline lysis, and cycle sequencing reactions were performed with forward and reverse sequencing primers using ABI BigDye v3.1 Terminator chemistry (Applied Biosystems, Foster City, CA). The reactions were ethanol precipitated, resuspended in 15 ul of dH2O, and sequence ladders were resolved on an ABI 3730 capillary instrument prepared with POP-5 capillary gel matrix.
Following elimination of any sequences encoding contaminating plasmid vector or host insect sequences, 38,755 shotgun reads were assembled using the Phusion assembler [48] using the paired-end sequences as mate-pair assembly constraints. Contig assemblies were viewed and edited in Consed [49], and reads with high quality (Phred>20) discrepancies were disassembled. After inspection and manual assembly to extend contigs, gaps were closed by iterative primer walking (895 primer walk sequence reads) and gamma-delta transposon-mediated full-insert sequencing of plasmid clones (6,165 sequence reads across 103 transposed plasmid clones) using an established protocol [50]. The average insert size of the plasmid library in the finished SOPE assembly was found to be 8.2 kb.
The SOPE fosmid library was constructed using the Epicenter EpiFOS Fosmid Library Production Kit (Epicentre Biotechnologies, Madison, WI), using SOPE total genomic DNA. 1,404 paired-end reads were generated from 702 fosmid inserts and mapped onto the assembly derived from the plasmid shotgun sequencing for validation (Figure S2).
Strain HS genomic DNA was isolated from liquid culture using the Qiagen DNeasy Blood & Tissue Kit (Qiagen, Valencia, CA). Five micrograms of total genomic DNA was used to construct a paired-end sequencing library using the Illumina paired-end sample preparation kit (Illumina, Inc. San Diego, CA) with a mean fragment size of 378 base pairs. This library was then sequenced on the Illumina GAIIx platform generating 26,891,485 paired-end reads of 55 bases in length.
Paired-end reads were quality filtered using Galaxy [51], [52] and low quality paired-end reads (Phred<20) were discarded. The remaining 17,054,405 reads were then assembled using Velvet [53] with a k-mer value of 37, with expected coverage of 119 and a coverage cutoff value of 0.296. The resulting assembly consisted of 271 contigs with an N50 size of 231,573 and a total of 5,135,297 bases. No sequences were found to share significant sequence identity with genes encoding plasmid replication functions, suggesting that strain HS does not maintain any extrachromosomal elements.
The assembled draft genome sequence of strain HS was annotated by automated ORF prediction using GeneMark.hmm [54]. The annotation was then adjusted manually in Artemis [55] using the published Sodalis glossinidius genome sequence [25] as a guide. ORFs were annotated as putatively functional only if (i) their size was ≥90% of the most closely related ORF derived from a free-living bacterium in the GenBank database, and (ii) they did not contain any frameshifting indel(s).
Curation of the strain HS genome sequence was performed in Artemis [55]. ORFs were classified into COG categories using the Cognitor software [56]. Syntenic links shown in Figure 2 were determined by pairwise nucleotide alignments between strain HS contigs and S. glossinidius (GenBank: NC_007712.1) or the finished SOPE genome using the Smith-Waterman algorithm as implemented in the cross_match algorithm [49]. Figure 2 was prepared from data obtained from these alignments using CIRCOS [57]. The metrics depicted in Table 1, Table 2, and Figure 8 were computed from pairwise nucleotide sequence alignments of strain HS, S. glossinidius and SOPE ORFs using custom scripts. Candidate genes were classified as intact orthologs when their alignment spanned >99% of the HS ORF length (or 90% for ORFs <300 nucleotides in size) and did not contain frameshifting indels or premature stop codons.
A simple Monte Carlo approach was implemented to simulate the evolution of pseudogenes in S. glossinidius and SOPE. The simulation facilitated the progressive accumulation of random mutations in all strain HS orthologs of both intact genes and pseudogenes identified in the current S. glossinidius or SOPE gene inventories. Mutations accumulated in proportion to ORF size in a randomly selected class of neutral genes of user-defined size over a defined number of mutational cycles. At preset cycle intervals, the simulation recorded (i) the difference in size between intact and disrupted sequences, (ii) the number of neutral genes that have accumulated one or more disrupting mutations, and (iii) the density of disrupting mutations, which was calculated based on the cumulative size of all neutral genes.
The GenBank accession numbers for sequences used in Figure 1 are as follows: Endosymbiont of Circulio sikkimensis 16S rRNA, (AB559929.1), groEL, (AB507719); Vibrio cholerae 16S rRNA, (NC_002506.1), groEL, (NC_002506.1); Dickeya dadantii 16S rRNA (CP002038.1), groEL, (CP002038.1); Escherichia coli 16S rRNA, (NC_000913.2), groEL, (NC_000913.2); Candidatus Moranella endobia 16S rRNA, (NC_015735), groEL, (NC_015735); Sodalis glossinidius 16S rRNA, (NC_007712.1), groEL, (NC_007712.1); Yersinia pestis 16S rRNA, (NC_008150.1), groEL, (NC_008150.1); Wigglesworthia glossinidia 16S rRNA, (NC_004344.2), groEL, (NC_004344.2); Candidatus Blochmannia pennsylvanicus 16S rRNA, (NC_007292), groEL, (NC_007292); Endosymbiont of Cantao ocellatus 16S rRNA, (AB541010), groEL, (BAJ08314); Endosymbiont of Columbicola columbae 16S rRNA, (AB303387), groEL, (JQ063388); Sitophilus zeamais primary endosymbiont 16S rRNA, (AF548142), groEL (JX444567); Sitophilus oryzae primary endosymbiont 16S rRNA, (AF548137), groEL (AF005236); Strain HS 16S rRNA, (JX444565), groEL (JX444566). The GenBank accession numbers for sequences used in Figure 4 are as follows: Strain HS Figure 4A (JX444569), Figure 4B (JX444571), Figure 4C (JX444572); Sitophilus oryzae primary endosymbiont Figure 4A (JX444568), Figure 4B (JX444570), Figure 4C (JX444573).
|
10.1371/journal.ppat.1005805 | Temporal Dynamics of CD8+ T Cell Effector Responses during Primary HIV Infection | The loss of HIV-specific CD8+ T cell cytolytic function is a primary factor underlying progressive HIV infection, but whether HIV-specific CD8+ T cells initially possess cytolytic effector capacity, and when and why this may be lost during infection, is unclear. Here, we assessed CD8+ T cell functional evolution from primary to chronic HIV infection. We observed a profound expansion of perforin+ CD8+ T cells immediately following HIV infection that quickly waned after acute viremia resolution. Selective expression of the effector-associated transcription factors T-bet and eomesodermin in cytokine-producing HIV-specific CD8+ T cells differentiated HIV-specific from bulk memory CD8+ T cell effector expansion. As infection progressed expression of perforin was maintained in HIV-specific CD8+ T cells with high levels of T-bet, but not necessarily in the population of T-betLo HIV-specific CD8+ T cells that expand as infection progresses. Together, these data demonstrate that while HIV-specific CD8+ T cells in acute HIV infection initially possess cytolytic potential, progressive transcriptional dysregulation leads to the reduced CD8+ T cell perforin expression characteristic of chronic HIV infection.
| Previous studies have demonstrated that HIV-specific CD8+ T cells are critical for the initial control of HIV infection. However, this control is typically incomplete, being able to neither clear infection nor maintain plasma viremia below undetectable levels. Mounting evidence has implicated CD8+ T cell cytotoxic capacity as a critical component of the HIV-specific response associated with spontaneous long-term control of HIV replication. CD8+ T cell cytotoxic responses are largely absent in the vast majority of HIV chronically infected individuals and it is unclear when or why this functionality is lost. In this study we show that HIV-specific CD8+ T cells readily express the cytolytic protein perforin during the acute phase of chronic progressive HIV infection but rapidly lose the ability to upregulate this molecule following resolution of peak viremia. Maintenance of perforin expression by HIV-specific CD8+ T cells appears to be associated with the expression level of the transcription factor T-bet, but not with the T-bet paralogue, Eomes. These findings further delineate qualitative attributes of CD8+ T cell-mediated immunity that may serve as targets for future HIV vaccine and therapeutic research.
| CD8+ T cells play a central role in the control of HIV replication. During acute infection the emergence of HIV-specific CD8+ T cells correlates with resolution of peak viremia [1, 2], and in the nonhuman primate model experimental depletion of CD8+ T cells prior to infection with simian immunodeficiency virus delays resolution of acute viremia until the CD8+ T cell pool is reconstituted [3]. Further evidence of the immunologic pressure exerted by CD8+ T cells is manifest by CTL escape mutations throughout all phases of HIV infection and the association of certain MHC class I alleles with superior control of viral replication [4–9]. However, for the vast majority of infected individuals control is incomplete and ultimately fails in the absence of therapy. A better understanding of the CD8+ T cell response to HIV may inform the design of vaccines, therapeutics, or eradication strategies designed to stimulate or potentiate the natural response to infection resulting in better, if not complete, control.
The CD8+ T cell response to viral infection is multifaceted, including the ability to proliferate, produce multiple cytokines and chemokines, degranulate, and induce cytolysis upon contact with infected targets [10]. During chronic progressive infection, HIV-specific CD8+ T cells have impaired proliferative potential [11–13], are less capable of multifunctional responses [14, 15], and have reduced cytotoxic capacity [16–20]. The primary mechanism by which CD8+ T cells kill virally infected cells is via exocytosis of granules containing the cytolytic proteins perforin and granzyme B [21, 22]. Control of HIV viremia has been associated with the ability of CD8+ T cells from chronically HIV-infected donors to upregulate these cytotoxic effector molecules following in vitro culture [18], and we have shown that CD8+ T cell cytotoxic potential, defined by the ability to rapidly upregulate perforin following brief stimulation ex vivo, correlates inversely with viral load [16].
Effector CD8+ T cell development is coordinated by an array of transcription factors [23]. Murine studies have identified the T-box transcription family members T-bet and eomesodermin (Eomes) as important regulators of the differentiation and function of cytotoxic effector T cells [24–26]. T-bet positively regulates genes associated with effector functions including perforin, granzyme B, and IFN-γ [27, 28], whereas Eomes is associated with the expression of perforin as well as proteins involved in maintenance of memory CD8+ T cells [24, 26, 29, 30]. While previous studies suggested a level of redundancy in the gene targets of these transcription factors, recent data show that the balance of T-bet and Eomes expression within a cell is a determinant of the differentiation pathway and functionality of the cell [30–34]. In the context of chronic HIV infection, HIV-specific CD8+ T cells with high levels of T-bet demonstrate greater overall functionality and maintain the ability to express perforin whereas cells with a T-betLoEomesHi phenotype are less differentiated, less functional, exhausted, and express little to no perforin [28, 32]. Notably, during chronic progressive infection the T-betLoEomesHi phenotype dominates the HIV-specific CD8+ T cell pool [32]. It remains unclear if low T-bet levels and the associated deficiency in perforin expression results from progressive loss on the part of responding HIV-specific CD8+ T cells or if responding cells are inherently dysfunctional throughout the infection period.
Much of our current knowledge regarding the dynamics of CD8+ T cell responses during acute infection is derived from murine models, particularly following infection with lymphocytic choriomeningitis virus, gammaherpesvirus, or influenza [35–37]. Infection by these viruses induces rapid and substantial activation and expansion of antigen-specific CD8+ T cells. Following resolution of acute viremia, the virus-specific population contracts, giving rise to memory cells that provide long-term protection. Human antiviral CD8+ T cell responses have primarily been assessed in the context of chronic infection, after the memory pool has been established [10, 38–41]. Recent studies have examined development of human CD8+ T cell responses to a range of primary infections, including attenuated yellow fever virus, attenuated vaccinia virus, influenza, tick-borne encephalitis virus (TBEV), hantavirus, and Epstein-Barr virus [42–47], demonstrating that antigen-specific cells have immediate cytotoxic capacity directly ex vivo during the acute phase of these infections. The few studies to examine the earliest responses to HIV showed that HIV-specific CD8+ T cells have limited functionality during the acute phase of infection but did not assess cytotoxic potential or regulation by T-bet or Eomes [48, 49], leaving the question unresolved as to whether these effector molecules are induced during acute infection.
Here, we examined the temporal dynamics of the CD8+ T cell effector response in peripheral blood of subjects experiencing acute primary HIV infection. We found that infection elicited a robust and highly activated response with immediate cytotoxic potential within the peripheral CD8+ T cell pool and that cells responding to short in vitro stimulation with HIV peptides were able to degranulate and rapidly upregulate perforin de novo. However, HIV-specific CD8+ T cells rapidly lost the ability to upregulate perforin following resolution of peak viremia. Loss of perforin expression coincided with a concurrent reduction in the expression of T-bet, but not Eomes, on a per-cell basis. Our data provide evidence of a robust and physiologically appropriate response during the earliest phase of acute HIV infection that is rapidly lost during progressive chronic infection, due in part to an inability to express sufficient levels of T-bet to properly drive effector differentiation.
Longitudinal samples were obtained from 32 subjects experiencing primary HIV infection (Fig 1A), 28 of whom had at least one acute time point (36 time points total; median 54 d from infection, range 23–100 d) and 23 with at least one chronic time point (40 time points total; median 551 d, range 367–880 d). Samples were drawn from three separate cohorts of acutely infected individuals: the CHAVI001 acute-infection cohort, the Montreal Primary Infection cohort, and the RV217/ECHO cohort. These cohorts provided broad geographical representation including North America, East Africa, Malawi, and Thailand (S1 Table). Subjects were antiretroviral therapy naïve at all time points, consistent with the standard of care at the time of study, and none controlled viral load to undetectable levels (Fig 1B). The mean peak viral load was 5.2 log10 RNA copies/ml for the entire study population (7.0 log10 RNA copies/ml for the better-characterized RV217 donors) and 4.42 log10 RNA copies/ml at set point. Peripheral blood CD4+ T cell counts and CD8+ T cell counts both declined over the study period (average rates of 80 cells/mm3 per year and 75 cells/mm3, respectively; Fig 1C and 1D). Samples from 41 seronegative healthy donors, including pre-infection time points for the 11 RV217 acute subjects (median -210 d from infection, range -41 to -478 d; Fig 1A and S1 Table), were analyzed for comparison.
To determine if different phases of infection were associated with changes in circulating CD8+ T cell differentiation and activation, we assessed the size and composition of the memory CD8+ T cell pool (S1 Fig). Relative to HIV-negative donors, HIV-infected subjects had a significantly larger memory (non-CCR7+CD45RO-) CD8+ T cell pool in both the acute and chronic phases of infection (Fig 2A and 2B). Of note, the frequency of total memory CD8+ T cells at the earliest post-infection time points inversely correlated with peak viral load, but not with set point viral load (Figs 2C and S2A). In addition to the larger memory pool we also observed a shift in the distribution of memory subsets in infected subjects, with significantly higher proportions of central memory (CCR7+CD45RO+) and, predominately, effector memory (CCR7-CD45RO-) subsets during acute infection (Fig 2D). Only the effector memory pool remained significantly elevated into the chronic phase. There was no difference in the proportion of the effector cell pool (CCR7-CD45RO-) during either phase of infection, although the relative frequency of these cells did appear to be larger as infection progressed (Fig 2D).
When we examined the activation state of the memory pool for four RV217 subjects by measuring surface expression of HLA-DR, we found massive levels of activation within the memory CD8+ T cell compartment following HIV infection (Fig 2E), in agreement with recent data from Ndhlovu et al. [48]. To determine if this population of highly activated cells expressed cytolytic molecules directly ex vivo we measured perforin content. We found that almost all HLA-DR+ cells expressed perforin during the acute phase (Fig 2F). In addition, we observed a significantly greater proportion of perforin+ cells in both acute and chronic phases of infection compared to healthy donors (Fig 2G). There was, however, no significant association between the frequency of perforin+ CD8+ T cells and viral load at any time point (S2B Fig). Together, these data show that during acute HIV infection a large proportion of the peripheral CD8+ T cell pool is highly activated and primed to exert cytotoxic effector activity but the absolute magnitude of total cytotoxic CD8+ T cells does not predict set point viral load.
We next examined if the large frequency of cytotoxic CD8+ T cells observed during acute HIV infection was consistent across other acute viral infections. We compared the total CD8+ T cell responses of subjects from the RV217 cohort with those of HIV-negative individuals who were vaccinated with attenuated vaccinia virus (VV) or attenuated yellow fever virus (YFV)-17D, or experimentally infected with a H1N1 strain of influenza virus (S3A–S3C Fig). Vaccination with VV or YFV elicits a robust and highly specific CD8+ T cell response that peaks approximately two weeks after inoculation and is largely resolved by four weeks [46]. The peripheral CD8+ T cell response to influenza is less robust, peaks at 1–2 weeks, and resolves by four weeks post-infection [43].
Consistent with the comparison between healthy donors and acute phase HIV infection (Fig 2B), both the total memory CD8+ T cell pool and the effector memory subset increased significantly from pre- to acute HIV infection (Figs 3A and S4E). There was also a significant increase in the proportion of perforin+ cells over the first thirty days of infection, with almost all (>90%) circulating memory CD8+ T cells expressing perforin in some donors (Fig 3E). When we examined the CD8+ T cell responses to in vivo stimulation following vaccination with VV or YFV, or infection with influenza, we did not observe significant changes in the size or distribution of the peripheral memory pool (Figs 3B–3D and S4). We did find increased levels of activated HLA-DR+ cells in some donors after vaccination with VV and YFV, but frequencies of perforin+ cells remained relatively stable throughout the entire vaccine course (Figs 3F, 3G and S5A–S5D). Only infection with influenza resulted in a slight but significant increase in perforin+ cells at d28 post-infection (Fig 3H). While these models of acute viral infections do have limitations in their use as comparators for our HIV-infected donors (e.g. different antigen loads, different localizations, and more precise timing of infection), overall these data show the dramatic increase in cytotoxic cells that takes place in the peripheral blood of HIV acutely infected subjects is significantly more pronounced compared to live-attenuated vaccination or influenza infection.
We next sought to determine if the cytotoxic potential of HIV-specific cells demonstrated similar dynamics to the total memory CD8+ T cell pool during acute to chronic HIV infection. To identify HIV-specific cells we focused on the detection of IFN-γ production and CD107a-marked degranulation following a short-term in vitro stimulation with peptides derived from the HIV-1 Gag and Nef proteins [15, 48–50]. In agreement with previous studies that evaluated HIV-specific cells longitudinally by functional responses or tetramer staining [49, 51], we found no difference in the absolute magnitude of responding cells for either protein over time (Figs 4A and S6A). Consistent with the memory distribution of the total CD8+ T cell pool, Gag-specific cells largely had an effector memory phenotype in the acute phase of infection but became more equally distributed between effector and effector memory subsets for early chronic time points (Fig 4B and 4C). Also in agreement with previous data, cells tended to degranulate more readily than upregulate IFN-γ in the acute phase of infection (Figs 4D and S6B)[48, 49]. The high proportion of degranulating cells suggested that the HIV-specific response might be cytotoxic over the course of infection, as analysis of the total CD8+ T cell pool had indicated. However, degranulation is not an absolute surrogate of cytolytic potential [16, 52], nor does it indicate whether the cells will continue to be cytotoxic following the initial granule release [53]. To assess cytotoxic potential more directly, we measured perforin expression levels within the Gag- and Nef- specific cells (Figs 4E and S6C). The majority of cells that responded to direct ex vivo stimulation rapidly upregulated perforin during the earliest time points following infection, suggesting that the early HIV-specific response was likely highly cytotoxic. In contrast to the bulk memory CD8+ T cell pool, however, as acute viremia was resolved there was a rapid loss of perforin expression by both HIV-1 Gag- and Nef-specific CD8+ T cells (Figs 4F and S6D).
A large proportion of HIV-specific CD8+ T cells have previously been shown to upregulate β-chemokines independently of degranulation during acute HIV infection [49]. To determine if β-chemokine-producing cells similarly expressed perforin, we assessed expression of MIP-1α by responding cells in a subset of subjects. Inclusion of MIP-1α did not significantly change the overall magnitude of Gag-specific cells detected over time, though it did identify a subset of cells not captured by IFN-γ or CD107a (S7A–S7C Fig). Importantly, the dynamics with which expression of perforin by Gag-specific cells was lost was the same with or without MIP-1α (S7D and S7E Fig). Combined, these data show similarities in the total and Gag-specific CD8+ T cell responses in both differentiation state and cytotoxic potential, suggesting the bulk of activated cells during acute HIV infection could be comprised of HIV-specific CD8+ T cells.
Studies in both murine models and humans have strongly linked the transcription factors T-bet and Eomes to the regulation of effector CD8+ T cell differentiation and function, including the expression of perforin [24–26, 28, 30, 42, 54]. To gain further insight into the evolution of the cytotoxic CD8+ T cell response to HIV we assessed the expression of T-bet and Eomes over the course of infection. For healthy donors, including HIV pre-infection time points, perforin expression was directly associated with T-bet and/or Eomes expression such that the majority of perforin+ cells were either T-bet+Eomes+ or T-bet+Eomes- (Fig 5A and 5B). In contrast, acutely HIV-infected individuals showed marked dissociation between perforin and both T-bet and Eomes resulting in significantly lower proportions of T-bet+Eomes+ and T-bet+Eomes- perforin+ cells (Fig 5A and 5B), and an expansion of perforin+ cells expressing neither T-bet nor Eomes. By the chronic stage these subsets had largely, though incompletely, returned to their normal distributions. When we analyzed T-bet and Eomes expression longitudinally for perforin+ CD8+ T cells within the HIV-infected cohort we found the proportion of T-bet+Eomes+ cells decreased over the first 30 days of infection and T-bet-Eomes- cells increased over the first 60 days before gradually returning to pre-infection levels (Fig 5C and 5D). We have previously shown that the level of T-bet expression within peripheral CD8+ T cells is directly associated with perforin expression, where perforin was found predominantly within T-betHi cells [28]. Consistent with those findings, perforin was most highly associated with a T-betHiEomes+ expression pattern in HIV negative donors and this subset experienced the largest drop during acute HIV (S8 Fig). Despite these shifts in expression patterns that appeared to coincide with the rise and fall plasma viremia, there was no association between the acute frequencies of T-bet or Eomes subsets and acute or set point viral loads (S9A–S9D and S10A–S10D Figs). However, frequencies of T-bet+ and T-bet-Eomes- CD8+ T cells at set point time points were inversely or directly associated with set point viral load, respectively (S9A and S9D Fig).
To determine if the dissociation between perforin, T-bet, and Eomes was unique to HIV, we examined T-bet and Eomes expression within total perforin+ cells following YFV and VV vaccination. While we found almost no dissociation for YFV, there was a transient dissociation following vaccination with vaccinia, although not to the same extent as observed during acute HIV (S11A and S11B Fig). We next examined expression of T-bet and Eomes within HLA-DR+ cells throughout the different vaccine courses. As noted above, during acute HIV infection the vast majority of HLA-DR+ cells are also perforin+ (Fig 2E); thus, it was unsurprising to find that perforin+ and HLA-DR+ cells showed almost identical dynamics in the loss of T-bet and Eomes expression for HIV (S11C Fig). Similarly, for both YFV and VV, activated cells showed a transient increase in the frequency of T-bet-Eomes- cells at day 14 post-vaccination. Together these data suggest that the transient expansion of highly activated bulk effector CD8+ T cells during acute viral infection in humans may not require expression and/or maintenance of T-bet and Eomes.
To determine if the transient loss of T-bet and Eomes within the bulk activated CD8+ T cell memory pool during acute HIV infection extended to HIV-specific CD8+ T cells, we assessed expression of these transcription factors in Gag-specific CD8+ T cells. In marked contrast to the highly activated bulk CD8+ T cell effector population during acute HIV infection, HIV-specific CD8+ T cells expressed T-bet and/or Eomes at the earliest detectable time point and throughout the course of infection (Fig 6A–6C). This indicates that despite their phenotypic similarities total and HIV-specific CD8+ T cells may be primed quite differently during acute infection and raises the possibility that the majority of expanded effector CD8+ T cells in early HIV infection may not be specific for HIV.
We next examined whether loss of perforin expression was related to changes in the level of T-bet expression during early HIV infection. Interestingly, the distribution of T-bet within Gag-specific CD8+ T cells changed over time from acute to chronic infection (Fig 6D). In the acute phase, responding cells were equally distributed between T-betHiEomes+ and T-betLoEomes+ expression patterns, which during the chronic phase began to be dominated by T-betLoEomes+ cells (Fig 6D). Furthermore, T-betHiEomes+ HIV-specific CD8+ T cells continued to express perforin as infection progressed, whereas T-betLoEomes+ cells gradually lost perforin expression over time (Fig 6E and 6F).
Finally, in contrast to the recent findings by Ndholuvu, et al. [48], we did not find the magnitude, proportion perforin+, or any T-bet- or Eomes-expressing subset of responding HIV-1 Gag-specific CD8+ T cells to be predictive of peak or set point viral load (S12 and S13 Figs). Despite this, our data suggest that in the earliest phase of infection, HIV-specific CD8+ T cells have both the transcriptional and functional properties associated with long-term control of HIV replication [16, 28], and that the inability to durably maintain high-level T-bet expression contributes to a qualitatively inferior response as infection progresses.
Mechanisms underlying the inability of CD8+ T cells to fully control HIV replication have remained unclear. Failure of antiviral immunity has been attributed in part to qualitative defects in total and HIV-specific CD8+ T cells [15, 16, 20, 55, 56]. However, the dysfunction observed within the CD8+ T cell pool has largely been defined in the context of chronic infection when the success or failure of the presumed response has already been determined. The question of whether CD8+ T cells in progressive infection were intrinsically less functional from the outset or if dysfunction arose over time has remained unanswered. To address this issue, we assessed the longitudinal CD8+ T cell responses of a diverse cohort of individuals experiencing acute/early HIV infection. We show that acute HIV infection elicits a robust cytotoxic CD8+ T cell response characterized by cells that express the cytolytic effector molecule perforin and the effector-associated transcription factors T-bet and Eomes. Importantly, the quality of the response quickly waned following the resolution of acute viremia, with a significant decrease in perforin expression by HIV-specific CD8+ T cells that was at least partially accounted for by a shift from T-betHiEomes+ cells to T-betLoEomes+ cells. The attenuation of the cytolytic response may help explain the failure of CD8+ T cells to control HIV replication in the long-term.
It is well documented that CD8+ T cell responses are elicited early in HIV infection and are associated with control of viral replication [1, 2, 48, 57]. Some of the strongest evidence of the CD8+ T cell-mediated immunologic pressure exerted during this period is the rapid emergence of viral escape mutations within known CD8+ T cell epitopes [4, 6, 9]. We found that HIV-specific cells had high cytotoxic potential at the earliest time points following HIV infection, but rapidly lost this function as disease progressed. This suggests a mechanism through which CD8+ T cells may exert a strong direct selective pressure on the virus resulting in the rapid selection of escape variants early in infection that ultimately have a reduced capacity to stimulate cytolytic CD8+ T cell responses [6, 9, 58, 59]. It should be noted that whereas perforin expression was lost over time almost all HIV-specific responding cells continued to produce MIP-1α. Thus, while cytotoxic CD8+ T cells play an important role in the resolution of acute viremia, as they lose their ability to express perforin they may be able to keep the virus partially in check through a combination of the remaining cytotoxic response and non-cytotoxic inhibitory effects exerted via the continued expression of β-chemokines or other non-cytolytic mechanisms [60]. This would be consistent with models suggesting CD8+ T cell cytotoxic mechanisms do not account for the entirety of CD8+ T cell-mediated viral suppression during chronic progressive SIV infection [61, 62]. It remains unclear if maintenance of perforin expression following acute infection would further enhance the level of control over viral replication CD8+ T cells provide as we would predict it should based on studies of CD8+ T cell responses in the chronic phase of infection [16, 18, 19]. Unfortunately, we were unable to find any direct associations between HIV-1 Gag-specific perforin, T-bet, or Eomes expression and the level of plasma viremia or CD4+ T cell numbers.
T-bet and Eomes are important regulators of effector CD8+ T cell differentiation and function for both mice and humans [24–26, 28, 30, 31, 33, 34, 42, 54]. Expression patterns of these transcription factors have been described for CD8+ T cells in the context of various human viral infections, including CMV, EBV, HBV, HCV, HIV, and TBEV [28, 32, 34, 42, 54, 63–66]. These studies demonstrated a high degree of variability in the relative levels of T-bet and Eomes expressed by virus-specific CD8+ T cells depending on time from infection, whether the infection was controlled, and tissue localization. CMV-specific cells express T-bet and Eomes during both acute and chronic phases of infection, but control of viral replication in the acute phase is associated with a higher ratio of T-bet+ versus Eomes+ cells [64, 66]. EBV- and TBEV-specific cells also express T-bet and Eomes during the earliest phase of their respective infections, but EBV-specific cells lose expression of both during convalescence whereas TBEV-specific cells retain T-bet expression and show a gradual reduction in Eomes [42, 63]. HCV-specific cells are T-bet+ in acute/resolving HCV infection and T-bet-Eomes- during acute/non-resolving infection. Post-acute phase, HCV-specific cells in the peripheral blood are T-bet-Eomes- for both resolved and non-resolved HCV infection, but T-bet+ within the livers of subjects with resolved infection and Eomes+ in livers of chronically infected subjects [34, 65]. Together, these results suggest expression of T-bet during the acute phase is a critical determinant of viral infection outcome. The differential outcomes associated with Eomes were also reflective of the relative expression level of T-bet, suggesting Eomes may not be as important for the resolution of acute viremia. Rather, Eomes expression may determine whether antigen-specific cells are fated to form a stable memory pool or become exhausted subsequent to the acute phase, dependent on whether or not the infection is ultimately cleared [34, 67].
Similar associations between T-bet, Eomes, and outcome have been demonstrated in chronic HIV infection. In this context, a high level of T-bet expression was associated with greater overall functionality of HIV-specific CD8+ T cells, including cytotoxic potential, and relative control of viral replication, whereas low T-bet levels and continued Eomes expression has been associated with lower overall functionality and persistent viremia [28, 32]. Our data show that HIV-specific cells have high cytotoxic potential during acute infection, but lose the ability to express or rapidly upregulate perforin in chronic infection. This loss of cytotoxic potential over time can at least partially be explained by a change in the relative expression levels of T-bet and Eomes: HIV-specific cells were equally T-betHiEomes+ and T-betLoEomes+ during acute infection and both subsets efficiently upregulated perforin initially but the proportion of T-betLoEomes+ cells increased significantly as infection progressed and cells with this phenotype had an inferior capacity to express perforin compared to T-betHiEomes+ cells. The expression of perforin by either phenotype during acute infection may be reflective of the high degree of inflammation and activation during this phase, a differential role for Eomes at different stages of infection, and/or the result of additional transcription factors not assessed here. Whatever the case may be, T-betHiEomes+ HIV-specific CD8+ T cells retain the ability to upregulate perforin following resolution of acute viremia and this subset declines during chronic progressive infection.
Recent data from Ndhlovu et al. suggests HIV infection elicits a massive antigen-specific CD8+ T cell response with limited bystander activation [48]. Similar observations have been reported after vaccination with vaccinia and yellow fever virus [46]. The similarities in differentiation state, activation, and immediate cytotoxic potential between total peripheral memory and Gag-specific cells reported here support the idea of a robust and specific response to HIV infection. However, we found a significant discrepancy between transcriptional control of HIV-specific CD8+ T cells versus the bulk activated perforin+ memory CD8+ T cell population. The degree to which these differences reflect a true lack of specificity, dysfunction on the part of the bulk activated cells, an inability to identify an appropriate functional marker, or an attempt by the host to mitigate immune-mediated pathology remains unclear. It is likely there area many more circulating HIV-specific CD8+ T cells than indicated by our findings using in vitro stimulation with only two HIV-1 proteins and a limited number of functional parameters to identify responding cells. However, it should be noted that CD8+ T cell bystander activation has been reported during acute HIV and EBV infection in humans and it is possible at least a subset of CD8+ T cells are activated non-specifically in our cohort [44, 68]. T cell receptor stimulation is required for upregulation of T-bet [69], but a large proportion of bulk activated perforin+ cells during acute HIV infection appear to express neither T-bet nor Eomes whereas all Gag-specific cells expressed one or the other. In addition, perforin can be upregulated in the absence of direct antigenic stimulation via exposure to IFN-α [70], levels of which are highly elevated during acute HIV infection [71]. Thus, the difference in T-bet and Eomes expression we observed between bulk perforin+ and responding HIV-specific CD8+ T cells raises the possibility that a significant number of bystander-activated cells are being induced in response to HIV infection. Alternatively, given the association between activation and the size of the T-bet-Eomes- pool across infections with vaccinia, yellow fever, and HIV, the absence of T-bet and Eomes expression in the bulk perforin+ CD8+ T cell pool may be a characteristic of the contraction phase that typically follows the initial CD8+ T cell response. This would be consistent with the pro-apoptotic phenotype of the majority of cells following peak HIV viremia and the timing of our samples [48]. Whether HIV-specific or bystander, the lack of T-bet and Eomes expression by these cells suggests they would be unable to sustain perforin expression upon encountering infected target cells. This may in part explain the inability of bulk peripheral CD8+ T cells from acutely HIV infected individuals to efficiently inhibit viral replication in vitro and further suggests they would not make a meaningful contribution to long-term control of viral replication in vivo [72, 73].
These data show how the peripheral CD8+ T cell response to HIV evolves over the course of progressive infection. HIV-specific CD8+ T cells are able to upregulate perforin and T-bet initially but begin to lose this capacity soon after peak viremia, demonstrating for the first time that there is not an initial intrinsic inability of HIV-specific CD8+ T cells to upregulate these molecules. It remains unclear how or if these responses differ from those of CD8+ T cells from subjects who go on to spontaneously control viral replication to very low levels in the chronic phase. While we did find frequencies of T-bet+ and T-bet-Eomes- total memory CD8+ T cells at set point time points were inversely or directly associated with set point viral load, respectively, we did not find any associations between viral load and the size of the total peripheral perforin+ pool or the magnitude or cytotoxic potential of HIV-1 Gag-specific cells at any time point. Nor did we find any subset of total memory or Gag-specific cells to be predictive of set point viral load for this group of subjects, possibly due to the limited number of very early time points and relatively narrow range of viral loads at set point. However, the fact that the initial phenotype of HIV-specific cells is similar to that associated with control during the chronic phase of infection suggests induction and maintenance of cells capable of upregulating high levels of T-bet and perforin could lead to subsequent control. Eliciting HIV-specific cells with these characteristics might serve as an important target for vaccination or therapeutic modalities seeking to fully control early viral replication or eradicate the chronic viral reservoir.
Blood specimens were acquired with the written informed consent of all study participants and with the approval of the institutional review board at each respective institution where patient materials were collected: University of Pennsylvania (IRB# 809316), McGill University Health Centre (REB# GEN-10-084), Human Subjects Protection Branch (RV217/WRAIR#1373), Kenya Medical Research Council (KEMRI/RES/7/3/1), The United Republic of Tanzania Ministry of Health and Social Welfare (MRH/R.10/18/VOLL.VI/85), Tanzanian National Institute for Medical Research (NIMR/HQ/R.8aVol.1/2013), Royal Thai Army Medical Department (IRBRTA 1810/2558), Uganda National Council for Science and Technology–National HIV/AIDS Research Committee (ARC 084), Uganda National Council of Science and Technology (HS 688), East London and City and the Southwest and Southwest Hampshire Ethics Review Committees, Duke University (IRB# Pro00006579 and IRB# Pro00007558), Emory University (IRB# 00009560), Oregon Health and Science University (IRB# 2470 and IRB# 2832). The study was conducted in accordance with the principles expressed in the Declaration of Helsinki.
Eleven HIV-1 acutely infected participants were enrolled as part of the RV217 Early Capture HIV cohort, nine were enrolled in the CHAVI 001 acute infection cohort, and twelve were enrolled in the Montreal Primary Infection cohort. Participant demographics are summarized in S1 Table. Acute HIV-1 infection was determined by measuring plasma HIV RNA content and HIV-specific antibodies using ELISA and Western blot. Fiebig staging [74] immediately following the first positive visit or at the screening visit was used to characterize the timing of infection for RV217 and CHAVI participants, respectively. The only exception was RV217 donor 40067 for which the estimated date of infection was taken as the midpoint between the last negative and first positive visit. For the Montreal Primary Infection cohort the following guidelines proposed by the Acute HIV Infection Early Disease Research Program sponsored by the National Institutes of Health were used to estimate the date of infection: the date of a positive HIV RNA test or p24 antigen assay available on the same day as a negative HIV enzyme immunoassay (EIA) test minus 14 days; or the date of the first intermediate Western blot minus 35 days. In addition, information obtained from questionnaires addressing the timing of high-risk behavior for HIV transmission was taken into account in assigning a date of infection when consistent with biological tests. The timing of visits relative to estimated date of infection for all acutely HIV infected donors used in this study is provided in Fig 1A. Study participants were antiretroviral therapy naïve at all time points analyzed, consistent with the standard of care at the time of study. HIV-1 viral loads were measured using the Abbot Real-Time HIV-1 assay (RV217; Abbot Laboratories, Abbott Park, IL), COBAS AMPLICOR HIV-1 monitor test, version 1.5 (CHAVI; Roche Diagnostics, Branchburg, NJ), or the UltraDirect Monitor assay (Montreal; Roche Diagnostics, Branchburg, NJ). HIV set point viral loads were defined as the average of all viral load measurements between 90 and 365 days post-infection in the absence of therapy with the requirement for at least two viral load measurements during this period.
For HIV-negative cohorts, volunteers were administered the live-attenuated YFV-17D vaccine (YF-Vax, Sanofi Pasteur), the live vaccinia smallpox vaccine (Dryvax, Wyeth Laboratories), or challenged with influenza A/Brisbane/59/07. YF-Vax was administered subcutaneously in the arm, Dryvax was administered by scarification of the upper arm with three pricks of a bifurcated needle, and influenza A virus was administered intra-nasally. Peripheral blood mononuclear cells (PBMCs) from pre-vaccination or pre-infection time points were available for most donors along with several time points post-vaccination or infection (S3A–S3C Fig). Pre-infection time points from all cohorts, including RV217 participants, along with PBMCs obtained from fifteen healthy human subjects through the University of Pennsylvania’s Human Immunology Core were combined for a total of 41 healthy donor data points.
Potential T cell epitope (PTE) peptides corresponding to the HIV-1 Gag and Nef proteins were obtained from the NIH AIDS Reagent Program (NIH, Bethesda, Maryland, USA). PTE peptides are 15 amino acids in length and contain naturally occurring 9 amino acid sequences that are potential T cell determinants embedded in the sequences of circulating HIV-1 strains worldwide, including subtypes A, B, C, D and circulating recombinant forms (CRF). As such, these peptide pools provided the coverage necessary for the T cell stimulation assays performed in this study given the broad geographical distribution of our study participants and diversity of infecting viruses (S1 Table). Lyophilized peptides were dissolved in dimethyl sulfoxide (DMSO, Sigma-Aldrich, St Louis/Missouri, USA), combined into two pools at 400 μg/ml, and stored at -20°C.
Cryopreserved PBMCs were thawed and rested overnight at 2x106 cells/ml in RPMI medium supplemented with 10% fetal bovine serum, 2 mM L-glutamine, 100 U/ml penicillin, and 100 mg/ml streptomycin. Cell viability was checked both immediately after thawing and after overnight rest by trypan blue exclusion. Costimulatory antibodies (anti-CD28 and anti-CD49d, 1 μg/mL each; BD Biosciences) and pre-titrated fluorophore conjugated anti-CD107a was included at the start of all stimulations. PBMCs were incubated for 1 hour at 37°C and 5% CO2 prior to the addition of monensin (1 μg/mL; BD Biosciences) and brefeldin A (10 μg/mL; Sigma-Aldrich) followed by an additional 5 hour incubation at 37°C and 5% CO2. For peptide stimulations, peptides from the two Gag PTE pools were added to a single tube of cells such that each individual peptide was at a final concentration of 1 μg/ml. As a negative control, DMSO was added to the cells at an equivalent concentration to the one used for peptide stimulation.
Antibodies for surface staining included CCR7 APC-Cy7 (clone G043H7; Biolegend), CCR7 APC-eFluor780 (clone 3D12; eBioscience), CD4 PE-Cy5.5 (clone S3.5; Invitrogen), CD8 BV711 (clone RPA-T8; Biolegend), CD8 Qdot 605 (clone 3B5; Invitrogen), CD14 BV510 (clone M5E2; Biolegend), CD14 Pacific Blue (clone M5E2; custom), CD14 PE-Cy5 (clone 61D3; Abcam), CD14 PE-Cy7 (clone HCD14; Biolegend), CD16 Pacific Blue (clone 3G8; custom), CD16 PE-Cy5 (clone 3G8; Biolegend), CD16 PE-Cy7 (clone 3G8; Biolegend), CD19 BV510 (clone HIB19; Biolegend), CD19 Pacific Blue (clone HIB19; custom), CD19 PE-Cy5 (clone HIB19; Biolegend), CD19 PE-Cy7 (clone HIB19; Invitrogen), CD45RO ECD (clone UCHL1; Beckman Coulter), CD45RO PE-CF594 (clone UCHL1; BD Biosciences), CD107a PE-Cy5 (clone eBioH4A3; eBioscience), CD107a PE-Cy7 (clone H4A3; Biolegend), and HLA-DR Pacific Blue (clone LN3; Invitrogen). Antibodies for intracellular staining included CD3 BV570 (clone UCHT1; Biolegend), CD3 BV650 (clone OKT3; Biolegend), CD3 Qdot 585 (clone OKT3; custom), CD3 Qdot 650 (clone S4.1; Invitrogen), Eomes Alexa 647 (WD1928; eBioscience), Eomes eFluor 660 (WD1928; eBioscience), IFN-γ Alexa 700 (clone B27; Invitrogen), Perforin BV421 (clone B-D48, Biolegend), Perforin Pacific Blue (clone B-D48; custom), Perforin PE (clone B-D48, Cell Sciences), T-bet FITC (clone 4B10; Biolegend), and T-bet PE (clone 4B10; eBioscience).
At the end of the stimulations, cells were washed once with PBS prior to be being stained for CCR7 expression for 15 min at 37°C in the dark. Cells were then stained for viability with aqua amine-reactive viability dye (Invitrogen) for 10 min at room temperature in the dark followed by addition of a cocktail of antibodies to stain for surface markers for an additional 20 min. The cells were washed with PBS containing 0.1% sodium azide and 1% BSA, fixed and permeabilized using a Cytofix/Cytoperm kit (BD Biosciences), and stained with a cocktail of antibodies against intracellular markers for 1 h at room temperature in the dark. The cells were washed once with Perm Wash buffer (BD Biosciences) and fixed with PBS containing 1% paraformaldehyde. Fixed cells were stored at 4°C in the dark until acquisition. Antibody capture beads (BD Biosciences) were used to prepare individual compensation controls for each antibody used in the experiment. ArC Amine Reactive beads (ThermoFisher Scientific) were used to generate a singly stained compensation control for the aqua amine-reactive viability dye.
For each stimulation condition, a minimum of 250,000 total events were acquired using a modified LSRII (BD Immunocytometry Systems). Data analysis was performed using FlowJo (TreeStar) software. Gating strategy is provided in the supplementary materials (S1 Fig). Reported antigen-specific data have been corrected for background based on the negative (no peptide) control, and only responses with a total frequency twice the negative control and above 0.01% of total memory CD8+ T cells (after background subtraction) were considered to be positive responses. By analyzing the data in this way, we examined cytolytic protein production resulting from antigen-specific stimulation and ensured that its expression was considered only within responding CD8+ T cells expressing at least one other functional parameter. Whereas IFN-γ, CD107a, and MIP-1α were used to identify antigen-specific CD8+ T cells for some donors, only IFN-γ and CD107a were used consistently for all donors and figures depicting antigen-specific data were derived from analysis of cells expressing these two markers unless otherwise noted.
All statistical analysis was performed using Stata (version 14.0). Graphs were generated using Stata or GraphPad Prism (version 5.0a). Generalized estimating equations (GEEs) with robust variances were used to test for changes while adjusting for repeated measurements on the same individuals [75]. In instances where many values were at 100% a random-effects tobit regression model was used to do a combined analysis of the percent of data points at 100% versus differences in values for data points below 100%. P values were Holm-adjusted for multiple comparisons. Bars represent approximations of the means generated by the models. Lowess smoothers were used to represent the mean over time for longitudinal data. Correlations were determined using Spearman’s rank correlation test (non-parametric; two-tailed).
|
10.1371/journal.pgen.1007424 | New insights into donor directionality of mating-type switching in Schizosaccharomyces pombe | Mating-type switching in Schizosaccharomyces pombe entails programmed gene conversion events regulated by DNA replication, heterochromatin, and the HP1-like chromodomain protein Swi6. The whole mechanism remains to be fully understood. Using a gene deletion library, we screened ~ 3400 mutants for defects in the donor selection step where a heterochromatic locus, mat2-P or mat3-M, is chosen to convert the expressed mat1 locus. By measuring the biases in mat1 content that result from faulty directionality, we identified in total 20 factors required for donor selection. Unexpectedly, these included the histone H3 lysine 4 (H3K4) methyltransferase complex subunits Set1, Swd1, Swd2, Swd3, Spf1 and Ash2, the BRE1-like ubiquitin ligase Brl2 and the Elongator complex subunit Elp6. The mutant defects were investigated in strains with reversed donor loci (mat2-M mat3-P) or when the SRE2 and SRE3 recombination enhancers, adjacent to the donors, were deleted or transposed. Mutants in Set1C, Brl2 or Elp6 altered balanced donor usage away from mat2 and the SRE2 enhancer, towards mat3 and the SRE3 enhancer. The defects in these mutants were qualitatively similar to heterochromatin mutants lacking Swi6, the NAD+-dependent histone deacetylase Sir2, or the Clr4, Raf1 or Rik1 subunits of the histone H3 lysine 9 (H3K9) methyltransferase complex, albeit not as extreme. Other mutants showed clonal biases in switching. This was the case for mutants in the NAD+-independent deacetylase complex subunits Clr1, Clr2 and Clr3, the casein kinase CK2 subunit Ckb1, the ubiquitin ligase component Pof3, and the CENP-B homologue Cbp1, as well as for double mutants lacking Swi6 and Brl2, Pof3, or Cbp1. Thus, we propose that Set1C cooperates with Swi6 and heterochromatin to direct donor choice to mat2-P in M cells, perhaps by inhibiting the SRE3 recombination enhancer, and that in the absence of Swi6 other factors are still capable of imposing biases to donor choice.
| Effects of chromatin structure on recombination can be studied in the fission yeast S. pombe where two heterochromatic loci, mat2 and mat3, are chosen in a cell-type specific manner to convert the expressed mat1 locus and switch the yeast mating-type. The system has previously revealed the determining role of heterochromatin, histone H3K9 methylation and HP1 family protein Swi6, in donor selection. Here, we find that other chromatin modifiers and protein complexes, including components of the histone H3K4 methyltransferase complex Set1C, the histone H2B ubiquitin ligase HULC and Elongator, also participate in donor selection. Our findings open up new research paths to study mating-type switching in fission yeast and the roles of these complexes in recombination.
| The fission yeast S. pombe exists as two haploid cell types, plus (P) and minus (M), that differ at the mat1 locus. When starved for nitrogen, haploid cells undergo sexual differentiation, mate with the opposite cell type and sporulate. These events are driven by master regulators expressed from the mat1-P and mat1-M alleles [1]. The regulators first drive sexual differentiation and mating and, when co-expressed in the zygote, meiosis and sporulation. Homothallic (h90) colonies sporulate very efficiently because they contain equal proportions of P and M cells due to frequent gene conversions at mat1. The genetic information at mat1 is replaced with genetic information copied from one of two silent loci, mat2-P or mat3-M [2]. The organization of the ~35 kb region of chromosome 2 that comprises mat1, mat2-P and mat3-M in h90 strains is depicted in Fig 1A. All three loci are flanked by short regions of sequence identity, the centromere-distal H1 box and the centromere-proximal H2 box. In addition, H3 homology boxes immediately adjacent to H2 are found exclusively at mat2-P and mat3-M. An alternative arrangement, known as h09, has mat2-M and mat3-P cassettes [3].
Mating-type switching follows the so-called ‘Miyata’s rules’ inferred from pedigree analyses of dividing cells [4]. A single h90 cell produces both ‘unswitchable’ and ‘switchable’ cells. According to the ‘one-in-four-rule’, illustrated in Fig 1B, one of two sister cells originating from an ‘unswitchable’ cell becomes ‘switchable’. That cell produces one switched daughter upon cell division and one unswitched, but switchable, daughter. Thus, only one cell out of four cousins displays a switched mating-type (the ‘one-in-four’ rule [4]) and lineages of ‘switchable’ cells are observed (the ‘recurrent switching’ rule [5]). The known mechanisms of mating-type switching provide an explanation for these rules.
Mating-type switching is initiated by an imprint at mat1 introduced during DNA replication [6, 7]. Replication stalls within mat1 at the MPS1 site and a nick or two ribonucleotides are incorporated into the lagging strand at the imprinting site situated nearby, at the junction of the mat1 cell-type specific information and H1 box [6–15]. This imprint creates a ‘switchable’ cell. During the next round of DNA synthesis, the imprint is converted into a double-strand break (DSB) that triggers homologous recombination and mating-type switching [10, 14]. At least seven factors (Swi1, Swi3, Pol1 (Swi7), Sap1, Lsd1, Lsd2 and Mrc1) are required for efficient DSB formation. The Swi1-Swi3 complex and Mrc1 are necessary for imprinting by pausing replication forks at MPS1 [7, 16]. The DNA primase Pol1 and a DNA element targeted by the essential DNA-binding protein Sap1 are not required for replication fork stalling at MPS1; hence Pol1 and Sap1 are believed to catalyze imprint formation downstream of the Swi1-Swi3 complex [7], while Lsd1 and Lsd2 are required upstream of Swi1-Swi3 [17, 18]. In addition, Swi1, Swi3 and Mrc1 block replication from the mat1 distal side at the replication termination site RTS1 to optimize mating-type switching [7, 16]. Thus, replication proceeds unidirectionally when leading-strand synthesis reaches the imprint and the nick is converted into a one-ended DSB.
The DSB end initiates repair by recombining with one of the heterochromatic and transcriptionally silent cassettes mat2-P and mat3-M [10, 19, 20]. The free DNA end can invade either mat2-P or mat3-M, however mating-type information opposite to the information present at mat1 is chosen with a ~90% probability, leading to Miyata’s observation that switchable cells nearly always switch to the opposite mating-type [4, 9]. Surprisingly, this strongly biased donor selection relies on the heterochromatic state of the mat2-mat3 region [3, 21–24]. In the wild-type, histone H3K9 methylation deposited between the inverted repeat boundaries IR-L and IR-R permits the binding of the key switching factor Swi6, an HP1 homolog [25]. Defective donor choice in swi6 mutants biases h90 cell populations towards the M mating-type due to preferred use of a mat3-M adjacent recombination enhancer over a mat2-P adjacent enhancer in the absence of Swi6 [22, 23, 26]. Also essential at this step of switching is the Swi2-Swi5 complex, capable of interacting with Swi6, and whose molecular role is inferred from the related Sfr1-Swi5 complex [27, 28]. Sfr1 shares sequence homology with the C-terminus of Swi2, in a domain that permits the interaction of either Swi2 or Sfr1 with the recombination mediator Swi5 and with the strand-exchange factor Rad51. The Sfr1-Swi5 complex stabilizes Rad51 filaments in vivo and promotes Rad51-mediated strand exchange in vitro [28–30]. Consistent with similar functions for the Swi2-Swi5 complex, Swi2-Swi5 interacts with Rad51 in two-hybrid assays [27]. Thus, mechanistically, the ability of Swi2-Swi5 to interact with Swi6 suggests that the complex participates in donor choice by biasing strand invasion [27]. This idea is supported by cell-type specific associations of Swi2 and Swi5 with the mating-type region [22]. Swi2 localizes to the mat2-P and mat3-M adjacent enhancers SRE2 and SRE3 (Swi2-dependent recombination enhancer element 2 and 3) in M cells, but only to SRE3 in P cells [22, 23]. In swi6 mutant cells, Swi2 localizes only to SRE3, as in P cells [22]. These observations indicate that the heterochromatin-mediated localization of Swi6 regulates Swi2-Swi5 localization to SRE2 to choose mat2-P [22, 23, 31].
After strand invasion into the homologous sequence at the H1 homology box, the donor locus information is copied by polymerase extension until it reaches the H3 homology box, then H3 is believed to form a hairpin loop structure [32]. The mismatch repair Msh2 (Swi8)-Msh3 (Swi4) complex recognizes this conformation and DNA synthesis stops at the donor cassette. The Rad51 mediator Rad55-Rad57 has been suggested to work together with Msh2 (Swi8), because mutants in these factors tend to form h+N rearrangements containing a duplication of the entire mat2-3 region at mat1 [33–36]. In addition to Rad55-Rad57, the homologous recombination factor Rad52 is required for DSB repair at mat1 [37, 38]. Presumably, Rad55-Rad57 and Rad52 are involved in annealing between two H2 boxes in mat1 and the donor cassette. The endonuclease Rad16 (Swi9)-Swi10 and its activator Pxd1 cleave the intermediate between the H2 and H3 boxes [37, 39]. This is followed by new DNA synthesis from H2 of mat1 to H1 to complete replication of this region and to thus switch mating-type.
This switching system has been utilized to study multiple aspects of replication, histone modification and recombination. Historically, the mating-type switching related genes were classified functionally by Southern blotting analysis of the effect of mutants on mat1 switching [40]. Class Ia genes (swi1, 3, and 7) are required for the imprinting step that leads to the DSB formation, hence mutants in these genes show no DSB in a Southern blot. Class Ib genes (swi2, 5 and 6) are not necessary for DSB formation, but required for efficient switching. The third group, Class II, (swi4, 8, 9, and 10) resolves the recombination intermediate, mutants in these genes contain frequent rearrangements of the mating-type region, in particular the h+N duplication. Subsequent screens identified additional factors (Table 1) and suggested that yet more might exist. In particular, aspects of imprint formation and donor choice are still not understood.
To identify yet unknown regulators of mating-type switching, we combined the deletion of 3420 nonessential genes (Bioneer gene deletion library version 5) with an h90 strain background. The strain also included a dual reporter system with CFP under the control of a P-specific promoter and YFP under the control of an M-specific promoter, so that the ratio of P-to-M cells was determined by comparing CFP and YFP fluorescence. As a secondary screening strategy, the genetic content at mat1 was quantified with multiplex PCR using genomic DNA isolated from the candidates that passed the initial screen. These extensive screens identified several new mating-type switching genes whose deletion results in a bias toward M cells within h90 populations instead of the balanced P:M ratio. In addition, analysis of h09 strains revealed that some strains showed clonal biases in independent colonies. As mentioned above, Swi6 is an essential directionality factor [3, 22, 23]. Epistasis analysis with swi6Δ suggests that Clr4, Sir2, Swd1, Clr3 work in the same pathway as Swi6, whereas Brl2, Pof3 and Cbp1 act via Swi6-dependent and -independent mechanisms. These observations provide new clues to understand the molecular mechanisms of mating-type switching.
We conducted a genome-wide screen for factors required for mating-type switching. The screen used an S. pombe gene deletion library (Bioneer) consisting of 3420 haploid strains, each of which lacks a non-essential gene. Mating-type switching occurs in h90 strains, yet the Bioneer library strains are heterothallic h+N strains for which a large duplication in the mating-type region abrogates mating-type switching. To construct h90 derivatives of the entire Bioneer collection, strain PG4045 was mated to the library. The h90 mating-type region of PG4045 could be selected in the progeny due to the linked LEU2 gene. In addition, PG4045 contains two fluorescent reporters specific for the P (CFP controlled by the map2 promoter) and M (YFP controlled by the mfm3 promoter) cell types, respectively. Both reporters are expressed from the leu1 locus where they were integrated together with the selectable ura4+ gene. Thus, we selected h90 LEU2 ura4+ segregants (S1 Fig). We obtained 3298 h90 deletion strains in which we monitored expression of the fluorescent reporters (Fig 2A). Efficient mating-type switching results in rapid homogenization of h90 cell populations to equal proportions of P and M cells (Fig 1B). Here, screening specifically for mutants that displayed biased cell-type ratios, differing by > 3 standard deviations from the mean, we isolated 105 candidates with skewed proportions (Fig 2B, S2 Fig, S2 Table). In several deletion strains, we detected co-expression of CFP and YFP in a cell. This phenotype was most likely caused by derepression of the mating-type information at mat2-P and mat3-M in these mutants [42]. In addition, 568 strains that could not be evaluated due to low fluorescence intensity or poor growth were examined by iodine staining, a stain for S. pombe spores that can be used as diagnostic for mat1 switching (Fig 2C, S3 Table). Wild-type h90 colonies are stained darkly by iodine vapors because of their high spore content while mutants with altered mating or sporulation are stained less. Here, 124 deletion strains among the strains tested showed a staining different from wild-type. They were analyzed by quantitative multiplex PCR for mat1 content alongside the 108 candidates that had passed the fluorescence microscopy screening.
Analyzing the content of mat1 permitted us to pinpoint mutants for which biased cell-type expression or poor sporulation is likely to result from switching defects. A number of candidates failed to show a biased mat1 content by multiplex PCR according to the chosen thresholds of P band intensity (Fig 2A and 2D). Indeed, many mutations might result in biased reporter gene expression or altered sporulation without affecting mating-type switching, for example genes located in the L region of the mat locus were eliminated from this screening. In addition, 35 deletion mutants that were diploid and/or heterothallic (h+N) were eliminated from the list of candidates (S4 Table). A bias in mat1 content was detected for 32 mutants, in all cases towards mat1-M, suggesting an increased use of mat3-M. The identity of the deleted gene in these mutants was confirmed using published barcode sequences or gene specific primers. Surprisingly, 9 genes encoding ribosomal proteins are in the list, possibly as a consequence of protein synthesis defects. We did not pursue the investigation of these mutants, but focused instead on the remaining 23 mutants. These included nearly all known non-essential switch-related genes, 16 of which were identified in total. A few switch-related genes (swi1, 5 or 10) were not deleted in the Bioneer library and were therefore not tested by the screen. A few other mutants might have escaped detection due to clonal variation, alternative switching phenotype with low frequency as for msc1Δ [49] or due to a switching bias close to the set thresholds, as for mrc1Δ [16]. Globally though the screening strategy was strongly validated by the identification of known switch-related factors alongside novel factors. The 7 newly identified factors include the F box protein Pof3, the CK2 family regulatory subunit Ckb1, the elongator complex subunit Elp6, the E3 ubiquitin protein ligase Brl2 and three subunits of the Set1/compass complex (Set1C), Swd1, Swd2 and Spf1 (Fig 2D, S2 Fig, Table 1). A protein interaction network analysis regrouping novel and previously known factors showed a high degree of connectivity (see below).
The fluorescence microscopy and multiplex PCR analyses for known and newly identified factors (S2 Fig) were confirmed by Southern blot analysis of mat1 content with the DdeI restriction enzyme (S3 Fig).
Southern blots can be used to detect the imprint at mat1 and to determine whether rearrangements have occurred in the mating-type region. A DSB results from breakage at the fragile imprint site during DNA preparation [40]. As mentioned in the Introduction, the mating-type switching factors can be subdivided into three groups by Southern blot analysis of mutants, reflecting the molecular function of each factor. Class Ia is required for DSB formation at mat1, Class Ib is involved in donor selection for mating-type switching or other steps in the use of the break, and Class II is required for processing the gene conversion intermediates [40]. The 23 strains selected for analysis were assayed by Southern blot (Fig 3). The analysis confirmed previous conclusions in the case of known factors, for instance swi3Δ abolished DSB, and rad57Δ, msh2Δ, msh3Δ, rad16Δ, and pxd1Δ caused high frequencies of rearrangements of the h+N type as expected for resolution-defective mutants [40]. The newly identified mating-type switching genes were assigned to Class Ib; elp6Δ, swd1Δ, spf1Δ, brl2Δ, pof3Δ and ckb1Δ strains were in that category together with swi2Δ, cbp1Δ, swi6Δ, sir2Δ and deletions of the Clr4 methyltransferase complex (CLRC), clr4Δ, rik1Δ and raf1Δ, or Snf/Hdac-containing repressor complex (SHREC), clr1Δ, clr2Δ and clr3Δ, subunits. As previously noticed for the swi6Δ mutant [51] a rearrangement producing an 8.2 kb HindIII fragment could be detected in several Class Ib mutants. The rearrangement could be a mat3:1 circle or a duplication of the mating-type region creating a mat3:1 cassette: mat1-L-mat2-P-K-mat3:1-L-mat2-P-K-mat3-M. The mat3:1 cassette would not be amplified by the primers used to detect mat1 content by PCR. The swd2Δ strain differed from the other mutants by showing a 9.9 kb HindIII fragment hybridizing to the mat1 probe, however reconstruction of the strain produced a Class Ib mutant lacking this additional band and the reconstructed deletion allele was used in further analyses.
In the h09 mating-type region, the contents of the silent cassettes are swapped to mat2-M mat3-P (Fig 4, S4 Fig). This arrangement results in inefficient heterologous switching and in a mat1 content biased towards mat1-M [3]. How mutations affect this bias provides insights into the directionality of mating-type switching. For example, deletion of swi6 biases the mat1 content towards mat1-P in h09 cells, and towards mat1-M in h90 cells, consistent in both cases with preferred selection of mat3 as a donor [3]. This likely reflects a preferential use of the SRE3 recombination enhancer in swi6Δ cells [23]. We created h09 strains to test whether the newly identified Class Ib factors contribute to mating-type switching in the same or similar way as Swi6. Each deletion mutant was crossed with the h09 strain PG4048. After the selection of recombinants, we analyzed four independent colonies of each h09 deletion strain by multiplex PCR for mat1 content. As a control, colonies originating from spores of self-mating PG4048 cells were analyzed. As expected, PG4048 contained a greater proportion of M cells than P cells (a mean of 86% M cells; Fig 4). All mutations tested affected this ratio. The effects varied. During these analyses, we created a fresh deletion of swd2 to eliminate the rearrangement detected in the Bioneer mutant (9.9 kb band in Fig 3), as mentioned above, and confirmed that the observed directionality defects were not a result of the rearrangement.
A first group of factors comprised Swi6, CLRC subunits (Clr4, Rik1 and Raf1) and the histone deacetylase Sir2 (Fig 4). Mutations affecting CLRC or Sir2 produced nearly identical values (around 70% P cells), similar to the loss of Swi6 (77% P cells). These mutants were also very similar to each other in the h90 background (around 20% P cells; Fig 2D). In the mating-type switching process, CLRC and its catalytic Clr4 subunit are believed to work by catalyzing the methylation of H3K9 in the mat2-mat3 heterochromatic domain and creating binding sites for Swi6 [24, 52–55]. The role of Sir2 in the process might be to remove acetyl groups from H3K9 [46], thus facilitating heterochromatin formation [46, 56–58]. Indeed, the methylation of H3K9 and Swi6 association are reduced several fold in the mating-type region of sir2Δ cells [46]. This is consistent with Sir2 acting upstream of CLRC and Swi6 through H3K9 deacetylation, without excluding that other actions of Sir2, e.g H3K4 deacetylation [59], might also be relevant to directionality.
The second group of mutants affecting mating-type switching in h09 cells were deletions of the genes for the ubiquitin E3 ligase Brl2, the Set1C subunits Swd1, Swd2 and Spf1, and the Elongator subunit Elp6 (Fig 4). These mutants resulted in a consistent increase in mat1-P content in h09 cells, from ~20% in wild-type background to ~35% in each of the three mutants. While not as pronounced as for Swi6 or CLRC mutants, the increase occurred to the same degree in all four isolates examined in each case. These mutants were unexpected because it has been reported that set1Δ has no effect on mating-type switching [60]. We examined each component of Set1C (Set1, Swd1, Swd2, Swd3, Spf1, Ash2, Shg1 and Sdc1 [61]) by iodine staining and multiplex PCR of mutants (Fig 5A–5D). Consistent with the previous report [60], deletion of set1 or other subunit genes showed little effect on iodine staining of either h90 or h09 colonies (Fig 5A and 5B). However, monitoring mat1 content clearly showed that individual gene deletions biased switching toward mat3-P in h09 cells and resulted in a correlated increased use of mat3-M in h90 cells for six Set1C subunits (Fig 5C and 5D). This trend is similar to mutations in the H3K9 methylation pathway, although not to the same amplitude (Fig 4). Interestingly, the iodine staining level of set1Δ colonies differed between h90 and h09 in spite of similar cell-type ratios (39% P cells in h90 and 36% in h09). Due to a different switching pattern, P and M cells might be less evenly mixed in h09 colonies compared with h90. This would lead to less efficient mating and spore formation in h09 even though the cell-type bias is only a little more pronounced than in h90. To investigate the functionality of each SRE element in set1Δ cells, we analyzed directionality in SRE element mutants (Fig 5E–5H). Deletion of set1+ did not significantly affect mat1 content in 2×SRE2 cells (where the SRE3 element is replaced with SRE2) (Fig 5E) or in SRE3Δ cells (Fig 5G). However, in populations of 2×SRE3 cells (where the SRE2 element is replaced with SRE3) deletion of set1 caused a small increase in M cells (31% P cells in 2×SRE3 set1Δ compared with 36% P cells in 2×SRE3 set1+) (Fig 5F). In the case of SRE2Δ, donor choice was more strongly biased towards mat3-M in set1Δ cells (6% P cells) than in set1+ cells (13% P cells) (Fig 5H). This suggests that Set1C normally inhibits the choice of mat3-M-SRE3 in M cells. It has similarly been observed that deletion of swi6 causes virtually no change in donor choice in the 2×SRE2 and SRE3Δ backgrounds, where SRE2 keeps being used, but decreases use of SRE3 in SRE2Δ cells (5% P cells in SRE2Δ swi6Δ compared with 16% P cells in SRE2Δ swi6+) [23]. Loss of the Brl2 ubiquitin ligase resulted in phenotypes similar to mutations compromising Set1C (Figs 4 and 5). Thus, like Swi6 and CLRC, Set1C and Brl2 appear to favor use of the SRE2 recombination enhancer over SRE3 when both enhancers are present, perhaps by inhibiting the use of SRE3, to result in balanced switching.
A third group of mutants displayed a variegated phenotype (clr1Δ, clr2Δ, clr3Δ, ckb1Δ, pof3Δ and cbp1Δ; Fig 4), with the proportion of P cells varying between independent cultures. In some isolates, the proportion of P cells was nearly wild-type whereas in others it was similar to the swi6Δ mutant. These phenotypes were not caused by rearrangements in the mating-type region (S4 Fig). Clr1, 2 and 3 are subunits of SHREC, the Snf2-histone deacetylase repressor complex [62]. Clr3 participates in the recruitment of Clr4 to the mating-type region, but in its absence a Clr3-independent, RNAi-dependent pathway accomplishes this function to some extent [63]. Clr3 localizes to three regions in the mating-type region, which are close to mat2-P (REII), cenH and mat3-M (REIII), respectively [62, 64]. In the cenH region, the heterochromatin platform is likely established by RNAi-mechanisms; on the other hand, at the REIII site, it is established by an RNAi-independent mechanism [63, 65, 66]. The clonal variations observed in clr1, clr2 and clr3 h09 mutants might reflect these distinct pathways of heterochromatin establishment similar to position effect variegation [67]. Heterochromatin would be partially formed and inherited in some clonal populations of the mutant strains but not in others, leading to populations differentially proficient for switching. It is also known that the CK2-dependent phosphorylation of Swi6 mediates Clr3 recruitment to centromeric regions [68], possibly accounting for the Ckb1 defects observed here.
To further address the mechanisms of mating-type switching, we analyzed the protein-protein interaction network linking newly identified and previously known mating-type switching factors in the STRING database (Table 1) [69]. The obtained interaction network correlates strongly with categories established by Southern blotting and phenotypic classification (Fig 6A). Among the newly identified factors, six subunits of Set1C and Brl2 are connected to Class Ib factors. While Pof3, Ckb1 and Elp6 show no direct interaction with Class Ib factors in the STRING analysis (Fig 6A), several studies have reported that Pof3 plays a role in heterochromatin silencing [70–72] and Ckb1 phosphorylates Swi6 [68]. Elp6 is an orthologue of a part of the six-subunit Elongator complex (Elp1-6) in S. cerevisiae [73, 74]. Elp3 also passed the initial screening (S2 Table), however the other subunits of Elongator complex did not. The main cellular function of Elongator is thought to be in tRNA modification, but Elongator has also been proposed to acetylate histones [75–77].
We tested the hypothesis that some factors might act through Swi6 by performing an epistasis analysis. Double mutants combining swi6Δ with each candidate gene deletion were constructed and four independent colonies were analyzed by multiplex PCR for each of them (Fig 6B and 6C).
The mating-type switching phenotypes of the swi6Δ clr4Δ and swi6Δ sir2Δ double mutants were quite similar to the single deletions of swi6, clr4 or sir2 (Figs 2D, 4, 6B and 6C). This is consistent with CLRC and Sir2 being required for heterochromatin establishment and Swi6 recruitment at the mat locus (Fig 7).
Double mutants combining swi6Δ with swd1Δ, lacking a Set1C subunit, or clr3Δ, lacking a SHREC subunit, also displayed a phenotype quite similar to the swi6Δ single mutant (Fig 6B and 6C). This indicates that Set1C and SHREC work in the Swi6 pathway. A previous study has shown that set1Δ does not affect Swi6 localization or silencing of a ura4+ marker gene at cenH in the mating-type region [60] but other assays have found that Set1C subunits participate in the repression of heterochromatic loci including the mating-type region [78]. This latter effect might be related to the occurrence of switching defects in Set1C mutants. Set1C may control Swi6 localization in a site-specific manner such as at SRE3 (Fig 5).
Donor preference in swi6Δ ckb1Δ cells indicated that swi6Δ is also epistatic to ckb1Δ even though phenotypes could only be assigned in the h90 background due to apparently high rates of rearrangement in the h09 background. Nevertheless, the data suggest that Swi6 phosphorylation by Ckb1 [68] is important for switching directionality controlled by Swi6.
More complex epistatic relationships were observed for the remaining mutants, brl2Δ, pof3Δ, cbp1Δ, elp6Δ, and swi2Δ, when these mutations were combined with swi6Δ (Fig 6B and 6C). The predominant mating-type had a tendency to vary between isolates, particularly with the h09 mating-type region, and rearrangements occurred. Double mutants combining brl2Δ and swi6Δ showed a mat1-M bias apparently even more pronounced than for the swi6Δ single mutant in h90, while in h09 two swi6Δ brl2Δ isolates were similar to swi6Δ and two had more balanced mat1 contents. Populations of h90 swi6Δ pof3Δ cells showed biases similar to the h90 swi6Δ single mutant for three isolates, whereas the fourth isolate was biased towards P cells rather than M cells. Populations of h09 swi6Δ pof3Δ cells had varied ratios of P and M cells, and two isolates were rearranged. The switching bias for swi6Δ cbp1Δ was similar to swi6Δ with the h90 mating-type region (Fig 6B), but two strains in four differed from swi6Δ with the h09 mating-type region (Fig 6C). This phenotype may be caused by Swi2 expression level, which is controlled by Cbp1 [26, 31]. Rearrangements in h90 swi6Δ elp6Δ and h09 swi6Δ elp6Δ mutants precluded analysis. These observations suggest that both Swi6-dependent and -independent pathways control switching directionality by Brl2, Pof3 and Cbp1. Finally, swi2Δ swi6Δ double mutants differed from swi6Δ in both h90 and h09. It has been reported that Swi2 can localize to SRE3 in the absence of Swi6 [22, 23, 27]. The more balanced cell populations in swi6Δ swi2Δ mutants are probably caused by loss of Swi6-independent function of Swi2.
In summary, our observations suggest that the factors, Clr4, Sir2, Swd1 and Clr3 probably work in the same pathway as Swi6, but Brl2, Pof3, Cbp1 and Swi2 have an effect on donor selection through Swi6-dependent and -independent mechanisms. Frequent DNA rearrangements in Class Ib mutants (Fig 3) and in double mutants with swi6Δ indicate that histone modifications not only direct donor choice, but also facilitate resolution steps or prevent unequal sister chromatid exchanges between cassettes.
We propose a model summarizing how each factor identified in this study might participate in the donor selection mechanism (Fig 7). In this model, the heterochromatin structure in M cells favors the cassette adjacent to SRE2 as a donor while structural changes in mutants and in P cells favor SRE3 [23]. In addition, chromatin structure prevents selection of the cassette adjacent to SRE3 in M cell (Fig 7A). It has been reported that SRE2 can facilitate donor choice efficiently not only in M cells but also in P cells whereas SRE3 is more active in P cells than in M cells [23]. These data indicate that the inhibition of donor choice does not affect SRE2.
Mating-type switching is initiated by a site-specific imprint during replication. In the replisome, Swi1-Swi3, Pol1 and Mrc1 are required for the imprint [16, 40]. One of the novel switch factors identified here is the ubiquitin ligase component Pof3. Pof3 interacts with the replisome, in an Mrc1- and Mcl1-dependent manner [71, 79]. However, rather than participating in imprint formation, we found that Pof3 affects donor selection. Pof3 is also required for heterochromatic silencing near mat3 [70]. We speculate that both effects are brought about by the Pof3-mediated degradation of replisome components [80] or of Ams2, a cell cycle-regulated transcription factor for histone genes [81] that also mediates long range chromosomal interactions [82] and interacts with Raf1, a component of CLRC [83]. Thus, Pof3 would couple the deposition of new histones and their modification by CLRC. During S phase, partly as a result of new histones deposition onto replicated DNA, Swi6 and H3K9me2 levels decrease at silenced loci [84, 85]. A wave of H3K9 acetylation, observed in other organisms in front of the replication fork [86], might further weaken heterochromatin. The current search expands on previous work to show that enzymatic complexes required for the restoration of heterochromatin, both NAD+-dependent and -independent HDACs and CLRC, are necessary for donor selection. Remarkably, lack of Sir2 or of a CLRC subunit phenocopied the swi6 deletion in both h90 and h09 cells (Figs 2D and 4) while the loss of SHREC components resulted in variegated phenotypes. We take these differences as reflecting the different substrate specificities and recruitment mechanisms of the two HDACs to heterochromatic regions [57, 87]. In both cases, our epistasis analysis with the Swi6 mutant points to defects in Swi6 recruitment. Our search also identified Cbp1 and CK2, both of which are thought to recruit Clr3 to heterochromatin regions [64, 68]. Cbp1 also controls expression of the swi2 gene and the cell-type specific protein isoform it produces, together with the M-specific protein Mc [26, 31].
A novel and intriguing outcome of our study is that multiple subunits of Set1C and the E3 ubiquitin ligase Brl2 are required for accurate donor selection. Set1C catalyzes the methylation of H3K4. A role in heterochromatin appears paradoxical, given that methylated H3K4 is strongly associated with expressed genes. One possibility is that Set1C regulates the expression of Swi2, central effector of switching directionality, or of other switching factors. RNA profiling analysis has revealed that switching genes are expressed to similar levels in wild type and individual Set1C mutants [87], however more subtle effects such as shifts in transcription initiation have not been ruled out. In addition, S. pombe Set1C appears directly required for silencing at various locations [78]. In the mating-type region, all subunits except for Shg1 are required for silencing of the cenH repeat. Evidence has also emerged for roles in meiotic recombination [88, 89] even though the effects of H3K4 methylation on recombination have been hard to unravel due to the prevalence of that modification genome-wide [90]. Here, we favor a simple model where Set1C affects donor choice directly. This could be through local, possibly temporally restricted methylation of H3K4 at SRE3 that would either act as such or by preventing H3K4 acetylation. Brl2 is part of HULC that catalyzes the ubiquitylation of histone 2B (H2Bub) [61, 91, 92]. H2Bub stabilizes the interaction of Set1 with chromatin in vitro [93]. The single brl2Δ deletion affected mat1 content similar to the deletion of Set1C components (Figs 2, 4 and 5). These connections and phenotypic similarities indicated that Set1C and HULC might co-operate to choose a correct donor. (Figs 4 and 6B). However, the phenotypes of two in four independent colonies of h09 swi6Δ brl2Δ strain differed from h09 swi6Δ swd1Δ (Fig 6C). Brl2 is also known to interact with Nse5, which is a part of the structural maintenance of chromosome 5/6 (Smc5-6) holocomplex [94, 95]. It may have multiple functions in mating-type switching.
Following the cell-type specific deposition of Swi6, Swi2-Swi5 localizes to SRE2 by interaction with Swi6 in M cells [22, 23, 27]. Together with the inhibited use of SRE3, this regulated recruitment of Swi2-Swi5 effectively directs the Rad51 strand-exchange protein to initiate homologous recombination at the proper cassette, providing an increasingly well understood model for the effects of chromatin structure on recombination.
S. pombe strains were generated and propagated according to standard protocols [96]. They were manipulated with a Singer RoToR plate-handling robot (Singer Instruments) for high throughput screens. To test for mating-type switching defects by fluorescence analysis, a query strain (PG4045: h90 (Blp1)::LEU2 leu1::ura4+-[mfm3p-YFP]-[map2p-CFP] ura4-D18 ade6-M216 was mated to the Bioneer gene deletion library (h+ leu1-32 ura4-D18 ade6-210 or 216, ORFΔ::kanMX4). Mating was performed on SPA plates supplemented with 200 mg/l leucine, 100 mg/l uracil, and 100 mg/l adenine. Cells were allowed to mate and sporulate at 30°C for two days. The mating plates were then moved to 42°C for three days to eliminate vegetative cells. Following heat treatment, spores were transferred onto YES plates with 100 mg/l G418 and allowed to germinate and divide for three days at 30°C. To select for h90 progeny, cells were then transferred from the YES plates to MSA plates with 100 mg/l G418 and 100 mg/l adenine and grown for a further three days at 30°C. This scheme selects for (Blp1)::LEU2, tightly linked to the h90 region, for ura4+, tightly linked to the fluorescent reporters, and for kanMX4 that marks each ORF deletion. Single colonies were isolated from bulk recombinants by streaking cells onto MSA plates containing G418 and adenine, incubated for three days at 30°C. The sdc1 deletion strain (h+ sdc1::Kanr ade6-M21? leu1-32 ura4-D18) was obtained from the National BioResource Project (NBRP ID: FY23769, strain name: P1-1G).
Single colonies of recombinants obtained as described above were inoculated into 50 μL MSA medium supplemented with 100 mg/l adenine and grown for two days at 30°C in 96 well plates. Cells were then diluted 24 times into MSA medium with adenine and grown for ~20 hr at 30°C. Sixty μl cell suspensions diluted 15 times with MSA medium with adenine were transferred to 384 well microplates with clear bottom (CellCarrier-384 ultra, Perkin Elmer). The fluorescence of cells was measured using an Opera high-content screening microscope (Perkin Elmer). The following settings were used [Filter sets: Camera 1: 475/50 for CFP signals, Camera 2: 540/75 for YFP signals, Camera 3: 690/50 for bright field, Light source: 405/488/635]. Twelve images were taken for each well, for a total of ~200 cells per strain. The proportion of P cells was then calculated using the Acapella software program (PerkinElmer). Selected strains were imaged again using a Delta Vision Elite microscope (GE Healthcare).
Cells were plated onto MSA medium supplemented with 100 mg/l adenine, allowed to form single colonies, and exposed to iodine vapors.
S. pombe cells were propagated in 2 mL liquid YES cultures at 30°C to saturation. Genomic DNA was prepared from wild-type and mutant cells as described [97]. The genomic DNA concentration was measured using QuantiFluor One dsDNA Dye System (Promega) and 4 μL genomic DNA (1.25–5.0 ng/μL) in TE was added to 16 μL PCR reaction reagent (total 20 μL) to perform multiplex PCR to determine the genetic content of the mat1 or mat3 locus. The primers used were FAM-MT1 (5’-AAATAGTGGGTTAGCCGTGAAAGG-3’) at 400 nM, MP1 (5’-ATCTATCAGGAGATTGGGCAGGTG-3’) at 200 nM and MM1(5’-GGGAACCCGCTGATAATTCTTGG-3’) at 200 nM (S3 Fig). The 5’ end of FAM-MT1 and FAM-MT3 were modified with 6-carboxyfluorescein (FAM). To reduce non-specific PCR products, 400 nM heat-stable RecA protein from a thermophilic bacterium, Thermus thermophiles, and 400 μM ATP were included in the PCR reaction buffer (10 mM Tris-HCl pH 8.3, 50 mM KCl, 2.5 mM MgCl2) [98]. The following amplification program was used: 2 min at 94°C—27 x [30 s at 94°C—30 s at 55°C—1 min at 72°C]—5 min at 72°C. PCR fragments corresponding to mat1-P and mat1-M alleles were resolved on 5% polyacrylamide gels. Fluorescence was detected and quantified using Typhoon FLA9500 (GE Healthcare) and ImageQuant (GE Healthcare).
Each gene knockout in the Bioneer collection contains ‘up-tag’ and ‘down-tag’ sequences that provide a unique barcode for each knockout. To confirm the identity of the mutants identified in our screen, we amplified the KanMX4 region using the U1 primer (5’-CGCTCCCGCCTTACTTCGCA-3’) and D1 primer (5’- TTGCGTTGCGTAGGGGGGAT). The PCR products were then sequenced using cpn1 (5'-CGTCTGTGAGGGGAGCGTTT-3') to read the up-tag and cpc300 (5'-AGACCGATACCAGGATCTTGCC-3') to read the down-tag. The results were compared with the barcode list.
S. pombe cells were propagated in 16 mL liquid cultures (YES) at 30°C to pre-saturation and genomic DNA for Southern blots was prepared as described above. Genomic DNA was digested with HindIII to classify the mating-type switching defective genes according to DSB formation or presence of rearrangements in the mating-type region (Fig 3), or with DdeI to assay mat1 content (S3 Fig). The digested samples were electrophoresed in 0.7% agarose gels. The probe to analyze mat1 content was a PCR product made with GTO-1369 (5’- GAGCCTACTGTTAATATAATAACATTATG-3’) and GTO-1370 (5’-CCTTCAACTACTCTCTCTTCTTTTCCTACCC-3’), corresponding to the centromere-proximal DdeI-NsiI fragment [23]. The probes to classify the mutants were 10.4 kb HindIII fragments containing mat1-P or mat1-M.
|
10.1371/journal.pbio.0060235 | A Mutual Support Mechanism through Intercellular Movement of CAPRICE and GLABRA3 Can Pattern the Arabidopsis Root Epidermis | The patterning of the Arabidopsis root epidermis depends on a genetic regulatory network that operates both within and between cells. Genetic studies have identified a number of key components of this network, but a clear picture of the functional logic of the network is lacking. Here, we integrate existing genetic and biochemical data in a mathematical model that allows us to explore both the sufficiency of known network interactions and the extent to which additional assumptions about the model can account for wild-type and mutant data. Our model shows that an existing hypothesis concerning the autoregulation of WEREWOLF does not account fully for the expression patterns of components of the network. We confirm the lack of WEREWOLF autoregulation experimentally in transgenic plants. Rather, our modelling suggests that patterning depends on the movement of the CAPRICE and GLABRA3 transcriptional regulators between epidermal cells. Our combined modelling and experimental studies show that WEREWOLF autoregulation does not contribute to the initial patterning of epidermal cell fates in the Arabidopsis seedling root. In contrast to a patterning mechanism relying on local activation, we propose a mechanism based on lateral inhibition with feedback. The active intercellular movements of proteins that are central to our model underlie a mechanism for pattern formation in planar groups of cells that is centred on the mutual support of two cell fates rather than on local activation and lateral inhibition.
| The patterning of the Arabidopsis root epidermis depends on a genetic regulatory network that operates within and between cells. Genetic studies have identified a number of key components of this network, but the functional logic of the network has remained unclear. In this work, we integrate genetic and biochemical data in a mathematical model that we use to explore both the sufficiency of known network interactions and the extent to which additional assumptions about the model can account for wild-type and mutant data. Our model shows that an existing hypothesis concerning the autoregulation of the transcription factor WEREWOLF does not account fully for observed expression patterns, and we confirm the absence of autoregulation experimentally in transgenic plants. We propose an alternative mechanism centred on the movement of transcriptional regulators between epidermal cells, and present experimental support for this mechanism. These movements underlie a novel mechanism for pattern formation in planar groups of cells, centred on mutual support of two cell fates rather than local activation and lateral inhibition.
| The cells of the Arabidopsis root epidermis emerge from the initial cells in the root meristem with the potential to adopt either of two cell fates—trichoblasts (cells that can go on to differentiate as root hair cells) or atrichoblasts (that differentiate into non–hair-bearing epidermal cells). In the wild-type seedling, the two cell types are arranged in a stereotyped spatial pattern, with files of trichoblasts overlying two cortical cells (the H position) separated by files of atrichoblasts in contact with only one underlying cortical cell (the N position) (Figure 1) [1,2]. This fixed pattern does not result from lineage restriction, but depends on a combination of positional information from the cortex and the operation of a genetic regulatory network within the epidermis [3–5]. At the core of this network lie protein complexes centred on the basic helix-loop-helix proteins GLABRA3 (GL3) and ENHANCER OF GLABRA3 (EGL3) and the WD40-repeat–containing protein TRANSPARENT TESTA GLABRA (TTG). These proteins can bind to the MYB proteins WEREWOLF (WER) and CAPRICE (CPC) to form two protein complexes (the WER- and CPC-complexes, respectively).
Genetic and biochemical studies have highlighted a number of basic features of the epidermal interaction network. First, the WER-complex represses GL3/EGL3 transcription and enhances CPC transcription [6–8]. The CPC-complex is believed to lack transcriptional activity, but CPC has been reported to repress WER transcription [8]. Second, the CPC and GL3 proteins exhibit striking mobility, moving freely between epidermal cells [9–11]. Third, the SCRAMBLED (SCM) receptor-like kinase is believed to play a role in the interpretation of a cortical signal that biases pattern formation by repressing WER transcription in the H position [12,13]. These network features have been proposed to underlie a pattern-forming mechanism based on lateral inhibition [8], but a detailed investigation of their sufficiency to account for experimental data has not been carried out. It has been suggested on theoretical grounds, however, that autoregulation of WER activity is necessary for epidermal pattern formation [14,15], although experimental support for this proposal is lacking [14]. In this paper, we show by a combination of mathematical modelling and experimental studies that WER autoregulation does not play a significant role in the epidermal patterning network, and propose a mechanism for patterning that depends on the mutual support of the two epidermal cell fates.
We have developed a mathematical model representing the core epidermal interaction network, in order to investigate the regulatory logic of epidermal patterning. Our model seeks to capture all key interactions and protein movements identified in experimental studies (Figure 2). The nature of the regulation of WER transcription is central to our model. WER transcription is repressed by both SCM and CPC, but no specific activators of WER transcription have been identified. To address directly the open question of the necessity for WER autoregulation, we consider two alternative forms of WER regulation. In the first version, we assume local WER self-activation, implemented by the enhancement of WER transcription by WER-complex (Figures 2A and 3A). In this scenario, CPC down-regulates WER indirectly via competition for TTG/GL3/EGL3. In the second version, we do not include local WER self-activation, assuming instead that WER transcription is activated uniformly in all epidermal cells, with both CPC and SCM (in the H-position cells) repressing WER transcription directly (Figures 2B and 3B). We refer to the genetic regulatory network containing the first version of WER regulation as the “local WER self-activation model,” and the regulatory network containing the second version as the “mutual support model.”
In order to focus more clearly on the core logic of the epidermal patterning network, the model incorporates a number of simplifying assumptions. First, since the expression pattern of TTG within the epidermis is not known, we assume that it is expressed uniformly and that it plays only a permissive role in allowing the formation of WER and CPC protein complexes with GL3/EGL3. On the basis of this assumption, we do not include an explicit TTG variable in our mathematical model. TTG is, however, present implicitly in all the cells of our model epidermis. Second, we do not include the CPC-complex explicitly in the model; rather, we represent the ability of CPC to compete with WER for binding to TTG/GL3/EGL3 [16] by a direct inhibition of WER-complex formation by CPC. The CPC-complex is implicitly present in all model cells that express both CPC and GL3/EGL3. Third, GL3 and EGL3, which act in a partially redundant manner [11], are represented by a single network component. Similarly, the three MYB proteins CPC, TRIPTYCHON (TRY), and ENHANCER OF TRY AND CPC1 (ETC1), which act in a partially redundant manner [17], are also represented by a single network component (denoted by CPC).
In order to incorporate the observed intercellular movement of the CPC and GL3/EGL3 proteins, we have imposed a specific mechanism in our model: both CPC and GL3/EGL3 proteins are moved actively out of the cells in which they are produced (translated). We adopt this active mechanism to reflect the observed accumulation of these proteins in the nuclei of cells neighbouring the cells in which they are produced. A GL3-YFP fusion protein, expressed under the GL3 promoter in a gl3 mutant background, accumulates in the nuclei of N-position cells, even though the corresponding mRNA is restricted to H-position cells [11]. Similarly, a HA-tagged CPC protein, expressed under the CPC promoter in a cpc mutant background, accumulates in the nuclei of H cells, even though its mRNA is restricted to N cells [10]. A CPC-GFP fusion protein can be observed in the nuclei of both cell types [9,10]. However, this protein is expressed at much higher levels than the endogenous CPC protein (due perhaps to protein stabilisation) and causes numerous cells in the N position to adopt the trichoblast fate [10]. These experimental results demonstrate that both CPC and GL3 proteins move away from their sites of production, but the mechanism by which they do this is not known. Given this uncertainty, we have incorporated in our model a simple movement scheme that captures the observed complementary patterns of protein production and accumulation. Possible molecular mechanisms underlying this scheme are discussed below.
We simulate a ring of 16 epidermal cells (which we refer to as the epi-net) following its emergence from the meristem. This represents the stereotypical number of cells found in each epidermal ring in the apical region of the seedling root in which patterning takes place [1,2]. As the cells age (and so move further away from the root apex), occasional anticlinal cell divisions can occur, increasing the number of cells in an epidermal ring [4,5] (the cross-section in Figure 1 shows an example of an older epidermal ring in which this has occurred). However, since we are here modelling the earliest stages of patterning in the epidermis, we do not consider these later events explicitly. Each simulated cell (referred to as a cell-net) contains all the components of the Arabidopsis root hair patterning network shown in Figure 2, and so in a simulated cell, any combination of components can be expressed, including the combinations specific to trichoblast or atrichoblast cells. Figure 3 shows the network state (expression of network components and active interactions) in cell-nets corresponding to epidermal cells that have adopted either a stable trichoblast or atrichoblast fate. The mechanistic differences between the local WER self-activation and mutual support models are clearly visible in Figure 3.
Since mechanistic details (such as rate laws and the corresponding kinetic parameters) of the epidermal interaction network are not known, a model based on differential equations would involve a large number of unknown parameters. Instead, we adopt a modelling framework that encodes the logical form of interactions. At a given time, the components of a cell-net are either expressed or not. Components that have only positive regulatory inputs (WER, GL3/EGL3, CPC, CPC, GL2, and GL2—see Figure 2) are expressed if their direct positive regulators are expressed. For example, if WER (mRNA) is expressed in a cell-net, then WER (protein) will be expressed. GL3/EGL3 has one negative input (the WER-complex) and is expressed if its input is not. To specify similar logical rules for the expression of the two components (WER and WER-complex) whose production is regulated by a combination of positive and negative regulators would involve making arbitrary assumptions about the dominance of activators or repressors (see Protocol S1). To avoid this, and to allow scope for investigating the effects of any assumptions we make about dominance, we adopt a stochastic formalism in which these components each have a time-evolving probability of expression. The probability of a component being expressed corresponds to the average abundance of that component in the cell. In our formalism, the change in probability over time is determined by the expression of the component's direct regulators and the corresponding activation/inhibition “rates” (which encode the relative strengths of the regulatory interactions). For example, the probability of the WER-complex being expressed is increased by a small amount if both GL3/EGL3 and WER are expressed, and decreased by a small amount if both GL3/EGL3 and CPC are expressed. The incorporation of stochasticity in our model not only increases the investigative scope, but also supplies a form of noise, which is an inherent feature of biological systems and is an integral part of cell differentiation. Furthermore, this stochasticity plays an important role in triggering fate assignment in our model of the scm mutant, which lacks positional cues from the cortical cells (see Protocol S1). However, the formalism that we adopt is not intended to provide a detailed representation of the stochastic nature of molecular dynamics in a cell. A detailed description of the modelling formalism and equations can be found in Materials and Methods.
Our stochastic Boolean formalism provides a versatile setting in which to investigate the effects of the relative strengths of combinatorial regulators for a specified regulatory logic. However, the results that we obtain from the model are not dependent on the use of this specific formalism. In particular, the behaviour of the model epidermis can be produced using Boolean models with appropriately chosen deterministic logical functions. In this case, the stochasticity needed to trigger patterning in the scm mutant epidermis can be introduced by adopting an asynchronous update scheme (see Protocol S1).
To assess the ability of the model networks to account for observed wild-type expression patterns, we simulated epi-nets in which all network components (except SCM) were initially expressed at the same level in all cells (i.e., all cell-nets are initially identical). To represent the positional bias received from the underlying cortex, SCM was set to be active only in cells located in the H position, resulting in a lower transcription rate of WER than in the N position. In an epi-net, H and N positions alternate: odd-numbered cell-nets are in the H position, while even-numbered cell-nets are in the N position (see Figure 4). With this imposed positional bias, both the local WER self-activation and mutual support models are capable of generating stable expression patterns that agree with the expression patterns observed in experimental data (Figure 4).
In scm mutant plants, experimental data show that epidermal cells adopt well-defined fates, but in a pattern that is not strictly correlated with position relative to the cortex [12,13]. To assess whether the model networks can also account for this phenotype, we set SCM to be inactive in all cells. In these simulations, the only patterning cues come from the stochasticity inherent in our modelling approach (we do not incorporate stochasticity in the initial conditions). Figure 5 shows a composite of the steady states resulting from 15 independent simulations in rings of cell-nets, aligned vertically to produce a virtual epidermis. However, it is important to note that such a picture does not represent the result of a full two-dimensional simulation, including aging and longitudinal signalling between cell rings. Both the local WER self-activation and mutual support models develop stable patterns in which each cell-net adopts a coherent state (either trichoblast or atrichoblast). For both models, the patterns produced are qualitatively comparable to those observed in scm mutant roots [12,13]. The total removal of cortical bias in our simulations may not be entirely equivalent to the situation pertaining in scm mutant roots, as the phenotypes of existing scm alleles suggest that some cortical positional information persists in these cases [13]. However, our simulations show clearly that both forms of the epidermal patterning network are capable of spontaneous pattern formation, even in the absence of spatial bias. For both local WER self-activation and mutual support, the stochasticity in our modelling formalism acts to break symmetry allowing a spatially patterned state to emerge from a spatially uniform initial state.
To simulate the effect of other mutations, we set the corresponding cell-net components to be inactive in all cell-nets. Simulations of a wer mutation (unpublished data) result in identical expression patterns for both models, in agreement with experimental data (namely, the uniform expression of GL3/EGL3) [8,11,17]. We simulate WER overexpression by imposing uniform expression of both WER mRNA and WER protein throughout the epi-net. The epi-net steady states resulting from 15 independent simulations of the two versions of the epidermal patterning networks are shown in Figure 6. The expression pattern of all network components other than WER mRNA and WER are as in the simulated scm mutant (Figure 5), with each cell-net adopting a coherent state corresponding to either a trichoblast or atrichoblast. This mirrors the expression patterns reported in [8,14] and reflects the fact that WER, when overexpressed uniformly, is no longer able to respond to an imposed cortical bias.
Figure 7 shows the expression of WER mRNA and WER protein in a simulated cpc mutant. Although both the local WER self-activation and mutual support models generate expression patterns for most network components that are in line with experimental data [11], they generate significantly different patterns of WER expression. In the local WER self-activation model (Figure 2A), the activation of WER expression by the WER-complex results in a wild-type pattern of WER expression even in the absence of CPC (Figure 7A, cf. Figure 4A). In contrast, the loss of CPC-mediated repression of WER in the mutual support model (Figure 2B) results in an increase in WER expression in the H positions, as it is only being repressed by SCM in the absence of CPC (Figure 7B). This corresponds to the pattern of WER expression observed experimentally [8]. This result suggests that the mutual support model, which does not incorporate local WER self-activation, more accurately reflects events occurring during the patterning of the epidermis.
Since the local WER self-activation model fails to reproduce the observed pattern of WER expression in the cpc mutant, we tested the ability of the WER-complex (or WER) to enhance WER expression by examining the expression of GFP driven by the WER promoter (WERpro::GFP) in a wer mutant background (using a null mutant in which no functional WER protein is produced). We found GFP expression to be the same in wild type and the wer mutant, showing that WER transcription does not depend on the presence of functional WER protein (Figure 8A and 8B). To test directly our alternative assumption that WER transcription is activated uniformly in all epidermal cells, we carefully examined WER promoter activity (as visualised by WERpro::GFP) in wild-type seedlings. Whereas WERpro::GFP is preferentially expressed in the N cell file in less apical cells of the meristem, it exhibits uniform activity between N and H cell positions in cells proximal to the initials (Figure 8C). These results show that the initially uniform activity of the WER promoter throughout the epidermis resolves rapidly into a pattern matching that of WER transcription in wild-type roots even in the absence of WER protein. This strongly suggests that the establishment of patterned WER transcription—a key event in epidermal patterning—does not depend on local WER self-activation. Since the pattern of WER promoter activity in both wild-type and wer mutant roots corresponds to the wild-type pattern of cell fate in the epidermis, there is no obvious role for posttranscriptional regulation of WER activity (since posttranscriptional regulation of WER can only occur in cells in which WER is transcribed). Taken together, our modelling and experimental results show that WER is initially activated uniformly in the epidermis, and suggest that its rapid repression in emerging trichoblasts is controlled by a combination of SCM-mediated positional information and CPC.
To explore further the differences between the two model networks, we simulated mutants that are incapable of forming the WER-complex. Since GL3/EGL3 and TTG are required for complex formation, both the gl3 egl3 double mutant and the ttg mutant should lack WER-complex. In this scenario, the local WER self-activation and mutual support models predict different patterns of WER expression. In the local WER self-activation model, the failure of WER-complex formation results in a uniform loss of WER expression in the model epidermis (Figure 9A). However, since WER expression does not depend on local self-activation in the mutual support model, WER is expressed in an essentially wild-type pattern in the model epidermis (with an increased probability of expression in cells in the H position due to the lack of CPC-mediated repression) (Figure 9B). To test this prediction experimentally, we examined the expression of GFP driven by the WER promoter (WERpro::GFP) in these mutant backgrounds. As predicted by the mutual support model, GFP expression is essentially the same in the wild-type and mutant epidermis (Figure 10). This supports our finding that WER self-activation does not play a significant role in the early stages of epidermal patterning, and provides direct experimental validation of the predictions of the mutual support model of epidermal patterning.
Taken together, our modelling and experimental studies support a mechanism for spatial pattern formation in the Arabidopsis root epidermis that depends critically on the movement of mobile proteins between cells—a lateral inhibition with feedback (LIF) mechanism. Importantly, this mechanism does not depend on local WER self-activation, but relies instead on the repression of WER transcription in emerging trichoblasts by CPC protein. Previous theoretical discussions of epidermal patterning [14,15] have suggested that local WER self-activation is a necessary feature of the patterning network—a local activation and lateral inhibition (LALI) mechanism [18,19].
Although both the LALI and LIF mechanisms can generate similar stable patterns of cell fate, the logical structure of the underlying networks is quite different. LALI mechanisms depend on interlinked positive feedback (short range) and negative feedback (long range) whereas LIF depends on a single “double-negative” feedback loop, mediated by intercellular signalling, and does not depend on local self-activation. The logical structure of the LIF mechanism is analogous to the Delta-Notch signalling system in animal epithelia, in which proneural activity in one cell represses proneural activity in its neighbours through the transmembrane ligand Delta and its receptor Notch, ensuring directional signalling. Models of the Delta-Notch system exhibit spontaneous patterning that does not depend on any local self-activation [20,21].
In the LALI mechanism, the “activated” cell state (atrichoblast) inhibits its neighbours, which adopt an alternative fate (trichoblast). In contrast, in the LIF mechanism, cells adopting one of the two epidermal fates are mutually supporting, producing factors required by cells adopting the alternative fate. Adoption of the atrichoblast fate (high WER-complex) requires GL3/EGL3 from neighbouring cells; adoption of the trichoblast fate (low WER-complex) requires CPC from neighbouring cells (to prevent accumulation of WER-complex). In other words, a cell can only have high levels of WER-complex if a neighbouring cell has a low level of WER-complex and vice versa. This model therefore predicts that “runs” of three or more epidermal cells with similar levels of WER-complex should not occur. In the root apical meristem, where the early patterning of gene expression in the epidermis occurs, each ring of epidermal cells contains 16 cells, with alternating cells in H and N positions, as encoded in our model [22]. We therefore observe a strict alternating pattern in our wild-type simulations that incorporate a positional bias from the cortex. In the simulated scm mutant, which lacks cortical bias, we do not observe more than two cells of the same fate neighbouring each other. In growing roots, the number of cells in an epidermal ring tends to increase as cells move away from the root apical meristem, due to occasional anticlinal cell divisions [4,5]. This is shown clearly in Figure 1, in which most H-position cells are separated by two N-position cells. In older epidermal rings, three or more adjacent cells are sometimes observed to have the same pattern of gene expression, which cannot be accounted for by our early patterning network in its current form. However, it is likely that once the basic pattern of expression of the core epidermal patterning components has been established, cell fate is stabilised by additional factors such as chromatin modification [23–25]. Such fate stabilisation mechanisms would allow cells to maintain their network state even when no longer supported by a neighbouring cell of the alternate fate.
We have shown by model simulation that the LALI mechanism (incorporating local WER self-activation) fails to account fully for the previously reported phenotype of a cpc mutant root, and by experiment that a specific form of local self-activation (WER-mediated up-regulation of WER transcription) does not operate in the early patterning of the root epidermis. Our combined modelling and experimental results favour an alternative mechanism (LIF) in which the two emerging cell fates mutually support each other through the active exchange of the mobile proteins CPC and GL3. The mutual support model predicts patterns of WER promoter activity in wer, gl3 egl3, and ttg mutant roots that are similar to wild type. We have verified these predictions experimentally, providing validation for the model and further support for our proposed patterning mechanism. Importantly, the model based on the LALI mechanism does not account for these new observations.
The mutual support model incorporates the active movement of the CPC and GL3 proteins from the cells in which they are produced to neighbouring cells. Such an active mechanism is suggested by the previously reported complementary patterns of production and accumulation of these proteins in the epidermis. We have adopted a modelling formalism based on binary states of expression (“on” or “off”). In this formalism, the patterning of the model epidermis depends on this active mechanism of protein movement. However, the possibility remains that the observed complementary patterns of protein production and accumulation could result from simple diffusion of the proteins between cells, together with sequestration of the proteins into nuclear-localised protein complexes (as occurs in the directed movement of the SHORTROOT protein in the root apical meristem [26]).
Previous theoretical discussions of epidermal patterning have proposed that local self-activation is a necessary feature of a patterning mechanism [14,15]. This conclusion is based on the theory of two-component activator–inhibitor models in which movement is purely diffusive. To explore the validity of this conclusion for the root epidermal patterning network, we have analysed two different mathematical representations of the mutual support model. First, we have developed a logical state (Boolean) model in which CPC and GL3 protein movement depends on a movement parameter, allowing both active and passive (diffusion-like) movement to be represented. Analysis of this model shows that passive GL3 movement is sufficient to account for patterning, so long as CPC moves actively (see Protocol S1). Second, we have developed a reaction–diffusion analogue of our logical model in which both GL3 and CPC move between cells by simple diffusion alone (see Protocol S1). When reduced to an effective two-component model for GL3 and CPC (by assuming that protein complex formation and WER dynamics reach equilibrium much faster than diffusive processes), we show that the model can take the form of a cross activator–inhibitor system, which is capable of spontaneous pattern formation via diffusion-driven instability [27]. This analysis shows that the mutual support mechanism we propose can generate pattern spontaneously by diffusive protein movement and protein complex formation, in the absence of any local self-activation reaction. Numerical simulation of both the full and reduced models confirms that the diffusive mechanism generates stable patterns with protein distributions that match those observed in the root epidermis (see Protocol S1).
Our results serve to highlight the importance of a detailed investigation of the mechanisms of the intercellular movement of proteins such as CPC and GL3/EGL3 [28]. A number of simple mechanisms might underlie an effective directionality of protein movement away from producing cells. For example, the movement of proteins through plasmodesmata could be dependent on a chaperone protein that is produced only in cells producing the mobile protein. Alternatively, passage through plasmodesmata could depend on localisation of the protein in the endoplasmic reticulum, which would favour movement away from the cells in which the protein is translated. An intriguing parallel is provided by the movement of small metabolites through small intercellular pores (microplasmodesmata) in the filamentous cyanobactoria Anabaena. A recent study has shown that the permeability of pores (and hence the mobility of metabolites) mirrors the states of differentiation of the two cell types in this system [29]. In particular, as individual cells in the filament move towards a differentiated heterocyst fate, the permeability of pores between emerging heterocysts and neighbouring vegetative cells decreases compared to that between two vegetative cells. Thus, in this very different system, differential permeability of intercellular channels, dependent on cell fate, can establish spatially patterned protein distributions. The widespread occurrence of cell-to-cell trafficking of macromolecules in plant and animal tissues [30] suggests that mechanisms of the type we describe—centred on mobile proteins that can be sequestered in protein complexes—may play a role in a range of pattern-forming processes operating in planar groups of cells.
The WERpro::GFP construct was previously reported in [31]. Briefly, it included a 2.5-kb WER promoter fragment 5′ to the GFP coding sequence and a 1.1-kb 3′ WER fragment, and faithfully reported the WER transcription pattern. To examine the expression of WERpro::GFP in the wer, gl3 egl3, and ttg mutant backgrounds, we used the published wer allele, wer-1 [31], the gl3–1 egl3–1 line [11], and the ttg1–13 mutant [2]. Plants homozygous for the WERpro::GFP insertion were crossed to plants homozygous for one of the mutant alleles. The resulting plants were self-pollinated, and F2 plants that were homozygous for the wer-1, gl3–1 egl3–1, or ttg1–12 mutations and the WERpro::GFP transgene were selected. These plants were in turn self-pollinated to produce a population of seed that were homozygous for the desired mutant allele and the WERpro::GFP transgene.
For confocal microscopy imaging, 4- or 5-d-old roots were stained with 10 μg/ml propidium iodide and visualised on a Leica TC5 SP confocal microscope. Images were assembled using Adobe Photoshop.
In our models, a ring of 16 epidermal cells (the stereotypical number found in the apical region of the seedling root in which patterning takes place) is represented by an epi-net comprising 16 identically composed cell-nets, indexed by the integer j = 1, 2, …, 16. The set of components in each cell-net, together with their interactions, is shown schematically in Figure 2. In the mathematical model, the state of mRNAs is represented by the corresponding gene name abbreviation (for example, CPCtj represents the state of CPC mRNA at time t in cell-net j). The state of the corresponding protein carries an appended “p” (for example, WERptj represents the state of WER protein at time t in cell-net j). The state of the WER-complex is denoted by WERc. In order to capture what we believe to be the essential logic of the epidermal patterning network, while keeping the number of distinct molecular species in the model to a minimum, a number of known network components have been left out of the model, or combined into a single model variable. Both GL3 and EGL3 are represented jointly by a single model element GL3 (comprising variables for mRNA and protein). We justify this simplification by noting that all published data suggest that GL3 and EGL3 are regulated similarly and exhibit functional redundancy. Similarly, we represent the three single-repeat R3 MYB proteins CPC, TRIPTYCHON (TRY), and ENHANCER OF TRY AND CPC1 (ETC1) by a single model element CPC, since experimental evidence supports the idea that they act collectively and redundantly to specify the trichoblast fate [17]. Furthermore, in the absence of experimental data to the contrary, we assume that the WD-repeat protein TRANSPARENT TESTA GLABRA (TTG), an essential component of the WER-complex, is expressed uniformly throughout the epidermis. This assumption renders the explicit representation of TTG in the models unnecessary, and our models do not contain TTG variables (although the protein is implicitly assumed to be present in all cells).
To investigate the patterning potential of the local WER self-activation and mutual support models (Figure 2), we use a discrete-time logical formalism. In this approach, the state of each network component is represented by a binary variable taking either value 1 (component expressed) or 0 (not expressed). Time evolution of the network state is modelled by the synchronous update of the state of each network component at equally spaced time points (t, t+1, t+2, …). For the network components whose state is regulated by only one other component type (WERp, GL3, GL3p, CPC, or CPCp), we adopt a conventional deterministic Boolean update formalism [32]. For the two components whose state is regulated by more than one input (WER and WERc), we adopt a novel formalism based on the probability Ptj[X] that the state of component Xtj will be 1 at time t. Rather than specifying a deterministic function for the time evolution of the states of these components, we instead specify a deterministic rule for the time evolution of the probability of expression. This form of update allows us both to vary the relative strengths of the inputs and to incorporate stochasticity in the update process. Although this approach directly introduces stochasticity into the evolution equations of only two network variables, the stochasticity filters through to the other components. Thus, whereas all components could be represented probabilistically, this would necessitate the introduction of many more undetermined parameters without adding further functionality to the model. In a simulation of the network, the actual values (0 or 1) of WERtj and WERctj are determined stochastically at each time step according to the probability of expression, P tj[WER] and P tj[WERc].
The parameters in our probabilistic update functions (see below) allow us to explore the robustness of patterning to changes in the relative strengths of the inputs. Furthermore, the incorporation of stochasticity into the system is important, since stochasticity is an inherent feature of biological networks and is required in our models to initiate patterning in the simulated scm mutant. However, our approach does not attempt to mimic any specific form of stochasticity found in biological systems, and we have shown that the results obtained using our probabilistic formalism can be reproduced by using a deterministic Boolean model with stochasticity introduced in the form of asynchronous state update (see Protocol S1).
Our probabilistic Boolean formalism provides a simple way of exploring the consequences of specific assumptions about the regulatory logic of the epidermal patterning network. However, the use of a logical (on/off) representation of the network state assumes that the regulatory interactions represented in the model (e.g., transcription and translation) are essentially “all or nothing.” Since our primary objective is to explore the differences between two alternative network structures, we believe that this assumption is appropriate. Other approaches to modelling regulatory networks, such as those based on differential equations, do not depend on such an assumption being made. However, these models require the specification of many more parameters than our model, to represent the details of specific interaction kinetics. Such models can provide more-realistic representations of the dynamical evolution of the state of the network. Given that there are currently no data, either from which appropriate parameters can be specified, or against which detailed network dynamics can be validated, we do not believe that these approaches currently have a significant advantage over our logical formalism.
The local WER self-activation and mutual support models are defined in Equations (1) and (2), respectively. The models are identical apart from the equation encoding the time-evolution of WER mRNA. The symbol ∨ represents the logical “inclusive OR” function (i.e., A∨B = 0 if and only if A = B = 0). c0, c1,...c5, are positive parameters that determine the relative strengths of the inputs in the probabilistic multi-input update functions for WER and WERc (Figure 2). The constant terms c1 and c5 represent either constitutive production or degradation, depending on their preceding signs. The regulatory inputs to WER and WERc specify the amount by which the probability of expression of these components changes during a single time step. This form of update rule is similar to the rate equations that form the basis of differential equation models (in which the rate of change of a component is determined by the values of its direct regulators). Values within the brackets ⌊ ⌋ are forced to remain between 0 and 1.
A positional bias from the underlying cortex is incorporated in the models via the state of the SCRAMBLED (SCM) receptor-like kinase, which is taken to be 1 in cell nets occupying the H position and 0 in cell-nets occupying the N position. Activity of SCM results in a reduction in the rate of transcription of WER, determined by the parameter c0. We assume the two positions to be arranged alternately, as is typically the case in the apical root epidermis (anticlinal cell divisions in the epidermis, which can increase the spacing between H-position cells, typically occur further from the meristem, where the expression pattern of network components has already stabilised).
The initial state of all components, bar SCM (see above), is identical in all cell-nets, representing the fact that the final stable state of each cell-net is determined by its position relative to the underlying cortical cells rather than cell lineage. As the state of each cell-net evolves in time, the cell-nets adopt stable patterns of expression corresponding to either the trichoblast or atrichoblast cell fate (Figure 4). A detailed discussion of the dependence of the behaviour of the models on initial conditions and parameter values can be found in Protocol S1.
|
10.1371/journal.pgen.1002513 | The Caenorhabditis elegans Eph Receptor Activates NCK and N-WASP, and Inhibits Ena/VASP to Regulate Growth Cone Dynamics during Axon Guidance | The Eph receptor tyrosine kinases (RTKs) are regulators of cell migration and axon guidance. However, our understanding of the molecular mechanisms by which Eph RTKs regulate these processes is still incomplete. To understand how Eph receptors regulate axon guidance in Caenorhabditis elegans, we screened for suppressors of axon guidance defects caused by a hyperactive VAB-1/Eph RTK. We identified NCK-1 and WSP-1/N-WASP as downstream effectors of VAB-1. Furthermore, VAB-1, NCK-1, and WSP-1 can form a complex in vitro. We also report that NCK-1 can physically bind UNC-34/Enabled (Ena), and suggest that VAB-1 inhibits the NCK-1/UNC-34 complex and negatively regulates UNC-34. Our results provide a model of the molecular events that allow the VAB-1 RTK to regulate actin dynamics for axon guidance. We suggest that VAB-1/Eph RTK can stop axonal outgrowth by inhibiting filopodia formation at the growth cone by activating Arp2/3 through a VAB-1/NCK-1/WSP-1 complex and by inhibiting UNC-34/Ena activity.
| The correct wiring of the nervous system depends on the ability of axons to properly interpret extracellular cues that guide them to their targets. The Eph receptor tyrosine kinases (RTKs) have roles in guiding axons, but their signaling pathways are not completely understood. In this study, we used the nematode Caenorhabditis elegans to study how the VAB-1 Eph RTK regulates the growth cone structure for axon guidance. Genetic and molecular data show that VAB-1 regulates the conserved molecules NCK-1, WSP-1/N-WASP, and UNC-34/Ena. Our study provides a model of how the VAB-1 Eph RTK modulates the growth cone structure to inhibit axonal outgrowth. We show that activated VAB-1 can inhibit an NCK-1/UNC-34 interaction by binding to the NCK-1 SH2 domain. We also show that NCK-1 and WSP-1 can physically interact and that VAB-1/NCK-1 and WSP-1 form a complex in vitro. We suggest that the VAB-1 Eph RTK can contribute to the termination of axon outgrowth by two methods: 1) The VAB-1/NCK-1/WSP-1 complex activates ARP-2/3 to change the actin growth cone dynamics to that of a branched structure thus reducing the number of filopodia, and 2) VAB-1 inhibits axon extension by inhibiting UNC-34/Ena's function in actin polymerization.
| During development, axons navigate to their final destination by interpreting extracellular guidance cues through their growth cone. The Eph receptor tyrosine kinases (RTKs) and their ephrin ligands are involved in directing axons to their proper location [1], [2]. Studies in vertebrate systems have identified a number of effectors in the Eph RTKs signaling pathway in axon guidance [2]. However, the molecular mechanism of how Eph RTKs regulate axon guidance is still incomplete. This is partly due to the large number of Ephrins and Eph RTKs that can engage in crosstalk [2], [3]. The presence of a single Eph RTK, VAB-1, in Caenorhabditis elegans can simplify the analysis of the signal transduction events from the receptor. The C. elegans VAB-1 Eph RTK is required for various aspects of neuronal development, including neuroblast movements, and axon guidance [4], [5], [6], [7]. The molecules involved in VAB-1 signaling in axon guidance are still unknown. To resolve this issue, we used a genetic suppressor approach as well as a physical protein interaction approach and identified NCK-1, WSP-1/N-WASP, UNC-34/Ena, and the Arp2/3 complex as molecules regulated by VAB-1 signaling in axon guidance.
The Nck adaptor proteins are known actin cytoskeleton regulators, and have been shown to function downstream of several axon guidance receptors including Robo, Dcc and the Eph RTKs [8], [9], [10], [11]. Although the function of Nck has been studied in various organisms, the biological function of NCK-1 in C. elegans has only been recently explored [12]. Furthermore, what molecules interact with the C. elegans NCK-1 is still unknown.
The WASP protein family (WASP and N-WASP) are scaffolds that integrate multiple signaling pathways, leading to the formation of short branched actin filaments through the activation of the Arp2/3 complex [13]. The C. elegans N-WASP homolog, WSP-1, functions in neuronal cell migration and axon guidance [14], [15]. However, a connection between WSP-1 and a guidance receptor has not yet been established.
The Ena/VASP proteins are involved in actin-dependent movements including neuronal migration and axon guidance, and are known for their role in promoting filopodia formation [16]. In C. elegans, the Ena/VASP homolog UNC-34 is required for proper neuronal cell migration, axon guidance and filopodia formation [14], [17], [18], [19], [20]. Previous work has shown that Ena/VASP proteins are versatile in their developmental roles and function in both repulsive and attractive cues. For example Ena/VASP are effectors of receptors for repulsive cues such as SAX-3/Robo, UNC-5/Netrin receptor and EphB4, but they can as also act as effectors for attractive cues downstream of receptors such as UNC-40/DCC [21], [22], [23], [24], [25].
The Arp2/3 complex is a conserved family of actin nucleators and when activated results in the formation of an elaborate network of branched actin filaments similar to those found in lamellipodia [26], [27]. In C. elegans, the Arp2/3 complex is required for axon guidance, and the initiation of growth cone filopodia downstream of an unidentified axon guidance signal [15], [20].
In this paper, we describe some of the molecular events that allow the VAB-1 Eph RTK to regulate actin dynamics for axon guidance. We provide genetic and biochemical evidence to show that VAB-1 signals through NCK-1 and WSP-1/N-WASP, and negatively regulates UNC-34/Ena. We propose a model for PLM (Posterior lateral microtubule) axon termination whereby the VAB-1 Eph RTK is able to prevent axon extension by inhibiting growth cone filopodia formation. This is accomplished by negatively regulating the activity of the filopodia elongator UNC-34/Ena, and simultaneously activating Arp2/3 through a VAB-1/NCK-1/WSP-1 complex.
To identify VAB-1 Eph RTK effectors, we utilized transgenic animals carrying mec-4::myr-vab-1 (quIs5) which encodes a constitutively active VAB-1 tyrosine kinase (myristoylated-VAB-1 termed MYR-VAB-1) in the mechanosensory neurons [6]. In wild-type young adults, PLM neuron cell bodies are located in the tail region and have axons that stop at the centre of the animal (Figure 1A). We previously showed that myr-vab-1 caused neuronal defects in the mechanosensory neurons, in particular the premature termination of PLM axons (Figure 1A, 1B) [6]. Since the MYR-VAB-1 behaves as a constitutively active VAB-1 RTK, we reasoned that mutations in effectors of the VAB-1 signal may suppress the neuronal defects. We used a candidate gene approach to examine genes with known roles in axon guidance and tested whether loss-of-function mutations could suppress the myr-vab-1 PLM premature termination phenotype. We identified nck-1 as a candidate effector of VAB-1 Eph RTK signaling. The nck-1(ok694) mutation partially suppressed the PLM axon premature termination (Figure 1B), indicating that other effectors are involved in the MYR-VAB-1 signaling. The C. elegans genome encodes for only one nck-1 adaptor protein, and is most similar to the human Nck2 and Drosophila DOCK [12]. NCK-1 has all the domain features of the NCK adaptor proteins, including three SH3 domains followed by a single SH2 domain. We previously reported that the deletion allele nck-1(ok694) is predicted to be a null allele, thus all of our genetic work was carried out using the ok694 allele [12].
If NCK-1 is an effector of VAB-1 signaling then we would expect the nck-1 loss-of-function mutation to have a phenotype similar to that of the vab-1 loss-of-function. Indeed, previous work showed that both vab-1 and nck-1 mutants have similar neuronal defects, including an overextension in PLM axons (Figure 1C) [6], [7], [12]. To further confirm that nck-1 and vab-1 are in the same pathway in the PLM neurons, we analyzed the effect of the double mutation on the PLM axons. The vab-1; nck-1 double mutation did not enhance the PLM over extension phenotype (Figure 1C), indicating that NCK-1 and the VAB-1 Eph receptor function in the same pathway to guide the PLM axons.
To determine if the PLM defects observed in vab-1 and nck-1 animals were present at an earlier stage, we examined the PLMs of the first larval stage (L1) (see Methods). Wild-type L1s had PLM axons that were 103–114 µm long, and terminated at a region anterior to the tip of the ALM cell body (93%) and is consistent with previous reports for L1 PLM lengths [28] (Figure 2A). Both vab-1 and nck-1 animals had PLM axons that significantly overgrew beyond the wild-type termination region (Figure 2A, 2B). This indicates that VAB-1 and NCK-1 are required at an early stage to prevent PLM axons from overgrowing beyond their normal termination region. We also showed that 96% of L1 myr-vab-1 transgenic animals had PLM axons that were undergrown when compared to wild-type (Figure 2A, 2C). The PLM undergrowth defects caused by MYR-VAB-1 were significantly reduced by nck-1(ok694) (57%) (Figure 2C). These results are consistent with our analysis carried out in early adults, and further confirm that NCK-1 is an effector of VAB-1 signaling in PLM axon guidance.
We previously showed that NCK-1 is expressed in various tissues including the nervous system [12]. In addition, like VAB-1, NCK-1 can function cell autonomously in the mechanosensory neurons for PLM axon guidance [6], [12]. If NCK-1 and VAB-1 function in the same pathway during neuronal development, then they should be localized in the same cells. Indeed, NCK-1 and VAB-1 were co-localized in some of the neurons, consistent with the role of NCK-1 as an effector of VAB-1 (Figure 3A, 3B). However, the expression pattern of VAB-1 and NCK-1 did not overlap exactly, suggesting that both NCK-1 and VAB-1 have independent roles during development (Figure 3A). Expression of NCK-1-GFP and activated VAB-1 (MYR-VAB-1) in the mechanosensory neurons showed that NCK-1 did co-localize with activated VAB-1 in the PLM axon and cell body (Figure 3B).
In a parallel approach we used yeast two-hybrid screens to identify effectors of VAB-1/Eph RTK signaling and identified the full length NCK-1 as a binding partner of the VAB-1 intracellular kinase region. Yeast two-hybrid analysis showed that the NCK-1 SH2 domain is sufficient to bind VAB-1 and that VAB-1 tyrosine Y673 is crucial for the interaction with the NCK-1 SH2 domain (Figure 4A).
To further confirm the NCK-1/VAB-1 interaction we used GST-pull down assays. Deletion analyses confirmed that the SH2 domain is necessary and sufficient to bind VAB-1 (Figure 4B). Furthermore, the NCK-1 interaction required an active tyrosine VAB-1 kinase since the NCK-1 SH2 domain did not bind a kinase inactive version of VAB-1 (G912E) (Figure 4C, 4D). Since SH2 domains are known to bind phosphotyrosines we wanted to test how specific the NCK-1 SH2 domain is for VAB-1. We found that four other SH2 domains (MIG-10, SEM-5, ABL-1, VAV-1) were unable to bind VAB-1 (Figure 4E). In summary, NCK-1 interacts with VAB-1 in a kinase dependent manner, the interaction is mediated via the NCK-1 SH2 domain and the VAB-1 Y673 juxtamembrane tyrosine, and VAB-1 has high specificity for the NCK-1 SH2 domain.
How does VAB-1 cause the PLM to stop once the VAB-1 Eph RTK is activated and adaptor proteins such as NCK-1 bind the receptor? A previous report indicated that Ena/VASP was required for repulsion caused by EphB4 signaling in fibroblasts, but it was unclear how the signal was conveyed [24]. The Ena/VASP family are composed of an N-terminal EVH1 domain, a central PRO region and a C-terminal Ena/VASP homology II domain (EVH2) [16]. We asked if NCK-1 could be the link between the Eph RTK and Ena/VASP. We first tested if NCK-1 and UNC-34 can directly interact. In vitro binding assays with bacterially expressed NCK-1 and UNC-34 confirmed that both proteins do indeed physically interact (Figure 4F, 4G). Furthermore, we found that the PRO-EVH2 domains are required together to bind NCK-1 (data not shown). We also showed that all three NCK-1 SH3 domains were able to bind UNC-34 (Figure 4G).
While nck-1 and vab-1 animals have overextended PLM axons, unc-34 animals have the opposite phenotype and have PLM axons that terminate prematurely (Figure 1D, Figure 2C). This suggests that UNC-34 is involved in PLM axon extension, and reflects a known role of Ena/VASP in actin filament formation and elongation [16], [29]. To understand the genetic nature of the interaction between nck-1 and unc-34, we analyzed the nck-1(ok694); unc-34(e566) double and found that nck-1 partially suppressed the unc-34 PLM termination defect, while unc-34 did not suppress the nck-1 overgrowth (Figure 1D, and data not shown). This suggests that, in PLM axon outgrowth, unc-34 may negatively regulate nck-1. To provide further evidence for this genetic interaction we over expressed NCK-1 (mec-4::nck-1) in the PLM neurons of unc-34(e566) animals and this resulted in a synergistic enhancement of the unc-34 PLM termination phenotype (Figure 1D). Although we cannot conclusively rule out that nck-1 inhibits unc-34, overall, our results suggest that UNC-34 can inhibit the function of NCK-1 and may do so by physically binding to it.
Since UNC-34 and NCK-1 physically interact, we wanted to examine whether VAB-1, NCK-1 and UNC-34 could form a complex in vitro. Surprisingly, although UNC-34 can bind strongly to NCK-1, the introduction of VAB-1 abolished the binding between UNC-34 and NCK-1 (Figure 5A Lane 4, 5). This result suggests that VAB-1 might be inducing its effect at the growth cone membrane by relieving the inhibition of NCK-1 that is caused by UNC-34. To provide in vivo support of this we over expressed UNC-34 in the mechanosensory neurons (mec-4::unc-34) and it significantly reduced the MYR-VAB-1 PLM premature termination phenotype (Figure 1B).
To gain more insight into the interaction between VAB-1 and UNC-34, we sought to analyze the effect of the vab-1;unc-34 double on PLM axons. We found that the vab-1;unc-34 double mutant is synthetic lethal (data not shown), so we used a mechanosensory specific unc-34 RNAi approach (see experimental procedures). The unc-34(RNAi) strain had PLM termination defects that were similar to unc-34(e566) (Figure 1D). Analysis of the vab-1;unc-34(RNAi) double showed that reducing the levels of unc-34 can rescue the PLM overextension defects seen in vab-1(dx31) (Figure 1C), which is consistent with vab-1 inhibiting unc-34 function. Since the genetic data suggested that vab-1 negatively regulates unc-34, we questioned if the activation of VAB-1 could affect the expression and/or localization of UNC-34. Induction of MYR-VAB-1 via heat shock promoter did not change the localization of UNC-34, but instead resulted in the reduction of UNC-34::GFP levels compared to wild-type animals (Figure 5B). To test whether VAB-1's negative regulation can function cell autonomously in the PLMs, we expressed UNC-34::GFP only in the mechanosensory neurons (via mec-4 promoter) and it is expressed at high levels. When we introduce constitutively active VAB-1 only in the touch neurons (mec-4::myr-vab-1) it reduced the UNC-34::GFP levels significantly (Figure 5C).
In summary, our binding assays and genetic analyses show that VAB-1 activation results in binding NCK-1 which in turn blocks the UNC-34 binding to NCK-1, freeing NCK-1 from the negative influence of UNC-34 and in addition VAB-1 negatively regulates UNC-34 protein levels.
Since mammalian Nck is known to physically bind and activate N-WASP to regulate actin filaments through the Arp2/3 complex [8], [30], [31], we questioned whether VAB-1 is linked to the cytoskeleton through WSP-1/N-WASP. If WSP-1 acts downstream of VAB-1, then the wsp-1 mutants should suppress the PLM termination defect caused by MYR-VAB-1. Two wsp-1 alleles are predicted to affect the WSP-1 protein. The wsp-1(tm2299) is not well characterized, but is homozygous lethal and is predicted to be a null allele. The embryonic lethality is due to wsp-1 pleiotropy as WSP-1 is also required for cytokinesis during embryogenesis [14]. The wsp-1(gm324) allele is a well characterized deletion that removes exons 2 and 3, furthermore, no WSP-1 protein nor mRNA can be detected, therefore wsp-1(gm324) is a strong loss-of-function allele [14]. wsp-1(gm324) displays some embryonic and larval lethality but can be maintained as a homozygote [14], [15], [32]. We chose to use the wsp-1(gm324) allele as it allowed us to bypass the embryonic lethality associated with the wsp-1 null allele. We found that wsp-1(gm324) could significantly suppress the MYR-VAB-1 PLM termination defect in young adults and L1s (Figure 1B, Figure 2C).
If WSP-1 is an effector of VAB-1 signaling then we would expect to see neuronal defects similar to vab-1 animals. It was previously reported that the wsp-1(gm324) had weak axon guidance defects, such as in the PDE and VD/DD neurons [15]. We report here that approximately 50% of wsp-1(gm324) animals have overextended PLM defects in young adults, and 42% PLM axon overgrowth in L1s (Figure 1C, Figure 2B). Since the wsp-1 PLM overextension frequency is much greater than vab-1 (Figure 1C), it implies that WSP-1 also functions independent of VAB-1 for PLM axon guidance. We also found that the vab-1(dx31);wsp-1(gm324) double mutants are synthetic lethal (data not shown), which is consistent with WSP-1 functioning in parallel pathways with VAB-1.
The presence of WSP-1 in the VAB-1 signaling pathway suggests the possibility that the PLM termination phenotype caused by MYR-VAB-1 could be due to the activation of the Arp2/3 complex. WSP-1, like its mammalian counterpart, is composed of an N-terminal Ena/VASP homology I domain (EVH1; also known as WASP-homology-1 domain (WH1)), a central section containing a basic region (BR), a GTPase binding domain (GBD) and a proline-rich region (PRO), and a C-terminal with two verprolin homology domains (V; also known as WH2), a cofilin homology domain (C) and an acidic domain (A) [13], [14], [32] collectively known as the VCA region. The C-terminal VCA regions of both WSP-1 and N-WASP have been shown to be sufficient for activating the Arp2/3 complex in vitro [32], [33]. We utilized the C-terminal VCA region of WSP-1 to selectively activate the Arp2/3 complex in the mechanosensory neurons (mec-4::wsp-1vca). The WSP-1VCA caused PLM premature termination defects that were very similar to MYR-VAB-1 (Figure 1D).
The activation of high levels of the Arp2/3 complex produces extensive short branched actin networks that prevent the formation of filopodia, and hence can inhibit axon extension [34], [35]. Ena/VASP, on the other hand, promotes axon extension through filopodia formation and elongation [16], [17], [20], [29]. Thus, activation of Arp2/3 complex and UNC-34/Ena have opposite roles in the axon growth cone, and perhaps Arp2/3 complex activation can antagonize the function of UNC-34/Ena. Since WSP-1/N-WASP is an activator of the Arp2/3 complex, we wanted to test genetically if wsp-1 can antagonize unc-34 function. Due to the synthetic lethality of wsp-1; unc-34 double mutants [14], [36], we analyzed the PLM axons in wsp-1; unc-34(RNAi) animals. Tissue specific unc-34 RNAi resulted in the partial suppression of PLM overextension defects caused by wsp-1(gm324) (Figure 1C), consistent with WSP-1/Arp2/3 activity antagonizing UNC-34 function.
In summary, we show that WSP-1 functions in PLM axon termination, through various signaling pathways, including the VAB-1 Eph RTK. Our results suggest that MYR-VAB-1 is exerting its effect by activating the Arp2/3 complex through WSP-1. We also suggest that WSP-1 can antagonize UNC-34 function by activating the Arp2/3 complex.
We used in vitro binding assays to ask whether VAB-1, NCK-1 and WSP-1 could form a complex. WSP-1 was able to bind NCK-1 (Figure 6A, Lane 6), but not VAB-1 (Figure 6A, Lane 5). However, WSP-1 was able to pull down VAB-1 in the presence of NCK-1, indicating that a VAB-1/NCK-1/WSP-1 complex can occur (Figure 6A, Lane 7).
Since NCK-1 is able to bind both UNC-34 and WSP-1, we wanted to determine whether all three molecules can form a complex, or do UNC-34 and WSP-1 compete for NCK-1 binding. We first confirmed that WSP-1 was unable to bind UNC-34 (Figure 6B, Lane 6). We found that although WSP-1 binds NCK-1, the presence of UNC-34 resulted in a 70% reduction of the NCK-1/WSP-1 complex (Figure 6B, Lane 8). This shows that UNC-34 can effectively compete with WSP-1 for NCK-1 binding. Furthermore we could not detect NCK-1/UNC-34/WSP-1 in a complex (Figure 6B, Lane 8). Interestingly, adding VAB-1 to the binding interaction increased the level of NCK-1 binding to WSP-1, indicating that VAB-1 eliminated UNC-34's ability to compete for NCK-1 binding (Figure 6B, Lane 9). In summary, our binding assays show that VAB-1, NCK-1 and WSP-1 form a complex, that UNC-34 competes with WSP-1 for NCK-1 binding, and that VAB-1 enables WSP-1 to outcompete UNC-34 for binding to NCK-1.
The VAB-1 RTK effectors NCK-1 and WSP-1 are known actin regulators and therefore implicate VAB-1 signaling in regulating actin cytoskeleton for axon guidance. To confirm this, we monitored the PLM growth cone of wild-type and myr-vab-1 transgenic animals at the time of hatching. In wild-type animals, most of the PLM growth cones exhibited dynamic changes and had many filopodia protrusions (70%; N = 20 movies) (Figure 7A, Video S1). Transgenic myr-vab-1 animals, on the other hand, had growth cones that were less dynamic and were usually void of filopodia like structures with only 25% (N = 16 movies) showing some filopodia structures (Figure 7B, Video S2). Since our molecular and genetic data suggest that VAB-1 inhibits UNC-34/Ena function we also observed the growth cones of unc-34(e566) animals. We found that unc-34(e566) mutants, like myr-vab-1 animals, had growth cones void of filopodia structures with only 25% displaying filopodia structures (N = 12 movies; not shown). Our results show that activated VAB-1 can affect the PLM growth cone morphology by inhibiting filopodia formation.
We previously reported a functional role for VAB-1 as a receptor for a repellent or stop signal in PLM axon guidance [6]. Here we describe some of the molecular events involved in VAB-1 signaling that allow the regulation of actin dynamics for PLM axon guidance. Our genetic and in vitro interaction analysis identified NCK-1, WSP-1 and UNC-34 as molecules regulated by VAB-1 Eph RTK signaling. Our data supports a model in which VAB-1 suppresses axon extension by negatively regulating UNC-34, and activating the Arp2/3 complex through a VAB-1/NCK-1/WSP-1 complex. Furthermore, using time-lapse analysis we show that activation of VAB-1 inhibits filopodia formation in the PLM growth cone.
Our results show that the C. elegans NCK-1 adaptor protein is an effector of the VAB-1 RTK signal in vivo. Several lines of evidence indicate that VAB-1 and NCK-1 act together to regulate axon guidance. First, nck-1 and vab-1 animals have similar neuronal defects. Second, NCK-1 and VAB-1 physically interact and co-localize in similar neuronal cells and axons. Finally, the nck-1 loss-of-function suppresses the defects caused by the constitutively active VAB-1. We found that NCK-1 binds the VAB-1 juxtamembrane tyrosine Y673 (YEDP) via its SH2 domain in a VAB-1 kinase dependent manner. This is consistent with the published binding specificity of the Nck SH2 domain, as well as reports of Nck1 binding to the second juxtamembrane tyrosine residue (YEDP) in EphA3 (Y602) and EphA2 (Y594) [37], [38], [39]. Interestingly, Nck adaptors have been reported to function downstream of Eph RTKs but it appears that the activated EphA RTKs are direct targets of Nck adaptors [38], [39], [40], [41], whereas Nck may indirectly interact with EphBs [11], [42], [43]. Considering that the intracellular region of VAB-1 is more similar to EphA receptors [4], our results in C. elegans provides relevant insight into how mammalian EphA receptors could regulate the actin cytoskeleton for axon guidance.
The Ena/VASP protein family is required in processes that involve dynamic actin remodeling such as platelet shape change, axon guidance and Jurkat T cell polarization [44]. The ability of Ena/VASP proteins to remodel actin stems from their ability to polymerize actin, which is required for filopodia formation and elongation [16], [45], [46]. In C. elegans, UNC-34/Ena functions in neuronal cell migration, axon guidance and epithelial filopodia formation [14], [18], [19], [20]. Our results further confirm the role of UNC-34 in axon extension, because we show that the unc-34(e566) PLM axons terminated prematurely. The cause of early termination is likely due to a reduction of filopodia elongation in the growth cone, resulting in the persistence of more densely branched filaments that can slow axon migration. This is supported by the finding that unc-34 mutants have fewer filopodia structures on growth cones, and a reduced rate of growth cone migration [20] (this work and our unpublished observations). In addition, mammalian studies show that depletion of Ena/VASP generates shorter and more densely branched filaments [47].
We propose that VAB-1 negatively regulates UNC-34 for PLM termination. This is supported by our observations that: 1) the loss-of-function unc-34 resulted in PLM axon defects similar to the hyperactive MYR-VAB-1; 2) over expressing UNC-34 in the PLM partially suppressed the MYR-VAB-1 phenotype; 3) tissue specific unc-34 RNAi suppressed the vab-1 PLM overextension defects; and 4) over expressing VAB-1 reduced the UNC-34 protein levels. Although we do not know the mechanism of the reduction of the UNC-34 protein levels displayed in the hyperactive VAB-1, it is possible that UNC-34, when removed from its adaptor NCK-1, is more prone to degradation. In this case NCK-1 may play a dual role and may also promote UNC-34 function as well. It is also likely VAB-1 signaling could affect the unc-34 transcriptional level. Future experiments should resolve how VAB-1 regulates UNC-34 protein levels.
Our finding that VAB-1 negatively regulates UNC-34/Ena is different from a previous report that shows mammalian EphB4 as an activator of Ena/VASP [24]. In fibroblast cells, the EphB4 receptor is thought to activate Ena/VASP to destabilize lamellipodia during cell repulsion and likely does so by promoting elongated actin filaments rather than a branched actin filament network. Although the Eph receptor signal transduction to Ena or UNC-34 is opposite (activates vs. inhibits) the role for UNC-34/Ena is conserved, because in both cases UNC-34 or Ena/VASP promotes actin filament elongation.
Our results provide evidence that VAB-1/Eph RTK can regulate the actin cytoskeleton through its interaction with NCK-1 and WSP-1. This is based on our observation that vab-1, nck-1 and wsp-1 mutants share the same phenotype of PLM axon overextension, that both nck-1 and wsp-1 were able to partially suppress the MYR-VAB-1 PLM termination defect, that VAB-1, NCK-1 and WSP-1 are able to form a complex in vitro, and that the activation of the Arp2/3 complex via the WSP-1 VCA domain resulted in PLM termination defects similar to MYR-VAB-1. The role of N-WASP as a negative regulator of axon elongation has been shown by two separate reports, where the reduction of N-WASP resulted in the enhancement of axon elongation [48], [49]. This phenotype is reminiscent of the PLM overextension defects we observed in wsp-1 animals. There have been conflicting reports on the role of the Arp2/3 complex in axon elongation, where some reports suggest that the Arp2/3 complex acts as a negative regulator of axon elongation [34], [49], while other reports show that the Arp2/3 complex is required for axon elongation [20], [50]. A paper by Ideses et al. (2008) provided a potential resolution to this paradox by looking at the characteristics of actin assembly in the presence of variable amounts of Arp2/3 complex in vitro [35]. It is proposed that high levels of the Arp2/3 complex prevent the formation of filopodia bundles by promoting the extensive branching networks of actin with short tips. On the other hand, at low concentrations of Arp2/3 the actin filaments have longer tips and are further apart making it easier to form filopodia bundles [35]. Therefore, it would be expected that the complete elimination of Arp2/3 would prevent any neurite elongation. Similarly, the excessive activation of Arp2/3 would also prevent neurite elongation due to the increased levels of short, branched networks of actin filaments. In the C. elegans epithelial cells unc-34 and wsp-1 function redundantly for epithelial cell migrations [14]. However our results in PLM neurons suggest that WSP-1 and UNC-34 have opposite roles. Why the apparent paradox? This is reminiscent of what has been observed for Ena/VASP proteins where some reports suggest Ena/VASP promotes actin dependent processes while others suggest Ena/VASP may inhibit actin dependent processes [51]. While the growth cones on axons and the leading edge of epithelial cells both require actin for movement, they might not be identical in the way the cell moves forward. Proteins such as Ena/VASP, N-WASP, and Arp2/3 are thought to promote actin polymerization, however these proteins also change the geometry of the actin filament network in addition to promoting actin assembly. Therefore the overall effects of such changes in the actin network may not be easy to predict with respect to cell movement since various concentrations of these actin regulators could lead to activation or inhibition of filopodia. Since WSP-1/N-WASP is an activator of the Arp2/3 complex and different levels of Arp2/3 can elicit different behaviors, WSP-1 may also have opposite effects depending on its level of activity. In addition, while most of our results are based on the PLM neurons it is very likely the roles of UNC-34 and WSP-1 and how they are regulated will be different in other neurons.
N-WASP has been shown to interact in a complex with the mammalian EphB2, through the adaptor molecule intersectin [52]. Furthermore, this complex of EphB2, intersectin and N-WASP is required for dendritic spine formation, which consists mainly of a meshwork of branched filaments caused by the activation of the Arp2/3 complex [52]. C. elegans intersectin (ITSN-1) is expressed in the nervous system, and it is enriched in presynaptic regions and has roles in neurotransmission [53]. Future work will determine whether the VAB-1/Eph interacts with ITSN-1 to connect WSP-1. Our current work shows that the VAB-1 Eph RTK can signal through WSP-1/N-WASP through a different adaptor molecule, NCK-1, and we propose, like the mammalian intersectin adaptor, this complex activates Arp2/3 to promote branched actin.
We propose a model of how the proteins VAB-1, NCK-1, UNC-34, WSP-1 and Arp2/3 function in axon growth cones for extension and termination (Figure 7C). During PLM axon outgrowth, the growth cone is stimulated by an attractive cue that results in the accumulation of UNC-34/Ena at the growth cone. The result is a net forward movement due to the role of UNC-34/Ena in inhibiting actin capping proteins, and allowing filopodia elongation by polymerizing F-actin at the leading edge. In addition, UNC-34/Ena binds to the NCK-1 SH3 domains to prevent it from interacting with WSP-1 and participating in a signaling pathway(s) that would otherwise inhibit axon extension. It is also possible that the UNC-34/NCK-1 heterodimer could function together for actin polymerization or that NCK-1 binding could stabilize the UNC-34 protein. In this case NCK-1 acts positively with UNC-34. However, since unc-34 and nck-1 mutants have opposite PLM axon phenotypes, it suggests that nck-1's role in axon outgrowth is more dispensable or redundant than its role in axon termination. Once the VAB-1/Eph RTK receives the signal to inhibit axon extension, VAB-1 is autophosphorylated and provides a docking site (Y673) for NCK-1. The NCK-1-SH2 domain binds the activated VAB-1 receptor and this disrupts the interaction between NCK-1 and UNC-34 to release the inhibitory effect of UNC-34 on NCK-1. Through an unknown mechanism, we also show that VAB-1 negatively regulates the UNC-34/Ena protein levels. VAB-1/NCK-1 can now recruit and activate WSP-1 and all three proteins form a complex, which results in high levels of Arp2/3 activation, ultimately leading to a branched meshwork of actin filaments. The combined actions of VAB-1/Eph blocking UNC-34/Ena activity, while activating Arp2/3 through NCK-1/WSP-1 contributes to the molecular events required to stop the growth cone forward movement.
All C. elegans strains were manipulated as described by Brenner [54]. All alleles were isolated in the standard wild type Bristol strain N2. All experiments were performed at 20°C unless otherwise indicated. The following strains were used in this study: N2 (var. Bristol) [54]; LGI: zdIs5[mec-4::gfp]; LGII: vab-1(dx31), quIs5[mec-4::myr-vab-1]; LG IV: wsp-1(gm324), LG V: unc-34(e566); LGX:, quIs6[unc-34::unc-34::gfp]; Unmapped: quIs16[hs::myr-vab-1] [55]; Extrachromosomal arrays (this study): quEx131[mec-4::nck-1A], quEx190[nck-1::nck-1A-gfp] [12], quEx215[mec-4::unc-34::gfp], quEx281[mec-4::unc-34], quEx283 [mec-4::nck-1A::gfp] [12], quEx321[mec-4::vca], quEx338[mec-4::unc-34 RNAi] (see tissue specific RNAi). Unless noted otherwise, all C. elegans strains were obtained from the C. elegans Genetics Stock Center, (U. of Minnesota).
To produce double stranded RNA (dsRNA) only in the mechanosensory neurons, we constructed a cloning vector (pIC659) with head to head Pmec-4 promoters on each side of a Multiple Cloning Site (MCS) such that the sense and antisense strands of an inserted cDNA would be transcribed. The mec-4::unc-34 RNAi construct (pIC727) was created by cloning an unc-34 cDNA fragment (ATG start to the first SalI site, 388 bp) into the pIC659 dual Pmec-4 RNAi cloning vector.
The mec-4::nck-1A construct (pIC313) was previously described in Mohamed and Chin-Sang (2011). The mec-4::unc-34 construct (pIC624) was generated by amplifying unc-34 cDNA and sub-cloning behind the mec-4 promoter. The same procedure was used to make the mec-4::unc-34::gfp (pIC540) construct, but unc-34 was fused to gfp amplified from pPD95.75. To create the mec-4::vca construct (pIC673), the VCA region of WSP-1 (9108–9606 of the wsp-1 gene; C07G1.4a in Wormbase) was amplified from genomic DNA and cloned behind the mec-4 promoter. The unc-34::unc-34::gfp translation reporter was generated by a PCR fusion approach [56] using the following pieces: 1. A ∼5 kb genomic region that includes 2 kb of 5′UTR and the first two exons of unc-34, 2. Exons 2–7 were amplified from RB2 cDNA library, and 3. a 868 bp GFP fragment amplified from pPD95.75 (gift from Dr. Andrew Fire). The expression of the UNC-34::GFP rescued the unc-34(e566) uncoordinated phenotype. Details of plasmid/PCR constructs and primer sequences are available upon request.
Transgenic animals were generated by germ-line transformation as previously described [57]. The unc-34::unc-34::gfp translational reporter was injected at a concentration of 20 ng/µL, and one of the unc-34 rescuing lines (quEx61) was integrated to create quIs6. The mec-4::unc-34 construct was injected at a concentration of 30 ng/µL into mec-4::gfp(zdIs5); mec-4::myr-vab-1(quIs5). The mec-4::unc-34::gfp construct was injected at a concentration of 30 ng/µL into N2. mec-4::unc-34RNAi, mec-4::vca and mec-4::nck-1 were injected into mec-4::gfp(zdIs5) at 30 ng/µL. mec-4::nck-1(quEx131) was later crossed into unc-34(e566), and mec-4::unc-34RNAi (quEx338) was crossed into vab-1(dx31) and wsp-1(gm324). Transgenic animals were identified by the co-injection marker pRF4/rol-6 (30 ng/µl), or odr-1::rfp (30 ng/µl) [57]. At least two independent lines were isolated and analyzed. The data shown are from one representative line.
Mixed stage animals were fixed and stained as described in Chin-Sang et al. (1999) [58]. Rabbit anti-VAB-1 antibodies (antigen VAB-1-HIS6) and chicken or mouse polyclonal antibodies against GFP (Chemicon) were used at 1∶100 dilutions. Texas Red-conjugated goat anti-rabbit and FITC conjugated goat anti-chicken or anti-mouse secondary anti bodies (Jackson's lab) were used at a 1∶100 dilution. For Western blot analysis, antibodies were used at the following dilutions: anti-NCK-1 at 1∶500, anti-VAB-1 at 1∶2500, anti-MBP-HRP at 1∶8000, anti-GST-HRP at 1∶4000 and 4G10 (Upstate Inc.) at 1∶2500. Goat-anti-rabbit-HRP and goat-anti-mouse-HRP were used as at 1∶10000 dilutions on western blots. Relative band intensities in Figure 6B were quantified using at least two independent blots and analyzed using the National Institutes of Health Image J program.
The mechanosensory neurons were visualized using the mec-4::gfp (zdIs5) reporter. Young adult animals were scored as having PLM axon overextension or premature termination as described previously [6]. Outgrowth of the PLM axon happens during embryogenesis and continues to grow after hatching and most of its growth happens at the L1 stage. From L2 onwards to adulthood PLM growth is maintained relative to its termination point along the body [28]. To measure the L1 PLM axons, newly hatched L1s were synchronized in the absence of food for up to 12 hours. We found that although the worms were born in the absence of food that the PLM was still able to grow and the PLM axon lengths were equivalent to the length of animals developing for 2–3 hours post hatching. This corresponds to the Phase 1 or fast growth PLM growth phase [28]. Our wild-type reference strain (zdIs5) had L1 PLMs with an average PLM length of 108.5 (±5.5) microns with a PLM length/total body length (from head to tail) ratio of 0.48 (±0.04). L1 PLM axons were scored as overgrown if they were longer than 114 µm and had a PLM/total body length ratio of greater than 0.52. L1 PLM axons were scored as under grown if the PLMs were shorter than 103 µm and had a PLM length/total body length ratio of length less than 0.44. The L1 PLM axons were traced from photograph and measured in NIH Image J software. The wild-type neuron morphology was defined by analysis of neuronal GFP reporters and is consistent with the electron microscopic reconstruction of the C. elegans nervous system [59]. Animals were anesthetized using 0.2% tricaine and 0.02% tetramisole in M9, and mounted on 3% agarose pads. Unless stated otherwise, fluorescent animals and images were analyzed using a Zeiss Axioplan microscope, Axiocam and Axiovision software.
PLM growth cones were visualized using a mec-4::gfp (zdIs5) reporter. Eggs were allowed to hatch for 5 minutes, and the newly hatched L1 animals were examined immediately on 3% agarose pads with a drop of 0.2% tricaine and 0.02% tetramisole in M9. PLM growth cones were imaged with a Zeiss LSM710 confocal microscope at intervals of 20–30 s. Axons were scored positive for filopodia if time-lapse movies revealed at least 2 protrusions, and there were dynamic movements (eg. growth and collapse) of the these structures within the 10–15 minutes of filming. See Videos S1 and S2 for examples.
Yeast cells were grown on standard and selective media as required [60]. The desired plasmids were transformed into yeast cells using the lithium acetate method [61]. For binding and deletion analysis, the pGBKT7 vector was used as bait and the pGADT7 vector (Clontech) as prey, and β-galactosidase activity was measured qualitatively by X-GAL overlay assays [62]. To identify interactions with VAB-1, the Kinase Region (669 aa-985 aa) of vab-1 was cloned into pGBKT7 (pIC187) and used in a screen against the RB2 cDNA library (gift from Dr. R. Barstead), and about 600,000 colonies were screened and two independent nck-1 cDNA clones were isolated. Site directed mutagenesis (QuickChange, Stratagene) of pIC187 was used to change the juxtamembrane tyrosine 673 changed to glutamic acid (Y673E). The SH2 domains of NCK-1, MIG-10, SEM-5, ABL-1 and VAV-1 were cloned into the activation domain of the pGADT7 vector. Primer sequences and details of plasmid constructs are available upon request.
The following constructs were created by cloning the desired cDNA fragment into Glutathione-S-Transferase (pGEX4T-2, Amersham): pIC282 – NCK-1 SH2 domain (298 aa–397 aa), pIC297 – all three NCK-1 SH3 domains (1 aa–308 aa), pIC308 – 1st NCK-1 SH3 domain (1 aa–72 aa), pIC593 – 2nd NCK-1 SH3 domain (112 aa–186 aa), pIC309 – 3rd NCK-1 SH3 domain (198 aa–308 aa), pIC324 – full length (F.L.) NCK-1 (1 aa–397 aa), and pIC606 – F.L. UNC-34 (1 aa–454). The following constructs were created by cloning the desired cDNA fragment into Maltose Binding Protein (pMALtm-p2X, New England Biolabs): pIC225 – F. L. intracellular region of wild-type VAB-1 (581 aa–1117 aa), pIC119 – F. L. intracellular kinase deficient VAB-1 (G912E), pIC603 – UNC-34 RPO-EVH2 domain (128 aa–454 aa), pIC605 – F.L. UNC-34 (1 aa–454 aa), pIC671 – UNC-34 PRO domain (128 aa–274), pIC674 – UNC-34 EVH2 domain (246 aa–454 aa), pIC670 – WSP-1 VCA domain (334 aa–607 aa). pIC582 – His-6::VAB-1 (581 aa–1117 aa) was described in Brisbin et al (2009). All fusion constructs were expressed in E. coli Tuner (DE3). For Figure 4B, 4C, 4F, 4G and Figure 5A, a GST ‘pull-down’ assay was used to confirm the VAB-1, NCK-1 and UNC-34 interactions. Soluble/purified (Load) MBP-VAB-1, MBP-VAB-1(G912E), MBP-UNC-34 F.L., MBP-UNC-34-PRO-EVH2, MBP-UNC-34-PRO or MBP-UNC-34-EVH2 were incubated for 2–3 hrs at 4°C with soluble extracts containing either GST, GST-NCK-1 F.L., GST-NCK-1-all SH3 domains, GST-NCK-1(1stSH3), GST-NCK-1(2ndSH3), GST-NCK-1(3rdSH3), GST-NCK-1(SH2), His-6::VAB-1(581 aa–1117 aa) (pIC582) or GST-NCK-1 F.L. coexpressed with pIC582 bound to 50 µl glutathione sepharose beads (GE healthcare). Unbound fractions were collected, protein bound to GST beads were washed four times (25 mM Hepes, 10% Glycerol, 0.1% Triton-X, 285 mM NaCl), and a proportional loading of each sample was analyzed by standard SDS polyacrlyamide gel, followed by western blotting. All loads fused to MBP were detected using anti-MBP conjugated to HRP (New England Biolabs). His6-VAB-1 was detected using Rabbit anti-VAB-1 antibodies (antigen VAB-1-His6) (Figure 5A). GST and GST-NCK-1 F.L., and GST-NCK-1 deletion domains were detected either by Ponceau S or anti-GST conjugated to HRP. For Figure 6, MBP ‘pull-down’ was used to confirm VAB-1, NCK-1, WSP-1 and UNC-34 interactions. Soluble extracts (Load) of GST-NCK-1 F.L., His-VAB-1 (pIC582), GST-NCK-1 F. L. coexpressed with pIC582, or GST-UNC-34 F.L. were incubated for 2–3 hours at 4°C with soluble extracts containing either MBP or MBP-WSP-1(334 aa–608 aa) bound to 100 µl amylose resin beads (New England Biolabs). Unbound fractions were collected, protein bound to amylose beads were washed four times (20 mM Tris-Cl [pH 7.5], 200 mM NaCl, 1 mM EDTA, 1 mM DTT), and a proportional loading of each sample was analyzed by standard SDS polyacrylamide gel, followed by western blotting. VAB-1 was detected by Rabbit anti-VAB-1, GST fused proteins were detected by anti-GST conjugated to HRP, MBP and MBP-WSP were detected by anti-MBP conjugated to HRP.
|
10.1371/journal.pntd.0007541 | Dosing pole recommendations for lymphatic filariasis elimination: A height-weight quantile regression modeling approach | The World Health Organization (WHO) currently recommends height or age-based dosing as alternatives to weight-based dosing for mass drug administration lymphatic filariasis (LF) elimination programs. The goals of our study were to compare these alternative dosing strategies to weight-based dosing and to develop and evaluate new height-based dosing pole scenarios.
Age, height and weight data were collected from >26,000 individuals in five countries during a cluster randomized LF clinical trial. Weight-based dosing for diethylcarbamazine (DEC; 6 mg/kg) and ivermectin (IVM; 200 ug/kg) with tablet numbers derived from a table of weight intervals was treated as the “gold standard” for this study. Following WHO recommended age-based dosing of DEC and height-based dosing of IVM would have resulted in 32% and 27% of individuals receiving treatment doses below those recommended by weight-based dosing for DEC and IVM, respectively. Underdosing would have been especially common in adult males, who tend to have the highest LF prevalence in many endemic areas. We used a 3-step modeling approach to develop and evaluate new dosing pole cutoffs. First, we analyzed the clinical trial data using quantile regression to predict weight from height. We then used weight predictions to develop new dosing pole cutoff values. Finally, we compared different dosing pole cutoffs and age and height-based WHO dosing recommendations to weight-based dosing. We considered hundreds of scenarios including country- and sex-specific dosing poles. A simple dosing pole with a 6-tablet maximum for both DEC and IVM reduced the underdosing rate by 30% and 21%, respectively, and was nearly as effective as more complex pole combinations for reducing underdosing.
Using a novel modeling approach, we developed a simple dosing pole that would markedly reduce underdosing for DEC and IVM in MDA programs compared to current WHO recommended height or age-based dosing.
| Lymphatic filariasis is a debilitating parasitic disease that affects over 65 million people globally. To eliminate this disease the World Health Organization (WHO) recommends treating entire communities with drug combinations that include ivermectin and diethylcarbamazine. Ideally, the amount of drug administered is determined by weight; however, obtaining accurate weight measurements in remote, resource limited areas is oftentimes not feasible. The alternatives currently recommended by the WHO based on height (ivermectin) and age (diethylcarbamazine) have maximal doses below that recommended by weight-based dosing. In this study we use statistical models, based on data from a large (5-country, >26,000 individuals) lymphatic filariasis clinical trial, to develop model-based dosing poles and compare dosing based on our dosing models to WHO recommendations. Our results showed that the WHO methods would have resulted in 32% (diethylcarbamazine) and 27% (ivermectin) of individuals in the clinical trial dataset to receive below the recommended weight-based dosage. Dosing poles based on our statistical models showed that our dosing pole would markedly reduce underdosing with 2% and 6% receiving below the recommended dosage for diethylcarbamazine and ivermectin, respectively. The dosing pole we propose has the potential to dramatically improve dosing and facilitate the elimination of lymphatic filariasis globally.
| Lymphatic filariasis (LF) is a disabling mosquito-borne parasitic disease that affects some 68 million people globally [1]; the WHO estimated more than 880 million people in 51 countries remained at risk for LF in 2017 [2]. In the year 2000, the World Health Organization (WHO) implemented a strategic plan (Global Programme to Eliminate Lymphatic Filariasis [GPELF]) to eliminate LF as a public health problem by 2020 [3, 4]. As part of GPELF, the WHO recommends using two-drug treatment combinations (diethylcarbamazine [DEC] + albendazole or ivermectin [IVM] + albendazole) in mass drug administration (MDA) programs that dose both “at risk” and infected individuals in LF-endemic areas [3, 4]. Recent studies have shown that a triple-drug treatment combination (IVM + DEC + albendazole, IDA) is more effective [5] and as safe [6] as a standard two-drug LF treatment (DEC + albendazole, DA). This has resulted in an updated WHO policy to include IDA as an option for LF treatment in certain settings [7] which will help to more rapidly achieve the GPELF goal of eliminating LF as a global public health problem.
Key components of GPELF include the use of MDA to deliver treatment to infected people and to reduce parasite transmission by reducing the reservoir of parasites required for mosquito transmission. The WHO recommends weight-based dosing for IVM and DEC [3, 4]. However, this is often not possible in remote, resource-limited areas. Consequently, when weight-based dosing is not feasible, the WHO has recommended alternatives such as the use of height-based dosing poles for IVM and age-based dosing for DEC. The current height-based IVM dosing pole recommended by the WHO was developed in the early 1990s [8–10]; it was originally employed in Nigerian communities based on 150 μg/kg and has a dosing range of 3 to 12 mg for four different height groupings (90 to 119 cm, 120 to 140 cm, 141 to 158 cm, and > 158 cm). The WHO age-based dosing for DEC ranges from 100 to 300 mg across three different age groupings (2 to 5 years, 6 to 15 years, > 15 years) [9]. Although height and age-based dosing are recommended by the WHO for LF MDA programs, the maximum recommended doses using these methods are lower than those indicated by weight-based dosing (6 mg/kg for DEC and 200 μg/kg for IVM). Furthermore, having separate dosing methods for DEC and IVM increases the complexity and challenge of implementing the new 3-drug treatment for LF MDAs, and an alternative based on a single height pole for both DEC and IVM would increase the efficiency and feasibility of administering this new treatment option.
The primary objectives of our study were to: (1) compare the WHO’s height-(IVM) and age-(DEC) based dosing recommendations for MDA programs to gold-standard weight-based dosing with data from a variety of LF endemic areas; (2) to use field data to develop and evaluate alternative height-based dosing poles for IVM and DEC; and (3) determine whether a single height-based dosing pole can be used to administer both IVM and DEC with the aim of reducing underdosing compared to current WHO recommended methods. To address these objectives, we analyzed height and weight data collected as part of a large LF drug-safety clinical trial [6].
The study protocols were reviewed and approved by independent Federal-Wide Assurance (FWA) registered ethical review boards in each country and at institutions of research partners who participated in the studies. The de-identified data used in this study were limited to gender, height, weight, age and country.
The data were collected during a community-based safety study of MDA for LF that enrolled more than 26,000 participants in 5 countries (Haiti, India, Indonesia, Fiji, and Papua New Guinea). A 21 CFR Part 11 compliant electronic data capture system allowed deidentified data to be entered directly into a hand-held tablet via a mobile data management solution (‘App’) called CliniTrial (CliniOps, Fremont, CA). Data were synchronized regularly over the internet from all study sites through a secured Amazon Virtual Private Cloud server and compiled into one complete dataset. Validation checks and automated alert checks were programmed into the electronic data capture system to maintain a high level of data quality at the points of entry. The current study included participants that met the inclusion/exclusion criteria and received treatment (IDA or DA) in the LF clinical trial [6], and were ≥ 90 cm in height. Age, sex, country, height and weight were the primary variables of interest in the present study.
Our modeling process involved three steps (Fig 1). The first step was to use quantile regression to predict weight based on height. Because of the myriad factors that influence weight as a person ages, weight and height become increasingly decoupled as individuals age. Quantile regression is a statistical approach that enables the data to be modeled across a range of quantiles [11, 12]. Quantiles correspond to the proportion of observations below a threshold. For example, in the context of quantile regression, the 0.25 quantile corresponds to the point at which ~25% of the observations fall below the regression line and ~75% of the observations are above the regression line. For our height-weight quantile regressions, the estimates for lower quantiles emphasize the prediction of lighter individuals and estimates for higher quantiles emphasize predictions towards heavier individuals. This approach allowed us to create and evaluate many different dosing pole scenarios (“dosing poles”) and choose the pole(s) that best meets the study objectives. In our quantile-regression models we employed a range of quantiles (0.10 to 0.90 at intervals of 0.05 for a total of 17 quantiles) to ensure that we captured a breadth of relationships between height and weight. In addition to Global models (no stratification of data, 1 model for each of the 17 quantiles), we performed analyses for multiple strata to determine whether predictions improved when data were stratified by Country (5 countries x 17 quantiles), Sex (2 sexes x 17 quantiles), or Country x Sex (5 countries x 2 sexes x 17 quantiles), resulting in a total of 306 quantile-regression models. Weight outcomes were log transformed prior to analysis and model predictions were then back-transformed (antilog [model prediction]) to obtain predicted weights on the original scale of the data. This allowed us to capture the nonlinear association between height and weight. Likelihood-ratio test statistics were used to obtain P-values. P-values < 0.05 were considered significant. All quantile-regression analyses were conducted using PROC QUANTREG in SAS version 9.4 (SAS Institute Inc., Cary, NC).
The second step was to use predicted weights from our step 1 models to create height-interval dosing poles. Our “gold standard” (hereafter recommended dosage) for weight-based dosing is based on the WHO GPELF recommendations and modified from the LF drug-safety clinical trial [6] so that the weight midpoints for the different dosages provided 6 mg/kg for DEC and 200 μg/kg for IVM (Table 1). The quantile-regression weight predictions were converted into tablet numbers using the weight ranges in Table 1. The predicted number of tablets for a given height were then operationalized as “full-dose” DOLF dosing poles.
The third step was to assess how well the DOLF dosing pole predictions (step 2), the WHO age-based DEC dosing [9], and the WHO height-based (dosing pole) IVM dosing [8] corresponded to the recommended weight-based dosing (Table 1) using the observed weights from the LF clinical trial dataset. We also created two “hybrid” dosing poles that combine criteria from the WHO IVM dosing pole and the full-dose DOLF (IVM) dosing pole providing in a single pole that can be used for both IVM and DEC dosing. The hybrid poles are identical to the full-dose DOLF IVM dosing poles, except the hybrid poles included a maximum number of 4 (Hybrid 4) or 6 (Hybrid 6) tablets, and all participants ≥ 90 cm receive at least 1 tablet (participants < 90 cm were excluded from the dataset prior to analysis). Using a single pole for DEC and IVM and administering a small number of tablets that constitutes an adequate dose is desirable from both an implementation standpoint (individuals may be more likely to adhere to treatment with fewer tablets and MDA is simplified by use of a single dosing pole) and from a cost perspective. If the hybrid dosing poles perform similarly to the full-dose poles, then the hybrid poles would be the preferred option. The different dosing options (WHO, and DOLF full-dose and hybrid poles) were assessed by estimating the percentage of subjects that would have received below (BRD), above (ARD), or the recommended weight-based dosage.
Height and weight data from 26,821 individuals were included in this analysis. Sex of participants was balanced within and among countries with the percentage of males ranging from 47% to 53%. There was a 12-year range in median age, with Haiti having the youngest participants (age: median [IQR] = 18 [11, 30] years; 51.6% adults) and Fiji having the oldest participants (age: median [IQR] = 30 [12, 49] years; 62% adults). There was substantial variability in height and weight across the different countries. Participants in Indonesia had the lowest mean height (144.5 cm) and weight (38.7 kg), and participants in Fiji had the greatest mean height (159.8 cm) and weight (69.5 kg) (Table 2). Histograms that provide a graphical summary of the distributions for the variables in Table 1 are included as a supplement (S1 Fig).
Very different dosing recommendations were obtained with different dosing algorithms (Fig 2). The WHO dosing pole for IVM agreed with the weight-based dosing in about 56% of participants; it resulted in below the recommended dosage (BRD) for 27% of the participants and above the recommended dosage (ARD) for 17% of the participants. Both BRD and ARD were more frequent in people with heights > 140 cm. Dosing discrepancies were more frequent with age-based dosing for DEC; it agreed with weight-based dosing in 47% of participants, with ARD and BRD rates of 21% and 32%, respectively. The percentage of participants receiving ARD was greatest for the 6 to 15-year age group and the percentage receiving BRD was greatest for people older than 15 years. When the analysis was restricted to adult (≥ 18 years) males, BRD percentages were 39% for IVM and 54% for DEC.
The quantile regression analyses revealed substantial heterogeneity in model predictions across quantiles. For all 306 quantile-regression models, the slope estimates showed significant positive associations between height and weight, regardless of quantile. Slope estimates and/or intercept estimates generally increased with quantile indicating a greater predicted weight for a given height in higher quantiles (S1 Table). Plots of the model predictions from the Global models (without stratification by country, sex, or age) showed that differences between model predictions increased with height, with the largest differences for the 10th and 90th quantile models for taller participants (Fig 3). For example, at a height of 95 cm the model predictions were relatively close for the 10th and 90th quantiles with predicted weights of 11.6 kg and 15.3 kg, respectively. However, at a height of 180 cm, the 10th quantile model predicted weight of 64.7 kg, while the 90th quantile model predicted a weight of 119.8 kg.
The country-specific models revealed marked variability in height-weight relationships between countries. Fiji and Haiti generally had the highest predicted weights for a given height, and Indonesia had the lowest, with India and PNG falling between these extremes (Fig 4). Weight prediction differences between countries increased both with subject height as well as quantile. Country-specific height-weight slope estimates indicated that for all quantiles Fiji weight predictions had the largest increase in predicted weight with height (S2 Fig).
When we stratified the data by sex, we found that females were heavier for a given height than males, as reflected in both the parameter estimates and the model predictions. Similar to the other strata, the differences between the quantile-regression predictions increased with height, and the predicted weight differences by sex were greater in larger quantiles (Fig 5). Sex-specific slope estimates indicated that females had steeper slopes than males for all quantiles (S3 Fig). Models stratified by both country and sex revealed some variability between sexes within a country, but generally the difference was small relative to inter-country differences (S1 Table).
Results from our weight predictions were used to make dosing poles based on observed subject heights. Compared to the WHO height-based IVM dosing pole and the WHO age-based DEC dosing, the “full dose” DOLF dosing poles had a greater range of dosages (0–6 tablets for DEC and 0–7 tablets for IVM). Results from our Global models (combined data from all countries) revealed that for both DEC and IVM, the minimum height required for a subject to receive any specific number of tablets was consistently lower for the higher quantile models (Fig 6). For example, at the 25th quantile the minimum height to receive a single tablet of IVM was 104 cm, whereas for the 90th quantile the minimum height to receive a single tablet of IVM was only 93 cm. For the WHO IVM dosing pole, the minimum height required to receive a specific dosage (≤ 4 tablets) was lower than the DOLF models for smaller quantiles (0.25 and 0.50) and higher for larger quantiles (0.75 and 0.90) except for “one-tablet”. Results from the hybrid poles that can be used for dosing DEC or IVM with either a 4-tablet maximum (Hybrid 4) or a 6-tablet maximum (Hybrid 6) showed that, in contrast to the full-dose poles, all participants 90 cm or greater would receive at least one tablet. Furthermore, for the highest quantile model (0.90) there was a much wider range of participants that would receive four tablets or more (147–200 cm) as compared to the WHO IVM pole (159–200). Dosing poles varied somewhat according to strata, with the most dramatic changes attributed to Fiji, which was the country with the lowest tablet thresholds of any of the dosing poles (S4 Fig).
Application of the model-based dosing poles to the DOLF dataset revealed a marked improvement in dosing compared to the current WHO recommendations. Using dosing poles from the stratified analyses (Country, Sex, Country x Sex) produced only marginal improvements compared to unstratified Global models with regards to dosing (See S5 Fig). Therefore, we decided to focus on results from the Global models here. For all models there was an inverse relationship between the percentage of participants above the recommended weight dosage (ARD) and below the recommended weight dosage (BRD): % BRD declined with quantile reaching a minimum at quantile 0.90 whereas the minimum % ARD was at quantile 0.10 (Fig 7). The Global DOLF models resulted in lower % BRD values for quantiles ≥ 0.50 compared to WHO height-based IVM dosing, and greater than quantile 0.35 compared to the WHO age-based DEC dosing. The percentage receiving recommended dosing for the DOLF IVM models were approximately equal to the percentage receiving recommended dosing for the WHO-IVM dosing between quantiles 0.3 and 0.6. For DEC, the percentage receiving recommended dosing for the DOLF model exceeded the percentage receiving recommended dosing for the WHO-DEC dose between quantiles 0.15 and 0.70. The dosing pole that minimized the percentage receiving BRD occurred at the 0.90 quantile, with 5% or less of participants BRD for both IVM and DEC resulting in 22% and 27% improvements over the existing WHO dosing methods, respectively. This dosing pole resulted in 69% (IVM) and 64% (DEC) participants ARD with only 5.8% of participants receiving more than two tablets ARD for IVM, and only 2.5% receiving more than two tablets ARD for DEC. Importantly, the 0.90 quantile DOLF dosing poles dramatically reduced BRD for adult males, with estimates of the percentage receiving BRD less than 3% for IVM and DEC compared to 39% and 54% underdosing for the WHO IVM height pole and WHO DEC age-based dosing, respectively.
The improvement (compared to WHO) in the percentage of participants receiving BRD for IVM with 4-tablet Global hybrid dosing poles was lower than that obtained with the full-dose DOLF IVM dosing pole. For DEC, the BRD percentage fell below the WHO DEC BRD at lower quantiles than the full-dose DOLF DEC dosing pole (Fig 8). The DEC 4-tablet hybrid pole performed better than the full-tablet pole (with respect to BRD) for many of the quantiles because we applied the DOLF IVM dosing pole to DEC which has lower weight thresholds for different dosing levels (see Table 1 and Fig 6). For IVM the hybrid 0.90 quantile dosing pole had 11% lower BRD compared to the WHO IVM pole, and for DEC the 0.90 quantile dosing pole had 22% lower BRD compared to the WHO age-based dosing.
Similar to the 4-tablet hybrid pole, the 6-tablet hybrid percentage receiving BRD for DEC fell below the WHO BRD at small quantiles, and the percentage BRD was the lowest of any of the dosing poles based on Global models—reaching as low as 1.9% (Fig 9). The trade-off is that for the 0.90 quantile dosing pole, 80% of individuals were ARD for DEC with about 7.1% receiving three or more tablets above the recommended dosage. For IVM, the 6-tablet hybrid dosing pole at the 0.9 quantile performed similarly to the full-dose IVM pole with 6% of the participants BRD, 27% recommended, and 67% ARD.
Estimates of the percentage of participants BRD from the Global dosing poles applied to specific countries indicated that at larger quantiles the DOLF dosing poles generally improved upon the current WHO age and height-based dosing (Fig 10, inset and lines). In India, Indonesia, and PNG the 4 and 6-tablet Global models % BRD estimates were approximately equal to the full-dose dosing poles for IVM. However, both hybrid poles resulted in lower estimates of the percentage receiving BRD as compared to the full-dose poles for DEC over all quantiles. In Haiti, the 4-tablet hybrid dosing pole had poorer performance for IVM than the full-dose and 6-tablet hybrid dosing poles for most quantiles, and the 4-tablet hybrid pole performed better than the full-dose DEC poles in all but the most extreme (>0.70) quantiles. The 6-tablet hybrid pole outperformed the 4-tablet and full-dose poles for DEC across all quantiles in Haiti. In Fiji, the full-dose poles for DEC and IVM performed much better than the 4-tablet hybrid poles, especially for larger quantiles. The 6-tablet hybrid poles outperformed the full-dose and 4-tablet hybrid poles for DEC, and the estimates of the percentage receiving BRD for the 6-tablet hybrid pole were similar to the full-dose poles for IVM.
We assessed whether use of the current WHO IVM dosing pole would have improved DEC dosing compared to age-based dosing. Our results indicated that the current WHO IVM pole would have resulted in 17% fewer participants receiving BRD compared to age-based dosing (15% vs 32%, Table 3). The consequence for that improvement was a 12% increase in ARD (33% vs 21%, Table 3). However, the WHO IVM pole did not perform as well as the Hybrid 4 and the Hybrid 6 DOLF dosing poles (0.90 quantile models) which had 22% and 30% fewer participants receiving BRD than age-based dosing, respectively. As indicated previously for IVM, the Hybrid 4 and Hybrid 6 poles reduce the percentage of participants receiving BRD compared to the WHO IVM pole by 11% and 21%, respectively (Table 3).
Our study considered data from more than 26,000 participants, and we believe this is the largest single multi-country study to date that has evaluated different dosing pole recommendations for treating a neglected tropical disease (NTD). The results suggest that the current WHO age (DEC) and height-based (IVM) dosing recommendations for treatment of LF result in an excessive number of individuals receiving lower than weight-based recommended dosing. Importantly, we found that BRD was more frequent for adult males. Adult males are of particular concern for LF elimination programs because they often have higher LF infection rates than other demographic groups [13]. Through a 3-step modeling process we were able to develop dosing poles that have the potential to substantially reduce BRD, a reduction whose benefits far outweigh the downside of an increase in ARD. When we applied a single DOLF hybrid dosing pole (based on our Global models) with 4 or 6-tablet maximal dosing for both DEC and IVM, the percentage of participants with BRD were similar to those obtained with the DOLF full-dose models (6 tablet maximum for DEC [600 mg] and 7 tablet maximum for IVM [21 mg]), and similar to more complex models that would require separate dosing poles for different countries and sexes. Because IVM and DEC have large safety margins, we consider the BRD risk to outweigh the ARD risk. Therefore, we recommend the use of a single 6-tablet maximum hybrid dosing pole based on the Global model from the 0.9 quantile (Figs 6–9).
The dosing poles described in this study improve upon age or height-based dosing methods currently recommended by the WHO. Alexander et al. (1993) developed the current IVM dosing pole that is used by the WHO to treat LF based on weight data collected as part of an onchocerciasis clinical trial in Nigeria. This dosing pole has been widely applied in national MDA programs [14]. The data from our study suggest that improvements can be made to the WHO IVM dosing pole by lowering the height thresholds for dosing. That would result in a marked reduction in BRD (21% reduction for the 6-tablet hybrid pole). The age-based dosing approach for DEC is recommended by the WHO in their guidance manual for preventative chemotherapy [9], and it is also widely implemented as part of community MDA programs. Our results showed that age-based dosing would have resulted in 32% participants receiving BRD in this study, whereas the lowest level of BRD for the 6-tablet hybrid pole was 2% of participants. This discrepancy is largely driven by the low maximum dosage of 300 mg (3 tablets) for WHO age-based dosing compared to maximums of 600 mg (6-tablets) or 400 mg (4-tablets) for the DOLF height-based dosing models. Furthermore, we found that applying the existing WHO IVM dosing pole for DEC dosing would also lower BRD. Although the existing WHO IVM pole did not perform as well as the DOLF hybrid poles, this result indicates that substantial improvements in DEC dosing could be achieved with the legacy IVM dosing pole.
Providing MDA below weight-based recommended dosing may constrain the ability of community MDAs to interrupt disease transmission, which is a principal goal of the GPELF program. Based on estimates from our study, the WHO height-based dosing pole for IVM and the age-based dosing for DEC are likely to provide BRD to as high as one third of the populations in LF-endemic areas. Adult males are a cohort of particular concern because males typically have higher infection rates than females [13]. BRD percentages for this group using WHO age-based dosing were higher than the overall study cohort, reaching 54% of participants for DEC. Although the consequences of BRD on treatment efficacy and transmission are uncertain, the current WHO recommended dosing strategy might prolong the time required to interrupt LF transmission. The dosing pole-approach outlined in our study should reduce the percentage of individuals who receive BRD in community MDA programs where weight-based dosing is not feasible. Using our modeling approach, the trade-off in reducing BRD is that it increases ARD. Of the participants that received ARD in our study, most would have received 2 tablets or fewer above the weight-based recommended dosage (only 3% and 7% of total participants would have received more than 2 tablets above recommended dosage for IVM and DEC, respectively). However, we do not consider ARD to be a major concern, because there is evidence that IVM [15] and DEC [16] are safe and have relatively few side effects even at much higher doses than the weight-based doses recommended by the WHO. Safety margins and other considerations would need to be considered for development of dosing poles for different drug combinations.
Our 3-step modeling process is a novel approach for creating dosing poles that could be applied for mass distribution of other medications. Previous studies have described using height-based dosing poles for treatment of other NTDs [8, 17–19]. However, a variety of factors can affect height-weight relationships (e.g., age, sex, geographic location) which can lead to inaccurate dosing [20]. Thus, use of a single dosing pole may not be appropriate in all settings. In our study, we employed quantile regression which allows for many different dosing pole scenarios to be created and evaluated. Quantile regression has been used in a variety of fields such as econometrics [11, 21, 22], ecology and evolution [23–25], and medicine [26–28] to address questions that cannot readily be answered by standard analytic methods. To our knowledge quantile regression has not been employed previously to develop dosing poles. Data from our study indicate that there is a progressive decoupling of height and weight in larger and older individuals that limits the accuracy of using only height to predict weight. Quantile regression enabled us to “bias” our model predictions by quantile to allow our models to emphasize heavier individuals (to lower the percentage of participants receiving BRD) or to emphasize lighter individuals (to lower the percentage receiving ARD). The choice of whether to weight predictions towards a specific group for dosing purposes will depend on the goals and objectives of a particular program or study, and the safety margins of the medications used for MDA. In our study, the primary focus was to determine whether our models could reduce the amount of BRD that occurs when current WHO height- and age-based dosing recommendations are used for IVM and DEC, respectively. Therefore, we chose models that minimized percentage of participants receiving BRD (larger quantiles that weighted predictions towards heavier individuals). The 306 different statistical models developed in our study can be used to generate both full dose (7 for IVM and 6 for DEC) or hybrid (4 or 6 tablet maximum) dosing poles. These dosing pole models are available as a supplement (S1 Table) and could be employed in a variety of different contexts with guidance.
Current WHO alternatives to weight-based dosing for mass treatment of LF are suboptimal. We have presented a modeling approach that offers an improved dosing method for administering IVM and DEC to LF-endemic populations. Our recommendation for mass treatment of LF is that a single 6-tablet maximum dosing pole from the 0.90 quantile should be used in all contexts. Areas with smaller individuals (e.g., in this study India and Indonesia) may be able to employ the 4-tablet dosing pole without appreciable increases underdosing. Results from our modeling effort go beyond recommending a single dosing pole solution for LF treatment. Users of our models can take into account a variety of competing goals and objectives to choose the dosing pole(s) that best corresponds to their setting. Improved dosing may enhance the efficacy of MDA and accelerate LF elimination.
|
10.1371/journal.ppat.1005120 | Interferon-γ Promotes Inflammation and Development of T-Cell Lymphoma in HTLV-1 bZIP Factor Transgenic Mice | Human T-cell leukemia virus type 1 (HTLV-1) is an etiological agent of several inflammatory diseases and a T-cell malignancy, adult T-cell leukemia (ATL). HTLV-1 bZIP factor (HBZ) is the only viral gene that is constitutively expressed in HTLV-1-infected cells, and it has multiple functions on T-cell signaling pathways. HBZ has important roles in HTLV-1-mediated pathogenesis, since HBZ transgenic (HBZ-Tg) mice develop systemic inflammation and T-cell lymphomas, which are similar phenotypes to HTLV-1-associated diseases. We showed previously that in HBZ-Tg mice, HBZ causes unstable Foxp3 expression, leading to an increase in regulatory T cells (Tregs) and the consequent induction of IFN-γ-producing cells, which in turn leads to the development of inflammation in the mice. In this study, we show that the severity of inflammation is correlated with the development of lymphomas in HBZ-Tg mice, suggesting that HBZ-mediated inflammation is closely linked to oncogenesis in CD4+ T cells. In addition, we found that IFN-γ-producing cells enhance HBZ-mediated inflammation, since knocking out IFN-γ significantly reduced the incidence of dermatitis as well as lymphoma. Recent studies show the critical roles of the intestinal microbiota in the development of Tregs in vivo. We found that even germ-free HBZ-Tg mice still had an increased number of Tregs and IFN-γ-producing cells, and developed dermatitis, indicating that an intrinsic activity of HBZ evokes aberrant T-cell differentiation and consequently causes inflammation. These results show that immunomodulation by HBZ is implicated in both inflammation and oncogenesis, and suggest a causal connection between HTLV-1-associated inflammation and ATL.
| HTLV-1 is a retrovirus which causes a cancer, ATL, and inflammatory diseases of several tissues, such as the spinal cord, eye, skin, and lung. Although these HTLV-1-mediated malignant and inflammatory diseases are recognized as distinct pathological entities, an increased number of HTLV-1 infected cells and enhanced migration/infiltration of infected cells into the lesions are common features of these diseases. Indeed, several clinical observations have suggested a causal link between inflammation and ATL (see Discussion). In order to investigate this issue, appropriate animal models are indispensable. Among HTLV-1-encoded regulatory/accessory proteins, HTLV-1 bZIP factor (HBZ) is thought to be critical to HTLV-1-mediated pathogenesis. We previously reported that HBZ transgenic (HBZ-Tg) mice which express HBZ in CD4+ T cells developed both systemic inflammation and T-lymphomas, indicating that they are suitable to evaluate the link, if any, between these phenomena. In this study, we generated several new genetically engineered strains by modifying HBZ-Tg mice, and found that IFN-γ is an accelerator of HBZ-induced inflammation. Importantly, we show that the incidence of inflammation is correlated with that of lymphomagenesis in HBZ-Tg. These findings indicate that modification of T-cell machinery by HBZ is closely associated with both HTLV-1-associated inflammatory diseases and ATL.
| Human T-cell leukemia virus type 1 (HTLV-1) infects to mainly CD4+ T cells [1], and the provirus is known to exist in effector/memory T cell and regulatory T cell (Treg) subsets [2, 3]. HTLV-1 induces clonal expansion of infected cells and consequently causes a malignancy of CD4+CD25+ T cells, adult T-cell leukemia (ATL) [1]. This virus also gives rise to inflammatory diseases including HTLV-1 associated myelopathy/tropical spastic paraparesis (HAM/TSP), HTLV-1 uveitis (HU), dermatitis, and HTLV-1-associated bronchoalveolitis (HABA)—diseases which are characterized by infiltration of T cells into the lesions [4–7]. In addition, the incidence of several infectious diseases, e.g., infective dermatitis [8] and strongyloidiasis [9], is higher in HTLV-1 carriers than uninfected individuals, suggesting the presence of HTLV-1-mediated cellular immunodeficiency. These findings indicate that HTLV-1 modifies the immunophenotypes of T cells in the host, and these diseases are induced or promoted by aberrant action of infected T cells. Importantly, some clinical observations imply that in HTLV-1-infected subjects, inflammation accelerates ATL development [10, 11], although a molecular basis connecting inflammation to leukemogenesis has not yet been elucidated. In order to understand the causal link between them, suitable animal models are necessary.
The HTLV-1 provirus encodes several regulatory/accessory genes in its pX region [12]. Among them, tax and HTLV-1 bZIP factor (HBZ), which are encoded in the plus- and minus-strand of the pX region respectively, are thought to be important in pathogenesis. HBZ is the only viral gene that is genetically conserved and constitutively expressed in ATL cells [13], whereas Tax is often inactivated by transcriptional silencing or genetic mutations [14, 15]. Moreover, HBZ-transgenic (HBZ-Tg) mice that express HBZ in CD4+ T cells develop systemic inflammatory diseases, cellular immunodeficiency, and T-cell lymphomas, suggesting that HBZ plays important roles in HTLV-1-mediated pathogenesis [16, 17]. In HBZ-Tg, the number of CD4+CD25+ T cells and effector/memory CD4+ T cells are increased as same as ATL cases [3]. Considering the similarities between phenotypes of HBZ-Tg mice and the clinical features of HTLV-1-infected individuals, the HBZ-Tg mouse model is useful for investigating the mechanisms of pathogenesis by HTLV-1.
We reported previously that the number of induced Tregs (iTregs) was increased in HBZ-Tg mice through upregulation of Foxp3, which is a master gene of Tregs [18]. On the other hand, expression of Foxp3 in HBZ-expressing iTregs is easily lost, whereupon these cells convert to IFN-γ-producing cells that are called exFoxp3 cells [19]. We hypothesized that the increase in iTregs and the concurrent induction of IFN-γ-producing cells are implicated in HBZ-mediated pathogenesis in vivo.
In this study, we focused on the significance of IFN-γ in HBZ-induced inflammation and lymphoma, and established HBZ-Tg/IFN-γ knock out (KO) mice. The incidence of dermatitis was significantly lower in HBZ-Tg/IFN-γ KO mice than HBZ-Tg mice, and importantly, HBZ-Tg/IFN-γ KO mice developed no T-lymphomas. In addition, since the intestinal microbiota have important roles in the development and proliferation of iTregs [20], we generated germ-free (GF) HBZ-Tg mice to evaluate the impact of the intestinal microbiota on the increase in Tregs. Even in aseptic circumstances, HBZ-Tg mice developed dermatitis and had the same pattern of T-cell immunophenotypes as specific pathogen free (SPF) HBZ-Tg mice, suggesting that HBZ causes inflammation in a cell intrinsic manner. We also found that the severity of dermatitis correlates with the development of lymphoma in HBZ-Tg mice. These results suggest a close link between inflammation and oncogenesis in HBZ-Tg mice, and demonstrate the important role of IFN-γ in the molecular mechanism of HBZ-mediated pathogenesis.
In order to analyze the impact of IFN-γ on HBZ-mediated pathogenesis, we crossed HBZ-Tg mice with IFN-γ KO mice to establish HBZ-Tg/IFN-γ KO mice (S1 Fig) [21]. We found that some HBZ-Tg mice developed dermatitis at only 8 weeks of age, and 90% of HBZ-Tg mice developed dermatitis within 2 years (Fig 1A and 1B), and these results are consistent with our previous observations [16]. In contrast, HBZ-Tg/IFN-γ KO mice did not suffer from dermatitis until 19 weeks or older, and after 2 years, only 50% of these mice had developed the skin disease (Fig 1B).
To evaluate the presence of systemic inflammation, we performed histological analysis of multiple organs from ten mice of each genotype at 24 weeks of age. The analysis revealed that 30% of HBZ-Tg mice showed infiltration of lymphocytes into the skin at the time point of analysis, whereas no HBZ-Tg/IFN-γ KO mice showed any abnormalities (Fig 1C and Table 1). Our previous study also showed that HBZ-Tg mice which became moribund had lymphomas [16]. Surprisingly, we found that 30% of HBZ-Tg mice had already developed lymphomas in spleen and lymph nodes at 24 weeks of age—earlier than we had guessed—and more importantly, the severity of inflammation correlated with lymphoma development (Fig 1D and Table 1). In contrast, no HBZ-Tg/IFN-γ KO mice had lymphoma. These data strongly suggest that IFN-γ has an important role in inflammation and lymphoma caused by HBZ, and that inflammation might accelerate oncogenesis in HBZ-expressing T cells.
The numbers of Foxp3+CD4+ T cells and effector/memory T cells are increased in HBZ-Tg [16]. To evaluate the influence of IFN-γ on CD4+ T cells, we performed flow cytometry and compared the patterns of T-cell subsets between HBZ-Tg and HBZ-Tg/IFN-γ KO mice. CD4+ T cells from HBZ-Tg/IFN-γ KO mice expressed Foxp3 at similar level to that of HBZ-Tg mice (Fig 2A and 2B and S2 Fig). Likewise, the effector/memory population was increased in HBZ-Tg/IFN-γ KO mice (Fig 2A and 2B and S2 Fig), indicating that these changes in CD4+ T-cell subset populations in HBZ-Tg mice are independent of IFN-γ production and not directly correlated with the inflammatory phenotypes of the HBZ-Tg mice.
Next, we analyzed the production of inflammatory cytokines. Splenic T cells from 24-week-old mice were stimulated by phorbol myristate acetate (PMA)/ionomycin and the expression of IL-17, TNF-α, IL-2, IL-4 and IFN-γ in CD4+ T cells was evaluated by flow cytometry. IFN-γ production was clearly increased in HBZ-Tg mice. Production of IL-17 and IL-2 were also increased in both HBZ-Tg and HBZ-Tg/IFN-γ KO mice (Fig 2C and 2D and S2 Fig). These findings show that loss of IFN-γ does not affect the production of these inflammatory cytokines by HBZ-expressing CD4+ T cells.
Recently, it has been reported that iTregs are most abundant in the colonic mucosa in mice, and that the number of mucosal Tregs is remarkably decreased in germ-free mice, indicating that the gut microbiota has important roles in the development and proliferation of iTregs [20]. Since both HBZ-Tg and HBZ-Tg/IFN-γ KO mice demonstrate increased numbers of iTregs, we asked if the microbiota affected HBZ-mediated iTreg expansion and subsequent inflammation as an extrinsic factor. In order to analyze the impact of microbiota on HBZ-mediated pathogenesis, we generated the germ-free (GF) HBZ-Tg mice, which are genetically the same as the HBZ-Tg mice we reported previously [16]. Contrary to our expectation, these GF HBZ-Tg mice were phenotypically no different than regular HBZ-Tg mice maintained in SPF conditions. The GF HBZ-Tg mice started developing skin inflammation as early as 9 weeks of age, and 16 of 28 (57%) GF HBZ-Tg mice suffered from dermatitis by 18 weeks of age (Fig 3A). Regarding the phenotypes of T cells, there were no significant differences between GF and SPF HBZ-Tg; the number of both effector/memory T cells and Tregs were higher than those in nontransgenic littermates, and the production of IFN-γ was upregulated in HBZ-Tg in both settings (Fig 3B and 3C and S3 Fig). These results imply that the intrinsic activity of HBZ is more important than the intestinal microbiota in influencing the immune modulation, inflammation, and lymphomas observed in HBZ-Tg mice.
In a previous study, we showed that a chemokine receptor, CXCR3, was highly expressed on HBZ-Tg CD4+ T cells and that most cells that had migrated into inflammatory lesions were CXCR3 positive [18]. CXCR3 is expressed in IFN-γ-producing Th1 cells [22]. Thus we hypothesized that the reduction of inflammation in HBZ-Tg/IFN-γ KO mice might correlate with reduced CXCR3 expression on their CD4+ T cells. We compared CXCR3 expression levels between HBZ-Tg and HBZ-Tg/IFN-γ KO mice, and found that HBZ-Tg/IFN-γ KO mice expressed high levels of CXCR3 on CD4+ T cells despite of the absence of IFN-γ (Fig 4A).
Furthermore, we carried out chemotaxis assay to evaluate the function of CXCR3 expressed on CD4+ T cells of HBZ-Tg and HBZ-Tg/IFN-γ KO mice. Murine recombinant CXCL10, which is a major ligand of CXCR3, was used as a chemoattractant [22]. CD4+ T cells were purified from HBZ-Tg and HBZ-Tg/IFN-γ KO mice, and these cells were placed in the upper chambers. The lower chambers were filled with media containing 200 or 500 ng/mL CXCL10 or control media. The migration capacity of CD4+ T cells from HBZ-Tg/IFN-γ KO mice was similar as that from HBZ-Tg mice (Fig 4B). From these results, we conclude that CXCR3 was inducible and functional in HBZ-Tg/IFN-γ KO mice.
Next, we evaluated the importance of CXCL10 in disease development in HBZ-Tg mice, since CXCL10 is one of the chemokines induced by IFN-γ [23]. To do this, we established HBZ-Tg/CXCL10 KO mice [24] (Fig 5A). HBZ-Tg/CXCL10 KO mice developed dermatitis beginning at 12 weeks old (Fig 5B and 5C). At 24 weeks of age, about 80% of the mice had developed dermatitis (Fig 5C). Histological analysis revealed that HBZ-Tg/CXCL10 KO mice also developed inflammation in several other organs (Table 2). In addition, HBZ-Tg/CXCL10 KO mice showed increases in the numbers of Tregs and effector/memory fraction compared to WT mice (Fig 5D). All phenotypes of HBZ-Tg/CXCL10 KO mice we analyzed were quite similar to those of HBZ-Tg mice. We thus concluded that the CXCR3/CXCL10 axis was not related to pathogenesis in our HBZ-Tg mouse model.
Although CD4+ T cells from HBZ-Tg mice and HBZ-Tg/IFN-γ KO mice were similar in their migratory responses to CXCL10, their abilities to infiltrate tissues in vivo may differ, because the HBZ-Tg/IFN-γ KO mice did not develop dermatitis to the same degree that the HBZ-Tg mice did. Therefore we looked for chemokine receptors or adherent molecules that are highly expressed on T cells in HBZ-Tg but not HBZ-Tg/IFN-γ KO mice. As shown in Fig 6A, most of the molecules studied were highly expressed on CD4+ T cells of both HBZ-Tg and HBZ-Tg/IFN-γ KO mice compared with wild type littermates. However, we found that the chemokine receptor CCR9 was upregulated only in HBZ-Tg mice (Fig 6B), suggesting that upregulation of CCR9 is involved in inflammation mediated by HBZ and IFN-γ.
In order to identify further cellular genes implicated in HBZ/IFN-γ-mediated inflammation, we performed DNA microarray analysis. We extracted RNA from CD4+ T cells of WT, HBZ-Tg, IFN-γ KO, and HBZ-Tg/IFN-γ KO mice and evaluated the profiles of gene expression. According to the result of microarray, we picked up several genes that were expressed higher in HBZ-Tg than HBZ-Tg/IFN-γ KO, and validated their expression profiles by quantitative RT-PCR. Among these genes, we further looked for the genes that are overexpressed in human ATL cases. Finally, we identified Neo1, Il1f9, Fgfr4, Hip1, Iklf2, and Nrxn3 that met these criteria (Fig 7A and 7B). Interestingly, human homologues of these genes were upregulated especially in the aggressive form of ATL (Fig 7B). They are likely to be divided into 2 groups by the pattern of the expression in healthy donor cells. One contains the genes which expression is unchanged or reduced in phytohaemagglutinin (PHA)-stimulated cells compared with resting cells, such as NEO1, NRXN3, IKZF2, and HIP1. In contrast, IL1F9 and FGFR4 belong to another group in which their transcription are enhanced by PHA, suggesting that they are inducible by potent mitogenic stimulation even in normal T cells. These genes were generally overexpressed in HTLV-1-transformed and ATL cell lines although there were several exceptions (S1 Table). Interestingly, it has been reported that most of them are aberrantly expressed in several types of cancer cells, suggesting that they are associated with the linkage between chronic inflammation and oncogenesis in HTLV-1-infected subjects.
Persistent inflammation is widely recognized as a tumor-promoting factor in many cancers, and it is estimated that about 15% of human malignancies are associated with chronic inflammation and infection [25]. For example, inflammatory bowel diseases, such as ulcerative colitis, are associated with colon cancer [26]. Chronic gastritis caused by Helicobacter pylori [27] and chronic hepatitis caused by hepatitis B virus or hepatitis C virus [28] are implicated in development of gastric cancer and hepatocellular carcinoma (HCC), respectively. In these solid tumors, infiltrating immune cells are thought to produce cytokines, chemokines, and growth factors that induce the proliferation of tumor cells [25]. In addition, those inflammatory cells produce reactive oxidative species resulting in genetic instability [29]. Activation of the TNF-α or the NF-κB pathway is important especially in the development of HCC [30] and colon cancer [31].
In the case of HTLV-1 infection, the virus itself dysregulates the functions of CD4+ T cells, modifies T-cell subsets, and triggers clonal expansion of infected cells. HTLV-1 causes both inflammation and a malignant disease, but a precise mechanism crosslinking these diseases was not clarified. Several clinical observations have suggested the correlation between HTLV-1-associated inflammatory diseases and ATL. It was reported that the frequency of ATL development in HTLV-1-infected patients with diffuse pan-bronchiolitis was significantly high among all HTLV-1 carriers [10]. In addition, the abundance of certain HTLV-1-infected clones is increased in HTLV-1 carriers with strongyloides and infective dermatitis [11], implying that these inflammatory diseases increase the risk of ATL development. In this study, we found T-cell lymphomas only in HBZ-Tg mice with dermatitis, and severity of inflammation tended to correlate with lymphoma development, suggesting that inflammatory signals induced by HBZ accelerate oncogenic processes. Since there is no immune reaction against HBZ in these mice, HBZ triggers inflammation only by its intrinsic action. This idea is compatible with the findings that, even in a germ-free environment, the number of Tregs was increased in HBZ-Tg mice and they developed systemic inflammation the same as under SPF conditions. These results suggest that the inflammatory phenotypes of HBZ-Tg mice are caused by an inherent function of HBZ, and that HBZ-mediated inflammation promotes oncogenesis in HBZ-expressing CD4+ T cells.
In addition, we show here that IFN-γ is an important molecule in the pathogenesis by HBZ. IFN-γ is conventionally recognized as a cytokine that acts in host defense against various pathogens and tumor rejection. IFN-γ is secreted by mainly activated CD4+ T cells (Th1 cells), cytotoxic CD8+ T lymphocytes, and natural killer cells, and has cytostatic/cytotoxic effects by inducing cell-mediated immune responses [32]. IFN-γ primarily activates the JAK/STAT signaling pathway through interaction with IFN-γR1, and induces the transcription of primary response genes such as IRF family genes. Many of these primary response genes encode transcription factors that induce a lot of secondary response genes to react to the stimulation. Previous studies showed that blockade of IFN-γ/IFN-γR signaling in mice compromised rejection of tumors by the immune system, indicating that IFN-γ functions in immune surveillance against tumors [33–35]. On the other hand, under certain circumstances, IFN-γ is also known to have a protumorigenic function involving proliferative and anti-apoptotic signals in tumor cells [32]. In this study, we found that knocking out of IFN-γ significantly decreased the incidence of inflammation and malignant lymphoma in HBZ-Tg mice, indicating that IFN-γ plays a supportive role in the development of both types of diseases caused by HBZ.
To understand how IFN-γ contributes to HBZ-associated pathogenesis, we looked for cellular factors differentially expressed in CD4+ T cells of HBZ-Tg compared with HBZ-Tg/IFN-γ KO mice. These genes are thus implicated in pathogenesis mediated by HBZ and IFN-γ together. CCR9 is an intestine oriented chemokine receptor [36]. This upregulation is consistent with our observation that massive infiltration of lymphocytes was observed in HBZ-Tg mice [18]. We also identified several cancer-related genes which are overexpressed in both HBZ-Tg and ATL patients. NEO1 encodes a cell surface protein that belongs to the immunoglobulin superfamily. It has been reported that overexpression of NEO1 in gastric cancer is involved in cell proliferation and migration [37]. IL1F9, also known as IL36gamma, is an IFN-γ-inducible gene that has been reported to activate NF-κB and MAPK signaling in human T cells [38]. FGFR4 encodes a member of the fibroblast growth factor receptor family, and implicated in the tumorigenesis of many types of cancers, such as HCC, prostate cancer, breast cancer, pancreatic cancer [39–43]. IKZF2 encodes a member of the Ikaros family of zinc-finger proteins, Helios, which is mainly expressed in T cell. A recent study showed that aberrant isoforms of IKZF2 are dominantly expressed in ATL cells, and function in T-cell proliferation and survival [44], suggesting that HBZ might dysregulate the expression pattern of IKZF2 in ATL cells. HIP1 is also overexpressed in several cancer tissues like breast cancer and possesses the oncogenic properties through BCL-2 and NF-κB pathways [45]. Taken together, it is possible that HBZ and HBZ-mediated inflammation induce these factors and subsequently trigger transformation in a part of HTLV-1-infected cells. In order to clarify the significance of each factor in HBZ-mediated pathogenesis, further experiments will be required. Interestingly, previous studies on Tax, which is another oncoprotein of HTLV-1, showed that transgenic mice expressing Tax under control of the granzyme B promoter developed LGL leukemia, and knocking out of IFN-γ in this strain enhanced the tumor formation [46, 47], suggesting that IFN-γ has the opposite effect on Tax-mediated oncogenesis that it has on HBZ-mediated oncogenesis. In these Tax-Tg mice, IFN-γ was shown to have an anti-angiogenic effect by suppressing the transcription of VEGF [47]. HBZ and Tax regulate several signaling pathways in opposite manners [1], suggesting that IFN-γ may differentially regulates the effects of HBZ and Tax on HTLV-1-infected cells or HBZ and Tax may regulate IFN-γ in opposite way, in response to the cellular context.
In HAM/TSP patients, IFN-γ-producing cells are increased in a CD4+FoxP3- subpopulation, and suggested to have a role in the pathogenesis of this inflammatory disease [48, 49]. A recent study showed that HTLV-1-infected cells in the cerebrospinal fluid expressed IFN-γ and CXCR3, and its ligand CXCL10 was expressed in astrocytes upon stimulation with IFN-γ, leading to an IFN-γ-CXCL10-CXCR3 inflammatory loop [50]. In our HBZ-Tg mice, however, CXCL10 is not associated with inflammation, since loss of CXCL10 didn’t affect the development of inflammatory diseases. In addition, the upregulation of CXCR3 observed in HBZ-Tg mice was independent of IFN-γ. Therefore CD4+ T cells from HBZ-Tg/IFN-γ KO mice still expressed high levels of CXCR3, and could react to its ligand. According to these observations, CXCL10/CXCR3 is unlikely to have strong effects on inflammation induced by HBZ. Indeed, the expression of several other adherent molecules and chemokine receptors such as CCR4, CD29, and CD49d, also showed the same pattern as CXCR3 (Fig 6). Induction of these molecules is mediated by HBZ, but not associated with IFN-γ, suggesting that these molecules might be involved in the inflammation that occurred late in HBZ-Tg/IFN-γ KO mice. Further studies are needed to test this hypothesis.
In conclusion, we showed that IFN-γ, which is secreted by Th1-like cells such as exFoxp3 cells, has important roles in HBZ-mediated inflammation. HBZ increases the number of Tregs in a cell intrinsic manner, and consequently induces IFN-γ in vivo. Importantly, inflammation is closely linked to the development of malignant lymphomas in HBZ-Tg mice. This is the first report showing the relationship between the immunomodulating function of HBZ and oncogenesis that might explain the clinical observations of ATL development in HTLV-1-infected subjects with chronic inflammations.
C57BL/6J mice were purchased from CLEA (Tokyo, Japan). Transgenic mice expressing the spliced form of the HBZ gene under control of the mouse CD4 promoter have been described previously [13, 16]. B6.129S7-Ifnγtm1Ts/J (Ifnγ-/-) [21] and B6.129S4-Cxcl10tm1Adl/J (Cxcl10-/-) [24] mice were purchased from The Jackson Laboratory (CA, USA). Mice used in this study were maintained under SPF conditions unless otherwise specified. GF HBZ-Tg and wild type mice were reconstituted from frozen embryos and reared at the Central Institute for Experimental Animals (Kawasaki, Japan). GF mice aged 18 weeks were transferred to Kyoto University, and analyzed within 24 hours.
HTLV-1-transformed cell lines (MT-2 and MT-4), ATL cell lines (MT-1, ED, TL-Om1, ATL-43T+, and ATL-55T+) were cultured in RPMI 1640 medium supplemented with 10% fetal bovine serum (FBS) and antibiotics at 37°C under a 5% CO2 atmosphere. For IL-2-dependent cell lines (ATL-43T+ and ATL-55T+), recombinant human IL-2 (100 U/ml) was added in the culture media.
Peripheral blood mononuclear cells (PBMCs) of ATL patients and healthy donors were collected by Ficoll-Paque PLUS (GE Healthcare). To obtain PHA-stimulated cells, PBMCs were treated with 10μg/ml PHA (Sigma) for 3 days.
The following antibodies were used for flow cytometric analysis of mouse lymphocytes:
Anti-CD3e (145-2C11), CCR5 (C34-3448), IFN-γ (XMG1.2), IL-2 (JES6-5H4), IL-17 (TC11-18H10), CD29 (Ha2/5), CD49d (9C10), and CD162 (2PH1) antibodies were purchased from BD Pharmingen. Anti-CD4 (RM4-5), CD8 (53–6.7), CD44 (IM7), CD62L (MEL-14), CXCR3 (CXCR3-173), CCR4 (2G12), and TNF-α (MP6-XT22) antibodies were from Biolegend. Anti-CD25 (pc61), Foxp3 (FJK-16s), CCR9 (eBioCW-1.2), and IL-4 (11B11) antibodies were from eBioscience. Anti-CCR10 antibody (248918) was purchased from R&D systems. In order to stain cytokines, splenocytes were stimulated with 50ng/mL PMA (Nakarai Tesque), 1μg/mL ionomycin (Nakarai Tesque) and a protein transport inhibitor, BD Golgi plug (BD Pharmingen) for 4 hours before harvesting cells. After cell surface staining, cells were fixed and permeabilized with Fixation/Permeabilization working solution (eBioscience) and intracellular antigens were stained. Flow cytometric analysis was carried out using a FACS Verse with FACSuite software (BD Biosciences) and Flow Jo (FlowJo, LLC).
Mouse tissues were fixed in 10% formalin in phosphate buffer (Nakarai Tesque) and then embedded in paraffin. Hematoxylin and eosin staining was performed according to standard procedures. Images were captured using a Provis AX80 microscope (Olympus) equipped an OLYMPUS DP70 digital camera, and detected using a DP manager system (Olympus).
Mouse CD4+ T cells were isolated from splenocytes by CD4 T lymphocyte enrichment Set-DM (BD Biosciences) and resuspended in RPMI containing 0.1% BSA. To evaluate migration activity, a Transwell insert (3.0um) (CORNING) was used. The lower chamber was filled with chemotaxis medium containing mouse recombinant CXCL10 (R&D systems). One million cells were added into the upper chamber. The chamber was incubated for 1 hour at 37C and 5% CO2. Cells that migrated towards CXCL10 were counted using Flow cytometry.
CD4+ T cells were isolated from WT, HBZ-Tg, IFN-γ KO and HBZ-Tg/IFN-γ KO mice as described above and lysed in TRIzol (Life Technologies). Total RNAs were extracted from these lysates with Direct-zol RNA MiniPrep (Zymo Research). RNA quality was checked using Agilent 2100 Bioanalyzer (Agilent Technologies). Microarray experiments were carried out with SurePrint G3 Mouse GE 8x60K (Agilent Technologies) according to manufacturer’s instructions. Data was analyzed with GeneSpring GX software (Agilent Technologies).
Splenocytes harvested from WT, HBZ-Tg, IFN-γ KO, and HBZ-Tg/IFN-γ KO mice and human PBMCs obtained from ATL patients and healthy donors were lysed with TRIzol reagent, and RNA was extracted as described above. cDNAs were synthesized from 1μg of total RNAs using random primers and SuperScript III Reverse Transcriptase (Life Technologies). The expression levels of candidate genes were quantified by the StepOnePlus real time PCR system (Life Technologies) using FastStart Universal SYBR Green Master (Roche). Relative expression levels of each gene were calculated by the delta delta Ct method [51]. The sequences of primers used in this study are listed in S2 Table. Human NRXN3 was quantified using Taqman Gene Expression Assays (Applied Biosystems, Hs01028186_m1).
Animal experiments were performed in strict accordance with the Japanese animal welfare bodies (Law No. 105 dated 19 October 1973 modified on 2 June 2006), and the Regulation on Animal Experimentation at Kyoto University. The protocol was approved by the Institutional Animal Research Committee of Kyoto University (Permit numbers are D13-02, D14-02, and D15-02). Experiments using clinical samples were conducted according to the principles expressed in the Declaration of Helsinki, and approved by the Institutional Review Board of Kyoto University (Permit numbers are G310 and G204). ATL patients provided written informed consent for the collection of samples and subsequent analysis.
|
10.1371/journal.pgen.1003850 | Evolutionary Change within a Bipotential Switch Shaped the Sperm/Oocyte Decision in Hermaphroditic Nematodes | A subset of transcription factors like Gli2 and Oct1 are bipotential — they can activate or repress the same target, in response to changing signals from upstream genes. Some previous studies implied that the sex-determination protein TRA-1 might also be bipotential; here we confirm this hypothesis by identifying a co-factor, and use it to explore how the structure of a bipotential switch changes during evolution. First, null mutants reveal that C. briggsae TRR-1 is required for spermatogenesis, RNA interference implies that it works as part of the Tip60 Histone Acetyl Transferase complex, and RT-PCR data show that it promotes the expression of Cbr-fog-3, a gene needed for spermatogenesis. Second, epistasis tests reveal that TRR-1 works through TRA-1, both to activate Cbr-fog-3 and to control the sperm/oocyte decision. Since previous studies showed that TRA-1 can repress fog-3 as well, these observations demonstrate that it is bipotential. Third, TRR-1 also regulates the development of the male tail. Since Cbr-tra-2 Cbr-trr-1 double mutants resemble Cbr-tra-1 null mutants, these two regulatory branches control all tra-1 activity. Fourth, striking differences in the relationship between these two branches of the switch have arisen during recent evolution. C. briggsae trr-1 null mutants prevent hermaphrodite spermatogenesis, but not Cbr-fem null mutants, which disrupt the other half of the switch. On the other hand, C. elegans fem null mutants prevent spermatogenesis, but not Cel-trr-1 mutants. However, synthetic interactions confirm that both halves of the switch exist in each species. Thus, the relationship between the two halves of a bipotential switch can shift rapidly during evolution, so that the same phenotype is produce by alternative, complementary mechanisms.
| In roundworms, the transcription factor TRA-1 controls sexual development. We show that TRR-1 is part of a complex of proteins that adds acetyl groups to its targets, and that this complex works with TRA-1 to initiate spermatogenesis. By contrast, a truncated form of TRA-1 blocks spermatogenesis. Because two different forms of this transcription factor oppose each other, TRA-1 is bipotential. To see if the interdependent relationship between these two forms has affected evolutionary change, we compared related species of roundworms. In one species, mutations that affect TRR-1 cause all germ cells to become oocytes, but mutations that affect three FEM proteins, which also regulate TRA-1, do not. In the other species, the roles of these regulatory genes are reversed. Thus, the relative importance of the two halves of this bipotential switch has changed during recent evolution.
| Most animals carry the genetic information needed to produce two alternate sexes, and must choose which program to employ. In the nematode C. elegans, for example, a signal transduction pathway controls the activity of the master transcription factor TRA-1 to determine sex [reviewed by 1], [2], [3]. The leading model is that hermaphrodites are produced when TRA-1 prevents the expression of male genes like fog-3 in germ cells [4], egl-1 in the HSN neurons [5], and mab-3 in the intestine and tail [6]. TRA-1 activity is itself controlled by interactions between the TRA-2 receptor and a complex of FEM proteins, which target TRA-1 for degradation [7]. As a result, XX animals have abundant TRA-1, which causes hermaphrodite development, and XO animals lack TRA-1, which allows male development.
In this model, the switch that controls sex-determination is part of a linear pathway. However, the fact that TRA-1 is a Gli protein [8] implies that its regulation might be more complex. Mammalian Gli proteins and their Drosophila homolog Ci are not only cleaved to form repressors or ubiquitinylated and degraded [9], but also function as activators [10]–[12]. As a consequence, these proteins constitute a key class of bipotential transcription factors [13].
Some of these regulatory interactions are also found in nematodes. For example, TRA-1 can also be cleaved to form a repressor [14], or eliminated by ubiquitinylation and degradation [7]. Although it is not known if TRA-1 is a direct activator of any male genes, older tra-1 mutants cannot maintain spermatogenesis [15], [16]. This phenotype and several other traits make the germ line an ideal tissue for elucidating how TRA-1 functions. First, both C. elegans and the related nematode C. briggsae make XX hermaphrodites, which resemble females but produce sperm before beginning oogenesis. Thus, both sexual fates can be studied in a single animal. Second, self-fertilization simplifies genetic screens, a fact that has been used for decades in C. elegans, and is now being exploited in C. briggsae [17]. Third, information about germ cell development in C. elegans provides the background needed for dissecting TRA-1 regulation [reviewed by 2]. Fourth, although some TRA-1 mutations promote spermatogenesis and others promote oogenesis, TRA-1 is not absolutely required for either fate, since null mutants make both sperm and oocytes [15], [16]. Thus, we can isolate mutations that affect all aspects of tra-1 activity by observing how they alter the sperm/oocyte decision.
This system is also a leading model for studying evolution. Hermaphroditic reproduction originated on several independent occasions in the genus Caenorhabditis [18]–[20]. It appears to have required regulatory changes that affect both sex-determination and sperm activation [21]. Hence, comparative analyses of C. briggsae and C. elegans could elucidate not only how TRA-1 functions, but also how the sex-determination pathway changed during recent evolution.
Here, we show that TRR-1, the homolog of mammalian TRRAP (TRansformation/tRanscription domain-Associated Protein), acts as part of the Tip60 Histone Acetyl Transferase (HAT) complex to control germ cell fates in C. briggsae. Genetic and molecular experiments imply that it works with TRA-1 as a co-activator to promote the expression of genes like fog-3. This activity occurs in parallel to the regulation of TRA-1 by the receptor TRA-2 and the FEM proteins, which control the levels of a cleaved TRA-1 repressor. Although these two functions have been conserved throughout Caenorhabditis, their relative importance has changed dramatically during recent evolution. Moreover, this shift shows that a novel trait can evolve by distinct but compensatory changes within the structure of a binary switch, and explains the puzzling differences between the functions of the C. elegans and C. briggsae fem genes [22].
In C. elegans, three fog genes were identified by screening for mutations that cause feminization of the germ line (the Fog phenotype) [23]–[25]. Because only two of these genes have homologs in C. briggsae, the regulation of germ cell fates must have changed during recent evolution [26]. When we screened for C. briggsae Fog mutants, we identified the new gene Cbr-she-1 [17], and also recovered the mutations v76 and v104. These alleles are recessive, fail to complement each other, and affect germ cell fates in both sexes — the XX animals only make oocytes and develop as females, and the XO males make oocytes instead of sperm, but are otherwise normal (Fig. 1). Thus, this gene controls the sperm/oocyte decision. Surprisingly, it maps to the left arm of chromosome II (Methods, Fig. 2A), which contains no homologs of C. elegans fog genes.
We used SNP mapping to clone this gene. Using the linked mutations Cbr-cby-15(sy5148) and Cbr-dpy(nm4), we identified recombinants between the AF16 and HK104 strains, and located each breakpoint using SNPs (Fig. 2B). Next, we studied a 150 kb region of the C. briggsae genome [27] that contained the new gene (Fig. 2C), and used RNA interference to test candidates. Knocking down Cbr-trr-1 activity caused a Fog phenotype. Finally, we sequenced Cbr-trr-1 genomic DNA, and found that both v76 and v104 were missense mutations (Fig. 2D, Table S1). Hence, C. briggsae trr-1 is a fog gene that regulates the decision of germ cells to become sperm or oocytes. It encodes the nematode homolog of the human TRRAP protein, which is a component of several Histone Acetyl Transferase (HAT) complexes [28].
Finally, we used RT-PCR and RACE to clone a complete Cbr-trr-1 transcript, which differs slightly from that predicted at WormBase (Fig. S1). The encoded protein is 4037 amino acids long, and is 67% identical and 82% similar to C. elegans TRR-1.
Because these alleles of Cbr-trr-1 were missense mutations, it was possible that neither caused a complete loss of function. Thus, we used a non-complementation screen to identify additional alleles (Fig. S2, Table S1). All of the new mutations were also Fog when homozygous, with the exception of the large deletion vDf4, which caused embryonic lethality. The Fog alleles include three small deletions, two of which shift the reading frame, and the deletion vDf3, which removes the entire gene (Fig. 2D). We conclude that eliminating Cbr-trr-1 causes the production of oocytes instead of sperm. Although all Cbr-trr-1 males made oocytes, 34% of v108 males and 44% of vDf3 males produced a few sperm before beginning oogenesis.
TRR-1, like human TRRAP [28], contains a PI-kinase like domain with an inactive catalytic site near its carboxyl terminus. This domain is altered by the new v128 missense mutation, which demonstrates its importance for TRR-1 activity.
Although C. elegans trr-1 regulates vulval development [29], none of the C. briggsae mutants showed vulval defects. Furthermore, Cbr-trr-1(RNAi) Cbr-lin-8(RNAi) animals had normal vulvae, which suggests that C. briggsae trr-1 is not a synthetic multivulva gene.
Finally, we looked for a maternal requirement by crossing homozygous females with heterozygous males. About 33% of the Cbr-trr-1(v76) progeny died as embryos, as well as all of the Cbr-trr-1(v108) progeny and some of the v108 heterozygotes (Table S2A). Since v108 is a deletion, this represents the null phenotype. To test for a strict maternal effect, we crossed homozygous females with wildtype males, and saw that some of their embryos died and some survived (Table S2B). Taken together, these results show that maternal and zygotic TRR-1 work together to promote viability during embryogenesis. In adults, the sperm/oocyte decision is primarily controlled by zygotic TRR-1.
TRR-1 is the sole nematode TRRAP protein [29]. In eukaryotes, TRRAP proteins are components of both the GNAT and MYST families of HAT complexes [reviewed by 28]. To see if TRR-1 acts through either of these complexes to control germ cell fates, we used RNA interference to knock down other components in C. briggsae (Table 1). Targeting Cbr-pcaf-1, which encodes the catalytic component of the GNAT family, did not affect germ cell fates, and knocking down three other components of this HAT complex also failed to produce Fog animals. However, targeting other components of the Tip60 HAT complex caused phenotypes like that of Cbr-trr-1. Knocking down four genes, including the catalytic component Cbr-mys-1, produced Fog animals at high frequency, and four other genes produced Fog animals at low frequency. For some genes, we saw embryonic lethality and some sterility. Thus, TRR-1 acts through the Tip60 HAT complex to control the sperm/oocyte decision in C. briggsae.
To determine where the Tip60 complex acts in the sex-determination pathway, we examined double mutants. First, we studied the upstream genes tra-2 and tra-3, which promote female development. Mutations in either gene cause XX animals to produce only sperm, and transform the body towards male fates [30]. Moreover, Cbr-tra-2(nm1) is a nonsense mutation that should act as a null allele, so it is ideal for epistasis. In each case, double mutants with Cbr-trr-1 restored the ability of XX animals to produce oocytes (Table S3), which indicates that Cbr-trr-1 acts downstream of these genes, or in parallel to them.
Next we studied mutations in the Cbr-fem genes, which are required for male somatic development. Although these genes act downstream of tra-2, null mutants produce sperm and oocytes, and develop as normal hermaphrodites [22]. As expected, double mutants with Cbr-trr-1 caused XX animals to produce only oocytes (Table S3). In addition, we saw a synthetic lethal interaction between Cbr-trr-1 and Cbr-fem-2. FEM-2 is a protein phosphatase [31], [32], which also causes synthetic lethality in C. elegans, in combination with Cel-mel-11 mutations [33].
Finally, we studied the interaction of Cbr-trr-1 with Cbr-tra-1, the master regulator of sexual development in nematodes (Fig. 3A–D). We used the Cbr-trr-1 deletion vDf3 and the point mutation v76, which has a strong germline phenotype but causes only low levels of embryonic lethality, so that we could study homozygous children of homozygous mothers. In all cases, we found that the Cbr-tra-1(nm2) mutation restores the production of sperm to Cbr-trr-1 mutants. This tra-1 allele is a nonsense mutation and behaves like a null allele [30]. Thus, the Tip60 complex promotes spermatogenesis, but only if TRA-1 is functional.
In C. elegans, TRA-1 acts directly on the fog-3 promoter [4], and these binding sites have been conserved in C. briggsae [34]. Moreover, C. briggsae fog-3 is required for germ cells to become sperm, and the level of fog-3 transcripts is correlated with spermatogenesis [34]. Since TRRAP proteins regulate transcription, we studied Cbr-tra-1 and Cbr-fog-3 transcripts, to see if either was affected by inactivation of Cbr-trr-1. We could detect no change in Cbr-tra-1 transcript levels or protein levels in Cbr-trr-1 mutants (Fig. 3E, F), so the Tip60 complex does not control the expression of TRA-1. However, loss of Cbr-trr-1 activity eliminates fog-3 transcripts, but only if TRA-1 is active (Fig. 3F). Thus, we propose that the Tip60 HAT complex cooperates with TRA-1 to increase fog-3 expression and promote spermatogenesis.
Although the transmembrane receptor TRA-2 acts through TRA-1 to promote female cell fates, two differences between these genes have been conserved among Caenorhabditis species. First, tra-2(null) XX mutants do not make complete male tails, but instead produce stubby tails that lack many of the sensory structures known as rays (Figure 4A); by contrast, tra-1(null) mutants develop perfect male tails [30], [35]. Second, tra-2(null) mutants only produce sperm, whereas tra-1(null) mutants make sperm early in life and then switch to oogenesis [15], [16], [30]. These differences do not reflect the degree of transformation, since tra-2(null) mutants have more strongly masculinized germ lines, but less masculinized tails.
We were surprised to find that Cbr-tra-2 Cbr-trr-1 double mutants resemble Cbr-tra-1 mutants in both of these traits. The double mutants make oocytes (Table S3), which shows that the Tip60 complex acts downstream of tra-2, or in parallel to it. However, oogenesis usually begins after an initial burst of spermatogenesis, as in Cbr-tra-1 XX animals. Moreover, the double mutants develop much better male tails than those seen in Cbr-tra-2 single mutants, with rays that are almost wildtype in both length (Figure 4B) and number (Figure 4C). Thus, the Tip60 complex is required for precisely the two tra-1 phenotypes that are not dependent on tra-2. As a consequence, Cbr-tra-2 Cbr-trr-1 double mutants appear to have lost all tra-1 activity.
The trr-1 gene regulates vulval development in C. elegans, and the embryos produced by null mutants die [29]. To see if Cel-trr-1 regulates the sperm/oocyte decision, we used RNA interference to knock it down in males, since the production of oocytes by a male would prove that a change in germ cell fates had occurred. At 20°C, 14% of Cel-trr-1(RNAi) males made oocytes (n = 58). This phenotype was sensitive to temperature, since 23% of males made oocytes at 25°C (n = 62), and an additional 8% failed to produce mature germ cells. Finally, knocking down trr-1 activity in the male/female species C. remanei also caused males to produce oocytes (Fig. S3), so we believe that the male function of the Tip60 complex in germ cell fates has been conserved during Caenorhabditis evolution.
We also tested the ability of C. elegans trr-1 to act synthetically with tra-2, by using RNA interference to knock down Cel-trr-1 activity in Cel-tra-2(e1095) null mutants. As in C. briggsae, lowering trr-1 activity produced better, more masculine tails in tra-2 XX animals (Fig. 4C). In addition, these double mutants also produced sperm and oocytes.
Finally, we tested C. elegans hermaphrodites. Rare Cel-trr-1(RNAi) XX animals developed as females at 25°C (6%, n = 135), although none of them did at 20°C (n = 332). This result suggested that trr-1 influences the sperm/oocyte decision in C. elegans hermaphrodites, but that it plays a smaller role than in C. briggsae. To measure its relative importance in each species, we assayed gene activity in sensitive genetic backgrounds. First, we studied partial loss-of-function mutations in either C. elegans fem-1 or fem-2. At permissive temperatures, these animals usually developed as hermaphrodites; however, if Cel-trr-1 activity was knocked down by RNAi, they became female (Table 2). Thus, trr-1 normally plays a minor role in C. elegans sex-determination because the activity of the fem genes is high.
Second, we studied partial loss-of-function mutations in C. briggsae trr-1. These mutants usually develop as hermaphrodites when maternal trr-1 activity is present (Table 2). Furthermore, null alleles of the C. briggsae fem genes are famous for not preventing hermaphrodite development, and Cbr-fem-3(nm63) does not alter the number of sperm produced by hermaphrodites [22]. However, all germ cells in Cbr-trr-1(weak); Cbr-fem-3(nm63) double mutants developed as oocytes rather than sperm (Table 2), and experiments with Cbr-fem-2 gave similar results. Thus, a loss of fem activity has no effect on C. briggsae hermaphrodites because trr-1 activity is normally high. We conclude that the relative importance of each regulator has shifted during the independent evolution of these self-fertile hermaphrodites.
A major thrust of evolutionary developmental biology has been comparing traits between model organisms like C. elegans or Drosophila melanogaster and closely related species [e.g. 17], [22], [36], [37]. These studies have focused on identifying key differences that provide a window into evolutionary mechanisms.
However, our results emphasize an equally important use of comparative evolutionary studies — conserved aspects of gene regulation are sometimes more easily detected in one species than in another. For example, we found that Fog mutations in the trr-1 gene are common and easy to work with in C. briggsae. By contrast, the role that trr-1 plays in spermatogenesis had not been detected in C. elegans, despite decades of work on sex-determination [reviewed by 3] and the existence of C. elegans trr-1 deletion mutants [29].
What factors might account for this difference? First, mutations in essential genes might be detected more easily in one species than another, because their pleiotropic activities differ. Second, the absence of redundant regulatory pathways might make it easier to identify mutants in one species. Third, the relative importance of the genes might differ in each species. All of these factors may have delayed the identification of the role Tip60 plays in sex determination.
We also note that the combination of forward genetic screens in C. briggsae with cloning by SNP mapping provides a powerful, unbiased approach to studying evolutionary differences. This technique has already been used to clone one other novel C. briggsae gene [17]. In addition, SNP mapping has helped assign new mutations to known genes from several well-characterized pathways [38], [39].
In nematodes, tra-1 plays a critical role in the regulation of sexual development. Null mutations in either C. elegans or C. briggsae cause XX animals to develop as fertile males [30], [35]; a transformation that involves many cell fates. Because TRA-1 also determines the sex of the distantly related nematode Pristionchus pacificus, this role is ancient [40]. Finally, TRA-1 is the sole nematode homolog of the Gli proteins from humans and of Cubitus interruptus from flies [8], and like them, it acts by regulating the transcription of numerous target genes [4]–[6].
The transmembrane receptor TRA-2 acts through TRA-1 to promote female development, but their effects differ in the germ line and male tail [15], [16], [35]. Thus, we were surprised to find that tra-2 trr-1 animals do not look like tra-2 mutants, but instead resemble tra-1 null mutants — they produce excellent male tails and make oocytes after an initial burst of sperm production. Since mutations in trr-1 and tra-2 affect both the germ line and cells of the developing tail, each of these genes must play a general role in the regulation of TRA-1, rather than a tissue-specific function in implementing one particular cell fate.
Although Tip60 works through TRA-1, knocking down Tip60 activity causes oogenesis, a phenotype normally associated with increased TRA-1 activity [16]. Since numerous papers have already shown that TRA-1 can repress some male genes [4]–[6], we infer that it is bipotential and can also activate genes needed for spermatogenesis.
To begin, two genetic experiments imply that repression by TRA-1 is mediated by a cleaved form of the protein, known as TRA-1100 [14]. First, the tra-1(e2272stop) mutant encodes a truncated form of the protein slightly shorter than TRA-1100 [41]; when its transcripts are protected from nonsense-mediated decay, they direct female development and oogenesis [42]. Second, a truncated from of TRA-1 created by the duplication eDp24 also directs female development [43]. Since the known targets of TRA-1 are male genes, truncated TRA-1 must work as a repressor. The levels of this repressor are protected by TRA-2, but decreased by three FEM proteins and CUL-2, which form part of a ubiquitin-ligase complex [7].
Previous studies hinted that TRA-1 has a second activity. First, wild-type males produce normal levels of full-length TRA-1, but lack the cleaved form [14]. Since old tra-1 males begin producing oocytes, full-length TRA-1 appears necessary to maintain spermatogenesis [15], [16]. Second, eliminating the TRA-1 binding sites from the promoter of C. elegans fog-3 prevents the transgene from directing spermatogenesis [4]. Taken together, these results imply that full-length TRA-1 might promote the expression of genes that activate spermatogenesis, just as full-length Gli proteins and Cubitus interruptus also work as activators [11], [12]. By identifying a co-factor required for the expression of fog-3, our current studies go farther. TRR-1 activity is needed for the expression of fog-3, but the loss of TRR-1 has no effect when TRA-1 is missing. The simplest explanation is that TRR-1 and the Tip60 HAT complex work with TRA-1 to promote the expression of fog-3, and that without Tip60 this activity is lost (Fig. 5). Because TRA-1 represses fog-3 in other circumstances [4], this bipotential regulation resembles the control of dpp by cleaved and full-length Cubitus interuptus [11]. Furthermore, Ci and Gli3 use CREB-binding protein as a co-activator [44], [45], a protein that also has intrinsic HAT activity.
We suspect that the specificity of the trr-1 phenotype in C. briggsae is due to two factors. First, the requirement for TRR-1 in embryogenesis is fully met by maternal supplies, so the homozygous mutants grow to adulthood. Second, TRA-1 only activates a few genes during development, so the trr-1 sex-determination defect is very specific.
Finally, two types of models could explain how TRR-1 and the Tip60 complex regulate gene expression. On the one hand, they might acetylate histones in target promoters, to open up chromatin conformation and promote transcription [28]. Indeed, several studies indicate that histone acetylation could be a part of the sex-determination process in nematodes. For example, NASP-1, a histone chaperone, and HDA-1, a histone deacetylase, work with TRA-4 to control sexual fates in C. elegans [46]. Moreover, TRA-1 interacts with SynMuv B genes in the development of the C. elegans vulva [47]; and many of these genes regulate chromatin structure [reviewed by 48]. Finally, a polymorphism in C. elegans NATH-10, an acetyltransferase, affects several reproductive traits, including the number of sperm made by hermaphrodites [49]. Although a broad survey of histone modifications in C. elegans did not uncover significant levels of acetylation in histones of the fog-3 promoter during larval development [50], only about 5% of the cells in these animals are becoming spermatocytes, so these experiments might not have been sensitive enough to detect variation in the developing germ line.
On the other hand, HAT complexes sometimes directly acetylate transcription factors, and the TRA-1 homologs Gli1 and Gli2 are acetylated in mammalian cells [51]. Thus, the Tip60 complex might directly acetylate TRA-1, to promote the activation of specific targets. Distinguishing these two models will require new methods for purifying TRA-1 complexes.
During evolution, some changes that accumulate in underlying regulatory pathways do not affect the phenotype [52]. In microevolution, this can produce populations with subtle differences in their regulatory architecture. For example, genetic variation that affects sex determination in C. elegans was revealed by different responses to the weak tra-2 allele ar221 [53], and genetic variation that affects vulval development was detected with a variety of weak mutations in the Ras pathway [54]. In macroevolution, this process can lead to species where significant regulatory differences underlie similar phenotypes. For example, the nematode vulva is induced by an EGF signal in C. elegans, but by a Wnt signal in Pristionchus pacificus [55], a distinction that involves many regulatory changes [56]. Because of these effects, mutations in orthologous genes sometimes have slightly different phenotypes in related species. For example, the Axin homolog PRY-1 plays similar roles in the development of the C. elegans and C. briggsae vulva, but the pry-1 mutant phenotypes in these species are not identical [38].
In theory, the internal constraints found in bipotential switches could prevent the accumulation of such changes. Indeed, several aspects of the TRA-1 switch have remained stable during Caenorhabditis evolution. For example, null mutants of tra-1 have similar phenotypes in both C. elegans and C. briggsae [15], [16], [30]. Moreover, the fog-3 gene is a conserved target of TRA-1 that promotes spermatogenesis in each species [4], [34]. Finally, FEM-3 not only regulates TRA-1 activity, but can influence germ cell fates by acting on a separate, downstream target in each species [57], [58]. Under these constraints, what types of regulatory change are possible?
We addressed this issue by studying mutations in genes that control the activating and repressing activities of TRA-1, and found that the relative importance of the Tip60 HAT complex and the TRA-2/FEM degradation pathway has shifted during recent evolution. In C. elegans, null alleles in the fem genes transform XX animals into females [57], but a null allele of trr-1 does not [29]. By contrast, null alleles have the opposite effects in C. briggsae. However, double mutants using weak alleles cause synthetic feminization in both species.
Although variation in sex-determination genes exists in nematode populations [59], there is no evidence it influences sexual development, and we observed no unexpected genetic interactions when mapping trr-1 mutations using different wild isolates. Thus, we believe that these differences between C. elegans and C. briggsae do not involve intra-specific variation, but instead represent unique solutions to the problem of hermaphrodite development.
We propose that the ancestor of these nematodes used Tip60 to regulate TRA-1's activator function, and the FEM genes to control the levels of the TRA-1100 repressor (Fig. 5B). During the independent evolution of self-fertility in C. elegans and C. briggsae, changes within this switch helped promote fog-3 expression and spermatogenesis in the XX animals of each species. We infer that C. elegans primarily relied on increased activity of the FEM proteins to eliminate TRA-1100 repressor in the larval germ line. This change required the novel FOG-2 protein, which increases FEM activity by down-regulating the translation of tra-2 [24], [26]. By contrast, C. briggsae might have relied primarily on increasing the activating activity of TRA-1 and the Tip60 complex.
Finally, the bipotential nature of TRA-1 might have influenced the origin of self-fertility in nematodes. Hermaphrodites evolved from male/female ancestors several times in the genus Caenorhabditis [20], and this transition can be modeled in the laboratory [21]. By contrast, large groups like the insects have not produced hermaphrodites. We suggest that the mutually antagonistic roles of TRA-1 in the sperm/oocyte decision might help explain this difference. In male/female species of worms, full-length and cleaved TRA-1 both have to be present in XX females. If small changes in the regulation of TRA-1 changed the balance between its activating and repressing activities, they might cause XX animals to make sperm, one of the changes needed to produce hermaphrodites. Furthermore, the complex web of regulators that control TRA-1 might have increased the opportunity for mutations that affect this switch, causing XX animals to make both sperm and oocytes. Thus, the structure of this regulatory switch might bias nematodes towards the evolution of self-fertility.
Strains were maintained as described by Brenner [60]. Wild type strains were AF16 and HK104. Mutant alleles were: tra-1(nm2), tra-2(nm1), tra-2(nm9), tra-3(ed24ts) [30], fem-1(nm27), fem-3(nm63) [22], dpy(nm4) [30], and cby-15(sy5148) (P. Sternberg, personal communication).
The tra-1(v56) allele was isolated while screening for she-1 mutations. Homozygous XX mutants have male bodies but make oocytes. The lesion is a 7 bp deletion, removing nucleotides 1244 to 1250 of the tra-1a coding region. Because it causes a frame shift and early stop, it was used in western blots, and no v56 protein was detected.
We counted self-progeny from cby-15 trr-1(v76)/+ + mothers, and observed 1044 wild type, 316 Cby Fog, 2 Fog, and 8 Cby hermaphrodites, implying a distance of 0.73 cM. We also counted self- progeny from trr-1(v76) dpy(nm4)/+ +, and observed 2303 WT, 698 Dpy Fog, 11 Fog, and 10 Dpy hermaphrodites, implying a distance of 0.69 cM.
Two mutants were isolated in screens for F2 females [17]. In the non-complementation screen (Fig. S1), we treated cby-15 tra-2/dpy(nm4) trr-1(v76) animals with 30 µg/ml trimethyl psoralen (TMΨ) followed by UV irradiation [61], or with 0.5% ethylmethane sulfonate [EMS, 60]. The P0 animals were transferred to new plates daily, and their F1 progeny screened for non-Dpy females. Each identified F1 female was crossed with wildtype males, and their F2 hermaphrodite progeny grown on individual plates. Those segregating F3 Cby Fog pseudo-males were identified and crossed with wildtype males. Finally, we screened the progeny of these crosses for recombinant Cby Fog animals that had lost the tra-2 mutation, and backcrossed each at least ten times.
We mapped v76 with SNPs that differ between the strains AF16 and HK104, as described [17]. Cby non-Fog recombinants were identified among the progeny of trr-1 cby-15 [AF16]/+ + [HK104] mothers, and Dpy non-Fog recombinants from dpy(nm4) trr-1 [AF16]/+ + [HK104] mothers. After establishing homozygous lines, we used the PCR to amplify and score DNA near each SNP from each of the recombinants, using primers in Table S4B [see www.briggsae.org/polymorph.php and 62].
Double strand RNA (dsRNA) was prepared as described [17]. Primers to clone the templates are listed in Table S5. RNA interference was performed by injection of young adults, using solutions of 1 mg/ml dsRNA.
PCR products were amplified with primers in Table S4A, purified on QIAquick PCR purification columns (Qiagen), and sequenced. To identify trr-1 lesions, we used genomic DNA templates.
At least 20 animals of each genotype were examined, using Nomarski microscopy.
For each genotype, groups of five L4 or young adult worms were collected and processed for RT-PCR as described [17]. Two independent samples were used to confirm reproducibility. PCR reactions were run for 32 cycles, using primers listed in Table S4C, and the conditions were shown to be in the linear range for fog-3 by testing serial dilutions of template.
Samples were made from 300 animals that were hand picked and added to 30 µl 2× SDS-PAGE buffer and boiled for 10 minutes. SDS-PAGE was run using the Bio-Rad mini-protein system II. Half of the sample was loaded to each lane, unless indicated otherwise. Gels were transfered to Immun-Blot PVDF membranes (Bio-Rad #162-0177) and blocked with 5% non-fat milk. Anti-TRA-1 antibody was a gift from Dr. David Zarkower, and was diluted to 1∶1000–2000 in 5% non-fat milk. Secondary antibody was HRP Goat anti-rabbit (Bio-Rad #170-5046), and was diluted to 1: 10000 in 5% milk. Finally, the Immun-Star HRP chemiluminescence kit (Bio-Rad) was used for detection.
|
10.1371/journal.pgen.1005866 | Angiotensin Converting Enzyme (ACE) Inhibitor Extends Caenorhabditis elegans Life Span | Animal aging is characterized by progressive, degenerative changes in many organ systems. Because age-related degeneration is a major contributor to disability and death in humans, treatments that delay age-related degeneration are desirable. However, no drugs that delay normal human aging are currently available. To identify drugs that delay age-related degeneration, we used the powerful Caenorhabdtitis elegans model system to screen for FDA-approved drugs that can extend the adult lifespan of worms. Here we show that captopril extended mean lifespan. Captopril is an angiotensin-converting enzyme (ACE) inhibitor used to treat high blood pressure in humans. To explore the mechanism of captopril, we analyzed the acn-1 gene that encodes the C. elegans homolog of ACE. Reducing the activity of acn-1 extended the mean life span. Furthermore, reducing the activity of acn-1 delayed age-related degenerative changes and increased stress resistance, indicating that acn-1 influences aging. Captopril could not further extend the lifespan of animals with reduced acn-1, suggesting they function in the same pathway; we propose that captopril inhibits acn-1 to extend lifespan. To define the relationship with previously characterized longevity pathways, we analyzed mutant animals. The lifespan extension caused by reducing the activity of acn-1 was additive with caloric restriction and mitochondrial insufficiency, and did not require sir-2.1, hsf-1 or rict-1, suggesting that acn-1 functions by a distinct mechanism. The interactions with the insulin/IGF-1 pathway were complex, since the lifespan extensions caused by captopril and reducing acn-1 activity were additive with daf-2 and age-1 but required daf-16. Captopril treatment and reducing acn-1 activity caused similar effects in a wide range of genetic backgrounds, consistent with the model that they act by the same mechanism. These results identify a new drug and a new gene that can extend the lifespan of worms and suggest new therapeutic strategies for addressing age-related degenerative changes.
| Age-related degeneration is a fundamental feature of animal biology and an important contributor to human disability and death. However, no medicines have been shown to delay human aging. To identify drugs that delay age-related degeneration, we screened FDA-approved compounds and discovered that the hypertension drug captopril significantly extended C. elegans lifespan. In humans, captopril inhibits angiotensin converting enzyme (ACE) to regulate blood pressure. The C. elegans homolog of ACE is encoded by the acn-1 gene. We discovered that reducing the activity of acn-1 also caused a robust extension of lifespan and delayed age-related changes in C. elegans. Captopril and acn-1 have a similar mechanism of action; both treatments displayed similar interactions with previously characterized pathways, and combining treatment with captopril and reducing the activity of acn-1 did not have an additive effect on life span extension. These results identify a new drug and a new gene that influence aging in C. elegans. They may be relevant to other animals such as humans because the pathway that includes ACE has been conserved during evolution. These findings establish a foundation for possible therapeutic interventions that can delay age-related degeneration.
| Animal aging is characterized by progressive, degenerative changes of tissue structure and function. In humans, these changes have profound negative effects on health by causing morbidity and mortality. An important goal of aging research is to identify interventions that can delay age-related degeneration and promote an extended period of vitality or healthspan. However, no interventions have been demonstrated to delay human aging. By contrast, a growing number of interventions have been demonstrated to delay age-related degeneration and extend lifespan in model animals such as worms, flies and mice [1]. These interventions include dietary changes such as caloric restriction, genetic changes such as reducing the activity of the insulin/insulin-like growth factor-1 (IGF-1) signaling pathway, and drugs such as rapamycin. These studies indicate that pathways that influence aging have been conserved during animal evolution [1]. Thus, model organisms are promising systems to identify and characterize interventions that promote healthy aging and may be beneficial in humans.
The terrestrial nematode Caenorhabditis elegans has emerged as an outstanding model organism for studies of aging. The biology of these animals is well suited for studies of aging because they have a rapid life cycle and a relatively short adult lifespan of about 15 days [2,3]. A wide variety of age-related degenerative changes have been documented, providing assays of aging and suggesting C. elegans undergoes mechanisms of aging similar to larger animals where progressive degenerative changes are well characterized [4]. Powerful experimental techniques are well established, including forward and reverse genetic approaches and molecular approaches facilitated by a fully sequenced genome [5,6]. C. elegans are well suited for pharmacological studies because they ingest compounds that are added to the culture medium. Molecular genetic studies have identified and characterized several pathways that substantially influence the rate of age-related degeneration. The insulin/IGF-1 pathway was first implicated in aging biology in C. elegans and has now been shown to play a conserved role in other animals, including flies and mammals [1]. Mutations that reduce the activity of the daf-2 insulin receptor or the age-1 phosphatidylinositol-3-OH (PI3) kinase substantially extend the adult lifespan, indicating that insulin/IGF-1 pathway activity promotes a rapid lifespan [7,8]; these mutant animals also display enhanced resistance to a variety of stresses such as UV light, oxidation, transition metals, and hypoxia [9–12]. A critical effector of the daf-2/age-1 pathway is the forkhead transcription factor DAF-16, which is activated and localized to the nuclei by low levels of daf-2 signaling [13,14]. The activity of daf-16 promotes an extended lifespan, and daf-16 is necessary for the extension of lifespan caused by mutations of daf-2 and age-1 [8,15]. Caloric restriction extends the lifespan of a wide range of organisms, including C. elegans, indicating that ad libitum feeding promotes a rapid lifespan. Mutations of genes that are necessary for pharyngeal pumping and food ingestion, such as eat-2, cause a substantial lifespan extension [16]. Mutations in multiple genes that are necessary for mitochondrial function, such as isp-1, cause a lifespan extension, indicating that wild-type levels of mitochondrial activity promote a rapid lifespan [17,18]. In addition to genetic approaches, C. elegans is emerging as a valuable system for pharmacological approaches that can be used to identify and characterize drugs that influence aging. Compounds that influence C. elegans aging have been identified by screening approaches and by testing candidate drugs based on a known mechanism of action [19–25].
To identify drugs that influence aging, we screened FDA-approved drugs for the ability to extend the lifespan of C. elegans hermaphrodites. Here we report that captopril, an angiotensin converting enzyme (ACE) inhibitor used to treat high blood pressure, extended mean lifespan. ACE is a protease that functions in a signaling cascade that is initiated by low blood pressure; in humans, ACE converts angiotensin I to angiotensin II, and angiotensin II binds the AT1 receptor, resulting in increased contractility of endothelial cells and thereby increasing blood pressure [26,27]. ACE inhibitors such as captopril are used by a large number of people to control hypertension [27]. The ACE gene has been conserved from bacteria to mammals, indicating it had a primordial function before the evolution of a closed circulatory system that creates blood pressure. The C. elegans homolog of ACE is encoded by the acn-1 gene; acn-1 is necessary for larval molting but has not been previously implicated in adult longevity [28]. We hypothesized that captopril inhibits acn-1 to extend lifespan, and here we present experimental evidence that supports this model. First, inhibition of the acn-1 gene by RNA interference extended lifespan and delayed age-related degenerative changes, indicating that acn-1 activity influences aging and longevity. Second, captopril treatment and reducing the activity of acn-1 caused very similar effects in a wide range of genetic backgrounds, indicating that these interventions have a common mechanism. Third, the lifespan extensions caused by captopril treatment and reducing the activity of acn-1 were not additive, indicating that these interventions may affect the same pathway. These results identify captopril as a new, FDA-approved drug that can extend the lifespan of C. elegans and acn-1 as a new gene that influences C. elegans aging. The findings establish acn-1 as the target of captopril in worms, connecting a pharmacological intervention that extends lifespan to its direct molecular target. In mammals, ACE regulates blood pressure, indicating there is a link between a system that controls aging in worms and physiology in mammals.
To identify drugs that influence aging, we selected 15 compounds that are Food and Drug Administration (FDA)-approved for human use, have known effects on human physiology, and represent different functional or structural classes (see Materials and Methods). Compounds were added to NGM agar at three different concentrations, and the lifespan of C. elegans hermaphrodites cultured at 20°C with E. coli OP50 as a food source was determined. We previously described a similar screening approach that was used to identify the lifespan extending compounds ethosuximide and valproic acid [19,20]. Captopril, an ACE inhibitor, caused a significant extension of lifespan (Fig 1A and 1C). To identify the optimal concentration for lifespan extension, we performed a dose–response analysis. A concentration of 2.5mM captopril in the medium caused the greatest lifespan extension, whereas concentrations of 1.9mM and 3.2mM caused smaller extensions (Fig 1B; Table 1, line 1–4). At the optimal concentration of 2.5mM, captopril treatment caused a significant 23% extension of mean adult lifespan and a significant 18% extension of maximum adult lifespan (Fig 1C; Table 1, line 5–6). We define maximum adult lifespan as the average lifespan of the 10% of the population that are longest lived. To determine the developmental stage when captopril functions to extend lifespan, we administered the drug beginning at the L4 larval stage. The drug was effective with this time of administration suggesting captopril functions in adults to delay age-related degeneration. To determine the temperature dependence of captopril, we analyzed animals cultured at 15°C, 20°C and 25°C. Captopril significantly extended the mean and maximum adult lifespan at all three temperatures, indicating that the effect is not temperature dependent (Fig 1D; Table 1, line 7–10; S1A Fig).
These experiments were conducted with live E. coli as a food source, raising the possibility that captopril may directly affect bacteria and indirectly affect worms. There are precedents for such a mechanism, since antibiotics can extend C. elegans lifespan by reducing the pathogenicity of bacteria [29–31], and the anti diabetic drug metformin was reported to extend C. elegans lifespan by altering bacterial folate and methionine metabolism [32]. To determine if the effects of captopril are mediated by an effect on live bacteria, we conducted the life span experiment with bacteria killed by exposure to ultraviolet light. Captopril extended the lifespan of C. elegans in these conditions, demonstrating that the mechanism of captopril-mediated lifespan extension is not dependent on live E. coli (S1C Fig; Table 1, line 11–12).
A large number of age-related degenerative changes have been characterized in C. elegans, including declines of physiological processes, such as body movement, pharyngeal pumping, and egg-laying, and changes in morphology, such as loss of tissue integrity [4,33,34]. Treatment with captopril caused a small delay in the age-related decline in pharyngeal pumping rate, although the change was not statistically significant with the sample size analyzed (S2 Fig). Several genetic manipulations that extend adult lifespan also affect reproduction. For example, caloric restriction and defects in insulin/IGF-1 signaling reduce total progeny production and increase reproductive span in self-fertile hermaphrodites [35]. To determine how captopril affects reproduction, we monitored progeny production of self-fertile hermaphrodites daily. Captopril did not significantly affect total brood size or reproductive span of self-fertile hermaphrodites (S3A and S3B Fig).
Captopril treatment in humans reduces blood pressure by inhibiting the activity of angiotensin converting enzyme (ACE) [36]. Therefore, we hypothesized that captopril treatment in C. elegans extends longevity by inhibiting the worm homolog of ACE. To investigate this hypothesis, we analyzed the acn-1 gene because it encodes a predicted protein that is most similar to human ACE [28]. To reduce the activity of acn-1, we used RNA interference (RNAi) [37]; worms were fed bacteria expressing dsRNA from the acn-1 gene, which is predicted to reduce the levels of the acn-1 transcript. Wild-type animals cultured with acn-1 RNAi beginning at the embryonic stage displayed a significant extension of mean and maximum lifespan of 21% and 18%, respectively (Fig 2A, Table 2, line 1–2). These results indicate that acn-1 activity is necessary to promote a rapid lifespan. To investigate the time of action of acn-1, we initiated the exposure to acn-1 RNAi at the L4 larval stage. Exposure only during adulthood caused a similar extension of mean and maximum lifespan of 22% and 20%, respectively, indicating that acn-1 functions in adults to promote a rapid lifespan (Fig 2B, Table 2, line 3–4).
Several mutations have been identified that increase the sensitivity of worms to feeding RNAi, including mutation of rrf-3 [38]. Feeding acn-1 RNAi bacteria to rrf-3 mutant animals beginning at the embryonic stage caused a significant increase of mean and maximum lifespan of 33% and 24%, respectively (Fig 2C, Table 2, line 5–6). Similarly, feeding acn-1 RNAi beginning at the L4 stage caused a significant extension of mean and maximum lifespan of 46% and 33%, respectively (Fig 2D, Table 2, line 7–8). The extensions caused by acn-1 RNAi in the rrf-3 background were greater than the extensions in the wild-type background, indicating that rrf-3 mutant animals are indeed more susceptible to the effect of the RNAi treatment. Moreover, acn-1 RNAi also caused a significant extension of mean and maximum lifespan of rrf-3 mutant animals at 25°C (S1B Fig, Table 2, line 9–10). To quantify how acn-1 mRNA levels are affected by feeding RNAi, we performed quantitative RT-PCR. acn-1 RNAi reduced mRNA levels about 50% compared to control RNAi in rrf-3 mutant animals (S4 Fig).
To characterize how acn-1 influences age-related degeneration, we monitored age-related declines of major physiological processes. Wild-type C. elegans hermaphrodites display coordinated, sinusoidal body movement as young adults, and the frequency and coordination of body movement display age-related declines. To analyze body movement quantitatively, we counted body bends on solid NGM using a dissecting microscope. Hermaphrodites cultured with acn-1 RNAi displayed a significantly higher rate of body movement beginning on day 4 of adulthood and extending to day 26 of adulthood (Fig 3A). To illustrate this difference, we exploited the fact that worms leave tracks in the bacterial lawn as they move. Five animals on day 15 of adulthood were transferred to fresh bacterial lawns, allowed to move for two hours, and the lawns were photographed. Fig 3B shows that hermaphrodites treated with control RNAi left a small number of tracks, and the tracks are suggestive of uncoordinated movement. By contrast, hermaphrodites treated with acn-1 RNAi left abundant tracks that were suggestive of coordinated sinusoidal movement.
We monitored the age-related decline in pharyngeal pumping rate quantitatively by direct observation using a dissecting microscope. Hermaphrodites treated with acn-1 RNAi displayed higher rates of pharyngeal pumping on days 12–20 of adulthood (Fig 3C). These results demonstrate that acn-1 is necessary to promote the rapid, age-related decline of body movement and pharyngeal pumping observed in wild-type animals.
To analyze the effect on reproduction, we monitored progeny production of self-fertile hermaphrodites. Wild-type animals treated with acn-1 RNAi did not display significant changes in total self fertile brood size or the daily production of progeny (S2C and S2D Fig). Captopril treatment only slightly delayed age-related changes of pharyngeal pumping, whereas acn-1 RNAi significantly delayed age-related changes of pharyngeal pumping and body movement, suggesting acn-1 RNAi may reduce the activity of acn-1 to a greater extent than captopril or the drug may have toxic effects.
Several C. elegans mutations that extend longevity also increase stress resistance [39,40]. To investigate the function of acn-1 in stress resistance, we analyzed heat and oxidative stress. Embryos were cultured at 20°C with control RNAi or acn-1 RNAi, and after 3 days animals were transferred to stressful conditions and monitored for survival. When exposed to continuous 34°C heat stress, control animals displayed a time dependent decrease in survival with a mean lifespan of 14.0 hours; animals treated with acn-1 RNAi displayed a significant, 12% extension of mean lifespan of 15.7 hours (Fig 4A, Table 3, line 1–2). When exposed to oxidative stress caused by 40mM paraquat, control animals displayed a time dependent decrease in survival with a mean lifespan of 47.8 hours; animals treated with acn-1 RNAi displayed a significant, 16% extension of mean lifespan of 55.4 hours (Fig 4B, Table 3, line 3–4). In addition, we observed a similar result of extended survival in oxidative stress when wild-type animals were treated with acn-1 RNAi (Fig 4C Table 3, line 5–6). To determine if the specific conditions or oxidation generating chemical are important for the results, we analyzed oxidative stress in liquid medium using the compound juglone to cause oxidative stress. After nine hours of juglone exposure, animals treated with acn-1 RNAi displayed a significant, 56% increase in survival compared to control animals (Fig 4D Table 3, line 7–8). These results indicate that the acn-1 gene is necessary to promote wild-type levels of sensitivity to multiple stresses including heat and oxidation.
To investigate the mechanism of action of captopril and acn-1 in lifespan extension, we analyzed how captopril treatment and acn-1 RNAi affects animals with mutations that alter longevity. Caloric restriction extends the lifespan of many organisms, indicating that ad libitum feeding during laboratory culture reduces longevity. Mutations of the eat-2 gene impair pharyngeal pumping, reduce food intake and cause a lifespan extension [16,41]. Captopril significantly extended the mean and maximum lifespan of eat-2(ad1116) mutant animals by 14% and 17%, respectively (Fig 5A; Table 1, line 13–14). Similarly, acn-1 RNAi significantly extended mean and maximum lifespan by 12% and 10%, respectively (Fig 6A, Table 2, line 11–12). Thus, the lifespan extension caused by caloric restriction was additive with captopril treatment and acn-1 RNAi.
Mutations of several genes that are important for mitochondrial function cause a lifespan extension in C. elegans, indicating that normal mitochondrial function promotes a rapid lifespan. The isp-1 gene encodes a iron sulfur cluster containing protein that is important for the function of complex III to catalyze electron transport from ubiquinol to cytochrome c, and isp-1 mutations extend lifespan [17,42]. Captopril treatment significantly extended the mean and maximum lifespan of isp-1(qm150) mutant animals by 23% and 20%, respectively (Fig 5B; Table 1, line 15–16). Similarly, acn-1 RNAi significantly extended mean and maximum lifespan by 14% and 13%, respectively (Fig 6B, Table 2, line 13–14). Thus, the lifespan extension caused by reducing mitochondrial function was additive with captopril treatment and acn-1 RNAi.
Overexpression of SIR2 (silent information regulator 2) is reported to extend the lifespan of several organisms, although this effect is not always observed [43,44]. In C. elegans, sir-2.1 is predicted to encode a nicotinamide adenine dinucleotide (NAD) dependent deacetylase that can extend the lifespan of C. elegans when overexpressed. We examined the null mutation sir-2.1(ok434) [45]. Captopril treatment significantly extended the mean and maximum lifespan of sir-2.1(ok434) mutant animals by 17% and 27%, respectively (Fig 5C; Table 1, line 17–18). Similarly, acn-1 RNAi significantly extended the mean and maximum lifespan of sir-2.1 mutants by 16% and 18%, respectively (Fig 6C, Table 2, line 15–16). Thus, sir-2.1 activity was not necessary for the lifespan extension activity of captopril or acn-1 RNAi.
The target of rapamycin (TOR) signaling network plays an important role in nutrient homeostasis and influences adult lifespan [46]. Loss-of-function mutations in rict-1 affect TOR signaling and cause a shorter lifespan. acn-1 RNAi significantly extended the mean and maximum lifespan of rict-1 mutants by 23% and 25%, respectively (Fig 6G, Table 2, line 17–18). hsf-1 encodes a transcription factor that is important for stress response; overexpression of hsf-1 extends lifespan and delays age-related protein miss folding [47], whereas reducing the activity of hsf-1 causes a shorter lifespan and proteotoxicity [31,48,49]. acn-1 RNAi significantly extended the mean and maximum lifespan of hsf-1(lf) mutants by 12% and 11%, respectively (Fig 6H, Table 2, line 19–20). Thus, the activities of rict-1 and hsf-1 were not necessary for the lifespan extension caused by acn-1 RNAi.
Mutations in the insulin/insulin-like growth factor (IGF) signaling pathway influence C. elegans lifespan [7,8,50–53]. Mutations that partially reduce the activity of daf-2, which encodes a protein homologous to the vertebrate insulin/IGF-1 receptor, or age-1, which encodes a protein homologous to the vertebrate PI3 kinase, extend lifespan. This signaling pathway controls the activity of a FOXO transcription factor encoded by daf-16, and daf-16 activity is necessary for the lifespan extension caused by mutations in upstream signaling genes [13,14]. Thus, daf-2 and age-1 activity promote a rapid lifespan and inhibit longevity, whereas daf-16 activity promotes longevity. Captopril treatment significantly extended the mean and maximum lifespan of daf-2(e1370) partial loss-of-function mutant animals by 11% and 9%, respectively (Fig 5D; Table 1, line 21–22). Similarly, acn-1 RNAi significantly extended mean and maximum lifespan by 11% and 9%, respectively (Fig 6D, Table 2, line 23–24). In combination with an age-1(hx546) partial loss-of-function mutation that causes an extended lifespan, acn-1 RNAi significantly extended mean and maximum lifespan by 30% and 19%, respectively (Fig 6F, Table 2, line 25–26). A similar result was obtained by analyzing age-1(am88) mutant animals (Table 2, line 27–28, S5 Fig) [54]. Thus, the lifespan extension caused by reducing daf-2 activity was additive with captopril treatment and acn-1 RNAi, and the lifespan extension caused by reducing age-1 activity was additive with acn-1 RNAi.
By contrast, captopril treatment did not extend the lifespan of daf-16 (mu86) loss-of-function mutant animals, but rather significantly shortened the mean and maximum lifespan by 10% and 11%, respectively (Fig 5E; Table 1, line 19–20). Similarly, acn-1 RNAi slightly shortened the mean and maximum lifespan by 3% and 2%, respectively (Fig 6E, Table 2, line 21–22). These findings indicate that the lifespan extension activity of captopril and acn-1 RNAi require daf-16 activity; however, the reduction of lifespan raises the possibility that the combination of captopril treatment or acn-1 RNAi and the daf-16 mutation causes toxicity.
To investigate the possibility that acn-1 functions upstream of daf-16, we analyzed additional phenotypes associated with the insulin/IGF-1 pathway. Upstream signaling proteins such as DAF-2 control the activity of DAF-16; specifically, daf-2(lf) mutations that cause a lifespan extension also cause DAF-16 protein to localize to the nucleus, where DAF-16 controls the activity of target genes [55]. To examine the nuclear localization of DAF-16, we used transgenic worms containing a DAF-16::GFP reporter construct [56]. Animals treated with captopril or acn-1 RNAi did not display a substantial nuclear localization of DAF-16::GFP compared to control animals (S6C Fig, S1 Table). Thus, acn-1 RNAi did not cause the same effect as a daf-2(lf) mutation, and acn-1 is not necessary to inhibit nuclear localization of DAF-16. It has been proposed that daf-16 is regulated by additional mechanisms that do not involve changes in subcellular localization, such as transcript levels [57], EAK-7 [58], and phosphorylation [59,60]. Our results do not exclude the possibility that captopril or acn-1 RNAi regulate daf-16 in a manner that does not change nuclear localization.
daf-2(lf) mutations cause a dauer constitutive (Daf-c) phenotype, indicating that daf-2 is necessary to inhibit dauer development. To analyze the role of acn-1 in dauer formation, we cultured worms with acn-1 RNAi bacteria at 20°C, shifted embryos to 27°C to stimulate dauer formation, and scored dauer larvae after 72 hours. acn-1 RNAi did not increase the frequency of dauer formation compared to control RNAi in wild-type animals or rrf-3 mutant animals (S6A Fig). To increase the sensitivity of the assay, we analyzed the function of acn-1 in daf-2(lf) mutants that display a partially penetrant, temperature sensitive Daf-c phenotype [61]. acn-1 RNAi was not different from control RNAi in this assay (S6B Fig). Thus, acn-1 RNAi did not cause the same effect on dauer formation as a daf-2(lf) mutation, and acn-1 is not necessary to inhibit formation of dauer larvae.
We hypothesized that captopril inhibits acn-1 to extend lifespan. This hypothesis predicts that the effects of captopril and acn-1 RNAi will not be additive, because our dose-response analysis indicates that we have identified the optimal dose of captopril for lifespan extension. To test this prediction, we combined treatment with captopril and acn-1 RNAi. Captopril treatment alone caused a 20% extension of mean lifespan to 18.7 days, whereas acn-1 RNAi alone caused a 38% extension of mean lifespan to 21.5 days (Fig 7A, Table 4, line 1–3). Combining captopril treatment and acn-1 RNAi resulted in a 18.9 day lifespan that was not significantly different from captopril treatment alone and significantly shorter than acn-1 RNAi treatment alone. Thus, captopril and acn-1 RNAi did not have an additive effect on lifespan extension, consistent with the model that both effects are mediated by a similar mechanism.
The identification of compounds that can delay age-related degeneration and extend lifespan is an important goal of aging research, because age-related decline is a major cause of disability and death in humans, and so far no compounds have been demonstrated to delay human aging. We reasoned that FDA-approved drugs used to treat human diseases might also influence aging and lifespan. To identify such drugs, we screened examples of different structural and functional drug classes. We previously described the identification of anticonvulsant drugs such as ethosuximide and the neuroactive drug valproic acid [19,20]. Here we identified the blood pressure medicine captopril as a way to extend C. elegans lifespan. The effect of captopril was dose dependent; at an optimal dose, captopril significantly extended mean lifespan 22–28% and maximum lifespan 18–32%. Captopril extended lifespan at a variety of temperatures and in a variety of mutant backgrounds, indicating that the effect is robust in the face of environmental and genetic variation. Captopril functioned in adult animals to extend lifespan, suggesting that it affects the rate of age-related decline rather than developmental processes.
The first of what is now a large class of ACE inhibitors, captopril is an oligopeptide derivative developed in 1975 based on a peptide found in pit viper venom [62]. ACE inhibitors modulate the renin-angiotensin-aldosterone system, a mechanism by which the body adapts to hypotension [63]. In response to a decline in blood pressure, the kidney releases renin, which cleaves angiotensinogen to angiotensin I. ACE converts angiotensin I to angiotensin II, and angiotensin II acts through a transmembrane receptor to stimulate aldosterone secretion and promote vasoconstriction to increase blood pressure. By blocking ACE and preventing the conversion of angiotensin I to angiotensin II, captopril lowers blood pressure.
Two strategies have been used to identify compounds that can extend C. elegans lifespan: screening chemical libraries and testing candidate compounds based on the hypothesis that the target of the drug may influence aging and longevity [64]. Compounds that have been tested included FDA-approved drugs, libraries of chemically defined molecules, and extracts of plants that contain a mixture of chemicals. Library screening resulted in the identification of antidepressant drugs [21,25]. Candidate compounds that have been reported to extend worm lifespan include resveratrol [65], trehalose [24], lithium [66] and garlic constituent [67]. Extracts of blueberries and ginkgo have been reported to extend worm lifespan [68,69]. ACE inhibitors such as captopril have not been previously reported to extend lifespan in worms, so our findings identify a new chemical entity that influences aging in C. elegans.
It is well established that ACE is the target that mediates the effect of captopril on blood pressure in humans [62]. ACE genes have been highly conserved during evolution, and acn-1 encodes the C. elegans homolog of ACE [28]. A major issue in aging pharmacology is the identification of the direct target of the drug, and in most cases the targets of drugs that extend C. elegans lifespan remain unknown. We hypothesized that captopril inhibits ACN-1 to extend longevity. This hypothesis makes three important predictions that were verified experimentally. First, it predicts that reducing the activity of acn-1 using genetic techniques can extend longevity. We showed that targeting acn-1 by RNAi increased mean lifespan 20–46% and maximum lifespan 18–33%. Second, it predicts that reducing the activity of acn-1 and treatment with captopril will cause similar effects in a variety of genetic backgrounds. Indeed, captopril treatment and reducing acn-1 activity gave very similar results in five genetic backgrounds (eat-2, isp-1, sir-2.1, daf-16 and daf-2) and at two temperatures. In addition, both treatments function in adults to extend longevity. Third, it predicts that the lifespan extension caused by captopril treatment and reducing acn-1 activity will not be additive. This prediction was also verified. While these results are consistent with captopril inhibition of ACN-1, they do not demonstrate that the drug directly binds ACN-1 protein or inhibits a biochemical activity of ACN-1. The biochemical activity of ACN-1 has not been established, and ACN-1 may not have protease activity because critical residues in the predicted active site have not been conserved during evolution [28]. Further studies are necessary to establish an assay for the biochemical activity of ACN-1 and directly test the effect of captopril.
The expression and function of acn-1 were analyzed by Brooks et al. [28] using a reporter gene encoding ACN-1::GFP and acn-1 RNAi, respectively. acn-1 is expressed in embryonic and larval hypodermis, in the vulva during organogenesis and in the ray papillae of the male tail. RNAi delivered by injection in the gonad caused larvae to arrest at the L2 stage and display evidence of molting defects. RNAi delivered by feeding to L1/L2 larvae caused a cuticle defective phenotype in L3/L4 larvae and adults. The failure to shed cuticle led to secondary defects such as vulva defects and constipation. These results indicate acn-1 is necessary for larval molting, mail tail development and formation of adult alae. Frand et al. [70] identified acn-1 in a genome-wide feeding RNAi screen for molting defects. An ACN-1::GFP transgene was expressed in the hypodermis, including the major body syncytium, hyp7, and hypodermal cells in the head and tail, the lateral seam cells, and the excretory gland cell. Neither of these studies describe aging phenotypes, so our results establish a new phenotype for acn-1 and a novel link between acn-1 and aging. The previously reported molting defects caused by acn-1 RNAi are partially penetrant [28,70]; we did not observe a significant penetrance of molting defects, which may indicate less extreme gene disruption in our studies resulting from differences between the feeding RNAi constructs or the conditions of RNAi delivery.
To elucidate the role of captopril and acn-1 in aging, we analyzed interactions with established pathways that influence longevity. Many mutations used in these experiments are not null alleles, and therefore the observation that the effects are additive does not exclude the possibility that two interventions act in the same pathway. Captopril treatment or reducing the activity of acn-1 was additive with the lifespan extensions caused by an eat-2 mutation that causes caloric restriction. Furthermore, these treatments did not reduce self-fertile brood size and reproductive span like caloric restriction, suggesting that captopril and acn-1 do not act by causing caloric restriction. Captopril treatment or reducing the activity of acn-1 was additive with the lifespan extensions caused an isp-1 mutation that reduces mitochondrial activity, suggesting these treatments do not reduce mitochondrial function. The lifespan extension caused by reducing the activity of acn-1 was not abrogated by loss-of-function mutations of sir-2.1, hsf-1 or rict-1, suggesting that acn-1 does not act by regulating these genes. Captopril treatment and reducing the activity of acn-1 displayed complex interactions with the insulin/IGF-1 pathway. These treatments were additive with the lifespan extensions caused by loss-of-function mutations of daf-2 and age-1. However, the lifespan extensions caused by both treatments were abrogated by a daf-16 mutation. To further analyze the relationship with daf-16, we demonstrated that reducing the activity of acn-1 did not cause dauer formation and did not promote nuclear localization of DAF-16, which are typical of reducing insulin/IGF-1 signaling upstream of daf-16. Thus, acn-1 does not appear to act upstream and regulate the nuclear localization activity of daf-16. It is possible that daf-16 is necessary because it functions in parallel to acn-1 or that toxicity develops in the absence of both daf-16 and acn-1. Overall, acn-1 defines a new gene that influences longevity, and interactions with known longevity pathways suggest that it functions by a mechanism that is distinct from those that have been characterized previously.
The ACE inhibitor enalapril and the angiotensin II receptor antagonist losartan have been reported to extend the life span of mice and rats [71–78]. Furthermore, these drugs delay the age-related degeneration of tissue structure and function in the kidney, cardiovascular system, liver and brain. Similarly, Santos et al., [79] showed that enalapril increased life span in rats. These interesting results indicate that the renin-angiotensin-aldosterone system promotes age-related degeneration, and blocking this system can extend longevity in rodents. The mechanism of these drugs in life span extension is not well defined–the affects are not well correlated with changes in blood pressure but may reflect preservation of mitochondrial number and function. Genetic studies reported by Benigni et al., [80] provide important support for these pharmacology studies, since disruption of the angiotensin II type I receptor (AT1) promotes longevity in mice. These results may be relevant to humans, since polymorphisms in the angiotensin II type I receptor gene are associated with extreme human longevity [81]. Overall, these studies suggest that the rennin-angiotensin-aldosterone system controls longevity in mammals. Thus, our discoveries in worms are likely to be relevant to mammalian biology. An important issue that has not been established by studies of mammals is the mechanism of action of this pathway in influencing aging and longevity. The results presented here provide new insights into the mechanism of action of captopril in lifespan extension and establish the powerful C. elegans system to investigate critical questions about the conserved activity of the pathway.
C. elegans were cultured on 6 cm Petri dishes containing NGM agar and a lawn of Escherichia coli strain OP50 at 20°C unless stated otherwise [2]. The wild-type (WT) strain was N2 Bristol. daf-2(e1370P1465S) is a partial loss-of-function mutation that affects the kinase domain of the DAF-2 receptor tyrosine kinase [50]. age-1(hx546P806S) and age-1(am88E725K) are partial loss-of-function mutations that affect the AGE-1 PI3 kinase [53,54,82]. daf-16(mu86) is a strong loss-of-function mutation caused by a deletion in the DAF-16 forkhead transcription factor [13,14]; eat-2(ad1116) is a change in a splicing site predicted to decrease the level of mRNA of the EAT-2 non-alpha nicotinic acetylcholine receptor [41]; isp-1(qm150P225S) is a loss-of-function mutation that affects an iron sulfur protein of mitochondrial complex III [42]; sir-2.1(ok434) is a deletion that causes a loss-of-function of the SIR-2.1 NAD dependent protein deacetylase [45]. rict-1(mg360G1067E) is a partial loss-of-function mutation of RICT-1, a component of the target of rapamycin complex 2 (TORC2) that encodes an ortholog of mammalian Rictor [83]. hsf-1(sy441W585stop) is a strong loss-of-function mutation of the HSF-1 transcription factor [84]. DAF-16 nuclear localization was analyzed using strain GR1352 containing the integrated array xrIs87 [DAF-16alpha::GFP::DAF-16B + rol-6(su1006)] [85]. The rrf-3(pk1426) mutation was used for RNAi feeding experiments [38].
Fifteen FDA-approved drugs were screened for extension of C. elegans lifespan using methods described by Evason et al., [19] (atropine, yohimbine hydrochloride, captopril, nicotinic acid, phenformin, haloperidol, acetazolam, adenosine, cimetidine, lidocaine, procainamide hydrochloride, caramazepine, 5’-5’-diphenylhydation Sodium, caffeine and imipramine). For each drug, we analyzed about 50 hermaphrodites cultured with three concentrations in the NGM medium (X, 10-100X, 1000X). The lowest dose (X) was approximately equivalent to the effective dose in humans [63]. Captopril was obtained from Sigma Aldrich (St. Louis, MO, USA), and a 30 mg/ml stock solution was prepared by dissolving the compound in water. Concentrated captopril was diluted to the desired final concentration in liquid NGM that had been autoclaved and cooled to 55°C, and 7–8 ml of medium was dispensed into 6 cm Petri dishes. Petri dishes were allowed to dry 1–2 days at room temperature and then seeded with E. coli OP50. Lifespan experiments using dishes containing drugs were always conducted in parallel with control dishes containing no drug in the same incubator to control for day-to-day variations in temperature and humidity.
Studies of lifespan were begun on day zero by placing approximately 30–40 L4 hermaphrodites on a Petri dish. Each hermaphrodite was transferred to a fresh Petri dish daily during the reproductive period (approximately the first seven days) to eliminate self-progeny and every 2–3 days thereafter. Each hermaphrodite was examined every day using a dissecting microscope for survival, determined by spontaneous movement or movement in response to prodding with a pick. Dead worms that displayed matricidal hatching, vulval extrusion or desiccation due to crawling off the agar were excluded from the data analysis. Average mean lifespan was calculated as the number of days from the L4 stage to the last day a worm was observed to be alive. To conduct experiments with dead bacteria, we seeded dishes with live E. coli OP50, cultured for 24 hours, and exposed the bacteria to ultraviolet light by placing dishes in a UV Stratalinker 2400 for 15 minutes. Death was confirmed by inoculating LB medium with treated bacteria and observing no growth.
To analyze progeny production, one L4 hermaphrodite was placed on a Petri dish (day one), transferred to a fresh dish daily until at least 4 days without progeny production, and progeny were counted after two days. Pharyngeal pumping and body movement were determined as described previously [33]. Briefly, we observed pharyngeal pumping using a dissecting microscope for a 10 seconds interval. Body movement was assayed by observation using a dissecting microscope for 20 seconds. Petri-dishes were tapped to stimulate animals to move before scoring.
RNAi interference was performed by feeding bacteria that express dsRNA as described by Kammath et al., [86]. Briefly, E. coli HT115 bacteria with the control plasmid (L4440) or a plasmid encoding acn-1 were obtained from the Ahringer library [37], and the identity of the clone was confirmed by DNA sequencing. The daf-2 RNAi bacterial strain was provided by M. Crowder. RNAi bacteria were streaked on LB dishes containing 50μg/ml ampicillin and 12.5 μg/ml tetracycline. Control and acn-1 RNAi cultures were grown for 6 hours in LB medium containing 50μg/ml ampicillin. Escherichia Coli expressing double-stranded acn-1 RNA did not form thick lawns on RNAi NGM agar dishes containing isopropyl β-d-1-thiogalactopyranoside (1 mM) and 50μg/ml carbenicillin, indicating double stranded acn-1 RNA might inhibit bacterial proliferation. To address this issue, we prepared 3X-concentrated liquid bacterial culture from both control and acn-1 RNAi bacteria, spread this on NGM RNAi dishes, and allowed dishes to incubate overnight. L4 stage larvae were transferred to RNAi dishes and cultured for one day, adults were transferred to a fresh RNAi dish and cultured for one day and then removed. Larva that developed on these plates were analyzed.
Dauer formation was assayed as described by Kimura et al., [50]. Briefly, we collected eggs from wild-type or rrf-3(pk1426) hermaphrodites cultured at 20°, transferred the eggs to 27°C with ample food, cultured for 72 hr, and examined hatched animals. Animals were classified as non-dauer (including adults and non-dauer larvae) or dauer on the basis of morphological criteria [61]. To analyze dauer formation of daf-2(e1370) mutant animals, we transferred eggs to 15°C, 17.5°C, 20°C, 22.5°C, or 25°C. For dauer formation experiments, we performed acn-1 RNAi using the feeding protocol described by Kammath et al., [86]. Briefly, L4 stage hermaphrodites were transferred to dishes with control (L4440) or acn-1 RNAi bacteria at 20°C, and embryos were transferred to fresh dishes with RNAi bacteria at the appropriate temperature and cultured for 3 or 4 days.
To analyze DAF-16::GFP localization, we used the strain GR1352 [85]. L4 stage animals were transferred to dishes seeded with control (L4440) and acn-1 RNAi bacteria. Progeny were analyzed at the one day old adult stage using an Olympus SZX12 dissecting microscope (Tokyo, Japan) equipped for fluorescence microscopy. To reduce bias, the scoring was done by an observer blind to the RNAi treatment status. We analyzed each worm as having (1) GFP diffusely localized in the cytosol, (2) GFP localized in nuclei displaying intensely fluorescing puncta throughout the entire body from head to tail or (3) intermediate nuclear localization of GFP, defined as puncta observed in at least one or more nuclei but not in most or all nuclei. To perform the data analysis, we combined the nuclear and intermediate nuclear categories.
Thermotolerance assays were performed as described by McColl et al., [87]. Briefly, L4 stage hermaphrodites were cultured at 20°C on control (L4440) and acn-1 RNAi dishes for 3 days. To perform the heat stress assay, we transferred adults to 34°C and scored the percentage of dead and live animals starting at 6 hours and continuing every hour until all animals died. Animals were scored as dead if they did not respond to a mechanical stimulus. To perform oxidative stress assays, we transferred day 3 adult hermaphrodites to NGM dishes containing 40 mM paraquat and scored for survival every 12 hours. For the heat stress and paraquat stress assays, animals that displayed matricidal hatching or vulval extrusion were not included in the data analysis. To perform oxidative stress assays with juglone, we transferred day 3 adult hermaphrodites to 2 ml of liquid M9 medium containing 240 uM juglone in an 18 well dish. Worms were scored for survival after 9 hours. Paraquat and juglone were obtained from Sigma Aldrich (St. Louis, MO, USA).
To quantify mRNA levels, we cultured rrf-3 adult worms on control and acn-1 RNAi dishes for 3–4 hours to obtain synchronized eggs, removed adult worms, and continued culture until the eggs developed into two day old adult worms. These adults were washed and collected for RNA isolation. RNA analysis was performed as previously described with modifications [88]. Briefly, RNA was isolated using Trizol (Invitrogen) and treated with DNAse 1 enzyme. cDNA was synthesized by using High Capacity cDNA Reverse Transcription kit (Applied Biosystems). Quantitative, realtime PCR was performed using an Applied Biosystems Step One Plus Real-Time PCR system and iTaq Universal SYBR Green Supermix (BioRad Laboratories, Hercules, CA). mRNA fold change was determined by comparing acn-1 mRNA levels with mRNA levels of the reference gene rps-23. Forward and reverse amplification primers were: rps-23 5′- aaggctcacattggaactcg and 5′- aggctgcttagcttcgacac; acn-1 5′- gtactacgagccactcatcaac and 5′- gaatctcctcgacagtgaatg.
All data were analyzed using the two-tailed student t-test for samples with unequal variances by using Excel and http://studentsttest.com. P values less than 0.05 were considered statistically significant. To determine if the choice of a statistical test affected the conclusions, we used the log rank (Mantel-Cox) method to analyze a subset of the lifespan experiments. Both tests produced similar P values.
|
10.1371/journal.pgen.1007993 | Regulation of anthocyanin accumulation via MYB75/HAT1/TPL-mediated transcriptional repression | Anthocyanin is part of secondary metabolites, which is induced by environmental stimuli and developmental signals, such as high light and sucrose. Anthocyanin accumulation is activated by the MYB-bHLH-WD40 (MBW) protein complex in plants. But the evidence of how plants maintain anthocyanin in response to signals is lacking. Here we perform molecular and genetic evidence to display that HAT1 plays a new breaker of anthocyanin accumulation via post-translational regulations of MBW protein complex. Loss of function of HAT1 in the Arabidopsis seedlings exhibits increased anthocyanin accumulation, whereas overexpression of HAT1 significantly repressed anthocyanin accumulation. We found that HAT1 interacted with MYB75 and thereby interfered with MBW protein complex. Overexpression of HAT1 suppresses abundant anthocyanin phenotype of pap1-D plant. HAT1 is characterized as a transcriptional repressor possessing an N-terminal EAR motif, which determines to interact with TOPLESS corepressor. Repression activity of HAT1 in regulation of gene expression and anthocyanin accumulation can be abolished by deletion or mutation of the EAR motif 1. Chromatin immunoprecipitation assays revealed that MYB75 formed a transcriptional repressor complex with HAT1-TPL by histone H3 deacetylation in target genes. We proposed that HAT1 restrained anthocyanin accumulation by inhibiting the activities of MBW protein complex through blocking the formation of MBW protein complex and recruiting the TPL corepressor to epigenetically modulate the anthocyanin late biosynthetic genes (LBGs).
| Anthocyanins, a class of flavonoids distributed ubiquitously in the plant kingdom, are induced by environmental stimuli and developmental signals, such as high light and sucrose. It is well established that anthocyanin accumulation is regulated by the MYB-bHLH-WD40 (MBW) protein complex in plants. But little is known about the regulation of MBW protein complex by other factors. Here, we show that an HD-ZIP II transcription factor HAT1 negatively regulates anthocyanin accumulation via post-translational regulation of MBW protein complex. Loss of function of HAT1 in the Arabidopsis seedlings exhibits increased anthocyanin accumulation, whereas overexpression of HAT1 significantly repressed anthocyanin accumulation. We reveal that HAT1 interacted with MYB75 and thereby sequestered MBW protein complex. Overexpression of HAT1 in pap1-D mutant suppresses abundant anthocyanin phenotype of the pap1-D mutant. HAT1 identified was as a transcriptional repressor possessing an N-terminal EAR motif, which determines the interaction with TOPLESS corepressor. The deletion or mutation of the EAR motif 1 of HAT1 partially eliminates the repression activity of HAT1 in regulation of gene expression and anthocyanin accumulation. Our results illustrate a new repressor HAT1 which helps plants fine-tune anthocyanin accumulation.
| Anthocyanins, one kind of flavonoids, are vital secondary metabolites widespread throughout the plant kingdom [1]. As water-soluble pigments, anthocyanins confer widest colors to flowers, leaves, and fruits [2]. Anthocyanin accumulation is stimulated by a variety of endocellular signals such as sucrose and phytohormone [3, 4], as well as exogenous environmental stresses including high light [5], drought [6], and nutrient depletion [7]. Anthocyanins can protect plants against excessive light [8] and drought [9], and defend from invasion by pathogens and herbivores [10].
Anthocyanin biosynthesis is derived from flavonoid synthetic pathway which is composed of multiple enzymes encoded by biosynthetic genes. Initially, the early flavonoid reactions catalyzed by early biosynthetic genes (EBGs) included chalcone synthase (CHS), chalcone isomerase (CHI), and flavonol 3-hydroxylase (F3H), which are regulated by three redundant R2R3 MYB transcription factors (TFs) MYB11, MYB12, and MYB111 [11]. Then the expression of anthocyanin-specific biosynthetic genes encoding dihydroflavonol-4-reductase (DFR), leucoanthocyanidin dioxygenase (LDOX), and UDP-glucose: flavonoid-3-O-glycosyl-transferase (UF3GT) is regulated by the ternary MYB-bHLH-WD40 (MBW) protein complex, which is composed of R2R3-MYB, basic helix-loop-helix (bHLH), and WD40-repeat proteins [2, 12]. In Arabidopsis, the identified R2R3-MYB transcription factors include PRODUCTION OF ANTHOCYANIN PIGMENTATION 1 (PAP1)/MYB75, PAP2/MYB90, MYB113, and MYB114 [13, 14]. The bHLH transcription factors include TRANSPARENT TESTA 8 (TT8) and ENHANCER OF GLABRA 3 (EGL3), and and only one WD40-repeat protein, TRANSPARENT TESTA GLABRA 1 (TTG1), has been identified [14–16].
Notably, some MYB TFs such as TRANSPARENT TESTA 2 (TT2), GLABRA1 (GL1), and WEREWOLF (WER) act as a component of MBW protein complex to transcriptionally regulate expression of multiple gene involved in proanthocyanin accumulation, trichome development, and root epidermal cell fate in Arabidopsis thaliana [17–19]. Recent studies demonstrated that post-translational modification of MBW proteins modulated the transcriptional activity of MBW protein complex. Degradation of MYB75 in the dark was mediated by E3 ubiquitin ligase COP1 [20], while MPK4 phosphorates MYB75 and increases its stability in response to light [5]. Ubiquitin protein ligase 3 (UPL3) regulates anthocyanin and trichome development by mediating the proteasomal degradation of GL3 and EGL3 [21]. GSK3-like kinase BIN2 controls root epidermal cell fate through suppressing the activity of MBW protein complex via phosphorylating both TTG1 and EGL3 [22]. The other post-translational regulation is preventing the formation of MBW protein complex. Single repeat R3-MYB transcription factors, including MYBL2, CAPRICE (CPC), TRIPTYCHON (TRY), ENHANCER OF TRY AND CPC 1 (ETC1), and ETC2, suppress anthocyanin accumulation or trichome initiation by disturbing the formation of MBW protein complex [23–27]. SPL9, a member of SQUAMOSA PROMOTER BINDING PROTEIN-LIKE family, negatively regulates anthocyanin accumulation by directly preventing expression of anthocyanin biosynthetic genes via destabilization of MBW complex [28]. Similarly, the JA-ZIM domain (JAZ) proteins suppress anthocyanin accumulation and trichome development by disturbing the MBW protein complex [4]. Additionally, MYBL2 and PhMYB27 transform the MBW complex from an activator to a repressor by replacing one of the R2R3-MYB components of MBW protein [23, 24, 29]. The transformation possibly depends on post-translational regulation. However, details of the regulation mechanism remain unclear.
Anthocyanin accumulation in plants is modulated by light conditions [20]. Plants accumulate less anthocyanin under shade conditions [29]. It is well established that the class II homeodomain-leucine zipper family participate in shade avoidance and their expression is rapidly induced by shade conditions [30]. The members of class II homeodomain-leucine zipper family all contain a conserved DNA-binding homeodomain (HD) that is closed to a leucine zipper motif (LZ), which is considered important to promote homo- or heterodimerization of HD-Zip protein [31, 32]. The CPSCE motif adjacent to LZ motif is comprised of five conserved amino acids Cys, Pro, Ser, Cys, and Glu. This motif is thought to form high molecular weight multimers through intermolecular Cys-Cys bridges under oxidant environment, which can not possibly be transported to the nucleus to play its role [33]. It was reported that the class II HD-Zip proteins participated in modulation of plant development and multiple stress response. HAT2 is strongly induced by auxin and affects lateral root and hypocotyl elongation [34]. HAT3 and ATHB4 impact the leaf polarity in Arabidopsis thaliana by repressing MIR165/166 expression [35]. ATHB4 and HAT1 participate in Brassinosteroid signaling [36, 37]. ATHB17 is involved in ABA response and plays an important role in protecting plants by adjusting expression of PhANGs and PEGs in response to abiotic stresses [38, 39].
To explore the regulatory mechanisms in anthocyanin accumulation, here we identify a new regulator of MBW protein complex. HAT1 interferes with the formation of MBW protein complex by interacting with MYB75. In Arabidopsis, TOPLESS (TPL) and TOPLESS RELATED (TPR) proteins generally mediate transcriptional repression in numerous pathways such as auxin and jasmonate signaling [40–42], as well as developmental pathways including leaf polarity and flowering time regulation [43, 44]. Meanwhile, we reveal that TPL can interact with HAT1. We propose that HAT1 represses anthocyanin accumulation by inhibiting the activities of MBW protein complex through recruiting the TPL/TPR corepressors and histone modification. Our study exposes that HAT1 acts as a new regulator of anthocyanin accumulation, and proffers a mechanism for repression of anthocyanin accumulation.
Our previous research has proved that HAT1 participated in drought response [45], and we noticed that transgenic plants overexpressing HAT1 (35S:HAT1) showed less anthocyanin accumulation compared with wild-type plants under drought stress. In Arabidopsis, the levels of anthocyanin accumulation are dependent on light intensity [46]. To understand how HAT1 influences anthocyanin accumulation, we maintained Arabidopsis seedlings under weak light of 40 μmol m-2 s-1 (hereafter called Control), a control light intensity at which wild-type plants accumulate very low levels of anthocyanins [47]. To induce anthocyanin biosynthesis, seedlings were then shifted to moderate high light (180 μmol m-2 s-1, hereafter called high light). Exposed to high light, two independent lines overexpressing HAT1 (35S:HAT1 #11 and #13) accumulated less anthocyanin than that of wild-type plants, while hat1 mutant showed higher anthocyanin accumulation (Fig 1A and 1B). We also detected the expression levels of anthocyanin-specific biosynthetic genes, DFR, LDOX and UF3GT. The expression of these genes was also reduced in 35S:HAT1 plants but induced in the hat1 mutant (Fig 1C–1E). Additionally, We also noticed that HAT1 transcription was notably repressed under high light conditions (Fig 1F). 35S:HAT1 #13 transgenic plants were selected for subsequent experiment because the similar phenotype was observed between 35S:HAT1 #11 and 35S:HAT1 #13 transgenic plants. As shown in S1 Fig, 35S:HAT1 #13 transgenic plants grown in soil exhibited less anthocyanin accumulation compared with wild-type under high light conditions.
Sucrose can specifically induce anthocyanin biosynthesis in Arabidopsis, thus we further investigate the anthocyanin accumulation in Arabidopsis seedlings grown with exogenous sucrose [3]. Similarly, the anthocyanin content of 35S:HAT1 #13 was significantly lower than that of wild-type under sucrose treatment (S2A and S2B Fig). Compared with wild-type, DFR transcripts were decreased in 35S:HAT1 #13 but increased in hat1 mutant. LDOX and UF3GT showed the similar expression pattern to that of DFR (S2C Fig). In Arabidopsis, some transcription factors modulate the expression of anthocyanin biosynthetic genes. We further examined the expression of several transcription factors including MYB75, MYB90, TT8, EGL3 and TTG1, which are characterized as regulators of anthocyanin biosynthetic genes (S2D Fig). The expression levels of MYB75 and MYB90 obviously decreased in 35S:HAT1 #13 when compared with wild-type, while a nearly 1.5-fold increased was recorded in hat1 mutant than that in the wild-type (S2D Fig). The transcript levels of TT8 and EGL3 were moderately increased in hat1 mutant (S2D Fig). In summary, these results suggest that HAT1 may play as a negative regulator in anthocyanin biosynthetic pathway.
To clarify how HAT1 regulates anthocyanin accumulation, we performed a yeast two-hybrid screen to identify its potential interaction partners. After screening, we identified MYB75 (At1g56650), an R2R3-MYB transcription factor, as its partner. Directed yeast two-hybrid assays validated that HAT1 interacted with MYB75, but not with MYB90, the paralog of MYB75 (Fig 2A, S3 Fig). To further determine the domains required for the interaction, truncated HAT1 and MYB75 were used. As shown in Fig 2B, C-terminal fragment of HAT1 was required for the interaction. The R2 domain of MYB75 could strongly interact with HAT1, but the C-terminal fragment of MYB75 weakly interacted with HAT1 (Fig 2B).
To verify whether HAT1 interacts with MYB75 in vivo, the bimolecular fluorescence complementation (BiFC) assay was performed for analysis. When MYB75-nYFP was coinfiltrated with HAT1-cYFP in tobacco (Nicotiana benthamiana) leaves, strong YFP fluorescence was observed in the nuclei (Fig 2C). Further, the interaction between HAT1 and MYB75 was also confirmed by a coimmunoprecipitation (Co-IP) assay (Fig 2D). These results suggest that HAT1 can interact with MYB75 in vivo.
Previous evidence demonstrated that ternary MYB75-TT8/EGL3-TTG1 protein complex can activate the expression of LBGs [14]. The results above suggest that HAT1 interacts with MYB75 and represses the transcripts levels of LBGs, hence, we speculate that HAT1 competed with bHLH proteins for interaction with MYB75 since no interaction between HAT1 and TT8/EGL3 was detected by BiFC assay (S4 Fig). To test this hypothesis, TT8-nYFP and MYB75-cYFP were transiently coexpressed in N. benthamiana leaves. As shown in Fig 3A, the strong YFP fluorescence were detected in N. benthamiana leaves, which is consistent with the previous study [48]. When HAT1-FLAG was coexpressed with TT8-nYFP and MYB75-cYFP, the fluorescence signal was visibly impaired (Fig 3A and 3B), whereas coexpression of empty vector (FLAG) with TT8-nYFP/MYB75-cYFP did not reduce the fluorescence intensity. Similar results were observed when HAT1-FLAG was coexpressed with EGL3-nYFP and MYB75-cYFP (Fig 3A and 3C). The expression of MYB75, TT8 and EGL3 exhibited no obvious difference in these combination respectively (Fig 3D and 3E).
In vitro competitive binding assays demonstrated that interaction between HIS-MYB75 and MBP-TT8 was gradually impaired by an increased amount of HIS-HAT1 (Fig 3F). Likewise, HIS-HAT1 also weakened the interaction between HIS-MYB75 and MBP-EGL3 (Fig 3F). Using protoplasts from 35S:HAT1 #13, we found that the interaction between TT8/EGL3 and MYB75 is counteracted by endogenous HAT1-GFP (Fig 3G). We further tested whether HAT1 interferes with the interaction between MYB75 and TT8 or EGL3 in a yeast three-hybrid assay. When yeast transformant that carried both AD-MYB75 and pBridge-TT8-HAT1 plasmids were grown on plates with high methionine concentrations (200 μM), which restrain the expression of HAT1, MYB75 strongly interacted with TT8 (Fig 3H). When the level of methionine was reduced from 200 μM to 50 μM, hence permitting HAT1 expression, yeast growth were consumingly suppressed in these transformants (Fig 3H). Similar results were observed when yeast transformed with AD-MYB75 and pBridge-EGL3-HAT1 (Fig 3H). In summary, these results prove that HAT1 interferes with the formation of MBW protein complex.
Next the transgenic line pap1-D overexpressing MYB75 was crossed with 35S:HAT1#13 to generate the pap1-D 35S:HAT1 #13 plants. As expected, overexpression HAT1 repressed anthocyanin accumulation in pap1-D background under control or high light conditions (Fig 4A and 4B). We also crossed hat1 null mutant with myb75-c that MYB75 knockout mutant using the CRISPR-Cas9 system. Under high light conditions, the myb75-c hat1 double mutant exhibited similar phenotype with myb75-c. Consistent with the phenotype, transcript levels of anthocyanin biosynthetic genes DFR and LDOX were also lower in pap1-D 35S:HAT1 #13 than that of pap1-D mutant under normal or high light conditions, although the expression levels of HAT1 in pap1-D 35S:HAT1 #13 were similar with pap1-D (Fig 4C–4E). Additionally, the anthocyanin content of the pap1-D hat1 exhibited no significant difference than that of pap1-D under high light conditions (S5A and S5B Fig), it might be due to high light repressed HAT1 expression in pap1-D plants (Fig 1D). Meanwhile, 35S:HAT1 #13 and 35S:HAT1 #13 myb75-c showed a similar anthocyanin accumulation phenotype (S5C and S5B Fig). Taken together, our results suggest that HAT1 represses anthocyanin accumulation through interacting with MYB75 and at least partially by interfering with the formation of MBW protein complex.
Although we have proved that HAT1 represses anthocyanin accumulation by sequestering the formation of MBW protein complex, we hypothesize that MYB75-HAT1 complex behaves as a repressor because overexpressing HAT1 in pap1-D represses anthocyanin accumulation when compared with pap1-D mutant. The previous study demonstrated that the members of HD-Zip II family act as a repressor in modulation of gene expression [31]. HAT1 protein possesses a leucine zipper motif (LZ, between amino acids residue 190 and 233) followed by a DNA-binding homeodomain (HD, between amino acids residue 134 and 188) (Fig 2B). Interestingly, two typical ERF-associated-amphiphilic repression (EAR) motif (DLGLSL and LQLNLK), which is proved as repression motif to repress transcription [49], are located at the N-terminal end of the HAT1 protein (Fig 5A). Many transcriptional repressors have been proved to regulate plant developmental process and signaling pathways by interacting with corepressor TPL/TRRs via EAR motif [40, 41, 43]. This promoted us to investigate whether HAT1 interacted with TPL/TRRs. Firstly, yeast two-hybrid assay showed that HAT1 interacted with TPL (Fig 5B). Furthermore, the interaction relied on the HAT1 EAR motif 1, because replacement of the three conserved Leu residues into Ala residues (LxLxL to AxAxA, designated as HAT1mEAR) abolished the interaction (Fig 5B). We also found that HAT1 interacted with TPR3, but not with TPR1, TPR2 and TPR4 (S6 Fig). Next we used Co-IP experiments to further check the interaction. Consistent with the yeast two-hybrid assays, Co-IP assays proffered that HAT1 interacts with TPL in plants, but HAT1mEAR1 could not form a complex with TPL (Fig 5C), suggesting that EAR motif 1 determined the interaction between HAT1 and TPL. Finally, we investigated whether HAT1 behaved as a transcriptional repressor to inhibit VP16-mediated transcriptional activation. As expected, wild-type HAT1 significantly inhibit the VP16-promoted LUC activity, but deletion of the EAR motif 1 (HAT1ΔEAR) or HAT1mEAR1 abolished the inhibitory effects (Fig 5D). Taken together, these results suggested that HAT1 interacts with TPL and plays as a repressor.
We next investigated whether the HAT1-dependent repression of anthocyanin accumulation is released by the loss of function of TPL. To clarify this, we crossed 35S:HAT1 #13 to the tpl-1 mutant, which is an N176H substitution and has a dominant-negative on the rest of TPRs [50]. Loss of function of TPL in 35S:HAT1 #13 (35S:HAT1 #13 +/tpl-1) largely rescued the lower anthocyanin accumulation phenotype of 35S:HAT1 #13 under high light conditions (Fig 5E and 5F). Consistently, repression of DFR, LDOX and UF3GT in 35S:HAT1 +/tpl-1 were largely released than that of 35S:HAT1 #13 (Fig 5G–5I). These data suggest that HAT1 represses anthocyanin accumulation partially by interacting with TPL/TPRs.
To further illustrate the mechanism how HAT1 regulated anthocyanin accumulation, we generated transgenic plants overexpressing variants of HAT1 protein (35S:HAT1mEAR1). This approach was used successfully in the previous study [43]. Two independent transgenic lines, 35S:HAT1mEAR1 #6 and 35S:HAT1mEAR1 #10, showed more anthocyanin accumulation than of 35S:HAT1 #13 under high light conditions (Fig 6A and 6B). Intriguingly, we noticed that the anthocyanin contents of 35S:HAT1mEAR1 #6 and 35S:HAT1mEAR1 #10 could not completely restore to that of wild-type plants. Consistently, the expression levels of these three transgenes were similar, but the transcript levels of DFR, LDOX and UF3GT were higher in the 35S:HAT1mEAR1 transgenic plants under high light conditions compared with 35S:HAT1 #13 plants (Fig 6C–6F). These data indicate the EAR motif 1 is required for HAT1-repressed anthocyanin accumulation.
It has been reported that transgenic plants overexpressing MYB75-SRDX fusion protein (35S:MYB75-SRDX) exhibited minimal anthocyanin under 3% sucrose [51]. Like EAR motif, the SRDX domain can convert transcriptional activators into dominant repressor when fused to transcription factors. In our study, the evidence that the C-terminal fragment of HAT1 (234 to 282 residues) interacts with MYB75 and that the EAR motif in the N-terminal region interacts TPL suggests that HAT1 could link TPL to MYB75 and convert MYB75 into a repressor. To check this hypothesis, MYB75 was fused with the N-terminal fragment of HAT1 (1 to 90 residues containing EAR motif), and the fusion construct driven by a CaMV 35S promoter was transformed into wild-type Arabidopsis (designated 35S:MYB75-HAT1N). As expected, two independent transgenic plants, 35S:MYB75-HAT1N #2 and 35S:MYB75-HAT1N #9, exhibited minimal anthocyanin accumulation compared with 35S:MYB75 transgenic plants under high light conditions (Fig 7A and 7B). Consistently, the levels of DFR, LDOX and UF3GT transcripts were also lower in 35S:MYB75-HAT1N #2 and 35S:MYB75-HAT1N #9 plants compared with 35S:MYB75 transgenic plants under high light conditions despite the similar expression levels of MYB75 in these transgenes (Fig 7C–7F).
Then we performed a transient transformation assay using the DFR promoter fused to the LUC gene as a reporter. HA-MYB75, HA-TPL and HAT1-GFP construct acted as effector and transfected together with the reporter construct into myb75-c mesophyll protoplasts. The LUC expression of DFR promoter was very low without MYB75 expression, but was activated by expression of MYB75 (S7 Fig). However, when HAT1 was coexpressed with MYB75, this activation was markedly decreased (S7 Fig). The relative luciferase activities were moreover decreased when TPL was co-transformed with HAT1, but the repression was alleviated when TPL was coexpressed with mutational HAT1, suggesting that HAT1 inhibits the transcriptional activity of MYB75 through TPL function. Taken together, these results indicate that HAT1 represses anthocyanin accumulation by connecting TPL with MYB75.
Previous studies have shown that TPL interacts with two histone deacetylase, HDA6 and HDA19, which function in chromatin modification and epigenetic regulation of developmental and hormone-responsive genes [52, 53]. The presence of ternary MYB75-HAT1-TPL protein complex led us to further analyze whether HAT1 represses the expression of anthocyanin biosynthetic genes via chromatin modifications. To prove this speculation, ChIP assays were performed by using antibody of acetylated H3 in different mutants. As shown in Fig 8A–8C, reduced histone H3 acetylation was verified in the transcription start sites (TSSs) of DFR, LDOX and UF3GT in 35S:HAT1 #13 under high light conditions, while elevated histone H3 acetylation was detected in hat1 mutant under the same condition. Simultaneously, 35S:MYB75-HAT1N #2 transgenic plants showed lower histone H3 acetylation in the TSSs of DFR, LDOX and UF3GT than that of 35S:MYB75 transgenic plants (Fig 8D–8F). These fingdings suggest that MYB75-HAT1-TPL inhibits LBGs expression through recruiting a histone modification complex.
Anthocyanins ara one kind of flavonoids induced by environmental stimuli and developmental signals. Exposed to extreme conditions, such as high light, high concentration of sucrose, and drought, plants evolved a range of mechanisms to regulate transcription of anthocyanin biosynthetic genes [29]. The spatial and temporal regulation of anthocyanin accumulation are determined by an MBW protein complex, while the R2R3-MYBs are pivotal to appoint the function of the MBW protein complex [14]. Here we illustrated that HAT1 negatively regulated anthocyanin accumulation through hindering the active function of MBW protein complex and epigenetic regulation.
Although members of HD-Zip family have been well described in plants, the possible function in anthocyanin accumulation remains to investigate. GL2, a member of HD-Zip subfamily IV, has been characterised as a negative regulator of anthocyanin accumulation in Arabidopsis [54]. In this study, our findings elucidate that HAT1 may function as a new repressor in anthocyanin accumulation through sequestering the MBW protein complex and epigenetic regulation in Arabidopsis. Additionally, MYB75, MYB90, TT8 and EGL3 transcript levels were increased in hat1 mutant and decreased in 35S:HAT1 plants (S2D Fig), presenting the possibility that HAT1 regulates anthocyanin accumulation by modulating MBW protein complex expression. Our results indicated that HAT1 regulated anthocyanin accumulation via post-translational regulation and transcriptional regulation of MBW protein complex.
Much work so far has evidenced that the expression of the members of HD-Zip II family is induced in response to simulated shade conditions. Interestingly, weak anthocyanin accumulation occurs under shade conditions in petunia plants and the conversion of light conditions change the members of HD-Zip II family abundance [30, 55]. It seems that the members of HD-Zip II family are involved in modulation of anthocyanin accumulation in response to light/shade. However, MYB75 did not interact with HAT2, HAT3, ATHB2, and ATHB4 in yeast (S8 Fig), suggesting that HAT1 is a unique gene of HD-Zip II family to repress anthocyanin accumulation.
It is likely that overexpression of one member will repress the expression of rest in HD-Zip family II [34]. Previous study described that the EAR motif exists in different members of HD-Zip II family [30]. We investigate the impact of EAR motif on anthocyanin accumulation. We proved that EAR motif is required for HAT1-repressed anthocyanin accumulation. Further, HAT1 suppresses anthocyanin accumulation via TPL-dependent histone deacetylation (Fig 8A–8F). Interestingly, our previous study proved that HAT1 negatively regulates hormone synthesis and response gene [36, 45].
It seems that HAT1 behaves as a transcriptional repressor in various hormonal signaling and metabolic process. Our recent study reported that phosphorylation of HAT1 by SnRK2.3 induced its degradation. And abscisic acid (ABA) positively regulated anthocyanin accumulation [56]. It raised the possibility that ABA induced anthocyanin accumulation by inhihiting HAT1 founction. Our previous studies also revealed that HAT1 was a positive regulator in BR pathway [36]. Under favorable conditions, HAT1 accumulated and interacted with BES1 to promote plant growth. Meanwhile, HAT1 also interacted with MYB75 and TPL to inhibit anthocyanin accumulation. It demonstrated that HAT1 was a key regulator in balancing plant growth and stresses adaption.
Numerous EAR motif-contained transcription factors repress gene expression through directly binding to the cis element. The previous study showed HAT1 interacts with BES1 [36]. Intriguingly, BES1-TPL complex mediates the inhibitory action of brassinosteroids on ABA responses during early seedling development [57]. We speculate that the these proteins can form a HAT1-BES1-TPL protein complex to repress BR-response gene. A recent study revealed that ATHB4 regulates leaf polarity and hypocotyl elongation by interacting with TPL protein [58]. Several studies proposed that some transcription factors repressed gene transcription by recruiting TPL and HDAs to the promoter of target gene [44, 53, 59]. It is important to highlight that HAT1 inhibits the expression of anthocyanin biosynthetic-specific genes through bridging MYB75 and TPL rather than directly binding to the promoter of these genes.
Anthocyanin accumulation occurs at the junction of the rosette leaves and stem during plant development in Arabidopsis [28]. We noticed the purple pigments of 35S:HAT1 #13 plants was much lower than that of wild-type, whereas hat1 mutant exhibited more purple pigments compared with wild-type (S9A and S9B Fig). The expression of DFR is confined within basal regions of stems during the transition from leaves to flowers [28]. MYB75 showed high expression in the lower part of the inflorescence stem and leaves in 6-week-old Arabidopsis [48]. Our results showed that transcript levels of HAT1 was minimal in the lower part of the inflorescence stem and highest in the upper part of the inflorescence stem, while DFR and MYB75 exhibited an inverse pattern in 6-week-old Arabidopsis (S10A Fig). Consequently, low expression of HAT1 eliminates the repression for MBW protein complex, thus resulting in anthocyanin accumulation in stem-rosette junction. On the other hand, HAT1 showed low expression in senescent leaves, while DFR and MYB75 displayed high expression in senescent leaves (S10B Fig). Consistent with this, gene expression analysis using the Arabidopsis microarray data displayed in the eFP browser indicated opposite expression patterns of MYB75 and HAT1, suggesting HAT1 serves as a repressor in senescent leaves during plant senescence (S11 Fig). MBW protein complex activates anthocyanin biosynthetic-specific gene expression when HAT1 expression is limited and thereby induces anthocyanin accumulation. These data proved that HAT1 represses anthocyanin accumulation in stem-rosette junction and leaves during plant senescence.
Present studies demonstrated that epigenetic modifications participated in regulation of anthocyanin accumulation [60, 61]. Recent research suggested DELLA promoted anthocyanin accumulation via sequestering MYBL2 and JAZ suppressors of the MBW complex in Arabidopsis thaliana [62]. To test the relationship between HAT1 and MYBL2 in suppression of anthocyanin accumulation, we crossed 35S:HAT1 #13 with mybl2 mutant. The 35S:HAT1 #13 mybl2 plants showed less anthocyanin accumulation (S12A and S12B Fig), indicating that HAT1 repressed anthocyanin accumulation is independent of the MYBL2-regulated pathway. We further observed that H3 acetylation levels in the promoter of LBGs were elevated in mybl2 mutant under high light conditions (S12C–S12E Fig). These results suggest that epigenetic modifications are prevalent in regulation of anthocyanin accumulation in plants. Interestingly, maize bHLH transcription factor R interacts with RIF1 and thereby forms a complex with MYB transcription factor C1. C1-R-RIF1 complex binds to A1 promoter and activates A1 expression by elevating the H3K9 and H3K14 acetylation levels in the promoter region [63]. Therefore, it remains challenging to identify the interaction between MBW protein complex and epigenetic regulators, which may explain why MBW protein complex activates the expression of biosynthetic genes.
To date, much work has characterized many repressors in modulation of anthocyanin accumulation. Several repressors regulate anthocyanin accumulation without direct interaction with members of MBW protein complex [7, 54, 64]. Notably, it has been proved that LBD37 interact with TPL via Y2H screening [65]. On the other side, MYB75, bHLH and TTG1 are expressed in various tissues of Arabidopsis under non-inductive conditions [15, 16, 19, 48]. Therefore, some repressors suppress anthocyanin accumulation by directly interacting with members of MBW protein complex [4, 28, 29, 66]. We prove that HAT1 suppresses anthocyanin accumulation by directly interacting with MYB75. Intriguingly, the proteins that interact with MBW protein complex all contain EAR moif except SPL9. Y2H screening proved that JAZ5/JAZ6/JAZ7/JAZ8, MYBL2 and HAT1 interact with TPL respectively [65]. We speculate that EAR moif containd-repressors inhibit anthocyanin accumulation by forming a protein complex with components of MBW and TPL in plants. The repressive activity of MYBL2 might play a critical role in suppression of anthocyanin accumulation [24]. In our studuy, HAT1mEAR1 may still interact with MYB75 and interfere with MBW protein complex because the C terminus of HAT1 determines the interaction with MYB75. We observe that anthocyanin level of 35S:HAT1mEAR1 transgenic plants was lower than that of wild-type. These results suggested that HAT1 inhibits anthocyanin accumulation partially by disturbing the formation of MBW protein complex. We suppose that interference of MBW protein complex (Passive repression) and epigenetic modification (Active repression) function synchronously in regulation of anthocyanin accumulation.
Under non-inductive conditions, such as low light conditions, plants possess high levels of HAT1. Meanwhile, HAT1 interferes with MBW protein complex by binding MYB75, as well as transforming the active MBW protein complex into a repressive complex by recruiting EAR motif-dependent TPL corepressor, thus preventing the expression of LBGs (Fig 8G). Under inductive conditions, such as high light conditions, HAT1 expression is suppressed. MYB75 and bHLH transcription factors are able to form MBW protein complex with TTG1 protein, which activates transcription of the target genes encoding LBGs (Fig 8H). Overall, our work together with other studies suggests that plants restrict the expression of anthocyanin biosynthetic genes via the multiple and intricate mechanisms.
The Arabidopsis thaliana 35S:HAT1 #11, 35S:HAT1 #13 and hat1 mutants were described as previously [36]. All wild-type, various mutants, and transgenic plants in this study are in Col-0 ecotype background. To avoid ecotype variability, the tpl-1 mutant, originally in Ler background [40], was introgressed into the Col-0. Arabidopsis seeds were placed on half-strength Murashige and Skoog medium. The plates were placed at 4°C for 3 d avant transfering to 22°C under different light conditions. Plates were put at control (40 μmol m-2 s-1) or high light (180 μmol m-2 s-1) conditions with a 16-h-light/8-h-dark photoperiod for high light-induced research [5]. For the BiFC assays, Nicotiana benthamiana was cultured in soil at 22°C under 16-h-light/8-h-dark conditions.
Anthocyanin levels were measured as described previously [67]. Briefly, arabidopsis seedlings were incubated in extraction buffer (methanol containing 1% HCl) overnight at 4°C in the dark. After extraction and centrifuged, the supernatants were collected and absorbance calculated at 530 and 657 nm. Relative anthocyanin content was quantified by (A530-0.25×A657) per gram fresh weight.
HAT1mEAR1 was amplified from pZP211-HAT1 using primers indicated in S1 Table and cloned into pZP211 vector to generate 35S:HAT1mEAR1-GFP [68]. To generate MYB75 constructs, the 1500 bp genomic sequence of MYB75 contained the coding area was cloned into pCM1307 vector to create 35S:HA-MYB75 [69]. The sequence coding N terminus of HAT1 was amplified from pZP211-HAT1 and cloned into pCM1307-MYB75 to generate 35S:HA-MYB75-HAT1N. The coding sequences (CDS) of TPL were amplified and cloned into pCM1307 plasmid to create 35S:HA-TPL. Oligo primers used for cloning are listed in S1 Table. Col-0 plants were transformed with these constructs by using Agrobacterium tumefaciens (strain GV3101)-mediated transformation [70].
The GAL4 reporter plasmid was generated from pUC19 [71], which contains the firefly LUC reporter gene driven by the minimal TATA box of the 35S promoter plus five GAL4 binding elements. For transcriptional inhibition assays, HAT1ΔEAR and HAT1mEAR were amplified from pZP211-HAT1 and cloned into pRT-BD to generate GAL4-HAT1ΔEAR and HAT1mEAR respectively [72]. The positive control (pRT-35S-BD-VP16) was constructed by insertion of VP16, a herpes simplex virus-encoded transcriptional activator protein, into pRT-BD. Plasmid pTRL was used as internal control. For transcription activity assays in protoplast, a 512-bp DFR promoter was amplified from genomic DNA and fused with pGreenII 0800-LUC. The internal control, effectors and reporter were co-transformed into Arabidopsis protoplasts by PEG/CaCl2-mediated transfection [73]. All transfection cultured for 16 h, then luciferase assays were performed using the Promega dual-luciferase reporter assay system and a GloMax 20–20 luminometer (Promega, http://www.promega.com). Relative LUC activity was defined as firefly LUC activity divided by Renilla LUC activity.
For yeast two-hybrid assays, the full-length CDS of MYB75 and HAT1 were amplified and cloned into pGADT7 (Clontech). The full-length CDS of MYB75 and N terminus of TPL were amplified and cloned into pGBKT7 (Clontech). The yeast strain (AH109) was transformed with pairs of plasmids and grown on Double DO supplement (SD-Leu/-Trp) for 3 days, then the cotransformants were shifted onto Quadruple DO supplement (SD-Leu/-Trp/-Ade/-His) to test for possible interactions.
For yeast three-hybrid assays, the complete CDS of TT8 and EGL3 were amplified and fused with pBridge-HAT1 plasmid to generate pBridge-TT8-HAT1 and pBridge-EGL3-HAT1 respectively [74]. Yeast three-hybrid experiments performed as described previously [20]. Briefly, pBridge-TT8-HAT1 or pBridge-EGL3-HAT1 were used in co-transformation with pGADT7-MYB75. pBridge contains a methionine (Met) suppressible promoter positioned upstream of a Gateway cassette. HAT1 expression was gradually suppressed using increasing methionine concentrations in Minimal Media Quadruple Dropouts (SD-Leu/-Trp/-His/-Met). For each combination, 3 colonies selected on dropout medium (SD -Leu/-Trp) were resuspended in water, the OD600nm was adjusted to 0.7 and 20 μl was streaked out on the respective plates.
For BiFC assays, full-length CDS of MYB75 and HAT1 was cloned into the pXY103-nYFP vector respectively [75]. The full-length CDS of MYB75, TT8, EGL3 and TTG1 were cloned into the pXY104-cYFP vector respectively [75]. The constructs were transformed into Agrobacterium tumefaciens strain (GV3101) respectively, and the lower epidermis of Nicotiana benthamiana plants were used for injection of different combination. The transfected plants were grown in the green house for at least 36 hours and fluorescent signals were observed by using scanning microsystem (Leica).
Different version of MYB75 were cloned into the pMAL-C2X and pET28a vectors with MBP tag and 6×His tag respectively. TT8 and EGL3 were also cloned into the pMAL-C2X vector respectively. Different version of HAT1 were cloned into the pMALc-B and pET28a vectors with MBP tag and 6×His tag respectively. In vitro pull-down assays performed as described [76]. Ni-NTA beads containing 5 μg HIS-HAT1 proteins incubated with 5 μg MBP-MYB75 by using 500 μl pull-down buffer (150 mM NaCl, 20 mM Tris, 1 mM PMSF, 0.2% Triton X-100, 1% protease inhibitor cocktail [pH 8.0]) at 4°C for 2 h. Equally, MBP-HAT1 was incubated with Ni-NTA beads contained HIS-MYB75. Beads were washed four times with the pull-down buffer and proteins were eluted from beads by boiling in 95°C with 30 μL SDS-PAGE loading buffer then separated by SDS-PAGE and analyzed by the anti-MBP antibies.
Competitive HIS pull-down analysis performed as described previously [77], 5 μg of MBP-TT8 or MBP-EGL3 mixed with 5, 10, or 20 μg HIS-HAT1 were incubated with Ni-NTA beads containing 5 μg HIS-MYB75 by using 500 μl pull-down buffer (150 mM NaCl, 20 mM Tris, 1 mM PMSF, 0.2% Triton X-100, 1% protease inhibitor cocktail [pH 8.0]) at 4°C for 2 h. Beads were washed four times with the pull-down buffer and proteins were eluted from beads by boiling in 95°C with 30 μL SDS-PAGE loading buffer then separated by SDS-PAGE and analyzed by the anti-MBP antibies.
For the competing MBP pull-down assay [78], samples from Col-0 or 35S:HAT1 protoplasts expressed HA-MYB75 were collected in protein extraction buffer containing 150mM NaCl, 50mM Tris-HCl (pH 7.5), 0.1% (v/v) NonidetP-40, 10% (v/v) glycerol, and 1×complete protease inhibitor cocktail (Roche). The lysate was centrifuged at 12,000 g for 5 min at 4°C, and the supernatant was taken for semi-in vivo pull-down assay. MBP-TT8 or MBP-EGL3 beads was added to 500 μL of total extracted protein and incubated at 4°C for 3 h. Beads were washed in extraction buffer five times, resuspended in SDS-PAGE loading buffer and analyzed using SDS-PAGE and immunoblotting with anti-HA antibody, anti-GFP and anti-MBP antibody.
Plants expressing different proteins as indicated were extracted with protein extraction buffer containing 150mM NaCl, 50mM Tris-HCl (pH 7.5), 0.1% (v/v) NonidetP-40, 10% (v/v) glycerol, and 1×complete protease inhibitor cocktail (Roche) [78]. After centrifuging at 12,000 g for 5 min at 4°C, and the supernatant was incubated for 3 h with Anti-HA Agarose Affinity Gel antibody at 4°C. Then the beads were washed six times using extraction buffer and then eluted with 50 μL of SDS-PAGE loading buffer for immunoblot analysis using Anti-HA and Anti-GFP antibody.
ChiP assays perform as described [75]. The 4-week-old plants collected in 50 mL tubes, and 37 mL 1% formaldehyde solution was used for cross-linked under a vacuum for 20 min. The chromatin was collected and sheared by sonication to reduce the average DNA fragment size to around 500 bps, then the sonicated chromatin complex was immunoprecipitated by specific antibodies anti-acetyl-histone H3 (Catalog no 06–599, Millipore). After reverse cross-linking, the immunoprecipitated DNA fragment was analysed by qPCR with specific primers shown in S1 Table.
Total RNA extraction, cDNA synthesis and qRT-PCR were performed as described before [79]. PCR analysis was carried out using SYBR Green PCR Master Mix was used as previously described. Three separate experiments were implemented, and technical triplicates of each experiment were implemented. Gene expression normalize to the transcript levels of ACTIN 8.
Samples were analyzed in triplicates, and the data are expressed as the mean ± SD unless noted otherwise. Statistical significance was determined using two-way ANOVA (LSD’s multiple-range test) or Student’s t-test. A difference at P<0.05 was considered significant.
The Arabidopsis Genome Initiative identifiers for the genes described in this article are as follows: HAT1 (At4g17460), HAT2 (At5g47370), HAT3 (At3g60390), ATHB2 (At4g16780), ATHB4 (At2g44910), MYB75 (At1g56650), MYB90 (At1g66390), TT8 (At4g09820), EGL3 (At1g63650), TTG1 (At5g24520), MYBL2 (At1g71030), TPL (At1g15750), TPR1 (At1g80490), TPR2 (At3g16830), TPR3 (At5g27030), TPR4 (At3g15880), DFR (At5g42800), LDOX (At4g22880), UF3GT (At5g54060), ACTIN 7 (At5g09810) and ACTIN 8 (At1g49240).
|
10.1371/journal.pgen.1006603 | Protecting cells by protecting their vulnerable lysosomes: Identification of a new mechanism for preserving lysosomal functional integrity upon oxidative stress | Environmental insults such as oxidative stress can damage cell membranes. Lysosomes are particularly sensitive to membrane permeabilization since their function depends on intraluminal acidic pH and requires stable membrane-dependent proton gradients. Among the catalog of oxidative stress-responsive genes is the Lipocalin Apolipoprotein D (ApoD), an extracellular lipid binding protein endowed with antioxidant capacity. Within the nervous system, cell types in the defense frontline, such as astrocytes, secrete ApoD to help neurons cope with the challenge. The protecting role of ApoD is known from cellular to organism level, and many of its downstream effects, including optimization of autophagy upon neurodegeneration, have been described. However, we still cannot assign a cellular mechanism to ApoD gene that explains how this protection is accomplished. Here we perform a comprehensive analysis of ApoD intracellular traffic and demonstrate its role in lysosomal pH homeostasis upon paraquat-induced oxidative stress. By combining single-lysosome in vivo pH measurements with immunodetection, we demonstrate that ApoD is endocytosed and targeted to a subset of vulnerable lysosomes in a stress-dependent manner. ApoD is functionally stable in this acidic environment, and its presence is sufficient and necessary for lysosomes to recover from oxidation-induced alkalinization, both in astrocytes and neurons. This function is accomplished by preventing lysosomal membrane permeabilization. Two lysosomal-dependent biological processes, myelin phagocytosis by astrocytes and optimization of neurodegeneration-triggered autophagy in a Drosophila in vivo model, require ApoD-related Lipocalins. Our results uncover a previously unknown biological function of ApoD, member of the finely regulated and evolutionary conserved gene family of extracellular Lipocalins. They set a lipoprotein-mediated regulation of lysosomal membrane integrity as a new mechanism at the hub of many cellular functions, critical for the outcome of a wide variety of neurodegenerative diseases. These results open therapeutic opportunities by providing a route of entry and a repair mechanism for lysosomes in pathological situations.
| This work is the result of our search for the mechanism of action of Apolipoprotein D (ApoD), a neuroprotective lipid-binding protein that confers cell resistance to oxidative stress. ApoD is one of the few genes consistently over-expressed in the aging brain of all vertebrate species, and no nervous system disease has been found concurring without ApoD over-expression. All evidence supports ApoD as an endogenous mechanism of protection. We demonstrate here that this extracellular lipid binding protein is endocytosed and targeted in a finely controlled way to subsets of lysosomes in need of protection, those most sensitive to oxidative stress. ApoD reveals the existence of biologically relevant lysosomal heterogeneity that conditions the oxidation state of cells, their phagocytic or autophagic capacity, and the final output in neurodegenerative conditions. The stable presence of ApoD in lysosomes is sufficient and necessary for lysosomes to recover from oxidation-induced membrane permeabilization and loss of proton gradients. ApoD-mediated control of lysosomal membrane integrity represents a new cell-protection mechanism at the hub of many cellular functions, and is critical for the outcome of a wide variety of neurodegenerative diseases. Therapeutic opportunities open, by providing a route of entry and a repair mechanism for lysosomes in pathological situations.
| Lysosomes are acidic intracellular vesicles that provide an optimal physicochemical milieu for enzymatic activities, most of them catabolic, which need to be controlled. A well-recognized lysosomal function is the degradation and recycling of defective cellular material through autophagy, and of extracellular material that reach lysosomes by endocytosis or phagocytosis. Newly documented functions such as energy and nutrient sensing, secretion, plasma membrane repair, immune response and cell death, reveal lysosomes as sophisticated organelles controlling fine decisions in the life of a cell [reviewed by 1,2,3].
Lysosome function is essential for human health, as clearly shown by the existence of numerous Lysosomal Storage Diseases [reviewed by 4], inherited metabolic disorders that affect a variety of tissues and organs and are particularly devastating for the nervous system. Moreover, lysosomal dysfunction is known to underlie the pathogenic mechanisms of neurodegenerative disorders, such as Alzheimer’s, Parkinson’s and Huntington’s diseases as well as physiological aging [5,6,7].
In the nervous system, proper lysosomal and autophagic functions are essential for the survival of postmitotic neurons, for the phagocytic activity of microglia and for the myelination process performed by Schwann cells and oligodendrocytes [8,9,10,11].
The stability of lysosomal membrane is a key factor determining cell survival-death signaling [12], and its composition is essential for an efficient lysosomal enzymatic activity [13]. Reactive oxygen species (ROS) compromise lysosomal integrity due to membrane lipid peroxidation [12,14]. Because of the critical role of its luminal pH, the alkalinization of lysosomes due to proton leakage is thought to contribute to many pathologies [15], and compensatory responses or therapeutic manipulations that restore lysosomal pH would result in clear benefits [16]. Interestingly, not all lysosomes are equally sensitive to oxidative stress or have the same luminal pH [14,17].
Apolipoprotein D (ApoD), a lipid binding protein of the Lipocalin family first known as part of bloodstream lipoprotein particles, is one of the few genes consistently over-expressed in the aging brain, and in all neurodegenerative and psychiatric diseases tested so far [reviewed in 18]. At the biochemical level, ApoD lipid binding properties are well known [19,20,21,22,23,24] and its ability to inhibit lipid peroxidation by reducing radical-propagating lipid hydroperoxides has been determined [25]. At the functional level, evidence for cellular protection and pro-survival roles for ApoD keep accumulating both, in animal models and cell systems. ApoD exhibits a functional pleiotropy, connecting nervous system response to oxidative stress [26,27,28,29,30], recovery after injury [31,32], brain aging [33,34,35], or diverse forms of neurodegeneration [36,37,38], as well as longevity regulation [39,40,41,42,43] or nervous system control of the metabolic adaptations to stress [44]. For instance, a lack of ApoD in mice compromise the process of myelin phagocytosis after peripheral nervous system injury [31,32], and the autophagic process was revealed as a key element in the degeneration rescuing activity of the ApoD Drosophila melanogaster homolog Glial Lazarillo (GLaz) [37]. How can an extracellular lipid binding protein accomplish these intracellular functions? ApoD is internalized by different cell types including astrocytes and neurons [26,28,45,46]. However, the ApoD-containing intracellular organelles or the dynamics and route of entry have not been identified. To achieve a mechanistic knowledge of ApoD neuroprotective actions, an in-depth analysis of its subcellular traffic is required.
In this work, we perform a comprehensive analysis of ApoD subcellular traffic and discover its role in lysosomal pH homeostasis and membrane stabilization under oxidative stress conditions. We demonstrate that ApoD is endocytosed and targeted to lysosomes in a stress-dependent manner. ApoD is functionally stable in this acidic environment, and actively helps to maintain lysosomal pH gradients both in astrocytes and in neurons, by maintaining lysosomal membrane integrity. Our analysis reveals ApoD as a specific marker for the subpopulation of lysosomes vulnerable to oxidation. We also test the notion that this lysosomal protecting mechanism is of wide biological consequences, by proving the role for this Lipocalin in two lysosomal-dependent biological processes such as myelin phagocytosis and optimization of neurodegeneration-triggered autophagy.
Using the Drosophila retina as a model system to assay neurodegeneration, we previously described that Type I Spinocerebellar Ataxia (SCA1) concurs with autophagic stress, showing an excessive or imbalanced induction of autophagy where autophagosome turnover is unable to keep pace with its formation [37]. GLaz, a Drosophila homologue of ApoD expressed by subsets of glial cells in the fly nervous system, has epistatic relationship with autophagy genes and optimizes clearance of aggregation-prone proteins such as the polyglutaminated form of human Ataxin 1 that is responsible for the SCA1 phenotype [37]. We concluded that GLaz rescues neurodegeneration by making autophagy more efficient, thus minimizing the negative effects of autophagic stress. We also proposed that the Lipocalin-mediated control of lipid peroxide levels influences autophagy at several steps, slowing down the process and ultimately making it more efficient. However, the mechanism for such an optimization of autophagy was not completely discerned.
Here we evaluate retinal degeneration in the SCA1 fly retina model (Fig 1I) using FLEYE, a method for unbiased quantification based on the acquisition of fly eye surface pictures [47]. We combined the expression of polyglutaminated human Ataxin 1 (hATXN182Q) with GLaz, and with DorRNAi, a knock-down of the HOPS complex subunit Vps18/Dor critical for tethering and fusing autophagosomes with lysosomes [48] (Fig 1). GLaz neurodegeneration rescue (Fig 1C and 1D, compare with controls in Fig 1A and 1B) is completely abolished by DorRNAi expression (Fig 1E and 1F). Dor down-regulation itself does not produce significant neurodegeneration in normal conditions (Fig 1G) nor modifies the neurodegeneration phenotype triggered by SCA1 in the fly retina (Fig 1H). These results demonstrate that the lysosome-autophagosome fusion event is required for GLaz optimization of autophagy that finally results in an efficient clearance of misfolded proteins in neurodegenerative conditions in vivo.
These results predict that Lipocalins must exert this function from the lysosome itself. Thus, a study showing the subcellular localization of ApoD-related Lipocalins once internalized in damaged neurons was required. To study the subcellular traffic of ApoD we switched to its native cell type, astrocytes, where we have reported its protective effect upon oxidative stress challenge, and its transcriptional regulation by the stress-responsive JNK pathway [26].
We studied ApoD subcellular localization by detecting the native protein in astroglial cells (1321N1 astroglioma cell line), avoiding transfections that could alter its physiological traffic. A set of nine different markers for intracellular organelles were used to evaluate colocalization with ApoD in control, low serum (LS) and treatment with the oxidative stress-inducing agent paraquat (PQ) at 2 and 24 hours after stimulus.
Fig 2 and S1 Fig summarize the results of our image analysis from confocal z-stacks of at least 20 cells per condition, selected randomly from two independent experiments with triplicate wells (see Methods). Following two rounds of principal component analysis (PCA), we selected the intensity correlation quotient (ICQ) index [49] referenced to ApoD signal (ApoD ICQ) and the % Pixel Overlap referenced to ApoD signal (ApoD Overlap) to quantify ApoD protein targeted to each organelle (S1 Fig) (see Methods). We use a 2xICQ threshold of 0.1 for a colocalization not to be considered due to chance [50]. Since most variables covariate between control and LS, only PQ referred to the control condition is shown for simplicity in most figures.
ApoD concentrates significantly in Clathrin, EEA1, Lamp-2 and LC3-positive organelles, with particular prominence in the late endosome-lysosome compartment (LELC) positive for Lamp-2 (Fig 2A–2G). Borderline average values of 2xICQ are detected for ApoD colocalization with Caveolin-1 (Fig 2G), though some cells in the sample studied show clear colocalization over the threshold (Fig 2B). A significant PQ-dependent enrichment is observed in ApoD colocalization with Clathrin, Lamp-2 and LC3 at 24h of treatment (asterisks in Fig 2G and S2A–S2E Fig). At shorter times (2h) no PQ-dependent enrichment of ApoD is observed in the lysosomal or autophagosomal compartments of 1321N1 cells, while a prominent colocalization is seen for ApoD-Clathrin (asterisk in Fig 2H).
In opposition, ApoD is not detected in mitochondria or peroxisomes, two organelles involved in oxidative stress generation and management (Fig 2G and S2F–S2G Fig).
As expected for a Lipocalin, 1321N1 cells secrete ApoD to the culture medium (S3A Fig). Also, in ApoD-transfected human embryonic kidney-293T (HEK293T) cells, extracellular ApoD is detected with a stable accumulation in the culture medium over time (S3B Fig). However, no significant colocalization was observed with rough endoplasmic reticulum (RER; S3C Fig) and only borderline values were obtained with the Golgi apparatus (Fig 1A and 1G) of 1321N1 cells. A rapid passage through RER and Golgi in astroglial cells might render ApoD concentration below detection levels in those organelles, a hypothesis supported by the evident ApoD-RER colocalization obtained when ApoD is overexpressed in HEK293T cells (S3D Fig).
To further analyze the spatial domain of ApoD overlap with organelle markers we calculated a percent pixel overlap (referenced to the ApoD signal) as a parameter independent of the fluorescence intensity taken into account in ICQ (Fig 2I). The representation of Clathrin pits, early endosomes, LELC and autophagosomes within the ApoD spatial domain is quite high (18–30%), indicating that these organelles are common residence sites for ApoD. No enrichment in spatial overlap is detected upon PQ treatment, suggesting that the elevated ApoD 2xICQ values reported above (Fig 2G and 2H) represent a stress- and time-dependent increase in ApoD concentration in those stable spatial domains.
Given the prominent colocalization of ApoD with the LELC marker Lamp-2 in 1321N1 cells, we performed a more detailed time-course analysis of their spatial overlap. PQ triggers specifically a significant and transient increase in pixel overlap (Fig 3A), which is accompanied by a change in the distribution of the ApoD signal (Fig 3B): an initial phase of large and less numerous ApoD objects is followed by a late phase with more objects of small average size. This analysis suggests a PQ-dependent early enrichment of ApoD in LELC that might coincide with organelle fusion, probably autophagolysosomes, followed by ApoD traffic to smaller vesicles.
Our colocalization studies, performed in conditions of membrane permeabilization (see Methods), uncovered that ApoD signal was also present in a peripheral dotted pattern after 24h of PQ treatment (Fig 3C, arrow in left panel). By performing ApoD immunolocalization in non-permeabilized conditions, we confirmed that this labeling is due to ApoD presence at the extracellular side of the plasma membrane (Fig 3C, arrowheads in right panel). Our object-based analysis of the plasma membrane located ApoD signal at 2 and 24h of PQ treatment (Fig 3D) shows two phases, which might be indicative of different cell biological processes: (i) An initial phase with increased number and size of objects is concordant with the early enrichment of ApoD in Clathrin-positive organelles (Fig 2H). (ii) Small, but still abundant, membrane-associated ApoD aggregates at 24h might point to a late traffic back to the membrane from the LELC.
The early distribution of ApoD signal in permeabilized cells upon PQ treatment (Fig 2B) suggests organelle fusion, and ApoD is a common resident in LC3-positive organelles of 1321N1 cells (Fig 2F). Therefore, we studied ApoD traffic in the autophagy process. As predicted by the Drosophila in vivo results, we show that ApoD entry in autophagolysosomes is dependent on a proper lysosome-autophagosome fusion since chloroquine (CQ), known to alkalinize lysosomal pH and prevent its fusion [51], completely abolished colocalization of ApoD with LC3 (Fig 3E). We then quantified the colocalization of ApoD with Lamp-2 or LC3 (2xICQ index referenced to ApoD signal; Fig 3F). The presence of ApoD in LELC is relatively stable, while colocalization with LC3 drops below random levels both, when lysosome-phagosome fusion is impaired by CQ or when autophagy initiation is blocked at an early step using 3-methyladenine (3-MA). Likewise, a significant increase in ApoD-LC3 colocalization is observed when autophagy is stimulated by the mTOR inhibitor Rapamycin (Rap).
An object-based analysis of ApoD-Lamp2-LC3 triple colocalization in 1321N1 cells exposed to PQ (Fig 3G and 3H), shows a transient increase in the volume of objects labeled by the three markers (ApoD-positive autophagolysosomes) and a decrease in their number. This suggests that a basal level of autophagic activity exists in these cells, and that after an initial phase of fusions of autophagolysosomes upon PQ treatment, autophagy is resolved with a net decrease of objects with the three markers. These data, together with the increased ApoD-LC3 colocalization after 24h PQ (Fig 2G) and a net PQ-dependent decrease in the number of LC3-positive vesicles (S4 Fig) suggest that ApoD is enriched in autophagolysosomes upon oxidative challenge.
The subcellular localization of ApoD in 1321N1 cells in control and PQ conditions was further confirmed by morphological criteria of immunoelectron microscopy (Fig 4). In control conditions ApoD was mostly detected in early endosomes, small and located close to the plasma membrane (Fig 4A), as well as in lysosomes, electron- dense vesicles close to the nucleus (Fig 4B). Clear differences are observed upon PQ treatment in the representation of labeled subcellular localizations. After 2h of PQ treatment ApoD signal was observed on the plasma membrane (Fig 4C), being endocytosed in Clathrin-coated pits (Fig 4D), in larger late endosomes (Fig 4E), electron-dense lysosomes (Fig 4F), secondary lysosomes (Fig 4G), and autophagolysosomes (Fig 4H). In most cases, ApoD signal is observed associated to membranes. Interestingly, lysosomes under oxidative stress show an enrichment of membrane-associated ApoD (Fig 4F) compared to control conditions (Fig 4B).
Since the LELC and autophagolysosomal compartments participate in protein degradation, we tested whether the ApoD passage through them simply reflects its degradation pathway. We used cells not expressing ApoD (HEK293T) subjected to a 2h pulse-chase experiments with purified human ApoD [39]. After a 2h period of ApoD exposure, ApoD labeling was evaluated in cells for up to 48h. The intracellular content of ApoD was very stable along the experimental period when we used native human ApoD (Fig 5A). However, a bacterially produced non-glycosylated human ApoD is rapidly endocytosed and quickly degraded (Fig 5B). Colocalization with Lamp-2 (Fig 5C and 5D) shows that bacterial recombinant ApoD reaches the LELC, but the signal disappears with a fast time course. Using immunoblot, we have estimated that only 30% of endocytosed native human ApoD is lost during a division cycle in HEK293T cells (Fig 5E and 5F), and a fraction of that loss corresponds to ApoD secretion to the culture medium (S3B Fig).
These experiments demonstrate a very stable presence of native ApoD in the intracellular compartments analyzed above, and particularly in LELC, which suggests a functional role there. It is well known that proteins with essential functions within lysosomes bear carbohydrate shields against proteolysis [52]. In addition to the stable Lipocalin folding, ApoD N-linked glycosylation [53] might be responsible for such biochemical stability in protease-rich environments.
To further assay whether the Lipocalin folding can be stable in the acidic lysosomal lumen, we performed ligand binding assays at pH 7.0 and 5.1 (Fig 5G; pH chosen in light of the pH distribution of ApoD-positive lysosomes under stress conditions, see below). Binding to retinoic acid, a generic hydrophobic ligand known to bind all human Lipocalins tested so far [19], indicates that ApoD ligand binding is functional under acidic conditions.
An acidic pH is a defining functional property of lysosomes that allow their protease and lipase activities to be tightly controlled within the cell in addition to influence lysosomal fusions and traffic. We set to study the functional relationships of ApoD to lysosomal pH by using the membrane permeable LysoSensor Yellow/Blue DND-160, a ratiometric dye specifically targeted to all lysosomes, and not only those reaching the LELC through the endocytic pathway ([54,55]; see Methods and S5 Fig). We used either excitation analysis in cell populations (S5A Fig), suitable for pH 4.0–6.0, or emission spectral analysis of single lysosomes in confocal microscopy optical sections combined with ApoD immunodetection (S5B–S5D Fig; Linear fit for pH 4.0–5.5).
We first analyzed the effects of the treatments used in our experimental paradigm on the average lysosomal pH in 1321N1 cell populations (Fig 6A). Chloroquine (CQ) was used as positive control. Two hours of PQ treatment resulted in significant lysosomal alkalinization (with an average increase of 0.5 pH units), in agreement with the reported sensitivity of lysosomal membranes to oxidative stress [14] resulting in proton leakage that counteracts pH gradient generation mechanisms.
When pH was measured at the single organelle level combined with ApoD labeling, we discover a striking difference in the frequency distributions of pH values (Fig 6B). Lysosomes without ApoD show a narrow pH distribution with a frequency maximum at pH 4.4, while ApoD-positive lysosomes show a broaden distribution (range: 4.4–5.5) and a mode at pH 4.7 still in the range of lysosomal pH. This difference observed in control conditions is maintained when lowering serum in the culture medium. When we apply PQ, ApoD-negative lysosomes suffer a mild alkalinization (average peak shifts 0.2 pH units; Fig 6C) that is persistent after 24h of treatment. However, when the ApoD-positive lysosome pool is analyzed (Fig 6D), a larger pH increase is observed (average of 0.5 pH units). This alkalinization is transitory, since lysosomal pH distribution returns after 24h of PQ treatment to more acidic values, within the range of ApoD-positive lysosomes in control conditions. These data support the existence of subsets of lysosomes differing not only in their pH and ApoD content, but also in their sensitivity and response to an oxidative insult.
Our results show that ApoD is present in a subset of lysosomes specifically sensitive to oxidative stress that undergo a large but reversible alkalinization in response to PQ. Is ApoD responsible for these changes or it locates in lysosomes passively subjected to the PQ-triggered pH shifts?
To test whether there is a causal relationship between ApoD presence and lysosomal pH we added exogenous purified human ApoD to non-expressing HEK293T cells, either simultaneous to PQ exposure (2h treatment) or sequentially, by first provoking lysosomal alkalinization with PQ (2h) and then adding exogenous ApoD (Fig 7A). We have already shown that by 2h ApoD has entered the LELC in HEK293T cells (Fig 5D). The addition of ApoD prevents the PQ-triggered alkalinization and is able to reverse an already established effect of PQ (asterisks Fig 7B).
We also added purified ApoD to differentiated SH-SY5Y neurons (Fig 7C). Here a longer simultaneous ApoD-PQ treatment was applied due to the dynamics of ApoD entry into neuronal LELC (see below, Fig 8). Again, ApoD is able to prevent (particularly prominently in the long 24h treatment), as well as to revert, the alkalinizing effects of PQ (asterisks in Fig 7C). Lack of ApoD labelling in the absence of exogenous addition of hApoD indicates that SH-SY5Y neurons do not express ApoD, as has been recently confirmed by RNAseq techniques (http://systemsbiology.uni.lu/shsy5y/). Therefore, the observed rescue of lysosomal pH can only be attributed to an endocytosis-mediated mechanism (Fig 8).
Aside of modifying ApoD presence by exogenous addition, we cultured murine primary astrocytes from wild type (WT) or knock-out (ApoD-KO) mice (Fig 7D). Using this model, we demonstrate: 1) The PQ-dependent alkalinization and re-acidification of 1321N1 lysosomes is also present in WT primary astrocytes. 2) ApoD-KO lysosomes do increase their pH upon oxidative stress, therefore indicating that the alkalinization itself is not related to ApoD presence. 3) No re-acidification is achieved in the absence of ApoD.
Summarizing, our data show that ApoD is present in a particular subset of PQ-sensitive lysosomes, which inevitably alkalinize in the presence of oxidative stress, an effect known to be due to lysosomal membrane damage [14,56]. We also show that ApoD is responsible for the pH recovery of PQ-challenged lysosomes. These data support the hypothesis that lysosomal membrane recovery is compromised in the absence of ApoD, and that this Lipocalin, with its structure and lipid-binding properties preserved, contributes to the repair of damaged lysosomal membranes in astrocytes and neurons.
We have shown that ApoD affects the differential response of a subset of lysosomes to oxidative stress by actively promoting pH recovery. Then, is the entrance of ApoD to the lysosomal compartment a constitutive or a regulated process?
A Lamp-2/ApoD colocalization analysis in differentiated SH-SY5Y neurons after a two-hour pulse of exogenous human ApoD (Fig 8A and 8B) shows that, in contrast with the observed rapid entrance into the LELC compartment in HEK293T cells (Fig 5), differentiated neurons show no colocalization in control conditions 2h after ApoD addition (Fig 8A and 8B). Colocalization is evident later on, after a 24h chase (Fig 8B). In contrast, a significant colocalization is observed at 2h in the presence of PQ, and keeps increasing during the 24h post-ApoD supplementation period (Fig 8B). Thus, ApoD entry into lysosomes in neurons is dependent on stress conditions, being slow in the absence of stress but very quick in its presence.
A similar phenomenon is observed with native murine ApoD in primary astrocytes (Fig 8C), though with a remarkable feature: In control condition WT astrocytes show a clear predominant plasma membrane localization of ApoD and no colocalization with Lamp-2. ApoD entrance into the LELC is PQ-dependent and has a fast time course, with a substantial colocalization at 2h.
We can conclude that the accelerated targeting of ApoD to the lysosomal compartment of neurons and primary astrocytes under stress conditions is a regulated process. To further explore the consequences of this process we measured the levels of 4-hydroxynonenal (4HNE) (Fig 9A, 9B and 9D), a lipid peroxidation-derived adduct, as a proxy for oxidative stress in each experimental condition. Interestingly, primary WT astrocytes have very low 4HNE levels in control conditions; they increase their oxidation levels by 2h of PQ treatment, but return to basal levels by 24h (Fig 9A and 9D). By contrast, ApoD-KO astrocytes show a significant basal level of oxidative stress and fail to recover from the insult (Fig 9B and 9D), further demonstrating the reported protective role of ApoD upon oxidative stress [26]. Starting with a higher oxidation state, the level achieved by 2h PQ in ApoD-KO astrocytes matches WT levels, indicating that cells have reached a 4HNE maximum. Therefore, ApoD entrance into lysosomes of WT astrocytes occurs at the peak of oxidative stress.
If ApoD targeting to lysosomes is regulated by oxidative stress, we wondered why there is a significant colocalization of ApoD with Lamp2 in 1321N1 astrocytoma cells under control conditions (Figs 2–4). Fig 9C demonstrates that 1321N1 cells have a high basal oxidative stress, coherent with the high metabolic rate of cancer cells. Thus, ApoD is targeted to lysosomes under these conditions. The levels of 4HNE further increase in response to PQ treatment, and are followed by a substantial clearance of 4HNE adducts. This rebound effect is consistent with the pH recovery effect observed in our single-lysosome study (Fig 6D), and suggests that oxidative-challenged cells transiently activate protective mechanisms (ApoD among them) that clear lipid peroxidation products. An additional analysis of the 1321N1 single lysosome data (Fig 9E) shows a low proportion of ApoD-positive lysosomes in control conditions, and a significant early increase upon PQ treatment that is maintained by 24h (overriding the proportion of ApoD-negative lysosomes). Thus, the mechanism controlling lysosomal targeting of ApoD also occurs in the astrocytoma cell line 1321N1. The efficient clearance of oxidized products might be the result of a very efficient lysosomal function, contributed by the ApoD-dependent re-acidification after PQ insult.
An active role of ApoD in oxidized lysosomes was further demonstrated by measuring lipofuscin, a reported readout of lysosomal pH dysfunction that results in cellular accumulation of damaged oxidized macromolecules [57]. A spectral analysis of confocal microscopy images of WT and ApoD-KO primary astrocytes (Fig 9F and 9G) shows that the absence of ApoD generates a significant increase in lipofuscin signal in astrocytes.
If the mechanism by which ApoD controls lysosomal pH is due to lysosomal membrane stabilization, two predictions can be made for lysosomal behavior in the absence of ApoD: 1) lysosomal proteases would decrease their activity, and 2) cytoplasmic proteins would aberrantly enter lysosomes. A Cathepsin B activity assay (Fig 10A–10C) demonstrate the first prediction. In WT astrocytes (Fig 10A and 10C), PQ-triggered oxidation reduces Cathepsin B activity, but a clear recovery takes place upon prolonged PQ exposure in parallel with lipid-peroxide clearance (Fig 9A and 9C). In the absence of ApoD (Fig 10B and 10C), Cathepsin B activity is significantly reduced in basal conditions, further deteriorated upon PQ treatment, and no recovery is obtained after 24 h treatment.
Galectin-3 immunocytochemistry (Fig 10C) demonstrate the second prediction. This lectin shows a diffuse cytoplasmic location in control WT astrocytes cultures, but translocates to leaky lysosomes undergoing oxidative stress-dependent membrane permeabilization upon PQ insult, giving a punctate labeling pattern [58]. Astrocytes lacking ApoD show evident Galectin-3 puncta in control conditions (arrows in Fig 9H), and the effect is further increased upon 2h PQ.
In summary, entrance of ApoD into the LELC, and particularly into lysosomes, is actively promoted in pro-oxidative conditions in neurons and astrocytes. Our data demonstrate that ApoD, contributing to lysosomal pH recovery upon oxidative stress, is recruited to a vulnerable subset of lysosomes, where it helps to keep lipid peroxides levels under control and to safeguard lysosomal functional integrity by avoiding lysosomal membrane permeabilization.
These findings represent a novel function for ApoD, and for Lipocalins. Their active role in the lysosomal compartment provides a clear explanation for the mechanism of GLaz rescue of polyglutamine-based neurodegeneration [37]: it requires fusion of autophagosomes to healthy lysosomes to optimize autophagy (Fig 1). Moreover, we have reported an protective role for ApoD in the functional recovery of injured mammalian peripheral nerves by a mechanism regulating myelin phagocytosis efficiency [32]. Taking into account our current results, the mechanistic link between these two apparently unrelated biological processes is clear: The direct control by ApoD of lysosomal function efficiency.
We have centered this work on astrocytes, the front line of defense against oxidative stress and one of the nervous system cell types that express ApoD. Since ApoD conditions the pH-dependent functionality of the lysosomal compartment, how does it affect biological processes where a lysosomal optimal function is important for astrocytes?
Recent studies reveal that astrocytes have phagocytic functions [59,60,61]. They digest the phagocytosed cargo through a process regulated by lysosome pH levels and autophagosome-lysosome fusion [62]. This is a slow process in astrocytes, and is proposed to regulate antigen presentation by these cells through lysosomal fusion to the plasma membrane. We have shown that lysosomal ApoD either enters autophagolysosomes or traffics back to the plasma membrane (Fig 3), and that lysosomal pH depends on the presence of ApoD (Figs 6 and 7). Therefore, ApoD might be a candidate regulator for the “digest-or-present” process in astrocytes.
Astrocytes are reported to start degrading phagocytosed cargo in acidic lysosomes at 6h after exposure to cell debris [62]. We thus exposed primary astrocytes to DiI-labeled myelin for 3 days and monitored DiI signal at 2 and 6 days after myelin removal (Fig 11). We estimated the phagocytosis potential of WT and ApoD-KO astrocytes by measuring the number and size of DiI-labeled myelin particles (see Methods). Starting with comparable initial levels of phagocytic activity (Fig 11A; no differences are observed in the number of myelin particles phagocytosed during the 3h exposure period), both astrocyte genotypes decrease the number of particles over time. However, this reduction in numbers is accompanied by a significant increase in large myelin particles at 6 days post-myelin exposure in ApoD-KO astrocytes only (Fig 11B, 11E and 11F), indicating that phagocytosis resolution is impaired in the absence of ApoD. These results agree with the high load of phagocytosed material observed in alkaline astrocyte lysosomes [62]. Taking into account that myelin phagocytosis induces ROS production [63], the more alkaline ApoD-KO lysosomes are expected to be less efficient, resulting in delayed myelin degradation. No differences are found at earlier times, suggesting that the absence of ApoD results in a deregulated processing of the already ingested myelin, requiring the maintenance of lysosomal optimal pH and functionality.
This study reveals for the first time a functionally complex stress-dependent traffic of the Lipocalin ApoD (Fig 12A and 12B) that is the base of a lysosomal protecting mechanism previously unknown (Fig 12C). While plasma membrane and endosomes are typical cellular locations for ApoD secreted under basal conditions (Fig 12A), lysosomes become an essential and stable niche for ApoD in a cell suffering from oxidative stress (Fig 12B). After a fast protein secretion through the canonical RER-Golgi pathway, stress conditions trigger ApoD endocytosis. Clathrin-dependent endocytosis is particularly favored early under oxidative stress conditions. ApoD then moves through the early endosomal compartment to reach the LELC. Subsequently, Lamp-2/ApoD-positive organelles either enter the autophagy pathway (early after oxidative stress stimulus), or ApoD is targeted back to the membrane at a late phase, possibly travelling within secretory lysosomes.
Our results explain important aspects of the ApoD neuroprotective mechanism previously unanticipated. 1) ApoD locates in a subset of lysosomes particularly sensitive to oxidative stress. These results support previous work reporting a functional heterogeneity of lysosomes according to their pH, vulnerability, pro-oxidative activity, or position within the cell [14,17,64]. 2) ApoD stability within lysosomes is dependent on its glycosylation state, which is functionally relevant given its reported heterogeneity in different tissues and cell types [65]. This result could also explain the absence of neuroprotective effects of bacterial recombinant ApoD against Aβ-challenged neuronal cells [66]. 3) ApoD targeting to lysosomes is a controlled process promoted by oxidative stress in ApoD-expressing astrocytes and ApoD-non-expressing neurons, which explains both autocrine and paracrine protective effects [26,36,37]. 4) ApoD behaves as an acute phase protein, finely tuned through a JNK pathway-dependent transcriptional expression [26,44] coordinated with a stress-dependent accelerated entry into lysosomes.
Another remarkable finding derived from our experiments is a functional link between the presence of intact ApoD within lysosomes and their pH. ApoD shows very low expression and seldom locates inside lysosomes of primary astrocytes cultured in control conditions. It is also found with more probability in endosomes than in lysosomes of control 1321N1 cells. However, it is rapidly up-regulated and mostly found in lysosomes upon oxidative stress. This situation is achieved in native cells after just 2h of PQ treatment, though it occurs basally in rapidly proliferating cell lines like the 1321N1 astroglioma cells. Oxidative stress makes ApoD to translocate quickly and specifically to lysosomes with a slightly alkalinized pH in basal conditions and high sensitivity to oxidation, which underlie their functional vulnerability. Nevertheless, the presence of ApoD in these lysosomes is necessary and sufficient for restoring lysosomal pH to normal values after oxidation-dependent alkalinization, as confirmed by the stably alkalinized PQ-challenged ApoD-KO cells. This result holds for cells that endogenously express ApoD, like astrocytes, and for non-expressing cells exogenously supplied with ApoD, like neurons.
Lysosomal alkalinization is known to result from ROS-induced membrane permeabilization [67], and we find clear signs of lysosomal dysfunction and membrane permeabilization in ApoD-KO lysosomes (Fig 10) together with high levels of lipid peroxidation that can be counteracted only in the presence of ApoD expression (Fig 9A–9D). Complex processes like bidirectional protein traffic along the endosome-lysosome compartments, or altered transcription, translation or trafficking of proton pumps to the lysosome, could result in pH changes similar to the observed ones. However, the time course of ApoD effects on lysosomal pH and the evidences of lysosomal permeabilization strongly support a direct effect of ApoD on lysosomal membranes. Such a mechanism is concordant with ApoD biochemical properties, its lipid binding properties (preserved at acid pH in the range of ApoD-positive lysosomes; Figs 5 and 6), and its membrane association, including lysosomal membranes (Fig 4). As a lipid peroxidation counteracting agent [25], ApoD restores the integrity of damaged lysosomal membranes (Fig 12C).
It is of special interest to compare the beneficial repair of lysosomal function by ApoD with the effects of other extracellular lipid binding proteins known to have effects on lysosomes. The Lipocalin Lcn2 reduces lysosomal degradative activity, resulting in insulin resistance in cardiomyoblasts [68]. Apolipoprotein E (also expressed by astrocytes and consistently related to neurodegenerative phenotypes) has deleterious effects on lysosomal function, as the ApoE4 allele causes lysosomal leakage and apoptosis [69].
That ApoD helps to maintain H+ gradients under oxidative stress in glial and neuronal lysosomes is a finding with high explanatory value in the understanding of lysosomal mechanisms of protection and of ApoD function. Lysosomal dysfunction does compromise cell resistance to oxidative stress, the major phenotypic hallmarks of all loss-of-function manipulations performed so far with ApoD and its related Lipocalins in animal models and cellular systems [26,29,33,36,38,39]. Moreover, a failure of lysosomal function is linked to inefficient toxic protein clearance in proteinopathies like SCA1 that ultimately leads to cell death and neurodegeneration. This study explains not only why neurodegeneration rescue by GLaz [37] depends on the lysosome-autophagosome fusion (Fig 1), but also why phagocytosis resolution in astrocytes (Fig 10) and after peripheral nervous system injury [32] is compromised. Interestingly, a recent report shows a similar delay in clearing myelin from injured nerves when lysosomal function is inhibited [70].
This study demonstrates that ApoD contributes significantly to the evolutionarily conserved mechanism of protecting cells by protecting their lysosomes. ApoD could be the first lysosomal marker known to be specific for a particular subset of lysosomes: the most vulnerable to oxidative stress. Therefore, ApoD localization assessment should provide a useful tool for characterizing many physiological and pathological situations.
Our work focuses on the endogenous mechanisms of protection in the nervous system, where astrocytes are central players and neurons are especially vulnerable cells. However, the known functional pleiotropy of ApoD warrants that it will be relevant to many other biological processes and pathological situations. Understanding ApoD actions in lysosomes will open the possibility of manipulating this mechanism for therapeutic purposes, using ApoD as a carrier to reach the lysosome.
Although previously considered to play a lipid transport function in various body fluids, the Lipocalin ApoD can now be recognized as a relevant acute phase protein contributing to the nervous system response to stress, injury, neurodegeneration and aging by stabilizing the membrane of vulnerable lysosomes. This finding identifies a new lipid binding protein-dependent cellular mechanism by which lysosomes are functionally protected against oxidative stress.
The cell lines 1321N1, HEK293T, and SH-SY5Y were obtained from Sigma-Aldrich and ATCC. Cells were grown at 37°C in humidity-saturated atmosphere containing 5% CO2. The culture medium was replaced twice a week, and cells were subcultured at 90% confluence.
The human astrocytoma cell line 1321N1 was cultured in Dulbecco-modified Eagle's medium (DMEM; Lonza), supplemented with heat-inactivated 5% fetal bovine serum (FBS), 1% L-glutamine (final concentration 2 nM), and 1% P-S-A stock (final concentration: 100 U/ml penicillin, 100 U/ml streptomycin, 0.25 μg/ml amphoterycin B).
The human neuroblastoma cell line SH-SY5Y was cultured in DMEM supplemented with 4.5 g/l glucose, heat-inactivated 10% FBS, 1% L-glutamine, 1% P-S stock (final concentration: 100 U/ml penicillin, 100 U/ml streptomycin) and 1% nonessential amino acids (Lonza). SH-SY5Y cells differentiation was achieved by culture on collagen-treated plates with medium supplemented with 3% FBS and Retinoic acid (10 μM). A 72h differentiation period was allowed before experiments.
HEK293T cells were cultured in DMEM supplemented with 4.5 g/l glucose, 1% L-glutamine, 1% P-S-A, and 10% FBS.
For the exogenous addition of ApoD, human ApoD purified from breast cystic fluid [39] or recombinant human ApoD from E. coli (ProSpec) were added (10 nM) to the cell cultures for 2h.
Cells treated with Paraquat (PQ; 500 μM; 1-24h), Chloroquine (CQ; 20 μM; 1h), Rapamycine (Rap; 2 μM; 2h), and 3-Methyladenine (3-MA; 5 mM; 2h) were cultured in phenol red-free DMEM supplemented with 1% L-glutamine, 1% P-S, and 0.2% charcoal stripped FBS. This medium without additives was used as our low-serum (LS) condition.
The expression constructs used in this work (pcDNA3.1-ApoD; pHSVer-GA; pHSVGA-cox8; pCDNA3.1-tgoGAm) were transiently transfected into cell lines using Lipofectamine LTX reagent (Invitrogen) according to the manufacturer’s protocol.
ApoD-KO mice were generated by homologous recombination [29] maintained in positive pressure-ventilated racks at 25±1°C with 12 h light/dark cycle, fed ad libitum with standard rodent pellet diet (Global Diet 2014; Harlan Inc., Indianapolis, IN, USA), and allowed free access to filtered and UV-irradiated water. In order to avoid potential maternal effects of ApoD, and to generate WT and ApoD-KO mice of homogeneous genetic background, the experimental cohorts used in this study are the F1 generation of homozygous crosses of ApoD −/− and ApoD +/+ littermates born from heterozygous crosses of an ApoD-KO line backcrossed for over 20 generations into the C57BL/6J background.
We used neonatal (0–1 days old) mice of two genotypes: ApoD-KO and their WT littermates. Cerebral cortices were quickly extracted, their meninges removed by rolling on a sterile filter paper, and pieces of cortex placed in Earle’s Balanced Salt Solution (EBSS) containing 2.4 mg/ml DNAse I and 0.2 mg/ml bovine serum albumin (BSA). Tissue was minced with a surgical blade; centrifuged (200 g, 2 min); incubated with 10 mg/ml trypsin for 15 min at 37°C (incubation terminated by 10% FBS addition); mechanically dissociated with a Pasteur pipette and centrifuged (200 g, 5 min). The last two steps were repeated, and the resulting cells were resuspended in DMEM with 10% FBS, 1% L-glutamine, 1% P-S-A. Cells were plated onto culture flasks and incubated at 37°C in 5% CO2 with 90–95% humidity, and the culture medium was replaced weekly. Cell cultures were used for experiments after two subculture steps, when >99% of cells are astrocytes [26].
Cells attached to poly-L-lysine (SIGMA) treated coverslips were fixed with 4% phosphate-buffered formaldehyde. Following washes in phosphate-buffered saline (PBS), the cells were blocked and permeabilized with Tween-20 (0.1%) and 1% non-immune (goat or donkey) serum. Cells were incubated overnight at 4°C with the following primary antibodies. Rabbit serum anti-human ApoD [custom made by Abyntek Biopharma against purified ApoD [39], or generated by Dr. C. López-Otin]. Goat serum anti-mouse ApoD (Santa Cruz Biotechnology). Mouse serum anti-clathrin LCA (Santa Cruz Biotechnology). Mouse serum anti-caveolin-1 (Santa Cruz Biotechnology). Goat serum anti-catalase (Santa Cruz Biotechnology). Mouse serum anti-EEA1 (BD Biosciences). Mouse serum anti-LC3 (MBL). Rat monoclonal anti-Galectin-3 (American Type Culture Collection, ATCC). Goat serum anti-4HNE (Alpha Diagnostic).
For immunolabeling with mouse monoclonal anti-human Lamp-2 and rat monoclonal anti-mouse Lamp-2 (DSHB), cells were blocked with 1% BSA, 10% normal goat serum, and 0.1% saponin in PBS, and incubated with primary and secondary antibodies for 1 h at room temperature.
Alexa Fluor 594 and 488 (Jackson Labs) or DyLight 405 (Thermo Scientific)-conjugated IgGs were used as secondary antibodies for fluorescence immunocytochemistry. After washes in PBS, the preparations were mounted with EverBrite Mounting Medium with DAPI, and sealed with CoverGrip Coverslip Sealant (Biotium).
Cell lysates, or culture media (either directly or concentrated twenty times by 0.22 μm filter centrifugation), were collected to analyze the amount of ApoD. Immunoblot analyses were performed with proteins transferred to PVDF membranes using standard procedures, and exposed to rabbit serum anti-human ApoD and HRP-conjugated goat-anti-rabbit (Santa Cruz Biotechnology). An HRP-conjugated anti-β actin antibody (Sigma) was used to normalize protein loads. Membranes were developed with ECL reagents (Millipore), and the signal visualized with a digital camera (BioRad). The integrated optical density of the immunoreactive protein bands was measured in images taken within the linear range of the camera, avoiding signal saturation.
These methods were performed as previously described [39]. Fluorescence measurements were conducted with a Shimadzu RF-5301PC spectrofluorometer in a quartz cuvette (105.251-QS, 3-mm path length; Hellma). Temperature was held at 22±0.1°C. Excitation wavelength was 295 nm (selective for tryptophan residues). Emission was recorded at 327–400 nm with slit width set at 5 nm. Purified human ApoD was diluted to 0.5 μM with 10 mM phosphate buffer (binding at pH 7.0), or 30 mM sodium citrate (binding at pH 5.1). The ligand retinoic acid (RA) was dissolved in dimethylformamide (DMF; Sigma). The mixture was equilibrated for 3 min in the dark before the fluorescence was recorded.
The fluorescence spectrum in the presence of ligand was subtracted from DMF baseline obtained mixing the protein with the same amounts of carrier without ligand. Binding was assayed with RA 5 μM (1:10 protein:ligand concentration) and a DMF final concentration of 0.005%.
Labeled cells were visualized with an Eclipse 90i fluorescence microscope (Nikon) equipped with a DS-Ri1 (Nikon) digital CCD camera. Images were acquired under the same conditions of illumination, diaphragm and condenser adjustments, exposure time, background correction and color levels.
Confocal images were obtained with a 63x oil immersion objective (HCX PL Apo CS NA = 1.4; Leica) attached to a confocal DMI 6000B microscope with a TCS SP5 confocal system (Leica) equipped with AOBS and AOTF systems. Fluorophores were excited with WLL laser (Leica) and a 405 line (Leica) controlled by LAS AF software (Leica). Emissions were collected with the AOBS system and three spectral detectors. Laser power and detection gains were set by scanning control samples labeled with secondary antibody alone. We ensured to obtain similar dynamic ranges in our images, and adjusted gain and offset using LUTs. In this manner, bleedthrough can be neglected. Negative control images showed very weak and homogeneous background. We obtained confocal sections under constant conditions to minimize image acquisition variation. Images were stored as 1024x1024 pixels and 8-bit TIFF files.
Z-series (xyz scan) were performed. The number of z-stacks was determined by observing the limits of the cell membrane. The focus plane was set to be 3 μm beneath the section surface. The optimal value of the step size was calculated for the wavelength used to fulfill the Nyquist theorem. The optical section thickness was 0.772 μm. Besides, images were taken with a 4x zoom, reducing field size. Pixel size corresponded to 0.06*0.06*0.3777 μm3. Scanning was performed with a 1.0 Airy unit pinhole size.
Images were processed with a Gaussian Blur filter [Sigma (Radius): 1.00], to facilitate object detection, and analyzed with the Colocalization Indices plug-in [50] and the 3D Object Counter tool using FIJI software. To analyze triple-colocalization experiments we used the Image Calculator and 3D Object Counter tools of FIJI.
A principal component analysis (PCA) was performed on the 54 different variables per cell retrieved from our image analysis to reduce its dimensionality. Nine components were found with informative value, and three of them explained over 55% of the data variability. Intensity correlation quotient (ICQ) [49] variables were heavily represented in the first component. Pixel overlap proportions, and number and volume of objects were also of interest. Thus, we run a second PCA with 11 variables (S1A Fig). ICQ (variables 1–3) and Pixel Overlap (variable 4) presented the largest weight in component 1 (accounting for 31.5% of variability), and relative Overlaps (variables 5–7) were the most important for component 2 (explaining 21.3% of variability). The first component score of this PCA was used to assess for global statistical differences due to the experimental conditions (S1B Fig). Only EEA1 and Lamp2 showed significant variation between control and PQ condition (Two-way ANOVA, Holm-Sidak post-hoc method, p < 0.05).
This multivariate analysis helped us to focus on two main colocalization variables relative to ApoD signal (1 and 5; arrows in S1A Fig) and the number and volume of ApoD-positive objects (variables 8 and 11) to understand the dynamics of ApoD traffic in the cell. It also helps us to focus on the PQ-dependent changes, since most variables covariate between control and LS conditions in our two-way ANOVA study.
Lysosomal pH was measured using the dye LysoSensor Yellow/Blue DND-160 (Life Technologies) as described [71,72,73,74]. As a ratiometric dye, the LysoSensor readout is independent of concentration. Moreover, because it is membrane permeable, its readout is representative of a broad range of lysosomes in comparison with a dextran-tagged probe that reach lysosomes by endocytosis. Experimental parameters such as incubation time and dye concentration have been set to minimize variation and to give the best signal-to-noise ratio [74].
Cells were grown to >80% confluence in black 96-well plates (NUNC), plated in alternating rows to control for any signal variation across the plate. After removal of medium and washes with PBS, cells were incubated for 3 min with 2 μM LysoSensor Yellow/Blue in isotonic solution (NaCl, 105 mM; KCl, 5 mM; HEPES-Acid, 6 mM; Na-HEPES, 4 mM; NaHCO3, 5 mM; mannitol, 60 mM; glucose, 5 mM; MgCl2, 0.5 mM; CaCl2, 1.3 mM; pH adjusted to 7.4). Dye loading and incubation steps were carried out at room temperature in the dark. After 3 min, cells were rinsed three times in isotonic solution, and incubated with either additional isotonic solution or with pH calibration buffers. After 10 min fluorescence was measured with a GENios Pro Fluorometer and recorded using the XFluor4GENiosPro software package (TECAN). Lysosomal pH was determined from the ratio of excitation light at 340 nm and 390 nm (F340 nm/F390 nm, 535 nm emission) [16]. Mean light levels at both excitation wavelengths were integrated over 2000 μs and recorded for each well. This step was repeated after 5 ms for each sample. The final calculated pH represents the mean of six measurements taken strictly 12 min after dye removal, as LysoSensor generates lysosomal alkalinization with longer incubation times.
Absolute pH levels were obtained by calibrating lysosomal pH against standards. Cells present in calibration wells were incubated with 15 μM monensin and 30 μM nigericin (SIGMA), proton-cationophores that permeabilize the lysosomal membrane to Na+ and K+, respectively. These ionophores were added in a solution of 20 mM MES (2-(N-Morpholino)ethanesulfonic acid), 110 mM KCl and 20 mM NaCl, with pH 4.0, 4.5, 5.0, 5.5 and 6.0, forcing lysosomes to equilibrate with those pH values (S5A Fig).
For LysoSensor staining, cells attached to poly-L-lysine coverslips were washed in warm isotonic solution, and incubated with isotonic solution or pH calibration buffers. Confocal fluorescence images were then obtained by exciting at 405 nm, and the emission collected at 420–700 nm (λ scan (xyz)) taking two averaged images every 10 nm.
Using the LAS AF Lite software (LEICA), we created ROIs for each lysosome and their emission spectra were obtained. Each spectrum was fitted to a five parameters Weibull’s equation (S5B and S5D Fig). The emission 470/524 nm ratio was then calculated for each sample. The lysosomal pH values were determined from the standard curve generated with the pH calibration samples (S5C Fig). Before cells were fixed and permeabilized a white field image was taken in the confocal microscope (S5D Fig step 4). Immunocytochemistry was then performed to detect ApoD in lysosomes, and each cell was position-identified to overlap ApoD to LysoSensor signals (S5D Fig steps 5–7). Images were analyzed with FIJI, and estimate the proportion of ApoD-positive and negative lysosomes we used the Image Calculator tool.
Lysosomal Cathepsin B activity was measured using the Magic Red assay as directed by manufacturer (InmmunoChemistry Technology, LLC). Briefly, cells are exposed in vivo to a membrane-permeable non-fluorescent substrate that yields a red-fluorescent product upon enzymatic cleavage. Both, confocal microscopy (590 nm excitation / 630–640 nm emission) and cell population analysis in 96-well plates (570 nm excitation / 612 nm emission) were performed. L-leucyl-L-leucine methyl ester (Cayman Chemical), 2.5 mM, was used as positive control for lysosomal membrane rupture.
1321N1 cells, under control or PQ conditions, destined for pre-embedding immunogold labeling of ApoD were fixed in 4% formaldehyde and 0.3% glutaraldehyde in 0.1 M phosphate buffer (PB) pH 7.4 for 30 min at 4°C. Following washes in 0.1 M PB, the cells were blocked with 0.1% cold water fish skin gelatin and permeabilized with Tween-20 (0.5%) in Tris-buffered saline (TBS; 20 mM Tris-HCL, 150 mM NaCl). Cells were incubated for 48h at 4°C with rabbit serum anti-human ApoD primary antibody [custom made by Abyntek Biopharma against purified ApoD [39]] diluted 1:500 in TBS containing Tween-20 (0.5%) and 0.1% cold water fish skin gelatin. Cells were later washed several times and incubated with ultra-small gold-conjugated goat anti-rabbit secondary antibodies (EMS, Electron Microscopy Sciences) diluted 1:50 in PBS for 48h at 4°C. After several washes with PBS, cells were post-fixed in 2% glutaraldehyde in PBS for 20 min, washed and the ultra-small gold particles were silver enhanced for 20 min at room temperature with AURION R-Gent SE-EM (Silver Enhancement for Electron Microscopy) (EMS, Electron Microscopy Sciences) following manufacturer’s indications. Later, cells were post-fixed with 0.5% OsO4 in PBS for 20 min at 4°C and washed with PBS. Cells were then dehydrated through a graded series of ethanol, and embedded in Epoxy EMbed-812 resin (EMS, Electron Microscopy Sciences). Ultrathin sections were obtained with an Ultracut E ultramicrotome (Reichert/Leica), contrasted with uranyl acetate and lead citrate, and analyzed using a JEOL JEM-1011 HR electron microscope with a CCD Gatan ES1000W camera with iTEM software.
Myelin isolation and labeling were previously described [32].
Fluorescence microscopy was used to visualize myelin uptake by astrocytes. For this purpose, 2.5×105 cells were cultured on 12 mm coverslips at 37°C in 5% CO2. Cells were incubated with 5 μg of DiI-labeled myelin (10 μg/ml) for three days. After incubation, non-endocytosed and unbound myelin was removed by washing twice with PBS, and cells were grown at 37°C for 2 or 6 days.
Cells were fixed in 4% formaldehyde in PBS, washed in PBS and mounted with EverBrite Mounting Medium with DAPI. DiI-labeled particles were visualized with a fluorescence microscope as described above. Number and size of myelin particles ingested by cultured astrocytes were measured from thresholded images with FIJI.
Flies were grown in a temperature-controlled incubator at 25°C, 60% relative humidity, under a 12 h light—dark cycle. They were fed on wet yeast 84 g/l, NaCl 3.3 g/l, agar 10 g/l, wheat flour 42 g/l, apple juice 167 ml/l, and propionic acid 5 ml/l. Fly females were used in all experiments. We used the line gmr:GAL4 to drive transgenes expression to the eye photoreceptors. UAS:hATXN182Q was used to trigger the neurodegenerative phenotype [75], combined with UAS:GLaz [37] and/or UAS:Dor-RNAi [48]. Recombination of two of the elements present in the second chromosome (gmr:GAL4 and UAS:GLaz) was required to obtain some of the experimental combinations. Two independent lines were used.
Flies were anesthetized with CO2 and immobilized with adhesive tape. Fly eyes were photographed with a Nikon DS-L1 digital camera, in a Nikon SMZ1000 stereomicroscope equipped with a Plan Apo 1× WD70 objective. Local intensity maxima were obtained with the FIJI program, and nearest neighbor distances were calculated for each ommatidium. A regularity index (IREG) was estimated as described [47], and a percent recovery was calculated considering 0% the average degenerated eye and 100% the control wild type eye. Samples of 20–35 flies (3 days old) per condition and genotype were used to calculate the mean±SEM regularity index. We have used the freely available FIJI plug-in FLEYE for automated analysis of fly eye pictures [47].
Experimental procedures applied to mice were approved by the University of Valladolid Animal Care and Use Committee, and followed the regulations of the Care and the Use of Mammals in Research (European Commission Directive 86/609/CEE, Spanish Royal Decree ECC/566/2015).
Statistical analyses were performed with SPSS v.19 (IBM) and SigmaPlot v.11.0 (Systat) softwares. A p value < 0.05 was used as a threshold for significant changes. The tests used for each experiment are stated in figure legends.
|
10.1371/journal.pmed.1002854 | Impact of nutritional supplementation during pregnancy on antibody responses to diphtheria-tetanus-pertussis vaccination in infants: A randomised trial in The Gambia | Exposure to a nutritionally deficient environment during fetal life and early infancy may adversely alter the ontogeny of the immune system and affect an infant’s ability to mount an optimal immune response to vaccination. We examined the effects of maternal nutritional supplementation during pregnancy on infants’ antibody responses to the diphtheria-tetanus-pertussis (DTP) vaccine included in the Expanded Programme on Immunisation (EPI).
The Early Nutrition and Immune Development (ENID) trial was a randomised, partially blinded trial conducted between April 2010 and February 2015 in the rural West Kiang region of The Gambia, a resource-poor region affected by chronic undernutrition. Pregnant women (<20 weeks’ gestation) with a singleton pregnancy (n = 875) were randomised to receive one of four supplements: iron-folic acid (FeFol; standard of care), multiple micronutrient (MMN), protein-energy (PE), or PE + MMN daily from enrolment (mean [SD] 13.7 [3.3] weeks’ gestation) until delivery. Infants were administered the DTP vaccine at 8, 12, and 16 weeks of age according to the Gambian Government protocol. Results for the primary outcome of the trial (infant thymic size) were described previously; here, we report on a secondary outcome, infant antibody response to vaccination. The effects of supplementation on mean DTP antibody titres measured in blood samples collected from infants at 12 weeks (n = 710) and 24 weeks (n = 662) were analysed with adjustment for confounders including maternal age, compliance to supplement, and infant sex and season. At 12 weeks, following a single dose of the vaccine, compared with FeFol (mean 95% confidence interval [CI]; 0.11 IU/mL, 0.09–0.12), antenatal supplementation with MMN or MMN + PE resulted in 42.4% (95% CI 20.1–64.6; p < 0.001) and 29.4% (6.4–52.5; p = 0.012) higher mean anti-diphtheria titres, respectively. Mean anti-tetanus titres were higher by 9.0% (5.5–12.5), 7.8% (4.3–11.4), and 7.3% (4.0–10.7) in MMN, PE, and PE + MMN groups (all, p < 0.001), respectively, than in the FeFol group (0.55 IU/mL, 0.52–0.58). Mean anti-pertussis titres were not significantly different in the FeFol, MMN, and PE + MNN groups but were all higher than in the PE group (all, p < 0.001). At 24 weeks, following all three doses, no significant differences in mean anti-diphtheria titres were detected across the supplement groups. Mean anti-tetanus titres were 3.4% (0.19–6.5; p = 0.038) higher in the PE + MMN group than in the FeFol group (3.47 IU/mL, 3.29–3.66). Mean anti-pertussis titres were higher by 9.4% (3.3–15.5; p = 0.004) and 15.4% (9.6–21.2; p < 0.001) in PE and PE + MMN groups, compared with the FeFol group (74.9 IU/mL, 67.8–82.8). Limitations of the study included the lack of maternal antibody status (breast milk or plasma) or prevaccination antibody measurements in the infants.
According to our results from rural Gambia, maternal supplementation with MMN combined with PE during pregnancy enhanced antibody responses to the DTP vaccine in early infancy. Provision of nutritional supplements to pregnant women in food insecure settings may improve infant immune development and responses to EPI vaccines.
ISRCTN49285450.
| Deficiencies of both macro- and micronutrients are common among pregnant women in low- and middle-income countries (LMICs) and are recognised as a key determinant of poor birth outcomes.
Nutritional deficiencies during pregnancy may also impact on the longer-term health of the offspring, including immune development and function.
Our trial was designed to investigate the impact of combined protein-energy (PE) and multiple micronutrient (MMN) supplementation given to women across pregnancy on vaccine responses in young infants.
We performed a randomised, partially blinded trial of nutritional supplementation to pregnant women in rural West Africa, to test whether nutritional supplementation during pregnancy improved antibody response to vaccination in their infants.
We enrolled 875 pregnant women (mean gestational age at enrolment = 13.7 weeks), who were randomised to receive either iron-folic acid (FeFol; standard of care) alone or combined with MMN, PE, or PE + MMN daily from enrolment until delivery.
Infants were vaccinated at 8, 12, and 16 weeks of age with the diphtheria-tetanus-pertussis (DTP) vaccine, as recommended by the WHO Expanded Programme on Immunisation (EPI), and their antibody responses were measured at 12 (n = 710) and 24 (n = 662) weeks of age.
We observed that supplementation with MMN combined with PE was the most efficient intervention tested, improving mean antibody titres against diphtheria and tetanus at 12 weeks, and tetanus and pertussis titres at 24 weeks.
Improving the nutritional status of women in areas at high risk of undernutrition has multiple benefits beyond immediate birth outcomes.
Understanding the mechanisms that link nutritional supplementation during pregnancy to improved vaccine responses in early infancy may lead to the development of novel interventions among vulnerable populations.
| The Expanded Programme on Immunisation (EPI), introduced in 1974 by WHO, established a standardised vaccination schedule for all children globally, initially including diphtheria-tetanus-pertussis (DTP), Bacillus Calmette-Guérin (BCG), oral polio, and measles vaccines [1]. This programme has been implemented widely and is estimated to prevent between 2 and 3 million deaths in children annually. However, despite this, there remains a heavy burden of childhood deaths globally, largely from infection-related causes and especially during the first year of life, when the immune system is most vulnerable [2,3]. In tandem, undernutrition during fetal life and early childhood has been estimated to contribute to 45% of all deaths in children globally [4,5]. Through a vicious circle, undernutrition heightens the risk of infections, while infections predispose to undernutrition [5]. In settings such as sub-Saharan Africa, where supplementary nutrition and routine childhood vaccinations are lifesaving [6], novel interventions are required to break this cycle of infection and undernutrition.
The immune system starts to develop early during fetal life so that at birth all components are in place to allow a rapid expansion to complete maturation in infancy and childhood [7]. Nutritional insults during these critical periods may lead to a breakdown in the complex pathways required for its development, with downstream consequences on functionality, including the ability to mount an adequate response to vaccination [8]. A limited number of studies have assessed the impact of direct nutritional supplementation on vaccine responses in children, with findings indicating mixed effects of protein-energy (PE) and micronutrient supplements, including vitamin A, zinc, and iron, across a range of vaccines [9]. Far fewer studies have investigated the effects of maternal nutritional supplementation during pregnancy with a single micronutrient, multiple micronutrients (MMN), or PE on vaccine responses in infancy [10–14]. Overall, the limited data available provide some evidence that nutritional supplementation during fetal development may improve immune response to childhood vaccination [10]. However, the heterogeneity in study design, lack of adjustment for confounders, and limited availability of high-quality data prevents any firm conclusions and supports the urgent need for well-designed trials looking at the effects of antenatal supplementation on vaccine responses in infants.
We report here findings from the Early Nutrition and Immune Development (ENID) trial, a randomised trial conducted in rural Gambia, in an area of widespread and seasonal undernutrition. The ENID trial examined the effects of prenatal and infant nutritional supplementation on infant immune development [15]. The primary outcome of ENID was thymic development during infancy, and antibody response to vaccination was a secondary outcome. We have previously published data showing an effect of infant micronutrient supplementation, but not maternal supplementation, on infant thymic size [16]. Here, we present the effects of maternal supplementation on the secondary outcome, antibody responses to DTP vaccinations in early infancy.
The ENID trial (ISRCTN49285450) was a partially blinded trial of prenatal and infant nutritional supplementation conducted in the rural West Kiang region of The Gambia between August 2009 (the date of the first participant consented) and February 2015 (the date the last infant born into the trial reached 12 months of age). S1 Fig presents the ENID trial design; full details of the trial are provided in the published trial protocol [15]. The primary outcome measure of the ENID trial was thymic size in infancy (findings reported previously; [16]). Infant antibody response to vaccination (presented here) was a secondary outcome. Other outcomes included measures of cellular markers of immunity in a selected sub-cohort and infant growth to 24 months of age.
Briefly, all women aged 18 to 45 years and registered in the West Kiang Demographic Surveillance System were invited to participate in the study [17], and written informed consent was obtained. Monthly surveillance of all participating women, including a short questionnaire on the date of the last menstrual period, enabled the identification of women with a possible pregnancy, subsequently confirmed by ultrasound examination. All women confirmed to be <20 weeks pregnant were then randomised into the trial, with supplementation commencing the following week until delivery. Women with a gestational age (GA) ≥20 weeks, a multiple pregnancy, severe anaemia (haemoglobin [Hb] <7 g/dL), or confirmed as HIV positive were excluded.
Pregnant women were randomised to one of four intervention arms: (1) Iron-folic acid (FeFol) tablets, representing the usual standard of care as per Gambian Government guidelines; (2) MMN tablets, a combination of 15 micronutrients designed for use during pregnancy as formulated by UNICEF/WHO/UNU (with the exception of FeFol, each tablet contained 2×RDA of each micronutrient [18,19]; (3) protein-energy and iron-folic acid (PE + FeFol) as a lipid-based nutritional supplement (LNS) providing the same level of FeFol as the reference arm, but with the addition of energy, protein, and lipids; (4) protein-energy and multiple micronutrients (PE + MMN) as the same LNS supplement fortified to provide the same level of micronutrients as the MMN arm. The composition of the four supplements is detailed in Table 1. Antenatal supplements were distributed on a weekly basis by community-based field assistants.
From six months of age, infants were further randomised to receive either an unfortified LNS paste or the same formulation fortified with MMN. However, for the current analysis, the infant intervention arms will not be considered, as the outcomes included were assessed before the infant supplementation commenced.
Randomisation was performed in blocks of eight using an automated system reflecting the eight combinations of prenatal and infancy supplements. The antenatal arms of the trial were partly open because it was not possible to blind project staff or study participants to the supplement type (tablet versus LNS); all the investigators, however, were blinded to participant allocation.
Women were invited for a standard antenatal examination at the Medical Research Council (MRC) Keneba clinic at enrolment, and then again at 20 and 30 weeks’ gestation. At each of these visits, maternal anthropometry and ultrasound measures of fetal biometry were taken by a study midwife. Fetal size at the enrolment visit was used to estimate GA. All measurements were performed using standardised and validated equipment and standard operating procedures.
Following delivery, a study midwife visited all women and their newborns for a standard health examination, and infant anthropometric measurements were taken (weight, length, mid-upper arm circumference [MUAC], and head circumference). Infants were subsequently seen at the MRC Keneba field station at 1, 8, 12, 24, and 52 weeks of age, and at home at 16, 20, 32, and 40 weeks for sample collections, health assessments, and anthropometric measurements. The same standard and regularly validated anthropometric equipment was used at each visit (clinic and home visits). At these infant visits, EPI vaccines were also administered, according to the Gambian Government protocol [15]. All study vaccines were acquired from the EPI Department of the Gambian Government and issued by a study nurse, following standard procedures. Briefly, infants were vaccinated at birth (within 72 hours) and at 8, 12, 16, and 40 weeks of age. At birth, they received BCG vaccine, Hepatitis B vaccine (HBV), and oral polio vaccine (OPV). At 8, 12, and 16 weeks of age, they were vaccinated with Penta (diphtheria-Tetanus-pertussis [DTP], HBV, Haemophilus influenzae type B [Hib], OPV, and pneumococcal conjugate vaccine [PCV]), and at 40 weeks of age with OPV, measles, and yellow fever vaccines. For the current analysis, only DTP vaccine responses were assessed, using blood samples collected at 12 and 24 weeks of age. DTP responses were measured at 12 weeks, reflecting responses to the first dose of vaccine given at 8 weeks; and at 24 weeks, reflecting antibody responses after up to three doses (at 8, 12, and 16 weeks) of the vaccine. A weekly questionnaire was used to collect data on maternal morbidity (during pregnancy only) and infant morbidity (from birth to 52 weeks) and on infant feeding practices.
A validated multiple immunoassay based on Luminex xMAP technology was used to measure serum-specific IgG antibody responses directed against the three components of the DTP vaccine: pertussis toxin (Ptx), diphtheria toxoid (Dtxd), and tetanus toxin (Ttx) [20,21]. This assay was chosen as it presents the advantage of specifically measuring Ptx, Dtxd, and Ttx antibodies in a single assay using a small sample volume (5 μL) and small amounts of antigen compared with other ELISA methods. In-house reference standards were used, calibrated against international standards (see below).
Reconstituted freeze-dried Ptx (National Institute for Biological Standards and Control, United Kingdom), Dtxd, and Ttx (Sigma Aldrich, Gillingham, UK) were conjugated to activated carboxylated microspheres (Bio-plex COOH beads, Bio-rad, Watford, UK) using a two-step carbodiimide reaction. The in-house pertussis standard (calibrated against the United States reference pertussis anti-serum human lot 3) was diluted 4-fold in six dilution steps (1:200–1:204,800), whereas the in-house diphtheria-tetanus standard (calibrated against the International Standards NIBSC code Di-10 and NIBSC code TE-3) was diluted 4-fold in eight dilution steps (1:50–819,200). The unknown sera were diluted 1:200 and 1:4,000, whereas the detection antibody, R-Phycoerythrin conjugated goat anti-human IgG (ƴ chain specific) (Jackson ImmunoResearch Laboratories, Westgrove, PA) was diluted 1:200. Results were generated using a Bio-plex 200 system with Bio-plex Manager software (version 4.1.1, Bio-rad, UK). Median fluorescent intensity for Ptx was converted to ELISA Unit (EU)/mL, and for Dtxd and Ttx to International Unit (IU)/mL by interpolation from a five-parameter logistic standard curve. Being responder to the vaccine was defined as presenting an antibody titre >0.1 IU/mL for diphtheria and tetanus, according to international standards (WHO) [20,22]. As for pertussis, an in-house antibody assay was used; an arbitrary threshold was established at >5.0 EU/mL. All antibody assays were performed at the MRC Unit in The Gambia.
Maternal date of birth and age were ascertained from the West Kiang Demographic Surveillance System [17]. Maternal parity was calculated as the number of previous pregnancies, including live births and stillbirths, using data from a questionnaire conducted at enrolment. Enrolled women were asked whether they had attended Arabic and/or English school and for how many years. A binary variable (Yes/No) was generated based on whether women went for at least a year in an Arabic/and or English school, because school attendance was low, with 77.3% (549/710) of women not having attended school. Maternal body mass index (BMI) was computed as weight (kg)/height (m)2. Maternal morbidity was calculated as the total number of morbidity episodes during pregnancy divided by the number of weeks enrolled in the study. A compliance score based on the amount of supplement remaining in the jars (empty, half-empty, and full) was generated for LNS products (PE and PE + MMN), and a count of tablets left in the bottle was performed for tablets (MMN and FeFol). For each woman, a compliance percentage was generated by dividing the number of LNS jars or tablets the woman consumed by the number she received and multiplying by 100.
Preterm birth (PTB) was defined as a GA at birth <37 completed weeks and low birth weight (LBW) as birth weight <2,500 g. Exclusively breastfed (EBF) was generated as a binary variable (yes/no) and defined as continuation of exclusive breastfeeding until 12 (or 24) weeks. The infant’s anthropometric measurements at 12 and 24 weeks were converted to z-scores using the WHO Child Growth standards [23]. In the models presented here, weight-for-length-z-scores (WLZs) were used. The infant’s morbidity was generated as total days of reported sickness by the guardian in a weekly questionnaire. As month of vaccination may influence antibody vaccine responses [24], to capture the influence of the month of the first DTP vaccination (at 8 weeks) on antibody responses, the monthly variation was fitted using the first two pairs of the Fourier terms: sin(θ) and cos(θ) and sin(2θ) and cos(2θ) [25]. For the present analyses, the Fourier terms were fitted in pairs and denoted as F1 = sin(θ) and cos(θ) and F2 = sin(2θ) and cos(2θ), with θ representing the angle in radians of the date in relation to its position on the annual cycle (on 1 January, θ = 2π/365; on 31 December, θ = 2π). For the infant’s season of birth, a binary variable was generated as rainy = June to October and dry = November to May, to avoid collinearity with the Fourier terms for the month of vaccination.
Statistical analyses were performed using STATA 15·0 (StataCorp LP TX). The study was powered based on thymic index as the primary outcomes [15]. For a power of 80% and a significance level of 5%, we estimated a required total sample size of 847 mother–infant pairs. Statistical significance was set at a two-sided alpha level of p < 0.05. Analyses were performed by intention to treat (ITT), including all mother–infant pairs with available antibody data (N = 710 at 12 weeks, N = 662 at 24 weeks, and N = 511 for the analysis of antibody ratios between 12 and 24 weeks). The latter was generated by dividing antibody titres measured at 24 weeks by those measured at 12 weeks for each infant. Descriptive statistics of participants were calculated for each intervention arm. Chi-squared tests for categorical variables or ANOVAs for continuous variables were performed for significant differences in participant characteristics across maternal supplementation groups.
Normality was tested using the Shapiro–Wilk test. Plasma antibody concentrations were skewed; therefore, we used the logarithms (base 10) values of the antibody titres. Linear regression models with robust standard errors adjusted with confounding factors were used to determine the means of antibody titres. Although the use of the robust analysis minimised potential deviations from the assumptions of the multiple linear regression models, further checks were made to the models for linearity and multicollinearity and the residuals were examined to check homoscedasticity and normality. Log-transformed mean antibody titres were compared by Student t test and back-transformed from the log scale. Confounding factors were defined a priori based on previous literature and biological plausibility [26,27]. Maternal variables considered as potential effect mediators in the association between supplement status and antibody response to vaccination and included in the linear models were as follows: age and BMI (at baseline), maternal education, supplement group, compliance to supplement, Hb at 30 weeks’ gestation, and maternal morbidity. Infant variables included in the models were GA at delivery, birth season (dry/rainy), month of vaccination, sex, infant size (WLZ), and Hb level at the visit preceding antibody measurements. Mode of feeding and infant morbidity scores were also calculated from birth to the time of antibody assessment (12 or 24 weeks).
Sensitivity analyses were performed to check for any potential bias in the data between infants missing antibody measurements at 12 weeks of age (n = 90) or at 24 weeks of age (n = 138) compared with those included in the analyses (S1 Table; S2 Table). No significant differences were observed between the characteristics of infants with antibody measurements compared with those missing antibody measurements at either time point. The results of this study, which is part of the ENID trial, are reported in accordance with Consolidated Standards of Reporting Trials (CONSORT) 2010 guidelines.
The ENID trial was approved by the Joint Gambian Government/MRC Unit The Gambia ethics committee (SCC1126v2). Written informed consent was obtained from all participants. The trial observed Good Clinical Practice Standards and the current version of the Helsinki Declaration.
A total of 2,798 participants were recruited for monthly surveillance of pregnancy between January 2010 and June 2013, and 1,195 participants were assessed for eligibility (Fig 1). Of these, 875 (73.2%) participants confirmed pregnant with singleton infants and with GA <20 weeks were randomised to enter the antenatal supplementation phase of the trial. Of the 800 live births, 710 (88.8%) infants had DTP antibody measurements at 12 weeks and 662 (82.9%) at 24 weeks and were included in the ITT analyses. There were no differences in participant characteristics between those remaining in the trial and those lost to follow-up, either from baseline to delivery, or from live births to infants included in the current analysis of antibody responses (S1 Table; S2 Table).
Table 2 presents characteristics of the 710 mother-infant by supplement group. At enrolment, mean maternal age was 30.3 (SD 6.8) years, mean GA was 13.7 (3.4) weeks, and 18.9% of women were underweight (<18.5 kg/m2). Women were administered antenatal supplements for an average of 26.5 (3.4) weeks. One in five women had received a formal education (22.7%). Mean birth weight was 3.01 (0.40) kg, with 8.9% of the infants being of LBW (<2.50 kg). Mean GA at delivery was 40.2 (1.4) weeks with 5.1% PTB. At 12 weeks of age, the majority of infants (93%) were EBF, compared with half (52%) of infants at 24 weeks. There were no significant differences in maternal or infant characteristics across the supplement groups, except compliance to supplement and maternal Hb levels at 30 weeks’ gestation. Compared with those in the PE or PE + MMN arms (LNS-based supplements), women who received FeFol or MMN (tablets) had a higher compliance to supplementation (93% versus 81%) and a higher mean (SD) Hb level at 30 weeks’ gestation (11.1 [1.2] g/dL versus 10.4 [1.3] g/dL) (both, p < 0.001).
Following the first dose of the vaccine (12 weeks), 55.5% of infants presented protective antibody titres for diphtheria, increasing to 96.8% at 24 weeks (after three doses). For tetanus, protective levels were 97.3% after the first dose, increasing to 99.6% at 24 weeks. Finally, for pertussis, this rate increased between 50.1% and 88.2% between doses. For each antigen, no significant differences in the proportion of responders were detected across the supplement groups at 12 or 24 weeks (S3 Table).
In Fig 2 and Table 3, we present the adjusted mean concentrations of DTP antibodies by supplement groups, along with the effect sizes for the differences between groups. At 12 weeks of infant age, mean anti-diphtheria titres were significantly higher in infants born to mothers who received MMN (0.16 IU/mL, 95% confidence interval [CI] 0.14–0.19) or PE + MMN (0.14 IU/mL, 0.12–0.16) during pregnancy compared with FeFol (0.10 IU/mL, 0.09–0.12). This corresponded to a mean (95% CI) increase in anti-diphtheria concentrations of 42.4% (20.1–64.6) in the MMN group and of 29.4% (6.4–52.5) in the PE + MMN group, compared with FeFol. We also observed that mean (95% CI) anti-diphtheria titres were 33.4% (11.5–55.3) higher in the MMN group compared with the PE group. Compared with the FeFol group, mean (95% CI) anti-tetanus titres were found to be higher by 9.0% (5.5–12.5), 7.8% (4.3–11.4), and 7.3% (4.0–10.7) in the MMN, PE, and PE + MMN groups, respectively. Mean (95% CI) anti-pertussis titres were lower in the PE group (4.1 EU/mL, 3.8–4.4) compared with FeFol (by 13.5%, 8.8–18.2), MMN (by 10.7%, 6.0–15.5), and PE + MMN (by 14.8%, 10.3–19.3) groups.
We then compared the mean concentrations of DTP antibodies measured at 24 weeks of infant age between the supplement groups, with effect sizes for each comparison (Fig 2; Table 3). No significant differences in mean anti-diphtheria titres were observed across the groups. In contrast, mean anti-tetanus titres were higher in the PE + MMN group (3.75 IU/mL, 3.56–3.95) compared with FeFol (by 3.4%, 0.19–6.5), MMN (by 4.0%, 0.96–7.0), and PE (by 4.4%, 1.1–7.7) groups. For pertussis, mean antibody titres were higher in the PE (by 9.4%, 3.3–15.5) and PE + MMN groups (by 15.4%, 9.6–21.2), when compared with the FeFol group (74.9 IU/mL, 67.8–82.8). Additionally, mean anti-pertussis titres were higher in the PE group (93.0 EU/mL, 83.7–103.4) compared with the MMN group (by 8.7%, 2.4–14.9) and higher in the PE + MMN group (106.7 IU/mL, 97.4–117.0) compared with both the MMN (by 14.6%, 8.7–20.5) and PE (by 6.0%, 0.05–11.9) groups.
To examine the effects of the supplements on the changes in antibody concentrations between 12 and 24 weeks, we compared the ratios of the adjusted means at 24 and 12 weeks across the supplement groups and calculated the effect sizes for each comparison (Table 4, with unadjusted data presented in S6 Table). The increase in mean anti-diphtheria titres did not differ significantly between the supplement groups. Mean (95% CI) anti-tetanus titres increased more in the FeFol group (6.6-fold, 6.1–7.1) compared with the MMN (5.1-fold, 4.7–5.5), PE (5.2-fold, 4.9–5.6), and PE + MMN groups (5.6-fold, 5.2–6.1). The mean anti-pertussis titre increased by 24-fold (21.8 to 27.5) between 12 to 24 weeks in the PE group, which was a 20.4% (12.9–28), 27.7% (20.5–34.8), and 12.7% (5.9–19.6) higher increase compared with the FeFol, MMN, and PE + MMN groups, respectively. There was also a higher increase in mean anti-pertussis titres in the PE + MMN group compared with FeFol (by 7.7%, 0.45–14.9) and MMN (by 14.9%, 8.1–21.8) groups. Overall, we note that there were few differences between the unadjusted and adjusted models (S4 Table; S5 Table; S6 Table).
In a food insecure environment, in rural sub-Saharan Africa, infants born to women who received nutritional supplementation during pregnancy containing a combination of MMN and PE had measurably better responses to routine vaccines given in early infancy compared with the standard of care (daily FeFol). To our knowledge, this is the first randomised trial to examine the impact of a comprehensive package of nutritional supplements given during pregnancy on antibody responses to vaccinations within the first 6 months of life. This corroborates previous findings indicating a direct role of maternal nutritional status and supplementation on the infant’s immune development and function.
Although several studies have examined the effects of childhood nutritional status and/or direct supplementation on vaccine responses [9], studies investigating the impact of maternal supplementation during pregnancy on vaccine responses in infants are scarce and have provided mixed results [9]. A recent systematic review identified only nine relevant studies exploring this association, including three observational studies embedded within controlled trials [12,28,29]. A small randomised controlled trial (N = 39) conducted in Bangladesh showed that supplementation with zinc (20 mg/day) from the second trimester of pregnancy to 6 months postnatally was weakly associated with antibody responses to HBV postpartum and at 6 months of age (r = 0.386; p < 0.10) [12]. Conversely, another study in Bangladesh investigating the impact of zinc (30 mg/day) supplementation during pregnancy on immune response to BCG and Hib vaccines in 405 infants found no significant differences in Hib polysaccharide antibody responses between the zinc-supplemented and placebo groups [29]. A study conducted in Birmingham, UK, among 149 Asian infants born to women who participated in a trial of supplementation with PE (10,000–30,000 kcal/trimester) during pregnancy [14] found that at 22 months of age, infants born to supplemented mothers showed an enhanced response to BCG vaccine, assessed by scar formation [28].
The ENID trial was purposefully designed to test whether nutritional repletion of women during pregnancy improved the immune development of their infants. We observed that supplementation with PE combined with MMN was the most efficient intervention tested, improving mean antibody titres against diphtheria and tetanus at 12 weeks, and tetanus and pertussis titres at 24 weeks. This finding supports our original hypothesis that provision of MMN in the format of a large-quantity LNS would be the most efficacious intervention, providing both additional micronutrients and macronutrients to this group of nutritionally vulnerable women. Furthermore, this observation corroborates other data from the literature that show small-quantity LNS (e.g., those with a lower kcal content, but the same quantity of micronutrients) have a mixed impact on birth outcomes [30], whereas high-quantity LNS appear more efficacious [31]. The MMN supplement used in the ENID trial also included a twice-the-RDA micronutrient requirements for pregnant women, with the exception of FeFol, which was set at 60 mg and 400 μg per day, respectively, in line with Gambian Government recommendations. The use of a higher dose of MMN was guided by previous findings from a similar West African population, in which 2×RDA during pregnancy was needed to impact on birth weight [18]. Our findings demonstrate that the provision of PE and MMN to pregnant women confers the greatest benefits to their infants and supports previous observations of a positive impact of combined MMN and early food supplementation on child survival in rural Bangladesh, an area with high rates of maternal and child undernutrition [32].
Our study showed that antenatal supplementation with MMN, compared with the FeFol standard of care, improved responses to diphtheria and tetanus vaccines at 12 weeks but had no measurable benefit on the response to pertussis vaccine at either 12 or 24 weeks. FeFol supplementation is currently recommended by WHO during pregnancy because of its known associations with a reduced risk of adverse pregnancy outcomes, including anaemia, intrauterine growth restriction (IUGR), preeclampsia, PTB, and LBW [33]. However, most pregnant women in developing countries suffer from deficiencies in a range of micronutrients, caused by poor and variety-restricted diets. A recent meta-analysis of individual patient data from 17 randomised trials showed that maternal supplementation with MMN, compared with FeFol alone, improves overall birth outcomes and reduces mortality in female neonates, especially in undernourished and anaemic pregnant women [34]. Our data add to this available evidence and demonstrate that antenatal MMN supplementation promotes immune function in infants born in food insecure settings.
Antenatal supplementation with PE enhanced antibody response to tetanus at 12 weeks and pertussis response at 24 weeks, although no measurable effect was observed with diphtheria. Women getting the LNS (both PE and PE + MMN) received an additional 746 kcal daily. The rationale for the use of this large-quantity LNS was based on the previous findings that protein-energy supplementation during pregnancy significantly improves birth outcomes, including birth weight and perinatal mortality, in this setting [35]. The findings from the current study support the beneficial effect of PE supplementation in pregnancy on immune function. In contrast, however, a negative effect of supplementation with PE was observed for vaccine responses to pertussis at 12 weeks compared with the other groups. This observation is difficult to explain, as it is inconsistent with the other effects observed but may be a consequence of the combined detrimental impact of poor compliance to the LNS products (so in contrast to the FeFol and MMN arms) and lack of MMN (in contrast to the PE + MMN arms). We do not believe the PE group was ‘harmful’; rather, there was some benefit to infants born to women consuming the FeFol as tablets, and so in this case the referent group outperformed the PE group, in which compliance was not as high as anticipated.
We observed for each supplement differences in effect sizes according to the vaccine, with the highest effect sizes measured with diphtheria responses at 12 weeks, suggesting the importance of vaccine-specific factors in modulating the impact of maternal supplements on vaccine responses in infants. A variety of factors may influence antibody responses to vaccination in infants, such as the presence of maternally derived antibodies, parameters inherent to the vaccine itself (e.g., adjuvants, routes of administration, immunogenicity) or to the capacity of the infant to respond to vaccination (e.g., genetics), and other environmental factors (e.g., ongoing infections) [36]. We reported that after the first dose of DTP vaccine, >90% of infants presented protective levels against tetanus compared with approximately 50% of infants against diphtheria and tetanus. Maternal tetanus vaccination during pregnancy is widely implemented in Africa including in The Gambia, and so the presence of maternally derived tetanus antibodies may explain the higher proportion of infants with protective anti-tetanus titres at 12 weeks than with protective anti-pertussis or anti-diphtheria titres at 12 weeks [37]. This may also explain the lower effect size measured with the tetanus response compared with diphtheria and pertussis responses, because there may only be a limited capacity to improve infant responses beyond the benefit conferred by prevaccination levels [37].
Additionally, parameters inherent to the vaccine itself may modulate the effects of the maternal supplements, but as the same adjuvant and route of administration were used in the combined DTP vaccine, the differences in effect sizes observed are likely linked to variations in immunogenicity. Tetanus toxoid vaccine is known to be a stronger immunogen than Dtxd and whole-cell pertussis, especially among immunocompromised individuals such as preterm infants [38]. Therefore, the finding of this study suggests that infant responses to vaccines with poor immunogenicity—such as the recently implemented oral rotavirus vaccine—may be enhanced by improved maternal nutritional status.
We reported significantly higher proportions of infants with protective DTP antibody titres at 24 weeks of age following boosting doses of the vaccine compared with 12 weeks after the priming dose of the vaccine. This increase in antibody titres is expected and related to the prime-boost strategy, which is commonly used to increase the efficacy and immunogenicity of vaccines and augment the numbers of responders after an initial immunisation, and a decline in specific memory B and T lymphocytes. We also observed generally greater effect sizes in relation to the first vaccination and a higher increase in antibody titres among infants, with the lowest responses after the initial immunisation. This suggests that antenatal supplementation may have a greater influence on the responses to the priming dose of a vaccine but few effects on the succeeding doses, which have been designed to allow all the infants, including those with low responses, to recover and reach protective levels.
The main strength of our study is the use of a randomised trial design using a comprehensive nutritional supplementation strategy with robust assessment of antibody responses to DTP, a vaccine which is universally administered in childhood as part of the EPI, and the use of a large representative sample population. DTP antibody titres were measured at 12 and 24 weeks of age, providing an assessment of supplement effects on capacity to mount immune responses after a priming dose and after two boosting doses of the vaccine. Detailed assessment of possible maternal, infant, and environmental effect modifiers enabled a robust evaluation of supplement-mediated effects.
Because the maternal transfer of antibodies through the placenta or breast milk may alter immune responses to vaccination in infancy [39], limitations of this study include the lack of any measures of breast milk antibody levels or prevaccination antibody levels in infants. Furthermore, we recognise that the measured antibody titres in infants may, to a certain extent, be a reflection of maternal antibody concentrations in response to their own vaccination prior to or during pregnancy and, for tetanus at least, the lack of any reliable data on maternal vaccination status reflects a further limitation of the current study. We also acknowledge that antibody measurements may not be an adequate indicator of longer-term immune protection; future studies should therefore incorporate the longitudinal surveillance of morbidity among vaccinated children.
We also note that compliance in the PE and PE + MMN arms of the trial (both given in the form of LNS) was significantly lower than in the FeFol and MMN arms (both given as tablets), and that this was matched by a significantly greater drop in maternal Hb between enrolment and 30 weeks’ gestation in the LNS groups (noting that the daily dose of iron was standard at 60 mg across arms). We have speculated elsewhere that this poorer compliance to this high-quantity LNS may reflect a lower acceptability of these products in this population, rather than a consequence of sharing of the supplement within the community [16]. However, we note that this lower compliance and difference in Hb levels between study arms does not explain the results observed. Finally, the findings from this trial are limited to populations with widespread nutritional deficiencies during pregnancy and with high rates of infection in infants and young children.
In conclusion, we have shown that giving nutritional supplements containing a combination of micronutrients and macronutrients to nutritionally vulnerable pregnant women in rural sub-Saharan Africa improves antibody response to vaccination in early infancy. The effects observed were most marked in response to the priming dose of each component of the DTP vaccine, with less of a marked effect after the full schedule was received. Although the observed effect sizes were modest, even small improvements in antibody response may help an infant to pass the thresholds of unprotective to protective antibody titres. Furthermore, because antibody titres naturally decrease with time following an initial peak after vaccination, higher initial antibody titres may also confer longer protection and improve vaccine effectiveness. Thus, our findings have potential clinical importance indicating that nutritional repletion of pregnant women may help to compensate for immunological vulnerability during the critical period of early infancy. The possibility that the benefits would be greater for vaccines that have lower efficacy in African populations is worthy of further investigation.
|
10.1371/journal.pcbi.1002787 | Weakly Circadian Cells Improve Resynchrony | The mammalian suprachiasmatic nuclei (SCN) contain thousands of neurons capable of generating near 24-h rhythms. When isolated from their network, SCN neurons exhibit a range of oscillatory phenotypes: sustained or damping oscillations, or arrhythmic patterns. The implications of this variability are unknown. Experimentally, we found that cells within SCN explants recover from pharmacologically-induced desynchrony by re-establishing rhythmicity and synchrony in waves, independent of their intrinsic circadian period We therefore hypothesized that a cell's location within the network may also critically determine its resynchronization. To test this, we employed a deterministic, mechanistic model of circadian oscillators where we could independently control cell-intrinsic and network-connectivity parameters. We found that small changes in key parameters produced the full range of oscillatory phenotypes seen in biological cells, including similar distributions of period, amplitude and ability to cycle. The model also predicted that weaker oscillators could adjust their phase more readily than stronger oscillators. Using these model cells we explored potential biological consequences of their number and placement within the network. We found that the population synchronized to a higher degree when weak oscillators were at highly connected nodes within the network. A mathematically independent phase-amplitude model reproduced these findings. Thus, small differences in cell-intrinsic parameters contribute to large changes in the oscillatory ability of a cell, but the location of weak oscillators within the network also critically shapes the degree of synchronization for the population.
| Circadian rhythms are daily, near 24-h oscillations in biological processes that nearly all organisms on Earth experience. Single cells contain a molecular clock that drives circadian rhythms in physiology and, when many cells synchronize in a population, daily behaviors. We hypothesized that small differences in intrinsic cellular properties allow for a diversity of circadian periods and amplitudes across cells. We observed circadian cells and their synchrony before, during, and after limiting communication between cells and then compared their intrinsic properties to their resynchronization behavior. We found that arrhythmic, weakly oscillating, and self-sustained circadian cells rejoined the rhythmic population independent of their cell-intrinsic oscillations. Using a mechanistic computational model of circadian cells, we found that resynchronization could be enhanced by including more weak oscillators or by placing weak oscillators at more connected nodes in the network. We conclude that intrinsic properties (e.g. oscillator weakness and responsiveness) and network structure (e.g. positions of weak oscillators) can independently buffer tissue rhythms from perturbations. This reveals how cellular and network properties impose rules on systems of circadian cells that must achieve synchrony from a desynchronized state, for example during perinatal development or when forced to overcome societal constraints on sleep-wake behavior, such as working early or late shifts.
| Circadian clocks generate the near 24-h oscillations that orchestrate daily behaviors in organisms throughout the kingdoms of life [1]. In mammals, the suprachiasmatic nucleus (SCN), a bilateral structure of 20,000 neurons in the ventral hypothalamus, functions as the master pacemaker with circadian cells driving rhythms in behavior and physiological processes, such as sleep-wake, locomotor activity, temperature, and hormone release [2]. It was hypothesized that every SCN neuron acts an autonomous clock, using molecular feedback loops to generate daily rhythms in gene expression and cellular output in the absence of external signals [3], [4], [5], [6], [7]. For example, the Period2 (Per2) gene, a clock gene found in humans and other animals, shows daily rhythms in transcription that appear to depend on daily repression by complexes including its protein (PER2) [8], [9]. Recent data, however, highlight that, when isolated, SCN neurons exhibit a range of behaviors including damped or unstable circadian oscillations [10], [11]. Therefore, although all cells may be capable of autonomous rhythmicity, they require stabilization from the SCN network to function as robust circadian oscillators. The potential source or sources of this cell-intrinsic variability, as well as its potential impact, are unknown. Whether the intrinsic properties of SCN oscillators independent of, or interactions amongst groups of oscillators within, the SCN network, or both, are responsible for the overall behavior is a current area of research [12], [13].
To first test the hypothesis that intrinsic differences between cells may affect how they resynchronize to each other, we followed daily oscillations of PERIOD2 protein levels in single SCN cells before, during, and after pharmacological blockade of intercellular signaling. The results revealed individual cells that differed in their intrinsic amplitude, level of gene expression, circadian period and ability to sustain rhythmicity, none of which predicted the cells' behaviors as they resynchronized to the population. Instead, we found that oscillations resumed and cells joined the rhythmic population at specific circadian phases, ultimately revealing the previously described daily waves of gene expression across the SCN [6], [14], [15]. Recent work has further suggested that the phase relationships of SCN cells across the network could be important for robust rhythmic behavior of the tissue [13], [16], [17].
To understand the complex behaviors of SCN cells, many studies have employed computational models. Both deterministic models detailing the molecular processes driving oscillations in single cells and stochastic models investigating the effects of noise on the system have aided in the understanding of mechanisms generating circadian rhythmicity in mammals [18], [19], [20]. Multi-cellular network models have been constructed from these single oscillators to describe synchrony across the SCN tissue, entrainment to light-dark cycles, and phase shifting behavior [21], [22], [23], [24]. Network models have also probed regional differences in the SCN [25], [26] and the phenomenon of splitting, in which synchronized regions in the SCN can oscillate with the same period but opposite phases [27]. We were interested in the relationship between cell-intrinsic rhythmicity and tissue synchronization, and found two major implications in the literature. First is that “smaller is better”: damped oscillators [23], [24], [28] and oscillators with short relaxation times [29] synchronize efficiently. Additionally, a recent study of fibroblast cells shows that cellular oscillators have small, but sustained amplitudes, and that their proximity to a bifurcation allows them greater control over their period [30]. The authors note that this could be advantageous for peripheral oscillators that need to be entrained by the pacemaker and suggest that similar properties in pacemaker cells could aid synchrony. Second, network topology also affects the quality of synchrony, and specifically, small-world type network topologies are beneficial for synchrony [22], [26]. It has not been shown, however, why small oscillations are good for synchrony or how cell-intrinsic behaviors and network topology together affect synchrony. Using a mathematical model provides us the flexibility to explain biological phenomena without constraints found in the physiology, e.g. the type, number, and location of oscillators within a network. We sought to address this by first assessing the roles of intracellular processes on intrinsic properties, such as rhythmic ability and phase-responsiveness. Next we assessed the effects of individual cell properties on network synchronization, and finally, how the location of key cells within the network affects synchrony.
We hypothesized that intracellular properties and intercellular interactions contribute to the resynchronization behaviors we observed in the tissue data. To test this prediction, we used a computational model to simulate clock gene transcription-translation feedback loops in single cells and found that small changes in parameter combinations produce the range of intrinsic oscillations observed in SCN cells. When placed in a network, these cells were able to synchronize, meaning that they were capable of adjusting their phases to align with the population. To understand this phenomenon, we computed velocity response curves (VRCs) for these cells [31], [32]. VRCs predict the phase velocity, i.e. how fast phase changes in response to intercellular signals. For our model, the VRCs suggested that cells with weaker oscillations could adjust their phase velocity more readily than cells with strong oscillations. These results were consistent with previous results that “smaller” is better to initiate synchrony, but with an alternative definition of smaller – we studied the effects of rhythmic, but low-amplitude (weak) cells, rather than initially rhythmic cells that lose amplitude, and eventually, all rhythmic ability, over the long-term (damped). We therefore tested the prediction that inclusion of weak circadian cells, which are highly responsive when isolated, would improve a network's ability to synchronize. We hypothesized that as weak cells establish rhythmicity and synchrony in the network, they lose responsiveness, becoming strong oscillators when coupled. By using a model of 400 coupled, heterogeneously oscillating cells, we found that increasing the proportion of weak oscillators or placing weak oscillators at more connected nodes in the network allowed for improved resynchronization.
Recent reports have shown that when SCN explants are treated with tetrodotoxin (TTX), a blocker of voltage-gated Na+ channels, the circadian rhythms of single cells gradually drift out of phase from one another [6], [10], [33], [34]. To understand the relative contributions of cell-intrinsic and network properties to these synchronization dynamics, we examined the bioluminescence recorded from single cells (n = 123 across two nuclei, slice 1; n = 90 within one nucleus, slice 2; for details see Text S1) in SCN explants from homozygous PERIOD2::LUCIFERASE (PER2::LUC) knock-in mice [35] during and after TTX treatment (Fig. 1). Although all cells appeared to gradually drift out of phase, only some expressed sustained circadian rhythms while others slowly or rapidly lost rhythmicity until the TTX was removed, at which point they began to regain rhythmicity and, eventually, synchrony. Looking at the timing of recovery of oscillations in slice 1, we found that approximately one-third of the cells that regained rhythms showed significant circadian oscillations within the first 35 h after removal of TTX. During the next 10 h another group of cells, similar in number, became circadian and began to synchronize to the first group. The remaining cells showed significant circadian rhythms starting around 45 h after removal of TTX, with the final cells entering by 96 h. Interestingly, the initial cohort of cells regained rhythmicity closely in phase while later cells regained rhythmicity with more broadly dispersed phases (Figs. 1B; Text S1; Rayleigh tests performed at the entrance time of the last cell in each cohort; Cohort 1, n = 39 cells, r = 0.68; Cohort 2, n = 43 cells, r = 0.55; Cohort 3, n = 32 cells, r = 0.43). In the second explant, we found a similar gradual restoration of rhythmicity to individual cells after TTX was removed (slice 2; Text S1). There was also a spatial pattern in each nucleus of the slices: lateral cells regain rhythmicity earlier than or phase lead medial cells (see Table S2 and Figure S10 in Text S1). In addition, lateral cells are, on average, smaller in amplitude than medial cells. This suggested a spatial organization of amplitude in the network during synchrony recovery. This led us to ask if there was something intrinsically different about the oscillations in cells that became rhythmic earlier or later after TTX was removed.
We reported previously that SCN cells uncoupled by TTX display diverse circadian behaviors both in terms of amplitude and period [10]. We acknowledge the possibility that TTX treatment itself can alter a cell's amplitude; however, we will assume that amplitude during TTX is reflective of amplitude that is independent of other feedback from other cells, and as such, is intrinsic to a cell. To determine whether or not intrinsic behaviors explain early or late restoration of rhythms, we compared amplitudes and periods of cells in TTX-treated SCN explants to the time when their rhythms reemerged and to the quality of synchrony within the group of circadian cells. In both slices, we found no significant correlations (R2 values of <0.2, Text S1) between intrinsic circadian properties, such as mean bioluminescence, total bioluminescence, bioluminescence amplitude and period, and when a cell joined in oscillations within a resynchronizing SCN network. We conclude that intrinsic properties alone fail to explain the dynamic emergence of rhythms and resynchrony of individual cells. Therefore network properties likely participate along with these intrinsic behaviors in synchrony. To explore the relationship, if any, between the intrinsic properties of the cells within the context of the network, we implemented a mathematical model.
First, we sought to reproduce the diversity of characteristics of isolated cells (i.e. PER-driven bioluminescence with patterns that could be described as strongly rhythmic, weakly rhythmic or arrhythmic over multiple days) by identifying potential molecular determinants of these circadian phenotypes. We utilized an existing model of the mammalian molecular clock to simulate SCN neurons [18] and focused on four parameters that regulate the output we recorded in the biological data (PER2::LUC): the rate of transcription of the Period (Per) gene, or translation, phosphorylation or degradation of the PERIOD (PER) protein. We categorized each cell as arrhythmic, weak (rhythmic but low in amplitude), or strong (rhythmic and high in amplitude; see Materials and Methods). We found that changing any of the four parameters by at least 10% moved simulated cells from arrhythmic to weakly rhythmic to sustained circadian gene expression (Fig. S2A). Regardless of whether they were varied alone or in combination, these parameters recapitulated the phenotypes found in SCN explants (Figs. S1, S2, S3).
We used a multi-dimensional visualization technique to evaluate the relative contributions of the four parameters to rhythm generation [36], providing a novel analysis of sensitivity of strength and sustainability of circadian oscillations to specific parameter combinations. By nesting parameter combinations into stacks, we arranged our data set with a large number of parameters in two dimensions that could be displayed easily (Materials and Methods). Based on the position of cells across the parameter space visualization, we found that small changes in rates of transcription of Per mRNA and degradation of PER protein produced larger effects than changes in translation and phosphorylation of PER on the circadian phenotype of simulated cells (Figs. S1 and S2B). Per rhythmicity was similarly more sensitive to Bmal1 transcription and BMAL1 degradation than BMAL1 translation and phosphorylation (Fig. S2B).
We ensured that individual model cells accurately represented individual cells from the slice. Amplitude was of particular importance because here we tested the effect of weak oscillators for the first time. When we compared the circadian periods and amplitudes of simulated and recorded cells we found no correlation between period and amplitude for either the model or the slices (Fig. S4; R2<0.02 for all). Further, the periods were similarly distributed (slice 1 std. dev. = 2.1 h, slice 2 std. dev. = 2.1 h, model std. dev. = 2.1 h) and the amplitude distributions were dominated by small values in both the model and the slices (Fig. S4, Text S1). This suggests that the period and amplitude values in model cells faithfully mimic behaviors we observe in the slice during TTX treatment. Neither this independence of period and amplitude, nor the dominance of small amplitudes has been described in other computational models. Here we are explicit in our modeling that the intrinsic amplitude is much smaller than the in-network amplitude. We concluded that by specifying small differences in key circadian parameters between cells, our simulated cells accurately represented the diverse rhythmic abilities, as well as realistic intrinsic properties such as period and amplitude, of SCN cells.
Another relevant property of a circadian oscillator is how it will adjust its phase velocity (speed) following a perturbation. We calculated the velocity response properties of the simulated cell set, including both strong and weak cells. Fig. 2C shows representative velocity response curves (VRCs) to a signal, where curves are plotted as a function of phase of oscillation. From the curve, we see that if the signal arrives early in the day (around circadian time, CT, 0) the cell will speed up, and if it arrives late in the day (between CT6 and CT12), the cell will slow down. To measure the cell's ability to shift, we computed the area under the absolute value of the VRC. We compared this VRC area to intrinsic oscillator amplitude (the sum of the peak to trough amplitude of all states) and found it inversely correlated with velocity response (Fig. 2D; R2 = 0.85). This indicates that oscillators with small intrinsic amplitude are more likely to have larger velocity response and therefore greater phase shifting ability compared to high-amplitude cells. Interestingly, we found that small oscillators in both the simulation and slice 1 have a broader distribution of periods compared to strong cells. The VRC results suggest a functional strategy to overcome this period variability: weaker cells are better at shifting their phase.
To test empirically if and how the proportion of weak oscillators contribute to the synchronization properties of a network like the SCN, we modeled a network of 400 SCN cells with diverse oscillatory abilities, including different periods and amplitudes, as well as network connections. Specifically, each cell had a unique set of parameters selected randomly to establish a population with defined proportions of arrhythmic, weak and sustained oscillators. We chose to include both local and global coupling between cells based on recent theoretical work [22]. Each cell was connected to its four nearest neighbors and 20% of cells connected to cells beyond their immediate neighbors (Fig. S5). Coupling was achieved in the model by simulating release of vasoactive intestinal polypeptide (VIP), a known synchronizer in the SCN [37], from all cells. Each network (n = 56 independent runs for each condition) was populated with 400 characterized cells and its overall response to uncoupling and recoupling was measured by calculating the synchronization index (SI) of all cells (see Materials and Methods). On average, we found that networks with more weak oscillators (total oscillator amplitude < = 8.4 a.u.) reliably reached higher levels of synchrony (SI> = 0.7 at days 15–18) with approximately 5-fold higher synchrony in networks comprised of mostly weak, compared to mostly strong, oscillators (Fig. 3A–B; ANOVA between populations, p<0.001). We found that networks with only strong oscillators failed to resynchronize (Fig. 3A–B; SI = 0.2 at days 15–18). We conclude that weak, highly shiftable cells can promote synchrony.
To test the importance of location within the network, we evaluated synchrony in networks of 50% weak and 50% sustained oscillators in which weak oscillators were assigned to hubs, i.e. nodes with more than the average of 10 outputs (range = 4–39 outputs). We found that when weak cells were placed in the more connected nodes of the network (n = 56 independent network runs for each condition), the population reached approximately 5-fold greater synchrony compared to networks with strong cells at these nodes or networks with oscillators distributed to nodes randomly (Fig. 3C–D ANOVA between populations, p<0.001). The quality of resynchrony, therefore, depended on both the number and placement of weak, shiftable oscillators in the network.
To test whether our findings regarding weak oscillators extend to damped oscillators, we repeated the simulations using cells that lose amplitude, and eventually, all rhythmic ability (see Materials and Methods). We found that the effects of weak circadian cells and of damped cells were nearly identical. For example, increasing the percentage of damped cells (Fig. S6A) or placing damped cells at network hubs (Fig. S6B) enhanced synchrony. Further, we verified that our results were not sensitive to our definition of weak. For the simulations used to generate Fig. 3 and Fig. S6, the weak cells were the smallest 30% in intrinsic amplitude. We repeated all simulations varying the percentage of rhythmic cells classified as weak. For each of these cut-offs, we measured the largest difference in SI between weak and damped cells (range = +0.07–0.69). The closer the weakly circadian cells were to the bifurcation, the more they acted like damped cells. We observed that as long as the cut-off is less than 50%, weak cells are similar to damped cells. To demonstrate that these behaviors could be generalized to other oscillatory systems, we constructed a phase-amplitude model, which functions as a reduced version of our mechanistic model (Text S2). We wanted to know if the benefit of weak cells for synchrony was evident in simpler systems and if a reduced model could further our understanding. The reduced model also showed that inclusion of more low amplitude, or small, oscillators or strategically placing them at more highly connected nodes increased synchrony. Thus, these results were robust across model compositions and types, and indicate that larger phase adjustments by small oscillators will, in general, produce higher synchrony.
Although physiologists and anatomists have described differences between SCN cells including their circadian amplitude, phase and waveform [38], [39], [40], the functional role of oscillator heterogeneity has been little studied. For example, the intrinsic daily oscillations of SCN neurons can be sustained, damped, or absent [10], [11], [41], but the consequences of these diverse circadian phenotypes remain unknown. Here, we found that the resynchronization of SCN cells following pharmacological blockade of cell-cell signaling involves waves of cells becoming rhythmic and adjusting their phases to join the daily cycling of the population. Previous theoretical studies have suggested that damped cells can aid network synchrony by entraining to a wider range of periods [23], [24] and relaxation oscillators can entrain faster if they have shorter relaxation rates or more spike-like waveforms [29], but have also highlighted that it is not yet possible to distinguish whether SCN cells should be modeled as damped oscillators or low amplitude, sustained oscillators [28]. What are the potential sources of these differences? By tuning a computational model, we found that small changes in a small set of parameters could produce a realistic distribution of cells that varied not only in period length, as has been generated previously [18], [21], but in qualities of the oscillations themselves. Using non-biased minimization techniques to represent multi-dimensional parameter space, we found that parameters associated with transcription rate and protein degradation of the Period gene were more likely to contribute to circadian changes than other parameters. We speculate that genetic differences in and the environmental modulation of these key rate constants between SCN cells could underlie the heterogeneity in their circadian properties. For example, it has been shown that the amplitude of Per transcription is altered in the absence of VIP [42] and that the stability of PER protein against degradation affects circadian period [43].
Because we found that the amplitude of our model cells is reliably and inversely related to their ability to adjust their phase velocities in response to natural signals, we tested the impact of both low (weak) and high (strong) amplitude oscillators on network synchrony. Previously, Bernard and colleagues suggested that a network comprised of damped circadian oscillators is capable of synchronizing and maintaining rhythmicity, and hypothesized that damped oscillators, when synchronized, induced rhythmicity in the population [24]. Locke and colleagues then performed a parameter optimization, searching to maximize the ability of a network of oscillators to synchronize. The best-synchronized networks were composed of damped cells [44]. Together, these results suggested that the driving force coupling cells together could arise from some inherent property found in damped cells.
In a similar fashion, we sought to identify inherent characteristics in both biological and modeled weak oscillators, including relationships between intrinsic amplitude and intrinsic shiftability. Published models using damped oscillators have been unable to mathematically quantify shiftability. By studying weak circadian oscillators, we measured larger changes in oscillator speed for smaller amplitude cells. Measurements of shiftability now provide a tool to study for the first time the kinetics of resynchronization. We posit that weakly oscillatory cells can send signals to other weakly oscillatory cells to readily adjust their phases. As the system synchronizes, the cells gain amplitude and thus lose the ability to make dramatic shifts. This suggests a strategy for neurons to resynchronize. The system can move from being sensitive to perturbations to being robust against them through a process of cell-cell amplification of rhythm amplitudes [33].
In our model networks we demonstrated that the total number of a specific oscillator type is critical and that there is an effect of the degree of connectivity of certain oscillator types on synchrony, such that, networks with more and more highly connected weak oscillators have improved synchrony during the recovery period following a perturbation. We concluded that heterogeneity arises from both cell intrinsic and network contributions, including the network topology and number of weakly circadian cells. The model does not account for all dynamics of resynchrony that we observed in the data, which will be addressed in the future. For example, though we observed populations of cells consistently ahead of or behind the mean phase of the network simulations, we observed no spatial pattern in these phase differences; the more homogeneous connections in the model networks led to most cells becoming rhythmic at the same time and together tighten in phase. Future work will use modeling to understand if and how spatial heterogeneity in network connections causes spatial patterns in the phase of oscillators across the slice. Future work will also take into account stochasticity in cell behavior. Preliminary results (data not shown) indicate that incorporating white noise into tissue simulations has no effect on the role of weak oscillators.
How the evolving differences within oscillators and amongst oscillator populations carry over to the behavior of networks is an open question for investigation. We return to the issue of whether rhythmicity and synchrony are due to intrinsic cell properties or are dependent on cell location and network structure. Recent studies have examined phase heterogeneity within the SCN [12], [13] and have concluded it is not a function of cellular properties. Foley and colleagues summarized their findings as “the tissue is the issue” – that placement within the SCN network (based on assigned phase) dictates whether and how an SCN neuron will oscillate [13]. We extend this to hypothesize specifically that cells, which are intrinsically different in their ability to maintain strong or weak rhythms, will impact the population rhythm differentially (e.g. the quality of synchronization increases with more weak oscillators), but also depending on their location within the network (e.g. cells at hubs have greater influence). Other theoretical studies have emphasized that the number of connections between cells could modulate the degree of synchrony in the network and argued for region-specific placement of particular oscillator types (e.g. sustained cells in the dorsal SCN and arrhythmic or gated cells in the ventral SCN) [25], [45], [46]. We find no evidence for specialized, localized populations of oscillators in the resynchronizing SCN slice following the removal of TTX. In contrast, our model shows the importance of weakly rhythmic, highly responsive oscillators at hubs where they can send coordinated phase information broadly throughout the network, becoming less responsive as they increase in amplitude, and that this is critical for improved synchrony. It is thought that SCN neurons establish rhythmicity and synchrony amongst each other and with the external light-dark environment late in gestation [47], [48]. Because these features are likely critical for the survival [49], we posit that the composition of the SCN, including a continuum of oscillator behaviors and connections, allows the tissue to adjust to shifts in environmental timing cues. These properties may be universal to all networks that include weak oscillators.
Single cells measured in SCN slices reported in this study were recorded as previously published [10]. Briefly, SCN explants from neonatal PER2::LUC mice were cultured for 3 days on MilliCell-CM (Millipore) membrane pieces in CO2-buffered medium supplemented with 10% newborn calf serum (Invitrogen) before being inverted onto polylysine/laminin coated coverslip dishes. All procedures were approved by the Washington University Animal Studies Committee and complied with NIH guidelines.
We conducted recordings in air-buffered medium containing 0.1 mM beetle luciferin (BioThema) at 37°C beginning at day 2 after slice transfer to coverslip dishes We temporally (1 h integration time) and spatially (2×2 pixel resolution) bioluminescence counts using a Versarray 1024 CCD camera (Princeton Instruments).
Following 6 days of baseline recording, we treated organotypic SCN explants with 0.5 µM tetrodotoxin (TTX, Sigma) as previously described [10]. TTX remained in the medium for 6 days before the medium was removed and we washed explants with 1 full volume exchange of fresh medium. Recording then continued for at least 6 days to examine rhythms as cells resynchronized after the restoration of cell-cell communication.
We used NIH ImageJ software to process all images by first subtracting background levels and then measuring pixel intensity over time in a region of interest above each cell. Cells were tracked manually from frame-to-frame and across treatments to account for any tissue movement. Cells were initially scored as rhythmic or arrhythmic if their gene expression rhythm oscillated with a period between 15 and 35 hours that was statistically significant by both Chi-squared periodogram [50] and FFT-NLLS [51]. We also used Wavos to determine period and phase information from the single cell traces [52].
A version of a previously published 16-ordinary differential equation model of the mammalian circadian clock was used to simulate rhythms in single model cells [18]. We altered parameters for rates of transcription, translation, phosphorylation, and degradation of either Period or Bmal1 genes, leaving 50 other parameters set to published basal values [18], and measured rhythms in gene output. We simulated 720 hours of gene expression from each cell, using initial conditions from a representative, high-amplitude sustained cell.
To measure the sensitivity of circadian cycling to clock gene parameters, we organized results from single cell simulations using clutter based dimensional reordering (CBDR), which applies minimization and dimensional stacking algorithms described below. These methods allow visualization of the underlying structure of clock parameter space and gauge the influence of tested parameters relative to output behavior. We utilized published Matlab code [36] to minimize differences between output scores (strong, weak, arrhythmic) and cluster behaviors together. The code arranged parameter combinations iteratively until the minimization requirement, i.e. cells with like behavior, were clustered together, was fulfilled. First, the code scans one pair of parameters over a range of values while the remaining parameters are set to basal values. Then we label this grid based on the output for each combination and add it to a larger montage of other parameter pairs. For a useful visualization, the code orders these dimensional stacks to group similar outputs together. Given a unique behavior and parameter combination for each cell, we minimize the stack so that differences between regions of varying outputs are small (in this case, strong, weak, or arrhythmic patterns in gene expression), and this provides an order ranking of “higher” versus “lower” parameters in the stack. Changes in parameter value that produce larger effects in output phenotype are higher in the stack order.
A velocity response curve (VRC) predicts the effect of parametric perturbation on the phase velocity of the oscillator. For a cell in the SCN, there is a single parameter (vsP) that is manipulated by VIP signaling. Hence, we consider the VRC associated with vsP, mapping the circadian time of VIP signaling to its effect on the phase velocity. Cells with higher-magnitude VRCs can be sped up or slowed down more by VIP signals than cells with lower magnitude VRCs. To quantify the “shiftability” of a cell, we compute the area under the absolute value of the VRC.
A VRC may be computed for any cell with a parameter set allowing for limit cycle oscillations. For details regarding computation, see [31].
Mathematically, the individual cells we have modeled can be categorized as rhythmic (those that converge to a periodic orbit) or arrhythmic (those that converge to a steady-state solution). However, simulations of single cells display a spectrum of behaviors, with some showing lower or higher amplitude than others. Using the peak to trough amplitudes of all model components (i.e. by summing the amplitudes of all states), we separated the rhythmic cells into two categories: 1.Weak cells are rhythmic with small amplitudes and 2. Strong cells are rhythmic with larger amplitudes.
For some simulations, we needed damped oscillators, which form a subset of the arrhythmic oscillators. We used the total amplitude at the end of a 15-day simulation (starting from high-amplitude initial conditions) to create the additional categories: 1. Flat cells are arrhythmic and have the smallest amplitudes in their final pseudo-cycle and 2. Damped cells are arrhythmic and have the largest amplitudes in their final pseudo-cycle.
For most simulations in the paper, we defined the smallest 30% (n = 228) of the oscillatory cells as weak and the remaining 70% (n = 595) as strong. For simulations that needed to distinguish between flat and damped, we chose the cut-off so there would be the same number of damped cells and weak cells.
Model cells were coupled together by VIP signaling, simulated as a drive on the rate of Per transcription, as previously published [21]. In our model, 20 percent of the 400 neurons were capable of sending a VIP signal and all neurons could respond to VIP. Connections between cells were organized with a small world network topology as in [22] where each VIP cell was coupled to its four nearest neighbors and then had a probability of sending unidirectional long-range connections to other cells in the network. We set the connection probability to p = 0.05, resulting in a synchronized system with a range of 4 to ∼40 outgoing connections in most networks. To mimic the TTX experiments, we simulated 6 days with VIP-mediated coupling followed by 6 days with coupling eliminated and then reinstated for 6 days. We assessed the intrinsic circadian expression of each cell as well as the rate of resynchronization of each cell and the ensemble. The network connections and parameter values for each cell did not change throughout the simulation.
The synchronization index (SI) provides a real-time measure of the phase dispersion across a population of oscillators, which ranges from 1 (all cells peak in phase) to 0 (all cells peak at uniformly-distributed times of the day). We defined SI at each time t by the radius r of the complex order parameter [53] according towhere N is the number of cells, φj(t) is the phase of the jth cell at time t, and ψ(t) is the average phase of all cells. We compute the instantaneous phase of each cell (simulated or real) by applying the continuous wavelet transform using a Morlet wavelet [54] to its trace of Period mRNA. The phase of the cell over time may be recovered from the ridges of the transform, which are extracted using a straight-forward algorithm ([52], [54]; Wavos Package). Briefly, the continuous wavelet transform (CWT) produces a complex-valued field over scales (which may be mapped to instantaneous frequency) and translations (which may be mapped to time). The magnitude of the complex number at a given translation and scale may be interpreted as the strength of oscillation of the signal at the frequency given by the scale and the time given by the translation. The phase of the complex number at a given translation and scale gives the phase of that oscillatory component. By selecting points with contiguous scales across a range of translations that maximize the magnitude of the CWT (the “wavelet ridge”), we may extract the dominant frequency of the oscillator over time, and from those points extract the phase evolution of the oscillator from the angles of the CWT coefficients.
Because the wavelet analysis requires a window (in time) around the model state in question, it is unable to calculate the phase during the first and last 34 hours of each simulation. We treat each experimental condition separately, which mean there are gaps in the SI plotted in Fig. 3.
|
10.1371/journal.pcbi.1002818 | The Evolution of Cell-to-Cell Communication in a Sporulating Bacterium | Traditionally microorganisms were considered to be autonomous organisms that could be studied in isolation. However, over the last decades cell-to-cell communication has been found to be ubiquitous. By secreting molecular signals in the extracellular environment microorganisms can indirectly assess the cell density and respond in accordance. In one of the best-studied microorganisms, Bacillus subtilis, the differentiation processes into a number of distinct cell types have been shown to depend on cell-to-cell communication. One of these cell types is the spore. Spores are metabolically inactive cells that are highly resistant against environmental stress. The onset of sporulation is dependent on cell-to-cell communication, as well as on a number of other environmental cues. By using individual-based simulations we examine when cell-to-cell communication that is involved in the onset of sporulation can evolve. We show that it evolves when three basic premises are satisfied. First, the population of cells has to affect the nutrient conditions. Second, there should be a time-lag between the moment that a cell decides to sporulate and the moment that it turns into a mature spore. Third, there has to be environmental variation. Cell-to-cell communication is a strategy to cope with environmental variation, by allowing cells to predict future environmental conditions. As a consequence, cells can anticipate environmental stress by initiating sporulation. Furthermore, signal production could be considered a cooperative trait and therefore evolves when it is not too costly to produce signal and when there are recurrent colony bottlenecks, which facilitate assortment. Finally, we also show that cell-to-cell communication can drive ecological diversification. Different ecotypes can evolve and be maintained due to frequency-dependent selection.
| Biological systems are characterized by communication; humans talk, insects produce pheromones and birds sing. Over the last decades it has been shown that even the simplest organisms on earth, the bacteria, communicate. Despite the prevalence of communication, it is often hard to explain how communicative systems evolve. In bacteria, communication results from the secretion of molecular signals that accumulate in the environment. Cells can assess the concentration of these signals, which indicate cell density, and respond in accordance. This form of cell-to-cell communication is responsible for the regulation of numerous bacterial behaviors, such as sporulation. Spores are metabolically inactive cells that are highly resistant against environmental stress. It is adaptive for a cell to sporulate when it struggles to survive. We show, via individual-based simulations, that cell-to-cell communication evolves because it allows cells to predict future environmental conditions. As a consequence, cells are capable of anticipating environmental stress by initiating sporulation before conditions are actually harmful. Furthermore, our model shows that cell-to-cell communication can even drive ecological diversification, since it facilitates the evolution of individuals that specialize on distinct ecological conditions.
| Complex systems in biology often come about through the communication of their parts, such as pheromone communication in insect societies and language in humans. Communication has been found to be ubiquitous in microorganisms as well [1]–[4]. Due to self-produced molecular signals that are secreted in the environment, cells can monitor the population density, which can quantitatively affect a cell's gene expression or trigger a differentiation process. In 1994, Fuqua and colleagues were the first to characterize this form of cell-to-cell communication as quorum-sensing signaling [5]. Quorum-sensing signaling has been shown to regulate a multitude of bacterial processes, such as extracellular enzyme production, antibiotic production and biofilm formation [6]–[11]. In one of the best-studied microorganisms, Bacillus subtilis, the differentiation of a number of cell types has been shown to depend on cell-to-cell communication [12]–[14]. These cell types emerge during the developmental process of biofilm formation and are presumably needed to survive the harsh environmental conditions that are present in the soil [10], [15], [16]. The most remarkable survival strategy among these cell types is that of the spore [17], [18].
A spore is a metabolically inactive cell that compartmentalized its DNA together with some essential proteins to survive starvation or other environmental stressors [18], [19]. Spore formation is an energy-expensive process that can take 6 to 8 hours and involves the expression of hundreds of genes [19], [20]. The initiation of sporulation is primarily dependent on the activation of a single transcription factor called Spo0A [14], [21]–[24]. When the level of activated Spo0A is sufficiently high, the sporulation process will be initiated [25]–[28]. The level of activated Spo0A is indirectly affected by a number of environmental and physiological cues, of which some are self-produced quorum-sensing signals [13], [22], [29]. These signals are assumed to accumulate in the environment and thereby give an indication of the cell density. As a consequence, the fraction of cells that initiate sporulation is higher for higher cell densities [29]–[33]. Even though these quorum-sensing signals affect the proportion of cells that initiate sporulation, they themselves are not sufficient for initiating sporulation since starvation is absolutely required [34]–[36]. Bischofs and colleagues (2009) mathematically modeled the regulatory mechanisms that integrate the quorum-sensing signals with other environmental cues, including those that are indicative of starvation [36]. They showed that the quorum-sensing signals allow for a density-dependent normalization of certain environmental cues. For example, when a cell can sense the amount of nutrients that are left in the environment, quorum-sensing signaling makes it possible to estimate the amount of nutrients that are left per cell. They concluded that these density-dependent normalizations might be adaptive for cellular decision-making, such as determining when to initiate sporulation (see also [34]).
However, despite the detailed knowledge of the regulatory mechanisms that underlie the sporulation process, little is known about their evolutionary origin. Why does cell-to-cell communication evolve and under which ecological and developmental conditions is it selected for? Here we examine, by using individual-based simulations, how three conditions, which inevitably relate to sporulation [15], [19], [20], [34], [37], [38], affect the evolution of cell-to-cell communication: environmental variation in nutrient conditions, costs of sporulation and time expenditure of sporulation. Even though our model is inspired by sporulation in B. subtilis, it is aimed to be conceptual and therefore does not include mechanistic details. The model is made such that it allows for the evolution of various developmental strategies, in which a cell's sensitivity and response to environmental cues can evolve.
Throughout the paper we discuss different versions of our model, which gradually increase in complexity. First we study the evolution of cell-to-cell communication under clonally-growing colonies. Next we allow for within colony-variation by initiating colonies with multiple individuals. Under these conditions multiple ecotypes evolve that transiently coexist over time due to negative frequency-dependent selection. Finally, we examine the evolution of cell-to-cell communication when signal production is costly. Under these conditions cooperative dilemmas emerge naturally and we find that different ecotypes evolve, which use different communicative strategies to time the onset of sporulation. The evolutionary significance of these strategies can only be understood by considering their ecological context.
We assume that cells are scattered throughout the soil. Only in a few locations these cells can grow and form colonies, because only in these areas there are nutrients available to do so. During colony growth cells consume nutrients in order to perform cell division and cell differentiation. A cell can differentiate into two cell types—a signal-producing cell or a spore—or it could remain undifferentiated. Eventually, all the nutrients will be depleted and a colony enters a starvation period. This period can only be survived by the spores. It is therefore crucial for a cell to initiate sporulation on time (i.e. when the nutrients that are needed to complete the sporulation process are still available). To decide when to initiate sporulation a cell could make use of two environmental cues: the nutrient concentration and the amount of quorum-sensing signal. The spores that eventually survive the starvation period migrate and germinate in new nutrient rich areas, where they form new colonies. Over evolutionary time, a cell's responsiveness to the environmental cues can evolve and thereby the timing of sporulation can evolve as well. We examine under which ecological and developmental conditions there is selection for cells that use quorum-sensing signaling to time the onset of sporulation. The system is studied by using individual-based simulations, which we describe in the following paragraphs.
We assume that the population of cells is divided into subpopulations, each representing a colony (i.e. biofilm or pellicle). Each colony is established by individuals. A colony is said to grow clonally when it is established by only one individual (). At the onset of colony growth there is a single nutrient input, which for each colony is taken from a normal distribution that is given by . Thus, the nutrient could be different for each colony. After receiving the nutrient input colonies are allowed to grow for a fixed number of time steps (); during this period cells consume nutrients in order to perform cell division and differentiation. At the end of a nutritional cycle all individuals (cells and spores) enter migration. The nutritional cycles of all colonies are synchronized such that the individuals from all colonies enter migration at the same time, forming a single migratory pool (see figure 1). Since migration occurs passively, we assume that all individuals have the same chance to establish a new colony. Thus, new colonies are established by choosing, for each colony separately, random individuals from the migratory pool. After this, the new colonies simultaneously start the next nutritional cycle.
Within a nutritional cycle three different cellular processes can occur at any time step (for each cell in the colony). First, a cell gets the opportunity to differentiate. A cell can differentiate into two different cell types—a signal-producing cell or a spore—or it could remain undifferentiated. A signal-producing cell secretes a fixed amount of signal in the environment. The more cells that produce signal, the higher the amount of extracellular signal. At the same time, the signal is degraded with a fixed rate . Thus, the amount of signal changes over time depending on the number of cells that are producing it. A cell could also initiate sporulation. Sporulation is an irreversible process that takes a fixed number of time steps () and during which a fixed amount of nutrients is consumed (), which is needed for making the spore. Thus, a sporulating cell consumes nutrients per time step. When there is an insufficient amount of nutrients in the environment, the sporulation process cannot be completed; in this case a cell inevitably dies. After completing the sporulation process, a mature and resistant spore is formed. A spore cannot divide, but has a much lower death rate than a cell. A spore germinates at the onset of a new nutritional cycle. Since sporulation requires time steps, a cell can be in one, out of , phenotypic states. It can be an undifferentiated cell, a signal producing cell or a sporulating cell, of which the latter is subsequently composed of states that indicate the number of time steps a cell has been sporulating (). At the final time step of sporulation () a cell turns into a spore. The cell's decision to differentiate into a signal-producing cell or spore depends in our model on two environmental cues—the amount of nutrients and signal—and on a cell's genotype (which we describe later).
The second cellular process that a cell can undergo, after having had the opportunity to differentiate, is division. All cells, excluding spores, have a certain chance of dividing. This chance is dependent on the amount of nutrients that are present in the environment (for details see equation S1). The more nutrients that are present in the environment, the greater the chance of cell division, with a maximum chance of . During each cell division a fixed amount of nutrients () is consumed. At each cell division there is a certain probability that the dividing cell incurs a mutation (the mutation process is described later).
The third and last cellular process that can occur at any particular time step is that of cell death. Both cells and spores have a fixed chance of dying, which is independent of the nutrient concentration. The death rate of a spore is much lower than that of a cell (). Hence, it is better to be a cell when nutrients are plentiful, because the chance of having cell division outweighs the chance of having cell death. On the contrary, when the nutrients are depleted, it is better to be a spore because spores have a smaller chance of dying than cells. The fitness of a genotype therefore depends on the timing of sporulation. When a genotype sporulates too early—at a nutrient concentration that is too high—it loses reproductive potential, since not all the nutrients are utilized. When a genotype sporulates too late—at a nutrient concentration that is too low—it has an increased risk of dying, especially when, due to nutrient scarcity, the sporulation process cannot be completed.
A crucial part of the model is the cell differentiation process. We aim to model it such that various developmental strategies can evolve. This requires to have sufficient degrees of freedom. On the other hand, we want to restrict the number of evolvable variables, in order to keep the model simple and tractable. The combination of these requirements resulted in a cell differentiation process that could be described by two Boolean decision-making steps, which are affected by the amount of nutrients and signal. The cell should decide to initiate sporulation or not and when it does not sporulate, a cell should decide if it wants to produce signal or not. These two decisions can be expressed by the following two inequalities (see figure 2):(1a)(1b)Inequality 1a shows when a cell initiates sporulation and inequality 1b shows when a cell initiates signal production. We assume that the decision to initiate sporulation is dominant over the decision to produce signal. Thus when both inequalities hold, only the sporulation process is initiated. The left hand side of each inequality contains the environmental cues: the amount of nutrients () and the amount of extracellular signal (). Since nutrients are consumed and signal can be produced and degraded over time, the values of these environmental cues change during colony growth. The effect of an environmental cue on the differentiation process depends on what we call the connection weight, ; here is the environmental cue (1 is the amount of nutrients and 2 is the amount of signal in the environment) that is affecting differentiation process (1 is sporulation and 2 is signal production). For example, determines how the amount of nutrients affects the initiation of sporulation. When a connection weight is positive, its corresponding environmental cue stimulates the differentiation process. When the connection weight is negative, the environmental cue inhibits the differentiation process. The absolute value of a connection weight shows the impact that a certain environmental cue has on the differentiation process. The right hand side of both inequalities is the activation threshold, ; here is the differentiation process to which the activation threshold belongs (1 is sporulation and 2 is signal production). The activation threshold shows how much stimulus from the environmental cues is required before the differentiation process is initiated. For example, when is positive a cell only sporulates when the stimulus from the nutrients () plus the stimulus from the signal () is bigger than the activation threshold (). On the contrary, when is negative a cell sporulates by default (when ) and sporulation can only be prevented if the environmental cues inhibit the sporulation process (i.e. negative connection weights). The activation thresholds could be viewed as a normalization of the connection weights. Namely, one could divide both sides of inequality 1a and 1b by the absolute values of, respectively, and , without altering the behavior of a genotype. Therefore the model could be simplified by fixing the activation thresholds (i.e. preventing mutations to occur in the activation thresholds), as long as it does not affect the strategies that can evolve. In the first two sections of the results we applied this simplification to the model and only allowed the connection weights to mutate. To show that this simplification did not affect the evolutionary outcome of the model we performed all simulations under non-simplified conditions and show the results in the supplementary information (figure S3). In the last section we did not fix the activation thresholds, because when signal production is assumed to be costly, the evolutionary outcome would be constrained by fixing the activation thresholds. We call the collection of connection weights () and activation thresholds () the genotype of an individual. In essence, the genotype describes how a cell responds to each combination of environmental cues.
When a cell division occurs each of the genotypic variables ( and ) has a certain chance to mutate (). When a mutation occurs, a small value taken from the normal distribution is added to the genotypic variable. Every mutation is taken independently from the same normal distribution, irrespective of the genotypic variable that mutates. All evolutionary simulations are initiated with the same monomorphic population of cells that do not produce signal and are not sensitive to it (). In addition, the initial cells are assumed to sporulate, to prevent the population from going extinct. The initial cells sporulate at a nutrient concentration of 500 ( and ; all input variables that are perceived by the cells are divided by 1000 as normalization, which is done consistently throughout the paper). Similar results would however be obtained if sporulation would occur at another nutrient concentration, as long as the initial population does not go extinct in the first growth cycle. By assuming that both and are negative, we assume that nutrients inhibit the sporulation process and that when this inhibition is too weak (e.g. when ) a cell initiates sporulation. Thus, we are not examining the evolution of sporulation, but the evolution of cell-to-cell communication as a mechanism to time the onset of sporulation.
A cell should turn into a spore when the growth rate of a spore exceeds that of a cell. The effective growth rate is given by the birth rate (i.e. chance of cell division; equation S1) minus the death rate (i.e. chance of cell death; and for respectively cells and spores). Since a spore cannot divide, its effective growth rate is , which is approximately equal to 0 (assuming that ). A cell should therefore turn into a spore when the chance of having cell death exceeds the chance of having cell division. The chance of cell division is subsequently dependent on the nutrient concentration (see equation S1). Thus, there is a critical nutrient concentration at which a cell should turn into a spore (see equation S2). However, sporulation costs time and during sporulation nutrients are consumed [19], [20]. In other words, the decision to sporulate has to be made in advance, before the critical nutrient concentration is reached. We examine why and when a cell uses quorum-sensing signals for its decision to sporulate. Moreover, we examine under which conditions cell-to-cell communication evolves. This is done for different variants of the model with increasing complexity. First, we examine if cell-to-cell communication evolves under the assumption that colonies grow clonally. Second, we examine how within-colony variation affects the evolution of cell-to-cell communication. Third and last, we examine if cell-to-cell communication evolves when signal production is costly.
In this section we examine the evolution of cell-to-cell communication under the assumption that colonies grow clonally, meaning that colonies are initiated by a single individual (). Genetic variation can only arise in these colonies via mutations. Moreover, for simplicity as explained before, we also assume that only the connection weights () can mutate (similar results are however obtained when the activation thresholds are allowed to mutate as well; see figure S3). Under these conditions, the timing of sporulation depends on and and the differentiation into a signal-producing cell solely depends on and (the activation thresholds, , are fixed over evolutionary time). To evolve cell-to-cell communication a cell should acquire two properties over evolutionary time. First, a cell should produce signal. Thus, before initiating sporulation a cell has to differentiate into a signal-producing cell. Second, a cell should be sensitive to the signal (), meaning that the nutrient concentration at which a cell initiates sporulation has to depend on the amount of signal. Irrespectively of the order in which these properties evolve, when both are present there is cell-to-cell communication. To examine if both properties can evolve in our model, we ran individual-based simulations that were initiated with a monomorphic population of cells that did not produce signal and were not sensitive to the signal (). Figure 3A shows two independent evolutionary trajectories projected on an adaptive landscape (for more replicates see figure S1).
The adaptive landscape is constructed by showing for each possible genotype—meaning each combination of and —the average colony size that is obtained at the end of a nutritional cycle. When solely examining the adaptive landscape, one expects that cell-to-cell communication would evolve, because the best-performing genotypes that are signal-sensitive () have a higher fitness than those that are signal-insensitive (). The two evolutionary trajectories that are plotted on the adaptive landscape are called run 1 and run 2 (both runs were performed under the same parameter settings). In both runs cell-to-cell communication evolved, which means that both signal-production and signal-sensitivity evolved. The evolutionary trajectories of figure 3A and S1 closely match the adaptive landscape and hence the adaptive landscape can be used to predict the outcome of evolution. The adaptive landscape only shows the selective advantage of cell-to-cell communication for and since nothing interesting happens outside this quadrant. In other words, nutrients are expected to inhibit sporulation (i.e. a cell only sporulates when there is nutrient scarcity), while signal is expected to stimulate sporulation (i.e. a cell sporulates earlier when it occurs in a bigger population). A limitation of the adaptive landscape of figure 3A is that it does not show the other two connection weights, and . and determine when a cell differentiates into a signal-producing cell (see figure 2). Signal production is, next to signal-sensitivity, essential for the evolution of cell-to-cell communication. To examine how signal production evolved we plotted the values of all connection weights (corresponding to the most-abundant genotypes), of run 1, along a time-axis (see figure 3B).
Figure 3B shows that signal production evolves after about 20.000 time steps ( becomes positive; as indicated by the green arrow). About 40.000 time steps later signal-sensitivity evolves as well ( becomes positive; as indicated by the blue arrow). In other words, signal production emerges before the occurrence of signal-sensitivity. Hence there was no selective advantage for signal production at the moment it evolved. Signal production evolved because a neutral mutation in hitchhiked along with a beneficial mutation in . Genetic hitchhiking is relatively prevalent, because there is no genetic recombination. In addition, there are no costs for signal production in this version of the model. Thus, cell-to-cell communication evolves by the sequential evolution of signal production and signal-sensitivity.
The question we are interested in though, is why cell-to-cell communication evolved at all. By sensing signal a cell can assess the colony size at the onset of sporulation. This estimate gives an indication of the amount of nutrients that will be consumed by the colony during sporulation. As explained before, a cell should turn into a spore when the chance of having cell death exceeds that of cell division, which is associated with a critical nutrient concentration (for details see equation S2). Since sporulation requires time, a cell has to anticipate or predict if the nutrient concentration at the end of sporulation matches this critical nutrient concentration. To make this prediction it is necessary to assess the amount of nutrients that will be consumed during sporulation. Since the total amount of nutrient consumption depends on the number of cells within a colony, it is advantageous for a cell to sense quorum-sensing signals. When the colony is big, a high amount of nutrients will be consumed during sporulation due to which a cell should initiate sporulation relatively early (i.e. at a high nutrient concentration). On the contrary, when the colony is small, a small amount of nutrients will be consumed and therefore a cell should initiate sporulation relatively late (i.e. at a low nutrient concentration). Thus, cell-to-cell communication allows a cell to predict the total amount of nutrient consumption during sporulation and, thereby, a cell can anticipate future environmental changes. There are three requirements that should be satisfied for cell-to-cell communication to evolve (corresponding to the parameter values in our model; see figure 4): (i) the colony size should affect the nutrient concentration during sporulation by, for example, nutrient consumption (); (ii) there should be a time-lag between the moment that a cell decides to sporulate and the moment that it turns into a mature spore (); and (iii) there should be environmental variation (). High values of , and (e.g. , and ) can result in a fitness advantage for cells that sense quorum-sensing signals over those that do not (figure 4).
The first requirement for the evolution of cell-to-cell communication is that the colony size should affect the nutrient concentration (figure 4A). For example, when each cell consumes a fixed amount of nutrients during sporulation (), the total nutrient consumption depends on the colony size. When there is no nutrient consumption during sporulation () the optimal time at which to initiate sporulation does not depend on the colony size and hence cell-to-cell communication does not evolve. Second, cell-to-cell communication only evolves when there is a time-lag between the moment that a cell decides to sporulate and the moment that it turns into a spore (figure 4B). In other words, sporulation should require time. When sporulation does not require time, there is no need to assess the nutrient consumption since a cell could turn into a spore instantaneously. Thus, cell-to-cell communication only evolves when . The third and last requirement for the evolution of cell-to-cell communication is the presence of environmental variation (figure 4). When there is no variation (), the amount of nutrients at the onset of a nutrient cycle is always the same. As a consequence, the changes in the nutrient concentration over time correlate with those of the colony size, since all colonies are initiated with the same number of cells, which reproduce at the same rate. Under these conditions, the nutrient concentration could be used as an accurate indication of the colony size, which makes the use of quorum-sensing signals superfluous, since these give an indication of the colony size as well. Only when the correlation between the nutrient concentration and colony size is relatively weak, the amount of signal could be used as a unique indication of the colony size. For this reason, there is stronger selection for cell-to-cell communication for higher levels of . Alternative conditions that weaken the correlation between the colony size and nutrient concentration can have a similar effect. For example, one could vary the initial colony sizes; colonies would still be clonal but different colonies would be initiated by different numbers of cells (see figure S8).
In most laboratory experiments sporulation is studied in isogenic populations. However, it is plausible that multiple genotypes can co-occur in a single colony [39]. In this section we examine how the developmental mechanisms that determine the onset of sporulation evolve when multiple genotypes can initiate a single colony (). This is done for the same conditions as those described in the previous section (i.e. only the connection weights, , are allowed to mutate; see figure S3 for simulations in which also the activation thresholds could mutate).
In figure 3C the evolutionary trajectory of a single run is shown on the adaptive landscape. Figure 3D shows, for the same evolutionary run, the connection weights of the most-abundant genotypes along a time-axis (for more replicates see figure S2). In contrast to the previous section, there is a bifurcation event during the evolutionary process that results in two coexisting ecotypes (an ecotype is a cluster of genotypes that is adapted to specific ecological condition). One of these ecotypes eventually goes extinct (see figure 3D and S2). Both ecotypes produce quorum-sensing signal and are sensitive to it. The ecotypes only differ in their responsiveness towards the nutritional conditions in the environment (). In one ecotype the value of is lower than in the other, meaning that the nutrients more strongly inhibit the sporulation process (see figure 3D and S2). This ecotype is therefore called the late sporulating ecotype (i.e. sporulation is initiated at a low nutrient concentration), while the other one is called the early sporulating ecotype (i.e. sporulation is initiated at a high nutrient concentration).
How can the late and early sporulating ecotypes stably coexist? In the absence of cell-to-cell communication, a genotype can only efficiently make use of the available nutrients for a limited range of nutrient inputs (i.e. nutrient concentration at the onset of a nutritional cycle; see figure S4 and S5B). When the nutrient input is higher than this particular range, a genotype would sporulate too late and when it is lower than this range a genotype would sporulate too early (see figure S4). When a genotype sporulates too early, not all the nutrients will be consumed. The leftovers can be used by other genotypes that sporulate slightly later and co-occur in the same colony. The late sporulating genotypes, in turn, cannot efficiently make use of the nutrients at high nutrient inputs, because they initiate sporulation too late. As a consequence, there is frequency-dependent selection in which the late sporulating ecotype has a selective advantage when the early sporulating ecotype is abundant and vice versa (see figure S6). Figure 3D shows that the early sporulating ecotype evolves first and later is accompanied by the late sporulating ecotype.
Over evolutionary time both the early and late sporulating ecotypes become more sensitive to the quorum-sensing signal (increase in ) and thereby evolve cell-to-cell communication (figure 3D). In other words, both ecotypes evolve the ability to adjust the timing of sporulation to the nutrient input. This increases the range of nutrient inputs at which an ecotype could efficiently make use of the nutrients (see figure S5C). As a consequence, there is an increasing overlap in the range of nutrient inputs at which both ecotypes grow efficiently, hence strengthening the competition between them. Ultimately, only a single ecotype survives (see figure 3D and S2). This ecotype is a generalist, since it grows efficiently at most nutrient inputs due to the evolved cell-to-cell communication. Thus, over evolutionary time, the evolved specialists—the early and late sporulating ecotypes—are replaced by a generalist—a signaling ecotype—that can grow efficiently at most nutrient inputs.
Not surprisingly, when there is no environmental variation (), a bifurcation event cannot occur. In that case only a single ecotype evolves that outcompetes all others (see figure S7). Branching is most likely to occur for high levels of (see figure S7); the same conditions that select for cell-to-cell communication (see figure 3 and 4). Another condition under which a bifurcation event cannot occur is clonal growth, since it hampers the presence of within-colony variation. Within-colony variation allows for competition at the cellular-level and hence for the coexistence of multiple ecotypes. However, allowing for within-colony variation can also result in a conflict between the genotypes that are selected for at the colony-level and those that are selected for at the cellular-level. In particular, when signal production is costly conflicts are expected, since cells that do not produce the costly signal have a fitness advantage at the cellular-level but undermine the performance of the colony. In the next section we examine whether cell-to-cell communication evolves when signal production is costly.
In this section we examine whether cell-to-cell communication can still evolve when signal production is costly. We assume that a signal-producing cell has a reduced chance of dividing by subtracting a fixed value () from the chance of having cell division (see equation S3). In contrast to the previous sections, all genotypic variables can mutate, to allow for a wider variety of communicative strategies. In this section we focus on a single representative evolutionary run (for more replicates see figure S9).
Figure 5 shows the outcome of this evolutionary run, by using a phenogram. The phenogram shows the dissimilarity between genotypes in a population that evolved for 550.000 time steps. The genotypes are named by letter-codes, which are ranked in alphabetic order and represent abundance, with genotype ‘AA’ being the most abundant and genotype ‘CH’ the least. Besides the letter-code, every genotype is connected to a small graph, which shows its phenotype for a range of environmental conditions. The population consists of multiple communicative strategies that cluster together. The three most-abundant genotypes partly reflect these clusters and are shown on the left side of the phenogram. Since, the phenogram does not show evolutionary descendance, the evolutionary lineages of the three most-abundant genotypes were used to construct an evolutionary tree. This tree is shown in figure 6. Hereafter, the phenotypes of the three most-abundant genotypes are called phenotype 1, 2 and 3; corresponding to the order in which they appear in figure 6.
All three phenotypes produce quorum-sensing signal for a range of parameter conditions (shown by the green areas in figure 6). Phenotype 2 produces quorum-sensing signal for all environmental conditions, except for those at which it sporulates. Since signal production is costly this phenotype is exploited by phenotype 1 and 3, which lack signal production for respectively high and low nutrient concentrations. As a consequence, phenotype 2 is always selected against at the cellular-level, irrespective of the population composition at the onset of a nutritional cycle. However, phenotype 2 is maintained in the population due to selection at the colony-level, in which the colonies that contain phenotype 2 often have a selective advantage over those that do not contain phenotype 2 (for details see table S1). This selective advantage results from the improved timing of sporulation. Thus, the selection pressures at the colony-level outweigh those at the individual-level. Since the other two phenotypes exploit phenotype 2 for different environmental conditions, they occupy different niches.
Figure 7 shows the selection pressures that act on each phenotype, given the frequency at which each phenotype occurs in the population (frequency over all colonies). The fitness measurements include the selection processes at the cellular- and colony-level. All phenotypes have a selective advantage when they are present in a low overall frequency. Thus, negative frequency-dependent selection is responsible for the stable coexistence of the three phenotypes. Since the three phenotypes are subject to a continuing process of evolution, it is unlikely that these specific phenotypes would coexist forever. Frequency-dependent selection does however assure the coexistence of multiple ecotypes, as shown by figure 5 and S9.
It is important to notice that the evolutionary simulation shown by figures 5, 6 and 7 assumes relatively low costs for signal production and a small bottleneck size. The costs of signal production are 2% of the maximal growth rate (), which means that a signal-producing cell has a 2% smaller chance to divide than an undifferentiated cell under the optimal growth conditions. The bottleneck size is given by the number of individuals that initiate a single colony (). Smaller bottleneck sizes facilitate assortment, because signal-producing cells are more likely to end up in a colony that only contains signal-producers. As a consequence, signal-producing cells are less likely to be exploited by cells that lack signal production. Figure 8 shows how the evolution of cell-to-cell communication depends on and , by showing the average amount of signal that is present in a population that evolved for 550.000 time steps. As expected, cell-to-cell communication is more likely to evolve for smaller signal costs and stronger population bottlenecks.
In conclusion, when signal production is costly, cell-to-cell communication can still evolve. However, signal-producing cells can be exploited by cells that lack signal production. This ultimately results in the evolution of ecological diversity, in which multiple ecotypes can coexist. Even though it is to be expected that signal production costs result in cheating (i.e. cells that do not produce signal), it is less intuitive that three ecotypes would evolve, including one that cheats for high nutrient inputs and another that cheats for low nutrient inputs. This coexistence is facilitated by negative frequency-dependent selection, which results from the selection processes at the cellular- and colony-level. Cell-to-cell communication only emerges in our simulations for relatively low costs of signal production and in the presence of population bottlenecks.
We demonstrated that cell-to-cell communication can evolve to regulate the timing of sporulation. The evolution of cell-to-cell communication requires both the evolution of signal production and signal-sensitivity. By sensing quorum-sensing signals a cell can predict future environmental conditions and thereby anticipate a starvation period by initiating sporulation. To predict the environmental conditions a cell has to assess the rate of nutrient consumption, which depends on the colony size. Our model shows that three conditions, which inevitably relate to sporulation, are sufficient to explain the evolution of cell-to-cell communication: (i) the population size has to affect the nutrient concentration (); (ii) a cell has to predict future environmental conditions (; see also [40]–[42]); and (iii) there has to be environmental variation (). Irrespectively of how these conditions come about, when all three are satisfied and signal production is not too costly, cell-to-cell communication evolves. It is not our claim that these conditions are strictly necessary, but rather that they are sufficient for the evolution of cell-to-cell communication. In nature, the requirements for the evolution of cell-to-cell communication in sporulating bacteria might be less stringent, since additional advantages, besides the timing of cell differentiation, can facilitate the evolution of cell-to-cell communication (e.g. colony-level properties; [2]).
In contrast to previous models on the evolution of cell-to-cell communication [43]–[46], our model shows that cell-to-cell communication can evolve as a mechanism to evaluate other environmental cues [34], [36]: neither the absolute signal concentration nor the absolute nutrient concentration determine the onset of sporulation. To understand when cell-to-cell communication evolves one has to understand how the information that results from quorum-sensing signaling is integrated with that of other environmental cues [47]–[49]. Moreover, we have demonstrated that cell-to-cell communication can even evolve when there is genetic variation within the colony and, in addition, when signal production is costly. Models on sporulation (or other persistence phenotypes) often exclude cell-to-cell communication as a mechanism to regulate sporulation [27], [50], [51]. This is because sporulation is mostly studied as a bet-hedging strategy: only a small fraction of genetically-identical cells sporulates under the same environmental conditions [28], [40]. Bet-hedging is a risk-spreading strategy that ensures the survival of a colony when there are severe and sudden environment changes [52], [53]. In our model a bet-hedging strategy cannot evolve, because cells always perceive accurate environmental information and lack developmental noise. Furthermore, bet-hedging is only beneficial when environmental changes are unpredictable [50], [54]. In our model, environmental changes might only become unpredictable when a cell is surrounded by different ecotypes, which differ in the amount of signal production and the timing of sporulation. It might therefore be interesting to extend the model, in order to examine how the evolution of bet-hedging affects that of cell-to-cell communication.
In our model, cell-to-cell communication represents a form of phenotypic plasticity, because it allows a cell to adjust the timing of sporulation in response to environmental changes [55]. Without cell-to-cell communication a cell can only grow efficiently for a limited range of nutrient inputs (figure S5). In that case, multiple ecotypes evolve that specialize on distinct ecological niches (e.g. the late and early sporulating ecotypes that evolved at the onset of our simulations, see figure 3C–D). However, by evolving cell-to-cell communication the range of nutrient inputs at which a cell grows efficiently increases. This ultimately results in competitive exclusion: the specialized ecotypes (i.e. narrow niche width)—such as the late and early sporulating ecotypes—are replaced by a single generalist (i.e. broad niche width) that can grow efficiently under most environment conditions due to cell-to-cell communication [56]–[58]. In our model phenotypic plasticity is a colony-level property, instead of a cellular property, since cells cannot respond to changes in environmental conditions without cooperation [59]: the amount of signal only gives an accurate indication of the colony size when all cells (or a constant fraction) produce quorum-sensing signals. The evolution of cell-to-cell communication therefore entails a cooperative dilemma (given that signal production is costly; [4], [60]–[62]). Cells that do not produce signal (i.e. public good) have an advantage over those that do, but at the same time they undermine the colony performance (see also [4], [63]–[66]). The cells that do not produce signal could therefore be called ‘cheaters’, while signal-producing cells are ‘cooperators’.
In our model, cheaters and cooperators evolved and stably coexisted due to frequency-dependent selection [43], [46], [67]–[71]. They have different communicative strategies [72] and therefore occupy distinct complementary niches (see figure 5 and 6). That is, the cheaters lack signal production for different subsets of environmental conditions. This emphasizes the importance of studying cell-to-cell communication under a wide range of environmental conditions, since a cooperator under one condition might be a cheater under another. The population structure (see figure 1), which results in two levels of selection, was essential for the maintenance of the different ecotypes [66], [73], [74]. Previous studies have shown that population structure can facilitate the evolution and maintenance of cooperation [69], [75]–[83]. The population structure makes individuals interact assortatively [84]: cooperators are therefore more likely to interact with other cooperators than cheaters. As a consequence, the benefits of cooperation mostly end up with cooperators, due to which there is a net selective advantage for cooperation. In our model the degree of assortment depends on the number of individuals that initialize a single colony () or, in other words, on the strength of the recurrent population bottlenecks [85], . We assumed that the colonies themselves are well-mixed, although within-colony structure—via the emergence of assortment—might have facilitated cooperation even more [87], [88]. When signal production is too costly, cell-to-cell communication does not evolve, because the selective advantage of cheaters at the cellular-level cannot be compensated by the selective advantage of cooperators at the colony-level. It is important to notice that our model only included signal production costs, even though plausible arguments could be made that the maintenance costs of a communicative system should be considered as well [89]. However, we do not expect that including maintenance costs would affect our results, since both cheaters and cooperators need to have a communicative system—and hence carry the associated costs—to sense the quorum-sensing signal.
Although our model is limited to sporulation, it could be extended to examine the role of cell-to-cell communication in the timing of other differentiation events as well, for example: motility, bioluminescence, conjugation, competence, matrix-production, biofilm formation, biofilm detachment, etc. (e.g. [11]–[14], [47], [90]–[94]). Every time there is a trade-off between the growth rate of two cell types (e.g. cells and spores) over two or more environmental niches that alternate over time (e.g. nutrient availability and nutrient scarcity), a cell has a selective advantage when it accurately times the developmental transitions between both cell types (see also [45]). When the population size affects the optimal time at which a cell should differentiate (e.g. when a cell must predict future nutrient conditions), cell-to-cell communication is expected to evolve in order to enhance a cell's developmental timing. The challenge for future studies is to unravel the developmental trade-off and ecological niches that underlie each of these differentiation events. Furthermore, our study emphasizes the importance of examining the integration of different environmental cues in cellular decision-making [49], [95]–[97]. The quorum-sensing threshold—and hence the critical population density—at which a differentiation event occurs can and mostly will strongly depend on other environmental conditions, such as nutrient availability [34], [48], [98], [99].
|
10.1371/journal.pntd.0005950 | Spatial and temporal trends of visceral leishmaniasis by mesoregion in a southeastern state of Brazil, 2002-2013 | Visceral leishmaniasis (VL) is expanding in Brazil and in other South American countries, a process that has been associated with the urbanization of the disease. This study analyzes the spatial and temporal distribution of VL in the Brazilian state of Minas Gerais and identifies the areas with higher risks of transmission.
An ecological study with spatial and time series analyzes of new confirmed cases of VL notified to the Brazilian Notifiable Disease Information System between 2002 and 2013, considering the 12 mesoregions of Minas Gerais. Two complementary methodologies were used: thematic maps of incidence and Poisson (log-linear) generalized linear model. Thematic maps using crude and smoothed cumulative incidences were generated for four trienniums. Poisson Regression measured the variation of the average number of cases from one year to the following, for each mesoregion.
The 5,778 cases analyzed revealed a heterogeneous spatial and temporal distribution of VL in Minas Gerais. Six mesoregions (Central Mineira, Jequitinhonha, Metropolitan area of Belo Horizonte, Northwest of Minas, North of Minas, and Vale do Rio Doce) were responsible for the expansion and maintenance of VL, with incidence rates as high as 26/100,000 inhabitants. The Vale do Rio Doce and Jequitinhonha mesoregions showed a considerable increase in the incidence rates in the last period studied. The other six mesoregions reported only sporadic cases and presented low and unsteady incidence rates, reaching a maximum of 1.2/100,000 inhabitants.
The results contribute to further the current understanding about the expansion of VL in Minas Gerais and may help guide actions for disease control.
| This article presents the spatial and temporal distribution of visceral leishmaniasis (VL) in Minas Gerais State and identifies the greater risk areas of transmission. This study is both timely and substantive because Minas Gerais is an important Brazilian state in the number of cases of visceral leishmaniasis. The results showed that during the 12-year time series the VL had a heterogeneous spatial and temporal distribution in the state of Minas Gerais. Among the 12 existing mesoregions, six (Central Mineira, Jequitinhonha, Metropolitan area of Belo Horizonte, Northwest of Minas, North of Minas, and Vale do Rio Doce) were responsible for the expansion and maintenance of VL in the state. Among them, the Vale do Rio Doce and Jequitinhonha mesoregions presented a considerable increase in the incidence rates of the disease in the last period. In the other six mesoregions only sporadic cases of the disease were reported during the study period. The results of in this study may contribute to a better understanding the dynamic of the disease in Minas Gerais. Also these findings can provide subsidies to assist the actions of the control program of VL.
| Until 1980, visceral leishmaniasis (VL) was considered a strictly rural disease in Brazil, the main parasite (Leishmania infantum) reservoirs being foxes (Dusicyon vetulus and Cerdocyon thous) and marsupials (Didelphis albiventris) [1]. However, the epidemiological profile of VL shifted with the urbanization of the disease and domestic dogs became the main reservoir (Canis familiaris) [1,2]. Since the first VL epidemic in 1981 in Teresina, the capital of the state of Piauí located in northeastern Brazil, various epidemics occurred in other major urban centers in the northeast (São Luís, Natal, and Aracajú), north (Boa Vista and Santarém), mid-west (Cuiabá and Campo Grande), and southeast (Belo Horizonte and Montes Claros) regions of the country [3].
The geographic expansion of VL is associated with the process of urbanization of the disease. Indeed, migration of people from rural endemic areas to urban centers, adaptation of the vector to the domestic environment, the presence of disease reservoirs such as domestic dogs, malnutrition, and the lack of basic sanitation are considered contributing factors to the urbanization and geographic expansion of VL [3–5]. In this context, the Visceral Leishmaniasis Control and Surveillance Program (VLCSP) was implemented in Brazil to reduce the risk of transmission, the lethality, and the morbidity rates of VL in urban and rural areas. The program has three main pillars: the treatment of human cases, control of canine reservoirs, and vector control [1].
Aiming to identify areas where VLCSP strategies should be prioritized, a temporal study (2001–2011) of the disease’s incidence was conducted and pointed out the southeastern Brazilian state of Minas Gerais as a priority [6]. The VL incidence rates in this state were 1.6 and 1.4 for 100,000 inhabitants in 2012 and 2013, respectively. These rates were superior to those found for the entire southeast region (0.6 e 0.5/100,000 inhabitants) and similar to the national rates (1.6/100,000 inhabitants) [7].
The first cases of human VL in Minas Gerais were registered from 1940 in the North of Minas mesoregion [8] and, in the 1960’s, in the Vale do Rio Doce mesoregion [9]. In 1989, the first autochthonous case in an urban area was registered in the municipality of Sabará [10], which belongs to the metropolitan region of Belo Horizonte. Later, in 1994, the first autochthonous case was registered in Belo Horizonte, the capital of the state [11].
Currently, some cities of the state of Minas Gerais are considered endemic for VL and have attracted studies regarding the disease. Among those, stands out Montes Claros [12] and Porteirinha [13] in the North of Minas mesoregion; Paracatu [14] in the Northwest of Minas mesoregion; and Belo Horizonte in the Metropolitan area of Belo Horizonte mesoregion [15,16]. Governador Valadares, located in the Vale do Rio Doce mesoregion, registered cases of human VL in the 1960’s [9] and, following disease control measures, the municipality became a silent area for VL [17]. Unfortunately, disease control actions were interrupted in the 1990’s [18,19] and reporting of VL cases resumed in 2008 [19,20] to the point that, currently, Governador Valadares is considered a re-emergent focus of intense VL transmission [19].
Several studies have attempted to understand VL urbanization and geographic expansion by means of spatial and temporal analyses. They studied the distribution and variation of the incidence rates of human [6,21–23] the distribution of canine infection cases [24,25] the abundance of phlebotomine sand flies [25–27], and the temporal trends of the disease [23,28,29]. Others identified areas where VL control measures could be prioritized [30–32].
Studies conducted in other Brazilian states such as Pernambuco [21], Mato Grosso do Sul [22], São Paulo [33,34] and Maranhão [35] evaluated the dispersion of VL over time. To our knowledge, no study to date has evaluated the VL dispersion in the state of Minas Gerais and therefore more robust and updated research is needed to unveil the VL profile in this area. Indeed, a combination of different methodologies for spatial and temporal analyses of VL may be useful to understand the aggregation, maintenance, and dispersion patterns of the disease, not only in Minas Gerais, but also in other regions of Brazil.
The present study analyzed the spatial and temporal distribution of VL in the state of Minas Gerais between 2002 and 2013 using two different methodologies to further understand, characterize, and quantify the expansion of the disease in the vast territory occupied by this Brazilian state. The results of this study identify within the state the areas that could be prioritized by the control and vigilance of VL, considering the specificities of each mesoregion of the state of Minas Gerais.
This study was approved by the Ethics Committee on Research of the Federal University of Minas Gerais (UFMG) under the number CAAE n.45497015.3.0000.5149. Only secondary information about the patients was collected from the Brazilian Notifiable Disease Information System (SINAN/VL). Data were analyzed anonymously.
This is an ecological study analyzing the spatial and temporal patterns of confirmed cases of VL notified to the SINAN/VL in the period between 2002 and 2013, in which the units of analysis were the 12 mesoregions of Minas Gerais.
Minas Gerais is one of the 27 federative units of Brazil, being located in the southeast region of the country. It has an area of 586,521.24 km² and ranks as the fourth largest state in territorial extension. With the highest number of municipalities (853) in the country, Minas Gerais is the second most populous Brazilian state with an estimated population of 20,997,560 in 2016, and a population density of 33.41 inhabitants/km2. The capital is Belo Horizonte [36].
The Brazilian Institute of Geography and Statistics (IBGE) divides Minas Gerais in 12 mesoregions (Fig 1): Campo das Vertentes; Central Mineira; Jequitinhonha; Metropolitan Area of Belo Horizonte; Northwest of Minas; North of Minas; West of Minas; South/Southwest of Minas; Triângulo Mineiro/Alto Paranaíba; Vale do Mucuri; Vale do Rio Doce; and Zona da Mata [36]. Each mesoregion is an aggregate of municipalities from the same geographic area presenting similar social and economic profiles and natural conditions.
In Brazil, VL is a disease of compulsory notification, i. e., in the case of clinically suspected VL, health professionals must fill in a specific SINAN/VL form to start up procedures to investigate the disease. Initially, data such as the home address of the patient, age, gender, schooling, occupation, date when the first symptoms erupted, date of notification, and clinical manifestations (signs and symptoms) are collected. Later, additional information such as laboratorial exam results, date of treatment commencement, medication used, and outcomes are added to the system.
During the period analyzed in this study (2002–2013), the SINAN/VL database changed platform from Windows-based version (the ILeishVi 2002–2006) to Net version (LEISHNET, 2007–2013). Therefore, the databases had to be standardized and a new unified dataset had to be generated with the same variables for analyzes using a Microsoft Office Excel 2013.
In the period analyzed herein, 13,409 suspected cases of VL were registered in the SINAN/VL. The number of confirmed cases was 6,158 (46%), 6,904 (51.5%) cases were discarded because the disease was not confirmed, and 347 (2.5%) cases had notification forms lacking information regarding the course of the disease. Among the confirmed cases, only new cases were included in the present study, totaling 5,778 cases (Fig 2).
The data were analyzed in two phases. First, thematic maps of the state of Minas Gerais were generated with the crude and the smoothed cumulative incidence rates of VL for each municipality within the mesoregions of the state. The software MapInfo 10.0 (MapInfo Corporation, Troy, New York) and TerraView 4.2.2 (Instituto Nacional de Pesquisas Espaciais, INPE, SP, Brazil) were used. The information regarding the estimated resident population of each municipality and the cartographic basis of the state were obtained from IBGE [37].
Because the analyzes covered 12 years (2002–2013) of notifications, the data involved was divided into four maps, each comprising three years of study: 2002–2004, 2005–2007, 2008–2010 and 2011–2013. The intervals of incidence rates used were chosen using the software MapInfo 10.0 considering the quartile, mean, median, and minimum and maximum values.
The cumulative crude incidence was calculated using Microsoft Office Excel 2013, and the smoothed cumulative incidence was obtained with the software TerraView 4.2.2. The local empiric Bayesian estimator, which allows for the estimation of the incidence of a municipality using the incidence rates of the neighboring municipalities converging to a local mean, was used. To evaluate the spatial variability of the data, a spatial proximity matrix using contiguity-based spatial weights was built. The elements of this matrix can take values 1 (if geographical analytical units are adjacent) or 0 (otherwise) [38]. Following the calculations, thematic maps for both the crude and the smoothed cumulative incidence rates were generated using MapInfo 10.0.
Next, Generalized Linear Models (GLM), through Poisson Regression, using STATA version 12.0 software (Stata Corp., College Station, TX, USA), was used to quantify the variation of the average number of VL cases from one year to the next in each of the 12 mesoregions of Minas Gerais. A curve with the annual incidence rates was generated for each mesoregion. Visual inspection of the resulting graphs allowed the establishment of cut off points according to the trends of increase or decrease of the number of VL cases over time. Consequently, the time interval analyzed was divided differently for each mesoregion.
The response (or dependent) variable was the “number of cases”, and the “year” was used as the independent variable in the temporal series. The logarithm of the population [log(pop)] was used as the “Offset” term, i. e., was a known component included for adjustment of the model. The equation used was as follows:
Log(μi)=log(popi)+α+β*year
Where:
The equation (eβ-1)*100% was used to obtain the variation of the mean number of VL cases from one year to the following.
The present study only included new cases of VL, which totaled 5,778. Of these, 89% were confirmed by at least one diagnostic test (ELISA and/or IFAT and/or parasitological). Also, the SINAN /VL’s platform (LEISHNET, n = 3349) included one variable informed that 95% of the cases were considered positive by clinical-laboratory criteria, while only 5% were confirmed by the clinical-epidemiological criteria. Therefore, we consider that the cases studied were correctly classified as VL. Of the cases analyzed, 91% were from urban areas.
The mesoregions of Central Mineira, Jequitinhonha, Metropolitan area of Belo Horizonte, Northwest of Minas, and North of Minas concentrated most VL cases (91%) and the highest number of municipalities with registered cases (66%). Among those, the Metropolitan area of Belo Horizonte mesoregion was the one with the largest population, the highest number, and the widest variation of VL cases, considering the population of its municipalities. On the other hand, the mesoregions of Campo das Vertentes, South/Southwest of Minas, Vale do Mucuri, and Zona da Mata presented the lowest number of cases (1%) and the lowest number of municipalities which reported VL cases (13%) (Table 1).
The variation of the incidence from 2002 to 2013 was obtained from the thematic maps of crude (Fig 3) and smoothed (Fig 4) cumulative incidence rates, for each municipality of the 12 mesoregions of Minas Gerais. The highest incidence rates concentrated in the mesoregions located in the north (Northwest of Minas, North of Minas, and Jequitinhonha), east (Vale do Rio Doce), and central (Central Mineira and Metropolitan area of Belo Horizonte) parts of the state (Table 2).
During the first three trienniums, the mesoregions of Northwest of Minas and North of Minas presented areas with the highest crude cumulative incidence rates. In the last triennium, however, a discrete decrease was observed in the rates in both these mesoregions. The highest incidence rate throughout the study, and among all the mesoregions, was observed in the Northwest of Minas mesoregion in the second triennium (67.7/100,000 inhabitants). This mesoregion presented a considerable increase in the incidence rates from the first to the second triennium and a subsequent reduction in the last two trienniums. Nonetheless, it was the mesoregion with the highest incidence rate within the state of Minas Gerais in the last triennium (31.2/100,000 inhabitants).
Paracatu, one of the most populous municipalities of the Northwest mesoregion, presented high VL incidence rates throughout the period of study (approximately 60/100,000 inhabitants). There was an increase in the incidence rates during the full study period in the municipalities of Unaí and Brasilândia de Minas, where VL cases were already observed between 2002 and 2004. Moreover, new cases arose in other municipalities (such as João Pinheiro and Guarda Mor), indicating the expansion of the disease in this mesoregion (Fig 3). This information is visualized in the map of smoothed incidence (Fig 4). In the last triennium (2011–2013), however, a slight reduction was observed in the number of municipalities that presented cases of VL (Figs 3 and 4).
The North of Minas mesoregion presented a high number of municipalities with VL cases during the four trienniums evaluated (17 municipalities). Among them, Montes Claros, Porteirinha, Matias Cardoso, Janaúba, Capitão Enéas, Montalvânia, Francisco Sá, and Nova Porteirinha showed the highest incidence rates. Figs 3 and 4 show the reduction in the incidence rates of VL in the municipalities of this mesoregion over the period analyzed. The incidence reduced from 29.9/100,000 inhabitants in the first triennium to 10.9/100,000 inhabitants in the last. This mesoregion showed the largest reduction in the incidence between the first and the second periods, as shown by the smoothed incidence rate maps (Fig 4).
The Vale do Rio Doce mesoregion had the largest increase in the incidence rates over the temporal series. Although incidence rates in this region remained stable during the first two trienniums (1.1 and 1.0/100,000 inhabitants, respectively), a considerable increase was observed in the last two (10.1 and 10.4/100,000 inhabitants, respectively). This increase reflected the expansion of VL to the east and the center of the mesoregion, particularly to the municipalities of Governador Valadares, Conselheiro Pena, Ipanema, Resplendor, and Tumiritinga (Figs 3 and 4).
The Central Mineira and Metropolitan area of Belo Horizonte mesoregions had small oscillations in the incidence rates of VL throughout the study period (Table 2). Nevertheless, the incidence rates reduced in the final triennium (10.3 and 8.5/100,000 inhabitants, respectively) in relation to the initial (11.8 and 11.1/ 100,000 inhabitants). The following municipalities showed incidence rates higher than 5/100,000 inhabitants in all four trienniums: Curvelo, Presidente Juscelino, and Inimutaba (all within the Central Mineira mesoregion); Belo Horizonte, Ribeirão das Neves, Sabará, Ibirité, Prudente de Morais, Vespasiano, Jaboticatubas, Sarzedo, Sete Lagoas, São Joaquim de Bicas, and Nova Lima (all within the Metropolitan area of Belo Horizonte mesoregion) (Figs 3 and 4).
Slight oscillations in VL incidence rates were also observed in the Jequitinhonha mesoregion. This oscillation is difficult to detect in the maps because different municipalities presented VL cases at various times (Figs 3 and 4). During all the time intervals evaluated, only seven municipalities (Berilo, Araçuaí, Jequitinhonha, Almenara, Itaobim, Diamantina, and Virgem da Lapa) stood out with very intense colors in the maps. In the last triennium (2011–2013) the incidence rate increased (15.8/100,000 inhabitants) in comparison with the three previous trienniums (10.7 for the first; 13.3 for the second, and 8.2/100,000 inhabitants for the third).
The other mesoregions (Campo das Vertentes, West of Minas, South/Soutwest of Minas, Triângulo Mineiro/Alto Paranaíba, Vale do Mucuri, and Zona da Mata) reported few VL cases and, consequently, low incidence rates throughout the study (Figs 3 and 4). Thus, these mesoregions were less important, as well as their variations over time in relation to VL.
Fig 5 shows the crude VL incidence rates obtained for each mesoregion between 2002 and 2013. These rates fluctuated over time and from one mesoregion to another. The following mesoregions reached elevated incidence rates, ranging from 0 to 26/100,000 inhabitants: Central Mineira, Jequitinhonha, Metropolitan area of Belo Horizonte, Northwest of Minas, North, of Minas and Vale do Rio Doce. On the other hand, the following mesoregions presented low VL incidence rates, reaching a maximum of 1.2/100,000 inhabitants: Campos das Vertentes, West of Minas, South/Southwest of Minas, Triângulo Mineiro/Alto Paranaíba, Vale do Mucuri, and Zona da Mata (Fig 5). The VL incidence rates in the whole state of Minas Gerais varied from 1.6 to 3.5/100,000 inhabitants.
The time periods used in the Poisson Regression model were selected by visual analysis of the graph presented in Fig 5. For each mesoregion, different time intervals were chosen, considering the intervals of increase or reduction in the incidence rates per mesoregion. The results of the adjustment of the model are shown in Table 3.
A single period (2002–2013) was chosen to analyze the mesoregions of Campos das Vertentes, Central Mineira, South/Southwest of Minas, Vale do Mucuri, and Zona da Mata). This was because it was not possible to observe trends in growth or reduction in the average number of VL cases occurring in these mesoregions, given the instability in the number of cases observed in the temporal series analyzed (Fig 5). The results obtained with the Poisson regression corroborated the analysis of the graph, as the results of the model were not significant, except in the case of the mesoregion South/Southwest of Minas. Indeed, the variation in the average number of VL cases per year was relatively stable during the study period (Table 3). However, the slope coefficient (β) obtained for the South/Southwest of Minas mesoregion was -0.22, indicating a reduction in the average number of cases between the beginning and the end of the period analyzed, despite the fluctuations observed over the years (Fig 5).
The Metropolitan area of Belo Horizonte, Northwest of Minas and North of Minas mesoregions presented similar results. Table 3 showed that, in the initial periods, the average number of cases per year rose 5%, 152%, and 43%, respectively. In the last periods, the average number of cases reduced in 15%, 12% and 13%, respectively. Of note is the mesoregion Northwest of Minas, which exhibited a marked increase in the average number of VL cases between the years of 2002 and 2005 (152%).
The time intervals analyzed in the West of Minas and Vale do Rio Doce mesoregions were divided into three periods. It was not possible to establish a significant increase or reduction in the average number of cases per year, in the first periods analyzed (Table 3). In the intermediate intervals, the coefficients were positive, demonstrating an increase in the average number of cases (108% in the West of Minas and 90% in Vale do Rio Doce). On the other hand, the last periods showed a reduction of 34% (West of Minas) and of 19% (Vale do Rio Doce) in the number of VL cases from one year to the next.
Two periods were chosen to analyze the Triângulo Mineiro/Alto Paranaíba mesoregion: 2002–2009 and 2009–2013. In the first period, the mesoregion presented a positive slope coefficient of the model (0.25), with an increase of approximately 28% in the average number of VL cases from one year to the next. In the interval between 2009 to 2013, the variations were not significant and it was not possible to establish a co-relation between the average number of cases from one year to the next. Thus, the average number of cases remained relatively constant during this period (Table 3).
Three periods were chosen to analyze the Jequitinhonha mesoregion: 2002–2005, 2005–2008, and 2008–2013. In the first interval, the model coefficient was 0.51, indicating an increase of approximately 66% in the average number of cases from one year to the next. On the other hand, the coefficient was negative (-0.39) in the period between 2005 and 2008, demonstrating a reduction of 32% in the average number of cases. During the final period, this mesoregion was the only one that presented a positive coefficient (0.16), with the average number of cases increasing by 17% from one year to the next.
The present study shows that VL had a heterogeneous spatial and temporal distribution in the state of Minas Gerais, in the period between 2002 and 2013. Among the 12 existing mesoregions, six (Central Mineira, Jequitinhonha, Metropolitan area of Belo Horizonte, Northwest of Minas, North of Minas, and Vale do Rio Doce) were responsible for the expansion and maintenance of VL in the state. Among them, the Vale do Rio Doce and Jequitinhonha mesoregions presented a considerable increase in the incidence rates of the disease in the last triennium (2011–2013). The North of Minas Gerais and Metropolitan area of Belo Horizonte mesoregions reduced the incidence rates in the last years of the study, despite the elevated number of VL cases. In the other six mesoregions (Campo das Vertentes, West of Minas, South/Southwest of Minas, Triângulo Mineiro/Alto Paranaíba, Vale do Mucuri, and Zona da Mata), only sporadic cases of the disease were reported during the study period and, consequently, these regions showed low and unsteady VL incidence rates.
VL is expanding in Brazil [2] and in other South American countries [39]. In Argentina [40,41] and Paraguay [42,43] a significant increase in the number of VL cases was observed in the last two decades, raising concerns about the spreading of the disease. Therefore, studies evaluating the process of expansion of VL and the spatial and temporal variation of its incidence are of great importance.
The Metropolitan area of Belo Horizonte mesoregion comprises a large number of municipalities (105) and stands out from the other mesoregions of Minas Gerais for being the most urbanized and the most economically developed, and where the political, financial, commercial, educational, and cultural centers of the state are concentrated [36]. Dissemination and urbanization of VL became even more evident in the Metropolitan region of Belo Horizonte after an autochthonous case of VL was registered in the municipality of Sabará [10]. The results presented herein reveals that since 2002 this municipality has higher, albeit oscillating, VL incidence rates than other municipalities of the same mesoregion considered of high VL transmission risk such as Sarzedo, Jaboticatubas, Sete Lagoas, and Vespasiano. Indeed, the 1990’s saw increasing numbers of VL cases being registered in the Metropolitan area of Belo Horizonte mesoregion and that trend persisted until the early 2000’s [44]. Previous studies in this region pointed out that the VL cases occurred in non-rural areas, reinforcing the urbanization process of the disease [44], which becomes more likely in localities of high population density (Table 1) and with houses close to one another [45]. The present study suggests that this trend persisted at least up to 2008 since map analysis detected a decreasing trend only in the last years analyzed. Indeed, the Poisson regression adjustment indicated an increase in the average number of cases between the years of 2002 and 2008 (5%) and a reduction between 2008 and 2013 (-15%).
Recent studies have shown that VL cases are heterogeneous in the capital Belo Horizonte [16,46]. This may be due to the city’s vast territorial extension, high population density, and different microenvironments [47]. The VLCSP implemented in Belo Horizonte in the 1990’s [48] is considered to take place in a systematic and orderly manner throughout the city [49] and this may explain the reduction in the VL incidence rates observed in the present study. It is possible that the situation of Belo Horizonte reflects in the adjacent municipalities as well as in the mesoregion of the Metropolitan area of Belo Horizonte. This may be due to the population densification in these areas [45,47], which compromises the execution and maintenance of measures to control the disease. Therefore, both Belo Horizonte and its mesoregion presented similar incidence rates, with a gradual increase in the incidence of VL in the first trienniums and a decrease in the last.
Different from the Metropolitan area of Belo Horizonte mesoregion, the North of Minas mesoregion presents more ancient cases of the disease [8]. Nevertheless, research performed in some municipalities of this mesoregion, such as Montes Claros [8,50] and Porteirinha [51] pointed out that the incidence of VL is decreasing in the last years. The authors attributed this decrease to the VL control actions that are being performed in these municipalities. This reduction in the number of cases in this mesoregion is in agreement with the results obtained herein. Accordingly, the results of the Poisson regression adjustment, revealed a reduction in the average number of VL cases in the North of Minas from 2004 onwards (-13%) and the thematic maps, presented similar data to those previously described in Montes Claros in the same period [8]. Indeed, while investigating the VL cases registered in Montes Claros between 2001 and 2007, these authors observed that, from 2005 onwards, the incidence of VL reduced, but the municipality remained endemic for the disease and representing a serious public health problem [8]. Because VL is associated with poor socioeconomic conditions [2,12,31] and that Montes Claros is considered the municipality with the best socioeconomic indexes in this mesoregion, one can assume that the problem regarding VL extends to the whole of the North of Minas mesoregion.
The thematic maps indicated that the Northwest of Minas and Vale do Rio Doce mesoregions had pronounced increase of VL cases during the study period. On the other hand, the Poisson regression showed that, despite the increase observed in some periods, there was a reduction in the incidence rates starting in 2005 in the Northwest of Minas and in 2010 in the Vale do Rio Doce. In 2005, VL expanded to other municipalities in the Northwest of Minas mesoregion. Initially, Unaí and Paracatu stood out as typical examples of VL urbanization [14]. Later, neighboring municipalities, such as João Pinheiro and Brasilândia de Minas, also began to show a high incidence of the disease. Noteworthy, the four mentioned municipalities occupy a vast territorial extension in the mesoregion and kept a cumulative incidence superior to 20/100,000 inhabitants in the last years. The high incidence of VL in the first three periods analyzed in the thematic maps and its reduction during the last period in the mesoregions North of Minas and Northwest of Minas may be explained by their geographical proximity.
The thematic maps generated for the Vale do Rio Doce mesoregion for the period between 2002 and 2004 reveal only five municipalities with cumulative incidence higher than 5/100,000 inhabitants. From 2008 onwards, expansion of VL is observed in this mesoregion with 10 municipalities presenting VL incidence rates higher than 5/100,000 inhabitants, in which stands out Governador Valadares. In this mesoregion, human VL cases started to be reported the 1960’s [9]. At that time, control measures such as the treatment of VL patients with Glucantime, elimination of dogs positive for VL, and use of the insecticidal dichlorodiphenyltrichloroethane–DDT in the domicile and peridomicile areas were established in the municipalities with an expressive number of cases. These measures resulted in a progressive reduction in the number of cases and the absence of VL in the years of 1978 and 1979 [52]. As an unfortunate consequence, disease control measures were interrupted in the 1990’s [18,19]. Since 2008, studies have identified cases of the disease in Governador Valadares [19,20], which may explain the increase in the incidence of VL in the mesoregion of Vale do Rio Doce in the same year and the increase in the average number of cases between 2006 and 2010. The urbanization process that is taking place in the municipality of Governador Valadares and the interruption of the disease control measures previously implemented in this mesoregion [19] may be the reasons underlying the observed increase in the number of cases VL. These results indicate the relevant role of Governador Valadares on the increase in the number of cases of the Vale do Rio Doce mesoregion in the last years analyzed.
The results presented herein showed the geographical expansion of VL in Minas Gerais between 2002 and 2013. The data reveal a trend for VL persisting in municipalities that already presented cases, even when there was oscillation in the disease incidence rates. Despite this expansion, the incidence of the disease does not spread from the North region of the state to the South, that is, to mesoregions such as Campos das Vertentes and South/Southwest of Minas, which presented a consistently low number of cases.
The persistence of the disease in the mesoregions located in the North of Minas Gerais may be related to socioeconomic [5,6,31] and environmental [6] factors. A study performed in Montes Claros showed that VL in this region is associated with poor domiciles and precarious sanitation conditions, which facilitates the accumulation of organic material and other factors that enable the life cycle of the main vector of VL, Lutzomya longipalpis [12]. The maintenance of the disease cycle explained by similar reasons was also observed in the Jequitinhonha (Araçuaí) [53]. The findings reported herein corroborate these previous observations, since the highest VL incidence rates were found in the North of Minas Gerais, a region that presents low Human Development Index [54]. Considering the environmental factors, several authors described what Sherlock (1996) [55] previously observed in the northeastern state of Bahia: that VL is occurring with higher frequencies in hot dry climates common to the northern areas of the Minas Gerais state [12,56,57]. These observations are in agreement with the results presented in the current study since it detected high incidence rates especially in the mesoregions located at the north of the state, which also have a hot and dry climate.
We suggest that VL presence in the northern region of Minas Gerais is mainly due to two reasons: climate and socioeconomic factors. Indeed, we observed that VL is distributed in contiguous mesoregions, namely Northwest of Minas, North of Minas, and Jequitinhonha, all of which have low socioeconomic development and display a warm climate favorable to the development of the vector [1]. On the other hand, the southern mesoregions, namely West of Minas, Campo das Vertentes, Zona da Mata, and South / Southwest of Minas, present lower temperatures and better socioeconomic conditions. Interestingly, although the mesoregion Triangulo Mineiro / Alto Paranaíba has a hot climate, it also presents few VL cases. This mesoregion is the second largest economy in the state and has the highest Human Development Index. The Human Development Index of the Belo Horizonte Metropolitan mesoregion is one the highest in the state of Minas Gerais, but there is a great socioeconomic disparity between the municipalities composing this mesoregion. The high levels of social inequality, as represented by the existence of slums within the municipality of Belo Horizonte itself, may underlie the high number of VL cases observed in this city.
This study has a few limitations that should be pointed out. Even though VL is of compulsory notification in Brazil, the real number of cases may be underestimated as the symptoms of the disease are unspecific, and some cases may go unreported. This might compromise calculations of the incidence and the estimates obtained through the Poisson regression. However, the extent of this underreporting is most probably minimum since SINAN covers all health systems (public and private) at its various levels of complexity. Furthermore, it is noteworthy that the medication used for treatment is solely dispensed by the Brazilian Health Public System and this measure has minimized underreporting.
Some variables of this system, such as “Final Classification” (cases that were confirmed, discarded or had incomplete forms) and “Type of entry” (New Case, Recurrence, Transference, and Ignored) are not filled after closure of the case, thus, increasing the number of losses (missing) and compromise the analyzes. Lastly, areas bordering the state of Minas Gerais were not evaluated. Furthermore, an empiric local Bayesian approach was used in one of the sets of the thematic maps created. This method estimates the mean local incidence in each municipality taking into account the incidence values of the neighboring municipalities. Thus, the rates generated are corrected, smoothed, and less unstable. Maps built using this approach are, therefore, more informative and interpretative.
The spatial and temporal epidemiology, together with the Poisson regression approach, allowed a more precise identification and quantification of areas of expansion and stabilization of VL, and the identification of regions that share similar spatial patterns. The findings reported herein may help to guide the implementation of actions for controlling the disease in the state of Minas Gerais. Given the worrisome expansion of VL in Brazil and in other South American countries, the results describing the spatial and temporal pattern of VL expansion in this vast geographic area may be relevant to researchers following the disease in other regions of Brazil and the world.
|
10.1371/journal.pcbi.1005006 | Genes as Cues of Relatedness and Social Evolution in Heterogeneous Environments | There are many situations where relatives interact while at the same time there is genetic polymorphism in traits influencing survival and reproduction. Examples include cheater-cooperator polymorphism and polymorphic microbial pathogens. Environmental heterogeneity, favoring different traits in nearby habitats, with dispersal between them, is one general reason to expect polymorphism. Currently, there is no formal framework of social evolution that encompasses genetic polymorphism. We develop such a framework, thus integrating theories of social evolution into the evolutionary ecology of heterogeneous environments. We allow for adaptively maintained genetic polymorphism by applying the concept of genetic cues. We analyze a model of social evolution in a two-habitat situation with limited dispersal between habitats, in which the average relatedness at the time of helping and other benefits of helping can differ between habitats. An important result from the analysis is that alleles at a polymorphic locus play the role of genetic cues, in the sense that the presence of a cue allele contains statistical information for an organism about its current environment, including information about relatedness. We show that epistatic modifiers of the cue polymorphism can evolve to make optimal use of the information in the genetic cue, in analogy with a Bayesian decision maker. Another important result is that the genetic linkage between a cue locus and modifier loci influences the evolutionary interest of modifiers, with tighter linkage leading to greater divergence between social traits induced by different cue alleles, and this can be understood in terms of genetic conflict.
| The theory of kin selection explains the evolution of helping when relatives interact. It can be used when individuals in a social group have different sexes, ages or phenotypic qualities, but the theory has not been worked out for situations where there is genetic polymorphism in helping. That kind of polymorphism, for instance cheater-cooperator polymorphism in microbes, has attracted much interest. We include these phenomena into a general framework of social evolution. Our framework is built on the idea of genetic cues, which means that an individual uses its genotype at a polymorphic locus as a statistical predictor of the current social conditions, including the expected relatedness in a social group. We allow for multilocus determination of the phenotype, in the form of modifiers of the effects of the alleles at a cue locus, and we find that there can be genetic conflicts between modifier loci that are tightly linked versus unlinked to the cue locus.
| Traditional theories of social evolution in structured populations use reproductive value to describe the fitness effects of variation in helping and harming traits [1–4]. They are applied to population structures such as the two sexes [1], juveniles and adults [3], dispersers and non-dispersers [5], and high- and low-quality individuals [4]. Individuals can, depending on their state, vary in their phenotype, which corresponds to a reaction norm [4], but genetic polymorphism in social traits is not explicitly included in the theory. Although it is recognized that frequency dependence is compatible with social evolution theory [6], questions of the emergence and maintenance of genetic polymorphism in social traits have not been given full attention. This absence is striking, as the possibility of such genetic polymorphism has attracted much interest. Examples of studies in the laboratory and the field span from work on cheater-cooperator polymorphisms [7–15] to investigations of genetic variation in microbial pathogens [16, 17]. The possibility that population structure contributes to polymorphism also has support [18–22].
It is already well understood that a social trait, such as an individual’s investment in helping, can evolve to different equilibria depending on the relatedness in social groups in different habitats, with more helping in habitats where there is higher relatedness. We use the concept of genetic cues to extend this insight to situations where there is dispersal between habitats and where the social trait is influenced by several, linked or unlinked, genetic loci. The basic idea of genetic cues [23–26] is that alleles can function as statistical predictors of coming selective conditions for an individual. As a consequence of selection, allele frequencies can differ between local environments, such that possessing particular alleles correlates with local conditions in a manner analogous to environmental cues. Using this insight one can integrate genetic polymorphism into theories of conditional phenotype determination.
If the environmental heterogeneity includes characteristics that are important for social evolution, like the size or composition of social groups, the heterogeneity could favor genetic polymorphism in social traits. If so, there will be a correlation between gene frequencies and social characteristics, and genes can act as cues of relatedness. To illustrate this general idea we develop a specific model with two habitats. We show that alleles at a cue locus can provide information about social circumstances, such as within-group relatedness and opportunities for cooperation, and that epistatic modifiers of the phenotypic effects of a genetic polymorphism can evolve to make use of this information. We also show that the evolutionary interests of epistatic modifiers can differ depending on their degree of linkage to a polymorphic locus, and we interpret this phenomenon in terms of genetic conflict.
There are two habitats, each containing a large number of groups. They are formed and dissolved by colonization followed by social interaction and the production of offspring that disperse, and again colonization. A group in habitat i, where i = 1, 2, is founded by Ni haploid individuals, randomly derived from a pool of dispersers in that habitat. To implement variation between habitats in average within-group relatedness, group members reproduce asexually following founding, forming Ni haploid offspring group members, such that each founding group member has an equal and independent chance of producing each of the Ni offspring (model details are given in S1 Text). A smaller Ni thus corresponds to higher relatedness. For a pair of group members, the probability of being identical by descent since founding is
r i = 1 N i , (1)
which follows [27] and [28]. The offspring group members engage in a social interaction, for instance a public goods game [29], and produce dispersing offspring in proportion to the payoff in the game. An individual’s phenotype z represents an investment (strategy) in the game, and we assume 0 ≤ z ≤ 1. The payoff to an individual with phenotype z in habitat i is a function w i ( z , z ¯ ) of z and the average investment z ¯ of the individual’s group. As a convenient example we will use w i ( z , z ¯ ) = W i + b i z ¯ - c i z 2, where the benefit b i z ¯ is proportional to the average investment and the cost ci z2 is assumed to increase quadratically with the individual’s investment. For polymorphic populations the group compositions will vary, and we are particularly interested in the expected payoff in habitat i to a randomly chosen rare mutant player of the game with phenotype z′, in a population where the resident phenotypes z1 and z2 occur with frequencies pi1 and pi2 (where pi1 + pi2 = 1). We write this as
w ¯ i ′ = E [ w i ( z ′ , z ¯ ) | z 1 , z 2 , p i 1 , p i 2 ] . (2)
Because a new group is founded by random dispersers, those groups containing mutant strategies will predominantly be founded by one mutant and Ni − 1 resident types. Some basic aspects of the model are illustrated in Fig 1.
To study the invasion of mutant traits, we need the derivative of the expected mutant payoff, which we write as
d i k = ∂ w ¯ i ′ ∂ z ′ z ′ = z k = b i N i 1 + ( N i - 1 ) r i - 2 c i z k , (3)
for habitat i and phenotype zk, i = 1, 2, k = 1, 2. We study evolutionary change of a dimorphism z1, z2 by examining the invasion of mutant modifiers. Let x1 and x2 denote two alleles at the cue locus. In the resident population, the genetic cue xk induces the phenotype zk, and nik is the number of individuals in habitat i with phenotype zk at a population dynamical equilibrium. The epistatic effect of a mutant modifier is that xk instead induces the phenotype z k ′. Letting n i k ′ denote the (small) number of mutant modifiers in habitat i with phenotype z k ′ (i.e., linked to cue allele xk), we can write down a population projection matrix for the mutant invasion. The invasion fitness of the mutant modifier is
F ( z 1 ′ , z 2 ′ ; z 1 , z 2 ) = log λ , (4)
where λ is the leading eigenvalue of the population projection matrix. Here we give an overview of the derivation of this matrix (details are given in S1 Text).
For simplicity, we assume that individuals are haploid over most of the life cycle. However, to explore the consequences of recombination between cue and modifier loci, we introduce sexual reproduction by assuming there is a brief sexual phase in the dispersal pool in a habitat. This involves diploid individuals and crossing over, with a recombination rate ρ between cue and modifier loci, to produce the haploid individuals that found the groups as described above. Mating is random with respect to the dispersal pool and occurs before the forming of groups in the habitat. As a census point, we specify the population composition at a time after the sexual phase, when groups have formed and the public goods game is about to start. The sequence of events in the life cycle, starting right after the census point, is as follows: (i) public goods game with offspring production in proportion to payoff, (ii) within- and between-habitat migration of these offspring, forming a dispersal pool in each habitat, (iii) mating and recombination, and (iv) the next episode of group formation, including one asexual generation. By putting these events together, one can write down the matrix (see S1 Text). Using the population dynamics we can also determine the region of coexistence of two phenotypes z1 and z2 for different sets of parameters, by determining when each phenotype can invade a monomorphism of the other (the condition is given in equation (S16) in S1 Text).
We compute a selection gradient from the invasion fitness Eq (4) using standard methodology of matrix population models [30]. Because we average over the group compositions Eq (2), our analysis is consistent with the structured population approach to adaptive dynamics [31], and it can also be seen as a direct fitness methodology for social evolution theory [1, 3], also referred to as a personal fitness methodology [6].
In order to check our analytical results, and to illustrate the effects of genetic conflict between cue and modifier loci, we have run individual-based evolutionary simulations corresponding to our model assumptions. As a genotype-phenotype mapping in these simulations, we used a sigmoid function
z = 1 1 + exp - a 0 - a g x (5)
of a ‘liability’ a0 + agx, where x is the effect of an allele at the genetic cue locus, and a0 and ag are parameters that are genetically determined by modifier loci (details are given in S1 Text).
To compute evolutionary equilibria numerically, we developed a C++ program that follows a path of small steps through z1 z2–space, each of which increase the invasion fitness Eq (4), until reaching an equilibrium. We used the Eigen C++ library [32] to compute eigenvalues. For the individual-based evolutionary simulations, we developed C++ programs that directly implemented the sequence of events in the life cycle, using pseudo-random numbers to handle stochastic events, such as recombination and mutation. In the simulations, we used a total populations size of 40 000 and time periods of 40 000 full life cycles or more.
We use the methodology of adaptive dynamics and matrix population modeling [30, 33] to compute the derivative of invasion fitness for a mutant modifier. The details of the derivation are given on pp. 8–10 of S1 Text, and here we focus on the interpretations in terms of information in a cue. The genetic cue provides information to an individual about its current habitat. The prior probability of being in habitat i is qi = ni/(n1 + n2), where ni = ni1 + ni2 is the number of individuals in habitat i at a population dynamical equilibrium and nik is the number of individuals in habitat i with phenotype zk. For an allele at a modifier locus, the probability of being in habitat i, conditional on being linked to allele xk at the cue locus is
q i k = p i k q i p 1 k q 1 + p 2 k q 2 = n i k n 1 k + n 2 k , (6)
where pik = nik/ni. The selection gradient is the derivative of invasion fitness Eq (4) with respect to mutant traits, and can be written
∂ F ∂ z k ′ z k ′ = z k = V 1 k d 1 k p k q 1 k + V 2 k d 2 k p k q 2 k . (7)
To interpret this expression, note that q1k and q2k are the respective probabilities of being in habitat 1 or 2, conditional on being linked to cue allele xk. The factor pk = (n1k + n2k)/(n1 + n2) is a ‘dilution factor’ that appears because the mutant z k ′ is only expressed in individuals with cue allele xk. The d1k and d2k are the derivatives of the expected payoff Eq (2) in habitats 1 and 2 with respect to the mutant trait, and are given in Eq (3). Finally Vik is the reproductive value of an offspring of a player in habitat i with cue allele xk. From the manner in which the conditional probability qik appears in the expression, we can conclude that the selection gradient describes changes in payoff to a ‘Bayesian decision maker at the modifier locus’. Eq (7) is an extension of the direct fitness approach of social evolution theory to situations with genetic polymorphism at the cue locus. Note that this selection gradient refers to the invasion of mutant modifiers, and not to the invasion of alleles at the cue locus, except for the special case of full linkage (ρ = 0), for which cue and modifier form a unit.
Completing the life cycle, through migration, mating and recombination, and group formation, we can express Vik in terms of reproductive values vjl at our census point:
V i k = v 11 ϕ 1 h 11 k m 1 i + v 21 ϕ 2 h 21 k m 2 i + v 12 ϕ 1 h 12 k m 1 i + v 22 ϕ 2 h 22 k m 2 i . (8)
Here, mji is the rate of migration from habitat i to j. The ‘cue inheritance’ is described by
h j l k = ( 1 - ρ ) δ l k + ρ p j l , (9)
so that with probability 1 − ρ the cue allele is passed to offspring and with probability ρ the offspring receives its cue allele through recombination with a random individual in the dispersal pool. Finally, ϕj is the probability for an individual in the dispersal pool in habitat j to become a founding group member.
We must also examine whether or not polymorphism can be maintained at the cue locus. This needs to be investigated as a separate question, by determining when each of the phenotypes z1 and z2 can invade a monomorphism of the other. The condition for this is given in equation (S16) in S1 Text.
Fig 2 shows how the migration rate m between habitats and the recombination rate ρ between cue and modifier loci influence dimorphic evolutionary equilibria, i.e. phenotypes where the selection gradient Eq (7) vanishes. The blue and red curves indicate phenotypes z1 and z2 suited to habitats with low and high relatedness. The selection gradient is illustrated in Fig 3 for a few values of m and ρ, and the shaded regions in this figure show where a polymorphism at the cue locus is maintained. In this example, the only difference between habitats is the number of founders of a social group, with N1 = 20 in habitat 1 and N2 = 2 in habitat 2, so it is appropriate to interpret the genetic cue as a cue of relatedness.
As seen in Fig 2, there is an interaction between the migration rate and the recombination rate, such that for very low migration rate (m = 0.01) the recombination rate has little influence on the equilibrium dimorphism, whereas for a higher migration rate (m = 0.10) the difference between z1 and z2 varies considerably from tight linkage to free recombination. For even higher rates of between-habitat migration, genetic polymorphism is not maintained at the cue locus, regardless of the cue-modifier recombination rate ρ, and the outcome is instead a monomorphism. For the parameter values in Fig 2, this happens for m = 0.15 or higher.
The divergence between z1 and z2 depends on ρ, as in Figs 2 and 3, because modifier alleles with different linkage to cue alleles have different demographic futures, and thus different evolutionary interests. A fully linked mutant modifier will remain more concentrated in one of the habitats, which tends to favor specialization to that habitat, whereas an unlinked one will fairly quickly become evenly distributed over cue alleles and habitats, which tends to favor less specialized phenotypes. This difference in evolutionary interest between modifiers follows the logic of genetic conflicts [34], in the sense that the invasion of a loosely linked modifier, reducing the divergence between phenotypes, creates the context for the invasion of a more tightly linked modifier that reverses this effect. The outcome of genetic conflicts can depend on such things as the of the availability of mutations, the genetic architecture of a trait, and the strength of selection.
For modifiers of polymorphic effects, genetic conflicts can have the further consequence of changing selection acting on the additive effects of alleles at a locus from stabilizing to disruptive, potentially giving rise to selectively maintained polymorphism at that locus. For instance, for the case of m = 0.10 in Fig 2, unlinked modifiers favor a very small divergence between z1 and z2, but once this outcome has been achieved, the selection on alleles at other loci with additive effects on z becomes disruptive (just as originally for the cue locus itself). Genetic polymorphism might then be transferred from an original cue locus to a new locus.
How this can happen is illustrated by the individual-based simulations in Fig 4. The genotype-phenotype mapping Eq (5) from the genetic cue x to the trait z has been changed from that in Fig 2, where the parameter a0 was fixed, to one where both parameters a0 and ag are genetically determined and can evolve. In Fig 4A, a0 and ag are each determined by a single locus, either fully linked or unlinked to each other and to the cue locus. When m is small or when ρ = 0, the outcome of the individual-based simulations remains in agreement with the predictions from the selection gradient Eq (7), but for m = 0.10 and ρ = 0.5, the outcome is instead the same as that for m = 0.10 and ρ = 0 (Fig 4B). The reason is that, starting with polymorphism at the cue locus, ag evolved to become small, reducing the divergence between the phenotypes from Eq (5), which in turn gave rise to disruptive selection on a0, causing polymorphism to evolve at that locus, while the polymorphism at the original cue locus collapsed. The end result was that the locus coding for a0 became a polymorphic cue locus, with phenotypes z1, z2 in accordance with the evolutionary interests of fully linked modifiers of this new polymorphism (Fig 4B). Other conceivable evolutionary outcomes of disruptive selection on a0 are shown in Fig 4C and 4D.
In Fig 4C, 5 unlinked loci have small positive effects on a0 and 5 have small negative effects, and each of these loci became polymorphic in the simulation, while at the same time the original genetic cue locus remained polymorphic. The overall effect was a fairly broad distribution of values for the investment z. In Fig 4D, the maximum expression at the loci with positive and negative effects was controlled by two separate unlinked loci, and one of these became polymorphic, giving rise to a bimodal distribution of values of z. In these examples, a notable amount of genetic variation in z evolved, but the width and shape of the distribution of z depended on the details of the genetic architecture of the trait. In all cases, an individual gains information about its current habitat from its genotype, and one can show that the clearcut polymorphism in Fig 4B is the most informative, with progressively less information on average in Fig 4C and 4D, as illustrated in Fig 5. The latter cases are intermediate between the evolutionary interests of fully linked and unlinked modifiers.
We have shown how adaptively maintained genetic polymorphism can be integrated into social evolution theory by making use of the concept of genetic cues. The selection gradient we derived (in eq (7)) parallels the direct, or personal, fitness approach to social evolution in class-structured populations [3, 6], with the distinction that the presence of a cue allele in an individual’s genotype, rather than a phenotypic state, defines the class structure. In our model, individuals use strategies that are conditional on a genetic cue, but our general approach can incorporate a combination of genetic, environmental and transgenerational cues [26, 35].
Just as is the case in standard social-evolution theory, relatedness enters into our model as a description of the genetic structure of social groups. The structure refers to genetic variation at epistatic modifier loci, rather than at genetic cue loci, so the relatedness parameter ri in the pay-off derivative Eq (3) refers to rare mutants at a modifier locus. This is in accordance with the general idea of treating genetic variation at a cue locus as input to a developmental or ‘decision-making’ system, and then to examine long-term evolution of the developmental system [24, 26]. The value of this perspective is that it guides the analysis and interpretation by forming a link to the study of conditional strategies, such as the study of phenotypic plasticity.
The different ways in which individuals gain information about themselves and their social partners has figured importantly in the study of social evolution [2]. For instance, migrant individuals arriving in a local population have different expectations of relatedness to their neighbors than non-dispersers [5]. The possibility that individuals can recognize kin through similarity in genetically polymorphic traits has been carefully investigated, with the conclusion that kin recognition can evolve in spatially structured populations [36–38]. Yet another possibility is that individuals estimate their degree of inbreeding, and thus how likely they are to be related to their neighbors, using their relative homozygosity as an indicator of local relatedness [39]. Our analysis of genetic cues of relatedness contributes to this general emphasis on the role of information, but differs from the other examples through an affinity to the study of local adaptation in the face of gene flow, which has been given much attention in evolutionary ecology but rather little in theories of social evolution.
We found that the genetic linkage between cue and modifier loci can influence the evolutionary outcome (Figs 2 and 3), and this gives rise to genetic conflicts. Genes unlinked to a genetic cue locus tend to favor phenotypes that are less specialized to particular habitats compared to tightly linked genes, because unlinked genes become adapted to exist in all habitats, be transferred between them, and to use the information in a genetic cue to adjust the phenotype in an optimal way for this situation. Tightly linked genes, on the other hand, might be selected to perform well in only one of the habitats, even at the expense of performance in another habitat. The reason is that a modifier allele tightly linked to a cue locus allele can become concentrated to one of the habitats, with the other habitat acting as a sink, to which little adaptation takes place [40, 41].
Our results on the role of genetic conflicts in giving rise to disruptive selection at modifier loci (Fig 4) extends the previous understanding of genetic conflicts when there is adaptively maintained genetic polymorphism [24]. Disruptive selection in heterogeneous environments can maintain genetic polymorphism [42], and genetic cue polymorphism is an example of this general phenomenon. So, if unlinked or loosely linked modifiers of a genetically polymorphic locus evolve to reduce or eliminate the divergence between phenotypes, there will be disruptive selection at loci with additive effects on the phenotype in question. Theoretical modeling has found that disruptive selection tends to favor genetic architectures where polymorphism is concentrated to a single locus [23, 43], but as we have shown, constraints on the set of genotype-phenotype mappings can lead to intermediate outcomes between a single-locus polymorphism and polygenic variation where each locus has a small effect (Fig 4).
A basic question for evolutionary theory is whether evolutionary change can be seen as optimizing some form of fitness for the organism or individual [44]. Our analysis of genetic conflicts throws new light on this issue. It is reasonable to regard unlinked modifiers of effects at a polymorphic locus as representing the evolutionary interest of the organism, because unlinked, small-effect mutant modifiers share their demographic future with the organism. Our results in Fig 4—that disruptive selection can act to diminish the control exercised by unlinked modifiers over the degree of phenotypic specialization—illustrate how individual optimization might be circumvented when there is genetic polymorphism. Furthermore, our individual-based simulations with multilocus genetic architectures resulted in evolutionary outcomes that were intermediate between the evolutionary interests of linked and unlinked modifiers (Figs 4C, 4D, 5C and 5D). This fits with the general idea of the organism as a compromise between different evolutionary interests [45, 46].
In conclusion, our framework broadens the scope of social evolution theory, by accounting for adaptively maintained genetic variation in heterogeneous environments and by incorporating evolutionary outcomes over the range from genetic specialism to generalism. Many instances of interactions between relatives in nature are likely to be found somewhere between the extremes of such a spectrum. A major insight from our work is that positions along this spectrum can correspond to the degree of genetic linkage between polymorphic loci and epistatic modifiers of the phenotype in question. Our analysis thus delivers potentially testable predictions about the evolution of epistasis between modifiers and polymorphic loci and can inspire empirical investigation of the importance of genetic cues of relatedness.
|
10.1371/journal.pgen.1002331 | Genetic Rearrangements Can Modify Chromatin Features at Epialleles | Analogous to genetically distinct alleles, epialleles represent heritable states of different gene expression from sequence-identical genes. Alleles and epialleles both contribute to phenotypic heterogeneity. While alleles originate from mutation and recombination, the source of epialleles is less well understood. We analyze active and inactive epialleles that were found at a transgenic insert with a selectable marker gene in Arabidopsis. Both converse expression states are stably transmitted to progeny. The silent epiallele was previously shown to change its state upon loss-of-function of trans-acting regulators and drug treatments. We analyzed the composition of the epialleles, their chromatin features, their nuclear localization, transcripts, and homologous small RNA. After mutagenesis by T-DNA transformation of plants carrying the silent epiallele, we found new active alleles. These switches were associated with different, larger or smaller, and non-overlapping deletions or rearrangements in the 3′ regions of the epiallele. These cis-mutations caused different degrees of gene expression stability depending on the nature of the sequence alteration, the consequences for transcription and transcripts, and the resulting chromatin organization upstream. This illustrates a tight dependence of epigenetic regulation on local structures and indicates that sequence alterations can cause epigenetic changes at some distance in regions not directly affected by the mutation. Similar effects may also be involved in gene expression and chromatin changes in the vicinity of transposon insertions or excisions, recombination events, or DNA repair processes and could contribute to the origin of new epialleles.
| In contrast to alleles, epialleles have identical DNA sequence and differ only in gene expression and chromatin features. Epialleles are heritable and can also contribute to phenotypes. How this variation originates is unclear. In this study, we analyzed two epialleles found in Arabidopsis for the difference between their chromatin features and their potential to change state. We mutagenized plants with the inactive epiallele and recovered mutants with restored gene expression. In several cases, this was connected with different rearrangements downstream of the epiallele that caused a switch of the epigenetic configuration further upstream. Therefore, sequence alterations, for example by transposon activity or recombination events, may trigger similar heritable changes of chromatin and gene expression in their proximity and could create new epialleles.
| Epialleles are heritable states of different gene expression from sequence-identical genes and have been described in several organisms [1]–[3]. Like genetically different alleles, epialleles contribute to phenotypic heterogeneity [4]–[5]. While the mutagenic processes creating DNA sequence allele variations are relatively well understood, little is known about how and when epialleles originate, and it is difficult to investigate this in statu nascendi. In plants, epialleles were described as natural variants [6]–[9], mutation-induced [10]–[12], or associated with tissue-culture [13]–[15]. Once established, epialleles can acquire stability over many generations; however, they have much higher reversion rates than genetic alleles. Therefore, analyzing the switch from one epigenetic state to the other at well-characterized epialleles can provide insight into their natural origin.
Pairs of epialleles are characterized by antithetic histone modifications at the associated nucleosomes, transcriptional activity at the expressed form, and transcriptional gene silencing (TGS) at the other. In some fungi, mammals, and higher plants, the latter is connected with cytosine methylation at the epiallele [e.g. 6], [16]–[17]. Several pairs of epialleles in plants define easily scored phenotypes like morphology [6], [10], development [9], pigmentation [7], [18], or reporter gene expression [19]–[20]. Some epialleles, as well as many other epigenetically controlled genes, have been used for mutant screens and have helped identify many different proteins and RNAs whose presence or absence can cause transient or stable changes of epiallele expression, or influence epigenetic regulation in general. There is also a wealth of data on the influence of drug treatments, sequence determinants, and the role of genomic neighborhood, on epigenetic regulation.
Arabidopsis thaliana has been the plant model of choice for genetic analysis of switching between epiallelic states, based on the rich genetic and genomic resources available. The experimental system in our study is based on a pair of epialleles in Arabidopsis thaliana containing either an expressed or silent hygromycin phosphotransferase gene (HPT). Active transcription confers resistance to the antibiotic while the inactive epiallele renders the plant sensitive. Gene expression can be selected for on antibiotic-containing medium but does not affect the plants during non-selective growth. The epialleles were found in tetraploid plants obtained by regeneration from protoplasts [20]. While some lines had resistant progeny and expressed the HPT gene, other lines had silenced the HPT and produced only sensitive progeny. The R and S epialleles (determining resistance and sensitivity on hygromycin, respectively) were maintained in their particular expression state after diploidization and for all generations of self-pollination analyzed so far (Figure S1). Beside their differences in transcription, they also differ in DNA methylation [21]. We screened for a switch between the epialleles, by scoring for restored hygromycin resistance after T-DNA mutagenesis of the diploid S line. We identified two trans-acting factors whose nature indicated an epigenetic ‘double lock’ at the silent epiallele [22]. In contrast to many other silent genes, silencing could only be released by simultaneous interference with methylation of DNA and histones. Six mutations from the same screen were mapped to the resistance gene itself. These cis-mutations provided the opportunity to study the nature and effect of DNA sequence changes on gene expression, chromatin organization, and genetic stability. We describe these new alleles in detail and compare them with the R and S epialleles. We show that different, and non-overlapping, sequence changes downstream of the HPT gene can restore the expression of the upstream promoter, to a similar extent as the mutations interfering with the chromatin factors in trans. Such small sequence alterations that cause epigenetic changes at some distance may also be involved in gene expression and chromatin changes in the vicinity of transposon insertions/excisions, recombination events, or DNA repair processes and may thereby contribute to the origin of new epialleles.
The HPT gene is inserted in an AT-rich intergenic region on Arabidopsis thaliana chromosome 3 [20]. Previous investigations, and published data from genome-wide screens for chromatin features [20], [23]–[24], indicated that the genomic localization itself is unlikely to influence the epigenetic state of the HPT gene, as no prominent epigenetic modifications are present in the neighborhood of the insertion. Resistant and sensitive Arabidopsis lines with the different epialleles had been generated from the same progenitor line homozygous for the HPT gene, thereby being supposedly isogenic. Nevertheless, the lack of transcription initiation in the hygromycin-sensitive lines could have been due to a DNA sequence mutation in a regulatory region, for example, a transcription factor binding site. Also, the structure of the insert had not been analyzed in detail. Therefore, active and inactive versions were amplified from genomic DNA of the respective lines. Both epialleles are potentially fully functional and have identical sequences. The 35S promoter (P1) is flanked upstream by a 661 bp fragment derived from the plasmid vector (V1). A rearrangement between two vector molecules prior to, or during, the integration of the transgene into the plant genome caused a duplication of the adjacent vector sequence (V2) and the 35S promoter (P2), resulting in two tandem repeats (Figure 1A). The polyadenylation signal from the CaMV 35S terminator following the HPT ORF lacks 151 bp compared to the transformation construct and has therefore lost its termination function (ΔT), causing read through of the P1 transcript into the flanking plant genome sequence (Figure 1A). P2 is followed by a 505 bp non-protein coding fragment (NC) harboring sequences of bovine carrier DNA used to assist PEG-mediated direct gene transfer to mesophyll protoplasts [25], interspersed with 54 nucleotides without homology to known sequences. This heterologous DNA is transcribed by P2, giving rise to a smaller non-coding transcript (P2 transcript) (Figure 1A). Resistant plants produce the longer P1 and the shorter P2 transcripts, while both promoters are inactive in sensitive plants (Figure 1B and Figure S6). Therefore, the isogenic inserts differ only by gene expression, and R and S represent true epialleles.
The different expression states were suspected to originate from distinct chromatin configuration, and previous studies had provided evidence for opposing DNA methylation at the epialleles, especially pronounced at the transcription factor binding sites ([20]–[21], Figure 1C). As DNA methylation and silencing are usually correlated with specific changes of the DNA-associated proteins, we investigated histone modifications and nucleosome occupancy at the epialleles by chromatin immunoprecipitation. This revealed significant differences between the epialleles along the whole transgenic insert. While expressing lines (R) were primarily marked by trimethylation of histone H3 at lysine residue 4 (H3K4me3), typically enriched in euchromatic regions, epialleles in silenced lines (S) have nucleosomes with a modification characteristic of heterochromatin, namely dimethylated lysines at position 9 (H3K9me2) (Figure 1D). These marks, also including low levels of H3 dimethylated at position 27 (H3K27me2), only extend a short distance from the transgene into the flanking plant DNA (Figure S2), indicating limited spreading in transcriptional direction. Beside the specific modifications, we also observed an overall reduced association with H3 in line R compared to S (Figure 1E), probably rendering the promoters more accessible for the transcription machinery. While the epialleles clearly differed in their local chromatin configuration, this did not have any effect on their nuclear localization (Figure S3).
Both epialleles were stably inherited over a minimum of eight generations of self-pollination, without any evidence for spontaneous switches in the germ line. To also study the stability of epialleles in undifferentiated cells, we initiated callus cultures, starting with cotyledons of resistant, sensitive, and non-transgenic plants, and propagated the calli for up to six months under non-selective conditions. We screened callus tissue at several time points for its ability to grow under hygromycin selection for up to 5 weeks. Calli derived from R lines were resistant whereas calli obtained from S or non-transgenic lines died on selection plates. We also determined chromatin modifications and DNA methylation in callus tissue grown on non-selective medium, with results comparable to those of leaf tissue (Figure S4). This demonstrates similar states and stable maintenance of epialleles even upon dedifferentiation.
We screened for the involvement of antisense and/or small RNAs in silencing maintenance. Significant promoter activity of the NC region was excluded (Figure S5A), and specific antisense RNA in line S could also not be detected, neither by northern blotting (Figure S5B) nor by RT-PCR (data not shown). Nevertheless, we generated libraries from size-fractionated 19 nt to 26 nt RNAs prepared from flower buds of plants containing either the sensitive or resistant epiallele. Both libraries were sequenced (Table S1) and the reads screened for alignment with the transgenic insert. The library from the R plants had only 59 reads (3 per 1 million reads) with only one sequence with a match in the epiallele (Figure 2A, Table S3). In line S, we found 2661 (129 per 1 million reads) matching the epiallele, with a predominant length of 24 nucleotides (Figure 2A, Table S2 and Table S3), the size class known to be primarily involved in RNA-directed DNA methylation (RdDM). This is significantly more than in R, but still relatively little, compared to an individual miRNA (820 reads per 1 million for miRNA165) or to siRNA from a repetitive sequence (>1000 reads per 1 million for TSI [26]). The reads in S were distributed along the epiallele but mostly outside the HPT coding region. Importantly, among all reads specific for the silent epiallele we found an sRNA peak (671 reads, 476 antisense and 195 sense) covering 61 bp in the middle of the 505 bp non-coding sequence of the P2 transcript (Figure 2B). The most abundant sRNAs overlap with the 54 nucleotides of unknown origin. However, this sequence encompasses 28 nucleotides that are homologous to the most 5′ end of the 35S promoter (Figure 2B).
In short, these results indicate very stable and completely isogenic epialleles that differ only in their transcriptional activity. DNA methylation, suppressing chromatin marks, and sRNAs, are specifically enriched at the transcriptionally inactive epiallele; while the counterpart produces high transcript levels, lacks DNA methylation and sRNAs, and carries modifications characteristic of open chromatin (Figure 2C).
In addition to the trans-acting mutants identified in a screen for restored HPT expression after mutagenesis of line S [22], we identified six hygromycin-resistant plants in which the mutant phenotype was genetically linked to the resistance gene itself (‘cis-mutations’, RΔ1-6). All these mutants produced progeny that could grow on hygromycin selection plates (Figure 3A), connected with restoration of variable amounts of P1 and P2 transcripts (Figure 3B). Northern blot analysis of cis-mutant RNA revealed P1 transcripts of smaller size in all cis-mutants compared to those from the active R line (Figure 3C). The length is reduced to different extents, indicating several independent mutational changes of the sequence. An extended northern blot analysis, with either total RNA or poly(A)-enriched RNA, showed that the P1 transcript in all lines besides RΔ6 is polyadenylated (Figure S6), likely due to a flanking sequence with some similarity to a polyA signal. While no P2 transcript from the second promoter is detectable in RΔ1, RΔ2, RΔ4, and RΔ6, there is a signal in RΔ3 and RΔ5, including in the poly(A) fraction (Figure S6C, S6D).
To characterize the P1 transcripts, and to identify the transcriptional termination sites in the cis-mutants, we performed 3′-RACE. We also analyzed the genomic DNA of all cis-mutants after amplification of the transgenic insert from genomic DNA and aligned DNA and RNA sequences (Figure 3D). This verified six different sequence rearrangements within the 3′ region: mainly deletions, but also one case of an inserted plant DNA fragment (RΔ3). The mutants RΔ1 and RΔ2 have both lost the duplicated promoter P2 and the NC sequence. The vector duplication was partially (RΔ1) or completely (RΔ2) deleted, as was part of the flanking plant sequence. The deletions in RΔ4, RΔ5, and RΔ6 did not or only partially affect the P2 promoter, and two of them maintain also the NC sequence. The rearrangement in RΔ3 is most complex: here, a 1243 bp plant DNA sequence derived from a position 1.2 kb upstream of the transgene location was inserted between the P1 transcript and the downstream vector fragment. In the mutants RΔ1, RΔ2, RΔ3, and RΔ4, the P1 transcripts are terminated at the (first) site of rearrangement, while the transcripts go beyond the breakpoints in RΔ5 and RΔ6. Only RΔ3 and RΔ5 are able to produce the P2 transcript, as in these cases, the P2 promoter is complete and the heterologous sequence downstream was only slightly affected by mutagenesis (Figure 3D). Nevertheless, the P2 transcript levels are much lower than in the R line (Figure 3B). Interestingly, there is no overlap between the deletions in all individual cis-mutants, but the rearrangements had either affected the second promoter copy (RΔ1, RΔ2, RΔ6), or the DNA template for the P2 transcript (RΔ1, RΔ2 and RΔ4), or the connection between both sequences (RΔ3, RΔ5).
All cis-mutants were tested for effects outside of the epiallele by analyzing the degree of genome-wide methylation at endogenous repeats and by introgressing a transcriptionally silent marker gene coding for β-glucuronidase from line L5, shown to be affected by other epigenetic mutations [27]–[28]. None of the cis-mutants changed the modification or expression of these markers (Figure S7). Therefore, it is unlikely that they have an effect outside of the epiallele.
Due to the hygromycin selection in the screen, all cis-mutants were expected to have a functional resistance marker gene. Indeed, the upstream promoter P1 and the HPT coding region were intact and identical in RΔ1-6 and hence potential new epialleles of the resistance gene. Therefore, we compared the chromatin state in this region. We found reduced DNA methylation levels in cis-mutants compared to S (Figure 4A), and a detailed bisulfite methylation analysis confirmed an overall reduction of DNA methylation in the promoter region of cis-mutants (Figure 4B, 4C). However, the degree of hypomethylation, and the distribution of the remaining methylated cytosine residues, do not support a direct and linear correlation with expression levels. Although RΔ2, RΔ3, and RΔ4 show the strongest reduction of CG methylation, especially at the transcription factor binding sites (Figure 4B, asterisk), and have expression levels comparable to R (Figure 3B), methylation in RΔ5 is similar to RΔ3 and RΔ4, although P1 transcript expression is much lower. Also, RΔ3 and RΔ4 have even gained CHH methylation in the 5′ region. Concomitant with the loss of DNA methylation, the modification specific for the silent state (H3K9me2) was changed in favor of the active mark (H3K4me3) in P1 and P1-transcribed regions, as demonstrated by ChIP (Figure 4D). One mutant (RΔ1) maintained a high level of H3K9me2 similar to that of the silent epiallele. Nonetheless, it also acquired a remarkable amount of H3K4me3, although less than other cis-mutants. Independent of the modifications, and similar to the resistant line, cis-mutants showed a decreased level of H3 association, indicating that the sequence rearrangements had also affected the nucleosome density (Figure 4E).
On the whole, the cis-mutants demonstrate that structural rearrangements can cause significant changes in transcriptional activation and chromatin configuration at the previously silent epiallele. These changes are surprisingly divergent and reflect specific effects of similar but not overlapping deletions.
The extreme stability of R and S epialleles through many generations and in callus cultures raised the question of expression stability in the cis-mutants. Most structurally rearranged derivatives displayed similar stability and provided comparable hygromycin resistance over several generations of homozygous cis-mutants (S4 to S6 tested). RΔ2, RΔ3, and RΔ4 produced resistant progeny in consecutive generations. Resistance in RΔ5 and RΔ6 was lower in S4 (56% and 61%, respectively), but maintained this level up to S6. In contrast, RΔ1 plants that were clearly hygromycin-resistant in S4 (84%) generated partially sensitive S5 and fully sensitive S6 progeny (Figure 5A). This correlates well with the loss of unmethylated sites at the transgenic insert (Figure 5B), similar to gradual loss of resistance over 5 generations described for another marker gene [29]. The instability in RΔ1 does not correspond with additional sequence changes, as the same rearranged structure (Figure 3D) is maintained in subsequent generations. Rather, it correlates with the epigenetic state, since RΔ1 was characterized by the bivalent histone modifications (Figure 4D).
The re-silencing in generation S6 of RΔ1 allowed us to compare silencing maintenance at promoter 1 between this line and the S epiallele. We tested plants of both lines after growth in the presence of zebularine [reducing DNA methylation, 30] or DZNep [reducing histone methylation and also DNA methylation via SAHH]-[inhibition, 22,31]. Zebularine alone did not reactivate promoter P1 in line S, but in RΔ1S6, and DZNep-induced activation was twice as high in RΔ1S6 compared to S (Figure 5C). This indicates that S and RΔ1S6 differ in the stringency of silencing, either due to presence or absence of the P2 promoter and transcript, or to the lineage history of RΔ1S6 from a recently active state. The presence of the P2 promoter in RΔ3 - 6 and the expression of the P2 transcript in RΔ3 and 5, which do not cause re-silencing in later generations, make the latter explanation more likely.
The thorough analysis of the HPT transgene in its two opposite expression states has revealed sequence identity over the full length of the insertion, significant differences in chromatin modifications and few, but silencing-specific, small RNA molecules. Chromatin differences are restricted to the affected sequence, with no hint of genome-wide changes or modified localization of the genomic region within the nucleus. Together with heritability of the expression states over many generations, and their maintenance even upon de-differentiation, the data prove the transcriptionally active and the silenced version to be authentic epialleles. Their occurrence in Arabidopsis, the best studied model for epigenetic research in plants, and the easy assay for the selectable hygromycin resistance conferred by the active state, made this pair of epialleles convenient tools for studying maintenance and switching of epigenetic states.
After mutagenesis, we identified several hygromycin-resistant plants in which mutations in the epiallele sequence downstream of the HPT coding region had reactivated the previously silenced epiallele. Combining DNA and RNA sequence analysis and characterization of chromatin modifications, we found that these structural changes of the DNA sequence caused substantial upstream changes in chromatin and transcriptional activity. Beyond the complex and mutually dependent interplay of chemical modifications of the DNA and the associated histones, and longer and small, coding and non-coding RNAs described in numerous cases, the results presented here have shown that even small and non-overlapping modifications of the genomic template, outside of the promoter and open reading frame, can modify transcription and chromatin states in the vicinity. These changes are not minor: the bacterial gene HPT coding for hygromycin phosphotransferase is a selectable marker gene applied in numerous plant transformation experiments [32], but plants need a significant amount of HPT transcript to produce enough protein to detoxify the antibiotic. Minor reactivation in the background of some epigenetic mutants tested in a reverse genetic approach (data not shown) was not sufficient. Therefore, the stringent assay for restored hygromycin resistance required a substantial change, as in the case of the trans-acting mutants from the same screen that revealed a double lock of two simultaneous chromatin modifications [22]. HPT expression levels are indeed similar between cis-and trans-acting mutants.
Although the transgenic marker allowed this convenient selection for drastic epigenetic switches, without affecting plants under non-selective conditions, it could have been considered not representative for other, plant-endogenous or general cases. However, a recent publication [33] describes an interesting mutation that affects expression of the gene for nodulation factor SUNN in Medicago truncatula. The mutation is closely linked to the SUNN gene, acts only in cis but does not change the DNA sequence of the SUNN gene itself. Although the nature of this mutation is not yet identified, it could exert its effect in a similar way to the cis-mutants described here, especially since the ‘like sunn supernodulator’ mutant phenotype is occasionally unstable, like the hygromycin resistance in RΔ1, 5, and 6. Other examples may be found upon further inspection of natural transcript level variation between regions with very similar gene sequences in plants [e.g. 8] or in the connection between chromatin structure and trinucleotide repeat expansion in mammals [for review 34].
Transcriptional gene silencing is often associated with the presence of homologous sequences in the genome [e.g. 35]–[37], and intentional rearrangements from complex inserts to single copies by site-specific recombinase eliminate silencing [e.g. 38]. Therefore, when we started the analysis of the sequence changes in the cis-mutants, we were expecting a clear dependence of reactivation on loss of the duplicated region. This is not the case, since all cis-mutants, with the exception of RΔ2, still retain some duplicated regions. Also against expectation, a loss of the non-coding sequence homologous to the most abundant small RNAs is not a prerequisite for reactivation (RΔ3, RΔ5, and RΔ6). Furthermore, a loss of the small transcript starting from the P2 promoter is not necessary (RΔ3 and RΔ5), although its level in these mutants is not as high as in R plants. It should be kept in mind that neither the tandem sequence duplications, nor either of the two transcripts, are sufficient to initiate silencing, since R plants (with the complete insert and substantial transcription from P1 and P2) are fully resistant and stable. This is distinct from the FWA gene where tandem repeats are necessary and sufficient for silencing and DNA methylation [39]. Considering the lack of DNA methylation and small RNAs at the HPT insert in R plants, it is possible that the initial steps of silencing do not occur, are not efficient enough to start the reinforcing mechanism [39], or are inhibited by efficient transcription [40]. However, such conditions must have been overruled on the rare occasions that produced the silent epiallele in the first place.
The deletions in the different cis-mutants do not overlap in a specific region, and the smallest change is the loss of just 65 bp (RΔ5). Apparently, rather than affecting a specific sequence, the rearrangements change the overall organization at this locus. These changes can have variable consequences for the upstream promoter, causing either decisive, stable epigenetic switches (RΔ2, RΔ3, RΔ4) or leading to ambivalent states that can later fall back into silencing (RΔ1). How such small genetic heterogeneity, that does not affect coding or regulatory regions, can cause extreme epigenetic diversity at a promoter elsewhere remains an open question. The sequence changes could exert their effect by modifying the distance to flanking regulatory regions, the nucleosome arrangement or density, the association with DNA-binding molecules, or any higher order structure within the DNA. It is clearly different from the ‘spreading’ effect of silencing often associated with RdDM [41]–[42]: it causes activation (not silencing), goes against (not along with) the direction of transcription, and the most abundant of the relatively few small RNAs does not match the affected sequence of the upstream promoter. The results emphasize the mutual dependence between genetic and epigenetic factors, while indicating that these do not necessarily act at overlapping genomic sites. Similar effects might explain some of the associated changes in gene expression in the vicinity of small or large sequence modifications by transposon or recombination events. One example at a similar distance might be the transposon-dependent loss and gain of DNA methylation and inverse gene expression regulating sex determination in melon, at a site just 1.5 kb away from the insertion/excision site [43].
The relatively high number of cis-mutants in the screen was plausible in retrospective: mutations outside of the epiallele released silencing only if they reduce two epigenetic marks simultaneously. This is achieved by a few special mutations [22] or theoretically by rare double mutations and explains the low number of trans-acting mutants. In the study here, the genetic changes were found after mutagenesis by Agrobacterium-mediated T-DNA transformation [22], although none of the cis-mutations was connected with an integrated fragment of the incoming T-DNA. T-DNA transformation is also known to create mutations unlinked, or independent, from the site of integration [44] and can cause complex chromosome rearrangements [45]–[46]. Successful, and possibly also attempted, integrations occur at sites of microhomologies between T-DNA and plant DNA [47]–[48]. The incoming T-DNA [49] has some homology with the terminator sequences in the epiallele (ΔT), and in fact, the deletion sites in two cis-mutants (RΔ2, RΔ3) are near, or in, this sequence. The other deletions are close to promoter copy P2 that has no homology with the T-DNA, but potentially reflect a recombination hotspot in the 35S promoter sequence [50]. Alternatively, the double strand breaks connected with completed or aborted integration might stimulate repair via homologous recombination between the duplicated sequences of the epiallele (RΔ3). This would indeed have selected for 3′ rearrangements since those affecting the upstream copy are likely to lose the functional HPT cassette.
All together, the R and S epialleles described here provide an example of identical DNA sequences with converse expression states and specific epigenetic configuration that are faithfully transmitted to progeny. However, sequence changes in the vicinity of the silent epiallele can induce an epigenetic switch to the opposite state. These can have different degrees of stability, depending on the complex interplay between the nature of the sequence alteration, the consequences for transcription and transcripts, and the chromatin organization (Figure 6). This also illustrates a tight dependence of epigenetic regulation on local structures and makes it likely that DNA rearrangements can potentially change or induce new epialleles outside the affected region.
Arabidopsis thaliana lines with R and S epialleles in accession Zürich and mutagenesis of line S were described previously [20], [22]. Stratified seeds were surface-sterilized with 5% sodium hypochlorite and 0.05% Tween-80 for 6 min, washed and air-dried overnight. Sterilized seeds were germinated and grown in Petri dishes containing agar-solidified germination medium (GM) in growth chambers under 16 h light/8 h dark cycles at 21°C. For drug treatments, seeds were sown and plants grown on GM plates with hygromycin (10 µg/ml, Calbiochem), zebularine (40 µM, Sigma) or 3-deazaneplanocin (DZNep, 2 µM, donated by Dr. Victor Marquez) under the conditions described above.
Genomic DNA was isolated from 3 week-old seedlings using either DNeasy Plant Mini Kit (Qiagen) or Phytopure (Amersham), following the manufacturers' protocols, except that genomic DNA was eluted in sterile water. Total RNA extraction from 3 week-old seedlings was performed with RNeasy Plant Mini Kit (Qiagen) including an on-column DNase I digest (Qiagen). For Southern blot analysis, 10 µg of genomic DNA were digested overnight with 20 U restriction enzymes. For methylation-specific Southern blot analysis, the methylation-sensitive restriction enzymes (HpaII, blocked by mCG and mCHG, and MspI, blocked only by mCHG) were used. Digested samples were electrophoretically separated on 1.2% TAE agarose gels, depurinated for 10 min in 250 mM HCl, denaturated for 30 min in denaturation solution containing 0.5 M NaOH and 1.5 M NaCl and neutralized twice in 0.5 M Tris, 1.5 M NaCl and 1 mM EDTA at pH7.2 for 15 min. For northern blot analysis of total and poly(A) RNA, 5 µg of RNA were denatured with 15% glyoxal and 50% DMSO for 1 h at 50°C and separated using 1.5% agarose gels in 10 mM sodium phosphate buffer pH7 in a Sea2000 circular flow electrophoresis chamber (Elchrom Scientific). DNA and RNA gels were blotted onto Hybond N+ (Amersham) membranes overnight with 20× SSC, washed and UV-crosslinked using a Stratalinker (Stratagene). Hybridization was performed as described [51]. Radioactively labeled sequence-specific probes were synthesized from 25 ng of DNA using the Rediprime labeling kit (Amersham) and 50 µCi dCTP-α-32P (Amersham or Hartmann Analytic) and purified on G50 Probequant (Amersham) columns. Signals were detected with phosphoimager screens (Bio-Rad) and scanned with a Molecular Imager FX (Bio-Rad).
3′-RACE was performed with the SMART RACE cDNA Amplification Kit (Clontech) according to the instructions. Total RNA (700 ng) was treated with DNaseI (Fermentas), then reverse-transcribed with RevertAidRT (Fermentas) with 3-RACE A primer (5–AAGCAGTGGTATCAACGCAGAGTAC(T)30V N–3) in a 20 µl reaction. Two µl of cDNA reaction were used as template in 3′-RACE PCR. For this, Advantage 2 PCR Kit (Clontech) was used according to instructions. A control primer (Actin, Act2F primer: 5-GCCATCCAAGCTGTTCTCTC-3) and gene-specific primers were used in combination with UniA_45 (5–CTAATACGACTCACTATAGGGCAAGCAGTGGTATCAACGCAGAGT–3).
RNA samples were treated with DNase I (MBI Fermentas) for 30 min at 37°C to remove residual DNA contamination. The reaction was inactivated by addition of EDTA and incubation at 65°C for 10 min. Reverse transcription was performed on 1 µg of RNA with 0.2 µg of random hexamer primers (MBI Fermentas) using 1 U RevertAid H Minus M-MuLV-RTase (MBI Fermentas) in the presence of 20 U RiboLock Ribonuclease inhibitor at 42°C for 1.5 h. Real time PCR analysis was performed with the 2× SensiMix Plus SYBR & Fluorescein Kit (Quantace) protocol using an iQ5 Real-Time-PCR System (BioRad Laboratories). The obtained Ct values were analyzed with the iQ5 Optical System Software Version 2.0 (Bio-Rad), applying the mathematical model for relative quantification in Excel (Microsoft) as described [52]. All primer sequences are listed in Table S4.
After treatment with RNase A and proteinase K, 1–2 µg of genomic DNA were digested overnight with BamHI (MBI Fermentas). Subsequent bisulphite conversion was carried out using the Epitect Conversion Kit (Qiagen) and controlled for completion as described [21], [53]. Converted DNA was used for PCR amplification. PCR-amplified DNA was cloned using pGEM-Teasy (Promega) and ligation mixes transformed into DH5α cells (Invitrogen) and sequenced by terminal-labeling using BigDye Terminator v3.1 (Applied Biosystems). The sequence information obtained was analyzed with CyMATE, www.gmi.oeaw.ac.at/cymate [54], and Excel (Microsoft).
ChIP was performed as described (http://mescaline.igh.cnrs.fr/EpiGeneSys/www/images/protopdf/p13.pdf) using 3 week-old seedlings. The chromatin was immuno-precipitated with antibodies to histone H3 (Abcam, ab1791), H3K4me3 (Upstate, 07-473), H3K9me2 (T. Jenuwein 4677 and Abcam ab1220), and H3K27me2 (Upstate, 07-473). Immunoprecipitated DNA was purified using a Qiagen PCR Purification Kit and eluted in 50 µl of EB buffer. Quantitative real-time PCR was carried out in a total reaction volume of 25 µl and qPCR conditions were according to the 2× SensiMix Plus SYBR & Fluorescein Kit (Quantace) protocol using an iQ5 Real-Time-PCR System (BioRad Laboratories). qPCR data were evaluated as a ratio to input DNA [55].
Small RNA was isolated from either pooled inflorescences or seedlings (21 days old) using the mirVana miRNA Isolation Kit (Ambion). Small RNA libraries were generated as previously described [56] and sequenced using the Illumina G2 platform. After clipping the adapter sequence by vectorstrip software from EMBOSS package [57], small RNA reads were screened for homology with the epiallele sequence using bowtie [58], allowing only perfect matches (Table S3). Reads homologous to tRNA, rRNA, snRNA, snoRNA, mitochondrial RNAs, and chloroplast RNAs were removed by custom Perl scripts. The total number of reads that mapped to a certain region was computed as sum of 1/N_i (N_i is the number of times the read i was mapped). It was then normalized to indicate the number of each read per million bp (adapted from the RPKM concept in RNA-Seq, [59]. A threshold of 10 reads was chosen for any sequence to be taken into account. For the epiallele region, the normalized number of mapped reads was computed at single bp scale. For a more detailed view on a selected region, the analysis was performed with SiLoMa [60].
Additional methods are described in Text S1.
|
10.1371/journal.pntd.0007416 | Salmonella Typhi, Paratyphi A, Enteritidis and Typhimurium core proteomes reveal differentially expressed proteins linked to the cell surface and pathogenicity | Salmonella enterica subsp. enterica contains more than 2,600 serovars of which four are of major medical relevance for humans. While the typhoidal serovars (Typhi and Paratyphi A) are human-restricted and cause enteric fever, non-typhoidal Salmonella serovars (Typhimurium and Enteritidis) have a broad host range and predominantly cause gastroenteritis.
We compared the core proteomes of Salmonella Typhi, Paratyphi A, Typhimurium and Enteritidis using contemporary proteomics. For each serovar, five clinical isolates (covering different geographical origins) and one reference strain were grown in vitro to the exponential phase. Levels of orthologous proteins quantified in all four serovars and within the typhoidal and non-typhoidal groups were compared and subjected to gene ontology term enrichment and inferred regulatory interactions. Differential expression of the core proteomes of the typhoidal serovars appears mainly related to cell surface components and, for the non-typhoidal serovars, to pathogenicity.
Our comparative proteome analysis indicated differences in the expression of surface proteins between Salmonella Typhi and Paratyphi A, and in pathogenesis-related proteins between Salmonella Typhimurium and Enteritidis. Our findings may guide future development of novel diagnostics and vaccines, as well as understanding of disease progression.
| With an estimated 20 million typhoid cases and an even higher number of non-typhoid cases the health burden caused by salmonellosis is huge. Salmonellosis is caused by the bacterial species Salmonella enterica and over 2500 different serovars exist, of which four are of major medical relevance for humans: Typhi and Paratyphi A cause typhoid fever while Typhimurium and Enteritidis are the dominant cause of non-typhoidal Salmonella infections. The proteome is the entire set of proteins that is expressed by a genome and the core proteome are all orthologous proteins detected in a given sample set. In this study we have investigated differential expression of the core proteomes of the Salmonella serovars Typhi, Paratyphi A, Typhimurium and Enteritidis, as well as the regulating molecules. Our comparative proteome analysis indicated differences in the expression of surface proteins between the serovars Typhi and Paratyphi A, and in pathogenesis-related proteins between Typhimurium and Enteritidis. Our findings in proteome-wide expression may guide the development of novel diagnostics and vaccines for Salmonella, as well as understanding of disease.
| The gram-negative bacterial genus Salmonella is divided in two species, Salmonella enterica and Salmonella bongori. Only the Salmonella enterica subspecies enterica is of clinical relevance for humans and is further classified into more than 2,600 serovars. The human restricted serovar Typhi (STY) and the closely related serovar Paratyphi A (SPTA) cause enteric fever [1], while the generalist serovars Typhimurium (STM) and Enteritidis (SENT) are the most important causes of non-typhoidal salmonellosis [2]. Enteric fever is a systemic disease that affects more than 27 million people worldwide and leads to more than 200,000 deaths annually [3,4]. While STY and SPTA both cause a systemic disease, SPTA causes a milder disease with a shorter incubation time [5]. In the last 20 years, the number of infections with SPTA has significantly increased in Asia [6]. The global burden of non-typhoidal Salmonella, a common cause of food poisoning that is usually characterized by localized gastroenteritis, is even higher with an estimated 93.8 million cases and 155,000 deaths each year [2]. Moreover, invasive non-typhoidal Salmonella has emerged as an important cause of bloodstream infection in Sub-Saharan Africa in both adults and children, and the incidence of invasive non-typhoidal Salmonella is estimated at 3.4 million cases with more than 600,000 deaths each year [7].
Comparative genomics of Salmonella enterica has revealed specific genetic fingerprints associated with invasive disease and host adaptation [8,9]. A comparative analysis of 8 typhoidal and 27 non-typhoidal Salmonella genomes demonstrated presence of typhoid-specific protein families which include virulence factors such as Vi polysaccharide pilus related proteins [10]. In addition, an in silico comparative analysis of Salmonella genomes identified 469 genes involved in the central anaerobic metabolism which was intact in gastrointestinal pathogens (SENT and STM among others) but decaying in extra-intestinal pathogens, such as STY and SPTA. This metabolic advantage might have a role in competing with other bacteria in the inflamed gut, thereby enhancing transmission of the gastrointestinal pathogens [11]. However, not all phenotypic differences in typhoidal and non-typhoidal Salmonella can be explained by presence or absence of functional genes. Investigating differential expression of the core proteomes (defined as all orthologous proteins quantified in a given sample set) between Salmonella serovars [12], and the regulating molecules involved, can reveal additional insights in the adaptations to different host environments and pathogenesis, as well as reveal the expression of potential vaccine and diagnostic targets.
In the last decade, mass spectrometry (MS) based proteomics has advanced rapidly and provides a comprehensive view on the proteins that are expressed by an organism. In clinical microbiology laboratories, MALDI-TOF MS is routinely used for bacterial genus and species identification [13]. In research, proteomics was used to characterize the proteomes of Salmonella Typhimurium and Enteritidis under specific in vitro culture conditions mimicking the phagosome [14,15], to identify proteins that were expressed by Salmonella Typhimurium isolated from infected macrophages [16], and to study antimicrobial resistance and virulence in Salmonella Typhimurium [17–19]. Next to proteome analysis within single serovars, comparative proteome studies have been conducted to assess the proteome variability between different Salmonella serovars. However, these studies used laboratory reference strains which may not represent the currently circulating clinical strains [20–22].
Here, we conducted a comparative analysis of the core proteomes of the clinically most relevant Salmonella enterica serovars: Typhi, Paratyphi A, Typhimurium and Enteritidis, using 20 Salmonella strains isolated from patients covering various geographical origins, as well as one reference strain per serovar. Our findings show that differential expression of the core proteome of the typhoidal serovars is mainly related to cell surface components and, for the non-typhoidal serovars, to pathogenicity.
Five clinical isolates per Salmonella serovar Typhi, Paratyphi A, Typhimurium and Enteritidis were selected from the strain collection at the clinical laboratory of the travel clinic of the Institute of Tropical Medicine, Antwerp, Belgium for shotgun proteome analysis. One ATCC reference strain for each Salmonella serovar was added to the sample set and for the Salmonella Typhi reference strain, a clinical strain was certified (Table 1). Given that the burden of typhoid fever and invasive non-typhoidal salmonellosis is highest in Asia and Africa respectively, we have selected representative strains from different countries covering both continents. All in vitro incubation was done at 37°C. Minimum and maximum temperatures were recorded and ranged between 35°C and 37°C. As all clinical strains have been isolated from patients, the strains were revived from Microbank cryogenic vials (Pro-Lab Diagnostics) on blood agar (BD Columbia Agar, 5% sheep blood) and grown overnight at 37°C. Single colonies were sub-cultured on MacConkey agar (BD MacConkey II Agar) and grown overnight at 37°C. Colonies were further solubilized into 3 ml of synthetic growth medium and supplemented with 1% glucose (Teknova HI-DEF Azure Media) until the OD was 0.06, and 250 μl of this suspension was inoculated into 5 ml of synthetic medium supplemented with 1% glucose and grown at 37°C with shaking at 220 rpm until mid-log phase (OD 0.5-OD 0.6). The Teknova HI-DEF Azure synthetic medium (S1 File) is based on the medium described by Neidhardt et al. [23].
Upon harvesting the bacteria, duplicate samples of 1 ml were taken from each culture and centrifuged at 5000 x g for 10 min at 4°C and the cell pellets were washed twice with phosphate buffered saline (PBS). Duplicate samples are thus further considered as technical replicates. Proteins were extracted from the bacterial pellets with the Qproteome Bacterial Protein Prep Kit (Qiagen) following the manufacturer’s instructions. Briefly, after snap-freezing on dry ice, bacterial cell pellets were thawed on ice for 15 minutes. Cell pellets were re-suspended 750 μl of lysis buffer supplemented with lysozyme and Benzonase Nuclease, all included in the extraction kit. EDTA-free protease inhibitor (Roche) was added to a final concentration of 2%. After incubation on ice for 30 minutes, lysates were centrifuged at 14,000 for 30 minutes to pellet the cellular debris, and the supernatant was collected. The protein concentration was determined with the BCA Protein Assay Kit (Pierce) (S1 Table). Proteins were reduced with 15 mM tris(2-carboxyethyl)phosphine hydrochloride (TCEP-HCl) and alkylated with 30 mM iodoacetamide (IAM) for 15 min in the dark while shaking at 37°C. The buffer was exchanged to digestion buffer (50 mM ammonium bicarbonate, pH 7.9) using G-25 illustra NAP-5 gel filtration columns (GE Healthcare). The eluates were then heated at 99°C for 5 min, put immediately on ice and, after cooling, sequencing grade modified trypsin (Promega) was added to a 1:100 trypsin to protein ratio upon which digestion proceeded at 37°C for 16 h. The trypsin activity was stopped by adding 60 μl of 10% trifluoroacetic acid (TFA) (0.6% final concentration).
The peptide mixtures were subjected to LC−MS/MS analysis using an Ultimate 3000 RSLC nano LC (Thermo Scientific, Bremen, Germany) in-line connected to a Q Exactive mass spectrometer (Thermo Fisher Scientific). The sample mixture was first loaded on a trapping column (made in-house, 100 μm internal diameter (I.D.), 20 mm long, filled with 5 μm C18 Reprosil-HD beads, Dr. Maisch, Ammerbuch-Entringen, Germany). After flushing from the trapping column, the peptides were loaded on an analytical column (75 μm I.D., 400 mm long and filled with 3 μm C18 Reprosil-HD beads (Dr. Maisch)) packed in the needle PicoFrit SELF/P PicoTip emitter (PF360-75-15-N-5 (NewObjective, Woburn, USA)). Peptides were loaded with loading solvent (0.1% TFA in water) and separated with a linear gradient from 98% solvent A’ (0.1% formic acid in water) to 40% solvent B′ (0.1% formic acid in water/acetonitrile, 20/80 (v/v)) in 130 min at a flow rate of 300 nL/min. This was followed by a 15 min wash reaching 99% solvent B’. The mass spectrometer was operated in data-dependent, positive ionization mode, automatically switching between MS and MS/MS acquisition for the 10 most abundant peaks in a given MS spectrum. The source voltage was 3.4 kV and the capillary temperature was at 275°C. One MS1 scan (m/z 400−2000, AGC target 3 × 106 ions, maximum ion injection time 80 ms) acquired at a resolution of 70,000 (at 200 m/z) was followed by up to 10 tandem MS scans (resolution 17,500 at 200 m/z) of the most intense ions fulfilling the defined selection criteria (AGC target 5 × 104 ions, maximum ion injection time 60 ms, isolation window 2 Da, fixed first mass 140 m/z, spectrum data type: centroid, underfill ratio 2%, intensity threshold 1.7xE4, exclusion of unassigned 1, 5–8, >8 charged precursors, peptide match preferred, exclude isotopes: on, dynamic exclusion time 20 s). The HCD collision energy was set to 25% normalized collision energy and the polydimethylcyclosiloxane background ion at 445.120025 Da was used for internal calibration (lock mass). The mass spectrometry proteomics data have been deposited to the PRIDE Archive (http://www.ebi.ac.uk/pride/archive/) via the PRIDE partner repository with the data set identifier PXD011154 (username: [email protected]; password: hN5SqXtY).
Raw MS files were analyzed by MaxQuant [24] version 1.5.0.25 and MS/MS spectra were searched against the translated protein sequences of the annotated genomes of Salmonella Typhi CT18 (NCBI accession number AL513382.1) [25], Paratyphi A ATCC 9150 (CP000026.1) [26], Typhimurium 14028S (CP001363.1) [27], and Enteritidis PT4/P125109 (AM933172.1) [28]. The following parameters were applied for the database search: enzyme specificity was set to trypsin/P allowing for a maximum of two missed cleavages; carbamidomethylation of cysteine was set as a fixed modification; methionine oxidation, N-terminal formylation on the protein level and conversion of N-terminal glutamine to pyroglutamate were set as variable modifications. The first search for precursor ions was performed with a mass tolerance of 20 ppm for calibration, while 6 ppm was applied for the main search. For protein identification, at least two unique peptides were required per protein group and the minimum peptide length was set to 7. The false discovery rate for peptide and protein identification was set to 1%. The minimum score threshold for both modified and unmodified peptides was set to 30. MS runs were analyzed with the “match between runs” option between samples of a given serovar. For matching, a retention time window of 42 s was selected. Protein quantification was based on the MaxQuant label-free (MaxLFQ) algorithm. For all other parameters, default settings were applied as advised by the developers.
The MaxQuant output file “proteinGroups.txt” was loaded into Perseus 1.5.0.8. The protein entries were filtered to remove potential contaminants, reverse hits and proteins only identified by site. Then, the LFQ intensities were log2 transformed and data were filtered for proteins containing a minimum number of valid values in 9 out of 12 samples. The log2 transformed data were then normalized by subtracting the median per sample within the dataset. To compare the different Salmonella serovars we used orthology mapping. Orthologous genes within the four serovars were retrieved from the Orthologous Matrix (OMA) database [29] with NCBI Taxonomy IDs 220341 (STY), 295319 (SPTA), 550537 (SENT) and 588858 (STM). Statistical significant differences in LFQ intensities were assessed using a two-sided t-test with Bonferroni adjusted P values using R. Proteins were considered differentially expressed if they showed a minimal 2-fold change in their overall levels with an adjusted P-value lower than 0.05. Principal component analysis (PCA) was done in Perseus 1.5.0.8 using default settings as advised by the developers.
Differentially expressed proteins were subjected to gene ontology (GO) term enrichment to investigate biological processes, molecular function and cellular compartment using the Database for Annotation, Visualization and Integrated Discovery (DAVID) bioinformatics resources 6.7 [30]. Briefly, we have uploaded the differentially expressed core proteins as an input list and performed GO term enrichment analysis against a background list with default settings (count threshold is 2 and EASE threshold is 0.1).
To infer regulatory interactions that can explain differential expression profiles we used the PheNetic web server (http://bioinformatics.intec.ugent.be/phenetic/#/index) with default settings (Cost is 0.1, Pathlength is 4 and k-best paths is 20) and upstream run mode [31]. Input data consisted of the available interaction network for Salmonella Typhimurium LT2 (http://bioinformatics.intec.ugent.be/phenetic/index.html#/network), the list of detected proteins that are shared by two groups, and the list of differentially expressed proteins with P<0.05.
The clinical Salmonella isolates were obtained through the project “Surveillance of antimicrobial resistance among consecutive blood culture isolates in tropical settings”, within the Third Framework Agreement between the Belgian Directorate of Development Cooperation (DGD) and the Institute of Tropical Medicine (ITM), Antwerp, Belgium. The partner institutes involved in this surveillance project that provided strains were: Sihanouk Hospital Centre of Hope, Phnom Penh, Cambodia and Institut National de Recherche Biomédicale, Kinshasa, Democratic Republic of the Congo. Ethical approval for the Microbiological Surveillance was granted by the Institutional Review Board at the ITM in Antwerp, by the Ethics Committees of the Antwerp University (Belgium). Ethical approval for the Microbiological Surveillance Study was granted by the Institutional Review Board of ITM, the Ethics Committee of Antwerp University and the competent ethical committees from the DR Congo and Cambodia respectively. The Salmonella Typhimurium isolate from The Gambia was received from the Medical Research Council (MRC) Keneba, MRC The Gambia. Ethical approval was granted by the Gambia Government/MRC Joint Ethics Committee. The remaining strains were obtained from patients presenting at the travel clinic at ITM, ethical approval was granted by the Institutional Review Board of ITM. All isolates and subsequent biological samples were anonymized.
The reference genomes of STY, SPTA, SENT and STM used in our analysis contain 4,600, 4,095, 4,318 and 5,372 protein-encoding genes, respectively. In total, 3596 orthologous genes in the four serovars were retrieved from the OMA database and 1,414, 1,558, 1,222 and 1,099 proteins were detected by LC-MS/MS analysis in the STY, SPTA, SENT and STM strains, respectively. Protein detection in technical replicates showed Pearson correlation coefficients higher than 0.92 for all samples, except for the STM strain from Ethiopia with a Pearson correlation of 0.86 (S2 Table). Intra-serovar PCA of the LFQ intensities of expressed proteins show little variation in expression levels between strains within the same serovar (S2 File). However, in order to conduct reliable intra-serovar comparisons, more strains should have been included per serovar.
In total, 418 orthologous proteins were detected in all serovars (Fig 1) and expression levels in the typhoidal (STY and SPTA) and non-typhoidal (STM and SENT) Salmonella serovars were compared by PCA of the LFQ intensities (Fig 2A). The first two components capture ~72% of the variability in the dataset and show that the typhoidal serovars do not separate from the non-typhoidal serovars based on the observed variability in LFQ intensities. When we compared the typhoidal with the non-typhoidal Salmonella strains, a total of 128 proteins showed a minimal 2-fold change in their overall levels with an adjusted P-value lower than 0.05 (S3 Table). GO term enrichment of these 128 proteins showed that all GO terms with a P value lower than 0.05 are related to translation and structural components of the ribosomes (Table 2).
A set of 810 core proteins were detected in Typhi and Paratyphi A and their LFQ intensities were used as input for PCA (Fig 2B). The first two components allow a clear separation of the STY from the SPTA strains, covering 80% of the total variation in expression levels. In addition, the PCA shows that clinical isolates do not separate from the reference strains in both serovars. A total of 230 proteins with a minimal 2-fold change in their overall levels and an adjusted P-value lower than 0.05 were considered significantly differentially expressed between STY and SPTA strains (S4 Table). GO functional enrichment analysis of these proteins indicated an enrichment of biological pathways that are related to carbohydrate and polysaccharide biosynthesis and metabolism, as well as the external encapsulating structure (Table 2). We have plotted our differential expression data set on the wide interaction network for Salmonella Typhimurium LT2. Using the upstream run mode, PheNetic searches for regulatory mechanisms that can explain our observed data set. The inferred sub-network (Fig 3) shows that many differentially expressed proteins are connected to each other by outer membrane, stress and carbohydrate metabolism regulatory proteins such as CpxR, YjeB and CRP, which are not necessarily differentially expressed themselves, but might have a post-translational serovar-specific effect. Moreover, the small regulatory RNAs OmrA and OmrB connect differentially expressed proteins involved in carbohydrate metabolism.
A set of 465 core proteins were detected in all strains of STM and SENT. PCA of the LFQ intensities of these proteins showed a clear separation of the STM isolates from the SENT isolates based on the observed protein expression levels where the first two components cover ~80% of the total variation in expression levels (Fig 2C). The PCA also shows that the reference strains and the clinical isolates do not separate in STM and SENT. A total of 192 proteins with a minimal 2-fold change in their overall levels and an adjusted P-value lower than 0.05 were considered significantly differentially expressed between STM and SENT strains (S5 Table). GO enrichment analysis of these proteins showed that all GO terms with P<0.05 are related to pathogenesis (Table 2). The inferred subnetwork (Fig 4) revealed that the flagellar biosynthesis sigma factor FliA and the flagellar transcriptional regulators FlhD and FlhC (STM1924.S) connect the upregulated flagellar synthesis and motility proteins in STM. HilA, the main regulator of Salmonella Pathogenicity Island 1 (SPI-1), is possibly involved in the upregulation of the type 3 secretion system (T3SS) structural protein Prgl and effector protein SipA in STM.
The genomes of typhoidal and non-typhoidal Salmonella have a high level of similarity with more than 98% of sequence identity [32]. However, these two groups cause different diseases, host-pathogen interactions and immune responses. Here, we conducted the first comprehensive analysis of the proteomes of the Salmonella serovars Typhi, Paratyphi A, Typhimurium and Enteritidis using five clinical isolates that cover different geographical regions and one reference strain per Salmonella serovar. We have compared the expression levels of proteins from the core proteome under in vitro conditions and identified regulators that may help to explain the differences between different Salmonella serovars.
The classification of the four serovars into typhoidal and non-typhoidal groups is largely based on clinical presentation, with systemic and gastrointestinal disease, respectively. However, PCA of the LFQ intensities of the 418 detected proteins shared by all four serovars did not separate the typhoidal from the non-typhoidal serovars. Out of these 418 detected core proteins, 128 were significantly differentially expressed between typhoidal and the non-typhoidal serovars. However, GO analysis showed enrichment for proteins involved in translation and ribosomal activity, and thus largely represent the house keeping machinery of the bacterial cells. PCA showed that the LFQ intensities of the reference and clinical isolates within the STY, SPTA, STM and SENT serovars do not cluster separately, and the reference strains can thus be considered as representative for the serovar.
Further analysis showed that 230 proteins were differentially expressed between STY and SPTA. GO analysis revealed that proteins involved in carbohydrate and lipopolysaccharide metabolism, and proteins involved in external encapsulating structures were most enriched. The regulators in the sub-network analysis connecting the differentially expressed proteins are implicated in the cell envelope stress response and in polysaccharide metabolism. For example, OmrA/B connect Dld and SdaB, two proteins that are involved in transport of sugars and carbohydrate biosynthesis in E.coli, respectively. It is plausible that a serovar-specific effect acts at the sRNA-level, which is not detected in our proteomic analysis. CpxR that is known to have a role in the response to alterations in the cell envelope in Salmonella [33], explains the expression of Psd and LpxA required for phospholipid and glycolipid metabolism, respectively [34,35]. RpoS, RpoE and RpoH are involved in the stress response to different environmental conditions and contribute to Salmonella virulence [36–38]. CRP regulates the transcription of different operons involved in the transport of sugars and in catabolic functions [39], and FruR is required for carbohydrate metabolism [40]. The observation that cell surface proteins are significantly differently expressed between STY and SPTA is relevant for the diagnosis of Salmonella as well as for vaccination purposes. While the reference diagnostic method for typhoid fever is microbiological culture (blood, bone marrow or stool) and subsequent serotyping, rapid diagnostic tests (RDTs) have been developed and are commercially available for STY antigen and antibody detection [41]. However, diagnostic accuracy of the current RDTs is low, ranging from 31–97% [42] and more performant RDTs are urgently needed, including RDTs for SPTA. It has recently been shown that Salmonella antigen-based RDTs can be successfully applied to blood culture broths for Salmonella identification [43]. Three currently available typhoid vaccines are recommended by the WHO: an oral vaccine based on a live attenuated mutant strain of STY Ty21a (Ty21a), the injectable Vi capsular polysaccharide (ViCPS) vaccine and the typhoid conjugate vaccine (TCV) (http://www.who.int/immunization/policy/position_papers/typhoid/en/). However, these Typhi vaccines do not provide protection against paratyphoid fever caused by SPTA [44], and hence, a vaccine that protects against typhoid and paratyphoid fever would be of high value. When selecting antigens for developing new diagnostics or vaccines for both STY and SPTA, one should take into account that although encoded in both serovars, membrane proteins can be differentially expressed between both serovars and this should be tested in vitro and in vivo.
Upon comparing the proteomes of STM and SENT, 465 core proteins were detected, of which 192 were differentially expressed between the two serovars. GO enrichment analysis revealed that flagellar proteins and proteins involved in pathogenesis were most differentially expressed between both serovars. Among the higher expressed proteins in STM over SENT, six proteins are directly related to Salmonella pathogenicity island 1-encoded Type III secretion system (InvJ, SipA, SipD, SipC, PrgI, SipB). The T3SS-1 is an important virulence machinery that controls penetration of the gut epithelium during the infection by injecting effector proteins directly into the cytoplasm of epithelial cells through a needle-like appendages [45]. The regulator proteins InvJ and PrgI are known to be involved in needle and inner rod assembly [46], while SipA induces actin cytoskeletal rearrangements [47] and the translocases SipB and SipC form a translocation pore into the host cell membrane which is connected to the needle complex [48]. The sub-network also shows that HilA is possibly involved in the observed activation of the invasion proteins (SipA and PrgI) in STM. In addition, in the inferred sub-network the regulators FlhC (STM1924.S), FlhD and FliA were identified as regulators that connect 8 differentially expressed flagellar proteins (FlgL, FliD, FlgE, FlgM, FlgK, FlgD, FlgN, FlgG), showing higher expression profiles in Typhimurium strains. Besides their role in motility, flagellins were shown to stimulate both the innate and adaptive immune system and to cause inflammation upon STM infection [49]. Moreover, loss of flagellin expression in Salmonella has been linked to increased virulence in mice [50].
Some limitations in our study should be considered. The Salmonella strains were grown in standard in vitro conditions which may not be representative for protein expression in the infected host [51]. The addition of glucose to the medium may have induced catabolite repression. However, the addition of glucose as carbon source in needed to permit the growth of bacteria. Moreover, growth temperatures ranged between 35°C and 37°C and may have impacted expression levels. For instance, pathogenicity related gene expression is known to be temperature-sensitive [52]. In addition, the protein extraction procedure might have minorly affected the observed protein profiles although all steps have been performed on ice or 4°C. However, all strains have been grown using the same in vitro culture conditions and underwent the same extraction procedure and any possible effects are thus very likely averaged out in the comparative analysis. In addition, our mass spectrometry set-up is not as sensitive as the newest instruments currently available, and we captured around 20 to 40% of the proteomes. Poorly expressed proteins in the standard in vitro culture conditions used may thus have been missed, such as virulence related proteins [53]. Finally, the aim of our study was to conduct a comparative analysis of orthologous proteins shared between the four Salmonella serovars, and as such, we do not present information on serovar-specific (non-orthologous) proteins.
In conclusion, to the best of our knowledge this is the first study that compared the core proteomes of a large panel of clinical Salmonella isolates, covering the four clinically most relevant Salmonella enterica serovars: Typhi, Paratyphi A, Typhimurium and Enteritidis. Our comparative proteome analysis indicated differences in the expression of surface proteins between STY and SPTA, and in pathogenesis-related proteins between STM and SENT. Our insights may guide future developed of novel diagnostics and vaccines, and understanding of disease progression.
|
10.1371/journal.pgen.0030179 | Meiotic Transmission of an In Vitro–Assembled Autonomous Maize Minichromosome | Autonomous chromosomes are generated in yeast (yeast artificial chromosomes) and human fibrosarcoma cells (human artificial chromosomes) by introducing purified DNA fragments that nucleate a kinetochore, replicate, and segregate to daughter cells. These autonomous minichromosomes are convenient for manipulating and delivering DNA segments containing multiple genes. In contrast, commercial production of transgenic crops relies on methods that integrate one or a few genes into host chromosomes; extensive screening to identify insertions with the desired expression level, copy number, structure, and genomic location; and long breeding programs to produce varieties that carry multiple transgenes. As a step toward improving transgenic crop production, we report the development of autonomous maize minichromosomes (MMCs). We constructed circular MMCs by combining DsRed and nptII marker genes with 7–190 kb of genomic maize DNA fragments containing satellites, retroelements, and/or other repeats commonly found in centromeres and using particle bombardment to deliver these constructs into embryogenic maize tissue. We selected transformed cells, regenerated plants, and propagated their progeny for multiple generations in the absence of selection. Fluorescent in situ hybridization and segregation analysis demonstrated that autonomous MMCs can be mitotically and meiotically maintained. The MMC described here showed meiotic segregation ratios approaching Mendelian inheritance: 93% transmission as a disome (100% expected), 39% transmission as a monosome crossed to wild type (50% expected), and 59% transmission in self crosses (75% expected). The fluorescent DsRed reporter gene on the MMC was expressed through four generations, and Southern blot analysis indicated the encoded genes were intact. This novel approach for plant transformation can facilitate crop biotechnology by (i) combining several trait genes on a single DNA fragment, (ii) arranging genes in a defined sequence context for more consistent gene expression, and (iii) providing an independent linkage group that can be rapidly introgressed into various germplasms.
| The production of transgenic maize has traditionally used techniques that integrate DNA fragments into a host chromosome. This can disrupt important native genes or can lead to poor expression of the added gene; consequently, large numbers of transgenic plants must be screened to find one suitable for commercial use. Further, there is a limit to the amount of DNA that can be integrated, making it difficult to add multiple genes at one time. Here, we describe a new system for delivering genes to maize. We constructed a minichromosome vector that remains separate, or autonomous, from the plant's chromosomes when introduced into maize cells. These minichromosomes were constructed from DNA sequences that naturally occur in maize centromeres, the chromosomal regions needed for inheritance. We characterized the behavior of Maize Minichromosome 1 (MMC1) through four generations, showing that it is efficiently inherited and that the genes it carries are expressed. This work makes it possible to design minichromosomes that carry several genes, enhancing the ability to engineer plant processes, including improving disease resistance, drought tolerance, or the production of complex biochemicals.
| Agricultural crops have the potential to meet escalating global demands for affordable and sustainable production of food, fuels, therapeutics, and biomaterials [1]. While standard integrative plant transformation can often meet these needs by safely introducing novel genes into plant chromosomes, they are limited in efficiency. Typically, biological delivery of DNA carried on an Agrobacterium T-DNA plasmid, or biolistic delivery of small DNA-coated particles is employed to transfer and integrate desired genes into a host plant chromosome [2]. Integration at random sites results in unpredictable transgene expression due to position effect variegation, variable copy number from tandem integrations, and frequent loss of gene integrity as a result of unpredictable breakage and end joining [2,3]. For highly characterized crops such as maize, transgene integration can also result in genetic linkage of the introduced genes to portions of the genome known to encode loci that confer undesired phenotypes, adding complexity when the transgenic locus is introgressed into other varieties [4,5]. Recent advances in gene integration technologies have aimed to surmount some of these difficulties. For example, zinc finger–mediated homologous recombination or site-specific recombination could eliminate the unpredictable expression that results from random insertion into the plant genome [6,7]. In addition, combining binary T-DNA elements with bacterial artificial chromosome (BAC) technology to produce BiBACs has the potential to introduce larger DNA fragments into the host genome [8,9]. In contrast to these systems, the maize minichromosomes described here remain separate from the host chromosomes, and thus provide an alternative approach with important benefits. Indeed, although precise integration into host chromosomes has long been a routine technique in Saccharomyces cerevisiae, the facile properties of autonomous vectors often make them a preferred choice for numerous applications, including commercial-scale protein production.
The first eukaryotic minichromosomes employed a simple centromere (CEN) sequence from the budding yeast S. cerevisiae, incorporated into versatile circular CEN and linear yeast artificial chromosome (YAC) vectors [10,11]. These yeast vectors were used to define a 125-bp DNA fragment sufficient for mitotic and meiotic centromere function [12]. While circular CEN vectors are most useful for carrying smaller DNA fragments, YAC vectors can carry megabase quantities of DNA and are convenient for manipulating large fragments of DNA [13]. Similarly, with carrying capacities of hundreds of kb, human artificial chromosomes (HACs) provide advantages over other in vitro–assembled vectors used in human cell transfection [14]. HACs containing tandem repeats of a 171-bp alpha satellite sequence can be maintained either as circular or linear, telomere-containing, episomes [15–19].
DNA sequences that can form stable minichromosomes are able to recapitulate centromere functions de novo by recruiting essential DNA binding proteins and epigenetic modifications. In human cells, different satellite arrays vary in their ability to efficiently form HACs, based on their satellite monomer sequence, chromosomal origin, array length, higher-order structure, and even vector composition [20–23]. These DNA sequences recruit centromere binding protein A (CENP-A), which substitutes for histone H3 to form centromeric nucleosomes; this protein marks active centromeres in S. cerevisiae (Cse4p), Schizosaccharomyces pombe (Cnp1), Drosophila melanogaster (Cid), Arabidopsis thaliana (HTR12), Zea mays (CENH3), and Homo sapiens (CENP-A) [24–29]. CENP-A complexes are maintained through mitosis and meiosis [30], resulting in an epigenetic mark that may be more important in perpetuating centromere activity than the underlying DNA sequence. Evidence for this role in centromere maintenance comes from human neocentromeres [31], where, at a very low frequency, ectopic centromeres are nucleated in regions that lack satellite DNA. Once formed, these neocentromeres are efficiently perpetuated. The ability to form centromeres on naked DNA also depends on cell type in mammalian systems; indeed, HAC formation has only been demonstrated in HT1080 fibrosarcoma cells. Yet once established, HACs can be transferred to other mammalian cell types, where they are stably maintained [32].
Maize centromeres contain repetitive sequences that are similar to those found in mammalian centromeres; for example, analogous to the tandem arrays of alpha satellite found in human centromeres, large tandem arrays of the 156-bp maize CentC satellite bind to CENP-A [33,34,28]. These satellite arrays are often interrupted by CRM, a centromere-specific retroelement that also binds CENP-A [28]; the significance of this arrangement for centromere function is unknown. Some maize varieties also have supernumerary B chromosomes with a distinct centromere satellite sequence, ZmBs [35,36]. These B chromosomes lack essential genes, and thus have been particularly useful for discerning the relationship between centromere structure and meiotic transmission [37–39]. A series of deletion derivatives of natural B chromosomes, derived from an A-B translocation event, showed a strong dependence on centromere size—the smallest functional derivative contained a 110-kb centromere and resulted in a meiotic transmission rate of 5%, yet showed a high stability in mitosis [39]. More recently, telomere-mediated chromosomal truncation was used to generate deletion derivatives from both A and B maize chromosomes [40]. Transgenes carried on these derivative chromosomes (or “engineered minichromosomes”) were expressed and meiotic inheritance ranged from 12% to 39% [40]. While this telomere-truncation approach can deliver both transgenes and sequences that promote site-directed integration, its utility for commercial applications may be limited—most commercial maize hybrids lack B chromosomes, and the duplications needed to maintain truncated A chromosomes may prove challenging for regulatory approval.
As described below, we developed autonomous minichromosomes that do not rely on alteration of endogenous chromosomes. We constructed plasmids carrying maize centromeric repeats, delivered purified constructs to embryogenic maize tissue, and assessed their ability to promote the formation of MMCs. MMC1 was characterized in detail; this CentC-based construct contained 19 kb of centromeric DNA and conferred efficient mitotic and meiotic inheritance through at least four generations when introduced into plant cells. This approach could be widely used in commercial corn production—a construct with a defined sequence will facilitate regulatory review, while MMC independence from the host genome reduces the risk of alternations that impair host fitness.
We probed a maize genomic BAC library with repetitive sequences, including those typically found in maize centromeres (Materials and Methods; Table 1). Clones enriched in satellite sequences, centromeric retroelements, and other repetitive sequences were chosen to assess whether they can form MMCs when delivered to plant cells. While our study did not explore the interactions between MMC DNA inserts and kinetochore or spindle proteins, we hereafter refer to these fragments as “centromeric,” based on the typical genomic location of the sequences they contain. In vitro Cre-lox recombination was used to fuse selected BAC clones to a circular vector containing a plant selectable marker (nptII) and a cell-autonomous reporter gene (nuclear-expressed DsRed), forming circular constructs. While circular and linear HACs containing repetitive centromeric DNA are mitotically transmitted in human cell lines, circular HACs can confer higher levels of meiotic transmission in transgenic mice [3,20,32]. Consequently, we focused our initial efforts on circular constructs, generating plants by bombarding embryogenic maize tissue with purified candidate MMC DNA, selecting transformed cells expressing the nptII marker and resistant to antibiotics, and propagating regenerated plants in the absence of selection (see Materials and Methods). Of the 102 constructs bombarded, 66 gave rise to regenerated plants; 52 of these constructs were randomly chosen and characterized as described below.
To evaluate whether the introduced constructs were maintained autonomously or instead had integrated into the genome, we preformed fluorescence in situ hybridization (FISH). We arrested root tip cells in mitosis, and stained chromosome spreads [41] with rhodamine and fluorescein-labeled probes corresponding to centromeric repeats and to MMC-encoded genes, respectively (Figure 1A–1I). FISH labeling of integrated control constructs resulted in adjacent pairs of metaphase FISH signals corresponding to replicated sister chromatids (Figure 1J). While some MMC constructs integrated (see below), we considered MMCs autonomous when (i) ≥70% of the cells examined (n ≥ 15) contained signals that were clearly distinct from the DAPI-stained host chromosomes, (ii) integrated signals were not detected, and iii) the fluorescent probe corresponding to the MMC-encoded genes colocalized with the probe to repetitive centromeric DNA, suggesting an intact construct and making it unlikely that the signal was due to noise. In many cases, the detection of a DAPI signal that colocalized with the FISH probes provided further evidence of MMC autonomy (Figure 1B–1D and 1F–1H).
Based on these criteria, 47/52 (90%) of the constructs we evaluated with FISH were able to form an autonomous MMC, and 43/52 (with centromeric inserts ranging in size from 7 to 190 kb) gave rise to plants that contained only an autonomous MMC (Table 2). This unexpectedly high rate of recovering autonomous MMCs suggests that embryogenic maize tissue readily establishes MMCs from purified DNA and that the BAC clones that yielded transformed plants contained sequences that efficiently promote MMC formation. The efficiency of forming an autonomous MMC increased slightly, although not significantly (t-test), as the size of the genomic DNA insert increased (Figure 1K). A similar analysis of human centromeric fragments showed that as little as 35 kb could generate a HAC, while larger fragments (70–220 kb) were required for efficient HAC formation [42]. As described below, MMCs were often efficiently inherited; nonetheless, MMC integration was detected only during the initial transformation event, and not in subsequent generations (T1 through T4, 0/312 metaphase spreads, 33 plants). Below, we report on the composition and behavior of one of the MMC constructs (MMC1) in detail.
Control transformations performed with a DsRed/nptII plasmid lacking a centromere-derived insert (pCHR758) contained a construct that integrated, as expected, into a native chromosome (7/7 events, Figure 1J). In contrast, for MMC1, 5/9 independent transformation events yielded solely an autonomous chromosome (Figure 1A–1H, see also Figure S1) and 4/9 generated both integrated and autonomous copies (Figure 1I). We tested the ability of these MMCs to confer inheritance by crossing T0 transformants to wild type, growing the progeny without selection, and monitoring nuclear-localized DsRed fluorescence (Figure 2A). Because we typically observed only one MMC per cell (monosomic), we expected these T0 plants to behave as hemizygotes; if the MMC obeyed Mendelian inheritance, then such crosses would yield DsRed progeny in a 1:1 ratio. Ten T0 plants (derived from three events) carrying solely an autonomous MMC1 copy were crossed to wild-type pollen. Two of the MMC1 events (V-1 and Q-1) indeed transmitted DsRed to T1 offspring in ratios that did not differ significantly from Mendelian predictions (Table 3). However, for a third MMC1 event (Q-2), we saw a significant reduction in DsRed+ progeny compared to expectations (52%, Table 3), suggesting genetic instability. PCR analysis of the progeny from this cross confirmed that the plants lacking DsRed expression also lacked DsRed sequences, indicating that the deviation from Mendelian assortment was not due to silencing of gene expression. Instead, the elevated MMC loss rate in this event could result from in planta modifications of the centromeric insert or from epigenetic effects that led to less robust segregation [43]. As expected, performing a similar analysis of six events carrying an integrated pCHR758 backbone yielded Mendelian inheritance ratios (118:119 DsRed+:DsRed−; p > 0.05).
FISH analysis showed that T1 plants from event V-1 retained an autonomous MMC: a DsRed-containing episome was present in 80% of root metaphase cells (n = 44), a detection level consistent with previous artificial chromosome studies [44]. Because we consistently observed DsRed expression in nearly every cell from these plants (see below), we conclude that the absence of an MMC FISH signal in 20% of root cells likely represents the challenges of retaining and detecting every MMC throughout the FISH protocol. To monitor MMC1 inheritance in subsequent generations and through both male and female gametes, we performed a series of crosses with T1, T2, and T3 plants derived from event V-1 and monitored DsRed transmission. When male or female monosomic MMC1 plants were crossed to wild type, DsRed segregation was not significantly different from Mendelian inheritance ratios (1:1, Table 3). For one exceptional T1 plant, however, such crosses yielded no progeny containing MMC1 (female: 0:48, male: 0:35; Table 3); the absence of DsRed-encoding DNA in these progeny was confirmed by PCR, supporting the view that this MMC was indeed autonomous. Interestingly, the leaf tissue of this plant had prominent mitotic DsRed leaf sectors, suggesting a high rate of MMC instability.
When we self-pollinated T2 and T3 hemizygous plants derived from event V-1, we observed DsRed+ inheritance in a ratio that did not significantly differ from a 3:1 Mendelian pattern. However, in a second case of non-Mendelian assortment, a self-cross in the T1 generation yielded a 1:1 DsRed+ inheritance ratio, suggesting loss of MMC1 from either the male or female floral tissue. Nonetheless, this cross was useful for generating plants that potentially carried two copies of MMC1 (homozygous disomes). Crossing pollen from a candidate T2 disome onto five different maize inbreds yielded 184 DsRed+:18 DsRed− offspring (p > 0.05 for disomy). Similarly, self-pollinating potentially disomic T2 or T3 plants produced 48:0 and 24:0 DsRed+:DsRed− offspring, respectively. Quantitative PCR (qPCR) analysis of the potentially disomic T2 plants confirmed 2.00 and 1.90 (standard error = 0.08) DsRed copies per cell, respectively (see Materials and Methods).
For most plants carrying an autonomous MMC, nuclear DsRed expression was observed in nearly every leaf cell, indicating stability through mitosis. In some cases, however, sectors that lacked DsRed expression were found (Figure 2B–2D); these were generally limited to a few cell files. In reproductive tissues, such sectors could be responsible for the aberrant meiotic MMC segregation described above. In total, mitotic sectors of DsRed expression from MMC1 were detected in 3.6% of T0 plants (n = 56), 3.0% of T1 plants (n = 404), 1.9% of T2 plants (n = 837), and no T3 (n = 738) or T4 (n = 250) plants. The reduced sectoring frequency as plants advanced through generations suggests a gradual increase in MMC stability due to changes in DNA composition, epigenetic modifications, or MMC copy number in mitotic cells. A similar stabilization through generations was observed in an oat-maize addition line [45]. We also found that 60 d of crowding and drought stress did not appreciably alter MMC1 stability; DsRed expression was found in every T2 and T3 plant from event V-1 grown under stress (151 and 159 plants, respectively). Moreover, pollen from stressed hemizygous T2 plants demonstrated Mendelian DsRed segregation (281:238 DsRed+:DsRed−; p > 0.05).
To assess the structure of MMC1 through generations, we performed Southern blot analysis, probing to detect all of the unique sequence bands contained in the MMC construct (Figure 2E and 2F). MMC structural alterations sometimes occurred during transformation, often involving the centromeric insert, rather than the gene cassette (Figure 2F). Additional rearrangements were typically not detected after the T1 generation (n = 5), although the repetitive nature of the centromeric fragment made it impossible to thoroughly evaluate its structure on these blots. In addition, Southern blot analysis showed centromeric alterations in event V-1 that were transmitted from the T1 parent to the T2 progeny. Event Q-2 suffered a larger alteration of the centromeric fragments (indicated by an asterisk in Figure 2F), potentially explaining its reduced meiotic stability. In contrast, an event carrying both integrated and autonomous MMC1 copies (V-4) showed a more complicated pattern, as did plants carrying integrated pCHR758. As expected for independently assorting loci, when plants from event V-4 were crossed to wild type, the autonomous and integrated copies segregated: FISH evaluation of DsRed-expressing T2 plants yielded a 1:4:2 ratio (autonomous:autonomous and integrated:integrated).
MMC1 was originally identified by its strong hybridization to a CentC probe, suggesting it contained a high percentage of this satellite repeat (Table 1). Sequence analysis confirmed the presence of CentC repeats arranged in an uninterrupted tandem array (GenBank accession number in Supporting Information; Figure 3A and 3B). The repetitive nature of CentC made a precise assembly of this array challenging; we used rare DNA polymorphisms within the repeats to aid in sequence assembly, and confirmed the overall length of the array (approximately 9 kb) with restriction enzyme digestion and gel electrophoresis. Based on these measurements and quantitative dot blot hybridization (see Materials and Methods) the CentC array contains between 59 and 64 (61.4 ± 2.3) copies. CentC repeat alignments showed that each base is conserved at an average frequency of 96.1% (Figure 3C and 3D), a level consistent with previously reported plant satellite conservation [46]. Clustering algorithms failed to detect higher order repeat patterns in MMC1 (unpublished data).
While the maize genome has an average GC content of 49.5%, the 5.6- and 4.8-kb regions flanking the CentC array of MMC1 reach 88% and 70% GC, respectively (Figure 3B). Overall, the GC content of the MMC1 centromeric insert is 48%; by comparison, published sequences from two maize centromeric BACs had 43% and 47% GC content [34] while Arabidopsis and rice centromere DNA averages 35%–40% and 39%–48%, respectively [47,48]. MMC1 encodes four regions with similarity to retrotransposons xilon, cinful, or ji [49], as well as a 453-bp open reading frame (MMC1.1) that potentially encodes a novel protein of unknown function, complete with a promoter and poly-A signal (Figure 3A). BLAST searches of GenBank revealed no evidence for MMC1.1 expression, but transcripts >95% identical to CentC and to the MMC1 retrotransposons were abundant. Transcription of centromeric repeats is important for centromere function in S. pombe [50], and Arabidopsis satellites are also transcribed [51]. The centromere-specific histone CENH3 binds to transcripts corresponding to CentC and to the retrotransposon CRM, suggesting a role for these RNAs in centromere function [52]; it is not clear if xilon, cinful, or ji transcripts play a similar role. Retrotransposons also can nucleate the formation of heterochromatin that can spread to nearby regions [53], although MMC1-encoded DsRed and nptII were readily expressed, despite their separation of 3.3 and 6.2 kb, respectively, from retrotransposons.
Taken together, the experiments described above strongly support the conclusion that MMC1 can be maintained as an autonomous chromosome: it remains distinct from host chromosomes, its gene cassette is structurally stable through at least four generations, the genes it carries are expressed and transmitted through meiosis and mitosis, and, in some cases, it can be lost from the genome at a frequency higher than that of a native chromosome. Interestingly, classical studies of plant trisomics typically reveal far greater defects in meiotic inheritance [54], while inheritance levels similar to those we observed with MMCs have been reported in other artificial chromosome systems. For example, a monosomic mouse artificial chromosome that showed <1% mitotic loss when carried in human, bovine, or mouse cell lines [55], suffered only 4% meiotic loss through the mouse germline [44]. Furthermore, while classically studied ring chromosomes are often unstable [56], circular MMC inheritance through four generations was reminiscent of that observed for circular chromosomes from yeast [57], mammals [32], and maize [58]. These data suggest that this MMC could be maintained indefinitely.
MMC centromere sequences, like those that make up endogenous centromeres, could rely on the kinetochore and spindle machinery for faithful segregation, or could be inherited through alternative mechanisms. For example, in plants, dense heterochromatic domains known as knobs or neocentromeres migrate to daughter cells by moving along the sides of the spindle, rather than by kinetochore-mediated association with the ends of microtubules. This process results in preferential assortment to gametes, and consequently greater than expected inheritance ratios (termed meiotic drive) [59]. Heterochromatin-based mechanisms of assortment have also been characterized in Drosophila, where chromosomes that lack evidence of meiotic exchange (chiasmata) are nonetheless inherited at Mendelian ratios [60]. Further, in S. cerevisiae, which lacks appreciable heterochromatin, the 2-μm circle plasmid is partitioned at an efficiency that rivals that of yeast chromosomes; this assortment relies on microtubule-mediated attachment of cohesin to 2 μm of DNA [61]. The possibility that MMC segregation might rely on alternative mechanisms is intriguing; indeed, the relatively small MMCs may differ from mammalian artificial chromosomes in which large alpha satellite arrays bind essential centromere proteins.
Epigenetic factors have been postulated to play a principal role in establishing higher eukaryotic centromeres [43], with studies of human neocentromeres [62] and Drosophila strains overexpressing CENH3 [63] suggesting a lack of dependence on specific DNA sequences. On the other hand, HACs are able to efficiently nucleate centromere activity in a sequence-dependent manner, and HAC sequences tend to expand in vivo [20], suggesting a selection for a preferred size and/or composition. The MMC1 DNA that we delivered to plant cells was purified from E. coli and thus lacked eukaryotic epigenetic marks, yet it formed autonomous chromosomes. This MMC construct contained only a 19-kb centromeric insert and is thus substantially smaller than the centromeric regions that were previously known to provide mitotic and meiotic inheritance. For example, the fully sequenced centromere of rice Chromosome 8 contains a satellite array measuring 69 kb [64], and a deletion derivative of the maize B chromosome that measures 110 kb is sufficient to confer meiotic inheritance, albeit inefficiently [39]. While HACs routinely expand to a larger size in vivo, we did not detect major rearrangements or expansions of MMC1 DNA through two plant generations, suggesting that its composition was adequate to establish a minichromosome. Nonetheless, our analysis was unable to fully assess the structure of the repetitive centromeric DNA, and it remains possible that these regions could expand, contract, or rearrange in some other manner.
While the total size of MMC1 is quite small (35 kb), other MMCs, some measuring over 200 kb, were successfully delivered to plants and transmitted through meiosis (unpublished data). This suggests that MMC1 has the capacity to serve as a platform to carry a large number of genes. As this MMC is optimized to commercial performance levels, it will provide an unprecedented opportunity to deliver gene combinations (“stacks”) that confer valuable traits to corn varieties. Long breeding programs are often required to introgress an integrated transgene into desired germplasm, while eliminating undesirable linked loci. Because an MMC forms an independent linkage group, these programs could be accelerated, allowing products to appear in the marketplace sooner. Moreover, the performance and expression of transgenic traits will likely become more predictable and reliable as MMC design rules are understood. Extensions of this minichromosome technology beyond traditional agriculture may enable the construction of multigene pathways to produce pharmaceuticals and other industrial products in plants.
A BAC library was created in pBeloBAC11 using MboI-digested DNA from the maize inbred B73. This library was arrayed on nitrocellulose filters and probed separately with repetitive sequences from maize that are often found in centromeres or neocentromeres: CentA, Cent4, CentC, CRM, MZEHETRO, and TR-1; 32P-labeled probes were hybridized for 14 h at 65 °C and washed with 0.5× SSC, 1% SDS three times at 65 °C. To identify clones carrying centromere DNA, phosphorimager scans of each hybridization experiment were digitally assembled into a MySQL database. BAC clones with strong hybridization signals to one or more of the repetitive sequences were selected for minichromosome construction (Table 1). First, a high copy number plasmid (pCHR758) carrying the Arabidopsis UBQ10 promoter to DsRed (Clonetech) and the yeast YAT1 promoter fused to nptII was constructed. An 8.5-kb fragment encoding the DsRed and nptII expression constructs (and lacking a bacterial origin) was liberated from pCHR758 with I-PpoI, purified from an agarose gel (QIAquick Gel Extraction Kit, Qiagen), and circularized by Cre-mediated exchange (New England Biolabs) at two loxP sites that flanked the gene expression cassette. BAC clones carrying putative centromere DNA insertions were recombined with this vector via the loxP site in pBeloBAC11, generating circular candidate MMC constructs (Figure 2E). These constructs were maintained in E. coli DH10B (Invitrogen).
MMC constructs grown in E. coli were purified using alkaline lysis or cesium chloride protocols and delivered to embryogenic H99 maize tissues by biolistic bombardment of DNA-coated gold particles as described [65]. Transformed events were identified by selection on Chu's N6 medium containing G418 sulfate (PhytoTechnology Laboratories) or paromomycin (Sigma) and regenerated. Transformed plants were subsequently grown without selection in a soilless mix (Sunshine LC1) in a greenhouse (16-h d, 26–28 °C). Seedlings were grown in 48-well flats (2 sq ft) with one plant per well to the V3 developmental stage and then transplanted into 1.6-gallon pots containing 1:1:1 soil:peat:perlite and grown to maturity. Plants subjected to stress conditions were maintained in 48-well flats for 60 d with watering limited to once per day. MMC containing plants have been advanced through four generations by backcrossing to H99, outcrossing to public maize inbreds, and by self pollination or sibling mating.
For DsRed expression, leaf 3 (V2 stage of development) was sampled across its entire width (minimally 2,500 cells per sample) and fluorescence was detected using a Zeiss SV-11 dissecting microscope equipped with a rhodamine filter cube (excitation: D540/25; dichroic 565LP; emission: D605/55). Background autofluorescence was detected with a GFP filter cube (excitation: BP 470/40; beamsplitter: FT495; emission: BP 525/50); bona fide DsRed fluorescence was not detectable at this excitation wavelength. DsRed expression in pollen was determined after fixing florets in 95% ethanol; aceto-carmine staining was subsequently used to assess pollen viability. For FISH, root tips were collected approximately 10 d after transplanting regenerated T0 plants to soil or after germination (T1 through T4 plants). Sampled roots (3–6 per plant) were moistened and exposed to nitrous oxide at 150 psi for 2.5 h to arrest chromosomes in metaphase [66]. Roots were fixed in 90% acetic acid and spread onto poly-lysine coated glass slides by squashing thin cross sections. FISH was performed essentially as described [41] using probes labeled with Alexa488 (pCHR758, Molecular Probes) and Alexa568 (CentC, Roche). Following hybridization, slides were counterstained with DAPI (0.04 mg/ml) and ≥15 metaphase cells were evaluated per plant using a Zeiss Axio-Imager equipped with rhodamine, FITC, and DAPI filter sets (excitation BP 550/24, emission BP 605/70; excitation BP 470/40, emission: BP525/50; and excitation G 365, emission BP 445/50, respectively). Extrachromosomal signals were only considered to indicate autonomous MMCs if ≥70% of the images (n ≥ 15 cells analyzed) showed colocalization of the Alexa488 and Alexa568 signals within one nuclear diameter of the endogenous metaphase maize chromosomes. Grayscale images were captured in each panel, merged, and adjusted with pseudo-color using Zeiss AxioVision (Version 4.5) software; fluorescent signals from doubly labeled MMCs were detected in both the red and green channels.
PCRs were carried out on genomic DNA isolated from young plants; qPCRs were performed in triplicate using a BioRad Chromo4 machine with TaqMan primers and probes (Sigma-Genosys). Amplification was achieved by incubating at 95 °C for 3 min, and 39 cycles of 95 °C for 15 s and 59 °C for 48 s, with a 1 s reduction per cycle. Copy number determinations were made by comparing qPCR signals from a control plasmid containing one copy of the maize Adh1 gene and DsRed to the signals obtained from MMC-containing plants. For Southern blots, genomic DNA was isolated from young leaf tissue using a Nucleobond Plant Genomic DNA extraction kit (Clontech). Ten micrograms of DNA was digested with BglII (New England Biolabs), separated on a 0.7% agarose gel, vacuum transferred to a nylon membrane (Amersham BioSciences), and probed with a mixture of nonoverlapping pCHR758 fragments labeled with 32P (Rediprime II, Amersham BioSciences). Hybridization was performed overnight at 65 °C and blots were washed three times (15 min each) with 0.25× SSC, 0.1% SDS at 65 °C; signals were detected with a Storm phosphorimager.
MMC1 was sequenced to an average of 30× coverage by shotgun sequencing (Lark Technologies) and 454 Technology (454 Life Sciences) and assembled with Phred/Phrap; a small gap was closed by primer walking, using direct dye-terminator cycling sequencing of MMC1. Quantitative dot blotting was used to calculate the total size of the CentC array. Briefly, two sets of blots, each containing samples in triplicate, were hybridized with CentC (CC) and vector specific (V) probes separately. Signals for each spot were captured with a Storm phosphorimager and CC/V ratios were calculated. Plasmids with the vector sequence and one, three, and eight copies of a cloned CentC repeat were used as standards. MMC1 assembly was verified by restriction mapping with panels of enzymes (BamHI, BmgBI, EcoRI and HindIII); this data was consistent with the calculated size of the CentC array. BLASTN (http://www.ncbi.nlm.nih.gov/BLAST/Blast.cgi) was used to assess sequence similarity, GENSCAN (http://genes.mit.edu/GENSCAN.html) to predict promoters and open reading frames, and repeat finder (http://tandem.bu.edu/trf/trf.basic.submit.html) to analyze CentC satellites.
The significance of MMC1 inheritance data was determined with a chi-square goodness-of-fit test. Differences from Mendelian segregation (based on a 1:1 segregation ratio in crosses to wild type or a 3:1 segregation ratio in self crosses or crosses to sibling plants) were considered significant at p < 0.05 (or a chi-square value greater than 3.84).
The National Center for Biotechnology Information (NCBI) GenBank database (http://www.ncbi.nlm.nih.gov/sites/entrez?db=Nucleotide&itool=toolbar) accession numbers for the sequences discussed in this paper are: Cent4, AF242891; CentA, AF078917; CentC, AY321491; contig carrying the Arabidopsis UBQ10 promoter, AL161503; CRM, AY129008; MZEHETRO, M35408; nptII-carrying pBSL97, U35136; S. cerevisiae YAT1 promoter, L28920; TR-1, AY083992; Z. mays Adh1, X04049; and Z. mays MMC1, EU053446. |
10.1371/journal.pgen.1002763 | Divergence of the Yeast Transcription Factor FZF1 Affects Sulfite Resistance | Changes in gene expression are commonly observed during evolution. However, the phenotypic consequences of expression divergence are frequently unknown and difficult to measure. Transcriptional regulators provide a mechanism by which phenotypic divergence can occur through multiple, coordinated changes in gene expression during development or in response to environmental changes. Yet, some changes in transcriptional regulators may be constrained by their pleiotropic effects on gene expression. Here, we use a genome-wide screen for promoters that are likely to have diverged in function and identify a yeast transcription factor, FZF1, that has evolved substantial differences in its ability to confer resistance to sulfites. Chimeric alleles from four Saccharomyces species show that divergence in FZF1 activity is due to changes in both its coding and upstream noncoding sequence. Between the two closest species, noncoding changes affect the expression of FZF1, whereas coding changes affect the expression of SSU1, a sulfite efflux pump activated by FZF1. Both coding and noncoding changes also affect the expression of many other genes. Our results show how divergence in the coding and promoter region of a transcription factor alters the response to an environmental stress.
| Changes in gene regulation are thought to play an important role in evolution. While variation in gene expression between species is common, it is hard to identify the phenotypic consequences of this variation since many changes in gene expression may have subtle or no phenotypic effects. In this study, we investigate changes in sulfite resistance and gene expression caused by the transcription factor, FZF1, that has evolved rapidly during the divergence of related yeast species. We find that divergence in the ability of FZF1 to confer sulfite resistance is mediated by changes in its expression as well as changes in its protein structure, both of which cause changes in the expression of other genes. Our results show how the combination of multiple changes within a transcription factor can produce substantial changes in phenotype and the expression of many genes.
| Transcriptional regulation plays a key role in development and an organism's response to physiological and environmental changes. However, changes in gene regulation that occur over the course of evolution are more difficult to interpret. Genome-wide patterns of gene expression divergence show that while many aspects of regulation are conserved between distantly related species [1]–[3], there is also extensive variation in gene expression levels within and between closely related species [4]. In many, but not all instances, gene expression divergence is consistent with a neutral model of evolutionary change [5]–[8]. Yet, understanding regulatory divergence requires identifying the genetic basis of divergence in gene expression and knowing which changes in gene expression translate into changes in phenotype and fitness.
Substantial progress has been made in understanding the genetic basis of regulatory divergence. Changes in gene expression are influenced by both cis-regulatory sequences and trans-acting factors, with cis-regulatory changes being enriched in interspecific comparisons [9], [10]. Expression changes caused by cis-regulatory elements frequently involve gain or loss of transcription factor binding sites, e.g. [11], [12], although other changes, such as nucleosome position, can also play an important role [13]. Even when changes in gene expression can be attributed to specific cis-regulatory elements, the phenotypic consequences of such changes are hard to know, especially if they depend on the combined effects of many cis-regulatory changes. While changes in trans-acting factors can simultaneously influence the expression of many genes, significant efforts are needed to identify the genetic basis of trans-acting changes in gene expression.
The phenotypic effects of changes in gene expression have in some cases been identified [14]. This has primarily been accomplished by mapping, association and transgenic studies that identify genetic changes underlying a phenotype. While these approaches typically identify changes in protein coding sequences, cis-regulatory changes are more frequently found to underlie interspecific compared to intraspecific differences [14]. Furthermore, changes in protein coding sequences can affect the expression of many genes [15], [16], and in some cases their phenotypic effects depend on multiple differentially expressed genes [17].
What has been more difficult to investigate is the combined influence of multiple regulatory changes. Multiple changes of small effect may frequently go undetected, at least individually, but together could have a substantial impact on divergence [18]. Evidence for adaptive evolution via multiple cis-regulatory changes has been found based on concerted changes in the expression of genes that function in the same pathway or biological process [19]–[21]. Multiple cis-regulatory changes at a single locus have also been found to make substantial contributions to phenotypic divergence between species [22]–[26].
Statistical tests of neutrality are particularly well-suited to identifying multiple adaptive substitutions at a single locus since multiple substitutions are often needed to detect a significant deviation from a neutral pattern of molecular evolution. Rapidly evolving noncoding sequences have been identified in a number of species [27]–[30], and in some instances are known to cause notable changes in gene expression [31], [32]. Although tests of neutrality rely on the concentration of multiple changes at single loci, clustering of changes may occur if there are genetic, developmental or selective constraints at other loci [33].
One mechanism by which multiple, coordinated changes in gene expression may arise is through changes in transcriptional regulators. However, changes in transcription factors can also be constrained by their pleiotropic effects on gene expression. The negative effects of pleiotropy may in some cases be eliminated by altering the regulation of a transcription factor; thereby limiting downstream changes in gene expression to specific times during development, within particular cells or tissues, or to certain environmental conditions [33], [34].
In this study, we investigated changes in gene expression and phenotype caused by a rapidly evolving transcription factor, FZF1. To directly target genes that have potentially accrued multiple cis-regulatory changes, we screened four Saccharomyces genomes for noncoding sequences with non-neutral patterns of divergence. FZF1 was among the genes identified and it also shows a non-neutral pattern of amino acid divergence [35]. To examine the phenotypic consequences of FZF1 divergence we used cross-species complementation assays and found divergence in both its coding and upstream noncoding sequence affect sulfite resistance. Whereas divergence upstream of FZF1 affects its expression in response to sulfites, divergence in the coding region of FZF1 affects the expression of SSU1, an efflux pump that mediates sulfite resistance [36]–[38]. Coincident with their effects on sulfite resistance, both the coding and noncoding regions of FZF1 affect the expression of many other genes. Our results show how divergence in the coding and promoter region of a transcription factor affect the response to an environmental stress.
To identify promoter sequences likely to have diverged in function, we screened the noncoding sequences of four Saccharomyces species for accelerated substitution rates. We used a likelihood ratio test to compare a model of sequence evolution where the ratio of the noncoding to synonymous substitution rate, dNC/dS, is constant across lineages versus a model where dNC/dS is free to vary across lineages. Out of 2,539 noncoding regions tested, we identified 145 that showed significant variation in the noncoding substitution rate across species (Likelihood ratio test, P<0.05, Bonferroni corrected, Dataset S1). In these regions, a higher noncoding substitution rate in one or more lineages may be the result of loss of constraint, or in some cases, positive selection.
One of the noncoding regions that we identified lies upstream of the transcription factor FZF1. We selected FZF1 for further analysis because it is known to function in sulfite resistance, a hypothesized adaptation to vineyard environments [39], and its potential role in gene expression divergence. The substitution rate upstream of FZF1 is characterized by an accelerated rate along the lineages leading to Saccharomyces cerevisiae and Saccharomyces paradoxus relative to that along the lineages leading to Saccharomyces mikatae and Saccharomyces bayanus (Figure 1). However, previous studies have shown that signals of selection are highly dependent on the alignment [40], [41]. To determine whether the evidence for rate heterogeneity upstream of FZF1 is dependent on the alignment used, we generated additional alignments using alternative alignment parameters and algorithms, and tested each for substitution rate heterogeneity. Both the alignment parameters and the algorithm affected the evidence for rate heterogeneity, with 9 out of 18 alignments showing evidence of rate heterogeneity (Table S1, Likelihood ratio test, P<0.05, Bonferroni corrected). Although the high substitution rate combined with uncertainty in the placement of insertions or deletions makes it difficult to know the correct alignment, dNC/dS along the S. cerevisiae and S. paradoxus lineage was consistently estimated to be greater than or equal to one (Figure 1).
The protein coding sequence of FZF1 also shows evidence for non-neutral evolution based on a sliding window analysis of the nonsynonymous to synonymous substitution rate ratio (dN/dS) between S. cerevisiae and S. paradoxus [35]. However, caution should be taken when interpreting the results of the dN/dS test in the context of a sliding window analysis since dS can vary for a number of reasons [42]. Upon re-examination of divergence in FZF1, we found that the window with the signal of positive selection, dN/dS = 1.95, is characterized by a synonymous substitution rate of 0.18, which is lower than the average of 0.46 across the entire gene, and a nonsynonymous substitution rate of 0.34, which is higher than the average of 0.14 across the entire gene. Despite some uncertainty regarding the evidence for non-neutral evolution, we decided that FZF1 was a reasonable candidate to test for functional divergence.
FZF1 encodes a five zinc finger transcription factor that activates the plasma membrane sulfite pump, SSU1 [37]. Gain of function mutations in FZF1 result in hyperactivation of SSU1 and increased sulfite resistance [36], [38]. To determine whether FZF1 has diverged in its ability to confer sulfite resistance, we tested FZF1 alleles from four Saccharomyces species: S. cerevisiae, S. paradoxus, S. mikitae, and S. bayanus, for their ability to complement a deletion of FZF1 in S. cerevisiae. The S. cerevisiae allele of FZF1 showed nearly complete complementation of the FZF1 deletion, as measured by the delay in exponential growth following sulfite treatment (Figure S1). In comparison, FZF1 alleles from the other three species all showed a shorter delay in growth relative to that of S. cerevisiae, indicating that these FZF1 alleles confer greater resistance to sulfites (Figure 2, Kruskal-Wallis test, P = 5.3×10−13).
To determine whether divergence in FZF1 activity resulted from changes in its protein coding sequence or upstream noncoding sequence, we also tested chimeric constructs containing each species' FZF1 upstream noncoding sequence combined with the S. cerevisiae FZF1 coding sequence. These FZF1 5′ noncoding chimeras conferred significant differences in sulfite resistance (Figure 2, Kruskal-Wallis test, P = 2.5×10−18), indicating that the 5′ noncoding region alone makes a significant contribution to FZF1 divergence. Both the S. paradoxus - S. cerevisiae and S. mikatae - S. cerevisiae chimeric alleles showed sulfite resistance intermediate to that of their full length parental alleles, although only the former chimera was significantly different from both parent alleles (Wilcoxon rank sum test, P = 1.9×10−14 for the S. cerevisiae parent and P = 4.2×10−8 for the S. paradoxus parent). In contrast, the S. bayanus 5′ noncoding region upstream of an S. cerevisiae coding sequence conferred greater resistance than either of the two full length parent alleles (Figure 2, Wilcoxon rank sum test, P = 4.6×10−16 for the S. cerevisiae parent and P = 2.4×10−8 for the S. bayanus parent).
The S. cerevisiae and S. paradoxus alleles of FZF1 confer the largest difference in sulfite resistance. This phenotypic divergence corresponds to the lineages showing the highest noncoding to synonymous substitution rates and the elevated nonsynonymous to synonymous substitution rate within a portion of the coding region. Thus, we further mapped the differences in sulfite resistance between the S. cerevisiae and S. paradoxus FZF1 alleles.
The S. cerevisiae FZF1 protein is 900 amino acids long and has 195 bases in the 5′ noncoding region. Between the S. cerevisiae and S. paradoxus FZF1 alleles there are 67 amino acid differences and 82 differences in the 5′ noncoding region, 31 of which are insertion/deletion differences. To delineate which subset of these differences are responsible for divergence in sulfite resistance, we generated ten sets of reciprocal chimeric constructs between the two species (Figure 3). The FZF1 chimeric breakpoints were located (1) in the middle of the 5′ noncoding region, (2) at the junction between the 5′ noncoding and the coding region, (3) in the coding region between the first zinc finger domain, known to bind DNA [37], and the region under positive selection [35], and (4) at the junction between the coding and 3′ noncoding region. Five sets of chimeric constructs contain a single region in the opposite background and the remaining sets of constructs contain five of the ten possible pairwise combinations of each region.
Including the full length S. cerevisiae and S. paradoxus alleles of FZF1, the 22 constructs show a nearly continuous distribution of sulfite resistance (Figure 4). Using an additive model, the estimated effects of the first three FZF1 regions individually account for 8.2%, 39.0%, and 49.5%, respectively, of the difference in sulfite resistance between the S. cerevisiae and S. paradoxus alleles (Table 1). The latter two regions are not statistically significant. Some of the variation in sulfite resistance can be attributed to non-additive interactions among regions. The additive model explains a total of 66% of the variance among alleles, significantly less than a model that allows for pairwise epistatic interactions, which explains 70% of the variance (Likelihood ratio test, 2Δln(L) = 56.48, 10 d.f., P = 1.7×10−8). However, out of all the pairwise interactions, only the interaction between the two coding regions is individually significant after correcting for multiple tests (Table 1). The interaction indicates that the two coding regions have a smaller effect in combination compared to that expected from each region individually.
FZF1-dependent changes in sulfite resistance may be mediated by changes in the expression of FZF1 or the expression of other genes. To characterize changes in gene expression caused by FZF1 divergence, we measured expression of FZF1 and SSU1, a sulfite efflux pump activated by FZF1 [37], [38]. Using quantitative PCR, we measured the expression of both genes before and after sulfite treatment of strains carrying an S. cerevisiae, S. paradoxus, or two reciprocal chimeric FZF1 alleles, which divide the coding and 5′ noncoding regions of the S. cerevisiae and S. paradoxus FZF1 allele.
All of the FZF1 alleles increased in expression following sulfite treatment. However at time-points 15, 30 and 60 minutes after sulfite treatment, the FZF1 alleles with an S. paradoxus promoter were expressed at higher levels than those containing an S. cerevisiae promoter (Wilcoxon rank sum test, P = 6.7×10−9, P = 1.7×10−4, P = 0.008, respectively, Figure 5A). No significant differences were found due to the FZF1 coding region alone from the two species. Yet, 30 minutes after sulfite treatment, the two FZF1 alleles with the S. paradoxus promoter showed significant differences in expression; the allele with an S. cerevisiae coding region remained at a higher level relative to the allele with an S. paradoxus coding region (Wilcoxon rank sum test, P = 0.0012). Similarly, the FZF1 allele with an S. cerevisiae promoter and S. paradoxus coding region showed higher expression at the 30 minute time-point relative to the full length S. cerevisiae allele, although this difference was not significant (Wilcoxon rank sum test, P = 0.15). Differences in gene expression that depend on changes within a coding region have previously been found in yeast [43] and could result from feedback regulation.
The FZF1 alleles also caused an increase in SSU1 expression after sulfite treatment (Figure 5B). Unlike FZF1 expression, SSU1 expression primarily depended on the origin of the FZF1 coding region. For both the 15 and 30 minute time-points, FZF1 alleles containing the S. paradoxus coding region caused higher levels of SSU1 expression relative to those containing the S. cerevisiae coding region (Wilcoxon rank sum test, P = 1.15×10−5, P = 8.94×10−6, respectively). No significant differences in SSU1 expression were found as a result of the FZF1 5′ noncoding region alone.
If FZF1-dependent differences in sulfite resistance are mediated by activation of FZF1 and SSU1, they may also be influenced by levels of FZF1 and SSU1 expression prior to sulfite treatment. Immediately prior to sulfite treatment, FZF1 alleles with the S. cerevisiae coding sequences were expressed at 1.5-fold higher levels than those with the S. paradoxus coding sequence (Wilcoxon rank sum test, P = 1.3×10−6). The 5′ noncoding region caused no significant differences in FZF1 expression prior to sulfite treatment. In comparison, expression of SSU1 prior to sulfite treatment was 1.09-fold higher for FZF1 alleles containing the S. cerevisiae coding region and 1.12-fold higher for FZF1 alleles containing the S. cerevisiae 5′ noncoding region relative to the corresponding S. paradoxus regions (Wilcoxon rank sum test, P = 0.011, P = 6.5×10−4, respectively). Because the S. paradoxus allele of FZF1 causes higher levels of sulfite resistance, levels of FZF1 expression prior to sulfite treatment do not appear to be related to sulfite resistance.
The effect of FZF1 divergence on SSU1 expression suggests that FZF1 may also affect the expression of other genes. To examine this possibility, we measured genome-wide changes in expression caused by the S. cerevisiae and S. paradoxus FZF1 alleles and the two reciprocal 5′ noncoding chimeras. Gene expression was measured using microarrays before and 15 minutes after addition of sulfites. Out of 6127 open reading frames queried, 655 showed FZF1-dependent differences in expression across both time-points and 648 showed FZF1-dependent differences in expression that varied by time-point (ANOVA, P<0.01 for both). For both tests, permutation resampling of the data indicated a false discovery rate of 9.8%. Out of the combined set of 1,096 genes that showed FZF1-dependent differences in expression, 87% showed significant changes following sulfite treatment (ANOVA, P<0.01), of which 219 and 271 showed a >2-fold decrease and increase, respectively, in expression following sulfite treatment. Consistent with other studies of the stress response [44], [45], many of the genes that decreased in expression are involved in ribosome biogenesis (64 genes) and many of the genes that increased in expression are involved in oxidation reduction (51 genes) and response to abiotic stimulus (49 genes)(Dataset S2). Overall, strains carrying the S. cerevisiae FZF1 allele showed more pronounced changes in expression than those carrying the S. paradoxus allele (Figure S2), consistent with the possibility that many of the expression differences are not due to direct differential activation or repression by FZF1, but rather a consequence of downstream differences in sulfite resistance initiated by FZF1. A small number of genes, including SSU1, showed a larger increase in expression in strains carrying the S. paradoxus compared to the S. cerevisiae FZF1 allele. Excluding two putative genes, SSU1 showed the largest differences in expression between the S. cerevisiae and S. paradoxus alleles at 15 minutes and was one of the most significant FZF1-dependent differences across both time-points.
FZF1-dependent changes in gene expression may be caused by protein coding changes or by regulatory changes in the FZF1 5′ noncoding region. To distinguish between these possibilities, we classified FZF1-dependent expression changes into those that can be attributed to the 5′ noncoding region, coding region, or an interaction between the two regions. Most of the genes that showed FZF1-dependent differences in gene expression across both time-points were characterized by an interaction between the coding and 5′ noncoding regions (ANOVA, P<0.01, Figure 6). Interestingly, in many cases, the chimeric alleles caused these genes to be expressed at higher or lower levels compared to both of the full length alleles of each species. In contrast, most of the genes showing allele-specific differences in gene expression that varied by time-point were characterized by effects that depended on the coding region of FZF1 (ANOVA, P<0.01, Figure 6). Together, these results suggest that both the FZF1 coding and 5′ noncoding region contribute to downstream changes in gene expression.
Identification of genes that have diverged in function between species is a key element to understanding species' diversity and evolution. Divergence in transcription factors are of particular interest as they can coordinately regulate the expression of many changes, but by doing so may be limited in how they can evolve. In this study, we used patterns of non-neutral sequence evolution to identify genes likely to have diverged in their regulation. We investigated one candidate, FZF1, by testing species-specific alleles for their ability to complement a deletion of FZF1 in S. cerevisiae. We found that FZF1 has diverged in its ability to confer resistance to sulfites, and used chimeric constructs to show that divergence in sulfite resistance is due to changes in multiple coding and upstream noncoding regions. Finally, we found that divergence at FZF1 affects the expression of FZF1, SSU1 and many other genes. Our results provide insight into how both phenotypic and regulatory divergence is caused by evolution of a transcription factor.
We identified FZF1 based on a genome-wide screen for patterns of non-neutral divergence. FZF1 shows evidence of non-neutral divergence in its promoter region based on an accelerated substitution rate in some lineages but not others. In the coding region, evidence of non-neutral divergence is also present and is based on an elevated ratio of nonsynonymous to synonymous substitutions. However, upon closer examination we found a number of uncertainties regarding the evidence for non-neutral patterns of divergence. In the noncoding region, the evidence for substitution rate heterogeneity depends on the alignment. In the coding region, the cause of the elevated nonsynonymous to synonymous substitution rate is ambiguous because the synonymous substitution rate decreases in the same region that the nonsynonymous substitution rate increases. Interestingly, the strongest evidence for non-neutral evolution comes from divergence between the S. cerevisiae and S. paradoxus alleles, which also show the greatest difference in sulfite resistance. Thus, the pattern of divergence for FZF1 is at least consistent with non-neutral evolution. With respect to a potential cause of non-neutral divergence, both positive selection and loss of constraint can result in elevated substitution rates. However, loss of constraint by itself does not provide a good explanation for the loss of sulfite resistance along the S. cerevisiae lineage and the gain of sulfite resistance along the S. paradoxus lineage relative to the intermediate levels of sulfite resistance in S. mikatae and S. bayanus.
While patterns of non-neutral divergence led us to test FZF1 alleles for functional divergence, the value of such an approach remains difficult to assess. First, the evidence for non-neutral evolution is not definitive. Second, we only tested a single candidate. Third, the coding region with the largest effect on sulfite resistance does not include the region with evidence for non-neutral evolution. One factor that may be critical in selecting candidates is whether there is a clearly defined phenotype to test. Many of the other genes that exhibit substitution rate heterogeneity are known to impact a variety of phenotypes, making it difficult to know which one to test. Testing FZF1 was facilitated by its narrowly defined function in sulfite resistance. Thus, while some fascinating examples have emerged, e.g. [46], further work is needed to evaluate whether non-neutral patterns of divergence provide an effective screen for genes that have diverged in function.
Chimeric FZF1 alleles from S. cerevisiae and S. paradoxus indicate that both upstream noncoding regions and the first coding region make additive contributions to divergence in FZF1 activity. A second coding region, including the region with the elevated nonsynonymous to synonymous substitution rate, contributes an epistatic effect through interaction with the other coding region. The number of regions underlying sulfite resistance is likely dependent on how we identified FZF1. Tests of neutrality based on rate heterogeneity and dN/dS only indicate deviations from expected rates of divergence based on multiple substitutions. The accumulation of multiple changes at a single locus has also been found in other studies of interspecific differences [22]–[26], so it may not be an uncommon result when multiple regions are individually tested.
A limitation of our study is that we only quantified the effects of five regions and did not narrow their effects to individual substitutions. This limitation is in part due to the sensitivity of our sulfite resistance assay. As such, we did not determine whether the regions with the largest effect are caused by single or multiple substitutions, and whether there are epistatic effects between substitutions within a region. Further dissection of FZF1 divergence is needed to more accurately quantify the number, effect size and interactions among mutations affecting sulfite resistance.
Transcription factors are often posited to be highly constrained during evolution due to their pleiotropic effects on the expression of other genes [47]. As such, many efforts to understand the evolution of gene regulation have focused on the evolution of cis-regulatory sequences rather than on trans-acting factors, e.g. [12], [48]. While changes in the expression of transcriptional regulators is hypothesized to be an important mode of evolutionary change [34], protein coding changes may also be important, e.g. [49]. We find that divergence in both the regulatory and coding sequence of FZF1 affects sulfite resistance and causes numerous downstream changes in gene expression. This raises the question of whether there have been any constraints on FZF1 divergence due to pleiotropy.
If FZF1 has been constrained by pleiotropy there must, at least under certain circumstances, be negative consequences to changes in FZF1 activity. Increased levels of FZF1 activity could reduce fitness in the absence of sulfites or after other exposures that activate FZF1, such as nitric oxide treatment [50]. The observation that the more potent S. paradoxus FZF1 allele is expressed at lower levels in the absence of sulfites provides some support for the idea that high levels of FZF1 activity may not always be advantageous. Assuming that there is some cost to constitutive increases in FZF1 activity, there are a number of ways in which this cost could be small enough to overcome or even eliminated.
One consideration is that SSU1 expression is likely the major determinant of sulfite resistance and so the benefit of increased SSU1 expression may outweigh any costs. In support of this possibility, SSU1 overexpression is able to rescue the effect of an FZF1 deletion (Figure S3) [38]. However, the expression of other genes may also be involved in sulfite resistance since divergence upstream of FZF1 affects sulfite resistance but only has a small, insignificant effect on SSU1 expression. Thus, coding changes in FZF1 that increase SSU1 expression may have outweighed any costs under other conditions, or may have been facilitated by lower levels of FZF1 expression in the absence of sulfites.
Another explanation for the lack of constraints on FZF1 divergence is compensatory changes in genes regulated by FZF1. In this scenario, slight changes in FZF1 activity may be compensated by cis-regulatory mutations in those FZF1 regulated genes where changes in gene expression are deleterious. A number of empirical studies have shown that transcription factors bind different targets between closely related species and even within species due in part to cis-regulatory sequence changes [11], [51]–[53]. Thus, it is also possible that cis-regulatory sequence evolution may have accommodated divergence in FZF1 activity.
A third explanation, suggested by the finding that transcription factors with few targets are less likely to be constrained by pleiotropy [54], is that FZF1 has few transcriptional targets and so is not greatly constrained by pleiotropy. In response to exogenously supplied nitric oxide, activation of only a small set of five genes, including SSU1, was found to specifically depend on the presence of FZF1 [50]. Another study found 21 upregulated and 37 downregulated genes two hours after sulfite treatment [55]. We found 1,096 FZF1-dependent expression changes, most of which showed the same direction of response to sulfite and only differed in magnitude. The observation that the sulfite-sensitive S. cerevisiae FZF1 allele caused more pronounced changes in gene expression relative to the S. paradoxus allele (Figure S2) is consistent with FZF1 causing indirect changes in gene expression mediated by its effects on sulfite resistance rather than by direct activation or repression of these genes. Furthermore, we found no enrichment of the FZF1 motif identified in the SSU1 promoter (TATCGTAT and CAACAA, [37]), defined by protein microarrays (CTGCTA, [56]), or by promoter bashing and response to nitrosative stress (YGSMNMCTATCAYTTYY, [50]) within the 271 genes showing a 2-fold significant increase in expression following sulfite treatment. Thus, most of the changes in gene expression that we observed may be an indirect consequence of a sulfite-induced stress response rather than a consequence of changes in direct targets of FZF1.
Regardless of the mechanism, the concentration of multiple sequence changes in FZF1 suggests that it may have evolved without many genetic, functional or evolutionary constraints. However, the apparent absence of constraints could be a consequence of low basal levels of FZF1 expression. Under this scenario, changes in FZF1 regulation may have facilitated changes within its protein coding sequence.
Even though FZF1 has diverged in its ability to confer resistance to sulfites, its impact on the evolution of sulfite resistance is hard to know. While there is substantial variation in sulfite resistance within and between species (Figure S4), divergence at other loci may be responsible for most differences in sulfite resistance and could compensate for any changes in FZF1. Within S. cerevisiae, variation in sulfite resistance is associated with a reciprocal translocation upstream of SSU1 that is more frequent in vineyard and wine strains than strains derived from other sources [39], [57]. The inferred loss of sulfite resistance conferred by changes in FZF1 along the lineage leading to S. cerevisiae, combined with the gain of sulfite resistance due to the translocation within some strains of S. cerevisiae, suggests that the evolution of sulfite resistance among species is not simple and compensatory changes may be involved.
In this study we find substantial divergence in function within the coding and upstream noncoding region of FZF1. Our finding that multiple regions underlie divergence in sulfite resistance is not unexpected given the patterns of non-neutral evolution, but differs from other studies that identify single changes of large effect based on genetic mapping or candidate gene approaches [14]. The contribution of both noncoding and coding regions to differences in sulfite resistance suggests that the distinction between evolution in noncoding and coding regions may be less important than the degree to which a gene has the capacity to evolve, unencumbered by constraints on its other functions [33]. In conclusion, our work supports a model whereby both gene expression and phenotypic divergence can be attributed to multiple mutations throughout the regulatory and protein-coding region of a single gene.
S. cerevisiae, S. paradoxus, S. mikatae, and S. bayanus noncoding regions [58] were tested for substitution rate heterogeneity using a likelihood ratio test implemented using HyPhy [59]. The likelihood ratio test was used to compare a constrained model with a single substitution rate across lineages to an unconstrained model where each lineage was allowed to have a different substitution rate. For both models we used the HKY85 substitution model implemented in HyPhy, the known phylogenetic relationship among the species, and either a single parameter (constrained) or branch-specific parameters (unconstrained) for the ratio of the noncoding substitution rate at the locus of interest to the substitution rate at four-fold degenerate sites across the genome. Noncoding alignments were removed if the total length of insertion/deletions was more than 15% of the length of the entire alignment. While this filter eliminated the noncoding region upstream of FZF1, we had already initiated our functional analysis of FZF1 based on preliminary rate heterogeneity results and so retained it in our list of candidates. To examine whether substitution rate heterogeneity upstream of FZF1 depends on the alignment, we aligned the 5′ noncoding region using 6 alignment programs: Clustalw [60], MUSCLE [61], TCOFFEE [62], MAFFT [63], PRANK [64], and DCA [65]. The resulting alignments were tested for rate heterogeneity using the likelihood ratio test described above. For the coding sequence of FZF1, a sliding window analysis of dN/dS was performed for FZF1 using the K-estimator software [66] as described in Sawyer and Malik (2006). K-estimator uses Monte Carlo simulations to estimate the confidence intervals for estimates of dN/dS.
FZF1 was deleted in YJF173 (S288c-background, Mat a, ho, ura3-52) using the KANMX deletion cassette [67]. FZF1 alleles were integrated into this strain at the ura3 locus by amplifying the entire FZF1 gene region, including the entire 5′ and 3′ noncoding regions along with 25 bases of ZRT1 and 45 bases of HXK2, using primers with homology to pRS306 and transforming the product along with the yeast integrative plasmid, pRS306 [68]. Integration of these constructs at the ura3 locus was achieved by selection on plates lacking uracil and each transformant was confirmed by PCR. Chimeras were generated using the same procedure but with FZF1 regions amplified from different species. The aligned ATG start site was used for all chimeras divided between the 5′ noncoding region and the coding region. A mutation of an alternate FZF1 start site in the S. paradoxus FZF1 allele did not significantly alter sulfite resistance compared to the non-mutated counterpart (data not shown). A subset of 2–5 transformants were sequenced to ensure that at least one transformant per construct contained no mutations.
All experiments were conducted using YPD+TA (1% yeast extract, 2% peptone, 2% dextrose, 75 mM L-tartaric acid buffered to pH 3.5) [69]. Sulfite resistance was measured by comparing growth in the presence and absence of sodium sulfite. Strains were grown overnight in YPD+TA, diluted 1∶1000 in YPD+TA, grown for 3 hours, treated with either water or sodium sulfite (final concentration 0.7–0.9 mM sodium sulfite), and then grown for 20 hours in an iEMS plate reader at 30° with 1200 rpm shaking (model no. 1400; Thermo Lab Systems, Helsinki, Finland). For each strain, the sulfite-dependent delay in growth was determined by comparing the time at which maximum growth rate was observed for strains treated with sulfite relative to a water-treated control [70]. For each FZF1 construct, 4 to 8 independent transformants were phenotyped. To compare the sulfite-dependent delay in growth within and between yeast species, three replicate measurements were obtained for 6 S. cerevisiae strains: S288c (source: laboratory, obtained from: D. Botstein), YPS163 (source: oak exudate, United States, obtained from: P. Sniegowski), M8 and M33 (source: vineyard, Italy, obtained from R. Mortimer), YJM440 (source: clinical, United States, obtained from: J. McCusker), K9 (source: saké, Japan, obtained from: Nami Goto-Yamamoto), and five S. paradoxus strains: YPS138 (source: oak soil, United States), N17 (source: oak exudate, Russia), N44 (source: oak exudate, Russia), Y7 (source: oak bark, United Kingdom), and NRRL Y-17217 all obtained from G. Litti and E. Louis. Additional yeast species included: S. mikatae (IFO1815, obtained from: E. Louis), S. bayanus (NRRL Y-11845, obtained from: C. Kurtzman, ARS Culture Collection), Saccharomyces castellii (NRRL Y-12630, obtained from: M. Johnston), Saccharomyces kluyverii (NRRL Y-12651, obtained from: M. Johnston), and Kluyveromyces lactis (FM423, a haploid MAT á strain obtained from M. Johnston). All strains are diploid except as noted.
Differences in sulfite resistance between species and species' chimeras were normalized for day effects and tested for significance using the nonparametric Kruskal-Wallis test. Pairwise differences between constructs were examined using the nonparametric Wilcoxon rank sum test with Bonferroni correction.
Differences in sulfite resistance among S. cerevisiae - S. paradoxus chimeric constructs of FZF1 were measured using linear mixed effect (lme) models to account for repeated measurements of the same construct. Sulfite resistance of each construct was measured three times and measurements on different days were standardized by a Z-score transformation. Sulfite resistance was fit to two models. The first model assumes each region from S. cerevisiae or S. paradoxus makes an additive contribution to differences in sulfite resistance: sulfite resistance = region1+region2+region3+region4+region5+(error | batch)+error, where each region has an effect that depends on the species the region came from and (error | batch) models random effects due to measurement of the same construct in different batches (96-well plates). The second model builds on the first model but adds in all pairwise interactions between regions: sulfite resistance = (region1+region2+region3+region4+region5)∧2+(error | batch)+error. The fit of the two models was compared using a likelihood ratio test with 10 degrees of freedom since the first and second models have 8 and 18 degrees of freedom, respectively. The percent variance explained by each model was calculated by R2 = 1−exp(−LR/n), where n is the sample size and LR is the likelihood ratio statistic defined by twice the difference in the log likelihood of the alternative relative to the null model [71]. The null model was fit using only an intercept: sulfite resistance = (error | batch)+error. For lme P-values, we tested whether the assumptions of the test were violated and resulted in inaccurate P-values by repeatedly permuting the data labels to obtain the distribution of P-values expected by chance. The permuted data showed no evidence for inaccurate P-values.
Gene expression was measured using four independent transformants of each FZF1 construct. Strains were resuspended in YPD+TA at an OD600 of 0.25 from an overnight YPD+TA culture and grown in 100 mL cultures at 30°C, 200 rpm. After 3 hours, each culture was sampled at 0, 15, 30 and 60 minutes after addition of sodium sulfite to a final concentration of 1 mM. Cells were centrifuged, washed and frozen in a dry ice/ethanol bath and stored at −80°C. RNA was isolated and cDNA prepared using Qiagen's RNaeasy Mini Kit and Omniscript RT Kit, respectively (Valencia, CA).
Quantitative PCR was used to measure expression of FZF1 and SSU1. A 20-fold dilution of cDNA reactions was used for the real-time PCR assays with gene specific primers and Strategene's Brilliant II SYBR Green QPCR Master Mix (Santa Clara, CA). Expression was assayed on Stratagene's MX3000P QPCR machine. For FZF1, species-specific primers were used and a plate specific correction factor, estimated for each plate from quantitative PCR measurements of DNA extracted from a heterozygous strain containing both the S. cerevisiae and S. paradoxus FZF1 alleles, was used to account for the difference in PCR efficiency between the S. cerevisiae and S. paradoxus primers. Data were mean normalized for day and batch effects and expression levels were measured relative to ACT1. The Wilcoxon rank sum test with Bonferroni correction was used to identify significant differences in expression due to FZF1 alleles.
Genome-wide measurements of gene expression were obtained using Agilent Technologies (Santa Clara, CA) yeast (V2) gene expression microarrays (8×15K, Catalog number: G4813A-016322) following the manufacturers protocols. Sample labeling, hybridization and microarray scanning was conducted by the Expression and Genotype Core at Washington University's Genome Center. Gene expression was measured for three independent replicates at the 0 and 15 minute time-points. Each sample was compared to a reference made up of a pool of all RNA samples. Expression data was deposited in the GEO database under accession GSE35308. After median normalization of each microarray, differences in gene expression were tested using an analysis of variance (ANOVA) with the model: expression = allele*time+technical replicate+error, where allele measures the effect of the different FZF1 alleles, time measures the effect of each time-point, and technical replicate accounts for differences between replicated features on the microarray. The rate of false positives was estimated by permuting the sample labels 100 times and repeating the analysis. For each gene showing a significant difference in expression, a second ANOVA was performed to identify expression changes that could be attributed to the coding or 5′ noncoding region or an interaction between the two regions. For genes showing expression differences that depended on the FZF1 construct we used the model: expression = noncoding*coding+error, and for genes showing differences that depended on an interaction between the FZF1 construct and time we used the model: expression = noncoding*coding*time+error. Gene sets enriched for gene ontology (GO) categories were identified using DAVID [72].
|
10.1371/journal.pgen.1007463 | Wdr62 is involved in meiotic initiation via activating JNK signaling and associated with POI in humans | Meiosis is a germ cell-specific division that is indispensable for the generation of haploid gametes. However, the regulatory mechanisms of meiotic initiation remain elusive. Here, we report that the Wdr62 (WD40-repeat protein 62) is involved in meiotic initiation as a permissive factor rather than an instructive factor. Knock-out of this gene in a mouse model resulted in meiotic initiation defects. Further studies demonstrated that Wdr62 is required for RA-induced Stra8 expression via the activation of JNK signaling, and the defects in meiotic initiation from Wdr62-deficient mice could be partially rescued by JNK1 overexpression in germ cells. More importantly, two novel mutations of the WDR62 gene were detected in patients with premature ovarian insufficiency (POI), and these mutations played dominant-negative roles in regulating Stra8 expression. Hence, this study revealed that Wdr62 is involved in meiotic initiation via activating JNK signaling, which displays a novel mechanism for regulating meiotic initiation, and mutation of WDR62 is one of the potential etiologies of POI in humans.
| Meiosis is a unique cell division process which is indispensable for the generation of haploid gametes. However, the regulatory mechanism of meiotic initiation is unclear. In this study, we demonstrated that Wdr62 is required for meiotic initiation in germ cells via activating JNK signaling. More importantly, we also found that mutation of WDR62 was a potential etiologies of POI in humans. Taken together, this study revealed a novel mechanism for regulating meiotic initiation, and a potential etiologies of POI in humans. The results of this study provide important information for better understanding the regulation of meiotic initiation.
| In mammals, the haploid gametes are generated via meiosis, a program of two successive cell divisions preceded by one round of DNA replication. The onset of this program is referred to as meiotic initiation. Several intrinsic and extrinsic factors have been demonstrated to play roles in meiotic initiation. Retinoic acid (RA), the active derivative of vitamin A, which is synthesized in the mesonephros and diffuses into the adjacent gonad, is one of the most important meiosis-inducing factors [1, 2]. As an extrinsic factor, RA induces the germ cells to express the gatekeeper gene of meiosis Stra8 (stimulated by retinoic acid gene 8) [3]. Although the molecular functions of Stra8 have not yet been identified, several studies have shown that it is the first detectable sign of a germ cell’s decision to enter meiosis and is essential for pre-meiotic DNA replication and subsequent meiotic initiation [4–6]. Additionally, the RNA-binding protein DAZL (deleted in azoospermia-like) is an intrinsic factor required for germ cells to initiate the process of meiosis. Dazl knock-out mice fail to express meiotic marker genes in germ cells and retain a PGC (primordial germ cells) -like state in both sexes [7, 8]. Although the morphological changes during meiosis have been extensively studied, the underlying mechanisms that initiate this process remain largely unknown.
Premature ovarian insufficiency (POI), which is characterized by menstrual disturbance (oligomenorrhea or amenorrhea), elevated gonadotropins and low estradiol before 40 years of age, affects approximately 1% of women of childbearing age [9]. POI is heterogeneous in etiology, and known causes include genetic, autoimmune, iatrogenic or idiopathic factors. Approximately 25% of cases are thought to be genetically associated, with mutations in more than 80 genes concerning gonadal development, DNA replication/meiosis, DNA repair, and hormone synthesis [10–15]. However, up to date, etiology in most patients remains unknown.
WDR62 was originally identified as a scaffold protein in the JNK signaling pathway. Wdr62 encodes a protein containing 13 WD40 domain repeats in its N-terminal half and MKK7/JNK binding domains and six potential JNK phosphorylation sites in its C-terminal half [16]. WDR62 is the second most common genetic alterations associated with microcephaly (MCPH) in humans [17]. The requirement of Wdr62 in brain development and neural stem cell expansion has also been confirmed in mouse models [18–20].
Strikingly, we found that inactivation of Wdr62 caused meiotic initiation defects and germ cell loss in this study. Further studies revealed that the meiotic defects in Wdr62-deficient mice could be partially rescued by JNK1 overexpression. Interestingly, two novel WDR62 mutations were detected in 2 sporadic cases with POI, suggesting that mutation of WDR62 is one of the potential etiologies of POI in humans.
The immunohistochemistry results showed that the WDR62 protein was abundantly expressed in the germ cells in both ovaries and testes during the embryonic stage (S1A and S1B Fig). The results of real time PCR showed that the mRNA level of Wdr62 had no significant difference between ovaries and testes at E13.5 and E15.5 (S1F Fig). Wdr62 mRNA levels gradually increased from E11.5 to E13.5, and dramatically decreased at E14.5 and E16.5 in female gonads (S1G Fig). In testes, Wdr62 expression significantly increased from P1 to P7 and dramatically decreased at P10 (S1H Fig). A knock-out mouse model was generated to investigate the function of Wdr62 in germ cell development (S2 Fig). Wdr62−/− mice were born with a normal Mendelian ratio, and no developmental defects were observed (Fig 1A). When male and female Wdr62−/− mice were crossed with wild-type mice, no pups were obtained within 6 months, indicating that Wdr62−/− mice were completely infertile (Fig 1C and 1D). The size of ovaries in 2-month-old Wdr62-deficient females (Fig 1B, right) was dramatically reduced compared with that of control littermates (Fig 1B, left). H&E staining results showed that the ovarian follicles were absent in the Wdr62-deficient mice (Fig 1F). Further study found that the number of germ cells in Wdr62-deficient ovaries (S3B Fig) was comparable to that in control ovaries (S3A Fig) at E12.5. Germ cell loss was first noted in the Wdr62-deficient ovaries at E13.5 (S3D Fig). Germ cell number was dramatically decreased in the Wdr62-deficient ovaries at E15.5 (S3F Fig), and only a few MVH-positive germ cells remained at P1 (S3H Fig). The results of quantitative analyses also showed that the germ cell number was significantly reduced in Wdr62−/− ovaries after E13.5 (S3I Fig). As shown in S4 Fig, the size and weight of testes in Wdr62-deficient males were comparable to control littermates at P1. The size and weight of Wdr62−/− testes were slightly reduced at P5 and dramatically reduced at P10. The development of germ cells in Wdr62−/− testes (S5B and S5D Fig) was not affected at E15.5 and P1. The germ cell number was significantly reduced at P5 (S5F and S5I Fig), and very few of them were observed at P10 (S5H and S5I Fig).
Further confirm that Wdr62 is involved in germ cell development with a cell autonomous function, Wdr62−/flox; Tnap-Cre mice were generated. In Tnap-Cre mice, Cre recombinase is specifically expressed in germ cells of both male and female gonads at approximately E8.5 [21]. We found that the germ cell number was dramatically reduced in both male and female gonads 7 days after birth (S6 Fig). We also noticed that the phenotype observed in Wdr62−/flox; Tnap-Cre mice was less severe than that in Wdr62−/− mice. This is most likely due to the relative low efficiency of Tnap-Cre activity. The leakage of Tnap-Cre is another issue which needs to be clarified. For this reason, we specifically deleted Wdr62 in Sertoli cells using AMH-Cre mice. As shown in the S7 Fig, germ cell development in Wdr62−/flox; Amh-Cre mice was not affected, a large number of mature sperm were observed in the epididymis and histology of testes is normal. These results indicated that the defect of germ cell development was directly caused by inactivation of Wdr62 in germ cells rather than in somatic cells.
The timing of germ cell loss in Wdr62-deficient mice was consistent with the developmental stage for meiotic initiation. To examine whether the germ cell loss in Wdr62-deficient mice was caused by the defect in meiotic initiation, the expression of meiosis-specific marker genes was analyzed by immunofluorescence and real-time PCR. STRA8 protein was detected in the germ cells of control ovaries at E12.5 (Fig 2A) and E13.5 (Fig 2B), but not in germ cells from Wdr62−/− ovaries (Fig 2E and 2F). SYCP3 (Fig 2C) and γH2AX (Fig 2D) proteins were detected in most of the germ cells in control ovaries at E13.5, but these proteins were virtually absent in germ cells from Wdr62−/− ovaries (Fig 2G and 2H). However, germ cell marker proteins, DAZL and MVH, were expressed in both control (Fig 2A–2D) and Wdr62-deficient ovaries (Fig 2E–2H) at E12.5 and E13.5. The results of quantitative analyses showed that the percentage of STRA8-, SYCP3- and γH2AX-positive germ cells was significantly decreased in Wdr62−/− ovaries compared with control ovaries at E13.5 (Fig 2I). As shown in Fig 2J, the mRNA levels of meiotic genes were all significantly reduced in the purified Wdr62−/− germ cells at E13.5, whereas the expression of the germ cell-specific genes Dazl and Mvh was not changed. Interestingly, the mRNA levels of pluripotency genes Oct4, Nanog, Sox2 and Stella were significantly increased in the purified Wdr62-deficient germ cells, indicating that Wdr62-deficient germ cells were retained in an undifferentiated state.
We also examined the expression of meiotic genes in male germ cells. In control testes, STRA8 protein (S8A Fig) was expressed in most of germ cells at P3, and SYCP3 (S8B Fig) and λH2AX (S8C Fig) proteins were detected in the germ cells at P5. By contrast, none of these proteins were observed in germ cells from Wdr62-deficient testes at these stages (S8D, S8E and S8F Fig). The percentage of STRA8-, SYCP3- and γH2AX-positive germ cells was dramatically decreased in Wdr62−/− testes compared with control testes (S8G Fig). The H&E staining results showed that the germ cells displayed patches of condensed chromatin at the periphery of the nucleus in both control and Wdr62−/− ovaries at E12.5 (S9A and S9C Fig). By E13.5, the nuclei in control germ cells showed thread-like chromosome condensation that represents preleptotene, an initial stage of meiotic prophase (S9B Fig), whereas the nuclei from Wdr62-deficient germ cells still retained the same morphology as at E12.5 (S9D Fig). All these results indicated that Wdr62 knockout caused a defect in meiotic initiation in germ cells.
To examine whether the germ cell loss in Wdr62-deficient ovaries is due to the defective proliferation or cell apoptosis, Ki67 immunostaining and TUNEL assay were performed. As shown in S10 Fig, a majority of germ cells were Ki67-positive in control and Wdr62-deficient ovaries at E11.5 and E12.5. Numerous germ cells were still Ki67-positive in Wdr62-deficient ovaries at E13.5 and E15.5. By contrast, most of germ cells in control ovaries were Ki67-negative at E13.5 and E15.5. The results of TUNEL assay showed that the number of apoptotic germ cells was slightly increased at E12.5 and dramatically increased at E13.5 in Wdr62-deficient ovaries. These results indicated that the loss of germ cell in Wdr62-deficient ovaries is not due to the defect of proliferation. The Wdr62-deficient germ cells were retained in an undifferentiated state and underwent apoptosis eventually.
In male gonads, meiotic genes are not expressed in the germ cells during the embryonic stage, but can be induced by exogenous RA treatment [1, 2]. To test whether Wdr62 is required for RA-induced meiotic genes expression, the testes from control and Wdr62−/− mice were dissected at E13.5 and cultured in the presence of 1 μM RA. As shown in Fig 3A–3D, STRA8 and SYCP3 proteins were detected in the germ cells from control testes, whereas no STRA8 and SYCP3 signals were noted in germ cells from Wdr62-deficient testes. The results of quantitative analyses showed that the percentage of STRA8- and SYCP3-positive germ cells was significantly reduced in Wdr62-deficient testes (Fig 3E). The mRNA levels of Stra8, Sycp3 and other meiotic genes were also dramatically increased in the control testes after RA treatment but not in the Wdr62-deficient testes (Fig 3F). An in vitro study also showed that Stra8 mRNA level was significantly induced by Wdr62 in F9 cells in the presence of RA, but could not be induced by only Wdr62 transfection, indicating that Wdr62 seems to be a permissive factor but not an instructive factor (Fig 4A). Further study revealed that the Stra8 promoter could not be activated by either the WD40 domain or the JNK binding domain of WDR62 (Fig 4B), indicating that both WD40 and MKK7/JNK binding domains are essential for the normal WDR62 function. All these results indicated that Wdr62 is required for RA-induced Stra8 expression as a permissive factor.
It has been previously demonstrated that JNK signaling is activated by Wdr62 [16, 18, 22]. It is reasonable to postulate that JNK signaling pathway may also be involved in Wdr62-dependent Stra8 expression. The phosphorylated JNK protein in germ cells at E13.5 was examined by immunostaining. We found that p-JNK was detected in a small portion of germ cells in control mice (S11A and S11B Fig). By contrast, very few p-JNK positive germ cell was noted in Wdr62-deficient mice (S11C and S11D Fig). These results suggest that the activation of JNK signaling is probably involved in germ cell meiotic initiation. Moreover, a luciferase assay with AP-1 promoter, which is a direct downstream factor of JNK signaling, was performed. As expected, the activity of the AP-1 promoter was significantly increased with Wdr62 transfection. Interestingly, the activity of the AP-1 promoter was synergistically activated by Wdr62 transfection and RA treatment (Fig 4C). Moreover, RA induced JNK signaling activation was significantly decreased when endogenous Wdr62 was knocked down with shRNA in F9 cells (Fig 4E–4G). We also found that the expression of Stra8 could be induced by JNK1 and JNK2 overexpression in F9 cells (Fig 4D).
To further examine the functions of JNK signaling in meiotic gene expression, ovaries and testes from E13.5 control embryos and P2 testes were cultured in vitro and treated with RA and/or JNK inhibitor SP600125. Immunostaining results showed that STRA8 protein level was dramatically reduced (Fig 5E) and that the germ cells were blocked at early leptotene stage in the ovaries when treated with 1 μM SP600125 (Fig 5F). The percentage of STRA8- and SYCP3-positive germ cells in SP600125-treated ovaries was significantly reduced compared to that of control group (Fig 5I). The mRNA levels of meiosis-related genes were also significantly reduced in SP600125-treated ovaries (Fig 5K). As expected, STRA8 and SYCP3 expression were detected in the testes after exogenous RA treatment (Fig 5C and 5D). However, no STRA8 and SYCP3 proteins were detected in the testes with combined RA and SP600125 treatment (Fig 5G, 5H and 5J). The mRNA levels of Stra8, Rec8 and Sycp3 were significantly reduced compared with that in RA-only treated testes (Fig 5L). In testes, after 3 days culture, the expression of STRA8 and SYCP3 proteins was dramatically reduced with SP600125 treatment (Fig 5O and 5P) compared with control group (Fig 5M and 5N). The results of quantitative analyses showed that the percentage of STRA8- and SYCP3-positive germ cells in SP600125-treated testes was evidently reduced compared with control group (Fig 5Q). The mRNA level of meiotic genes was also significantly decreased in SP600125-treated testes compared with that in control testes (Fig 5R). These results indicated that JNK signaling plays important roles in germ cell meiotic initiation, and demonstrated that the defect in germ cell development in Wdr62 knockout mice was most likely mediated by JNK signaling.
To test this hypothesis, Wdr62−/−; CAJNK1±/flox; Tnap-Cre mice (referred to as rescued mice thereafter) were obtained. In this mouse model, Wdr62 was completely inactivated and JNK1 was constitutively activated in germ cells from approximately E8.5. Strikingly, the number of germ cells was significantly increased in the ovaries from rescued mice at E13.5 compared with ovaries from Wdr62−/− mice (S12J Fig). Meiotic gene (STRA8, SYCP3 and γH2AX) expression was significantly increased in the germ cells from rescued mice at E13.5 (S12B, S12E, S12H and S12K Fig). A large number of developing follicles were observed in the ovaries of rescued mice 7 days after birth (Fig 6C and 6D). Because the defect of germ cell development in rescued mice was only partially recovered, it is very hard to get pregnant spontaneously. To test the functions of oocytes from rescued mice, superovulation experiment was performed. After super-ovulation and mating with wild-type males, fertilized oocytes were obtained from 8-week-old rescued females. Live pups were obtained from both control and rescued females (Fig 6H and 6I) after embryo transplantation, and the quantitative information was shown in S2 Table. These results indicated that the defect of germ cell development in Wdr62−/− mice was partially rescued by JNK1 overexpression. Based on these results, we concluded that JNK signaling pathway plays important roles in Wdr62-dependent RA-induced meiotic gene expression.
The phenotype observed in Wdr62-deficient females correlate well with the pathological changes in PA patients, who exhibit many more severe defects in ovarian function compared with those with secondary amenorrhea in POI patients. We found two WDR62 heterozygous mutations by the screening of the whole exome sequencing in two patients with primary amenorrhea, and then we verified these mutations through Sanger sequencing. As shown in Fig 7A, Sanger sequencing revealed two novel mutations in the WDR62 gene in two patients. The missense mutation c.G1796A (p. C599Y) located on exon 14 and the frameshift-deletion c.3203_3206del (P.T1068fs) located on exon 26. Although the software recognizes the base "G" in the sequence diagram on the left, it can be seen from the sequence diagram that it is A and G heterozygous (green curve represents A base, and black curve represents G base). A mouse homolog was identified, indicating conservation of this protein. No other mutation associated with infertility was detected in POF916 patient. Two mutations of BRCA2 and one mutation of SPTB were detected in POF1072 patient. The BRCA2-deficient mice fail to progress complete meiosis, whereas meiotic initiation is normal [23]. The SPTB mutation information regarding infertility in mouse model can be found in Mouse Genome Informatics (http://www.informatics.jax.org). The detailed genetic profiles for these two patients were shown in supplementary material. Neither of the WDR62 mutations was reported in either 1000 Genomes or dbSNP database. Further Sanger sequencing confirmed that both variants were absent in the 192 healthy controls (Fig 7A and 7B).
To test whether the normal function of WDR62 was affected by these mutations, a luciferase assay was performed with a Stra8 promoter reporter vector and Wdr62-expressing vectors carrying a missense mutation (M1) or a frameshift-deletion (M2). The Stra8 promoter was significantly activated by wild-type WDR62 when RA was present, but it was not activated by mutant WDR62 (M1 and M2). Additionally, WDR62-induced Stra8 promoter activity was attenuated by co-transfection with mutant WDR62 (M1 or M2) (Fig 7C). These results indicated that the mutations observed in human patients played a dominant-negative role in regulating Stra8 expression and that mutation of WDR62 is a potential etiology of POI in humans.
Germ cell loss and infertility caused by meiotic defects have been reported previously [8, 24, 25]. In the Wdr62-deficient mouse model, germ cell loss was observed in both female and male mice, and the timing of germ cell loss in Wdr62-deficient mice was consistent with the developmental stage for meiotic initiation. The absence of meiosis-specific gene expression and chromatin condensation in Wdr62-deficient germ cells further confirmed the defects in meiotic initiation after Wdr62 inactivation. Although the expression of meiotic genes was absent, the expression of germ cell-specific genes (such as Dazl and Mvh) was not affected in Wdr62-deficient germ cells. Germ cells maintain pluripotency before meiotic initiation. Upon meiotic initiation, the pluripotency program is switched off and a set of genes is turned on to enable their differentiation. In Wdr62-deficient germ cells, the expression of core pluripotency genes (Oct4, Nanog, Sox2 and Stella) was all significantly increased compared with the control germ cells. This phenomenon is also observed in Stra8 and Dazl knock-out germ cells [5, 8], indicating that the process of meiosis is blocked and that the germ cells are retained in an undifferentiated state.
Organ culture experiments showed that Stra8 expression could not be induced by RA treatment in Wdr62-deficient germ cells. However, an in vitro study showed that Stra8 mRNA level was significantly induced after RA treatment without Wdr62 transfection. This discrepancy is most likely due to the endogenous Wdr62 in F9 cells.
Stra8 expression was completely absent, whereas Dazl expression was not affected in Wdr62-deficient germ cells, suggesting that Wdr62 is required for RA-induced Stra8 expression, which is probably independent of Dazl. Based on these results, we concluded that Wdr62 is required for germ cell meiotic initiation and that the germ cell loss in Wdr62-deficient mice is most likely a consequence of meiotic defects.
Although diverse cell types are exposed to RA during embryo development [26], meiotic initiation is limited to the germ line. Moreover, embryonic germ cells do not respond to RA induction until they migrate into the developing gonad, suggesting that RA alone is not sufficient to induce the temporal and cell-type-specific meiotic initiation. Other intrinsic factors are required for this process. Dazl has been demonstrated to play an essential role for meiotic initiation as an intrinsic factor [7]. Inactivation of Dazl leads to a lack of Stra8 expression and meiotic initiation defects [7]. However, whether other signaling pathways are also involved in meiotic initiation is unknown. JNK signaling pathway has been reported to play important roles in multiple organ development, and inactivation of JNK1 and JNK2 causes embryonic lethality in mouse model [27]. In this study, we found that JNK signaling was induced by RA treatment and Wdr62 was required for RA-induced JNK signaling activation in germ cells. The function of JNK signaling in meiosis has been further confirmed by organ culture experiments. We found that RA-induced meiotic gene expression in germ cells was abolished by treatment with a JNK inhibitor. Most importantly, the defect in germ cell development from Wdr62-deficient mice could be partially rescued by overexpression of constitutively activated JNK1. Our study demonstrated for the first that JNK signaling is involved in germ cell meiotic initiation. However, the underlying mechanism is still unclear and need further investigation.
POI is highly heterogeneous, with no single underlying dominant gene deficiency. More than 80 candidate genes have been identified, with approximately 25% demonstrating causative function. Elucidating the etiology and molecular basis of POI is of paramount importance for understanding ovarian physiology. The two mutations in the WDR62 gene identified in POI patients with a dominant-negative role provide clinical evidence for the role of WDR62 in folliculogenesis.
The gene trap mouse model of Wdr62 has been reported previously. In that mouse model, a β-geo reporter was inserted in the intronic region between exon 14 and 15, which caused a down-regulation of Wdr62 expression [19]. The homozygous mutant mice exhibited growth retardation and reduced brain size with spindle instability and mitotic arrest in neural progenitor and MEF cells. In our mouse model, exon 2 was deleted which caused a frame-shift of the Wdr62 gene. Surprisingly, we found that Wdr62−⁄− mice were born at a normal Mendelian ratio, and no obvious developmental defects were observed. This discrepancy between the two mouse models is probably caused by the different strategies used for generating the mouse models.
In summary, this study demonstrated that Wdr62 is required as a pivotal permissive factor for meiotic initiation in germ cells via activating JNK signaling and that mutation of WDR62 is one of the potential etiologies of POI in humans. The results from this study reveal a novel mechanism for regulating meiotic initiation, which will allow us to better understand the regulation of meiotic initiation.
All animal experimental procedures involved were performed in accordance with protocols approved by the Institutional Animal Care and Use Committee (IACUC) of the Institute of Zoology, CAS (AEI-09-02-2014). This human research study was approved by the Institutional Review Board of Reproductive Medicine of Shandong University ([2012] IRB No.18). Written informed consent was obtained from all subjects.
A Wdr62+/flox mouse model was generated based on methods reported previously [28]. In brief, the Wdr62flox allele was generated by inserting two loxp sites at both sides of exon 2 via homologous recombination. The Wdr62+/− mouse strain was obtained by crossing with ZP3-Cre transgenic mice. In ZP3-Cre mice, Cre recombinase is specifically expressed in oocytes [29]. In this mouse model, exon 2 was deleted, which caused a frame shift. The strategy for gene targeting in ES cells and genotyping was shown in S2 Fig.
All mice were maintained on a C57BL/6;129/SvEv mixed background. Wdr62−/flox; Tnap-Cre mice were obtained by crossing Wdr62+/−; Tnap-Cre males with Wdr62flox/flox females. The rescued mice (Wdr62−/−; CAJNK1+/flox; Tnap-Cre) were obtained by crossing Wdr62+/−; Tnap-Cre males with Wdr62+/−; CAJNK1flox/flox females. DNA isolated from tail biopsies and fetal tissues was used for genotyping as described previously [30, 31].
Gonads dissected from knock-out and control mice immediately after euthanasia, were fixed in 4% paraformaldehyde for up to 24 hours, stored in 70% ethanol, and embedded in paraffin. Five-micrometer-thick sections were cut and mounted on glass slides. After deparaffinization, sections were processed for immunohistochemistry and immunofluorescence analysis.
IHC and IF procedures were performed as described previously [30]. Antibodies were diluted as follows: MVH (1:500, Abcam, ab13840), WDR62 (1:400, Bethyl, A301-560A), DAZL (1:100, AbD Serotec, MCA2336), STRA8 (1:200, Abcam, ab49405), SYCP3 (1:200, Abcam, ab15093), γH2AX (1:400, Millipore, 05–636), Ki67 (1:200, Abcam, ab15580). After staining, the sections were examined with a Nikon microscopy, and images were captured with a Nikon DS-Ri1 CCD camera. The IF sections were examined using a confocal laser scanning microscope (Carl Zeiss Inc., Thornwood, NY). TUNEL assay was performed using the DeadEnd Fluorometric TUNEL System (Promega, G3250).
For quantitative analyses, more than three biological replicates from the control and experimental groups were performed. Paraffin-embedded ovaries and testes were serially sectioned and at least three sections apart were stained for observation. Within the group, at least three cross sections from each animal were examined. Around 1000 germ cells were counted in each group. The quantification of germ cells number (GCs number) was normalized to the control group. We considered the number of germ cells in the control group as 1, then the number of germ cells in other groups were quantified relative to the number of germ cells in the control group.
Germ cells were isolated from E13.5 genital ridges using SSEA-1 antibody as previously described [32]. Briefly, the digested female gonads were pooled and transferred to an Eppendorf tube containing 0.25% trypsin-0.02% EDTA for 5 min at 37°C. After digestion, a single-cell suspension was obtained by repeated pipetting. Then, 20 μL of SSEA-1 microbeads (Miltenyi) were added to the single-cell suspension and incubated for 15 min at 4°C. A magnetic separation was used to collect the magnetically labeled germ cells by applying the cell suspension onto a column (Miltenyi) placed in a MiniMACS separation unit (Miltenyi).
Total RNA was extracted from MACS sorted cells, cultured F9 cells or gonads using a Qiagen RNeasy kit in accordance with the manufacturer’s instructions. Two micrograms of total RNA was used to synthesize first-strand cDNA. To quantify gene expression, a SYBR Green real-time PCR assay was performed with the isolated RNA. All gene expression analyses shown in Fig 2, Fig 4 and S2 Fig were quantified relative to Gapdh expression, and Mvh was used as an endogenous control for gene expression analysis in the other figures (Fig 3, Fig 5 and S1 Fig). The relative concentration of the candidate genes was calculated using the formula 2-ΔΔCT as described in the SYBR Green user manual. The primers used were listed in supplementary S1 Table.
Tissue and cell were extracted in cold RIPA buffer (25 mM Tris-HCl, pH 7.6, 150 mM NaCl, 1%NP-40, 1% sodium deoxycholate, and 0.1% sodium dodecyl sulfate), which supplemented with 1 mM phenylmethylsulfonyl fluoride and a protease inhibitor cocktail (Roche, Indianapolis, IN, USA). The protein lysates were resolved by SDS–polyacrylamide gel electrophoresis (PAGE), transferred onto a nitrocellulose membrane and probed with the primary antibodies. The images were captured with the ODYSSEY Sa Infrared Imaging System (LI-COR Biosciences, Lincoln, NE, USA). Antibodies were diluted as follows: WDR62 (1:1000, Abcam, ab154044), Phospho-SAPK/JNK (Thr183/Tyr185) (1:1000, CST, 9251).
The promoter region of the Stra8 and Ap-1 genes was amplified by PCR using pfu DNA polymerase (NEB) as reported previously [33–35]. The amplified fragments were cloned into pGL3-basic plasmids. The Wdr62 expression vector was amplified by PCR using pfu DNA polymerase (NEB) and cloned into a pcDNA3.1-HA vector. The WDR62 mutation vectors carrying p.C599Y and p.T1068fs mutations were constructed by point mutagenesis using KOD-Plus-Neo polymerase (TOYOBO). The primers used for cloning were listed in S1 Table.
Cells of the mouse embryonal carcinoma F9 cell line were plated in 24-well plates and transfected with plasmids using Lipofectamine 3000 (Invitrogen) according to the manufacturer’s instructions. Forty-eight hours after transfection, the cells were lysed, and whole-cell extracts from triplicate wells were analyzed. pRL-TK (Promega) containing the Renilla luciferase gene was used as an internal control. The luciferase activity was measured with a luminometer (Promega). The results were normalized against Renilla luciferase activity.
Agarose gel stands (1.5% w/v, placed in 24-well plates) were pre-incubated with culture medium for more than 24 hours. The gonads with mesonephroi were dissected from control and Wdr62-deficient E13.5 embryos and P2 testes were dissected, then placed them on agarose stands. The gonads were cultured in DMEM/F12 containing 10% FBS at 37°C and 5% CO2. Then, 1 μM RA and/or 1 μM SP600125 were added to the culture medium. The cultured gonads were collected 72 hours later, and the gene expression was analyzed by real-time PCR and IF analysis.
Each female mouse at 8-week-old was injected with 7.5 IU of PMSG followed by 7.5 IU of hCG 48 hours to promote ovulation. The fertilized oocytes were collected from the ampulla of the oviduct of superovulated female mice after mating with male mice. The fertilized oocytes were transferred to the oviduct of surrogate mother in ICR background.
A total of 50 unrelated Han Chinese POI women with primary amenorrhea (PA) were recruited between April 2003 and Nov 2016 from the Center for Reproductive Medicine, Shandong Provincial Hospital Affiliated to Shandong University. Among patients with POI, women with PA exhibit much more severe defects in ovarian function compared with those with secondary amenorrhea. The inclusion criteria were absence of menstruation by the age of 16 with a serum follicle stimulating hormone (FSH) level >40 IU/L, measured on two occasions at least one month apart. Patients with known chromosomal abnormalities, previous chemo-/radiotherapy or ovarian surgery, autoimmune disorders, or somatic anomalies (particularly any reported as associated with syndromic POI) were excluded. In addition, 192 age-matched women with regular menses and a normal FSH level were enrolled as controls. Clinical characteristics of sporadic patients with PA and controls were shown in S3 Table.
We preformed whole exome sequencing (WES) in the 50 patients with PA. Genomic DNA was extracted from peripheral blood with QIAamp DNA Blood kit (QIAGEN, Hilden, Germany) according to standard protocols. Whole exome capture was carried out with SureSelect Target Enrichment System for Illumina Paired-End Sequencing Library (Agilent Technologies, Santa Clara, CA). DNA sequencing were performed on the Illumina platform (Illumina Hiseq, San Diego, CA). Reads were mapped to the hg19 reference genome with Burrows-Wheeler Alignment (BWA), and variants were called and annotated using ANNOVAR. Protein-coding variants were checked against established databases (1000 Genomes Project and dbSNP, version138). Novel nonsynonymous variants, which were predicted to below 0.05 on SIFT website and above 0.95 on PolyPhen-2 website, were confirmed through Sanger sequencing on two occasions. Nomenclature of variants identified was established according to Human Genome Variation Society (HGVS, www.hgvs.org/mutnomen).
Genomic DNA was extracted from peripheral blood according to standard protocols. The target fragments of WDR62 (NM_001083961) were amplified by PCR using primers listed in S1 Table. The PCR products was purified, labeled by BigDye (Terminatorv3.1 Cycle Sequencing Kits, Applied Biosystems), and sequenced by ABI 37306l DNA Analyzer (Applied Biosystems, Foster City, CA). The two variants were confirmed by two independent PCR runs and sequenced in both forward and reverse strands.
All experiments were repeated at least three times and at least three individual animals of each genotype were performed. The quantitative results were presented as the mean±SEM. The data were analyzed with Student’s t-test and one-way ANOVA.
|
10.1371/journal.pgen.1006522 | A Kinase-Phosphatase Switch Transduces Environmental Information into a Bacterial Cell Cycle Circuit | The bacterial cell cycle has been extensively studied under standard growth conditions. How it is modulated in response to environmental changes remains poorly understood. Here, we demonstrate that the freshwater bacterium Caulobacter crescentus blocks cell division and grows to filamentous cells in response to stress conditions affecting the cell membrane. Our data suggest that stress switches the membrane-bound cell cycle kinase CckA to its phosphatase mode, leading to the rapid dephosphorylation, inactivation and proteolysis of the master cell cycle regulator CtrA. The clearance of CtrA results in downregulation of division and morphogenesis genes and consequently a cell division block. Upon shift to non-stress conditions, cells quickly restart cell division and return to normal cell size. Our data indicate that the temporary inhibition of cell division through the regulated inactivation of CtrA constitutes a growth advantage under stress. Taken together, our work reveals a new mechanism that allows bacteria to alter their mode of proliferation in response to environmental cues by controlling the activity of a master cell cycle transcription factor. Furthermore, our results highlight the role of a bifunctional kinase in this process that integrates the cell cycle with environmental information.
| Free-living bacteria are frequently exposed to various environmental stress conditions. To survive under such adverse conditions, cells must induce pathways that prevent and alleviate cellular damages, but they must also adjust their cell cycle to guarantee cellular integrity. It has long been observed that various bacteria transform into filamentous cells under certain conditions in nature, indicating that they dynamically modulate cell division and the cell cycle in response to environmental cues. The molecular bases that allow bacteria to regulate cell division in response to fluctuating environmental conditions remain poorly understood. Here, we describe a new mechanism by which Caulobacter crescentus blocks division and transforms into filamentous cells under stress. We find that the observed cell division block depends on precise regulation of the key cell cycle regulator CtrA. Under optimal conditions, the membrane-bound cell cycle kinase CckA activates CtrA in response to spatiotemporal cues to induce expression of genes required for cell division. Our data suggest that external stress triggers CckA to dephosphorylate and inactivate CtrA, thus ensuring the downregulation of CtrA-regulated functions, including cell division. Given that CckA and CtrA are highly conserved among alphaproteobacteria, the mechanism found here, might operate in diverse bacteria, including those that are medically and agriculturally relevant.
| The bacterial cell cycle has been studied extensively in the past. Genetics, biochemistry and more recently, advanced microscopy techniques have provided important insight into the processes of DNA replication, chromosome segregation and cell division, and numerous regulatory mechanisms have been identified that precisely coordinate these processes in time and space. Most of this research has focused on cell cycle regulation under standard and stable laboratory growth conditions. However, in nature bacteria are exposed to drastic environmental changes, where they have to constantly adjust their growth rate and mode of proliferation [1,2]. It has frequently been reported that various bacteria transform into multi-chromosome containing filaments in response to certain environmental conditions [2–4], indicating that bacteria dynamically modulate cell division and the cell cycle in response to environmental cues. Nevertheless, the precise mechanisms transducing environmental information into the cell division machinery and how these mechanisms help cells to survive under adverse conditions are not well understood.
Cell cycle regulation has been studied in several model bacteria. One prominent example is the asymmetrically dividing alphaproteobacterium Caulobacter crescentus, a freshwater bacterium that mainly occurs in oligotrophic aquatic environments, but also in organically rich environments such as wastewater [5]. The Caulobacter cell cycle is characterized by asymmetric cell division and well-defined, morphologically distinct cell cycle phases, offering the possibility to examine cell cycle progression with high spatial and temporal resolution. Past work has identified a suite of key regulatory proteins required for cell cycle progression and important progress has been made in understanding how these factors are wired in higher-ordered circuits to drive cell cycle progression under optimal conditions [6,7]. However, how the Caulobacter cell cycle is modulated in response to environmental changes is only at the beginning of being explored.
One major cell cycle regulator is the conserved response regulator CtrA, which regulates the transcription of nearly 100 genes involved in cell division, cell cycle regulation and morphogenesis [8,9]. By binding to the origin of DNA replication CtrA also serves as a negative regulator of DNA replication initiation [10]. CtrA activity is strictly regulated and oscillates in a cell cycle-dependent manner [11]. In G1-phase CtrA is active and represses the origin [10]. At the G1-to-S transition it is inactivated and rapidly proteolysed allowing DNA replication to initiate [12,13]. During S-phase, active CtrA accumulates again to induce the expression of cell division and morphogenesis genes that are required to complete the cell cycle by cell division [9]. The oscillations of CtrA depend on its precise regulation by the CckA-ChpT phosphorelay, which is comprised of the polarly localized bifunctional histidine kinase CckA and the phosphotransferase ChpT [14,15]. In response to spatiotemporal cues, CckA phosphorylates CtrA via ChpT, resulting in CtrA activation. Reversal of the phosphorelay leads to CtrA dephosphorylation and hence its inactivation [15,16]. CtrA inactivation is tightly coupled to its degradation by the ClpXP protease that depends on CpdR, an adaptor protein, whose activity also depends on CckA-ChpT [17]. In contrast to CtrA, CpdR is only active when dephosphorylated [17,18]. Thus, under conditions when CtrA is inactivated, it is also targeted for degradation by ClpXP and CpdR. In addition to CpdR, the cell-cycle dependent degradation of CtrA involves the adaptors RcdA and PopA that promote CtrA proteolysis by ClpXP in a second messenger and phosphorylation dependent manner [19–21].
Recent work has provided important insight into the mechanisms regulating the CckA kinase in a cell cycle-dependent manner. It was shown that protein interactions with the upstream regulators DivL and DivK at the cell poles allow CckA to switch between kinase and phosphatase activity [22,23]. In addition, CckA activity is also modulated by the second messenger cyclic diguanylate (c-di-GMP) [24]. This molecule, which accumulates at the G1-S transition [25], directly binds CckA and stimulates its phosphatase activity, and thus CtrA inactivation and proteolysis [24]. As a transmembrane protein CckA may also directly respond to external signals, as typical for most other histidine kinases that have important sensing functions and directly transduce environmental information into cellular responses. However, external conditions that influence CckA activity have not been identified until now.
The present study started with a systematic analysis of the effects of different stress conditions on C. crescentus cell cycle progression. This analysis revealed that under certain conditions Caulobacter specifically blocks cell division and grows to filamentous cells. Interestingly, the stress-induced cell filamentation that we observed is not mediated by induction of small cell division inhibitors, as previously described for conditions inducing DNA damage [1,26,27]. Instead, we found that stress leads to rapid downregulation of CtrA. Our data suggest that the stress-induced regulation of CtrA stems from stimulation of CckA phosphatase activity and that it provides a growth advantage under stress.
To analyze the consequences of different stress conditions on cell cycle progression in C. crescentus, we challenged bacterial cultures with carbon starvation, stationary phase, heat shock, DNA damage, oxidative stress, salt stress, sucrose stress, ethanol stress and changes in pH while monitoring changes in cell morphology and chromosome content using phase-contrast microscopy and flow cytometry, respectively. As previously shown, carbon starvation, growth in stationary phase and acute heat shock lead to a block of DNA replication initiation and a G1-arrest while cell size is largely maintained (Fig 1A) [28–30]. Interestingly, several other stress conditions caused C. crescentus cells to respond in a clearly different manner. Most conspicuously, upon treatment with 100 mM NaCl or 4% ethanol (EtOH) cells transformed into long filaments and accumulated multiple chromosomes (Fig 1A). Consistent with the flow cytometry data, we found by using a fluorescent repressor-operator system (FROS), which fluorescently marks the origins of replication [31], that the elongated cells contained multiple well-segregated origins (S1 Fig), demonstrating that cells continue to undergo DNA replication, chromosome segregation and cellular growth under these conditions, but that they stop dividing. Cells treated with NaCl or ethanol increased to up to eight to ten-fold of their original length and often accumulated six to seven chromosomes within eight hours, suggesting that growth and the cell cycle continued for three to four doubling times in the absence of division. The effects on cell division were observed within two hours of NaCl or EtOH treatment, however the phenotype became more pronounced over time (Fig 1B). We also observed that the filamentous phenotype only occurred in a narrow range of NaCl or EtOH concentrations (Fig 1B), in which cells are still viable and able to grow (S2 Fig).
Similar to the phenotypic response observed upon NaCl or EtOH treatment, we observed that incubation at 40°C also caused cell elongation. At this temperature C. crescentus is still able to grow and to replicate its DNA, whereas temperatures above 40°C lead to a growth and DNA replication arrest (Fig 1, S2 Fig). The chromosome content was not as strongly increased at 40°C as under the EtOH and NaCl stress conditions.
We did not observe significant changes in cell morphology or chromosome content when cultivating C. crescentus at different acidic or alkaline pHs (Fig 1A). Treatment of cells with sublethal doses of H2O2 led to an increased proportion of cells (54.1%) with a DNA content that equals a single chromosome compared to the non-stress condition (38.2%) (Fig 1A), suggesting that under oxidative stress many cells arrest in G1-phase, similar to starvation conditions, stationary phase and acute heat shock.
To assess whether cell filamentation in response to salt, ethanol or mild heat shock stress is reversible, we followed the fate of stress-treated filamentous cells by time-lapse microscopy upon shifting them to non-stress conditions. Filamentous cells from each of the three conditions were able to resume growth and to initiate cell division at multiple sites shortly after release to fresh growth medium with first cell division events occurring within two hours (Fig 2). Within six hours most bacteria were able to propagate and divide normally yielding typically sized Caulobacter daughter cells. Notably, a fraction of daughter cells that arose from the filamentous cells maintained elongated until the end of our time-lapse study, indicating that these cells were severely damaged.
Altogether our data demonstrate that C. crescentus changes its morphology and cell cycle status in response to changing external conditions. Different stress conditions affect the cell cycle at different stages, either by transiently blocking DNA replication initiation and cellular growth or by delaying cell division.
The filamentous phenotype of cells treated with salt, ethanol and mild heat shock is similar to the phenotype of cells treated with DNA damaging agents such as mitomycin C (Fig 1A) [26]. Previously, it has been shown that in C. crescentus and other bacteria DNA damage induces the expression of small division inhibitors, which directly interfere with components of the cell division apparatus and thereby block the process of cell division in response to DNA damage [1,26,27]. To test whether the transient block of cell division upon treatment with NaCl, EtOH or increased temperature is due to induction of the SOS response we monitored the promoter activity of the sidA gene, which encodes the SOS induced division inhibitor in C. crescentus [26]. In contrast to mitomycin C treatment, which caused strong induction of PsidA-gfp within two hours, the reporter was not turned on upon exposure to NaCl or EtOH stress (Fig 3A and 3B, S3 Fig). A mild induction in a subpopulation of cells was observed at 40°C, which was however clearly lower compared to mitomycin treatment. These data show that the filamentous phenotype induced by salt stress, ethanol stress or mild heat shock does not appear to depend on the SOS response and SidA.
Another previous study reported that a subpopulation of C. crescentus cells form helical filaments during long-term growth in stationary phase [32]. It was shown that in these filaments the level of the major cell division protein FtsZ was strongly reduced [32]. Therefore we wanted to test if FtsZ abundance and localization was affected under NaCl stress, EtOH stress and mild heat shock. However, Western Blot analysis showed that FtsZ protein abundance was not significantly altered within eight hours of stress treatment (Fig 3D). Furthermore, using a strain expressing ftsZ-eYFP showed that FtsZ still localized in distinct foci along the length of the cells (Fig 3C), suggesting that the observed block of cell division is not caused by a failure of FtsZ to localize to division sites.
To better understand the molecular basis underlying the observed stress-induced morphological changes, we analyzed global changes in gene expression by RNA-sequencing (RNA-seq). We focused on EtOH and NaCl stress as the effect on cell division was most pronounced under these conditions.
Treatment with 100 mM NaCl for 60 minutes resulted in a >2-fold induction of 472 genes (11.6%) and a >2-fold downregulation of 282 genes (6.9%). In the case of EtOH stress, 570 genes were >2-fold upregulated (14.0%) and 489 genes (12.0%) were >2-fold downregulated after 60 minutes. Statistical analysis revealed that in each case, the gene expression profile after 60 minutes was highly similar to the profile obtained after 30 minutes of stress treatment with significant scores (z-scores) greater than 40 (Fig 4A). We also observed a strong overlap when comparing the gene expression data sets from the NaCl and EtOH conditions to each other for both upregulated and downregulated genes (z-scores >20), suggesting that salt and ethanol stress result in similar changes in gene expression. Consistent with our data obtained with the PsidA-gfp reporter, the SOS regulon was not strongly induced under ethanol and salt stress (Fig 4A).
Interestingly, among the most downregulated genes were many involved in flagella biosynthesis, pili biogenesis, chemotaxis, and cell cycle progression (Fig 4B). In C. crescentus, these genes are under the direct control of the master cell cycle regulator CtrA [8]. Consistently, we found that gene expression changes in a divLts mutant, which upon shift to 37°C results in loss of CtrA function [22,33], were similar to the gene expression changes induced by salt and ethanol stress in wild type cells (Fig 4A and 4B). Genes that were downregulated in divLts cells, for example flagella, pili and chemotaxis genes, cell cycle regulators (ccrM, sciP, divK) or cell division genes (ftsA, ftsQ, ftsW) also showed significant downregulation in response to NaCl or EtOH treatment. By contrast, CtrA repressed genes, which were upregulated in the divLts mutant also showed upregulation in the NaCl treated and EtOH treated cultures. These data demonstrate that the regulon of CtrA is strongly influenced in response to salt and ethanol stress, suggesting that CtrA activity is affected under these conditions. Loss of CtrA function results in strong chromosome accumulation and cell filamentation [13,22], phenotypes similar to those observed during salt and ethanol stress (Fig 4C).
The observed effect on the expression of CtrA-regulated genes strongly points to altered CtrA function in response to certain stress conditions. Consistently, when we monitored steady-state protein levels of CtrA by Western blotting, we found that the level of CtrA strongly and rapidly decreased upon exposure to salt, ethanol stress or mild heat shock (Fig 5A). Most conspicuously, treatment with 4% EtOH resulted in complete elimination of CtrA within only 15 minutes. Exposure to NaCl or 40°C for 60 minutes led to a drop in CtrA levels to 15% and 10%, respectively. By contrast, in response to mitomycin-induced DNA damage CtrA levels were not significantly affected. In a strain depleted of the protease subunit ClpX and in a ΔcpdR mutant CtrA was stabilized (Fig 5B), indicating that the ClpXP protease along with the CpdR adaptor, which are responsible for the cell cycle-dependent degradation of CtrA under optimal growth conditions [12,17], are also required to rapidly downregulate CtrA in response to environmental stress.
To test if the stress-induced reduction of CtrA steady-state levels was caused by increased proteolysis, we measured CtrA degradation rates by in vivo stability assays under the different stress conditions. The degradation rate of CtrA depended strongly on the external condition under which cells were cultured (Fig 5C). Under optimal growth conditions CtrA had a half-life of approximately 26 minutes, similar to previously published results [22]. Exposure to increased external salt concentrations, or incubation at 40°C led to a strong decrease in half-life to 12.8 minutes or five minutes, respectively. Most remarkably, upon incubation with 4% EtOH the half-life was as short as two minutes (Fig 5C, S4 Fig). The decrease in CtrA stability under these conditions contrasts the changes in CtrA stability induced by carbon starvation [29]; under this condition the half-life was prolonged to 173 minutes (Fig 5C). Taken together, our data demonstrate that salt, ethanol and mild heat shock induce rapid proteolysis of CtrA. More generally, our data indicate that the rate of CtrA proteolysis is subject to modulation by environmental signals allowing for the integration of external information into the cell cycle.
How do environmental stress conditions affect the rate of CtrA proteolysis? Because CtrA stability is tightly linked to its phosphorylation [15,17], we thought that the observed stress-induced decrease in CtrA stability might result from its dephosphorylation and inactivation involving the CckA-ChpT phosphorelay. To test this possibility, we measured CtrA activity following stress treatment in the ΔcpdR strain, in which CtrA is stabilized but growth rate unchanged under non-stress conditions (Fig 5B, S5 Fig). As a read-out for CtrA's transcription factor activity we measured the expression of CtrA-dependent genes by quantitative RT-PCR. Our data showed that as in wild type cells the mRNA abundance of the genes sciP, fljO, ccrM, ftsQ and pilA rapidly dropped in ΔcpdR cells following EtOH or NaCl addition (Fig 5D), suggesting that despite being stabilized, CtrA is not able to activate the expression of these genes after stress exposure. In agreement with our qRT-PCR data, we found that the phenotype of the ΔcpdR strain was indistinguishable from the filamentous phenotype of wild type cells upon treatment with EtOH or NaCl (Fig 5E, Fig 1A). These data let suggest that the increased proteolysis of CtrA under stress conditions follows its inactivation as a transcription factor.
Under non-stress conditions CckA dynamically switches between its kinase and phosphatase activities [16,34]. We reasoned that the observed inactivation of CtrA could either be due to complete loss of CckA function or to a more active mechanism, in which CckA switches from a kinase into a phosphatase. To distinguish between these possibilities we investigated if CckA phosphatase activity is required to induce rapid downregulation of CtrA upon EtOH treatment, the stress condition that impacted CtrA stability most strongly. To this end, we used a point mutant of CckA, CckA(V366P), that retains kinase activity, but lacks significant phosphatase activity [16]. Indeed, constitutive expression of this mutant in a cckA deletion background partially prevented CtrA downregulation upon EtOH addition, whereas expression of wild type CckA from the same construct led to rapid removal of CtrA similar to the wild type (Fig 6A). Monitoring CtrA degradation rates in cckA(V366P)-expressing cells showed that the half-life of CtrA during EtOH exposure was markedly increased in the mutant (7.7 min) compared to the wild type (2.2 min) (Fig 6B). Similar results were also obtained under NaCl stress (S6 Fig). Together, these data indicate that CckA phosphatase activity is critical to ensure rapid degradation of CtrA in response to external stress.
We next tested if increasing the kinase activity of CckA results in a similar effect by using a CckA variant containing the substitution G319E, which has been shown to render CckA hyperactive as a kinase [16]. Because constitutive expression of cckA(G319E) leads to severe cell cycle and growth defects, we expressed this variant from an inducible promoter in a wild type background. Western blots showed that CtrA levels were completely stabilized in this mutant following EtOH treatment (Fig 6A). Moreover, the half-life of CtrA in cells expressing this mutant was strongly increased in the presence of stress (22.3 min) compared to the half-life in cells expressing wild type CckA (4.5 min) (Fig 6B, S6 Fig). Consistent with the strong stabilization of CtrA in cckA(G319E)-expressing cells we observed that cells neither formed filaments nor accumulated chromosomes when treated with salt, EtOH or mild heat shock, but instead arrested with a single chromosome in G1-phase (Fig 6C, S7 Fig). Expression of a stable and active ctrA allele, ctrA(D51E)Δ3Ω [13] led to a similar phenotype: cells no longer accumulated multiple chromosomes in response to EtOH and NaCl stress and instead arrested in G1-phase (Fig 6D, S8 Fig). These data suggest that the replication phenotype of cckA(G319E) cells is due to increased CtrA phosphorylation and activity. We observed that stress-induced cell filamentation and chromosome accumulation was also less severe in cckA(V366P) cells than in the wild type (Fig 6C, S7 Fig). However, compared to the cckA(G319E) and the ctrA(D51E)Δ3Ω mutants, the change of phenotype was not as strong, likely because CtrA was only partially stabilized in the cckA(V366P) mutant (Fig 6A).
Together our data are consistent with a model, in which environmental stress causes CckA to switch to phosphatase activity, leading to CtrA dephosphorylation and rapid proteolysis. As an additional verification of this model, we analyzed CtrA activity by measuring its occupancy at the origin of replication (Cori) upon stress treatment by using quantitative chromatin immunoprecipitation (qChIP). We compared CtrA occupancy between the ΔcpdR mutant and the cckA(G319E) mutant, in both of which CtrA is completely stbilized upon stress exposure (Fig 5B, Fig 6A). In the ΔcpdR mutant addition of EtOH led to a decrease in CtrA occupancy to 18% (Fig 6E), indicating that stress treatment results in CtrA dephosphorylation and inactivation. By contrast, in cells expressing cckA(G319E), CtrA occupancy at Cori remained high even after stress exposure (Fig 6E). These results reinforce our conclusion that stress triggers the inactivation and proteolysis of CtrA by stimulating CckA phosphatase activity.
How is CckA phosphatase activity stimulated under stress conditions? Previous work has reported different mechanisms that allow CckA to switch between its kinase and its phosphatase mode under standard conditions [22,24]. One such mechanism depends on the upstream regulatory factors DivL and DivK. While DivL directly interacts with CckA at the swarmer pole in stalked and predivisional cells and activates its kinase activity [23], DivK acts as an antagonist of DivL-CckA favoring the phosphatase mode of CckA at the stalked pole [22].
Consistent with DivL's role in activating CckA kinase activity, previous work showed that loss of DivL function leads to a strong reduction in the level of phosphorylated CtrA and rapid CtrA degradation [22]. In agreement with these data we found that shifting a divLts mutant strain to the restrictive temperature resulted in rapid proteolysis and downregulation of CtrA (Fig 7A and 7B) [22]. Because the rate of CtrA degradation in the divLts strain was similarly fast as in wild type cells upon EtOH treatment, we hypothesized that stress-dependent inactivation and proteolysis of CtrA might be caused by a failure to accumulate and localize DivL. However, our microscopy data showed that fluorescently tagged DivL (DivL-GFP) correctly localized in a cell cycle-dependent manner even in the presence of EtOH (Fig 7C, S9 Fig). Similarly, we did not observe that EtOH treatment induced significant changes in the localization pattern of CckA that changes during the cell cycle (Fig 7D, S9 Fig) [35]. By contrast, loss of DivL function has previously been shown to cause mislocalization of CckA [22,23]. Together, these data suggest that CtrA inactivation in response to adverse conditions is not due to failure to accumulate or localize DivL.
To investigate whether the switch of CckA into its phosphatase mode is caused by increased activity of DivK, we measured CtrA levels following EtOH treatment in a strain in which DivK was inactivated. When shifting a cold-temperature sensitive mutant of DivK (divKcs) to the restrictive temperature 20°C [22,36], we observed the same rapid downregulation of CtrA upon EtOH treatment as in the wild type (Fig 7E), suggesting that the stress-dependent regulation of CtrA activity is independent of DivK.
In addition to the upstream regulators DivK and DivL, it has recently been shown that c-di-GMP can modulate CckA activity through a direct interaction. Binding of c-di-GMP to CckA inhibits the kinase and stimulates the phosphatase activity of CckA [24]. To test the possibility that EtOH affects CckA activity through c-di-GMP, we investigated if the EtOH-dependent changes in CtrA abundance and stability were abolished in a CckA(Y514D) mutant that is compromised for c-di-GMP binding [24]. However, just like in the wild type, addition of EtOH resulted in rapid CtrA downregulation in this mutant (Fig 7E). Similarly, we found that in a strain lacking all diguanylate cyclases (cdG0) that is devoid of c-di-GMP [25], EtOH addition resulted in the same drop in CtrA abundance as in the wild type (Fig 7E). Consistent with the downregulation of CtrA in divKcs, CckA(Y514D) and cdG0 cells, we observed that EtOH-treatment led to the similar phenotypic changes as observed in wild type cells; cells became filamentous and accumulated multiple chromosomes (Fig 7F, Fig 1A).
Altogether these data suggest that external EtOH stress affects CckA and CtrA activity neither via the DivK-DivL pathway nor through c-di-GMP signaling. Hence, the histidine kinase might directly respond to external changes.
We wondered whether the stress-dependent downregulation of CtrA provides a selective advantage and helps the culture to survive and grow under adverse conditions. To investigate the physiological relevance of the rapid shut-down of CtrA function during ethanol stress, we compared the growth rate of wild type cells with that of cckA(V366P) mutant cells, in which CtrA is partially stabilized due to lower CckA phosphatase activity (Fig 6A and 6B). Under optimal conditions, the wild type and the mutant had identical growth rates suggesting that CckA phosphatase activity is not critical for cell growth and cell cycle progression in the absence of stress (Fig 8A). Interestingly however, in the presence of 4% EtOH the cckA(V366P) mutant strain showed a clearly reduced growth rate compared to the wild type. This result suggests that under stress conditions the shut-down of CtrA function due to increased CckA phosphatase activity indeed provides a growth advantage.
This work reports a new mechanism by which bacteria delay cell division and consequently transform into filamentous cells under stress. Unlike previously described mechanisms that transiently block cell division through the induction of small division inhibitors [26,27], the mechanism that we describe depends on a phospho-signaling system that modulates the stability and activity of a master cell cycle regulator required for cell division.
The bifunctional histidine kinase CckA plays a central role in this regulation as it makes the decision of whether to divide or not by integrating the cell cycle with environmental information. It is well established that under optimal conditions CckA drives oscillations of the master cell cycle regulator CtrA through dynamically switching between its kinase and phosphatase activities [16,22,34,37]. Our new data suggest that environmental stress locks CckA in its phosphatase mode leading to the rapid inactivation of CtrA, its elimination through the protease ClpXP and consequently a block of CtrA regulated functions including cell division (Fig 8B). Although our data show that CckA phosphatase activity is critical for the stress-dependent inactivation of CtrA, the detailed molecular process by which stress signals modulate CckA function remains to be elucidated. Our results rule out the involvement of the small signaling molecule c-di-GMP (Fig 7E and 7F), which promotes CckA phosphatase activity at the G1-to-S transition and promotes CtrA degradation by ClpXP under non-stress conditions [21,24,37]. Similarly, the stress-dependent regulation does not appear to be mediated by the upstream regulators DivL and DivK (Fig 7), which are critical for the cell cycle-dependent regulation of CckA in the absence of stress [22].
The result that none of the known CckA regulators were involved in the stress-dependent regulation of CckA together with the rapid response time that we observed (Fig 5) argues for a direct sensing mechanism. CckA has two PAS (Per Arnt Sim) domains, PAS-A and PAS-B, which sense distinct spatiotemporal signals and thereby mediate the cell cycle-dependent regulation of CckA activity [34]. These PAS domains might also perceive information about the environment, for example by binding molecules that are present under certain conditions. Alternatively, stress-sensing could be mediated in a fashion independent of CckA's PAS domains and instead involve its periplasmic or transmembrane regions. Indeed, previous work on other kinases demonstrated that environmental information can be directly sensed through the membrane [38]. For example, the histidine kinase DesK from Bacillus subtilis was shown to respond to temperature changes by sensing membrane thickness [39]. It is well documented that salt, EtOH and increased temperature induce changes in membrane properties, for example membrane fluidity or lipid composition [40–42]. These changes might directly induce changes in CckA conformation and activity. A direct sensing mechanism by CckA would provide an efficient means to transduce environmental information into the cell cycle. Nevertheless, we do not rule out the involvement of unidentified regulatory proteins that may interact with CckA to control its activity in response to stress.
Other studies have analyzed the response of C. crescentus to increased salt concentrations. One of them investigated gene expression and proteome changes upon treatment with 60 mM NaCl [43]. Under this condition growth rate and CtrA regulated genes were hardly affected [43], which is consistent with our finding that the salt-induced filamentous phenotype occurred only in a narrow range of concentrations (Fig 1B). Another recent study investigated the response of individual cells of C. crescentus to repeated salt exposure using a microfluidics system [44]. The authors observed that an initial exposure to moderate NaCl concentrations (80 mM) led to a cell division delay and cell-cycle synchronization and that the response of individual cells to a subsequent exposure to a higher NaCl concentration (100 mM) was dependent on the cell cycle state [44]. It is possible that the stress-induced changes in CtrA activity that we report here contribute to these behaviors.
Noticeably, while salt stress, EtOH stress and mild heat shock lead to rapid elimination of CtrA, our data as well as previously published results show that carbon starvation causes an increase in CtrA stability by a mechanism involving the small signaling molecule (p)ppGpp [28,29,45]. Although the precise mechanism of the starvation-dependent increase in CtrA stability remains unclear, it likely ensures, in combination with the downregulation of the DNA replication initiator DnaA, a block of DNA replication initiation under this condition [1,28,29].
Besides elucidating the mechanisms of how cell division is environmentally controlled, another important question concerns why cells inhibit cell division under stress. Cell division is a vulnerable process involving extensive membrane and cell wall remodeling [46]. Initiating this process under stress conditions, in particular those impacting the cell membrane, potentially causes cell lysis and consequently death. Preventing cell division in the presence of stress might thus provide a mechanism to preserve cell integrity. Continued global macromolecule synthesis, which still can take place under the conditions that we described (Fig 1A, S2 Fig), allows for the production of new cell mass, enabling the rapid generation of new daughter cells when conditions improve. It is also possible that the filamentous morphology of division-inhibited cells provides an adaptive advantage under certain conditions in nature. A filamentous cell shape is expected to influence various cell properties, including surface area, mobility and adhesive forces, and is thus expected to affect the interaction with other species and the attachment of cells to biotic and abiotic surfaces [3,4].
Finally, while we have focussed in this study on C. crescentus, other alphaproteobacteria might employ similar mechanisms to control CtrA and cell division in response to external cues. Previous work in the nitrogen-fixing plant symbiont Sinorhizobium meliloti demonstrated that CtrA is strongly downregulated during the early steps of symbiosis [47]. In the pathogen Brucella abortus the CckA-ChpT-CtrA-CpdR pathway was shown to be required for intracellular survival in human macrophages [48]. Therefore, precise environmental regulation of CtrA abundance and activity likely plays an important role for the diverse functions that different alphaproteobacteria perform in the environment.
Wild type C. crescentus NA1000 and its mutant derivatives were routinely grown in PYE (rich medium) or M2G medium (minimal medium containing 0.2% glucose). When necessary, growth medium was supplemented with 0.3% xylose, 0.2% glucose or 1 mM IPTG. Cultures were grown at 30°C with 200 rpm, temperature sensitive mutants were cultivated at 30°C and sensitivity was induced either at 37°C or 20°C depending on the mutant allele. Antibiotics were added as previously described [30,49]. Rifampicin was used at concentrations of 2.5 μg/ml (liquid media) and 5 μg/ml (solid media) for C. crescentus and 25 μg/ml (liquid media) and 50 μg/ml (solid media) for E. coli. E. coli strains were routinely grown in LB medium at 37°C, supplemented with antibiotics as required.
For induction of stress, mixed (non-synchronized) Caulobacter cultures grown overnight to exponential phase in PYE medium (OD 0.1–0.4) were shifted to medium supplemented with NaCl, EtOH, mitomycin C, sucrose or H2O2 at the indicated concentrations. To induce heat shock, cultures grown at 30°C were diluted in pre-heated medium and cultivated at 37, 39, 40, 42 or 45°C for the indicated time. Carbon starvation was induced by shifting cells from M2G to M2 medium supplemented with 0.02% glucose as previously described [28]. Low and high pH stress medium was prepared by adjusting the pH with HCl or Na2CO3 and NaHCO3 [32] to the pH values of 4.9 and 9.1, respectively. If necessary, cultures were backdiluted during the stress treatment to keep them in exponential phase.
Strains used in this study are listed in S1 Table. Strain ΔcpdR::rif, KJ798, was created by using the two-step recombination procedure [50] with plasmid pKJ808. Strain KJ799 was generated by introducing the plasmid pKJ809 into C. crescentus NA1000 by electroporation. To construct strain KJ800, plasmid pKJ810 was introduced instead. To construct strain KJ811 the empty vector pJS14 was introduced into C. crescentus NA1000 by electroporation.
Samples were prepared as earlier described [28] and analyzed using a BD LSRFortessa flow cytometer or the BD LSR II (BD Biosciences). Data were collected for at least 30000 cells. Flow cytometry data were analyzed with FlowJo. Each experiment was repeated independently and representative results are shown.
For living cell analysis and time-lapse microscopy, cells were transferred onto a PYE 1% agarose pad with supplementation as required. Otherwise cells were fixed with 1% formaldehyde, pelleted, resuspended in an appropriate volume of ddH2O and mounted onto 1% agarose pads. Phase contrast and fluorescence images were taken using a Ti eclipse inverted research microscope (Nikon) with a 100x/1.45 NA objective (Nikon). Fiji (ImageJ) was used for image processing.
One ml culture was harvested and resuspended in 1 ml ddH2O. The optical density and the GFP fluorescence of 100 μl cells was measured in a SpectraMax i3x (Molecular Devices) plate reader. The fluorescence / OD ratio was calculated after blanking and the WT auto fluorescence signal was subtracted from the GFP signal.
Pelleted cells were resuspended in 1X SDS sample buffer, normalized to the optical density of the culture and heated to 95°C for 10 min. Protein extracts of cell lysates were then subject to SDS-PAGE for 90 min at 130 V at room temperature on 11% Tris-glycine gels and transferred to nitrocellulose membranes. To verify equal loading, total protein was visualized using the TCE in-gel method [52] prior to blotting. Proteins were detected using primary antibodies against CtrA (kindly provided by M. Laub), DnaK or FtsZ (kindly provided by M. Thanbichler) in appropriate dilutions, and a 1:5000 dilution of secondary HRP-conjugated antibody. SuperSignal Femto West (Thermo Scientific) was used as detection reagent. Blots were scanned with a Chemidoc (Bio-Rad) system. Images were processed with Bio-Rad Image Lab, Adobe Photoshop, Image J and the relative band intensities quantified with Image Lab software.
To measure protein degradation in vivo, cells were grown under the desired conditions. Protein synthesis was blocked by addition of 100 μg/ml chloramphenicol. Samples were taken as indicated, every 10 min (for 1 hour) or at the time points 0, 3, 6, 9, 12, 20, and 30 minutes, and pellets were snap frozen in liquid nitrogen before being analyzed by Western blotting.
RNA was collected from bacteria that were grown under the appropriate conditions and extracted using the RNeasy mini kit (Qiagen). RNA-sequencing was performed by GENEWIZ, South Plainfield, NJ. For statistical analysis, the transcriptome data of the EtOH and NaCl stress conditions were compared to each other, to divLts transcriptome data and to previously published DNA damage microarray [26]. In order to integrate RNA-seq and microarray transcriptomics data we applied a fold-change cut-off of two to identify up- and down-regulated genes. For each condition, genes with a fold-change > 2 in the stress / non-stress sample comparison were assigned to the up-regulated group and genes with a fold-change < 0.5 to the down-regulated group. The similarity between two conditions is reflected by the extent of intersection of the genes in the respective up-regulated and down-regulated gene groups. As the intersection depends strongly on the size of the groups we used Monte-Carlo simulations to statistically evaluate group intersections. The z-score indicates the number of standard deviations from a random intersection. Hence, a high z-score means a strong deviation from a random similarity value.
Samples were prepared as previously described [53] with the following modification: Lysates from one independent experiment were sonicated simultaneously (Bioruptor Plus, diagenode, Ougrée, BE) using 40 cycles of 30 seconds sonication (High) and 30 seconds pause. Real-time PCR was performed with a StepOnePlus real-time PCR system (AppliedBiosystems, Foster City, CA) using 5% of each ChIP sample and the SYBR green PCR master mix (Bio-Rad) in a 20 μl volume and 10 pmol primers (OKH79 and OKH80), amplifying a 88 bp region spanning the c and d CtrA binding boxes within Cori. The cycle threshold (Ct) of the input DNA was adjusted to 100%. The percentage of the input DNA was calculated (100*2^(adjusted input DNA-Ct(IP))) for every condition and mutant. Each qPCR reaction was performed in triplicate.
RNA was collected from bacteria that were grown under the appropriate conditions as described above. Equal amounts of isolated RNA were reverse transcribed into cDNA using the iScript cDNA synthesis kit (Bio-rad). The cDNA was used as template for the real-time PCR reaction using the iTaq universal SYBR Green Supermix (Bio-rad) and primers as listed in S2 Table. Analysis was performed with a qTower instrument (Analytik Jena) using the standard run mode. For detection of primer dimerization or other artifacts of amplification, a dissociation curve was run immediately after completion of the real-time PCR. Individual gene expression profiles were normalized against 16S RNA, serving as an endogenous control. Relative expression levels were determined with the comparative Ct method. Each qPCR reaction was performed in triplicate.
Complete data sets for the RNA-seq experiment are provided in S3 Table in the supplemental material and are available through GEO (accession number GSE90030).
|
10.1371/journal.pcbi.1007273 | Integrating Hi-C links with assembly graphs for chromosome-scale assembly | Long-read sequencing and novel long-range assays have revolutionized de novo genome assembly by automating the reconstruction of reference-quality genomes. In particular, Hi-C sequencing is becoming an economical method for generating chromosome-scale scaffolds. Despite its increasing popularity, there are limited open-source tools available. Errors, particularly inversions and fusions across chromosomes, remain higher than alternate scaffolding technologies. We present a novel open-source Hi-C scaffolder that does not require an a priori estimate of chromosome number and minimizes errors by scaffolding with the assistance of an assembly graph. We demonstrate higher accuracy than the state-of-the-art methods across a variety of Hi-C library preparations and input assembly sizes. The Python and C++ code for our method is openly available at https://github.com/machinegun/SALSA.
| Hi-C technology was originally proposed to study the 3D organization of a genome. Recently, it has also been applied to assemble large eukaryotic genomes into chromosome-scale scaffolds. Despite this, there are few open source methods to generate these assemblies. Existing methods are also prone to small inversion errors due to noise in the Hi-C data. In this work, we address these challenges and develop a method, named SALSA2. SALSA2 uses sequence overlap information from an assembly graph to correct inversion errors and provide accurate chromosome-scale assemblies.
| Genome assembly is the process of reconstructing a complete genome sequence from significantly shorter sequencing reads. Most genome projects rely on whole genome shotgun sequencing which yields an oversampling of each genomic locus. Reads originating from the same locus are identified using assembly software, which can use these overlaps to reconstruct the genome sequence [1, 2]. Most approaches are based on either a de Bruijn [3] or a string graph [4] formulation. Repetitive sequences exceeding the sequencing read length [5] introduce ambiguity and prevent complete reconstruction. Unambiguous reconstructions of the sequence are output as “unitigs” (or often “contigs”). Ambiguous reconstructions are output as edges linking unitigs. Scaffolding utilizes long-range linking information such as BAC or fosmid clones [6, 7], optical maps [8–10], linked reads [11–13], or chromosomal conformation capture [14] to order and orient contigs. If the linking information spans large distances on the chromosome, the resulting scaffolds can span entire chromosomes or chromosome arms.
Hi-C is a sequencing-based assay originally designed to interrogate the 3D structure of the genome inside a cell nucleus by measuring the contact frequency between all pairs of loci in the genome [15]. The contact frequency between a pair of loci strongly correlates with the one-dimensional distance between them. Hi-C data can provide linkage information across a variety of length scales, spanning tens of megabases. As a result, Hi-C data can be used for genome scaffolding. Shortly after its introduction, Hi-C was used to generate chromosome-scale scaffolds [16–20].
LACHESIS [16] is an early method for Hi-C scaffolding that first clusters contigs into a user-specified number of chromosome groups and then orients and orders the contigs in each group independently to generate scaffolds. Thus, the scaffolds inherit any assembly errors present in the contigs. The original SALSA1 [21] method first corrects the input assembly, using a lack of Hi-C coverage as evidence of error. It then orients and orders the corrected contigs to generate scaffolds. Recently, the 3D-DNA [20] method was introduced and demonstrated on a draft assembly of the Aedes aegypti genome. 3D-DNA also corrects the errors in the input assembly and then iteratively orients and orders contigs into a single megascaffold. This megascaffold is then broken, identifying chromosomal ends based on the Hi-C contact map.
There are several shortcomings common across currently available tools. They are sensitive to input assembly contiguity and Hi-C library variations and require tuning of parameters for each dataset. Inversions are common when the input contigs are short, as orientation is determined by maximizing the interaction frequency between contig ends across all possible orientations [16]. When contigs are long, there are few interactions spanning the full length of the contigs, making the true orientation apparent from the higher weight of links. However, in the case of short contigs, there are interactions spanning the full length of the contig, making the true orientation have a similar weight to incorrect orientations. Biological factors, such as topologically associated domains (TADs), also confound this analysis [22].
SALSA1 [21] addressed some of these challenges, such as not requiring the expected number of chromosomes beforehand and correcting assemblies before scaffolding them with Hi-C data. We showed that SALSA1 worked better than the most widely used method, LACHESIS [16]. However, SALSA1 often did not generate chromosome-sized scaffolds. The contiguity and correctness of the scaffolds depended on the coverage of Hi-C data and required manual data-dependent parameter tuning. Building on this work, SALSA2 does not require manual parameter tuning and is able to utilize all the contact information from the Hi-C data to generate near optimal sized scaffolds permitted by the data using a novel iterative scaffolding method. In addition to this, SALSA2 enables the use of an assembly graph to guide scaffolding, thereby minimizing errors, particularly orientation errors.
SALSA2 is an open source software that combines Hi-C linkage information with the ambiguous-edge information from a genome assembly graph to better resolve contig orientations. We propose a novel stopping condition, which does not require an a priori estimate of chromosome count, as it naturally stops when the Hi-C information is exhausted. We show that SALSA2 produces fewer orientation, ordering, and chimeric errors across a wide range of assembly contiguities. We also demonstrate its robustness to different Hi-C libraries with varying levels of intra-chromosomal contact frequencies. When compared to 3D-DNA, SALSA2 generates more accurate scaffolds across most conditions. To our knowledge, this is the first method to leverage assembly graph information for scaffolding Hi-C data.
Fig 1(A) shows the overview of the SALSA2 pipeline. SALSA2 begins with a draft assembly generated from long reads such as Pacific Biosciences [23] or Oxford Nanopore [24]. SALSA2 requires the contig sequences and, optionally, a GFA-formatted assembly graph [25] representing the ambiguous reconstructions. Hi-C reads are aligned to the contig sequences, and contigs are optionally split in regions lacking Hi-C coverage. A hybrid scaffold graph is constructed using both ambiguous edges from the GFA and edges from the Hi-C reads, scoring edges according to a “best buddy” scheme. Scaffolds are iteratively constructed from this graph using a greedy weighted maximum matching. A mis-join detection step is performed after each iteration to check if any of the joins made during this round are incorrect. Incorrect joins are broken and the edges blacklisted during subsequent iterations. This process continues until the majority of joins made in the prior iteration are incorrect. This provides a natural stopping condition, when accurate Hi-C links have been exhausted. Below, we describe each of the steps in detail.
Hi-C methods first crosslink a sample (cells or tissues) to preserve the genome conformation. The crosslinked DNA is then digested using multiple restriction enzymes (targeting in this case the restriction sites GATC and GANTC). The single-stranded 5’-overhangs are then filled in causing digested ends to be labeled with a biotinylated nucleotide. Next, spatially proximal digested ends of DNA are ligated, preserving both short- and long-range DNA contiguity. The DNA is then purified and sheared to a size appropriate for Illumina short-read sequencing. After shearing, the biotinylated fragments are enriched to assure that only fragments originating from ligation events are sequenced in paired-end mode via Illumina sequencers to inform DNA contiguity.
Hi-C paired end reads are aligned to contigs using the BWA aligner [26](parameters: -t 12 -B 8) as single end reads. First, the reads mapping at multiple locations are ignored as they can cause ambiguities while scaffolding. Reads which align across ligation junctions are chimeric and are trimmed to retain only the start of the read which aligns prior to the ligation junction. After filtering the chimeric reads, the pairing information is restored. Any PCR duplicates in the paired-end alignments are removed using Picard tools [27]. Read pairs aligned to different contigs are used to construct the initial scaffold graph. The suggested mapping pipeline is available at http://github.com/ArimaGenomics/mapping_pipeline.
As any assembly is likely to contain mis-assembled sequences, SALSA2 uses the physical coverage of Hi-C pairs to identify suspicious regions and break the sequence at the likely point of mis-assembly. We define the physical coverage of a Hi-C read pair as the region on the contig spanned by the start of the leftmost fragment and the end of the rightmost fragment. A drop in physical coverage indicates a likely assembly error. In SALSA1, contigs are split when a fixed minimum coverage threshold is not met. A drawback of this approach is that coverage can vary, both due to sequencing depth and variation in Hi-C link density.
Fig 2 sketches the new contig correction algorithm implemented in SALSA2. Instead of the single coverage threshold used in SALSA1, a set of suspicious intervals is found with a sweep of thresholds. For a sweep of thresholds, we find the continuous stretches of regions which have lower physical coverage. Note that there can be multiple intervals for a particular threshold that have multiple stretches of low coverage. In such case, we only consider the interval of the maximum size. These intervals denote the regions of potential misassembly on the contig. Using the collection of these intervals as an interval graph, we find the maximal clique, which is the maximal set of intervals intersecting at any location along the contig. This maximal clique represents the region of the contig which had low coverage for the majority of the tested cutoffs. This can be done in O(NlogN) time, where N is the number of intervals. For a maximal clique, the region between the start and end of the smallest interval in the clique is flagged as a mis-assembly and the contig is split into three pieces—the sequence to the left of the region, the junction region itself, and the sequence to the right of the region. The intuition behind choosing the smallest interval is to accurately pinpoint the location of assembly error. Note that this algorithm finds only one misassembly per contig. For more rigorous misassembly detection, same algorithm can be run multiple times on each contig until no more drops in physical coverage are found.
For our experiments, we use the unitig assembly graph produced by Canu [28] (Fig 1(C)), as this is a more conservative assembly output than contig sequences that represent various traversals of this graph. SALSA2 requires only a GFA format [25] representation of the assembly. Since most long-read genome assemblers such as FALCON [29], miniasm [25], Canu [28], and Flye [30] provide assembly graphs in GFA format, their output is compatible with SALSA2 for scaffolding.
The scaffold graph is defined as G(V, E), where nodes V are the ends of contigs and edges E are derived from the Hi-C read mapping (Fig 1B). The idea of using contig ends as nodes is similar to that used by the string graph formulation [4].
Modeling each contigs as two nodes allows a pair of contigs to have multiple edges in any of the four possible orientations (forward-forward, forward-reverse, reverse-forward, and reverse-reverse). The graph then contains two edge types—one explicitly connects two different contigs based on Hi-C data, while the other implicitly connects the two ends of the same contig.
As in SALSA1, we normalize the Hi-C read counts by the frequency of restriction enzyme cut sites in each contig. This normalization reduces the bias in the number of shared read pairs due to the contig length as the number of Hi-C reads sequenced from a particular region are proportional to the number of restriction enzyme cut sites in that region. For each contig, we denote the number of times a cut site appears as C(V). We define edges weights of G as:
W ( u , v ) = N ( u , v ) C ( u ) + C ( v )
where N(u, v) is the number of Hi-C read pairs mapped to the ends of the contigs u and v. By the ends, we mean the first and second half of the contig if divided at the midpoint along its length.
We observed that the globally highest edge weight does not always capture the correct orientation and ordering information due to variations in Hi-C interaction frequencies within a genome. To address this, we defined a modified edge ratio, similar to the one described in [20], which captures the relative weights of all the neighboring edges for a particular node.
The best buddy weight BB(u, v) is the weight W(u, v) divided by the maximal weight of any edge incident upon nodes u or v, excluding the (u, v) edge itself. Computing best buddy weight naively would take O(|E|2) time. This is computationally prohibitive since the graph, G, is usually dense. If the maximum weighted edge incident on each node is stored with the node, the running time for the computation becomes O(|E|). We retain only edges where BB(u, v) > 1. This keeps only the edges that are the best incident edge on both u and v. Once used, the edges are removed from subsequent iterations. Thus, the most confident edges are used first but initially low-scoring edges can become best in subsequent iterations.
For the assembly graph, we define a similar ratio. Since the edge weights are optional in the GFA specification and do not directly relate to the proximity of two contigs on the chromosome, we use the graph topology to establish this relationship. Let u ¯ denote the reverse complement of the contig u. Let σ(u, v) denote the length of shortest path between u and v. For each edge (u, v) in the scaffold graph, we find the shortest path between contigs u and v in every possible orientation, that is, σ(u, v), σ ( u , v ¯ ), σ ( u ¯ , v ) and σ ( u ¯ , v ¯ ). With this, the score for a pair of contigs is defined as follows:
S c o r e ( u , v ) = min x ′ ∈ { u , u ¯ } - { x } , y ′ ∈ { v , v ¯ } - { y } σ ( x ′ , y ′ ) min x ∈ { u , u ¯ } , y ∈ { v , v ¯ } σ ( x , y )
where x and y are the orientations in which u and v are connected by a shortest path in the assembly graph. Essentially, Score(u, v) is the ratio of the length of the second shortest path to the length of the shortest path in all possible orientations. Once again, we retain edges where Score(u, v) > 1. If the orientation implied by the assembly graph differs from the orientation implied by the Hi-C data, we remove the Hi-C edge and retain the assembly graph edge (Fig 1D). Computing the score graph requires |E| shortest path queries, yielding total runtime of O(|E|*(|V| + |E|)) since we do not use the edge weights.
Once we have the hybrid graph, we lay out the contigs to generate scaffolds. Since there are implicit edges in the graph G between the beginning and end of each contig, the problem of computing a scaffold layout can be modeled as finding a weighted maximum matching in a general graph, with edge weights being our ratio weights. In a weighted maximum matching, a set of edges from a graph is chosen in such a way that they have no endpoints common and the sum of edge weights is maximized. In the case of scaffolding, a maximum weighted matching implies a layout of contigs, where no end can be used twice, that is maximally consistent with the data being used for scaffolding (Hi-C in our case). If we find the weighted maximum matching of the non-implicit edges (that is, edges between different contigs) in the graph, adding the implicit edges to this matching would yield a complete traversal. However, adding implicit edges to the matching can introduce a cycle. Such cycles are prevented by removing the lowest-weight non-implicit edge. Computing a maximal matching takes O(|E||V|2) time [31]. We iteratively find a maximum matching in the graph by removing nodes found in the previous iteration. Using the optimal maximum matching algorithm this would take O(|E||V|3) time, which would be extremely slow for large graphs. Instead, we use a greedy maximal matching algorithm which is guaranteed to find a matching within 1/2-approximation of the optimum [32]. The greedy matching algorithm takes O(|E|) time, thereby making the total runtime O(|V||E|). The algorithm for contig layout is sketched in Algorithm 1. Fig 1(D)–1(F) show the layout on an example graph. Contigs which were not scaffolded are inserted in the large scaffolds with the method used in SALSA1. If unitigs are used as an input and the layout of unitigs along contigs is provided as an input to SALSA2, it can replace unitig sequences by contigs in the final scaffolds.
Algorithm 1 Contig Layout Algorithm
E: Edges sorted by the best buddy weight
M: Set to store maximal matchings
G: The scaffold graph
while all nodes in G are not matched do
M* = {}
for e ∈ E sorted by best buddy weights do
if e can be added to M* then
M* = M* ∪ e
end if
end for
M = M ∪ M*
Remove nodes and edges which are part of M* from G
end while
Since the contig layout is greedy, it can introduce errors by selecting a false Hi-C link which was not eliminated by our ratio scoring. These errors propagate downstream, causing large chimeric scaffolds and chromosomal fusions. We examine each join made within all the scaffolds in the last iteration for correctness. Any join with low spanning Hi-C support relative to the rest of the scaffold is broken and the links are blacklisted for further iterations.
We compute the physical coverage spanned by all read pairs aligned in a window of size w around each join. For each window, w, we create an auxiliary array, which stores −1 at position i if the physical coverage is greater than some cutoff δ and 1, otherwise. We then find the maximum sum subarray in this auxiliary array, since it captures the longest stretch of low physical coverage. If the position being tested for a mis-join lies within the region spanned by the maximal clique generated with the maximum sum subarray intervals for different cutoffs (Fig 2), the join is marked as incorrect. The physical coverage can be computed in O(w + N) time, where N is the number of read pairs aligned in window w. The maximum sum subarray computation takes O(w) time. If K is the number of cutoffs(δ) tested for the suspicious join finding, then the total runtime of mis-assembly detection becomes O(K(N + 2*w)). The parameter K controls the specificity of the mis-assembly detection, thereby avoiding false positives. The algorithm for mis-join detection is sketched in Algorithm 2. When the majority of joins made in a particular iteration are flagged as incorrect by the algorithm, SASLA2 stops scaffolding and reports the scaffolds generated in the penultimate iteration as the final result.
Algorithm 2 Misjoin detection and correction algorithm
Cov: Physical coverage array for a window size w around a scaffold join at position p on a scaffold
A: Auxiliary array
I: Maximum sum subarray intervals
for δ ∈ {min_coverage, max_coverage} do
if Cov[i] ≤ δ then
A[i] = 1
else
A[i] = −1
end if
sδ, eδ = maximum_sum_subarray(A)
I = I∪{sδ, eδ}
end for
s, e = maximal_clique_interval(I)
if p ∈ {s, e} then
Break the scaffold at position p
end if
We created artificial assemblies, each containing contigs of same size, by splitting the GRCh38 [33] reference into fixed-sized contigs of 200 to 900 kbp. This gave us eight assemblies. The assembly graph for each input was built by adding edges for any adjacent contigs in the genome. So the simulated assembly graph was linear with edges between two adjacent contigs for each contig in the graph.
For real data, we use the recently published NA12878 human dataset sequenced with Oxford Nanopore [34] and assembled with Canu [28]. We use a Hi-C library from Arima Genomics (Arima Genomics, San Diego, CA) sequenced to 40x Illumina coverage (SRX3651893). Table 1 shows the statistics for this library. We compare results with the original SALSA(commit—833fb11), SALSA2 with and without the assembly graph input(commit—68a65b4), and 3D-DNA (commit—3f18163). We did not compare our results with LACHESIS because it is no longer supported and is outperformed by 3D-DNA [20]. SALSA2 was run using default parameters, with the exception of graph incorporation, as listed. For 3D-DNA, alignments were generated using the Juicer alignment pipeline [35] with defaults (-m haploid -t 15000 -s 2), except for mis-assembly detection, as listed. A genome size of 3.2 Gbp was used for contiguity statistics for all assemblies.
For evaluation, we also used the GRCh38 reference to define a set of true and false links from the Hi-C graph. We mapped the assembly to the reference with MUMmer3.23 (nucmer -c 500 -l 20) [36] and generated a tiling using MUMmer’s show-tiling utility. For this “true link” dataset, any link joining contigs in the same chromosome in the correct orientation was marked as true. This also gave the true contig position, orientation, and chromosome assignment. We masked sequences in GRCh38 that matched known structural variants from a previous assembly of NA12878 [37] to avoid counting true variations as scaffolding errors.
Table 2 lists the metrics for NA12878 scaffolds. We include an idealized scenario, using only reference-filtered Hi-C edges for comparison. As expected, the scaffolds generated using only true links had the highest NGA50 value and longest error-free scaffold block. SALSA2 scaffolds were generally more accurate and contiguous than the scaffolds generated by SALSA1 and 3D-DNA, even without use of the assembly graph. The addition of the graph further improved the NGA50 and longest error-free scaffold length.
We also evaluated the assemblies using Feature Response Curves (FRC) based on scaffolding errors [40]. An assembly can have a high raw error count but still be of high quality if the errors are restricted to only short scaffolds. FRC captures this by showing how quickly error is accumulated, starting from the largest scaffolds. Fig 5(D) shows the FRC for different assemblies, where the X-axis denotes the cumulative % of assembly errors and the Y-axis denotes the cumulative assembly size. The assemblies with more area under the curve accumulate fewer errors in larger scaffolds and hence are more accurate. SALSA2 scaffolds with and without the graph have similar areas under the curve and closely match the curve of the assembly using only true links. The 3D-DNA scaffolds have the lowest area under the curve, implying that most errors in the assembly occur in the long scaffolds. This is confirmed by the lower NGA50 value for the 3D-DNA assembly (Table 2).
Apart from the correctness, SALSA2 scaffolds were highly contiguous and reached an NG50 of 112.8 Mbp (cf. GRCh38 NG50 of 145 Mbp). Fig 6 shows the alignment ideogram for the input contigs as well as the SALSA2 assembly. Every color change indicates an alignment break, either due to error or due to the end of a sequence. The input contigs are fragmented with multiple contigs aligning to the same chromosome, while the SALSA2 scaffolds are highly contiguous and span entire chromosomes in many cases. Fig 7(A) shows the contiguity plot with corrected NG stats. As expected, the assembly generated with only true links has the highest values for all NGA stats. The curve for SALSA2 assemblies with and without the assembly graph closely matches this curve, implying that the scaffolds generated with SALSA2 are approaching the optimal assembly of this Arima-HiC data.
We also evaluated the ability of scaffolding short-read assemblies for both 3D-DNA and SALSA2. We did not include SALSA1 in this comparison because it is not designed to scaffold short-read assemblies. We observed that use of the assembly graph when scaffolding significantly reduced the number of orientation errors for SALSA2, increasing the scaffold NGA50 and largest chunk almost two-fold. When compared to 3D-DNA without input assembly correction, SALSA2 with the assembly graph generates scaffolds of much higher NGA50 (7.99 Mbp vs. 1.00 Mbp). The number of scaffolding errors across all the categories was much lower in SALSA2 compared to 3D-DNA.
We computed the CPU runtime and memory usage for both the methods while scaffolding long and short read assemblies with Arima-HiC data. We excluded the time required to map reads in both cases. While scaffolding the long-read assembly SALSA2 was 30-fold faster and required 3-fold less memory than 3D-DNA (11.44 CPU hours and 21.43 Gb peak memory versus 3D-DNA with assembly correction at 318 CPU hours and 64.66 Gb peak memory). For the short-read assembly, the difference in runtime was even more pronounced. SALSA2 required 36.8 CPU hours and 61.8 Gb peak memory compared to 2980 CPU hours and 48.04 Gb peak memory needed by 3D-DNA without assembly correction. When run with assembly correction, 3D-DNA ran over 14 days on a 16-core machine without completing so we could not report any results.
We next tested scaffolding using two libraries with different Hi-C contact patterns. The first, from [42], is sequenced during mitosis. This removes the topological domains and generates fewer off-diagonal interactions. The other library was from [43], are in vitro chromatin sequencing library (Chicago) generated by Dovetail Genomics (L1). It also removes off-diagonal matches but has shorter-range interactions, limited by the size of the input molecules. As seen from the contact map in Fig 8, both the mitotic Hi-C and Chicago libraries follow different interaction distributions than the standard Hi-C (Arima-HiC in this case). Table 1 shows the mapping statistics for these libraries. We ran SALSA2 with defaults and 3D-DNA with both the assembly correction turned on and off.
For mitotic Hi-C data, we observed that the 3D-DNA mis-assembly correction algorithm sheared the input assembly into small pieces, which resulted in more than 25,000 errors and more than half of the contigs incorrectly oriented or ordered. Without mis-assembly correction, the 3D-DNA assembly has a higher number of orientation (250 vs. 81) and ordering (215 vs. 54) errors compared to SALSA2. The feature response curve for the 3D-DNA assembly with breaking is almost a diagonal (Fig 5(B)) because the sheared contigs appeared to be randomly joined. SALSA2 scaffolds contain longer stretches of correct scaffolds compared to 3D-DNA with and without mis-assembly correction (Fig 7(B)). SALSA1 scaffolds had a similar error count to SALSA2 but were less contiguous.
For the Chicago libraries, 3D-DNA without correction had the best NGA50 and largest correct chunk. However, the scaffolds had more chimeric join errors than SALSA2. SALSA2 outperformed 3D-DNA in terms of NG50, NGA50, and longest chunk when 3D-DNA was run with assembly correction. 3D-DNA uses signatures of chromosome ends [20] to identify break positions which are not prominently present in Chicago data. As a result, it generated more chimeric joins compared to SALSA2. However, the number of order and orientation errors was similar across the methods. Since Chicago libraries do not provide chromosome-spanning contact information for scaffolding, the NG50 value for SALSA2 is 5.8 Mbp, comparable to the equivalent coverage assembly (50% L1+L2) in [43] but much smaller than Hi-C libraries. Interestingly, SALSA1 was able to generate scaffolds of similar contiguity to SALSA2, which can be attributed to the lack of long range contact information in the library. SALSA2 is robust to changing contact distributions. In the case of Chicago data it produced a less contiguous assembly due to the shorter interaction distance. However, it avoids introducing false chromosome joins, unlike 3D-DNA, which appears tuned for a specific contact model.
To evaluate the effectiveness of SALSA2 on a non-model organism, we used Hi-C data from recently published Anopheles funestus genome assembly which was scaffolded using an independent method (Phase Genomics or LACHESIS [16]) and manually curated using Illumina mate-pair support as well as FISH information [44]. This genome had high heterozygosity as the data was sequenced from a colony of mosquitoes rather than a single individual. Due to this, the assembly had a high duplication rate and was almost double the expected genome size. We scaffolded both the full assembly and the assembly after running purge haplotigs [45] using SALSA2 and 3D-DNA. For the post purge assembly, 3D-DNA generated an assembly with higher continuity but with more errors and a similar NA50 to SALSA2. (S3 Table). However, neither method performed well for the full assembly. SALSA2 was more contiguous than 3D-DNA (S4 Table) but was still very fragmented and much larger than the expected genome size. We conclude that heterozygous genome scaffolding remains a challenge and assemblies must either be de-duplicated beforehand or improved algorithms for scaffolding, such as [46] are needed.
In this work, we present the first Hi-C scaffolding method that integrates an assembly graph to produce high-accuracy, chromosome-scale assemblies. Our experiments on both simulated and real sequencing data for the human genome demonstrate the benefits of using an assembly graph to guide scaffolding. We also show that SALSA2 outperforms alternative Hi-C scaffolding tools on assemblies of varied contiguity, using multiple Hi-C library preparations.
SALSA2’s misassembly correction and scaffold misjoin validation can be improved in several ways. The current implementation does not detect a misjoin between two small contigs with high accuracy, mainly because Hi-C data does not have enough resolution to correct such errors. Also, we do not account for any GC bias correction when using the Hi-C coverage for detecting misjoins. Addressing these challenges in misjoin detection and misassembly correction is the immediate next step to improve the SALSA2 software.
The human genome is relatively homozygous compared to many other species. Assembly of many species is further complicated by DNA input requirements which necessitates pooling multiple individuals. SALSA2 does not remove duplication present in an input assembly and thus requires pre-processing by another tool, such as Purge Haplotigs [45] or haplomerger [47]. Once contigs are classified into the “primary” and “haplotig” sets, SALSA2 could be run on each of the sets independently.
Hi-C scaffolding has been historically prone to inversion errors when the input assembly is highly fragmented. The integration of the assembly graph with the scaffolding process can overcome this limitation. Orientation errors introduced in the assembly and scaffolding process can lead to incorrect identification of structural variations. On simulated data, more than 50% of errors were due to inversions, and integrating the assembly graph reduced these by as much as 3 to 4 fold. We did not observe as much improvement with the NA12878 test dataset because the contig NG50 was much higher than in the simulation. However, it is not always possible to assemble multi-megabase contigs. In such cases, the assembly graph is useful for limiting Hi-C errors.
Most existing Hi-C scaffolding methods also require an estimate for the number of chromosomes for a genome. This is implicitly taken to be the desired number of scaffolds to output. As demonstrated by the Chicago, mitotic, and replicate [48] Hi-C libraries, the library as well as the genome influences the maximum correct scaffold size. It can be impractical to sweep over hundreds of chromosome values to select a “best” assembly. Since SALSA2’s mis-join correction algorithm stops scaffolding after the useful linking information in a dataset is exhausted, no chromosome count is needed as input.
Obtaining the chromosome-scale picture of the genome is important and there is a trade-off between accuracy and continuity of the assembly. However, we believe that manual curation to remove assembly errors is an expensive and involved process that can often outpace the cost of the rest of the project. Most of the assembly projects using Hi-C data have had a significant curation effort to date [19, 49]. Thus, we believe that not introducing errors in the first place is a better strategy to avoid the later burden of manual curation of small errors in chromosomes. The Hi-C data can be used with other independent technologies, such as optical mapping or linked-reads to reach accurate chromosome-scale scaffolds. 3D-DNA was recently updated to not require the chromosome count as input but the algorithm used has not been described. Interestingly, it no longer generates single-chromosome scaffolds in our experiments, a major claim in [20], supporting a conservative scaffolding approach. Even while scaffolding short-read assemblies, we observed that SALSA2 generated more accurate scaffolds than 3D-DNA, indicating the utility of SALSA2 in scaffolding existing short-read assemblies of different genomes with the newly generated Hi-C data.
As the Genome10K consortium [50] and independent scientists begin to sequence novel lineages in the tree of life, it may be impractical to generate physical or genetics maps for every organism. Thus, Hi-C sequencing combined with SALSA2 presents an economical alternative for the reconstruction of chromosome-scale assemblies.
|
10.1371/journal.pntd.0003204 | Assessment of Anthelmintic Efficacy of Mebendazole in School Children in Six Countries Where Soil-Transmitted Helminths Are Endemic | Robust reference values for fecal egg count reduction (FECR) rates of the most widely used anthelmintic drugs in preventive chemotherapy (PC) programs for controlling soil-transmitted helminths (STHs; Ascaris lumbricoides, Trichuris trichiura, and hookworm) are still lacking. However, they are urgently needed to ensure detection of reduced efficacies that are predicted to occur due to growing drug pressure. Here, using a standardized methodology, we assessed the FECR rate of a single oral dose of mebendazole (MEB; 500 mg) against STHs in six trials in school children in different locations around the world. Our results are compared with those previously obtained for similarly conducted trials of a single oral dose of albendazole (ALB; 400 mg).
The efficacy of MEB, as assessed by FECR, was determined in six trials involving 5,830 school children in Brazil, Cambodia, Cameroon, Ethiopia, United Republic of Tanzania, and Vietnam. The efficacy of MEB was compared to that of ALB as previously assessed in 8,841 school children in India and all the above-mentioned study sites, using identical methodologies.
The estimated FECR rate [95% confidence interval] of MEB was highest for A. lumbricoides (97.6% [95.8; 99.5]), followed by hookworm (79.6% [71.0; 88.3]). For T. trichiura, the estimated FECR rate was 63.1% [51.6; 74.6]. Compared to MEB, ALB was significantly more efficacious against hookworm (96.2% [91.1; 100], p<0.001) and only marginally, although significantly, better against A. lumbricoides infections (99.9% [99.0; 100], p = 0.012), but equally efficacious for T. trichiura infections (64.5% [44.4; 84.7], p = 0.906).
A minimum FECR rate of 95% for A. lumbricoides, 70% for hookworm, and 50% for T. trichiura is expected in MEB-dependent PC programs. Lower FECR results may indicate the development of potential drug resistance.
| Soil-transmitted helminths (STHs; roundworms, whipworms, and hookworms) infect millions of children in sub-tropical and tropical countries, resulting in malnutrition, growth stunting, intellectual retardation, and cognitive deficits. To fight against STH, large-scale deworming programs are implemented in which anthelmintic drugs (either albendazole (ALB) or mebendazole (MEB)) are administered. Currently, these large-scale programs are intensifying, highlighting the need to closely monitor the efficacy of anthelmintic drugs to detect changes in drug efficacy that may arise through the evolution of anthelmintic drug resistance in the parasites. We have previously defined the minimum expected efficacy of ALB based on the fecal egg count reduction (FECR) rate, but these reference values are lacking for MEB. Therefore, we therefore evaluated the FECR rate of MEB against STHs in six STH endemic countries. In addition, we compared the results of the FECR rate for MEB with those we obtained previously for ALB. The results confirm that MEB treatment was highly efficacious against roundworms, and to a lesser extend against hookworms, but not against whipworms. Compared to ALB, MEB is less efficacious against hookworm, but equally efficacious against roundworms and whipworms. Based on this study we propose the minimum expected FECR rate for MEB-dependent large-scale deworming programs.
| The soil-transmitted helminths (STHs, Ascaris lumbricoides, Trichuris trichiura, and the two hookworm species, Necator americanus and Ancylostoma duodenale) are responsible for the highest burden among all neglected tropical diseases (NTDs) [1]. Recent global estimates indicate that in 2010 more than 1.4 billion people were infected with at least one of the four STH species, resulting in a global burden of approximately 5.2 million disability-adjusted life years (DALYs) (20% of the total number of DALYs attributable to NTDs) [2]. Mass drug administration (MDA) programmes in which a single oral dose of albendazole (ALB) or mebendazole (MEB) - the drugs of choice for STHs - are periodically administered to pre-school and school aged children, are the main strategy for controlling the morbidity caused by STH [3], [4], and these programmes have recently received increased political and scientific attention [5], [6].
While the laudable long-term aim is to eliminate soil-transmitted helminthiasis as public problem by 2020 [7], the pledges of drug donations on this scale re-enforce the necessity for thoroughly designed monitoring systems that allow detection of any changes in anthelmintic drug efficacy that may arise through the evolution of anthelmintic drug resistance in these parasites. Both ALB and MEB belong to the same pharmaceutical group (benzimidazole drugs, BZ) sharing the same mode of action (the inhibition of the polymerisation of microtubules). Thus, development of resistance against any BZ drug would most likely by accompanied by poor anthelmintic drug efficacy of the other BZ drugs. It is pertinent also that there is a paucity of anthelmintic drugs licensed for the treatment of STH infections in humans and available commercially, and hence should anthelmintic resistance against BZ drugs eventually emerge and spread, chemotherapy based control of STHs will be even more limited than at present with few acceptable alternative options [8].
Currently, assessment of the reduction in fecal egg counts following drug administration (fecal egg count reduction (FECR) syn. egg reduction rate) is the recommended method for monitoring the efficacy of anthelmintic drugs against STHs [9]. In contrast to other available assays, it allows for the assessment of drug efficacy against all three STHs (vs. in vitro assays and molecular assays) with a minimum of laboratory equipment (vs. molecular assays) [10]–[12]. However, the interpretation of the results from the FECR tests remains difficult, since reliable reference efficacy values for BZ drugs (which can act as standard reference points for comparison with new data) for each of the STH species are still lacking. In a systematic review and meta-analysis of published efficacy trials targeting STH infections, Keiser and Utzinger (2008) [13] highlighted that there is a lack of high-quality trials to determine these reference values. The available efficacy data have been obtained through a variety of widely differing study protocols, including protocols that used different diagnostic methods, different durations in follow-up periods, differing origin of the drugs (i.e., different manufactures and therefore different quality), and statistical analyses, all of which impede a robust meta-analysis of drug efficacy based on FECR [13]. As a response to these earlier limitations, our consortium has recently reported on the efficacy of a single oral dose of ALB (400 mg) against STH in seven trials across sub-Saharan Africa, Asia, and Latin America based on a protocol standardized in respect to the origin of the drug (Zentel, GlaxoSmithKline, batch N° L298), the follow-up period (14 to 30 days), the egg counting method (McMaster egg counting method), the statistical analysis, and the interpretation of the data (group based FECR using arithmetic means) [14]. This study suggested that a FECR of 95% for A. lumbricoides and 90% for hookworm should be the expected minimum in all future drug efficacy studies, and that FECR rates below these levels following a single dose of ALB, should be viewed as danger signs of potential development of drug resistance. For T. trichiura, reference FECR values could only be provided for specified mean fecal egg count (FEC) values at pre-intervention, as the drug efficacy measured by FECR decreased as a function of increasing mean FEC at the pre-intervention survey. For T. trichiura, we therefore expect FECRs of at least 90%, 70%, and 50%, in populations where the mean FECs are below 275 eggs per 1 g of stool (EPG), 550 EPG, and 800 EPG, respectively, and even lower in settings where baseline infection intensities are higher [15]. Data derived from such standardized multi-center efficacy trials that establish reliable reference FECR values and assess the impact of infection intensity are still currently lacking for MEB. Therefore, in the present study we assessed the efficacy measured by means of FECR of a single oral dose of MEB (500 mg) against STHs in six trials in sub-Saharan Africa, Asia, and Latin America using a protocol that we previously standardized in assessing the drug efficacy of a single dose of ALB, and compared the drug efficacy of both BZ drugs against each of the STHs.
The overall protocol of both the ALB and MEB trials was approved by the Ethics Committee of the Faculty of Medicine, Ghent University (reference nos. B67020084254 and 2011/374) and was followed by local ethical approval for each trial site. For Brazil, approval was obtained from the Institutional Review Board from Centro de Pesquisas René Rachou (no. 21/2008), for Cambodia from the National Ethics Committee for Health Research (no. 185 NECHR), for Cameroon from the National Ethics Committee (nos. 072/CNE/DNM08, 147/CNE/DNM/11), for India from the Institutional Review Board of the Christian Medical College (Vellore) (no. 6541; participated in the ALB study only), for Ethiopia from the Ethical Review Board of Jimma University (Jimma) (no. RPGE/09/2011), for United Republic of Tanzania from the Zanzibar Health Research Council (nos. 20, ZAMREC/0003/JUNE/2012), and for Vietnam by Ethical Committee of National Institute of Malariology, Parasitology and Entomology (Ha Noi) and the Ministry of Health (no. 752/QD-VSR). The parents of all subjects included in the studies signed an informed consent form. In Brazil and Ethiopia an informed consent form was obtained from children aged 10 or 11 years and above. In Cambodia and Ethiopia, a verbal assent was obtained from all children, and this procedure was approved by the respective ethics boards. Our studies assessing ALB and MEB are registered under ClinicalTrials.gov, identifiers nos. NCT01087099 and NCT01379326, respectively (CONSORT Checklist S1).
The MEB multi-center study reported here was carried out in six countries located in sub-Saharan Africa (Cameroon, Ethiopia, and United Republic of Tanzania (Zanzibar)), Asia (Cambodia and Vietnam), and Latin America (Brazil). However, it is important to note that, while we refer to individual countries to identify results from particular trials, names of countries are used only to distinguish between six separate trials that were conducted in six geographically distinct regions of the world. These six STH-endemic countries were selected because of the presence of investigator groups/institutions with extensive experience in the diagnosis, and control of STH. These same six investigator groups were also involved in the earlier evaluation of the efficacy of a single-oral dose 400 mg ALB against STH in children, based on an identical standardized protocol [14].
Since study designs can have a significant effect on the subsequent calculation of efficacy and to ensure that our values for ALB and MEB were not confounded by study design, the protocol described by Vercruysse and colleagues for assessment of ALB was also used here to evaluate the drug efficacy of MEB [14]. In short, schools were selected based on previous STH surveys. Within schools, children were recruited on a voluntary basis. School children aged 4–18 years at each of the different trial sites were asked to provide a stool sample during a pre-intervention survey. For the initial sampling the aim was to enroll at least 250 STH-infected children for at least one species. This sample size was selected based on statistical analysis of study power, using random simulations of correlated over-dispersed FEC data reflecting the variance-covariance structure in a selection of real FEC data sets. This analysis suggested that a sample size of up to 200 individuals (alpha = 0.05, power = 80%) was required to detect a 10% point drop from a null efficacy of 80% (mean FECR individual) over a wide range of infection scenarios. Standard power analyses for proportions also indicated that the detection of a 10% point drop from a null cure rate required sample sizes up to 200 (the largest samples being required to detect departures from null efficacies of around 50%). Given an anticipated non-compliance rate of 25%, a total of at least 250 individuals was therefore considered necessary at each study site. A single oral dose of 500 mg MEB (Vermox) from the same manufacturer (Janssen-Cilag, Latina, Italy, batch no: BCL2F00) was administered to the subjects at all study sites. Seven to 15 days after the pre-intervention survey (Brazil: 7–14 days; Cambodia: 11–15 days; Cameroon: 9–11 days; Ethiopia: 14 days; United Republic of Tanzania: 14 days; Vietnam: 11–12 days), stool samples were again collected from the subjects. Subjects who were unable to provide a stool sample at follow-up, or who were experiencing a severe intercurrent medical condition or had diarrhea at the time of the first sampling, were excluded from the study (Study Protocol S1).
All stool samples were individually processed by the McMaster egg counting method. McMaster is a flotation technique that is commonly used in veterinary parasitology both to assess intensity of gastro-intestinal parasite infections and to evaluate drug efficacy against these parasites. For the diagnosis and enumeration of STHs in public health, it has been found to be user-friendly (vs. FLOTAC [16]) robust (vs. Kato-Katz thick smear [17]) and accurate for enumeration of STHs, but less sensitive when intensity of infection is low (vs. Kato-Katz and FLOTAC [16], [17])) Complementary data indicate that FECR estimates obtained by the McMaster are comparable to those using the Kato-Katz thick smear [18], [19].
The standard operating procedure to perform a McMaster on human stools is described in more detail elsewhere [15]. Briefly, 2 g of stool were suspended in 30 ml of saturated salt (NaCl) solution at room temperature (density: 1.2). The fecal suspension was poured three times through a tea sieve to remove large debris. After thorough mixing 10 times, 0.5 ml aliquots were added to each side of a McMaster slide chamber. Both chambers were examined under a light microscope using 100x magnification and the FEC, expressed as EPG for each helminth species, was obtained by multiplying the total number of eggs counted under the microscope by a factor 50. A detailed tutorial can be found on http://www.youtube.com/watch?v=UZ8tzswA3tc.
The efficacy of a single dose of MEB (500 mg) against each of the three STH species (the two hookworm species were treated as one species since the eggs of A. duodenale and N. americanus cannot be distinguished by conventional microscopy), as measured by FECR, was calculated for the different trials. We have not summarized the efficacy of MEB by means of cure rate (CR; the proportion of the subjects who are not excreting eggs after drug administration). This is because an intervention may fail to cure STH infections (CR = 0%), but may result in a FECR of 99%, which is satisfactory. Second, it has been shown that estimates of CR are highly affected by both sampling and diagnostic effort, estimates being overestimated when the sampling and diagnostic effort is minimized. This was in sharp contrast to FECR estimates, which remained unchanged regardless of both sampling and diagnostic effort [20].
To-date, a wide range of formulae has been used to calculate FECR, each differing in terms of the statistical unit (individual vs. group) and how the mean FEC is calculated (arithmetic vs. geometric). However, recent studies suggest that the group-based formula using the arithmetic mean, as described below, is a suitable metric for evaluating drug efficacy. Compared to the other formulae, it represents a robust indicator (vs. individual-based formula [14]) that provides accurate estimates of drug efficacy (vs. group-based formula using geometric mean) [14], [21]. Moreover, it is important to note that there is no common formula for calculating the variance for each of these FECR formulae. The formula used here to calculate variance for the group-based FECR using the arithmetic mean is described below and is based on the Delta method [22].
We first analyzed the trial outcomes for each of the three STHs separately using FECR data at the trial level, and its corresponding variance and sample size, and secondly we compared data from both the current trials with MEB and our earlier trials with ALB, employing meta-analytical approaches. In addition, the impact of infection intensity at the pre-intervention survey on the FECR rate of both BZ drugs was evaluated. To assess the FECR rate of MEB, generalized mixed effect models were fitted for each of the three STH species in turn, with the FECR rate at the trial level as the dependent variable, and the mean FEC at the pre-intervention survey as a covariate. To compare the FECR rates between ALB and MEB, generalized mixed effect models were fitted for each of the three STH species with FECR at the trial level as the dependent variable, and BZ drug (two levels: ALB and MEB) and mean FEC at the pre-intervention survey as covariates, and the interaction between these covariates. For both analyses, we only included trials for which at least 50 subjects who had been treated and provided stools at both pre- and post-intervention surveys were available. The different MEB and ALB trials included in the analyses are listed in Table 1 and Table 2, respectively. These tables also describe for each trial the sample size, mean age, sex ratio (number of female subjects/number of male subjects), mean FEC, and the level of infection intensity at the pre-intervention survey for each of the three STH species separately. The levels of infection intensity correspond to the low, moderate, and high intensity of infection ranges, as described by WHO [23]. For A. lumbricoides these were 1–4,999 EPG, 5,000–49,999 EPG, and ≥50,000 EPG; for T. trichiura these levels were 1–999 EPG, 1,000–9,999 EPG, and ≥10,000 EPG; and for hookworm these were 1–1,999 EPG, 2,000–3,999 EPG, and ≥4,000 EPG, respectively. The ALB trials have been described previously by Vercruysse et al. (2011) [14] and Mekonnen et al. (2013) [24], targeting A. lumbricoides (five trials), T. trichiura (five trials), and hookworm (7 trials). Each of these trials report FECR rates of a single dose ALB (400 mg), all were based on the aforementioned trial design, and with the exception of one trial (India) the same laboratories which assessed the FECR rates in current study were also involved in these ALB trials. The meta-analysis was carried out using the ‘metafor’ package of the statistical software R [25]. The level of significance was set at p<0.05.
The numbers of subjects enrolled for the study, allocated to the intervention, receiving treatment, and then providing samples at both pre- and post-intervention surveys are summarized in Figure 1. For infections with each species of STH we also provide the number of subjects who were included in the statistical analysis. Trials in which fewer than 50 subjects carried a specific parasite were not included in the statistical analysis.
Figure 2 illustrates the outcome of the analyses of FECR of MEB by means of forest plots for A. lumbricoides, T. trichiura, and hookworm infections. Overall, the estimated FECR rate [95% confidence interval [CI]] of MEB was the highest for A. lumbricoides (97.6% [95.8–99.5]), followed by hookworm (79.6% [71.0–88.3]). For T. trichiura, the estimated FECR rate was lower, namely 63.1% [51.6–74.6]. An association between the mean FECs at pre-intervention and the FECR rate was observed for A. lumbricoides only. For this STH species, the model predicted that the FECR rate would drop by 0.4% as mean FECs at pre-intervention increase by increments of 1,000 EPG (z = -2.97, p = 0.003). For the remaining two STHs, there was no significant relationship between FECR rate and mean FEC at pre-intervention (T. trichiura: z = 1.46, p = 0.144; hookworm: z = 1.00, p = 0.316).
Figures 3 to 5 illustrate the outcome of the meta-analyses of the FECR rate against STHs for ALB and MEB by means of forest plots for A. lumbricoides, T. trichiura, and hookworm infections, respectively. Overall, ALB resulted in statistically higher FECR rates against A. lumbricoides (ALB: 99.9% [99.0–100] vs. MEB: 98.0% [96.9–99.1], z = −2.5, p = 0.012) and hookworm infections (ALB: 96.2% [91.1–100] vs. MEB: 80.6% [74.4–86.8], z = 37.4, p<0.001). For T. trichiura there was no significant difference in FECR rate (ALB: 64.5% [44.4–84.7] vs. MEB: 62.7% [40.8–84.6], z = −0.1, p = 0.906).
Associations between mean FEC at pre-intervention and the FECR rate were observed for both A. lumbricoides and T. trichiura but in respect of different AE, (Figure 6). For A. lumbricoides, the model predicted that the FECR rate after treatment with ALB should remain unchanged across mean FEC at pre-intervention, but that the FECR rate after MEB treatment should fall on average by 0.4% for each 1,000 EPG incremental increase in mean FEC at pre-intervention (interaction term between BZ drug and mean FEC at the pre-intervention survey, z = −2.93, p = 0.033). For T. trichiura, the model predicted that the FECR rate following treatment with ALB should decrease on average by 7.8% per incremental increase of 100 EPG in mean FEC at pre-intervention ( = main effect of mean FEC at the pre-intervention, z = −9.1, p<0.001). However, there was a significant interaction between BZ drug and mean FEC at pre-intervention (z = 6.9, p<0.001), indicating a significant difference between the two BZ drugs in the rate of fall of the FECR rate with increasing pre-intervention FEC. In contrast to the 7.8% fall/100 EPG increment for ALB, for MEB, the model predicted a net fall in FECR rate of only 1.1%/100 EPG increment, which as shown earlier when assessed separately from ALB, did not represent a significant association. For hookworm, the FECR rate of both BZ drugs did not depend significantly on the mean FEC at pre-intervention.
This is the first study that has generated a robust, reliable estimation of the FECR rate following treatment with MEB, and has compared thoroughly the efficacy of MEB with that of ALB, against STH infections. Although we must acknowledge some variation in follow-up period across the trials, both the ALB and MEB trials were standardized at a level unprecedented in the scientific literature [14]. Moreover, most previous studies evaluating drug efficacy of BZ drugs against STHs have generally not summarized their efficacy results by means of the group-based FECR formula, using the arithmetic mean and its corresponding 95% CI, which are now recognized as a suitable, indeed the most informative metric, for the outcome of such trials [14] and are needed to enable a meta-analysis of drug efficacy against STHs [13], .
Overall, the results of this study indicate that a single oral dose of MEB is most efficacious against A. lumbricoides infections, followed by hookworm, but that it is relatively inefficacious for infections with T. trichiura, thus confirming the earlier efficacy studies reviewed by Bennett and Guyatt [26] and Keiser and Utzinger [13]. The relatively poor efficacy of a single dose treatment with either MEB or ALB in reducing T. trichiura FECs is not a novel finding [13], [26], and has resulted in ongoing research on the development and evaluation of new drugs or drug combinations that reduce T. trichiura worm burdens more effectively following single dose application, e.g. pyrantel/oxantel [27], mebendazole/ivermectine [28], oxantel, [29], and papaya cysteine proteinases [30]. Based on the overall drug efficacy results for the three STH species, we recommend that monitoring programs of single-dose MEB-dependent PC use a minimum FECR (group-based formula using arithmetic mean) of 95% for A. lumbricoides, 70% for hookworm, and 50% for T. trichiura as appropriate reference values (as they are below the lower limit of the 95% CI of overall estimates), and that efficacy levels below this should raise concern about the possible emergence of drug resistance.
Compared to a single oral dose of ALB, MEB was significantly less efficacious against hookworm and to a lesser extent against A. lumbricoides infections, but equally inefficacious for T. trichiura infection. In addition, the efficacies of both ALB and MEB were dependent on the intensity of A. lumbricoides and T. trichiura infection, decreasing with increasing infection intensity. However, the magnitude of this loss of efficacy as a function of increasing infection intensity differed between the two BZ drugs and the STH species. Between the BZ drugs, the change in drug efficacy was more pronounced for MEB with A. lumbricoides, whereas for T. trichiura the decrease was more pronounced for ALB. Among STHs, the overall impact of infection intensity on treatment with BZ was pronounced for T. trichiura (1.2–7.8% per 100 EPG), but almost negligible for A. lumbricoides (0.4% per 1,000 EPG). For hookworm, the efficacy did not depend on the infection intensity. This could be explained by either a true constant efficacy across infection intensities or a low number of moderate and high infection intensities in these trials (see Table 1). The bases of these differences in efficacy between the BZ drugs and their effects on STHs remain unclear, mainly due to the paucity of detailed pharmacokinetic and pharmacodynamics studies in pediatric populations in STH-endemic countries [31].
Currently, recommendations in PC programs are solely based on the overall prevalence of STHs, with these drugs being administered once a year when the STH prevalence is ≥20% and <50%, and twice a year when the prevalence is ≥50% [4]. Although this study indicates that the best choice of BZ drug depends on the relative prevalence and species of STH infections (ALB: hookworm> T. trichiura; MEB: T. trichiura> hookworm), practical experience with both drugs in the field over several years indicates that both are equally effective in controlling all three STH species irrespective of their initial prevalence and intensity of infection [32], [33]. However, future studies are required to (i) evaluate the difference between BZ drugs in long-term impact (prevalence, infection intensity, and occurrence of single nucleotide polymorphisms in the β-tubulin gene associated with BZ resistance); (ii) determine STH-specific thresholds for infection intensity to justify choice of BZ drugs; and (iii) assess the cost-effectiveness of distributing more than one class of BZ to different regions in a country [12], [34].
In conclusion, our findings suggest that FECR rates exceeding 95% for A. lumbricoides, 70% for hookworm, and 50% for T. trichiura should be expected in all future surveys, and that any FECR rate below these levels following a single oral dose of MEB (500 mg) should be viewed with concern in light of potential development of drug resistance. In addition, the study highlights the need for detailed pharmacokinetic/pharmacodynamic studies for single-oral dose of BZ drugs in pediatric populations in countries where STHs are endemic to gain additional insights into the observed differences in drug efficacy between ALB and MEB across the various STH species. Finally, additional recommendations advising those running PC programs about which of the BZ drugs to administer in a given setting (i.e., depending on the extent of T. trichiura and hookworm infections in a specific location/population) may improve the long-term benefits accruing from PC programs.
|
10.1371/journal.pgen.1000318 | Neuropeptide Y Gene Polymorphisms Confer Risk of Early-Onset Atherosclerosis | Neuropeptide Y (NPY) is a strong candidate gene for coronary artery disease (CAD). We have previously identified genetic linkage to familial CAD in the genomic region of NPY. We performed follow-up genetic, biostatistical, and functional analysis of NPY in early-onset CAD. In familial CAD (GENECARD, N = 420 families), we found increased microsatellite linkage to chromosome 7p14 (OSA LOD = 4.2, p = 0.004) in 97 earliest age-of-onset families. Tagged NPY SNPs demonstrated linkage to CAD of a 6-SNP block (LOD = 1.58–2.72), family-based association of this block with CAD (p = 0.02), and stronger linkage to CAD in the earliest age-of-onset families. Association of this 6-SNP block with CAD was validated in: (a) 556 non-familial early-onset CAD cases and 256 controls (OR 1.46–1.65, p = 0.01–0.05), showing stronger association in youngest cases (OR 1.84–2.20, p = 0.0004–0.09); and (b) GENECARD probands versus non-familial controls (OR 1.79–2.06, p = 0.003–0.02). A promoter SNP (rs16147) within this 6-SNP block was associated with higher plasma NPY levels (p = 0.04). To assess a causal role of NPY in atherosclerosis, we applied the NPY1-receptor–antagonist BIBP-3226 adventitially to endothelium-denuded carotid arteries of apolipoprotein E-deficient mice; treatment reduced atherosclerotic neointimal area by 50% (p = 0.03). Thus, NPY variants associate with atherosclerosis in two independent datasets (with strong age-of-onset effects) and show allele-specific expression with NPY levels, while NPY receptor antagonism reduces atherosclerosis in mice. We conclude that NPY contributes to atherosclerosis pathogenesis.
| Early-onset coronary artery disease (CAD) has a very strong genetic component as evidenced by the heritable nature of this disease. However, little is known about the actual genes underlying disease risk. Neuropeptide Y (NPY) is an abundant protein in humans that has been implicated in cardiovascular disease pathophysiology, but comprehensive evaluation of the gene coding for this protein has never been pursued in cardiovascular disease. Therefore, using gene-wide evaluation of variants within the NPY gene in a family-based as well as a non-familial study, we have shown that a cluster of six related NPY genetic variants is associated with early-onset CAD risk. We then show that one of these variants, which resides within the promoter region of this gene, is associated with higher NPY levels. Finally, to further support the functional role of this gene in CAD, we find that antagonism of the primary receptor of this gene results in marked attenuation of atherosclerosis in a mouse model. In conclusion, these findings demonstrate the role of the NPY gene in cardiovascular disease risk and add important additional information about the genetic architecture of this complex disease.
| The prevalence of early-onset cardiovascular disease in Americans (under 40 years of age) is approximately 10–15% [1] and the hereditary nature of coronary artery disease (CAD) is well-established [2]. The relative risk of developing CAD in a first degree relative is 3.8 to 12.1, with higher risk correlating with earlier age-of-onset [3]. Recent successes suggest that CAD genes can be identified through comprehensive genetic and functional studies [4]–[6]. However, the genetic architecture of CAD remains complex and poorly understood.
To identify genetic risk factors in early-onset CAD, we implemented a strategy combining the strengths of family-based studies with validation by case-control association, in conjunction with careful consideration of phenotype and functional data. In our own GENECARD linkage study of early-onset CAD, we have found five genomic regions of linkage with multipoint linkage odds (LOD) scores >1.0 [7]. The neuropeptide Y gene (NPY) is located adjacent to the peak microsatellite marker in the 7p14 peak. Because of its proximity to the peak marker, and because NPY has been implicated in disorders of vascular smooth muscle cell proliferation [8],[9], we sought to examine NPY further as a candidate gene for early-onset CAD.
NPY is the most abundant peptide in the heart and brain, and is produced by sympathetic neurons, endothelial cells [10], and platelets [11]. NPY has diverse functions including roles in sympathetic nerve stimulation through co-release with norepinephrine; immune function [12]; regulation of food consumption [9]; and modulation of heart rate, vasoconstriction, coronary blood flow and ventricular function [13]. These cardiovascular functions are primarily mediated through the NPY1 receptor [12],[14],[15]. In injured rat carotid arteries, non-atherosclerotic neointimal hyperplasia is aggravated by local delivery of NPY, and ameliorated by treatment with NPY1 receptor antagonist [8],[9]. In humans, NPY levels predict cardiovascular complications in end-stage renal disease [16], and NPY is implicated in congestive heart failure [17]. An NPY variant rare in most populations has been associated in Scandinavian populations with hyperlipidemia and carotid atherosclerosis [18],[19], CAD in type 1 diabetics [20], and MI in hypertensive patients [21]; however, the effects of this variant on NPY expression remain obscure.
To date there have been no systematic studies of the role of the NPY gene, or of the functional consequences of genetic variation at this locus, in cardiovascular disease pathogenesis. In response to the results of the genome-wide linkage analyses, the phenotypic correlations, and the strong but limited prior published work, we proposed to test the hypothesis that NPY variants affect atherosclerosis through effects on NPY plasma levels. We pursued a comprehensive gene-wide approach to correlating NPY variants with CAD and plasma NPY levels in humans, and tested the effects of NPY1 receptor blockade on atherosclerosis in mice.
Table 1 presents baseline characteristics of the three datasets: GENECARD (N = 946 affected, 37 unaffected individuals); CATHGEN (N = 556 cases, 256 controls); and GENECARD probands versus CATHGEN controls (N = 221 cases, 256 controls). Despite GENECARD families being selected on early age-of-onset, genetic heterogeneity manifest as differences in age-of-onset could still be present, as observed in the discovery of the BRCA1 breast cancer gene [22]. Hence, we performed ordered subset analysis (OSA) to understand the effect of age-of-onset on linkage to CAD. We found increased linkage on the chromosome 7p14 peak in a subset of 97 families with the youngest age-of-onset (overall LOD = 1.04; subset LOD = 4.22; p = 0.004 for increase, Figure 1). The mean age-of-onset in these families was 37.8 years, and they had significantly higher mean total- and LDL-cholesterol and were more often male, compared with affected members of the remaining 323 families. No other genomic regions showed a correlation between linkage and age-of-onset. The NPY gene resides within this linkage peak and is a strong biological candidate. As a result we aimed to evaluate NPY polymorphisms, NPY levels, and age-of-onset of CAD, along with evaluating the role of NPY in a mouse model system.
Twenty-four SNPs were genotyped and were in Hardy-Weinberg equilibrium. Rs5571 was monomorphic and not analyzed further. Consistent with microsatellite results, nine NPY SNPs had LOD>1.0 in GENECARD, with higher LODs for several SNPs in the subset of 97 very-young-age-of-onset OSA families. Six of these linked SNPs also showed family-based association with CAD by PDT in GENECARD (Table 2). These SNPs are in varying degrees of LD (Figure 2). Association with CAD was validated in the non-familial dataset CATHGEN for five SNPs within the 6-SNP block; the sixth SNP (rs16147) demonstrated borderline significance (Table 3). CATHGEN results also validated our GENECARD findings for age-of-onset effects, with stronger association in the youngest age-of-onset CAD cases (<38 years of age, threshold defined by the GENECARD OSA subset). Although the CATHGEN sample size decreased using this age-of-onset threshold, SNPs within the 6-SNP block remained significantly associated to CAD (Table 3), with odds ratios higher than those obtained with the full dataset, and with several SNPs showing greater significance than the overall cohort (e.g., rs16119, allelic OR 2.20, p = 0.0004). These very-young-age-of-onset cases had a higher prevalence of family history of CAD (p = 0.003) and a trend for higher prevalence of dyslipidemia (p = 0.07).
All six key NPY SNPs were also associated with early-onset CAD in the comparison of GENECARD probands with CATHGEN controls (Table 4). Although we had adjusted for race in our regression models, to further address the potential for population stratification we performed analyses stratified by race. In these analyses, the association between NPY SNPs and early-onset CAD remained consistent and often more significant in self-reported Caucasians (Table S1), suggesting that the results are not confounded by population stratification from race. However, there was no association between NPY SNPs and CAD in non-Caucasians, most likely due to the lower power to detect such a difference given the small sample size (N = 162 CATHGEN non-Caucasian CAD cases, 72 non-Caucasian controls). Haplotype analysis in the CATHGEN dataset showed association of several two-SNP haplotypes with CAD, recapitulating the LD structure of the individual SNPs, but not providing additional information. The haplotype most strongly associated was composed of rs5574 and rs16474 (p = 0.005).
Because rs16147 is a promoter SNP with established effects on expression of the gene [23]–[25], and all other key NPY SNPs are in LD with rs16147, we will focus the remainder of our results primarily on rs16147. However, as expected, findings were consistent across all six SNPs. As with the other five SNPs, there was consistency of the CAD-associated allele for rs16147 (G allele) and of its allele frequency across all datasets (CATHGEN cases 0.49; very-young CATHGEN cases 0.55; GENECARD probands 0.55; CATHGEN controls 0.43) (Table S2).
To test the possibility that associations between NPY SNPs with CAD are mediated through traditional CAD risk factors, we performed multivariable regression in our case-control datasets, using hypertension, dyslipidemia, diabetes, BMI, and smoking as covariates with NPY SNP genotype. In this analysis, the association of rs16147 with CAD remained significant in our comparison of GENECARD probands with CATHGEN controls (genotype OR 1.60 [95% CI 1.11–2.31], p = 0.01; allele OR 2.00 [1.11–3.59], p = 0.03), suggesting that NPY genotype contributes to CAD risk independently of traditional risk factors, at least in family-based cohorts. In the CATHGEN cohort, however, adjusting for risk factors attenuated association for rs16147 (genotype OR 1.15 [0.87–1.51], p = 0.34; allele OR 1.24 [0.82–1.88], p = 0.32), but left the association of rs16120 with CAD intact (allele model, OR 1.54 [1.01–2.35], p = 0.04). Race-stratified analyses in Caucasians also showed attenuation of association between NPY SNPs and early-onset CAD, although three of the SNPs remained significantly associated (Table S1).
To understand which CAD risk factor(s) most mediate association of NPY genotype with CAD, we performed a forward stepwise logistic regression, first fitting a model with genotype and then adding each risk factor. The greatest attenuation of association was caused by dyslipidemia, with minimal additional attenuation by other variables. Table S3 lists risk factors significant in the multivariable model for the SNP remaining significant after adjustment (rs16120), demonstrating the independent strength of effect of genotype, as well as those for intermediate risk factors, in a full fitted model. Neither BMI nor cholesterol levels varied significantly by NPY genotype (data not shown). Thus, it appears that NPY SNPs may contribute to atherosclerosis risk either independently of traditional risk factors, especially in high-risk families, or via mechanisms that may relate to dyslipidemia in non-familial CAD.
An alternate way to examine the relationship between quantitative covariates and genetic variants is comparing means of traits by genotype (measured genotype analysis). We performed this analysis in CATHGEN to elucidate further the influence of NPY variants on age-of-onset. We found that the risk allele for rs16147 is indeed associated with a younger age-of-onset (mean age-of-onset: 45.8 (SD 6.5) in subjects with 1 or 2 copies of risk allele versus 47.3 (5.9) in subjects with 0 copies, p = 0.03). These results support our GENECARD findings and suggest NPY may affect the course of atherogenesis such that atherosclerosis manifests earlier in life.
To examine allele-specific effects, we assessed the relationship of NPY variants with peripheral NPY levels in 220 subjects randomly selected from the CATHGEN case/control dataset. We found that rs16147 was associated with higher NPY levels for the minor (risk, G) allele compared with the major (A) allele (46.3 vs. 41.0 pmol/L, p = 0.04, Figure 3). Concordantly, NPY levels were higher in CAD cases compared with controls (46.8 vs. 41.1 pmol/L, p = 0.02).
To determine whether NPY-evoked signaling contributes to atherosclerosis, we attenuated the vascular effects of NPY [26] with an NPY1 receptor antagonist that does not cross the blood-brain barrier: BIBP 3226 (BIBP) [8],[26],[27], which has been shown to reduce non-atherosclerotic neointimal hyperplasia in rats [8],[26]. In order to accelerate typical atherosclerosis in a focal manner, we employed carotid endothelial denudation in apoe−/− mice, as reported by other groups [28],[29]. With this approach, we could apply BIBP just focally to the carotid artery in a peri-arterial Pluronic gel. Atherosclerosis developed typically in this carotid model, with abundant evidence of macrophage foam cells, SMC-like cells constituting fibrous caps of complex intimal lesions, and extracellular cholesteryl ester (Figure 4A–4C). Peri-carotid application of Pluronic gel by itself had no effect on extent of atherosclerosis (data not shown), and BIBP in the gel did not engender cell toxicity, as judged by apoptosis: cleaved caspase-3 levels were 70±20% higher in control than in BIBP-treated carotids (p<0.02, Figure 4D–4E). While control specimens demonstrated greater apoptosis than BIBP-treated specimens, they also demonstrated 1.7±0.3-fold greater cell proliferation, as judged by immunofluorescence for proliferating cell nuclear antigen (PCNA, Figure 4F, 4G). However, while BIBP did not affect the prevalence of macrophages, SMCs or collagen in atherosclerotic plaques, it reduced total plaque cell number, collagen content and plaque area by 56% (p<0.03, Figure 5). Thus, it appears that NPY, through the NPY1 receptor and perhaps other NPY receptors, contributes to atherogenesis.
Our data provide the first evidence that NPY gene variants associate with CAD in humans, particularly those with early-onset CAD, and that NPY contributes to atherosclerosis in mice through its NPY1 receptor. Our results supporting the role of NPY in CAD pathogenesis emerge from multiple lines of evidence, including linkage, family-based and case-control association studies in multiple cohorts, as well as allele-specific differential NPY levels. Because we observed increased linkage and association between NPY SNPs and CAD in the very youngest age-of-onset cases, NPY may make a particularly appealing therapeutic target for CAD prevention in families with early-onset disease. Together with our mouse data, our results support the following scenario by which NPY variants promote atherosclerosis: presence of the rs16147 risk allele leads to increased plasma NPY levels, which, acting on NPY1 (and perhaps other) receptors, promotes arterial smooth muscle cell proliferation [30], thereby promoting atherogenesis.
To our knowledge, ours is the first study to take a gene-wide approach to NPY in cardiovascular disease. Furthermore, the NPY variant rs16147 has never been reported for any cardiovascular phenotypes. Because rs16147 is not in LD with upstream SNPs, it appears that the LD block tagged by rs16147 does not extend beyond the gene (www.hapmap.org). This inference is supported by our own data: the upstream rs7800861 is not in LD with rs16147 or other key SNPs.
The allele-specific effects of the NPY promoter SNP rs16147 (A/G) on NPY expression have been documented in three previous studies [23]–[25]. In accord with our work, two of these studies have observed that the rs16147 G allele (reverse strand C allele) increases NPY expression [23],[24]. Furthermore, a putative SP1 transcription factor binding site within the rs16147 stretch of NPY sequence is lost with the rs16147 A allele [24]; consequently, we would expect the A allele to demonstrate lower NPY expression. In contrast, Zhou et al recently studied NPY expression in post-mortem brain and virally transformed lymphoblastoid cells and found that the rs16147 A allele results in higher, rather than lower expression of NPY [25]. Zhou et al. also found that haplotypes reporting on the rs16147 A allele result in higher NPY plasma levels [25]. However, it is important to note that Zhou et al. collected plasma from patients under resting conditions, whereas we collected plasma from patients under arguably stressful conditions, immediately before they underwent coronary angiography. Furthermore, differences in cell and tissue types examined in these studies may underlie the discordant results obtained for NPY expression. Given previous studies showing that NPY levels are higher in cardiovascular disease [16],[31] (findings that were corroborated in our study), and the strong concordance of our results showing a higher frequency of the rs16147 G allele in CAD cases, it follows that the G allele would result in increased NPY expression, as observed in the previous NPY studies [23],[24] and in our study. Taken together, all of these studies nevertheless imply a functional role for rs16147.
One prior linkage scan has implicated chromosome 7p in atherogenesis, showing modest linkage to carotid plaque (LOD 0.60–0.78, p = 0.03–0.05) [32]. A recently published study has also shown linkage on chromosome 7p15 (LOD 3.3) to pulse pressure, a measure of central arterial stiffness and a predictor of cardiovascular mortality [33]. None of the genomewide association studies (GWAS) of atherosclerosis have published association with variants on chromosome 7p. However, we reviewed individual SNPs through the Framingham SHARe GWAS database (dbGAP, http://www.ncbi.nlm.nih.gov/projects/gap) and found association of several flanking NPY SNPs with cardiovascular phenotypes (Text S1, p = 0.04–0.0005). These results provide further support for a role for NPY in human atherosclerosis.
Although previously associated with atherosclerosis primarily in Scandinavian subjects [18]–[20], the NPY SNP rs16139 was not associated with CAD in our sample. Our inability to replicate association for rs16139 may have resulted from the low minor allele frequency of rs16139 in our cohorts (similar to other non-Scandinavian populations [34]), and hence the small power of our studies for rs16139. Alternatively, the relationship between rs16139 and atherosclerosis in Scandinavians, which is as yet mechanistically obscure, may not pertain to our genetically more diverse cohorts. Concomitantly, we note that rs16147 is very common, and despite a modest relative risk conferred, this variant may have a higher population attributable risk in genetically diverse cohorts. In fact, 21% of our CATHGEN subjects are homozygous for the rs16147 risk allele. This prevalence is similar in magnitude to that observed with the recently identified chromosome 9p21 CAD susceptibility variant discovered through GWAS [5].
We cannot exclude the possibility that our CATHGEN data are affected by population stratification; however we adjusted for race and CATHGEN subjects were recruited from only one site. Furthermore, race-stratified analyses showed consistent, and often more significant, association between NPY SNPs and CAD in Caucasians, though our analyses were underpowered to assess such relationships in non-Caucasians. Importantly, there is no evidence of population stratification in previous GENECARD analyses [7] and, unlike the non-familial case-control regression analyses which are sensitive to population stratification, the family-based association analyses are robust to population stratification.
The allele frequencies of the NPY SNPs were remarkably similar in very-young-age-of-onset CATHGEN cases and GENECARD probands. Consequently, these independent cohorts appear to be genetically as well as clinically similar, in that they demonstrate a strong genetic component to their risk profiles. These very young individuals represent a prime target for prevention. In identifying the association between NPY variants and atherosclerosis, our results highlight the importance of pursuing a comprehensive gene-wide SNP survey. The replication of our GENECARD findings in our CATHGEN validation dataset is very consistent, not just for the NPY gene itself but also for the association of the individual associated alleles of each SNP, suggesting it is extremely unlikely that results are due to chance alone. It remains to be determined whether association between NPY and CAD is influenced by traditional CAD risk factors, particularly dyslipidemia, and whether a particular constellation of risk factors along with NPY SNPs is related to the very early-onset CAD associations observed in both GENECARD and CATHGEN. A more complete understanding of the relationship between NPY, CAD risk factors, and CAD will require additional studies.
In our study, the correlation between SNPs in the form of LD makes a Bonferroni correction for multiple comparisons too conservative. Using a method that corrects for LD [35] resulted in 11 minimally redundant SNPs (corrected p<0.0047). Thus, one SNP for our CATHGEN very-young-age-of-onset group (rs16119, p = 0.0004) and most SNPs for the GENECARD proband comparison group would remain significant. More importantly, we have used additional statistical and functional studies to demonstrate that association of NPY SNPs with CAD does not result from chance alone: the convergence of evidence from linkage and association, as well as replication of the same SNP/CAD association in independent cohorts. The possibility that NPY contributes to atherosclerosis is supported by the correlation of plasma NPY levels with NPY SNP risk alleles, and the inhibition of murine atherosclerosis with an antagonist of the NPY1 receptor, which mediates most cardiovascular effects of NPY [8].
In summary, NPY is a strong candidate gene for early-onset CAD. NPY SNPs may help refine CAD risk estimates and target therapies for young members of families with CAD.
The GENECARD study enrolled 920 families with early-onset CAD (age-of-onset before 51 years for men, 56 years for women) to perform affected-sibling-pair linkage [7]. GENECARD families were recruited based on the presence of at least two siblings with early-onset CAD defined by: stress testing with ischemia; myocardial infarction (MI); coronary revascularization; or angiography with ≥50% stenosis in one major vessel. Unaffected family members had no clinical evidence of CAD and age >55 years for men (>60 years for women). The 420 families from the initial linkage scan [7] are included in this report, including a limited number of unaffected family members (N = 37).
An independent non-familial validation cohort was selected from the CATHGEN biorepository, consisting of subjects recruited sequentially through the cardiac catheterization laboratories at Duke University Medical Center (Durham, NC). Fasting whole blood and plasma samples were obtained from the femoral artery during cardiac catheterization and frozen prior to use. CAD cases were defined as CAD-index≥32 (at least one vessel with ≥95% stenosis) with age-of-onset <56 years. CADi is a numerical summary of angiographic data directly related to outcome [36]. Age-of-onset was defined as age at first MI, coronary revascularization, or catheterization meeting CADi threshold. Controls were defined as age-at-catheterization >60 years; CAD-index≤23, no history of cardiovascular disease, and no clinically significant CAD. Dyslipidemia was defined as a previous diagnosis and/or treatment of hypercholesterolemia (yes/no), confirmed by review of medical records. We also constructed a third case-control group comprising GENECARD probands included in the original linkage scan from United States sites (N = 221) and CATHGEN controls. Institutional Review Boards approved all study protocols. Informed consent was obtained.
Tagged NPY SNPs were identified using the SNPSelector program, which employs a linkage disequilibrium (LD) tagging algorithm that prioritizes functional SNPs [37], and were then genotyped using either Taqman or Illumina BeadArray (www.illumina.com). The 7900HT Taqman SNP genotyping system incorporates a standard PCR-based, dual fluor, allelic discrimination assay with a dual laser scanner. Assays were purchased through Applied Biosystems (www.appliedbiosystems.com). QC samples, composed of 12 reference controls, were included in each quadrant of the plate. Illumina BeadStation genotyping was performed using the 500G system. All SNPs examined were successfully genotyped for ≥95% of the individuals in the study. Error rate estimates for SNPs meeting QC benchmarks were <0.2%.
Plasma NPY levels were measured by an NPY-specific radioimmunoassay (Alpco Diagnostics, Salem NH) on 220 CATHGEN subjects, randomly selected (regardless of case/control status) from all subjects included in the genotyping studies. NPY levels were measured in fasting unextracted plasma collected at time of cardiac catheterization (prior to administration of supplementary anticoagulants if given), which was subsequently frozen and stored at −80°C, and analyzed in a single RIA run. Cross-reaction of the NPY RIA assay as reported by the company is: human NPY 100%; human PYY <2.0%; human pancreatic polypeptide (PP) <1.0%; NPY 1–21 <0.1%; NPY 20–36 <0.4%. The assay was characterized as a mean recovery of 82% (range 75–88%) for NPY-spiked plasma. Precision was characterized by an intra-assay CV of 2.6–3.9% in our samples, which were run in one batch. NPY levels were approximately normally distributed.
All animal experiments complied with institutional guidelines. Gender-matched apolipoprotein E-deficient (apoe−/−) C57BL/6 mice were used, at the age of 10±2 weeks. To accelerate atherosclerosis, we performed endothelial denudation of the left common carotid artery with a 0.36-mm flexible angioplasty guidewire (Johnson and Johnson), inserted via the external carotid artery as described [28],[38]. After ligating the external carotid artery, we encased the common carotid in 150 µl of 22.5% (w/v) Pluronic gel, containing either 1% (v/v) water that lacked (control) or contained the NPY1 receptor antagonist BIBP 3226 [27] (21 µmol/L, [final]; Sigma-RBI, Inc.). The skin was closed after the Pluronic gelled (virtually instantaneously). Postoperatively, mice were fed a Western diet (Harlan Teklad #88137) ad libitum for two-six weeks (until arterial harvest), and there were no differences in weight between mice treated with vehicle or BIBP 3226. BIBP 3226 does not affect blood pressure in rodents [26]. As we previously described [39], arteries were perfusion/fixed and embedded in paraffin, or embedded in OCT and frozen. Five-µm sections were stained with a modified connective tissue stain (paraffin sections), Sudan IV (frozen sections, for cholesteryl ester), or immunostained (frozen sections) as previously described [38],[39]: for macrophages (FITC-conjugated rat IgG1κ anti-mouse Mac3 or negative control FITC-conjugated rat IgG1κ (Pharmingen, Inc.)); smooth muscle cells (SMCs) (Cy3-conjugated 1A4 IgG targeting SMC α-actin (Sigma)); for apoptotic cells (monoclonal rabbit IgG targeting cleaved (activated) caspase-3 (Cell Signaling, Inc.)); or for proliferating cell nuclear antigen (Santa Cruz Biotechnology, Inc). All sections were digitally photographed with fixed light settings for all immunofluorescence specimens within a single staining batch, to facilitate quantitation [39]. Atherosclerosis, cellular (or protein antigen) prevalence, and PCNA-positive nuclear prevalence [38] were quantified with Scion Image (www.scioncorp.com) or by manual counting by an observer blinded to specimen identity [39], from three cross sections obtained from the distal, middle, and proximal portions of each specimen, as described [39]. Data from these locations were averaged for each carotid specimen.
Ordered subset analysis (OSA) examines the impact of covariates by defining homogeneous subsets of families for linkage [40], and has been successfully used for gene mapping in complex diseases [41]–[44]. Similar methods were used to identify the BRCA1 breast cancer susceptibility gene, where evidence for linkage was only present in earlier-onset families [22],[45]. Subsets of families with earlier ages-of-onset may help evaluate genetic heterogeneity and define subgroups with a stronger genetic effect. A priori specification of the age-of-onset threshold is unnecessary, as OSA uses maximal linkage evidence to define subsets. OSA was conducted using nonparametric multipoint family-specific LODs from microsatellite genotypes from the original screen as input [7], with age-of-onset as a covariate. A permutation procedure provides empirical p-values for the significance of the increase in the maximum-subset-LOD from the overall LOD. To assess SNP linkage, two-point LODs were calculated using Merlin [46]. The Pedigree Disequilibrium Test (PDT) and Genotype-PDT were used for family-based association. PDT, an extension of the transmission-disequilibrium-test (TDT), allows incorporation of extended pedigrees and is valid even with population substructure [47]. Power calculations (QUANTO:http://hydra.usc.edu/) showed power ≥0.80 to detect effect sizes of ≥2.45 for the lowest allele frequency SNP (0.02, rs16139), and effect sizes ≥1.35 for the highest allele frequency SNP (0.49, rs16147).
In CATHGEN, association was assessed using logistic regression models adjusting for race and sex; and for race, sex, hypertension, diabetes, dyslipidemia, smoking and body-mass-index (BMI) (multivariable model). Measured genotype analysis using generalized linear models was performed to compare differences in means of quantitative traits (NPY levels, age-of-onset) by NPY genotype. Baseline differences were assessed using a chi-square or t-test. The Graphical Overview of Linkage Disequilibrium (GOLD) program [48] was used to assess LD. Haplotype analysis used HaploStats 1.1.0 (Mayo Clinic, Rochester, MN). We report p-values uncorrected for multiple comparisons, but also present results in the context of correction for LD between SNPs [35]. The extent of atherosclerosis was compared between control and NPY1 receptor antagonist groups in mice with one-way ANOVA and Tukey's post-hoc test for multiple comparisons. SAS 9.1 (SAS Institute, Cary, NC) was used for statistical analysis.
|
10.1371/journal.pcbi.1006645 | Chirality provides a direct fitness advantage and facilitates intermixing in cellular aggregates | Chirality in shape and motility can evolve rapidly in microbes and cancer cells. To determine how chirality affects cell fitness, we developed a model of chiral growth in compact aggregates such as microbial colonies and solid tumors. Our model recapitulates previous experimental findings and shows that mutant cells can invade by increasing their chirality or switching their handedness. The invasion results either in a takeover or stable coexistence between the mutant and the ancestor depending on their relative chirality. For large chiralities, the coexistence is accompanied by strong intermixing between the cells, while spatial segregation occurs otherwise. We show that the competition within the aggregate is mediated by bulges in regions where the cells with different chiralities meet. The two-way coupling between aggregate shape and natural selection is described by the chiral Kardar-Parisi-Zhang equation coupled to the Burgers’ equation with multiplicative noise. We solve for the key features of this theory to explain the origin of selection on chirality. Overall, our work suggests that chirality could be an important ecological trait that mediates competition, invasion, and spatial structure in cellular populations.
| Is it better to be left- or right-handed? The answer depends on whether the goal is making a handshake or winning a boxing match. The need for coordination favors the handedness of the majority, but being different could also provide an advantage. The same rules could apply to microbial colonies and cancer tumors. Like humans, cells often have handedness (chirality) that reflects the lack of mirror symmetry in their shapes or movement patterns. We find that cells gain a substantial fitness advantage by either increasing the magnitude of their chirality or switching to the opposite handedness. Selection for specific chirality can overcome differences in growth rate and is mediated by the formation of bulges along the colony edge in regions where cells with different chiralities meet.
| Living systems have harnessed a variety of physical principles to design and exploit spatial patterns [1–3]. Many biological patterns are chiral, i.e. they break left-right symmetry. While the mechanism of chiral symmetry breaking have been elucidated in some systems [4–10], the functional role of chirality remains largely unexplored [10–13].
Chirality exists at all scales: from molecules to populations [7, 13–18]. The origin of molecular chirality is typically attributed to chance [4, 5, 19, 20]. For nucleotides and amino acids, the classical explanation of homochirality posits two steps: a fluctuation that slightly breaks the left-right symmetry and a self-amplifying process that increases the asymmetry further [6, 19]. More recent work demonstrates that the amplification step may not be necessary because intrinsic noise in chemical reactions is sufficient to establish and stabilize the symmetry breaking [20]. The existence of many chiral components within the cell then serves as a natural explanation for macroscopic chirality [15, 21, 22]. Consistent with this view, chiral body plans arise early in the development due to a symmetry breaking event at a microscopic scale, which is amplified further during the subsequent growth [8, 9, 16, 21, 23, 24]. Similarly, the macroscopic chirality of bacterial colonies is typically explained by the chirality of individual bacteria [13, 25, 26].
The existing theory explains how, but not why chirality emerges. Indeed, a lot of effort went into elucidating the mechanism of the chiral symmetry breaking [7, 13, 14, 27], but the relationship between chirality and fitness has received much less attention [11, 12, 28–30]. Several lines of evidence, however, do suggest that a change in chirality could be advantageous [11, 12, 28–30]. Experiments with Arthrospira showed that this bacterium changes from a right-handed to a left-handed helix following the exposure to grazing by a ciliate [11, 28]. Extensive work with Paenibacillus demonstrated that this microbe switches between a chiral and a non-chiral forms to optimize its fitness in different environments [13, 30]. Human cells are also known to form chiral patterns. The handedness of these patterns is the same across all tissue types; except, it is reversed in cancer [29]. Thus, in a variety of systems, a change in chirality co-occurs with the evolution of higher growth, dispersal, or competitive ability.
Motivated by these striking examples, we decided to explore whether chirality could be a product of natural selection rather than a historical accident. We asked this question in the context of growing cellular aggregates and found that chirality could directly affect fitness through a pattern-formation mechanism. Our results show that, depending on the growth conditions, it could be advantageous for a cell to increase its chirality or to switch its handedness relative to that of the ancestral population. These dynamics often lead to the coexistence of the left- and right-handed forms, which is a major departure from the classic theories of homochirality [6, 19, 20]. Coexisting cell types may enjoy additional benefits of chirality because they develop unique spatial structure that facilitates cross-feeding and other social interactions [31–34].
Microbial colonies and cancer tumors exhibit a variety of complex morphologies including smooth and rough compact disks, concentric rings, radiating branches, and many others [13, 36–39]. For aggregates that grow as a network of filaments or branches, chirality manifests as clockwise or counterclockwise bending of the branches [27]. Although chiral growth is not as easily detected for other morphologies, it could be present even if the overall colony shape shows no left-right asymmetry.
The “hidden” chirality can be revealed by growing a colony from an initially well-mixed population of two strains that are identical, except they express different fluorescent proteins. As the colony expands, demographic fluctuations at the colony edge lead to local extinctions of one of the strains creating a characteristic pattern of sectors shown in Fig 1A [17, 18, 35, 40, 41]. In the absence of chirality, the boundaries between the sectors are approximately radial, but the boundaries twist consistently clockwise or counterclockwise for chiral cells. The direction of this twisting stays the same across replicate colonies [17, 18, 26]. So far, this method has been applied only to a few model organisms; two of them, Escherichia coli and Bacillus subtilis, were found to grow in a chiral fashion [17, 18, 26]. Since most organisms have not been examined, hidden chirality could potentially be quite prevalent in cellular aggregates.
The twisting of the boundaries can be quantified by the increase of the polar angle with the distance from the colony center. Using the data from Ref. [18], Fig 1C shows that this dependence is logarithmic, i.e. the boundaries twist as Bernoulli spirals [42]. The origin of the logarithmic twisting can be explained by a simple phenomenological description that combines a constant velocity of the sector boundary with a linear increase of the colony radius in time [35]. The molecular mechanism responsible for cellular chirality has not been fully determined, but we do know that chirality in not due to flagella and is mediated by outer membrane proteins such as antigen 43, extracellular structures including pili, and the interaction with the substratum [26].
Because the factors that mediate chirality also contribute to other components of cell phenotype, it could be challenging to create two strains that differ only in their chirality. This difficulty, however, can be easily overcome in a computational model, where chirality can be tuned without affecting the growth and motility (see Methods and SI). A large number of approaches has been developed to model cellular aggregates from analytic equations to mechanistic simulations [43–47]. We chose to study a minimal reaction-diffusion model because it involves few parameters and is more likely to capture the universal behavior that generalizes across diverse cellular populations. We also focused on the simplest morphology of a compact disk because it is both common and well-understood.
For strains with equal chirality, our simulations (described below) showed excellent agreement with the experimental observations (Fig 1). The simulations not only reproduced the formation and bending of sectors, but also exhibited the same logarithmic twisting of sector boundaries as in the experiments. Thus, the few ingredients in our model are sufficient to describe the chiral growth in compact microbial colonies.
In simulations, cells grow and move in a two-dimensional habitat. The movement is stochastic and short-ranged, but potentially biased relative to the direction of the local density gradient. For non-chiral populations, the bias is along the gradient, in the direction of the outward growth. This bias accounts for the effects of chemotaxis towards nutrients and higher pressure within the colony. For chiral cells, the direction of movement is not collinear with the applied force [48], so the bias in cell movement makes a nontrivial angle with the local gradient of the population density. The sign and magnitude of this angle control the handedness and the strength of chirality. A detailed description of the simulation procedure is provided in Methods and SI.
In the deterministic limit, our simulations are described by the following reaction-diffusion equation:
∂ n ( α ) ∂ t = g ( α ) n ( α ) - ∇ · J ( α ) , (1)
where t is time, n(α) is the population density of strain α, g(α) is a density-dependent per capita growth rate, and the flux J is given by
J i ( α ) = - D ( α ) ∇ i n ( α ) - n ( α ) ∑ j ( S ( α ) δ i j - A ( α ) ϵ i j ) ∇ j n . (2)
Here, the indexes denote the Cartesian components of vectors, δij is the unit tensor, and ϵij is the totally antisymmetric tensor, also known as Levi-Civita symbol.
The density-dependent diffusion and advection are described by D(α)(n) and S(α)(n) respectively. A(α)(n) is the strength of the chiral term, which is the only term that changes sign under the mirror symmetry. To the lowest order in the gradient expansion, no other term that breaks the left-right symmetry can be added to Eq (2), which suggests that all chiral patterns in compact aggregates are described by our coarse-grained theory regardless of the microscopic origin of chirality.
To test whether chirality confers a selective advantage, we competed a chiral strain vs. a non-chiral strain. The growth and motility rates of these strains were identical and, as a result, they expanded at the same rate when grown separately (Fig 2A and 2B); see SI for a more quantitative comparison of the velocities. The chiral strain, however, had a clear selective advantage when the competition occurred within the same colony. Fig 2C illustrates this by showing how the chiral strain took over the population starting from a small, localized patch representing a mutation or an immigration event. This selective advantage of chirality was not specific to the competition between a chiral vs. strictly non-chiral strains. In simulations, we explored a wide range of microscopic parameters and invariable found that the more chiral strain outcompeted the less chiral strain when both strains had the same handedness.
In contrast, the competition between two strains with opposite handedness often resulted in stable coexistence. As an example of this, Fig 3 shows the competition between two strains with chiralities that equal in magnitude, but opposite in direction. Both strains invaded when introduced in the population of the opposite handedness, but did not completely take over (Fig 3A and 3B). Instead, the population approached a steady state where both strains were equally abundant. To confirm this observation of negative frequency-dependent selection, we performed simulations starting from well-mixed initial conditions with different fractions of the left-handed strain. As expected from symmetry, the fraction of the left-handed strain converged to 50% (Fig 3C) suggesting that left- and right-handed strains can stably coexist. Strains with opposite, but not exactly equal chirality were also found to coexist, but the equilibrium fractions deviated from the 50:50 ratio in favor of the more chiral strain.
The effects of chirality persist even when the strains have different growth rates. To demonstrate this, we competed a chiral vs. a faster growing non-chiral strain (Fig 4A). The chiral strain completely excluded the non-chiral strain provided its growth rate penalty was less than 2%. For growth rate differences between 2% and 7%, the two strains stably coexisted. The chiral strain went extinct only when its growth rate penalty exceeded 7%. Similar dynamics occurred during the competition between two strains with opposite handedness (Fig 4B). As we increased the difference in the growth rate between the strains, their steady-state abundance started to deviate from the 50:50 ratio. The coexistence was lost only when the growth rate difference exceeded 7%. The 7% threshold is representative for the parameters used in our reaction-diffusion model, but its value in an actual biological system could be affected by the details of the cell-cell and cell-surface interactions, which we did not explicitly model. Nevertheless, our results demonstrate that chirality can influence the outcome of the competition between strains with substantial differences in growth and motility.
Selection mediated by chirality may seem quite surprising. In the following, we first explain the origin of this phenomenon using the competition between strains with opposite handedness as a simple example. After that, we return to the general case and explore the transition between exclusion and coexistence as the chirality of the strains is varied. We also describe the role of demographic fluctuations and characterize the spatial patterns that emerge in populations of chiral cells.
Why does chirality affect competition? To answer this question, we developed an analytical theory that explains the spatial patterns shown in Figs 1–3. For this purpose, we reduced the reaction-diffusion model (Eq (1)) to a simpler effective theory that describes only the overall shape of the colony edge and its genetic composition (see SI). This effective theory can also be derived purely from the symmetry considerations and is therefore more general than the underlying reaction-diffusion model (see SI). Below, we use the effective theory to explain how the competition between two strains is affected by their chiralities. Our main result is that the difference in chiralities leads to changes in colony shape, which in turn influence the relative abundance of the strains.
Because little growth occurs inside cellular aggregates [47, 52], their population dynamics can be largely described in terms of only two variables: the position of expansion front and the relative abundance of the strains at the edge of the colony. To fix the coordinate system, we take the x-axis to be a straight line along the average direction of the colony edge (Fig 2A). The y-axis then points in the direction of colony growth. We denote the y-coordinate of the expansion front by h(t, x), and the fraction of the first strain by f(t, x). In terms of these quantities, the effective theory is given by the following set of equations:
∂ h ∂ t = v 0 + v 0 2 ( ∂ h ∂ x ) 2 + D h ∂ 2 h ∂ x 2 + α ∂ f ∂ x + noise , ∂ f ∂ t = D f ∂ 2 f ∂ x 2 + β ( f * - f ) ∂ f ∂ x + v 0 ∂ h ∂ x ∂ f ∂ x + noise . (3)
The first equation in Eq (3) is an extension of the Kardar-Parisi-Zhang (KPZ) equation of surface growth [53–55]. Here, the first term, v0, is the expansion velocity of the strains grown in isolation. The second term accounts for the fact that the expansion of a tilted front (∂ h ∂ x ≠ 0) occurs perpendicular to the front and, therefore, at a angle with the y-axis. The third term arises because fronts that are curved outward expand more slowly and because of effective surface tension at the edge of the microbial colony [56]. The last term couples the dynamics of f and h and is a new term that describes chirality. Its magnitude is controlled by parameter α, which is proportional to the difference in the chiralities of the strains. We show below that this last term changes colony shape and mediates the competition between chiral strains.
The second equation in Eq (3) is an extension of the Burgers’ equation used to describe fluid and traffic flow [57–60]. The first term describes random, diffusion-like movement, while the second term accounts for the directional motion due to a chiral bias in motility. Here, the factor of −β(f* − f) can be viewed as a local advection velocity, which depends on the relative abundance of the strains and two parameters that describe the chiral properties of the strains. The first parameter, β, is proportional to the difference in the chiralities of the strains. The second parameter f* is the ratio of the chirality of the first strain to the difference in the chiralities of the strains.
We choose the most left-handed (also the least right-handed) strain to be the first strain, which ensures that α and β are positive (see SI). With this convention, the first strain that is left-handed for f* > 0 and right-handed for f* < 0. When f* ∈ (0, 1), the two strains have opposite handedness, while, when f* ∈ (−∞, 0) ∪ (1, + ∞), the two strains have the same handedness. Furthermore, f* = 0 corresponds to a non-chiral and a right-handed strain, and f* = 1 corresponds to a left-handed and a non-chiral strain. The special case of equal, but opposite chiralities corresponds to f* = 1/2.
The third term in the equation for ∂ f ∂ t describes how colony shape affects the relative abundance of the strains. This term is non-zero only in tilted regions of the front, where cells at higher h displace cells at lower h as the colony grows. In other words, the relative abundance of the strains changes because the growth of the colony proceeds in the direction perpendicular to the front and, therefore, induces the movement of cells along the x-axis whenever ∂ h ∂ x ≠ 0. Below we demonstrate how this coupling between colony shape h(x) and genetic composition f(x) mediates the competition between the strains in compact aggregates.
Finally, the noise terms account for demographic fluctuations and genetic drift. In the first equation, the noise is the regular additive noise present in the KPZ equation; it arises due to local fluctuations in the growth velocity. In the second equation, the noise accounts for genetic drift, so it is multiplicative with the strength proportional to f ( 1 − f ). Such dependence on f is typical for population dynamics [41] and is necessary to ensure that f = 0 and f = 1 are absorbing states.
Genetic drift in the equation for f leads to local extinctions of one of the strains and the formation of sector boundaries [41]; see Fig 1. When these boundaries separate strains with different chirality, h(t, x) develops this characteristic shape that ultimately controls the competition between the strains. For simplicity, we first discuss the behavior of two strains with exactly opposite chiralities (f* = 1/2). In this special case, the mirror symmetry ensures that the boundaries between the strains do not have a net bias and, therefore, remain stationary when ∂ h ∂ x = 0.
The dynamics of a strain boundary depends on whether the chiral biases of the strains point towards or away from it. We term a boundary an in-flow boundary when the strains move towards each other, i.e. the left-handed strain is to the right of the boundary, and the right-handed strain is to the left of the boundary (Fig 5). Boundaries with the opposite arrangement of the strains are termed out-flow boundaries.
Eq (3) predicts that the two types of boundaries have a diametrically opposite effect on the colony shape. For in-flow boundaries α ∂ f ∂ x > 0, and we expect a bulge due to local overgrowth of h. In contrast, a dip in the front is expected at out-flow boundaries, where α ∂ f ∂ x < 0. Fig 5 shows that these shapes indeed develop in our simulations.
We exactly solved the chiral KPZ equation without noise and obtained an analytical expression for the shapes of the bulges and dips (see SI) in the limit of sharp boundaries between the strains. After a transient, the bulges assume an approximately triangular shape given by
h ( t , x ) - v 0 t = { 0 , | x - x b | ≥ v 0 α 4 D h t , v 0 α 2 8 D h 2 t - α 2 D h | x - x b | , | x - x b | < v 0 α 4 D h t (4)
for an in-flow boundary located at xb. The slope of the bulge stays constant, but its height and width increase linearly in time. The depth of the dips, on the other hand, increases only logarithmically in time, so they remain quite small on the time scale of our simulations (see SI). As a consequence, the front primarily consists of titled regions near the bulges and flat regions away from the strain boundaries (Fig 5).
One immediate consequence of Eq (4) is that a mixture of two strains with opposite handedness expands faster than either strain in isolation. Indeed, d h ( t , x b ) / d t = v 0 + v 0 α 2 / ( 8 D h 2 ), which is greater than v0. In our simulation, this increase was typically on the order of a few percent; see SI for further details.
The expansion of bulges changes the relative abundance of the strains (Fig 6). Initially, no bulges are present because we start our simulations with a flat front to mimic the coffee-ring effect that creates a smooth edge around a microbial colony [61]. As the colony expands, small bulges form and grow around the in-flow boundaries. In the beginning, a small bulge has no effect on the genetic composition of the front because it is completely enclosed within the two sectors surrounding the in-flow boundary. However, as the bulge expands, it comes in contact with the out-flow boundaries on its sides and then starts “pushing” them outwards. The subsequent movement of the out-flow boundaries changes sector sizes and, therefore, the relative abundance of the strains.
For the case of f* = 1/2 that we are considering now, the left and the right ends of the bulge are equidistant from the in-flow boundary. Hence, the two strains have equal abundance within the bulge. The expansion of the bulge then brings the global fraction of the first strain, f ‾, towards 1/2. The change in f ‾ ceases only when the out-flow boundaries stop moving, which occurs when they are locked between the two neighboring bulges (see Fig 6). At this point, the entire front consists of bulges, so f ‾ = 1 / 2.
The argument above explains the mutual invasion and coexistence for strains with exactly opposite chiralities shown in Fig 3. For f* ≠ 1/2, the dynamics are essentially the same with only two minor modifications (see SI). First, in-flow and out-flow boundaries do not remain stationary within the regions of flat front. Instead, the boundaries move with velocity v∥ = β(1/2 − f*), which reflects the unequal chiral biases of the two strains. Second, while the bulges remain triangular, they are no longer symmetric relative to the y-axis. The steeper slope occurs on the side that leads the forward motion of the bulge (Fig 7A).
As before, natural selection occurs due to the expansion of bulges, and the steady state is reached when bulges occupy the entire front. In this state, the relative abundance of the strains is determined by the ratio of bulge slopes, and we find that (see SI)
f ¯ eq = { 0 , f * ≤ 1 2 - α v 0 2 D h β , 1 2 + β D h α v 0 ( f * - 1 2 ) , | f * - 1 2 | < α v 0 2 D h β 1 , f * ≥ 1 2 + α v 0 2 D h β . (5)
Here, the middle line describes the relative abundance of the strains when they coexist. The first and the last line describe to the exclusion of the less chiral strain. Exclusion occurs when one of bulge slopes becomes horizontal, and, therefore, no steady state can be reached.
These theoretical conclusions are supported by the simulation results summarized in Fig 7B. The data shows a clear transition from coexistence to competitive exclusion and a nearly linear dependence of f ‾ on f* in the coexistence region as predicted by Eq (5). Quantitative comparison between the theory and the simulations is described in S1 Fig. Close to the extinction transitions, however, there are noticeable deviations from linearity. Such nonlinearities are typical for non-equilibrium phase transitions and are described by critical exponents [31, 62]. To obtain the critical exponent and characterize the nature of the phase transition, one would need to account for the stochastic creation and annihilation of domain boundaries, which we have neglected in our analysis.
The size of the coexistence regions depends on the model parameters (see S2 Fig for a narrower coexistence range). In all of our simulations, we found that the transition between exclusion and coexistence occurs for f* ∈ (0, 1). Therefore, the competition between a chiral and a non-chiral strain falls outside the coexistence region, which explains the competitive exclusion shown in Fig 2.
So far, our analysis has relied on the existence of sharp boundaries between the strains. Such boundaries appear readily due to genetic drift both in microbial colonies [17] and in our simulations (Figs 2 and 5–7). Previously, it has been shown that any non-zero genetic drift prevents diffusive broadening and ensures a finite size of a boundary between two neutral, non-chiral strains [63]. We found that the behavior is unchanged when the two strains have non-zero, but identical chirality (S3 Fig). This observation is not surprising because the front of the colony remains flat, and the chiral motion of the strains can be removed by changing into a reference frame moving along the front. The boundaries also have a finite width (at least on the time scale of our simulations) when one of the strains is outcompeted due to the differences in growth rates or chirality; see S4 Fig and Refs. [35, 64]. In this case, the boundary width is controlled both by genetic drift and selection.
When there is selection towards coexistence, there are two distinct possibilities: Either the boundaries are sharp as in Fig 6 or the boundaries widen over time leading to a intermixed state as in Fig 3. We found that there is a well-defined transition between these two regimes, which is controlled by the strength of chirality and genetic drift (Fig 8 and S4 Fig). For strong genetic drift or weak chirality, the boundaries between the strains remain sharp, and population dynamics are completely described by the theory developed above. For weak genetic drift or strong chirality, the strains become intermixed. Our main conclusions remain the same even in this regime (see S5 Fig). In particular, we still observe either coexistence or exclusion depending on the relative chiralities of the strains. The spatial patterns are also similar. A large, non-triangular bulge forms around the intermixed region between the two strains, and small bulges are visible around individual in-flow boundaries (Figs 3 and 8).
The transition between demixed and intermixed regimes appears to be continuous (second order) as seen from Fig 8A. Identification of the universality class of this phase transition and the quantitative description of the mixing regime, however, require a careful analysis of the interplay between stochastic and deterministic terms in Eq (3) and are left for future work.
Spatial intermixing could be especially important for species that participate in social interactions such the exchange of metabolites. In such situations, chirality could not only stabilize the coexistence of the species, but also ensure that they are sufficiently close to each other. When no special mechanism exists to ensure spatial proximity, mutualistic interactions can be easily destroyed by genetic drift [31–34]. Hence, some microbes may rely on different or fluctuating handedness to ensure that the separation between the species does not exceed the maximal distance over which they can interact.
Selection for a particular chirality may seem impossible because an object and its mirror image have identical physical properties. This apparent paradox is however easily resolved by noticing that natural selection always favors a change in chirality relative to that of the ancestral population rather than an absolute, pre-defined value of chirality.
A vivid example of how evolution drives a change in chirality comes from Satsuma snails. Most species in the Satsuma genus are dextral (clockwise coiled), but they often have sister species that are almost identical except for the opposite direction of coiling [12]. These sinister (counterclockwise coiled) species enjoy a distinct selective advantage because they are essentially resistant to the predation by Pareatidae iwasakii, a snake that is common in the rage of Satsuma [12]. Resistance to predation comes from the left-right asymmetry in the jaw of P. iwasakii, which has adapted to the coiling direction of its most common prey. Similarly, a reversal of handedness provides protection to Arthrospira, cyanobacteria that forms helical trichomes, from the predation by a ciliate [11, 28]. In both examples, the mutants enjoy the advantage of being in the minority. This mechanism does not require the presence of a predator and can occur due to a large number of factors. For example, a mutant with chiral motility may spatially segregate from the rest of the population and thereby escape from an intense competition for resources [65].
Our main finding is that selection for chirality can also be mediated by the formation of non-trivial spatial patterns. Mismatch in the chiral bias makes cells move towards each other near in-flow boundaries and away from each other near out-flow boundaries. As a result, the colony edge becomes populated with bulges and dips, which grow over time and alter the relative abundance of the strains. One consequence of these dynamics is that it pays off to be different from the majority of the population: A mutant with the opposite handedness can invade when rare and stably coexist with the ancestor due to negative frequency-dependent selection. For strains with the same handedness, the more chiral strain typically wins the competition because it creates a one-sided bulge that overgrows the less chiral strain. Thus, we identified a distinct selection mechanism that can explain both the evolution toward stronger chirality and sudden reversal of handedness. The predicted effects of chirality are observable even in the presence of moderate growth rate differences and can be tested experimentally by comparing the competition between cells with different chirality to our predictions for colony shape and composition.
Selection for chirality could also come from the indirect benefits of the emergent spatial pattern. One possibility is that pointed bulges might facilitate the invasion of host tissue or other environments. The other possibility is that strain intermixing could promote social interactions that rely on cell contact or the exchange of diffusible metabolites. We found that intermixing between the strains with opposite handedness is stable only when their chiralities exceed a certain threshold. Below this threshold, genetic drift creates macroscopic sectors that grow over time and spatially segregate the strains. As a result of this process, social interactions are either suppressed or completely abolished [31–34].
All of our results can be explained by a simple effective theory that describes population dynamics in terms of colony shape and composition. This description is simpler and much more intuitive than the full two-dimensional growth encoded by reaction-diffusion equations or other mechanistic models. Therefore, our theory could provide a valuable framework to study competition and cooperation in compact aggregates such as microbial colonies and cancer tumors.
Shape undulations inevitably occur when aggregates contain strains that grow at different rates [35, 66]. So far, most theoretical studies have neglected this complexity and assumed that colonies have a flat front [31, 34, 41]. Front undulations, however, are known to profoundly change the nature of competition, for example, by allowing the regions with cooperating strains to overgrow the regions where cooperation has been lost [66, 67]. Our theory is an important first step towards understanding this interplay between evolution in compact aggregates and their shape.
The effective theory also provides an interesting extension of the KPZ equation to systems that break the mirror symmetry. Such symmetry breaking could occur in a variety of systems within the KPZ universality class, for example, during the simultaneous deposition of two homophilic molecules with opposite handedness.
In summary, we have identified a new mechanism of selection for chirality and developed a theory to explain it. Our findings describe the chirality of cells while most of the previous work focused on the emergence of homochirality in biological molecules [4–6, 19, 20]. Unlike the frozen homochirality of nucleotides and amino acids, the chirality of cells continues to evolve, often on the time scale of a few generations [11, 28–30]. Our work suggests that some changes in cellular chirality could be adaptive and, therefore, deserve further study.
Lattice-based simulations were performed on a two-dimensional rectangular grid with periodic boundary conditions. The lattice spacings Δx and Δy were both set to 1. The length of each time step Δt was set to 1 as well. Each time step, we first performed a deterministic update to account for growth and migration and then performed a stochastic update to account for demographic fluctuations and genetic drift. To speed up simulations, only a small rectangular region at the expanding edge was updated while the bulk remained frozen. This did not not affect the results since both migration and growth are zero in the colony bulk. The heterozygosity reported in Fig 8, was calculated within this rectangular region at each time step.
During the deterministic update, we computed an auxiliary quantity ρ(α) equal to the expected value of n(α) at the next time step:
ρ ( α ) ( t , x , y ) = n ( α ) ( t , x , y ) + G ( α ) ( t , x , y ) Δ t + M ( α ) ( t , x , y ) Δ t . (6)
The growth term G describes the increase in the population density due to logistic growth:
G ( α ) ( t , x , y ) = g ( α ) n ( α ) ( t , x , y ) ( 1 - n ( t , x , y ) N ) , (7)
where n = n(1) + n(2) is the total population size, and N is the carrying capacity. The growth rates g(α) were typically the same for the two strains.
The migration term M describes the change in the population size due to migration:
M ( α ) ( t , x , y ) = - m ( x , y ) → ( x + Δ x , y ) ( α ) - m ( x , y ) → ( x , y + Δ y ) ( α ) - m ( x , y ) → ( x - Δ x , y ) ( α ) - m ( x , y ) → ( x , y - Δ y ) ( α ) + m ( x + Δ x , y ) → ( x , y ) ( α ) + m ( x , y + Δ y ) → ( x , y ) ( α ) + m ( x - Δ x , y ) → ( x , y ) ( α ) + m ( x , y - Δ y ) → ( x , y ) ( α ) , (8)
where m ( x 1 , y 1 ) → ( x 2 , y 2 ) ( α ) Δ t is the expected number of migrants of strain α from the site at (x1, y1) into the site at (x2, y2). The first four terms in the equation describe migration out of the lattice site (x, y) and the last four terms describe the migration into the lattice site (x, y). Note that the number of cells leaving a particular site into the direction of its nearest neighbor is equal to the number of cells arriving into that neighboring site, i.e. migration conserves the number of cells.
The migration fluxes m ( x 1 , y 1 ) → ( x 2 , y 2 ) ( α ) were nonzero only between the four nearest neighbors and were defined as follows
m ( x , y ) → ( x + Δ x , y ) ( α ) = n ( α ) ( t , x , y ) ( 1 - n ( t , x + Δ x , y ) N ) × ( m 0 ( α ) + m s ( α ) n ( t , x , y ) N + m d ( α ) n ( t , x + Δ x , y ) N + m l ( α ) n ( t , x , y + Δ y ) N + m b ( α ) n ( t , x - Δ x , y ) N + m r ( α ) n ( t , x , y - Δ y ) N ) , m ( x , y ) → ( x , y + Δ y ) ( α ) = n ( α ) ( t , x , y ) ( 1 - n ( t , x , y + Δ y ) N ) × ( m 0 ( α ) + m s ( α ) n ( t , x , y ) N + m d ( α ) n ( t , x , y + Δ y ) N + m l ( α ) n ( t , x - Δ x , y ) N + m b ( α ) n ( t , x , y - Δ y ) N + m r ( α ) n ( t , x + Δ x , y ) N ) , m ( x , y ) → ( x - Δ x , y ) ( α ) = n ( α ) ( t , x , y ) ( 1 - n ( t , x - Δ x , y ) N ) × ( m 0 ( α ) + m s ( α ) n ( t , x , y ) N + m d ( α ) n ( t , x - Δ x , y ) N + m l ( α ) n ( t , x , y - Δ y ) N + m b ( α ) n ( t , x + Δ x , y ) N + m r ( α ) n ( t , x , y + Δ y ) N ) , m ( x , y ) → ( x , y - Δ y ) ( α ) = n ( α ) ( t , x , y ) ( 1 - n ( t , x , y - Δ y ) N ) × ( m 0 ( α ) + m s ( α ) n ( t , x , y ) N + m d ( α ) n ( t , x , y - Δ y ) N + m l ( α ) n ( t , x + Δ x , y ) N + m b ( α ) n ( t , x , y + Δ y ) N + m r ( α ) n ( t , x - Δ x , y ) N ) , (9)
where the factors of n(α) ensure that the number of migrants is proportional to the local abundance of the strain, and the factors of 1 − n N ensure that migration cannot occur into occupied lattice sites. As a result of these choices, the spatial distribution of the strains remains “frozen” behind the growing front just as in microbial colonies, where the growth in the bulk of the colony is suppressed. The last factor in each of the equations describes the dependence of the migration rates on the local population population density and its spatial gradients; this can be seen by expanding population densities into Taylor series.
Note that our definitions preserve the equivalence of all four lattice direction because the migration coefficients are chosen according to the position of the lattice sites relative to the direction of the migration rather than relative to a particular lattice direction; see Fig 9. To emphasize this fact, we use the index labels that refer to source site, destination site, left site, back site, and right site—all specified with respect to the migration direction. For simplicity, we limited the dependence on n to the lowest order of the Taylor expansion that is sufficient to produce chiral growth.
The relationship between the model parameters and the coefficients in the continuum description is provided below:
D ( α ) ( n ) = [ m 0 ( α ) + n N ( m s ( α ) + m d ( α ) + m l ( α ) + m b ( α ) + m r ( α ) ) ] ( 1 - n N ) Δ x 2 Δ t S ( α ) ( n ) = [ 2 ( m b ( α ) - m d ( α ) ) ( 1 - n N ) + ( m 0 ( α ) + m s ( α ) + m d ( α ) + m l ( α ) + m b ( α ) + m r ( α ) ) ] Δ x 2 Δ t A ( α ) ( n ) = 2 ( m l ( α ) - m r ( α ) ) ( 1 - n N ) Δ x 2 Δ t (10)
From Eq (10), it is clear that A(α) depends on m l ( α ) − m r ( α ) while D(α) depends on m l ( α ) + m r ( α ). Thus, one can vary the chirality of a strain without affecting its motility. We used this freedom to isolate the effects of chirality from other components of strain fitness in most of our simulations by keeping m l ( α ) + m r ( α ) fixed.
The stochastic update consisted of two rounds of binomial sampling.
The first round accounted for the demographic fluctuations in the total population size. We drew n(t + Δt, x, y) from a binomial distribution with N trials and (ρ(1)(t, x, y) + ρ(2)(t, x, y))/N probability of success. This procedure ensures that (i) the expectation value of n is consistent with the deterministic dynamics, (ii) the size of a typical fluctuation scales as n for n ≪ N, and (iii) the population size never exceeds the carrying capacity N.
The second round accounted for genetic drift. We drew n(1)(t + Δt, x, y) from a binomial distribution with n(t + Δt, x, y) trials and ρ(1)(t, x, y)/(ρ(1)(t, x, y) + ρ(2)(t, x, y)) probability of success. The abundance of the other strain was set to n(2)(t + Δt, x, y) = n(t + Δt, x, y) − n(1)(t + Δt, x, y). This stochastic update does not change the relative fractions of the two strains on average, and the typical fluctuation in the relative abundance of the strains scales as n.
To ensure that our results do not arise because of the lattice effects, we developed off-lattice simulations of our reaction-diffusion model. In these simulations, cells reproduced stochastically depending on the local population density and performed short-range jumps. The magnitude of the jump was controlled by the population density and the direction of the jump depended on the local density gradient and the chirality of the cell. The functional forms of the growth rates and the jump kernels are provided in the SI.
Off-lattice simulations confirmed the predictions of our theory and lattice-based simulations. Specifically, we observed stabilizing selection and the formation of bulges between strains with opposite handedness. These results are shown in the SI. Because of computational efficiency, most of the analysis was carried out using lattice-based simulations.
|
10.1371/journal.pntd.0007109 | The impact of imperfect screening tools on measuring the prevalence of epilepsy and headaches in Burkina Faso | Epilepsy and progressively worsening severe chronic headaches (WSCH) are the two most common clinical manifestations of neurocysticercosis, a form of cysticercosis. Most community-based studies in sub-Saharan Africa (SSA) use a two-step approach (questionnaire and confirmation) to estimate the prevalence of these neurological disorders and neurocysticercosis. Few validate the questionnaire in the field or account for the imperfect nature of the screening questionnaire and the fact that only those who screen positive have the opportunity to be confirmed. This study aims to obtain community-based validity estimates of a screening questionnaire, and to assess the impact of verification bias and misclassification error on prevalence estimates of epilepsy and WSCH.
Baseline screening questionnaire followed by neurological examination data from a cluster randomized controlled trial collected between February 2011 and January 2012 were used. Bayesian latent-class models were applied to obtain verification bias adjusted validity estimates for the screening questionnaire. These models were also used to compare the adjusted prevalence estimates of epilepsy and WSCH to those directly obtained from the data (i.e. unadjusted prevalence estimates). Different priors were used and their corresponding posterior inference was compared for both WSCH and epilepsy. Screening data were available for 4768 individuals. For epilepsy, posterior estimates for the sensitivity varied with the priors used but remained robust for the specificity, with the highest estimates at 66.1% (95%BCI: 56.4%;75.3%) for sensitivity and 88.9% (88.0%;89.8%) for specificity. For WSCH, the sensitivity and specificity estimates remained robust, with the highest at 59.6% (49.7%;69.1%) and 88.6% (87.6%;89.6%), respectively. The unadjusted prevalence estimates were consistently lower than the adjusted prevalence estimates for both epilepsy and WSCH.
This study demonstrates that in some settings, the prevalence of epilepsy and WSCH can be considerably underestimated when using the two-step approach. We provide an analytic solution to obtain more valid prevalence estimates of these neurological disorders, although more community-based validity studies are needed to reduce the uncertainty of the estimates. Valid estimates of these two neurological disorders are essential to obtain accurate burden values for neglected tropical diseases such as neurocysticercosis that manifest as epilepsy or WSCH.
ClinicalTrials.gov NCT03095339.
| Epilepsy and progressively worsening severe chronic headaches are the two most common clinical manifestations of neurocysticercosis, a form of cysticercosis. To understand where the prevalence of these neurological disorders is highest for targeted infection control, valid prevalence estimates are needed. Most neuroepidemiological studies conducted in low-resource settings use a two-step approach to identify cases (screening questionnaire followed by physician examination among screened positives) to obtain prevalence and burden estimates. We found that this most commonly-used two-step approach in community-based studies leads to an underestimation of the prevalence in our study. Our paper provides an analytic solution to reduce errors in estimating the prevalence of neurological disorders in community-based studies when using the two-step approach. Our proposed approach provides more valid estimates of the prevalence of these neurological disorders and could be used to better reflect the consequences that neglected tropical diseases manifesting as epilepsy or headaches have on people’s health and disabilities.
| The Global Burden of Disease Study estimates epilepsy and migraines to be among the 30 leading causes of years of life lost due to disability [1]. The prevalence of epilepsy and headaches varies globally, likely due to different risk factors across populations or epidemiological biases [2,3]. A meta-analysis estimated the median prevalence of lifetime epilepsy in developing countries at 15.4‰ in rural and 10.3‰ in urban areas, higher than the estimated 5.8‰ in developed countries [4]. The estimated prevalence of lifetime epilepsy shows substantial variation within sub-Saharan Africa (SSA) [4–8], ranging from 7.3‰ to 29.5‰. In contrast, the prevalence of migraine in adults is reported to be higher in developed countries than that in developing countries, with an estimate of 15% in Europe and 5% in Africa [3]. The limited data collected on headaches in SSA suggest that the one-year period prevalence ranges from 3.0% to 5.4% for migraines, and 1.7% to 7.0% for tension-type headaches [9–11]. Reasons for the opposite trends in the frequency of epilepsy and headaches may be due to the distribution of socio-economical, cultural, and infectious risk factors and genetic susceptibilities, as well as important methodological differences between studies [1,3,12]. In particular, whereas prevalence data on epilepsy and headaches often originate from national surveys or nation-wide electronic health records in high income countries, it is not the case in SSA, where data originate from research projects conducted in a small number of communities. The way these neurological disorders are measured in different settings could result in important misclassification error bias which, in turn, could explain some of this variation in estimates.
Rural community-based studies in SSA often use a two-step approach to identify neurological disorder cases for prevalence estimates [13,14]. In step one, a screening test, often in the form of a questionnaire, is used to identify positive cases for the neurological disorder(s) of interest in the study population. The participants in this step can be randomly selected or identified via door-to-door visits. In step two, a physician or sometimes a neurologist confirms the given neurological disorder(s) through a medical examination, often only among those screened positive [15–19]. Diagnosis through electroencephalography is often not possible in rural community-based settings due to limited resources [13], and therefore, the physician/neurologist’s diagnostic is often considered as the gold standard.
The two-step approach is frequently used for studies in low-resource settings due to its convenience and cost-effectiveness. Despite its frequent use, the prevalence estimates from the two-step approach can be seriously biased from failure to account for the imperfect validity of the tests, referred to as misclassification error hereinafter, employed either at step one or step two, or both. The validity of a test is typically assessed by its sensitivity and specificity, and the correct sensitivity and specificity must be used to obtain unbiased prevalence estimates. However, for questionnaires used as the screening tool for epilepsy and headaches, especially those used in rural areas of SSA, little is known about their validity. In most studies, standardized screening questionnaires are used, but their validity is not determined in the study population [8–10,19]. To our knowledge, only two studies have reported validity estimates of epilepsy screening questionnaires in the general population; however, neither addresses potential verification bias in their validity estimates [20,21]. Verification bias occurs when the participants from step one have different probabilities of being selected for step two [22]. For example, individuals screened positive at step one have a much higher chance of being confirmed by a physician or neurologist, compared to those screened negative, who are rarely, if at all, examined at step two. As a result, the selected individuals at step two are not representative of the entire study population, which may lead to biased prevalence estimates.
These biases can have important consequences in evaluating the global burden of neglected tropical diseases. For example, epilepsy and progressively worsening severe chronic headaches (WSCH) are the most frequently observed clinical signs of neurocysticercosis (NCC), a preventable infection with the eggs of Taenia solium [23], which is present in communities with poor sanitation and free roaming pigs [24,25]. Most epidemiological studies evaluating the prevalence of NCC in communities will first identify people with epilepsy, and rarely with headaches, and invite them to obtain brain imaging to diagnose NCC. The proportion of NCC among people with epilepsy (or rarely headaches) is then used to estimate the prevalence of epilepsy-associated (or headaches-associated) NCC in the study population. Such estimates may then be combined to the prevalence of epilepsy or headaches in the population to estimate Disability Adjusted Life Years (DALYs) associated with NCC [26–28]. The global burden of disease initiative used this approach to estimate DALYs associated with cysticercosis [29]. Therefore, to obtain accurate estimates of the global burden of NCC, or of any neglected tropical disease causing neurological signs, we first need valid prevalence estimates of epilepsy and WSCH. From there, we may obtain more reliable assessments of the relative burden of NCC, or other neglected tropical diseases manifesting as epilepsy or headaches, compared to other infections or chronic diseases.
In this paper, we aimed to quantify the bias introduced in the prevalence estimates when failing to account for the verification bias and the imperfect validity of the screening tool using data collected from 60 villages in Burkina Faso. We also investigated the validity of screening questionnaire by estimating the sensitivity and specificity to detect epilepsy and WSCH.
Baseline cross-sectional data collected from February 2011 to January 2012 for a cluster randomized controlled trial were used. The aim of the parent study was to estimate the effectiveness of a community-based educational intervention to reduce the cumulative incidence of human and porcine cysticercosis in 60 villages of Burkina Faso [30]. From 70 to 80 individuals aged 5 years or above were sampled in each village using a cluster random sampling approach described elsewhere [30].
The University of Oklahoma Health Sciences Center Institutional Review Board and the Centre MURAZ ethical review panel (Burkina Faso) approved this study. The field staff read the written consent forms to participants and answered all their questions. Consent forms were signed, marked with a cross or a fingerprint by those who were unable to write. All consent forms were signed by a witness. Parents of individuals 5 to 16 years of age gave consent for their children. Individuals aged 10 to 15 years were invited to give their assent. All participants were given a bar of soap as incentive. The parent study was registered through clinicaltrials.gov (NCT03095339).
In step one of the two-step approach, each participant was screened for epilepsy and WSCH using a screening questionnaire (S1 Questionnaire). Questions related to epilepsy were based on the International League Against Epilepsy screening of epilepsy questionnaire developed by Preux et al.[31], and was previously used in three villages in the same study area [32]. Questions related to headaches were designed to capture NCC-related headaches [33].
In step two, all individuals screened positive for either epilepsy or WSCH from the screening questionnaire were invited to be examined by a study physician. In addition, 231 screened-negative individuals were randomly selected to be examined by the physician. The medical examination results were collected on a medical examination questionnaire (S2 Questionnaire). Two medical examination rounds took place with the second round aimed at capturing individuals who were absent during the first round and to examine patients from the more remote province of Nayala. The physicians discussed any uncertain diagnosis with the neurologist on the phone at the time of the medical examination. At the end of the study, the neurologist reviewed all diagnoses and his final diagnosis was considered as the gold standard. The diagnostic result that confirmed whether the individual had epilepsy and/or WSCH was used to assess the validity of the screening questionnaire.
Epilepsy was defined as having more than one seizure of central nervous system origin without apparent cause [34]. Individuals not meeting the epilepsy case definition were considered as epilepsy-free (i.e., screened negative for epilepsy). Six individuals diagnosed with single epileptic seizures were excluded from all analyses.
WSCH was defined as having symptoms arising more than weekly for two weeks or more, with each episode lasting at least 3 hours, and progressively worsening in severity with time. Headaches had to be severe enough to require analgesic or to prohibit working, playing, attending school, or partaking in daily activities [35]. Individuals not meeting the WSCH definition were considered as WSCH-free (i.e., screened negative for WSCH).
The statistical analyses aimed at assessing the bias from failing to correct for both verification bias and misclassification error, when estimating the prevalence of epilepsy and WSCH in the study population. To assess the degree of the bias, we first calculated the unadjusted estimates, where no correction was made for verification bias and misclassification error. The unadjusted estimates were obtained by running a Bayesian binomial model, where the number of confirmed cases of either epilepsy or WSCH was assumed to result from a binomial distribution with a probability corresponding to the unadjusted prevalence and the number of individuals screened. The prior choice for the prevalence parameter is discussed below.
The adjusted prevalence estimate was obtained by running a Bayesian latent-class model [22]. In this model, the probabilities that participants were selected for the confirmation test at step two were specified by a set of conditional distributions. Participants with different selection probabilities had different conditional probability distributions. This way, the selection probabilities were correctly accounted for, eliminating verification bias. To obtain the specificity and sensitivity estimates of the screening questionnaire while correcting for verification bias, we provided prior information on the model parameters of sensitivity and specificity. Different priors, including both informative and non-informative (i.e., vague) priors, were investigated in modeling either epilepsy or WSCH. Specifically, for epilepsy, one set of informative priors based on the sensitivity and specificity estimates for similar epilepsy screening questionnaires from two previous community-based studies [25, 26] were used. In these studies, sensitivity and specificity were estimated as 92.9% and 79.3%, and 99.6% and 72.4%, respectively. To allow some variability, the prior sensitivity and specificity values in our analysis were assumed to follow beta distributions with mean based on these validity estimates and a standard deviation of 0.05. This led to a Beta(54.8, 17.5) prior for sensitivity, and a Beta(12.9, 0.5) for specificity. We also considered the vague priors of Unif(0.5, 1) for both parameters. Since epilepsy was highly stigmatized in SSA [6] and some forms of partial epilepsy may be difficult to identify, we also used a vague prior of Unif(0.3, 1) for the sensitivity of epilepsy screening. Due to the lack of validity studies for WSCH screening, the same informative priors as epilepsy were adopted for WSCH. We also ran the model with vague priors of Unif(0.5, 1) for both the specificity and sensitivity, and results were compared with those obtained using informative priors.
When modeling each epilepsy and WSCH, we used a vague Unif(0, 0.3) for the unadjusted prevalence estimate. For the adjusted prevalence parameter, a vague prior of N(0, 10) was used (on logit scale).
We used WinBUGS [36] for all the Bayesian analyses and reported the posterior mean and 95% credible intervals. The bias was then evaluated by the ratio of the adjusted to the unadjusted estimates, with the Bayesian credible interval for the ratio excluding 1 indicating the existence of bias.
We examined three scenarios reflecting different screening strategies often used in community-based studies conducted in low-resource areas. In Scenario 1, we estimated the unadjusted prevalence using the screening information for only one neurological disorder instead of for epilepsy and WSCH, to reflect a common situation in the existing literature where only one neurological disorder was studied at a time instead of both disorders simultaneously. In Scenario 2, we estimated the unadjusted estimate using the information for participants that were only examined in the first medical round. This scenario was considered to reflect situations where there were insufficient personnel and monetary resources to find people absent from the village during the initial visit. Scenarios 1 and 2 are fairly common in field studies conducted in resource-poor contexts. In Scenario 3, we only used the screening information for the unadjusted prevalence estimate, reflecting a situation where the validity of the screening questionnaire cannot be assessed. We calculated the adjusted estimates resulting from each scenario, and evaluated the bias estimating the ratio of the verification bias and misclassification error-adjusted to the unadjusted estimates.
In this exercise, the impact of verification bias and misclassification error on community-based estimates of NCC-associated epilepsy and WSCH prevalence and number of cases was evaluated. To obtain the estimated prevalence of NCC-associated epilepsy and WSCH for our study population, the adjusted and unadjusted estimates of epilepsy and WSCH prevalence were multiplied by the proportion of NCC among people with epilepsy reported in a meta-analysis [37] and the proportion of NCC among people with headaches reported in a case-control study [38]. We then investigated the difference between the adjusted and unadjusted prevalence estimates of NCC-associated epilepsy and WSCH. To obtain the difference between the number of NCC-associated epilepsy and WSCH cases in our study population, the estimated adjusted and unadjusted prevalence estimates were multiplied by the study population size. We assumed that the mean number of NCC-associated cases of epilepsy and WSCH would follow beta distributions with parameters based on the estimated means in the published studies (29% for epilepsy and 4.7% for headaches) [37,38] and a standard deviation of 2.75%. All estimates were run using the set of informative priors described earlier. The adjusted prevalence and number of cases of NCC-associated epilepsy and WSCH were also estimated under the three screening strategies described above.
A total of 4794 individuals were sampled at baseline, including the analytical sample of 4768 with complete screening data. Of these, 669 (14.0%) screened positive for epilepsy (7.1%), WSCH (2.8%) or both (4.2%) (Table 1). A physician examined 609 (91.0%) screened-positive and 231 (5.6%) screened-negative individuals. The higher proportion of screened-positive examined by a physician compared to those screened-negative showed evidence for the need to adjust for potential verification bias (Fig 1).
Of those examined in step two, 748, 57 and 35 were seen at the first, second and both medical rounds, respectively. Perfect agreement was observed for those examined at both rounds. Table 1 describes the characteristics of the analytical sample, and the screening and medical examination results. The majority of participants were either a farmer or housewife and a high proportion did not complete primary school.
Table 2 provides the posterior estimates of the unadjusted prevalence and the corresponding adjusted prevalence with the associated sensitivity and specificity estimates, using the different sets of priors. The epilepsy screening questionnaire showed posterior sensitivity and specificity estimates of moderate variation with different priors used, while the specificity estimates remained robust. The posterior sensitivity estimate was 44.7% (95%BCI: 33.0%;60.0%) with the most vague prior and increased to 66.1% (95%BCI: 56.4%,75.3%) with the informative priors based on previous literature. Posterior sensitivity and specificity estimates of the WSCH screening questionnaire were less affected by the priors. Despite the variation in the estimated sensitivity and specificity, the unadjusted prevalence estimates were consistently lower than the adjusted ones for both epilepsy and WSCH, as indicated by the posterior bias distribution lying under the value of 1. Somewhat less bias was observed using the informative priors.
Bias (i.e., where the unadjusted estimate was smaller than the adjusted estimate) was introduced for both epilepsy and WSCH (Figs 2 and 3) under the two most common screening strategies found in the literature, namely Scenarios 1 (i.e. when we assumed that the study would only screen for one neurological disorder) and 2 (i.e. when resources would not allow for returning to communities to examine those absent during a first visit). The magnitude of the bias was similar to that observed in the main analyses for epilepsy, but more marked for WSCH. For Scenario 3 (i.e. when only screening results were used), an important positive bias was present where the unadjusted prevalence estimate of 11.2% (95%BCI: 10.4%; 12.2%) was considerably larger than the adjusted prevalence for epilepsy when more informative priors were used. For WSCH, Scenario 3 resulted in a negligible positive bias. The magnitude of the bias did not vary extensively by the priors used for epilepsy and WSCH, except for epilepsy in Scenario 3.
Using the study data, the differences between the adjusted and unadjusted prevalence and number of NCC-associated epilepsy were 0.6% (95%BCI: 0.3%; 1.0%) and 29 cases (95%BCI: 13; 49), respectively.
This was the first study to estimate the sensitivity and specificity of a screening questionnaire for epilepsy and WSCH in the community while adjusting for both verification and misclassification error bias. This was also the first study to quantify the bias from failing to account for verification and misclassification error bias when estimating the prevalence of epilepsy and WSCH in a community-based study.
Our sampling strategy was not designed to provide population prevalence estimates of epilepsy and WSCH in Burkina Faso as a whole, in the three study provinces, or even the villages selected for the parent study. Indeed, our sampling strategy, which favored concessions with pigs, was likely to result in higher estimates of prevalences than what would have been observed if a simple random sampling strategy had been adopted. Nonetheless, we provided below prevalence estimates from other community-based studies conducted in resource-poor settings using the two-step approach. In our study sample, the unadjusted prevalence of epilepsy was higher than that of 1.6% (95%CI: 1.2%;2.0%), 1.1% (95%CI: 0.9%; 1.4%) and 0.5% (95%CI: 0.2%;0.8%) found in Benin, Tanzania and Nigeria, respectively [7,15,39]. Our estimate was also higher than that of 0.6% (95%CI: 0.5%;0.7%) found in Cambodia where a screening questionnaire similar to ours was used [40]. Similarly, our unadjusted estimate was higher than the first study conducted in rural Burkina Faso, which estimated an epilepsy prevalence of 1.1% in 18 villages [41]. Our prevalence estimate was most similar to a recent Burkinabé study, conducted by the same research group, which estimated a lifetime epilepsy prevalence of 4.5% (95%CI: 3.3%;6.0%) in three villages purposely sampled to have a high prevalence of epilepsy [32]. This supported the suspicion that our study sample might represent people at higher risk of epilepsy than the general population.
Our unadjusted estimate of lifetime WSCH prevalence was similar to the lifetime migraine prevalence estimate of 3.3% (95%CI: 2.4%;4.6%) in Benin [42], and lower than the lifetime migraine prevalence estimate of 5.3% (95%CI: 5.0%; 5.6%) in Nigeria [43]. Lifetime prevalence estimates of tension-type headaches in SSA were unavailable for comparison. Compared to published one-year prevalence estimates of tension-type headaches, ours was higher than the 1.7% (95%CI: 1.5%; 1.9%) prevalence in Ethiopia among adults 20 years and older, and lower than the 7.0% (95%CI: 6.5%;7.6%) prevalence in Tanzania among all ages [9,11]. This suggested that our sampling strategy might have not influenced the estimated frequency of WSCH as much as it did for epilepsy.
The posterior estimates of the specificity for the neurological screening questionnaire were similar for epilepsy and WSCH, and were consistently around 88%. These estimates were slightly lower than those previously reported [20,21], possibly because our study population had a higher proportion of false negatives compared to the previously published validity studies. Study participants in our study may also have been less likely to report their symptoms compared to those in Ecuador and Bolivia, perhaps due to the stigmatizing effects of epilepsy in SSA [6]. The posterior estimates of sensitivity varied depending on the priors used, particularly for epilepsy. The sensitivity posterior median was higher when prior knowledge [20,21] information was used and lower for vague priors; although the credible intervals overlapped. A similar observation where the sensitivity parameter was more affected to the prior choice was also found in a study assessing the validity measures of a screening test for human papillomavirus [44].
Our sensitivity estimates were lower and specificity estimates higher for epilepsy than that found in validation studies conducted in clinical settings with estimates of sensitivity between 91% and 100% and specificity between 51% and 85% [19,20,45]. Placencia et al. found lower sensitivity and similar specificity estimates when the same screening questionnaire was used in the community as compared to the clinic [20]. Such discrepancies may be due to spectrum bias where populations in clinical settings have more severe disease and acknowledge their symptoms more, thereby increasing the test sensitivity. The observed estimates from clinical settings may also result in part from verification bias, which typically leads to an overestimation of the sensitivity and underestimation of the specificity [22]. Our sensitivity estimates for epilepsy were similar to the two other community-based studies with sensitivities of 72.4% (95% CI: 52.8–87.3) and 79.3% [20,21]. To our knowledge, prevalence studies using the two-step approach for WSCH have not reported validity estimates.
Even in a situation where 90% of those screened positive were examined by a physician, significant biases were observed. The unadjusted prevalence estimates were consistently lower than the adjusted prevalence estimates regardless of the priors used. However, the adjusted prevalence estimates were highly dependent on priors, particularly for epilepsy, which introduced considerable uncertainty. Such uncertainty could be reduced by conducting more community studies assessing and reporting validity estimates of screening. The validity estimates were expected to vary from one community to the next, due to how epilepsy and WSCH were reported by patients and to the way interviewers were trained to use the questionnaire. This would also result in varied bias estimates. Therefore, the reported sensitivity and specificity estimates of the screening questionnaire may not be applicable to other communities.
We chose the Bayesian framework [46,47] to simultaneously correct for verification bias and misclassification error in the two-step approach for two reasons. First, verification bias can be treated as a missing data problem, and therefore be corrected in a straightforward manner. Second, verification bias and misclassification error can be addressed simultaneously by including an additional level in the Bayesian model estimating the specificity and sensitivity of the invalid test(s). The capture-recapture method is an alternative approach for obtaining corrected prevalence estimates [13]. This method combines multiple sources of information independently, such as medical records and non-medical interviews with community members, with the two-step approach. This method yielded higher prevalence estimates than the two-step approach alone in two studies in Benin [7,42]. Our bias estimation method using their data yielded similar results (between 0.5 and 0.3). However, the capture-recapture method has multiple difficult-to-meet assumptions: closed population, statistical independence between sources, identical case definitions across multiple sources and requires more personnel resources [7,42]. Our study was less resource-intensive and provided an alternative to the capture-recapture method.
When screening strategies commonly used in community-based studies were explored, the unadjusted prevalence was lower than the adjusted prevalence in the two most frequently encountered scenarios, and the bias estimates were similar to that observed under the screening strategy used in the main analysis of this study. In our study, we had resources to capture those missed through the first medical round and we screened for two neurological outcomes, which resulted in more individuals being confirmed in step two compared to most community-based studies. Despite this more complete examination of individuals screened positive, the level of bias was similar to that estimated for screening strategies commonly used in community-based studies. In the last scenario explored, the prevalence relied entirely on the screening questionnaire, which led to a large number of false positive cases of epilepsy. This was especially true because the prevalence of epilepsy was relatively low, and hence, there were relatively more people without epilepsy who were false positives than people with epilepsy who were false negatives, even if the screening test’s specificity was much better than its sensitivity. As opposed to the other scenarios, failure to use physician confirmation led to an overestimate of the adjusted prevalence of epilepsy. This scenario could occur when economical and personnel resources are too limited for physician confirmation. For WSCH, the estimated negative bias (i.e. when the unadjusted estimate was lower than the adjusted estimate) was more marked in the two scenarios most often encountered in community-based studies than when using the screening strategy in the main analysis of this study. This suggested that by only screening for WSCH with or without two rounds, more WSCH cases were missed. There is rising evidence that seizures might increase the risk of headaches [48,49]. Hence, it is possible that screening for both outcomes improved the detection of WSCH through the neurological examination of people screening positive for epilepsy. The opposite (i.e. the screening of people with WSCH increases the detection of epilepsy) might not be as marked in our study because a lot less participants screened positive for WSCH than for epilepsy or both epilepsy and WSCH. In the last scenario where the prevalence estimate relied solely on the screening questionnaire, we observed negligible positive bias for WSCH. This finding illustrated that screening for both epilepsy and WSCH led to a less biased estimate of the prevalence compared to the more commonly used approach of only screening for one neurological condition. This was not observed for the last scenario for epilepsy because screening for WSCH along with epilepsy might not increase the detection of epilepsy.
When we examined the impact of verification and misclassification bias on the estimation of the prevalence and frequency of NCC-associated epilepsy and WSCH, we found that the prevalence and number of NCC were underestimated when using two-step approach regardless of the scenario. In the scenario without confirmation by a physician or neurologist, we found that the number of NCC-associated epilepsy cases were over-estimated, while we did not observe a difference between the unadjusted and adjusted prevalence and number of NCC-associated WSCH cases. These results could have important consequences in the estimation of the monetary and non-monetary burden of NCC locally and globally. For example, the estimated global DALYs of NCC-associated epilepsy were estimated to be 2,788,426 (95% Uncertainty Estimates (UI): 2,137,613–3,606,582) by the Foodborne Epidemiology Research Group in 2010 [50] and to be 468,100 (95% UI: 322,900–625,800) in 2016 by the Global Burden of Disease 2016 Collaborative [51]. Moreover, the 2015 DALYs for migraine were estimated to be 32,899,000 (95%UI: 20,295,000–48,945,000) [52]. If these estimates used underlying NCC-associated epilepsy and headaches frequencies which were underestimated to a similar level as in our study, it could have important consequences on how NCC would rank among all diseases locally and globally and on policy making. Indeed, if the burden of NCC were higher than currently believed, more resources should be allocated to control it.
Our study has several limitations. First, the physicians examined a small proportion of individuals screened negative. Increasing this proportion would have reduced the variance of our posterior sensitivity estimates. Second, our study used two medical rounds with different physicians. However, perfect agreement between the two physicians was observed and all diagnoses were reviewed and confirmed by the neurologist, minimizing the possibility for bias. Third, we could not conclude that our prevalence estimates for the study population were reflective of the prevalence for the villages due to the sampling scheme. Since the aim of our study was to quantify the bias that rose from failure to account for misclassification error and verification biases by comparing the unadjusted estimates to the adjusted estimates, we believe our findings were still of importance.
Our results suggest that the burden of epilepsy and WSCH in low-resource settings might be much higher than previously reported. Future studies should consider using the statistical models presented here to account for the imperfect nature of screening questionnaires. Bias-adjusted prevalence estimates of these two neurological disorders will improve our understanding of the burden of these conditions and help identify where cysticercosis may be present. More valid prevalence estimates will allow for the development of targeted cysticercosis control programs in those communities.
|
10.1371/journal.ppat.1006251 | Host cell-derived lactate functions as an effector molecule in Neisseria meningitidis microcolony dispersal | The development of meningococcal disease, caused by the human pathogen Neisseria meningitidis, is preceded by the colonization of the epithelial layer in the nasopharynx. After initial adhesion to host cells meningococci form aggregates, through pilus-pilus interactions, termed microcolonies from which the bacteria later detach. Dispersal from microcolonies enables access to new colonization sites and facilitates the crossing of the cell barrier; however, this process is poorly understood. In this study, we used live-cell imaging to investigate the process of N. meningitidis microcolony dispersal. We show that direct contact with host cells is not required for microcolony dispersal, instead accumulation of a host-derived effector molecule induces microcolony dispersal. By using a host-cell free approach, we demonstrated that lactate, secreted from host cells, initiate rapid dispersal of microcolonies. Interestingly, metabolic utilization of lactate by the bacteria was not required for induction of dispersal, suggesting that lactate plays a role as a signaling molecule. Furthermore, Neisseria gonorrhoeae microcolony dispersal could also be induced by lactate. These findings reveal a role of host-secreted lactate in microcolony dispersal and virulence of pathogenic Neisseria.
| The human restricted pathogen Neisseria meningitidis is a major cause of bacterial meningitis and sepsis worldwide. Colonization of the mucosal layer in the upper respiratory tract is essential to establish invasive disease. The initial interaction with host cells is characterized by bacterial proliferation and adhesion as aggregates, called microcolonies. Detachment from microcolonies in the nasopharyngeal epithelium facilitates crossing of the cell barrier that can result in invasive disease, yet this process is poorly understood. Here we demonstrate that lactate, an abundant molecule in host mucosal environments, induces N. meningitidis microcolony dispersal. Interestingly, metabolic utilization of lactate by the bacteria was not required for the process, suggesting that lactate play a role as a signaling molecule in pathogenic Neisseria. We propose that the microcolony dispersal in pathogenic Neisseria is influenced by environmental concentrations of lactate. These findings will assist in better understanding the transition from asymptomatic carriage to invasive disease.
| Humans serve as the sole reservoir for the pathogen Neisseria meningitidis. The bacteria asymptomatically colonize the upper respiratory tract, with a prevalence of carriage ranging from 10 to 35% [1]. However, meningococci occasionally cross the epithelial mucosa and blood-brain barrier, causing life-threatening septicemia and meningitis [2]. Meningococcal adhesion to the epithelium in the nasopharynx is a prerequisite for colonization and pathogenicity [3], and this process can be divided into two steps. The initial interaction with cells is characterized by proliferation and adhesion as aggregates, called microcolonies. This stage is followed by the detachment of individual bacteria from the microcolonies, enabling relocation to new colonization sites or intimate adhesion to the cells in a single bacterial layer. Dispersal from microcolonies enables the meningococci to invade the mucosa and enter the circulation [4, 5].
The process of neisserial autoaggregation is highly dynamic and driven by interactions between type IV pili (Tfp). Tfp are one of the most multifaceted prokaryotic virulence factors mediating adhesion, autoaggregation, DNA uptake, biofilm formation, and twitching motility (reviewed in [6]). The dynamics of aggregation have been shown to be affected by several factors, including the minor pilin PilX, PilW and recently characterized polynucleotide phosphorylase (PNPase) [7–9]. The abundance of Tfp has also been shown to be critical for aggregation [10]. In addition to the amount of pili present on the surface of the bacterium, post-translational modifications of the major pilin subunit PilE have been shown to affect aggregation and dispersal. These modifications include O-linked glycosylation and addition of phosphocholine, phosphoethanolamine, or phosphoglycerol moieties [11–13]. The induction of PilE phosphoglycerol transferase B (PptB) and increased phosphoglycerol moieties on PilE favor the dissociation of pili bundles and subsequent microcolony dispersal [14]. A recent study showed that neisserial microcolony stability was affected by the oxygen concentration. Depletion of oxygen resulted in a loss of proton motive force (PMF), which affected the pilus-pilus interaction and led to a subsequent detachment of bacteria from microcolonies in liquid [15]. Although several factors have been shown to contribute to detachment of bacteria from microcolonies, the underlying mechanism is poorly understood.
In this study, we investigated the role of host cell-derived molecules in microcolony dispersal. Bacterial pathogens can sense host signaling molecules to acquire information about the host physiological status and alter virulence properties accordingly to adapt to distinct niches. In the gut, Escherichia coli O157:H7 (EHEC) sensing of ethanolamine, a source of carbon and nitrogen, and the sugar fucose can play an important role in regulating virulence genes involved in colonization. This mechanism occurs independently of metabolic utilization of the nutrients by the bacteria [16, 17]. Additionally, Salmonella typhimurium can utilize ethanolamine signaling to modulate metabolism and virulence [18]. N. meningitidis responds to a component present in human saliva, 4-hydroxyphenylacetic acid (4HPA), by altering gene regulation of the adhesins nadA and mafA, as well as metabolic pathways; 4HPA binds directly to and inhibits the activity of the FarR regulator [19–22]. This compound is a catabolite of aromatic amino acid metabolism, but it is unclear whether 4HPA is from the host cells or microbiota [23].
Here, we demonstrate that accumulation of a host cell-derived molecule, identified as lactate, is capable of inducing a rapid synchronized dispersal of meningococcal microcolonies. The dispersal was not dependent on the utilization of lactate by the bacterium. Additionally, the dispersal was independent of the previously reported mechanisms of microcolony dispersal mediated through depletion of the proton motive force. Altogether, our findings suggest that host-derived lactate plays an important role as a signaling molecule that mediates microcolony dispersal in pathogenic Neisseria.
The aim of this study was to investigate the process of N. meningitidis microcolony dispersal. To evaluate the necessity of direct interaction with host cells, we monitored the timing of the dispersal upon infection of pharyngeal epithelial FaDu cells at different confluences using live-cell time-lapse microscopy. The dispersal phase began after approximately 4.5 h in the presence of cells and after 6.5 h in the absence of cells (Fig 1A). The total time the bacteria spent in the dispersal phase was drastically shorter in the presence of cells, lasting approximately 20 min, while in the absence of cells bacteria were still detaching from microcolonies after 8 h of incubation (Fig 1A). Microcolonies on cells at confluences of 100%, 80%, 50% and 20% showed no differences in dispersal phase length. Interestingly, the microcolonies that were not in direct contact with host cells (Fig 1B) also showed a synchronized and short dispersal phase similar to microcolonies that were in direct contact with host cells (S1A Fig).
To further examine the importance of direct contact prior to microcolony dispersal, we carried out the infection using an adhesion-deficient ΔpilC1 mutant [24]. In an infection assay with FaDu cells, the ΔpilC1 mutant showed a short and synchronized dispersal beginning after approximately 4 h of incubation, similar to that of the wild-type, supporting that cell contact is not required (Fig 1C). The adhesive properties of both the wild-type and the ΔpilC1 mutant strains was confirmed by performing an adhesion assay. As expected, the ΔpilC1 mutant adherence decreased by 10-fold in a comparison to the wild-type (S1B Fig).
The importance of live epithelial cells was further investigated by observing dispersal on FaDu cells that were fixed before infection. The microcolonies on fixed cells showed a long dispersal phase (Fig 1D). Microcolony dispersal was also monitored on the epithelial cell lines A549 (lung), Detroit 562 (pharynx), and Hec-1B (endometrium) to confirm that the observed effect of human cells was not specific to the FaDu cells. The length of the dispersal phase during the A549, Hec-1B and Detroit 562 infections was short, resembling that observed with the FaDu cells (S2A Fig).
Taken together, the data indicated that live host epithelial cells are important for N. meningitidis microcolony dispersal. The detachment of meningococci from microcolonies was synchronized and rapid in the presence of cells. The microcolony dispersal did, however, appear to be independent of direct contact between the bacteria and the cells.
Since direct contact with host cells was not necessary for microcolony dispersal, we chose to study whether a secreted compound(s) might stimulate this response. Since N. meningitidis in liquid without cells did not disperse as quickly, the compound(s) must originate from the host cells themselves or be generated during the cell-bacteria interaction. As bacteria had dispersed after 5 h in the presence of host cells, we collected medium after 5 h of incubation with both infected and uninfected FaDu cells. The conditioned medium (CM) from the infected or uninfected cells was added at a 1:1 volume ratio to bacterial suspensions in which microcolonies had been allowed to form for 3 h (Fig 2A). Regardless of whether the CM was from the infected or uninfected cells, a short dispersal phase was initiated after approximately 20 min and finished 20 min later (Fig 2B, 2C and S1 Movie). The microcolony dispersal of microcolonies, which had been treated with control medium, began after 1 h and was followed by a dispersal phase that lasted several hours (Fig 2B, 2C and S2 Movie). Induction assays performed with CM from other human cell lines induced dispersal similar to that of the CM from FaDu cells (S2B Fig).
We next investigated the time required for the medium to accumulate a sufficient amount of the microcolony dispersal factor(s). While CM collected between 5 min and 5 hours all accelerated the dispersal, the effect increased with time, with the fastest dispersal in CM collected after 5 h (Fig 2D). To determine whether the CM could also prevent microcolony formation, N. meningitidis were resuspended in CM at the beginning of the experiment. Interestingly, the bacteria formed microcolonies after approximately 2 h, similar to the bacteria in control medium, but dispersed 30 min later (Fig 2E). This indicates that N. meningitidis must first aggregate and can then respond to the CM.
In response to CM, we detected significant changes in expression of the pilus-associated genes pilE, pilT, pilC1, pilC2, pilX, pilV, pilW, pptB [4, 7, 9, 14, 25, 26] and the transcriptional regulator crgA [27, 28], (Fig 3A). However, no significant changes were observed in expression of the transcriptional regulators misR or pnp [8, 29]. Additionally, no significant changes were observed in the genes pglC, pglI, pglB2, pglH and pglL, encoding for pilus posttranslational modification enzymes [30], (Fig 3A). Moreover, we analyzed the protein level of PilE, PilT, PilC, PilX and PilW at 10 min after addition of CM to microcolonies. We did not observe any significant changes on the protein level upon addition of CM (Fig 3B). The results indicate that expression of the proteins PilE, PilT, PilC, PilX and PilW, previously associated with meningococcal colonization, was not changed upon induction of microcolony dispersal. To summarize, these results indicate that N. meningitidis in microcolonies disperses in response to one or more compounds derived from host cells. The activity of the CM increased with incubation time, suggesting that the active molecule(s) accumulate over time.
To characterize the cell-derived compound(s) actively inducing meningococcal microcolony dispersal, we treated the CM with a protease inhibitor cocktail, proteinase K, trypsin, chymotrypsin, DNase I, RNase A, EDTA, or EGTA before addition to the microcolonies. The treated CM samples were fully active after the different treatments, as well as after heat inactivation (Fig 4A). These results indicated that the active component was not a protein, DNA, RNA, a divalent metal ion or a heat-sensitive molecule. Negative controls consisting of treated control medium did not affect microcolony dispersal (S3 Fig). In addition, the pH of the CM was measured and remained unchanged from the control medium. To further analyze the compound(s), we passed the CM over a 3 kDa cut-off filter. As shown in Fig 4B, the flow-through retained the ability to trigger microcolony dispersal, while the retentate, containing molecules larger than 3 kDa, did not, suggesting that the molecule is small in size. The treatments of the CM indicate that the compound(s) is smaller than 3 kDa, heat stable and is not a protein, a metal ion, DNA or RNA.
The accumulation of a low-molecular weight molecule over time in the CM led to the hypothesis that it may be a host cell metabolic end product. One such molecule is lactate, which is produced and excreted as a part of glucose fermentation in human cells. Lactate has previously been shown to stimulate metabolism and oxygen consumption in pathogenic Neisseria [31, 32]. A number of studies have examined a link between lactate metabolism and virulence in pathogenic Neisseria (reviewed in [33]). To test the ability of lactate to induce microcolony dispersal, we performed induction assays using L-lactate, which is the predominant lactate isoform found in the human body [34]. This assay showed that L-lactate was able to induce microcolony dispersal, similar to CM (Fig 5A). Interestingly, D-lactate was also able to induce dispersal in the same manner as L-lactate (Fig 5B). The minimum concentration required to induce dispersal was 0.4 mM for both L-lactate (Fig 5A) and D-lactate (Fig 5B). Since most lactate produced by human cells is derived from glucose fermentation, we collected CM from cells that were grown in absence of glucose. While the glucose depletion did not fully abolish the activity of the CM, it prolonged the dispersal phase by 1.5 hours compared to normal CM (Fig 5C).
Using reversed-phase HPLC, we separated the CM using an acetonitrile and water gradient. Fractions were collected every three minutes for a total 12 min (Fig 5D). The collected fractions were dried in a Speed-Vac and then resuspended in water, and the activity was examined in an induction assay. In addition, lactate was separated in the same way as CM. For both CM and lactate, fraction 2 was able to induce fast dispersal (Fig 5E and S4 Fig). The remaining fractions did not induce dispersal (Fig 5E and S4 Fig). The presence of lactate in the CM was confirmed by quantification. The concentration in the CM was 2 mM, and in the CM from cells grown in absence of glucose, the concentration was 0.15 mM (S5 Fig). This further supports the hypothesis that lactate is the effector molecule in the CM that induces rapid dispersal of microcolonies. To examine that the observed lactate-induced dispersion was not due to bactericidal effects exerted by lactate itself, we examined the viable count during the induction assays. We did not observe any difference in bacterial viability when DMEM containing 50 mM lactate was added to the microcolonies (S6 Fig).
To establish that lactate-induced microcolony dispersal was not only restricted to the strain FAM20, we tested N. meningitidis serogroup W-135 strain JB515 and N. gonorrrhoeae strain MS11. Lactate induced similar effects on microcolony dispersal of N. meningitidis serogroup W strain JB515 and Neisseria gonorrhoeae strain MS11 (Fig 6).
Together, these data indicate an important role of lactate in microcolony dispersal of pathogenic Neisseria. Both L- and D-lactate at millimolar concentrations was sufficient to induce rapid dispersal.
Meningococcal lactate metabolism is dependent on lactate permease (LctP) to take up lactate and a lactate dehydrogenase (LDH) to oxidize it to pyruvate [35, 36]. N. meningitidis is known to encode at least 3 LDHs. Two are membrane-bound respiratory LDHs, specific for either L-lactate (lldA) or D-lactate (ldhD), and one is a cytoplasmic D-lactate LDH (ldhA) [37–39]. To further assess the importance of lactate in microcolony dispersal, we created ldhD, lldA, ldhA and lctP deletion mutants of FAM20. Growth analysis was performed to determine the ability of the strains to grow on lactate as a carbon source (Fig 7A–7C). Consistent with previous studies, the ΔlctP mutant was unable to utilize either L- or D-lactate but grew normally in the presence of glucose. The ΔldhD mutant was unable to utilize D-lactate but grew normally on both L-lactate and glucose. With L-lactate as a carbon source, the ΔlldA mutant was unable to grow and the ΔldhA mutant was impaired in growth (Fig 7A–7C).
To investigate whether lactate uptake and metabolism were required for induction of dispersal, ΔlctP, ΔldhA, ΔlldA and ΔldhD mutants were used in induction assays. Interestingly, deletion of lctP accelerated microcolony dispersal in control medium (Fig 7D). Although ΔldhD mutant was unable to utilize D-lactate, microcolony dispersal could be induced by addition of D-lactate (Fig 7E). The same could be observed by addition of L-lactate to microcolonies formed by ΔldhA and ΔlldA mutants (Fig 7F–7G). This suggests that lactate does not need to be metabolized to induce rapid dispersal.
One factor that controls microcolony dispersal in N. gonorrhoeae is the oxygen concentration. Oxygen depletion leads to dispersal through pili retraction, which is mediated by depletion of the proton motive force (PMF) [15]. Since it has been shown that lactate stimulates growth and oxygen consumption in the presence of glucose [32], one hypothesis was that this increased oxygen consumption leads to oxygen depletion and dispersal. Changes in PMF affect the ATP level and the NAD+/NADH ratio. To determine if depletion of the PMF caused dispersal in response to lactate addition, we measured the concentration of ATP and the ratio of NAD+/NADH after induction of microcolony dispersal with either lactate or CM. Our results showed that the ATP concentration and the NAD+/NADH ratio remained similar or increased slightly (Fig 8A and 8B). The addition of carbonyl cyanide m-chlorophenyl hydrazone (CCCP, 25 μM) caused a decrease in ATP concentration (S7 Fig). This suggests that a depletion of the PMF caused by oxygen depletion was not the cause of dispersal. In addition, we were unable to induce dispersal by decreasing the environmental oxygen concentration of bacteria grown in DMEM (Fig 8C), in contrast to bacteria grown in GC as shown by Dewenter et al. [15]. However, one of the reasons for the observed differences in GC and DMEM for the role of oxygen depletion in microcolony dispersal may be due to differences in the rate at which oxygen depletion takes places in both the medium. Further investigation is required to conclusively determine the role of oxygen depletion in the lactate-induced accelerated microcolony dispersal.
Induction assays were also performed with addition of pyruvate since it is the metabolite directly downstream of lactate and has been shown to have the same stimulatory effect on Neisseria metabolism as lactate [40]. Addition of pyruvate did not induce microcolony dispersal, which further suggests that the effect is not due to an increase in metabolic rate (Fig 8D).
These data indicate that the observed effect on dispersal is independent of the previously described mechanism that relied on depletion of oxygen and the membrane potential.
The transition from nasopharyngeal colonization to an invasive infection is a crucial step in meningococcal pathogenicity. The detachment of meningococci from microcolonies allows bacteria to colonize new sites and to act as single cells that can cross the epithelial barrier [5, 14]. In this study, we investigated the importance of epithelial cells and cell-derived factors for microcolony dispersal. We demonstrated that the previously observed short and synchronized dispersal of microcolonies [41] requires the presence of live epithelial cells but not direct contact between cells and bacteria. Microcolony dispersal could be induced by a low-molecular weight host cell-derived factor that accumulated in cell-conditioned medium (i.e., CM) in absence of infectious agent. Furthermore, we showed that lactate is the active inducer of rapid microcolony dispersal in both N. meningitidis and N. gonorrhoeae. We propose that the microcolony dispersal in pathogenic Neisseria is influenced by environmental concentrations of lactate (Fig 9). Our data reveal a potential role of lactate as an effector molecule in colonization of pathogenic Neisseria.
To survive and proliferate within the host, bacteria must be able to sense and respond to changes in the environment. Many pathogens are known to respond to nutrient availability in the environment to regulate virulence-related gene expression (reviewed in [42]). Several studies have highlighted the importance of energy metabolism for meningococcal colonization [35, 43]. Neisseria is restricted to glucose, pyruvate and lactate as carbon sources [36]. While the concentrations of pyruvate generally are low, glucose and/or lactate are present in the tissues colonized by Neisseria. During normal conditions, L-lactate is the dominant isoform found, but D-lactate can also be present as a result of fermentation from bacteria such as lactobacilli [44]. Lactate is the major carbon source on mucosal surfaces where lactic acid bacteria colonize, and the concentration of lactate in nasopharyngeal tissue can be higher than 1 mM [35]. Lactate is also present at high concentrations in the female genital tract colonized by N. gonorrhoeae, where it can go up to more than 6 mM. In the bloodstream and cerebrospinal fluid, glucose serves as the major carbon source, although the lactate concentration can be higher than 1 mM [36]. Lactate and pyruvate predominate as carbon sources within phagocytes [45]. Our results showed that both lactate isoforms were able to induce microcolony dispersal at 0.4 mM. This is within the range of physiological concentrations encountered by pathogenic Neisseria in vivo [35, 36]. As pathogenic Neisseria metabolize and grow faster in the presence of lactate, receiving nutritional signals from the environment during favorable growth conditions and detaching from microcolonies could be a survival strategy.
Our results showed that lactate induces microcolony dispersal independent of the presence of the bacterial LDH that oxidizes it. These results show that the observed effect is not due to the previously characterized burst in respiration and growth after lactate addition in the presence of glucose [32]. This is also supported by the observation that pyruvate did not induce dispersal but promoted growth in a similar manner to lactate. These results indicate that the previously described mechanism of neisserial microcolony dispersal triggered by oxygen depletion and subsequent membrane depolarization is not the mechanism in this case [15]. To test this, we measured the concentration of ATP and the ratio of NAD+/NADH during dispersal. We did not observe any decrease in either the ATP concentration or the NAD+/NADH ratio in the dispersed bacteria, suggesting that no dramatic decrease in the PMF occurred during the process.
In the absence of lactate the dispersal phase lasted for several hours. However, when the gene encoding for the lactate permease (lctP), required for lactate uptake, was deleted we observed accelerated dispersal even in absence of lactate. This result suggests a role for LctP as a negative regulator of microcolony dispersal since deletion of the lctP resulted in microcolony dispersal. It is worthwhile to mention that lactate permease has been shown to function in a PMF-dependent manner in bacteria [46, 47]. Disruption of the PMF may therefore affect microcolony dispersal, as shown by Dewenter et al [15], indirectly by a change in lactate permease activity. However, there is also a possibility that the absence of LctP can affect lactate export and lead to accumulation of intracellular lactate produced by the bacterium and cause dispersal.
Deletion of lctP in N. meningitidis has previously been shown to attenuate nasopharyngeal colonization despite for increased initial adherence to host cells. It has also been shown to affect the ability to cause bloodstream infection and cause an increase in sensitivity to complement-mediated killing [31, 35]. The incompetence to acquire and utilize nutrients, like lactate, important for growth plays a crucial role in colonization. Since microcolony dispersal seems to be a highly regulated in pathogenic Neisseria, deregulation of the process like we have observed with the ΔlctP mutant, could also affect the ability of the bacteria to colonize in efficient way.
Although lactate is able to induce microcolony dispersal, we cannot exclude the possibility that other host derived factors present in the CM and in vivo might have the same stimulatory effect as lactate. However, collection of CM from host cells grown in absence of glucose, and thus inhibited in their ability to produce and secrete lactate, significantly reduced the stimulatory activity.
Earlier studies have shown that meningococci bound to epithelial cells modulate their gene expression, enabling them to form intimate contacts with the cells [14, 27–29, 48]. However, we did not detect any upregulation of the transcription regulator misR or the MisR/S-regulated genes pptB, pilC1 and crgA after inducing dispersal with CM. One possible explanation is that meningococci respond differently when interacting with cells than when they are treated with CM. However, our results from the infection assays using different cell confluences and the adhesion-deficient ΔpilC1 mutant showed that direct contact with cells is not essential for the timing of the dispersal phase or for its short time span. Additionally, our qPCR analysis included genes important for meningococcal piliation and pilus aggregative properties (pilC1, pilC2, pilE, pilT, pilX, pilV and pilW) [4, 7, 9, 10]. We did observe significant changes in their mRNA levels upon induction with CM. However, when we analyzed the protein level of PilE, PilT, PilC, PilX and PilW we did not detect any changes.
Post-translational modifications of the PilE subunit can also modulate Tfp function. Tfp of Neisseria can be further grouped into class I and class II pili. In majority of our study we used the meningococcal C strain FAM20 that is known to express class II pili. Recent findings by Gault et al indicate that class II pili can contain up to 5 glycosylation sites and deletions of glycosylation genes in these strains affect adhesion, aggregation and successful pilus assembly. The multiple glycosylation sites are suggested as a way to avoid immune recognition as class II pilus expressing strains lack the ability to vary the antigenicity of the pilus fiber [30]. By using two additional strains, meningococcal W-135 strain JB515 and N. gonorrhoeae MS11, we observed that lactate induced dispersion occurs in both class I and class II pilus expressing strains. Although no upregulation was observed in pptB expression and no changes were observed in the expression of genes involved in glycosylation modifications we cannot exclude that post-translational modifications play a role in response to CM and lactate stimulation. Moreover, PilE can also be modulated by phosphocholine and phosphoethanolamine [49]. The absence of these modifications has been shown to affect pilus bundling but no changes in functionality have been observed [50].
In conclusion, the work presented here demonstrates that lactate, secreted from host cells, can stimulate the dispersal of microcolonies in pathogenic Neisseria. The metabolic utilization of lactate by N. meningitidis was not required for induction of rapid dispersal. In the future, it would be interesting to identify adhesive and general properties of dispersed bacteria upon interaction with lactate. This study provides a basis for future research to further investigate the role of lactate as a signaling molecule influencing disease progression in pathogenic Neisseria.
The Neisseria meningitidis FAM20 serogroup C strain and mutant deficient in PilC1 have been described previously [24, 51]. The Neisseria meningitidis JB515 serogroup W strain and Neisseria gonorrhoeae MS11 strain have been described previously [24, 52]. The strains were grown on GC agar (Acumedia) supplemented with 1% Kelloggs’ for 18 h at 37°C in a 5% CO2 environment. Antibiotics for selection of FAM20 mutant strains were used in following concentrations: tetracycline 1 μg/ml, kanamycin 50 μg/ml and spectinomycin 40 μg/ml.
The human epithelial cell lines FaDu (ATCC HTB-43), A549 (ATCC CCL-185), and Detroit 562 (ATCC CCL-138) and the human endometrial epithelial cell line Hec-1B (ATCC HTB-113) were maintained in Dulbecco’s modified Eagle’s medium containing GlutaMAX and pyruvate (DMEM; Thermofisher) and supplemented with 10% heat inactivated fetal bovine serum (FBS; Sigma-Aldrich). For live-cell imaging, the cells were grown to 100% confluence unless stated otherwise. Prior to the experiments, the cells were washed and the medium exchanged to fresh DMEM without FBS. For fixation, the FaDu cells were treated with 3.7% paraformaldehyde (Sigma-Aldrich) for 10 min and then washed extensively.
For live-cell imaging, the cells were seeded in 24-well poly-D-lysine-coated glass-bottom plates (MatTek) precoated with collagen type 1 from calf skin (Sigma-Aldrich). Wells were coated with 0.5 ml of a 0.01% collagen solution for 1 h at room temperature and then washed with PBS and water. The plates were dried under laminar airflow for 30–60 min and stored at 4°C until they were seeded with cells.
N. meningitidis FAM20 and its isogenic mutants were resuspended (2 × 106 CFU/ml) in prewarmed medium, filtered through a 5-μm pore filter to break apart preexisting bacterial aggregates, and used to infect cells at confluences of 20%, 50%, 80% and 100% (MOI of 10 for 100% confluent cell layer) and 0% (absence of cells) in 24-well collagen coated poly-D-lysine-coated glass-bottom plates (MatTek). Infection assays were performed in DMEM without FBS, and the bacteria were gently centrifuged onto the cell surface (200 × g; 5 min). When observing aggregation in liquid during induction assays, bacteria were resuspended in prewarmed DMEM (107 CFU/ml) containing 1% FBS, filtered through a 5-μm pore filter and added to 24-well glass-bottom plates (MatTek). While FBS was included in the medium for the assays without cells to facilitate the formation of microcolonies, it was excluded from the assays with cells to facilitate the later purification of soluble compounds triggering microcolony dispersal.
The bacteria were observed under a live-cell microscope (Axiovert Z1, Zeiss) at 37°C in a 5% CO2 environment. Three images per well were acquired every 5–10 min for 8 h using a 40× objective. The aggregation phase was defined as the time from the start of incubation to the start of microcolony dispersal. The dispersal phase was defined as the time from when the bacteria in the microcolonies began to spread out until microcolonies were no longer present. The planktonic phase was defined as the time when all bacteria had dispersed from the microcolonies and were present as single cells.
N. meningitidis FAM20 and ΔpilC1 were resuspended (2 × 106 CFU/ml) in prewarmed medium, filtered through a 5-μm pore filter to break apart preexisting bacterial aggregates, and used to infect FaDu cells at 100% confluence at MOI of 10 in 24-well plates. The bacteria were gently centrifuged onto the cell surface (200 × g; 5 min) and incubated for 4 h at 37°C in a 5% CO2 environment. After incubation, unbound bacteria were removed by washing 3 times. The adherence was estimated from the viable counts by lysing the FaDu cells with 1% saponin in DMEM for 10 min and plating serial dilutions on GC agar.
Cells were grown in either 75 cm2 or 25 cm2 cell culture flasks to 90–100% confluence, washed once with PBS, and, unless stated otherwise, incubated in fresh DMEM (5 ml per 25 cm2) for 5 h to obtain conditioned medium (CM). CM was collected from uninfected cells unless stated otherwise. For the CM from infected cells, FaDu cells were infected with FAM20 at an MOI of 10 as described above for 5 h. The CM obtained were sterile-filtered and used in the induction assays. For the analysis of the active compound(s) in the CM from uninfected cells, the following treatments were applied: (a) heat inactivation by incubation at 95°C for 5 min, followed by sterile filtration to remove precipitates; (b) protease inhibitor cocktail (1:100 final dilution; P1860, Sigma-Aldrich), added to the CM 10 min before induction; (c) proteinase K (100 μg/ml; P2308, Sigma-Aldrich) added to CM, incubated overnight at 56°C and inactivated at 95°C for 20 min before sterile filtration to remove precipitates; and (d) trypsin (200 μg/ml; T1426, Sigma-Aldrich) in CM for 1 h at 37°C or chymotrypsin (200 μg/ml; C3142, Sigma-Aldrich) in CM for 1 h at 30°C, followed by inactivation with the protease inhibitor cocktail (1:100) for 10 min before induction. For the DNA and RNA digestions, final concentrations of 10 U/ml DNase I (D5025, Sigma-Aldrich) or 10 μg/ml RNase A (R4875, Sigma-Aldrich) were incubated with the CM for 2 h at 37°C. EDTA (E5134, Sigma-Aldrich) and EGTA (E4378, Sigma-Aldrich) were used as metal ion chelators at final concentrations of 2 mM. As negative controls, the same treatments were performed on DMEM and used in the induction assays. The CM was also filtered through a 3 kDa cut-off Amicon Ultra centrifugal filter device (Millipore).
N. meningitidis strain FAM20 and its isogenic mutants, N. meningitidis strain JB515 or N. gonorrhoeae strain MS11 were resuspended (107 CFU/ml) in prewarmed medium containing 1% FBS and filtered through a 5-μm pore filter to break apart preexisting bacterial aggregates. The bacteria were grown in 24-well glass-bottom plates (MatTek) in either 0.5 ml (wild-type compared to isogenic mutants) or 1 ml and allowed to form microcolonies for 3 h in the live-cell microscope at 37°C in a 5% CO2 environment. Prewarmed DMEM, CM, DMEM supplemented with pyruvate (final concentration 10 mM, sodium pyruvate, 11360070, Thermo Fisher scientific) or lactate (sodium D-lactate; L7022, sodium D-lactate; 71716, Sigma Aldrich) at the indicated concentrations were then gently added to the microcolonies at a 1:1 volume ratio. The 3 h time point was chosen because the microcolonies had formed, but at least 1 h remained until they would spontaneously begin to disperse. Three images per well were acquired every 5–10 min for 8 h using a 20x or 40× objective.
Induction assays with CM from non-infected cells and control DMEM were performed on FAM20 wild-type as indicated above in 12-well glass-bottom plates in 2 ml. Samples were collected 10 min after addition of CM or DMEM and resuspended in twice the volume of RNAprotect Bacteria Reagent (Qiagen). The sample:RNAprotect mixtures were vortexed at high speed (5 s), incubated for 5 min at room temperature, pelleted by centrifugation for 15 min at 4000 × g and purified using the RNeasy plus mini kit (Qiagen) according to the manufacturer’s protocol. The RNA yield and purity were assessed using a NanoDrop 8000, and reverse transcription was performed with random hexamers using a SuperScript VILO Master Mix (Thermofisher). The resulting cDNA was amplified using LightCycler 480 SYBR Green I Master mix (Roche) in a LightCycler 480 Real-Time PCR System. The PCR program was adapted from the manufacturer using an annealing temperature of 55°C or 60°C. The 30S ribosomal protein rpsJ was used as a reference gene. Table 1 lists the primers used in the assay. Primers were used at a final concentration of 500 nM except the primer pairs for pilT and pilC2, which were diluted to 250 nM. The analysis was performed with the LightCycler 480 Real-Time PCR System software using the comparative cycle threshold method. The target mRNA levels in the samples were normalized to the reference gene and then compared to the value of the non-induced DMEM control sample. Primer-pair specificity was controlled for by analyzing the melting curves.
Induction assays with CM from noninfected cells were performed on FAM20 wild-type as indicated above in 12-well in glass-bottom plates (MatTek) using DMEM as a negative control. Samples were collected after 10 min, centrifugated and resuspended in 1x sample buffer containing β-mercaptoethanol. The samples were heated for 5 min at 95°C and separated on a 4–15% gradient SDS-PAGE gels (Bio-Rad). After a transfer to Immobilon-P membrane (Millipore) rabbit polyclonal PilE (1:5000), PilT (1:10000), PilC (1:1000), PilX (1:1250) and PilW (1:2000) antibodies were used to detect protein expression levels [7, 9, 51, 53]. After detection the membrane was stripped and monoclonal mouse antibody (Hycult Biotech) diluted 1:2000 was used to detect EF-Tu elongation factor as a loading control. Infrared (IR)- reactive dye conjugated goat anti-mouse or goat anti-rabbit antibodies (Li-Cor) were used as a secondary antibodies. Membranes were examined in Odyssey IR scanner at standardized 700 and 800nm. Image J analysis software (version 1.48) was used for quantification of band intensity.
Fractionation of CM and D-lactate (50 mM) was performed using a preparative liquid chromatography (LC) method. A Varian Prostar 230 HPLC pump (Walnut Creek, USA) delivered the binary mobile phase (MP) at the flow rate of 4.0 mL/min according to the programmed gradient (MP A: water and MP B: acetonitrile) as follows: 0.0 min, 2% B; 2.0 min, 2% B; 10.0 min, 98% B and 12.0 min, 2% B. The LC system was equipped with an Atlantis T3 Prep column (250×10 mm, 5 μm particle size, Waters, Ireland), and the T-connection split after the column, which divided the efflux 1:9, directing it to YL9181 ELSD superior sensitivity detector (Young Lin Instruments Co. Ltd., Korea) and Waters fraction collector I (Advantec, Japan) accordingly. The injection volume was 100 μl. The fraction collector was set to collect fractions every three minutes for a total of 12 minutes. Fractions were dried in a Speed-Vac and then resuspended in 100 μl water. The activity of fractions was examined in an induction assay as described above. The assay was performed in a 96-well MatTek plate (100 μl:100 μl).
Cells were grown for 5 h in DMEM in the presence or absence of glucose. CM was collected as previously described. The concentrations of lactate in the CM were quantified using a lactate assay kit (MAK064-1KT, Sigma-Aldrich) according to the manufacturer´s instructions.
Mutations in the genes involved in lactate metabolism were created using fusion PCR. All PCR reactions were performed using high-fidelity Phusion DNA polymerase (Thermo Scientific). Primers used for generation of mutants are listed in Table 1.
First, up- and downstream regions of ldhA (ldhA_up_fw:ldhA_up_rev and ldhA_dn_fw:ldhA_dn_rev), ldhD (ldhD_up_fw: ldhD_up_rev and ldhD_dn_fw: ldhD_dn_rev), lldA (lldA_up_fw:lldA_up_rev and lldA_dn_fw:lldA_dn_rev) and lctP (lctP_up_fw:lctP_up_rev and lctP_dn_fw:lctP_dn_rev) were amplified from FAM20 chromosomal DNA using the indicated primer pairs. For the ldhA and lctP mutants, a tetracycline resistance gene (tetA) was amplified from plasmid pACYC184 with primer pairs ldhA_tet_fw:ldhA_tet_rev and lctP_tet_fw:lctP_tet_rev, respectively. For the ldhD mutant, a kanamycin cassette was amplified from the pDONR KmR plasmid [55] with primers ldhD_kan_fw and ldhD_kan_rev. For the lldA mutant, a spectinomycin cassette [14] was amplified using lldA_spc_fw and lldA_spc_rev. With all acquired PCR products containing overlapping sequences, a fusion PCR reaction was performed in 2 steps. First, a fusion reaction was performed without primers to fuse the upstream and downstream fragments with the resistance cassette. The resulting fusion products were then amplified with the upstream forward primer and the downstream reverse primer for the respective constructs. The constructs were integrated into the genome of N. meningitidis FAM20 using homologous allelic replacement following spot transformation with purified PCR fragments. The correct location and sequence of the strains were confirmed by sequencing (MWG Eurofins).
Wild-type FAM20, ΔlctP, ΔldhA, ΔlldA and ΔldhD mutants were resuspended in DMEM (A1443001, Thermofischer) supplemented with glutamax and either 25 mM glucose, 10 mM L- or D- lactate at an OD600 of 0.1 and grown under agitation at 37°C in a 5% CO2 environment. The OD600 was measured every hour for 10 h.
To determine if lactate exerts bactericidal effect, induction assays were performed as previously described with initial concentration of 107 CFU/ml. The bacteria were grown for 3 h under static conditions. After the 3 h incubation DMEM or DMEM containing L-lactate at a final concentration of 50 mM was added to the suspension at a 1:1 volume ratio and bacterial viability was determined at the end of 8 h.
Samples were prepared as described above for induction assays. After 3 hours of incubation, prewarmed DMEM, CM or DMEM supplemented with 10 mM L-lactate was added. Samples were taken 10, 20 and 40 minutes after addition and frozen at -80°C. The OD600 of the samples was also recorded. The total ATP concentration was measured using a BacTiter-Glo Microbial Cell Viability Assay (Promega) according to the manufacturer’s instructions. Carbonyl cyanide m-chlorophenyl hydrazone (CCCP, Sigma) was dissolved in DMEM and used as a positive control at a final concentration of 25 μM. The NAD+/NADH ratio was determined with the NAD/NADH-Glo Assay (Promega) according to the manufacturer’s instructions.
N. meningitidis FAM20 was resuspended (107 CFU/ml) in prewarmed medium containing 1% FBS and filtered through a 5-μm pore filter to break apart preexisting bacterial aggregates. The bacteria were incubated in 24-well glass-bottom plates (MatTek) in 1 ml at 37°C in a 95% air/5% CO2. At 1.5 h, the oxygen concentration was set to 0%. Oxygen depletion was conducted by a 95% N2/5% CO2 flow. At 5 h, the oxygen concentration was set to 95% air/5% CO2. The bacteria were observed under a live-cell microscope (Axiovert Z1, Zeiss). Three images per well were acquired every 10 min for 8 h using a 40× objective.
Differences between two groups were analyzed using two-tailed and unpaired Student t-tests. Differences between multiple groups were analyzed with an analysis of variance (ANOVA) followed by a Bonferroni post-test. For time-lapse microscopy, statistical analysis was used to analyze the differences between the lengths of the dispersal phases. P-values below 0.05 were considered statistically significant. The statistical analysis was performed using Microsoft Excel (2011) or GraphPad Prism software, version 5.
|
10.1371/journal.ppat.1000026 | Pneumococcal Pili Are Composed of Protofilaments Exposing Adhesive Clusters of Rrg A | Pili have been identified on the cell surface of Streptococcus pneumoniae, a major cause of morbidity and mortality worldwide. In contrast to Gram-negative bacteria, little is known about the structure of native pili in Gram-positive species and their role in pathogenicity. Triple immunoelectron microscopy of the elongated structure showed that purified pili contained RrgB as the major compound, followed by clustered RrgA and individual RrgC molecules on the pilus surface. The arrangement of gold particles displayed a uniform distribution of anti-RrgB antibodies along the whole pilus, forming a backbone structure. Antibodies against RrgA were found along the filament as particulate aggregates of 2–3 units, often co-localised with single RrgC subunits. Structural analysis using cryo electron microscopy and data obtained from freeze drying/metal shadowing technique showed that pili are oligomeric appendages formed by at least two protofilaments arranged in a coiled-coil, compact superstructure of various diameters. Using extracellular matrix proteins in an enzyme-linked immunosorbent assay, ancillary RrgA was identified as the major adhesin of the pilus. Combining the structural and functional data, a model emerges where the pilus RrgB backbone serves as a carrier for surface located adhesive clusters of RrgA that facilitates the interaction with the host.
| Streptococcus pneumoniae (pneumococcus) is one of the most important human pathogens and a major cause of morbidity and mortality worldwide, causing respiratory tract infections, community acquired pneumonia, and invasive diseases. Although the pneumococcus is a well-studied bacterial pathogen, first described in the late 19th century, pili on its surface were discovered only recently. Pili are elongated structures extruding from the bacterial surface and were found to be important virulence factors of both Gram-positive and Gram-negative bacteria. Bacterial pili are considered to participate in bacterial adhesion to a host, a crucial step in bacterial infection. In contrast to Gram-negative pili, little is known about the structure of native Gram-positive pili. We used native purified pili of pathogenic pneumococcus TIGR4 to study its structural composition, mainly by the use of cryo EM techniques. Pili were found to be composed of protofilaments that are arranged in a coiled-coil, compact superstructure of various diameters. Adhesive properties of pilus surface located ancillary protein RrgA to selected compounds of the extracellular matrix might be part of the pilus mediated host–pathogen interplay. Analysis of native pneumococcal pili revealed structural basics of a Gram-positive pilus that could also serve as a basis for effective vaccine design.
| The Gram-positive bacterium Streptococcus pneumoniae, also known as pneumococcus, is one of the most important human pathogens causing respiratory tract infections such as sinusitis, otitis media, and community acquired pneumonia, but also invasive diseases such as septicemia and meningitis. Together with HIV, malaria, and tuberculosis the pneumococcus represents one of the four major infectious disease killers [1]–[4]. Even though pneumococcus is a devastating pathogen, it is also a member of the human commensal flora and is known to asymptomatically colonize the nasopharynx [1]. A major virulence factor of Streptococcus pneumoniae is the polysaccharide capsule, by which pneumococci are grouped into at least ninety different serotypes [5]. Other genetic factors, such as CbpA (choline-binding protein A) and pneumolysin, have been described to be of importance for virulence [6]–[8]. Infection by Streptococcus pneumoniae leads to invasive disease triggered by initial colonization of the nasopharynx, but the mechanisms of adhesion are not well understood [9]. Recently, pilus harboring pneumococci were discovered and results obtained indicate a key role for these structures in virulence and disease [10],[11]. Furthermore, in a mouse model of intraperitoneal infection Gianfaldoni et al. [12] reported protective immune responses after active and passive immunization with recombinant pilus subunits of Streptococcus pneumoniae Type 4 strain TIGR4 (T4). Previously, similar pili-like surface structures had been identified in other Gram-positive bacteria, such as Corynebacterium diphtheriae [13],[14], Actinomyces spp. [15], group A streptococci (GAS) [16], group B streptococci (GBS) [17] and recently Mycobacterium tuberculosis [18] where they were shown to play an important role in the interaction with the host at different stages of infection.
The Streptococcus pneumoniae pilus was found to be encoded by the rlrA pathogenicity islet [10],[19], initially discovered in T4, a clinical, serotype 4 strain, of which the genome is known [20]. Sequencing of various pneumococcal strains revealed, that not all isolates contain this genetic element [21],[22]. The rlrA operon encodes, besides a Rof-A-like transcriptional regulator (RlrA), 3 sortases (SrtB, SrtC and SrtD) and 3 structural proteins RrgA (Swiss-Prot Q97SC3), RrgB (Swiss-Prot Q97SC2) and RrgC (Swiss-Prot Q97SC1) containing a LPxTG motif (or variants thereof) [10],[19],[23].
In contrast to Gram-negative pili, which are composed of non-covalently linked subunits, Gram-positive pili studied so far are thought to be extended polymers formed by a transpeptidase reaction involving covalent cross-linked subunit proteins containing specific amino acid motifs, which are assembled by specific sortases. Sortases are also responsible for the covalent attachment of the pilus to the peptidoglycan cell wall [24]. Fundamental work on this was done by Schneewind and co-workers studying Corynebacterium spp. pili [13],[14],[25],[26] and recent reviews summarize the more general knowledge on Gram-positive pili [27]–[29].
In Corynebacterium diphtheriae, in addition to a N-terminal signal sequence and a C-terminal cell wall sorting signal, two motifs are considered to be important for the major pilus component, i.e. the so called pilus backbone forming protein: the pilin motif and the so called E-box [13]. Following the corynebacterial system, pneumococcal RrgB was proposed to form the backbone of the pneumococcus T4 pilus structure, as its sequence contains homologues of both motifs. For pneumococcal T4 RrgA and RrgC, a role as ancillary proteins was suggested [10]. These observations are supported by initial electron microscopy (EM) analysis on pneumococcal cells containing pili [10],[11]. Although the precise mechanism of incorporation of RrgA, RrgB and RrgC into the pneumococcal pilus is not yet understood, one hypothesis is that incorporation of the three subunits is specifically catalysed by each of the 3 sortases present in the rlrA islet: in line with this are results found by LeMieux et al. [11] that showed that SrtD is needed for RrgA incorporation into the typical high molecular weight (HMW) structure. In addition, the incorporation of RrgA is dependent on the presence of RrgB but not RrgC.
Whereas genetically based functional studies regarding Gram-positive pili such as those of Streptococcus pneumoniae are emerging, structural information of the native entire pilus in Gram-positives is lacking and its significance in infectious disease is not clear. Very recent data based on crystal structures of single pilus subunits of Gram-positive pili in Streptococcus agalactiae and Streptococcus pyogenes stimulated novel insights into Gram-positive pilus composition [30],[31]. The elucidation of the structure of the native pilus is of great interest not only to increase our understanding of the biology of Gram-positive bacteria, but also as potential tool to develop proper therapeutics and vaccines against pathogenic bacteria like Streptococcus pneumoniae [32],[33]. Our approach consists in using native, purified pneumococcal pili of a pathogenic T4 strain to study structure and properties of these Gram-positive surface appendages. We provide for the first time structural evidence of the pneumococcal pilus, which is composed of protofilaments arranged in a coiled-coil superstructure. Structural proteins RrgA, RrgB and RrgC localized to different regions of the same pilus, confirming RrgB as the major compound, followed by clustered RrgA and single RrgC molecules on the pilus surface. RrgA was identified as major adhesion protein towards selected extracellular matrix (ECM) compounds. Structural and functional data indicate the pneumococcal pilus to be a flexible carrier of functional groups able to cross pneumococcal polysaccharide capsule, promoting host cell interaction.
In order to study the pneumococcal pilus in detail, a multi step purification procedure was set up to obtain pure native pili preparations. Pneumococcal pili were isolated from strain T4, bacteria that in low-dose EM showed individual pili and bundles of individual pili on the bacterial cell surface (Figure 1). Both types of appendages were distributed on the bacterial surface, the majority of which (∼65%) belonging to the individual pilus type. The same appearance was found analysing purified pili preparations with cryo electron microscopy (cryo-EM).
Briefly, for pili purification, bacteria were grown on blood agar plates. Harvested bacteria were washed and subjected to mutanolysin treatment. N-acetyl muramidase treatment released pneumococcal pili into the supernatant. Subsequently, supernatants containing pili were applied to a sucrose gradient to separate them from other cellular impurities and to concentrate their relative amount in the sample preparation (Figure 2A). Pili-positive gradient fractions were identified by dot blot analysis with anti-RrgB antibodies. Western analysis (anti-RrgB) of SDS-PAGE separated pili samples was performed to identify HMW material in fraction number 5–8 at the top of the separation gel. Dialysed sucrose pools were concentrated and applied to gel filtration to further purify the pili preparations. As shown in Figure 2B, size exclusion chromatography allowed the separation of HMW pili (peak A) from lower molecular weight material (peak B and C) as proven by SDS-PAGE and western blot analysis with anti-RrgB antibodies (data not shown). As a control, the same purification procedure was performed with a pneumococcus T4 Δpil strain. As expected, HMW pili (peak A) that were eluted in the void volume were not found in the respective delta pilus preparation. A summary of the purification strategy of pneumococcal T4 pili is shown in Figure 2C: Silver stained SDS-PAGE analysis of the different purification steps and the correspondent western blot anti-RrgB analysis show purified HMW pili after gel filtration at the entry of the gel pocket. Purified pili were used for further analysis and to study the structure of pneumococcal pili.
HMW pili, following size exclusion chromatography, were applied to SDS-PAGE and HMW band was subjected to mass spectrometry analysis and N-terminal sequencing (Edman analysis). Tryptic peptide sequence of HMW pili were analysed by MALDI-TOF. Results identifying the cell-wall surface anchor family protein RrgB of Streptococcus pneumoniae T4 (gi|15900379) were confirmed by MS/MS analysis: the fragmentation of a peptide with a mass of 2064 Da matched the peptide sequence LAGAEFVIANADNAGQYLAR that is part of pneumococcal T4 RrgB. The Edman analysis resulted in the peptide sequence AGTTTTSVTVHXL, which could be identified as part of T4 protein RrgB. The identified N-terminal starting amino acid corresponds to an alanine, which is located 30 residues downstream of the Met of the RrgB sequence, in agreement with the predicted cleavage site of the signal sequence.
The detailed composition of pili was investigated by IEM with antibodies raised against recombinant HisTag-RrgA, -RrgB and -RrgC. Initially, single and double IEM were performed with different combinations of the three antisera on both bacteria and isolated pili in order to reveal the presence of all three pilus components. Triple immunogold staining was then performed on the same pilus preparation and on whole bacteria to observe the type of distribution and the relative amount of the 3 structural proteins. Figure 3A shows RrgB distributed evenly along the entire pilus polymer while RrgA and RrgC were present at non-regular intervals along the pilus shaft. An approximate estimation of the relative amounts of the three proteins based on triple, double and single IEM observations indicated that roughly 90% of the gold particles corresponded to RrgB. The remaining 10% circa were composed of RrgA and RrgC, with a higher occurrence of RrgA in comparison to RrgC. In particular, the IEMs showed that RrgA was organized in small clusters, as found by particulate aggregates of 2–3 anti-RrgA antibody units, distributed along the entire pilus surface. Interestingly, RrgC protein was found in single copies and often co-localized with the RrgA clusters. Triple IEM performed on whole bacteria (Figure 3B) confirmed that purified pili conserved the same structural characteristics as native pili attached to bacteria.
Purified HMW pili preparations, observed by cryo-EM (Figure 4) and freeze drying/metal shadowing techniques (Figure 5A, 5B and 5C) showed that pili were elongated structures of up to 1 µm in length. They were identified as elongated and adhesive structures with the tendency to form a net on the EM grids (Figure 4). Generally present as individual single pili of different thickness, they were also found to form bundles of individual pili (data not shown). Among the individual single pili several classes could be identified based on their diameter and morphological variability (Figure 5). The most subtle filaments (Figure 5A and 5D) showed a linear morphology with no evident periodicity. A corresponding IEM displayed a linear distribution of gold particles binding to RrgB backbone proteins (Figure 5D). We defined this type of filament as the pilus protofilament. Increasing filament widths (Figure 5B and 5C) resulted in an increasing complexity of the filaments, as clearly indicated by the higher number of gold particles decorating the filaments in a non-linear spacing (Figure 5E and 5F). The majority of the pili could be assigned to the class of thin pili (∼63%) (Figure 5B and 5E) with an observed average diameter of 9.5 nm, as calculated from cryo-EM data (Figure 4). The remaining ∼37% of the isolated pili were of larger diameters, the majority of which, individualised as class of thick pili, showed a width of about 10.5 nm (data not shown). Further structural analyses were performed on the thin pili, more than 200 individual thin pilus segments were selected [34] from digitized micrographs (Figure 4) obtained by cryo-EM on vitrified samples by using 300×300 pixel size boxes. Thus working with shorter segments that resulted to be approximately straight for the chosen box size was possible. Pili segments were treated as discrete single particles, 300 pixel in length and processed [35] by first aligning them rotationally and translationally to a reference cylinder centered into the image box. All segments that did not align with the reference were eliminated. Finally 124 segments of the thin pili (Figure 6A) were kept and used to generate averaged thin pili segments with an increased signal-to-noise ratio (Figure 6B). Subsequently the diameter of the averaged thin pilus segments could be calculated from its density profile, by creating the profile (IMAGIC5) [35] of the averaged segment generated from the 2D image of the pilus (red line). The results showed a diameter of 9.6±0.3 nm for the thin pilus. Moreover the shape and the values of the density profile clearly indicated that thin pili were rather compact structures.
Interestingly when a double Gaussian filter, where both, high and low frequency transmission were cut off, was applied on original thin pili data (Figure 7A), the filtered 2D image (Figure 7B), clearly showed that thin individual pili were composed of at least two protofilaments arranged in a coiled-coil superstructure. The Gaussian filtered image of the thin pilus showed zones where the 3.5 nm diameter protofilaments were tightly intersected (crossovers), resulting in a pilus diameter of 6.8 nm. This was alternated by zones where protofilaments had a more relaxed intersection with a pilus diameter of 9.5 nm. The average distance between two neighbouring crossovers was approximately 13 nm. Preliminary results suggest that thick pili are also composed of similar protofilaments arranged in coiled-coil manner (data not shown).
In order to investigate adhesive properties of isolated pili and of single pilus subunits RrgA, RrgB and RrgC in vitro binding assays were performed to study the interaction to selected ECM components using this approach to provide a proof in principle [36],[37]. In particular, fibrinogen, fibronectin, laminin, lactoferrin and collagen I were selected, as these cellular compounds are known to be recruited by pathogenic agents [9], [38]–[40]. Apart from overall pili adhesive properties special interest was drawn to a potential role of distinctly surface exposed RrgA and RrgC. For this purpose, serial diluted samples of recombinant proteins RrgA, RrgB and RrgC, as well as native purified pili and a pilus negative control were added to 96-well plates coated with the selected ECM components. Binding was detected using polyclonal sera raised against the single recombinant pilus subunits and quantified by enzyme-linked immunosorbent assay (ELISA). As demonstrated in Figure 8 (lane A), RrgA showed very pronounced dose depending binding to most of the tested ECM compounds, whereas binding results obtained for RrgC and RrgB are negligible. In addition, RrgA binding was observed for lactoferrin and fibrinogen whereas no binding was detected to vitronectin coated plates (data not shown). Bovine serum albumin (BSA) was used as negative control in all the assays. Binding studies performed using purified pili (Figure 8, lane B) showed binding to ECM components clearly distinguishable from the Streptococcus pneumoniae delta pilus negative control.
Pili are considered important key players in bacterial pathogenesis and disease [10],[27]. To date structural information of the native pilus in Gram-positive bacteria is lacking, therefore the elucidation of their structure and function are of great interest. Our approach consisted in obtaining native purified pili from a pathogenic strain of Streptococcus pneumoniae to study pilus structure and function. Special emphasis was drawn on the overall structural principle of the native pilus and the role of the individual structural proteins RrgA, RrgB and RrgC. As opposed to pili attached to the bacterial surface, isolated pili allow a broader spectrum of analyses and at the same time permit a comprehensive characterization of their structure in sufficient detail to describe the function at the quasi-molecular level. We developed a multi step purification procedure to obtain native pilus material that allowed to perform the desired analyses.
T4 bacteria were examined by low dose negative stain EM, IEM and cryo-EM, showing that the bacterial surface is covered with elongated, flexible and rather sticky pilus-like appendages of up to 1 µm long. Interestingly, we observed pili of various morphologies: individual single pili, distinguishable into different classes by their diameter (ranging from 9.5 nm up to 10.5 nm), and bundles of individual pili. Whether or not this has a physiological role has yet to be evaluated. The established purification method allowed for the isolation of pure HMW material that showed pili morphotypes having the same features as those found for wild-type pili expressed on whole bacteria. Structural analysis based on cryo-EM data of vitrified, purified single pili revealed that they are organised in coiled-coil superstructures made by at least two protofilaments. The observed range in pilus diameters could either reflect a difference in the degree of packaging of the identical protofilaments into the pili superstructures or a higher number of protofilaments composing the larger pili. The protofilaments of the thin pilus type are organized to form a rather compact superstructure. However no distinct internal cavity could be identified within the thin pilus structures. Preliminary results on the thick type of pilus suggest also a protofilament based structure.
The picture of the individual pili that emerges from our analysis indicates that the Streptococcus pneumoniae pilus does not exist in a single structural state but rather in several structural states that are underlaying, among other things, the flexibility and elasticity of these polymers while keeping the same protein composition and proteins roles: RrgB forming the backbone, surface located clustered RrgA being the major ancillary protein involved in adhesion and RrgC as minor ancillary protein of still unknown role.
Additional biochemical analysis of isolated pili supports RrgB as the main pilus building block: Mass analysis of native pili revealed clear signals only for peptides that could be assigned to structural protein RrgB. Neither RrgA nor RrgC related signals could be identified, which is probably due to their minor abundance and the overall hindered protease digestibility of the isolated HMW pili. Similarly, the determination of the N-terminal amino acid sequence of the purified pili by Edman analysis, matched only with the sequence following the predicted signal sequence of RrgB. The observation that purified pili show a free N-terminal part of RrgB starting exactly after the signal sequence may reflect properties of pilus biosynthesis and subsequently its structure. Purified pili of a Streptococcus pneumoniae delta RrgA background show a similar overall pili structure composed of protofilaments. This is in accordance with studies showing that a pneumococcal RrgA mutant strain is still able to form pili, whereas a ΔrrgB ΔrrgC strain is not [11] and fits with the detected structural organisation of RrgA clusters on a coiled-coil RrgB based scaffold.
Gram-positive and Gram-negative pili differ substantially in their assembly mechanisms (Gram-negative pili: non-covalently linked protein subunits versus Gram-positive pili: covalently linked subunits), interestingly both types of bacterial pili share a common arrangement, the coiled-coil superstructure. Our work supports that, also for Gram-positive bacteria, adhesive pili extending from the bacterial surface are the most appropriate structures to promote biological function like adherence to the host due to their structural arrangement leading to flexibility and elasticity. Until now this could be only observed in Gram-negative bacteria like Haemophilus influenzae type b pili and Escherichia coli P-pili [41] or Actinobacillus actinomycetemcomitans [42]. Results by Kang et al. [30] identified a novel principle of stabilization of long and thin pilus filaments by isopeptide linkage between pilus subunits of Gram-positive Streptococcus pyogenes. Further work will have to show whether similar design can also be found in other Gram-positive pili, like those of Streptococcus pneumoniae. Our results suggest that the coiled-coil arrangement of the protofilaments, forming the pneumococcal pili, might be an additional principle, other than isopeptide bond formation, to confer stability and flexibility to subtle surface structures in order to withstand mechanical rigors outside the cell.
Research on bacteria and therefore also the study of the pneumococcal pilus should be seen in the context of bacterial life cycles within specific ecological niches and e.g. in the interaction with its host. Pneumococcal infection of the host occurs mainly via the mucosal route [1], thus bacteria need to develop strategies to adhere and resist actions of the human immune system like mucosal clearance [43]. Studies performed using pilus negative mutants of T4 clearly demonstrate a positive correlation between bacterial virulence and colonization and the presence of the pilus [10]. We therefore wanted to study, if the structural data found for the isolated pili help us to better understand the functionality of pilus mediated pneumococcal behaviour within a host, and whether structural properties of the pneumococcal pilus could be function derived.
How does the pneumococcal pilus mediate interaction with its host? Our data suggest that pneumococcal pili are flexible protofilament-based structures composed of ancillary proteins RrgA and RrgC and the RrgB backbone (Figure 9). Recently, proteins of group B streptococcal pilus were found to facilitate the interaction with endothelial cells [44]. Our data elucidate the adhesive properties of RrgA to fibronectin, laminin and collagen, suggesting that the clusters containing RrgA are the adhesive regions of pili. In silico analysis of RrgA (T4) sequence identified domains important for adhesion, like MSCRAMM motifs [19] and homologues of the von Willebrand factor A (vWFA) [45]. Interestingly, PapG, the adhesin of Escherichia coli P-pili, that binds to uroepithelial cells in its human host was also found to be located on the pilus surface, but only at the very distal end of the pilus fiber [46]. Streptococcus pneumoniae is a mucosal commensal, a mucosal pathogen and an invasive pathogen. Colonization of the nasopharynx by Streptococcus pneumoniae is a prerequisite for the development of pneumococcal disease and the result of a complex interplay between host and pathogen factors. Respiratory pathogens are known to release products which interfere with mucosal defences, causing epithelial disruption and cell death [47],[48]. Streptococcus pneumoniae was seen to adhere in particular to damaged cells and extruded cells [47], and bacteria were often found to be associated with damaged epithelium and exposed ECM [49]. Pathogen-ECM interactions have been found to be associated with adhesion and subsequent invasion of the pathogen [9]. Adhesive properties of pilus surface located ancillary protein RrgA to selected compounds of the ECM might therefore be part of the pilus mediated host-pathogen interplay. Flexibility of the pilus, as suggested by the protofilament-based structure, supports its functionality under in vivo conditions. Interestingly, recent work done by Nelson et al. [50] identified adhesive properties of pneumococcal pilus RrgA in cell-based assays. This together with data showing the impact of RrgA on pneumococcal virulence in mice [19],[50] indicate that the polypeptide may function at more than one stage in the infection process.
In summary, this report presents support for the structural composition of the Streptococcus pneumoniae pilus as an oligomeric appendage with adhesive properties and future work will help to further improve our understanding of the structure and function of the pilus and its main components.
Streptococcus pneumoniae type 4 strain TIGR4 has been described [20]. Mutants TIGR4Δpil (rrgA-srtD) [10] and TIGR4Δ(rrgA) [50] were kindly donated by B. Henriques-Normark (Karolinska Institute, Stockholm). The pneumococcal strains were stored at −80°C in 12% glycerol and routinely grown at 37°C in 5% CO2 on Tryptic Soy Agar (Becton Dickinson) supplemented with 5% defibrinated sheep blood or in Tryptic Soy Broth (Becton Dickinson). When appropriate, erythromycin (Sigma-Aldrich) as selection marker was used.
Standard recombinant DNA techniques were used to construct all expression plasmids. Vector pET 21b+ was purchased from Invitrogen. Full length sequence of T4 pili proteins RrgA (TIGR annotation No. sp0462), RrgB (TIGR annotation No. sp0463) and RrgC (TIGR annotation No. sp0464) with exception of their N-terminal signal sequence and C-terminal cell wall sorting signal motif, hydrophobic stretch and charged tail was cloned into pET21b+: pellets of IPTG induced recombinant Escherichia coli BLR(DE3) cultures, containing expressed His-tagged RrgA, RrgB and RrgC proteins respectively, were subjected to lysis by lysozyme in a BugBuster (Novagen), Benzonase Nuclease (Novagen) solution containing proteinase inhibitors. After centrifugation at 35000 rpm for 1 h at 4°C, the soluble fraction was subjected to metal chelate affinity chromatography on His-Trap HP columns (GE Healthcare) equilibrated and eluted according to manufacturer's instructions. Pooled fractions were dialysed overnight (ON) against 0.9% NaCl and stored at −80°C until further use. Protein concentration and purity was determined by scanning densitometry of Coomassie Blue-stained SDS-PAGE using a BSA standard and measuring 280 nm absorption of the protein solution (NanoDrop®).
Streptococcus pneumoniae T4 was chosen as starting material as far as the bacteria belong to a clinical relevant serotype 4 isolate, the sequence of which is known [20] and it represents a well characterized pneumococcal strain.
Streptococcus pneumoniae T4 glycerol stock (−80°C) was grown on Tryptic Soy Agar supplemented with 5% defibrinated mutton blood (ON at 37°C in 5% CO2). Fresh bacteria were used to incubate new agar plates and cultivated for about 12 h (at 37°C in 5% CO2). Harvested bacteria of about 10 plates were washed once in 35 ml PBS, and resuspended in 2 ml protoplast buffer PPB (10 mM MgCl2, 50 mM NaPPi pH 6.3, 20% sucrose) containing protease inhibitors. About 450 U of mutanolysin in 100 mM NaPPi pH 6.3 were added to each half of the suspension and incubated at 37°C for about 5 to 8 h with gentle shaking until protoplast formation was detected (microscopic control). Supernatant, containing digested pilus material was loaded on a sucrose gradient (25 to 56% in 10 mM MgCl2, 50 mM NaPPi pH 6.3) and run for about 20 h at 38000 rpm (4°C). All further steps were performed at 4°C using buffers containing protease inhibitors. In addition, benzonase nuclease (Novagen) was added to remove DNA and RNA impurities. Collected gradient fractions were tested for pilus material using anti-RrgB antibodies. Pilus containing fractions were pooled and dialyzed against 10 mM MgCl2, 50 mM NaPPi pH 6.3 for about one day to remove sucrose. When necessary, additional chromatography steps were added to reduce polydispersity and pooled pilus preparations were concentrated before loading them on a Superose 6 10/300 GL column (Amersham Biosciences). Gel filtration resulted in separation of pilus containing material of different molecular weight. Purified pilus fractions were judged to be homogeneous based on EM and SDS-PAGE. Samples were stored at −80°C or liquid nitrogen until further use.
Protein spots corresponding to HMW pili material were excised from SDS-PAGE gels (3–8% TA, Invitrogen), washed with 100 mM ammonium bicarbonate/ACN 50/50 v/v, and dried using a SpeedVac centrifuge (Savant, Holbrook, NY, USA). Dried spots were digested for 2 h at 37°C in 12 ml of 0.012 µg/ml sequencing-grade modified trypsin (Promega, Madison, WI, USA), in 50 mM ammonium bicarbonate. After digestion, 5 µl of 0.1% Trifluoroacetic acid (TFA) were added, and the peptides were desalted and concentrated with Zip-Tips (C18, Millipore). Samples were eluted with 2 µl of 5 g/l 2,5-dihydroxybenzoic acid in 50% ACN/0.1% TFA onto the mass spectrometer Anchorchip 384 (400 µm, Bruker Daltonics, Bremen, Germany), and allowed to air-dry at room temperature. MALDI-TOF spectra were acquired on a Bruker Ultraflex MALDI-TOF instrument (Bruker Daltonics). Protein identification was carried out by both automatic and manual comparison of experimentally generated monoisotopic values of peptides in the mass range of 700–3000 Da with computer-generated fingerprints using MASCOT software running on proprietary databases. Identifications were confirmed by MS/MS analysis: after denaturing the samples in a MS-compatible detergent (RapiGest SF, Waters) and boiling for 15 min, in-solution digestion was performed by adding 2 µg of trypsin, and allowing digestion ON. MS/MS spectra were acquired using an ESI-q-TOF Micro mass spectrometer (Waters), coupled to a nano-LC on a CapLC HPLC system (Waters). A MS survey scan was used to automatically select multicharged peptides over the m/z range of 400–2000 for further MS/MS fragmentation. After data acquisition, the MS/MS spectra were combined, smoothed and centroided by MassLynx software, version 4.0 (Waters). Search and identification of peptides were performed with a licensed version of MASCOT, in a local database, after converting the acquired MS/MS spectra in .pkl files.
Identification of the N-terminal amino acid sequence of HMW pili material by Edman degradation was performed according to standard conditions. HMW pili material, following size exclusion chromatography, was applied to SDS-PAGE (3–8% TA; Invitrogen). After western transfer to PVDF membrane, HMW pili band was cut out and used for Edman analysis.
SDS-PAGE analysis was performed using NuPAGE™ 3–8% Tris-Acetate Gels (Invitrogen) according to the instructions of the manufacturer. HiMark™ pre-stained HMW protein standard (Invitrogen) served as protein standard. Western analysis was done using standard protocols. Antibodies against recombinant HisTag-RrgB were used at 1/10000 dilution. Secondary goat anti-mouse HRP antibodies were used at 1/30000.
Antibodies against recombinant HisTag-RrgA (mouse; guinea pig), -RrgB (mouse), and -RrgC (mouse; rabbit) were produced in our lab and tested for specificity. Secondary goat anti-mouse HRP antibodies were obtained from Bio-Rad. Gold labelled antibodies for IEM were purchased of BBInternational: anti-mouse (5 nm), anti-rabbit (10 nm) and anti-guinea pig (15 nm).
96-well MaxiSorp™ flat-bottom plates (Nunc, Roskilde, Denmark) were coated for 1 h at 37°C followed by an ON incubation at 4°C with 2 µg/well of respective ECM vitronectin (from human plasma, BD Biosciences), lactoferrin (from human milk, Sigma), collagen I (from human lung, Sigma) and fibrinogen (from human plasma, Sigma) and with 1 µg/well with laminin (from human placenta, Sigma) and fibronectin (from human plasma, Sigma) in phosphate-buffered saline pH 7.4 (PBS). A BSA coated plate served as negative control. Plates were washed 3 times with PBS/0,05% Tween 20 and blocked for 2 h at 37°C with 200 µl of PBS/1% BSA followed by 3 washing steps with PBS/0,05% Tween 20. Recombinant protein samples (HisTag-RrgA, -RrgB and -RrgC) were initially diluted to 4 µg/ml with PBS. 200 µl of protein solution or 100 µl of wild type pilus preparation (53 µg/ml) and 100 µl T4Δpil sample (35 µg/ml), diluted in 200 µl total volume with PBS, and respective controls were transferred into coated-blocked plates in which the samples were serially diluted two-fold with PBS, obtaining a final volume of 100 µl/well. Plates were incubated for 2 h at 37°C and ON at 4°C. The plates were washed 3 times and incubated for 2 h at 37°C with respective primary mouse anti-HisTag-Rrg antibodies (1/10000 dilutions); pilus coated plates were incubated with anti-HisTag-RrgB antibodies. After another 3 washing steps, antigen-specific IgG was revealed with alkaline phosphatase-conjugated goat anti-mouse IgG (Sigma Chemical Co., SA Louis, Mo.) after 2 h of incubation at 37°C, followed by addition of the phosphatase alkaline substrate p-nitrophenyl-phosphate (Sigma). Read out was performed at 405 nm by an ELISA plate reader.
Formvar-carbon-coated nickel grids were charged with 5 µl of purified sample and let stand for 5 min. The grids were then fixed in 2% paraformaldehyde (PFA) in Phosphate Buffered Saline 0.1 M pH 7.4 (PBS) for 5 min, and placed in blocking solution (PBS containing 1% normal rabbit serum and 1% BSA) for 1 h at room temperature. The grids were then floated on drops of polyclonal antibodies α-RrgA (guinea pig), α-RrgB (mouse) and α-RrgC (rabbit) at dilution of 1∶10 in blocking solution for 1 h at room temperature, washed with 5 drops of blocking solution for 5 min, and floated on secondary gold-conjugated antibodies (goat anti-mouse IgG, 5 nm; goat anti-rabbit IgG, 10 nm; goat anti-guinea pig IgG, 15 nm) diluted 1∶20 in blocking buffer for 1 h. The grids were then washed with five drops of PBS and fixed in 2% PFA/PBS for 5 min at room temperature. Finally samples were washed with 8 drops of distilled water. Grids were stained with 1% buffered phosphotungstic acid (PTA) (pH 6.5) for 15 s, the excess of solution was soaked off by Whatman filter paper. The grids were examined in a CM10 Transmission Electron Microscope (TEM, Philips Electronic Instruments, Inc) operating at 80 kV. The command boxer from software EMAN was used to isolate and to count the single gold particles of different sizes.
20 µl of the solution containing purified pili were transferred onto a cover slip that had been previously cleaned by immersion in chromic acid solution followed by several rinses in distilled water. Pili were allowed to sediment on the glass surface then the cover slips were rinsed in distilled water to remove the excess of material. Immediately before freezing each cover slip was rapidly rinsed in distilled water, and a thin meniscus of solution was left on the glass to prevent dehydration of the samples. While the freezing machine was brought to its lowest temperature 4oK, the tiny glass was placed onto a thin slice of aldehyde fixed lung for support during freezing. This was accomplished by slamming the samples onto the liquid helium-cooled copper block of a quick freezing device (Cryopress; Med-Vac, Inc., St. Louis, MO). The frozen samples were freeze dried in a freeze etching unit (Baf 301; Balzers S.p.A., Milan, Italy) for 20 min at −80°C. Pilus absorbed to the cover slip were rotary replicated with ∼2 nm of platinum applied from an angle of 24° above the horizontal and then backed with 25-nm-thick film of pure carbon. Replicas were separated from the glass by immersion in concentrated hydrofluoric acid then cleaned with sodium hypochlorite. After several rinses in distilled water replicas were picked up on 75-mesh formvar-coated microscope grids. Samples were viewed in a transmission electron microscope (CM10; Philips Electronic Instruments, Inc, Mahwah, NJ) operating at 80 kV.
5 µl aliquots of whole bacteria were applied to 200-mesh copper grids coated with a thin carbon film and let stand for 5 min. The grids were first washed by streaming several drops of PBS over the grids. They were subsequently negatively stained by two drops of 1% buffered PTA (pH 6.5). The last drop was left on the grids for 17 s. Finally the grids were washed with several drops of ddH2O, the excess of liquid was soaked off by Whatman filter paper and quickly air dried. The grids were observed using a CM200 FEG Philips Electron Microscope (FEI, Eindhoven, The Netherlands), equipped with a GATAN GIF 2002 postcolumn energy filter (Gatan, Pleasanton, California, United States), and images were collected at an accelerating voltage of 200 kV and a nominal magnification of 50000×, on Kodak SO163 film.
5 µl of purified pili preparation were loaded onto a glow discharged Quantifoil holey carbon grid with 2 µm holes. After being blotted from the front side with a slip filter paper (Whatman No. 4), the grid was flash frozen into liquid ethane as described [51].
Micrographs taken at 50000× of magnification were digitized on a IMACON 949 scanner at spacing of 7.95 µm resulting in a nominal sampling of 1.6 Å/pixel-1. Pili were picked from digitized images using the command “helixboxer” from the software EMAN [34]. Digitized pili images were cut into individual repeats by using boxes of 300×300 pixels, with overlapping ends, using 10 pixel shift for each box, so that adjacent boxes had 90% overlap. The isolated repeats were treated as single particles. In a first analysis, the straightest pili segments were selected and pre-aligned interactively, subsequently the pre-aligned repeats were aligned using alignments with only limited angular ranges (−5°, +5°), finally a vertical alignment has been performed using as a future-less reference the projection of a model cylinder followed by translational alignment perpendicular to the cylinder axis only. Aligned repeats were than subjected to high-pass and low-pass filtrations before the density profiles were calculated (the densities across the filament axis of the pili were projected onto the short axis) using different commands of IMAGIC 5 [35] and of Bsoft software [52]. All the aligned and filtered images were consistent: they all presented centred rods with similar diameters. The only major differences were the surrounding stain distributions.
Swiss-Prot (http://www.expasy.org/sprot/) accession numbers for pilus proteins mentioned in the text are:
SP_0462, RrgA (TIGR4) Q97SC3
SP_0463, RrgB (TIGR4) Q97SC2
SP_0464, RrgC (TIGR4) Q97SC1 |
10.1371/journal.pntd.0001774 | Significantly Reduced Intensity of Infection but Persistent Prevalence of Schistosomiasis in a Highly Endemic Region in Mali after Repeated Treatment | Preventive chemotherapy against schistosomiasis has been implemented since 2005 in Mali, targeting school-age children and adults at high risk. A cross-sectional survey was conducted in 2010 to evaluate the impact of repeated treatment among school-age children in the highly-endemic region of Segou.
The survey was conducted in six sentinel schools in three highly-endemic districts, and 640 school children aged 7–14 years were examined. Infections with Schistosoma haematobium and S. mansoni were diagnosed with the urine filtration and the Kato-Katz method respectively. Overall prevalence of S. haematobium infection was 61.7%, a significant reduction of 30% from the baseline in 2004 (p<0.01), while overall prevalence of S. mansoni infection was 12.7% which was not significantly different from the baseline. Overall mean intensity of S. haematobium and S. mansoni infection was 180.4 eggs/10 ml of urine and 88.2 epg in 2004 respectively. These were reduced to 33.2 eggs/10 ml of urine and 43.2 epg in 2010 respectively, a significant reduction of 81.6% and 51% (p<0.001). The proportion of heavy S. haematobium infections was reduced from 48.8% in 2004 to 13.8% in 2010, and the proportion of moderate and heavy S. mansoni infection was reduced from 15.6% in 2004 to 9.4% in 2010, both significantly (p<0.01). Mathematical modelling suggests that the observed results were in line with the expected changes.
Significant reduction in intensity of infection on both infections and modest but significant reduction in S. haematobium prevalence were achieved in highly-endemic Segou region after repeated chemotherapy. However, persistent prevalence of both infections and relatively high level of intensity of S. mansoni infection suggest that more intensified control measures be implemented in order to achieve the goal of schistosomiasis elimination. In addition, closer monitoring and evaluation activities are needed in the programme to monitor the drug tolerance and to adjust treatment focus.
| Schistosomiasis or bilharzia is caused by infection with the trematode Schistosoma spp. It is widely prevalent in Mali, causing a major public health problem. One of the major control measures is mass treatment of population at risk, particularly school-age children, using the drug praziquantel. Mali started such mass drug treatments in 2005 and a cross-sectional survey was conducted in 2010 in highly endemic region Segou to evaluate the impact of the programme comparing with the baseline data collected in 2004. The repeated treatment significantly reduced the intensity of infection which is closely related to the morbidity of schistosomiasis, however, the prevalence persists with little reduction in the highly endemic region. The results suggest that more intensified control measures be implemented in order to achieve the goal of schistosomiasis elimination, possibly more frequent treatments, improved coverage and improved water and sanitation.
| Schistosomiasis or bilharzia, caused by infection with Schistosoma spp, is one of the major neglected tropical diseases (NTDs). It remains a major cause of morbidity in developing countries, especially in sub-Saharan Africa. It has been estimated that around 200–209 million people worldwide may be infected with schistosomiasis and about 85% of these infections occur in Africa [1], [2], [3]. The mainstay of the current control strategy recommended by World Health Organization (WHO) against schistosomiasis is preventive chemotherapy (PCT) with praziquantel (PZQ) [4], [5]. The aim is to reduce morbidity due to schistosomiasis through regular treatment. Praziquantel is safe and efficacious against both Schistosoma haematobium (causing urogenital schistosomiasis) and S. mansoni (causing intestinal schistosomiasis), two major human species in sub-Saharan Africa. Repeated chemotherapy, usually as an annual regular treatment, can reduce infection and associated morbidity [6], [7], [8], [9]. WHO recently estimated that about 237 million people, including 109 million school-age children, globally require annual treatment for schistosomiasis, of which 220 million are in Africa, and only 12.7% of this estimated population have received treatment in 2010 [10].
In Mali, both urogenital and intestinal forms of schistosomiasis are prevalent throughout the country with geographically varying degrees of prevalence [11], [12], [13], [14]. The most recent survey in 2004–2006 showed a prevalence of 38.3% (ranging 0.0–99.0%) for S. haematobium and 6.7% (ranging 0.0–94.9%) for S. mansoni [14]. Mali was one of the first countries in sub-Saharan Africa to initiate a national schistosomiasis control programme. Control efforts were initiated in the Bandiagara district (Mopti region) as a component of a dam-building project in 1978 and it became a national programme in 1982 [11], [14]. The initial control program with PZQ distribution was implemented by the Ministry of Health, through the National Institute of Public Health Research (INRSP), in collaboration with WHO and with funds from the German Technical Co-operation (GTZ) [11]. GTZ's support for the control programme ceased in 1992 and the Malian government took over the financial responsibilities. However, lack of resources led to control activities being considerably reduced.
In 2004, a new initiative to resume the national control activities was set up with technical and financial support from the Schistosomiasis Control Initiative (SCI) [15], [16]. Mass drug administration (MDA) with PZQ commenced in 2005 in four endemic regions, targeting only school-age children (7–14 years old) attending schools and in 2006 in two other regions, targeting all school-age children (5–15 years old) [16]. In 2007, the schistosomiasis control programme became part of the integrated national control programme on NTDs, funded primarily by the United States Agency for International Development (USAID) NTD Control Program managed by RTI International and implemented initially by International Trachoma Initiative (2007) and latterly by Helen Keller International (since 2008) [17]. Since then, MDA with PZQ has been steadily scaled up to cover seven regions plus Bamako, targeting all school-age children and adults at high risk in hyper-endemic regions, achieving 100% geographical coverage and 72–100% programme coverage [17].
To monitor the progress and impact of the national schistosomiasis control programme, limited parasitological surveys were conducted in selected sentinel schools. The current paper presents the parasitological impact of repeated treatment on schistosomiasis in the highly endemic region of Segou and the recommendations for future control efforts.
Mali's schistosomiasis control programme is a national disease control programme using the WHO endorsed preventive chemotherapy strategy [5], and therefore MDA with praziquantel did not require specific ethical clearance. Data collection at sentinel sites was within the framework of the national control programme and was approved by the Ministry of Health. Before doing the survey at each sentinel school, informed verbal consent was first obtained from the chief of the village and the parents of children during the village meeting and was also obtained from school teachers at schools prior to the recruitment of children. These were recorded by the survey team leader. During recruiting, informed verbal consent was obtained from each child with the presence of school teachers. Any children who did not want to participate were free to leave. In Mali, for traditional cultures and low literacy rate in villagers, verbal consent was deemed accepted procedures and approved by the Ministry of Health.
Mali's national schistosomiasis control programme adopts the preventive chemotherapy strategies recommended by WHO [4], [5]. An MDA campaign is designed each year and health personnel at regional, district and community health centres are mobilized. Drugs are distributed through the school-based delivery in schools targeting school-going children and through the community-based delivery in villages targeting non-school-going children and community members at risk. School-based drug delivery is carried out by trained school teachers. Community-based drug delivery is carried out by trained community drug distributors. PZQ tablets (600 mg) were delivered using the WHO dose pole method to determine the dosage for each child [18].
At the start of the national control programme, a national mapping of schistosomiasis was conducted [14]. A number of sentinel sites were randomly selected across the country and the details of the overall study design have been described elsewhere [19], [20]. Baseline data collection before MDA and longitudinal cohort follow-up surveys after MDA were conducted during 2004–2006 [19], [20]. Such longitudinal follow-up surveys stopped after 2006 due to cessation of the SCI funding. In 2010, further cross-sectional survey was undertaken in three health districts (Segou, Macina, and San) in the Segou region, which is highly endemic with schistosomiasis according to the previous mapping surveys [12], [13], [14]. MDA in this region started in 2005. Before the survey in 2010, Segou and Macina districts received four rounds of MDA in 2005, 2006, 2008 and 2009, while San district received three rounds of MDA in 2005, 2006 and 2008. Segou district is located on the Niger River basin; Macina district is an irrigation system area for rice cultivation; and San district reflects a Sahelian environment. The main activities of the population are agriculture, with predominantly rice cultivation and vegetable growing (Macina) and cultivation of millet (San, Segou). The three districts were selected based on their ecological transmission profiles, thus the sentinel sites represent different transmission settings. Two sentinel sites/schools were selected from each district. A total of six sites were surveyed in 2010. The location and the baseline prevalence of schistosomiasis for these six sites are shown in Figure 1. The sampling method for children of 7–14 years old within each school was similar to that used in the baseline survey in 2004 [19], [20]. Briefly, within each site/school, 14 children (7 males and 7 females) were randomly selected from each of eight age groups of 7–14 years, giving a total number of approximately 110 children per school.
Cross-sectional analysis was conducted on the baseline data collected in 2004 before MDA and the data collected in 2010 after a number of rounds of MDA. The 7-year olds were the first year pupils in the schools and these children were supposed to be treated by community-based drug delivery before they joined the schools. The infection status in this group of children would represent the quality of treatment in the community [21]. Therefore, we also compared the data in the 7-year olds only from the baseline (2004), after one treatment (2005) and after 3–4 rounds of treatments (2010). Prevalence and intensity of infection with 95% confidence intervals were calculated using the SPSS (IBM, version 19) Complex Samples module taking into consideration the cluster nature of school children using district as strata and school as clusters. Samples were weighted according to the population sizes in the surveyed districts when calculating the overall prevalence and intensity of infection. Arithmetic mean intensity of infections from all the subjects examined (including both positive and negative) was used in the analysis [22], [23]. The individual infection was categorized as heavy (≥50 eggs/10 ml of urine) and light (<50 egg/10 ml of urine) infections for S. haematobium, and heavy (≥400 epg), moderate (100–399 epg) and light (1–99 epg) infections for S. mansoni. The Chi-squared test was used to compare differences in prevalence and the Kruskal–Wallis test was used to compare differences in mean intensities. The coordinates of each survey site was identified on Google maps and the site location map was drawn in ArcMap version 10 (ESRI, Redlands, CA).
The expected prevalence and intensity of infection in 2010 was predicted by mathematical models. The dynamic schistosomiasis transmission model, EpiSchisto, predicts the impact of chemotherapy and has been validated for S. haematobium and for S. mansoni [24], [25]. For easier manipulation, the equations in EpiSchisto were coded into the Berkeley Madonna modelling software (version 5.0) [26].
Schistosomiasis infection typically displays a high degree of overdispersion (or aggregation, whereby the majority of the parasites are harboured in a minority of human hosts) and the degree of this overdispersion often varies significantly between locations and between species. The distribution of parasites amongst hosts is often taken to be adequately approximated by the negative binomial distribution [27], which is described by two parameters: m, the mean intensity of infection, and k, the inverse of the overdispersion. In the EpiSchisto software, the overdispersion parameter k is allowed to vary linearly with the mean intensity of infection via the following relationship:
Given the high levels of heterogeneity observed in schistosomes in different epidemiological settings, the k values were estimated separately for each of the three districts (Macina, San, Segou) and for each of the two species (Table 1). The parameters were fitted using a maximum likelihood estimation approach [28]. The observed baseline mean intensity of infection (m) and the estimated k values were input into the models for prediction, using the treatment schedules in each district. Treatment was assumed to have 95% efficacy [24], [29], and we used 75% as the minimum treatment coverage required as per WHO recommendations each year for each district.
Table 2 summarizes the prevalence and the intensity of infection in school children from the baseline in 2004 and the survey in 2010. The data collected from 640 school children aged 7–14 years in 2010 from six sentinel schools were compared with the baseline data from the same age groups (648 school children) in the original cohort of the same schools.
In 2004, S. haematobium infection was extremely prevalent (75.9–98.7%) and heavy (mean intensity of infection over 50 eggs/10 ml of urine) in all three districts surveyed. Macina district also had heavy S. mansoni infection (prevalence 66.5% and mean intensity of infection 372.1 epg). Overall prevalence of S. haematobium infection was 88% and that of S. mansoni infection was 17.3%.
In 2010, overall prevalence of S. haematobium infection was 61.7%, a significant reduction of 30% from the baseline in 2004 (p<0.01), while overall prevalence of S. mansoni infection was 12.7% which was not significantly different from the baseline in 2004 (p>0.05). Overall mean intensity of S. haematobium and S. mansoni infection was 180.4 eggs/10 ml of urine and 88.2 epg in 2004 respectively. These were reduced to 33.2 eggs/10 ml of urine and 43.2 epg in 2010 respectively, significant reductions of 81.6% and 51% (p<0.0001). Among three districts, different degrees of reduction were seen in S. haematobium infection in 2010, with San showing 61.7% and 92.0%, Macina 33.5% and 91.7% and Segou 11.0% and 79.3% in prevalence and intensity of infection respectively.
There were no statistically significant differences between boys and girls in prevalence and mean intensity of S. haematobium infection either in 2004 or in 2010 and of S. mansoni infection in 2004 (p>0.05). However, there was a significant difference between boys and girls in prevalence and mean intensity of S. mansoni infection in 2010 (p<0.05). As between boys and girls, similar pictures were seen among different ages in the prevalence and the mean intensity of S. haematobium or S. mansoni infection in 2004 and 2010 (details not shown).
The overall proportion of heavy S. haematobium infections was reduced from 48.8% in 2004 to 13.8% in 2010, and the overall proportion of moderate and heavy S. mansoni infection was reduced from 15.6% in 2004 to 9.4% in 2010, both significantly (p<0.01). The shift of the categories of intensity of infection in three districts is illustrated in Figure 2.
The prevalence and intensity of infection observed in 2010 were compared to those predicted from mathematical models. The model-derived changes in the prevalence and intensity of infection were estimated for two separate groups: the 7–14 years old and the entire population. The predicted results of the 7–14 year group allow for direct comparison with the observed results, while the predicted results in the wider age range (0–60 years) indicate the expected treatment impact in the entire communities. The results are presented in Figure 3 for S. haematobium and in Figure 4 for S. mansoni in Macina district. S. manosni in the other two districts was not analyzed using mathematical models as the prevalence was very low. Overall, the predicted changes in the prevalence and intensity of infection were within or close to the 95% confidence intervals of the observed results in each district. This suggests that the observed results in each district were largely expected using the current MDA strategy.
As shown in Table 3, S. haematobium prevalence in 2010 was 59.4%, significantly lower than 90.5% in 2004 (p<0.001) but higher than 47.2% in 2005 (p<0.05), with an overall reduction of 34.4%. S. haematobium mean intensity of infection was 25.6 eggs/10 ml of urine in 2010, significantly lower than 194.9 eggs/10 ml of urine in 2004 (p<0.001) but not significantly different from 38.0 eggs/10 ml of urine in 2005 (p>0.05). There was a significant overall reduction of 86.9% in S. haematobium mean intensity of infection.
S. mansoni prevalence was 26.5% in 2010, not significantly different from 20.2% in 2004 (p>0.05) but significantly higher than 16.0% in 2005 (p<0.05). S. mansoni mean intensity of infection was 71.7 epg in 2010, not significantly different from 85.0 epg in 2004 (p>0.05) but higher than 49.5 epg in 2005 (p<0.05). There was no significant overall change in either prevalence or mean intensity of S. mansoni infection.
The baseline data showed that three districts surveyed in the Segou region were highly endemic with schistosomiasis. These three districts represent three different epidemiological settings: river basin, irrigated rice field and Sahelian area. Two sites in Segou district along the Niger River showed the highest level of infection and school children were nearly universally infected with S. haematobium; two sites in Macina district in the irrigation area showed high level of infection with both S. haematobium and S. mansoni; and two sites in San district in the Sahelian area showed lower but still high level of infection of S. haematobium (Figure 1). This represents significant disease burden caused by schistosomiasis in children in this region.
Before the survey in 2010, schoolchildren in Segou and Macina districts received four rounds of treatment and those in San district received three rounds of treatment. And yet the overall prevalence of S. haematobium infection in 2010 remained unacceptably high, though a significant 30% reduction compared with the baseline, while overall prevalence of S. mansoni infection did not show a significant reduction. Looking at three epidemiological settings, S. haematobium prevalence at two sites along the Niger River was almost at the baseline level; prevalence in the irrigation area showed a modest reduction (around 30%) for both S. haematobium and S. mansoni; while S. haematobium prevalence in the Sahelian area showed the biggest drop (61.7%). Mathematical modelling of the changes in prevalence and intensity of infection in three districts showed that the observed results were in line with the expected changes in each of three epidemiological settings. Such varying level of reduction in different epidemiological settings may be explained by different transmission levels and patterns in these settings: permanent transmission along the big river versus seasonal transmission in temporary water bodies. In the Sahelian environment in Burkina Faso and Niger, one round of MDA significantly reduced S. haematobium prevalence to a very low level, which remained low for 2–3 years [21], [30], [31]. However, along the Niger River in Niger, a similar situation as in Segou district, one year after treatment the S. haematobium prevalence bounced back to nearly the pre-treatment level [32], suggesting high level of transmission and frequent water contact activities in such locations. In Ugandan national control programme, S. mansoni prevalence also showed less reduction in highly endemic areas along the lake shores after repeated MDA [33].
Regardless of persistent prevalence, significant reduction in overall intensity of infection was seen in 2010 for both S. haematobium (by 81.6%) and S. mansoni (by 51%). The proportion of heavy S. haematobium infection was reduced from 48.8% to 13.8%, and the proportion of moderate and heavy S. mansoni infection was reduced from 15.6% to 9.4%. This is important as it is well known that the severity of morbidity caused by schistosomiasis is closely related to the intensity of infection. The heavier infections, in general, cause the more severe morbidity [34]. It is therefore anticipated that significant morbidity would have been prevented or reverted due to MDA in the national control programme. This is in line with what was achieved in other national MDA programmes in both East and West Africa through preventive chemotherapy [8], [9], [21], [33], [35]. However, there still exist a significant proportion of children with relatively low intensity of infections: 27.2–52.9% with S. haematobium and 0–18.8% with S. mansoni in three districts as shown in Figure 2, not including those undetected due to the low sensitivity of the diagnostic techniques. Such light infections have long been overlooked in terms of the morbidity consequences, and recent findings suggest that light infections can cause considerable morbidity due to anaemia, chronic pain, diarrhoea, exercise intolerance and undernutrition [34], [36]. The objective was to reduce morbidity due to schistosomiasis by regular treatment according to the WHO recommendations [4], [5], and regular MDA in the highly endemic regions in Mali may have just served this purpose, however, the persistent infection in the Segou region highlights the need for a persistent effort.
Schistosomiasis is one of the five major NTDs currently targeted for preventive chemotherapy through the integrated MDA strategy [5]. WHO published its first NTD report in 2010 [37], and has just launched a new roadmap for overcoming the burden of the NTDs, including elimination of schistosomiasis [38]. To achieve the objectives, comprehensive control measures are recommended including preventive chemotherapy, intensified case management, vector and intermediate host control, veterinary public health at the human-animal interface, and provision of safe water, sanitation and hygiene [37], [38]. However, in the current schistosomiasis control in the integrated national NTD control programmes, such as the one in Mali, the funding focus has been almost exclusively on preventive chemotherapy, while other components such as case management, snail control, and safe water, sanitation and hygiene are not implemented due to lack of funding. The current results, together with others, suggest that MDA alone in such highly endemic areas is simply not enough. In sub-Saharan countries, such as Mali, it is almost impossible to implement such other components in schistosomiasis control programme without proper external funding. It is therefore understandable that schistosomiasis prevalence persists in such high transmission areas even after several rounds of MDA. To achieve the said objectives in the new WHO roadmap towards schistosomiasis elimination [38], funds must be made available for the national programmes to implement the recommended comprehensive control measures.
Persistent prevalence of S. haematobium and S. mansoni infections and relatively high level of intensity of infection for S. mansoni (in the highly endemic Macina district) after several rounds of MDA indeed raised some concerns in schistosomiasis control in Mali. As shown in the 7-year olds, the first MDA reduced both prevalence and intensity of infection; however, prevalence for both species and intensity for S. mansoni had significantly rebounded since 2005. One concern is the treatment coverage and quality in Segou region. The reported program coverage in Segou region for schistosomiasis treatment was 56.4% in 2005, 75.1% in 2006, 0% in 2007, 76.4% in 2008 and 71.6% in 2009 according to the national schistosomiasis control programme. However, a post-MDA survey conducted in 2009 showed a significantly lower coverage than the reported coverage (details not shown). Lessons have been learned and measures have since been taken to increase the drug coverage. Another concern may be the possible drug tolerance, particularly for S. mansoni. PZQ has been used at large scale in Mali for rather a long time, first between 1982 and 1992 and second since 2005. Mounting drug pressure may have caused parasites to establish a certain level of tolerance. A similar situation was observed in a recent treatment trial in Niger [32]. These together suggest that a closer monitoring of drug efficacy in the national MDA programme is required.
It is noted that the current results are limited due to lack of more sentinel site surveys and only represent areas with similarly high level of transmission as in Segou region. Schistosomiasis is endemic in all regions of Mali with varying levels of endemicity, and the current situation in other regions is not clear. After several rounds of MDA, there is a need for a national survey in order to monitor the treatment impact, adjust the treatment focus and increase cost-efficiency of control measures.
MDA with PZQ has been conducted in Mali since 2005 and several rounds of treatment have been delivered. Sentinel site surveys in Segou region showed that significant reduction in intensity of infection on both infections and modest but significant reduction in S. haematobium prevalence were achieved in highly endemic regions. Most importantly, proportion of moderate and heavy infections was reduced in school-age children. Significant prevention and reversion of morbidity due to schistosomiasis was anticipated in the Mali NTD control programme. However, persistent prevalence of both infections and relatively high level of intensity of S. mansoni infection suggests that more intensified control measures be implemented, possibly more frequent treatments, improved coverage and improved water and sanitation. In addition, closer monitoring and evaluation activities are needed in the programme to monitor the drug tolerance and to adjust treatment focus.
|
10.1371/journal.pntd.0000738 | Comparison of Three Commercially Available Dengue NS1 Antigen Capture Assays for Acute Diagnosis of Dengue in Brazil | Dengue is associated with explosive urban epidemics and has become a major public health problem in many tropical developing countries, including Brazil. The laboratory diagnosis of dengue can be carried out using several approaches, however sensitive and specific assays useful to diagnose in the early stage of fever are desirable. The flavivirus non-structural protein NS1, a highly conserved and secreted glycoprotein, is a candidate protein for rapid diagnosis of dengue in endemic countries.
We aimed to evaluate the potential use of 3 commercial kits in a panel of 450 serum samples for early diagnosis of dengue in Brazil. The PanBio Early ELISA (PanBio Diagnostics) showed a sensitivity of 72.3% (159/220) and a specificity of 100%, while the sensitivity of the Platelia™ NS1 assay (Biorad Laboratories) was 83.6% (184/220). However, the highest sensitivity (89.6%; 197/220) was obtained by using the NS1 Ag Strip (Biorad Laboratories). A lower sensitivity was observed in DENV-3 cases by all 3 kits. Serum positive by virus isolation were more often positive than cases positive by RT-PCR by all three assays and a higher detection rate was observed during the first four days after the onset of the symptoms. The presence or absence of IgM showed no influence in the confirmation by the pan-E Early ELISA (P = 0,6159). However, a higher confirmation by both Platelia™ NS1 (Biorad) and Dengue NS1 Ag Strip (Biorad) in the absence of IgM was statistically significant (P<0,0001 and P = 0,0008, respectively). Only the Platelia™ NS1 test showed a higher sensitivity in confirming primary infections than secondary ones.
The results indicate that commercial kits of dengue NS1 antigen are useful for the laboratory diagnosis of acute primary and secondary dengue. It can be used in combination with the MAC-ELISA for case detection and as screening test to complement viral isolation.
| Dengue is the one of the most prevalent arthropod-borne viral diseases in tropical regions of the world. Manifestations may vary from asymptomatic to potentially fatal complications. Laboratorial diagnosis is essential to diagnose dengue and differentiate it from other diseases. Dengue virus non-structural protein 1 (NS1) may be used as a marker of acute dengue virus infection. Our results, based in the comparison of three NS1 antigen capture assays available, have shown that this approach is reliable for the early diagnosis of dengue infections, especially in the first four days after the onset of the symptoms. A lower sensitivity was observed in DENV-3 cases. Serum positive by virus isolation were more often detected than those positive by RT-PCR by all three assays. Only the Platelia™ NS1 test showed a higher sensitivity in confirming primary infections than secondary ones. In conclusion, NS1 antigen capture commercial kits are useful for diagnosis of acute primary and secondary dengue infections and, in endemic countries where secondary infections are expected to occur, may be used in combination with MAC-ELISA to increase the overall sensitivity of both tests.
| Dengue is associated with explosive urban epidemics and has become a major public health problem [1]. Annually, the World Health Organization estimates that 50–100 million people are infected with dengue virus (DENV) worldwide with estimated 250,000–500,000 cases of dengue haemorrhagic fever (DHF) and dengue shock syndrome (DSS) with about 25,000 deaths occurring. One or more of four serotypes of DENV (DENV1–4), a mosquito-borne, positive-strand RNA virus in the genus Flavivirus, family Flaviviridae cause the disease in more than 100 endemic countries in tropical areas [2].
The geographical spread of all four DENV serotypes throughout the subtropical regions of the world has led to larger and more severe outbreaks and the accurate and efficient diagnosis of the disease is important for clinical care, surveillance, pathogenesis studies and vaccine research. Furthermore, an efficient diagnosis is an important tool to support Epidemiological Surveillance Programs considering the difficulties in confirming dengue cases based only on the clinical symptoms, especially during inter-epidemic periods.
Dengue is an enveloped virus with a single-stranded, positive sense RNA genome of about 11 kb containing a single open reading frame enconding a single polyprotein co- and pos-translationally cleaved into 3 structural (C, prM and E) and 7 nonstructural proteins (NS1, NS2A, NS2B, NS, NS4A, NS4B and NS5) [3].
Dengue is a major public health problem in many tropical and subtropical countries in the world. The accurate and efficient diagnosis of dengue is important for clinical care, surveillance, pathogenesis studies, and vaccine research.
The most used techniques use for dengue serodiagnosis are based on the anti-DENV IgM and IgG detection by using MAC-ELISA and IgG-ELISA [4]. However, one of the limitations consists in the variations on the detection rate during the acute phase of the disease. Usually, it takes from 3 to 5 days after the onset of the symptoms to detect anti-DENV IgM and from 1 to 14 days to anti-DENV IgG to become detectable, depending on whether the patient has primary or secondary infections [5].
During the acute phase, however, the NS1 exists as secreted as well as a membrane-associated protein and both forms are demonstrated to be immunogenic [6], [7], [8], [9], [10]. High NS1 level was demonstrated to circulate in the acute phase of dengue by antigen capture ELISAs, found in the sera of patients with primary and secondary DENV infections, up to the ninth day after the onset of the symptoms [10], [11].
The availability of commercial kits for the detection of anti-DENV NS1 in acute serum provides an alternative to the existing methods such as PCR, serology and virus isolation. Previous studies have shown the sensitivity and specificity of NS1 capture commercial kits for the laboratorial diagnosis of dengue infections [12], [13], [14], [15], [16], [17], [18], [19].
Recently, the Brazilian Ministry of Health has establish this new approach in sentinel clinics throughout the country after the 2008 dengue epidemic, however without a full evaluation of the commercial tests available. In the study, we aimed to evaluate the sensitivity and specificity of 3 commercially-available dengue NS1 antigen kits to demonstrate its potential use for the early laboratory confirmation of acute dengue infection in Brazil. This constitutes the first report of a comparison of NS1 antigen capture assays performed in the country.
The samples belong to a previously-gathered collection from an ongoing Project in the Laboratory approved by the Ethics Committee on Human Research (CEP: 274/05).
Laboratory-positive DENV infection was defined in patients experiencing a febrile illness consistent with dengue according to WHO criteria [20] in which infection was confirmed by DENV isolation [21], detection of DENV RNA by RT-PCR [22], detection of anti-DENV IgM antibodies by MAC-ELISA[23], and/or a >4-fold rise in anti-DENV IgG-ELISA titer in paired acute and convalescent sera [24]. Individuals negative for DENV infection by using all the methods described above and health individuals were classified as non-dengue cases.
The serum samples (days 1st to 9th after the onset of the symptoms) analyzed in this study by the pan-E Early ELISA (PanBio Diagnostics, Brisbane, Australia- first generation), Platelia™ (Biorad Laboratories, Marnes-La-Coquette, France) and NS1 Ag Strip (Biorad Laboratories, Marnes-La-Coquette, France) belong to a previously-gathered serum collection of the Laboratory of Flavivirus at Oswaldo Cruz Institute, FIOCRUZ, Brazil, from epidemics occurred from 1986 to 2008. A panel of 450 sera (220 dengue positive sera and 230 non-dengue sera) was divided into eleven Groups as follows: Groups A to C, sera from patients infected with DENV-1 (n = 50), DENV-2 (n = 50), and DENV-3 (n = 58), respectively; Group D, sera from patients with dengue infection serologically confirmed by MAC-ELISA with negative virus isolation and RT-PCR (n = 62); Group E, sera from healthy individuals (n = 30); Group F, sera from individuals negative for dengue (n = 86); Group G, sera from yellow fever positive individuals (n = 20); Group H, sera from individuals vaccinated for yellow fever and negative for anti-DENV antibodies (n = 44); Group I, sera from measles patients (n = 16) and Group J, sera from rubella patients (n = 34).
Virus isolation was performed by inoculation into C6/36 Aedes albopictus cell line [21] and isolates were identified by indirect fluorescent antibody test (IFAT) using serotype-specific monoclonal antibodies [25].
RT—PCR for detecting and typing DENV was performed as described previously [22]. Briefly, consensus primers were used to anneal to any of the four DENV types and amplify a 511-bp product in a reverse transcriptase-polymerase reaction. A cDNA copy of a portion of the viral genome was produced in a reverse transcriptase reaction. After a second round of amplification (nested PCR) with type-specific primers, DNA products of unique sizes for each DENV serotype were generated and analyzed by gel electrophoresis.
The in-house MAC-ELISA was carried out for dengue cases confirmation as described previously [23].
The IgG—ELISA previously described by Miagostovich [24] was performed for the characterization of dengue immune response as primary or secondary infections in dengue cases previously confirmed by virus isolation, RT—PCR and/or MAC-ELISA. Briefly, 96-well plates were coated with hyper immune ascitic fluid (a mixture of anti-DENV-1 to 4), followed by the addition of a mixture of the four DENV antigens. Serum diluted 1∶40 was added to the first well and four-fold dilutions were carried out up to the eighth well. After incubation, anti-human IgG conjugated to horseradish peroxidase was added Acute phase serum samples (<6 days after onset of symptoms) with IgG-ELISA titers of 1∶160 or greater are considered to be secondary infections. Likewise, samples with titers >1∶10, 240 on days 6–9, or >1∶40, 960 on days 10–15 after onset are secondary responses.
The sensitivities, specificities, efficiency, negative and positive predicted values were calculated as follows:
Sensitivity: a/a+c X 100%
Specificity: d/d+b X 100%
Efficiency: a+d/a+b+c+d X 100%
Negative Predicted Value: d/d+c X100%
Positive Predicted Value: a/a+b X 100%; where: a = number of true positive, b = number of false positive, c = number of false negative and d = number of true negative.
The derived data was tabulated in appropriate worksheets using the Microsoft Excel and evaluated by chi-square test using the Epi Info 6 (Center for Disease Control and Prevention, Atlanta) for any statistical significant association.
A panel of 450 (n = 220 dengue cases and n = 230 non-dengue cases) was used to evaluate three NS1 antigen capture tests commercially available. The overall sensitivities were 72.3% (159/220) for the pan-E Early ELISA (PanBio) test, 83.6% (184/220) for the Platelia™ NS1 (BioRad) kit, and 89.6% (197/220) for the NS1 Ag Strip kit (BioRad), Table 1. The differences observed in the sensitivities between the three kits analyzed were statistically significant (P = 0.0009).
The pan-E Early ELISA (PanBio) showed a higher sensitivity in confirming DENV-2 infections (Group B; 82.0%) than confirming DENV-1 and DENV-3 infections. The Platelia™ NS1 kit (BioRad) was more sensitive in the detection of DENV-1 cases (Group A; 98.0%) than in the detection of DENV-2 and DENV-3 infections. The Dengue NS1 Ag Strip kit (BioRad) showed the same sensitivity in confirming DENV-1 and DENV-2 infections (98.0%). DENV-3 infections were the detected less often by all the three kits tested (65.5%, 86.2% and 88.0% for the pan-E Early ELISA (PanBio), the Platelia™ NS1 kit (BioRad) and the Dengue NS1 Ag Strip kit (BioRad), respectively (Table 1).
Specificities were 100%, 98.7% and 99.1% for the PanBio kit, for the Platelia™ NS1 kit (Biorad) and for the NS1 Ag Strip (BioRad), respectively, based on the analysis of sera of healthy individuals (Group E) and individuals negative for dengue (Group F), Table 1. No cross-reactivity was observed with sera from yellow fever infected patients (Group G); however both Biorad kits showed cross-reactivity with one yellow fever vaccinee (Group H). None of the measles sera (Group I) were recognized by the NS1 tests. One rubella positive case (Group J) showed cross-reactivity with both Platelia™ NS1 kit (Biorad) and for the NS1 Ag Strip (BioRad) kit. The overall evaluations according to the different Groups analyzed are shown in Table 1.
A higher sensitivity (71.5%, 94.8% and 98.7%) was observed in cases positive by virus isolation only than in cases previously positive by RT-PCR (62.3%, 82.3% and 82.3%) for the pan-E Early ELISA, Platelia™ NS1 and Dengue NS1 Ag Strip, respectively (Table 2).
The detection rate by the pan-E Early ELISA, Platelia™ NS1 and Dengue NS1 Ag Strip in the presence of IgM was 69.4%, 64.5% and 77.4%, respectively (Table 2). In this study, the presence or absence of IgM did not influence detection by the pan-E Early ELISA (P = 0,6159). However, a higher detection rate by both Platelia™ NS1 (Biorad) and Dengue NS1 Ag Strip (Biorad) in the absence of IgM was statistically significant (P<0,0001 and P = 0,0008, respectively).
The sensitivities of all NS1 tests were evaluated according to the number of days of illness. A higher detection rate by the three tests analyzed was during the first four days after the onset of the symptoms (Day 3, in Figure 1, considering day 0 as the first day of fever). The sensitivity of Platelia™ NS1 (Biorad) decreased to 75% of detection after that and maintained the same rate until day 6 of illness. However, after the 4th day, the NS1 Ag Strip (Biorad) showed 89.0% of sensitivity up to the 7th of symptoms. From day five to the 7th, the pan-E Early ELISA (Panbio) confirms about 60.0% of the cases (Figure 1). Although dengue NS1 antigen detections up to the 9th day are observed, here we plotted cases only up to the 7th day due to the low number of samples representing 8th and 9th days in our population.
We also aimed to compare the cases confirmation by the dengue NS1 antigen capture to the confirmation by other methodologies used in this study according to the number of the days of illness. In this comparison, we considered a NS1 positive case, as a case positive in any of the three tests used. Figure 2 shows NS1 confirmation around 90% of the cases up to the 7th day of illness, as previously shown. RT-PCR and virus isolation detections rate were around 80% in the first three days of illness, decreasing after that. However, on the other hand IgM detection rates increase only after the 4th day of illness.
The serologic response could be characterized by IgG-ELISA in 54 samples, where a second specimen was available. There were 40 primary and 14 secondary infections. No differences were observed by the pan-E Early ELISA (Panbio) (P = 0.96) and by the NS1 Ag Strip (Biorad) (P = 0.76) in confirming primary and secondary infections (Table 3). However, the Platelia™ NS1 test showed a higher sensitivity in confirming primary infections than secondary ones (P = 0.01).
In our study, the pan-E Early ELISA test (Panbio) was less efficient in detecting acute dengue infections (86.1%) when compared to the Platelia™ NS1 test (91.3%) and the NS1 Ag Strip (95.0%). Positive predictive values were 98.3%, 99.5% and 100% for the Platelia™ NS1 (Biorad), NS1 Ag Strip (Biorad) and pan-E Early ELISA tests (Panbio), respectively. However, the pan-E Early ELISA (Panbio) showed the lowest negative predictive value (78.3%), followed by the Platelia™ NS1 test (Biorad) with 86.3% and the NS1 Ag Strip (Biorad) with 91.1% (Table 4).
The techniques of dengue serologic diagnosis which have been widely used are based on the detection of IgM antibodies by MAC-ELISA and IgG by IgG-ELISA. However, one of the limitations of these techniques is the inability to detect antibodies to DENV in the acute phase of disease [5], [26]. It takes 3 to 5 days for IgM antibodies and anti-DENV 10–14 days for IgG anti-DENV to become detectable. Moreover, primary and secondary infections have different profiles of production of these antibodies [27].
According to previous studies the presence of NS1 in human sera can be confirmed between days 0 to 9 [28], [29], [30] and with a peak at days 6 to 10 [31]. Currently, commercial kits such as the Dengue EARLY ELISA (Panbio Diagnostics, Brisbane, Australia), Platelia™ Dengue NS1 Ag-ELISA and Dengue NS1 Ag STRIP (BioRad Laboratories Marnes La Coquette, France) are available for early diagnosis of dengue based on NS1 antigen capture and several studies have been conducted in many laboratories [12], [15], [18], [19], [29], [31], [32], [33], [34], [35], [36].
In this study, we had the opportunity to evaluate and compare three NS1 antigen capture kits available with a panel of samples (n = 450) from cases occurred since the introduction of dengue in Rio de Janeiro, Brazil in 1986 to 2008. The NS1 Ag Strip test (Biorad) was the most sensitive in confirming dengue cases, followed by Platelia™ NS1 (BioRad). The least sensitive was the pan-E Early ELISA (PanBio) with 72.3% of sensitivity. However, in this study PanBio kit was the most specific (100%) while both kits from BioRad showed 98.7% and 99.1% of specificity, respectively. A recent evaluation in Malaysia showed that the NS1 Ag Strip had 90.4% of sensitivity and 99.5% of specificity [17]. Studies performed in Vietnam [18] and French Guyana [29] showed sensitivities of 82% and 88%, respectively for the Platelia™ NS1 test. However, sensitivities varying from 63.2% to 93.3% have also been reported for this kit [12], [15]. Even though different DENV genotypes may circulate in the Americas and Asia, NS1 kits evaluations in countries from those area show the ability of those tests to detect DENV in infected patients. Our observations are consistent with previous studies in which the pan-E Early ELISA had lower sensitivities [13], [14], [16], [37]. However, to increase diagnostic performance, Panbio has recently released an improved second generation for their NS1 capture kit with changes in key reagents and procedure [38].
All NS1 tests were more sensitive in confirming cases positive by virus isolation than in cases positive by RT-PCR. Dussart [29] confirmed 94.1% of cases positive by virus isolation and 85% of the cases RT-PCR positive using the Platelia™ NS1 test. Recently, McBride [16] showed that the NS1 antigen capture was positive in 87% of the cases positive by RT-PCR. In our study, the Dengue NS1 Ag Strip confirmed 98.7% of the cases positive by virus isolation and 82.3% of RT-PCR positive cases, results similarly observed by Zainah [17]. In the presence of IgM antibodies, the Dengue NS1 Ag Strip confirmed more cases (77.4%) than the pan-E Early ELISA (69.4%) and the Platelia™ NS1 (64.5%). The presence or absence of IgM did not influence in the cases confirmation by the pan-E Early ELISA (P = 0,6159). However, the higher confirmation by both Platelia™ NS1 and the Dengue NS1 Ag Strip in the absence of IgM were statistically significant. Sekaran [32] showed that the NS1 detection rates decrease as IgM levels rise, in agreement with our results.
The pan-E Early ELISA (PanBio) showed a higher sensitivity in confirming DENV-2 infections and the Platelia™ NS1 kit (BioRad) in DENV-1 infections. However, the Dengue NS1 Ag Strip kit (BioRad) showed the same sensitivity in confirming DENV-1 and DENV-2 infections. DENV-3 infections were the least confirmed by all three kits. The apparent inability in confirming infection by this serotype has been shown previously [33]. Furthermore, differences in the inter-serotype sensitivities have been reported for all three kits. McBride [16] recently showed lower sensitivities by the pan-E Early ELISA (PanBio) in DENV-2 and DENV-4 infections. The latter was also found in previous studies performed by Bessoff [37] and Dussart [14] and most recently in a study performed in Venezuela [36]. Due to the absence of DENV-4 circulating in Brazil, we were not able to access the assays sensitivities in cases infected by this serotype. Both Biorad kits (Platelia™ NS1 and Dengue NS1 Ag Strip) showed a lower sensitivity in DENV-2 infections from Vietnam [18] and Venezuela [36].
A higher detection rate by the three tests was found during the first four days after the onset of the symptoms. Although dengue NS1 antigen detections up to the 9th day are described, here we analyzed cases only up to the 7th day due to the low number of samples representing 8th and 9th days in our population. The lack of later samples in this study did not allow us to determine when NS1 detection would decrease. However, previous studies found NS1 antigen in 82% to 83% of patients with dengue from day 1 to 9th after the onset of fever [11], [30].
The Platelia™ NS1 test showed a higher sensitivity in confirming primary infections than secondary ones, as previously observed [12], [15], [17], [18], [32], [34]. False negative results by NS1 antigen capture in secondary infections may also be due to the immune-complexes formation by the anti- DENV IgG sequestration [39]. Efforts to dissociate immune complexes by acid treatment can enhance the assays sensitivities, as previously shown [15]. However, in our study no attempts were made to dissociate those complexes. To further analyze the sensitivity of those tests in confirming secondary cases, a larger number of cases should be tested.
Among the kits evaluated, the Dengue NS1 Ag Strip (BioRad) was the most efficient in confirming dengue infections by capturing NS1 antigen from infected patients. Moreover, it was more convenient to be used, as the results can be obtained in 15 minutes, easy to perform and its performance does not involve the use of special laboratory equipment.
Previous studies have demonstrated a diagnostic strategy combining NS1 Ag detection in acute-phase sera and DENV IgM detection in early-convalescent-phase sera, providing a sensitivity of about 90% for dengue diagnosis [29], [34].
In conclusion, this evaluation has shown that NS1 antigen capture assays are indeed an alternative tool for the early diagnosis of dengue infections, may be used as a screening test prior virus isolation and used in combination with IgM capture can increase the rate of cases confirmation, especially in endemic areas where secondary infections are expected to occur due to the co-circulation of the different DENV serotypes, such as seen in Brazil.
This evaluation was performed for research purposes only and authors have no financial interest. The pan-E Early ELISA from Panbio and the Dengue NS1 Ag Strip from BioRad were kindly provided for evaluation.
|
10.1371/journal.pntd.0007685 | The division of labour between community medicine distributors influences the reach of mass drug administration: A cross-sectional study in rural Uganda | Despite decades of community-based mass drug administration (MDA) for neglected tropical diseases, it remains an open question as to what constitutes the best combination of community medicine distributors (CMDs) for achieving high (>65%/75%) treatment rates within a village.
Routine community-based MDA was evaluated in Mayuge District, Uganda. For one month, we tracked 6,148 individuals aged 1+ years in 1,118 households from 28 villages. Praziquantel, albendazole, and ivermectin were distributed to treat Schistosoma mansoni, lymphatic filariasis, and soil-transmitted helminths. The similarity/diversity between CMDs was observed and used to predict the division of labour and overall village treatment rates. The division of labour was calculated by dividing the lowest treatment rate by the highest treatment rate achieved by two CMDs within a village. CMD similarity was measured for 16 characteristics including friendship network overlap, demographic and socioeconomic factors, methods of CMD selection, and years as CMD. Relevant variables for MDA outcomes were selected through least absolute shrinkage and selection operators with leave-one-out cross validation. Final models were run with ordinary least squares regression and robust standard errors.
The percentage of individuals treated with at least one drug varied across villages from 2.79–89.74%. The only significant predictor (p-value<0.05) of village treatment rates was the division of labour. The estimated difference between a perfectly equal (a 50–50 split of individuals treated) and unequal (one CMD treating no one) division of labour was 39.69%. A direct tie (close friendship) between CMDs was associated with a nearly twofold more equitable distribution of labour when compared to CMDs without a direct tie.
An equitable distribution of labour between CMDs may be essential for achieving treatment targets of 65%/75% within community-based MDA. To improve the effectiveness of CMDs, national programmes should explore interventions that seek to facilitate communication, friendship, and equal partnership between CMDs.
| Community-based mass drug administration (MDA) uses volunteers within at-risk communities to distribute preventive chemotherapies en masse for neglected tropical diseases. Treatment rates achieved by community medicine distributors (CMDs) vary widely and can undermine morbidity control. We studied routine community-based MDA in 28 villages near Lake Victoria in Uganda. There were two CMDs per village who were tasked with moving from home-to-home to administer praziquantel, albendazole, and ivermectin for schistosomiasis, lymphatic filariasis, and soil-transmitted helminths. We observed treatment outcomes for 6,148 eligible individuals aged 1+ years. Here we identified the best combination of CMD characteristics for achieving high village-level treatment rates. We found that a more equal division of labour (e.g. 50–50 split between how many people each CMD treated) was associated with higher treatment rates when compared to CMDs with an unequal division of labour (e.g. one CMD treating no one). CMDs who were friends were more likely to have a division of labour that was nearly twofold more equal than CMDs who were not friends. The similarity of CMDs with respect to network, demographic, or socioeconomic characteristics did not influence village treatment rates. National programmes should explore interventions that seek to facilitate communication, friendship, and equal partnership between CMDs.
| For public interventions, the first-informed individuals influence the spread of information and uptake within the target population [1]. Understanding who should be the first-informed individuals or the deliverers of an intervention is a widespread challenge for any area of public policy, but in particular for global health programmes [2–4]. Little is known about how best to introduce and to maintain global health programmes in resource-poor settings where access to formal medical care and health-seeking behaviours are limited. Effective global health programmes rely on identifying the appropriate starting points for an intervention, e.g. who should deliver drugs, who should be treated first, and who should act as health promoters.
One successful and extensively used model for identifying the starting points for global health programmes is mass drug administration (MDA) [5]. MDA is the blanket, diagnosis-free distribution of single dose preventive chemotherapies to individuals within at-risk areas for neglected tropical diseases (NTDs) [6]. The frequency and implementation of MDA is determined by the prevalence of infection within a geographical catchment, school, or community and varies by disease [6]. Several methods of MDA implementation exist, utilizing communities, primary schools, or child health days. The most common method of implementation is through community-based MDA, which is used to treat schistosomiasis, lymphatic filariasis, trachoma, and onchocerciasis with some communities also benefiting from the treatment of soil-transmitted helminths (STHs) because of coendemicity with lymphatic filariasis. To promote local ownership of MDA, national programmes instruct individuals within NTD-endemic areas to select local community medicine distributors (CMDs) through open, community-wide meetings [7]. CMDs serve as volunteers, apart from the reimbursement for travel costs to attend annual training sessions, and are tasked with either moving from home-to-home (e.g. schistosomiasis) [8] or with mobilizing individuals to retrieve drugs from a central post (e.g. lymphatic filariasis) [9]. Progress towards NTD control, including community-based MDA, has been proposed as a platform for measuring access to universal health coverage [10]. In 2017, nearly 1/3rd of school-aged children, who are included in the World Health Organization (WHO) Roadmap for NTDs [11] and require preventive chemotherapies for schistosomiasis or STHs remained untreated [12]. For example, after 10 years of community-based MDA in Mayuge District, Uganda, CMDs treated only 56.66% of eligible individuals with at least one drug for schistosomiasis, lymphatic filariasis, or STHs [13]. Therefore, a better understanding is needed of how to increase the effectiveness of CMDs.
The context in which CMDs have been studied in order to improve MDA includes 1) how best to alleviate the opportunity costs of time volunteered [14, 15], 2) how to reduce capacity constraints resulting from a limited number of CMDs [14, 16, 17], 3) the impact of financial or in-kind incentives for CMDs [18], 4) the role of knowledge, attitudes, and practice as well as available health system support in promoting CMD motivation [19], 5) the social biases that manifest in the CMD’s decision on whom to treat [13, 20], and 6) the personal characteristics of CMDs that determine their performance during MDA [21]. Despite the wide variation of treatment rates across communities [4], there is a limited understanding from both national MDA programmes and communities of how and whether CMDs should be replaced before they choose to resign, and specifically of what combination of CMDs is best for achieving the highest treatment rates. Two aspects of CMD selection have been studied and associated with increased treatment rates: exploiting local social network structures to choose well-placed CMDs [4, 22] and including CMDs with diverse kinship affiliations so that a CMD treats only individuals with a shared clan membership [16, 17]. The kinship studies [16, 17] identify similarity between CMDs and MDA recipients rather than measure the similarity between CMDs, and do not consider network or socioeconomic similarity. Thus, it remains an open question as to if/how the similarity of CMDs affects MDA outcomes.
To identify the best combination of CMDs, there is a need to understand how network, demographic, and socioeconomic similarity between CMDs affects their performance during MDA. Similarity has been widely shown elsewhere to determine peer effects, i.e. how one person influences another person (either directly or indirectly) [23–28]. Yet, how CMDs influence one another or how shared CMD affiliations affect MDA, to our knowledge, has not been studied. Here we conduct the first analysis of CMD similarity by comparing the networks and personal attributes of CMDs to identify what combination of CMDs best facilitates the reach of MDA. Moreover, to further delve into peer effects, we provide the first study of how CMD similarity influences the division of labour between CMDs. We answer the following question. How does CMD similarity affect the division of labour and treatment rates achieved during MDA?
This study was reviewed and approved by the Uganda National Council of Science and Technology (SS4077), and the University of Cambridge School of Humanities and Social Sciences (HSSREC2016.6). Written informed consent was obtained from all respondents. For respondents who indicated they were unable to write or who preferred to provide fingerprints, verbal informed consent and a fingerprint signature were obtained.
Using methods described and validated in Chami et al. [4, 20], routine community-based MDA was tracked in 31 villages in Mayuge District, Uganda from mid-July to mid-August 2016. The study area predominantly comprises fishing villages along Lake Victoria, which are hyperendemic (>50% prevalence) with Schistosoma mansoni [29]. To remove administrative barriers that may delay the start of MDA, researchers provided local District Vector Control Officers—the individuals responsible for routinely training CMDs—with cars to start MDA within three days in July for all study villages. Study surveys were conducted after one month of MDA. Preventive chemotherapies were only available from the community-based MDA programmes during the study period. Two CMDs were tasked with approaching all households, i.e. moving door-to-door, and administering preventive chemotherapies. There were no limits on treatment rates achievable by CMDs due to insufficient medicine supplies. Researchers provided the Vector Control Officers with enough pills/tablets for all CMDs to treat all eligible individuals within their villages. Survey teams conducted surprise checks of CMD homes after the one-month MDA tracking and pills/tablets for all medicines remained with all CMDs. Praziquantel was distributed to treat school-aged children and adults (individuals aged 5+ years) for potential infections with S. mansoni. Albendazole and ivermectin were administered to treat school-aged children and adults (all individuals aged 5+ years old) for potential lymphatic filariasis infections. Due to hookworm endemicity, albendazole was provided to pre-school aged children, school-aged children and adults (all individuals aged 1+ years old), although albendazole was not donated for treating hookworm through community-based MDA [29]. The most common method of MDA implementation for schistosomiasis and STHs in Uganda is the distribution of medicines through primary schools, i.e. excluding adults for treatment and using schoolteachers as distributors instead of CMDs [8]. The high prevalence of S. mansoni infections and the endemicity of lymphatic filariasis enabled community-wide treatment for schistosomiasis and STHs, respectively. When lymphatic filariasis treatment stops in our study area then community-wide treatment may stop for STHs.
MDA was community-based as opposed to community-directed in that communities did not lead the design of MDA, which was completed by the national programmes. Communities selected CMDs, but did not choose the dates, time period, or method of distribution for MDA. Communities also were not formally involved in the monitoring of CMDs, which was the task of the District Vector Control Officers. The national MDA programme instructed communities to select systematically two CMDs through a community-wide meeting and to choose individuals who were literate and able to fill in NTD registers. Communities also were encouraged to have gender balance between CMDs, i.e. one female and one male CMD per village. No other instructions for the selection or replacement of CMDs were provided by the national MDA programme. Communities did not strictly follow national recommendations. Village leaders (local government members or village health team members) directly selected more than half of the CMDs instead of holding community-wide meetings [21].
Systematic random sampling of households was conducted [21]. Village registers of households—ordered by year of settlement—were used to select 40 households per study village. Household heads and lead wives were interviewed to provide information on all members of the household aged 1+ years—the minimum criteria for MDA eligibility. In addition to the systematic random sampling, all CMDs and their household heads were interviewed. Households of CMDs only were included in the calculation of treatment outcomes if selected by chance through the systematic random sampling.
Using a structured questionnaire [21], two sets of treatment outcomes were examined for participants who were selected through systematic random sampling: village treatment rates and the division of labour between CMDs. Village treatment rates comprised the overall level of treatment within a village, i.e. the work of both CMDs, and were calculated at both the individual and household levels. Treatment responses were recorded by an independent team of surveyors who conducted surprise visits to villages after one month of undisturbed MDA, as described in Chami et al. [4, 20]. At the individual level, treatment rates were measured as the percentage of eligible individuals who were offered and had ingested at least one drug of praziquantel, albendazole, or ivermectin. This indicator most closely aligns with the WHO’s indicator of surveyed coverage [4]. We used a conservative measure of treatment with at least one drug to reduce the dimensionality of the analysis (number of models run) and to account for endogeneity that arises with individual drug outcomes, i.e. drug-specific treatment rates are strongly positively correlated [21]. At the household level, treatment rates were measured as the percentage of households with at least one eligible person who was offered and had ingested at least one drug of praziquantel, albendazole, or ivermectin. Household-level treatment rates were of interest as the Uganda MDA programme instructed CMDs to move from home-to-home within our study area and treatment rates for individuals are strongly positively correlated within a home [20]. Also, household-level treatment rates represent the percentage of homes approached by CMDs [4]. WHO disease-specific treatment targets include treatment of 75% of eligible individuals with praziquantel for schistosomiasis and albendazole for hookworm, and 65% of eligible individuals with albendazole plus ivermectin for lymphatic filariasis. As a note, treatment outcomes were determined by drug delivery efforts from CMDs since only less than 1% of MDA recipients refused to ingest offered medicines [21]. Hence, in our study, the offer and ingestion of medicine (treatment) also can be thought of as indicative of evidence of contact with CMDs.
We undertook a proof-of-principle investigation into the relevance of the division of labour for predicting village treatment rates, which are used to assess progress towards WHO treatment targets [30]. To develop the first measure of the division of labour for CMDs, we sought a simple indicator that 1) did not interfere with routine MDA, 2) captured the primary objective of MDA, i.e. maximizing the number of people treated, and 3) could be applied in various geographical or social contexts. In this respect, the division of labour was outcome-based and focused on the number of people treated by each CMD. Importantly, neither the national MDA programme nor local health facilities provided instructions to CMDs for dividing labour. The national MDA programme only indicated to CMDs that they should treat all eligible individuals within their village. Hence, CMDs were not pre-allocated households or geographical areas of a village. CMDs did not hold discussions with their communities about the division of labour. Consequently, we assumed here that how best to divide labour to meet programmatic goals was the sole decision of CMDs.
The entire village was considered for assessing the labour of each CMD. In Uganda, the village is the lowest administrative unit; there are no further formal subdivisions that could have been exploited for the division of labour. Moreover, no intra-community spatial divisions were considered due to the small size of study villages and previously shown irrelevance of the number of homes for explaining village treatment rates in our study area [4]. On average, there were only 238 homes per village (range 87–535 homes). Concerning spatial aspects, the village ecology, such as the number of roads or swamps, has been shown to be uninformative for MDA in our study area [4]. The spatial spread/diameter of the study villages also has been shown to be unrelated to village treatment rates [4]. The furthest distance in metres between two homes has been shown to be on average only 1.26 km (std. dev. 428.29 m) [4]. The mean distance between two homes within our study villages has been measured at 400.11 m (std. dev. 142.27 m) [4].
For the division of labour, the percentage of eligible individuals or eligible households treated by each CMD was examined. For drugs offered to eligible individuals, the household respondent provided the name of the CMD who offered treatment. All respondents knew who treated whom. Few eligible individuals (<12%) were offered treatment by both CMDs, which included either CMDs separately approaching the same individual or both CMDs being present at the same time to treat the same individual. More detailed methods on the calculation/attribution of individual CMD treatment rates are provided in Chami et al. [21].‘Treated’ was defined as described for the village-level treatment rates. The division of labour was calculated as a ratio of treatment rates for the two CMDs in each village. For the two CMDs, the lowest treatment rate was divided by the highest treatment rate to create a normalized village-level outcome. The division of labour was an indicator from 0–1 where 0 was a perfectly unequal division of labour (one CMD treated no one) and 1 was a perfectly equal division of labour (CMD treatment rates were equal). There were no villages where both CMDs treated no one. The division of labour was calculated for treatment rates at both the individual and household levels. Although MDA consists of separate tasks such as registering households, sensitizing individuals, and mobilizing the community, CMDs in our study area perform these tasks whilst they treat individuals [4, 13, 21]. Thus, in our study context, the division of labour for treatment outcomes also represents the division of labour dedicated to diverse MDA tasks.
CMDs were interviewed and asked to provide the names of their close friends, using the following structured prompt [4, 13, 22].
The individuals named as close friends by CMDs also were interviewed. The friends of CMDs were provided with a list of names, which included all individuals who were named by both CMDs as well as the names of the CMDs. Friends of CMDs were then asked to indicate with whom they had close friendships. Hence, CMDs could belong to the same network component, i.e. a path could exist between the two CMDs, due to either a friendship between CMDs or a friendship between the friends of CMDs. Moreover, a CMD could have more than 10 ties due to the friends of the other CMD naming the CMD of interest. All friendship networks were analyzed as undirected; if an individual was named or had named someone then there was a tie between those two individuals.
Sixteen indicators of similarity between the two CMDs in each village were examined, which included three network characteristics and 13 personal attributes. For network similarity (structural equivalence), three variables that captured both direct and indirect ties were calculated using NetworkX in Python v2.7 [31]. A direct tie was a binary indicator of friendship between CMDs. The Jaccard index captured indirect ties between CMDs and the similarity of their network neighbourhood, i.e. common friends. It was calculated as the number of common friends divided by the total number of friends across both CMDs. To gain insight into the cohesion between CMDs and to account for the fact that influence between CMDs may travel further than two network steps (beyond common neighbours) [27, 32], the minimum node cut was calculated. The minimum number of nodes (friends) that would need to be removed from the network to disconnect CMDs, i.e. to remove all paths between CMDs, was counted then normalized by dividing by the total number of friends for both CMDs. A comparison of direct ties versus indirect ties was of interest to understand by what means could peer effects occur between CMDs [24]. With direct ties, peer effects occur due to an existing channel of communication between two individuals [28]. Alternatively, indirect ties have been shown to influence two individuals of interest through either competition or comparison [25]. For competitive influences, it has been shown that two individuals vie for the attention of the same friends (due to the overlapping friendship group) and this competition is what drives similar behaviours [25, 26]. Alternatively, indirect ties may signal shared friendships that are used as a reference for behaviours, i.e. an individual compares themself to their group of friends and two individuals with shared friends will compare themselves to the same group [4, 26, 27].
CMDs were interviewed using a structured questionnaire [21] in order to observe 13 personal attributes. The method of CMD selection was recorded as a categorical variable and included nominations from a community-wide meeting, the village health team, or a local council (village government) member. The total number of years as a CMD was noted. Eleven demographic and socioeconomic characteristics were observed. Age was rounded to the nearest year. Gender was a binary variable and equal to one if the CMD was female. Education was measured as a categorical variable to represent the highest level of education attained and included primary school, secondary school, or post-secondary school diplomas. Binary indicators for majority tribe and religion were equal to one if the CMD belonged to the majority tribe or religion of their village, respectively. Occupation was a categorical variable that included values for farmer, fisherman/fishmonger, and ‘other’ jobs; occupation was coded to capture the main occupations in the study area [29]. Formal status was a binary indicator that was equal to one if the CMD was a religious/tribe/clan leader, on the beach management committee, or a member of the local council (village government). Two binary indicators of preventative health behavior were measured using WHO and United Nations International Children’s Emergency Fund (UNICEF) guidelines [33]. A CMD belonged to a household that used a protected water source if drinking water was retrieved from piped water, village taps, boreholes, or protected wells. Private home latrine ownership included only covered pit latrines with privacy. Home quality score was the total sum of scores (min. 3, max. 12) for the floor, walls, and roof materials (four for each category, ranked 1–4) [13]. The ‘years in village’ was a count of the total years since the CMD’s household had settled in the current village.
To calculate attribute similarity between CMDs, binary indicators were constructed for all 13 CMD attributes, including MDA-related variables, and equal to one if CMDs were similar [34]. For all binary or categorical variables, if CMDs shared the same value/category then the attribute indicator was equal to one. For the number of years settled in the village, CMDs were coded as similar if their years of settlement were within +5/-5. For years as CMD, age, and home quality score variables, CMDs were classified as similar if their values were within +3/-3. In addition to the CMD similarity indicators, we accounted for variation in village and network size [4, 27]. The natural log of total homes in the village and the natural log of the average CMD degree (total friendship ties for each CMD) were calculated.
Statistical analyses were completed at the village level and conducted in R v3.2.3 and Stata v13.1. With a limited number of village observations and no previous work on CMD similarity, we employed an unsupervised approach. Leave-one-out-cross-validation (LOOCV) with least absolute shrinkage and selection operators (LASSO) [35, 36] were used to select the predictors of the division of labour and village treatment rates. This approach is a commonly used method for dimension reduction in statistical analyses. LOOCV LASSO was run with simple ordinary least squares (OLS) [35, 36]. For the selection of predictors for the division of labour, all 16 CMD variables as well as the village and network sizes described in the previous section were candidates. In addition to these 18 variables, the division of labour was included for the selection of predictors of village treatment rates. The predictors that were selected through LOOCV LASSO were then entered in OLS regressions with robust standard errors [37]. To test for potential endogeneity of the division of labour and village treatment rates, i.e. an incorrectly specified direction of association where village treatment rates may determine the division of labour or more generally an association of the two outcome equations through the error terms, a Durbin-Wu-Hausman test was conducted and seemingly unrelated regressions were run [38]. For the Durbin-Wu-Hausman test, no evidence was found to indicate that the two outcome equations were correlated (F-stat = 2.73, p-value = 0.111 for individual-level outcomes and F-stat = 0.74, p-value = 0.399 for household-level outcomes). Similarly, no support for simultaneous equations was found from the seemingly unrelated regressions (Chi2 = 0.498, p-value = 0.481 for individual-level outcomes, and Chi2 = 0.602, p-value = 0.438 for household-level outcomes). Thus, separate OLS regressions were run. LOOCV was run for both the selection of the predictors and for the final models.
For the statistical analyses, three villages (IDs 20, 24, 30) were not included because one CMD in each of those villages was missing network information. Thus, 56 CMDs from 28 villages had complete data. For the target population, two households were excluded due to having no eligible individuals for MDA or missing information regarding treatment. In total, 1,118 households and 6,148 eligible individuals within those households were observed.
In 28 villages, 47.87% (2943/6148) and 24.77% (1523/6148) of eligible individuals were treated with at least one drug and all three drugs, respectively. Only 56.71% (634/1118) of households had at least one eligible person treated with praziquantel, albendazole, or ivermectin. Treatment rates achieved by individual CMDs ranged from 0–84.25% (std. dev. 22.09%) and 0–87.50% (std. dev. 23.49%) for individuals and households, respectively (Obs. 56). Village treatment rates also varied widely within the study area. The percentage of eligible individuals treated in each village varied from 2.79–89.74% (std. dev. 26.15%). Similarly, the percentage of households with at least one eligible person treated ranged from 7.50–97.50% (std. dev. 25.10%). WHO treatment targets for each disease were not necessarily met when village treatment rates (as measured here) met WHO-recommended levels. For schistosomiasis and hookworm, five communities had village treatment rates of at least 75% (Village IDs 1,14,18,21,31) and three communities had praziquantel and albendazole treatment rates of at least 75% (Village IDs 1,18,31). For lymphatic filariasis, 10 communities (Village IDs 1,2,12,14,17,18,21,25,26,31) had village treatment rates of at least 65% yet only three communities had both albendazole and ivermectin treatment rates of at least 65% (Village IDs 1, 18, 31). Therefore, only 10.71% (3/28) of villages met WHO-recommended treatment targets for each disease.
The division of labour across villages was highly unequal. The average division of labour for both the percentage of individuals and households treated was 0.327 (std. dev. 0.277 for individuals and 0.285 for households). In other words, when two CMDs were compared within the same village, one CMD treated on average only one third as many people or households as their counterpart. Wide variation in the division of labour was observed. The divisions of labour for individual and household level outcomes ranged from perfectly unequal to nearly a perfectly equal 50–50 split (range 0–0.967, std. dev. 0.278 for individuals; and 0–0.957, std. dev. 0.285, for households). There were six villages with one CMD who treated no one. Despite the wide variation in the division of labour, high inequality between CMDs was most common. For example, for the percentage of individuals treated, 75% of villages had a division of labour between CMDs where one CMD treated twice as many people as the other CMD. The unequal division of labour was not due to both CMDs treating few people, i.e. one CMD treating marginally more individuals (e.g. 10% versus 5%). The average absolute difference in the percentage of eligible individuals treated between CMDs within a village was 30.23% (std. dev. 20.85%).
A summary of personal attributes of CMDs, similarities between CMDs, and village sizes are presented in Tables 1–3. Within the study area, there was no single characteristic that was shared by all CMDs. When two CMDs within the same village were compared by personal/observable characteristics, CMDs were most often similar with respect to preventative health behaviours and socioeconomic status. A large majority (81.14%) of villages had two CMDs with the same ownership status of private home latrines; in all but one of these villages (22/23) both CMDs owned a private home latrine. In 75.00% of villages, both CMDs had the same formal status (where 4/21 villages had both CMDs with formal status) and similar home quality scores. Approximately 50% or more of villages had two CMDs who differed with respect to gender, educational attainment, membership in the majority tribe, and the number of years settled in the village. There were 25.00% (7/28) and 17.86% (5/28) of villages, respectively, where CMDs were either both females or both males. With respect to MDA-related characteristics, only 53.57% of villages had CMDs who were selected through the same means and only 42.86% of villages had two CMDs who had volunteered for MDA for a similar number of years. CMDs selected in the same manner were not necessarily selected through community-wide meetings. Only 40.00% (6/15) of villages who selected both CMDs in the same manner did so through community-wide meetings whilst the remainder of those villages had both CMDs selected by a member of the local council (village government).
Network similarity between CMDs is illustrated in Fig 1 and summarized in Table 3. On average, each CMD had 8.43 friends. CMDs were close friends in only 5 of 28 villages (17.86%). Yet, CMDs did not belong to distinct friendship groups. Amongst the total number of friends named by both CMDs, an average of 84.40% of friends were shared between the two CMDs. Moreover, in every village, each CMD was no further than two steps apart, i.e. each CMD had at least one common friend. Over an average of seven friends had to be removed from the network to completely disconnect CMDs. The cohesiveness of CMDs was maintained through indirect ties because the friends of CMDs also were well connected. When both CMDs were removed from the network, density remained high amongst the friends. Here density is defined as the proportion of ties that exist amongst the maximum possible number of ties. The density amongst friends of CMDs was on average 0.784 (std. dev. 0.135, range 0.526–1). There was only one village (ID 28) where the removal of CMDs resulted in one friend becoming isolated from the network.
Table 4 presents the determinants of the percentage of individuals treated at the village level. Neither CMD similarity nor the sizes of the friendship networks and villages were associated with village treatment rates. Only one variable—the division of labour—was selected through LOOCV LASSO as a potential predictor of village treatment rates. The division of labour was positively correlated (p-value = 0.008) with village treatment rates. For the percentage of eligible individuals treated at the village level, there was a remarkable absolute difference of 39.69% between the treatment rates of CMDs with a perfect division of labour compared to CMDs with a perfectly unequal division of labour. The predicted village treatment rates against the range of values for the division of labour, i.e. the marginal effects of increasing equity in the division of labour, are shown in Fig 2. When the percentage of households treated was examined, the results for the division of labour were upheld despite LOOCV LASSO selecting two predictors in addition to the division of labour (Table 5). The discrepancies amongst village treatment rates at the household level for CMDs with and without equal divisions of labour were as large as 50.51% (p-value = 0.006). The additional predictors of the percentage of households treated included the similarity in the number of years spent as CMD and a village-level variable of the total homes. The total number of homes in the village was negatively related to village treatment rates at the household level (p-value = 0.001), although this effect was modest. A 10% increase in the total number of homes in a village was estimated to decrease the percentage of households treated by only 1.86%.
Tables 6 and 7 present the predictors of the division of labour between CMDs. For both individual and household level outcomes, the only predictor selected by LOOCV LASSO was the friendship between CMDs. A direct tie between CMDs was predicted to substantially increase the division of labour by 0.444 and 0.393, respectively, at the individual or household level when compared to CMDs without a direct tie. Hence, workload equity between CMDs was estimated to increase by just under twofold. Notably, the friendship between CMDs only predicted the division of labour and was not associated with village-level treatment rates (Tables 4 and 5, and ρ = 0.144, p-value = 0.464 at the individual level; and ρ = 0.098, p-value = 0.619 at the household level). There were no missed effects of other characteristics of CMD similarity influencing the division of labour due to indirectly affecting the presence of a direct tie (Table 8).
In the context of rapidly expanding community-based MDA and WHO disease-specific goals of elimination [39–41], there is an urgent need to increase the effectiveness of CMDs. In accord with previous work [4], here we showed that treatment rates varied widely across villages in rural Uganda. The percentage of eligible individuals treated varied from 2.79–89.74% across 28 villages and worsened from an average of 59.05% per village in 2013 [4] to 47.77% in 2016 (our study). To gain a better understanding of how CMDs work and cooperate, we examined the influences of CMD similarity on the division of labour and village treatment rates.
The division of labour was highly unequal between CMDs, with one CMD treating on average only one third as many eligible individuals as the other CMD within the same village. The equality in the division of labour was positively associated with overall village treatment rates. The estimated difference between a perfectly equal (a 50–50 split of individuals treated) and unequal (one CMD treating no one) division of labour was remarkable. An equal division of labour was associated with the treatment of an additional 39.69% more of the eligible population or 50.51% more households approached when compared to an unequal division of labour. Considering that WHO-recommended treatment rates for effective morbidity control are 65% for lymphatic filariasis and 75% for schistosomiasis or STHs, our results suggest that treatment targets are not achievable without an equitable distribution of labour between CMDs. These findings also highlight that a discussion of CMD capacity constraints [14] is unfounded within villages where one CMD treats few or no people. It is unknown whether adding more volunteer CMDs would facilitate MDA or simply contribute to an even more unequal division of labour and idle labour. Efforts to reduce CMD attrition rates [42] may be counterproductive if the result is that poorly performing CMDs are retained [21]. National MDA programmes should focus on the quality rather than the numbers of CMDs. Characteristics that may be used to improve the selection of CMDs can be gleaned from hardworking CMDs—defined here as CMDs who treated many individuals. In our study area, hardworking CMDs have been shown to be individuals who engage in good preventative health behaviours, belong to high-risk groups for endemic NTDs (e.g. fishermen for schistosomiasis), are male, and have supportive friendship networks [4, 21]. Ultimately, the selection/replacement criteria for CMDs should align with factors that are of interest to NTD-endemic communities.
The only determinant of the division of labour was a direct tie, i.e. close friendship between CMDs. Friendship was neither indicative of contact between CMDs nor of who knew whom. CMDs belonged to villages that were small with respect to population size and geographical spread. The variation in village size (total homes and total population) was uncorrelated to the presence of a direct tie. This result might suggest that village size also was unrelated to the frequency of contact between CMDs, assuming that the presence of a direct tie was partially determined by the frequency of in-person contact. CMDs knew each other well; they were both selected by individuals within the same village, trained together annually for MDA, and shared many common friends. Hence, CMDs had similar social networks. Yet, few villages (5/28) had CMDs who themselves were friends. Therefore, improving the division of labour is not as trivial as introducing two CMDs. Communication between CMDs was essential to improving the equity in the distribution of work related to MDA [28]. Here we only examined the ties between CMDs within the same village. Future research is needed to understand how CMDs are connected across villages. National MDA programmes do not hold regular meetings to bring together CMDs apart from the annual training. Encouraging CMDs both within and perhaps across villages to meet more frequently—maybe monthly—to compare and submit data, collect additional MDA supplies (registers, medicines, etc.), and simply to socialize may lead to new friendship connections that could facilitate communication between CMDs.
CMD friendship was only associated with village treatment rates indirectly, i.e. through the division of labour. This finding accords with previous research that tracked MDA in our study area [4] and found that a friendship tie between CMDs was not directly correlated with village treatment rates. Surprisingly, similar CMDs were not more likely to be friends than CMDs with different attributes. Conventional wisdom on social networks [23] suggests that direct ties exist between individuals in part due to homophily, which is the tendency of individuals to connect with others most like themselves. However, here we showed that shared socioeconomic characteristics did not predict the presence of a direct tie between CMDs. The presence of a direct tie is one indicator of CMDs belonging to the same cluster within the broader village social network. Stifling of an intervention hypothetically could occur by trapping its spread (information or uptake) within a confined set of closely-knit, clustered individuals [43]. There was no support that the reach of MDA was stifled when two CMDs were within the same network cluster. In contrast, we found that implementing MDA with CMDs in the same network cluster indirectly was positively associated with village treatment rates by improving the division of labour. It remains an open question as to whether negative ties (active dislike) exist between CMDs who are not close friends. The positive effect of direct ties on the division of labour suggests that CMDs are complements rather than substitutes with respect to their labour input. Complementary inputs result in additive or multiplicative effects on cooperation and equality between CMDs whereas substitutes might suggest a crowding out effect of one CMD working harder thereby causing the other CMD to work less. We cannot rule out that CMDs planned their division of labour, perhaps trading off efforts where one CMD agrees to take on the majority of responsibility for MDA this year whilst the other CMD resumes duties in the following year. If CMDs negotiated the division of labour then we would expect a direct tie, which represents a channel of communication between CMDs, to be negatively related to village treatment rates. Yet, no such association was observed. There remains the possibility of a motivational imbalance between CMDs [19] that is unrelated to CMD similarity. Regardless of the reason for a highly unequal division of labour, this inequality undermined community-based MDA and was correlated with low village treatment rates.
Our definition of the division of labour captured a number of features relevant to our study area and similar contexts. There were two CMDs. MDA was conducted in rural, small villages that did not have further geographical sub-units. There were no MDA programme stipulations for if/how labour between CMDs should be divided. Importantly, our measurement of the division of labour was outcome-based, i.e. dependent on the number of people to be treated. This definition is in light of WHO treatment targets and thus, from the perspective of the agency rather than from the view of the community. To investigate the division of labour in other MDA contexts, there is a need to develop a typology of the division of labour that includes a range of options for dividing the population as well as methods for evaluating labour/effort from MDA volunteers.
A provisional typology of the division of labour might include, as studied here, the number of people to be treated or instead could be focused on tasks, community geography, or social groups. CMDs performed MDA tasks of registering households, sensitizing individuals, and mobilizing the community whilst administering treatment instead of before treatment as instructed by the national programmes [4, 13, 20]. Thus, in our study area, there was no distinction between tasks and treatment. However, it remains an open question as to if creating a distinction between tasks and treatment, and dividing by task increases CMD productivity. Community participation is needed to understand the value of each MDA task and to identify tasks that are conducted by CMDs and the wider community but not recognized by national MDA programmes. The identified tasks might be enumerated in a form that is used to monitor CMDs by both national programmes and NTD-endemic communities. Beyond MDA, CMDs are members of the village health team, which is responsible for a wide range of primary health care tasks including bed net distribution, the management of childhood illnesses, and individual referrals to government health services [44]. There is a need to better understand the process in which CMDs divide responsibilities for MDA tasks versus other primary health care activities. In our study, we did not have information on whether CMDs who treated no one during MDA were actively engaged in other primary health care interventions.
Although village size did not affect the division of labour in our study area [4], spatial dimensions, population size, and local ecology will need to be considered in urban or peri-urban settings. A system could be devised using local knowledge of ecological or administrative divisions to assign the population to be treated. Local knowledge is critical, as—anecdotally in our study district—government records of villages did not accord with existing villages. This misalignment was not due to inaccuracies in reporting but rather the quickly change shape of local boundaries as determined by the communities themselves.
Potential social groups that could be used to divide labour include kinships, friendship groups, gender, occupations, burial societies, saving cooperatives, tribes, clans, or religious associations. In our study area, both kinships [16] and friendship networks [4, 13] have been shown to affect treatment rates, whereas gender has not been a barrier to who treats whom [21]. Women and men are just as likely to treat individuals from another gender when compared to how many people they treat of the same gender [21]. We found no support that similarities/differences in demographic or socioeconomic characteristics affected the division of labour or village treatment rates. This result was surprising in that social imbalances did not lead to a highly unequal division of labour, i.e. there was no evidence that high status CMDs were free riding off the efforts of lower status CMDs [13]. Diversity between CMDs did not translate into higher village treatment rates [16]. Villages with CMDs who represented more social groups, assuming CMD attributes were indicative of social group memberships, did not achieve higher village treatment rates than villages with CMDs from the same social groups. Although CMD diversity did not translate into higher village treatment rates, future work should examine whether CMD diversity affects the treatment rates of marginalized, underrepresented populations [20]. The usefulness of social divisions for the treatment of marginalized populations will depend foremost on the homogeneity of the communities to be treated and personal attributes of CMDs. Homogenous populations will not have natural social divisions. Additional research is needed to 1) develop indicators for assessing the variation of homogeneity between NTD-endemic communities, 2) measure if/how homogeneity impacts treatment rates, 3) develop methods to ‘match’ CMDs with individuals to be treated using a wide range of socioeconomic characteristics, and 4) analyze how aligning CMDs with similar social groups affects their treatment rates.
Developing a typology for dividing labour is only the first step in fully evaluating CMD equity. A limitation of our study is that we did not directly assess the effort expended by CMDs. For example, we require an understanding of how difficult it is to complete a particular task or traverse different terrains. The number of hours spent per task has been enumerated elsewhere from the perception of CMDs [14]. If time constraints were determined by occupation or familial demands (related to age and gender) then we would expect differences in CMD attribute similarity to capture differences in CMD time constraints. In that case, our results might suggest that differences in time constraints between CMDs did not explain the division of labour. Additional studies are needed to assess differences in time spent on MDA versus other primary health care tasks and the effect on the division of labour and MDA treatment rates. All CMDs received the same remuneration for attending MDA training and no other payments from MDA programmes. However, CMDs who engage in other primary health care tasks may be remunerated and these additional sources of income might affect CMD effort during MDA. To identify indicators in addition to time and remuneration for evaluating effort, community expectations for CMDs also should be examined.
Improving the division of labour between CMDs is a social problem. The Declaration of Alma-Ata in 1978 emphasized the need to account for the social determinants of treatment and disease [45]. Our study emphasizes the importance of the message of Alma-Ata for community-based MDA. Social relations between CMDs should be improved and, to do so, community involvement in MDA must be increased. The shift in terminology from ‘community-directed’ MDA, which originated in the 1970s, to the now widely used ‘community-based’ MDA suggests a reduction in the role of NTD-endemic communities [46]. Reducing the role of communities from actively designing/monitoring treatment programmes (community-directed) to nominating CMDs (community-based) risks turning communities into passive actors during MDA. Even a slight deviation from community-based MDA towards community-directed MDA, whereby CMDs involve their close friends in monitoring or disseminating information, has been associated with increased treatment rates [21]. To move from passive to active community involvement, there is a need to reimagine the concept of equity as posed in this study. Here we examined the equity between CMDs in light of their ability to meet programmatic goals (treatment targets). Equity instead could be explored by examining to what extent CMDs and communities, or communities and the national programme are equal partners in designing and running MDA. An ultimate goal for MDA might be the integration into local health systems to empower communities to manage their own health [46].
To improve the effectiveness of community-based MDA, national programmes should facilitate the equal division of labour between CMDs. The similarity of personal attributes of CMDs was unrelated to the best combination of CMDs with the most equal division of labour and, in turn, highest treatment rates. National MDA programmes may instead aim to foster friendships between CMDs or to encourage the selection of CMDs who already are close friends in order to promote open communication between CMDs. Alternatively, guidelines might be trialed where an equal division of labour is encouraged between CMDs; treatment rates may be recorded for each CMD and monitored to identify inactive CMDs. Future interventions also might seek to explore other avenues to increase community involvement in the design of MDA. A more equal division of labour between CMDs may assist NTD programmes with achieving treatment targets.
|
10.1371/journal.pntd.0002775 | Structures of Trypanosoma brucei Methionyl-tRNA Synthetase with Urea-Based Inhibitors Provide Guidance for Drug Design against Sleeping Sickness | Methionyl-tRNA synthetase of Trypanosoma brucei (TbMetRS) is an important target in the development of new antitrypanosomal drugs. The enzyme is essential, highly flexible and displaying a large degree of changes in protein domains and binding pockets in the presence of substrate, product and inhibitors. Targeting this protein will benefit from a profound understanding of how its structure adapts to ligand binding. A series of urea-based inhibitors (UBIs) has been developed with IC50 values as low as 19 nM against the enzyme. The UBIs were shown to be orally available and permeable through the blood-brain barrier, and are therefore candidates for development of drugs for the treatment of late stage human African trypanosomiasis. Here, we expand the structural diversity of inhibitors from the previously reported collection and tested for their inhibitory effect on TbMetRS and on the growth of T. brucei cells. The binding modes and binding pockets of 14 UBIs are revealed by determination of their crystal structures in complex with TbMetRS at resolutions between 2.2 Å to 2.9 Å. The structures show binding of the UBIs through conformational selection, including occupancy of the enlarged methionine pocket and the auxiliary pocket. General principles underlying the affinity of UBIs for TbMetRS are derived from these structures, in particular the optimum way to fill the two binding pockets. The conserved auxiliary pocket might play a role in binding tRNA. In addition, a crystal structure of a ternary TbMetRS•inhibitor•AMPPCP complex indicates that the UBIs are not competing with ATP for binding, instead are interacting with ATP through hydrogen bond. This suggests a possibility that a general ‘ATP-engaging’ binding mode can be utilized for the design and development of inhibitors targeting tRNA synthetases of other disease-causing pathogen.
| Infection by the protozoan parasite Trypanosoma brucei causes sleeping sickness, also called human African trypanosomiasis. Without treatment, the disease is fatal yet current therapeutic options are inadequate and better medicines are needed. We have previously reported several potent inhibitors of T. brucei methionyl-tRNA synthetase, an essential enzyme involved in the protein biosynthesis. Recently, a new series of the inhibitors was synthesized which has improved membrane permeability over the earlier inhibitors. When applied to mouse with T. brucei infection, the new compounds are orally available and reach the central nervous system to reduce parasite loads, and therefore are promising molecules to be developed into antitrypanosomal drug. Here, more inhibitors from this series are reported and tested for their activities. High resolution crystal structures were determined that revealed how these inhibitors bind to the target enzyme. The binding pockets of these inhibitors are thoroughly explored, providing profound insights which are beneficial for further development of MetRS inhibitors against sleeping sickness. A ternary complex of the enzyme, an inhibitor, and an ATP analogue was also determined, indicates that the inhibitor does not compete with ATP for binding. Based on this, a general approach to use inhibitors that engage ATP for binding to tRNA synthetases is proposed.
| Human African trypanosomiasis (HAT), also called sleeping sickness, is a disease caused by the protozoan parasite Trypanosoma brucei. Up to 60 million people in sub-Saharan Africa are estimated to be at risk for the infection [1]. The disease usually occurs in two stages. In the first, haemolymphatic, stage, the parasites multiply in blood and lymph. In the second, meningoencephalitic, stage, the parasites cross the blood-brain barrier (BBB) to invade the central nervous system. Most of the reported cases of HAT are caused by T. brucei gambiense which progresses slowly in months to years. In contrast, HAT caused by T. brucei rhodesiense progresses very rapidly in weeks [2]. HAT is uniformly fatal if left untreated. However, currently available treatment options for HAT are largely inadequate mainly due to drug toxicity. All treatment regimens require parenterally administered drugs and only two (melarsoprol and eflornithine) cross the BBB for treatment of late stage HAT [2], [3]. Therefore, new oral antitrypanosomal drugs that are affordable, effective and safe are urgently needed. It is crucial that a new drug is orally available for ease of storage and administration, and crosses the BBB for effective treatment of the late stage of the disease.
The aminoacyl-tRNA synthetases (aaRS) are essential enzymes involved in protein synthesis and hence attractive targets for anti-infective drug design [4]–[6]. Generally, aaRS recognize a specific amino acid and charge it to its cognate tRNA through a two-step reaction: (1) recognition of the amino acid and ATP to form an aminoacyl-adenylate intermediate, and (2) recognition of the cognate tRNA to transfer the aminoacyl group to the 3′-terminal adenosine of the tRNA. In addition, various proofreading or editing mechanisms can be involved to increase the fidelity of translation [7].
Based on an analysis of available structural and functional information of the parasite and human tRNA synthetases, methionyl-tRNA synthetase of T. brucei (TbMetRS) was selected as a target for antitrypanosomal drug design. Most eukaryotes have at least two genes encoding MetRS enzymes that are respectively targeted to cytoplasm on the one hand, and mitochondria and chloroplasts on the other hand, forming two subfamilies. Mitochondrial MetRS belongs to MetRS1 subfamily, harbors one so-called connective peptide (CP) ‘knuckle’ – although residues capable of coordinating Zn2+ may or may not be present. In contrast, cytoplasmic MetRS belongs to MetRS2 subfamily and has two knuckles, with either one or two Zn2+ ions bound [8]–[10]. Remarkably, the T. brucei genome revealed only one gene encoding for a MetRS and this belongs to the MetRS1 subfamily [11]. The presence of only one MetRS in T. brucei enables the inhibition of protein translation in the cytosol as well as in the mitochondrion through the inhibition of a single enzyme, boosting the potential of targeting TbMetRS for drug development.
TbMetRS has been validated as a possible drug target through RNAi experiments [12]. In addition, a series of aminoquinolone-based inhibitors (ABIs) was shown to have potent antitrypanosomal activity in vitro and in vivo [12]. To guide further development of the inhibitors, well diffracting crystals of TbMetRS, which only grow in the presence of methionine, were soaked in solutions of ABIs and resulted in high resolution views of protein•ABI complexes [13]. Despite being in the crystalline state, the enzyme showed a large degree of flexibility, resulting in substantially different conformational states when bound with substrate, product or inhibitors. This, along with analysis of other available structures of MetRS, led to the conclusion that conformational selection is the main mechanism by which these compounds bind to TbMetRS. Major conformational states of the enzyme are the ligand-free “F-state”, the Met-bound “M-state”, the intermediate product methionyl-adenosine monophosphate (MAMP)-bound “P-state”, and the inhibitor-bound “I-state”. The binding pockets for ABIs are likely to be present in the F-state, but not M-state and P-state, and are stabilized by the ABIs to drive the population of conformations of the enzyme towards the I-state [13].
Unfortunately, despite their potency, the ABIs have poor membrane permeability and are unlikely to cross the BBB for treating late stage HAT, rendering them of limited use as antitrypanosomal drugs [14]. Therefore, another series of TbMetRS inhibitors, with a urea moiety connected to an aryl group, was designed to replace the aminoquinolone moiety [14]. These urea-based inhibitors (UBIs) inhibit the blood stream form of T. brucei in culture with an EC50 as low as 150 nM while having minimum toxicity to mammalian cells. Representative UBIs, such as Chem 1433 and Chem 1356, appear to have excellent membrane permeability, enter the central nervous system, have reasonable oral bioavailability, and suppressive activity against the parasite in a mouse model [14]. In culture, resistance to both ABIs and UBIs can be induced, but resistance development is slower than eflornithine and pentamidine [15]. However, the binding mode of UBIs is not obvious without determination of crystal structures, due to the conformational flexibility of TbMetRS. To ascertain their binding mode, and to further understand the structural plasticity of TbMetRS, a total of 15 crystal structures of TbMetRS•UBIs complexes were determined. The collection of UBIs explores a larger chemical, structural and functional diversity than the previously reported ABIs, thus providing a more complete picture of the binding mode of the inhibitors and the concomitant conformational changes of TbMetRS when a spectrum of inhibitors is bound. By analyzing the 23 structures of TbMetRS•Met, TbMetRS•MAMP, TbMetRS•ABI and TbMetRS•UBI complexes now available, general principles of inhibitor binding to TbMetRS, a flexible enzyme, are derived. These include the filling of the two subpockets in the enlarged methionine pocket (EMP), the importance of planarity in the auxiliary pocket (AP) binding moiety, and the hydrogen bonds with a completely conserved Asp. In addition, a ternary complex of TbMetRS with an UBI and the ATP analogue β,γ-methyleneadenosine 5′-triphosphate (AMPPCP) was also determined, providing structural evidence that UBIs are not competing with ATP for the inhibition of TbMetRS.
Methods for TbMetRS expression and purification are as previously reported [13]. Full-length and truncated (237–773) TbMetRS was cloned into the AVA0421 vector for expression in E. coli. Based on the truncated protein, site directed mutagenesis of surface residues 452KKE454 to ARA, all remote from the active site, was required to obtain well diffracting crystals, but solely in the presence of methionine.
Protein was purified by a Ni-NTA affinity column followed by overnight cleavage of the N-terminal hexa-histidine tag using N-terminally histidine tagged 3C protease at 4°C. Cleaved protein was purified by a second Ni-NTA step followed by size-exclusion chromatography on a Superdex 75 column (Amersham Pharmacia Biotech) using a buffer containing 25 mM HEPES, 500 mM NaCl, 2 mM DTT, 5% glycerol, 0.025% NaN3 and 10 mM L-methionine at pH 7.0. Purified protein retained five residues of the 3C protease cleavage site (GPGSM) at the N-terminus.
Unless otherwise stated, all chemicals were purchased from commercial suppliers and used without further purification. The final purity of all compounds was determined by analytical LCMS with Phenomenex Onyx Monolithic C18 column (4.6 mm×100 mm). The products were detected by UV at the detection wavelength of 220 nm. All compounds were determined to be >95% pure by this method. The purification by preparative HPLC was performed on Waters Xterra Prep RP18 OBD 5 µM (19 mm×50 mm) with CH3CN/H2O and 0.1% TFA as eluent. The mass spectra were recorded with an Agilent Liquid Chromatograph - Ion Trap Mass Spectrometer. NMR spectra were recorded with either a Bruker 500 MHz spectrometer or a Bruker 300 MHz spectrometer at ambient temperature. Synthesis of Chem 1433, Chem 1356, Chem 1387, Chem 1392, Chem 1444 and Chem 1415 have been previously reported [14]. The synthesis of Chem 1472, Chem 1473, Chem 1475, Chem 1476, Chem 1469, Chem 1478, Chem 1509 and Chem 1540 are similar and given in detail in the Supporting Information.
The thermal shift assay was performed as previously described [12], [14] using 0.5 mg/mL for TbMetRS, 100 µM of inhibitor, and 5% DMSO. The assays were repeated three times independently.
Compound IC50s were determined in the T. brucei methionyl-tRNA synthetase aminoacylation assay as previously described [12], [14] except 10 U/mL of pyrophosphatase was used per reaction.
T. brucei brucei (bloodstream form strain 427 from K. Stuart, Seattle BioMed, Seattle, WA) were used for EC50 measurements as previously described [12], [14].
The truncated TbMetRS surface mutant was crystallized following procedures reported earlier [13]. Briefly, the crystals were obtained by vapor diffusion using sitting drops equilibrated against a reservoir containing 2.0 to 2.3 M (NH4)2SO4, 0.2 M NaCl and 0.1 M sodium cacodylate pH 6.2 to 6.8. The drops consisted of 1 µL protein at 10 mg/mL plus 1 µL of the reservoir solution and additional 10 mM L-methionine and 1 mM tris(2-carboxyethyl)phosphine. Crystals grew in 1–2 days at room temperature.
To obtain enzyme•inhibitor complexes, TbMetRS•Met crystals were soaked in a cryo-solution containing the inhibitors as previously described [13]. Briefly, crystals were soaked in a 10 µL solution obtained by mixing 1 µL of 20 mM inhibitor in 20% DMSO, 4 µL reservoir solution and 5 µL 60% glycerol (as cryoprotective agent) in protein buffer. Crystals usually disintegrated when soaked longer than several minutes and, typically, required to be flash frozen in liquid nitrogen within one minute.
All data were collected under cryogenic conditions. For crystals soaked with compounds Chem 1433, Chem 1469 and Chem 1540, data were collected in home source facility using a MicroMax-007 HF rotating anode (Rigaku) equipped with VariMax HF (Osmic) monochromator and a Saturn 994 (Rigaku) CCD detector at a wavelength of 1.54 Å. For crystals soaked with compounds Chem 1356, Chem 1472, Chem 1473, Chem 1475, Chem 1476, Chem 1478, Chem 1509 and with Chem 1433•AMPPCP•Mg2+, data were collected at Stanford Synchrotron Radiation Lightsource synchrotron beamlines 9-2 and 12-2 at wavelength of 1 Å. For crystals soaked with compounds Chem 1387, Chem 1392, Chem 1444 and Chem 1415, data were collected at Advanced Light Source synchrotron beamlines 8.2.1 and 8.2.2 at wavelength of 1 Å. All data were processed with HKL2000 (Table S1) [16]. Previously reported structures of TbMetRS [13] were used as search models for phase determination by molecular replacement using the program Phaser [17]. Iterated building/rebuilding and refinement of models, including the use of translational/libration/screw (TLS) groups [18] in refinement, were performed using Coot [19] and REFMAC5 [20], respectively. The refinement restraints for all ligands were generated by the Grade web server [21] and modified based on our survey of crystal structures deposited in the Cambridge Crystallographic Database Center. The structure validation server MolProbity was used throughout the process to monitor the progress of structure determination [22]. All refined structures showed good statistics with no outliers in Ramachandran plots according to MolProbity. The final crystallographic refinement statistics are given in Table S1. The interplanar angles between the urea and the R2 group as listed in Figure 1 were calculated with the program Geomcalc in the CCP4 program suite [23]. Figures were created and rendered with Pymol [24]. Superposition of structures was carried out with Pymol, which first aligns proteins by sequence, followed by structural alignment using five cycles of refinement to improve the fit by discarding pairs with high relative variability.
Coordinates and structure factors for TbMetRS in complex with compound Chem 1433, Chem 1356, Chem 1387, Chem 1392, Chem 1444, Chem 1415, Chem 1472, Chem 1473, Chem 1475, Chem 1476, Chem 1469, Chem 1478, Chem 1509 and Chem 1540 are deposited in the Protein Data Bank under PDB ID: 4MVW, 4MVX, 4MVY, 4MW0, 4MW1, 4MW5, 4MW2, 4MW4, 4MWE, 4MW6, 4MW7, 4MW9, 4MWB and 4MWC, respectively. Coordinates and structure factors for TbMetRS in complex with Chem 1433 and AMPPCP are deposited in the Protein Data Bank under PDB ID 4MWD.
All 14 UBIs reported here have two aromatic moieties, R1 and R2, connected by a N-methylpropylamine linker, as shown on the first line of Figure 1. R1 is typically a substituted phenyl ring, a dichlorothiophene or a benzothiophene moiety. R2 is linked directly to a urea moiety which in turn is connected to the linker. R2 is either a phenyl ring, a hydroxyl-phenyl group, or a thiophene ring (Figure 1). The UBIs were tested for their ability to inhibit growth of T. brucei in cell culture (EC50) and aminoacylation activity of purified TbMetRS (IC50). Changes in melting temperatures (ΔTm) of TbMetRS in the presence of the inhibitors were also determined and range from 0.6 to 9.2°C (Figure 1). The UBIs cover a wide range of potencies, with IC50 values vary from 19 nM to more than 10,000 nM (Figure 1).
Crystal structures of 15 TbMetRS•UBIs complexes were determined with good statistics and geometry, without outliers in Ramachandran plots according to the MolProbity server (Table S1). The structures contain two subunits of TbMetRS per asymmetric unit. The Rossmann fold catalytic core of the enzyme contains an inserted connective peptide (CP) domain which can be further divided into two parts: (1) the ‘CP base’, which is formed by two antiparallel strands spanning from Asp353 to Tyr363 and from Thr398 to Arg408; and (2) the ‘CP knuckle’, formed by residues Ser364 to Val397 between the CP based strands (Figure S1). The single CP knuckle of TbMetRS lacks residues needed for the coordination of Zn2+ and hence, although the CP knuckle is well defined in all structures, there is no zinc ion bound.
Soaking of UBIs into the TbMetRS•Met crystals results in drastically different responses of the two subunits in the asymmetric unit: Met is retained in subunit A but inhibitor is bound in subunit B (Figure S2). Binding of inhibitors is accompanied by extensive conformational changes in subunit B [13]. Movement of multiple residues near the active site enlarges the initially small Met pocket, forming an enlarged methionine pocket (EMP). In addition, a new, previously unobserved, auxiliary pocket (AP) is formed next to the EMP. UBIs bind to both pockets, with R1 occupying the EMP and urea-R2 filling the AP (Figure 2A). The linker is crucial for properly connecting R1 and R2 such that these moieties can be optimally inserted into the EMP and the nearby AP, but is itself mostly solvent exposed. Chem 1433 will be discussed as the prototypic UBI since it is one of the most active UBIs (IC50 = 28 nM), exhibits excellent ability to cross the membrane, and has moderate activity in a mouse model of trypanosomiasis when administered orally [14].
The EMP is formed mainly by hydrophobic TbMetRS residues, some of which form the original Met pocket. Insertion of R1 into the EMP prevents the binding of Met. A core of 11 residues is in contact with the R1 group in all the complexes (Figure S3A). Within the EMP, two sub-pockets can be discerned. One sub-pocket, lined mainly by Pro247, Ile248, Asn480, Tyr481 and His523, is initially filled by the sulfur atom of Met in the Met pocket, hence will be termed the ‘EMP-S’. The displacement of Val473, Trp474 and Phe522, together with Leu478, form the ‘enlarged’ part of the EMP, the sub-pocket ‘EMP-E’ (Figure 2B).
The more potent UBIs typically are doubly-substituted at their meta-positions of the phenyl group in R1, filling both sub-pockets. For example, the prototypic inhibitor Chem 1433 is 3,5-diCl substituted. One of the meta-Cl atom occupies essentially the same position as the Met sulfur atom (within ∼0.6 Å), filling the EMP-S. The other meta-Cl atom fits into the EMP-E (Figure 3A). A few other UBIs, such as Chem 1444 and Chem 1415, have the asymmetrical meta-substitution of 3-Cl,5-OMe groups. In both inhibitors, the 3-Cl atom is preferred over the 5-OMe in the EMP-S (Figure 3A).
Five UBIs used in this study contain a tri-substituted R1 phenyl ring. These UBIs all have a Cl atom at one meta-position with either an iodine or an allyl group at the other meta-position. In addition, one of the ortho positions is substituted with either a 2-OH, 2-EtOH, 2-butoxy or 2-benzoxy. They are less active than inhibitors with a di-meta-substitution but these five structures provide information about the size of substituents that can be tolerated in the two sub-pockets of the EMP. The larger meta-substituents of an iodine or an allyl group appear to be favored in the EMP-S compared to the EMP-E, if the ortho-substituent is small (2-OH in Chem 1472 and Chem 1473) (Figure 3B). This is because next to the EMP-S, the space around the ortho-position is quite confined by the carbonyl O of Ile248 and C-β of Pro247, which are at distances of 3.7 Å and 4.5 Å, respectively, from the ortho carbon of R1 in the TbMetRS•Chem 1433 complex. In contrast, the space around the ortho-position next to the EMP-E is open. Therefore, larger ortho substituents do not fit into the EMP-S sub-pocket but can be accommodated by the space next to EMP-E sub-pocket (2-EtOH in Chem 1469, 2-butoxy in Chem 1475 and 2-benzoxy in Chem 1476) (Figure 3C). This is likely to result in the insertion of large meta substituents in the EMP-E (3-I in Chem 1469, and 3-allyl in Chem 1475 and Chem 1476), which is a less favorable binding pose as reflected in their high IC50 values (Figure 1).
Interestingly, the binding of large 2- and 3-substituents in and around the EMP-E, as seen in the complexes with Chem 1475 and Chem 1476, respectively, is accompanied by conformational adjustments of the protein. Residues Asp518 to His523 in the N-terminal part of helix α-9 are displaced by an average of 0.7 Å for the Cα atoms, and by as much as 1.8 Å for the side chain of Phe522, to create the space needed for the insertion of the allyl group of Chem 1475 and Chem 1476 into the EMP-E (Figure 3D).
Structures of three inhibitors that lack a halide-substituted R1 phenyl ring were also determined. The R1 moieties of Chem 1478, Chem 1509 and Chem 1540 are 3-phenylethyne, 2,5-dichlorothiophene and 2-methyl-1-benzothiophene, respectively. While the EMP-S in these complexes is filled, the fit in the EMP-E is sub-optimal, in agreement with their weaker activities compared to the prototypic inhibitor Chem 1433 (Figure 3E). The distances between inhibitor atoms facing the EMP-E to protein atoms forming the pocket are between 3.6 Å to 6.1 Å, creating a hydrophobic void in the pocket.
The AP is not present in the M-state of TbMetRS but is observed in the I-state after binding to inhibitors, as a result of a large number of structural changes [13]. The AP, delineated by 10 residues, is also largely hydrophobic like the EMP (Figure S3B). There are three types of R2 moieties in the complexes determined in this study. Eleven UBIs have a 3-thiophene ring, two a phenyl ring, and two a hydroxylated phenyl ring. All the urea-R2 moieties in UBIs bind to the AP in a similar mode. The R2 group of the inhibitors is inserted between helices α-2 and α-7 of the Rossmann fold core. The near planar urea-R2 moieties slide deep into the AP mainly through stacking interactions with, on one side, the planar Tyr250 side chain and the His289-Gly290 peptide unit and, on the other side, the Tyr472-Val473 peptide unit and the Val473 side chain (Figure 4A). The urea moiety is directly connected to the R2 group and occupies the entrance to the AP. Crucially, Asp287, which originally formed hydrogen bonds with the NH of the substrate Met [13], shifts towards the inhibitor and forms new hydrogen bonds with an NH group of the urea moiety (Figure 4B). A groove connecting the EMP and the AP in the I-state is blocked in the M-state by Tyr250. A flip of the Tyr250 side chain in the I-state allows the placement of the N-methylpropylamine urea linker of the inhibitors.
The N-methylpropanamine linker is the most flexible part of the inhibitors. There are four rotatable bonds connecting the five atoms in this linear moiety. The flexibility of the linker is evident even after the inhibitors are bound to the protein, as densities of the linker in all complexes are almost always weaker than the other parts of the inhibitors (Figure S2). Also, linkers can adopt different conformations – in particular in compounds Chem 1472, Chem 1475 and Chem 1476, the linker deviates from the most common linker conformation observed in other UBIs.
In most structures, the secondary amine group in the linker is bound to a water molecule, which is in turn H-bonded to the carboxylate of Asp287 and the carbonyl oxygen of Ile248 (Figure 4B). This water-mediated interaction with Asp287 could be particularly important in optimally positioning the Asp residue. The linker is otherwise mostly solvent exposed.
Superposition of the TbMetRS•UBIs and TbMetRS•MAMP (PDB code 4EG3 [13]) complexes revealed that the binding sites of UBIs and ATP may not overlap. Therefore, a soak was performed with Chem 1433, the ATP analogue AMPPCP and Mg2+, resulting in a structure of the ternary TbMetRS•Chem 1433•AMPPCP complex. The binding sites for ATP in both subunits in the asymmetric unit are accessible in the crystals, demonstrated by the formation of MAMP in both active sites when ATP and Mg2+ are soaked into TbMetRS•Met crystals [13]. Yet surprisingly, after soaking, AMPPCP does not bind to subunit A and Met remains bound essentially as without AMPPCP. In contrast, the density for AMPPCP is clear in subunit B, along with density for Chem 1433 (Figure 5A). Chem 1433 binding to the protein is essentially identical whether or not AMPPCP is present. Of possible significance is that one of the β-phosphate oxygen in AMPPCP is only 2.6 Å away from the amine group of the linker, with an NH…O angle of ∼135° (Figure 5B). Hence, a favorable interaction between this UBI and AMPPCP is likely to exist.
Superposition of subunit B in the current TbMetRS•Chem 1433•AMPPCP structure and the TbMetRS•MAMP structure [13] shows that the AMPPCP binding site overlaps with that of the AMP moiety of MAMP (Figure 5C). A few side chains (His256, His259 and Trp547) in the pocket are shifted slightly, by less than 2.5 Å, but the overall shape of the two pockets is rather similar to each other. However, the adenine ring and ribose sugar from MAMP bind ∼1.5 Å deeper into their respective pockets, compared to AMPPCP (Figure 5C). In the MAMP complex, stacking of the adenine ring with the Trp547 indole is also tighter than in the AMPPCP complex. Distances between atoms of the two rings range from 3.5 to 3.8 Å for MAMP in the binary TbMetRS•MAMP complex, and around 4.0 to 4.5 Å for AMPPCP in the TbMetRS•1433•AMPPCP complex.
Thermal melt experiments on TbMetRS in the presence of different combinations of ligands were performed to determine the nature of the interactions between Chem 1433 and ATP when they are bound to TbMetRS. A similar analysis has been used to establish synergism between an inhibitor of human ProRS, halofuginone, and ATP [25]. An increase in Tm of the TbMetRS•1433 complex upon adding ATP may be an indication of synergism between Chem 1433 and ATP in binding. Our results showed that the Tm of TbMetRS does not change significantly in the presence of ATP-Mg2+ or AMPPCP-Mg2+. Instead, the addition of ATP-Mg2+ or AMPPCP-Mg2+ to the complex of the TbMetRS•Chem 1433 complex result in increase of Tm by an average of 1.6°C (n = 4, P<0.02) and 1.2°C (n = 4, P<0.05), respectively, compared to the Tm of TbMetRS•Chem 1433 alone (Figure S4).
The 15 new UBI structures reported here widen our understanding of key structural features of this class of inhibitors which govern their affinity for TbMetRS. Some general rules based on the current complexes between TbMetRS and the R1-linker-R2 type of inhibitors include:
These insights might be used to improve further the affinity of next generations of inhibitors or to maintain high affinity when modifications are introduced to improve pharmacokinetic or other properties of current compounds.
The two sub-pockets in the EMP appear to be important determinants for high affinity inhibitors. Inhibitors with the highest potency have a Cl atom that virtually coincides with the sulfur atom position in the EMP-S. Methionine is the only natural amino acid with a thiol ether side chain. It is likely that the sulfur atom is utilized as a feature for specific recognition by MetRS. Considering the similar van der Waals radii of sulfur and chlorine [26], it is tempting to suggest that a meta-Cl in inhibitors mimics the physico-chemical properties of sulfur in Met and therefore occupies a strikingly similar position in the EMP-S as the S of Met. In contrast, UBIs with other types of R1 group (Chem 1478, Chem 1509 and Chem 1540) are unable to fill both EMP sub-pockets (Figure 3E), resulting typically in weaker activity. Large meta-substituents and additional ortho-substituents on R1 phenyl are also not well tolerated, possibly due to steric clashes (Figure 1).
We have previously proposed that ABIs bind to TbMetRS through conformational selection [13]. The new complexes are consistent with this hypothesis. Despite a wide diversity of inhibitor structures being explored, including R1 groups that are either ‘oversized’ (such as Chem 1472 and Chem 1469) or ‘undersized’ (such as Chem 1478, Chem 1509 and Chem 1509), the binding pockets of all complexes are essentially identical. This supports the notion that the inhibitors ‘select’ for a pre-existing, low energy conformation in the F-state of the enzyme for binding, instead of inducing conformations of binding pockets for different inhibitors. Yet, Chem 1475 and Chem 1476 appear to be the exception in that the EMP is subtly changed by these inhibitors. This represents, most likely, a realistic picture of the dynamics during the inhibitor binding process where the distinction between conformational selection or induced-fit mechanism is not always sharp. An “extended conformational selection” model, in which conformational adjustment following initial conformational selection has been suggested [27]. In this case, the common conformation selected by all other inhibitors is probably not compatible with binding of Chem 1475 and Chem 1476 due to steric clashes. However, a subsequent minor adjustment of the EMP upon binding, albeit with a higher energy protein conformation, is possible. This unfavorable component of the free energy of binding is manifested in the higher IC50 values for these two inhibitors.
Structurally, the key difference between ABIs and UBIs is the replacement of the aminoquinolone group in ABIs by the urea-R2 moiety in UBIs. This results in a considerable gain in membrane permeability that was reported recently although accompanied by a modest drop in activity [14]. For example, Chem 1433, a UBI, which corresponds to Chem 1312, an ABI (Figure 6A), crosses the membrane more efficiently in MDR-1 MDCKII assays but has a ∼3.5 fold less favorable IC50. Both aminoquinolone and urea-R2 groups bind to the AP. The hydrogen bonds between the carboxylate of Asp287 and the amines of either aminoquinolone or urea are conserved between ABIs and UBIs.
Nonetheless, two differences can be observed between UBIs and ABIs complexes. Firstly, the 4-ketone group of the aminoquinolone moiety, which is missing in the urea-R2 moiety, forms at least two water-mediated hydrogen bonds with the protein (Figure 6B). The loss of these interactions, albeit indirectly with the protein, could be part of the reason for the drop in activity. Secondly, the replacement of aminoquinolone with urea-R2 groups introduces an extra rotatable bond between the urea and the R2 groups. Given the larger number of rotatable bonds in UBIs than ABIs, and the larger flexibility of urea-R2 moiety than the conjugated system in aminoquinolone, it is reasonable to assume that UBIs have a higher conformational entropy in solution than ABIs. Therefore, upon binding, the loss of entropy of UBIs is likely higher than of ABIs, contributing to an overall less favorable free energy of binding.
The effect of the extra rotatable bond also manifests itself in a slight difference in the planarity of the AP binding moiety of inhibitors, as measured by the inter-planar angle between urea and R2 groups (Figure 1). When other parts of the inhibitors are equal, this angle is zero in ABIs with aminoquinolone, but slightly larger in UBIs with urea-thiophene, and even larger in UBIs with urea-phenyl (Figure 6C). The IC50 values of the inhibitors appear to be following the same trend. For example, Chem 1312 (aminoquinolone, flat, IC50 = 8 nM) has a lower IC50 than Chem 1433 (urea-thiophene, inter-planar angle = 9.1°, IC50 = 28 nM), which in turn has a lower IC50 than Chem 1356 (urea-phenyl, inter-planar angle = 17°, IC50 = 57 nM). Another pair of comparable inhibitors, Chem 1444 (urea-thiophene, inter-planar angle = 4.5°, IC50 = 19 nM) and Chem 1415 (urea-phenyl, inter-planar angle = 16.7°, IC50 = 45 nM), also follows the same trend. Therefore, a planar system is probably preferred for insertion into the AP due to the stacking interactions in the pocket, although there is no clear evidence how much the small deviations from planarity may contribute to differences in affinity. It is also to be noted that the UBIs with hydroxylated-phenyl as R2 (Chem 1387 and Chem 1392) appear to be deviating from the trend, possibly due to a gain of water-mediated interactions between the hydroxyl group and the protein. These differences in the planarity of the AP binding moieties may warrant more studies in the future.
The MetRS1 subfamily, but not the MetRS2 subfamily, was previously reported to be susceptible towards a group of synthetic inhibitors similar to the ABIs [8], [9]. The AP appears to be fully accessible in the F-state of all MetRS1, but is occluded in MetRS2, mainly due to the position of the gating residue Val473 [13]. The side chain of Val473 and its flanking residues Tyr472 and Trp474 form one side of the wall for the AP. Interestingly, these residues are strictly conserved among all MetRS. Based on the now available TbMetRS structures, Tyr472 and Val473 are not part of the Met or MAMP-binding pockets. The other side of the wall for the AP is formed mainly by His289 and Gly290, in which Gly290 is similarly found to be strictly conserved among all MetRS despite being more than 9 Å away from any atoms of the bound Met. Therefore the AP appears to be a well-conserved pocket, at least among all MetRS1 enzymes where Val473 does not occlude the entrance [13]. This raises the interesting questions regarding the role of the AP, located right next to the Met-binding pocket, in the normal functioning of MetRS. One option is that the pocket is involved somehow in tRNA binding, but unfortunately the available tRNA-bound structures of MetRS (PDB code 2CT8 and 2 CSX) have their tRNA acceptor arms located ∼25 Å away from the active site and provide little information about tRNA near the active site [28]. Therefore, the TbMetRS•Chem 1433 complex was superimposed onto the structure of another Class I aaRS-tRNA complex, E. coli LeuRS•LAMP analogue•tRNALeu (PDB code 4AQ7) [29], to determine if the AP might be involved in acceptor arm binding during aminoacylation. Although LeuRS does not possess an equivalent pocket in the position of the AP, it is interesting that the tRNA sugar moieties of both A76 and C75 are located near the entrance to the AP when superposed onto TbMetRS structure (Figure 6D). Consequently, it remains possible that in MetRS1 enzymes, either A76 or C75 of tRNAMet could rotate around their phosphodiester bond and place either one of its bases in the AP. Interestingly, in a recent report on the complex between human ProRS, which is a Class II aaRS, and the inhibitor halofuginone, a similar binding mode was suggested, where the inhibitor simultaneously bound to Pro- and A76-binding pockets [25].
The observation of simultaneous binding of AMPPCP and UBIs could have important implications for the development of inhibitors. The ternary TbMetRS•Chem 1433•AMPPCP complex showed that both inhibitor and ATP can bind to TbMetRS simultaneously, consequently, the inhibition is not competitive with respect to ATP (Figure 5). In addition, the favorable H-bond between AMPPCP and Chem 1433 observed in the structure also raises the possibility of co-operativity between the two ligands. Thermal melt studies on TbMetRS•Chem 1433 in the absence and presence of ATP and AMPPCP showed an increase in melting temperature (Figure S4) in agreement with the interaction seen between AMPPCP and Chem 1433 in the crystal structure (Figure 5). This is also supported by the kinetic data reported for similar types of inhibitor (like e.g. REP8839) in the binding of Staphylococcus aureus MetRS, where the inhibitor is competitive with Met but uncompetitive with ATP [9]. For example, the IC50 of REP8839 is 30- to 50-fold lower when a physiological concentration of ATP (2500 µM) is used, than when 25 µM of ATP is used [9]. Further, ATP is similarly reported to be synergistic regarding the binding of the inhibitor halofuginone to human ProRS [30]. Recently, a crystal structure of the ternary complex of human ProRS, halofuginone and an ATP analogue (AMPPNP), an enzyme•inhibitor•nucleotide complex similar to the TbMetRS•Chem 1433•AMPPCP complex, was reported [25]. Potentially, the interaction of ATP with inhibitors when bound to TbMetRS could be one of the considerations to be kept in mind for further optimization of these inhibitors. The high concentration of ATP present in the cell could be an advantage for the binding of this type of inhibitors. Developing inhibitors to exploit such an ATP-engaging binding mode might be also further used as a general strategy for targeting other aaRSs for inhibition.
This series of structures ascertain the binding mode of UBIs to the highly flexible target TbMetRS. The structures also reveal important features of the interactions of UBIs with its target MetRS which will guide future improvement of this group of active, orally available TbMetRS inhibitors. Further, the ability of UBIs to cross the BBB makes them exciting leads for obtaining drug candidates for treatment of HAT. Finally, the amino acid and possible tRNA pocket (EMP and AP)-targeting nature of these inhibitors, along with a potentially ATP-engaging binding mode, constitute a general approach to develop inhibitors of all tRNA synthetases of disease-causing pathogens, a large family of valuable drug targets.
|
10.1371/journal.pbio.1001904 | Evolution of Sexes from an Ancestral Mating-Type Specification Pathway | Male and female sexes have evolved repeatedly in eukaryotes but the origins of dimorphic sexes and their relationship to mating types in unicellular species are not understood. Volvocine algae include isogamous species such as Chlamydomonas reinhardtii, with two equal-sized mating types, and oogamous multicellular species such as Volvox carteri with sperm-producing males and egg-producing females. Theoretical work predicts genetic linkage of a gamete cell-size regulatory gene(s) to an ancestral mating-type locus as a possible step in the evolution of dimorphic gametes, but this idea has not been tested. Here we show that, contrary to predictions, a single conserved mating locus (MT) gene in volvocine algae—MID, which encodes a RWP-RK domain transcription factor—evolved from its ancestral role in C. reinhardtii as a mating-type specifier, to become a determinant of sperm and egg development in V. carteri. Transgenic female V. carteri expressing male MID produced functional sperm packets during sexual development. Transgenic male V. carteri with RNA interference (RNAi)-mediated knockdowns of VcMID produced functional eggs, or self-fertile hermaphrodites. Post-transcriptional controls were found to regulate cell-type–limited expression and nuclear localization of VcMid protein that restricted its activity to nuclei of developing male germ cells and sperm. Crosses with sex-reversed strains uncoupled sex determination from sex chromosome identity and revealed gender-specific roles for male and female mating locus genes in sexual development, gamete fitness and reproductive success. Our data show genetic continuity between the mating-type specification and sex determination pathways of volvocine algae, and reveal evidence for gender-specific adaptations in the male and female mating locus haplotypes of Volvox. These findings will enable a deeper understanding of how a master regulator of mating-type determination in an ancestral unicellular species was reprogrammed to control sexually dimorphic gamete development in a multicellular descendant.
| Sexual differentiation in eukaryotes is manifested in two fundamentally different ways. Unicellular species may have mating types where gametes are morphologically identical but can only mate with those expressing a different mating type than their own, while multicellular species such as plants and animals have male and female sexes or separate reproductive structures that produce sperm and eggs. The relationship between mating types and sexes and whether or how an ancestral mating-type system could have evolved into a sexually dimorphic system are unknown. In this study we investigated sex determination in the multicellular green alga Volvox carteri, a species with genetic sex determination; we established the relationship of V. carteri sexes to the mating types of its unicellular relative, Chlamydomonas reinhardtii. Theoretical work has suggested that sexual dimorphism could be acquired by linkage of gamete size-regulatory genes to an ancestral mating-type locus. Instead, we found that a single ancestral mating locus gene, MID, evolved from its role in determining mating type in C. reinhardtii to determine either spermatogenesis or oogenesis in V. carteri. Our findings establish genetic and evolutionary continuity between the mating-type specification and sex determination pathways of unicellular and multicellular volvocine algae, and will enable a greater understanding of how a transcriptional regulator, MID, acquired control over a complex developmental pathway.
| In many unicellular and simple multicellular eukaryotes sexual interactions are governed by mating types. Among sexually reproducing organisms mating types are thought to have evolved before gamete size differences and separate sexes evolved [1],[2]. Mating types (defined below) control sexual differentiation and specialized roles of cells that function as gametes in a diverse range of taxa including fungi, algae, ciliates, and cellular slime molds [3]–. In mating-type systems gametes can be isomorphic but can only mate with partners that express a different mating type than their own. Male and female gametes, on the other hand, are a hallmark of multicellular organisms such as metazoans and land plants. Males and females have developmentally specialized gamete types: large immotile eggs that are produced by females or female reproductive organs, and small motile sperm produced by males or male reproductive organs. The groundbreaking theory proposed by Parker and colleagues [7] modeled the evolution of anisogamy (asymmetric-sized gametes) from a starting population of isogametes (i.e., mating types) and identified the evolutionary forces that might cause a mating-type system to evolve into anisogamy (large and small gamete types) or oogamy (eggs and sperm). Additional theories and modifications to the original ideas of Parker and colleagues have been proposed (reviewed in [1],[8]–[10]), but very little attention has been given to the mechanism through which natural selection might act on a mating-type system to drive the transition to anisogamy or oogamy. One model involves the establishment of genetic linkage between a polymorphic locus that affects gamete size and a mating-type locus [11]. However, the genetic basis for the evolution of anisogamy/oogamy has not been determined in any experimental system, and it is not known whether it requires the addition of size control genes or other genes to an ancestral mating locus as the model proposes.
Volvocine algae are an excellent model for investigating the evolution of sexual dimorphism. They form a monophyletic clade encompassing a progression from unicellular species to multicellular forms with increasing organismal size and cell-type specialization [12],[13]. Volvocine algae all have a haploid vegetative reproductive cycle, but under specific conditions can be induced to undergo sexual differentiation and mating to form dormant diploid zygospores. Zygospores undergo meiosis and produce haploid progeny that reenter the vegetative phase [4],[14]. Chlamydomonas and smaller colonial volvocine genera are isogamous, while larger colonial forms are anisogamous or oogamous as is the case with the genus Volvox [15],[16]. Some species of Volvox and other anisogamous volvocine algae are heterothallic with genetically determined male and female sexes, while others are homothallic with a single clone producing a mixture of all-male and all-female colonies (dioecy), or homothallic with a single clone producing colonies containing both male and female gametes (monoecy) (reviewed in [16]). Previous studies have made use of volvocine algae to evaluate theories relating to the evolution of anisogamy and oogamy [13],[17]–[19], but the genetic basis for sexual dimorphism in this clade is still unclear [4],[20],[21].
In C. reinhardtii, the two genetically determined mating types, plus and minus, are morphologically similar, but express mating-related genes that allow fusion with a partner of the opposite mating type [14],[22]. Gametic differentiation in C. reinhardtii is triggered by absence of nitrogen (−N) and is governed by a mating locus (MT) whose two haplotypes, MT+ and MT−, are large, rearranged multigenic regions, which are suppressed for recombination and therefore segregate as Mendelian alleles [23],[24]. The C. reinhardtii gene MID (CrMID) is present only in the MT− haplotype and encodes a putative RWP-RK family transcription factor whose expression is induced by −N and that governs gametic differentiation [25]. The presence of MID activates the minus differentiation program and represses the plus program, while the absence of MID causes activation of the plus program and repression of the minus program. A second MT− gene, MTD1, also contributes to MT− gametic differentiation but is not essential for it [26]. MID is a rapidly evolving gene [27], but orthologs have been found in MT− strains or in males of all volvocine algae examined to date including VcMID in V. carteri (Figure S1A) [20],[21],[27]–[30]. However, the role of MID in sex determination has not been investigated outside of Chlamydomonas.
Volvox carteri f. nagariensis (hereafter V. carteri) is a spheroidal multicellular alga whose vegetative form is identical for males and females (Figure 1A). Each vegetative spheroid contains ∼2,000 sterile flagellated somatic cells on the periphery that provide motility, while inside the spheroid are ∼16 large immotile reproductive cells called gonidia. All of the cells are embedded within a clear extracellular matrix that comprises most of the spheroid volume. The two-day vegetative reproductive cycle begins with mature gonidia undergoing embryogenesis to form new miniature juvenile spheroids. During embryogenesis a programmed series of symmetric and asymmetric cleavage divisions occurs to produce a hollow ball of 2,000 small cells with 12–16 large cells on the anterior surface. The process of inversion then turns the embryo inside out so that the large cells end up on the interior of the spheroid where they will differentiate into new gonidia, and the small cells end up oriented with their basal bodies facing outward, and will begin to grow flagella as they undergo somatic differentiation. Over the next 1.5 days the juveniles grow, mature into adults, hatch, and begin the cycle again (reviewed in (Figure 1B) [31],[32]). Unlike C. reinhardtii that uses a nutrient trigger for gametogenesis, sexual differentiation in V. carteri is triggered by a diffusible glycoprotein hormone called sex-inducer that is active on both sexes [33]–[35]. In response to sex-inducer, gonidia from vegetative females and males undergo modified embryogenesis programs to produce sexual spheroids (Figure 1C) [36],[37]. Sexually induced female spheroids have ∼2,000 somatic cells similar to vegetative females, but inside contain 32–48 large egg cells that are formed during embryogenesis through altered timing of asymmetric cell divisions. Sexually induced male spheroids develop with 128 somatic cells and 128 large cells called androgonidia that are also produced through modification of asymmetric embryonic division patterning. The day after male sexual embryogenesis each androgonidial cell undergoes additional cleavage divisions to form a packet of 64 or 128 sperm cells. Sperm packets hatch and swim together to a sexual female where they break apart into individual sperm that enter the female through a fertilization pore. Sperm swim within the female until they find an egg and then fuse with it to form a diploid zygospore. Upon germination a single vegetative meiotic progeny is formed while the remaining three meiotic products are discarded as polar bodies (Figure 1C) [38].
Sexual differentiation in V. carteri is controlled by a dimorphic sex-determining locus (MT) with haplotypes designated MTM (male) and MTF (female). V. carteri MT occupies an equivalent chromosomal position to C. reinhardtii MT based on flanking syntenic gene content, but is at least 5-fold larger. Compared with C. reinhardtii MT V. carteri MT contains more sequence rearrangements between haplotypes, more repeat sequences, and has gametolog pairs (genes with an allele in both MT haplotypes) that are far more differentiated from each other [20],[24]. V. carteri MTF and MTM haplotypes can thus be considered a UV sex chromosome pair [39]. As described above, it has been proposed that anisogamy or oogamy could evolve through a size-regulatory gene becoming linked to an ancestral mating locus [11]. Both MTM and MTF haplotypes contain a putative cell-size regulatory gene, MAT3, whose alleles are highly dimorphic in sequence and expression between the sexes [20]. However, it is now apparent that anisogamy and oogamy in volvocine algae predate the appearance of MAT3 allelic dimorphism in the lineage meaning that other mating locus genes probably underlie the origins of anisogamy and oogamy [40]. Although MTM contains a MID homolog, VcMID (Figure S1A), its role in sexual differentiation is unclear because VcMID mRNA is expressed constitutively in both vegetative and sexual stages of males [20]. The apparent uncoupling of VcMID expression from the sexual cycle suggests that the VcMid protein might have a function outside of the sexual cycle or that its function might be regulated differently than that of CrMID whose expression is induced by −N.
In this study, we tested the role of VcMID in V. carteri sex determination by making transgenic females that express VcMid protein or by knocking down its expression in males using RNAi. We found that expression of VcMID in females is sufficient to convert eggs to sperm packets, while its absence in males causes androgonidial cells to differentiate into eggs. However, alteration of VcMid expression did not affect female or male early embryonic patterning during which the number and location of germ-cell precursors is established. We found that VcMID mRNA is expressed in all cell types, but VcMid protein accumulation is regulated by cell type and its subcellular localization is restricted to nuclei of differentiating and mature male gametes. Swapping experiments with CrMid demonstrated that the VcMid DNA binding domain and N-terminal domain are both required for its function in directing spermatogenesis in V. carteri. Crosses with sex-reversed strains revealed sexually antagonistic interactions between genes in MT and the sexual development pathway controlled by VcMid that negatively impacted reproductive fitness when gamete type did not match the mating locus genotype.
We tested the role of VcMid protein in sexual differentiation by generating female transgenic lines with an autosomally integrated VcMID transgene (pVcMID-BH) expressed under its own promoter and fused to a blue fluorescent protein (BFP) and a hemagglutinin (HA) epitope tag at its C-terminus to detect expression (Eve::VcMID-BH) (Figures 2A and S1B). Female transformants carrying an untagged version of VcMID (Figure S1C) had identical phenotypes as those carrying the tagged version, and all subsequent work was done with tagged strains and untagged transformants as a negative control for detection of VcMid protein. Eve::VcMID-BH lines showed a normal vegetative phenotype and constitutively expressed the mRNA for the VcMID-BH transgene, an expression pattern identical to the endogenous VcMID mRNA in males (Figure S2) [20]. As described above, when wild-type vegetative female gonidia are exposed to sex inducer they undergo modified embryogenesis and develop into sexual spheroids with 32–48 eggs and ∼2,000 sexual somatic cells (Figure 2B). When wild-type vegetative male gonidia are exposed to sex inducer they undergo modified embryogenesis and develop with 128 sperm packets and 128 somatic cells (Figure 2C). When vegetative gonidia from Eve::VcMID-BH lines were exposed to sex inducer they developed into progeny spheroids with a novel pseudo-male sexual phenotype: They produced ∼2,000 somatic cells and 32–48 sexual germ-cell precursors in a pattern similar to that of female eggs; but each of the 32–48 germ-cell precursors in the Eve::VcMID-BH lines underwent additional cleavage divisions like male androgonidia to produce sperm packets (Figure 2D). Moreover, the sperm produced in Eve::VcMID-BH lines were capable of fertilizing wild-type female eggs to produce characteristically orange-pigmented, thick-walled zygotes (Figure 2E). However, male function was incomplete as fertility defects were noted (see next section). Similar to wild-type sperm (and unlike wild-type female eggs) the Eve::VcMID-BH sperm cells were terminally differentiated and could not revert back to vegetative growth if left unfertilized. Another male-specific phenotype exhibited by Eve::VcMID-BH lines was frequent spontaneous occurrence of sexual differentiation in vegetative cultures [36], a trait whose underlying basis is not clear, but which appears to be under the control of VcMID.
Although some of the Eve::VcMID-BH sperm were functional and could fertilize wild-type eggs, the sperm packets and sperm cells from these strains had multiple defects including heterochronic delays in maturation and hatching defects (Figure S3A). The sperm packets were four times larger than those from wild-type males and contained about four times as many sperm cells (256 sperm/packet) (Figure 2F and 2G), some of which were aberrantly formed in contrast with wild-type sperm cells that had uniform morphology (Figures 2H, 2I, and S3B–S3F). Nonetheless the crosses between Eve::VcMID-BH pseudo-males and wild-type females produced MTF/MTF diploid zygotes (Figure 2E) that could germinate and produce haploid progeny. Forty-one progeny from one such cross were genotyped, half of which (19/41) inherited the VcMID-BH transgene and developed as pseudo-males, and half of which lacked the transgene and developed as normal females (22/41).
The absence of a vegetative phenotype in Eve::VcMID-BH transgenic lines despite constitutive expression of the VcMID-BH mRNA (Figure S2) suggested that VcMid protein expression or localization might be under posttranscriptional control. Although we could not detect BFP fluorescence in Eve::VcMID-BH strains, we could detect the HA epitope tag by Western blotting at all stages of vegetative and sexual development (Figures 3A and S4). Using immunofluorescence (IF) a nuclear-localized signal was detected for VcMid-BH protein in cleaving androgonidia and in mature sperm packets (Figures 3B–3J, S5A–S5R, and S6C–S6H), a result similar to earlier findings of Mid protein localization in sperm nuclei of Pleodorina [21]. However, nuclear VcMid-BH was not detected during early stages of sexual male embryogenesis prior to androgonidia cleavage (Figure S7).
We also examined VcMid-BH expression and localization in gonidia and somatic cells from vegetative spheroids. RNA was prepared from purified gonidial or somatic cells and reverse transcription and PCR (RT-PCR) detected VcMID-BH mRNA at similar levels in both cell types from transgenic females (Eve::VcMID-BH) and males (AichM::VcMID-BH) (Figure S8B and S8D). The endogenous VcMID transcript from wild-type males was also expressed in both vegetative cell types (Figure S8C and S8D). Whole cell extracts were prepared from purified Eve::VcMID-BH or AichM::VcMID-BH somatic cells and gonidia and subjected to SDS-PAGE and Western blotting to detect VcMid-BH protein (Figures 3K, S9A, and S9B). In contrast to VcMID-BH mRNA that was present in both vegetative cell types, VcMid-BH protein was only detected in vegetative somatic cells indicating that there is cell-type–specific regulation of VcMid protein synthesis or stability that restricts its accumulation to somatic cells during vegetative growth (Figures 3K and S6B). However, unlike the case for androgonidia and sperm cells, the VcMid-BH protein signal in vegetative somatic cells was excluded from the nucleus and was instead observed only in the cytosol and peri-nuclear region (Figures 3L–3O, S6I–S6P, and S9C–S9J). Together these data suggest that cell-type–limited expression and regulated nuclear localization of VcMid restrict its function to developing and mature sexual male germ cells in V. carteri.
In order to test whether VcMID is necessary for sexual differentiation of males we developed a new strategy for gene knockdown on the basis of RNAi-inducing hairpin constructs. The hairpin-forming portion of the construct corresponding to VcMID sequences was inserted directly into the 3′ UTR of the nitA selectable marker gene to allow direct and simultaneous selection of both NitA+ and hairpin expression (Figures 4A and S10; Text S1). This strategy has been successful for other loci besides VcMID (unpublished data), but only the results for VcMID are presented here. We also note that the VcMID knockdown phenotype was gene specific and did not occur when hairpins targeting other loci were introduced into V. carteri (unpublished data). Two hairpins targeting VcMID (VcMID-hp1 and VcMID-hp2) were introduced into a V. carteri wild-type male strain AichiM to generate AichiM::VcMID-hp1 and AichiM::VcMID-hp2 transgenic lines (see Materials and Methods for details). All transgenic male strains had normal vegetative phenotypes, and both hairpin constructs reduced VcMID expression; but AichiM::VcMID-hp1 lines had lower VcMID transcript levels than AichiM::VcMID-hp2 lines (Figure S2). We note that unlike wild-type males or pseudomales (see above), vegetative cultures of AichiM::VcMID-hp1 lines did not undergo spontaneous sexual induction. Sexually induced AichiM::VcMID-hp1 lines with strong knockdowns showed a novel phenotype: Their early sexual development proceeded as it would for a wild-type male strain and resulted in spheroids that contained 128 small somatic cells and 128 large cells that resembled uncleaved androgonidia (Figure 4B), but the large cells never underwent further cleavage into sperm packets. Instead, many of them could be successfully fertilized with wild-type male sperm to make MTM/MTM diploid zygospores (Figure 4C). The ability to differentiate as zygospores when fertilized indicates that the presumptive androgonidia in AichiM::VcMID-hp1 strains were converted to functional eggs and that these strains were behaving as pseudo-females. However, unlike normal zygotes from a wild-type cross, 30%–50% of the zygotes from AichiM×AichiM::VcMID-hp1 pseudo-female crosses died and bleached shortly after fertilization (Figure 4C), a phenotype that depended on addition of exogenous sperm. The surviving zygotes from these crosses produced some viable meiotic progeny, but germination and survival of the progeny were reduced compared with normal wild-type zygotes (Table S1). 29/43 viable progeny inherited the VcMID-hp1 transgene and developed as pseudo-females, while 14/43 lacked the transgene and developed as normal males. The apparent deviation from a 1∶1 inheritance pattern of the transgene was noted but was not pursued further in this study. The high mortality of pseudo-female eggs—whose mating loci are genetically male—suggest that MTF contains genes that promote female gamete and/or zygote fitness that are absent from MTM. If left unfertilized, the eggs from AichiM::VcMID-hp1 lines could de-differentiate and reenter the vegetative reproductive cycle as do unfertilized female eggs. A similar phenotype as our pseudo-male strain was reported previously for a male mutant [41], but the mutant strain is no longer available for characterization.
AichiM::VcMID-hp2 lines were not as severely knocked down for VcMID expression as AichiM::VcMID-hp1 lines (Figure S2), and had a distinct hermaphrodite phenotype in which sexual spheroids developed with a mixture of normal-looking male sperm packets and pseudo-female eggs (Figure 4D). The hermaphrodite lines exhibited self-fertility as evidenced by zygospores that formed in sexually induced monocultures (Figure 4E). These results indicate that V. carteri sex determination is highly sensitive to VcMid dosage where either a male or female fate is established depending on the level of VcMid.
In several instances genes from C. reinhardtii have been shown to function interchangeably with their V. carteri orthologs in developmental processes such as inversion and asymmetric cell division (reviewed in [42],[43]). Mid proteins have at least two domains: The C-terminal region has a predicted RWP-RK motif DNA binding domain, while the N-terminal region does not show similarity to characterized protein domains from other organisms (Figure S1) [28]. We tested whether either of the domains from CrMid could substitute for those of VcMid to control spermatogenesis in V. carteri. To do so we generated three constructs in which all or part of the CrMID genomic coding region was substituted for VcMID sequences in pVcMID-BH. One construct contained the entire CrMID gene (pCrMID-BH) (Figure 5A) while the other two contained the CrMID N-terminal domain fused to the VcMID DNA binding domain (pMID-VNCC-BH) (Figure 5B), or the VcMID N-terminal domain fused to the CrMID DNA binding domain (pMID-CNVC-BH) (Figure 5C). All three constructs as well as pVcMID-BH were introduced into wild-type females, and transformants that expressed the different predicted proteins were identified (Figures 5D–5F). Unlike Eve::VcMID-BH transformants that showed a pseudo-male phenotype (Figure 1D), none of the transformants that expressed CrMid or Mid chimeras had this phenotype, but instead always developed as wild-type females (Figure 5G; Table S2). The CrMid-BH protein was expressed as a full-length form and two shorter isoforms (Figure 5D), possibly due to proteolytic cleavage or incorrect pre-mRNA processing. However, lines with either of the two chimeric constructs expressed only full-length predicted proteins at levels comparable to VcMid-BH (Figure 5E and 5F). We conclude from these experiments that CrMid and VcMid are not functionally interchangeable, and that both the N-terminal and RWP-RK domains of VcMid are required to activate spermatogenesis in sexual germ cells.
Although it was reasonable to predict that the route for evolving sexual dimorphism would be through addition of new genetic functions and pathways to a core mating locus as proposed originally by Charlesworth for the evolution of anisogamy [11], we found instead that a single conserved volvocine algal mating locus gene, MID, is largely responsible for controlling male versus female sexual differentiation in V. carteri. It is notable that VcMID controls multiple sexual traits in V. carteri that have no analogs in C. reinhardtii. The male traits controlled by VcMID include specialized cleavage divisions of androgonidia, sperm packet inversion, sperm packet hatching, specialized sperm cell morphology, and gamete recognition without flagellar adhesion.
The term deep homology is used to describe ancient and conserved genetic mechanisms that control traits that, on the surface, appear disparate or have no obvious homology relationship [44]. Metazoan eye development is a classic example; across phyla with very different eye architecture, it is controlled by the conserved transcription factor, eyeless/Pax6 [45]. In the case of volvocine algae Mid proteins control two very different manifestations of sexual reproduction in C. reinhardtii and V. carteri whose MID orthologs have been diverging for as long as 200 million years [46]. Sexually dimorphic gametes have evolved from isogamous ancestors several times in independent multicellular taxa, and the mechanism could be similar to, or different from, that in volvocine algae where a master regulatory gene acquired the ability to direct sperm-egg dimorphism. Oogamy was likely established very early in the Streptophyte lineage before the split between Charophyte algae and land plants, but the origins and bases of oogamy in Charophytes remain unclear [47]. The recent identification of an anisogamous sexual cycle in a Choanoflagellate—a unicellular relative of metazoans—introduces the potential for investigating the early evolution of gamete size dimorphism in animals [48].
Our results demonstrate for the first time a function for Mid protein in a volvocine algal sexual cycle outside of the genus Chlamydomonas and suggest that Mid could be the master regulator of mating type, gamete size, and gender throughout the volvocine lineage where MID genes have been identified in most genera [21],[28],[30]. Although MID sequences were previously shown to evolve rapidly, the finding that Mid protein from C. incerta (now reclassified as C. globosa [49]) can substitute for C. reinhardtii Mid indicates that functional conservation can be retained after speciation [27]. However, C. reinhardtii Mid protein could not substitute for the V. carteri ortholog, which appears to require both its native DNA binding and N-terminal domains to function in sex determination (Figure 5). Future work using cross-species complementation will help clarify whether Mid protein function co-evolved with sexual dimorphism in volvocine algae as our data suggest might be the case.
Transcriptional regulatory network evolution has been studied in various developmental contexts [50]–[53], but very little is known about how regulatory networks are modified or coopted during unicellular to multicellular transitions [54]. The Mid system in volvocine algae represents a new opportunity for understanding how a cell-type specification pathway in a unicellular ancestor evolved to control a complex developmental program in a multicellular descendant. While mating-type differentiation in Chlamydomonas appears to involve differential expression of a small number of genes between the plus and minus gametes [14],[22], a larger set of differentially expressed genes might be expected to specify sperm and eggs, which are developmentally very different from each other [37],[55]. An important future goal will be to identify and compare the direct and indirect targets of Mid proteins in both C. reinhardtii and V. carteri that are predicted to be more numerous and diverse in V. carteri.
Another interesting question is whether MID-like genes function in sex determination in green algae outside of the volvocine lineage. The molecular bases for sex determination in most green algae and protists are poorly understood, but RWP-RK family proteins are found throughout the green eukaryotic lineage [56], including small Mid-like proteins in Prasinophyte algae [57], and even in distantly related Cryptophyte algae [58]. It remains to be seen whether these Mid-like proteins in non-volvocine species function in sex determination.
As described in the Introduction, volvocine algae exhibit wide diversity in their sexual cycles: There are isogamous, anisogamous, and oogamous species with sexual cycles that can be heterothallic or homothallic. Among homothallic species some are dioecious (producing a mixture of all male and all female sexual offspring) or monoecious (producing sexual offspring containing gametes of both sexes in one individual) [16],[55].
V. carteri is a heterothallic species with genetically determined sexes; but, a remarkable phenotype was produced by a partial RNAi knockdown of VcMID in males (Figure 4D). Rather than developing as male or female, the partial-knockdown male spheroids became self-fertile monoecious hermaphrodites that produced sperm and eggs in a single individual. On the basis of this observation, it can be inferred that sexual differentiation in V. carteri is bi-stable and highly sensitive to initial MID dosage. Once the Mid-sensitive step of development is initiated (which may coincide with nuclear translocation of VcMid protein), positive and negative feedback loops may be used to lock the sex determination program into a male or female state. This state could be achieved either independently of VcMid concentration, or through positive/negative reinforcement of initial expression states. The phenotype of hermaphroditic development by partial VcMID knockdown may be relevant to the evolution of homothallism, which appears to have arisen in all three major clades of Volvox [16],[55],[59]. We speculate that naturally evolved homothallic volvocine algae possess a MID gene whose expression is insufficient to specify 100% male gamete production—much like our VcMID-hp2 strains. Moreover, the timing of MID expression in homothallic species of Volvox could play a role in determining monoecious versus dioecious reproductive development: If the Mid-sensitive developmental switch is triggered relatively late in development after germ-cell precursors are formed then a mixture of male and female gametes could develop within a single spheroid as we found with VcMID-hp2 strains (i.e., homothallism, monoecy). In contrast, if the Mid-sensitive step occurred very early in development before individual germ cells were established, then the fate of all germ cells within a mature spheroid might be locked into a male or female program and the resulting population would produce a mixture of all-female or all-male spheroids (i.e., homothallism, dioecy).
The bi-stability of sex determination at intermediate levels of MID expression in V. carteri is also reminiscent of the iso1 mutant phenotype in C. reinhardtii where MT− iso1 cells differentiate into a mixture of plus and minus gametes that iso-agglutinate but cannot self-fertilize [60]. The self-infertility of iso1 MT− strains is due to the absence of FUS1, a gene from the MT+ haplotype that is required for gamete fusion [61]. In contrast, there are no essential genes in MTF of V. carteri that are absolutely required for fertilization and subsequent germination of progeny from matings between pseudo-females (VcMID knockdown strains) and males. This lack of essential female MT genes for completing the sexual cycle may have facilitated transitions from heterothallism to homothallism in volvocine algae.
In volvocine genera other than Volvox—including the anisogamous genus Pleodorina— sexual differentiation and MID expression are both triggered by the absence of nitrogen (−N) [21],[25],[28],[62]. In contrast, V. carteri f. nagariensis and other Volvox species use species-specific pheromones called sex inducers to trigger sexual differentiation [16],[55],[63]. A seemingly parsimonious evolutionary route for rewiring input into the Mid pathway in V. carteri would simply place VcMID transcription under the control of sex inducer instead of nitrogen availability. Unexpectedly, however, VcMID mRNA and VcMid protein are both expressed constitutively at all life cycle stages (Figures 3A and S4) [20], and unlike the case in C. reinhardtii, VcMid appears to be under at least three types of posttranscriptional control: (i) Although VcMID mRNA is present in both vegetative cell types (somatic and gonidial cells), VcMid protein is only translated or stably produced in somatic cells and is absent from vegetative gonidia and vegetative embryos (Figures 3K and S6B). (ii) The VcMid protein produced in somatic cells is excluded from the nucleus (Figures 3L–3O, S6I–S6P, and S9C–S9J). (iii) In response to the presence of sex inducer, VcMid protein accumulates in the nuclei of cleaving androgonidial cells and in the nuclei of sperm cells where it is presumed to function in specifying sperm development (Figure 3B–3J, S5, and S6C–S6H).
In depth study will be required to determine how cell-type–regulated production and localization of VcMid are achieved, but it seems reasonable to infer that one or more factors are produced in sexually induced spheroids that promote the translation and/or stability of VcMid and its nuclear localization in androgonidia and sperm. The factor may interact directly with VcMid as a partner and may help specify its localization at promoters of target genes, but this idea remains to be tested. The absence of VcMid protein in vegetative gonidia despite its message accumulating to the same extent as in somatic cells seems puzzling at first glance. However, the block in stability or translation of VcMid in gonidia may have evolved as a failsafe mechanism to prevent sexual differentiation during vegetative embryogenesis. Such a mechanism would be unnecessary in vegetative somatic cells that are already terminally differentiated, but could potentially be important for vegetative gonidial cells because male sexual differentiation is irreversible and would be fatal if it occurred at the wrong time. However, why vegetative somatic cells express VcMid remains a mystery. As noted in Results, we observed no obvious vegetative phase phenotypes in AichiM::VcMID-hp strains whose somatic cells were missing VcMid, or in Eve::VcMID-BH strains that contained VcMid in somatic cells, but a definitive conclusion about whether cytoplasmic VcMid has a role in vegetative somatic cells awaits more in depth examination.
Our results show that the presence or absence of VcMID is the key determinant of differentiation in V. carteri sexual spheroids; yet the MT locus of this alga has around 70 additional genes, many of which show sex-biased gene expression [20]. Some of the MT genes are present only in the male or only in female haplotype, while the majority are male and female gametolog pairs that are highly diverged in sequence and expression pattern [20]. Our ability to uncouple gender from sex chromosome identity by manipulation of VcMID expression allowed us to uncover potential contributions of male and female MT genes to sexual dimorphism and reproductive fitness. Haploid sex chromosome systems have received less attention than diploid systems, but there are several predictions about them that our results begin to address. Under haploid dioecy, recessive mutations in sex-linked genes are not sheltered from selection as they are in the heterogametic sex of diploid systems, and are therefore expected to degenerate equally and lose only genes required for the opposite sex and not their own [64]. We note, however, that in V. carteri MT there is no evidence for an allele of a gametolog pair having been eliminated from one mating haplotype and retained in the other [20]. Haploid sex chromosomes are predicted to be similar to diploid systems in that both should accumulate sexually antagonistic alleles that benefit one sex, but harm the other [39],[65], and should accumulate repeat sequences, as appears to have occurred in both Volvox and in the bryophyte Marchantia [20],[66]. Accumulation of sexually antagonistic alleles has not been tested for haploid sex chromosomes, but the extensive divergence between V. carteri MT gametologs suggests that this phenomenon may contribute to the developmental and fitness defects we observed in pseudo-male and pseudo-female strains.
Although none of the sex-limited MT genes besides VcMID appear to be essential for V. carteri sex determination and completion of the sexual cycle, they clearly impact sexual development and reproductive fitness. A striking phenotype for both the pseudo-male and pseudo-female strains was the patterning of their germ-cell precursors formed during sexual development (Figures 2B–2D and 6A–6C). It is clear from these phenotypes that sexual germ cell patterning (i.e., the number and distribution of germ-cell precursor cells and ratio of germ-cell precursors to sexual somatic cells) is separable from germ cell differentiation and is not controlled by the VcMid pathway. Instead, this sexual patterning trait must be controlled by other MT genes (Figure 6). One candidate for this male-female patterning difference is the MAT3 gene that encodes the retinoblastoma-related homolog in V. carteri [20],[67]. In Chlamydomonas, Mat3 protein controls the multiple fission cell cycle by establishing the threshold size at which division can occur and by coupling the extent of cell division to mother cell-size [68],[69]. Although not directly involved in determining gamete cell-size as we had originally predicted [40], it is possible that the male and female alleles of VcMAT3 dictate the timing of asymmetric cell divisions by coupling embryonic blastomere cell-size to the asymmetric division machinery. Future work will be aimed towards determining the role of male and female VcMat3 gametologs in sexual cell division or other parts of the V. carteri sexual cycle.
In addition to defects in germ cell patterning in Volvox pseudo-males and pseudo-females, these strains show other reproductive defects. These include abnormal sperm cell shape and morphology (Figures 2H, 2I, and S3B–S3F), low efficiency of sperm packet hatching (Figure S3D), and delayed timing of androgonidial cleavage into sperm packets (Figure S3A). In depth study may reveal other defects in pseudo-male sperm related to cytoskeletal organization, gamete recognition, motility, and fertilization dynamics. Although egg cells lack distinct morphological features like sperm, we noted a very high mortality rate in pseudo-female eggs that occurred when we attempted fertilization. The mortality we observed under these conditions could be due to pre-zygotic defects in the eggs caused by their smaller size than wild-type female eggs or by a mismatch between the male mating locus and the female sexual differentiation pathway. The mortality might also be due to zygotic defects that occur when a copy of the female mating locus is absent from the zygote immediately after fertilization. Future work aimed at developing quantitative assays for distinct steps of fertilization will allow us to document in more detail the fitness contributions of male and female MT genes to the sexual cycle.
Detailed Materials and Methods are provided in Text S1. Materials used in this study will be made available upon request with the completion of a Materials Transfer Agreement from Donald Danforth Plant Science Center.
Eve (Volvox carteri. f. nagariensis UTEX 1885) and AichiM (Volvox carteri. f. nagariensis NIES 398) were obtained from stock centers http://web.biosci.utexas.edu/utex/ and http://mcc.nies.go.jp/, respectively. The strains that were used for transformations are described below and in Text S1. Other strains are described in Table S3. Growth of strains was in standard Volvox medium (SVM) or urea-free standard Volvox medium (UF-SVM) with growth conditions described in more detail in Text S1.
All plasmid constructs were made using standard molecular cloning methods and manipulations [70]. PCR amplifications used for plasmid construction were done with Phusion Polymerase (Thermo Scientific) according to the manufacturer's guidelines (see also PCR amplification conditions below). Primers used in this study are described in Table S4.
Constructs were introduced into NitA− female strain E15 or male strain A18. Transformation or co-transformation of E15 or A18 with nitrate-reductase (NitA) encoding plasmid pVcNR15 [71], pVcMID hp1, or pVcMID hp2 was done as previously reported [72] with minor modifications described in Text S1. Transformed gonidia cells were selected in UF-SVM.
For each assay three vegetative juvenile-stage spheroids were placed in a well of a six-well microtiter plate with ∼9 ml SVM per well and 10 µl of sex inducer with a titer of 106, and maintained at 32°C in a 16 h∶8 h light∶dark cycle [34]. The phenotypes of the 40–50 sexual progeny spheroids that resulted from sexual development of gonidia in the three starting spheroids were scored visually under a dissecting microscope and documented using a compound light microscope (Leica DMI6000B, 40× objective, differential interference contrast [DIC] optics) after 3–7 days.
Mating and zygote germination were performed as previously described with minor modifications [73]. Parental strains were grown to a density of ∼300 unhatched juveniles in 350 ml of SVM at which point sex-inducer was added. In the subsequent cleavage cycle sexual spheroids were produced and allowed to mature. Egg-bearing females were released from their parental spheroid by gentle pipetting 3–5 hours prior to hatching and mixed with males that had their sperm packets released from their vesicles within the parental spheroid by a similar procedure. Matings took place in a glass 150 mm×25 mm petri dish at a density of 10–15 spheroids/ml. The petri dish was placed on a light box within a 30°C growth chamber for 8–16 hours, and then the dishes were wrapped with aluminum foil and left at 32°C for at least one week and typically for three weeks prior to germination.
Drawn-out Pasteur pipets were used to manipulate zygotes. In order to remove residual sex-inducer and any other potential inhibitors of germination, zygotes were put into 1.5 ml tubes with ∼1 ml SVM medium and washed as follows: Tubes were vortexed for 10 min, then spun briefly at ∼1,000 rpm in a microcentrifuge to pellet zygotes. The supernatant was removed and the washing step was repeated three times. The washed zygotes were transferred to a sterile glass depression slide, washed briefly with SVM, and allowed to settle to the bottom. Ten to 50 zygotes were transferred to a sterile glass depression well containing SVM with 60 µg/ml carbenicillin, and incubated in a 16 h∶8 h light∶dark 30°C growth chamber. Zygotes germinated after two to six days. The germling colonies were individually transferred to six-well microtiter plates for growth and clonal expansion.
Total Volvox RNA was prepared as previously described [20]. cDNA was prepared from 5 µg total RNA following the manufacturer's protocol for Thermoscript (Invitrogen) using a 10∶1 mixture of oligo dT and random hexamer for priming and the following cDNA synthesis reaction temperatures: 25°C 10′, 42°C 10′, 50°C 20′, 55°C 20′, 60°C 20′, 85°C 5′, after which the reactions were treated with RNaseH. Reactions were diluted 1∶10 with 10 mM Tris pH 8.0, 1 mM EDTA (TE), and stored at −20°C. S18 and VcMID expression was measured using amplification with primer sets VcMid.f1 and VcMID.r1 and S18.1 and S18.2 as described previously [20] and in Table S4.
Vegetative gonidia and somatic cell separation was done by mechanical disruption and differential centrifugation. Cultures grown in four 350 ml SVM flasks with ∼5,000 spheroids/flask were collected with a magnetic filter funnel with 25 µm nylon mesh filter, and transferred to a 40 ml Kimble Kontes Dounce homogenizer. Spheroids were broken with a tight-fitting pestle (B type) with six strokes. Broken spheroids were transferred to a 50 ml Falcon tube, and the volume adjusted to ∼40 ml with SVM, after which 2.8 ml Percoll was added and the tubes spun at 200g for 5 min at room temperature. The supernatant containing somatic cells was transferred to a beaker and diluted to 200 ml with SVM. The pellet containing gonidia or embryos was washed two times with 50 ml SVM and gonidia collected after each wash by centrifugation at 200g for 5 minutes. Pure gonidia were then collected in a filter funnel using 10 µm nylon mesh, which allows any remaining somatic cells to pass through. To obtain pure somatic cells the diluted supernatant from the Percoll step above was spun at 460g for ∼3 min to pellet any contaminating gonidia. The low speed spin supernatant was then spun at 3,220g for 5 min to obtain a pure somatic cell pellet, which was washed twice with 50 ml SVM prior to extraction of RNA or protein.
Approximately 1,000 synchronized spheroids at designated stages were hand picked for protein sample preparation. Pelleted spheroids were mixed 1∶1 with Volvox Lysis Buffer (1× PBS supplemented with 1% NP40 [IPEGAL], 1× Sigma Plant Protease Inhibitor Cocktail [catalog number P9599], 5 mM PMSF, 10 mM benzamidine, 5 mM EDTA, 5 mM EGTA). Spheroids and cells were disrupted using a Covaris S220 ultrasonicator according to the manufacturers instructions with the following program settings: PP = 200, DF = 20, CpB = 300, T = 6°C, and t = 300 s in TC 12×12 tubes at 4°C. After lysis the samples were centrifuged at full speed in a microfuge to pellet debris and the supernatant mixed with sample buffer and boiled prior to gel fractionation. SDS-PAGE and Western blotting were performed using standard procedures [70]. SDS-PAGE gels were blotted to Immobilon-P PVDF membranes (Millipore) prior to immunodetection. The rat monoclonal antibody 3F10 (Roche) was used for detection of the HA epitope on immunoblots and for IF. Tubulin was detected with an anti-α-tubulin antibody purchased from Sigma-Aldrich (clone B-5-1-2, catalog number T6074). Antibodies were used at the following dilutions: 250 mg/ml anti-HA, 1∶2,000; and 2 mg/ml anti-α-tubulin, 1∶20,000, all diluted in PBS with 0.05% Tween 20. Blocking was performed with 40 ml of 5% nonfat dry milk in PBS with 0.05% Tween 20 for one hour. Blots were incubated overnight at 4°C min (anti-HA antibody). Blots were washed three times with PBS and 0.05% Tween 20 at room temperature for 10 min, then horse radish peroxidase (HRP)-conjugated goat anti-rat secondary antibody (Thermo Scientific) was used at 1∶2,500 dilution and incubated with blots at room temperature in PBS, 5% nonfat dry milk, 0.05% Tween 20 for 1 h. Blots were then washed three times with PBS and 0.05% Tween 20 at room temperature for 10 min and then briefly rinsed with deionized water. Antigen was detected by chemiluminescence (Luminata Forte Western HRP Substrate, Millipore) using autoradiographic film (HyBlot CL autoradiography film, Denville Scientific Inc.)
Eve::VcMID-BH or Eve::VcMID (as a negative control) samples were processed in parallel and imaged under identical conditions. Spheroids were collected using a magnetic funnel with a 25 µm nylon mesh filter (Pall Scientific). Sexual spheroids were fixed with 2% or 4% paraformaldehyde with 1× plant protease inhibitor cocktail (Sigma) (10 µl/ml), 1 µm ALLN (catalog number 208719, VWR International), 1 µm MG132 (catalog number 133407-82-6, Cayman Chemical), 1 mM dithiothreitol (DTT) for 1 h on ice. Spheroids or cells were then washed with PBS and resuspended in cold methanol (−20°C) for five minutes. The methanol wash was repeated two more times and then samples were washed again with PBS and rehydrated in PBS at room temperature for 15 min after which they were adhered to poly-L-lysine–coated cover slips. Cover slips were blocked for 30 min in blocking buffer (5% BSA and 1% cold-water fish gelatin) and incubated for 30 min in the same buffer with 10% (v/v) normal goat sera (Antibodies Incorporated). Cover slips were incubated overnight in anti-HA 1∶500 (1∶300 for somatic cell staining) in 20% blocking buffer at room temperature or at 4°C and washed 6×10 min with 20% blocking buffer in PBS 0.05% Tween 20. After washing out unbound primary antibody cover slips were incubated with AlexaFluor 488 conjugated goat anti-mouse secondary antibodies (Invitrogen) 1∶500 (1∶300 for somatic cells staining) in 20% blocking buffer and incubated for 1 h at room temperature (for 4 h at 4°C for somatic cells staining) in the dark. Following six washes with 20% blocking buffer diluted in PBS and 0.05% Tween20, the cover slips were incubated for 15 min in 2 µg/ml DAPI and then washed for 5 min in PBS. Excess liquid was removed, and the cover slips were mounted with VectaShield (Vector Labs) or Mowiol∶PPD (PPD = p-phenylenediamine 1,4-Benzenediamine hydrochloride [Sigma P1519]) with a 9∶1 ratio. Microscopy was performed with a Leica DMI6000 B using DIC optics or using the following filter cube sets and illumination with a Prior Lumen 200 light source: A4: excitation BP 360/40; dichroic 400; emission BP470/40, L5: excitation BP 480/40; dichroic 505; emission BP 527/30. Where indicated Z stacks were subject to deconvolution using the Leica Advanced Fluorescence Application Suite. Images are representative of results from at least three independent experiments.
|
10.1371/journal.pgen.1001017 | A Forward-Genetic Screen and Dynamic Analysis of Lambda Phage Host-Dependencies Reveals an Extensive Interaction Network and a New Anti-Viral Strategy | Latently infecting viruses are an important class of virus that plays a key role in viral evolution and human health. Here we report a genome-scale forward-genetics screen for host-dependencies of the latently-infecting bacteriophage lambda. This screen identified 57 Escherichia coli (E. coli) genes—over half of which have not been previously associated with infection—that when knocked out inhibited lambda phage's ability to replicate. Our results demonstrate a highly integrated network between lambda and its host, in striking contrast to the results from a similar screen using the lytic-only infecting T7 virus. We then measured the growth of E. coli under normal and infected conditions, using wild-type and knockout strains deficient in one of the identified host genes, and found that genes from the same pathway often exhibited similar growth dynamics. This observation, combined with further computational and experimental analysis, led us to identify a previously unannotated gene, yneJ, as a novel regulator of lamB gene expression. A surprising result of this work was the identification of two highly conserved pathways involved in tRNA thiolation—one pathway is required for efficient lambda replication, while the other has anti-viral properties inhibiting lambda replication. Based on our data, it appears that 2-thiouridine modification of tRNAGlu, tRNAGln, and tRNALys is particularly important for the efficient production of infectious lambda phage particles.
| In this study, we took advantage of a new genetic resource for E. coli mutants to screen for previously undiscovered lambda phage host-dependencies. We then assessed the dynamics of infection in these different E. coli mutants and applied a mathematical model of infection in an attempt to further classify the role of these novel interactions. This model-driven approach to biological discovery led us to identify the previously uncharacterized gene yneJ as a regulator of lamB gene expression. In addition, we identified two highly conserved pathways involved in post-transcriptional modification of tRNA—one pathway was required for efficient lambda replication, while the other has anti-viral properties inhibiting lambda replication. This finding is important as it illustrates a new potential anti-viral strategy that could be applied broadly to other viruses.
| Viral infections present a deadly paradox: in spite of the apparent simplicity of the viral genome, the complexity of the infection process has, for the most part, thwarted our attempts to prevent or cure it. Viral infections pose a serious threat to populations in both developing and developed countries. Additionally, viral infection is a serious problem for the bioprocessing industry, threatening production of items ranging from food to pharmaceuticals [1]. Increasing our understanding of viral infection would therefore have a major impact on human health, industry, and quality of life.
One resolution of this paradox is that the complexity of infection is not limited by the scope of the viral genome, but by the host machinery that the virus must commandeer in order to replicate. Recently, several genome-scale experimental studies have sought to identify these host-dependencies in viral replication. Research groups studying HIV [2]–[4], Influenza virus [5], [6], West Nile virus [7], Hepatitis C [8], [9], yeast virus [10], and T7 bacteriophage [11] have made use of newly constructed host knockout or siRNA knockdown libraries, in order to perturb the host and identify host dependencies. These forward-genetic screens have identified hundreds of host factors involved in viral infection and have provided a greater appreciation for the host's contribution to viral infection.
Today, the best-characterized model of viral infection remains bacteriophage lambda and its host—E. coli. Lambda is a temperate phage with two possible outcomes upon cell entry. In lytic growth (also known as productive growth), phage quickly replicate and lyse the cell, releasing new phage particles into the surrounding environment. In lysogenic growth, the injected phage DNA integrates into the attB site of E. coli genome and becomes a prophage [12]. The inserted prophage lies dormant until a later time, when upon induction the prophage genome excises itself from the host genome and begins productive growth. The determinants of lytic versus lysogenic growth appear to depend on several factors, such as multiplicity of infection [13], temperature [14], [15], and host cell physiology (e.g., nutrient state and size) [16], [17]. The lambda-E. coli system has also been a central player in elucidating and helping to understand host-virus interactions. Many genetic screens have been used to understand the infection phenotypes of different virus and host mutants [18]. These studies have greatly increased our knowledge pertaining to viral infection.
In this study, we focused on determining the interactions between E. coli and lambda phage during the infection process. We began with a forward-genetic screen to identify the E. coli genes whose absence results in a significantly reduced infection by phage lambda. We then performed higher resolution measurements of the infection time course for each gene and used a combination of bioinformatics and mathematical modeling in an effort to more rapidly identify likely roles in the lambda lifecycle.
Our screen to determine E. coli genes involved in lambda phage infection made use of the “Keio Collection”, an in-frame single-gene knockout strain collection, which contains 3,985 strains corresponding to all the genes which are non-essential during growth in rich medium [19] (see Figure 1A). We grew each knockout strain, as well as the “wild-type” K-12 MG1655 strain (K-12 WT), together with lambda phage on an agar plate with nutrient broth (NB) and 24 hours later assessed the resulting plaque morphology. In the first pass, 152 knockout strains appeared to affect lambda replication efficiency, producing either no visible plaques or smaller plaques relative to K-12 WT. All of the strains that appeared to inhibit phage replication, along with an additional 88 strains that were difficult to assess—primarily due to lawn defects—were considered further in two replicate experiments.
In all, 57 strains were identified with significantly different plaque morphology from K-12 WT (see Figure 1B and Table S1). The genes corresponding to these strains fall into three categories with respect to annotation: (i) genes with a known role in lambda infection, (ii) well-characterized genes whose products had nevertheless not been implicated in lambda infection, and (iii) unannotated genes.
The first group—well-known E. coli genes involved in lambda infection—included 19 genes. These include genes involved with lambda transport including lamB, which encodes a membrane protein required for the phage to bind E. coli [20], as well as transcriptional regulators of lamB: malT [21], malI [22] and cyaA [23], [24] (see Figure 2A). The inner membrane transporter manZ is part of a mannose PTS permease that is thought to be used by lambda phage to transport its genome into the cytoplasm [25], [26]. cyaA plays a dual role in lambda infection as it regulates expression of lamB and is involved in the lysis-lysogeny decision [24], [27].
In addition to cyaA, other genes involved in the lysis-lysogeny switch were found, including the proteases encoded by hflC, hflK and hflD. The FtsH-HflKC complex contributes to E. coli lysis by degrading the cII lambda transcription factor [28]. HflD directly interacts with cII, facilitating its degradation as well as disrupting its DNA binding ability [29], [30]. ihfA and ihfB were also found in the screen, and their gene products form a complex called integration host factor (IHF), which has been shown to induce sharp bends in DNA and is required for the integration of the prophage into the E. coli genome [31]. IHF has also been shown to play a role in phage DNA maturation [32].
The chaperone DnaJ contributes to a complex that works to destabilize the lambda P-DnaB complex bound to the ori site, thus allowing the DNA to be unwound and replicated [33]. The antiterminator gene nusB is known to play an important role in transcriptional dynamics of phage infection [34].
gmhA, gmhB, hldD, hldE, rfaC, rfaF, rfaH, and rfaP are all involved in the synthesis of the inner core of lippopolysaccharide (LPS) and were shown to affect the ability of phage to infect each strain (see Figure 2B). ADP-L-glycero-ß-D-manno-heptose is a key precursor to the inner core of LPS and is synthesized from D-sedoheptulose-7-phosphate by the gene products of gmhA, gmhB, hldD, and hldE. The heptose is then covalently bound to the KDO-Lipid A region by rfaC. All of the genes in the pathway, up to and including rfaP (with the exception of rfaG, whose activity may be replaced by rfaJ), showed reduced infectivity in our screen. This corresponds to the point in the pathway where the first heptose is phosphorylated. These eight LPS biosynthesis genes represent ∼73% of the genes found in a recent screen for E. coli genes required for T7 infection [11].
Previous work has shown that K-12 strains with varied LPS compositions have altered lambda receptor activity [35], [36]. Two possible explanations of this observation have been suggested. One study showed that mutations for LPS synthesis reduce the number of surface LamB proteins [35] while it has also been suggested that altered LPS configuration leads to non-optimal arrangement of LamB receptors, by analogy to the T4 phage receptor OmpC [36], [37]. These explanations are not mutually exclusive.
The second group of E. coli genes identified in our screen consisted of 34 genes with known functional roles but no previous link to lambda phage infection. Interestingly, the largest functional category in this group appears to be metabolic. Several of the identified genes play a role in central metabolism, including several key enzymes (pgi, pgm, atpA, talB, fucA), transporters (crr, lamB, manZ), and regulators (cyaA, malT, malI, fruR, bglG, see Figure 2C). Furthermore, the glucokinase encoded by glk was not included in our initial list but likely inhibits lambda replication as only one relatively small plaque was observed in our screen.
iscS, tusA, tusE, and mnmA were all found in our initial screen and are involved in the thiolation of tRNA [38]. iscS is necessary for thiolation of all nucleosides while tusA, tusE, and mnmA gene products contribute specifically to 5-(carboxy)-methylaminomethyl-2-thiouridine modification of tRNAGlu, tRNAGln, and tRNALys.
Several other interesting genes emerged in the second group. spr is an outer membrane lipoprotein, which for certain mutants, show thermosensitivity [39]. Intriguingly, mutants of nlpI, also found in our screen, have been shown to suppress this sensitivity [40]. pepA expresses aminopetidase A/I, which is involved in gene regulation, maintaining plasmid monomers, and preventing plasmid trans-recombination [41]. RlmE is the methyltransferase responsible for methylating U2552 of 23S rRNA [42]. Mutants for rlmE show reduced growth rate, protein synthesis activity, [43] and can modulate translational accuracy [44].
Four genes found using the screen are essentially uncharacterized. The structure of YbeD has been determined and may play a role in lipoic acid biosynthesis or the glycine cleavage system [45]. Global bioinformatic analysis revealed that YneJ has a LysR-type DNA binding domain and may therefore be a transcription regulator [46]. YfiM might act as an ABC transporter [47] and yecR has appeared in a computational screen to identify genes regulated by the flagellar master regulator FlhD2C2 [48].
We wondered how best to characterize the functional roles of new genes implicated in lambda infection. One weakness of the previous screens for host gene requirements in viral infection is that the information produced by these screening experiments is highly limited, generally involving only a few measurements per gene.
We hypothesized that higher time-resolution monitoring of infection dynamics would facilitate validation and further characterization of the roles these host genes play in the lambda phage lifecycle. We monitored E. coli growth and lysis over a full course of lambda infection in liquid culture (see Figure 3A). Infected K-12 WT bacteria grow exponentially for about 3 hours, after which the rate of bacterial lysis briefly outpaces growth. During this time, many phage have also induced lysogeny in their E. coli hosts, which then become resistant to further lytic infection. The lysogenic strains eventually take over the culture, growing exponentially until stationary phase.
We monitored growth of infected and uninfected cultures for all 57 strains that showed reduced infectivity in our plaque assay. We found that the infection dynamics varied significantly between strains. Figure 3B shows selected time course data, normalized by growth rate and maximum carrying capacity (i.e., optical density at stationary phase) to highlight the difference between infected and uninfected strains (The infection time courses for all 57 strains can be found in Figure S1).
We observed that genes with directly cooperative roles in lambda infection often exhibited very similar growth and clearance dynamics. As a simple example, Figure 3C shows the infection time courses for ΔlamB, ΔmalI, and ΔmalT. As mentioned above, all of these genes work together—malT and malI regulate the expression of the lamB transcript and functional protein product.
This observation suggested that similar infection time courses between knockout strains might be an indicator that the corresponding gene products act together in a pathway required for lambda infection. We performed agglomerative hierarchical clustering on the processed time course data to help identify knockout stains with similar infection dynamics (see Figure 4). To reduce the effects of varying growth rate between strains and to focus on the key transitions in the infection time courses, we pre-processed our data to obtain a normalized time course of the infection curve derivative for each strain (see Materials and Methods).
In all, we identified 18 separate clusters, separated primarily by: (i) the time point at which cell lysis outpaced cell growth, (ii) the length of time to achieve maximum clearance, and (iii) the re-growth or lysogenic growth rate. As an example, Clusters 17 and 18 show similar dynamics for properties (i) and (ii), but Cluster 17 shows limited secondary/lysogen growth. Genes in clusters with limited secondary growth are particularly interesting because they may play a role in lysogenization, as exemplified by Cluster 9, which contains hflC and hflK—two known regulators of lysogenization. However, this interpretation comes with the caveat that the secondary growth occurs under slightly different nutrient conditions than the primary growth phase.
Two major clusters bookended the dataset. At one end, Cluster 18, which was largely comprised of K-12 WT samples, showed rapid growth and lysis followed by robust secondary growth. Interestingly, two genes, yfiM and rutE, which were identified in our initial screen and consistently showed small plaques in the plaque assay, clustered with this group.
The largest cluster, Cluster 3, falls at the opposite end of the spectrum from the K-12 WT cluster and is characterized by little or no lysis over the entire time course. This cluster includes many of the previously known E. coli genes whose absence prevents lambda infection, such as lamB. There is a strong correlation between the plaque assay and infection time course as eight (ΔdnaJ, ΔyneJ, ΔmalI, ΔmalT, ΔlamB, ΔmanZ, ΔrfaF, and ΔrfaC) of the nine strains that had zero plaques in all trials fell into this cluster (see Table S1). The one exception corresponds to the outer membrane lipoprotein named spr (Cluster 15). Of the remaining genes in this cluster, Δtpx, ΔihfB, ΔgmhA, and ΔhdlD showed zero plaques in the initial screen and small plaques in the validation plaque assays. The two remaining genes in this cluster, Δcrr and ΔihfA, produced small plaques in each plaque assay and very little clearance in the liquid culture assay.
Several other genes involved in the same biological pathway were also found in the same cluster, for example, hflC and hflK (Cluster 10); lamB, malT, and malI (Cluster 3); and many of the LPS pathway genes (Cluster 3). Interestingly, pgi and talB (Clusters 14 and 15) both catalyze formation of fructose-6-phosphate and demonstrate similar infection dynamics. However, some genes involved in shared pathways did not cluster together. For example, ΔpdxH and ΔpdxA showed very different dynamics and clustered at opposite ends of the clustergram.
The finding that strains from multiple pathways all exhibited similar dynamics in Cluster 3 was intriguing, and we wondered if we could further discriminate between members of this cluster. We monitored the total E. coli concentration during infection at higher (10- and 100-fold) multiplicity of infection (MOI) for each strain in the cluster (see Figure 5A and Figure S2). After normalization, we compared the difference in the population between the uninfected and infected samples (see Figure 5B). We found that the higher MOI values appeared to have no effect on several strains, including ΔlamB, ΔdnaJ, ΔmalI, ΔmalT, ΔnusB, Δtpx, and ΔyneJ. tpx and yneJ are particularly interesting as their roles in lambda phage infection are entirely unknown. Those strains that did appear to be affected by higher MOIs corresponded to the genes involved in biosynthesis of inner core LPS, ihfA, ihfB, and manZ.
At the higher MOIs, the genes involved in LPS synthesis appeared to separate into two groups based on peak times (see Figure 5C and 5D). With the exception of gmhB, the genes required for attaching heptose to the KDO group had nearly identical peak times. Others have shown that ΔgmhB does not produce an entirely heptoseless form of LPS and conclude that there is another phosphatase that can catalyze this reaction [49]. ΔrfaF, ΔrfaP, and ΔrfaH had peak times much closer to K-12 WT.
We found the identification of unannotated gene yneJ to be of particular interest. Infected ΔyneJ showed no visible plaques and growth dynamics nearly identical to uninfected samples. We decided to apply mathematical modeling to interpret our infection time courses and help direct our efforts in characterizing the role of yneJ in lambda phage replication.
The population level interaction of phage with bacterial hosts has previously been phrased as a predator-prey system of differential equations [50], [51]. Following these previous efforts, we constructed a model that defines three populations as concentrations: uninfected bacteria, lysogens, and infectious phage (see Figure 6A). We then considered the effects of three key parameters on infection dynamics: (i) the burst size b, meaning the average number of infectious phage released upon host cell lysis, (ii) the fraction f of infection events that proceed down the lytic pathway, and (iii) the rate ki at which infection occurs. We found that we could recapitulate the infection time courses we observed simply by varying the parameters in our model (see Figure 6B, compare to Figure 3B).
To determine how model parameters could create different phenotypes, we simulated an infection time course for five levels of each of these three key parameters (125 simulations in total). The parameter combinations led to simulations that strongly resemble virtually all of the experimental time courses we observed (see Figure S3). We clustered the derivatives of the simulated time courses and found that the variation between simulations likewise resembled the variation between experimental strains (see Figure 6C and Figure S4). The one significant exception to this observation was that many of the simulations actually led to enhanced phage infection.
Importantly, the same types of variation that were found in the experimental time course cluster were also found in the simulated data cluster. We therefore wanted to determine which parameter combinations contributed to which type of variation. We found that the variation in f primarily contributes to the re-growth of the lysogenic population (see Figure 6C, right). In contrast, there is some correlation between b and ki (see Materials and Methods), but we found that the product b*ki varies inversely with the time at which lysis outpaces cell growth (see Figure 6C, far right).
This relationship between the parameters b and ki undermines the hypothesis that strains with a similar infection time course are involved in the same pathway, because multiple parameter combinations can lead to identical time courses. For example, the no-infection phenotype exhibited by members of the lamB cluster can be created computationally by setting f, b, or ki to very low values (see Figure 7A for the simulations and Figure 3C for representative data). However, the computational scenarios that produce equivalent total E. coli concentration time courses (see Figure 7A) create significantly more variation in the lambda phage concentration time courses (see Figure 7B). Specifically, the situation when lambda cannot infect the host at all (equivalent to setting ki to zero) leads to a stable concentration of phage over time, and when no viable phage are produced by infection (a low or zero value for b), the phage concentration is steadily reduced over time. In the third scenario, viable phage is produced at a rate exceeding its absorption (initially) and thus accumulates, however rate of production is very low (low f, ki, or b) and insufficient to produce host lysis on an observable scale. Accumulated phage populations in the latter case are orders of magnitude lower than observed for wild type.
To test these model predictions, we compared them to phage concentration time courses that we generated for members of the lamB cluster (see Figure 7C). We found that the simulated time courses compared favorably to our experimental data, in that infection of the lamB strain led to a nearly constant phage concentration, as would be expected if phage coul;d not infect the host. We also found that infection of the dnaJ strain, which is necessary for replication of lambda DNA, led to a decrease in phage concentration. In contrast, the ΔmanZ, ΔmalI, and ΔyneJ strains all produced significant amounts of phage over time, albeit less than the K-12 WT.
Our results of the phage concentration time courses suggest that the effect of ΔyneJ on lambda infection is within the low f, ki, or b range. To investigate if the loss of yneJ might affect the lytic-lysogenic decision (f) or the relative infection rate (ki), we performed single cell analysis of K-12 WT and ΔyneJ cultures infected by GFP-expressing lambda phage (see Figure 8A). We observed a large decrease in the percentage of ΔyneJ cells initially infected relative to K-12 WT (see Figure 8B), while lysis occurred at approximately the same time relative to K-12 WT. In addition, K-12 WT demonstrated a very low frequency of lysogeny, where only 2 of 27 infected cells appeared to become lysogens. We did not observe any lysogens among the 10 infected ΔyneJ cells. Based on these results, it does not appear that the reduced infectivity of lambda phage in ΔyneJ is due to a high frequency of lysogens (low f), but more likely, a reduced infection rate (low ki). This is consistent with ΔyneJ showing similar dynamics to other knockouts that are known to affect infection rate (i.e., ΔmanZ, ΔmalI, and ΔmalT) for both the infection and phage concentration time courses.
We hypothesized that similar to MalI and MalT, YneJ might play a role in the regulation of lamB. To test this hypothesis we used quantitative real-time RT-PCR to examine lamB mRNA expression in K-12 WT, ΔlamB, ΔmalT, ΔmalI, and ΔyneJ cultures. We found that ΔyneJ had a significantly reduced expression of lamB (see Figure 8C). lamB mRNA levels in ΔyneJ were comparable to levels found in ΔmalI mutants.
Sulfur used for the thiolation of tRNA nucleosides in E. coli is first transferred to IscS. Depending on the tRNA and the particular base modification there are two distinct pathway classes: an iron-sulfur ([Fe-S]) protein-independent and a [Fe-S] protein-dependent pathway [52]. Our plaque assay results found several genes in the [Fe-S] independent pathway responsible for thiolation of U34 for tRNAGlu, tRNAGln, and tRNALys. In this pathway, sulfur is transferred from IscS to TusA. TusB, TusC, and TusD form a heterotrimer complex where Cys78 of TusD is able to form persulfide, facilitating transfer of sulfur from TusA to TusE [38]. Sulfur is then passed from TusE to MnmA, which thiolates uridine 34 of the tRNA. Our plaque assay screen showed reduced infectivity for all strains in this pathway with the exception of ΔtusB, ΔtusC, and ΔtusD. For ΔtusB, ΔtusC, and ΔtusD, we do observe a slight reduction in efficiency of plating and the plaques are generally smaller then the K-12 WT controls (data not shown).
In the [Fe-S] dependent pathway, sulfur is transferred from IscS to IscU. This transfer is facilitated by several proteins, including IscA, Fdx, and the chaperone/co-chaperone pair HscA and HscB. As it is thought that IscU binds to IscS, potentially competing with TusA for sulfur, we looked more closely at the proteins in the [Fe-S] dependent pathway. We found that ΔiscU, ΔhscA, and ΔhscB showed significantly larger plaques compared to K-12 WT (data not shown).
In addition to the four tRNA genes identified in the original screen, we examined the infection dynamics for ΔtusB, ΔtusC, ΔtusD, ΔiscU, ΔhscA, ΔhscB and a ΔtusBCD strain (see Figure 9 and Figure S5). The ΔtusB, ΔtusC, and ΔtusD strain infection dynamics were indicative of reduced infectivity, showing a significant delay in the peak. ΔtusBCD had very similar dynamics to the individual knockouts for tusB, tusC, and tusD. The infection dynamics for ΔiscU, ΔhscA, and ΔhscB cleared relatively quickly and had very little secondary growth.
In summary, to determine the E. coli gene requirements for lambda infection, we performed a screen of 3,985 non-essential gene knockout strains and found 57 strains with impaired lambda infectivity. In addition to identifying many genes with established roles in lambda phage infection, we found a surprising number of previously unassociated genes, four of which are currently unannotated. In addition, several of the genes found in our screen have human orthologues (see Table S2).
Central metabolism was the largest shared functional category among our results. In particular, we found many genes involved in multiple entry points to and regulators of glycolysis. This observation is in part related to lambda phage's dependence on LamB and the factors that regulate lamB expression. Several other metabolic genes identified in our screen are known to play a role in regulating the lytic-lysogenic decision. Based on this observation, it is tempting to speculate that many of these genes may influence lysogeny, whether by sensing or altering the cell's nutritional state. Our results potentially expand this lytic-lysogenic network and provide a framework for a deeper understanding of this decision as well as the metabolic requirements for lambda phage replication. It is interesting to note that many of the host-dependencies identified in the recent mammalian virus screens are also involved in energy metabolism and enzymes involved in amino acid and nucleic acid synthesis.
Second, we demonstrated that the knockout strains for genes with common roles in infection often showed similar infection dynamics. Some examples include the hfl, ihf, LPS biosynthesis and lamB-related genes. Our observation is complicated by at least three factors. One complication arises when the effect of gene deletion within a pathway varies depending on the step within the pathway, as in the LPS biosynthesis pathway. Another confounding factor is that some proteins may be partially redundant, where in the absence of one protein another can adequately compensate thus producing a mild or no change in phenotype. This may be the case for GmhB and an unidentified phosphatase. While we did not identify the complementary phosphatase in this study, one might look among the strains clustering with ΔgmhB. Finally, many proteins have pleiotropic effects and the observed dynamics may arise from the modulation of multiple pathways. An example of this is the protein CyaA, which has a metabolic function but also plays a role in the lytic-lysogenic decision. Notwithstanding these exceptions, dynamic infection data was extremely useful as a validation tool and a “first-step” in assessing host gene functionality in lambda infection.
We used a combination of computational modeling and further experimentation to characterize many of the strains in more detail. This was best exemplified by our efforts to characterize the previously unannotated gene, yneJ, and its apparent critical role in infection. We observed that ΔyneJ closely resembled the ΔmanZ, ΔmalT, and ΔmalI strains in terms of growth phenotype, lambda phage production, and response to higher phage concentration. Using single-cell imaging of GFP expressing lambda phage, we found that ΔyneJ is not immune to infection and does not appear to regulate the lytic-lysogenic decision. We suspected that yneJ may play a role in attachment or entry, possibly through the regulation of LamB or another unidentified membrane-associated protein and tested the expression levels of lamB in several strains including ΔyneJ. Our data demonstrated a reduced level of lamB mRNA in ΔyneJ similar in magnitude to ΔmalI. Given YneJ's LysR-type DNA binding domain, we speculate that YneJ is an upstream regulator of lamB transcription—possibly directly or indirectly regulating malI expression. Given its effect, it is intriguing that yneJ was not identified over the several decades of reverse-genetic screening of phage lambda.
Another surprising result of this work was the identification of two important pathways for lambda phage replication with a common node at cysteine desulferase IscS. Interestingly, inactivation of one of these pathways (including tusA, tusB, tusC, tusD, tusE, and mnmA) leads to impaired lambda infection, while inactivation of the other pathway (including iscU, hscA, and hscB) facilitates lambda infection.
Others have shown that the 5-(carboxy)-methylaminomethyl-2-thiouridine modification at the tRNA wobble position increases frame maintenance and prevents codon-specific frameshift in E. coli [53]. We hypothesize that the rapid and proper synthesis of lambda phage proteins is particularly dependant on this modified nucleoside. Viruses of infected cells lacking this modification likely show decreased production of functional viral proteins and thus produce a smaller number of infectious particles. In concept, this hypothesis resembles the lethal mutagenesis therapeutic strategy for treating viral infection [54], albeit at the protein level. Based on our data, it appears that 2-thiouridine modifications of tRNAGlu, tRNAGln and tRNALys are particularly important for the efficient production of infectious lambda phage particles. The s2U34 modification is conserved for several tRNAs across all organisms [55], raising the question of whether these genes or related pathways might present novel anti-viral targets for mammalian viruses.
Our screen was performed using the established plaque assay method [56]. K-12 WT (ATCC, 47076) and all strains of the “Keio Collection” were cultured overnight (∼16 hours) in nutrient broth (NB, 8 g/L, Fluka Analytical, N7519) using 96 deep-well plates (Nunc, #278743). Incubation was done at 37°C while rotating at 225 RPM. Each E. coli strain was incubated with bacteriophage lambda (ATCC, 23724-B2) for 30 minutes at 37°C—optimal titers for bacteriophage lambda were determined beforehand using a dilution series. 2% 2,3,5-Triphenyltetra-zolium chloride (Sigma, T8877) was added to NB top agar (0.04% final concentration) and incubated at 55°C for 30 minutes. The top agar was then added to the incubating E. coli and phage samples and 80 µl was plated onto 24-well agar plates. The agar plates were made the previous day with 500 µl of NB bottom agar for each well. Included on each plate were two replicates of infected K-12 WT and one well of ΔlamB. After 24 hours, the number of plaques generated by infection of each strain was counted and compared to K-12 WT counts. Wells with significantly fewer or smaller plaques for two replicate experiments were considered “hits”.
To monitor the E. coli concentration of an infected batch culture over time, we used an incubated plate reader (Perkin-Elmer Victor3, 2030-0030). Strains of the “Keio Collection” were inoculated and grown overnight (∼16 hours) in 2 ml NB in 5 ml round-bottom tubes. The next morning the samples were diluted 1∶100 in fresh NB media and grown for 3 hours. After 3 hours the samples were measured for absorbance at 600 nm and diluted in NB to 0.1 OD. 15 µl of 0.1 OD E. coli and 15 µl of ∼104 plaque forming units/ml (pfu/ml) lambda stock were added to 170 µl of NB in 96-well plates—representing an MOI of ∼2×10−4 pfu/bacteria. Four replicates for each low infectivity strain (infected and uninfected) were assayed. Strains ΔatpA, ΔthyA, and ΔiscS did not show significant growth rates over the time course, and therefore provided no additional information concerning phage infection. The “Keio Collection” was created using E. coli BW25113 as the background strain, so we compared the infection dynamics of this strain with E. coli MG1655 and found them to be essentially identical (see Figure S6).
Included on each plate were replicates of K-12 WT. The incubation protocol included an initial 10 minute shake (double orbital, 1.5 mm diameter, normal speed), followed by 38 cycles consisting of the following actions: a one second absorbance measurement at 600nm (Perkin Elmer, 600/8nm, 1420-521), 5 µl injection of milliQ water into each well (to counter volume loss due to evaporation), and a 10 minute shake (double orbital, 1.5 mm diameter, normal speed). The time course was performed at 37°C for approximately eleven hours.
The K-12 WT, ΔlamB, ΔmanZ, ΔmalT, ΔmalI, ΔdnaJ, and ΔyneJ strains were inoculated in NB and grown overnight at 37°C while rotating at 225 RPM. The next morning the samples were diluted 1∶100 in fresh NB media and grown for 3 hours. After 3 hours the samples were measured for absorbance at 600 nm and diluted in NB to 0.1 OD600. 375 µl of E. coli, 375 µl phage stock, and 4.25 ml of NB at 37°C were combined in 14 ml tubes. Aliquots were taken from each culture at 90, 173, 205, 236, 276, 322, 364, and 415 minutes post-infection and filtered using 0.2 µm PES filters (Nalgene, 180-1320). The filtrate was then diluted in SM + gelatin (0.058% (w/v) NaCl, 0.02% (w/v) MgSO4-7H2O, 50 mM Tris (pH 7.5), 0.01% gelatin). 100 µl of these dilutions were incubated with an equal quantity 2.0 OD600 E. coli in 10 mM MgSO4 for 10 minutes before combining with 1 ml NB top agar 500 µl onto 6-well plates. After incubating overnight, the plaques were counted to determine the pfu/ml at each time point.
A key challenge in our data processing was to determine a metric to compare two infection time courses. Observing that uninfected knockout strain time courses cluster by pathway based solely on the variation in background growth phenotype, normalization of infected growth curves was necessary in order to avoid clustering these based on the similar behavior of their bounding uninfected curves. To prevent clustering based on growth rate and the eventual population limit, we non-dimensionalized the growth time courses for each strain using parameters of logistic population growth fit to uninfected data (model presented, simulated as uninfected using initial condition [λ] = 0 at t = 0). The carrying capacity, K, was considered the characteristic population and inverse growth rate 1/μ the characteristic time. This treatment allowed us to compare the key features of the infected time courses relative to their background, for example allowing distinction between a strain that is overtaken by phage late due to generally weak/slower growth/metabolic function and one that is overtaken late despite quick/normal growth potentially indicating a more phage specific knockout impact. Although deemed necessary for the meaningful data interpretation via our method, anytime rescaling is performed to clarify certain dataset features, the potential exists to deemphasize or obscure other features of that data set, and we have thus presented our entire raw dataset in Figure S1. Data was smoothed using the robust lowess implementation in MatLab (window size 0.1 as a fraction of total data points), followed by analytical calculation of the time derivative from a cubic smoothing spline to avoid noise amplification. A cosine distance metric was used in clustering, 1-θ, where theta is the included angle between time course derivatives treated as vectors, emphasizing similarities in population change direction rather than magnitude. Agglomerative hierarchical cluster tree was constructed with average linkage, and clusters defined by a linkage of less than 0.16.
We constructed a model that defines three populations as concentrations: uninfected bacteria [E], lysogenically-infected bacteria [E*], and infectious phage [λ]. Bacterial growth is assumed to occur logistically with rate μ towards a limiting population [K]. We used analogous but separate parameters for lysogens (denoted by *) [57].
The ordinary differential equations are as follows:Mass-action kinetic infection of bacteria by phage, with rate ki, is well established [58]. Superinfection of lysogens occurs with no effect other than to reduce the infection phage population [59]. The amplification factor parameter b represents the average number of infectious phage released upon host cell lysis. Lysogen induction to the lytic pathway by DNA damaging stresses is included at rate ks for the sake of completeness, but is assumed negligible under our conditions and henceforth neglected. f is the fraction of infection events that proceed down the lytic pathway. Given the complex nature of the lambda decision and switch (and neglect of ks), f as a constant fraction is necessarily an average parameter containing information on the ease of lysogeny establishment and maintenance.
For the purposes of well-scaled numerical simulation the model was converted to a non-dimensional form using the same characteristic time and population as for the experimental time courses. Replacing the time with its product with μ, and dividing each population variable by K, both maximum growth rate and maximum population reduce to unity. The parameters f and b are retained as fraction and amplification factor respectively, while lysogen growth reduce to ratios to their analogous uninfected growth parameters, and ki becomes the dimensionless group kiK/μ. All simulations presented were completed using the scaled form, and values of parameters listed are the corresponding non-dimensional form or group.
All numerical integration was completed using the standard MatLab implementation of rk45, with default tolerances. The initial condition [E]t = 0 = 0.02 was chosen and used for all simulations based on values fit simultaneously as logistic kinetics were fit to experimental data (least squares distance), with [λ]t = 0 set using the experimentally determined MOI, and no initial lysogen population ([E*]t = 0 = 0). The simulation results for the E. coli population presented are the sum [E]+[E*], as these populations are not distinguished experimentally during our infection time course observation. Simulations were clustered using the same method as experimental data (though smoothing was unnecessary). Parameter values used in the simulations, displayed in Figure S3 and Figure S4, were all possible combinations of ki = {0, 0.25, 0.75, 2, 5}, f = {0, 0.5, 0.75, 0.95, 1}, b = {0, 10, 20, 50, 100}. Values for f and b were chosen as physically reasonable values. The set of values for ki were then chosen from the range that produced model behavior resembling our experimental observations. It should be noted that for a given infection time course simulation, a practically equivalent bacterial population time course can be generated using alternate combinations of the parameter ki and b values. The lysogen growth character was not varied, using constant μ* = K* = 0.7 for all simulations, both because our primary focus was on the character of the infection process that leads to the initial overtake of the bacterial population by phage, and also because at later times in batch culture it is possible that our experimental observations of lysogen growth character are dominated by nutrient limitation.
Exponentially growing cells were infected with lambda b::GFP kanR [17]. Adsorption of lambda (MOI≈0.5) was done at 4°C for approximately 20 minutes. 1 µl of sample was pipetted onto a 2% agarose pad. The agarose pad was made using equal volumes of 4% LMP agarose and 2× EZ-RDM (Tecknova, M2105) supplemented with 2% maltose, 20 mM MgSO4, and 1mM of IPTG to induce expression of the GFP construct. The pad was then inverted onto a cover slide and a plastic lid was placed on top to help prevent drying of the pad over the time course. The assembled slide was maintained at 37C during imaging. Both phase (30 ms) and GFP (50 ms) images were taken every 2 minutes at 60× magnification. Cells were manually assessed for infection.
Exponentially growing cells were treated with RNALater (Ambion, AM7020) according to the manufacturer's recommendations. The RNA was then isolated using Qiagen's RNeasy Mini Extraction kit (Qiagen, 74104). DNase digestion was performed on the RNA sample using Deoxyribonuclease I (Invitrogen, 18068-015). First-strand cDNA synthesis was performed (Invitrogen, 18080-051) followed by RNase H digestion. 100 ng of template cDNA and primers (200 nM final concentration) were combined with the SYBR Green PCR master mix (Applied Biosystems, 4367659). LamB (left primer, 5′- ATGAGCACCGTGATGGAAAT -3′; right primer 5′- AGCGTTACCGGTGTAGTCGT -3′) and rrsA (left primer, 5′- CGGTGGAGCATGTGGTTTAA -3′; right primer, 5′- GAAAACTTCCGTGGATGTCAAGA -3′) primers were run in quadruplicate for each strain. K-12 WT cDNA dilutions (2×, 4×, 10×, and 100×) were used to calculate the primer amplification efficiency.
The ΔtusBCD strain was constructed according to methods described elsewhere [60]. Primers were designed to knockout all 3 genes as one continuous fragment and replace it with the chloramphenicol acetyl transferase gene from pKD3. The upstream primer was 5′-TACATCCGCCAGTTCAAGAGCGGTGATTTCCAGGGGCAAGATAAGTAATGATGGGAATTAGCCATGGTCC-3′ and the downstream primer used was 5′-GTGTCAAGAAATATACAACGATCCCGCCATCACCAGGCCATCTGGCTGGGGTGTAGGCTGGAGCTGCTTCG-3′.
|
10.1371/journal.pcbi.1002540 | The CanOE Strategy: Integrating Genomic and Metabolic Contexts across Multiple Prokaryote Genomes to Find Candidate Genes for Orphan Enzymes | Of all biochemically characterized metabolic reactions formalized by the IUBMB, over one out of four have yet to be associated with a nucleic or protein sequence, i.e. are sequence-orphan enzymatic activities. Few bioinformatics annotation tools are able to propose candidate genes for such activities by exploiting context-dependent rather than sequence-dependent data, and none are readily accessible and propose result integration across multiple genomes. Here, we present CanOE (Candidate genes for Orphan Enzymes), a four-step bioinformatics strategy that proposes ranked candidate genes for sequence-orphan enzymatic activities (or orphan enzymes for short). The first step locates “genomic metabolons”, i.e. groups of co-localized genes coding proteins catalyzing reactions linked by shared metabolites, in one genome at a time. These metabolons can be particularly helpful for aiding bioanalysts to visualize relevant metabolic data. In the second step, they are used to generate candidate associations between un-annotated genes and gene-less reactions. The third step integrates these gene-reaction associations over several genomes using gene families, and summarizes the strength of family-reaction associations by several scores. In the final step, these scores are used to rank members of gene families which are proposed for metabolic reactions. These associations are of particular interest when the metabolic reaction is a sequence-orphan enzymatic activity. Our strategy found over 60,000 genomic metabolons in more than 1,000 prokaryote organisms from the MicroScope platform, generating candidate genes for many metabolic reactions, of which more than 70 distinct orphan reactions. A computational validation of the approach is discussed. Finally, we present a case study on the anaerobic allantoin degradation pathway in Escherichia coli K-12.
| The discovery of the various metabolic functions catalyzed by enzymes encoded by the genes from the exponentially increasing number of sequenced genomes is one of the main focuses of bioinformatics tools today. However, most of these tools rely on already identified enzyme-coding gene or protein sequence information to predict known enzymatic activities in new genomes. Therefore, they cannot be used to reveal metabolic activities without any corresponding sequenced genes, dubbed “sequence-orphan activities”. In such cases, the best approach is the bioanalysis of target genes by human expert curators, manually integrating so-called “context-based information” (such as gene co-localization on the genome, or the presence of incomplete metabolic pathways) to infer novel functions. Few bioinformatics tools exploit such information and render accessible results in an automated way. Here, we present “CanOE”, a strategy that uses contextual information to propose and rank Candidate genes for Orphan Enzymes in Bacteria and Archaea. Beyond the merit of extending our knowledge and comprehension of prokaryote metabolism, identifying coding genes for sequence-orphan activities opens new opportunities for functional annotation (homology-based transfer made accessible), drug design (new metabolic targets), synthetic biology (new building blocks) and biotechnology applications (new biocatalysts).
| Approximately 27% of all enzymatic activities recognized by the IUBMB [www.iubmb.org] are still sequence-orphan metabolic activities (dubbed “orphan enzymes” for short) in the UniProt databank [1], a number that has decreased slowly over the past years [2]–[4]. It would, of course, be too time-consuming and costly to conduct wet-lab experiments to test all known activities against all genes from the exponentially increasing number of sequenced genomes. Instead, bioinformatics tools have been developed in order to help annotate newly sequenced genes and to guide biologists in selecting the right candidate genes for further experimental testing. These tools can be classified into two types: 1) those transferring existing annotations between genes belonging to different organisms on the basis of detected homology (inferred using clues such as high sequence similarity, domain conservation, or feature-based similarities), and 2) those using “context-based” methods capable of inferring functions from existing gene annotations in the same organism, on the basis of detected functional dependence (inferred from clues such as those presented in the following paragraph). Due to the lack of any sequence data, tools based on sequence similarity detection cannot be used to solve the “orphan enzyme” problem, and research has turned to context-based approaches.
Various indicators of prokaryote genes being functionally dependent have been devised in the literature. The foremost of these are collectively termed as “genomic context”, and include gene clustering [5], phylogenetic profiles [6], and gene fusion/fission [7], [8]. “Metabolic context”, for its part, refers in an informal way to the sum of all metabolic knowledge for the genes of a given genomic context. Many ways of exploiting these contextual indicators have been imagined. Manual integration of diverse comparative genomics data sources by expert bioanalysts is an obvious approach, formalized (amongst others) in [5] and [9]. Such strategies have since been put into application in various bioinformatics platforms such as IMG [10], MicroScope [11], 12, the SEED [13] and ERGO [14].
Only a few tools based on these context-based methods have been developed over the past decade with the specific goal of solving the “orphan enzyme” problem. The PathwayHoleFiller-GenomicContext [15] is an improvement over a previous method [16] that allows genomic context similarity measures (gene neighbors, gene clusters, gene fusion, or phylogenetic profile methods, see [17]) as well as metabolic context to be taken into account in a Bayesian classifier. ADOMETA [18] uses various scores (based on gene co-expression, phylogenetic profile similarity, gene clustering, and protein interaction data) integrated using a simple likelihood approach to fill in the missing reactions for three organisms having specifically reconstructed metabolic networks. Yaminishi et al. [19] use a kernel approach to integrate two data sources (gene proximity and phylogenetic profiles) to build a global network onto which they project an organism's known reaction set. They then search manually for candidate genes corresponding to orphan reactions based on their operon-like results. Chen et al. [20] combine gene sequence similarity and gene proximity across many genomes to establish path-based scores as a functional dependence measure, which is then used to rank candidate genes for pathway holes, including orphan enzymatic activities. Other resources can be exploited manually for finding candidate genes for orphan enzymes using context-based functional dependency measures, such as the STRING [21].
Inspired by the modus operandi of human expert research conducted at the Genoscope [22]–[24], we have developed CanOE (Candidates for Orphan Enzymes), an automated strategy that exploits genomic and metabolic contextual information by a graph-based algorithm. This strategy has been integrated into our in-lab genome annotation platform, called MicroScope, and uses its set of expert curated annotations as input, with the objective of improving the reconstructed metabolic networks from the MicroCyc component of the platform [11], [12]. Its results are available via a web interface at the following URL: http://www.genoscope.cns.fr/agc/microscope/metabolism/canoe.php
The principle of our strategy lies in the continuity of previous works [25]–[27]. A first step involves searching for groups of genes corresponding to groups of reactions participating in a same metabolic process. This is done by looking for groups of adjacent genes encoding enzymes catalyzing connected reactions, allowing for gene and reaction gaps. We called the functional units thus identified “genomic metabolons” (in reference to biological metabolons [28]), and they form the basis for the proposition of potential associations (i.e. hypothetical annotations) between gene gaps and reactions gaps in the second step. The third step integrates known and potential associations over all available genomes by building gene families and calculating family-reaction association scores. Finally, these scores are used to rank candidate gene-reaction associations. This is particularly interesting when a reaction gap actually corresponds to an orphan enzymatic activity, but can also be used as additional support when transferring annotations on the basis of limited sequence similarity. In this article, we detail the strategy's primary data and operational steps, as well as the evaluation of the performance of our association scores with a benchmarking test. We present a biological case study showing the usefulness of our approach, and finally highlight in which ways our strategy sets itself apart from previous methods.
The first step of our strategy requires three types of input data: 1) a gene graph, modeling gene contiguity in a target genome; 2) a reaction graph, modeling the global (i.e. pathway- and organism-independent) metabolic network we wish to work with; and 3) the set of all already-known gene-reaction associations, i.e. all current functional annotations in the target genome. These data sources are detailed hereafter.
Gene graphs are built separately for each genome from the MicroScope database (1117 available at the time of writing). The gene graph represents all protein-coding genome features (“genes” here) of a single prokaryote organism as vertices. In this work, we use gene contiguity as an indicator of functional dependence. Immediately consecutive genes are thus connected by edges, ignoring gene transcription direction and intergenic distance, as bidirectionally-translated operons have been observed [29], [30]. We thus are independent of operon, regulon, stimulon and über-operon definitions [31], though our metabolons will be still able to capture some parts of such structures.
The reaction graph represents metabolic reactions as vertices. We link two reactions by an edge when the product of one is a substrate of the other. However, to avoid the high connectivity problems that are common when building such metabolic networks, we limited such shared compounds to “main compounds”, i.e. metabolites deemed biologically relevant to both reactions in at least one metabolic pathway (for example, phosphoenolpyruvate, but not water, in the glycolysis pathway). Main compounds are arbitrarily defined as such by biochemists on the basis of atom-tracing experiments, molecular structure conservation, or other data. The modeled reactions were extracted from MetaCyc 15.0 [32], but any other generalist metabolic database (preferentially one containing main compound data) can be used (e.g. Rhea [33], KEGG reactions [34]…). The metabolic network is global, as it contains all known metabolic reactions and is not split into separate, disconnected pathways. It is thus not organism-specific, guaranteeing maximal metabolic freedom.
Finally, we retrieved functional annotations from the MicroScope platform (and in the case of Escherichia coli K-12, additionally from EcoCyc [35]) to benefit from its high level of expert manual curation. The MicroCyc component of MicroScope gathers a set of Pathway Genome DataBases (PGDBs) which were built using the Pathway Tools software [36] and with the MetaCyc database [32] as a reference. Gene-reaction associations are extracted from these PGDBs and used to link elements from the gene graph to those of the reaction graph. This creates a new graph, called the “data graph”, which has two types of vertex (genes and reactions) and three types of edge (gene-gene edges, reaction-reaction edges, and gene-reaction edges). The previously described gene-reaction associations are flagged as “Known”, as they correspond to the current state of biological knowledge. The metabolic network is thus populated specifically for each organism by reactions known to be catalyzed within them. It should be noted that multiple reactions can be linked to a same gene (e.g. bi-functional genes or enzymes with wide substrate specificity), and conversely, multiple genes can be linked to a same reaction (e.g. enzymes with several subunits). Details on graph construction can be found in [Text S1].
Two kinds of “reaction knowledge hole” can be formalized in metabolic networks [37]. The first kind is the gap reaction, i.e. a missing reaction in an organism-specific metabolic network reconstruction whose presence appears necessary for the network to be complete (without spurious dead-end metabolites). Basically, no experimental results necessarily confirm its presence in an organism, but metabolic context within the organism suggests it. The other kind of “reaction knowledge hole” is the orphan reaction, i.e. an enzymatic activity thought to be present in an organism (preferably with experimental evidence) but without any known coding genes. Reactions can be orphans in a specific target organism (local orphan), or for all known organisms (global orphan). In this article, we work exclusively with prokaryote organisms from the MicroScope platform; a reaction is thus considered as a global orphan when it has no known coding genes in any of the platform's prokaryote organisms (even though it may have coding genes in eukaryote species). In an organism-specific metabolic network, global orphan reactions (if present) may appear as gap reactions. On the other hand, gap reactions may be either local or global orphan reactions.
The CanOE strategy will first detect potential gap reactions by computing metabolons in a global metabolic network populated by reactions known to be catalyzed in the target organism. It will then propose candidate genes for these gap reactions, be they local orphan reactions or global orphans across all of MicroScope's genomes.
In our benchmarking experiment, we considered the set of all metabolic reactions having at least one Known gene-reaction association involved in a metabolon (since the method does not make predictions for genes and reactions not involved in metabolons). For each reaction from this set, we removed all the gene-reaction associations involving that reaction in all organisms (effectively rendering it a sequence-orphan reaction), and recalculated all gene- and family-level association scores, for which we consider the rank of the genuine gene-reaction associations. Results were pooled across all reactions from the set.
A recovered gene-reaction association is considered as a positive hit when its rank (according to a chosen score) is below a certain threshold k. All recovered associations can be declared as positive hits by taking k = ∞.
For a level of k, we defined true positive associations (TP) as the number of genuine gene-reaction associations that were recovered in the experiment in respect to the original CanOE run, false negative associations (FN) as those that were not recovered, and false positives (FP) as Potential associations that were proposed that did not correspond to Known associations in the original run. We then classically compute the recall (or sensitivity) as the fraction of recovered associations (TP/(TP+FN)) and the precision as the fraction of correctly predicted associations (TP/(TP+FP)).
In order to gauge how indicative our family-reaction descriptors are of gene-reaction association strength, we examined the evolution of the recall and the precision while varying the rank threshold k (i.e. keeping only the k-best associations for each gene), thus generating a precision-recall curve for each score.
The genomes of 1,090 prokaryotes from the MicroScope platform produced a total of 61,670 metabolons, leading to an approximate average of 57 metabolons per organism (see [Figure S2]). E. coli K-12, at 105 metabolons, is comparatively rich. A brief analysis showed that 78 of these metabolons (74%) shared at least two genes with operons defined by RegulonDB [38]. All in all, the density of genomic metabolons is consistent with previous findings, given the current state of functional annotation amongst bacterial genomes [25].
These organisms contained a total of 4,646,851 genes, of which 1,088,330 (32.6%) had metabolic annotations (i.e. genes coding for enzymes represented in the MicroCyc database). The metabolons themselves covered 215,968 of these genes (19.8%). When considering the well-annotated genome of E. coli K-12, 1,441 out of 4,414 genes (30.7%) were annotated with metabolic activities, and 399 of these (27.7%) were in a metabolon. The per-organism gene coverage of the metabolons varies between 2.5% and 7.5% as shown in [Figure S3]. The distributions per phylum are given in [Figure S4]. The genes were grouped into 8,629 gene families by our clustering method, of which 616 (7.1%) were declared non metabolic.
Our local installation of the MetaCyc database (version 15.0) contains 9,531 reaction instances. Using pathway-specific main compound definitions, reaction-reaction edges were added between these. A total of 5,157 reactions were connected in this way in our global metabolic network. Of these, an impressive 2,839 (55.1%) are sequence-orphan activities across all MicroScope prokaryote genomes. 1,626 (31.5%) reactions were found in at least one metabolon, showing that the coding genes of almost two thirds of available reactions are never clustered throughout prokaryote genomes, and thus cannot be captured by metabolons based on simple gene neighborhood. Furthermore, only 104 (6.4%) of these were global orphans, showing that well-known metabolic reactions are generally surrounded by other well-known reactions. Finally, at the time of writing, 72 of these had potential gene candidates, and 50 of these had candidate genes belonging to a gene family, allowing the calculation of the family-level association scores. Only one prokaryote orphan reaction was found with candidate genes in E. coli K-12, and is described in the case study section. The list of all proposed candidate genes for all found global orphan enzymatic activities is available from the CanOE main web page by simply selecting the “Consult global orphan reactions” radio button and clicking “Go!”. A manual bioanalysis of these cases is summarized in [Text S6]. We determined that 31 of the global orphan reactions may have plausible candidate gene predictions. Of particular interest, 20 of these have good CanOE-independent supporting evidence (e.g. circumstantial literature, sequence similarities with enzymes of related reactions, predicted domains…).
To evaluate the CanOE strategy in a systematic way, we undertook a benchmarking experiment allowing us to a) quantify how well Known gene-reaction associations were recovered after having been removed from the input data, and b) show how informative our gene- and family-level scores are. 1,469 of the MetaCyc reactions that had been observed in metabolons were sequentially rendered orphan in our benchmarking experiment.
The global (i.e. for a rank cut-off of k = ∞) gene-level recall is 80.5%, meaning that over four out of five Known gene-reaction associations that were removed by the orphan experiments were successfully recovered. The global gene-level precision is 45.2%, meaning that a little less than half of all Potential associations generated in the experiments are indeed true associations. At the family level, the global recall is 80.6% and the precision is 45.4%, showing a very slight global improvement imputable to family-wise association redefinition.
However, more importantly, we show in the precision-recall curves of [Figure 3] that the family-level ScoreR→F rank outperforms the gene-level ScoreR→G rank. We can observe that the precision can be increased with minimal impact on the recall by keeping no more than the best 3 or 4 candidates according to the former, when the recall is more heavily impacted for less precision improvement in the latter. Individual TP, FP, FN counts can be found in [Table 1, Table 2]. This illustrates a definite advantage of exploiting integration over all available organisms to help rank proposed gene-reaction associations, even if the maximal precision and recall values (obtained when considering all propositions) hardly differ. The precision-recall curve and its associated tables for ScoreF→R and ScoreG→R can be found in [Figure S5, Table S1, Table S2], showing that it is also informative, though with a lower performance. To be comparable to self-rank tests as in [42], we show the fraction of recovered TP associations as a function of a maximal rank cutoff in [Figure S6]. Over 99% of TPs are found amongst the first 5 ranks, outperforming ADOMETA [18], consistent with results obtained by Chen et al. [20], though possible ranks go up to around 80.
We conclude that these descriptors are sufficiently informative for use by biologists, biochemists and bioanalysts in determining which candidates are the best bets to test experimentally when considering potential annotations with orphan reactions.
To illustrate the usefulness of our method, we investigated a predicted metabolon in E. coli K-12 containing a prokaryote orphan reaction with candidate genes [Figure 4]. This metabolon is composed of 6 reactions covering the complete pathway for the anaerobic degradation of allantoin, in which two reactions are orphans in E. coli according to the EcoCyc resource [35]: oxamate carbamoyltransferase (OXTCase, global prokaryote orphan) and carbamate kinase (CKase, local orphan). In the CanOE metabolon [Figure 4], the CKase is shown to be catalyzed by the ECK0514/ybcF gene: this association is absent from EcoCyc, despite the latter being a heavily-curated resource, but is supported by the MicroScope annotation of this gene which shares more than 50% amino acid identity with an experimentally-validated CKase from Pseudomonas aeruginosa (P13982 UniProt entry). This first point demonstrates how the CanOE strategy can aid bioanalysts to confirm putative annotations for local orphan reactions by automatically mining the wealth of a metabolic context.
The second missing activity in E. coli K-12 (the OXTCase) has yet to be associated to any genes in any organism and is thus a global orphan activity, despite its presence having been biochemically demonstrated in Streptococcus allantoicus [43], and even reported in E. coli [44], [45]. The CanOE metabolon bearing this reaction [Figure 4] contained 5 gap genes (ECK0506-507 and ECK0511 to 0513) that could serve as candidate genes. The first one, ECK0506/ybbY, belongs to an Intepro family defined by the presence of a permease domain and is annotated as a putative purine permease according to the UniProt resource [1]. This gene was declared non-metabolic by CanOE and thence not considered as a potential candidate for the OXTCase activity. However, the purine permease function is quite consistent with trans-membrane transport of allantoin or another intermediate of purine metabolism, of which allantoin degradation is a part.
The second gene, ECK0507/glxK, was experimentally demonstrated to encode a glycerate kinase involved in the aerobic degradation of allantoin via glyoxylate metabolism [45]. This gene was a non-gap member of a neighboring CanOE metabolon (genes ECK0500 to ECK0507) that contains three known gene-reaction associations involved in glyoxylate degradation. ECK0507 was thus not be considered by our strategy as a candidate for OXTCase activity either. It may be interesting to note that the genes behind both the anaerobic and aerobic degradation of allantoin are neighbors in E. coli K12's genome.
The remaining three candidate genes (ECK0511 to ECK0513) were ranked at the family-level using CanOE-generated family-level scores; values are given in [Table 3]. ECK0513/ylbF and ECK0512/ylbE belong to two distinct Pfam families harboring domains of unknown function (DUF2877 and DUF1116, respectively) which are conserved over half a thousand proteins from other organisms; either could be good candidates. We have noticed that the gene sequence of ECK0512/ylbE presents a frameshift which is absent in all other sequenced E. coli strains and may be due, according to UniProt, to a sequencing error. The sequence analysis of the third candidate gene (ECK0511/fdrA) gave more clues about its potential molecular function. Indeed, this gene belongs to a family defined by the presence of a conserved domain (PF00549 Pfam domain), many members of which are annotated as CoA-ligase enzymes. The reaction mechanism of the OXTCase activity resembles in no way that of a CoA-ligase activity, suggesting that this gene does not catalyze the former activity. We hypothesize that, if the Pfam assignation is correct, this gene encodes a CoA-ligase which transfers a coenzyme A group to the oxamate produced by the OXTCase enzyme for its degradation by a yet-unknown catabolic pathway (oxamate is currently a dead-end metabolite in the E. coli metabolic network).
None of the three candidate genes (ECK0511 to 0513) share any significant sequence similarities with known carbamoyltransferases. Thus, the candidate genes proposed by CanOE suggest that the OXTCase activity may be catalyzed by a previously-unknown family of carbamoyltransferases. This hypothesis is consistent with a recent study which did not observe any OXTCase activity for gene ECK2866/ygeW, the last E. coli K-12 member of the known carbamoyltransferase family whose function remains to be elucidated [46]. Starting with the best-ranked CanOE candidate (ECK0513/ylbF), protein expressions and biochemical assays are currently under way.
Needless to say, given that the genomic metabolon-defining step of the CanOE strategy is based on the modus operandi of bioanalysts, any respectable bioanalyst could propose candidates genes for reaction gaps after a manual examination of our metabolons. However, the added value of CanOE results are multiple: 1) metabolons are established by an automated procedure, and are distinguished as functional units of a target genome, saving the bioanalyst the effort of locating and building it in his mind; 2) Potential gene-reaction associations are also generated automatically, akin to the hypotheses a bioanalyst formulates during his work; 3) results are integrated across a thousand genomes, a very difficult task for a human to undertake, in the form of a few scores and ranks that can be easily interpreted; 4) all CanOE results are available to the bioanalyst community via a MicroScope platform web interface, making them easily exploitable.
Due to its independence to sequence similarity in its first step and its usage of genomic and metabolic contexts, our CanOE strategy is capable of detecting reaction gaps and proposing candidate genes for them, even in the case of orphan reactions. Calculated metabolons have a relatively good genome coverage (approximately 5% of genes, out of an estimated possible maximum of 30%) and even better metabolic network coverage (1,628 out of 5,157, i.e. 55%). Results are integrated over more than 1,000 prokaryote organisms. We show in a benchmarking experiment that our family-based association scores are informative for the selection of the most promising gene candidates for orphan enzymatic activities; indeed, when keeping the 3 best-ranked associations, precision is 52% for a recall of 76.5%. Out of 72 global orphans with CanOE-proposed candidate genes, 20 of these seemed particularly promising after manual bioanalysis. Even the highly-curated E. coli K-12 genome yielded one orphan reaction with candidate genes, for which biochemical testing is under way.
Other methods exploiting genomic and sometimes metabolic context have been designed in previous works to propose candidate genes for orphan enzymatic activities. Most of them [15], [19], [20] are pathway-dependent in that they require the presence of a predicted pathway (i.e. in which at least one reaction is assigned to one gene in the target organism) to propose candidate genes for the remaining unassigned reactions of that pathway. ADOMETA [18], on the other hand, is not a pathway-dependent method, but orphan reactions must be explicitly described in the organism-specific metabolic reconstructions to be used as targets for candidate genes. Furthermore, ADOMETA requires a filtering step to reduce their metabolic network connectivity: they remove reaction-reaction edges corresponding to the 15 most connected compounds, taking the risk of losing important edges. In comparison to these methods, CanOE uses main compounds defined in metabolic pathways to create a sparser and more biologically relevant global network. This network is thus pathway-dependent in its scope (no reactions not assigned to a pathway are included in it), but is independent in its use (i.e., metabolons can cross multiple pathways). This scope currently limits us to 2,839 of all MetaCyc prokaryote orphan reactions, though this should improve as additional metabolic data is integrated into the MetaCyc database. Also of note is the fact that our strategy predicts gap reactions under the constraint of necessarily anchoring a metabolic context by at least two known reactions to two co-localized genes, making the approach more robust in respect to the quality of the organism's predicted metabolic network.
Unlike previous methods, CanOE is an approach that explicitly integrates its results across several organisms in order to refine and rank its predictions. In the approach developed by Yamanishi et al. [19], candidates are not prioritized, and results across a small number of organisms must be integrated manually. Methods like ADOMETA and the STRING [18], [21] do propose functional association scores that can be used to rank candidates on a per-organism basis; the PathwayHoleFiller-GC method [15] gives an association probability extracted from a Bayesian network. However, even though phylogenetic profile similarity measures are used as input, the results of these methods do not explicitly take into account results across many genomes. Making our family-level selectivity scores available at gene level allows CanOE to have greater power distinguishing false positive associations by favoring conserved associations, as we have shown in our benchmarking experiment.
In the CanOE strategy presented here, we exploit the simplest of genomic context indicators, gene neighborhood, the use of which relies on the observation that genes involved in a same biochemical process tend to cluster on prokaryote genomes, forming operons or regulons. We chose this indicator as it is the most visible and easiest to interpret, and has been shown to outperform other genomic context indicators [47]. It does, however, have some shortcomings. In our graph-based algorithm, gene gaps are located, by construction, between non-gap genes within the metabolon; therefore, genes flanking the metabolon are not proposed, excluding possible interesting candidates. However, systematically proposing all metabolon-flanking genes as candidates leads to many false positive propositions. So far, we argue that our necessity of anchoring a metabolon between at least two genes is a guarantee of the quality of the metabolon; in the worst cases, the manual bioanalysis of a metabolon in situ on the genome may reveal interesting candidates nearby.
Another limitation of our approach is that genes participating in a same biological process might not be clustered on the chromosome because they are linked by other, more complex regulatory mechanisms. In this case, CanOE will obviously not be able to find a metabolon, and hence be unable to propose candidate associations. The previous works discussed above have the advantage of being able to propose candidate genes that are not clustered on the chromosome thanks to their use of other functional dependence measures (such as gene co-regulation, phylogenetic profile similarity, co-citation…). We plan to include additional genomic context indicators within our strategy to extend its scope, if they prove to be informative. For example, it would be possible to use phylogenetic profiles calculated across all organisms, linking genes with high similarities. This modification would allow metabolons to span groups of genes scattered throughout the genome, capturing larger biological processes and opening up new possibilities for gene candidates.
Our metabolons currently cover over 1,060 prokaryote organisms, which is much more than previous strategies achieved [18], [19], [25], and results are integrated a posteriori using gene families. The actual use of gene families for functional annotation remains debatable [48]. Here, we argue that our gene families are not designed to serve as accurate representations of true ortholog families, but only as a means of reinforcing CanOE-proposed associations across several genomes, associations which were, after all, generated based on data other than sequence similarity. Another possibility would be to directly integrate gene-reaction association scores and gene-gene sequence similarity scores rather than use gene families, if computational tractability problems can be overcome.
CanOE results are of interest to the bioanalyst community at four different levels. Firstly, the many metabolons generated by CanOE are independent functional units of a target genome, and each can be represented as easily interpreted graphs. As such, they can be used as an aid to annotation. Secondly, a large fraction of these metabolons are exploited to automatically generate potential gene-reaction association hypotheses, with informative cross-organism integrated scores and ranks to further guide manual annotation. Thirdly, some of these are automatically transformed into Inferred associations, thus helping with automated functional transfer. Finally, a small number of the generated associations concern reactions that are sequence-orphan activities, and are thus of paramount interest; being automatically created, bioanalysts can focus on these specific cases. The web interface allows MicroScope platform users to exploit CanOE results to each of these aims. Altogether, these four levels should help metabolons become a reference in annotating prokaryote genomes. Indeed, it is our hope that this strategy will be employed in wider, systematic enzymatic activity screenings; interacting with projects such as COMBREX [49] and the Enzyme Function Initiative [50] should be productive. Iteratively computing gene-reaction association predictions before validating or invalidating them in wet-lab assays should gradually help cover the metabolism of any prokaryote genome.
|
10.1371/journal.pntd.0003094 | Sporotrichosis in Rio de Janeiro, Brazil: Sporothrix brasiliensis Is Associated with Atypical Clinical Presentations | There have been several recent changes in the taxonomy of Sporothrix schenckii as well as new observations regarding the clinical aspects of sporotrichosis. In this study, we determined the identification of the Sporothrix species associated with both classic and unusual clinical aspects of sporotrichosis observed in the endemic area of sporotrichosis in Rio de Janeiro, Brazil.
To verify whether S. brasiliensis is associated with clinical manifestations of sporotrichosis, a cross-sectional study was performed in which Sporothrix isolates from 50 patients with different clinical manifestations were analyzed and their isolates were studied by phenotypic and genotypic methods. Data from these patients revealed a distinct clinical picture and therapeutic response in infections caused by Sporothrix brasiliensis (n = 45) compared to patients with S. schenckii sensu stricto (n = 5). S. brasiliensis was associated with disseminated cutaneous infection without underlying disease, hypersensitivity reactions, and mucosal infection, whereas patients with S. schenckii presented with less severe and more often localized disease, similar to the majority of previously described sporotrichosis cases. Interestingly, S. brasiliensis-infected patients overall required shorter durations of itraconazole (median 16 weeks) compared to the individuals with S. schenckii (median 24 weeks).
These findings suggest that Sporothrix species are linked to different clinical manifestations of sporotrichosis and that S. brasiliensis is effectively treated with oral itraconazole.
| Sporothrix brasiliensis is a dimorphic fungus that is responsible for an ongoing epidemic of cat-transmitted sporotrichosis in Rio de Janeiro, Brazil. More than 4,100 human cases have been diagnosed in only one health institution since 1998. Most patients are children or housewives with frequent contact with domestic and/or stray cats. The patients usually live under poverty conditions in suburban regions of the metropolitan area with poor access to health care and unsanitary living conditions. For instance, most patients report that they need to have cats in their houses as a control against invasion by rodents. It is important to study the clinical aspects of S. brasiliensis infection in order to improve patient management, including optimizing therapeutic and prophylactic approaches. We have found that S. brasiliensis is responsible for some unusual clinical manifestations of sporotrichosis, such as disseminated infection in immunocompetent patients and hypersensitivity reactions. Also, treatment with itraconazole appears to be extremely effective in most cases of infection by S. brasiliensis. Our study will contribute for the management of the infection caused by S. brasiliensis, bringing benefits to the patients with sporotrichosis.
| Sporotrichosis is a subcutaneous mycosis with a worldwide distribution that is currently notable for areas of especially high endemicity in Latin America [1]–[3]. Some authors classify sporotrichosis as an implantation mycosis, because this infection may also involves other sites beyond the skin and the subcutaneous tissues, such as lymphatic vessels, muscles, fascia, cartilage, and bones [3]. Historically, sporotrichosis has been attributed to a single species, Sporothrix schenckii [1]. However, phenotypic and genotypic analyses by Marimon and coworkers [4] have led to the identification of four new species in the Sporothrix complex: (i) S. globosa, a globally distributed fungus [5]–[7]; (ii) S. brasiliensis, the species related to the zoonotic epidemic of sporotrichosis in Rio de Janeiro, Brazil [8]; (iii) S. mexicana, initially limited to Mexico [4], but with recent cases reported in other regions [9], [10]; and (iv) S. luriei, formerly S. schenckii var. luriei [11].
Classical infection is associated with traumatic subcutaneous inoculation of soil, plants, or organic matter contaminated with fungus, with rare cases of transmission occurring from infected animals [1]. However, in Rio de Janeiro state, Brazil, sporotrichosis is currently largely occurring via transmission from infected cats to humans [12]. Recently, our group performed a georeferencing survey of sporotrichosis cases that revealed a transmission belt along the border between Rio de Janeiro city and adjacent counties in the Greater Metropolitan Area [13]. Genotypic analyses have shown that isolates from the Rio de Janeiro epidemic have a high genetic similarity, which is suggestive of a common niche [14], [15].
Although some studies have described several clinical aspects of this epidemic [12], [16], [17], taxonomic analyses have not been correlated with disease presentations. Therefore, the main purpose of this study was to investigate a possible association between manifestations of sporotrichosis in Rio de Janeiro and the different genomic species of S. schenckii sensu lato.
This study was approved by the Research Ethics Committee of Fundação Oswaldo Cruz (FIOCRUZ), under the number CAAE-0024.0.009.000-10. All patient samples and data were analyzed anonymously after receiving a random number during database construction.
A cross-sectional study was performed in 50 patients with different clinical forms of sporotrichosis. They were selected from a database of 246 patients [8] who had Sporothrix strains isolated from clinical specimens and stored at the Pathogenic Fungal Collection of the Laboratório de Micologia at IPEC, and which were part of a cohort of 1,563 patients with sporotrichosis treated from 1999 to 2008 at the Instituto de Pesquisa Clínica Evandro Chagas (IPEC). Patients were submitted to a protocol that included clinical evaluation, mycological examination of clinical specimens and blood tests (blood count, biochemistry and liver function). In the absence of disseminated disease, oral itraconazole 100 mg/day was prescribed. Higher intraconazole doses were used if the lesions worsened or remained unchanged after eight weeks. The duration of treatment was determined by clinical cure (lesion healing defined as epithelization and absence of crusts, infiltrates, or erythema). Clinical cure of extracutaneous sites was defined as the disappearance of preexisting lesions in cases of conjunctival, nasal, or oral mucosa involvement. Patients with disseminated sporotrichosis received amphotericin B at a total dose of 1–2.5 g. Follow-up was 4–12 weeks after clinical cure. The data were collected by review of medical charts and were recorded on a standardized case report form, containing demographic, epidemiologic, clinical and follow-up information. Since the Rio de Janeiro sporotrichosis outbreak is massive, we had to establish some criteria to select patients related to common and unusual manifestations of sporotrichosis. Inclusion criteria for selection in this study were: patients who lived in Rio de Janeiro city or in other cities from Rio de Janeiro state in Brazil, patients with common (fixed cutaneous and lymphocutaneous) and unusual (disseminated cutaneous, extracutaneous, and disseminated) clinical forms of sporotrichosis [1], patients with and without hypersensitivity manifestations (eythema nodosum or erythema multiforme), patients co-infected with HIV, and patients treated with itraconazole as well as patients with spontaneous regression of lesions. However, for the less common variables (e.g., patients outside the endemic area), all available cases were included.
The fungal isolates were cultured from different body sites, such as skin, eyes, nose, or cerebrospinal fluid. Each isolate was previously identified by classical microbiological phenotypic techniques as S. schenckii sensu lato. Additionally, control strains CBS 120339 (S. brasiliensis) [4], ATCC 16345 (S. schenckii), IPEC 27135 (S. globosa) [7], and MUM 11.02 (S. mexicana) [9] were included in identification tests.
Filamentous fungal colonies for each isolate were grown on Sabouraud Dextrose Agar and slide cultures were mounted with Lactophenol Cotton Blue (Fluka Analyted, France) for Sporothrix identification [4]. Dimorphism was demonstrated by conversion to the yeast-like form on Brain Heart Infusion Agar slants for 7 days at 37°C. Furthermore, colonies were sub-cultured on Potato Dextrose Agar plates and Corn Meal Agar slants, and incubated at 30 and 37°C in the dark to study fungal growth and sporulation respectively [4], [8]. Carbohydrate assimilation tests were performed using freshly prepared yeast nitrogen base (YNB) medium supplemented with sucrose or raffinose, using YNB supplemented with glucose as positive control and YNB without carbohydrates as a negative control. Experiments were performed at least three times on different days and, in the case of discordant results, repeated two additional times. All culture media were from Difco (Becton, Dickinson and Company/Sparks MD, USA). Results were interpreted according to the identification key detailed by Marimon and coworkers [11].
Genomic DNA was extracted and purified from Sporothrix spp mycelial phase by chloroform/isoamyl alcohol method as described [7]. The gene encoding for the nuclear calmodulin was used for molecular differentiation of the isolates because this locus has a high number of parsimony informative sites, allowing Sporothrix differentiation in several genotypes [18]. For partial sequencing of the nuclear calmodulin (CAL) gene, we used the primers CL1 (5′-GA(GA)T(AT)CAAGGAGGCCTTCTC-3′), and CL2A (5′-TTTTTGCATCATGAGTTGGAC-3′) under previously described conditions [7]. Automated sequencing was done using the Sequencing Platform at PDTIS/FIOCRUZ, Brazil [19]. Sequences from both DNA strands were generated and edited with the Sequencher ver. 4.6 software package (Genes Codes Corporation, USA), followed by alignment with Mega version 4.0.2 software. Our sequences were compared by BLAST (Basic Local Alignment Search Tool) with sequences available from NCBI GenBank (Sporothrix AM 398382.1/AM 398393.1/AM 117444.1/AM 116899.1/AM 116908.1). All phylogenetic analyses were performed as previously described [7], [8].
All sequences from isolates included in genotypic analysis were deposited in the GenBank database under accession numbers GU456632, HQ426928 to HQ426962, and KC463890 to KC463903.
Data were processed and analyzed using the SPSS 17.0 software. Frequencies and median values were calculated for each group of this study.
Of the 50 patients, 16 were male and 34 female, with ages ranging from 9 to 83 years (median = 47). Lesions were located at upper limbs (n = 31, 62%), lower limbs (n = 6, 12%), face (n = 1, 2%), trunk (n = 1, 2%), and more than one segment (n = 11, 22%). Fifteen patients (30%) presented with a fixed cutaneous form, 24 (48%) lymphocutaneous form, 6 (12%) disseminated cutaneous form, and 5 (10%) disseminated (involving internal tissues) sporotrichosis. Additionally, six of these patients also presented with erythema nodosum and four with erythema multiforme. Table 1 summarizes the clinical and mycological information for each patient.
Of the 50 strains, 45 (90%) were classified by molecular methods as S. brasiliensis and 5 (10%) as S. schenckii. In 21 (42%) isolates, results from phenotypic tests were inconclusive, precluding species differentiation; these strains were phenotypically classified as Sporothrix spp. Interestingly, phenotypic identification of 10 (20%) isolates did not match to the genotypic results. Eight (16%) strains phenotypically classified as S. schenckii, DNA sequencing clustered them amid S. brasiliensis. The strain phenotypically classified as S. mexicana was genotypically identified as S. schenckii, and one S. brasiliensis was classified as S. schenckii by CAL sequencing.
Forty-two (93.3%) of the strains identified taxonomically as S. brasiliensis were from the Rio de Janeiro endemic area of sporotrichosis, including Rio de Janeiro city, Duque de Caxias, Belford Roxo, Sao João de Meriti, Nova Iguaçu, Nilópolis, and Mesquita (Fig. 1). The other three (6.7%) S. brasiliensis strains were isolated from patients who lived in Teresópolis, a county 91 km away from Rio de Janeiro city. S. brasiliensis was isolated from 32 of 34 women (94%). Forty (88.9%) patients with S. brasiliensis had documented contacts with cats. Two additional S. brasiliensis-infected patients (4.4%) reported plant and glass trauma preceding the development of sporotrichosis.
With respect to the five patients with S. schenckii, four of them (80%) were isolated from patients who lived in three different rural regions and one urban area (in Itaboraí, Barra do Piraí, Casimiro de Abreu, and Teresópolis, respectively; 45, 100, 127, and 91 km away from Rio de Janeiro city), which are outside of the endemic area. S. schenckii was also isolated from a patient who lived within the zoonotic endemic sporotrichosis area in Rio de Janeiro. Three patients were male and two female.
Hypersensitivity reactions such as erythema nodosum or erythema multiforme (10 cases), disseminated cutaneous forms (6 cases), and all but one case of lymphocutaneous sporotrichosis were all attributed to infection with S. brasiliensis. Localized cutaneous forms were observed in patients infected with either S. brasiliensis (n = 12, 26.7%) or S. schenckii (n = 3, 60%). Disseminated disease occurred due to S. schenckii in one patient with AIDS, S. brasiliensis in two patients with AIDS, and S. brasiliensis in one patient without any history of immunosuppression. Finally, there was one case of fixed cutaneous sporotrichosis caused by S. brasiliensis in a HIV infected patient with CD4>200 cells/µL.
Four patients infected with S. brasiliensis were lost to follow-up. The three patients with AIDS and disseminated disease were excluded from analysis since they received amphotericin B as part of their antifungal regimen. Spontaneous regression was observed in one patient infected with S. schenckii (fixed form) and three with S. brasiliensis (two fixed and one disseminated cutaneous forms). The remaining 3 cases of S. schenckii required more than 24 weeks of itraconazole, and two of them required increased doses (200 and 400 mg/day). Most of the 35 patients infected by S. brasiliensis included in this analysis (82.9%) resolved with less than 24 weeks of treatment, regardless of their clinical form. For eight S. brasiliensis-infected patients, up to 400 mg/day itraconazole were necessary for clinical cure. The median time to cure for patients with hypersensitivity reactions was similar to the patients without these manifestations (16 weeks), and their diseases resolved with 100 mg/day of itraconazole.
The clinical presentations of sporotrichosis caused by Sporothrix spp are highly variable and poorly understood. Kong and collaborators [20] have demonstrated that S. schenckii genotypes can be correlated with clinical forms of disease, as mice challenged with S. schenckii isolates from patients with fixed cutaneous, lymphocutaneous or disseminated sporotrichosis developed more severe disease according to the severity of the manifestations in the originating patient. However, they did not define the relationships between genotype and treatment outcome or other unusual manifestations. In the present work, we show the direct association between unusual clinical presentations of human sporotrichosis with infection by S. brasiliensis. Although S. brasiliensis caused typical manifestations of fixed cutaneous and lymphocutaneous sporotrichosis, all 10 patients with hypersensitivity reactions and 6/7 patients with disseminated disease were infected with S. brasiliensis. To the best of our knowledge, this is the first work that demonstrates an association between genotypic identification of Sporothrix species and several clinical aspects of human sporotrichosis. Given the recent changes in the nomenclature and advances in the molecular taxonomy of Sporothrix, it is even more important to understand the clinical implications of these advances [21].
As expected, the majority of our isolates have been identified as S. brasiliensis by DNA analyses. Our group has previously characterized S. brasiliensis in 230 (93.5%) of 246 isolates obtained from this endemic zoonotic transmission area [8]. A study by Marimon and coworkers of 127 Sporothrix strains collected from several parts of the world reported only S. brasiliensis among the tested isolates from Rio de Janeiro [4]. There are also a few reports of S. brasiliensis in Brazilian states other than Rio de Janeiro [10], [22], [23], but, in these states, the frequency of S. brasiliensis appears to be lower than that for the other Sporothrix species, with S. schenckii predominating [22].
As noted above, S. brasiliensis genotype caused typical clinical forms of sporotrichosis (lymphocutaneous and fixed cutaneous). However, except for 1 case of disseminated S. schenckii in a patient with AIDS, all of the unusual clinical forms of sporotrichosis were attributed to infection with S. brasiliensis, including disseminated cutaneous sporotrichosis in the absence of an underlying immunosuppressive condition, mucosal involvement affecting nasal cavity or conjunctiva, and hypersensitivity reactions. Regarding the hypersensitivity manifestations, our finding are consistent with the previously reported cases of erythema nodosum and erythema multiforme associated with zoonotic sporotrichosis [24], [25] due to S. brasiliensis. Recently, Sweet syndrome has also described in 3 patients with sporotrichosis [26], and studies are underway to determine, by calmodulin sequencing, the species involved in these cases.
Another interesting finding for disease due to S. brasiliensis is the 32/13 female/male ratio, since there is a predominance of male over female patients with sporotrichosis caused by S. schenckii. This can be explained by the fact that the most affected group in the endemic area of sporotrichosis in Rio de Janeiro are housewives that interact with or take care of S. brasiliensis infected cats [1].
Barros and coworkers [27] studied the effects of itraconaozle treatment on cutaneous sporotrichosis in 645 patients from the Rio de Janeiro epidemic, including 87 patients with erythema nodosum or erythema multiforme. Interestingly, they observed that of the patients with hypersensitivity reactions resolved their disease more rapidly compared with patients without these conditions. Although our current study did not find differences regarding treatment between these groups, we did determine that most patients with hypersensitivity reactions presented with fixed cutaneous sporotrichosis. Hence, we believe that hypersensitivity reactions may indicate a robust host response to the S. brasiliensis yeast cells and play a protective role in sporotrichosis, as observed in coccidioidomycosis [28].
The small number of S. schenckii cases in our study calls our attention to the infection caused by this species in Rio de Janeiro. The majority of these cases occurred in rural counties where inhabitants are engaged in agricultural activities, and, therefore, they have frequent and protracted contact with soil. Moreover, in two of these cases, patients denied cat contact. However, S. schenckii was identified in one case from the endemic zoonotic transmission area. Our results suggest that S. schenckii also circulates, in minor proportions, in this endemic area. New studies with a large number of S. schenckii infected patients are necessary to verify the clinical meaningful of sporotrichosis caused by this species.
Several factors could influence the different outcomes of sporotrichosis, such as the size of initial inoculum, the host immune response status, depth of traumatic inoculation and fungal virulence [29]. Virulence studies in a mouse infection model have shown that S. brasiliensis is significantly more lethal and results in higher fungal burdens compared to S. schenckii as well as other examined Sporothrix spp. [30]. This same study concludes that lesional mechanisms could be species-specific, which supports our results. Zoonotic transmission of sporotrichosis by cats results in high Sporothrix inoculums for humans, since these animals have high fungal burdens [31]. In some of the endemic sporotrichosis cases, fungal inoculation is presumably repetitive, due to constant bites and scratches suffered by owners [32]. These factors, coupled with the purported higher virulence of S. brasiliensis [30], could account for the unusual and more severe clinical manifestations observed with this species.
Itraconazole is the drug of choice for sporotrichosis treatment [27]. It is interesting to note that, regardless the clinical form, there was a trend toward shorter treatment durations in patients with sporotrichosis caused by S. brasiliensis, (median = 16 weeks) than the cases due to S. schenckii (median = 24 weeks), for our study. Although we are comparing 46 cases of S. brasiliensis to only 4 infections due to S. schenckii, we propose that our finding might have a therapeutic implication. The response of S. brasiliensis to treatment is consistent with the fact that S. brasiliensis is more susceptible to antifungal drugs, such as itraconazole, posaconazole, and ravuconazole, than S. schenckii [33]. Moreover, previous results of our group, which included eight S. brasiliensis from this study, showed that strains from the zoonotic endemic area were highly susceptible to itraconazole [14]. Nevertheless, clinical, randomized studies should be performed to confirm these findings.
Since different Sporothrix species appear to be related to distinct clinical manifestations and treatment responses, we propose that speciation should become standard laboratory practice. However, it will require a significant effort to make this recommendation a reality in most clinical laboratories. Phenotypic fungal identification is easier than molecular methods to routinely apply. However, in the present work as well as in a previous study from our team [8] and studies from other groups [10], phenotypic description too often fails to be corroborated by genotypic results. Since the differences between the species of the S. schenckii complex were observed at the molecular level [18], we considered DNA sequencing as the gold standard on species identification for the present study. Unfortunately, at present, DNA sequencing is not a suitable methodology for routine clinical laboratories. We recently described a simple and reliable T3B DNA fingerprinting methodology to identify the S. schenckii species complex at the DNA level [34], making it an alternative identification methodology for clinical microbiology laboratories.
Our study intended to perform a molecular analysis of the most peculiar clinical cases that we observed in this epidemic as well as typical cases, and also patients presenting from areas outside the sporotrichosis belt of zoonotic transmission, which corresponds to almost 20% of our stored samples. We have checked the medical data from the other 196 patients and they were very similar to the cases we included here. The small size of genotyped strains is a weakness of this study, but in our opinion, these cases illustrate the association of S. brasiliensis to the unusual presentations of sporotrichosis. Also, molecular genotyping with sequencing of the calmodulin gene is laborious and expensive (at present), with significant financial impact in our severely limited funding situation.
In conclusion, we have used molecular analysis to clearly demonstrate that S. brasiliensis is the primary cause of endemic sporotrichosis in Rio de Janeiro state. Moreover, S. brasiliensis causes both classic and unusual manifestations of sporotrichosis, including severe disease in otherwise immunocompetent individuals. We also have documented that local and invasive S. brasiliensis disease responds well to itraconazole therapy, with shorter durations of therapy compared to the patients studied with sporotrichosis caused by S. schenckii. This study adds new information to our knowledge base on S. brasiliensis disease and supports the careful speciation of Sporothrix isolates to guide clinical care.
|
10.1371/journal.pgen.1000916 | Two New Loci for Body-Weight Regulation Identified in a Joint Analysis of Genome-Wide Association Studies for Early-Onset Extreme Obesity in French and German Study Groups | Meta-analyses of population-based genome-wide association studies (GWAS) in adults have recently led to the detection of new genetic loci for obesity. Here we aimed to discover additional obesity loci in extremely obese children and adolescents. We also investigated if these results generalize by estimating the effects of these obesity loci in adults and in population-based samples including both children and adults. We jointly analysed two GWAS of 2,258 individuals and followed-up the best, according to lowest p-values, 44 single nucleotide polymorphisms (SNP) from 21 genomic regions in 3,141 individuals. After this DISCOVERY step, we explored if the findings derived from the extremely obese children and adolescents (10 SNPs from 5 genomic regions) generalized to (i) the population level and (ii) to adults by genotyping another 31,182 individuals (GENERALIZATION step). Apart from previously identified FTO, MC4R, and TMEM18, we detected two new loci for obesity: one in SDCCAG8 (serologically defined colon cancer antigen 8 gene; p = 1.85×10−8 in the DISCOVERY step) and one between TNKS (tankyrase, TRF1-interacting ankyrin-related ADP-ribose polymerase gene) and MSRA (methionine sulfoxide reductase A gene; p = 4.84×10−7), the latter finding being limited to children and adolescents as demonstrated in the GENERALIZATION step. The odds ratios for early-onset obesity were estimated at ∼1.10 per risk allele for both loci. Interestingly, the TNKS/MSRA locus has recently been found to be associated with adult waist circumference. In summary, we have completed a meta-analysis of two GWAS which both focus on extremely obese children and adolescents and replicated our findings in a large followed-up data set. We observed that genetic variants in or near FTO, MC4R, TMEM18, SDCCAG8, and TNKS/MSRA were robustly associated with early-onset obesity. We conclude that the currently known major common variants related to obesity overlap to a substantial degree between children and adults.
| Genome-wide association studies (GWAS) have successfully contributed to the detection of genetic variants involved in body-weight regulation. We jointly analysed two GWAS for early-onset extreme obesity in 2,258 individuals of European origin and followed-up the findings in 3,141 individuals. Evidence for association of markers in two new genetic loci was shown (SDCCAG8 on chromosome 1q43–q44 and between TNKS/MSRA on chromosome 8p23.1). We also re-identified variants in or near FTO, MC4R, and TMEM18 to be associated with extreme obesity. In addition, we assessed the effect of the markers in 31,182 obese, lean, normal weight, and unselected individuals from population-based samples and showed that the variants near FTO, MC4R, TMEM18, and SDCCAG8 were consistently associated with obesity. For variants of TNKS/MSRA, the obesity association was limited to children and adolescents. In summary, we detected two new obesity loci and confirmed that the currently known major common variants related to obesity overlap to a substantial degree between children and adults.
| Recent genome-wide association studies (GWAS) conducted in adult population-based samples assessed for body mass index (BMI) or in case-control designs for extreme obesity led to the discovery of genetic loci relevant for body weight regulation. The first genetic loci were detected via variants in intron 1 of the FTO (fat mass and obesity associated gene; e.g., [1]–[4]) and variants approx. 200 kb downstream of MC4R (melanocortin 4 receptor gene; [5]–[8]) reported by the GIANT (Genetic Investigation of ANthropometric Traits) consortium. This consortium subsequently detected six additional genetic loci relevant for BMI in a meta-analysis of 15 GWAS based on 32,387 probands and large confirmation samples (>58,000 individuals; with single nucleotide polymorphisms (SNP) in or near TMEM18, transmembrane protein 18 gene; KCTD15, potassium channel tetramerization domain containing 15 gene; GNPDA2, glucosamine-6-phosphate deaminase 2 gene; SH2B1, SH2B adapter protein 1 gene; MTCH2, mitochondrial carrier homologue 2 gene; NEGR1, neuronal growth regulator 1 gene). In parallel, a combined analysis of 34,416 individuals from Iceland, the Netherlands, North America (European and African descent) and Scandinavia revealed 11 regions of genome-wide significance at ≤1.6×10−7 (in or near FTO; MC4R; TMEM18; KCTD15; SH2B1; NEGR1; SEC16B, SEC16 homologue B gene; ETV5, ets variant gene 5; BDNF, brain-derived neurotrophic factor gene and two gene rich loci on chromosome 6p21.33 and 12q13.13 with the closest genes AIF1, allograft inflammatory factor 1 gene, and BCDIN3D, BCDIN3 domain containing gene, respectively). Finally, shifting to the analysis of extremely obese subjects, Meyre et al. [9] analyzed GWAS data from 1,380 Europeans with early-onset and morbid adult obesity and 1,416 age-matched normal-weight controls and reported three new risk loci in NPC1 (Niemann-Pick disease, type C1 gene), near MAF (v-maf musculoaponeurotic fibrosarcoma oncogene homolog gene) and PTER (phosphotriesterase related gene), which were followed-up in 14,186 European subjects. Altogether, 16 genetic loci relevant for body weight regulation have been identified by these three GWAS approaches [10]–[12].
While meta-analytic combinations of multiple GWAS were highly successful in population-based samples, no such approach has up to now been applied to case-control designs for obesity. Here we combined GWAS based on two samples that were specifically ascertained for the analysis of paediatric extreme obesity [3], [9]. We aimed to identify genetic loci that are relevant for early onset extreme obesity and to determine effect sizes of such loci for obesity in adults and in population-based samples including both children and adults (see Figure 1 for the general design of the study).
In particular, our study design was based on two steps to enable hypothesis-free SNP identification and confirmation. In the DISCOVERY step, we screened 2,239,392 genotyped or imputed SNPs and tested 1,596,878 SNPs (after quality control) for association in a combined French and German sample of 1,138 extremely obese children and adolescents and 1,120 normal- or underweight controls as based on a minor allele frequency above 1%. Next, we (de novo) genotyped all SNPs with strong evidence for an association to obesity (according to p-value ranking; for details see “Materials and Methods” and Text S1) in independent samples of 1,181 obese children and adolescents and 1,960 normal- or underweight controls and in up to 715 nuclear families with at least one extremely obese offspring. In the GENERALIZATION step, we extended the focus of our study in two dimensions - (i) from children and adolescents to adults and (ii) from (extreme) obesity to the population level (in sum we (de novo) genotyped 31,182 individuals in the GENERALIZATION step).
In addition to our hypothesis-free step-wise design, we aimed to re-confirm the associations of the recently reported GWAS-based genetic loci for body weight regulation [9], [13], [14] in our paediatric extreme obesity GWAS meta-analysis.
In our GWAS meta-analysis based on the German and French study groups encompassing both young obese cases and normal weight or lean controls we discovered three SNPs with genome-wide significance (Table 1 and Figure 2, Figure S1) even when applying the conservative Bonferroni correction at αBF≈3.1×10−8 for all 1,596,878 SNPs. While two markers are located in the previously reported FTO (intron 1; rs1421085; p = 2.99×10−8) and downstream of MC4R (rs17700144; p = 2.40×10−8), rs473034 indicates a new genetic locus for early onset extreme obesity located on chromosome 8p23.1 (p = 2.77×10−8) with the closest genes TNKS (tankyrase, TRF1-interacting, ankyrin-related ADP-ribose polymerase gene; ∼135 kb upstream of rs473034) and MSRA (methionine sulfoxide reductase A gene; ∼178 kb downstream of rs473034). In addition to the three genome-wide significant regions, the GWAS data revealed 18 genomic regions of interest which were defined by (i) two-sided p-values of a lead SNPs ≤10−5 and (ii) more than a single SNP within a locus (lead SNP ±500 kb) showing evidence for association as defined via a p-value rank <1,500 (roughly corresponding to p≤5×10−4; for details see Text S1).
As part of our DISCOVERY step, we subsequently (de novo) genotyped 44 SNPs representing these 21 genomic regions of interest in independent 1,181 obese children and adolescents and 1,960 normal- or underweight controls and in up to 715 nuclear families with at least one extremely obese offspring (Table 1; Table S3). For 5 out of the 21 regions the association was directionally consistent (i.e. we observed the same obesity risk effect allele as in our GWAS meta-analysis) and the minimum combined p-value for each region across the samples was p≤5×10−4 (Table 1; for details see Text S1). These 5 genomic regions included three known loci on chromosome 2p25.3 (TMEM18), 16q12.2 (FTO), 18q21.32 (3′ of MC4R) as well as two new loci on chromosome 1q43-q44 and on chromosome 8p23.1 (Figure 2, Figure S2). The SNPs of the first new locus on chromosome 1q43-q44 are located within introns of the SDCCAG8 (serologically defined colon cancer antigen 8 gene) whereas the second new locus on chromosome 8p23.1 between the TNKS and MSRA had already showed evidence for an association at the genome-wide level in the initial paediatric extreme obesity GWAS meta-analysis.
Based on these results, we extended the focus of our study in two dimensions - from children and adolescents to adults and from the extremes to the population level - looking for GENERALIZATION of the replicated 5 regions represented by 10 SNPs (Table 1). Comparing children and adolescents to adults using case-control designs with overweight and obese cases vs. normal weight controls revealed directionally consistent (see above) findings for the variants of FTO, TMEM18 and the novel SDCCAG8 (Table 1). Similarly the odds ratios for the respective obesity risk effect alleles did not vary strongly by group (children and adolescents vs. adults) with point estimates ranging between 1.35–1.45 (FTO), 1.35–1.45 (TMEM18) and 1.10–1.19 (SDCCAG8). For the SNPs related to MC4R and the new TNKS/MSRA locus, however, we observed age dependent differences: For MC4R, we confirmed the findings by Loos and co-workers [6] by finding a stronger effect size estimator in children and adolescents as compared to adults (1.44 vs. 1.14 for rs17700144 of MC4R; p = 9.39×10−3 for the interaction of genotype and group). For TNKS/MSRA, we found an effect in children and adolescents but no effect in adults (e.g., 1.12 vs. 0.97 for rs516175). These differences in obesity risk effects between children and adolescents as compared to adults, however, were not due to large differences in allele frequencies as based on the population-based samples with a maximum difference of 0.82% for rs11127485 of TMEM18. We then compared (extreme) obesity assessed in case-control designs to the analyses of quantitative BMI data derived from population-based samples in the GENERALIZATION step (3,525 children and adolescents and 25,958 adults of European origin; Table 1, Table 2). BMI analyses revealed that the two SNPs in FTO and TMEM18 would have also been detectable using population-based samples of the given sizes from children/adolescents and adults (p-values 7.87×10−4 and 9.99×10−16 for FTO and 0.01 and 9.97×10−12 for TMEM18 with the values in the adults being even significant at a stringent genome-wide significance level of α = 5×10−8). The MC4R SNP, however, would have been harder to detect (p-values of 0.02 for children and adolescents and 1.10×10−4 for adults); detection of the two new loci SDCCAG8 and TNKS/MSRA would have been impossible (Table 2).
In sum, our hypothesis-free step-wise design revealed three known (FTO, MC4R and TMEM18) and two new loci (SDCCAG8 and TNKS/MSRA) with estimated odds ratios that ranged from ∼1.07 to ∼1.44 in children and adolescents and from ∼1.17 to ∼1.45 in adults with the strongest overall signals related to the FTO locus. Modelling of the joint and epistatic effects revealed that <1% of the BMI (or BMI-SDS when BMI is expressed as standard deviation score) variance can be attributed to the five variants analyzed in or near TNKS/MSRA, SDCCAG8, TMEM18, FTO, and MC4R. For children and adolescents this value did not change upon inclusion of gender, age and age2 as covariates whereas it changed to 11% for the adult sample (KORA S2-S4). Applying the model including the same covariates derived in one population-based data set of adults (KORA S2-S4) to a second population-based data sets of adults (Heinz-Nixdorf Recall Study) r2 dropped from 11% to ∼2%. Proceeding similarly for epistatic effects, we found no evidence for strong epistatic effects using regression tree analyses (Figure S3, Figure S4).
In addition to our hypothesis-free step-wise design, we investigated our paediatric extreme obesity GWAS meta-analysis data focussing on recently reported GWAS-based candidate markers [9], [13], [14]. For the 16 confirmed genetic loci for which quality controlled genotyped or imputed SNPs were available, two loci on chromosome 1 (1p31.1–NEGR1, 1q25.2 - SEC16B), a locus on 11p14.1 near BDNF, and a gene-rich locus on 12q13.13 near BCDIN3D all showed directionally consistent effects of the respective SNPs (all p≤.005). Details on all analysed candidate gene SNPs are provided in Table S4 and Table S5. Note that the 16 confirmed genetic loci [9], [13], [14] correspond to 46 SNPs in our GWAS meta-analysis; in case of multiple markers at the same locus all showed evidence for strong LD (r2>.9).
We identified two new genomic loci associated with paediatric obesity on chromosomes 1q43–q44 and 8p23.1 by a meta-analysis of two GWAS for early onset extreme obesity with a total 2,258 individuals of European origin. In addition, we confirmed the three known loci FTO, MC4R and TMEM18 using a hypothesis-free step-wise design. Leaving the hypothesis-free approach and focussing on known GWAS-based candidate markers, we were able to substantiate another four loci (NEGR1, SEC16B, BDNF and BCDIN3D) of the 16 obesity loci previously detected in GWAS [6], [9], [13], [14]. Thus, we demonstrate that the currently known major common variants related to obesity overlap to a substantial degree between children and adults confirming previous observations for FTO, MC4R, TMEM18, NEGR1 [2], [6], [14] and extending this observation to SEC16B, BDNF and BCDIN3D; [13], [14]. As our meta-analysis includes data from Meyre et al. [9] an independent well-powered replication of NPC1, MAF and PTER was not possible here.
The new chromosome 1q43–q44 locus was represented by three SNPs in strong pairwise LD (r2>.9) which are located in introns 6, 9 and 10 of SDCCAG8. There is no obvious indication for an involvement of SDCCAG8 in body weight regulation. Data on this gene are scarce. It has been shown that SDCCAG8 is located in centrosomes during interphase and mitosis in human and murine cells. N- and C- terminal truncations of the human protein alter this location; a possible role of SDCCAG8 (alternative name: NY-CO-8) in centrosomal organization has been suggested [15]. It is considered to be a naturally occurring autoantigen [16]. SDCCAG8 is ubiquitously expressed, amongst other tissues in thymus, small intestine, colon mucosa, liver and brain (http://www.genecards.org/cgi-bin/carddisp.pl?gene=SDCCAG8). Hypothalamus, pituitary and adrenals have been shown to have a particularly high transcript abundance. This pattern indicates a role of SDCCAG8 in this pivotal hormonal axis that is well-known for its impact on body weight regulation [16]. Other candidate genes in proximity of the three SNPs include CEP170 (centrosomal protein 170 kDa gene, ∼95 kb downstream of rs12145833) and AKT3 (v-akt murine thymoma viral oncogene homolog 3 (protein kinase B, gamma) gene, ∼168 kb upstream of rs12145833) with the latter being the more interesting candidate. The protein encoded by this gene is a member of the AKT family known to regulate cell signalling in response to insulin and growth factors. In particular AS160, an Akt substrate of 160 kDa, and TBC1D1 (TBC1 domain family, member 1) have been suggested to have complementary roles in regulating vesicle trafficking in response to insulin [17] with TBC1D1 being persuasively linked to body weight regulation [18]–[20]. However, we observed no evidence for strong pairwise LD (r2>.9) to any likely functional relevant variant in a region of ±1 Mb around the lead SNP (rs12145833) using Ensembl (version 56; GRCh37, 02/2009; Figure S6).
The new chromosome 8p23.1 locus, for which we observed genome-wide significance in our GWAS meta-analysis (Figure 1, ), was also represented by three SNPs with strong pairwise LD (r2>.9). TNKS and MSRA are the genes located closest to our association finding. MSRA encodes a repair enzyme for oxidative damage in proteins by enzymatic reduction of methionine sulfoxide. Oxidation of methionine residues in proteins is considered to be an important consequence of oxidative damage to cells [21]. Oxidation of proteins by reactive oxygen species (ROS) is generally associated with oxidative stress, aging and many neurodegenerative diseases such as Alzheimer's disease [21]. Also, obesity is associated with oxidative stress in the mitochondrion, with the chronic excess of ROS resulting in mitochondrial dysfunction in liver and skeletal muscle contributing to insulin resistance [22]. MSRA is mainly expressed in kidney followed by liver, brain, and adipose tissue (http://biogps.gnf.org/#goto=genereport&id=4482). The other candidate gene at the chromosome 8p23.1 locus is TNKS which is ubiquitously expressed (http://biogps.gnf.org/#goto=genereport&id=8658). Tankyrase is a Golgi-associated poly-ADP-ribose polymerase, which is involved in the regulation of GLUT4 trafficking in 3T3-L1 adipocytes. Mice lacking Tnks show increased energy expenditure, fatty-acid oxidation, and insulin-stimulated glucose utilization; they are lean even with excessive food intake [23]. In other GWAS, the 8p23.1 genomic region has been related to increased triglyceride levels [24] and to waist circumference in adults [21]. The variants with the strongest reported association signals (rs7819412; rs7826222 which is now labelled rs545854) are about 1.3 and .08 Mb downstream of our best finding (rs473034). For the former, the association to obesity was moderate in our GWAS meta-analysis data (p = 0.02) whereas for the latter no genotype data were available (with pairwise LD between rs545854 and rs473034 of r2<.01 (D' = .03) according to Ensembl version 56). Thus, further research is needed to elucidate if our finding for TNKS/MSRA detected in paediatric extremes of the quantitative trait BMI and the finding for waist circumference in adults [21] point to the same underlying genetic mechanism.
In our study we used two steps to enable hypothesis-free SNP identification and confirmation covering the extremes and the population distribution of BMI in paediatric as well as adult samples. Both dimensions of our design are related to statistical power considerations and the genetic architecture of the phenotype studied. A case-control design with highly selected individuals outperforms a design using unselected population-based individuals if the same number of individuals are genotyped and if the same alternative hypothesis holds true (see Text S1). This contrast will be aggravated the more extreme the selection and possibly also the younger the subjects [25]. In addition the selection of extremes may lead to the detection of genetic variations that are rare in the population, that accumulated in families and that might result in stronger effect sizes. Nevertheless, the power of our GWAS meta-analysis sample is still limited for small effects (see Text S1) and growing consortia like GIANT [14] will be best suited to detect them. Not surprisingly, we confirmed the strongest effects (odds ratio for the obesity risk effect alleles of ∼1.4) reported for children and adolescents near FTO, MC4R and TMEM18 [12] but also found support for variants near NEGR1, SEC16B, BDNF and BCDIN3D. Thus, one may speculate, that the genetic architecture in the paediatric extremely obese is in part similar to the BMI findings based mainly on adults from large population-based assessments (e.g. [13], [14]). On the other hand, some of the related effect sizes of these variants seem to vary longitudinally as shown here for MC4R and previously stressed by others [6], [26] while other genetic loci might only be relevant for (paediatric) extreme obesity such as TNKS/MSRA.
In conclusion, two new loci related to body weight regulation were identified using highly selected paediatric samples from the extremes of the quantitative phenotype BMI. By showing that one locus is relevant across all age groups whereas the impact of a second is limited to childhood and adolescence, our data support previous studies showing the importance of age-related aspects upon interpretation of GWAS signals.
Our study design consisted of two steps (Figure 1). As first part of the DISCOVERY step we performed a meta-analysis of two genome-wide association studies (GWAS) including 1,370 individuals of French and 888 of German ancestry, defined by self-reported ethnicity. Ascertainment in both GWAS was very similar with a focus on extremely obese children and adolescents and normal weight or lean controls (Table S1). Body-mass-index (BMI in kg/m2) was calculated and the extremes were defined using percentile criteria of large population-based samples of the general population [27], [28]. We applied the cut-offs ≥97th percentile and ≥90th percentile to define ‘obesity’ and ‘overweight’ in children and adolescents; most of the cases with extreme obesity had a BMI ≥99th percentile (Table S1; [29]). Whole-genome genotyping was carried out using the Illumina Human CNV370-Duo array (French GWAS) and the Affymetrix Genome-Wide Human SNP Array 6.0 (German GWAS). Genotype data quality measures, e.g. genotype calling rates, were similar in both GWAS (Table S2). To combine both datasets, the GWAS genotypes were imputed using publicly available HapMap CEU (release 22; http://www.hapmap.org). From this GWAS meta-analysis, we selected 44 SNPs covering 21 loci (Table S3; Figure S5) which we (de novo) genotyped in 1,181 overweight and obese children and adolescents and 1,960 normal weight or lean children and adolescents and young adults (controls) of European ancestry and up to 715 nuclear families with obese offspring of European ancestry were examined. The SNP selection was based on (i) an unadjusted two-sided p-values ≤10−5 and (ii) more than a single SNP within a locus (lead SNP ±500 kb) showing evidence for association (with a p-value rank <1,500 roughly corresponding to p≤5×10−4; for details see Text S1). Sub-whole genome SNP genotyping was performed using by the MALDI-TOF mass spectrometry-based iPLEX Gold assay. In the GENERALIZATION step, 10 SNPs, for which DISCOVERY step had revealed consistent observations (Table 1; Table 2), were further investigated for generalizability to adults and to unselected population-based samples. Thus, 711 overweight and obese children and adolescents (Datteln Paediatric Obesity sample), 3,525 children and adolescents from the general population (GINI, LISA, Berlin School Girls), 988 obese adults (Marburg Adult Obesity sample) and 25,958 adults from the general population (EPIC-Potsdam Study, KORA S2-S4, SHIP, Heinz-Nixdorf Recall Study) each of European ancestry were genotyped. SNP genotyping was performed by the MALDI-TOF mass spectrometry-based iPLEX Gold assay at the Helmholtz Zentrum, München and at the Department of Genomics, Life & Brain Center, Bonn or by KBioscience, Hoddeston, UK. All were assessed for genotype calling rates and deviations from Hardy–Weinberg equilibrium (for details see Text S1).
The RefSeq accession numbers for the reported genes are: FTO: NM_001080432; MC4R: NM_005912; TNKS: NM_003747; SDCCAG8: NM_006642.2; TMEM18: NM_152834; CEP170: NM_014812; AKT3: NM_181690.
After similar quality control analyses of both GWAS, the imputed GWAS were jointly analysed using the inverse normal method to combine p-values of allele-based chi-square tests. Details on the imputation and on the marker selection for the follow-up are described in Text S1. In the paediatric extreme obesity GWAS meta-analysis data set we also explored genetic variants for obesity recently derived from other GWAS [9], [13], [14] and variants for ‘classical’ obesity candidate genes [3], [30] by testing the best SNP reported in Scuteri et al. [4].
In both the DISCOVERY and the GENERALIZATION part of the study either log-additive or additive genetic models were applied. Case-control samples were analysed using logistic regression (both with and without gender and age as covariates). The nuclear families were analysed using UNPHASED (Version 3.0.13; [31]) which addresses the correlation among sibs and provides estimators; nuclear family data and case-control data sets were combined using a method described in [32]. In the GENERALIZATION step, BMI in adults of population-based samples was analysed using linear regression with gender and age as covariates. Similarly, we used linear regression analyses for the population-based samples of children and adolescents. However, as phenotype we used a normalized version of the BMI applying Cole's least mean square method [33] to express BMI as a standard deviation score (BMI-SDS) which is comparable to the BMI z-score as e.g. used by the Center for Disease Control and Prevention (http://www.cdc.gov/). As BMI-SDS already includes information on gender and age additional sensitivity analyses were performed where these covariates were omitted. Note that the case-control analyses in GENERALIZATION step are not completely independent from the population-based analyses. In particular, controls in GENERALIZATION were individuals from the population-based samples which either had a BMI<25 for adults or a BMI percentile below the median. Due to the similarity to the original design it was nevertheless decided to report both analyses.
As secondary sensitivity analyses, we performed gender stratified analyses in all GENERALIZATION samples for the markers which we followed-up. We explored the recessive and dominant genetic model, investigated the impact of the control group cut-off for the case-control analyses (results not shown as they did not alter the conclusions drawn here) and explored joint and epistatic effects (multiple linear regression and regression trees using lm, rlm, and party of R.2.9.1) of all five loci (see Figure S3, Figure S4). To address, to some extent, problems of the ‘bias-variance trade-off’ and the ‘winners curse’ [34], the largest GENERALIZATION population-based sample KORA (n = 12,002) was chosen for this modelling. The model was tested in the Heinz-Nixdorf Recall Study sample (n = 4,646). These two samples were chosen due to their largest similarities in the recruitment and due to the availability of directly genotyped SNPs. In addition, we also explored the sample of population-based children and adolescents (GINI, LISA, Berlin School Girls; n = 3,525) separately.
Unless otherwise stated, all reported p-values are nominal, two-sided and not adjusted for multiple testing. To address multiple testing in the paediatric extreme obesity GWAS meta-analysis we applied a Bonferroni-corrected αBF≈3.1×10−8 to the quality controlled SNPs on autosomes. Confidence intervals were calculated with coverage of 95% (abbreviated 95%CI). More details on quality control and power considerations are provided in Text S1.
The study, including the protocols for subject recruitment and assessment, the informed consent for participants, were reviewed and approved by all local IRB boards.
|
10.1371/journal.ppat.1002138 | An Interaction between KSHV ORF57 and UIF Provides mRNA-Adaptor Redundancy in Herpesvirus Intronless mRNA Export | The hTREX complex mediates cellular bulk mRNA nuclear export by recruiting the nuclear export factor, TAP, via a direct interaction with the export adaptor, Aly. Intriguingly however, depletion of Aly only leads to a modest reduction in cellular mRNA nuclear export, suggesting the existence of additional mRNA nuclear export adaptor proteins. In order to efficiently export Kaposi's sarcoma-associated herpesvirus (KSHV) intronless mRNAs from the nucleus, the KSHV ORF57 protein recruits hTREX onto viral intronless mRNAs allowing access to the TAP-mediated export pathway. Similarly however, depletion of Aly only leads to a modest reduction in the nuclear export of KSHV intronless mRNAs. Herein, we identify a novel interaction between ORF57 and the cellular protein, UIF. We provide the first evidence that the ORF57-UIF interaction enables the recruitment of hTREX and TAP to KSHV intronless mRNAs in Aly-depleted cells. Strikingly, depletion of both Aly and UIF inhibits the formation of an ORF57-mediated nuclear export competent ribonucleoprotein particle and consequently prevents ORF57-mediated mRNA nuclear export and KSHV protein production. Importantly, these findings highlight that redundancy exists in the eukaryotic system for certain hTREX components involved in the mRNA nuclear export of intronless KSHV mRNAs.
| Herpesviruses hijack cellular components to enhance viral gene expression. This is particularly important for the efficient nuclear export of herpesvirus intronless mRNAs to allow the production of viral proteins. We have previously demonstrated that Kaposi's sarcoma-associated herpesvirus encodes a conserved protein, ORF57, which recruits essential cellular mRNA export proteins onto the viral intronless mRNAs to form an export competent viral ribonucleoprotein particle. Specifically, we have shown that ORF57 interacts directly with the cellular export adaptor protein, Aly, to recruit other cellular mRNA export proteins. Surprisingly however, depletion of Aly has a limited effect on both cellular and viral mRNA nuclear export levels, suggesting a degree of redundancy in the export pathways and the existence of other export adaptor proteins. Here we have identified a novel interaction between ORF57 and a second export adaptor protein, UIF. We show for the first time that the ORF57-UIF interaction allows the recruitment of the essential cellular mRNA export proteins onto viral intronless mRNA, in cells lacking Aly. However, depletion of both Aly and UIF prevents the formation of an export competent viral ribonucleoprotein particle, suggesting that either Aly or UIF must be present for efficient KSHV intronless mRNA nuclear export and protein production.
| Post-transcriptional events which regulate mRNA biogenesis are fundamental to the control of gene expression [1]. A nascent mRNA is therefore steered through multimeric RNA-protein complexes that mediate its capping, splicing, polyadenylation, nuclear export and ultimately its translation [2], [3]. A key aspect of these post-transcriptional events is that they are intrinsically linked [4]. For example, the act of splicing is coupled to the deposition of two distinct multiple protein complexes onto the mRNA which are involved in further processing events, namely the human transcription and export complex (hTREX) [5]–[7] and the exon-exon junction complex (EJC) [8]. The hTREX complex associates with the 5′end of the first exon by virtue of interactions with the cap-binding complex, and facilitates the nuclear export of the bulk mRNA through the TAP-mediated pathway [6]. In contrast, the EJC is deposited 20–24 nucleotides upstream of each exon-exon boundary and plays a role in mRNA surveillance [9] and translation enhancement [10]–.
The TREX complex is conserved from yeast to metazoans [3], [13], [14]. The human TREX complex comprises several core components: Aly (a NXF/TAP adaptor protein); UAP56 (a DEAD-box helicase); Tex1 (a protein of unknown function) and the stable multi-protein hTHO complex (hHpr1, hTho2, fSAP79, fSAP35 and fSAP24) [3]. Moreover, recent proteomic analysis has identified CIP29/Tho1 as a hTREX component that is conserved in both yeast and metazoans [15]. The precise mechanism of how hTREX is assembled onto the mRNA is not fully understood or characterised. UAP56 is thought to associate with mRNA at an early stage during the assembly of the spliceosome and functions to mediate the recruitment of Aly, CIP and the THO complex in an ATP-dependent manner to form hTREX [15], [16]. This involvement of the spliceosome in hTREX assembly reflects the splicing-dependent nature of mRNA nuclear export [16]–[18]. In addition to splicing, a functional 7-methylguanosine 5′ cap is also essential for hTREX recruitment, due to an interaction between Aly and the cap-binding complex protein, CBP80 [6]. Such cap-dependent recruitment of the export complex affords the mRNA polarity upon exiting the nuclear pore. Once assembled onto the mRNA, hTREX then instigates the recruitment of the nuclear export factor TAP, and its heterodimeric partner, p15, at the nuclear periphery, via a direct interaction with Aly [18]–[20]. TAP binding then elicits a RNA handover mechanism which results in the remodelling of the protein-mRNA interactions within the ribonucleoprotein complex [21]. Subsequently, TAP associates with the nucleoporins through central and carboxy-terminal domains, directing the ribonucleoprotein though the nuclear pore complex into the cytoplasm [22].
Surprisingly, considering the central role played by Aly in TAP recruitment, gene knockdown experiments performed in Drosophila melanogaster and Caenorhabditis elegans have shown that only UAP56, in contrast to Aly and THO-complex proteins, is required for bulk mRNA nuclear export [23]–[25]. Moreover, a genome-wide RNAi study in D. melanogaster reported that the conserved THO-complex was only required by a subset of transcripts for nuclear export [26], [27]. This data indicates a degree of redundancy is present in these pathways and suggests the existence of additional export adaptor proteins which are involved in bulk mRNA nuclear export. In support of this idea, a novel mRNA export adaptor protein has recently been identified that utilises the UAP56/TAP-mediated pathway. UAP56-interacting factor (UIF) was initially identified in silica, by virtue of sequence similarity to the characterised UAP56-binding domain found in Aly [28]. Notably, cellular expression levels of UIF appear to be linked in vivo to the relative expression of Aly, as miRNA-mediated depletion of Aly led to a dramatic increase in UIF expression. Importantly, simultaneous depletion of both Aly and UIF leads to a dramatic nuclear accumulation of bulk mRNA [28]. Therefore, it is believed that Aly and UIF bind independently to the same mRNA providing multiple export adaptor proteins to recruit multiple TAP molecules to ensure efficient mRNA nuclear export. Moreover, the observation that UIF expression increases in Aly-depleted cells is believed to be a redundancy mechanism that ensures cellular survival should Aly expression be compromised.
Given the importance of the formation of multimeric mRNA-protein complexes in mRNA biogenesis, it is not surprising that viruses manipulate and exploit these pathways. This is particularly important for herpesviruses which replicate in the host-cell nucleus and express numerous lytic intronless mRNAs. Due to the reliance of herpesviruses on the host cell machinery for efficient processing of their mRNAs, an immediate issue arises concerning the mechanism by which the viral intronless mRNAs are efficiently exported from the nucleus, given that the majority of cellular bulk mRNA nuclear export is intimately linked, and dependent upon, splicing [29]. We have investigated this potential roadblock to herpesvirus gene expression and replication in the gamma-2 herpesvirus, Kaposi's sarcoma-associated herpesvirus (KSHV) [30], which is associated with the AIDS-related malignancies Kaposi's sarcoma (KS), primary effusion lymphoma (PEL) and multicentric Castleman's disease [31]–[33]. To circumvent the roadblock of efficient intronless viral mRNA nuclear export, KSHV encodes a multi-functional protein termed ORF57/Mta. KSHV ORF57 is a functionally conserved protein found in all herpesviruses that plays a pivotal role in enhancing viral gene expression at a post-transcriptional level [34], [35]. ORF57 has been implicated in multiple steps of RNA biogenesis, including enhancing viral splicing, protecting viral RNAs from degradation to enhancing viral mRNA nuclear export and translation [36]–[39].
We have demonstrated that KSHV ORF57 promotes the nuclear export of intronless viral mRNAs via the TAP-mediated pathway, by directly interacting with the hTREX export adaptor, Aly [37]. Moreover, we investigated the composition and assembly of these export-competent intronless KSHV ribonucleoprotein particles (vRNP) and showed that ORF57 functions to recruit the complete hTREX complex to intronless viral mRNA, an event that is essential for viral intronless mRNA export and KSHV replication [37]. Furthermore, these properties are also conserved in other gamma-2 herpesvirus ORF57 homologues, such as the Herpesvirus saimiri (HVS) ORF57 protein [40], [41]. These data suggest that Aly is essential for ORF57-mediated KSHV intronless mRNA export, as well as playing an important role in mRNA nuclear export in other herpesviruses. However, experiments involving siRNA-mediated depletion of Aly report only a modest effect on ORF57-mediated KSHV intronless mRNA export, although only partial depletion of Aly was achieved [42]. This data correlates with depletion-related studies on the role of Aly in mRNA export in higher eukaryotes where, surprisingly, Aly has been shown to be dispensable in mRNA export [23], [24]. Similar stories are also evident for other herpesviruses mRNA export proteins. For example, an observed interaction between ICP27 (the HSV-1 ORF57 homologue) and Aly was initially reported as important for HSV-1 mRNA export [43]. However, subsequent functional studies using siRNA-mediated depletion of Aly led to the authors suggesting that Aly is not essential for ICP27-mediated HSV-1 mRNA export [44]. This suggests that additional cellular mRNA export proteins play important roles in herpesvirus intronless mRNA export. Indeed, recently it has been demonstrated that the SR proteins, SRp20 and 9G8, can contribute to efficient export of herpes simplex virus 1 mRNAs [45].
Herein we report a novel interaction between the KSHV ORF57 protein and the recently identified mRNA export adaptor protein, UIF. Moreover, we provide data to suggest that ORF57 may preferentially bind Aly compared to UIF. Furthermore, we investigate whether the linked expression of UIF and Aly plays a role in the apparent redundancy of Aly in herpesvirus intronless mRNA nuclear export. We provide the first evidence that the ORF57-UIF interaction enables the recruitment of the complete hTREX and the nuclear export factor, TAP, to KSHV intronless mRNA in Aly-depleted cells. Strikingly, we demonstrate that depletion of both Aly and UIF inhibit the formation of an ORF57-mediated nuclear export competent ribonucleoprotein particle and consequently prevent ORF57-mediated nuclear export of intronless viral mRNAs and KSHV protein production. Importantly, these findings highlight that redundancy exists in the eukaryotic system for certain hTREX components involved in the mRNA nuclear export of intronless KSHV mRNAs.
KSHV ORF57 interacts directly with the cellular export adaptor protein Aly to recruit cellular hTREX, comprising UAP56 and the hTHO complex, onto a viral intronless mRNA to form an export competent ribonucleoprotein particle [37]. However, ORF57 and homologues can mediate nuclear export of an intronless viral mRNA in Aly-depleted cells [42], suggesting that alternative export pathways may be targeted by the ORF57 protein. Therefore, to determine whether ORF57 interacts with alternative export adaptor proteins, GST-pulldown and co-immunoprecipitations assays were performed to assess if ORF57 interacted with the recently identified UAP56 interacting protein, UIF. Initially, recombinant GST-, GST-UAP56 or GST-ORF57 fusion proteins were produced and used in GST-pulldown assays. It must be noted however, that although full length GST-ORF57 is produced, a large proportion of the product is degraded as previously observed [37]. GST-pulldown experiments were therefore performed using equal amounts of total protein from each GST construct immobilised to beads followed by incubation with 293T cell lysates transfected with pUIF-Flag. Analysis showed that UIF interacted with both UAP56 and KSHV ORF57 (Figure 1A). To confirm these results co-immunoprecipitation experiments were also performed. 293T cells were transfected with either pEGFP, pUAP56-myc or pORF57GFP in the presence of pUIF-Flag and used in co-immunoprecipitation experiments with GFP or UAP56-specific antibodies. Results confirmed the interaction between UIF and KSHV ORF57 (Figure 1B).
We have previously identified an ORF57 mutant protein, ORF57PmutGFP, which is unable to interact with Aly and therefore recruit the remainder of the hTREX complex onto viral intronless mRNAs [37]. Moreover, we demonstrated that this mutant is unable to efficiently export viral intronless mRNA from the nucleus suggesting that the recruitment of a complete hTREX complex is required for ORF57-mediated nuclear export. ORF57PmutGFP contains site-directed alterations of two proline residues within a PxxP poly-proline motif, situated in the previously identified minimal Aly-binding domain encompassing residues 181–215. We have previously demonstrated that although ORF57PmutGFP is unable to bind Aly, it still retains features of the wild type ORF57 protein, namely localising to nuclear speckles, the ability to homodimerise, bind KSHV RTA and bind intronless viral mRNA [37]. To assess whether this mutant could interact with UIF, GST-pulldown experiments and co-immunoprecipitation experiments were performed as described above using GST-ORF57Pmut and pORF57PmutGFP, respectively. In both cases the mutant ORF57 protein, which fails to bind Aly, also lacks the ability to interact with UIF (Figure 1A and 1B). Importantly, this suggests that the failure of ORF57PmutGFP to recruit hTREX and efficiently export intronless viral mRNAs from the nucleus may be due to the inability to bind either Aly or UIF.
To determine if the interaction between ORF57 and UIF depended on RNA bridging, co-immunoprecipitation experiments were repeated in the absence and presence of RNase. 293T cells were transfected with either pEGFP or pORF57GFP in the presence of pUIF-Flag and co-immunoprecipitation assays were performed using a polyclonal Flag-specific antibody. In addition, no antibody and a negative control antibody (α-SC-35) were also used in the analysis. ORF57 was readily precipitated using the Flag-specific antibody in contrast to negative controls. Moreover, the observed interaction was still detected in the presence of RNase suggesting the interactions are not due to RNA bridging (Figure 1C). To ensure the RNase treatment was effective the immunoprecipitations were also blotted with an Aly-specific antibody. Results show that the UIF-Aly interaction is RNA dependent as previously described [28], [36].
In order to address potential overexpression artefacts of the above co-immunoprecipitation experiments and also determine whether ORF57 interacts with UIF during KSHV lytic replication, latently-infected BCBL-1 cells remained uninduced or reactivated using the phorbol ester, TPA. Lytic expression was confirmed by the detection of ORF57 using Western blot analysis in the reactivated samples (Figure 2). Uninduced and reactivated cell lysates were then incubated with no antibody control, ORF57- or UIF-specific antibodies. Reciprocal western blot analysis using the antibodies in reverse demonstrated that ORF57 interacts with UIF during KSHV lytic replication (Figure 2). Therefore, these data provide the first evidence of a viral protein associating with UIF.
ORF57 is a nucleocytoplasmic protein that is predominately observed in the nucleus, specifically colocalising with nuclear speckle and nucleoli-associated proteins [42], [46]. Therefore, we were interested to determine whether ORF57 colocalises with UIF in either of these subnuclear domains. To this end, 293T cells were cultured on poly-L lysine coated coverslips and transfected with either pORF57-mCherry or pUIF-GFP alone or in combination. The subcellular localisation of ORF57 and UIF were observed via direct fluorescence, in addition indirect-immunofluorescence was performed to identify nuclear speckles and the nucleolus using SC35- (Figure 3Bii) and B23- (Figure 3Dii) specific antibodies, respectively. As previously observed ORF57 colocalises with both subnuclear domain markers (Figure 3Bii and 3Dii). Moreover, UIF was also found to localise with these subnuclear structures and also colocalises with the ORF57 protein (Figure 3B and 3D). However, results demonstrate that the majority of ORF57 and UIF colocalise in the nucleolus whereas only a proportion of ORF57 and UIF colocalise with the nuclear speckle marker, SC35.
One major difference between the mRNA export adaptor proteins Aly and UIF is the mechanism they utilise to be loaded onto mRNA. Aly has been shown to associate with mRNA in a UAP56 and splicing-dependent manner [47], in contrast UIF is loaded onto mRNA via the histone chaperone FACT [28]. We have previously demonstrated that ORF57 is required for the recruitment of Aly and the remainder of the hTREX complex onto viral intronless mRNAs, therefore we were intrigued to determine if UIF could associate with intronless viral mRNAs in an ORF57-independent manner using RNA-immunoprecipitation (RNA-IP) assays. A vector expressing KSHV ORF47 (a late structural intronless gene) was transfected into 293T cells with either pEGFP or pORF57GFP. Total cell lysates were then used in immunoprecipitations performed with either No, Y14- (negative control), UIF- or GFP-specific antibodies and the amount of ORF47 precipitated was measured by qRT-PCR. RNA-IPs performed on cell extracts transfected with pORF47 and pEGFP failed to show an interaction between UIF and the viral intronless ORF47 mRNA (Figure 4). In contrast, extracts from cells transfected with both pORF47 and pORF57GFP displayed a clear interaction between UIF, ORF57GFP and the intronless viral ORF47 mRNA (Figure 4). These data show that although UIF can associate with cellular spliced and unspliced single exon cellular mRNAs, ORF57 is required for the recruitment of UIF onto intronless viral mRNA.
We have previously demonstrated that the nuclear export adaptor protein, Aly, is recruited to viral intronless mRNAs in a splicing-independent manner by directly interacting with ORF57. Once bound it then leads to the recruitment of the remaining components of hTREX, in turn leading to efficient nuclear export of these viral intronless mRNAs via a TAP-mediated pathway [37]. We therefore next sought to determine if UIF can perform a similar function by linking ORF57 to hTREX components such as UAP56. Initially, we determined whether ORF57 interacted with UIF directly using GST-pulldown assays. Recombinant GST- and GST-ORF57 proteins were immobilised to beads and incubated with purified recombinant UIF-6xHis or a negative control purified recombinant HVS ORF73-6xHis protein [48]. UIF-6xHis was precipitated by GST-ORF57 but not the negative GST control, moreover ORF73-6xHis failed to interact with either GST or GST-ORF57 (Figure 5A). These data provide further support for the direct interaction between ORF57 and UIF.
Given the fact that ORF57 and UIF interact directly, we next determined whether UIF can bridge the interaction between ORF57 and hTREX components, such as UAP56, which we have previously shown fails to interact with ORF57 directly [37]. Reconstitutive GST-pulldowns were therefore performed using recombinant GST- and GST-UAP56 proteins immobilised to beads and incubating with either purified recombinant ORF57-6xHis or purified recombinant UIF-6xHis alone or in combination. No interaction with GST or GST-UAP56 was observed in the presence of ORF57-6xHis alone. In contrast, an interaction between GST-UAP56 and ORF57 was observed in the presence of purified UIF-6xHis protein (Figure 5B), suggesting that UIF can facilitate the formation of the ORF57-hTREX complex. This provides the first evidence to demonstrate that UIF could function to assemble the hTREX complex on viral intronless mRNAs.
The above data demonstrate that UIF interacts directly with ORF57 and suggest that it can function to bridge an interaction between ORF57 and the remaining hTREX components. This mechanism is similar to our previous observations regarding the functional significance of the Aly-ORF57 interaction, and therefore leads to the intriguing question of whether ORF57 has a preference for Aly binding over UIF or vice versa. To address this question we performed competitive GST pulldown assays. Recombinant GST-ORF57 protein was immobilised to beads and incubated with a constant amount (1 µg) of purified recombinant Aly-6xHis protein, in addition the pulldown was spiked with increasing amounts of purified recombinant UIF-6xHis protein (0, 0.5, 1, 2, 3 µg). Western blot analysis was then performed using an Aly-specific antibody. Results demonstrate that the binding of Aly to GST-ORF57 is only slightly reduced in the presence of increasing amount of UIF (Figure 6Ai), suggesting that UIF cannot out-compete Aly for ORF57 binding. Similar spiked experiments were performed using a constant amount of UIF and increasing amounts of Aly. In contrast, results showed that even low quantities of Aly led to a dramatic loss of UIF binding to the ORF57 protein (Figure 6Aii). These results reveal that Aly can out-compete UIF for ORF57 binding, suggesting that ORF57 may preferentially bind Aly to form an export competent ribonucleoprotein particle.
However as shown in Figure 1A, although bacterial expression of full length GST-ORF57 results in a full length ORF57 protein, a large proportion of degraded products are also produced. Therefore, to further assess the possibility that ORF57 may interact with Aly preferentially over UIF, dose-dependent coimmunoprecipitation assays were performed. To this end, 293T cells were cotransfected with 0.5 ug of pORF57GFP and 0.5 ug of pAly-myc, in addition to increasing amounts of pUIF-Flag (0, 0.1, 0.5, 0.8, 1.2 ug). After 24 hours, cell lysates were incubated with GFP-TRAP-Affinity agarose beads and the amount of precipitated Aly was identified by immunoblotting with a Myc-specific antibody. Results again show that the binding of Aly is only slightly reduced in the presence of increasing amounts of UIF (Figure 6Bi). Moreover, reciprocal dose-dependent coimmunoprecipitations were performed using 0.5 ug of pORF57GFP and 0.5 ug of pUIF-Flag, in addition to increasing amounts of pAly-myc (0, 0.1, 0.5, 0.8, 1.2 ug). In contrast, results suggest that higher concentrations of Aly can significantly reduce the amount of precipitated UIF (Figure 6Bii). These results corroborate the above GST pulldown assays and suggest that ORF57 may preferentially bind Aly over UIF to form an export competent ribonucleoprotein particle.
Having established that both Aly and UIF can bridge an interaction between ORF57 and hTREX components, such as UAP56, we next sought to determine the effect of depleting Aly and UIF either singularly, or in combination, on the ability of ORF57 to form an export competent ribonucleoprotein particle containing the complete hTREX complex and the nuclear export factor TAP. To this end, we have utilised doxycycline inducible 293 cell lines expressing miRNAs targeting Aly, UIF or both Aly and UIF [28]. Effective depletion of Aly, UIF or both proteins can be observed after 72 hours post doxycycline induction (Figure 7A). However, a caveat of this type of experiment is that depletion of multiple mRNA export factors in combination may firstly be toxic to the host cell and second inhibit the expression of ORF57 itself as recently reported [49]. Characterisation of the cell viability and growth of the cells depleted with both Aly and UIF has previously been performed and results show they are viable and grow for 4 days post knockdown prior to cell death at day 6 [28]. Therefore all experiments using these cell lines were performed in this 4 day window. Moreover, to ensure ORF57 protein production, cells were transfected at 48 hours prior to complete Aly or UIF depletion at 72 hours.
To assess viral ribonucleoprotein particle formation, wild type 293 cells and each miRNA-targeted cell line were induced with doxycycline to deplete the respective proteins and after 48 hours' induction, each cell line was transfected with pORF57GFP. After a further 24 hours when maximum Aly and UIF depletion has occurred, cell lysates were used in co-immunoprecipitation experiments using GFP-TRAP-Affinity agarose beads. Western blot analysis was then performed using UAP56-, FSAP79- (a hTHO complex component) and TAP-specific antibodies. As a negative control, GFP was also transfected into the wild type 293 cell line and co-immunoprecipitations performed using GFP-TRAP-Affinity agarose beads, no interactions were observed with any of the hTREX components or TAP. However, results showed that expression of ORF57 in the wild type 293 cell line led to the precipitation of UAP56, FSAP79 and TAP suggesting that ORF57 expression leads to the formation of an export competent ribonucleoprotein particle (Figure 7B). Similar complex formation was observed in cell lines depleted singularly for Aly and UIF, where ORF57 can precipitate UAP56, FSAP79 and TAP (Figure 7B). In contrast, depletion of Aly and UIF in combination significantly reduced the interaction between ORF57 and the hTREX components and the nuclear export factor TAP. Importantly, these data demonstrate that either Aly or UIF are required for the formation of an ORF57-mediated nuclear export competent ribonucleoprotein particle.
The above data suggest that ORF57 must interact with either export adaptor protein, Aly or UIF, to recruit hTREX and the nuclear export protein TAP, to form an export competent ribonucleoprotein particle. Therefore, we next determined whether both UIF and Aly were required for efficient ORF57-mediated nuclear export of viral intronless mRNAs. To this end, we assessed the ability of ORF57 to enhance the nuclear export of the KSHV intronless ORF47 mRNA, using a previously described assay to compare the accumulation of ORF47 mRNA in the cytoplasm [46]. Essentially, cells are transfected with a plasmid expressing the intronless KSHV ORF47 gene in addition to either GFP or wild type ORF57 constructs. After 24 hours, RNA is extracted from total and cytoplasmic fractions and RNA levels quantified using qRT-PCR. Total RNA levels are assessed to ensure similar expression levels of the ORF47 mRNA in each sample, where an increase in cytoplasmic levels of ORF47 mRNA signifies an increase in ORF57-mediated mRNA export levels. Therefore, to assess the ability of ORF57 to export ORF47 mRNA from the nucleus in the absence of either UIF or Aly or both, wild type 293 cells and each miRNA-targeted cell line were induced with doxycycline to deplete the respective proteins and after 48 hours induction, each cell line was transfected with pORF57GFP and pORF47. Again, this allowed sufficient time to express ORF57 prior to optimal export adaptor protein depletion. After a further 24 hours, RNA was extracted from total and cytoplasmic fractions and ORF47 levels assessed by qRT-PCR. Results demonstrated that ORF47 mRNA levels from total cell fractions are similar in wild type and the depleted cell lines. Moreover, in the control 293 cell line ORF47 mRNA accumulates in the cytoplasm in the presence of ORF57 as previously described [46]. Similarly, mRNA can accumulate in the cytoplasm of cells depleted singularly for Aly and UIF, however, a reduction in export efficiency was observed of approximately 40% and 23% of wild type levels, respectively (Figure 8A). In contrast, depletion of both Aly and UIF together led to a dramatic reduction of ORF47 mRNA accumulation in the cytoplasm with an 80% decrease compared to wild type levels (Figure 8A).
We next tested whether the observed reduction in the ability of ORF57 to export intronless mRNAs from the nucleus in cell lines depleted for Aly and UIF had any effect on KSHV protein production. To this end, the wild type 293 cells and each miRNA-targeted cell line were induced with doxycycline to deplete the respective proteins and after 48 hours induction, each cell line was infected with recombinant KSHV at a MOI = 1. This time point was used to allow sufficient time to express ORF57 prior to optimal export adaptor protein depletion. After a further 48 hours, the cell lysates were analysed by immunoblotting using KSHV glycoprotein B- and ORF4-specific antibodies. Results showed that gB protein expression in cell lines singularly depleted for either Aly or UIF was reduced by ∼42% and ∼10%, respectively, whereas little or no reduction was observed for ORF4 protein levels in the singularly depleted cells. Strikingly, however depletion of both Aly and UIF led to a dramatic reduction in both gB and ORF4 expression levels of 78% and 79%, respectively (Figure 8B). These results suggest that depletion of UIF has limited if any effect of virus replication, however, depletion of UIF together with Aly had a dramatic negative effect on KSHV protein production. However, it must be noted that this reduction in protein levels may also stem from altered levels of one or more key cellular proteins involved in KSHV lytic protein production.
Taken together, our data suggest that either one of the cellular nuclear export adaptor proteins, Aly or UIF, is required for the formation of an ORF57-mediated nuclear export competent ribonucleoprotein particle which is essential for KSHV protein production.
The nuclear export of bulk mRNA is mediated by the conserved heterodimeric export receptor, TAP/p15 [3]. Cellular mRNAs gain access to TAP/p15 via interaction with a group of RNA-binding proteins termed export adaptors. The first mRNA export adaptor to be identified in higher systems was Aly/REF, and subsequent work from a number of groups led to the current model where Aly is recruited to the 5′ cap of spliced mRNA along with several other proteins to form a multimeric protein complex termed hTREX [6]. The hTREX complex facilitates the association of bound mRNAs with TAP/p15 thus licensing nuclear export. In addition to Aly, several other hTREX components have been identified including the DEAD-box helicase UAP56, hTex1, the multi-protein THO complex and recently, CIP29 [15]. While the underlying mechanism of hTREX-mediated mRNA export is loosely understood, the individual functions of the hTREX components remain elusive.
Perhaps the greatest enigma surrounding TAP/p15-mediated mRNA export is the apparent redundancy that exists for certain hTREX proteins. This is particularly true for Aly, where a number of different studies have shown that the metazoan homologue, REF1, is not required for the bulk export of mRNA [23], [24]. These studies suggest that additional mRNA export adaptors must exist which can function to link nascent mRNA to the TAP/p15 heterodimer. Moreover, this raises the intriguing possibility that, via the use of numerous different mRNA export adaptor proteins, a further layer of control may exist to regulate gene expression. Indeed, several recent reports have highlighted that differences exist within component members of mRNA export complexes associated with different classes of mRNAs. For example, HSP70 mRNA only requires Aly and the co-adaptor Thoc5 to mediate TAP recruitment [50]. Moreover, an alternative mRNA export (AREX) complex, distinct to hTREX has recently been identified which comprises the related RNA helicase URH49, instead of UAP56 [51]. Interestingly, each helicase regulates a specific set of mRNAs associated with distinct subsets of key mitotic regulators. In addition, members of the SPEN family of proteins, RBM15 and OTT3 are functionally similar, in that they can bind RNA and physically interact with TAP. However, the association of OTT3 with TAP is attenuated compared to RBM15, leading to speculation that strong and weak variants exist that may function during developmental or tissue specific mRNA processing events [52]. These data galvanise the hypothesis that ultimately it is the recruitment of TAP/p15 that is required for nuclear export, and that one function of the export adaptor proteins is to provide selectivity to this system. Such a hypothesis is consistent with, and offers an explanation to, conflicting data regarding the nuclear export of KSHV intronless mRNAs.
Herpesviruses hijack the TAP/p15-mediated mRNA export pathway in order to enhance the nuclear export of viral intronless mRNA. We have previously shown that during KSHV replication the virus-encoded ORF57 protein procures the hTREX complex (and subsequently TAP/p15) via a direct interaction with Aly, facilitating the efficient export of KSHV intronless mRNAs [37]. We proposed therefore, that as the ORF57-Aly interaction provides the link between the virus mRNA and hTREX, it was likely that Aly would be essential for KSHV mRNA export. This hypothesis was supported by data showing that an ORF57 mutant, ORF57PmutGFP, unable to bind Aly was no longer functional in virus mRNA export. However, similarly to previous studies in D. melanogaster and C. elegans, siRNA-mediated depletion of Aly did not translate to a decrease in ORF57-mediated nuclear export of KSHV intronless mRNA, although only partial knockdown of Aly was observed [42]. Correspondingly, the HSV homologue of ORF57, ICP27, was shown to directly interact with Aly. Moreover, studies in Xenopus laevis oocytes showed ICP27 dramatically stimulated the export of intronless viral mRNAs, and a mutant ICP27 protein that failed to interact with REF is inactive in viral mRNA export [43]. Again however, siRNA-mediated depletion of Aly has been shown not to affect HSV-1 mRNA export [44].
Herein, we demonstrate that redundancy exists in the eukaryotic system for certain hTREX components involved in the mRNA nuclear export of intronless KSHV mRNAs. Evidence for such redundancy in export adapter proteins was recently provided by the identification of a second mRNA export adaptor protein, UIF [28]. Importantly, cellular expression levels of UIF appear to be linked in vivo to the relative expression of Aly, as depletion of Aly leads to a dramatic increase in UIF expression. This would therefore account for the modest reduction in mRNA nuclear export in Aly-depleted cells. Indeed, as shown in Figure 1 and 5, ORF57 interacts directly with UIF and thus is able to recruit hTREX/TAP/p15 allowing efficient intronless virus mRNA nuclear export in Aly-depleted cells (Figure 8).
Recent analysis has also suggested that additional mechanisms exist to ensure the nuclear export of viral transcripts in other herpesviruses. For example, ICP27 can bind directly to TAP, suggesting ICP27 can bypass nuclear export adapter proteins [53]. However, although analysis of ICP27 mutants unable to interact with TAP export showed greatly reduced intronless viral mRNA export, it was not completely abolished suggesting other cellular proteins may have a role. Indeed, recent analysis has shown that nuclear accumulation of HSV-1 mRNA is reduced when cells were treated with siRNAs specific for the SR proteins, SRp20 and 9G8, confirming that other cellular export proteins, such as SR proteins, can contribute to HSV-1 mRNA nuclear export [45]. Similarly, the EBV ORF57 homologue, SM/EB2, can interact with SRp20, although to date, this interaction has been implicated in alternative splicing mechanisms [54]. However, EBV SM/EB2 has been previously shown to interact with alternative cellular export factors, such as CRM-1 [55]. An alternative approach may be employed by the hCMV ORF57 homologue, UL69, which interacts with other hTREX proteins required for bulk mRNA nuclear export, such as UAP56 [56]. However, current work is ongoing to determine if these homologues interact with UIF. Moreover, the role of UIF may also have wider implications in the field of virology. Influenza A virus produces capped and polyadenylated mRNAs in the nucleus of infected cells that resemble mature cellular mRNAs, which require export by the TAP-mediated pathway [57]. Depletion of Aly had little effect on viral mRNA export, but reduction of UAP56 levels strongly inhibited trafficking and/or translation of influenza mRNAs [58]. It will now be interesting to determine whether UIF also substitutes for Aly function in this viral system.
There are however, some important mechanistic differences between Aly and UIF which have implications for KSHV intronless mRNA nuclear export. The hTREX component, CIP29, bridges the Aly-UAP56 interaction to form a trimeric complex that is assembled in an ATP-dependent manner [15]. Importantly, the recruitment of Aly to the mRNA requires an interaction with the 5′ cap and is dependent on splicing [6]. However, UIF appears to be co-transcriptionally loaded onto burgeoning mRNAs via an interaction with the histone chaperone, FACT [28]. It appears therefore that Aly and UIF are deposited onto the same mRNA separately and independently, a hypothesis supported by ribonuclease-treated co-immunoprecipitation experiments, which show that the interaction between Aly and UIF is facilitated by RNA-bridging [28], [36]. These data suggest that there are two distinct cellular mechanisms that can each recruit TAP to an mRNA. This raises a number of interesting questions with regards to how ORF57 orchestrates the recruitment of hTREX (and ultimately TAP/p15) via UIF. As seen in Figure 4, UIF is recruited to KSHV intronless mRNA only in the presence of ORF57, this is in stark contrast to the mechanism by which UIF is loaded onto cellular mRNA. Why UIF is not loaded onto KSHV intronless transcripts via FACT is unclear. One possible explanation is that FACT does not interact with RNA polymerase II during the transcription of ORF47 mRNA in this assay, possibly due to incomplete chromatinisation of vector DNA. Alternatively, recruitment of UIF to both spliced and unspliced mRNA maybe partially dependent on UAP56 and we have previously shown that UAP56 recruitment to KSHV mRNA is dependent on the ORF57 protein [37].
As mentioned above, Aly and UIF are loaded separately onto the same cellular mRNA via different mechanisms and both function to ultimately recruit TAP/p15 to the mRNA via interactions with hTREX. Intriguingly, we show in Figure 6, that ORF57 may preferentially bind to Aly over UIF, using both competitive GST-pulldown and dose-dependent coimmunoprecipitation assays. Why KSHV ORF57 would evolve to preferentially bind Aly over UIF is at present uncertain. One possibility is that Aly is the major export adaptor protein and UIF forms a backup or default pathway. This is not without precedent as proteins expression levels suggest that Aly is more abundantly expressed than UIF and UIF protein levels significantly increase in Aly-depleted cells [28]. Alternatively, it is possible that ORF57 may have a higher affinity for Aly due to important functional differences in how the Aly export adaptor recruits the remaining hTREX components to virus mRNA, compared with UIF. Alternatively, Aly and UIF could recruit different components of the hTREX complex to a KSHV mRNA, highlighted by the Aly-specific recruitment of CIP29, and that the export of KSHV intronless mRNA is more reliant on these Aly-recruited hTREX proteins.
As discussed earlier, a number of siRNA-mediated studies have proposed that Aly is not essential for KSHV intronless mRNA export. However, we have previously described an ORF57 mutant protein, ORF57Pmut, which is unable to interact with Aly and failed to export viral intronless mRNAs [37]. The region mutated in ORF57Pmut maps to a PxxP motif in the N-terminal region of the protein. It is not known whether the PxxP motif mutated in ORF57Pmut is a direct interaction site for Aly, or if this mutant confers some structural change of ORF57 in the Aly binding region. Importantly, herein we have shown that this mutant is also unable to interact with UIF, suggesting that ORF57Pmut is ‘dead’ with regards to export adaptor interaction. This explains therefore why this mutant is unable to export viral intronless mRNAs, as it is unable to bind to either Aly or UIF (Figure 1). This result is also confirmed by depletion of both these export adaptors which lead to a block in KSHV mRNA nuclear export. Importantly, Aly depletion in these and previous studies have shown that UIF expression is increased and therefore UIF probably replaces Aly as the dominant export adaptor protein. It is tempting to speculate that the link between increased UIF expression in Aly-depleted cells is a redundancy mechanism that ensures cellular survival should Aly expression be compromised.
The fact that ORF57Pmut is unable to interact with both Aly and UIF would suggest that the PxxP motif is either the complete ORF57 interacting motif, or part of the interacting motif, for Aly and UIF binding, and that the binding sites for the two proteins are either identical or overlap to some degree. Alternatively, the PxxP motif may cause a loss of interaction of both Aly and UIF by altering the structure of each of the binding sites. Importantly, our competition assays demonstrate that ORF57 may preferentially bind to Aly over UIF. These observations suggest that Aly and UIF may compete for a binding site on ORF57, and further studies are now required to determine if this is the case. Interestingly, we have recently identified the key residues that interact directly with Aly in both HSV-1 ICP27 and herpesvirus saimiri (HVS) ORF57 using solution-state NMR and mapped this interaction to a WRV/A motif [59]. Due to the sequence differences between ORF57 homologues this motif does not appear in KSHV ORF57, although the region of KSHV ORF57 that interacts with Aly has been mapped to the N terminus (aa 1–215). We are currently investigating the interacting residues for both Aly and UIF within this N-terminal region of KSHV ORF57 using solution-state NMR.
In summary, our results demonstrate the first known interaction between a viral protein and the newly described export adaptor protein, UIF. Importantly, the ORF57-UIF interaction is sufficient to recruit the hTREX complex onto viral intronless mRNAs and highlights that redundancy exists in the eukaryotic system for certain hTREX components involved in the mRNA nuclear export of intronless KSHV mRNAs. It now seems clear that the events which lead up to TAP/p15 recruitment to the mRNA are not linear. Indeed, it appears that multiple pathways exist by which an mRNA can bind TAP/p15 and be licensed for nuclear export. The existence of numerous export adaptor proteins may partly be explained in terms of redundancy but there is strong evidence to suggest that this also generates specificity within the system.
Details of oligonucleotides used for qRT-PCR have been described previously [37], [46]. KSHV, hTREX and UIF-related plasmid constructs have been described previously [6], [28], [37]. KSHV ORF57- and ORF4- specific antibodies were a kind gift from Gary Hayward (Johns Hopkins, Baltimore) and Brad Spiller (Cardiff University), respectively. Antibodies against SC-35, Flag, Myc and Aly (Sigma), GFP and mCherry (Clontech), B23 (Santa Cruz), KSHV gB (Abcam) and GAPDH (Abcam) were purchased from their respective suppliers. Western blot analysis was carried out using specific antibodies at 1∶1000 dilution, except for UIF-specific antibody (1∶250) and GFP-specific antibody (1∶5000). Antibodies used for immunofluorescence studies were at a dilution of 1∶250.
293 inducible cells lines which specifically deplete Aly, UIF and both Aly and UIF have been previously reported [28]. They were produced using the FLP-In T-REX 293 cells (Invitrogen) system to express miRNAs to each specific export adapter protein, miRNA sequences are detailed in Hautbergue et al., 2009. HEK-293T cells, HEK-293T BAC36 cells harbouring a recombinant KSHV BAC36 genome and FLP-In T-REX 293 cells were cultured in Dulbecco's modified Eagle medium (DMEM, Invitrogen) supplemented with glutamine, 10% foetal calf serum (FCS, Invitrogen) and penicillin-streptomycin. 293T BAC36 cells were reactivated using TPA (20 ng/ml) for the designated time. miRNA expression from FLP-IN T-REX 293 cells was induced with 2 µg/ml doxycyclin (Sigma) for the designated time. Plasmid transfections were carried out using Lipofectamine 2000 (Invitrogen) or GeneJuice (Novagen) and were carried out as per the manufacturer's instructions. rKSHV.219 (KSHV) was produced from the latently infected Vero line [60]. This virus specifies red fluorescent protein (RFP) from the KSHV lytic PAN promoter, green fluorescent protein (GFP) from the EF-1α promoter, and encodes a puromycin resistance gene. Vero cells stably infected with rKSHV.219 were maintained in MEM Eagles medium, 2.2 g/L NaHCO3, 10% fetal calf serum, puromycin (5 ug/ml) (Sigma-Aldrich, Poole, UK) and penicillin and streptomycin (Invitrogen). To induce KSHV lytic replication in these cells, they were infected with BacK50, a baculovirus construct encoding the lytic switch RTA protein, and treated with 1.25 mM sodium butyrate (Sigma). 48 h after KSHV reactivation, the supernatant was harvested, centrifuged (500g, 15 mins) to remove cell debris, and the virions concentrated by centrifugation (65,000g, 4 h). The virion pellet was resuspended overnight in EBM2 medium (Lonza, Clonetics). The rKSHV.219 titre was determined on 293 cells, quantifying GFP-positive cells by fluorescence microscopy. 293 and 293 derived cells were infected with KSHV. To this end, cells were plated at 1.25×105 cells per well in 24-well plates for infection and cultured overnight. The culture medium was then removed and replaced with virus diluted in EBM2 basal media after 24 hrs. Cells were then centrifuged for 30 min at 420× g at room temperature. Cells were transferred to a 37°C incubator (5% CO2, humidified) for 90 min. Virus supernatant was removed and cells were washed once in cell culture media and incubated for 48 hrs before being harvested.
Recombinant GST, GST-ORF57, GST-ORF57pmut, GST-UAP56 and UIF-His, Aly-His and ORF73-His were expressed and purified as previously described [36],[37],[61]. Purification of Baculovirus recombinant ORF57-6xHis was as per the manufacturer's instructions (Invitrogen) using the pFASTBac protocol.
GST pull-down experiments and co-immunoprecipitations were performed as described previously [62], [63]. GFP-TRAP-Affinity (Chromotek) experiments were performed as per the manufacturer's instructions. RNA immunoprecipitation experiments were carried out as follows: 1×107 adherent 293T cells were transiently transfected with appropriate GFP-containing plasmid DNA. After the appropriate amount of time cells were washed in ice-cold PBS and UV irradiated (900 mJ/cm2) using a Stratalinker 2400 (Stratagene) to crosslink protein and RNA. Cells were then scraped, transferred to an RNA-free tube and pelleted at 300× g for 3 min. Cell pellets were then resuspended in 2 ml lysis buffer [Dulbecco's PBS, 1% Nonidet P-40 (v/v), 1 µl/ml RNaseOUT (Invitrogen), 1× Complete EDTA-free Protease inhibitor (Roche)]. Cells were left on ice for 30 min before being centrifuged for 10 min at 15,000× g. The clear lysate was then transferred to a clean RNA-free tube. 1 ml of the cleared lysate was added to 30 µl pre-washed GFP-TRAP-Affinity agarose beads (Chromotek) per IP and immunoprecipitated at 4°C with end-over-end mixing for 4 hrs. Beads were washed 3 times in ice-cold PBS containing 1× Complete EDTA-free protease inhibitor (Roche) followed by a further 2 times in PBS. Beads where then incubated in protease buffer (Dulbecco's PBS, 1% Nonidet P-40 (v/v), 0.1% SDS (w/v), 0.5 mg/ml Proteinase K) for 30 min at 50°C. RNA was extracted using TRIzol reagent (Invitrogen) as per the manufacturer's directions. cDNA was then produced from 10 µl of RNA using Superscript II RT (Invitrogen) and qPCR performed to analyse the relative levels of cDNA. RT-ve samples were used as controls.
Bacterially expressed GST-tagged ORF57 was immobilised to GST beads and used for GST pulldown competition assays. Recombinant His-tagged Aly or UIF was expressed and purified as previously described [36], [37]. Equal amounts of Aly-His (1 µg) were used in the pull-downs with increasing amounts of UIF-His (0, 0.5, 1, 2, 3 µg). The converse experiments were also performed with equal amounts of UIF-His (1 µg) and increasing amounts of Aly-His (0, 0.5, 1, 2, 3 µg).
To assess ORF57-mediated ORF47 mRNA export efficiency, 293T and inducible cells were cotransfected with ORF47 and ORF57 expression constructs. After 24 hours RNA was extracted from total and cytoplasmic fractions using TRIzol (Invitrogen) as described by the manufacturer. Cytoplasmic fractions were produced by lysis of cells in 200 µl of PBS 1% Triton-X 100(v/v) containing 40 U of RNaseOUT (Invitrogen), prior to TRIzol purification. RNA was DNase treated using the Ambion DNase-free kit, as per the manufacturer's instructions, and RNA (1 µg) from each fraction was reverse transcribed with SuperScript II (Invitrogen), as per the manufacturer's instructions, using oligo(dT) primers (Promega). 10 ng of cDNA was used as template in SensiMixPlus SYBR qPCR reactions (Quantace), as per manufacturer's instructions, using a Rotor-Gene Q 5plex HRM Platform (Qiagen), with a standard 3-step melt program (95°C for 15 seconds, 60°C for 30 seconds, 72°C for 20 seconds). With GAPDH as internal control mRNA, quantitative analysis was performed using the comparative CT method as previously described [46].
Immunofluorescence staining and visualisation by microscopy was carried out as previously described [64]. Visualisation was performed on an LSM 510 Meta confocal microscope (Zeiss) and images were analysed using the LSM imaging software (Zeiss).
|
10.1371/journal.pntd.0005081 | Knowledge, Attitude and Practices of Vector-Borne Disease Prevention during the Emergence of a New Arbovirus: Implications for the Control of Chikungunya Virus in French Guiana | During the last decade, French Guiana has been affected by major dengue fever outbreaks. Although this arbovirus has been a focus of many awareness campaigns, very little information is available about beliefs, attitudes and behaviors regarding vector-borne diseases among the population of French Guiana. During the first outbreak of the chikungunya virus, a quantitative survey was conducted among high school students to study experiences, practices and perceptions related to mosquito-borne diseases and to identify socio-demographic, cognitive and environmental factors that could be associated with the engagement in protective behaviors.
A cross-sectional survey was administered in May 2014, with a total of 1462 students interviewed. Classrooms were randomly selected using a two-stage selection procedure with cluster samples. A multiple correspondence analysis (MCA) associated with a hierarchical cluster analysis and with an ordinal logistic regression was performed. Chikungunya was less understood and perceived as a more dreadful disease than dengue fever. The analysis identified three groups of individual protection levels against mosquito-borne diseases: “low” (30%), “moderate” (42%) and “high” (28%)”. Protective health behaviors were found to be performed more frequently among students who were female, had a parent with a higher educational status, lived in an individual house, and had a better understanding of the disease.
This study allowed us to estimate the level of protective practices against vector-borne diseases among students after the emergence of a new arbovirus. These results revealed that the adoption of protective behaviors is a multi-factorial process that depends on both sociocultural and cognitive factors. These findings may help public health authorities to strengthen communication and outreach strategies, thereby increasing the adoption of protective health behaviors, particularly in high-risk populations.
| Although dengue fever has been a focus of many awareness campaigns in Latin America, very little information is available about beliefs, attitudes and behaviors regarding vector-borne diseases among the population of French Guiana. At the initial onset of the first chikungunya outbreak, a quantitative survey was conducted among 1462 high school students aiming to study experiences, practices and perceptions related to mosquito-borne diseases and to identify factors that could be associated with protective behaviors. Chikungunya was less understood and perceived as a more dreadful disease than dengue fever. Students were clustered in three different groups according to their level of protection: “low” (30%), “moderate” (42%) and “high” (28%). Protective health behaviors were found to be performed more frequently among students who were female, lived with a parent who had a higher educational status, lived in an individual house, and had a better understanding of the disease. The results revealed that the adoption of protective behaviors is a multi-factorial process that depends on both socio-economic and cognitive factors. These findings may help the public health authorities to strengthen their communication and outreach strategy, thereby increasing the adoption of protective health behaviors, particularly in endemic countries and high-risk populations.
| Chikungunya fever is a re-emerging arboviral disease caused by chikungunya virus (CHIKV), an alphavirus transmitted to humans primarily via the bite of an infected mosquito (Aedes spp. mosquito) [1]. Clinical onset is abrupt with high fever, headache, back pain, rash, myalgia and arthralgia, and symptoms generally resolve within 7–10 days [2]. The illness is usually self-limiting and resolves with time. Nevertheless, acute or chronic complications (e.g., polyarthralgia) can occur [3,4]. There is no specific treatment for chikungunya, and no vaccine is currently available [2]. Since its first identification in the early 50s in Tanzania, the spread of CHIKV has been the cause of many large outbreaks in Africa, Asia, and the Pacific Islands, especially during the last decade [1,5]. Before December 2013, CHIKV transmission had never been documented in the Americas despite annually reported imported cases and the presence of the main vectors Ae. albopictus and Ae. aegypti [6–9]. In December 2013, autochthonous cases were detected in the French overseas territory of Saint Martin and led to the rapid spread and transmission of CHIKV in the Caribbean countries and the Americas, including French Guiana, within 9 months [10,11].
In French Guiana, where Ae. aegypti mosquito was responsible for several dengue fever outbreaks [12–14], the first autochthonous cases of chikungunya were reported in February 2014 [15]. By June 2014, the epidemiological situation had evolved to moderate autochthonous transmission, epidemic clusters and a localized transmission chain. In September 2014, the situation worsened with an increasing number of clusters in the region [16]. Since the introduction of CHIKV in the territory, health authorities have reactivated the dengue fever control vector plan, based on an integrated vector management strategy promoted by the World Health Organization (WHO) and applicable to all vector-borne diseases. This strategy includes different approaches combining an environmental management program aimed at reducing breeding sites, using insecticides safely, biological control using organisms that reduce target species, providing education, increasing public awareness and promoting personal protection [17]. The active and persistent participation of the individuals and communities is a key factor in the achievement and sustainability of vector control programs, as the punctual interventions are generally ineffective at preventing outbreaks of vector borne diseases [18]. One important target group for such programs is the young generation, who can become more easily involved in community-based vector-source reduction campaigns [19,20]. In addition, participation at the individual level, such as use of insect repellent, mosquito netting or elimination of the indoor breeding sites, may also play an important role. Although education campaigns have increased public awareness of health risks related to dengue fever, which has strongly affected French Guiana, especially since 2006, it remains unclear to what extent this knowledge is associated with better preventive practices. Notwithstanding, no study had been previously conducted in French Guiana about the knowledge, attitude and practices related to vector-borne diseases. Additionally, studies conducted on health and illness perceptions indicated that these perceptions can significantly change over time according to health events [21–24]. Therefore, it was opportune to evaluate beliefs and practices of disease prevention with the increasing risk of arbovirus in the Americas [25,26], and follow their evolution according to epidemiologic settings.
The increasing risk of arbovirus transmissions in French Guiana prompted the need to document beliefs and behaviors related to vector-borne diseases and to determine the extent that socio-demographic, cognitive and environmental factors are associated with adequate protective behaviors.
The aims of the study were to describe and explore experiences, practices and perceptions of a new health threat related to vector-borne diseases among students of French Guiana; identify the main factors that are associated with the practice of protective behaviors; and quantify those associations.
French Guiana is a French overseas department located on the north-eastern coast of South America between Brazil and Surinam. At the time of the 2012 census, the population of the department was estimated at 239,500 individuals. This region is characterized by an extremely high birth rate and a high proportion of youth: 44% are under 20 years-old and only 4% are more than 65 years-old [27]. In 2014, the territory had 15 high schools localized on the coastline (Fig 1) where 12,400 students were registered (approximately 5.2% of the population). These students represented the target population of the study.
A cross-sectional survey about “beliefs, attitudes and practices” among students of French Guiana was conducted from May 12–21, 2014, based on a random cluster, stratified sample of French Guiana students, with the aim of investigating 1,600 students using a 2-stage selection procedure to have a 95% confidence level associated to a 2.5% precision. The population was first divided into two strata: professional high schools and general high schools. Between those two strata, 77 classrooms (primary sampling unit) were selected from a simple random sampling. All of the individuals were investigated in every selected classroom using standardized self-questionnaires administered by local high school nurses.
The data were collected according to the rules established by the National Data Protection Authority (CNIL Declaration N°aaH1243499A), which is responsible for the ethical issues and protection of individual data collected in France. All of the information was collected anonymously, and an information letter was sent to the parent through the student’s liaison notebook informing them about the right to oppose the survey and to access the data. A signature was requested and checked by the school nurse before the survey begins.
Aside from socio-demographic variables, collected data were grouped into five general categories: 1) environmental variables and exposure, 2) beliefs about disease, 3) perception of the risk associated with infectious diseases, 4) perception of the effectiveness of protective measures and 5) self-reported protective behaviors.
The questionnaire contained a wide range of items such as the type of housing, the presence or absence of potential breeding sites and potential factors associated with breeding sites, such as farming. In addition, respondents were asked how frequently they were bitten by mosquitoes (response options: ‘Never’, ‘Seldom’, ‘Sometimes’ and ‘Often’). Participants were also asked if they had ever seen or heard about “Aedes mosquitoes”, how frequently they practiced outdoor activities, and during what time of the day mosquito bites occurred. Participants were asked to report the occurrence of an acute febrile illness consistent with presumptive dengue virus and/or CHIKV infections.
A large range of beliefs were investigated; in particular, perceptions of the health threat, i.e., qualitative and quantitative judgments that individuals expressed when asked to evaluate a specific illness and the risk of contracting it [28]. To characterize these perceptions within the population, questions were drawn from the existing literature using the Brief Illness Perception Questionnaire (B-IPQ). This questionnaire measures five components: the identity- the symptom the patient associates with the illness; the cause- the personal ideas about etiology; the time-line- the perceived duration of the illness; the consequences- the expected effects and outcome; and the treatment control- the method of recovery from the illness [29]. Others questions were adapted from the methodological literature devoted to transmissible infectious diseases to complement the survey and assess perceived exposure, severity and susceptibility [30]. With the exception of the cause, the identity, the time-line and the mean of awareness, respondents answered each item by scoring them on a numeric scale of 0 to 10, with the meaning of each end-point indicated on the questionnaire.
Participants were asked to answer those questions in relation to dengue, chikungunya and three other infectious diseases occurring in French Guiana, malaria, human immunodeficiency virus (HIV) and yellow fever, by similarly giving a score of 0 to 10 for each question.
For protective health behaviors, participants were asked how often they applied personal protective measures, such as wearing long-sleeved clothes or using repellent, and vector control measures, such as covering water receptacles. Then, participants were asked whether those behavioral recommendations were appropriate to prevent mosquito bites (response options: ‘Ineffective’, ‘A bit effective’, ‘Quite effective’, ‘Very effective’ and ‘Not sure’). Finally, respondents were asked to assess the constraint of each protective measure (response options: ‘Very constraining’, ‘Quite constraining’, ‘Bit constraining’ and ‘Not constraining’).
Data were recorded using Access, and statistical analyses were performed using STATA12 software (Stata Corp., College Station, TX, USA) [31] and SPAD8 [32]. Primary sampling units and strata were taken into account to calculate estimations according to the design effect, and all estimations were obtained using STATA “svy” commands.
A multiple correspondence analysis (MCA) was used to display the relationships among the individual and structural factors, examining the association of protective behaviors with socio-economic and environmental factors. This method allows the simultaneous analysis of a large number of variables and their respective categories. The MCA method plots all the information represented by variables and individuals on a graph based on multiple factorial axes, and searches for patterns in the dataset, helping to identify the variables more closely associated with different groups. A matrix of eigenvalues was determined to identify a combination of variables that presented more stability in the factorial plan and explained the largest percentage of variability in the dataset. The number of dimensions was chosen by analyzing the decrease in eigenvalues. Variables were grouped as active or supplementary. Active variables were the respective frequency of each protection mean cited in the questionnaire with 4 modalities (never, seldom, sometimes and often) and the overall frequency of protection. Supplementary variables were the type of house, the high school sector, the parent’s level of education, the presence of a pool, an air-conditioning system, a yard and the name of the high school.
Following the MCA, a hierarchical cluster analysis was performed to determine the natural groupings of observations regarding protective behaviors. This cluster analysis encompassed a variety of mathematical methods for classifying groups of similar entities regarding the adoption of protective behaviors. We used a hierarchical agglomerative clustering algorithm that initially placed each individual in a separate cluster and then iteratively joined the two most similar clusters.
Finally, a logistic ordinal regression was performed to identify the differences among the socio-economic, environmental, exposure and cognitive variables associated with the level of protective behaviors. The clusters identified above were used as the dependent variable for the analysis. Factors that were determined significant in the univariate analysis were tested in a stepwise multivariate model as independent variables. The level of statistical significance was set to (P = 0.05).
In total, 1462 students from 13 high schools were included in this study, representing 12% of the whole student’s population. Two high schools were not surveyed in the study because of refusal from their director. All the students present at the time of the survey participated to the study. Respondents were aged 16.7 years old on average (range: 15–24).
The proportions of respondents who declared having or having had chikungunya and dengue were 0.8% (95% Confidence Interval (CI): 0.4–1.6) and 44.6% (95% CI: 41.5–47.6) respectively.
Mean threat perception scores are reported in Fig 2. We observed that chikungunya was a disease that worried more students than dengue fever. In fact, chikungunya displayed a higher mean score than dengue, malaria and yellow fever in relation to feeling worried, perceived severity and perceived consequences, although those scores remained lower than those associated with HIV. Students reported understanding less about chikungunya than dengue or yellow fever. Dengue and chikungunya displayed the same level of exposure, whereas students felt more exposed to chikungunya than malaria and yellow fever. In terms of control, students perceived chikungunya to be as avoidable as dengue and yellow fever, but more avoidable than malaria.
Students expressed the same perceived exposure for dengue (5.37, 95% CI: 5.12–5.63) and chikungunya (5.38, 95% CI: 5.14–5.63). However, regarding the time of illness, the perceived duration of chikungunya was longer than the duration of dengue, with 51% (95% CI: 47.3–54.1) of students thinking that chikungunya could last to three weeks to several months and only 32% (95% CI: 29.2–35.2) of students believing dengue fever had the same duration. Finally, the perceived level of information was higher regarding dengue (5.93, 95% CI: 5.69–6.14) than chikungunya (4.91, 95% CI: 4.50–5.31).
Malaria, which is primarily endemic to isolated and land areas, was reported with a higher mean score than dengue or yellow fever regarding feelings of fear and was less understood than dengue and chikungunya. HIV had the highest mean score for disease control, suggesting that students identified vector-borne disease protection more difficult than sexually transmitted disease protection. With the exception of perceived treatment efficacy, all HIV mean scores were higher than the midpoint value on the response scale adapted from the B-IPQ. Finally, in term of perceived treatment efficacy, yellow fever had the highest score among the reported values.
Participants identified different diseases that they believed to be transmitted by mosquitoes. More than half of the participants properly identified dengue (81%) and chikungunya (64%), more than a third properly identified yellow fever (40%) and malaria (36%), and 10% incorrectly identified HIV as transmitted by mosquitoes. Among students who correctly identified that mosquitoes could transmit dengue and chikungunya viruses, more than 80% could mention at least one symptom of these diseases. The most commonly mentioned symptom was tiredness (88%), followed by headache (82%). Overall, more than two-thirds of the participants reported that tiredness, headache, myalgia, arthralgia and skin rashes could be attributed to a mosquito-borne disease, in accordance with biomedical evidence on the clinical manifestation of these diseases.
Regarding the availability of a vaccine, 91% of the students reported that there was a vaccine against yellow fever, followed by 39% for a dengue vaccine, 34% for a malaria vaccine, and 16% for a chikungunya vaccine; 7% of the students thought that HIV was a vaccine-preventable disease.
Finally, when students were asked which method of awareness outreach they believed was the most appropriate for mosquito-borne diseases, school intervention, television and social networks were the most frequently cited (82, 76 and 74%, respectively).
Although 77% of the student population was bitten by mosquitoes sometimes or often, only 54% of them reported taking preventive measures to reduce the number of mosquito bites. As shown in Fig 3, when asked about effectiveness of several preventive measures, the most commonly reported measures were bed nets (60.6%), followed by sprays (60.5%), window nets (58.6%) and removal of stagnant water from containers (58.5%). A perceived lack of effectiveness was expressed with regard to wearing protective clothing (a long-sleeved shirt and long trousers) (35%) and closing windows (33%).
Self-reported behaviors to protect against mosquito bites are shown in Fig 4. The most frequently reported preventive measures were using indoor insecticide sprays (34.7%), closing windows (32.9%), removing stagnant water containers (32.3%) and using an air conditioner (31.5%). Surprisingly, when asked to note if one of the preventive measures was restrictive, students primarily noted insecticide sprays (54.4%). The most restricted option was the use of an indoor insecticide electric diffuser (67%) and the less constraining measure was the use of an air conditioner (26.1%). The most frequently expressed reasons for lack of protection were that students were not reminded to adopt protective behaviors (57%) and did not feel annoyed by mosquito bites (17%).
Graphic presentation of the variables, constructed in a series of 2-dimensional spaces is shown in Fig 5. This two-dimensional space enabled the mapping of protective behaviors as active variables and socio-economic as supplementary variables. Only the two first principal factors derived from the analysis were kept to plot the coordinates of the 62 variables. Factorial axis 1 captured 19.18% of the variability and distinguished individuals who adopted preventive measures from those who did not. The second axis captured 16.71% of the variability and showed a gradient in the level of protection. As shown in Fig 5, professional sector, collective housing, stay-at-home parent, parent with a low educational attainment, absence of yard or an air conditioner system were variables that were negatively associated with axis 1, as the response “never” was related to the frequency of several means of protection. Other variables that corresponded to a higher socio-economic status were positively associated with axis 1 as a positive attitude regarding protective behaviors. Finally, two high schools were negatively associated with axis 1.
The hierarchical cluster analysis identified three clusters related to the level of protection. Based on the usage patterns and the socio-economic and risk perception variables, this method offered a typology of the students as shown in Fig 6. The first cluster contained 450 individuals that never used protective means against mosquito bites. This category was characterized by the absence of a yard, a pool or an air conditioner system compared with the others clusters. Moreover, students registered in professional high schools and students with parents who had low educational level belonged to this cluster. The largest cluster accounted for 611 individuals (cluster 2) and concerned students that predominantly reported “seldom” or “sometimes” when asked about the frequency of usage protective measures. The last cluster, cluster 3, grouped together 401 individuals that reported “often” usage of protect measures. These last two clusters contained students with similar socio-economic patterns, such as the presence of a yard, a pool or an air conditioner system, parents with high educational level and registration in a general high school. Nonetheless, clusters could be distinguished with differences in risk perception variables. In cluster 3, students were very worried about chikungunya and dengue diseases, considering them very severe, but understandable diseases. Finally, in cluster 1, students primarily perceived a low efficacy related to several protective measures, whereas in cluster 3, students perceived these measures as very effective.
To identify socio-economic, environmental and individual factors associated with the level of self-reported protective behaviors, we performed univariate and multivariate ordinal regressions. Clusters resulting from the hierarchical cluster analysis were used as the dependent variables. The odds ratios (OR) resulting from the analysis and their significance are shown in Table 1. The results from the multivariate regression indicated different types of variables associated with the level of frequency of vector-borne disease protective behaviors. Among students, those who were female, lived with a parent that received an education higher than primary school, lived in an individual house and easily understood the diseases were more likely to report protective behaviors than other students.
To our knowledge, this is the first published study on protective behaviors related to mosquito-borne diseases conducted in French Guiana. The survey was conducted during the emergence of chikungunya virus in French Guiana and nearly one year after the last dengue outbreak. Even if the student population was not representative of the entire population, the high school environment provided us an opportunity to conduct a large survey in the aftermath of the first reported confirmed cases of chikungunya in the region within an accessible and diversified population in terms of socioeconomic status. Additionally, the study represented an opportunity to investigate a perceived prevalence of the disease a few weeks after detection of the first autochthonous transmission clusters in the territory.
The method used in this study measured the perceived risk of chikungunya compared to various other health risks facing the public in French Guiana. The study showed that, with the notable exception of HIV, chikungunya was associated with the highest scores in term of perceived severity, worry and consequences, which is not surprising in a context of an emergence of a new virus. Our findings highlight that students in French Guiana were concerned and aware of the characteristics about vector-borne diseases. Approximately 81% and 64% of students knew that mosquitoes transmit dengue and chikungunya virus, respectively, and more than 80% were able to correctly mention at least one of the symptoms associated. Participants appeared to have good information about common symptoms and transmission. This awareness is likely attributed to a former infection of one of the mosquito-borne diseases or a result of a mass media campaign about dengue prevention. In fact, media coverage and the extent of reporting events may have exerted an influence on public perception of these diseases and therefore to produce positive changes in health-related behaviors [33,34].
It should be noted that among those who were bitten by mosquitoes, only half adopted protective behaviors. This result is somewhat unexpected since the perceived exposure to vectors (seeing bugs, being bitten) has been repeatedly found as one of the most important triggers for taking health protective actions [35,36]. This cannot be explained by an important perceived lack of behavioral control because this variable was not found to be associated with the adoption of preventive measures against mosquitoes. Nevertheless, we cannot exclude a sentiment of fatigue regarding the emerging infectious diseases and the continuous individual and collective efforts that are required to control them, given the growing number of public health warnings experienced by the local population over the past years. Moreover, many students expressed in additional comments that the means of protection were too much expensive and that they could not afford it. Several studies have explored cognitive factors and underlined the importance of public health beliefs in the adoption of effective preventive behaviors. Findings have notably shown that a lack of perceived behavioral control and a low perceived susceptibility to the threat inhibited sustained protective actions against vector-borne diseases; however, an increased understanding of the disease transmission led to better dengue prevention practices [37–41], and this factor has been importantly highlighted in our analysis.
Our findings hold important implications for the prevention of a new threat showing the importance for public health authorities to accommodate their strategies to rapidly strengthen the disease understanding in populations at risk.
The results showed that the adoption of protective behaviors is a multi-factorial process that depends both on socio-economic, environmental and cognitive factors. These factors should be further reviewed and considered in the development and implementation of future large-scale mosquito-borne disease prevention programs. In previously conducted infectious disease studies, these social and cognitive factors have consistently been found to influence the engagement in health protective behaviors [35,42,43]. However, in our study, only a few cognitive factors were associated with the protective health behaviors recommended by the public health authorities at the early stage of the epidemic. Unlike previous studies, this survey was conducted among students, thus, 80% of individuals were between 16 and 18 years old. This result can be explained by an age group effect in which the variability of cognitive variables was too low to have an impact on behaviors.
In contrast, a number of socioeconomic factors were found to shape the behavioral response to this emerging health threat. The multivariate model showed that girls with a parent who received education higher than primary school were more likely to report health protective behaviors than were other students. There was also an influence of the environmental variables on protective behaviors; people living in individual houses were more likely to have gardens and, therefore, were more likely to be exposed to mosquito bites during outdoor activities and likely required to take extra precautions to keep mosquitoes away. Finally, the results revealed that a moderate or high level of disease comprehension supported the adoption of protective health behaviors. This finding is rather surprising because many previous studies only revealed a weak association between health literacy and practices aiming to reduce the risk associated with vector borne diseases [36,44–46]
Last but not least, this study enabled us to identify two high schools associated with a high level of non-protective practices and provided to local education authorities the opportunity to target and modify health messages delivered in these institutions. Indeed, this finding highlighted also that school intervention was the best mode of awareness mentioned by the participants.
The survey helped to characterize the public apprehension of several vector-borne diseases, as well as the nature and frequency of health protective behaviors among students to control and prevent their dissemination in the population. Given the importance of the public understanding of illnesses in the adoption of effective protective behaviors, this study shed light on the value of education campaigns aiming to improve the lay comprehension of the diseases. They may be a useful pre-requisite for the programs encouraging community participation in vector control.
|
10.1371/journal.pntd.0005483 | Dengue in Bali: Clinical characteristics and genetic diversity of circulating dengue viruses | A high number of dengue cases are reported annually in Bali. Despite the endemicity, limited data on dengue is available for Bali localities. Molecular surveillance study was conducted to explore the clinical and virological characteristics of dengue patients in urban Denpasar and rural Gianyar areas in Bali during the peak season in 2015. A total of 205 adult dengue-suspected patients were recruited in a prospective cross-sectional study. Demographic and clinical information were obtained, and dengue screening was performed using NS1 and IgM/IgG ELISAs. Viral RNA was subsequently extracted from patients’ sera for serotyping using conventional RT-PCR and Simplexa Dengue real-time RT-PCR, followed by genotyping with sequencing method. We confirmed 161 patients as having dengue by NS1 and RT-PCR. Among 154 samples successfully serotyped, the DENV-3 was predominant, followed by DENV-1, DENV-2, and DENV-4. Serotype predominance was different between Denpasar and Gianyar. Genotyping results classify DENV-1 isolates into Genotype I and DENV-2 as Cosmopolitan Genotype. The classification grouped isolates into Genotype I and II for DENV-3 and DENV-4, respectively. Clinical parameters showed no relationship between infecting serotypes and severity. We observed the genetic diversity of circulating DENV isolates and their relatedness with historical data and importation to other countries. Our data highlights the role of this tourist destination as a potential source of dengue transmission in the region.
| Dengue is the most significant mosquito-borne viral disease affecting humans. Up to one third of the world population is at risk of dengue virus (DENV) infection, transmitted through the bite of Aedes mosquitoes. Bali, a well-known international tourist destination, is regularly ravaged by dengue disease. This disease impacts the health of both local people and visitors thus imposing a heavy economic burden. Bali has a constant flow of travelers and labors that contribute to the spread of DENV infection. Detailed characterization of DENV from Bali is limited; most reports are from travel-acquired cases. Here, we study dengue clinical and virological aspects in local Balinese people. We presented the clinical spectrum of the disease and the virological characteristics, observing the circulation of genetically diverse endemic virus strains including strains which are closely related to imported viruses in neighboring countries. The circulation of a lineage of DENV-2 proposed to cause outbreak in the past is also identified. Our study provides data on the genetic of circulating DENV in Bali which are useful for further applications, such as to monitor the virus transmission and outbreak investigation in the region.
| Dengue is the most important arthropod-borne disease affecting humans with high incidence in tropical and subtropical countries. It is estimated that 390 million infections occur annually and over 70% of the world population is at risk of being infected by dengue viruses (DENVs) [1]. Dengue can manifest complex clinical features; infection with any of the four antigenically distinct DENVs may lead to a range of clinical manifestations, which vary in severity from classic dengue fever (DF) to a more severe and fatal dengue hemorrhagic fever (DHF) and dengue shock syndrome (DSS) [2]. DENV, a member of the Flaviviridae family, consists of a 10.7 kb single-stranded positive-sense RNA genome encoding three structural (C, prM/M, E) and seven non-structural (NS1,NS2A, NS2B, NS3, NS4A, NS4B, NS5) proteins [3]. The substantial genetic diversity of DENV is shown by the presence of various genotypes within the four DENV serotypes (DENV-1, -2, -3, and -4) [4,5].
Bali is a well-known international tourist destination located in the tropical country of Indonesia and is regularly affected by dengue disease. This disease affects the health of both local people and visitors imposing a heavy economic burden [1,6]. It has been reported that Bali has a constant flow of labor and travelers that contribute to the spread of DENV infection [7]. Major outbreaks occurred in 2010 and 2015 with 12,574 (including 35 fatalities) and 10,704 (28 fatalities) reported dengue cases, respectively (Dinas Kesehatan Provinsi Bali/Bali Provincial Health Office).
Previous reports documented the hyperendemic transmission of all four DENV serotypes in Bali during 2010, and the circulating DENV included the dominant local strains which had circulated for several years, as well as strains more recently introduced into Bali [8]. This transmission created substantial DENV diversity and serve as a hub for dengue transmission and mixing in Bali [8]. Most of the reports of dengue in travelers returning from Indonesia have implicated Bali as the source of importation [8–10]. All four DENV serotypes were detected from travelers entering Western Australia between 2010–2012, mostly from Bali [8]. It has also been reported that highest proportion (24.6%) of imported dengue cases of travelers in Queensland, Australia originated from Indonesia [11]. Despite the year-round transmission and large cyclical outbreaks observed in travelers from Bali, no substantial data of clinical and genetic features of dengue in local Balinese people are currently available.
To obtain comprehensive data on dengue disease in local Balinese, we have conducted molecular surveillance to characterize clinical aspects and genetic diversity of the DENVs circulating in Bali during the high dengue season in 2015. We report here the demographic, clinical, hematological, and virological characteristics of dengue in adult hospitalized patients in two geographically different regions, namely the urban Denpasar municipality and rural Gianyar regency. The data will be beneficial for the control of this disease in Bali and also highlights the potential of this island as the source of imported dengue cases to other parts of the world.
The study protocol was approved by the Medical Research Ethics Committee of Faculty of Medicine Udayana University/Sanglah Central General Hospital, Bali Indonesia with approval No. 122/UN.14.2/Litbang/2015.
The cross-sectional prospective study was conducted during the period of March to May 2015. Samples were collected from Wangaya (WGY) and Sanjiwani (SJN) General Hospitals which are located in Denpasar municipality and Gianyar regency, respectively. Denpasar and Gianyar were recorded as the regions with the highest dengue incidents in Bali in 2014 [12]. Inpatients (above 14 years) presenting at the adult wards with fever >38°C accompanied by at least one sign of dengue such as malaise, arthralgia, rash, retro-orbital pain, DHF or DSS were enrolled in the study after providing written informed consents. We excluded patients with history of chronic illnesses, such as chronic liver disease, diabetes mellitus, chronic kidney disease, chronic lung disease, human immunodeficiency syndrome, and cardiac disease. Sera were collected during the acute phase (within the first five days of illness) and before discharge from the hospital. Each patient’s demographic, clinical and hematological data as well as disease severity according to the WHO-SEARO 2011 guidelines were recorded [13].
The preliminary screening for DENV in patients’ sera was performed using the SD Bioline NS1 rapid test (Alere, Australia), according to the manufacturer’s instructions. Serological tests were performed using Panbio Dengue Duo IgM/IgG Capture ELISA (Alere). The serological test results were used to determine the primary/secondary infection status of patients, as described previously [14]. Confirmation using Panbio IgG indirect ELISA (Alere) was also performed, in which the presence of previous IgG antibody to DENV would indicate secondary infection.
DENV viral RNA was extracted from acute serum samples using QIAmp Viral RNA Mini kit (Qiagen, Hilden, Germany) according to the manufacturer’s instructions. The DENV detection and serotyping was carried out according to the two step protocol as previously described by Lanciotti et al.[15] with some modification [16]. The more sensitive Simplexa Dengue (Focus Diagnostics, Cypress, USA) qRT-PCR was used as a confirmatory assay to simultaneously detect and serotype DENV as previously described [17].
Samples were genotyped based on Envelope (E) gene. DENV RNA was reverse-transcribed into cDNA using Superscript III reverse transcriptase (Invitrogen-Life Technologies) and then PCR-amplified using Pfu Turbo Polymerase (Stratagene-Agilent Technologies)[18]. PCR products were purified from 0.8% agarose gel using the QIAquick gel extraction kit (Qiagen) and cycle sequencing reactions were performed using six overlapping primers for each serotype from both strands with Big Dye Dideoxy Terminator sequencing kits v3.1 (Applied Biosystems-Life Technologies), as described previously [18]. Purified DNA was subjected to capillary sequencing performed on 3130xl Genetic Analyzer (Applied Biosystems). Sequence reads were assembled using SeqScape v.2.5 software (Applied Biosystems) with manual inspections employed whenever ambiguities were present. Sequence contigs were generated and used in subsequent analyses.
Genotyping of DENV was based on classifications by Goncalvez et al. [19], Twiddy et al.[20], Lanciotti et al. [21], and Lanciotti et al. [22] for DENV-1, -2, -3 and -4, respectively. Envelope gene sequences of Bali isolates together with other representative sequences downloaded from GenBank were aligned using MUSCLE in MEGA 6.0 software (http://www.megasoftware.net/). The initial dataset was prepared using BEAUti graphical interface and the tip of each isolate calibrated using the year of isolation as calibration point. Phylogenetic tree was inferred based on selection of statistical model for likelihood calculation optimized for Maximum Likelihood (ML) tree using jModelTest v.2.1.4 [23]. Phylogenetic reconstruction and evolutionary rate analysis were performed using Bayesian Markov chain Monte Carlo (MCMC) method as implemented in BEAST v.1.8.2 [24]. Runs were performed using General Time Reversible (GTR) model with four gamma parameters (GTR + Γ4) and relaxed uncorrelated lognormal molecular clock using the initial estimated evolutionary rate of 7.6 x 10−4 substitutions per site per year, as previously described [25]. The tree prior was set as coalescent bayesian skyline prior, to facilitate the fewest demographic assumptions. One hundred million chains were run and sampled for every 1000th iteration, with 10% burn-in employed. The convergence of parameters was analyzed using Tracer v.1.5.0 to ensure adequate Effective Sampling Size (ESS) for all parameters. Maximum clade credibility (MCC) tree was created using TreeAnnotator v.1.8.2 and visualized in FigTree v.1.4.0.
Pearson’s Chi-square or Fisher’s Exact tests were used to compare univariate categorical data. Parametric One-way ANOVA or non-parametric Kruskal-Wallis tests were used to compare groups of laboratory test results within DENV serotypes. The regression analyses were performed using modified Poisson regression or the ordinal logistic regression with adjustment of clinically relevant potential covariates, i.e. age, gender, recruitment site, infection status, and fever at day of presentation. A probability value of p < 0.05 was considered statistically significant. All statistical analyses were performed using Stata version 12 (StataCorp, TX).
The complete E gene sequences of 28 DENV isolates were deposited in GenBank repository and granted accession numbers KY006129 to KY006156 (Supplementary S1 Table).
A total of 205 dengue-suspected cases were recruited from two hospitals in Bali Province. This comprises of 99 (48.3%) and 106 (51.7%) samples from Wangaya General Hospital, Denpasar, and Sanjiwani General Hospital, Gianyar respectively. The majority of the patients were recruited on day 4 or 5 of illness. The median age of the patients was 29 years, with a total range of 14–80 years. Equal proportions of male and female were observed with no significant differences in terms of gender and age (Table 1). Two patients did not meet the inclusion criteria and were excluded from analysis.
The NS1 antigen detection confirmed 154 (75.9%) cases of dengue infection. Further confirmatory tests using RT-PCR were performed on NS1-negative samples and resulted in seven additional dengue-confirmed samples, for a total number of 161 (79.3%). Serology testing revealed that the majority of the cases were of secondary infection (83.8%) while the remaining 16.2% were of primary infection (Table 1).
We successfully serotyped 154 out of 161 (95.6%) dengue-confirmed samples. All four serotypes were found to circulate in Bali during the study period. DENV-3 was predominant (48%), followed by DENV-1 (28%), DENV-2 (17%), and DENV-4 (4%). Five samples (3%) were detected as mixed infection of two different serotypes (DENV-1 and -2 and DENV-1 and -3). The predominance of DENV-3 was evident in both study sites. However, there were differences in the proportions of the other serotypes (Table 1). While six DENV-4 were found in Denpasar, the presence of DENV-4 in Gianyar was only detected as one case of mixed infection with DENV-3. The distribution of dengue serotypes between two regions was significantly different (p = 0.008) based on univariate statistical analysis. Other variables i.e. sex, age, and infection status were not significantly different between serotypes (Table 1).
Analysis of dengue clinical manifestations and hematological parameters were performed excluding the DENV-4 and mix infection cases due to the small sample size. Following the adjustment based on age, gender, infection status, site of study, and day of fever by the time of presentation, we observed no correlation between clinical manifestations and the infecting serotypes. The statistical analysis indicated that DENV-2-infected patients were likely to have 0.23 times the risk of loss of appetite compared to infection with other serotypes (95% CI = 0.07–0.75) (Table 2). The only hematological parameter difference between dengue serotypes was higher diastolic blood pressure in patients infected with DENV-1 (p = 0.042) (Table 3). In terms of disease severity, 75 patients (52.4%) were DF with the remaining 68 (47.6%) patients classified as DHF, where four patients were identified as DHF grade III (DSS) with hematemesis. Similarly, we did not find any correlation between disease severity and the infecting serotype (Table 4).
To determine the genotypes of DENV within each serotype in Bali in 2015, we performed genotyping based on E gene sequences. We successfully obtained complete sequences of E gene from 28 patients’ sera. Of 43 DENV-1 isolates, 10 (23.3%) were successfully PCR-amplified for their E genes. Phylogenetic analysis revealed that all 10 isolates were grouped into Genotype I based on Goncalvez [19] classification (Fig 1). Although grouped in a single genotype, the 10 Bali DENV-1 isolates were further differentiated into six lineages. It was also notable that the Genotype IV isolates that were present in Bali in 2010 (Fig 1) were not found in this study.
For DENV-2, five isolates were successfully genotyped out of 26 isolates. Phylogenetic analysis revealed that all of these five isolates belonged to Cosmopolitan genotype according to Twiddy [20] classification (Fig 2). These five viruses of three distinct lineages formed a monophyletic clade with previous Bali isolates circulating in 2010 [8].
We also genotyped 10 isolates of DENV-3 which were grouped as Genotype I based on Lanciotti [21] classification and were further differentiated into two major lineages (Fig 3). Both lineages appeared to have different ancestral origin from other DENV-3 isolated in Bali in 2010.
For DENV-4, from six isolates serotyped, three isolates were successfully sequenced and genotyped. Following Lanciotti classification [22], all three viruses were grouped as Genotype II (Fig 4). Two separate lineages were observed, and the Bali 2015 isolates formed a monophyletic clade with Bali 2010 isolates as well as other Indonesia isolates from Sukabumi, Makassar, and Jakarta. A grouping with DENV isolates imported to Taiwan was also observed.
Bali Province is an island of approximately 5,780 km2 in area and located in the tropical climate zone (latitude -8.4095178 and longitude 115.188916). Having one municipality and eight regencies, the province is inhabited by 3,995,281 residents [12]. The data provided by the Provincial Health Office showed fluctuating numbers of dengue cases during the period of 2009 to 2015. We conducted the first virological investigation of dengue in Bali to determine the genomic diversity and its relation to the clinical manifestations.
In this study, we confirmed 79.3% (161/203) of patients as dengue positive, suggesting the considerable burden of dengue in the community. The majority (83.8%) of the confirmed cases were of secondary dengue infection based on serology results. This result was as expected since the recruited patients were adults with more prolonged exposure to dengue infection in the past. This number confirmed the endemicity of dengue in Bali.
All four DENV serotypes were found circulating in Bali in 2015. The most prevalent serotype was DENV-3, followed by DENV-1, DENV-2, and DENV-4. With lack of molecular data for the dengue virus in Bali, no comparison could be made to the current predominant serotypes identified in our study. The predominance of DENV-3 has been reported in Indonesia [26–28]. In our molecular surveillance, we observed the rise of DENV-1 infection as the second most common serotype. Serotype replacement has been described in a number of reports [14,18,29,30]. However, further surveillance is needed to monitor the dynamics of DENV in Bali in order to confirm this phenomenon. Although the two study sites (the Denpasar municipality and Gianyar Regency) are only about 28 km apart, statistical analysis showed that the distribution of the serotypes between the two regions was significantly different (p = 0.008). This discrepancy may have resulted from the different demographic profile of the sites. Sanjiwani Hospital is a referral hospital for the Gianyar regency in eastern part of Bali which mostly consisted of rural areas with less dense population (1345 people per km2), while the patients admitted to Wangaya Hospital, Denpasar mostly resided in densely populated areas around Denpasar (population density 6891 people per km2) [12] which is also a major domestic and international tourist destination. The different population density may account for the different transmission profile of particular serotypes, as has been observed in Viet Nam [31].
On the clinical aspect of dengue in Bali, a higher number of patients with DF was observed rather than DHF (Table 1), even though the majority of the patients were adults with secondary infection. It was reported that secondary infection is a risk factor for increased severity [32]. However, we did not observe a correlation between the higher number of secondary infection and increased severity as in other studies [33].
The effect of serotypes on clinical manifestations of dengue fever in adults has been reported [34,35]. A previous study in adults in Singapore reported that joint pain and red eyes were associated with DENV-2 and DENV-1, respectively [36]. Within all the clinical variables observed, we found the loss of appetite as the only parameter with significant correlation with DENV-2 (Table 2). Similarly, the only vital signs and hematological parameters that significantly correlated with different serotypes was higher diastolic blood pressure observed in DENV-1-infected patients (Table 3) which was not reported in the published literature. Interestingly, we did not find a correlation between thrombocytopenia and infecting serotype, as observed in other studies conducted in adults [36,37]. In addition, there was no correlation between other clinical parameters and the disease severity (Supplementary S2 Table).
In terms of virological aspects, the DENV E gene sequences for 28 representative isolates were generated. This provides DENV genetic data from Bali that will be useful for various applications such as molecular epidemiology studies and outbreak investigations. Phylogenetic analysis revealed that all of the 10 DENV -1 isolates were grouped into Genotype I (Fig 1). The isolates were closely related to Bali strains from imported cases to Australia in 2010 [9] and 2011 [8] and Japan [10] and strains from other cities in Indonesia i.e. Makassar [18] and Surabaya [29]. The genetic diversity of DENV-1 in Bali is extensive as shown by the presence of multiple lineages within Genotype I group (Fig 1). In this study, we did not find the other DENV-1 genotype (Genotype IV) known to circulate in Bali in 2010 [38] and other cities in Indonesia as well as in imported cases to other countries. The absence of Genotype IV in Bali together with genotype replacement is similar to that seen with the Jambi dengue outbreak [30], and other cities of Indonesia [18,29]. Altogether, our data suggest the ongoing replacement of DENV-1 Genotype IV by Genotype I in Bali.
The Cosmopolitan DENV-2 isolates were further clustered into different lineages and closely related to the imported dengue cases to Australia during 2009–2011 [8]. Several lineages have been reported within the Cosmopolitan genotype of DENV-2 from imported cases to Australia, mostly from Bali [8]. Following the lineage numbering, Bali 2015 isolates were clustered into lineages 3, 4, and 5 (Fig 2). Of the five Bali DENV-2 isolates sequenced, three isolates belong to this lineage. The lineage 4 has been described to have emerged during a major outbreak in Bali in 2011–2012 [8]. Isolates from Denpasar and Gianyar were grouped into this lineage, reflecting the spread of this lineage in Bali. Our study confirms the active circulation of this particular lineage in Bali and the potential active transmission and exportation to other regions.
Contrary to the data reported by Ernst et al [8] which showed the predominance of DENV-2 in Australian travelers visiting Bali in 2010, DENV-3 was the predominant serotype in Bali in 2015, indicating the shifting of serotypes in Bali. DENV transmission is dynamic and serotypes have been known to show cyclical predominant pattern which may correlate with herd immunity [39]. Genetically, ten isolates of Bali DENV-3 were grouped into Genotype I (Fig 3) which is the common DENV-3 genotype found in Indonesia [14,18,30,40,41]. The Bali 2015 isolates were grouped together with isolates of imported cases to Australia and Taiwan and those from Surabaya and Jakarta [9,41–43] The Bali 2015 DENV-3 isolates apparently were not as divergent as DENV-1 and -2, in which only two major lineages were observed. This might suggest that the DENV-3 that caused outbreaks in Bali were the local/endemic strains that have been circulating in the region for decades and not those introduced from outside of Indonesia. DENV-4 was the least prevalent in Bali, in which only six isolates were found. Among these, three isolates were grouped into Genotype II (Fig 4) which is commonly found in Indonesia [14,18,40,41]. The isolates were closely related to imported dengue cases to Australia and isolates from Sukabumi, a city in West Java Province [9,40]. Again, we did not observe any introduced DENV-4 strains in Bali which suggests that the DENV-4 in Bali were the local and endemic strains.
In summary, our study provides the first detailed information on the clinical and virological features of dengue in two areas in Bali with the highest dengue cases in 2015. Our study has some limitations related to potential selection bias based on recruitment criteria as only adults were enrolled, a relatively small number of samples were collected; and the time period of collection was limited. Nevertheless, we confirmed the hyperendemicity of all four DENV serotypes where the circulating DENV included dominant local strains which were in circulation for several years and were related to recent imported dengue cases to other countries. Our study highlights Bali as a place with prominent genetic diversity of DENV and supports previous reports on its role in dengue transmission and mixing. Further studies on active molecular surveillance of DENV should be done in Bali to monitor dengue dynamics.
|
10.1371/journal.pntd.0000579 | Increasing Trends of Leptospirosis in Northern India: A Clinico-Epidemiological Study | Leptospirosis, a zoonosis associated with potentially fatal consequences, has long been a grossly underreported disease in India. There is no accurate estimate of the problem of leptospirosis in non-endemic areas such as north India.
In order to understand the clinical spectrum and risk factors associated with leptospirosis, we carried out a retrospective study in patients with acute febrile illness in north India over the last 5 years (January 2004 to December 2008). There was increased incidence of leptospirosis (11.7% in 2004 to 20.5% in 2008) as diagnosed by IgM ELISA and microscopic agglutination titer in paired acute and convalescent sera. The disease showed a peak during the rainy season (August and September). We followed up 86 cases of leptospirosis regarding their epidemiological pattern, clinical features, laboratory parameters, complications, therapy, and outcome. Mean age of patients was 32.6 years (2.5 years to 78 years) and males (57%) outnumbered females (43%). Infestation of dwellings with rats (53.7%), working in farm lands (44.2%), and contact with animals (62.1%) were commonly observed epidemiological risk factors. Outdoor workers including farmers (32.6%), labourers (11.6%), para-military personnel (2.3%), and sweepers (1.2%) were commonly affected. Modified Faine's criteria could diagnose 76 cases (88.3%). Renal failure (60.5%), respiratory failure (20.9%), the neuroleptospirosis (11.6%), and disseminated intravascular coagulation (DIC) (11.6%) were the commonest complications. Five patients died, giving a case fatality rate of 5.9%.
There has been a rapid rise in the incidence of leptospirosis in north India. Severe complications such as renal failure, respiratory failure, neuroleptospirosis, and DIC are being seen with increasing frequency. Increased awareness among physicians, and early diagnosis and treatment, may reduce mortality due to leptospirosis.
| Leptospirosis is often not suspected by physicians in patients with acute febrile illnesses reporting from supposedly “non-endemic areas,” including north India. Clinical manifestations are protean, and complications can affect most organ systems, including liver, kidneys, lungs, and the central nervous system. Timely diagnosis and specific therapy can reduce severity of illness and, in turn, mortality. In this study conducted at a tertiary care center in north India, we find how a much-neglected disease entity has emerged as a major cause of acute febrile illness in a so called “non-endemic area.” Incidence is increasing yearly. The majority of patients were from a rural background, and were farmers or farm labourers. Poor hygiene, contact with animals, rat infestation of houses, and contact with stagnant dirty water are the major determinants of disease. Apart from the usual symptoms of intermittent fever with chill and rigor, hepatosplenomegaly, renal decompensation, muscle pain and tenderness, and conjunctival suffusion, signs and symptoms indicating involvement of the respiratory and central nervous systems were also commonly observed. Severe complications resulting in mortality do occur and is especially due to late suspicion among primary level physicians, and the resulting inappropriate therapy.
| Leptospirosis, a worldwide zoonosis associated with sinister complications and fatalities, has been recognized in India since 1931 [1]. It is especially rampant in southern, central, eastern and western India, where heavy monsoon, animal rearing practices, unplanned urbanization and agrarian way of life predispose to this infection [1],[2],[3],[4],[5],[6],[7]. Leptospirosis has long been recognized as one of the foremost causes of acute febrile illness in those parts of the country [2],[5]. Though similar conditions exist in the north, reports of this disease from north India are few and only of recent origin (two of them from our center) [8],[9],[10],[11],[12]. Lack of awareness, clinical suspicion and active surveillance could be the probable reason. Leptospirosis has been a neglected disease even in developed countries like USA [13]. There is a wide spectrum of clinical presentations for leptospirosis. While most patients with Leptospira infection present only with mild fever and recover without complications, a small proportion develops various complications due to involvement of multiple organ systems. We have previously highlighted the importance of leptospirosis in pyrexia of unknown origin (PUO) cases [8],[9] in our center (Post Graduate Institute of Medical Education and Research, Chandigarh, India) situated in north India. Being a tertiary care center, undiagnosed and complicated patients are referred from all across the northern states.
Recently we observed an increase in the number of leptospirosis cases. We retrospectively reviewed the records of 86 such cases and observed the varied clinical manifestations and course of the disease in these patients. Apart from usual cases of acute icteric and anicteric febrile illnesses, severe manifestations of the disease, like neuroleptospirosis, haemorrhagic pneumonitis, and adult respiratory distress syndrome, were observed.
This present study is a retrospective review of records of leptospirosis cases diagnosed at our institute during the last 5 years (2004 to 2008). During this period, the microbiology laboratory received 1391 blood samples from suspected cases with pyrexia of unknown origin for leptospira serology. Paired acute and early convalescent (10–15 days into illness) serum samples were tested for specific anti-leptospira IgM antibody using the PanBio IgM ELISA (Panbio diagnostics, Brisbane, Australia). The test procedure was performed according to the protocol provided along with the kit. The results were interpreted according to manufacturer's instructions, i.e. values <9 PanBio ELISA units were considered negative, 9–11equivocal, and >11 positive. For samples showing equivocal results, another blood sample was drawn after a period of 10 days, and the test was repeated. Negative and positive controls were kept with each test run. Microscopic agglutination test (MAT) could be done in a few samples only (paired sera). For this test these samples were sent to the National Reference Laboratory for Leptospirosis, Port Blair, Andaman Islands, India. MAT was carried out following standard procedure [14] using 10 live leptospiral reference strains as antigens. The strains belonged to serogroups Australis, Autumnalis, Ballum, Bataviae, Canicola, Grippotyphosa, Icterohaemorrhagiae, Javanica, Pomona, and Tarassovi. The criteria for a positive MAT test was a titre of ≥1∶400 in a single sample, four-fold rise in titre or seroconversion in paired samples. Partial autopsy was done on one of the five patients who died, after obtaining informed written consent from close relatives. Representative tissue samples were processed for paraffin embedding and histological evaluation.
Eighty six patients positive for leptospira serology were evaluated noting their clinical history, presentation, radiological features, laboratory parameters, management, course and outcome. Utilizing clinical, epidemiological, and laboratory parameters modified Faine's criterion was scored and assessed [13]. Scoring system by modified Faine's criteria is detailed in Table 1.
The study was approved by the Institute ethical committee, PGIMER, Chandigarh. Informed consent for blood samples and for autopsy was not needed in the study. Blood samples are received in the laboratory routinely as part of patient care. Autopsies are routinely done at our center; institute ethical committee approval is not needed for the autopsy because it was not done for research purposes.
During the study period, from 2004 to 2008, there was a sustained rise of leptospirosis cases from (11.7% to 20.5%) (Figure 1A). In all we detected 232 cases of leptospirosis in the five years of study period (9 in 2004, 17 in 2005, 25 in 2006, 74 in 2007, and 107 in 2008). Cases were more common in the months of July-October for most of the years (Figure 1B). The patients resided in different states of north India, however majority of our patients were from Punjab, Haryana and Himachal Pradesh (Figure 2). Mean age of patients was 32.6 (±0.7) years with a range from 2.5 years to 78 years. Male patients (49, 57%) outnumbered female patients (39, 43%). Most of the patients (∼70%) were young adults in their 2nd, 3rd, and 4th decades of life (Table 2). Sixty six (76.7%) patients were from rural areas, travelling to endemic area was suggestive in 2 patients (Table 2). Major epidemiological risk factors noted in our patients include wet environmental living conditions, lack of protective footwear, infestation of dwelling with rats, working in farm lands, contact with animals, especially cattle, bathing in public places, history of unprotected contact with dirty stagnant water, alcohol addiction, and smoking (Table 3). Most of the patients by occupation were farmers (28, 32.6%), followed by housewives (19, 22.1%), students (11, 12.8%), labourers (10, 11.6%), indoor non-manual workers (10, 11.6%), para-military personal (2, 2.3%), sweeper (1, 1.2%), carpenter (1, 1.2%); and 4 were children below 4 years (unemployed); two of the female patients were pregnant.
Fever was seen in all 86 individuals, being intermittent and associated with chills and rigor in most of the patients. Icterus, abdominal pain, hepatomegaly, muscle pain and tenderness, headache, vomiting, breathlessness, splenomegaly, subconjunctival effusion, oliguria and altered sensorium were the common manifestations of the disease. Meningism, lymphadenopathy, arthralgia were seen in fewer cases (Table 4). The most frequently observed alterations in laboratory parameters in these patients included leukocytosis, anemia, thrombocytopenia, elevated hepatic enzymes [alanine aminotransferase (ALT) and asparate aminotransferase (AST)] in the range of 100 to 200 IU/dl, elevated serum bilirubin levels in the range of 2–8 mg/dl, thrombocytopenia, increased prothrombin time, and D-dimer positivity (Table 5). Laboratory parameters suggestive of DIC (derangement of any three of prothrombin time, activated partial thromboplastin time, D-dimer, thrombin time, fibrinogen level, fibrinogen degradation products, and fragmented RBCs on a peripheral blood film+a low platelet count) were present in ten patients. Modified Faine's criteria could diagnose 76 cases (88.3%).
By far the commonest complication was renal failure (serum creatinine >1.4 mg/dl). Other common complications observed among these patients included respiratory failure requiring mechanical ventilation, neuroleptospirosis, ascitis and pleural effusion. Laboratory confirmation of disseminated intravascular coagulation (DIC) could be noted in 10 cases; however, obvious external bleeding, petechiae and echymoses were seen in 19 cases (Table 6). Majority of complications occurred in the second week of illness or later, however 18 cases of renal impairment (34.5% of 52), three cases of respiratory failure (16.7% of 18) and one case of neuroleptospirosis (10% of 10) occurred in the first week of illness. Three cases of mixed infection with P. vivax, 2 with P. falciparum, one each with dengue and hepatitis A virus were also observed. No specific clinical feature or complication correlated with the geographic area of incidence of the disease.
Treatment with once a day ceftriaxone therapy was given to 66 cases, doxycycline therapy alone to 3 patients; and combined doxycycline and ceftriaxone therapy to 17 patients. All these patients had documented hospital acquired septicemia or were suspected of having such superinfection. High dose corticosteroid therapy was instituted in 7 cases, all of them with respiratory failure. Six of them survived. Of all 86 cases, 5 patients died (5.9%). Pathological autopsy was done in one of these cases. Leptospires were demonstrated in post-mortem kidney (Figure 3) and lung specimens of the patient using the Warthin-Starry stain. By the microscopic agglutination test the following serovars gave highest titres (>1∶400) against patients' sera: Pomona, Ballum, Gryppotyphosa, and Autumnalis.
The increase in leptospirosis cases during the last few years is possibly the result of greater awareness of this disease in the north and more drier parts of the country. WHO estimates the incidence of leptospirosis between 0.1–1 cases/100000 population/year in temperate, non-endemic areas and between 10–100 cases/100000 population/year in humid, tropical, endemic areas. Though north India receives less rainfall compared to the coastal regions and the south, most areas still receive ≥100 cm rainfall in the monsoon season between July and October. Flooding and unseasonal heavy precipitation are not uncommon like the August floods in Punjab, Bihar and Himachal Pradesh in 2008. A large proportion of the population depends on the agrarian way of life. Intimate contact with animals, unprotected entry into waterlogged fields, and bathing in contaminated community ponds are a part of rural life in across north India [2]. These are precisely the conditions most suitable for the survival and transmission of Leptospira [2]. Thus the stupendous rise in the number cases seen in this study should not come as a surprise. Also, previous reports from Chandigarh, Ludhiana, New Delhi, and Uttar Pradesh point to the fact that leptospirosis is present all over India [8],[9],[10],[11],[12]. Reports from Italy, Bulgaria and certain centers in south India point at a decreasing incidence of leptospirosis, this is however, not true at our center [15]. Like our study, a study conducted in Chennai (south India) too has seen a rapid increase in leptospirosis cases between 2004 and 2006 [3].
We used the Pan Bio IgM ELISA to screen for leptospirosis in well timed acute and convalescent blood samples, and performed MAT on few samples. It is unlikely that a large number of leptospirosis cases were missed since sensitivity of PanBio IgM ELISA as used in this study has been reprted to be as high as 76–90% [16],[17]. Though dark ground microscopy and culture in EMJH media may be performed from blood and urine samples during the acute phase of the disease, these have relatively poor sensitivity in detection of the disease [1]. In India, MAT is performed only in the reference laboratory at Andaman. Most laboratories hence prefer IgM ELISA formats for the diagnosis of leptospirosis [1]. Further this test is reactive even in early cases of leptospirosis when MAT may be negative [1].
The modified Faine's criteria could diagnose leptospirosis in 76 patients. MAT being unavailable in our center, the original Faine's criteria is not an option for our physicians. Instead the modified Faine's criterion is a useful guide. Sivakumar S et al., 2004 modified the original Faine's criteria to include local factor like rainfall, and newer investigations like IgM ELISA and Slide Agglutination test (SAT) [13]. No modifications were however made to the clinical criteria. Rainfall has been added because of the observation that most cases of leptospirosis are reported in monsoon and post-monsoon period. Compared to MAT, IgM ELISA and SAT are simpler and more sensitive tests that can be used to diagnose acute leptospira infections including milder forms which are associated with low clinical scores [13],[18]. The differential diagnosis of leptospirosis is very long, and this disease easily confuses with other viral, parasitic and bacterial infections. We suggest that the modified criterion be used by physicians in this regard.
Farmers and farming labourers (32.6% and 11.6% in our study) are the ones most commonly infected in the rural setting and the disease is associated with sowing and harvesting seasons and meteorological phenomena like monsoons. Presence of farming animals, and rodents, some of them leptospiral carriers, in the farmland, wet and humid environmental conditions for Leptospira survival, and frequent human agricultural and animal rearing activity form the core determinants of Leptospira transmission [2]. That leptospirosis occurs in those living in unhygienic conditions was evident when 34.7% of the patients lived in mud and part-mud houses and 12.8% gave history of entering waterlogged areas barefoot.
Intermittent fever with chills and rigor was the most common manifestation; however continuous fever and lack of chills and rigor were also seen in a few patients. Though the incidence of icteric and severe disease with renal failure has decreased in certain centers of south India [15], the present study found several cases manifesting with severe icteric disease and renal failure. Prabhakar MR et al. point out that epidemiological and clinical pattern of infectious disease change in course of time and leptospirosis is no exception to this rule [15]. We further argue that the pattern may vary from region to region. Knowledge of such changing epidemiological and clinical profile of leptospirosis is essential for successful prevention, early diagnosis and treatment [15].
Typically a biphasic illness, complications ensue in the second immune phase of the disease [1],[19]. Renal failure is the commonest complication noted, both in anicteric and icteric leptospirosis [1]. Azotemia, oliguria and anuria commonly occur during the second week of illness, but may appear as early as 3 to 4 days after onset [1].
Cough, pleural effusion, respiratory failure, and hemorrhagic pneumonia were commonly observed in our study. Respiratory symptoms are known to occur commonly in severe leptospirosis [19],[20],[21]. However, when respiratory complications are predominant, chances of misdiagnosis as community acquired pneumonia increase. Pulmonary complications have especially been noted to occur early and more frequently in the Andamans, and has been associated with higher mortality [21]. The higher frequency of respiratory presentation is in contrast to our earlier studies [8],[9]. Many patients with severe respiratory problems progress to develop multiple organ dysfunction syndrome and are admitted in intensive care [20] resulting in further complications from hospital acquired infections.
Majority of our patients with a diagnosis of neuroleptospirosis presented with an encephalitic syndrome, with altered sensorium, and headache. A minority also had signs of meningism, and experienced generalized seizures. This finding is similar to the case series reported by Mathew P et al., 2006 [22]. Neurological manifestations were seen in 10%–15% of leptospirosis patients [22]. Such manifestations are varied and often lead to misdiagnosis, unless strongly suspected. Most frequent manifestations include altered sensorium and neck stiffness [22]. More importantly leptospirosis is responsible for 5%–13% of all cases of aseptic meningitis [23]. Generalised tonic-clonic seizures with altered sensorium, encountered in 3 patients in our study, are a manifestation of the encephalitis [23]. Less common manifestations include hemiplegia [22],[23], intracranial bleed [23], cerebellitis [23], movement disorder [23], myelitis [23], acute flaccid paralysis including Guilain Bare syndrome [23], mononeuritis [23], facial palsy [22],[23], and neuralgias [23]. CT scan may be normal [22], however diffuse cerebral edema may be seen in a minority of patients. An early and specific diagnosis is mandatory as effective and specific therapy is available. Mathew P et al., 2006 observe that neuroleptospirosis should be considered in the differential diagnosis of neuroinfections associated with hepatorenal dysfunction in endemic areas [22].
Thrombocytopenia is especially common, while minority of patients also present with prolonged prothrombin time due to hypoprothrombinemia [1],[23]. Activation of the coagulation system is an important feature of leptospirosis [23]. Direct or indirect signs of the DIC and intravascular platelet aggregation are characteristic of malignant forms of leptospirosis [23]. Concentrations of fibrinogen, D-dimer, thrombin-antithrombin III complexes, and prothrombin fragment 1,2 are significantly elevated in leptospirosis patients [23]. In our study, patients with leptospirosis had significantly longer prothrombin times (9.3%), were D-Dimer positive (7%), and had lower platelet counts (18.3%). In all 10 patients could be categorized as having DIC. Mild leukocytosis with a shift to the left is also commonly observed in leptospirosis [1],[23]. 61.6% of our cases had an increased leucocyte count.
These severe and unusual manifestations are often not recognized as leptospirosis, and other infectious etiology like viruses looked for. Such is especially common in areas wrongly thought to be non-endemic. This leads to delay in appropriate therapy and further progression of the disease. Further, renal complications can ensue even in the first week of illness [24], although they occur more often in the second or third week of illness as observed in our study. Including leptospirosis in the differential diagnosis and instituting early empirical therapy could reduce inadvertent deaths. Treatment with high dose corticosteroids in cases with severe complications, where the immune phase of disease has begun, is controversial [25]. Only 7 of our cases were put on high dose systemic corticosteroids. Further study on the role of systemic steroids in this disease is warranted. We also describe cases with unusually high ALT and AST levels (>200 IU/ml, 2 cases) and very high bilirubin levels (>8 mg/dl, 2 cases). Chronic alcoholism (32.7% in our study) may predispose to additional liver damage and symptomatic disease [2]. Coinfection with malaria, dengue and other viruses may present diagnostic dilemmas to the treating physician. Inevitably such cases have severe manifestations and result in high morbidity and mortality.
Once a day ceftriaxone therapy has been documented to have equal efficacy to penicillin therapy, and seemed to be the preferred therapeutic regimen in our center. The mortality in treated cases was high (5.9%), however it was comparable to that seen by Jayakumar M et al. (9.5%) [26]. Severe disease and mortality from Leptospira infection is not always due to the bacteria itself, but usually due to the destructive activity of the immune system of the host [1],[23],[25]. Before death, irreversible multiple organ dysfunction syndrome usually sets in, although leptospiraemia has ceased.
A significant rise in the incidence of leptospirosis in north India was documented. Clinical manifestations and laboratory abnormalities were protean; severe complicated disease with renal or respiratory failure, neuroleptospirosis, and DIC was also observed. The increased awareness among physicians of protean clinical manifestations of leptospirosis and early laboratory diagnosis will help reduce morbidity and mortality associated with disease.
|
10.1371/journal.pntd.0005430 | Significance of major international seaports in the distribution of murine typhus in Taiwan | International seaports are hotspots for disease invasion and pathogens can persist in seaports even after ports are abandoned. Transmitted by fleas infected by Rickettsia typhi, murine typhus, a largely neglected and easily misdiagnosed disease, is known to occur primarily in large seaports. However, the significance of seaports in the occurrence of murine typhus has never been validated quantitatively.
We studied the spatial distribution of murine typhus, a notifiable disease, in Taiwan. We investigated whether risk of infection was correlated with distance to international seaports and a collection of environmental and socioeconomic factors, using a Bayesian negative binomial conditionally autoregressive model, followed with geographically weighted regression. Seaports that are currently in use and those that operated in the 19th century for trade with China, but were later abandoned due to siltation were analyzed. A total of 476 human cases of murine typhus were reported during 2000–2014 in the main island of Taiwan, with spatial clustering in districts in southwest and central-west Taiwan. A higher incidence rate (case/population) was associated with a smaller distance to currently in-use international seaports and lower rainfall and temperature, but was uncorrelated with distance to abandoned ports. Geographically weighted regression revealed a geographic heterogeneity in the importance of distance to in-use seaports near the four international seaports of Taiwan.
Our study suggests that murine typhus is associated with international seaports, especially for those with large trading volume. Thus, one of the costs of global trade in Taiwan might be elevated risks of murine typhus. Globalization has accelerated the spread of infectious diseases, but the burden of disease varies geographically, with regions surrounding major international seaports warranting particular surveillance.
| Globalization has hastened the spread of infectious diseases, with seaports as hotspots for disease invasion. Transmitted by fleas infected with the rickettsia Rickettsia typhi, murine typhus occurs worldwide, but its significance as a common causative agent of illness in tropical regions remains largely neglected. Although it is recognized that murine typhus is prevalent primarily in large seaports, the significance of seaports in the occurrence of murine typhus has never been validated quantitatively. We thus investigated whether distribution of murine typhus in Taiwan was associated with international seaports. Notably, abandoned international seaports (abandoned in the 19th century due to siltation) were also studied to see whether the causative agent of murine typhus might still circulate around the ports even after being abandoned. We found that infection risk of murine typhus was negatively associated with distance to operating seaports but was uncorrelated with nearness to abandoned seaports. In addition, the importance of distance to operating seaports for risk of murine typhus infection varied spatially. Our study highlights elevated disease risk as a cost of international trade and suggests particular surveillance in regions surrounding major international seaports.
| Trade is commonly accompanied by the spread of infectious diseases and international seaports have long been hotspots for disease invasion [1]. The great expansion in trade and international networks in recent history has seen seaports increasingly receive imported pathogens and vectors [2, 3]. For example, yellow fever has devastated seaports in the Americas due to the importation of the virus-infected mosquito Aedes aegypti (a competent vector for yellow fever) by ships [4]. Another new disease vector originating in Asia, Aedes albopictus, has also spread to seaports in both the Old and New Worlds [5, 6].
Successful introduction of exotic diseases involve arrival, establishment of local transmission, and subsequent spatial dispersal [7]. In suitable environments, exotic pathogens or parasites can persist in invaded regions even though these pathogens or parasites have ceased to arrive at the seaports. For instance, plague introduced to the USA through San Francisco in 1899–1900 still circulates among prairie dogs in the deserts of the Southwestern United Sates [8, 9] despite the absence of current importations. Likewise, helminths introduced by exotic rats have spread to indigenous mice on the California Channel Islands, with transmission persisting even after eradication of the rat hosts [10]. The probability of ongoing transmission following introduction to a new area is dependent on habitat suitability: for example, the availability of host species and/or vectors which may, in turn, be influenced by environmental conditions [11]. Thus, one legacy of past shipping events might be continuing circulation of exotic pathogens near receptive seaports; that is, although seaports have ceased to function, imported pathogens may persist in proximity to the abandoned seaports, if the conditions are suitable.
Murine typhus is a rickettsial disease with a worldwide distribution, but its significance as a common causative agent of illness in tropical regions remains largely neglected [12]. Transmitted by fleas infected with Rickettsia typhi, people typically acquire murine typhus via contaminated flea faeces near the bite sites instead of directly from the flea bites [13]. The life cycle of R. typhi commonly involves the oriental rat flea Xenopsylla cheopis and commensal rats, particularly Rattus rattus and Rattus norvegicus [14]. However, in suburban Southern California and Southern Texas, R. typhi is instead maintained by the cat flea Ctenocephalides felis, the opossum Didelphis marsupialis and domestic cats [15–18], and in Spain, dogs were found to host R. typhi [19]. It is well recognized that murine typhus is prevalent primarily in large seaports, probably due to the repeated introduction of infective fleas and rats [20]. Nevertheless, the significance of seaports in the occurrence of murine typhus has never been validated quantitatively. Likewise, while incidence of murine typhus is associated with the abundance of fleas, which is affected by climatic factors such as temperature, precipitation and humidity [20, 21], spatial analysis of the relationship between murine typhus and environmental variables remains very rare. The spatial distribution of murine typhus has been investigated in Lao PDR to confirm whether murine typhus is more common in urban areas, but only socio-economic risk factors have been included in the study [22]. Spatial clustering of murine typhus was also studied in Texas, but focusing on a comparison of clustering detection methods [23] instead of environmental correlates.
In Taiwan, murine typhus is an endemic disease, with 13 to 44 human cases annually during 2005–2014 (Taiwan Centers for Disease Control (CDC); http://nidss.cdc.gov.tw/). The spatial pattern of murine typhus occurrence and the reasons for geographic heterogeneity have never been explored in Taiwan; instead, past studies have focused on clinical manifestations of the disease [24–28]. We conducted a retrospective investigation of the spatial distribution of murine typhus in Taiwan and explored its association with environmental and socioeconomic factors. Notably, we sought to determine whether murine typhus incidence was higher in areas closer to international seaports. Seaports that are currently in use and abandoned seaports were analyzed to identify the public health consequences of historical international trade. Occurrence of murine typhus could also be related to the presence of cats, dogs and cat fleas, as recently found in Spain and the United States of America [16, 19]. However, the lack of information on the number of cats and dogs (particularly stray ones) and the spatial distribution of cat fleas in Taiwan hindered incorporation of this non-classic infection route in this research. The current study therefore focused on the classic rat-flea transmission cycle, which remains the primary route of infection all over the world [15].
The case records were retrieved from the Taiwan National Infectious Disease Statistics System administrated by Taiwan Centers for Disease Control (Taiwan CDC) and no personally identifiable information were used as part of this study.
This study focused on the main island of Taiwan. Small associated islets were excluded (Kin-men, Ma-tou, Peng-hu, Little Liu-chiu, Ci-jin, Green, and Orchard islands) because they frequently differ with regard to potentially important ecological characteristics (e.g., animal communities [29]). The basic geographical units used in this analysis were administrative districts (within urban cities) and townships (within rural counties); these are the smallest administrative areas to which murine typhus cases can be assigned. In this study, we use “district” to refer to both the urban districts and the rural townships.
Human incidence of murine typhus from 2000 to 2014 was retrospectively analyzed in this study. Murine typhus is a notifiable disease in Taiwan. Blood samples from patients with suspected murine typhus are collected and sent to the Taiwan CDC for laboratory diagnosis. Samples were considered positive for murine typhus based on a positive real-time polymerase chain reaction (PCR) test or the detection of R. typhi-specific antibodies based on the indirect immunofluorescent assay (IFA). The real-time PCR test targeted the 17-kDa antigen in Rickettsia spp. and the PCR products were sequenced and then assessed with the Basic Local Alignment Search Tool (www.ncbi.nlm.nih.gov) for resemblance to known Rickettsia spp. For IFA, each serum sample was applied to slides coated with R. typhi antigens (Focus Technologies, Inc., Cypress, CA, U.S.A.). Two IFA criteria were applied: (1) four-fold increase in R. typhi-specific immunoglobulin M (IgM) or IgG antibody in paired sera (each for the acute and convalescent phase, with interval >14 days); (2) positive for patient with IgM 1:80 dilution and IgG 1:320 dilution. Because infection may occur away from a patient’s residence, starting in 2003, the presumptive location of infection was recorded as well as the patient’s residence. These data, along with gender, age, and date of symptom onset, are available from the Taiwan CDC. To more accurately assess the relationship between infection and environmental factors, we allocated cases of murine typhus (2003–2014) to the presumed district in which the infection occurred rather than the district in which the patient resided. For incidences during 2000–2002, patient’s residence was used instead. The presumed district of infection and district of residence were the same for 97.1% of cases from 2003 to 2014, so the use of patient’s residence from 2000 to 2002 is not considered problematic. Because yearly variation (2000–2014) in district population size was low (3.7%, average value of (standard deviation divided by mean) for all districts), population size for each district was represented by the mean value from 2000–2014. Population size was obtained from the Department of Statistics of the Taiwan Ministry of the Interior (http://sowf.moi.gov.tw/stat/month/list.htm), and the murine typhus incidence rate (IR, number of cases per 100,000 people per year) was calculated for inter-district comparisons.
The presence of spatial autocorrelation of the murine typhus IR (incidence rate) was assessed using Moran’s I [30]. The locations of spatial clusters of murine typhus incidence were identified using local indicators of spatial association (LISAs). LISAs can be treated as a local version of Moran’s I [30], and can be used to detect local clusters of observations with similar or dissimilar values [31]. A map of LISAs clusters, thus, allowed the assignment of each district to one of five categories: high-high, which indicates a district with high IR surrounded by districts with high IR (also called a hot spot); low-low, a district with low IR surrounded by low-IR districts (a cold spot); low-high or high-low, a district with low IR surrounded by high-IR neighbors and vice versa; and not significant, which indicates a district with no significant local autocorrelation [32]. Inference for significance of Moran’s I and LISAs was based on 99,999 permutations using the GeoDa 0.9.5 software [33], and empirical Bayes (EB) was applied to correct for large variation in population size among districts [32], with population size as the base variable. The threshold of significance was set at P = 0.05, and maps were displayed using QGIS 2.12 (QGIS Development Team).
We selected variables for analysis based on the availability of data and our knowledge of the study system. Twelve explanatory variables (seven environmental variables, two socioeconomic variables, and three port distance variables) were included in the study. Environmental variables included elevation (elevation, meters), total annual rainfall (rainfall, mm), mean annual temperature (temperature, °C; calculated as the mean of 12 monthly mean temperatures), number of days with temperature higher than 30°C within a year (daysT30), relative humidity (%) and a selected list of land cover categories. Elevation was derived from a 40-m digital elevation model (Aerial Survey Office of Taiwan Forestry Bureau). The four meteorological variables were obtained from Central Weather Bureau of Taiwan (n = 390 meteorological stations, the Data Bank for Atmospheric Research is available at https://dbahr.narlabs.org.tw/) and were calculated over the period 1991 to 2013. The spatial layers of climatic variables were generated at a spatial resolution of 1 km by interpolation (390 stations) using Kriging in ArcGIS with a spherical variogram model [34]. We overlaid administrative district boundaries and calculated the mean values for elevation, rainfall, temperature, days over 30°C and relative humidity for each district. Land cover data were obtained from the Globcover database [35] using a spatial resolution of 30 arc seconds (ca. 1 km) and the initial land cover classes were merged to create a smaller number of land cover types likely to be important for R. typhi transmission. These include artificial structure and forests (artificial surface and forest) because human infection of R. typhi occurs mainly inside buildings [20] and we were interested in the potentially protective effects of forests. The proportion of each district that consisted of each of these land cover classes was calculated to provide a quantitative characterisation of the land cover.
To assess the role of socioeconomic factors, average income (income) of each district for the year 2005 was obtained from the Fiscal Information Agency of the Taiwan Ministry of Finance (http://www.fia.gov.tw/). Population density for each district was obtained by dividing population size by the respective administrative area.
In this study, distance to three different types of international seaports were analyzed for comparison with R. typhi infection: (1) currently in use (n = 4); (2) operated mainly during the 19th century for the trade of commodities with mainland China, where murine typhus has long been prevalent along the coast [36], but were largely abandoned later because of siltation (n = 26); and (3) including both in-use and abandoned international seaports (n = 28). Two seaports which were operational during the 19th century remain in operation today, and so are included in all three of these categories.
International seaports that are currently operated include Keelung, Taichung, Kaohsiung, and Hualien seaports (Fig 1). Keelung and Kaohsiung seaports have been in use since the 19th century while Taichung and Hualien seaports have been operated since the 1970s. 19th century Taiwanese seaports have been classified into ten categories based on the volume of seaborne goods handled [37]. Some of the ports with the largest amount of cargo handled were deemed international ports in this study, because each had direct marine traffic with mainland China [37]. These 26 international ports are mostly located along the coast although some are situated along rivers (Fig 1) and only two of them (Keelung and Kaohsiung) continue to engage in international trade. Distance to international seaports was represented by the Euclidean distance from the geographical centroid of each district to the nearest ports. Overlay of the district boundaries on grids of environmental variables and the calculation of the nearest distance to ports were implemented in ArcGIS 10.2.
Correlation analysis was applied to assess the strength and direction of the association amongst the explanatory variables. Where variables were highly correlated with one another, only one of the variables was retained for subsequent non-spatial multivariate regression analysis to avoid multi-collinearity. Lastly, significant variables in the final non-spatial multivariate model were analyzed separately with a Bayesian spatial regression model and geographically weighted regression.
A total of 476 human cases of murine typhus were recorded during 2000–2014, with an incidence rate of 0.14 cases per 100,000 residents per year; this was higher in males than in females (0.20 vs. 0.08; Chi-square test with Yates’ correction, χ2 = 85.0, P < 0.001). The incidence rate also varied with age (χ2 = 167.1, P < 0.001) and was higher in the 50–79 age range (Fig 2A). There was also a significant seasonal variation in incidence rate (χ2 = 114.6, P < 0.001), with rates higher in later spring and summer than in other seasons (Fig 2B).
Among the 349 districts, the number of cases of murine typhus during 2000–2014 ranged from zero to 16 cases, with more cases occurring in southwest and central-west Taiwan (Fig 3A). The IR varied from zero to 3.1 cases per 100,000 residents per year and was higher in southwest and central-west Taiwan, along with central Taiwan (Fig 3B).
Incidence of murine typhus was not randomly distributed in Taiwan (Moran’s I = 0.35, P<0.0001). Instead, the LISA map revealed that hot spots were present in southwest and central-west Taiwan while cold spots occurred in eastern Taiwan (Fig 4).
This research has examined the spatial distribution of murine typhus in Taiwan, and possible explanatory factors for this distribution, for the first time. We found spatial clustering of human cases of murine typhus in southwest and central-west Taiwan. The risk of infection was higher in areas closer to international seaports that are currently in use, particularly near Kaohsiung and Taichung seaports. However, the probability of infection was not significantly associated with distance to abandoned international seaports. Risk of murine typhus was also negatively associated with rainfall and temperature, after controlling for distance to in-use international seaports.
It has been stated that ports are the primary foci of murine typhus transmission [20]. Nevertheless, to the best of our knowledge, this is the first study to quantitatively validate a negative association between risks of R. typhi infection and distance to seaports, based on an advanced spatial modeling approach. Higher risks of infection near (active) seaports suggest that these may be the source of infection, as a consequence of repeated introduction of infective rats and/or fleas from abroad in combination with the mild weather typically enjoyed by coastal cities that is also hospitable for rats and fleas [20]. In spite of the negative association between disease incidence and distance to operating international seaports, the IR of murine typhus and the importance of distance varies considerably among the four ports. Distinctly, negative association between IR of murine typhus and distance to operating seaports occurs primarily near the Kaohsiung and Taichung seaports (Fig 6C). There have been no cases surrounding the Hualien seaport in eastern Taiwan and there are very few cases along the eastern coast of Taiwan (Fig 3), in stark contrast with the high prevalence along the western coast, particularly near the Kaohsiung and Taichung seaports. This is consistent with the finding of a higher seropositivity rate of R. typhi infection in shrews and rodents trapped in Kaohsiung seaport (26.1%) and Taichung seaport (18.1%) than the other eight seaports or airports (including Keelung seaport of 0.7% and Hualien seaport of 1.7%, [44]). Such geographical variation could be due to the remarkable difference in trading volume among the four ports, with Kaohsiung dealing with the lion’s share of international cargo (an annual mean of 112 million tons during 2011–2013), followed by Taichung (60 million tons), Keelung (18 million tons), and Hualien (4 million tons) (Taiwan International Ports Corporation; http://www.twport.com.tw/). The lack of cases in proximity to Hualien might be related to the smaller cargo volumes providing fewer opportunities for pathogen introduction although this could also be related to higher temperature and rainfall near this port (Supporting information S1 Fig) so that pathogen transmission cannot be easily sustained after being imported. It was also found that the spatial distribution of murine typhus differed from that of scrub typhus, another rickettsial disease transmitted by mites. In Lao PDR, murine typhus was more common in urban areas while scrub typhus was more common in rural regions [22]. This contrasting spatial distribution also occurs in Taiwan, where scrub typhus is much more prevalent in less developed eastern areas than industrialized western areas of Taiwan [45, 46]. Although R. typhi was not detected in fleas in eastern Taiwan [47], rickettsial strains similar to R. typhi have been detected in ticks and rodents in the same region [48, 49], indicating that murine typhus might also circulate in this region but may be overlooked by physicians. This could be due to low prevalence as revealed by the low seropositivity rate of R. typhi infection in shrews and rodents in Hualien seaport (1.7%, [44]). Our study suggests that murine typhus should be considered as a possible diagnosis when patients close to the Hualien seaport present with suspected rickettsial infections. Indeed, clinical manifestations of many rickettsial diseases (e.g. high fever, headache, rash) are so similar that identification of the etiologic agent is very challenging, especially in the tropics [50]. Under-reporting is thus likely to be common, particularly in Hua-lien, where the other rickettsial disease (scrub typhus) is very prevalent [45, 46] and murine typhus might be readily excluded.
Although it is expected that poorer hygiene in the 19th century vessels might render rats infested with fleas more likely to board ships and invade ports, we did not find quantitative evidence supporting higher risks of infections near ports that operated in the 19th century, but which have subsequently been abandoned. This suggests that local conditions might not be suitable for long-term sustained transmission, and as these ports have largely been abandoned since the late 19th century, there was little opportunity for recent introduction at these locations. Because more people work in operational than abandoned seaports, more food might be available to sustain a higher population of rats and fleas in operational than abandoned ports. However, whether the lower risks were the result of lower survival of rats, fleas or R. typhi in abandoned ports remains to be investigated due to a lack of systematic studies on these ports. Another possibility is that the transmission cycle is sustained to the current date in so few abandoned ports that the significance of abandoned ports cannot be established statistically. In other words, contemporary infection might continue in a few abandoned ports since the 19th century, but because infection has ceased in most abandoned ports, we were unable to recognize its significance when all obsolete ports are considered in spatial analysis. For example, while the two hotspots in southwest and central-west Taiwan are also close to abandoned seaports (Fig 4), the majority of obsolete ports have very low incidence; although a similar spatial pattern could also arise where there are so many abandoned ports that hotspots coincidently occur close to a few of them.
It is very difficult to unpick the significance of in-use versus abandoned seaports although our results suggest that some in-use seaports are of more importance for contemporary murine typhus incidence than abandoned ports. One potential solution to this issue is to investigate the population genetic structure of R. typhi in Taiwan. Cargos moving through abandoned and operating seaports came from different locations: abandoned seaports are likely to have dealt with cargo mainly from coastal China, while in-use seaports are likely to deal with cargo mainly from other countries. Therefore, the genetic composition of R. typhi should differ based on origin prior to introduction to Taiwan. This information would allow a more comprehensive assessment of the importance of abandoned seaports in the contemporary spatial distribution of infection. Studying the population genetic structure would also help discern whether R. typhi is mostly imported (i.e. genetic composition differs among international seaports) or is spread from within Taiwan (i.e. no spatial structure in genetic composition is observed). The current status of rats and fleas (species and abundance) in operational and abandoned ports could also be better understood when trapping rodents to investigate the genetic structure of R. typhi in fleas; this could help reveal how the non-sustained transmission of R. typhi in abandoned ports could be related to the survival of rats or fleas. Another limitation of the current study is that due to the lack of data on trading volume at abandoned seaports [37], the probability of importation of R. typhi at each seaport is considered identical, even though the volume of trade varies considerably among ports. Anping, Lukang and Wanhua were regarded as the largest ports in the 19th century in Taiwan, but there was no evidence of spatial clustering around these obsolete ports (Fig 4), suggesting that historical trading volume might not be the primary determinant of contemporary murine typhus infection risks.
Lastly, whereas serological assay is the primary method for diagnosis of murine typhus [51], cross-reactivity can occur in human sera against R. typhi and R. felis antigens [52–55], although it is unclear why similar cross-reactivity does not always occur [e.g. 56, 57]. Potential serologic cross-reactivity suggests that confirmed cases of murine typhus based solely on IFA diagnosis may include some misdiagnosed R. felis infections (also a flea-borne rickettsial disease). In Taiwan, molecular methods have detected R. felis or R. felis-like organisms in one patient [58] as well as in small mammals [49] and fleas [47, 59, 60]. Therefore, we cannot exclude the possibility that murine typhus cases confirmed by the Taiwan CDC may include some cases of R. felis and the case data might more accurately reflect infections of flea-borne rickettsial diseases (caused by R. typhi or R. felis) instead of infectious associated with R. typhi only. In Taiwan, however, sera of confirmed cases of murine typhus were not found to cross-react with R. felis antigen [61] and serum from the single patient detected with R. felis nucleotides did not cross-react with R. typhi antigen [58]. Based on this, the extent of R. felis infections in patients diagnosed with murine typhus is presumed to be minimal in Taiwan, but this warrants further investigation. Also awaiting validation is the importance of R. felis as the causative agent of human illness, which is recently questioned for its widespread distribution in cat fleas but few and spatially restricted human cases of flea-borne rickettsioses in California [62, 63] and its ubiquity in a diverse array of arthropods and also in healthy people in Africa [51]. Whether R. felis is simply a symbiont of arthropods similar to Wolbachia [64] would therefore determine if infection of R. felis is required to be considered in epidemiological studies of murine typhus.
Across Taiwan, rainfall and temperature were also significantly associated with murine typhus incidence, after controlling for the influence of distance to operating international seaports. Given the same distance to seaports, murine typhus incidence decreased with increasing rainfall and temperature. This suggests that after R. typhi was introduced at the ports, the probability of further inland invasion of rats, fleas or R. typhi may have been determined by local climates. It was found that in eastern Taiwan, fleas were more abundant in months with less rainfall and lower temperature, but the underlying mechanism still awaits investigation [47]. In fact, climatic effects on flea-borne diseases are complex and context dependent. For example, similar to murine typhus, transmission of plague also involves bacteria, fleas, and rodent hosts. While it is generally thought that fleas prosper under hot and humid conditions, Yersinia pestis, the etiologic agent of plague, can persist in arid regions (such as Central Asia and Western USA), but is less likely to sustain transmission in humid tropical areas [21]. Likewise, fleas were predominantly collected in dry rather than humid regions. However, in the dry part of Reunion Island, fleas were more abundant during the hot-wet season [65]. This is akin to our finding that geographically, murine typhus tended to occur in cooler and drier areas, but seasonally murine typhus was more prevalent in the warmer season (late spring and summer). Apart from fleas, climate could also influence the abundance of rodents and human behavior [21], both of which could affect the infection risk of murine typhus. The precise relationship between rainfall, temperature and murine typhus incidence in Taiwan (given the same distance to seaports) is, therefore, complex and necessitates great care to disentangle it. This is further supported by the geographical heterogeneity in the importance and direction of relationship for temperature and rainfall (Fig 6A and 6B). Moreover, it should be emphasized that while an association between infection risk of murine typhus and climatic variables is identified, such correlation does not definitely represent that climate does determine the risk of infection; other not-recognized variables correlated with rainfall and temperature might instead be the main determinant.
Our study demonstrates that one of the costs of international trade in Taiwan might be an elevated risk of murine typhus. This can be exemplified by Kaohsiung seaport, whose container traffic ranks 13th globally (World Shipping Council; www.worldshipping.org). Kaohsiung is not only a hotspot for murine typhus (this study); dengue and scrub typhus, both vector-borne diseases, are also common in this port city [45, 66]. To prevent potential importation of exotic diseases, regulations mandated by Taiwan CDC that govern quarantine at international ports require all arriving ships to report occurrence of rodents and disease vectors on the ships. Small mammals, fleas and seroprevalence of R. typhi in rodents are also monitored constantly and the eradication of rats has been attempted in these four international seaports by Taiwan CDC [44]. Globalization has hastened the spread of infectious diseases [67, 68], but the burden of diseases varies geographically, and as this study has shown, regions surrounding international seaports should warrant particular surveillance. Also needed is the assessment of whether eradication programmes implemented in seaports do indeed mitigate the risks of targeted diseases.
|
10.1371/journal.pbio.2000016 | Regulation of the Human Telomerase Gene TERT by Telomere Position Effect—Over Long Distances (TPE-OLD): Implications for Aging and Cancer | Telomerase is expressed in early human development and then becomes silenced in most normal tissues. Because ~90% of primary human tumors express telomerase and generally maintain very short telomeres, telomerase is carefully regulated, particularly in large, long-lived mammals. In the current report, we provide substantial evidence for a new regulatory control mechanism of the rate limiting catalytic protein component of telomerase (hTERT) that is determined by the length of telomeres. We document that normal, young human cells with long telomeres have a repressed hTERT epigenetic status (chromatin and DNA methylation), but the epigenetic status is altered when telomeres become short. The change in epigenetic status correlates with altered expression of TERT and genes near to TERT, indicating a change in chromatin. Furthermore, we identified a chromosome 5p telomere loop to a region near TERT in human cells with long telomeres that is disengaged with increased cell divisions as telomeres progressively shorten. Finally, we provide support for a role of the TRF2 protein, and possibly TERRA, in the telomere looping maintenance mechanism through interactions with interstitial TTAGGG repeats. This provides new insights into how the changes in genome structure during replicative aging result in an increased susceptibility to age-related diseases and cancer prior to the initiation of a DNA damage signal.
| Telomerase is very tightly regulated in large, long-lived species such as humans. Telomerase is expressed during early human fetal development, turned off in most adult tissues, and then becomes reactivated in most human cancers. However, the exact mechanism(s) regulating these switches in expression are not fully known. We recently described a phenomenon in which genes near chromosome ends (telomeres) are regulated by telomere length-dependent loops (telomere position effect—over long distances; TPE-OLD). Interestingly, the TERT gene is only a megabase from the human chromosome 5p end. We observed that when telomeres are long, TERT gene expression is repressed and the 5p sub-telomeric region and the TERT locus are spatially co-localized. When telomeres are short, at least one of the TERT alleles is spatially separated from the telomere, developing more active histone marks and changes in DNA methylation in the TERT promoter region. These findings have implications for how cells turn off telomerase when telomeres are long during human fetal development and how cancer cells reactivate telomerase in cells that have short telomeres. These studies add to the growing support for the role of telomeres in regulating gene expression via TPE-OLD. Furthermore, telomere length may be one of the mechanisms of how cells time changes in physiology without initiating a DNA damage response.
| All mammalian telomeres (the ends of linear chromosomes) are composed of large tracts of repeated 5ʹ-TTAGGG sequences. Telomeres are well-conserved DNA end structures from yeast to mammals, and it is believed that the primary role of telomeres, in combination with shelterin proteins, is to provide protection of the linear chromosome ends from being recognized as damaged or broken DNA [1] and to facilitate the completion of DNA replication each cell cycle. Telomeres prevent DNA end-joining, DNA recombination, and loss of essential genetic information during DNA replication. Telomeres are maintained by many essential genes, including the six-component shelterin (TRF1, TRF2, POT1, TIN2, RAP1, and TPP1) and the CST (CTC1-STN1-TEN1) complexes [1,2]. Impairment of these genes is closely associated with age-related clinical pathology and defects in germ cell and stem cell maintenance [3–5].
It is well established that hTERT, the catalytic core reverse transcriptase component, protein levels are rate-limiting for telomerase activity and telomere length homeostasis [6]. Human embryonic stem cells and transit amplifying adult progenitor stem-like cells express hTERT and have active/functional telomerase that can fully or partially maintain telomeres during the substantial number of cell divisions required in fetal development [7]. While telomerase is present from the blastocyst stage in early human development, at approximately 16–18 wk of gestation, telomerase activity is silenced in the vast majority of somatic cells [8]. The molecular mechanisms (i.e., transcriptional regulation, alternative splicing changes, epigenetic modifications, or other regulatory processes) that trigger the silencing of telomerase at specific times during human development remain uncertain. Irrespective, telomerase largely remains silent throughout adult life except for tumor development. In ~90% of human tumors, telomerase is upregulated or reactivated for the maintenance of telomeres during the numerous rounds of cell divisions required for the emergence of malignant and metastatic disease [9]. Thus, tight regulation of telomerase and progressive telomere shortening are thought to be an initial barrier to the early onset of cancer.
High resolution mapping of the three-dimensional chromatin interactome addresses many unanswered questions about the cis-regulatory long-range looping interactions important in gene regulation. The human genome is composed of continuous chromosome loops and TADs (topologically associating domains), forming gene territories [10,11]. Distal enhancers and/or insulators are believed to be responsible for the regulation of genes along the genome via chromatin folding. Recently, we demonstrated that telomeres also loop to specific loci to regulate gene expression, which we have termed TPE-OLD (telomere position effect—over long distance) [12–14]. In the examples characterized so far, genes close to telomeres are silenced in young cells (with long telomeres) and become expressed when telomeres are short. Importantly, re-elongation of cells with short telomeres by exogenous expression of the hTERT gene (active telomerase) results in expression patterns that mirror the expression of these genes in cells with long telomeres [12–15]. As we have observed genes between the TPE-OLD regulated genes that are not regulated by TPE-OLD, this mechanism is clearly distinct from classic TPE, which regulates genes proportional to the proximity to the telomeric repeats [15]. In the present study, we show that the expression of the hTERT gene itself is also regulated by TPE-OLD. The ability to regulate genes by telomere length without induction of a DNA damage signal from a single or a few critically short telomeres has potential explanatory value for what regulates the maximum length of human telomeres during fetal development and ways to regulate major age-associated transitions as well as to activate or repress genes as part of normal aging without requiring a DNA damage signal.
Long-ranged genomic interactions between telomeres and distal loci may play important roles in the regulation of gene expression, a phenomenon that we previously referred to as TPE-OLD [12,13]. Through previous microarray analyses [12], we identified the human CLPTM1L (cleft lip and palate-associated transmembrane protein 1-like) gene that is ~1.3 mega bases apart from the chromosome 5p telomere as a putative TPE-OLD candidate gene. CLPTM1L is frequently upregulated in cancer cells [16] and shows preserved colocalization with the TERT locus for a shared synteny in many species (Fig 1A). We analyzed mRNA expression of the genes at this locus, including CLPTM1L and hTERT, in BJ human fibroblast clones with long and short telomeres, to determine if the expression of this locus is regulated by TPE-OLD. CLPTM1L was expressed in normal young passaged cells but showed increased gene expression with progressive telomere shortening (S1A Fig). Historically, it is generally believed that hTERT is not actively transcribed in normal telomerase silent cells; however, expression of hTERT splice variants does occur [17]. The reason for this misconception is that most investigators use primer pairs designed to measure transcripts containing only the RT domain of TERT (exons 5–10), while exons outside of the RT domain are not measured (i.e., exons 1–4 and 11–16). It is now known that hTERT transcripts can be detected in a variety of telomerase-negative cells and tissues, but the mRNA produced is not full-length mRNA capable of producing active telomerase [17]. To test if replicative age or telomere length influenced hTERT expression, we measured hTERT gene expression using a primer pair targeting the 5ʹUTR to exon 1 of hTERT. We observed that hTERT is expressed at higher levels in two human fibroblast strains with short telomeres compared to the same cells with long telomeres (Fig 1B, S1 Fig). As previously described, we did not detect any transcripts that contain the RT domain of hTERT (Fig 1B); thus, transcripts that could code for active telomerase were not observed. We also analyzed protein expression of CLPTM1L (S1B Fig) and observed that the expression of CLPTM1L protein significantly increased during progressive telomere shortening, but the expression was greatly decreased when we re-introduced hTERT in old BJ cells and re-elongated telomeres (S1B Fig). We also examined mRNA expression of genes located between the 5p telomere and the hTERT-CLPTM1L locus (S1A Fig) in young and old BJ cells. The expression of the intermediate genes on chromosome 5p showed no significant increase in BJ cells with short telomeres (S1A Fig). We explored if telomere repeat containing RNA, TERRA, was also altered and potentially important in TPE-OLD. Consistent with previous reports [18,19] we observed an increase in TERRA expression from three subsets of chromosomes (1q-21q, 5p, and 9p-15q-Xq-Yq; Fig 1C) when telomeres were short compared to long. The TERRA data support our observations that the chromatin environment surrounding chromosome 5p and hTERT change when telomeres are short. Overall, this implies that the hTERT locus may be influenced by the length of telomeres through long-ranged chromatin interactions.
Perhaps not surprising, but potentially significant, is that the location of the TERT gene is also evolutionarily conserved (Fig 1D). TERT genes are located at the very end of their chromosomes, near the telomere, in higher primates including humans and most other large long-lived mammals. However, the location of the TERT gene in rodents and many other smaller shorter-lived mammals is non-telomeric. The local genome structure around the TERT locus in rodents is quite different from primates, implying they may have developed different strategies to regulate telomerase expression [20,21]. Based on these observations, we decided to test if there is a functional role for TERT being localized at the end of human chromosome 5p. As the distance between the hTERT locus and the telomere is only ~1.3 mega bases, we postulated that hTERT might also be regulated in part by a long-ranged telomere looping mechanism in human cells.
We designed two specific BAC probes to visualize the hTERT locus and the sub-telomeric 5p region for three-dimensional fluorescence in situ hybridization (3D-FISH) (Fig 2A). We measured the distance between the hTERT locus and the sub-telomeric 5p region, and the pairs of alleles were divided into adjacent to (A) or separated (S) by the three-dimensional location (S2A and S2B Fig). We first stained the sub-telomeric BAC region, the hTERT locus and telomeres in old BJ cells, with short telomeres (Fig 2B). The telomere staining was detected at the hTERT locus with sub-telomere 5p in the adjacent allele pair. However, we observed at least one hTERT allele that was spatially separated from the sub-telomere 5p probe in old BJ cells without telomere staining. We measured the distance between the hTERT locus and the closest telomere (Fig 2C). The results showed that the hTERT locus colocalized with the telomere when it is adjacent to the sub-telomeric 5p region (Fig 2B and 2C). This implies that the telomere is likely to be adjacent to the hTERT locus for potential long-ranged looping interactions.
We next tested if telomere looping close to the hTERT locus changes when telomeres became short. We measured and compared the distance between the hTERT locus and sub-telomeric 5p in young BJ fibroblasts at 20 population doublings (PDs) with long telomeres versus old BJ fibroblast at PD90 with short telomeres (Fig 2D). More than 70% of allele pairs were adjacent in BJ cells at PD20, implying that the telomeric heterochromatin might affect the expression of the hTERT locus in young BJ fibroblasts. BJ cells are telomerase-negative, but non-catalytic alternatively spliced variants are expressed, as shown in Fig 1 and as previously described [17]. This might explain why a small proportion of alleles is separated from the telomere in telomerase-negative young BJ cells with long telomeres, based on the assumption that the looping interactions suppress transcription. In old BJ cells at PD90, we found that the percentage of adjacent allele pairs was significantly reduced. Almost 60% of alleles were separated in the old cells with short telomeres, indicating that there is at least one hTERT locus spatially separated from the telomere in each cell. Importantly, we confirmed these 5p/TERT looping interactions in a second fibroblast cell strain, IMR90 (S2C and S2D Fig). We measured the number of separated and adjacent alleles in IMR90 cells young (PD 22) and old (PD 52) and show a shift from the majority of alleles being adjacent (76%) in young cells compared to the majority of alleles being separated (88%) in old cells. The looping data and the expression of hTERT are consistent. We suggest that old cells (with short telomeres) lose one control mechanism in regulating the hTERT locus (i.e., telomere chromatin looping) that helps repress the expression of hTERT. However, while we observed increased transcription of exon 1 of hTERT, there must be additional mechanisms preventing the inclusion of exons critical to produce active telomerase. There is substantial evidence that alternative splicing of hTERT may also have a major role in suppressing the production of active telomerase in old cells [22–24]. Furthermore, we performed 3D-FISH analysis in transformed SW26 and SW39 cells. SW cells are SV40 antigen expressing clones of IMR90 cells that have spontaneously immortalized using either telomerase (SW39) or an alternative lengthening of telomeres (ALT; SW26) mechanism to maintain telomeres (S2E Fig). In both cell lines, the majority of the alleles were separated (SW39 = 72%; SW26 = 66%), indicating that short telomeres due to replicative aging are likely responsible for the change in chromatin conformation and that a secondary change occurs to cause the production of full-length TERT or engage ALT.
It has been suggested that hTERT shows mono-allelic expression in cancer, which is sufficient to preserve constant telomere length [25,26]. Our results support this assumption, as we observed that, on average, only one hTERT allele was generally in the open configuration during in vitro aging well before the onset of cancer. As controls for global conformational changes at chromosome 5p, we performed two additional FISH experiments. In the first experiment, we stained intermediate genomic region between the hTERT locus and the 5p telomere (S3A–S3C Fig). In addition, we also stained cells for two loci located 25.5 MB and 30.6 MB away from hTERT (S3D–S3F Fig). There were no changes in distances between the control loci in young and old cells, demonstrating that the conformation change occurs at the specific genomic region encompassing hTERT during in vitro aging, and this change is not due to classic TPE.
To determine if we could artificially shorten telomeres and recapitulate the aging phenotype, we utilized CRISPR/Cas9 (clustered regularly interspaced short palindromic repeat-associated 9) to remove a large portion of the telomere and subtelomere region from chromosome 5p. This experiment allows testing the role of chromosome 5p’s telomere in regulating the looping observed in cells with short and long telomeres. As illustrated in Fig 2D, we also infected young BJ cells with a lentivirus expressing Cas9 protein and single guide-RNA targeting the sub-telomeric region of 5p to specifically disturb telomeres at chromosome 5p for a short period of time [27]. We also added an NHEJ inhibitor, SCR7, simultaneously during the infection to suppress repair of the double strand breaks induced by the Cas9 protein [28]. The targeted cells showed an unstable end structure of chromosome 5p (S5 Fig), and the specific disturbance of the 5p telomere significantly diminished telomere looping at the end of the chromosome 5p.
We further examined if the proposed mechanism was present in BJ cell clones in which both young and old cells were passaged the same amount of time in culture. This approach was necessary to eliminate the possibility that young and old cells that were in culture for vastly differing times could introduce artifacts. To accomplish this, we expressed a floxable hTERT in BJ clones, followed by excision at different time points in order to make isogenic cells with different length of telomeres but passaged similar times in cell culture [12,29]. Telomere length of the early-excision clone was 9 kb, and this was extended up to 13 kb in the late-excision clone. The telomere length (terminal restriction fragment [TRF]) results are presented in Figure 1B. Population doublings were evenly matched between clones (to avoid confounding effects of passage of time in culture), and we also analyzed telomere looping. Similar to our observations in normally passaged BJ cells, the isogenic clones also showed decreased levels of telomere looping with telomere shortening (Fig 2E). Importantly, there were only background levels of DNA damage signaling during telomere shortening (Fig 2F) indicating that the change in genome structure occurred before initiation of DNA damage responses from critically short telomeres. To ensure that our staining protocol was robust, we induced DNA damage (double strand breaks) by treating long and short telomere BJ cells with zeocin and assaying for DNA damage (S2F Fig). These data can be interpreted to indicate that our staining protocol is robust and that we are analyzing cells before telomere-DNA damage induced foci are present or significant DNA damage occurs in the cells.
We next performed droplet digital 3C (chromatin conformation capture) to detect the genomic interactions between the 5p telomere and the hTERT locus in young and old BJ cells (Fig 2G, left side). The results showed that the hTERT locus has specific genomic interactions with the 5p telomere, and the interaction was reduced during in vitro aging and telomere shortening. A proximity control primer which is 10kb away from the fixed primer at the hTERT locus was selected for normalization of 3C results (Fig 2G, right side). Taken together, telomere looping exists between the hTERT locus and the sub-telomeric 5p in normal human cells, and this looping is greatly reduced by gradual telomere shortening.
It has been shown that cis-elements upstream of the hTERT locus may play important roles in the tight regulation of human telomerase [30]. Thus, we decided to test if telomere looping could affect the epigenetic status of the hTERT proximal promoter region. We first analyzed DNA methylation of the region from -720bp to +90bp of the hTERT promoter in isogenic BJ cells with different lengths of telomeres but similar times in cell culture (Fig 3A). The relationship between DNA methylation and transcription in the hTERT promoter remains controversial in normal and cancer cells [31,32], but the transcription start site of hTERT retains little or no methylation in telomerase-active cancer cells for active transcription [33]. We found that the level of DNA methylation is significantly higher in BJ cells with long telomeres at several regions associated with hTERT and the hTERT region in comparison to cells with shorter telomeres. The largest differences were observed at -580bp, -250bp, -30bp, and +20bp of the hTERT promoter, including the E-box motif (a putative Myc binding sequence). It has also been reported that the proximal region of the hTERT promoter, including exon 1 and 2, regulates the activity of the hTERT promoter and that the methylation of this region is responsible for binding of several proteins [34,35]. Therefore, our results can be interpreted to indicate that telomere length-associated changes in methylation levels of the hTERT proximal promoter might affect transcriptional regulation of this locus. We next analyzed active and inactive histone marks on the hTERT proximal promoter using chromatin immunoprecipitation combined with droplet digital polymerase chain reaction (ChIP-ddPCR; [12]) (Fig 3B). We measured two histone marks associated with active chromatin H3K4 trimethylation (H3K4me3) and H3K9 acetylation (H3K4ac) and two histone marks associated with repressed chromatin H3K27 trimethylation (H3K27me3) and H3K9 trimethylation (H3K9me3), which have key roles in regulating gene expression [36]. We observed an increase in both H3K4me3 and H3K9ac across the TERT promoter in aged cells with short telomeres (Fig 3B). We also observed an increase in the repressive histone mark H3K27me3, but did not observe any significant differences in young or old BJ cells for the repressive histone mark H3K9me3. Collectively, this shows that the chromatin status of the hTERT promoter in old BJ cells with short telomeres is different and may be more transcriptionally permissive compared to young BJ cells with long telomeres. These data correlate well with the increased hTERT transcription we observed in cells with short telomeres. Furthermore, we analyzed chromatin at the promoters of three genes surrounding TERT that could also be affected by the altered chromatin environment with aging. We analyzed the proximal promoter regions of CLPTM1L, SCL6A18, and SCL6A19 for the same histone marks described above in the same cells and preparations used for TERT ChIP. At the CLPTM1L promoter we observed significant increases in histone marks indicating active transcription (Fig 3B). These data correlate well with an increase in CLPTM1L transcripts and protein levels (S1 Fig). We also observed significant changes in the chromatin surrounding the solute/amino acid transporter genes (SCL6A18 and SCL6A19), even though these genes are not expressed above basal/background levels in old/short telomere BJ cells. Specifically, we observed that both the repressive histone marks were increased in old cells (short telomeres) compared to young cells (long telomere). However, there was an increase in the activation marks as well. This indicates an intricate balance between chromatin modifications, methylation status, telomere length, and the expression of tissue-specific transcription and splicing factors that dictates the activation or repression of genes with replicative aging (telomere shortening—TPE-OLD).
While we demonstrated that telomere shortening induced conformation changes between the hTERT locus and the sub-telomeric 5p resulting in up regulation of exon 1, presumably containing spliced hTERT transcripts in normal BJ cells (see Fig 1B), it did not result in full-length telomerase activity competent transcripts. Thus, we suggest that telomere shortening may render the hTERT locus more permissive and under oncogenic stress may lead to the production of full-length hTERT mRNA transcripts that could in turn produce telomerase activity. To test this, we simulated a step in spontaneous cancer transformation by knocking down p21 (CDKN1A) and analyzing mRNA expression level of hTERT (Fig 3C). The knockdown of p21 was previously shown to de-repress hTERT expression [37]. Thus, we tested if the knockdown of p21 would increase the expression of hTERT mRNAs and result in the inclusion of exons 7/8 in the short-telomere old BJ cells but not in the young BJ cells with long telomeres. We measured the expression level of hTERT transcripts in young and old BJ cells with and without p21 stable knockdown; mRNA containing exons 7/8 (exons coding for critical residues in the reverse transcriptase domain of TERT) and exon 15/16 (most splice variants of hTERT contain exons 15 and 16), responsible for putative active hTERT and total hTERT variants respectively. Both the active and the total hTERT transcript variants significantly increased with the knockdown of p21 in old BJ but not in young BJ cells; however, we did not detect telomerase activity (S6 Fig). While we observed an increased portion of transcripts that contain exons 7/8 of the TERT RT domain, other critical regions such as exon 2 may be spliced out [38]. Further work into the regulation of hTERT splicing is necessary to more fully understand the complex regulatory network surrounding hTERT and why the majority of transcripts are inactive splice variants as opposed to full length. While this result does not prove a causal role during cancer development, this series of experiments does demonstrate that telomere shortening in cells that bypass replicative senescence leads to the hTERT locus entering into a more permissive state (e.g., increased hTERT mRNA expression) in the presence of oncogenic stresses, consistent with the disengagement of telomere looping.
Characterization of cis- or trans-acting factors responsible for telomere looping will be important to understand this novel mechanism for telomerase regulation. A recent report showed that TRF2 (telomeric repeat-binding factor 2) protein is essential for the functional organization of chromosome ends, including human fibroblasts [39,40]. There is also mounting evidence for off-telomere functions of the shelterin components [41]. While a recent whole genome sequencing study found 2,920 interstitial TTAGGG repeats throughout the human genome [39], we also found frequent internal (interstitial) telomeric sequences (ITS) near the TERT locus in higher primates but not in rodent cells (Fig 4A). Thus, we first checked for a putative role of TRF2 in telomere looping in BJ cells as a candidate approach. We knocked down TRF2 by siRNA and performed 3C to directly assess the genomic interactions between the telomere and the hTERT locus (Fig 4B). The knockdown of TRF2 significantly reduced the genomic interactions between the telomere and the hTERT locus in young PD30 BJ cells, implying TRF2 may have a role in telomere looping interaction on hTERT locus.
As shown in Fig 4A, a region 100 kb downstream of the hTERT (Chr5: 1,154,047–1,154,347) contains a series of internal telomeric sequences that may recruit TRF2 shelterin protein (hereafter termed hTERT-ITS). Thus, we reasoned that this region would be a putative binding site for TRF2 and may be responsible for the telomere looping interaction between the telomere and the hTERT locus in cells with long but not short telomeres. ChIP-qPCR analysis showed that the TRF2 protein associates with the hTERT-ITS region in young and old BJ cells as proposed (Fig 4C). We next performed 3C to further clarify that hTERT-ITS interact with the hTERT promoter by genome folding to affect transcriptional permissiveness as shown in Figs 1 and 3. Within 200kb, we found more than 20 HindIII restriction enzyme sites were in the hTERT/CLPTM1L locus (Fig 4D). Droplet digital PCR (ddPCR)-mediated amplification showed specific interactions between the 5ʹ end of hTERT and the hTERT-ITS (Fig 4E). Moreover, the interaction was weakened in old BJ cells, implying there might be a transition from a more repressive state to a more active state of this TAD location during in vitro aging, consistent with the increased hTERT mRNA, altered methylation, and chromatin. This result also shows that there is an additional genome folding between the hTERT locus and the hTERT-ITS at an intermediate region between the SLC6A18 and SLC6A19 loci. The hTERT promoter is not close to the hTERT-ITS on a linear genome map, but the unique genome folding at this region potentially positions the hTERT promoter close to the ITS, followed by putative TRF2-mediated telomere recruitment to the hTERT promoter only in cells with long telomeres.
In Fig 4F, we demonstrate that TRF2 protein is also enriched in the hTERT promoter region using ChIP-qPCR approaches. While TRF2 protein was enriched at proximal regions on the hTERT promoter, the interaction was significantly decreased in old BJ cells at the genomic regions containing -350bp to -50bp of the hTERT promoter. This shows TRF2 protein can occupy the hTERT promoter region, but the interaction is weakened during in vitro aging and telomere shortening. Together, we interpret these experiments to indicate that TRF2, and perhaps upregulated TERRA, may have at least a partial mechanistic role in telomere looping at the hTERT locus through interaction with the conserved interstitial telomeric repeats.
Because we have shown the interaction between the 5p telomere and the hTERT locus, we modeled one possibility for the detailed local genome structure of this locus (Fig 4G). In this model, the hTERT promoter is close to the hTERT-ITS by genome folding in young cells with long telomeres. In addition, this model shows that TRF2 protein is recruited to hTERT locus and hTERT-ITS, which makes this interaction potentially dependent on telomere length. In summary, the hTERT promoter has specific interactions with the hTERT-ITS through gene looping, which may also recruit telomere length-dependent looping (TPE-OLD) mechanisms through TRF2 protein.
We next performed 3D-FISH to visualize the genomic structure changes between the hTERT locus and the sub-telomeric 5p (Fig 4H). Control PD17 BJ cells showed that 89% of the hTERT and sub-telomeric 5p allele pairs were adjacent, but knockdown of TRF2 reduced this down to 34%. We also knocked down CTCF (CCCTC-binding factor) and LDB1 (LIM domain-binding protein 1), which are proposed to be essential proteins in global gene looping maintenance [42,43]. CTCF and LDB1 knockdown also significantly reduced the adjacent allele pairs, implying that the general gene looping mechanisms may also be involved in telomere looping. Western blotting was also performed to show knockdown efficiency (Fig 4I). Taken together, TRF2, part of the shelterin complex, may be mechanistically involved in the establishment of telomere looping near the hTERT locus through ITS together with general chromosome looping mechanisms.
In almost all primary human cancers, telomere length is very short compared to adjacent normal tissues [44]. It is likely that short telomeres, in combination with oncogenic alterations, result in the hTERT gene becoming more permissive for protein expression and enzyme activity. Thus, we next investigated how telomere length affects hTERT expression in telomerase-active cancer cells. We first infected hTERT and hTR (hTERC) into the SW39 cell line (SV40 immortalized human telomerase expressing fibroblasts) and analyzed mRNA expression of the endogenous hTERT by examining the 3ʹ untranslated region. We observed that the extended telomere length reduced endogenous expression of hTERT mRNA in qPCR analysis (Fig 5A) implying TPE-OLD remains engaged at least in this tumor cell line.
We further established isogenic HeLa cell clones with different telomere lengths by excising a floxable hTERT cDNA at different time points. We examined expression of splice variants of hTERT mRNA containing total, full-length (indicative of telomerase activity), and minus beta alternative spliced forms through ddPCR analysis (Fig 5B). All three splice variants showed significantly decreased expression in the long-telomere HeLa clone. We also performed 3D-FISH to analyze changes in genomic structure between the hTERT locus and the sub-telomeric 5p after the extension of telomeres in HeLa cells (Fig 5C). The long-telomere HeLa clone showed a higher percentage of adjacent allele pairs compared with the short-telomere HeLa clone. This indicates that the expression of hTERT may also be influenced by the length of telomeres through TPE-OLD in telomerase-positive cancer cells.
The local genome structure around the hTERT locus may be important for the tight regulation of human telomerase. For example, introduction of proximal cis-elements of the hTERT promoter sufficiently inhibits the activity of the TERT promoter [45]. In addition, chemicals perturbing chromatin structure, including trichostatin A and 5-aza-2ʹ-deoxycytidine, induce changes in hTERT expression [46]. Moreover, chromosomal translocation and gene duplication of the hTERT locus can occur as part of the immortalization process in primary cultured cells [47,48]. Here, we reasoned that the hTERT locus might recruit telomeric heterochromatin to regulate its own gene expression, especially in large, long-lived mammals where tumor suppression mechanisms are perhaps more important. We showed that telomere looping exists in long-telomere young fibroblasts and that telomere looping was reduced by in vitro aging. This is one possible explanation for why higher primates preserved the location of the TERT gene at the end of one of their chromosomes. We speculate that, in addition to other conserved tumor suppressor mechanisms, higher primates also developed a mechanism to suppress the undesired expression of TERT. For example, it is well established that during human fetal development, full-length telomerase transcription is repressed and correlates with increases in nonfunctional alternative splicing changes in hTERT [8]. Thus, during early human development, when telomerase is active, telomeres elongate. Our current results are consistent with the idea that longer telomeres can fold back on the TERT locus and repress or significantly reduce transcription. Our results also show that replicative senescence, while initially a tumor suppressor mechanism, may paradoxically impinge on the predisposition to cancer through telomerase transcriptional de-repression.
While still preliminary, the hTERT locus is arranged in a local chromatin domain that is regulated by telomere length and the interstitial telomere sequences in the vicinity of the hTERT locus. We showed that expression of the CLPTM1L gene (adjacent to hTERT) is also regulated by the length of telomeres and predicts transcriptional permissiveness of this locus. However, because hTERT re-activation is an extremely rare event, there may be additional levels of regulation. We propose that, upon telomere shortening, the hTERT region becomes permissive (as indicated by increased transcription of exon 1 containing RNAs), but this first step is not sufficient to support full-length hTERT transcripts at an adequate level to produce telomerase enzyme activity. We further propose that there is another biological role for telomere looping at this locus during development to repress telomerase when telomere length homoeostasis is reached (i.e., suggesting that having too-long telomeres may be detrimental).
Here, we demonstrated a novel epigenetic mechanism regulating hTERT expression during in vitro aging (Fig 5D). Cells with long telomeres at the end of chromosome 5p in young passaged cells form a chromatin loop in the region of the hTERT locus. Importantly, we demonstrated that the chromatin loop is disengaged in cells with short telomeres, leading to partial increased expression of hTERT mRNA during in vitro aging and in response to p21 knockdown; however, telomerase activity was not detected, and, alternatively, spliced variants were likely produced [17,22–24,38]. Finally, we demonstrated that, in old cells with short telomeres, re-introduction of hTERT and elongation of telomeres results in a re-engagement of TPE-OLD. We found that DNA methylation and histone modifications in the hTERT promoter region showed significant changes as cells developed shorter telomeres, and that TRF2 and, perhaps, TERRA, may have important roles in these age-dependent genomic changes. These observations offer a model and a partial explanation for how age-dependent changes in the genome structure affect the regulation of hTERT without initiating a DNA damage response from a critically shortened telomere.
BJ, SW39, HeLa, HEK293FT, IMR90, and Phoenix A cells were maintained in a 4:1 ratio of Dulbecco’s modified Eagle’s medium to Medium 199 containing 10% of fetal bovine serum (Hyclone, Logan, UT, USA) under 5% CO2 in a humidified incubator. Retrovirus containing human TERT cDNA was infected into BJ cells and HeLa cells, followed by adenoviral infection for transient expression of Cre recombinase at different time points to produce cells with different lengths of telomeres that had been passaged in vitro for similar times [12,29]. Retrovirus was prepared by transfecting viral vectors into Phoenix A cells for 48 h. Medium containing virus was filtered through a 0.45 μm pore and provided to cells in the presence of 2 μg/ml polybrene. Lentivirus was prepared by transfecting viral vectors into HEK293FT cells with two packaging vectors (pMD2 and psPAX2) for 48 h. Medium containing virus was filtered through a 0.45 μm pore and cells exposed to lentivirus in the presence of 2 μg/ml polybrene. Selection for hygromycin was performed using 100 μg/ml and for puromycin using 1 μg/ml. CRISPR-Cas9 introduction for 5p genomic editing was performed by infecting cells with lentivirus carrying sgRNA target sequence of 5ʹ-GCCTCACTCCTTACGGAGTG-3ʹ.
3D-FISH was performed as described previously [12]. 104 BJ cells were seeded into 4-chambered slides. Cells on slides were fixed with 4% paraformaldehyde, followed by permeabilization with 0.1% Triton X-100 in PBS. Repeated liquid nitrogen freezing-thawing cycles were performed for further permeabilization with preservation of intact nuclear structure under 20% glycerol in PBS. After 5 d of incubation of with 50% formamide in 2X SSC, cells were stained with indicated probes at 37°C for overnight. Slides were washed with 0.1% SDS in 0.5X SSC at 70°C for 5 min, followed by 2 rounds of PBST (Phosphate-buffered saline with Triton X-100) washing for 10 min. Images were acquired using a LSM780 confocal microscope (Carl Zeiss, Jena, Germany), and analyzed by Imaris deconvolution software (Bitplane, Zurich, Switzerland). The proximity of allele pairs was determined visually and quantitated. At least 30 nuclei were counted for the statistical analyses. We used the following criteria for the analyses: adjacent ~0.5 μm space or less between probes, separated ~1.0 μm between probes or more. The length was determined by calculating the 3D distance between each center of deconvolved fluorescent spots. Probes were prepared using nick translation kits (Abbott Laboratories, Abbott Park, IL, USA) from each BAC following manufacturer’s instructions. BAC plasmids were purchased from CHORI (Children’s Hospital Oakland Research Institute, Oakland, CA, USA); RP11-990A6 for hTERT locus staining and RP11-44H14 for sub-telomeric region 5p staining. Quality of probes was assessed by metaphase spread analyses and PCR.
DdPCR and ddTRAP were performed as previously described [12,49]. Messenger RNA was prepared from RNeasy plus mini kit (Qiagen, Valencia, CA, USA) following the manufacturer’s instructions. 100 ng of RNA was reverse-transcribed from cDNA synthesis kit (Bio-Rad, Hercules, CA, USA) by following the manufacturer’s instructions. Ten percent of synthesized cDNA was used for the ddPCR reaction. For ddTRAP, harvested cells were lysed in NP40 lysis buffer (1mM Tris-Cl pH8.0, 1mM MgCl2, 1mM EDTA, 1% NP40, 0.25mM sodium deoxycholate, 10% glycerol, 0.15M NaCl, 0.05% 2-ME) for ddTRAP. Lysate was used for TS extension, and the extended products were analyzed with ddTRAP. Endogenous levels of 3ʹUTR were assessed with EvaGreen dye (Bio-Rad, Hercules, CA, USA). Probes were purchased from Roche (Basel, Switzerland), and the primer sequences are described below:
100 ng of gDNA was modified using the EpiTect Bisulfite kit by following the manufacturer’s instructions (Qiagen, Valencia, CA, USA). Modified DNA was PCR-amplified and cloned into the T vector system (Promega, Madison, WI, USA). 7~10 bacterial clones were sequenced for methylation analysis. Primers for the hTERT promoter region amplification were designed as previously described [33].
Chromatin conformation capture (3C) was performed as previously described [12]. Five million cells were washed with PBS and fixed with 25 ml of medium containing 1% formaldehyde for 10 min at room temperatures. To quench the crosslinking reaction, 1.5 ml of 2.5 M glycine was added and incubated for 10 min at room temperature, followed by an additional 15 min of incubation at 4°C. Cells were washed with PBS and harvested into 1 ml of cold-PBS with protease inhibitor. Cells were next lysed by homogenization, and the nuclear pellet was collected by centrifugation. The nuclear pellet was washed and resuspended in 500 μl of ice-cold NEBuffer 2 (NEB, Ipswich, MA, USA). 15 μl of 10% SDS was added and incubated at 37°C for 1 h, followed by addition of 46.35 μl of 20% Triton X-100 for 1 h on a shaking incubator. HindIII (400U) was added and incubated overnight. Enzyme reaction was stopped by adding 88 μl of 10% SDS at 65°C for 20 min. Samples were next transferred to DNA ligation mix containing 50 mM Tris-Cl, pH 7.5, 10 mM MgCl2, 1 mM ATP, 10 mM DTT, and 50 μg/ml BSA. 372 μl of 20% Triton X-100 was added and incubated at 37°C for 1 h. 2,000 U of ligase (NEB, Ipswich, MA, USA) was added and incubated for 5 h at 16°C. 40 μl of 20 mg/ml Proteinase K was added to the ligation mix at 65°C overnight. DNA extraction was performed by phenol-chloroform extraction and ethanol precipitation. Quality of the libraries were determined by checking for a single DNA band under agarose gel electrophoresis. Taq-man probe and 5ʹ primers were selected to amplify constant regions at the 5p telomere regardless of genome conformation. 3ʹ primers were selected to amplify the genomic interaction between 5p telomere and subtelomeric genes up to 1.3 mega base pairs from 5p containing hTERT. Primer binding regions are 100 base pairs apart from a HindIII recognizing motif. Primer and probe sequences are described below;
Chromatin immunoprecipitation was performed as previously described [12]. Antibodies against total histone H3 and a 1:1 mixture of rabbit and mouse IgG isotypes were used as pulldown positive and negative controls of ChIP analyses, respectively. Relative occupancy was determined by first normalizing the target results with amplification signals from total H3 and then dividing by 1% input chromatin extracts. Antibodies against H3K4me3, H3K27me3, H3K9me3, H3K9ac, and LDB1 were purchased from Abcam (Cambridge, MA, USA). Antibody against TRF2 was purchased from Novus biologicals (Littleton, CO, USA). Antibody against CTCF and histone H3 was purchased from Cell signaling (Cell signaling technology, Danvers, MA, USA). Primers for hTERT promoter amplification were described in a previous study [33]. Primers for detection of CLPTML1, SLC6A18, CLC6A19, and the hTERT-ITS are described below;
The TIF assay is based on the co-localization detection of DNA damage by an antibody against gamma-H2AX and telomeres using FITC-conjugated telomere sequence (TTAGGG)3-specific peptide nucleic acid (PNA) probe. Briefly, BJ cells with long and short telomeres (100,000 cells) were seeded to four-well chamber slides, and, after the cells attached to the surface (next day), slides were rinsed twice with 1xPBS and fixed in 4% formaldehyde (ThermoScientific, IL) in PBS for 10 min. Then, cells were washed twice with PBS and permeabilized in 0.5% Triton X-100 in PBS for 10 min. Following permeabilization, cells were washed three times with PBS. Cells were blocked with 10% goat serum in 0.1% PBST (TritonX-100) for 1 h. Gamma-H2AX (mouse) (Millipore, Billerica, MA) was diluted 1:1,000 in blocking solution and incubated on cells for 2 h. Following three washes with PBST (1x PBS in 0.1% Triton) and three washes with PBS, cells were incubated with Alexaflour 568 conjugated goat anti mouse (1:500) (Invitrogen, Grand Island, NY) for 40 min, then washed five times with 0.1% PBST. Cells were fixed in 4% formaldehyde in PBS for 20 min at room temperature. The slides were sequentially dehydrated with 70%, 90%, and 100% ethanol. Following dehydration, denaturation was conducted with hybridization buffer containing FITC-conjugated telomere sequence (TTAGGG)3-specific peptide nucleic acid (PNA) probe (PNA Bio, Thousand Oaks, CA), 70% formamide, 30% 2xSSC, 10% (w/v) MgCl2.6*H20 (Fisher Sci), 0.25% (w/v) blocking reagent for nucleic acid hybridization and detection (Roche) for 7 min at 80°C on heat block, followed by overnight incubation at room temperature. Slides were washed sequentially with 70% formamide (Ambion, Life Technologies, Grand Island, NY) / 0.6 x SSC (Invitrogen) (2 x 1 h), 2 x SSC (1 x 15 min), PBS (1 x 5 min), and sequentially dehydrated with 70%, 90%, and 100% ethanol, then mounted with Vectashield mounting medium with DAPI (Vector Laboratories, Burlingame, CA). Images were captured with Deltavision wide-field microscope using the 60X objective. TIFs were quantified using Image J and representative pictures were prepared in Imaris software after deconvolution using Autoquant X3.
The average length of telomeres (terminal restriction fragment lengths) was measured as described in [50] with the following modifications. DNA was transferred to Hybond-N+ membranes (GE Healthcare, Piscataway, NJ) using vacuum transfer. The membrane was air-dried and DNA was fixed by UV-crosslinking. Membranes were then probed for telomeres using a DIG-labeled telomere probe [51], detected with an HRP-linked anti-DIG antibody (Roche), and exposed with CDP-star (Roche).
|
10.1371/journal.pntd.0003212 | Melarsoprol Sensitivity Profile of Trypanosoma brucei gambiense Isolates from Cured and Relapsed Sleeping Sickness Patients from the Democratic Republic of the Congo | Sleeping sickness caused by Trypanosoma brucei (T.b.) gambiense constitutes a serious health problem in sub-Sahara Africa. In some foci, alarmingly high relapse rates were observed in patients treated with melarsoprol, which used to be the first line treatment for patients in the neurological disease stage. Particularly problematic was the situation in Mbuji-Mayi, East Kasai Province in the Democratic Republic of the Congo with a 57% relapse rate compared to a 5% relapse rate in Masi-Manimba, Bandundu Province. The present study aimed at investigating the mechanisms underlying the high relapse rate in Mbuji-Mayi using an extended collection of recently isolated T.b. gambiense strains from Mbuji-Mayi and from Masi-Manimba.
Forty five T.b. gambiense strains were used. Forty one were isolated from patients that were cured or relapsed after melarsoprol treatment in Mbuji-Mayi. In vivo drug sensitivity tests provide evidence of reduced melarsoprol sensitivity in these strains. This reduced melarsoprol sensitivity was not attributable to mutations in TbAT1. However, in all these strains, irrespective of the patient treatment outcome, the two aquaglyceroporin (AQP) 2 and 3 genes are replaced by chimeric AQP2/3 genes that may be associated with resistance to pentamidine and melarsoprol. The 4 T.b. gambiense strains isolated in Masi-Manimba contain both wild-type AQP2 and a different chimeric AQP2/3. These findings suggest that the reduced in vivo melarsoprol sensitivity of the Mbuji-Mayi strains and the high relapse rates in that sleeping sickness focus are caused by mutations in the AQP2/AQP3 locus and not by mutations in TbAT1.
We conclude that mutations in the TbAQP2/3 locus of the local T.b. gambiense strains may explain the high melarsoprol relapse rates in the Mbuji-Mayi focus but other factors must also be involved in the treatment outcome of individual patients.
| Sleeping sickness, or human African trypanosomosis, constitutes a serious health problem in sub-Sahara Africa. Treatment is a key factor in the control of this disease, not only to save the lives of the individual patients but also to stop transmission. As for other infectious diseases, drug resistance forms a constant threat for sleeping sickness control. Understanding the mechanisms underlying drug resistance is of uppermost importance for evidence-based adaptation of treatment protocols as well as for the development of new drugs. In this study we investigated the phenotype and genotype of more than 40 recently isolated strains of Trypanosoma brucei gambiense, the parasite that causes sleeping sickness in West and Central Africa. By comparing the parasites from cured and from relapsing patients in two separated sleeping sickness foci, we can explain the difference in treatment failure rates between these two sleeping sickness foci in D.R. Congo. We also provide evidence that treatment outcome in the individual patient is not exclusively defined by the drug resistance genotype of the infecting parasite but that other factors must be involved.
| Human African trypanosomosis (HAT) is a parasitic disease transmitted by tsetse flies (Glossina sp) and caused by Trypanosoma brucei (T.b.) gambiense and T.b. rhodesiense. This disease constitutes a serious public health problem in sub-Saharan Africa, particularly in central African countries like the Democratic Republic of the Congo (DRC), Central African Republic (CAR), the Republic of the Congo and the Republic of South Sudan [1]. The disease evolves from an early stage with trypanosomes invading blood, lymph and peripheral tissues towards the late or neurological stage with invasion of the brain. Chemotherapy of HAT relies on five drugs (eflornithine, melarsoprol, nifurtimox, pentamidine and suramin). Early-stage HAT is treated with pentamidine (T.b. gambiense) or suramin (T.b. rhodesiense). For treatment of the neurological stage, drugs that are able to pass the blood-brain-barrier, such as melarsoprol, nifurtimox or eflornithine, are necessary [2]. Until recently, melarsoprol was the first line treatment of late stage gambiense and rhodesiense HAT but for gambiense HAT, the nifurtimox-eflornithine combination therapy (NECT) is now recommended by the World Health Organization as first line treatment [3]. This recommendation follows the observation that NECT is as effective as melarsoprol monotherapy with less severe side effects and that NECT is able to cure patients that experienced a relapse after treatment with melarosprol monotherapy [4]. Traditionally, 5–10% of gambiense HAT patients treated with melarsoprol could not be cured but in the last decade up to >50% relapse rates were reported in Angola, Central African Republic, Democratic Republic of the Congo, Republic of South Sudan and Uganda [5]–[11]. Particularly problematic was the situation in the HAT focus of Mbuji-Mayi, East Kasai Province in DRC, where a 57% failure rate was observed in patients treated with the 10 days abridged melarsoprol regimen [8]. Various studies have been conducted to explain these unexpectedly high relapse rates, considering either the parasite or the human host being responsible for this phenomenon. Although known mechanisms may be involved, such as mutations of the P2 adenosine transporters and of aquaglyceroporin transporters in the trypanosome membrane, the explanation for treatment failure is probably more complex, including parameters of the parasite (reduced drug sensitivity and higher tissue tropism), the drug (content of active principle and correct administration) and the host including individual differences in pharmacokinetics, co-infections and disease stage [9], [12]–[24].
To investigate the mechanism underlying the high failure rates in the Mbuji-Mayi focus, Pyana and co-workers undertook the large scale isolation of the trypanosome from HAT patients in that focus [25]. Thus, we established a collection of 85 T.b. gambiense type I strains from cured and relapsed patients, some of which from the same patient before treatment and after relapse. Among these strains, 41 were adapted to Mus musculus in order to test their sensitivity to melarsoprol in an in vivo mouse infection model and to analyse some genetic features that may be related to reduced sensitivity to melarsoprol. Recently, some of them were shown to be resistant to pentamidine and melarsoprol by in vitro drug sensitivity testing and to carry a TbAQP2/3 chimera that was supposed to lead to a reduced uptake of pentamidine in the trypanosome flagellar pocket [20]. In this study, we aimed at 1° investigating the in vivo melarsoprol sensitivity phenotype of all 41 mouse adapted T.b. gambiense strains isolated in the Mbuji-Mayi focus, 2° investigating some of their genotypic characteristics and comparing them with 4 strains isolated from a sleeping focus with low relapse rates in Masi-Manimba, and 3° relating their phenotype and genotype with treatment outcome of the patients from whom they were isolated.
The study in mice was approved by the Veterinary Ethics Committee of the Institute of Tropical Medicine, Antwerp, Belgium (protocol PAR-022) and adhere to the European Commission Recommendation on guidelines for the accommodation and care of animals used for experimental and other scientific purposes (18 June 2007, 2007/526/EG) and the Belgian National law on the protection of animals under experiment.
The parasite strains included in this study belong to the cryobank of the World Health Collaboration Center for Research and Training on Human African Trypanosomiasis Diagnostics at the Institute of Tropical Medicine in Antwerp, Belgium. Their isolation and use for research purposes was approved by the Ethical Committee of the Institute of Tropical Medicine (04441472) and of the Ministry of Health of DRC [25]. The strains are anonymized by using international codes and alias names.
A list of the T.b. gambiense strains from the Mbuji-Mayi focus in East Kasai Province, DRC, is given in Table 1. The isolation history of these strains is described elsewhere [25]. The alias name of each strain indicates whether it was isolated before treatment (BT) or after treatment (AT). Eleven strains were isolated from patients that were cured after melarsoprol treatment. Thirty strains were isolated from patients that relapsed after treatment. Among these 30 strains, twenty belong to “couples”, i.e. isolated from the same patients, before treatment and after relapse. The melarsoprol sensitive and resistant strains included as reference in the in vivo drug sensitivity experiment, T.b. brucei 427 wild type and T.b. brucei 427 AT1/P2 KO (P2 adenosine transporter knock out) were received from the Swiss Tropical and Public Health Institute. For the genotype analysis, 7 extra T.b. gambiense type I strains were added (Table 1). Four of these strains were isolated in 2011 from cured patients in the Masi-Manimba focus, Bandundu Province, RDC where high relapse rates after melarsoprol treatment were never observed. Three strains are “old” isolates. LiTat 1.3 is a cloned population of the Eliane strain, isolated in Côte d'Ivoire, and previously shown to be sensitive to melarsoprol and pentamidine in vitro, while MBA and KEMLO are two Congolese strains with unknown drug sensitivity profile [26]. All the strains were kept as 250 µl cryostabilates in liquid nitrogen.
Each stabilate was thawed in a water bath at 37°C and immediately, 250 µl of phosphate buffered saline glucose (PSG, 7.5 g/l Na2 HPO4 2H2O, 0.34 g/l NaH2 PO4 H2O, 2.12 g/l NaCl, 10 g/l D-glucose, pH 8) were added. This mixture was kept on ice until inoculated intraperitoneally (IP) into two 1–2 months old female OF-1 mice (Charles River, Belgium). Two days before infection, these mice had been immunosuppressed by IP injection with 200 mg/kg body weight (BW) of cyclophosphamide diluted in water (Endoxan, Baxter, Lessing, Belgium). Parasitaemia was monitored three times a week on a fresh preparation of 5 µl of tail blood according to the matching method of Herbert and Lumsden (1976). If needed, immunosuppression was repeated after 5 days, until the parasitaemia reached 107.5/ml or more and the trypanosome population was large enough to inoculate 12 mice for the in vivo melarsoprol sensitivity experiment. For the inoculation of the mice in the in vivo drug sensitivity experiments (see below), 30 to 40 µl of infected tail blood was diluted in about 3 ml of PSG, trypanosomes were counted in a Uriglass cell counting chamber (Menarini Diagnostics) and the suspension was further diluted in PSG to obtain a concentration of 250 trypanosomes/µl.
For the treatment of mice, melarsoprol (Aventis, 5 ml vials of 180 mg in propylene glycol, lot nr. 725) was freshly diluted in 50% polyethylene glycol (PEG400, Sigma Aldrich, Belgium) to concentrations of 1 mg/ml.
For each experiment, twelve female OF-1 mice (1–2 months old, 20–30 g body weight (BW)) were immunosuppressed as described above, two days before inoculation and subsequently at days 3, 11, 25 and 85 post-infection (DPI). Each mouse was inoculated IP with 200 µl PSG containing 5×104 trypanosomes. Parasitaemia was monitored as described above from day 4 post-infection onwards. As soon as trypanosomes were detected in all mice (between 4 and 6 DPI), one group of 6 mice received IP injections of melarsoprol at 10 or 12 mg/kg BW during 4 consecutive days. On day 5–10 post infection, the mice of the control group were euthanised with an IP injection of sodium pentobarbital at 350 mg/kg BW (Nembutal, CEVA Santé Animale, Brussels, Belgium) and blood was taken by heart puncture on heparin to prepare sediments of pure trypanosomes for DNA extraction and genetic analysis (see below). The mice of the melarsoprol treated group were checked for the presence of trypanosomes in tail blood two times a week during the first two weeks and subsequently once a week for maximum 100 days. As soon as a trypanosome was detected in at least one mouse of this group, the experiment was stopped and the strain was considered “resistant”. Relapsing mice were sacrificed and blood was collected on heparin to separate the trypanosomes from the blood via DEAE chromatography for further genotypic analysis. On days 90 to 100 postinfection, all mice that remained trypanosome negative were sacrificed and blood was collected on heparin where after it was run over two mini Anion Exchange Centrifugation Technique columns to detect subpatent parasitaemia [27].
Infected blood from mice was passed over a DEAE cellulose column (1∶6 blood∶gel ratio) to separate the trypanosomes from the blood [28]. The trypanosomes eluting from the column were washed three times with 5 ml ice-cold PSG by centrifugation. After the last centrifugation, the supernatant PSG was discarded and the trypanosome sediment was frozen at −80°C until DNA extraction. After thawing and addition of 200 µl of phosphate buffered saline, pH 8, genomic DNA was extracted from the trypanosome sediment with the Maxwell 16 Tissue DNA Purification kit on the Maxwell 16 robot (Promega, Madison, WI, USA) and DNA was stored at −20°C. DNA concentrations were measured with the Nanodrop ND-1000 UV-Vis spectrophotometer (NanoDrop Technologies, Wilmington, USA) and adjusted to 10 ng/µl if appropriate.
In a preparatory experiment, mice were infected with strain 348BT, isolated from a patient who had been cured, and were treated with melarsoprol at 1, 2, 3, 4, 5, 8 and 10 mg/kg BW. The minimum melarsoprol dosage needed to cure the mice from the infection appeared to be 10 mg/kg BW. This dosage was then used to treat mice infected with all the 41 T.b. gambiense strains and with the two T.b. brucei strains. Some mice infected with T.b. gambiense strains 15BT, 163AT and 346AT experienced a relapse that was detectable only about three months after the infection and after immunosuppression with cyclophosphamide (Table 2). All mice infected with the other T.b. gambiense strains and with both T.b. brucei strains (including the AT1/P2 KO strain) got cured by melarsoprol at 10 mg/kg/BW, as defined by the absence of detectable relapse up to 100 days after infection. To confirm the apparent melarsoprol resistance of 15BT, 163AT and 346AT, the experiment was repeated with these strains and with melarsoprol treatment at 10 and 12 mg/kg BW. All mice infected with 15BT remained without detectable parasitaemia after treatment. Among the mice infected with 163AT, one mouse relapsed at DPI 28 after treatment with 10 mg/kg BW and one mouse relapsed at DPI 31 after treatment with 12 mg/kg BW melarsoprol. All mice infected with 346AT and treated with 10 and 12 mg/kg BW melarsoprol, relapsed at DPI 20 (Table 2).
This study was undertaken to investigate the mechanisms underlying the high relapse rates observed in second stage gambiense HAT patients treated with melarsoprol in Mbuji-Mayi, DRC. The in vivo melarsoprol sensitivity experiment showed that a minimum dose of 10 mg/kg BW melarsoprol was needed to cure mice infected with trypanosomes that were isolated from a cured patient. This is 4 times higher than the dose needed to cure mice infected with T.b. gambiense strains isolated from Ibba in South Sudan, another HAT focus known for high melarsoprol relapse rates [9]. On the other hand, Kibona and co-workers considered 3 out of 35 tested T.b. rhodesiense strains as resistant when mice relapsed after treatment with 5 mg/kg BW melarsoprol [30]. Among the 41 strains from Mbuji-Mayi, 2 induced infections that could not be cured with 12 mg/kg BW melarsoprol. Both were isolated from patients after treatment with melarsoprol.
When setting up the in vivo drug sensitivity experiment, we were confronted with the lack of a standardised protocol, especially for T.b. gambiense. Most studies have been dealing with T.b. brucei or T.b. rhodesiense, both behaving quite virulent in laboratory mice. Although melarsoprol is a drug that can cure the chronic phase of trypanosomosis, we opted for an acute phase in vivo model for several reasons: i. the concentration of melarsoprol that reaches the central nervous system is only a minor fraction of what reaches the plasma and is more prone to uncontrolled variations among individual outbred animals, ii. the acute phase model is expected to correspond better with the standard in vitro model, iii. since T.b. gambiense can cause subclinical or even silent infections, assessing treatment outcome in a chronic model via examination of blood and organs, including the brain, is unreliable [31]. The protocol we used here is mainly based on the study that Maina et al. carried out on recent isolates of T.b. gambiense from Sudan [9]. We also immunosuppressed the mice before inoculation with 5×104 trypanosomes to guarantee that all mice would become infected. Some major differences however are to be noted. During the preparatory experiments, we noted that melarsoprol precipitates immediately when diluted in water, the usual diluent in other in vitro and in vivo studies [7], [9], [30]. In our final protocol, we diluted melarsoprol in polyethylene glycol to keep it in solution facilitating correct dosage when treating the mice. In contrast to the custom 60 days after treatment follow-up, we monitored the mice for up to 100 days after treatment and we immunosuppressed them on day 85 after treatment. In addition, at the end of the follow-up period, we sacrificed the mice and passed all the blood on DEAE cellulose columns instead of checking only a few drops of tail blood with the microhaematocrit technique. This allowed us to observe relapses at days 85–90 after treatment that otherwise would have been missed. Still, some treated mice showed paralysis but without any detectable trypanosome in the blood, suggesting that the real relapse rate was higher than what we actually can report based on trypanosome detection only. A weakness in our study is the absence of well documented T.b. gambiense melarsoprol resistant control strain. In the absence of such a strain, we had to rely on the T.b. brucei 427 AT1/P2 KO strain and its corresponding T.b. brucei 427 wild-type strain. Both strains appeared to be sensitive to melarsoprol at 10 mg/kg BW which is consistent with what has been observed in previous studies on a Tbat1 null mutant [14]. On the other hand, for 5 strains included in our in vivo experiment, it was shown in vitro that they were 2–4 times less sensitive for melarsoprol than the reference sensitive T.b. gambiense strain STIB 930 [20]. Basing our treatment dose for the in vivo study solely on the dose required to cure a stabilate isolated from a cured patient proved to be a limitation for further interpretation of the in vivo results. Our initial hypothesis, including the rationale for the isolation of couples, was that strains originating from cured patients would be more sensitive to melarsoprol than strains from relapsed patients. It was clearly not expected that all strains would carry drug resistance markers. The different in vivo melarsoprol sensitivity phenotypes observed in our experiment do not correspond with the low variability observed within the two studied genetic markers associated with melarsoprol resistance. Indeed, all strains from Mbuji-Mayi, irrespective of their isolation from cured or relapsing patients, carry the wild type TbAT1 allele. In addition, in all strains from Mbuji-Mayi, the TbAQP2 and TbAQP3 are replaced by chimeric TbAQP2/3 variants, of which one has been reported previously to correlate with in vitro pentamidine and melarsoprol resistance by Graf and co-workers [20]. The latter study included 5 strains from the Mbuji-Mayi collection (40 AT, 45 BT, 130 BT, 349 BT, and 349 AT). According to Graf and co-workers, these 5 strains contained the TbAQP2/3 chimera and the wild type TbAT1, what is confirmed in our study, and showed decreased sensitivity for pentamidine and for melarsoprol in vitro. In our study we found a second variant of a chimeric AQP2/3 gene in these strains, which was only observed after cloning and not by direct sequencing, possibly indicating the presence of such variant outside the known AQP2 locus. Surprisingly, the strains from Masi-Manimba were heterozygous for the AQP2 locus. One allele contained the wild-type sequence, but the second allele contained a yet undescribed TbAQP2/3 chimera. Munday and co-workers recently described that the presence of a functional wild-type AQP2 sequence renders strains sensitive to melarsoprol and pentamidine [24]. However, the effect on pentamidine and melarsoprol uptake of the newly described AQP2/3 chimeras is unknown. All variants were cloned in a trypanosomal expression vector for future evaluation and are available upon request. That this probable drug resistant genotype is found in all strains from Mbuji-Mayi and not in the strains from Masi-Manimba could be sufficient to explain the difference in melarsoprol relapse rates observed in East-Kasai (high) and in Bandundu (low). Within this context, it is interesting to note that we observed yet another TbAQP2/3 chimera genotype in two “old” T.b. gambiense type I strains isolated in 1974 in Kinshasa and in Nord Equateur Province. Their AQP2/3 variant is similar, but shorter, than the AQP2/3 mutation described by Baker and co-workers and is therefore possibly capable of reducing melarsoprol and pentamidine uptake [19]. However, in contrast to the locus described by Baker and co-workers, AQP3 seems not preserved in these strains. Both “old” isolates also contain a new variant of TbAT1 with unknown effect on drug uptake. Due to the fact that both AQP2/3 and TbAT1 genes are different, these “old” isolates are probably not closely related to the contemporary strains circulating in Masi-Manimba and Mbuji-Mayi. Appearance of resistance to arsenicals, including melarsoprol, and to pentamidine in several HAT foci in DRC (former Congo Belge and Zaire) has already been described decades ago and was considered a result of mass treatment and chemoprophylaxis [32], [33]. The finding of the pentamidine/melarsoprol resistant genotype in the “old” T.b. gambiense strains may be the basis for molecular studies into the appearance and spread of pentamidine/melarsoprol resistant T.b. gambiense strains. In the present study, we didn't carry out microsatellite analysis to verify the similarity between strains from the Mbuji-Mayi focus, in particular from strains isolated from the same patient before treatment and after relapse. However, results from mobile genetic element PCR (MGE-PCR) as described by Simo et al suggest only very small differences between all strains from Mbuji-Mayi (figure S4) [34]. Thus, most probably, all strains isolated in Mbuji-Mayi are probably the clonal progeny of one strain acquiring the AQP2/3 chimeras. This is the first large scale in vivo drug sensitivity study on T.b. gambiense strains that were isolated within a short period of time, in one single HAT focus, from cured as well as from relapsing patients and for 10 cases from the same patient before and after treatment. As such, the fact that the homogeneity observed within the TbAQ2/3 and the TbAT1 loci does not correspond with the heterogeneity in treatment outcome of the patients and of the mice, suggests that other factors, such as virulence and tissue tropism of the parasite or genotype and phenotype of the individual patient, may influence the treatment outcome. For example, in several independent studies it was found that high cell count in the cerebrospinal fluid (>100 cell/µl) is arisk factor for relapse [8], [23], [35]–[37]. This may indicate that patients in advanced second stage of the disease are less responsive for the standard melarsoprol treatment schedule. It is even not excluded that such patients are also less responsive for other drugs or therapeutic regimes such as NECT, the current first line treatment of second stage gambiense HAT. Therefore, further investigations into treatment failure in HAT and into alternative drugs or treatment regimes should not only focus on differential genotypes of the parasites but also on differential virulence and tissue tropism.
In conclusion, this study confirms that the high melarsoprol relapse rates observed in the Mbuji-Mayi focus can be explained by mutations in the TbAQP2/3 locus of the trypanosomes that circulate in that focus. However, other factors will also influence the treatment outcome of individual patients.
|
10.1371/journal.pntd.0003118 | High Fat Diet Modulates Trypanosoma cruzi Infection Associated Myocarditis | Trypanosoma cruzi, the causative agent of Chagas disease, has high affinity for lipoproteins and adipose tissue. Infection results in myocarditis, fat loss and alterations in lipid homeostasis. This study was aimed at analyzing the effect of high fat diet (HFD) on regulating acute T. cruzi infection-induced myocarditis and to evaluate the effect of HFD on lipid metabolism in adipose tissue and heart during acute T. cruzi infection.
CD1 mice were infected with T. cruzi (Brazil strain) and fed either a regular control diet (RD) or HFD for 35 days following infection. Serum lipid profile, tissue cholesterol levels, blood parasitemia, and tissue parasite load were analyzed to evaluate the effect of diet on infection. MicroPET and MRI analysis were performed to examine the morphological and functional status of the heart during acute infection. qPCR and immunoblot analysis were carried out to analyze the effect of diet on the genes involved in the host lipid metabolism during infection. Oil red O staining of the adipose tissue demonstrated reduced lipolysis in HFD compared to RD fed mice. HFD reduced mortality, parasitemia and cardiac parasite load, but increased parasite load in adipocytes. HFD decreased lipolysis during acute infection. Both qPCR and protein analysis demonstrated alterations in lipid metabolic pathways in adipose tissue and heart in RD fed mice, which were further modulated by HFD. Both microPET and MRI analyses demonstrated changes in infected RD murine hearts which were ameliorated by HFD.
These studies indicate that Chagasic cardiomyopathy is associated with a cardiac lipidpathy and that both cardiac lipotoxicity and adipose tissue play a role in the pathogenesis of Chagas disease. HFD protected mice from T. cruzi infection-induced myocardial damage most likely due to the effects of HFD on both adipogenesis and T. cruzi infection-induced cardiac lipidopathy.
| Infection with Trypanosoma cruzi, the etiologic agent in Chagas disease, may result in heart disease. There has been an increase in obesity, diabetes, hypertension and ischemic cardiovascular disease in endemic areas. Previously, we demonstrated that adipose tissue is an early target and a reservoir for T. cruzi. T. cruzi has high affinity for lipoproteins, and that infected tissues there is an increase in intra-cellular cholesterol levels. It is likely that adipocytes and lipoproteins play a key role in the pathogenesis of Chagas disease. The role of host lipids in the pathogenesis of Chagas disease is understudied. Diet plays a major role in the regulation of systemic and whole body lipid levels including adipogenesis and lipogenesis. We report, for the first time, the effect of diet on myocardial inflammation and damage observed during acute T. cruzi infection and provide data on the role of parasite associated LDL/HDL in the regulation of systemic lipid homeostasis in white adipose tissue (WAT) and in the heart. Interestingly, we demonstrate that a high fat diet protects mice from the consequences of infection-induced myocardial damage through effects on adipogenesis in adipose tissue and reduced cardiac lipidopathy.
| Chagas disease, caused by the parasite Trypanosoma cruzi, is classified by WHO as a neglected tropical disease and is a major cause of morbidity and mortality in Latin America [1]. Globalization has led to the increased recognition of this infection among immigrants from Latin America in non-endemic countries [2]. It has been estimated that 18 to 20 million people have Chagas disease [2]. Symptoms of infection are varied, but include heart disease and megasyndromes of the gastrointestinal tract. Chagas disease has acute, indeterminate and chronic phases. Acute systemic infection is often asymptomatic, but, in those that are symptomatic the disease is characterized by myocarditis and/or meningoencephalitis [1]–[3]. T. cruzi infection causes an intense systemic pro-inflammatory response in many organs including the heart. Following infection the majority of patients develop an asymptomatic latent infection termed the indeterminate (or latent) stage of infection. As many as thirty percent of infected individuals may progress to chronic disease characterized by cardiomyopathy and/or mega syndromes. Myocardial dysfunction is associated with extensive remodeling caused by the initial infection and ensuing fibrosis [4].
The low density lipoprotein receptor (LDLr) is involved in LDL internalization and regulation of cholesterol homeostasis [5]. We have demonstrated that T. cruzi utilizes LDLr to invade host cells and that LDLr likely plays an important role in the pathogenesis of Chagas disease [6]. T. cruzi has high affinity for LDL and HDL and the rate of invasion increases in the presence of lipoproteins [7]. T. cruzi primarily targets lipid rich adipose tissue as their reservoir and causes lipolysis during acute infection [8], [9]. Altered serum triglyceride and cholesterol levels are associated with acute infection [6].
The role of host lipids in the pathogenesis of Chagas disease is understudied. Diet plays a major role in the regulation of systemic and whole body lipid levels including adipogenesis and lipogenesis [10]. Recent changes in diet, underlying the well-recognized obesity epidemic, in regions endemic for Chagas Disease are likely to have significant effects on the interaction of this parasite with its human host. Herein, we report, for the first time, the effect of diet on myocardial inflammation and damage seen during acute T. cruzi infection. We also provide data on the role of parasite associated LDL/HDL in the regulation of systemic lipid homeostasis in white adipose tissue (WAT) and in the heart.
All animal experimental protocols were approved by the Institutional Animal Care and Use Committees (IACUC) of Albert Einstein College of Medicine (No. 20130202) which is adhered to the National Research Council guidelines (Guide for the Care and Use of Laboratory Animals: Eight Edition, Washington, DC: The National Academies Press, 2011).
The Brazil strain of T. cruzi was maintained by passage in C3H/Hej mice (Jackson Laboratories, Bar Harbor, ME). Male CD-1 mice (Jackson Laboratories) were infected intraperitoneally (i.p.) at 8–10 weeks of age with 5×104 trypomastigotes of the Brazil strain [8]. Mice were maintained on a 12-hour light/dark cycle. Mice, starting at the day of infection, were randomly divided into two groups (n = 15 per group) and fed on either high fat diet (HFD; 60% fat) or Regular diet (RD, 10% fat) (D12492 or D12450 Research Diets, Inc., New Brunswick, NJ). Uninfected mice were fed on either HFD (n = 15) or RD (n = 15) and used as respective controls in all the experiments. For each replication of this experiment the same numbers of mice were used in all groups.
Plasma samples were obtained from 75 µl of blood collected from the orbital venous sinus (using isoflurane anesthesia) at 10, 15, 20, 25 and 30 days post infection (dpi). Parasitemia was evaluated by counting in a Neubauer hemocytometer as described previously [8]. Thirty five days after infection the mice were euthanized and heart and epididymal white adipose tissues (WAT) were harvested for analysis. At this time-point there was no peripheral parasitemia and mice appeared normal.
Colorimetric assays were performed using colorimetric assay kit for non-esterified fatty acid (NEFA from Cell Biolabs, Inc. CA), triglyceride (TG from Cayman Chemicals, MI), and low density lipoprotein (LDL), high density lipoprotein (HDL) and total cholesterol (TC) (Enzychrom (E2HL-100), Bioassay system, CA) in serum samples6. Cholesterol levels were quantified in the hearts and WATs of mice at d35pi using a colorimetric assay kit and samples were prepared and assayed following manufacturer's protocol (Total cholesterol colorimetric assay kit, Cell Biolabs Inc., CA). The lipid content of frozen WAT was quantified using Oil Red O staining. Frozen sections stained with Oil Red O were washed 3 times in isopropyl alcohol for 3 minutes each. Bound Oil Red O was eluted by incubating in isopropyl alcohol (5 mL) for 16 hours; the eluted lipid content stained with Oil Red O was then measured at 540 nm in a spectrophotometer (Shimadzu uv-1201) as previously reported [9].
Tissue lysates were prepared as previously described [6]. An aliquot of each sample (40 µg protein) was subjected to a 4–15% gradient SDS-PAGE and the proteins transferred to nitrocellulose filters for immunoblot analysis. LDLr specific rabbit monoclonal antibody (1∶1000 dilution, AB52818 Abcam, Cambridge, MA), lipoprotein lipase (LPL) specific mouse monoclonal antibody (1∶1000 dilution, AB21356, Abcam), or ABCA1 specific rabbit monoclonal antibody (1∶1000 dilution, AB18180, Abcam) were used as primary antisera. Horseradish peroxidase-conjugated goat anti-mouse immunoglobulin (1∶2000 dilution, Amersham Biosciences, Piscataway, NJ) or horseradish peroxidase- conjugated goat anti-rabbit immunoglobulin (1∶5000 dilution, Amersham Biosciences) were used to detect specific protein bands (explained in Figure Legends) using a chemiluminescence system [6]. GDI (1∶10000 dilution, 71-0300, and rabbit polyclonal, Invitrogen, CA) and a secondary antibody horseradish peroxidase conjugated goat anti-rabbit (1∶2000 dilution, Amersham Biosciences) were used to normalize protein loading. Rabbit Adiponectin antibody was produced in the laboratory of Dr Scherer as described previously [8].
Heart and white adipose tissue were collected from mice on 35 days post-infection and stored at −80°C. A quantitative real-time polymerase chain reaction (qPCR) was used to quantify parasite load employing PCR SYBR Green Master Mix (Roche Applied Science, CT) containing MgCl2 employing an iQ5 LightCycler (Bio-Rad). Isolation of DNA, preparation of standard curves for host and epimastigote DNA, and qPCR analysis was performed as previously published [8]. Host 18srRNA gene was used for normalization [18S forward: 5′-AGGGTTCGATTCCCGGAGAGG-3′, reverse, 5′-CAACTTTAATATACGCTATTGG-3′].
An RT2 Profiler (SA Biosciences, Valencia, CA) custom designed PCR array for mouse genes involved in LDL internalization, cholesterol metabolism, fatty acid and triglyceride metabolism, glucose metabolism and inflammatory signaling was used to analyze gene expression. Data analysis was performed normalized to the expression of 18sRNA using the ΔΔCT method according to the manufacturer's protocol (SABiosciences) and statistical analysis was performed as suggested [9].
Freshly isolated tissues were fixed with phosphate-buffered formalin overnight and then embedded in paraffin wax (n = 5). Hematoxylin and eosin (H&E) staining was performed and the images were captured as previously published [8]. Four to six sections of each heart were scored blindly. For each myocardial sample, histologic evidence of myocarditis and inflammation was classified in terms of degree of degenerating cardiac muscle fibres, inflammation, fibrosis and adipocyte presence and was graded on a five point scale ranging from 0 to 4+. A zero score indicated lowest or negligible changes and 4 the most damaged state. IFA was performed on the frozen sections using anti-LDL and the images were captured as previously published [6]. The fluorescent intensities of the images were quantified using NIH-Image J program for four to six images of each heart.
All mice were imaged after 3 hours of fasting. Mice were administered 300–400 uCi (12–15 MBq) in 0.1 mL normal saline, [18F] fluoro-2-deoxyglucose (FDG), via tail vein and imaging was started at 1 hour after injection. This period permits the tracer to be delivered throughout the body and trapped by the glycolytic pathway. Prior to FDG administration the mice were anesthetized with 1.5% isoflurane-oxygen mixture, which continued throughout the imaging portion of the procedure. After MicroPET imaging, the animals were housed in the imaging facility for ten half lives (18F has a half-life of 110 minutes) until they could be safely moved back to the Animal Institute for continued housing.
The mice were imaged by an Inveon Multimodality scanner (Siemens, Knoxville, TN) using its PET module. PET imaging is performed using the PET gantry, which provides a 12.7 cm axial and 10 cm transaxial active field of view. The PET scanner has no septa, and acquisitions are performed in the 3D list mode. A reconstructed full-width-half-max resolution of <1.4 mm is achievable in the center of the axial field of view. List mode acquisition of data is performed to permit dynamic re-framing for kinetic evaluation of the radiotracer uptake, where indicated. After each acquisition, data were sorted into 3D sinograms, and images were reconstructed using a two dimensional (2D)-Ordered Subset Expectation Maximization algorithm. Data were corrected for decay, dead time counting losses, random coincidences and the measured nonuniformity of detector response (i.e. normalized), but not for attenuation or scatter. Analysis was performed by using the Inveon Research Workplace 4.1 software (Siemens). All imaging studies were inspected visually in a rotating 3D projection display to identify interpretability and image artifacts. Regions of interest (ROI) were manually defined. Successive scrolling through 2D slices (each 1.2 mm thick in the axial images) permitted measurement of radioactivity within defined volumes. Corrected counts per cc within this volume divided by the counts per gram of total body mass of injected radioactivity determined the SUV. SUVmax, the maximum value of SUV within the heart was determined. The SUVmax is the maximum value of the percent-age injected dose per gram of cardiac tissue multiplied by the body weight of each animal. The SUVmax has been validated in numerous animal and human models as a reproducible and robust measure of radioactivity in longitudinal studies.
Cardiac gated MRI was performed on uninfected and infected mice at 26 dpi were imaged using a 9.4 T Varian Direct Drive animal magnetic resonance imaging and spectroscopic system (Agilent Technologies, Inc. Santa Clara, CA) as previously published [10]. Briefly, anesthesia was induced with 2% isoflurane in air, mice were positioned supine inside an MR compatible holder and positioned within a 35-mm ID quadrature 1H volume coil (Molecules2Man Imaging Co., Cleveland, OH). Body temperature was maintained at 34∼35°C using warm air with feedback from a body surface thermocouple. A respiratory sensor balloon was taped onto the abdomen. Cardiac (ECG electrodes inserted subcutaneously in front left paw and rear right paw) and respiratory signal (from sensor balloon taped to the abdomen) were continuously monitored and used for MR gating/triggering by an SA Monitoring and Gating System (Small Animal Instruments, Inc., Stony Brook, NY). Ten to fourteen 1-mm-thick slices without gap was acquired in short-axis orientation covering the entire heart using an ECG-triggered and respiratory gated multi-frame tagged cine sequence. The imaging parameters used were field of view (FOV) of 40×40 mm2, matrix size of 256×256, TE of 2.6 ms, TR of 5.5 ms, flip angle of 25°, number of averages of 2. The number of frames was twelve to eighteen. Data were transferred to a PC and analyzed using MATLAB-based software. Left ventricle (LV) and right ventricle (RV) dimensions in millimeters were determined from the images representing end-diastole. The left ventricular wall is the average of the anterior, posterior, lateral, and septal walls. The right ventricular internal dimension is the widest point of the right ventricular cavity.
Immunoblot, immunofluorescence and quantification of parasite load studies were performed at least three times and representative data are presented in the figures. Lipid profile analysis and gene arrays were done in duplicates. Data were pooled and statistical analysis was performed using a Student's t-test (Microsoft Excel) as appropriate and significance of difference was determined as p values between <0.05 and <0.005.
Cxcl16 NM_023158.6, Stab1 NM_138672, Vldlr NM_013703, Lrp6 NM_008514, Ldlr NM_010700, Scarf1NM_001004157, Apoa1NM_009692, Apob NM_009693, Apoe NM_009696, Acaa1aNM_130864,
Acad9 NM_172678, Acad10NM_028037, Acox1NM_015729, Fabp1NM_017399, Acsbg1NM_053178,
Lipe NM_010719, Npc1NM_008720, Lcat NM_008490, Abca1NM_013454, Abcg1NM_009593, Cyp39a1NM_018887, Cyp7a1NM_007824, Hmgcr NM_008255, Insig1NM_153526, Lep NM_008493,
Ppara NM_011144, Pparg NM_011146, Adig NM_145635, Adipoq NM_009605
To investigate the impact of diet on the course of acute T. cruzi infection we initially studied CD-1 mice that were placed on either a high fat diet (HFD (20 kcal% protein, 20 kcal% carbohydrate and 60 kcal% fat)) or regular diet (RD (20 kcal% protein, 70 kcal% carbohydrate and 10 kcal% fat)) at the time of infection (i.e. on the first day of infection). HFD fat content is composed of saturated (81.5 g), monounsaturated (91.5 g) and polyunsaturated (81.5 g) fat and RD fat content is composed of saturated (9.9 g), monounsaturated (13 g) and polyunsaturated (20.7 g) fat. No significant differences in the body weight were observed between the groups at the start of infection. As time progressed, the infected RD fed mice displayed fat loss (60%), as quantified by oil red O staining [9] compared to uninfected mice and this was associated with edema on 35 days post infection (dpi) as previously described [8]. Analysis suggested that infected RD fed mice gained weight due to edema and the infected HFD fed mice mainly due to fat. In contrast, infected HFD fed mice displayed no signs of edema and had only a 20% fat loss compared to uninfected HFD fed mice. No significant difference was observed in the food take between RD-fed or HFD-fed mice during infection.
Peak parasitemia was observed between 15 and 28 dpi in all groups. There was a twofold decrease in parasitemia in HFD fed mice compared to RD fed mice (Fig. 1a). HFD fed mice had a significantly increased survival rate (85%) compared to RD fed mice (40%) during acute infection (Fig. 1b). In addition, a higher parasite load was seen in the myocardium of RD fed mice as determined by qPCR (Fig. 1c). However increased parasite loads were detected in white adipose tissue (WAT) of HFD compared to RD fed mice (Fig. 1c).
We next investigated the effect of T. cruzi infection on the rate of mortality and parasitemia in mice that were placed on a HFD or RD for 30 days before infection (i.e. diet pre-fed mice) and were then continued on the respective diets for 35 dpi. HFD pre-fed mice had a 92% survival compared with a 40% survival of the RD pre fed mice due to acute infection. These data demonstrate that HFD has a protective effect on the mortality seen during acute T. cruzi infection in this murine model.
Similar to what was seen in mice started on HFD and RD at the time of infection, we also observed a significant decrease in body weight with HFD pre fed mice (86%) and an increase in body weight with RD pre-fed mice (112%) at 35 dpi compared to their respective uninfected mice (100%). This appeared to be due to edema which was seen in infected RD pre-fed mice, but which did not occur in infected HFD pre-fed mice. However, uninfected HFD mice tended toward obesity (25% greater body weight) compared to uninfected RD fed mice at d35pi.
There were no significant histological differences in the hearts obtained from HFD and RD fed uninfected mice (Fig. 2a). In contrast, the hearts obtained from infected RD fed mice displayed an intense inflammatory reaction most prominent at the right ventricle and left ventricle junction associated with vasculitis and fibrosis. Hearts obtained from infected mice fed a HFD had fewer parasites and a reduction in inflammation and fibrosis (Fig. 2a). Histological scoring of hearts ranged from 0 to 4+ in each of the categories of degenerating cardiac fibers, inflammation, fibrosis and presence of adipocytes. Infected RD animals had higher scores than infected HD fed animals (figure 2b). Summing all scores (maximum cumulative score 16) using this method, the hearts from infected RD fed scored 13 and the hearts from HFD fed infected mice 5.2.
Lipolysis and adipolysis are observed during acute T. cruzi infection [8], [9]. Adipogenesis is regulated by adipokines such as adiponectin, leptin, PPAR-γ, and TNF-α [11], [12]. These adipokines contribute to the regulation of fatty acid oxidation. The protein encoded by the gene Adig (adipogenin) has been implicated in adipocyte differentiation and fat accumulation.
Examination of the cellular morphology of WAT of infected RD- and HFD-fed mice demonstrated that WAT from the uninfected HFD-fed mice displayed enlarged (280±25 µm) and lipid enriched adipocytes compared to WAT of uninfected RD-fed mice (170±60 µm (Fig. 2b, top). WAT from both the infected RD- and HFD-fed mice demonstrated enlarged adipocytes (2.0–2.5 fold enlarged) surrounded by adipocytes with smaller lipid droplets. The average size of inflamed adipocytes of HFD fed mice was 500±150 µm and RD fed mice 400±125 µm during infection. HFD-fed mice had lipid droplets that were significantly smaller compared to RD-fed mice (70±20 µm and 150±30 µm respectively,). There was an increase in the number of dead cells in the WAT of RD-fed mice (Fig. 2c). The number of dead cells was 680±72 and 206±40 in RD fed and HFD fed infected WAT respectively surrounding inflamed adipocytes (for n = 200 adipocytes). The fat loss in WAT was determined using Oil red O staining [9]. RD-fed infected mice displayed 60% fat loss and HFD-fed infected mice 20% compared to their respective uninfected control mice (p<0.05). qPCR demonstrated a significant decrease in the mRNA levels of the genes involved in adipogenesis such as adiponectin, (−11.5 fold), Peroxisome proliferator activated receptor gamma (PPAR-γ, −6.0), Adipogenin (Adig, −3.5 fold) and leptin (−55 fold) in WAT of RD-fed mice compared to HFD-fed mice during acute (d35pi) infection (Table 1). There was an upregulation of adipogenic genes, except for PPAR-γ, in the heart of infected HFD-fed mice compared with RD-fed mice. These data suggest that HFD increases adipogenesis during infection and that this alteration affects the course of acute T. cruzi infection. It is likely that adipogenesis or lipolysis directly affects serum lipid homeostasis.
T. cruzi has a high affinity for LDL and HDL, and the rate of invasion increases in presence of lipoproteins [7]. We have reported that there is a decrease in serum TG and TC during acute infection in mice fed with a standard chow diet (17% fat) [6]. To analyze whether the HFD would improve the serum lipid levels during infection, we measured TG, FFA and TC levels in HFD fed mice and compared it with RD fed mice at different time points of infection. Serum levels of TG, FFA and TC were significantly decreased in T. cruzi infected mice irrespective of the type of the diet fed (Fig. 3). Serum TG content of uninfected RD fed mice was higher (38%) than uninfected HFD fed mice; however the TC was significantly higher (25%) in the uninfected HFD fed mice compared to uninfected RD fed mice. Such alterations in the serum lipid level could potentially regulate myocardial lipid levels and the associated inflammation.
The rate of T. cruzi invasion has been shown to depend on serum cholesterol levels [7]. Invasion results in elevated intracellular/tissue cholesterol levels (Fig. 3d) which could affect triglyceride and cholesterol metabolism (Table 2). Analysis demonstrated a significant increase in the cholesterol levels in the hearts (260%) and the WATs (394%) of the infected RD fed mice compared to uninfected mice (100%). Though infection increased cholesterol levels in the HFD-fed infected mice, this analysis demonstrated a significant decrease in their level in both the hearts (202%) and WAT (211%) compared to uninfected RD-fed control mice. IFA of the myocardium using anti-LDL displayed decreased accumulation of LDL (30%) in infected HFD-fed mice compared to infected RD-fed mice on d35 pi (Fig. 4).
qPCR analysis demonstrated increased mRNA levels of the inflammatory markers such as TNF-α (5 fold) and IFN-γ (40 fold) in the WAT of HFD-fed mice at d35pi compared to their respective control groups. WAT of RD-fed mice had mostly necrosed cells and showed 2 fold increase in TNF-α and 9.0 fold increase in IFN-γ which was significantly different than HFD-fed mice. The myocardium of the infected RD-fed mice demonstrated a significant increase in TNF-α (92 fold compared to uninfected) whereas, infected HFD-fed mice showed only 28 fold increase in TNF-α. This data supports our histological observations (Fig. 2). The mRNA level of IFN-γ was higher in the myocardium of both the infected HFD and RD-fed mice (185 and 195 fold respectively, compared to their respective controls). There was no statistical difference between the mRNA level of IFN-γ seen in infected HFD and RD fed mice.
Inflammatory cells use glucose as a primary source of metabolic energy, and thus increased uptake of glucose and high rates of glycolysis are characteristics of inflammatory cell infiltration into tissues. Using microPET technology, the metabolic state of the myocardium was assessed by determining the regional uptake of the glucose analogue, 18F-FDG. The mean value of the myocardial SUVmax was used to compare the microPET data between the uninfected age-matched controls and infected groups fed on either HFD or RD 30 dpi. Tracer uptake was significantly higher in the myocardium of the RD-fed (14.5±2.2) infected mice compared with infected HFD-fed mice (5.5±1.03) and compared to uninfected control mice (Fig. 5 a). We believe this increased tracer uptake indicates increased inflammation and myocarditis in the RD-fed infected mice [13].
MRI examination of the heart during both diastole and systole revealed a significant decrease in the left ventricle internal diameter (LVID) and an increase in the wall thickening of the ventricles during acute infection compared with uninfected mice (Fig. 5b) (Table 3). The alterations in heart morphology were less pronounced in the infected mice on the HFD compared to those on the RD. The right ventricle internal diameter (RVID) was increased in infected mice and no significant difference was observed between RD- and HFD- fed mice (Table 3). During the acute stage of infection the percent fractional shortening was significantly increased in the infected mice fed RD and likely reflects hypertrophic cardiomyopathy and possibly hypokinetic motion or dyskinesis of the heart wall [14].
Our results demonstrate an alteration in the whole body and systemic lipid homeostasis during acute infection. We further investigated the role of diet on the expression levels of scavenger receptors, lipoproteins, and the proteins involved in lipid metabolism both by qPCR and Immunoblot analysis.
T. cruzi utilizes the host LDLreceptor to invade cells and invasion upregulates LDLr levels [6], [7]. Analysis of the mRNA levels of other LDL receptors (scavenger receptors), such as stabilin 1 (Stab 1), and scavenger receptor class F member 1 (Scarf 1) in the heart and WAT of both the HFD and RD fed infected mice demonstrated a significant increase compared to their respective uninfected control mice (Table 2). The expression of the oxidized LDL receptor and the chemokine (C-X-C motif) ligand 16 (Cxcl16) was significantly upregulated in WAT of infected mice (Table 2). The fold increase in the expression of Cxcl16 and Stab1 in WAT was 10-fold and 60-fold respectively, in RD-fed infected mice compared with HFD-fed infected mice. Infected heart tissue displayed a significant increase in LDLr and Scarf1 mRNA levels in both RD- and HFD-fed mice, however, a significant increase in the expression of Stab1 was observed only in RD-fed mice (19 fold increase) (Table 2).
Apolipoproteins play a major role in lipid metabolism and cholesterol homeostasis. We measured the mRNA levels of apolipoproteins such as ApoA, ApoB and ApoE in the heart and WATs of mice fed a HFD and RD during infection. Analysis of the mRNA levels demonstrated a significant difference in the expression levels of these genes in WAT and heart in HFD and RD fed mice (Table 2). RD-fed mice down regulated Apo A1 (−165 fold) and Apo B (−54 fold) expression in WAT and had increased expression of ApoA1 and ApoB in the heart (2 and 23-fold respectively) compared to their respective uninfected mice. In these RD-fed mice Apo E was up regulated both in WAT and heart tissue. Infected HFD mice had increased Apo A1 (5-fold) and Apo B1 (7-fold) in the WAT and increased levels of Apo A1, Apo B1 and Apo E in heart tissue. The expression levels of Apo A1, Apo B1 and Apo E were significantly lower than that seen in RD-fed mice. It is likely that changes in apolioprotein levels affect systemic cholesterol homeostasis in these RD- and HFD-fed mice.
Fatty acid (FA) transport and β-oxidation are important in the functioning of WAT and the heart. The genes responsible for FA transport, TG metabolism and β-oxidation such as acetyl-Coenzyme A acyltransferase 1A (Acaa1a), mitochondrial acyl-CoA dehydrogenase family members (Acad 9 &10), acyl-CoA oxidase (Acox1), glycerol-3-phosphate dehydrogenase (Gpd1), FA acid binding protein (Fabp1) and acyl-CoA synthetase bubblegum family member 1 (Acsbg1) were analyzed (Table 2). WAT from the infected RD-fed mice demonstrated a significant fold decrease in the expression of these genes; however infected HFD-mice showed no significant change in the majority of these genes (the only exceptions being Acad9 (2 fold) and Acsbg1 (3 fold) where there was increased expression compared to uninfected control mice). In the hearts of both RD and HFD fed infected mice the levels of these genes were significantly increased especially in HFD compared to control mice (Table 2).
The Niemann–Pick disease type C1 (Npc1) gene encodes a large protein that resides in the limiting membrane of endosomes and lysosomes and mediates intracellular cholesterol trafficking via binding of cholesterol to its N-terminal domain [15]. Analysis of the mRNA levels demonstrated an elevated expression of Npc1 mRNA (3-fold) in the WAT of both the RD- and HFD-fed infected mice. No significant difference, however, was seen in Npc1 in the hearts of these mice. The gene lecithin-cholesterol acyltransferase (LCat) encodes the extracellular cholesterol esterifying enzyme Lcat, which is required for cholesterol esterification [14] is down regulated in infected RD-fed mice (7 fold) compared to the infected HFD-fed mice (Table 2). The esterification of cholesterol is required for cholesterol transport [16]. We investigated the expression of genes involved in cholesterol efflux like ATP binding cassette (Abc) transporters such as Abca1 and Abcg1 in WAT and heart [17]. With cholesterol as their substrate, these proteins function as cholesterol efflux pumps in the cellular lipid removal pathway. Abca1 is significantly increased in both the WAT and heart of infected mice (Table 2). Abcg1 is mainly associated with macrophage cholesterol efflux. Abcg1 is significantly upregulated (1270 fold) in the WAT of infected RD fed mice compared with infected HFD fed mice which suggests that increased macrophage infiltration is associated with RD-fed mice. We have published that there is an increased influx of macrophages in WAT during acute infection [8], [9]. Abcg1 is significantly higher in the hearts of both HFD- and RD-fed infected mice (Table 2). Intracellular cholesterol is mainly converted to bile acids in liver. Cytochrome p450 monooxygenases like Cyp39a1 and Cyp7a1 were up regulated (3- and 5-fold respectively) in WAT of HFD-fed infected mice compared with infected RD-fed mice (Table 2). Cyp39a1 and Cyp7a1are endoplasmic reticulum proteins involved in the conversion of cholesterol to bile acids [18], [19]. Cyp7a1 is the rate limiting enzyme in the primary pathway of bile acid synthesis [19]. Even though adipocytes are not a classical bile acid synthesizing tissue, it has been shown that farnesoid x-receptor (FXR) a nuclear receptor which is involved in bile acid synthesis is expressed in adipose tissue during metabolic dysfunction [20]. The mRNA levels of HMG co A reductase (Hmgcr), a rate limiting enzyme in the cholesterol biosynthesis is upregulated in both the WAT and the hearts of both RD-and HFD-fed mice during acute infection (Table 2).
Immunoblot analysis demonstrated an upregulation of LDLr in infected mice. LDLr levels in the hearts of infected RD-fed mice were significantly higher compared to that of infected HFD-mice (Fig. 6). In WAT of infected mice the reverse was seen (Fig. 7). We also analyzed the expression of LOX1 (oxidized LDLr) in heart tissue and found results similar to that seen with LDLr in RD and HFD fed infected mice (Fig. 6). Hearts from infected mice displayed increased macrophage infiltration (probed with anti-F4/80), lipoprotein lipase activity (anti-LPL) and cholesterol efflux levels (ant-Abca1) (Fig. 6). However, the expression levels of F4/80 (−2 fold), LPL (−2.5 fold) and Abca1 (−0.7 fold) were lower in the infected HFD mice compared with the infected RD mice.
The enzyme involved in the rate limiting step of cholesterol biosynthesis HMGCR is significantly increased (400-fold) in the hearts of infected mice. In the hearts of uninfected HFD-fed mice there was an increase in HMGCR compared to uninfected RD- fed mice. This suggests that HFD induces HMGCR but that infection further increases the expression of HMGCR in heart tissue (Fig. 6). Interestingly, infected heart tissue had increased amount of adiponectin multimers, but decreased monomers (Fig. 8), whereas we observed no multimers in infected WAT. Also the amount of adiponectin monomers were decreased in infected WAT (Fig. 7 and Fig. 8). Uninfected HFD fed mice heart tissue had higher adiponectin levels compared to uninfected RD fed mice hearts which was similar to what was seen in WAT from these mice (Fig. 7 and Fig. 8).
T. cruzi has a high affinity for lipoproteins/cholesterol [7] and depends on host lipid molecules for invasion and survival [6], [7], [21]. Adipose tissue is the largest endocrine organ in the body and is a rich source of lipids and is involved in energy homeostasis. Previously we reported that adipose tissue is an early target of T. cruzi infection and serves as a reservoir for parasites [8], [9]. T. cruzi infection-induced lipolysis is a hallmark of acute infection [8], [9] and lipolysis is known to alter lipid homeostasis. Diet plays a major role in adipogenesis and in lipid homeostasis. In the present study, we systematically analyzed the impact of high fat (HFD, 60% fat) and regular diet (RD, 10% fat) on an acute model of Chagas disease and demonstrated a link between diet, adipogenesis and myocarditis. HFD increased adipogenesity and reduced lipolysis which affected peripheral parasitemia and parasite load in the heart. Even though a significant difference in the parasite load of heart was demonstrated between the RD- and HFD- fed infected mice, the adipose tissue of HFD fed mice had a significantly higher number of parasites. We believe the increased fat tissue in HFD mice resulted in a sequestration of parasites (i.e. a sponging effect) that may have led to a reduction in the parasite load in the heart. Consistent with this hypothesis, HFD-fed mice displayed an increased survival rate (95%) with diminished myocardial damage during acute infection compared with RD-fed mice. HFD induced a modest obese condition (as seen with uninfected mice) may bring metabolic changes during infection. Overall, this increase in survival with HFD supports the “so-called” obesity paradox hypothesis [21].
The various techniques we employed to investigate cardiac structure and function clearly demonstrated that in HFD-fed mice there was a significant amelioration of myocardial dysfunction compared to RD-fed mice. This is likely due to the role of adipogenesis and lipolysis during infection. Lipolysis is a characteristic marker of acute infection in mice where there is a significant decrease in total body fat between d10 and d35d pi. HFD-fed mice increased adipogenesis and reduced the rate of adipocyte lipolysis compared to RD-fed mice. The parasite load in the WAT was higher in these HFD mice compared to RD. The pro-inflammatory marker TNF-α was higher in the WAT of HFD-fed mice at this time point of infection. Adipocyte lipolysis was significantly higher in the infected RD-fed mice as indicated by serum triglyceride and fatty acid levels. Overall, this change in fatty acid metabolism probably contributes to increased parasitemia in RD-fed mice and a higher parasite load in heart.
Serum cholesterol levels were reduced with a concomitant increase in intra-organelle LDL/cholesterol levels (Fig. 3). We have demonstrated the accumulation of LDL/cholesterol in adipose tissue and the hearts [22] during infection. Accumulated lipids need to be degraded (through lipases) before undergoing further degradation through the β-oxidation pathway. Increased lipolysis activation through LPL was observed both in the hearts and WAT of infected mice. There was a significant increase in the LPL expression in the hearts of RD-fed mice compared with HFD-fed mice suggesting that more lipids accumulated in RD-fed mice due to increased lipolysis and increased parasite load. Hormone sensitive lipase (Lipe) is significantly down regulated only in the WAT of RD-fed mice. No significant change was observed in the hearts of either RD- or HFD-fed mice at d35pi.
FA transport and β-oxidation are important signaling pathways in the functioning of WAT and heart. White adipocytes are not mitochondrial rich cells, unlike brown adipocytes which are mitochondrial rich, and the elevation of infection induced lipids may cause a burden on mitochondrial oxidative capacity in WAT leading to necrosis. qPCR demonstrated a down-regulation of the mRNA levels of many of the genes involved in triglyceride and FA metabolism which reflects the oxidative state in WAT of RD-fed mice. Previously, we reported a significant loss of WAT during acute infection [8], [9]. The heart, however, could sustain this load as it is rich in mitochondria and we observed increased FA and triglyceride metabolism both in the RD- and HFD-fed mice. Interestingly, it has been recently reported that host FA metabolism is essential for the persistence of T. cruzi amastigotes [23]. Decreased expression of Insig1, a regulator of cholesterol biosynthesis through SREBP in both the hearts and WAT of RD fed mice is also responsible for an increased HMGCR expression and cholesterol biosynthesis.
De novo cholesterol biosynthesis is highly regulated and depends on intracellular cholesterol levels [24]. When there is a depletion in intracellular cholesterol, cells respond with a SREBP-activated increase in HMGCR and LDLr expression which results in endogenous cholesterol biosynthesis and LDL-mediated uptake of cholesterol. In the present study, we found increased SREBP levels and increased HMGCR and LDLr protein levels in the tissues when the cells already had elevated intracellular LDL/cholesterol levels due to parasite invasion suggesting that infection results in dysfunctional cholesterol homeostasis in tissues and changes in the regulation of these key cholesterol homeostasis genes.
Expression of the ABC transporters is highly upregulated during macrophage differentiation and cholesterol loading, and they synergize to mediate cholesterol transport to Apo-A1 [25]. The expression of these genes was significantly altered during infection.
Adiponectin, an adipokine secreted by the WAT, plays an important role in regulating glucose and lipid metabolism and controlling energy homeostasis in insulin-sensitive tissues [26]. Adiponectin is considerably reduced in the WAT, but, importantly produced in the heart during acute infection. Previously, we demonstrated a reduction in serum adiponectin levels at 15 days post infection (dpi) returning to normal by 30 dpi [8], [9]. WAT (fat cells) significantly decreases during infection and thus, the amount of WAT in infection may not be sufficient to maintain normal serum adiponectin. As the adiponectin multimers are functionally active, further studies are required to confirm that the heart secretes adiponectin during T. cruzi infection, the mechanism by which this occurs and the significance of elevated adiponectin levels in the heart during acute infection.
Lipogenic and adipogenic markers such as leptin, adiponectin, adipogenin and PPAR are greatly reduced in the WAT of RD-fed mice compared to HFD-fed mice, but significantly increased in the heart especially in the HFD-fed mice. This suggests that there are tissue specific of acute T. cruzi infection related to the altered lipolysis and lipogenic status of different tissues. Previously, we demonstrated the role of host LDLr in T. cruzi invasion [6]. qPCR analysis revealed (Table 2) the association of other LDL receptors and modified LDL receptors such as classical LDLr, very low density lipoprotein receptor (VLDLr), STAB1, CXCL16 and SCARF1 in the hearts and WAT during acute infection. The significant increase and the decrease in the LDLr protein levels in the WAT and heart respectively, of the HFD-fed infected mice compared with RD-fed mice, suggests that WAT of HFD-fed mice and the hearts of RD-fed mice are targets of T. cruzi during the late phase of the acute infection. Thus, an increased parasite load in the hearts of RD-fed mice resulted in elevated cardiac LDL/cholesterol levels, macrophage infiltration (F4/80 staining) and inflammation (TNF-α) compared with HFD-fed mice. Overall, it is clear that HFD modulates the effect of T. cruzi infection on myocarditis and mortality in this acute model and this was confirmed by microPET and MRI analyses.
HFD is known to alter the metabolic state of the host leading to diabetes and obesity. HFD induces the metabolic syndrome and this alteration in the host affects the pathogenesis of acute Chagas disease resulting in a decreased heart parasite burden and an increased survival rate. This observation in the mouse model is consistent with the “obesity paradox” that has been described for some chronic infections reflecting an observation that obesity can have a positive effect on disease outcome. To this end, the metabolic syndrome may have evolved as a response to “times of plenty” when the extra calories could be used for metabolic changes that would allows the host to better deal with chronic infectious diseases such as T. cruzi [21]. Herein, we report, for the first time on alterations in the lipid signaling net work due to diet, adipogenesis, and lipolysis in the setting of acute T. cruzi infection. We believe that these alterations contribute to the pathogenesis of acute Chagas disease. The rate of survival and the severity of the myocardial damage are related to the adipogenic and the lipolytic status of adipose tissue and the heart. Lipid and cholesterol homeostasis is completely altered by the infection which warrants further mechanistic studies to understand the pathogenic role of LDL/cholesterol in the progression of Chagasic cardiomyopathy which can now be considered, in part, to be a lipidopathy (onset of cardiomyopathy due to abnormal intracellular lipid level).
|
10.1371/journal.pgen.1002792 | Induction of Cytoprotective Pathways Is Central to the Extension of Lifespan Conferred by Multiple Longevity Pathways | Many genetic and physiological treatments that extend lifespan also confer resistance to a variety of stressors, suggesting that cytoprotective mechanisms underpin the regulation of longevity. It has not been established, however, whether the induction of cytoprotective pathways is essential for lifespan extension or merely correlated. Using a panel of GFP-fused stress response genes, we identified the suites of cytoprotective pathways upregulated by 160 gene inactivations known to increase Caenorhabditis elegans longevity, including the mitochondrial UPR (hsp-6, hsp-60), the ER UPR (hsp-4), ROS response (sod-3, gst-4), and xenobiotic detoxification (gst-4). We then screened for other gene inactivations that disrupt the induction of these responses by xenobiotic or genetic triggers, identifying 29 gene inactivations required for cytoprotective gene expression. If cytoprotective responses contribute directly to lifespan extension, inactivation of these genes would be expected to compromise the extension of lifespan conferred by decreased insulin/IGF-1 signaling, caloric restriction, or the inhibition of mitochondrial function. We find that inactivation of 25 of 29 cytoprotection-regulatory genes shortens the extension of longevity normally induced by decreased insulin/IGF-1 signaling, disruption of mitochondrial function, or caloric restriction, without disrupting normal longevity nearly as dramatically. These data demonstrate that induction of cytoprotective pathways is central to longevity extension and identify a large set of new genetic components of the pathways that detect cellular damage and couple that detection to downstream cytoprotective effectors.
| Many mutations that increase animal lifespan also confer stress tolerance, suggesting that cytoprotective mechanisms underpin the regulation of longevity. It has not been established, however, whether the induction of individual cytoprotective pathways is essential for lifespan extension, or merely correlated. To establish whether the regulatory pathways for the induction of cytoprotective responses are key in the extension of lifespan, we performed an RNAi screen for gene inactivations that decouple the activation of cytoprotective pathways from xenobiotic stimuli that normally induce them. The screen identified 29 genes that constitute the regulatory cascades of the unfolded protein response, oxidative stress response, and detoxification. These upstream regulatory genes are critical to stress tolerance and the extension of lifespan conferred by decreased insulin/IGF-1 signaling, disruption of mitochondrial function, or caloric restriction, but have little effect on normal longevity.
| Lifespan can be extended in C. elegans and other organisms by a variety of ostensibly deleterious interventions: disruption of mitochondrial function, disruption of translation, disruption of insulin/IGF-1 signaling, caloric restriction, exposure to xenobiotics and others. The counterintuitive benefits of these stressful stimuli suggest a hormetic mechanism rooted in the beneficial induction of cytoprotective pathways that respond to environmental challenges, such as starvation, heat, or exposure to xenobiotics. These cytoprotective pathways may represent the mechanisms that drive lifespan extension. While the correlation of stress tolerance and longevity is well established, the underlying cytoprotective pathways have not been fully explored.
Many of the gene inactivations that extend lifespan encode core, conserved components of cells, such as translation factors or mitochondrial proteins, many of which are the molecular targets of known xenobiotics [1]. Lifespan-extending inactivation of cytochrome C reductase, ATP synthase, F59C6.5 in electron transport chain (ETC) complex I, or cytochrome C oxidase may induce the same cytoprotective responses as the xenobiotics that target them, which include antimycin, oligomycin, rotenone and sodium azide, respectively. Similarly, a wide variety of xenobiotics disrupt translation, including hygromycin, genetecin and emetine. Disruption of endoplasmic reticulum (ER) function also extends longevity and may induce cytoprotective mechanisms effective against ER-targeted xenobiotics such as tunicamycin or thapsigargin. The parity of essential cell components targeted by xenobiotics and those that extend longevity upon inactivation suggests that long-lived animals engage cytoprotective mechanisms that evolved as cellular homeostatic and detoxification responses to xenobiotics and virulence factors produced by other organisms.
Cytoprotective responses, including chaperones, antioxidants and pathogen response genes, as well as xenobiotic detoxification mechanisms, can protect extant components of the cell and may contribute to lifespan extension. Genetic studies have identified over 50 mutations that extend the lifespan of C. elegans, and each is resistant to one or more stressors, such as oxidative damage, heat stress or irradiation [2], [3]. The oxidative stress theory of aging has driven extensive analysis of oxidative damage in particular, and while long-lived animals are resistant to compounds that generate ROS, such as paraquat, identification of underlying mechanisms has proven challenging [4], [5], [6], [7], [8], [9], [10], [11], [12]. Longevity is also correlated with thermotolerance, and expression of the heat shock response gene hsp-16.2 predicts longevity in C.elegans [13], [14], [15], [16], [17], [18]. In a genetic screen for enhanced thermotolerance, the majority of isolated mutants were long-lived by at least 15% [19]. Other protein folding mechanisms, such as the ER and mitochondrial unfolded protein responses, contribute to longevity as well [20], [21], [22], [23]. Cellular damage may result from the production of toxic metabolic byproducts or exposure to xenobiotics, consistent with the extension of lifespan by overexpression of the detoxification transcription factor skn-1, a gene that is also required for lifespan extension in daf-2 mutants [24], [25], [26], [27], [28], [29], [30], [31]. The potential influence of diverse cytoprotective functions on longevity is underscored by the heat, ROS and toxin resistance of long-lived animals.
Consistent with the stress-tolerant phenotypes of long-lived animals, cytoprotective mechanisms are activated in long-lived mutants. Disruption of insulin/IGF-1 signaling induces heat shock (hsp-16.49, hsp-16.11, hsp-16.1, hsp-16.2 and hsp-12.6), antioxidant (ctl-1, ctl-2 and sod-3), and pathogen response (lys-7, lys-8 and spp-1) genes [32]. The long-lived mitochondrial mutants isp-1, clk-1 and cyc-1 induce the mitochondrial unfolded protein response [33]. Inactivation of the translation initiation factor ifg-1 induces the transcription of 51 stress responsive genes, including daf-16 and skn-1 [34]. In each of these long-lived mutants, evidence suggests concurrent induction of detoxification mechanisms. The detoxification of xenobiotics in many systems, including C. elegans, involves the upregulation of cytochrome P450's (CYPs), UDP-glucuronosyltransferases (UGTs), and glutathione S-transferases (GSTs). Transcriptional profiling of the long-lived daf-2 insulin/IGF-1 signaling mutant reveals daf-16-dependent upregulation of these functions [32], [35]. A xenobiotic response is similarly induced in long-lived mitochondrial mutants and lifespan extension by disruption of translation requires skn-1, which participates in xenobiotic stress tolerance [36], [37]. While the response to xenobiotics is, in part, the upregulation of detoxification, other cytoprotective mechanisms, such as chaperones, mitigate cellular damage; detoxification and cytoprotection may both be components of a xenobiotic response apparatus mobilized by various aging interventions.
Mechanistic evidence supports the causality of cytoprotective gene activation in lifespan extension. Loss of chaperone expression through inactivation of hsf-1, the transcriptional regulator of the heat shock response genes, abrogates lifespan extension in a daf-2 mutant, while overexpression extends the lifespan of wild-type animals [38]. The ER unfolded protein response (UPR) underlies lifespan extension in daf-2 mutants and in response to caloric restriction [22], [39]. The mitochondrial UPR is required for lifespan extension in the mitochondrial mutants isp-1 and clk-1 [21]. Lifespan regulatory factors, including daf-2, hif-1, skn-1 and hsf-1 are required for pathogen defense, further suggesting that these pathways coordinate critical elements of cytoprotection [40], [41], [42], [43], [44], [45], [46], [47], [48], [49]. These findings highlight the potential contributions of a range of cytoprotective pathways to lifespan extension, but a systematic genetic analysis of the regulation of cytoprotective mechanisms in diverse models of lifespan extension has not been conducted.
We tested the hypothesis that the regulation of cytoprotective gene expression in long-lived animals underlies lifespan extension across mechanistically diverse models of this phenotype. We utilized xenobiotic and genetic stimuli in an RNAi screen to identify regulatory genes required for the appropriate induction of hsp-6, hsp-4, gst-4 and sod-3 in long-lived animals. We directly addressed the activity of these cytoprotection regulatory genes in lifespan extension by inactivation and subsequent lifespan analysis in three functionally diverse long-lived mutants, isp-1, eat-2 and daf-2 (overview in Figure S1). We find that these cytoprotective regulatory genes are critical to lifespan extension in all three longevity backgrounds. These results provide mechanistic support for the hypothesis that lifespan extension occurs through the activation of cytoprotective pathways triggered by xenobiotic or genetic means.
Long-lived mutants express cytoprotective genes at elevated levels. To identify the cytoprotective pathways induced by the diverse conditions that confer lifespan extension, we analyzed the induction of 13 stress-responsive GFP fusion genes functioning in the response to heat, ER stress, mitochondrial stress, oxidative damage, pathogenesis, osmotic stress, xenobiotics, and decreased insulin/IGF-1 signaling (Table S1) by each of 160 gene inactivations found to increase longevity in high-throughput RNAi screens (Table S2) [1], [20], [23], [32], [50], [51], [52], [53], [54], [55], [56], [57]. Clustering the expression of phsp-6::gfp (Mt UPR), phsp-60::gfp (Mt UPR), phsp-4::gfp (ER UPR), pgst-4::gfp (detoxification), psod-3::gfp (ROS), pF55G11.7::gfp (pathogenesis) and pgpdh-1::gfp (osmotic stress) generates distinct groups of longevity gene inactivations (Figure S2). Overall, cytoprotective gene activation is a hallmark of the most potent lifespan extension mechanisms. The average mean lifespan extension amongst the 88 gene inactivations that induce at least one fusion gene (Figure S2) is 27.3%, while the 72 gene inactivations that do not activate a single fusion gene exhibit an average extension of 12.5% (t-test p = 7.7E−12) (Figure S2, Table S2) [1], [50], [51].
To discern how interventions that extend longevity couple to the activation of cytoprotective pathways, we sought to identify the genes required for the activation of hsp-6 (Mt UPR), hsp-4 (ER UPR), sod-3 (ROS response) and gst-4 (detoxification) by drugs or genetic triggers. Activation of cytoprotective genes requires the capacity to detect the disruption of an essential cell function, such as translation or mitochondrial function, and to generate signals that activate downstream responses. The mechanisms by which these events occur remain largely unknown. We reasoned that components of longevity signaling might emerge from an RNAi screen to identify gene inactivations that disrupt the induction of cytoprotective pathways. Because many gene inactivations that confer longevity extension encode targets of naturally occurring xenobiotics, analysis of toxin response may explore the same cytoprotective pathways activated in long-lived mutant animals. We raised C. elegans to young adulthood before treating the animals with toxins to induce the expression of longevity-correlated cytoprotective genes. Tunicamycin, an antibiotic that disrupts N-linked glycosylation in the ER, was employed for the activation of the ER UPR reporter phsp-4::gfp [20]. Antimycin disrupts complex III of the electron transport chain, activating the mitochondrial UPR reporter phsp-6::gfp. Sodium azide treatment activates the skn-1 target pgst-4::gfp, and the activity of the daf-16 target psod-3::gfp was modulated via a temperature sensitive allele of daf-2 [53], [58]. We used RNAi to screen for gene inactivations that blocked the expected cytoprotective response to each drug or genetic stimulus.
The screen encompassed ∼1500 gene-inactivating RNAi constructs, assayed for inhibition of each of the four cytoprotective responses. This library was composed of 395 kinases and 610 transcription factors, as well as two cherry-picked small RNA and longevity sublibraries. Because small RNA pathways have been implicated in stress responses, we screened 317 gene inactivations that have emerged from screens for defects in microRNA or RNAi functions [59], [60]. Gene inactivations that abrogate the increase in longevity conferred by low insulin/IGF-1 signaling have also been identified and potentially regulate cytoprotective functions. We therefore included this set of 179 gene inactivations as well [61].
The primary screen identified 73 gene inactivations required for appropriate activation of cytoprotective responses (Table 1, Table S3). Known stress response regulatory factors (ire-1, skn-1, daf-16) score strongly and each is highly specific to its known function (Table 1). Quantification of fluorescence intensity for 32 gene inactivations that scored most strongly in the primary optical screen confirmed the role of 29 genes in cytoprotective gene induction (Table 1). Induction of phsp-16.2::gfp following heat shock and the expression of a non-stress-induced fusion gene, psur-5::gfp, were quantified to control for generic transgene silencing phenotypes; none of these gene inactivations were potent transgene silencers (Table S4). Expression of the chromosomal loci corresponding to the screened fusion genes was analyzed by quantitative PCR to distinguish transgene dysregulation from regulation of the endogenous loci (Table S5). Results confirmed that the majority of these gene inactivations decouple the chromosomal cytoprotective loci from activation by toxins. While results were largely consistent, measured decreases in fluorescence from the GFP fusion genes were more dramatic than those detected by quantitative PCR for the corresponding genetic loci. The use of multi-copy transgenic constructs may contribute to this observation. Tissue specificity may do so as well, since quantitative PCR averages gene induction over all C. elegans cells while cytoprotective gene induction may be isolated to particular tissues. In addition, the efficacy of RNAi is reduced in neurons and other excitable cells.
The 29 regulators of cytoprotection identified are annotated to function in RNA processing (cpsf-2, cpsf-4, cpf-2), protein degradation (pas-3, let-70, ufd-1, skr-1, cul-1), deacetylation (sdc-2, hda-1, dcp-66, lin-40), phosphorylation (wnk-1, F18F11.5, let-92, kin-1, nekl-2), transcription (mdt-26, dpy-22, elt-2) and other activities (Table 1). Fourteen have been annotated as candidate cofactors for miRNA function, four as cofactors of RNAi and eight as positive regulators of lifespan extension in the long-lived daf-2 mutant [59], [60], [61]. Of the eight putative insulin/IGF-1 signaling factors, five were found to potently regulate the transcription of sod-3 downstream of daf-2 [61].
Sixteen gene inactivations that disrupt the coupling of cellular dysfunction to cytoprotective gene activation demonstrate specificity to one of our four tested cytoprotective gene fusions, including the canonical stress response regulatory factors daf-16, skn-1, and ire-1 (Table 1, Figure 1, Figure S3). The canonical factors score most strongly amongst these, with the exception of nekl-2, which regulates the expression of gst-4::gfp 50% more potently than skn-1. F53F4.11, cpf-2 and dcp-66 are also noteworthy as the most potent previously unidentified pathway-specific gene inactivations, regulating the expression of hsp-4, hsp-6 and sod-3 fusion genes, respectively. We observe the greatest degree of regulation, however, amongst the 16 regulators of cytoprotective gene expression that regulate 2 or more of the tested cytoprotective pathways. lin-40 gene inactivation disrupts psod-3::gfp induction 17-fold, and pgst-4::gfp 2-fold. let-92 gene inactivation results in the most potent disruption of pgst-4::gfp induction, decreasing expression 25-fold, and inhibiting induction of psod-3::gfp and phsp-6::gfp by 3- and 4-fold, respectively. ima-3 and elt-2 gene inactivations both dramatically decrease induction of phsp-6::gfp by antimycin. ima-3 gene inactivation additionally inhibits the induction of pgst-4::gfp (4-fold). elt-2, like let-70 and cpsf-2, is required for the appropriate regulation of all four tested cytoprotective pathways (Table 1, Figure 1). We conclude that the cytoprotective pathways upregulated by conditions that increase longevity are regulated by both distinct and shared genetic components (Table 1, Figure S3). We speculate that shared regulatory genes may be upstream of pathway-specific factors, though the complexity of this regulatory network remains unexplored.
If the cytoprotective pathways normally induced by conditions that confer increased longevity are essential for that increase, decoupling their induction might shorten the lifespan of long-lived mutants more than that of wild-type animals. To test this hypothesis, we asked whether the 29 gene inactivations that disrupt cytoprotective gene induction also abrogated the increase in lifespan conferred by mitochondrial dysfunction (isp-1;ctb-1), reduced feeding (eat-2) or disruption of insulin/IGF-1 signaling (daf-2).
Inactivation of an idealized lifespan regulatory gene would reduce the lifespan of a long-lived strain to that of the control strain (N2) without perturbing wild-type lifespan. In these experiments, 12 of 29 tested gene inactivations abrogate 2/3 or more of the lifespan extension observed in eat-2, isp-1 and/or daf-2 mutants (Table 2). These gene inactivations shorten wild type lifespan much less dramatically, differentiating these lifespan-regulatory gene inactivations from generalized sickness. While dcp-66, pas-3 and arf-3 exert their largest suppression of lifespan in isp-1, inactivation of cpf-2, wnk-1 and nekl-2 are most potent in the eat-2 mutant (Table 2, Figure 2). Of these, however, only dcp-66, which reduces lifespan extension in a mitochondrial mutant (isp-1) by 87%, does not significantly influence lifespan extension in at least one additional mutant. The remaining 6 gene inactivations, including phi-50, ima-3, gob-1, ufd-1, let-70, and elt-2, are critical to lifespan extension in both the isp-1 and eat-2 mutants (Table 2, Figure 2). Two of these, phi-50 and ima-3, also reduce the lifespan of daf-2 mutants by more than 2/3 (Table 2, Figure 2). These phenotypes represent the most potent inhibitions of lifespan extension. Applying a less conservative standard of 15% reduction in lifespan extension, 25 of 29 regulators of cytoprotective gene induction suppress the extension of lifespan in at least one long-lived strain (Table 2). Cumulatively, we find that genes required for the appropriate transcriptional response to xenobiotic stress and disruption of insulin/IGF-1 signaling contribute to diverse axes of lifespan extension, and in some cases, in particular phi-50 and ima-3, to all three studied lifespan extension axes (Table 2, Figure 2).
Many gene inactivations that inhibit lifespan extension also decrease stress tolerance. To reveal the role of cytoprotective response regulatory genes in the tolerance of xenobiotic stress, we inactivated the 29 genes identified in the screen and challenged animals with sublethal (LD30) doses of antimycin, sodium azide, cadmium chloride and paraquat. These toxins parallel the conditions utilized in the screen. Antimycin targets the function of mitochondrial ETC complex III and activates phsp-6::gfp. Sodium azide disrupts the final step of electron transport, blocking energy production and releasing reactive oxygen species, leading to the induction of pgst-4::gfp [58]. Cadmium, like tunicamycin, induces the ER stress response and cadmium tolerance is dependent upon a functional ER UPR [62]. Paraquat survival has been utilized as a measure of ROS tolerance, known to result from the activation of cytoprotective genes downstream of insulin/IGF-1 signaling, such as the superoxide dismutase sod-3 [53].
Inactivation of 16 of 29 genes that disrupt induction of the cytoprotective GFP fusion genes, also disrupted the ability of animals to survive exposure to xenobiotics (Figure 3). Eleven of the sixteen xenobiotic-sensitive gene inactivations (phi-50, wnk-1, nekl-2, mdt-26, let-70, arf-3, elt-2, dpy-22, let-92, F18F11.5, and C06A8.2) enhance sensitivity to the xenobiotic that pairs with the compromised cytoprotective response. The strongest examples of this correlation include phi-50 and nekl-2 (pgst-4::gfp/sodium azide), elt-2, wnk-1 and mdt-26 (phsp-4::gfp/cadmium chloride), let-92, elt-2 and mdt-26 (phsp-6::gfp/antimycin) and dpy-22 (psod-3::gfp/sodium azide). Of the eleven total gene inactivations that pair in this way, seven are also susceptible to additional xenobiotics, suggesting that the pathways examined serve cytoprotection more extensively than previously predicted. In addition, five gene inactivations (dcp-66, pas-3, kin-1, cpf-2 and cul-1) are sensitive only to xenobiotics that do not directly pair with the observed deficit in cytoprotective gene induction, further demonstrating the complexity of stress responsive gene networks and their protective functions. None of the 29 gene inactivations significantly decreased survival following treatment with drug solvent controls alone (Figure S4).
Cumulatively, phi-50, ima-3, elt-2, nekl-2, wnk-1, let-92, mdt-26, and let-70 stand out amongst the 29 genes that disrupt cytoprotective response (Table S6). wnk-1, phi-50 and elt-2 are severely sensitive to multiple xenobiotic stress conditions, but not control conditions, and modulate lifespan extension in all three tested axes (insulin/IGF-1 signaling, mitochondrial function and caloric restriction). While ima-3 and let-92 do not demonstrate xenobiotic sensitivity under the tested conditions, and let-70 only a subtle sensitivity to sodium azide, they are amongst the most robust suppressors of cytoprotective transcription and suppress lifespan extension in all three long-lived mutants. mdt-26 and nekl-2 are sensitive to all four xenobiotic treatments and suppress lifespan extension in two of the three long-lived mutants tested (daf-2 and eat-2 or isp-1;ctb-1 and eat-2, respectively). In total, we identify 15 regulators of cytoprotection that are required for tolerance of xenobiotic stress and lifespan extension (Table S6).
Stress tolerance and lifespan extension are remarkably correlated. The contradictory extension of lifespan by ostensibly deleterious conditions, and the concomitant induction of stress tolerance, suggests that lifespan extension may occur through the hormetic induction of damage-buffering cytoprotective mechanisms. We have identified the cytoprotective pathways that are upregulated by conditions that extend lifespan. In a screen of 160 gene inactivations that increase lifespan, the most potent lifespan extension phenotypes were defined by the induction of suites of cytoprotective genes (Figure S2) [32], [35], [63]. To identify upstream regulatory genes in xenobiotic responses, we designed an RNAi screen to detect gene inactivations that disrupt the normal induction of phsp-6::gfp, pgst-4::gfp and phsp-4::gfp by toxins, and the activation of psod-3::gfp by low insulin/IGF-1 signaling. The induction of cytoprotective longevity-modulatory pathways by toxins may be the normal biological context in which these pathways function, having evolved as countermeasures to the xenobiotic and environmental challenges that animals encounter. Because xenobiotics and lifespan extending gene inactivations engage the same cytoprotective, physiological and behavioral responses, xenobiotic responses may be triggered by direct surveillance of cell functions, which would provide broad, adaptive utility in toxin detection [64], [65], [66].
We identified 29 gene inactivations that decouple normal transcriptional responses to toxins and environmental stress. While some gene inactivations were specific to one toxic modality, such as mitochondrial dysfunction, others affected multiple, distinct toxin response pathways. The identified genes may act in damage surveillance, signaling, or the transcriptional coordination of cytoprotective responses by acting either within cells and tissues or across tissues by an as-yet-undefined endocrine mechanism. Gene inactivations specific to one toxin may act in dedicated surveillance pathways, while those that affect multiple, distinct responses may identify points of signal convergence.
Many of the gene inactivations that disrupt the coupling of cellular dysfunction and the transcription of cytoprotective responses in the screen are annotated phosphorylation or transcription factors. The kinase nekl-2, which we identify in the regulation of gst-4 expression, is necessary for the nuclear localization of skn-1 following oxidative stress [67]. The kinase wnk-1, which we identify as a regulator of the ER stress response, has previously been placed upstream of effector genes in the osmotic stress response [68]. let-92, the catalytic subunit of protein phosphatase 2A, stands out as a potent regulator of gst-4 expression, suggesting it may play a critical role in the regulation of skn-1 activity. The transcription factor elt-2 is expressed exclusively in the intestine, a critical tissue in xenobiotic detection and detoxification in C. elegans. Targets of elt-2 include osmoprotective and innate immune responses, detoxification and oxidative defenses and metal detoxification, as well as the transcription factor pha-4 and, potentially, skn-1. This multitude of key cytoprotective functions is consistent with our finding that elt-2 is required for appropriate expression of hsp-6, hsp-4, sod-3 and gst-4.
We identified five components of the ubiquitin proteasome system (pas-3, let-70, ufd-1, skr-1, cul-1). Proteasome regulatory components like aip-1, and potentially skn-1, are believed to prolong lifespan by stimulating the degradation of damaged proteins [69]. Others, such as the E3 ubiquitin ligase vhl-1, serve regulatory functions by degrading key signaling components [70]. The E2 ubiquitin conjugation factor let-70 regulates all four tested cytoprotective responses (phsp-6::gfp, phsp-4::gfp, psod-3::gfp, pgst-4::gfp), but not the heat shock response (phsp-16.2::gfp) or a constitutively expressed control (psur-5::gfp). It has been previously shown that let-70 interacts with numerous chaperones, contributes to the DNA damage response, and that inactivation of let-70 increases the size of aggregates in a polyglutamine model of protein aggregation [71]. The diverse functions of let-70, one of 22 E2 ubiquitin conjugation factors in C. elegans, are consistent with its interaction with a much larger pool of target-specifying ubiquitin-protein ligases (E3s).
We identify three deacetylases (hda-1, dcp-66, lin-40), all of which regulate the expression of psod-3::gfp downstream of insulin/IGF-1 signaling. Studies of chromatin modification have demonstrated that both silencing and desilencing epigenetic marks regulate lifespan [72], [73].
Other genes of interest include ima-3 and phi-50 (Table 1, Table 2). ima-3, one of three importin alphas that channel NLS-tagged molecules into the nucleus, stands out as the most potent regulator of the mitochondrial UPR we identify. Our results suggest that ima-3 may participate in the transport of stress regulatory factors into the nucleus. phi-50, which regulates psod-3::gfp and pgst-4::gfp, is orthologous to the human hydroxymethylglutaryl-CoA synthases 1 and 2 (HMGCS1, 2) [74]. HMGCS1 is required for the prenylation of proteins such as GTPases, which are essential to organelle homeostasis and many signaling cascades, while HMGCS2 mediates the response to fasting [75], [76].
Eighteen genes found to disrupt the induction of cytoprotective responses serve putative microRNA or RNAi functions, suggesting that small RNAs regulate stress response. As most microRNAs in C. elegans are not individually essential for viability under standard laboratory conditions, the possibility that they fulfill conditional functions such as stress response is particularly appealing [77]. Small RNAs are attractive candidates for stress response regulation, as they are rapidly inducible and the suppression of protein levels they mediate is rapidly reversible [78], [79]. Small RNAs may be generated without translation and spread easily amongst cellular compartments or from cell to cell [79], [80]. Roles for small RNAs in stress responsive gene regulation are emerging in bacterial, plant and mammalian systems [81], [82], [83], [84], [85], [86], [87]. The capacity of small RNA pathways to mediate the expression of duplicated and genetically linked genes may contribute to their potential to act as key regulators of xenobiotic response, as detoxification genes are known to form long tandem arrays by duplication [88]. Like canonical stress response regulatory genes, some small RNAs have been found to regulate longevity [89]. Regulation of cytoprotective genes by siRNA- or miRNA-mediated silencing is likely indirect under the tested conditions because inactivation of genes required for silencing would be expected to increase, not decrease, the expression of a direct target. Additionally, the transgenes utilized were promoter fusions.
The complexity of stress response is increasingly evident, challenging the presumption that cytoprotective pathways are genetically independent. Translation initiation factor 2 (eIF2α) integrates signals from four stress-activated kinases, each responding to diverse stress stimuli including oxidative damage, amino acid starvation, infection and ER stress. Another example is found in the transcription factor slr-2, which co-regulates a suite of diverse stress response genes, including hsp-16.2 (heat shock), hif-1 (oxidative stress) and gpdh-1 [90]. The insulin/IGF-1 signaling pathway contributes to the tolerance of heat, radiation, osmotic stress, oxidative damage and heavy metals, as well as pathogens. It is not surprising that some of the genes identified in our study engage more than one cytoprotective mechanism.
The regulators of cytoprotection we identified contribute to lifespan extension in three distinct long-lived mutants: isp-1 (mitochondrial function), eat-2 (caloric intake) and daf-2 (insulin/IGF-1 signaling). Nearly all of the identified regulators of cytoprotection contributed to lifespan extension in at least one of these mutants and many modulate lifespan extension in at least two conditions. Because previously identified positive regulators of lifespan, such as daf-16 and hsf-1, manifest the cumulative benefit of large suites of co-regulated genes, we suggest that the stress response and lifespan regulatory genes identified here similarly abrogate the induction of many downstream cytoprotective effectors. Our finding that nearly all gene inactivations that disrupt the induction of cytoprotective pathways by toxins also disrupt longevity extension suggests a tight coupling of these pathways. Increased longevity may be the cumulative result of cytoprotective pathway induction or, alternatively, a coregulated output of xenobiotic response analogous to the hormonal pathways of lifespan regulation engaged by insulin/IGF-1 signaling mutants. Our results suggest that xenobiotic and environmental response mechanisms underpin diverse models of longevity extension, with the potential to unify the study of long-lived animals.
Lifespan poses an evolutionary conundrum, as the genetic determination of lifespan ostensibly suggests post-reproductive selection. Our data suggests that lifespan-determining genes do not specify lifespan per se, but rather the activity of damage-buffering cytoprotective pathways normally engaged only in response to stress stimuli, such as toxins. Cytoprotective programs must be subject to Darwinian evolution, selected pre-reproductively to maintain the viability of larval and young adult animals in the presence of xenobiotic and environmental challenges. Post-reproductive adults could engage the same programs. The functions of these pathways are expected to be highly regulated, since they marshal essential resources, such as iron for cytochrome p450s or ATP for chaperones and transporters, away from anabolic pathways and reproduction; organisms that upregulate these pathways continuously would be outcompeted by those who regulate them conditionally [91].
We have identified upstream regulatory components of longevity and xenobiotic response pathways, the overlap of which supports our hypothesis that longevity pathways evolved as xenobiotic and environmental stress response programs. Our results reveal the complex networking of cytoprotective gene regulation. The genes we have identified may act in the detection of stress stimuli, the transduction of a resulting signal or the direct regulation of the transcription of stress response effectors. We find that these upstream regulators play central roles in both xenobiotic stress tolerance and the extension of lifespan in several canonical long-lived strains, including eat-2, isp-1 and daf-2. Xenobiotic and environmental stress response pathways may underpin many current models of longevity extension. The xenobiotic hypothesis of aging invokes hormesis, a phenomenon observed from microorganisms to humans, highlighting the possibility of a xenobiotic approach to longevity extension in humans.
Fluorescent strains phsp-6::gfp(sj4100), phsp-60::gfp(sj4058), phsp-4::gfp(sj4005), psod-3::gfp(cf1553), pgst-4::gfp(cl2166)., phsp-16.2::gfp(tj375) and pfat-7::gfp(bc15777) were obtained from the C. elegans genome center (CGC). pgpdh-1::gfp(vp198) was obtained courtesy of Kevin Strange. Bristol(N2), rrf-3(pk1426) and daf-16::gfp(gr1352) were obtained from the Ruvkun laboratory. pF55G11.7::gfp(hd92), plys-1::gfp(hd102), plys-7::gfp(hd100) and pnlp-29::gfp(hd101) were obtained courtesy of Scott Alper. Long lived strains were daf-2(e1370), eat-2(ad465) and isp-1;ctb-1(mq989).
RNAi clones were grown overnight in LB with 100 µg/ml ampicillin and seeded 100 µl/well to 24-well 5 mM IPTG worm plates. Clones were induced overnight at room temperature. Synchronized L1 worms were raised on RNAi at 20°C, 50 animals/well. Fluorescence was assayed at 48, 72 and 96 hours. Scores were recorded from 0 (no expression) to 4 (strong expression). For lethal clones, worms were grown to young adulthood at 20°C on empty-vector RNAi (L4440) before treatment, with subsequent scoring after 24, 48 and 72 hours. Each clone was scored in three trials. Resulting data was clustered using the open source software Cluster 3.0 with hierarchical uncentered correlation of average linkage and visualized using Java TreeView.
RNAi clones were cultured overnight in LB with 100 µg/ml ampicillin and seeded 100 µl/well to 24-well 5 mM IPTG worm plates, each well containing 1.5 mL agar. Synchronized L1 transgenic strains (SJ4100, sj4005, cf1553;e1370, cl2166) were distributed to RNAi, 50 animals/well, and raised to young adulthood (56 hours) at 20°C. At this time, each well was treated with 0.5 µl 20 mg/ml antimycin in EtOH (sj4100), 1.8 µl10 mg/ml tunicamycin in DMSO (sj4005), 17 µl1 mg/ml sodium azide in water (cl2166) or 1 h 37°C heat shock (tj375). Toxins were diluted in water to 20 µl/well and applied directly to the agar wells. Expression was assayed after 8 hours (tunicamycin, heat), or 24 hours (antimycin, sodium azide). For sod-3(cf1553), psod-3::gfp;daf-2(e1370)ts was raised to young adulthood at 15°C and shifted to 25°C with GFP expression assayed after 12 hours. All clones in the primary screen were scored in three replicates. Candidate stress response regulatory genes were subsequently verified in five or more additional replicates.
Treated worms (see activation of fusion genes by stress in materials and methods) were washed into 200 µl M9 containing 0.3 mg/ml lavamisole and 0.005% Triton X-100, and concentrated in a 96-well pate by centrifugation for 1 minute at 500 RPM, then transferred to 96-well glass slides with a final liquid volume of 5 µl/well. Imaging of slides was automated using Molecular Devices ImageXpress Micro imaging platform with MetaXpress software. Device captures four images per well, which are tiled to construct full wells. Images are captured in both in both GFP and bright field channels. Custom MATLAB (The Mathworks, Natick, MA) scripts distinguish well boundaries by blurring the image and applying a threshold of L*F where L is determined by Otsu's method and F = 0.9. To identify worms, bright field images are bottom hat filtered to decrease variability in background intensity. Otsu's method and a size filter are applied to distinguish objects from background and debris. An outlier method is applied in place of Otsu's when effectiveness is low (<0.7). Fluorescence of worm objects is averaged from the median intensity of each well after background subtraction. Results were averaged from four to eight replicates for each experimental condition and the psur-5::gfp control, with two replicates for the phsp-16.2::gfp control. Significance was determined by a one-tailed t-test, p = 0.05, and fold decreased expression >1.5× without multiple tests correction. Venn diagrams were generated using the online BioInfoRx Venn diagram tool available at <bioinforx.com/free/bxarrays/venndiagram.php>.
Wild-type N2 animals were raised on HT115 bacteria on 10 cm agarose plates at 20°C and treated with cytoprotective response-inducing toxins (see activation of fusion genes by stress in materials and methods). Animals were harvested after 8 hours (antimycin, tunicamycin) or 24 hours (sodium azide). In the case of sod-3, e1370 animals were raised to young adulthood at 15°C and shifted to 25°C and harvested after 12 hours. To isolate total RNA, animals were washed, resuspended in trizol, frozen and homogenized by grinding. RNA was isolated by chloroform extraction followed by ethanol precipitation. Reverse transcription was carried out with the Ambion AM1710 Retroscript RT-PCR kit. Quantitative PCR reactions utilized 12.5 µl Bio-Rad iQ SYBR Green PCR Supermix (170-8880) with 2 µl template, 5.5 µl water and 5 µl each of two of the following paired oligos: hsp-4 GAGAACACAATTTTCGACGCC/GACTTGTCGACGATCTTGAACGG; hsp-6 GATAAGATCATCGCTGTCTACG/GTGATCGAAGTCTTCTCCTCCG; sod-3 CACTATTAAGCGCGACTTCGG/CAATATCCCAACCATCCCCAG; gst-4 GCCAATCCGTATCATGTTTGC/CAAATGGAGTCGTTGGCTTCAG. Fold change was measured in comparison to Y45F10D.4 using oligos GTCGCTTCAAATCAGTTCAGC/GTTCTTGTCAAGTGATCCGACA as described by Hoogewijs et al. 2008 [92]. Quantitative RT-PCR was carried out using a Bio-Rad C1000-CFX96RT thermocycler (3 m 95°C; 44 cycles of 95°C 10 s, 60°C 30 s, 72°C 30 s; 5 m 72°C). Experiments were carried out with four replicates of approximately 4,000 animals per replicate.
Long-lived strains daf-2(e1370), eat-2(ad465) and isp-1;ctb-1(mq989) were raised to young adulthood at 20°C on 10 cm worm plates with HT115 E. coli. RNAi clones were cultured overnight in LB with 100 µg/ml ampicillin, seeded 400 µl/well to 6-well 5 mM IPTG worm plates and induced overnight. The young adult animals were transferred to the prepared IPTG worm plates, 40 animals/well, and FuDR was immediately applied to the plate to a final concentration of 80 µg/ml agar. On day 4 adulthood, wells were supplemented with 400 µl of additional bacteria concentrated to 10× in 5 mM IPTG M9 with 100 mg/ml ampicillin and induced for 2 hours at room temperature before application to worm plates. Lifespan was scored by touch response on alternate days with censoring. Survival statistics were calculated using SPSS Kaplan-Meier. All analyses are based upon mean lifespan. Experiments were performed with three replicates per condition and an average of 103 worms scored per condition. Significance was held to p = 0.05 within each tested strain; significance of differences in lifespan extension are based upon a threshold of 15% decrease. The DNA and RNA synthesis inhibitor 5-fluoro-2′-deoxyuridine (FUdR) was used to inhibit progeny production. Although the use of FUdR is well established and does not affect the lifespan of wild-type animals, FUdR can affect lifespan of particular mutants [93], [94], [95], [96]. For example, mutation of tub-1 or gas-1 extends lifespan in the presence of FUdR, but not in its absence [93], [97]. Our experiments included FUdR in both control and experimental trials, so that any FUdR effects on lifespan were controlled.
L1 rrf-3(pk1426)ts worms were synchronized overnight in M9 and raised to young adulthood at 25°C on 10 cm worm plates with HT115 E. coli. RNAi clones were cultured overnight in LB with 100 µg/ml ampicillin, seeded to 6-well 5 mM IPTG worm plates and induced overnight. Young adult rrf-3 worms were distributed to RNAi, 50 animals/well, and raised for 3 days at 25°C. Worms were transferred to wells of M9 containing 24 mg/ml paraquat, 5.2 mg/ml cadmium chloride, 22 µg/ml sodium azide, or 696 µg/ml antimycin, with a total volume of 230 µl per well in a 24-well format. Animals were incubated in solution for 16 hours. Survival was then analyzed by scoring for spontaneous movement. Experiments were conducted with three replicates and an average of 94 animals scored per condition. Significance of proportion survival was determined by a one-tailed t-test, p = 0.05 without multiple tests correction. Error bars display S.D.
|
10.1371/journal.pcbi.1005010 | Colony Expansion of Socially Motile Myxococcus xanthus Cells Is Driven by Growth, Motility, and Exopolysaccharide Production | Myxococcus xanthus, a model organism for studies of multicellular behavior in bacteria, moves exclusively on solid surfaces using two distinct but coordinated motility mechanisms. One of these, social (S) motility is powered by the extension and retraction of type IV pili and requires the presence of exopolysaccharides (EPS) produced by neighboring cells. As a result, S motility requires close cell-to-cell proximity and isolated cells do not translocate. Previous studies measuring S motility by observing the colony expansion of cells deposited on agar have shown that the expansion rate increases with initial cell density, but the biophysical mechanisms involved remain largely unknown. To understand the dynamics of S motility-driven colony expansion, we developed a reaction-diffusion model describing the effects of cell density, EPS deposition and nutrient exposure on the expansion rate. Our results show that at steady state the population expands as a traveling wave with a speed determined by the interplay of cell motility and growth, a well-known characteristic of Fisher’s equation. The model explains the density-dependence of the colony expansion by demonstrating the presence of a lag phase–a transient period of very slow expansion with a duration dependent on the initial cell density. We propose that at a low initial density, more time is required for the cells to accumulate enough EPS to activate S-motility resulting in a longer lag period. Furthermore, our model makes the novel prediction that following the lag phase the population expands at a constant rate independent of the cell density. These predictions were confirmed by S motility experiments capturing long-term expansion dynamics.
| Collective motility is a key mechanism bacteria use to self-organize into multicellular structures and to adapt to various environments. An important example of such behavior is social (S) motility in the gram-negative bacterium Myxococcus xanthus. S-motile cells are restricted to movement in groups and do not move as individual cells. S-motility is powered by type IV pili (TFP)–multi-subunit filaments, which extrude from the cell poles, adhere to the substrate and retract, pulling the cell forward. TFP retraction or adhesion is suggested to be triggered by extracellular exopolysaccharides (EPS) deposited by cells on the substrate. As individual cells synthesize both pili and EPS, it is unclear why S-motile cells only exhibit group movement. Moreover, the experimentally observed initial cell-density dependence of S-motility remains unexplained. To understand these phenomena, we developed a mathematical model for the colony expansion of S-motile cells. Our model hypothesizes that the EPS level regulates the TFP activity that initiates collective cell movements. With this assumption, the model quantitatively matches the density-dependent expansion rate. Moreover, the model predicts two phases during colony expansion: an initial density-dependent lag phase with a slow expansion rate, followed by a faster expansion phase with a density-independent rate. These model predictions were confirmed by long-term colony expansion experiments.
| New interest in the study of microbial collective behaviors has been ignited by recent discoveries that are critical to bacterial pathogenesis and multicellular developmental processes in these single-cell organisms, including quorum sensing [1, 2], phenotypic heterogeneity [3], and biofilm formation [4]. The soil bacterium, Myxococcus xanthus is the premiere bacterial model organism for investigations of self-organization and multicellular development [5]. Different M. xanthus multicellular behaviors emerge depending on their environmental conditions. In nutrient-rich conditions, M. xanthus cells spread in a coordinated manner forming organized groups [5]. When spreading over prey microbes, M. xanthus cells self-organize into bands of traveling waves termed ripples [6–8]. When nutrients are scarce, M. xanthus executes a multicellular developmental program in which roughly 100,000 cells aggregate into a hay stack-shaped fruiting body within which many of the cells sporulate [9, 10].
M. xanthus cells move exclusively on solid surfaces and this movement is essential for all their multicellular behaviors. M. xanthus possesses two genetically distinct types of motility: gliding or adventurous (A) motility and twitching or social (S) motility [5, 11, 12]. Single cell movement is facilitated by A-motility, which is most efficient at high agar concentrations. In contrast, group movement is facilitated by S-motility, which is most efficient at low agar concentrations. S-motile cells only move when they are within a cell length of a neighbor [11]. Wild-type cells exhibit these two motility systems simultaneously. This most likely has been a selective advantage enabling M. xanthus to adapt to a variety of physiological and ecological environments. Both motility systems enable these rod-shaped cells to move along their long axis and periodically reverse direction by switching polarity, i.e. the leading cell pole becomes a lagging pole and vice versa [13–15].
The molecular basis of A and S motility has been studied extensively [5, 11, 12, 16–19]. Recent studies have proposed a ‘focal adhesion complex’ model for A motility in which intracellular motors interact with adhesion complexes on the membrane that are bound to substrate and power movement by pushing against the substrate [19]. S motility has been determined by genetic and behavioral analysis to require interaction between type IV pili (TFP) that powers movement and extracellular matrix polysaccharide (EPS) [16, 20–23]. The lipopolysaccharide (LPS) O-antigen is also required for social motility, yet its contribution is currently unclear [23]. The TFP are filaments 5–7 nm in diameter and 3–10 μm in length composed of PilA monomers encoded by the pilA gene. Each step of S movement, involves TFP extension and retraction, which is achieved by polymerization and depolymerization of PilA monomers. Secreted EPS is the anchor and/or trigger for TFP retraction [24]. Consequently, M. xanthus mutants lacking TFP or EPS fail to display S motility [16, 18, 22, 25].
Although it is clear that TFP and EPS are essential for social motility, many aspects of S motility-driven colony expansion remain unexplained [26]. For example: why, despite of cell reversals, does the colony radius increase [27, 28] and why does the observed colony expansion rate depend on the initial cell density [11, 27, 29]? To explain the S motility-driven colony expansion dynamics of M. xanthus cells we have developed a mathematical model that accounts for the interaction between TFP and EPS. This model makes two novel predictions that are confirmed experimentally in this report.
To study social motility-driven colony expansion of M. xanthus cells we developed a reaction-diffusion model. The major assumptions and ingredients of the model are summarized and justified in this section and the technical details are included in the Methods section.
Experimentally, social motility in M. xanthus is usually studied by placing a specific number of liquid-grown cells on an agar plate and measuring the increase in the colony diameter over time. Notably, these colony expansion experiments start with over 105 cells and the cell population further increases over time [27, 29]. The colony dimensions (∼10–30 mm) are orders of magnitude larger than the single cell length (4–5 μm). These conditions make it impractical to simulate the expansion using agent-based modeling [30]. Therefore, we focused on continuous approaches formulating the equation for ρ(r,t)–cell density at a given location and time. Furthermore, the experimental studies are conducted over a long observation period (∼10–100 hr), which is much longer than the single-cell reversal period (∼5–10 min) [27–29]. Under these conditions, we can approximate the cell movement over a time-scale of multiple reversals as diffusion [31]. An effective diffusion coefficient can be estimated based on the single-cell speed and reversal period (see Methods section) [31].
For S-motile cells, TFP adhesion and/or retraction is stimulated by the presence of EPS [24, 25, 32, 33] and therefore, effective diffusion should increase with increasing EPS. To incorporate this into the model, we used the following expression for the effective diffusion coefficient,
D(e)=D0+Dpϕ(e).
(1)
Here D0 is a S motility-independent diffusion coefficient. For strains lacking A-motility (A−S+ strains) or wild-type cells under conditions in which A motility is ineffective (low agar concentrations) this term is small and can arise from the mechanical cell-cell repulsion during growth [34, 35]. It can be estimated from the expansion of mutants lacking both A and S motility [27]. In the second term, Dp, the maximal diffusion coefficient due to S-motility is multiplied by a dimensionless factor 0≤ϕ(e)≤1. This factor is a function of the local EPS concentration (e) and can be interpreted as a probability of pilus retraction. We assume that in the absence of EPS retractions fail (ϕ(0) = 0) and at high EPS concentrations retractions always succeed (ϕ(∞)→1). For our model we have chosen a phenomenological Hill-function of ϕ(e):
ϕ(e)=eme0m+em
where m is the Hill coefficient and e0 is half-saturation concentration.
To compute the local EPS concentration, e(r,t) we assume that each cell produces EPS at a constant rate (α) with a resulting production flux being a linear function of cell density. We assume that EPS is not diffusible as it consists of large macromolecules that bind to the agar surface and that it degrades/dries with a constant rate (β).
If cell diffusion (random motion) was the only factor contributing to cell expansion, we would expect that the colony radius would increase as a square root of the expansion time [36] and correspondingly the expansion rate would gradually decrease. However, this is not experimentally observed [27, 28]; instead, an approximately constant expansion rate is seen. This apparent contradiction can be resolved by our observations that during long-term expansion experiments the cells continue to grow (the M. xanthus generation time of ~4–5 hr [37–39] is shorter than the typical spreading experiment time-scale). Thus, since cell growth can substantially change the expansion dynamics [31, 39], it must to be accounted for in the model. The growth rate is modeled using the Monod equation
g(N)=gmaxNN0+N
(2)
where gmax is the maximum growth rate, N is a local density of growth-limiting nutrients and N0 is half-saturation coefficient [40]. The nutrients will also diffuse through the agar [41] (the corresponding diffusion coefficient is denoted as DN).
When the assumptions described above are combined together, the following set of three coupled partial differential equations describe S motility-driven colony expansion,
∂∂tρ(r,t)=1r∂∂r(rD(e)∂∂rρ)⏟cellmotility+g(N)ρ⏟cellgrowth
(3)
∂∂tN(r,t)=1r∂∂r(rDN∂∂rN)⏟nutrientdiffusion−g(N)ρ⏟nutrientconsumption
(4)
∂∂te(r,t)=αρ⏟EPSproduction−βe⏟EPSdrying
(5)
where D(e) and g(N) are given by Eqs (1) and (2), respectively. To reduce the number of unknown parameters we can without loss of generality set the half-saturation EPS level e0 = 1. This is done by rescaling the EPS level and production rate to e → e/e0 and α → α/e0, respectively. M. xanthus motility parameters were estimated in a modeling study that showed existence of traveling waves during colony expansion [31]. In this model, a cell density-dependent diffusion rate for cell movement was assumed irrespective of their motility (A or S) type. Whereas our model is based on the experimental observation that the TFP motility (or the diffusion rate) is regulated by the self-produced cellular EPS.
We numerically solved the set of equations described above with the appropriate initial and boundary conditions (details are provided in the Methods section). Fig 1A shows the numerical solution of the population density and nutrients at different times. In our simulation, cells enter from the outer edge of the initial colony (at distance r = 0 in Fig 1A) into an empty region and grow by consuming the available nutrients (Fig 1A). As the cell density increases, the level of EPS rises, which in turn increases the diffusion rate of the cells and causes the population to spread outward. The population profile shows a sharp increase in density at the colony front (defined as the advancing part of the population profile), which is a consequence of the sharp increase in diffusion rate with an increasing EPS density. The existence of a sharp profile is consistent with the colony patterns observed during social motility [5, 11, 27], where there are no single cells at the colony edge (defined as the low density region leading the advancing colony front). At longer incubation times, the shape of colony front becomes fixed and the colony expansion rate becomes constant, i.e. there is a traveling wave solution. Such properties of the reaction diffusion model with population growth are traditionally observed in Fisher’s equation (also known as Fisher-Kolmogorov equation), which is widely used in theoretical ecology [42]. The equation was first formulated by Fisher to describe the spread of advantageous genes in spatial populations and assumed logistic growth and constant diffusion [43, 44],
∂∂tρ(x,t)=D∂2ρ∂x2+gρ(1−ρ).
This equation admits a traveling wave solution of the form ρ(x,t) = ρ(x − c t), where the wave speed is given by [43, 44]
c=2Dg
An extended form of this equation in which the growth rate depends on the nutrient concentration (via Eq (2)), also displays a traveling wave solution [45] with a speed that can be shown (for non-diffusing nutrients, i.e. DN = 0) to be
c=2Dgmax(NinN0+Nin)
[45] (see S1 Text) where Nin is the initial nutrient concentration. In these examples, the expansion rate is determined by the maximum growth rate at the tip of the wave [42] where the population density is low (ρ~0) and nutrients are high N~Nin.
In contrast to these cases, in our model, the diffusion rate is non-linear and it increases from a low value (D0) at the outer edge to a higher value (D0+Dp) in the interior as EPS levels increase from the outer edge to the interior of the colony (Fig 1B). Therefore, the colony interior determines the expansion rate in our model. For Fisher’s equation with non-linear diffusion, an analytical expression of the expansion rate is often not straightforward. However using simple scaling (see S2 Text), we show that the expansion rate is proportional to
Dpgmax
(as shown in Fig 1D and 1E), therefore four-fold changes in effective diffusion or growth leads to two-fold changes in the wave speed. The proportionality coefficient may differ from the value of 2 for the Fisher equation and depends on model parameters D0, N0 and Nin. The dependence of the expansion rate on the cell diffusion rate and growth rate is a common feature displayed in colony expansion models for different bacteria [35, 42, 46] including M. xanthus [31].
By applying traveling wave solutions to our model, we can numerically calculate the exact wave speed using a method of phase-space analysis (see Methods section). The expansion rates determined by the phase-space analysis are in agreement (solid and dashed lines in Fig 1D and 1E) with the expansion rates calculated by measuring the advance of colony front over time with variations in the model parameters (circles in Fig 1D and 1E). For instance, the expansion increases from a minimum value cmin to a maximum cmax as the initial level of nutrients is increased (as shown in Fig 1F). In biologically relevant conditions, colony expansion is observed on nutrient-rich agar and therefore the nutrients are sufficient for the cells to produce EPS at least to its half saturation value to enable S motility-driven movement. Therefore, for S motility-driven colony expansion, the expansion rate will be between c' (dotted line in Fig 1F) and cmax, which are the expansion rates for constant EPS at half saturation levels and for large nutrients level.
Given that the effective diffusion coefficient increases with increasing EPS density, which in turn increases as more cells produce EPS, we hypothesized that these effects could be responsible for the increase in the expansion rate at higher cell densities. To test this hypothesis, we used experimental data from two papers which measured M. xanthus S motility-driven colony expansion [27, 29]. Briefly, in these experiments M. xanthus cells at different densities are placed on an agar substrate and as the cells move and divide, the colony expands and its radius increases. The expansion rate is quantified by measuring the difference between the initial and final radius of the colony. The final time corresponded to 8 hr (for expansion rate estimation) in the Kaiser et al. experiments [27] and 24 hr in the Berleman et al. experiments [29]. Using our model equations, we simulated each set of experiments by adjusting our model parameters.
The results reveal that our model reasonably matches the data from Berleman et al. (Fig 2A) and Kaiser et al. (Fig 2B) which represent a > 100-fold variation in the initial cell density. Notably, the best fit to each data set was achieved using a set of parameters that was identical, except for the maximum diffusion coefficient, Dp, and EPS production rate, α. The difference in the effective diffusion coefficients can be easily attributed to the differences in experimental conditions. The Berleman et al. experiments [29] were performed on soft agar (0.5% agar) and led to a high effective diffusion coefficient (Dp = 220 μm2min-1), whereas the Kaiser et al. experiments [27] were performed on harder agar (1.5% agar; Dp = 16 μm2min-1). This is consistent with the fact that S-motile cells perform better on soft agar surfaces. This difference can be achieved with about 3.5-fold differences in the cell speed (see Methods section).
Furthermore, our model predicts differences in EPS-related parameters for the two experiments. To match the data we needed to include an approximately 8-fold difference in the EPS production rate. The same effect can be achieved by changes of the effective EPS drying/degradation rate or the EPS threshold to enable social motility. These results indicate that the production rate or degradation or threshold of EPS could be different for different agar conditions or for the different bacteria strains (A-S+ [27] and A+S+ [29]) used in these two experiments. We also noted that the fit was best when the effective cell diffusion D(e) sharply changes with the EPS level, i.e. for a high value of Hill’s coefficient (m≥4). At lower values of Hill’s coefficient, the model does not fit the experimental data in Fig 2A as it lacks the sharp increase in expansion radius above a threshold initial density (see S3 Text and S1 Fig). This result indicates the existence of a sharp threshold in the EPS level above which TFP are able to attach and/or retract. This sharp threshold is another model prediction that can be tested in the future.
To further explore the effects of the different initial cell densities on the colony expansion, we used our model to compute how the expansion rate (defined as the time derivative of the position of the leading edge, where ρ is very low ~0.01) depends on time and on the initial cell density for the parameters estimated to match the experimental data used in Fig 2. The results of our simulation (Fig 3) show that the colony expansion rate has a transient slow expansion phase with a duration that depends on the initial cell density, followed by a constant expansion phase at longer incubation times. Our results indicate that a population with a low initial cell density will lag behind higher density populations due to its slower transition to a steady-state expansion rate. This effect is mediated through the production of EPS, which is low for low initial densities. This leads to reduced motility and thereby a slower expansion driven only by the basal diffusion rate D0. The cells start moving with an effective diffusion rate close to Dp, only when the EPS density reaches a threshold value. At steady state, the expansion rates for populations that had different initial cell densities are similar because the initial cell density does not affect the cell density of the advancing colony edge.
Previous experiments found the rate of expansion to be different for different initial cell densities and these data suggested that it would remain constant over time [27, 29]. Our model, in contrast, shows that the expansion rate eventually becomes independent of the initial cell density and the density dependence is observed only during a transition period before the constant expansion phase. This contradiction could be due to the fact that previous experiments measured expansion rates for short periods [27], during which some populations were still in their slower expansion phase. Thus, our model, which quantitatively reproduces the previous results, generated a new prediction that could not be confirmed with the previous experimental data. Therefore, new experiments were needed to determine how long the effect of the initial cell density persists during colony expansion.
The prediction of a transient density-dependent lag phase followed by a density-independent expansion rate motivated us to conduct systematic long-term colony expansion experiments. Previous M. xanthus S motility-expansion experiments at different initial cell densities only reported net expansion after 8 hr [27] or 24 hr [29], without reporting any later time-points. To test the prediction we decided to extend these experiments to > 4 days, which according to our model is sufficient to reach the steady-state expansion rate.
The expansion assays were performed on 0.5% agar CTT plates with 0.2% yeast extract and initial cell numbers ranging from about 6 x 104 to 1.2 x 107 cells per initial spot (initial spot radius ~1.7mm). To this end, we inoculated cultures of A−S+ cells (strain DK1218) and incubated them until an exponentially growing density of ~4 x 108 cells/ml was reached. The cultures were 10-fold concentrated and then diluted to achieve densities ranging from 2 x 107 to 4 x 109 cells/ml. Three μl drops of cells at each density were spotted onto the agar plates and incubated at 32°C. To quantify the colony expansion images of the colonies were collected for at least 96 hr using a stereo microscope and digital camera.
The increase in colony diameter commenced at different times for different initial cell densities indicating the presence of a density-dependent lag phase. The populations with lower initial cell densities began expansion later than the populations with higher densities, shown in Fig 4A. We performed three replicates of this experiment and quantified the colony expansion by computing the net increase in the colony radius as a function of time for different initial cell densities. We observed that the model fits our data using the same set of parameters as in Fig 2B with the diffusion rates Dp = 200 μm2min-1 due to the use of 0.5% agar. As predicted by the model, the colony expansion during the longer incubation times occurred at a constant rate in our experiments regardless of the initial density (all lines in Fig 4B have equal slopes). Furthermore, we observe that the expansion curve for high initial cell numbers (6 x109 cells & 12 x109 cells in Fig 4B) nearly overlap indicating there is either no lag phase or a very short lag phase at high initial cell densities. These data indicate that cell motility rapidly becomes active due to high EPS production. This scenario directly corresponds to the saturation in the Hill-function (ϕ(e≫e0) ~1) at high EPS concentrations and justifies our choice of function ϕ(e) to represent TFP activity. Similarly, at low initial cell numbers (0.6 x108 cells & 1.2 x108 in Fig 4B) the differences in the lag phase duration are small, resulting in near overlap of the expansion curves, which suggests that the TFP activity below half saturation (ϕ(e≪e0)≈(e/e0)m) is low until a threshold EPS level is reached. Therefore, our long-term experiments validate the assumptions of our model and confirm its predictions. Moreover, these data reveal the regulation of cell motility by EPS as a mechanistic basis of M. xanthus social motility.
Reaction-diffusion models have been widely used in many biological systems to study various spatial and temporal patterns [47, 48], including the expansion of microorganisms on surfaces [46, 49, 50]. In this paper, we formulated a deterministic reaction-diffusion model with a key characteristic that cell motility depends on EPS deposition. Our model successfully explains several salient features of S motility-driven colony expansion in M. xanthus. Specifically, it predicts that M. xanthus colonies expand as a traveling wave with a sharp front. The speed of expansion scales with
Dpgmax
and its absolute value depends on the EPS half-saturation (e0) and the diffusion rate ratio (D0/Dp). Using a calibrated set of parameters the model recapitulates the experimental trends showing density-dependent colony expansion. Our model suggests that, in order to achieve agreement between the modeling results and the experimental trends, a sharp increase in cell diffusion rate at a threshold EPS level is critical. As a consequence the model predicts that populations starting at low initial densities have a lag phase until sufficient EPS accumulates. Our model further predicts that this lag phase ends after longer incubation times and then the population advances at a constant rate irrespective of the initial cell density. To validate these predictions we performed long-term colony expansion experiments with S-motile cells. Our results confirm the presence of a lag phase that depends on the initial density followed by density-independent steady-state expansion.
To explain the experimental data the model assumes that the activity of TFP motility is triggered by the EPS level. To date no definitive evidence proves this relationship. However, experimental studies have reported the loss of social motility in mutants lacking EPS production [25, 29] and the gain of social motility when EPS is complemented externally [29, 33] or when cells are subjected to specific conditions which overcome the EPS requirement (e.g., polystyrene substrate submerged in 1% methylcellulose) [51]. However, it would be useful to show a direct relationship between S motility-driven colony expansion and EPS production. A systematic study could be performed in which EPS mutants are placed on various concentrations of EPS purified from M. xanthus wild-type cells or mixed with strains producing different amounts of EPS. According to our model, expanding colonies will achieve different expansion rates depending on the EPS level. Using such expansion rate data, the EPS-dependent diffusion rate can be extracted using the relation D(e*) = c2/2g for each known EPS level (e*). Furthermore, examining cell behavior at the colony edge might provide additional insights into this relationship. We expect the cells at the very edge of the colony move less compared to the cells in the interior until the sufficient EPS accumulates. Future quantitative experiments will provide stronger evidence for the role of EPS in the regulation of cell motility.
Although our model is used here to explain social motility-driven expansion in M. xanthus, the model can also be applied to other bacterial species. For example, a similar density-dependent lag phase is observed during swarming motility in undomesticated strains of Bacillus subtilis [52]. Swarming motility is multicellular movement on solid surfaces (soft agar plates in a narrow concentration range: 0.3%-0.5% agar) powered by rotating flagella. B. subtilis swarming depends on surfactin, which is a surfactant produced by the cells that acts as a lubricant to reduce the surface tension between the cells and substrate, and thereby promotes surface spreading. The surfactant production depends on the local cell-density and appears to regulate colony expansion in a similar fashion as EPS does during social in M. xanthus. Moreover, the long-term colony expansion rate of B. subtilis cells lacking surfactant production is shown to be dependent on the amount of the externally provided surfactant levels [52]. Despite these similarities the underlying biophysical mechanism behind the extracellular-component-dependent expansion may be somewhat different. For instance, the surfactin dependency could arise from the fluidic properties of the colony itself. It has been observed that as non-flagellated B. subtilis colonies expand on hard agar, the colony height (thickness) increases. This transiently increases the osmotic pressure, which eventually decreases as the colony expands. In this case, a non-linear dependence of diffusion rate on the cell density originates from a fluid dynamic model [53].
We considered most model parameters from the literature [31]. The gliding speed of Myxococcus xanthus is reported to be in the range vg = 4–7 μm min-1 [54]. The reversal period of a single cell varies between 5–10 min [13], giving a reversal frequency of f = 0.1–0.2 min-1. Single cell movement with periodic reversals can be considered as a velocity-jump process, in which a cell moves along its length at a velocity vg and reverses direction according to a Poisson process with a constant rate (1/f). A diffusion equation can be obtained for such a velocity-jump process and the corresponding diffusion rate can be approximated as Dp≈(vg)2/2f~80–245 μm2 min-1 [31, 55]. The nutrient diffusion rate in agar media is faster than the cell diffusion rate and is taken to be DN = 104 μm2min-1[31, 56]. The doubling time is 4 hr (DZ2 strain) giving the growth rate g = 0.173 hr-1. Other parameters are reported in Fig 1A.
We solve the model equations starting from the edge of the initial spot, which is taken to r = r0 (the initial radius of the colony ~1300–1700 μm) to a boundary (rb = r0+30 mm). The initial conditions are set as ρ(r,0) = e(r,0) = 0 and N(r,0) = Nin. The boundary conditions are
(∂ρ∂x)r=r0=c0ρ0,(∂ρ∂x)r=rb=0,(∂N∂x)r=r0,r=rb=0,(∂e∂x)r=r0,r=rb=0
where c0 = 0.003 μm-1 (10% of vg/Dp = 2f/vg, the net flux of cells in the presence of EPS) is the initial flux at which the cells disperse from the edge of the colony and grow by consuming the available nutrients. The initial nutrient profile is set to a uniform value of Nin = 3 a.u. per μm2 and the nutrient half-saturation is chosen to be N0 = 0.1 a.u per μm2 (so that cells stop growing in low nutrient conditions). We numerically solved the partial differential equations (Eqs (3–5)) using the Crank-Nicolson method [57].
A phase-space analysis method was used for calculating the steady expansion rate of the traveling waves formed in our model. To simplify the analysis we neglected the radial part of the diffusion term as it decays inversely with expansion distance. In addition we assumed that the EPS concentration quickly reaches its steady state level e* = αρ/β. As a result the model can be reduced to two equations,
∂ρ∂t=∂∂x(D(ρ)∂ρ∂x)+gmaxρ(NN0+N)
(6)
∂N∂t=DN∂2N∂x2−gmaxρ(NN0+N)
(7)
where
D(ρ)≡D(e*)=D0+Dp((αρ/β)m1+(αρ/β)m)
is a density-dependent diffusion rate.
In steady state the two model equations display traveling wave solutions. Thus, the following property for ρ(z) = ρ(x − c t) and N(z) = N(x − c t) can be considered, where c is the speed of the traveling wave. Using these properties, the equations become
−cdρdz=ddz(D(ρ)dρdz)+gmaxρ(NN0+N)
−cdNdz=DNd2Ndz2−gmaxρ(NN0+N)
Adding the above equations, results in the following
−cdρdz−cdNdz=ddz(D(ρ)dρdz)+DNd2Ndz2
Integrating once the equations become
−cρ(z)−cN(z)=D(ρ)dρdz+DNdNdz+IC
where IC is an integration constant that can be set by substituting the values ρ(z) = 0 and N(z) = Nin at z → +∞ (in the unpopulated region). Note that their derivative also vanishes at z ± ∞, i.e., ρ(±∞) = 0 and N(±∞) = 0. As a result IC = −cNin. Therefore, we arrive at the following set of equations
N″(z)=ρgmax(NN0+N)−cN′DN
(8)
ρ′(z)=−c(ρ+N−Nin)+DNN′(z)D(ρ)
(9)
In the moving z-frame, the traveling wave starts from a fixed point (ρ(z) = Nin, N(z) = 0) at z → −∞ and approaches another fixed point (ρ(z) = 0, N(z) = Nin) at z → +∞. To determine the wave speed numerically, we cast the equations above into the following first order autonomous equations,
Ω′(z)=ρgmax(NN0+N)−cΩDN
N′=Ω
ρ′(z)=−c(ρ+N−Nin)+DNΩD(ρ)
The fixed points of the equations listed above are (ρ,N,Ω): (0,Nin,0) (S2 Fig stable node, closed circle) and (Nin,0,0) (S2 Fig saddle node, open circle). The traveling wave solution connects the saddle node or initial state at z = −∞ to the stable node or the final state at z = +∞ in the phase plane (ρ,N).
Numerical analysis shows that the solution behavior transitions from oscillatory to negative and then to non-negative values in cell density (ρ) as the wave speed c increases. As a result, for physically realistic solutions, we need to identify the minimal c value for which the solution becomes non-negative (rather than non-oscillatory). This determines the expansion rate for our model’s equations.
M. xanthus strain DK1218 (A-S+) was grown overnight in CTT broth (1% Difco Casitone, 10 mM Tris-HCl pH 8.0, 8 mM MgSO4 and 1 mM KHPO4 pH 7.6) at 32°C with shaking [58]. When the M. xanthus culture reached mid-log phase (4x108 cells/ml, 100 Klett units), the cells were harvested in 1.5 ml micro-centrifuge tubes at 13,000 rpm and ten-fold concentrated to Klett 1000 by resuspension in CTT broth. The cells were then diluted in CTT broth to Klett 5, 10, 25, 50, 125, 250, and 500. For microscopy a 3-μL drop of each dilution of M. xanthus cells was placed onto one 10 cm CTT 0.5% agar plate that also contained 0.2% yeast extract. The plates were incubated at 32°C in an in-house-designed humidity-controlled chamber (a plastic shoe box with a lid in which the bottom was covered by wet paper towels) for more than 96 hrs. Each spot was imaged using an Olympus SZH10 stereo microscope and OptixCam Pinnacle series digital camera with OCView7 software. The distance moved by each colony edge was measured at 2, 4, 6, 8 hrs, and at least twice a day for at least 4 days. Each image was quantified using imaging software GIMP. Specifically, a circle was fitted to the colony in the image to obtain the radius. The expansion distance was computed as a difference between the colony radius at any given time t to the initial colony radius. The average and the standard deviation of the expansion distance from three experimental repeats were calculated in an Excel spreadsheet (plotted in Fig 4B). The cell initial density used in the simulation was determined by calculating the number of cells in the drop used for inoculation divided by the average initial colony radius (∼1700 μm).
|
10.1371/journal.pntd.0002499 | Identification of Seroreactive Proteins of Leptospira interrogans Serovar Copenhageni Using a High-Density Protein Microarray Approach | Leptospirosis is a widespread zoonotic disease worldwide. The lack of an adequate laboratory test is a major barrier for diagnosis, especially during the early stages of illness, when antibiotic therapy is most effective. Therefore, there is a critical need for an efficient diagnostic test for this life threatening disease.
In order to identify new targets that could be used as diagnostic makers for leptopirosis, we constructed a protein microarray chip comprising 61% of Leptospira interrogans proteome and investigated the IgG response from 274 individuals, including 80 acute-phase, 80 convalescent-phase patients and 114 healthy control subjects from regions with endemic, high endemic, and no endemic transmission of leptospirosis. A nitrocellulose line blot assay was performed to validate the accuracy of the protein microarray results.
We found 16 antigens that can discriminate between acute cases and healthy individuals from a region with high endemic transmission of leptospirosis, and 18 antigens that distinguish convalescent cases. Some of the antigens identified in this study, such as LipL32, the non-identical domains of the Lig proteins, GroEL, and Loa22 are already known to be recognized by sera from human patients, thus serving as proof-of-concept for the serodiagnostic antigen discovery approach. Several novel antigens were identified, including the hypothetical protein LIC10215 which showed good sensitivity and specificity rates for both acute- and convalescent-phase patients.
Our study is the first large-scale evaluation of immunodominant antigens associated with naturally acquired leptospiral infection, and novel as well as known serodiagnostic leptospiral antigens that are recognized by antibodies in the sera of leptospirosis cases were identified. The novel antigens identified here may have potential use in both the development of new tests and the improvement of currently available assays for diagnosing this neglected tropical disease. Further research is needed to assess the utility of these antigens in more deployable diagnostic platforms.
| Leptospirosis is an infectious zoonotic disease that causes non-specific signs and symptoms in humans, which hampers the clinical diagnosis and treatment by physicians. Complications can occur if the proper treatment is not initiated early in the course of illness. Although the early diagnosis is critical for preventing unnecessary complications, currently available tests do not exhibit sufficient diagnostic sensitivity in the beginning of disease. We took advantage of high throughput techniques to perform an embracing study of the humoral immune response to the bacteria in order to identify antigens that could be used in a new test for the diagnosis of leptospirosis. A protein microarray chip containing 2,241 leptospiral proteins was constructed and probed with serum samples from patients and healthy individuals. We identified 24 proteins that are recognized by patients' sera but not by healthy individuals. These proteins are potential diagnostic markers, especially the ones identified for acute-phase patients, which can discriminate between a positive and a negative leptospirosis case within a few days after onset of symptoms. This work establishes the protein microarray approach for improving our understanding of the serological response to leptospirosis. Further research is needed to assess the performance of these antigens in the clinical setting.
| Leptospirosis is one of the most common zoonotic infectious diseases worldwide. Humans usually become infected through occupational, recreational or domestic contact with the urine of reservoir animals, either directly or through contaminated soil or water [1]–[3]. Pathogenic leptospires frequently produce an asymptomatic infection in wild rodents and other reservoirs; however, in humans and other accidental hosts, it can cause hepato-renal failure, pulmonary hemorrhage syndrome and even death depending on bacterial virulence and the host immune response [1], [2]. Such complications can be prevented if the proper antibiotic therapy is initiated at the onset of the disease [3]–[6]. Nevertheless, the lack of a rapid and reliable diagnostic test is a major barrier to providing an early diagnosis.
Clinical diagnosis of leptospirosis is hindered by the overlapping clinical manifestations with other febrile illnesses [2], [4], [7]. Therefore, the diagnosis depends upon laboratory tests and different methods have been developed. Recovery of leptospires from clinical specimens such as tissue and blood by culture is considered a definitive diagnosis. This diagnosis is hampered, however, by the slow growth rate, the long incubation period until culture is established, low sensitivity and high cost due to the human and laboratory resources required [2]. Nucleic acid-based amplification techniques to detecting leptospiral DNA in biological specimens have also been developed but sensitivity usually decreases as patient progresses to the late stages of disease [2], [4]. Hence, serology is the most frequently used diagnostic approach for leptospirosis.
The gold standard assay is the microagglutination test (MAT), which may use a panel of 19 live leptospires representing the major serogroups for the detection of agglutinating antibodies [2]–[4], [7], [8]. Despite the high specificity, the MAT usually requires paired acute- and convalescent-phase samples, thus being insensitive in the beginning of the disease. To overcome the drawbacks of MAT, numerous serological assays have been developed, particularly ELISA tests based on either whole cell extracts or recombinant proteins [3], [4], [7]. However, these assays have similar performance characteristics, with sensitivity and specificity rates that match those of MAT. Among other serological approaches whose accuracy has been described are agglutination, dipstick, and lateral flow assays [7]. Together, these assays demonstrated low sensitivity during acute phase, so the need for an efficient method to diagnose early infection remains urgent.
High-density protein microarrays are an effective approach to perform large scale serological studies and define antigen-specific antibody responses to infectious agents on a whole proteome scale. They can be produced and probed in a high-throughput manner, allowing for the screening of hundreds of serum samples thus improving the statistical power and generating more accurate conclusions. Additionally, unlike cell extracts, a set of antigens can be identified with optimal sensitivity and specificity. The aims of this research approach are to understand the breadth, intensity and diversity of the antibody response to leptospirosis disease and to discover novel antigens that can be employed in diagnostic tests and subunit vaccines.
Here we report the results of a study probing more than 250 human serum samples, including healthy controls and leptospirosis cases from the state of Bahia, Brazil, against a partial proteome microarray chip containing 2,421 proteins from Leptospira interrogans serovar Copenhageni strain L1–130, which was isolated in Bahia, Brazil in 1996. The reason for choosing this specific strain relies on the availability of L. interrogans serovar Copenhageni complete genome sequence. Moreover, leptospirosis is an emerging health problem in developing countries. In Brazil, 4832 laboratory-confirmed cases were reported in 2011, distributed among the North (484 cases), Northeast (890 cases), Southeast (1762 cases), South (1673 cases) and Central-West (23 cases) regions [9]. Our group has shown that urban transmission of leptospirosis in Brazil is related to the presence of domestic rats in the environment [10]–[12]. Accordingly, >90% of the leptospirosis cases there are caused by L. interrogans serovar Copenhageni, which is commonly associated with Rattus species reservoirs [10], [11]. The homogeneity of pathogen exposure and availability of sequenced genomic material from a related strain makes this clinical setting ideal for an initial proteomic study.
The study protocol was approved by the institutional review board committees of Yale University and Oswaldo Cruz Foundation. Samples from infected patients and healthy individuals living in a community with high endemic transmission of leptospirosis came from the following projects: “Epidemic Urban Leptospirosis in Salvador, Brazil: A Study of the Clinical Presentation and Development of Rapid Diagnostic Methods” and “Natural History of Urban Leptospirosis”. The participants involved in both projects provided written informed consent. Blood donors from the city of Salvador were anonymous. Sera from U.S. healthy individuals were obtained from anonymous volunteers at the General Clinical Research Center at the University of California, Irvine. After collection, a code number was designated to each patient so that all samples were rendered anonymized for researchers before its use.
The evaluation was performed with a collection of 114 control human serum samples and 160 laboratory-confirmed sera of leptospirosis cases. Control samples were (i) 29 sera from healthy volunteers from California/US, where endemic transmission of leptospirosis does not exist; (ii) 35 sera from blood donors from Salvador/Brazil, city with endemic transmission of leptospirosis and (ii) 50 sera from healthy subjects who were enrolled in a cohort study in a high risk urban slum community in the same city [12]. Cases were identified during active hospital-based surveillance in the same state of the slum community, including patients from the city of Salvador and from the country side, from April 1996 to August 2010. During this period, 1529 MAT-confirmed cases of severe leptospirosis were identified, of which we selected 80 acute- and 80 convalescent-phase sera to conduct this study. Serum samples were randomly selected and therefore acute and convalescent samples are not necessarily paired. Acute-phase samples were collected upon patient admittance at the hospital and convalescent-phase samples were collected from recovering patients at least 14 days after hospital admittance and that may or may not have received standard antibiotic therapy. Laboratory confirmation was defined according to the criteria for seroconversion, a four-fold rise in titer or a single titer of 1∶800 in the MAT.
Selection of the ORFs that would compose the array was performed considering the Leptospira interrogans serovar Copenhageni strain Fiocruz L1–130 genome annotations available at National Center for Biotechnology Information (NCBI) and at John Craig Venter Institute (JCVI) databases. The criteria used included proteins with potentially biological importance [13], [14] and also with potential antigenic features [15]–[18] (Table S1).
The selected ORFs were attempted to be amplified by PCR and cloned into pXI vector using a high-throughput PCR recombination cloning method described elsewhere [19]. Briefly, ORFs were amplified using 5 ng of L. interrogans serovar Copenhageni strain Fiocruz L1–130 with Accuprime Taq DNA Polimerase System (Invitrogen) according to the manufacturer's protocol. Cycling conditions were as follows: 94°C-2 min, 31 cycles of 94°C-90 s, 55°C-15 s, 50°C-15 s, 68°C-2 min and a final extension of 68°C-10 min. Primers contained a 20 bp ORF-specific sequence and a unique 20 bp “adapter” sequence, which becomes incorporated into the 5′ and 3′ termini flanking the amplified gene and is homologous to the cloning sites of the linearized pXI vector (ACGACAAGCATATGCTCGAG and TCCGGAACATCGTATGGGTA, respectively). Genes larger than 3 kb were cloned as smaller segments, maintaining an overlap of at least 150 nt between the sequences, since high throughput cloning efficiency declines when genes are larger than ∼2,500 bp. The segmented ORFs were named with the gene ID followed by the letter “s” and the number of the segment, e.g. LIC10502-s4. The ligA and ligB genes (LIC10465 and LIC10464, respectively) were fragmented considering the repeated Big domains present in the proteins' structures (LigB Repeats 7–12, LigA Repeats 7–13 and LigA/B Repeats 1–6) [20], which have been previously described as diagnostic markers and/or vaccine candidates [21]–[24]. Up to 3 additional rounds of amplification were attempted for failures, which were usually recovered by adjusting the PCR conditions. All PCR reactions were confirmed for correct insert size by gel electrophoresis before cloning.
The pXI plasmid encodes an N-terminal 6×His-tag and a C-terminal hemagglutinin (HA) tag. The plasmid was linearized by digestion with BamH1 and amplified by PCR to generate the acceptor vector as described previously [19]. A reaction containing 40 ng of linearized pXI vector, 1 uL of ORF PCR reaction and 10 uL of super-competent Escherichia coli DH5-α cells (McLab) was incubated on ice for 30 min, heat-shocked at 42°C for 1 min and chilled on ice for 1 min. One hundred and eighty microliters of S.O.C medium were added and cells were cultured for 1 hour at 37°C. The entire reaction mixture was added to 1.1 mL of LB supplemented with kanamycin 50 ug/mL and incubated overnight at 37°C with vigorous aeration. Plasmids were extracted with QIAprep 96 Turbo Kit (Qiagen) without colony selection and analyzed by gel electrophoresis to confirm insert size. Up to 2 additional rounds of cloning were performed to increase efficiency and were resumed by doubling the PCR volume for transformation. All plasmids carrying inserts <500 bp and some randomly selected ones were confirmed for insert presence by PCR using the insert specific primers. After probing the microarrays with the serum samples, the seroreactive antigens were identified and the corresponding plasmids were sequenced. The insert was confirmed in all cases.
For array fabrication, purified minipreparations of DNA were used for expression in an E. coli based in vitro transcription-translation (IVTT) reaction system (RTS Kit, Roche) according to the manufacturer's instructions. Ten-microliter reactions were performed in 384-well plates and incubated for 16 hours at 26°C under 300 rpm shaking. Control reactions were performed in the absence of DNA (“NoDNA” controls) to assess the background given by the IVTT reaction itself. Protease inhibitor mixture (Complete, Roche) and Tween-20 to a final concentration of 0.5% v/v were added to the reactions, which were then mixed and centrifuged to pellet any precipitates and remove bubbles prior to printing. Unpurified supernatants were immediately printed onto nitrocellulose coated glass FAST slides using an Omni Grid 100 microarray printer (Genomic Solutions). In addition, arrays were printed with multiple negative control reactions, positive control spots of an IgG mix containing mouse, rat and human IgGs (Jackson ImmunoResearch) and purified Epstein-Barr Virus Nuclear Antigen 1 (EBNA1) protein, which is recognized by the majority of humans thus serving as a marker for serum quality.
Protein expression was verified by probing the array with monoclonal anti-polyhistidine (Sigma Aldrich) and anti-hemaglutinin (Roche Applied Science) against the respective tags. First, arrays were blocked for 30 min with Protein Array Blocking Buffer (Whatman) and probed overnight with anti-tag antibodies diluted 1/400 in Blocking Buffer. Arrays were then incubated for one hour in biotinylated secondary antibodies (Jackson ImmunoResearch) diluted 1/1000 followed by one-hour incubation with streptavidin-conjugated SureLight P3 (Columbia Biosciences). After each incubation, slides were washed 3 times with Tris-buffered saline containing Tween-20 0.05% v/v (TTBS). Additional washes with TBS and distilled water were performed and the slides were air-dried by brief centrifugation before scanning. Slides were scanned in a Perkin Elmer ScanArray confocal laser and intensities were quantified using QuantArray package.
For probing with human serum, samples were diluted 1/100 in Protein Array Blocking Buffer containing E. coli lysate 10 mg/mL (McLab) at a final concentration of 10% v/v and incubated for 30 min at room temperature under constant mixing to remove background reactivity to E. coli proteins in the IVTT reactions. E. coli protein-antibody complexes were removed from the sample dilution mix via centrifugation prior to addition to the microarray. Arrays were blocked for 30 min with Protein Array Blocking Buffer and then incubated with diluted samples overnight at 4°C, with gentle rocking. Biotinylated anti-human immunoglobulin G (Fc-γ fragment specific, Jackson ImmunoResearch) was diluted 1/2000 in Blocking Buffer and added to the arrays for one-hour incubation at room temperature. Slides were washed 3 times with TTBS after each incubation and bound antibodies were detected by one-hour incubation with streptavidin-conjugated SureLight P3, as described above. Finally, slides were scanned for intensity quantification.
Eleven clones, corresponding to the 10 most differentially reactive antigens for either acute or convalescent groups (see results), were submitted to a five-hour IVTT reaction (RTS, Roche) according to the manufacturer's instructions. Protease inhibitor mixture (Complete, Roche), Tween-20 and methanol were added to final concentrations of 0.5% and 10% v/v respectively. The reactions were mixed and centrifuged to remove bubbles. Unpurified supernatants were printed on Hi-Flow Plus HF240 membrane (Millipore) using a BioJet dispenser (BioDot) at 1 uL/cm and cut into 3 mm strips. Individual strips were then blocked in TTBS 5% non-fat milk for 30 min. Sera samples were diluted 1/250 in TTBS 5% nonfat milk containing E. coli lysate at a final concentration of 20% v/v and incubated for 30 min at room temperature under agitation. Blocked strips were then incubated with diluted sera during 1 hour and washed 6 times with TTBS. Alkaline phosphatase-conjugated anti-human IgG (Jackson ImmunoResearch) was diluted 1/5000 in TTBS 5% nonfat milk and applied to each strip for 1 hour at room temperature under agitation. After washing 6 times with TTBS, 3 additional washes with TBS were performed and reactive bands were visualized by incubation with 1-step Nitro-Blue Tetrazolium Chloride/5-Bromo-4-Chloro-3′-Indolyphosphate p-Toluidine Salt (NBT/BCIP) developing buffer (Thermo Fisher Scientific) for 2 min at room temperature. Enzymatic reaction was stopped with tap water and the strips were air-dried before scanning at 2,400 dpi (Hewlett-Parckard scanner). Images were converted to gray scale and band intensities were quantified using the ImageJ software (found at http://rsbweb.nih.gov/ij/).
Spot intensities were quantified using QuantArray software. Raw data were obtained as the mean pixel signal intensity for each spot and all intensities were automatically corrected for spot-specific background. For each array, the average of control IVTT reactions (NoDNA controls) was subtracted from spots' signal intensities in order to minimize background reactivity. Proteins were considered to be expressed when signal intensity for either tags was above the NoDNA control reactions mean plus 2.5 standard deviations. The same cut-off was applied to identify the reactive proteins using the sera collection. Data analysis was performed using the R statistical software (found at http://www.r-project.org). To stabilize the variance, VSN normalization was applied to the raw data and groups were compared by a Bayes regularized t test adapted from Cyber-T for protein arrays [25], [26]. Benjamini and Hochberg (BH) method was used to control the false discovery rate [27] so that p-value smaller than 0.05 was considered significant and the corresponding protein was considered differentially reactive. For plotting the histogram, BH corrected p-values smaller than 1E-14 were assigned as 1E-16. Multiplex classifiers were generated using linear and nonlinear Support Vector Machines (SVMs) using the “e1071” R package. SVM is a supervised learning method that has been successfully applied to microarray data characterized by small samples sizes and a large number of attributes. The SVM approach, as any other supervised classification approach, uses a training dataset to build a classification model and a testing set to validate the model. To generate unbiased training and testing sets, leave one out cross-validation (LOOCV) was used. With this methodology, each data point is tested with a classifier trained using all of the remaining data points. Plots of receiver operating characteristic (ROC) curves were made with the “ROCR” R package. Sensitivity and specificity were determined from the resulting ROC curves. Clinical characteristics of the leptospirosis patients whose acute and/or convalescent serum samples were selected for this study were described using frequencies and medians with interquartile (IQR) ranges (Table S2). The Chi square test or the Mann-Whitney/Wilcoxon test was used to compare clinical presentations of acute-phase leptospirosis patients with convalescent-phase patients. An association between patients' clinical characteristics and the intensity of acute sera signal against the three antigens that presented the best performance in the protein microarray were evaluated by the Kruskal-Wallis test.
The raw and normalized array data used in this study have been deposited in the Gene Expression Omnibus archive (http://www.dtd.nlm.nih.gov/geo/), accession number GSE42720.
Characterization of the serological response to Leptospira exposure and infection on a whole proteome scale with protein microarrays has not been previously done. To evaluate the feasibility of this approach for leptospirosis, we identified a subset of proteome more likely to be immunoreactive. The selection criteria used to choose the proteins included in the array provided 2,241 ORFs, which corresponded to 61% of Leptospira interrogans proteome. The basis for selecting this particular subset of proteins took advantage of empirical mass spectrometry and RNA expression data available for Leptospira interogans [13], [14] and also from proteome microarray data from other Gram negative bacteria [17], [18] (Supplementary Table S1).
In total, the array contained 2361 antigens, including full length proteins and protein segments. Protein expression was evaluated by probing the array with anti-His and anti-HA, and over 97% of protein spots were confirmed positive for either His or HA tags (Figure S1A).
Sera used in this study were classified into 5 groups, summarized in Table 1 and described in the methods section. Table S2 shows the clinical characteristics of the leptospirosis patients who provided sera for this study. The majority of them (88%) were male and the median age was 34 (IQR: 24–45) years old. Median duration of symptoms before hospitalization was 6 (IQR: 5–8) days. Jaundice and acute respiratory distress syndrome occurred in 87% and 13% of the patients, respectively. Renal impairment was frequent (median creatinine: 4.0 [IQR: 2.0–6.4] mg/dL) and 30% of the patients received peritoneal or hemodialysis. Intensive care was provided for 20% of the patients and 3% died.
Representative microarray images of L. interrogans infected and control samples are shown in Figure S1B. The heatmap in Figure 1 gives an overview of the reactivity of the 42 reactive antigens for each of the 239 individual samples. Brazilian blood donors are not shown in this figure. Individual specimens are in columns and grouped by healthy controls from USA, healthy controls from the high endemic area group, acute-phase patients and convalescent-phase patients. The antigens, in rows, are organized according to those that are significantly more reactive in the cases than in the healthy controls. These antigens are termed ‘differentially reactive’ (DR) and are divided in 3 sections: antigens identified as differentially reactive for both acute- and convalescent-phase patients, antigens identified as differentially reactive only for acute patients and differentially reactive antigens only for convalescent patients. There is a second set of antigens that were equally as reactive in healthy controls and the cases, and these antigens are termed ‘cross-reactive’ (CR). Although there was some reactivity seen in the healthy controls against the differentially reactive antigens, there was more IgG response against these antigens after acute infection, and still more in the convalescent specimens. The background reactivity seen from the cross-reactive antigens was similar between all three groups.
Here, we aim to identify antigens that can discriminate between positive and negative leptospirosis cases and for that we based our analysis on comparing acute and convalescent-phase patients to healthy individuals from an area with high endemic transmission (Figure 2). Since healthy individuals living in this area show some background reactivity to leptospiral LPS [12] and proteins (Figure 3, described later in this section), we find that the identification of antigens with sero-reactivity among patients but not among those healthy individuals distinguish a current leptospirosis case. All the high endemic controls used in this study were MAT-negative for leptospirosis and in order to avoid bias in our analysis, we compared the IgG reactivity detected on the microarray by probing 10 MAT-positive and 10 MAT-negative healthy individuals living in this area. The overall reactivity seen for both groups was low (Figure S2 A) and most of the reactive antigens detected for infected patients (described later in this section) were not reactive (average signal intensity below the cut-off, Figure S2 B) for either MAT-positive or MAT-negative healthy individuals. Therefore, we used the MAT-negative high endemic controls for the following analysis.
There were 30 reactive antigens, ∼1.3% of all of the antigens printed on the array, of which 18 detected significantly more IgG antibody in the convalescent samples compared to control individuals from the high endemic area group (Table S3). For the acute-phase samples, the IgG antibody response detected 35 seroreactive antigens or 1.5% of the array, of which 16 discriminate between acute and negative cases. LipL32, LigA Repeats 7–13 and LigB Repeats 7–12 antigens were the three most reactive targets on average for both convalescent- and acute-phase groups. Ten differentially reactive antigens overlap between the acute and convalescent groups.
In order to investigate background reactivity among healthy individuals living in an area with endemic transmission of leptospirosis, we compared the cumulative antigen reactivity for the 3 control groups, from USA, Brazilian blood donors and healthy individuals from the high endemic area groups. The heatmap in Figure 3A shows the reactivity of all antigens with average signal intensity above the cut-off for any of the control groups. We observed a higher overall reactivity in the high endemic area group compared to USA controls and Brazilian blood donors. Accordingly, when we analyzed the cumulative signal intensity against all antigens on the array (Figure 3B), USA healthy subjects showed the lowest total reactivity followed by blood donors from Salvador and healthy individuals from high endemic area. Blood donors living in endemic area had slightly higher reactivity than USA naïve subjects, but the difference was not statistically significant. However, the total background reactivity in healthy individuals residing in the area with high endemic transmission was significantly greater (p<0.05) than either the blood donors from Brazil or the USA controls.
Finally, we compared the average signal intensity of all the reactive antigens for each patient to the patient's MAT titer. MAT is based primarily on agglutinating antibodies that bind to leptospiral LPS [28], [29] and does not differentiate between IgM and IgG subtypes. All acute and convalescent samples used in this study were laboratory confirmed for infection by MAT and we observed a 3-fold increase in the median titer for convalescent samples compared to the acute group (from 800 to 3,200, Table 1). Although we have also observed a general increase in antigen signal intensities for the convalescent group compared to acute group (Figure 1 and Figure 2), we were unable to draw a correlation between these two approaches (Figure S3) indicating that MAT antigen and protein antigens identify different antibody pools in these patients.
To determine the accuracy of the differentially reactive antigens in distinguishing a leptospirosis case, individual antigen ROC curves were generated and the AUC for each antigen was determined. Acute and convalescent-phase samples were analyzed separately against the high endemic area control group and sensitivity and specificity were calculated for both groups using the SVM computational approach. Antigens were then ranked by decreasing AUC and multiple antigens ROC curves generated. Single antigen ROCs for acute-phase group are shown in Figure 4A and for convalescent-phase group are shown in Figure S4. For both cases, the false positive rate was calculated considering the high endemic area healthy control group.
For acute-phase patients, the non-identical domains of the Lig proteins (LigA Repeats 7–13 and LigB Repeats 7–12) provided best sensitivity and specificity (AUC = 0.894-0.857), followed by LipL32 (LIC11352, AUC = 0.841, Table 2). As disease progresses to convalescence, the accuracy of these antigens increases so that LipL32 achieves best performance (AUC = 0.986) followed by LigA Repeats 7–13 (AUC = 0.965) and LigB Repeats 7–12 (AUC = 0.968, Figure S4). None of the three antigens with better accuracy (LigA Repeats7–13, LigB Repeats7–12 and LipL32) had the signal intensities from the acute serum sample associated with patients' clinical characteristics (Table S4). A heat shock protein of the GroEL family (LIC11335) was also identified as seroreactive, with high sensitivity for both acute- and convalescent-phase patients (90.0% and 92.0%, respectively) but low specificity (53.8% and 62.5%). DnaK (LIC10524), another heat shock protein, showed seroreactivity for the convalescent group, although we could not detect significant levels of IgG against this antigen in the acute group (average signal intensity below the established cut-off). The virulence-associated protein Loa22 (LIC10191) showed very low sensitivity for acute-phase patients (36.0%) and was considered not seroreactive for the convalescent group. Similarly, the IgG response against LipL31 (LIC11456) was detected only among acute patients, with a diagnostic accuracy of 82% sensitivity and 68.8% specificity.
Several novel antigens, for which no seroreactivity has been previously described, were identified in this study. The hypothetical protein LIC10215 provided 92.0% and 86.0% sensitivity and 67.5% and 83.8% specificity for distinguishing healthy from either acute- or convalescent-phase patients, respectively. LIC10215 was the best antigen for distinguishing an acute case from a healthy individual after the domains of the Lig proteins and LipL32. Regarding the convalescent group, LIC20087, antigen annotated as outer membrane protein, provided best accuracy after the domains of the Lig proteins and LipL32, with 96.0% sensitivity and 86.3% specificity (Table 2).
The combination of 11 differentially reactive antigens allowed for best sensitivity and specificity for the acute cases (78.0% and 87.5%, respectively) whereas the combination of 4 antigens provided best accuracy (98.0% sensitivity and 94.0% specificity) for convalescent cases (Figure 4B).
Eleven differentially reactive antigens, corresponding to the most significant antigens for either acute- or convalescent-phase groups were printed onto a nitrocellulose membrane and cut into 3 mm strips which were probed with 20 highly endemic, 20 acute and 20 convalescent randomly selected samples. Healthy individuals showed lower reactivity against these antigens whereas leptospirosis patients reacted strongly against most of the antigens (Figure 5). Antigen intensities were quantified and groups were compared using Bayes regularized t test adapted from Cyber-T. A total of 6 antigens with significant BHp-values (BHp<0.05) were identified as differentially reactive for both acute and convalescent groups, of which 4 overlap (Table S5). For both acute- and convalescent-phase groups, the domains of the Lig proteins provided the best single antigen discrimination, followed by LipL32. LIC10215, LIC10486, LIC11271, LIC20087 and LIC11573 showed no sero-reactivity on immunostrips. The lower reactivity observed for these proteins on immunostrips may be due to technical differences between both platforms.
Protein microarrays are a powerful tool to describe pathogen-specific antibody responses produced after exposure to infectious agents. Our group has applied this approach to more than 25 agents of medical relevance, including viruses, bacteria, protozoan and helminthes and some of the antigens identified by our methodology were successfully employed in different diagnostic platforms [18], [30]–[33]. No currently available approach enables such a complete understanding of the humoral immune response to infection. Here, we constructed a protein microarray comprising 2,421 proteins, 61% of the proteome of L. interrogans serovar Copenhageni, to examine the IgG response to leptospirosis. Our focus in the present study was to profile the immune response associated with leptospirosis exposure and infection, and to identify seroreactive and serodiagnostic antigens.
Our results showed distinct IgG reactivity against dozens of differentially reactive leptospiral antigens in both acute- and convalescent-phase sera. The high reactivity detected in most of the acute-phase patients led us to speculate how the IgG response could rise so quickly after infection. The first exposure to an infectious agent in a previously naïve individual is expected to take 10–14 days before mounting an IgG response and the onset of symptoms vary according to the pathogen's incubation period. The incubation period for leptospirosis ranges from as few as 2 to as many as 30 days and the onset of symptoms usually comes together with the appearance of agglutinating antibodies, which increase with disease progression [1]. In this study, the acute-phase patients had a mean of 6 (IQR 5–8) days of symptoms onset before blood sampling and no correlation was observed between IgG reactivity and numbers of days of symptoms before sample collection (Table S4). Therefore, we speculate that the symptomatic individuals with less reactivity in the acute group may have experienced a shorter incubation period before becoming symptomatic compared to those with a broader and more intense response. Alternatively, rapid onset of the IgG responses in acutely infected subjects may be an anamnestic response from a previous clinical or subclinical exposure to the organism. Previously exposed individuals can produce antibody more rapidly from the memory pool within a few days post-exposure.
Here we showed that healthy individuals living in an area with endemic transmission of leptospirosis have higher antibody responses than those from outside the endemic environment. Previous exposure can lead to background reactivity and false positive results, interfering with identification of true active leptospirosis cases especially among those individuals living in areas with endemic transmission. It has been previously reported a 15% overall prevalence of anti-leptospire antibodies detected by MAT in healthy individuals living in that urban slum community (high endemic area group) [12]. Most of the antibodies detected by MAT are directed against leptospiral LPS. Here, we show that MAT-negative healthy individuals living within a community with high endemic transmission of leptospirosis present higher overall seroreactivity against leptospiral proteins than healthy individuals from outside the endemic area, suggesting that protein antigens may also play a role in background reactivity. The shifts in background reactivity between groups of healthy individuals are small compared to the large increases in reactivity seen after acute infection and convalescence. Our results also show that the reactivity against the proteins on the chip doesn't differ between MAT-positive and MAT-negative healthy individuals.
Despite the background reactivity seen for the high endemic area group, we were able to identify several individual antigens that were differentially reactive for acute- and/or convalescent-phase patients when compared to that control group. These antigens can be considered for use alone in single antigen ELISAs or together in a multiplex assay. The diagnostic accuracy was assessed when several antigens were used together in combination. The most accurate test results to distinguish acutely infected subjects from controls were obtained when 11 antigens were combined together and 14 antigens, for convalescent cases. The use of a minimal set of antigens in an assay would represent the best option in terms of production complexity and manufacturing costs. However, our group has previously shown that the addition of antigens can reduce the effect of noise in the data introduced from variables such as executing it in different locations, at different times and by different operators [34]. A multiplex test using several antigens could minimize the effect of these variables and justify the development of a more robust assay of this kind.
Five of the leptospiral proteins identified here have been previously reported reactive in patients' sera including the non-identical domains of the Lig proteins, LipL32, chaperonin GroEL, DnaK and Loa22 [20], [35], [36]. Different platforms have been developed to employ the Lig proteins as serodiagnostic markers for human leptospirosis with promising results [20], [24], [37], [38]. Lig-based immunoblot assays for IgM detection showed superior performance than MAT and superior performance than a commonly used whole-cell ELISA in Brazil during early acute phase [21]. A new Lig-based rapid serological test, the DPP assay, was recently developed and also outperformed the whole-cell IgM ELISA assay for severe acute cases, particularly for patients tested early in the course of the disease [24]. For LipL32, GroEL, DnaK and Loa22, however, the findings were not as encouraging [34], [39], [40], even though LipL32 in combination with LipL21 and OmpL1 [41] improved its diagnostic performance in ELISA platforms. The identification of these previously reported reactive antigens is proof-of-concept for the protein microarray antigen discovery platform.
In this study, the well-known antigens LipL32 and the non-identical domains of the Lig proteins had the best sensitivity and specificity of all antigens probed. The next best differentially reactive antigen for detecting acute-phase patients was the novel hypothetical protein LIC10215. Several other hypothetical proteins also found to be differentially reactive antigens identified in this work were LIC11222, LIC11955, LIC10486, LIC11271, LIC10483 and LIC20301. Although no previous functions have been assigned to these proteins, here we show that they are part of the L. interrogans immunoproteome and can elicit a host immune response as they are recognized by sera from infected subjects. We also discovered numerous differentially reactive antigens that are not hypothetical and have been functionally annotated including LIC20042 (BatC), LIC11889 (FlbB), LIC11573 (GspG), LIC12180 (methyltransferase), LIC11456 (LipL31), LIC11437 (adenylate/guanylate cyclase), LIC12544 (DNA binding protein), LIC20087 (outermembrane), LIC10623 (MotB), LIC11570 (GspD).
The results reported here were from a protein microarray derived from one leptospire serovar, L. interrogans serovar Copenhageni, probed with sera from acute- and convalescent-phase patients from a well-characterized model epidemiological setting in Salvador [10]–[12]. This study was limited by the restricted number of antigens selected for the array and also by the prevalence of one specific serovar at our study site. Further research is needed to investigate the diversity of the antibody profile after exposure to different serovars. All the samples used here corresponded to hospitalized leptospirosis patients, but the immune response may be different for mild presentations. Finally, we recognize the importance of also evaluating the IgM antibody response to understand the kinetics of the humoral immune response.
In other protein microarray studies of kind we have found that proteins are not randomly selected for recognition by the immune system and antigens share proteomic features that increase their likelihood to be seroreactive and serodiagnostic [15], [16]. Interrogating the antibody response in a whole proteome scale allows molecular features related to antigenicity to be classified. Proteomic feature enrichment analysis for antibody recognition of leptospiral antigens will be the focus of a separate study using the full leptospire proteome consisting of 3,667 proteins, in which we will also assess the IgM reactivity profile to leptospirosis. We also aim to probe with more diverse specimen collections worldwide to better characterize the antibody repertoire against different leptospire species and serovars, and from different mammalian hosts.
In summary, we reported a protein microarray approach for L. interrogans serovar Copenhageni and discovered a limited set of 24 differentially reactive antigens. The antigens identified could be applied to improve the accuracy of rapid tests to diagnose leptospirosis in resource-limited settings. The results show that this is a feasible approach that can be applied in the future to study the humoral immune response in other epidemiological settings worldwide, to examine the antibody response after exposure to different leptospire species and determine the antibody profiles elicited by the pathogen in domestic animals and reservoir hosts.
|
10.1371/journal.pntd.0001728 | Trypanosoma brucei gambiense Group 1 Is Distinguished by a Unique Amino Acid Substitution in the HpHb Receptor Implicated in Human Serum Resistance | Trypanosoma brucei rhodesiense (Tbr) and T. b. gambiense (Tbg), causative agents of Human African Trypanosomiasis (sleeping sickness) in Africa, have evolved alternative mechanisms of resisting the activity of trypanosome lytic factors (TLFs), components of innate immunity in human serum that protect against infection by other African trypanosomes. In Tbr, lytic activity is suppressed by the Tbr-specific serum-resistance associated (SRA) protein. The mechanism in Tbg is less well understood but has been hypothesized to involve altered activity and expression of haptoglobin haemoglobin receptor (HpHbR). HpHbR has been shown to facilitate internalization of TLF-1 in T.b. brucei (Tbb), a member of the T. brucei species complex that is susceptible to human serum. By evaluating the genetic variability of HpHbR in a comprehensive geographical and taxonomic context, we show that a single substitution that replaces leucine with serine at position 210 is conserved in the most widespread form of Tbg (Tbg group 1) and not found in related taxa, which are either human serum susceptible (Tbb) or known to resist lysis via an alternative mechanism (Tbr and Tbg group 2). We hypothesize that this single substitution contributes to reduced uptake of TLF and thus may play a key role in conferring serum resistance to Tbg group 1. In contrast, similarity in HpHbR sequence among isolates of Tbg group 2 and Tbb/Tbr provides further evidence that human serum resistance in Tbg group 2 is likely independent of HpHbR function.
| Human African Trypanosomiasis, or sleeping sickness, is caused by two different parasites: Trypanosoma brucei gambiense (Tbg) and T. b. rhodesiense (Tbr). Each parasite employs a different mechanism to resist trypanosome lytic factor (TLF), the active innate immune component of human serum. In Tbg group 1, which causes the vast majority of disease cases, the mechanism is thought to involve the reduced activity of a receptor involved in binding and internalizing TLF. In this study, we investigate genetic variation in this receptor across a broad geographic sample of Tbg and closely related trypanosomes to test whether unique polymorphisms in the receptor from Tbg may explain its altered function. We identified a single mutation in all copies of the receptor gene sequenced from Tbg but not in any other closely related species. This finding suggests that this single mutation could play a key role in conferring human infectivity to Tbg. Given the possible consequences for drug development and diagnostics, we suggest that future functional studies target this mutation to fully elucidate its role.
| Trypanosomiasis, a deadly disease of humans and livestock in sub-Saharan Africa, is caused by protozoan parasites of the genus Trypanosoma, which are transmitted between mammalian hosts by insect vectors of the genus Glossina (tsetse). Human-infective members of the Trypanosoma brucei complex cause the human form of the disease, Human African Trypanosomiasis (HAT), or sleeping sickness. T. b. rhodesiense (Tbr) causes an acute form of HAT in eastern Africa, while T. b. gambiense group 1 (Tbg1) causes a chronic form of the disease in western and central Africa and accounts for over 90% of reported cases (Figure 1a). T. b. gambiense group 2 (Tbg2), a rare form described from West Africa in the 1970s and 1980s, also causes human disease but the trait of human-infectivity is not stable [1], [2], [3]. The final member of the brucei complex, T. b. brucei (Tbb), is not infective to humans, but, together with other animal trypanosome species, causes the livestock wasting disease, Nagana, across a range that overlaps with that of the human-infective parasites.
Humans possess an innate resistance to some trypanosomes through the action of trypanosome lytic factors (TLFs) in their serum [4]. TLF-1 is a high-density lipoprotein complex that includes the active toxin apolipoprotein L-I (apoL-I) in association with haptoglobin-related protein (Hpr). In the primary immune pathway [5], [6], [7], TLF-1 is bound and internalized via a haptoglobin haemoglobin receptor (HpHbR) on the surface of susceptible trypanosomes. Uptake of TLF-1 is followed by disruption of the lysosomal membrane by apoL-I and eventual cell lysis. While Tbb is susceptible to lysis by human TLF-1, Tbr, Tbg1 and Tbg2 are resistant. In Tbr, the serum-resistance associated (SRA) protein confers resistance to TLF-1 [8] by binding directly to apoL-I after it has been internalized into the cell, inhibiting its lysosome-lytic capacity [9]. Tbg1 and Tbg2, on the other hand, lack the gene encoding SRA and are thought to have evolved an independent mechanism to prevent lysis by TLF.
In Tbg2, apoL-I is also internalized, but lysis is prevented by an unidentified mechanism [10]. In Tbg1, the mechanism is better understood and appears to involve reduced expression and altered function of the parasite HpHbR [11]. Sequencing of a few isolates of Tbg and Tbb led Kieft et al. [11] to suggest that mutations in HpHbR may have altered TLF-1 binding in Tbg1. Specifically, the authors identified five non-synonymous substitutions shared by the four sequenced isolates of Tbg1, but not present in two Tbb isolates. This work has helped to narrow the universe of possible structural differences in HpHbR that could, for example, eventually be exploited to design novel drugs to overcome Tbg1 resistance. However, the small number of isolates examined to date is not sufficient to determine whether the mutations are really Tbg1-specific. While genetic variation in Tbg1 is extremely limited [12], [13], the remainder of the T. brucei complex exhibits relatively high variation, most of which does not partition into neatly defined geographic or taxonomic units [14], [15], [16], [17]. Thus, characterizing the genetic differences that contribute to a critical epidemiological trait such as human infectivity requires that those differences be evaluated in a comprehensive geographical and taxonomic context.
In the present study, we tested if the five non-synonymous substitutions previously hypothesized to alter HpHbR activity in Tbg1 [11] are both conserved in Tbg1 isolates and also absent from other T. brucei subspecies by examining HpHbR gene variation in T. brucei s.l. sampled across the entire range of the species complex. By narrowing the pool of substitutions that are specific to Tbg1, we expect to facilitate future functional studies aimed at understanding the contribution of HpHbR to conferring human serum resistance.
Isolates of Tbb, Tbg1, Tbg2 and Tbr, were selected to incorporate representative genetic diversity from the entire geographic range of the T. brucei complex (Table S1, Figure 1b). When available, we included isolates of all co-occurring taxa from each country sampled (Figure 1b).
For each isolate, DNA was extracted as described in [17]. PCR was performed using primers designed from Tbb (TREU927) and Tbg1 (Dal972) TriTrypDB database sequences (Tb927.6.440 and Tbg972.6.120, respectively) to amplify a 1297 base pair (bp) fragment that encompassed the entire HpHbR gene (HpHbR_F 5′ CGGGAAAGTTGTACGCAAG, HpHbR_R2 5′ CGACCACTTAATGTTACGAGG). For each PCR, 2–4 µL of a 1∶10 dilution of DNA extract were used. PCR reactions were performed using the reagents provided with GoTaq® DNA Polymerase and Green Master Mix. Difficult templates were amplified using Failsafe PCR 2X PreMixes Buffers (Epicentre Biotechnologies, Madison WI). All PCR reactions used the following cycle: Initial denaturation 95°C for 2 min, 50 cycles of 95°C for 35 s, 58°C for 35 s, and 72°C for 1 min 20 s and a final extension at 72°C for 7 min. PCR success was verified with 1% agarose gel electrophoresis. PCR products were purified and then sequenced (Yale DNA Analysis Facility) using two internal primers located in the middle of the sequence (HpHbR_F2in 5′ TGCTCGAGATATTCCTCAAG, HpHbR_Rin 5′ CTCCCACTGAAGCATTAGAC). The sequenced fragment included 22 nucleotides upstream of the HpHbR start codon, the entire HpHbR gene and 62 nucleotides downstream of the HpHbR stop codon.
Sequences generated using the internal primers overlapped by approximately 200 bp permitting the assembly of an entire contiguous sequence of the HpHbR gene. Contiguous sequences were constructed and chromatograms from each isolate were manually examined for double peaks using the CLCBio DNA Workbench 5.7 (Cambridge, MA). Sites with double peaks were assigned the appropriate nucleotide ambiguity code. Sequences were aligned manually using MacClade 4.08 [18].
Samples with double peaks were considered heterozygotes. We used the programs SeqPhase [19] to format files and PHASE 2.1.1 [20] to resolve individual alleles from heterozygous sequences. To assess evidence for recombinant alleles and to relax the assumption of a stepwise mutation model, we employed the recombination model (MR) and the parent-independent models, respectively. Each run used 1000 iterations and a burnin of 500 generations and thinning interval = 1. The dataset was run twice with different random starting seeds and checked for consistency. The replicate with the best average goodness-of-fit was selected for subsequent analyses.
Nucleotide DNA sequence alignments were generated from phased alleles in MacClade 4.08. Haplotype networks were constructed in the program TCS [21]. DNA sequences were translated to amino acids and aligned in MacClade 4.08. Non-coding regions were removed from sequences and amino acid sequences were compared to those generated by [11].
We collected 1296 bp of sequence from each of 65 T. brucei isolates: 32 from Tbb, 15 from Tbg1, five from Tbg2 and 13 from Tbr. In addition, we generated sequence for one isolate each of Trypanosoma equiperdum and Trypanosoma evansi (Table S1), both of which are also members of the subgenus Trypanozoon but are not human infective (reviewed in [3]).
Of the 67 isolates sequenced in this study, 30 were heterozygous at the locus sequenced. PHASE 2.1.1 inferred a total of 34 alleles present in the 67 isolates. For all heterozygotes, allele pairs had Bayesian posterior probabilities of 1.0 across replicate runs, indicating that no alternative allele sequences could be inferred from the heterozygotes.
The 34 alleles recovered in this study exhibited a total of 40 variable sites, of which four were located outside the HpHbR coding region. Each of these four sites occurred in a distinct allele (f2, c, u3, z1) across a total of five isolates (Boula (Tbg1), STIB338 (Tbr), STIB386 (Tbg2), STIB777AE (Tbb), and KP13 (Tbb)). The remaining 36 variable sites were found within the coding region of HpHbR (File S1).
Most allelic diversity (28 alleles) was found in isolates of Tbb, Tbr and Tbg2 and much of this diversity was common to two or more taxa. Five of the seven alleles recovered from Tbr were identical to those found in Tbb. Likewise, four of the five alleles recovered from Tbg2 were also identical to alleles in Tbb. The most common allele in this study (u1; Table S1) was recovered from Tbb, Tbr and Tbg2. In contrast to these observations, we recovered four distinct alleles from Tbg1, but none of these were shared with any member of the subgenus Trypanozoon. Allelic diversity in Tbg1 was relatively low. Allele z1 (identical to the Tbg1 sequence reported in Kieft et al. [2010]) was the most common Tbg1 variant and was recovered from 26 of 30 sampled chromosomes. The remaining Tbg1 alleles differed by only one nucleotide from this common variant, z1. Trypanosoma equiperdum sequences were more similar to Tbb and Tbr sequences, though both alleles from T. equiperdum were unique. In T. evansi, alleles were identical to the most common allele found in Tbb, Tbr and Tbg2 (Fig. 1b).
The HpHbR protein consists of 403 amino acids. In silico translation of the DNA sequences of the 34 alleles described above yielded 25 unique protein sequences (Figure 1, Figure S1). Notably, the single nucleotide difference in HpHbR that distinguished all isolates of Tbg1 from all other T. brucei isolates sampled in this study was non-synonymous, resulting in the replacement of leucine with serine at position 210 (L210S; Figure 1b, Figure S1). With one exception, all Tbg1 isolates possessed two copies of HpHbR that coded for just this single amino acid sequence (Z). In the exception, isolate ITMAP020578, one allele coded for amino acid sequence Z and the second allele coded for a second peptide (Y) differing from Z by just one amino acid at position 212. All other variation in HpHbR amino acid sequences partitioned to differences within and among Tbb, Tbr and Tbg2.
The primary goal of this study was to examine the genetic diversity of HpHbR in a broad geographical and taxonomic context to better characterize the mutations that potentially give rise to differences in HpHbR function and that may contribute to the phenotype of human serum resistance observed in Tbg1. An earlier study of HpHbR genetic diversity in a limited sample of parasite isolates identified five non-synonymous substitutions shared by Tbg1, but not found in Tbb, suggesting that these five differences could play an important functional role [11]. By sampling more broadly across the subgenus Trypanozoon and across Africa, we have demonstrated that just one of these substitutions (L210S) is conserved in Tbg1 and also absent from the most closely related trypanosome taxa, all of which are either susceptible to human serum (Tbb) or known to possess an alternative resistance mechanism (Tbr or Tbg2). Although our sample size remains relatively limited compared to the vast number of parasites distributed widely across Africa, the extremely low genetic diversity observed in Tbg1 HpHbR is consistent with prior population genetic studies [12], [13], [17] and we hypothesize that the mutation L210S is likely fixed in the taxon. This could be extended to field-circulating Tbg1 by using either allele specific PCR primers or a restriction fragment length polymorphism (RFLP) that targets the single nucleotide substitution (e.g., enzyme PleI).
To the extent that the unique substitution in Tbg1 HpHbR prevents the uptake of TLF-1, this single amino acid change may play a key role in conferring serum resistance to this parasite. A role for HpHbR in facilitating lytic activity of human serum was originally established by experiments demonstrating that loss of HpHbR in Tbb (through RNA interference or gene knockout) conferred resistance to TLF-mediated lysis [22]. Later work demonstrated that Tbb selected to be TLF-1-resistant exhibited reduced HpHbR expression. Furthermore, the ectopic expression of Tbg1 HpHbR (using an allele identical to the most common Tbg1 allele identified in our study) in these serum resistant Tbb was not sufficient to restore human serum susceptibility, providing evidence for the altered function of Tbg1 HpHbR [11]. Our data indicate that this altered function likely stems from the L210S mutation in Tbg1, a substitution that effects an approximate 20-fold reduction in the affinity of HpHbR for HpHb [23]. Given that L210S appears to be the single mutation that distinguishes Tbg1 HpHbR from the HpHbR of all closely related members of the Trypanozoon subgenus, we hypothesize that this single mutation could play a major role in the serum resistance of Tbg1. However, this mutation is unlikely to be the sole factor. As noted previously, reduced expression levels of HpHbR are also likely to play a role in Tbg1 serum resistance [10], [11]. Also, while HpHbR is likely to be the main route of entry into the cell for TLF-1, poorly characterized alternative routes appear to exist for both TLF-1 and TLF-2, a second HDL particle that also exhibits trypanolytic activity [6]. Finally, an in vitro study has demonstrated that, regardless of receptor function, Tbg1 may be inherently resistant to apoL-1, the active trypanolytic factor in human serum [10]. While HpHbR may only be one component of Tbg1 serum resistance, the possible benefit of designing new drugs targeted to this receptor variant warrants further functional study to fully circumscribe its effect on serum resistance.
In contrast to Tbg1, the mechanism of Tbg2 resistance to human serum is thought to be independent of HpHbR, based on the finding that HpHbR from Tbg2 internalizes TLF-1 at a rate similar to that observed in Tbb and Tbr [10]. While that study included just a single strain of Tbg2 (STIB386), our results, which include data for several additional strains, suggest that this conclusion is likely to hold more broadly in Tbg2. Sequencing of HpHbR indicated that several isolates of Tbg2 shared sequence identity with isolates of both Tbb and Tbr, while exhibiting no overlap with isolates of Tbg1, a result that is consistent with previous surveys of neutral genetic markers [13], [17]. The genetic similarity of HpHbR observed among a large collection of isolates of Tbb, Tbr, and Tbg2 suggests that the function of HpHbR in Tbg2 is more likely to reflect that of Tbb and Tbr than Tbg1 and further supports the conclusion that Tbg2 serum resistance is independent of HpHbR. Our study surveyed only five strains of Tbg2, but even these five strains exhibited substantially more diversity than Tbg1 at both the nucleotide and amino acid level. The genetic variability of HpHbR in Tbg2 reiterates the fact that Tbg2, unlike Tbg1, is not genetically homogeneous and suggests that future studies should consider this diversity when examining functional differences among parasite subgroups.
|
10.1371/journal.pntd.0006298 | Spatio-temporal coherence of dengue, chikungunya and Zika outbreaks in Merida, Mexico | Response to Zika virus (ZIKV) invasion in Brazil lagged a year from its estimated February 2014 introduction, and was triggered by the occurrence of severe congenital malformations. Dengue (DENV) and chikungunya (CHIKV) invasions tend to show similar response lags. We analyzed geo-coded symptomatic case reports from the city of Merida, Mexico, with the goal of assessing the utility of historical DENV data to infer CHIKV and ZIKV introduction and propagation. About 42% of the 40,028 DENV cases reported during 2008–2015 clustered in 27% of the city, and these clustering areas were where the first CHIKV and ZIKV cases were reported in 2015 and 2016, respectively. Furthermore, the three viruses had significant agreement in their spatio-temporal distribution (Kendall W>0.63; p<0.01). Longitudinal DENV data generated patterns indicative of the resulting introduction and transmission patterns of CHIKV and ZIKV, leading to important insights for the surveillance and targeted control to emerging Aedes-borne viruses.
| Over the past decades, Aedes-borne viruses (dengue, chikungunya, Zika) have become a major source of morbidity within urban areas. Worldwide, public health response to these viruses is reactive to the occurrence of symptomatic cases (a small proportion of all infections). Here we used geocoded passive surveillance data to determine if areas of historically persistent dengue transmission fuel the introduction and propagation of other Aedes-borne viruses. This article provides quantitative evidence of the strong spatio-temporal overlap that occurs between dengue, chikungunya and Zika, all transmitted by Aedes aegypti mosquitoes in the city. Additionally, it emphasizes the value of analyzing long-term geo-coded passive surveillance information to help identify areas for prioritizing surveillance and control. Findings from this article open a window for considering historical DENV data to make predictions of likely sources of invasion for other emerging Aedes-borne viruses, as well as to the consideration of spatially-targeted approaches for delivery of vector control and surveillance. Arbovirus control in complex urban environments can greatly benefit from exploiting existing spatial information for better delivery of interventions.
| The unprecedented global emergence of infectious disease threats has demonstrated how ill-prepared the world is to predict, rapidly respond to, and contain pandemics [1–3]. Zika virus (ZIKV), an emerging arthropod-borne flavivirus, has become a global public health threat since its emergence outside Africa in 2007 [4,5]. A large outbreak in French Polynesia in 2013–2014 showed that an otherwise clinically mild illness (common symptoms include mild fever, rash, arthralgia, and conjunctivitis) could also be responsible for severe neurological complications in adults (e.g., Guillain-Barre syndrome) [6,7]. Once active ZIKV transmission was recognized in Brazil in 2015–2016, the spectrum of disease severity expanded to include neurodegenerative complications in newborns (e.g., microcephaly) as well as other developmental sequelae grouped within the diagnostic term “congenital Zika syndrome” [4,8].
The World Health Organization declaration of Zika as a public health emergency of international concern on February 1, 2016, led to a global coordinated effort to improve vaccine development and scale-up vector control actions against the primary ZIKV vector in the Americas, Aedes aegypti. Failure in previous years to contain the transmission of chikungunya (CHIKV) and dengue (DENV) viruses, also primarily transmitted by Ae. aegypti, demonstrated the challenge in scaling-up vector control tools and interventions such as ultra-low volume (ULV) insecticide applications, larviciding and community mobilization, all with limited efficacy in preventing disease [9]. Although most DENV endemic countries in the Americas anticipated the arrival of ZIKV, they still performed only reactive interventions once symptomatic cases (assumed to be ~20% of all ZIKV infections, [4]) were detected. This led to significant delays in response and limited impact on virus containment, as seen in the rapid geographic spread of epidemics [5]. While some vaccine candidates may soon enter into Phase IIb/III trials [10], vector control and community education still remain the main preventive approaches to minimize the severe and devastating health outcomes associated with ZIKV infection.
DENV, CHIKV and ZIKV are all primarily transmitted by Ae. aegypti mosquitoes, leading to the early assumption that ZIKV would follow the path of the other two viruses in the Americas [11]. While this hypothesis is supported at regional scales, with the three viruses following similar transmission seasonality and regional circulation (e.g., [12]), it is unknown how their transmission pattern could differ within urban centers. DENV tends to show strong heterogeneity within cities, with some neighborhoods reporting higher sero-prevalence or transmission than others (e.g., [13–17]). Daytime human movement and aggregation patterns can influence exposure and are drivers of both human-Ae. aegypti contacts and rapid virus propagation, affecting DENV, CHIKV and ZIKV transmission [18–21]. Given the shared vector and ecology between the three viruses, we conducted spatio-temporal analyses on historical (2008–2015) DENV passive surveillance data and recent (2015–2016) CHIKV and ZIKV virus invasion events within the city of Merida, Mexico, and used the results to answer the following questions: Are DENV transmission hot-spots within cities also likely to be CHIKV and ZIKV transmission hot-spots? Do areas of historically persistent DENV transmission fuel the introduction and propagation of the other two viruses?
The Yucatan Peninsula of Mexico is endemic for all dengue serotypes, with DENV1 and DENV2 dominating during 2008–2015 and DENV4 invading in 2013. Over the same period, 40,028 probable DENV cases have been reported from the city of Merida (S1 Text), the most important urban center (population ~1 million) and Yucatan State’s capital, and major contributor to the burden of virus transmission in the region [22]. Merida is located in a subtropical environment with mean temperatures ranging from 29°C in December to 34°C in July. The rainy season occurs from May to October and overlaps with the peak dengue transmission season between July and November, although cases occur year-round [23]. Dengue virus is widely distributed throughout the Yucatan peninsula, and the vector control strategies used by local authorities include ultra-low volume (ULV) spraying with the organophosphate insecticides chlorpyrifos and malathion and indoor space spraying with pyrethroids (deltamethrin) and organophosphates (malathion).
Recent operational innovations in the surveillance and control of Aedes-borne viruses in Mexico led to the development of a comprehensive online system for the capture, visualization and analysis of entomologic and epidemiologic data [24]. The National System of Epidemiological Surveillance of the Mexican Ministry of Health (operational since 2008) includes a nationwide, web-based, geographic information system where disease-specific reported DENV, CHIKV and ZIKV cases are automatically geocoded and mapped at the geographic scale of households [24]. The following information was obtained for the city of Merida: probable and laboratory confirmed cases geocoded to the household level and with information of age, epidemiologic week of the onset of symptoms for every case; disease infection status (DENV, CHIKV, ZIKV) for each case; DENV serotype (for a subset of laboratory-confirmed cases). Demographic information was obtained for the city of Merida (2010 census) from Mexico’s National Census and aggregated at the level of census tracts, called Area Geoestadistica Basica (AGEB). Tracts aggregate up to 50 city blocks (average area, 0.5 km2; Standard Deviation, 0.3 km2), and Merida has 540 tracts.
Epidemiologic information was aggregated at the spatial scale of census tracts and the temporal scale of weeks, with separate datasets generated for each virus. The weekly count of cases was normalized, to allow for comparisons between years and viruses, by calculating, for each virus, weekly z-scores per tract as follows:
zi=Ni−μσ
(1)
where Ni is the number of cases in tract i, μ the mean case count for the city, and σ the standard deviation of case counts for the city. For a given week, the obtained z-score was rescaled dividing it by census tract’s maximum value of case counts.
The weekly trend in DENV cases was analyzed to separate, during the years 2008–2015, the epidemic from non-epidemic transmission periods of the time-series using change point detection methods [25]. A G* local spatial clustering test [26] was applied to annual normalized case counts separately for epidemic and non-epidemic periods (for DENV) and by year for CHIKV and ZIKV. The significance of the computed G* was evaluated by comparing observed values to the null hypothesis of random case distribution by randomly re-assigning the tract labels to the case counts. This was performed using 100,000 Monte Carlo randomizations. Tracts with statistically significant (p<0.05, calculated adopting Bonferroni correction) high z-scores, defined as hot-spots, were mapped for each virus and epidemic period (DENV) or year (CHIKV, ZIKV). For DENV, G* output was aggregated to estimate a count of time-periods that a tract was identified as a hot-spot.
A Kendall W test evaluated the spatio-temporal overlap between DENV, CHIKV and ZIKV z-values by tract. W measures concordance between two datasets, ranging from +1 (complete agreement) to -1(no agreement) and was calculated for all combinations of virus pairs (DENV-CHIKV, DENV-ZIKV, CHIKV-ZIKV). A second approach to evaluate spatio-temporal overlap between viruses consisted in overlaying the persistence of DENV hot-spots per tract with the hot-spots identified for CHIKV (2015, 2016) and for ZIKV (2016).
Data from a longitudinal DENV cohort study following 1,666 residents aged 1–60 from Merida was utilized to contrast DENV seroprevalence of those living in areas identified as hot-spots versus those living outside. Briefly, 17 elementary schools distributed across Merida in areas thought to differ in DENV incidence (low incidence in the north and high incidence in the south) were chosen as the hubs for participant enrollment and follow-up. From the school student list, 464 children were selected (up to 50 per grade per school) for the serological follow-up, which also included all family members living in the same residence and who consented to the study. Annual blood draws were performed in 2015 and positivity to any DENV serotype was assessed by Yucatan State Laboratories. Prior exposure to dengue and age-specific serostatus were determined using Panbio IgG indirect ELISA. Standard cut-off points were used for defining positive (≥ 12 Panbio units) from negative samples (<9 Panbio units) and indeterminate for those in between. From the people enrolled in the cohort, we selected those whose residence was found inside the areas identified as DENV hot-spots and randomly selected a similar number of individuals enrolled in the cohort living outside the area. As our retrospective analysis covered the period 2008–2015, we focused our analysis on children aged 8 years or younger. A mixed-effects logistic regression model quantified the odds ratio for the following fixed effects: DENV cluster (i.e., living outside or inside the 2008–2015 DENV hot-spot areas) and age. A random intercept was included at the level of the household to account for the occurrence of nested samples within each house. Models were run with the R package lme4 [27].
Protocols for processing and analyzing data were approved by Emory University’s ethics committee under protocol ID: IRB00088659. The protocol was also approved by the Ethics and Research Committee from the O'Horan General Hospital from the state Ministry of Health, Register No. CEI-0-34-1-14. All cases were anonymized before being analyzed.
A total of 40,028 clinically apparent DENV cases were reported to the public health system of Merida during 2008–2015, 94.5% (37,894) occurring in eight epidemic periods lasting on average 34 weeks (range: 24–44) (Fig A in S1 Text). Approximately 30% of all cases were confirmed by laboratory assays (IgM and PCR). During this period, three DENV serotypes circulated in Merida; DENV1 predominated in most years, DENV2 invaded in 2009 and DENV4 in 2013 (Fig B in S1 Text). About 40% of reported cases occurred in <15 year olds, with the proportion of cases by age-class not differing significantly between years (Fig C in S1 Text) (Fisher’s exact test, p < 0.05).
The weekly spatio-temporal progression of standardized DENV case counts within the city was heterogeneous across census tracts, showing a complex pattern of emergence and persistence over the 8-year period (Fig 1A and 1B, Fig D in S1 Text). When summarized over the entire period, standardized case counts aggregated in nearby tracts (Fig 1C) suggested the occurrence of strong autocorrelation. The Getis-Ord G* statistic, calculated separately for epidemic and non-epidemic periods (Fig E and Fig F in S1 Text), identified tracts within Merida with persistent DENV transmission up to 6 years (epidemic, Fig 1D) and 4 years (non-epidemic, Fig 1E) concentrated in the center of the city. Altogether, the hot-spot areas contained 41.9% (16,773) of all reported symptomatic cases within 27.8% of the city’s 540 census tracts.
Serological status for DENV infection in 505 children aged 8 or younger belonging to a longitudinal cohort study was compared between those residing inside and outside the hot-spot areas (244 children inside and 261 children outside the area) (Fig 2A). Living inside the hot-spot area was associated with a significantly higher infection probability than living outside (odds ratio, 1.71 [95%CI, 1.08–2.20]; p < 0.05), after adjusting for the age of the child and the occurrence of nested cases within households (Table 1). Model-predicted DENV infection probabilities were significantly higher in children aged 6–8 years living inside a hot-spot area, compared to children the same age living outside (Fig 2B).
In 2015, the first CHIKV case was reported in a tract within the persistent DENV clustering area (Fig G in S1 Text). From this tract, the virus propagated rapidly throughout the city (counting 1,101 probable cases) (Fig G in S1 Text), leading to difficulties in containing both DENV and CHIKV. ZIKV was first reported in April 2016, also inside the persistent DENV clustering area (Fig G in S1 Text). The 2016 ZIKV outbreak included 2,273 reported cases (of which ~50% were in <15 year olds, Fig H in S1 Text). No Zika-related microcephaly cases were yet detected in newborns to mothers infected during this first wave of ZIKV invasion. The tracts where the first 10 CHIKV and ZIKV cases were reported overlapped 100% and 75%, respectively, with the DENV persistent clustering area (Fig G in S1 Text).
The DENV, CHIKV and ZIKV case counts by tract showed a significantly (p<0.01) high coherence, with Kendall’s W values of 0.75 for DENV-ZIKV comparisons, 0.72 for DENV-CHIKV and 0.63 for CHIKV-ZIKV. The high similarity among DENV, CHIKV, and ZIKV are also evident when comparing distribution of standardized case counts using quantile-quantile (Q-Q) plots (Fig 3A). A 35.6% (392) of all reported CHIKV cases and 66.9% (1,509) of all ZIKV cases were found within the area of persistent DENV clustering. When comparing the location and extent of hot-spots, 9 (64.3%) CHIKV clusters for 2015 and 2016 fully overlapped with the 2008–2015 DENV clustering, and 3 (14.3%) clusters of high ZIKV standardized case counts overlapped with the DENV clustering area (Fig 3A).
Combining powerful analytic methods with detailed retrospective data provides evidence of the significant within-city spatio-temporal overlap between three Aedes-borne viruses. While forecasting the exact timing and introduction points of CHIKV and ZIKV within a city would be bound with significant uncertainty [28], historical spatio-temporal DENV surveillance data can be analyzed to identify the extent of high transmission areas that would serve as important transmission foci for novel Aedes-borne viruses.
Surveillance data has characterized the temporal pattern of CHIKV and ZIKV invasion in areas with active DENV transmission in great detail (e.g., [6,12]). Such descriptive studies provide evidence of high temporal overlap between the three viruses, with transmission peaks of CHIKV and ZIKV occurring during the same time periods as DENV peaks. In Merida, the three viruses not only overlapped in time, but they also exhibited strong spatio-temporal coherence in their distribution. The first CHIKV and ZIKV cases occurred within the DENV clustering area with a strong match between pathogens. It is important to note the partial mismatch between ZIKV and DENV hot-spots. While stochasticity in the initial virus introduction point may contribute to this finding, it is worth mentioning that in 2016 the Yucatan Ministry of Health intensified vector control interventions (i.e., indoor space spraying, ULV spraying and the use of larvicide) in the same areas predicted as hot-spots (based on our preliminary maps) as a way to potentially slow-down the spread of the virus. Whether this enhanced control contributed to the reduced frequency of ZIKV clustering in DENV hot-spots (even in the early presence of cases in the area) will require further study.
Mathematical models predict that, for pathogens that show strong spatial or social heterogeneity, targeted interventions would represent a more effective means of disease mitigation as compared to blanket control [29,30]. Focusing efforts in high transmission areas has the additional benefits of increased efficiency by better allocating limited personnel and resources [31]. Recognizing areas of persistent DENV, CHIKV or ZIKV transmission can enable the pre-emptive deployment of vector control before transmission is apparent [32]. The possible benefits of such pro-active, targeted control may actually be experienced on a larger spatial scale, as interventions may additionally protect individuals who visit the hot-spot areas but do not reside in them [33]. Thus, in settings where Aedes-borne diseases are seasonal, targeting vector control to traditional hot-spot areas in an anticipatory fashion during the period of low transmission can lead to efficiencies in intervention coverage (when personnel workload is at its lowest) and theoretically provide the highest intervention effectiveness (~60–90% of cases prevented) [34]. Alternatively, vector habitat modification strategies such as housing improvement [35,36] may be more cost-effective and justifiable if undertaken within hot-spots. Before targeted control can be broadly recommended, our findings will have to be validated in other urban centers (ideally, of differing size and level of DENV endemicity), and the epidemiological impact of targeting hot-spots will need to be empirically evaluated and weighed against the value of reactive interventions. Additionally, any targeted intervention will have to be monitored for potential future changes in epidemiological trends, like the emergence of new hot-spots in the city due to shifts vector abundance or socio-environmental conditions.
Several issues limit the quality and power of passive surveillance data and the potential for their use to forecast DENV transmission [28,37]. The data do not capture the disproportionately large number of infections that occur sub-clinically or that are mild enough not to result in a visit to a doctor. In many areas, DENV may be misdiagnosed as other febrile illnesses (e.g., influenza, leptospirosis [38]), adding uncertainty to observed trends. Additionally, given Ae. aegypti is a day-biting mosquito, mapping the place of residence may not correspond to the location where infection occurred [18,39]. Finally, in cities where Ae. aegypti and Ae. albopictus co-inhabit, spatial patterns of arbovirus transmission may be potentially influenced by their apparent disjoint distribution within urban areas (e.g., [40]). In Merida, where Ae. aegypti is the only vector and Ae. albopictus is not present, aggregating case data to relatively small (up to 50 city blocks) geographic units and analyzing standardized data over multiple years provided a strong signature of the spatial pattern of DENV circulation. About 27.8% (150) of the city districts accounted for 41.9% of cases, and such transmission heterogeneity was independently captured epidemiologically in the age-dependent sero-prevalence of DENV infection in children aged 8 years or younger. The ability of spatial clustering tests performed on passive surveillance data to capture strong trends in virus sero-prevalence provide a prospect for the development of city-wide risk maps, particularly if analyses involve multiple years and several virus introduction events.
Several spatial epidemiology studies have determined the set of census-derived environmental and socio-economic variables associated with DENV hot-spots. For instance, in Machala (Ecuador) hot-spots of dengue incidence at the level of census tracts was associated with older age and female gender of the head of the household, greater access to piped water in the home, poor housing condition, and less distance to the central hospital ([41]). In an endemic city in Venezuela, DENV hot-spots at the block level were associated with living in crowded conditions, having an occupation of domestic worker/housewife and not using certain preventive measures against mosquitoes ([42]). The contrast between these and other spatial epidemiology studies point to an important issue related to the finding of a generalized explanation to the elevated risk of DENV infection in hot-spots. Census variables vary from country to country, studies use data at different levels of aggregation, many census variables are likely auto-correlated, and there is an almost absence of entomological information. Thus, our study excluded imputing hot-spots to specific census variables because we will be performing a comprehensive environmental and anthropological assessment of the root drivers of the epidemiological trends observed in Merida.
We acknowledge several limitations to our study. While we used residential addresses in our analyses, we are aware that this excludes the potential for infection in areas other than the home [18,21]. Thus, our study was not able to explicitly account for movement and exposure in locations other than the residence, which may have obscured both the passive surveillance and the serological data. Future human mobility studies in Merida will help elucidate the role of hot-spots as foci of infection from both their residents and visitors from other parts of the city. The lack of Zika-related microcephaly cases in Merida at the point of analysis prevented evaluating any possible overlap between them and the historical DENV hot-spot areas. Given the possibility for immune interaction between viruses ([43]), evaluating if areas of intense DENV transmission lead to higher risk of microcephaly in newborns can lead to important insights for disease surveillance and control. While laboratory diagnosis of all three viruses is performed routinely in Mexico, current policy dictates that testing should be reduced once outbreaks are declared. This prompted us to include in our analyses clinically diagnosed (i.e., without laboratory confirmation) cases in addition to those with a laboratory confirmation.
While vaccines are seen as an ultimate disease mitigation strategy, their development can lag years after pathogen emergence [44], leaving case management and outbreak containment as the first line of defense. Thus, public health efforts face a challenging prospect: identifying vulnerable populations or likely transmission hot-spots as an approach to pro-actively mitigate emerging pathogen introduction and propagation. Recognizing the critical need to improve surveillance to prevent and rapidly contain pathogen transmission, the World Health Organization (WHO) is developing a global coordination mechanism for the containment of emerging pathogens [3]. Crucial to such effort is the integration of surveillance data and robust analytical methods to anticipate and better respond to disease threats [45,46]. Leveraging historical DENV, CHIKV and ZIKV data can help endemic countries improve control and surveillance programs by acknowledging and accounting for inherent spatial variability in risk within urban areas. Downscaling existing global risk maps (e.g., [47,48]) to capture within-city transmission risk is an important next step in disease mapping, as this can provide policy-makers with a more precise tool for intervention planning and strategic deployment at an operational scale.
|
10.1371/journal.pntd.0005603 | Ileus in children presenting with diarrhea and severe acute malnutrition: A chart review | Severely malnourished children aged under five years requiring hospital admission for diarrheal illness frequently develop ileus during hospitalization with often fatal outcomes. However, there is no data on risk factors and outcome of ileus in such children. We intended to evaluate predictive factors for ileus during hospitalization and their outcomes.
This was a retrospective chart review that enrolled severely malnourished children under five years old with diarrhea, admitted to the Dhaka Hospital of the International Centre for Diarrhoeal Disease Research, Bangladesh between April 2011 and August 2012. We used electronic database to have our chart abstraction from previously admitted children in the hospital. The clinical and laboratory characteristics of children with (cases = 45), and without ileus (controls = 261) were compared. Cases were first identified by observation of abnormal bowel sounds on physical examination and confirmed with abdominal radiographs. For this comparison, Chi-square test was used to measure the difference in proportion, Student’s t-test to calculate the difference in mean for normally distributed data and Mann-Whitney test for data that were not normally distributed. Finally, in identifying independent risk factors for ileus, logistical regression analysis was performed. Ileus was defined if a child developed abdominal distension and had hyperactive or sluggish or absent bowel sound and a radiologic evidence of abdominal gas-fluid level during hospitalization. Logistic regression analysis adjusting for potential confounders revealed that the independent risk factors for admission for ileus were reluctance to feed (odds ratio [OR] = 3.22, 95% confidence interval [CI] = 1.24–8.39, p = 0.02), septic shock (OR = 3.62, 95% CI = 1.247–8.95, p<0.01), and hypokalemia (OR = 1.99, 95% CI = 1.03–3.86, p = 0.04). Mortality was significantly higher in cases compared to controls (22% vs. 8%, p<0.01) in univariate analysis; however, in multivariable regression analysis, after adjusting for potential confounders such as septic shock, no association was found between ileus and death (OR = 2.05, 95% CI = 0.68–6.14, p = 0.20). In a separate regression analysis model, after adjusting for potential confounders such as ileus, reluctance to feed, hypokalemia, hypocalcemia, and blood transfusion, septic shock (OR = 168.84, 95% CI = 19.27–1479.17, p<0.01) emerged as the only independent predictor of death in severely malnourished diarrheal children.
This study suggests that the identification of simple independent admission risk factors for ileus and risk factors for death in hospitalized severely malnourished diarrheal children may prompt clinicians to be more vigilant in managing these conditions, especially in resource-limited settings in order to decrease ileus and ileus-related fatal outcomes in such children.
| Childhood malnutrition and diarrhea are important problems in lower and middle-income countries, including Bangladesh. Severe acute malnutrition (SAM) and diarrhea are responsible for more than one third of all deaths in children below five years old globally. Most of these deaths occur because of complications of SAM and/or diarrhea. SAM and diarrhea may simultaneously present in a child, often with serious complications. Ileus is a common fatal complication in such hospitalized children, and is accompanied by abdominal distension and hyperactive, sluggish, or absent bowel sounds heard using a stethoscope. Ileus is confirmed by radiologic evidence of multiple gas-fluid levels in the abdomen. However, the underlying factors contributing to ileus and its management in diarrheal children with severe acute malnutrition is unknown. Our study aimed to ascertain these risk factors and the outcome of ileus in such children by reviewing the data of previously admitted children between April 2011 and August 2012. Of 306 severely malnourished Bangladeshi under-five-year-old children with diarrhea enrolled for 17 months, 45 had ileus. Ileus was independently associated with a reluctance to feed, septic shock, and hypokalemia and had a higher case-fatality rate compared to those without ileus.
| Globally, diarrhea accounted for 9% of an estimated 5.9 million deaths in children under five years old in 2015 [1]. Most of the deaths occurred in lower and middle-income countries including Bangladesh [2]. The case-fatality rate (CFR) from diarrhea in Bangladesh was 6% among an estimated 119,000 deaths in children less than five years old [1]. The majority of the deaths occurred due to a number of immediate complications of diarrhea such as dyselectrolytemia, sepsis, and ileus [3]. However, severe acute malnutrition (SAM) was one of the important causes of death in under-five-year-old children and the risk of death from any cause was 9 times higher for SAM compared to non-SAM children [4]. CFR among children hospitalized with SAM has virtually remained unchanged over the past several decades in many centers [5]. Some centers have been able to reduce CFR to less than 5% by minimizing complications from SAM via the implementation of World Health Organization (WHO) guidelines [6]. These complications were life-threatening in children presenting with severe malnutrition and were often associated with death [3]. Ileus, a condition accompanied by abdominal distension and hyperactive, sluggish, or absent bowel sounds and radiologic evidence of multiple gas-fluid levels in the abdomen, is considered as one of the serious ramifications of diarrhea [7]. It is suggested that ileus could be more deleterious to the health of diarrheal children with severe malnutrition compared to those without severe malnutrition.
Ileus was found to be common in critically ill children and adults and most often reflected the severity of underlying disease [8]. Ileus usually presents with distension and tenderness of the abdomen and abnormal bowel sounds, especially in hospitalized children [9]. Recent studies have revealed that a significant proportion of hospitalized children experience ileus due to sepsis, dyselectrolytemia, and pseudo-membranous colitis in malnutrition [10,11]. On the other hand, ileus in adults mostly presents as an acute abdomen characterized by abdominal distension and an ischemic bowel with or without perforation [12]. A previous study from Bangladesh revealed that 12% of children with diarrhea developed ileus and 25% of them died [9]. The prevalence and fatal outcome of ileus is considered to be higher in children with SAM compared to those without SAM. However, WHO has not recommended any management plan for ileus in such children because of a lack of evidence regarding underlying factors that contribute to ileus in children having diarrhea with SAM. Identifying the factors contributing to ileus in children having diarrhea with SAM may help clinicians to deploy early and appropriate intervention that should further help to reduce deaths, especially in developing countries.
In the Dhaka hospital of the International Centre for Diarrhoeal Disease Research, Bangladesh (icddr,b), diarrhea patients with and without severe malnutrition are treated [13], and a substantial number of these patients develop ileus during hospitalization [7]. If such cases are promptly identified and factors associated with ileus in severely malnourished children with diarrhea are treated, it might be possible to prevent the serious complications and death due to ileus. However, to our knowledge, there are no published data on the risk factors and outcome of ileus in children with diarrhea presenting with SAM. Thus, our objective was to identify the factors that predicted ileus and to evaluate the outcome in such children.
The data used in this study were retrieved from the case records of patients of the Dhaka hospital of icddr,b. Data were entered in an anonymized manner prior to analysis and used for the improvement of the quality of care of the hospital patients. The study was approved by the Ethical Review Committee of icddr,b.
The hospital in this study is an urban diarrheal hospital situated in Dhaka, the capital city of Bangladesh. Annually, approximately 150,000 patients are seen and the presence of diarrhea is compulsory for admission. There are 350 beds in this hospital; however, during diarrheal epidemics, extra beds are often used to treat the additional patients. Two epidemics are commonly observed in this region: one in the hot summer of April-May and another in September-October. More than 90% patients are treated in the Short-Stay Ward (SSW). Those without diarrheal complications or associated problems receive treatment in the SSW and their median duration of stay is 24 hours. Patients with diarrheal complications or associated problems are treated in the Longer-Stay Ward (LSW) or in the Intensive Care Unit (ICU). The median duration of stay at the LSW and ICU is 3 days and 5 days, respectively. A detailed description of Dhaka hospital has been provided elsewhere [14].
We employed a retrospective chart analysis with an unmatched case-control design. A matched case-control design was not conducted as there was not enough power for a matched design because of the small sample size. The study enrolled children aged 0–59 months, with severe malnutrition and diarrhea, admitted to the ICU or LSW of the Dhaka ICDDR,B hospital, and with information regarding abdominal examination including ileus as a part of standard care in the ICU between April 2011 and August 2012. The enrolled children with diarrhea who developed ileus during hospitalization constituted our cases, whereas those who did not develop ileus constituted controls. Ileus was defined as the development of abdominal distension with hyperactive, sluggish, or absent bowel sounds and radiologic evidence of abdominal gas-fluid level [7] during hospitalization. Severe malnutrition was defined following WHO anthropometry as described elsewhere [15]. We used the electronic database of the hospital to perform our chart abstraction, from children previously admitted in the hospital. Those involved in the abstraction were not involved in the care of these patients when they were hospitalized. Events in the ICU were stored in an electronic medical record. Socio-demographic history, clinical features, and laboratory investigations reports are automatically collected in this database. First, the database was queried to identify children below 5 years old admitted for diarrhea, and then these charts were reviewed for the diagnosis of severe malnutrition. Regarding the severely malnourished children with diarrhea, the database was further queried to establish whether these children had clinically and radio-graphically confirmed ileus, and the cases and controls were differentiated via the information received. At different time points, we had collected data from more than one individual in the Information and Technology (IT) Department of icddr,b and during the collection of our data we did not find any inconsistency among the datasets.
All the children with ileus received standard conservative medical treatment following hospital guidelines derived from Ahmed et al [3]. No patient required surgical intervention. In managing ileus, food was given at 3-4-hourly instead of the routine 2-hourly intervals. If the condition was not resolved within 6–8 hours of spacing of diets, a single intramuscular injection of magnesium sulphate 50%, 0.3 ml/kg was administered, up to a maximum of 2 ml. This bolus dose was in addition to the daily maintenance dose of magnesium sulphate injection. If symptoms resolution was not achieved within 2 hours, feeding was discontinued and intravenous fluid was administered; 1/2 strength normal saline with 5% dextrose, 72 ml/kg per 24 hours (3 ml/kg per hour). Usually, 20 mmol/L of injection potassium chloride was added to 1 L of the infusion. Other treatments (antibiotics) received in the wards have been described elsewhere [14,15].
Case report forms were developed, pretested, and finalized for the acquisition of study-relevant data. The demographic information on admission such as age, sex, vaccinations, socioeconomic status, and lack of breastfeeding during the neonatal period were analyzed. Clinical features such as duration of diarrhea; vomiting, with duration; dehydration; weight for age; and weight for length/height z-score; reluctance to feed, with duration; fever, with duration; convulsion; oral thrush; hypoxemia; septic shock; hospital acquired infection; and outcome were also analyzed. The analyzed laboratory characteristics were hypoglycemia (random blood sugar <3.0 mmol/L), bacteremia (bacterial isolate from a single blood sample culture), hypokalemia (serum potassium < 3.5 mmol/L), hyperkalemia (serum potassium > 5.5 mmol/L), hyponatremia (serum sodium < 130.0 mmol/L) and hypernatremia (serum sodium >150.0 mmol/L), elevated serum creatinine (> 35.0 mmol/L in children <12 months old, <65.0 umol/L in children ≥12 months old), metabolic acidosis (serum TCO2 < 17.0 mmol/L), hypocalcemia (serum calcium < 2.12 mmol/L), and hypomagnesemia (serum magnesium < 0.70 mmol/L). Septic shock was defined as severe sepsis that was unresponsive to fluid resuscitation. Severe sepsis in diarrheal children has been defined in recent publications from Bangladesh [14,16]; originally adopted from the surviving sepsis guideline recommended by the American Pediatric Association [17,18], with minor modifications for children with diarrhea. Severe sepsis was defined as sepsis plus the presence of poor peripheral perfusion (weak or absent peripheral pulses), and a capillary refilling time greater than 3 seconds or age-specific hypotension. Sepsis was defined as the presence or presumed presence of infection with hyperthermia or hypothermia (rectal temperature >38∙5°C or <35∙0°C, respectively) and tachycardia in the absence of dehydration or after correction of dehydration.
Using the available aseptic precautions, the attending physician performed venipuncture and collected blood for culture; 10% povidone iodine followed by 70% rectified spirit was used to disinfect the puncture site, which was dried for 30 to 60 seconds. A standard pediatric blood culture bottle (BacT/Alert PF, Organon-Teknika, Durham, NC) was used to collect 2–5 mL of blood [19]. After the collection of blood, the microbiology laboratory, situated in the 2nd floor of the hospital, immediately processed the bottles. Blood culture bottles were incubated in the automated BacT/Alert system. Initial positivity of specimens was flagged by the system and the flagged specimen was subsequently sub-cultured onto blood agar, MacConkey agar, and chocolate agar. Isolation and identification of pathogens were performed using standard bacteriologic procedure [20]. Antimicrobial susceptibility testing was performed with Mueller-Hinton agar plates, using a disc diffusion method.
All data were entered into SPSS for Windows (version 17.0; SPSS Inc., Chicago) and Epi-Info (version 6.0, USD, Stone Mountain, GA). Differences in proportion were compared by the Chi-square test. Student’s t-test was used to compare the means of normally distributed data and the Mann-Whitney test was used for the comparison of data that were not normally distributed. A p-value less than 0.05 was considered statistically significant. The strength of association was determined by calculating the odds ratio (OR) and its 95% confidence interval (CI). For the identification of risks for ileus in children having diarrhea with SAM, we initially analyzed the relevant variables in a univariate model (Table 1). Finally, for the adjustment of the covariates, we performed a multivariable logistic regression analysis (Table 2), where the dependent variable was ileus and the independent variables were those associated with ileus in the univariate model. We performed a second multivariable logistic regression analysis (Table 3) to evaluate the independent predictors of death, where death was the dependent variable and the variables associated with death in a univariate analysis acted as independent variables.
We also performed pathogen-specific data analysis and compared the differences in the proportion of bacterial pathogens between children with and without septic shock. We also compared the differences in the proportion between the overall and different bacterial pathogens, between children with and without ileus (Table 4). Only the Chi-square test was used for the comparison of the proportion.
The review identified 45 cases and 261 controls. The cases, more often, were reluctant to feed, had septic shock, hypokalemia, and hypocalcemia on admission, and received blood transfusion during hospitalization, compared to the controls (Table 1).
In the logistic regression model, the dependent variable was ileus and the independent variables were septic shock, reluctance to feed, hypokalemia, hypocalcemia, and blood transfusion. In this model, ileus was independently associated with reluctance to feeding, septic shock, and hypokalemia (Table 2).
In a univariate analysis, the mortality rate was significantly higher in those with ileus compared to those without ileus (Table 1); however, in a multivariable logistic regression analysis, after adjusting for potential confounders such as septic shock, no association between ileus and death was noted (OR = 2.05, 95% CI = 0.68–6.14, p = 0.20). Moreover, in the multivariable logistic regression model shown in Table 3, death was the dependent variable and ileus, septic shock, reluctance to feed, hypokalemia, hypocalcemia, and blood transfusion were the independent variables. After adjusting for potential confounders, septic shock was identified as an independent predictor of death in our study population (Table 3).
Among the 26 children who had septic shock, 5 (19%) had bacteremia, and in the remaining 280 children without septic shock, 25 (9%) had bacteremia; however, the difference in bacteremia between the two groups was not statistically significant (p = 0.156). Stool culture was performed in 29 of 45 (64%) cases and 137 of 261(52%) controls (Table 4).
The proportion of the total number of bacterial pathogens isolated from stool was comparable between children with and without ileus (Table 4). In the pathogen-specific isolation, the proportion of Vibrio cholera, Shigella, Non-typhoidal Salmonella, Campylobacter species, and mixed bacterial pathogens (Shigella and Vibrio cholerae) was also comparable between the groups (Table 4).
To date, no study has attempted to identify the risk factors and outcome of ileus in children aged under five years with diarrhea and severe malnutrition. Some important observations were made in this study: first, children with SAM and diarrhea who developed ileus during hospitalization more often were reluctant to feed, had septic shock, and hypokalemia on admission; and second, these children had a higher mortality rate compared to those without ileus in a univariate analysis, however, after adjusting for potential confounders such as septic shock in a multivariable regression analysis, we observed no association between ileus and death. These observations need to be clarified for the formulation of better future interventions that may help to reduce deaths in children with diarrhea and SAM, especially in resource-poor settings.
The association of septic shock with ileus in children under five years old with diarrhea and severe malnutrition is understandable. Septic shock in children often stimulates oxidative stress, the endogenous production of nitric oxide [21], and the simultaneous surge of serum lactate [22–24]. This eventually results in abandoned vasodilatation followed by hypotension and reduced splanchnic circulation [25,26]. Absorption of food is often perceived to be greatly reduced in the gut because of compromised circulation in septic shock and, thus, contributes to the development of abdominal distension followed by ileus.
The association of hypokalemia with ileus in children having diarrhea and SAM is consistent with a number of previous studies [27,28]. The independent association of ileus with children being reluctant to feed is also understandable. Children with diarrhea and severe malnutrition who were reluctant to feed, more often presented with severe gut infection, leading to an edematous small bowel [18]. This phenomenon might contribute to the narrowing of the small bowel lumen and gaseous abdominal distension resulting in ileus.
The non-association between ileus and death in a multivariable regression analysis, despite an initial observation of higher deaths in a univariate analysis in children with diarrhea and SAM is critically important. Although a number of previous studies reported a high mortality in children with ileus but without diarrhea [7,29], ileus failed to predict deaths in this study population after adjusting for potential confounders (Table 3). In the regression analysis shown in Table 3, septic shock was found to be independently associated with death, underscoring the confounding effect of septic shock on the ability of ileus to predict death. Septic shock has been shown to be independently associated with death in a number of previous studies [30–32].
The lack of performance of viral isolation, including rotavirus, was one of the limitations of this study, although, the prevalence of rotavirus infection in SAM children is quite low due to the lack of receptors required for rotavirus attachment to villus cells in the intestine [13,33]. Potential misclassification bias in enrolling our study population during the chart analysis was another limitation of the study. Moreover, the retrospective nature of the study and the small sample size might have prevented the observation of an association between some of our variables of interest and ileus.
The current study; however, has several strengths. First, the data were abstracted from the electronic database of the largest diarrhea hospital in the world and during the data collection from different individuals at different time points, we did not find any inconsistency among the datasets. Second, this is the only study that has evaluated the risk factors and outcome of ileus in children with diarrhea and severe malnutrition. Third, there was scrupulous adherence to available standard treatment protocols for ileus in the ICU. Finally, the results of this study may be used in resource-limited settings where diarrhea and malnutrition are common.
In conclusion, our study noted that the deaths were significantly higher among children with diarrhea and severe malnutrition who had ileus, compared to those without ileus. Children below five years old who were hospitalized for diarrheal illness and presented with reluctance to feed, septic shock, and hypokalemia on admission, were at higher risk of developing ileus during hospitalization. Thus, the identification of these simple parameters in severely malnourished children with diarrhea on admission may prompt clinicians to be more vigilant in managing these conditions, especially in resource-limited settings in order to cure ileus and prevent further complications. A prospective study involving a larger sample of such children is crucial for the evaluation of our findings.
|
10.1371/journal.pgen.1005996 | Separase Is Required for Homolog and Sister Disjunction during Drosophila melanogaster Male Meiosis, but Not for Biorientation of Sister Centromeres | Spatially controlled release of sister chromatid cohesion during progression through the meiotic divisions is of paramount importance for error-free chromosome segregation during meiosis. Cohesion is mediated by the cohesin protein complex and cleavage of one of its subunits by the endoprotease separase removes cohesin first from chromosome arms during exit from meiosis I and later from the pericentromeric region during exit from meiosis II. At the onset of the meiotic divisions, cohesin has also been proposed to be present within the centromeric region for the unification of sister centromeres into a single functional entity, allowing bipolar orientation of paired homologs within the meiosis I spindle. Separase-mediated removal of centromeric cohesin during exit from meiosis I might explain sister centromere individualization which is essential for subsequent biorientation of sister centromeres during meiosis II. To characterize a potential involvement of separase in sister centromere individualization before meiosis II, we have studied meiosis in Drosophila melanogaster males where homologs are not paired in the canonical manner. Meiosis does not include meiotic recombination and synaptonemal complex formation in these males. Instead, an alternative homolog conjunction system keeps homologous chromosomes in pairs. Using independent strategies for spermatocyte-specific depletion of separase complex subunits in combination with time-lapse imaging, we demonstrate that separase is required for the inactivation of this alternative conjunction at anaphase I onset. Mutations that abolish alternative homolog conjunction therefore result in random segregation of univalents during meiosis I also after separase depletion. Interestingly, these univalents become bioriented during meiosis II, suggesting that sister centromere individualization before meiosis II does not require separase.
| Sexual reproduction depends on meiosis, a special cell division that occurs in two steps, meiosis I and II. Meiosis is distinct in males and females that produce two very different forms of compatible gametes, the sperm and egg, respectively. In the fly Drosophila melanogaster, sex-specific differences are pronounced. While pairing of homologous chromosomes into bivalents before the first meiotic division proceeds in a canonical manner in females, males use an alternative system. After production of bivalents and their biorientation in the meiosis I spindle, they need to be split up again in both sexes so that chromosomes can be segregated to opposite spindle poles during anaphase I. Here we demonstrate that separase, a protease that separates canonical bivalents by cleavage of a cohesin subunit, is also required during male meiosis I, even though alternative homolog conjunction does not depend on cohesin. Moreover, our results suggest that the separation of sister centromeres into functionally distinct units that has to occur between the first and second meiotic division does not depend on separase. Centromeric cohesin might therefore either not enforce the crucial sister centromere coorientation during meiosis I, or be removed in a separase-independent manner in preparation for meiosis II.
| After their production during S phase, sister chromatids remain paired. This sister chromatid cohesion is crucial for proper bipolar chromosome orientation within mitotic spindles during early M phase. Sister chromatid cohesion is maintained primarily by cohesin, a protein complex composed of an SMC1/3 heterodimer and accessory subunits including an α-kleisin protein [1, 2]. However, during late metaphase after biorientation of all chromosomes within the spindle, cohesion between sister chromatids needs to be released for chromosome segregation during anaphase. This cohesion release depends on separase, an endoprotease which specifically cleaves α-kleisin just before the metaphase to anaphase transition [1–5]. Loss of cohesin or separase function results in chromosome segregation errors due to premature separation of sister chromatids or failure of their separation, respectively.
In comparison to mitosis, chromosome segregation during meiosis is more elaborate [6]. After pre-meiotic S phase, homologous chromosomes pair up and form bivalents. Maintenance of homologous chromosome pairs usually depends on chiasmata generated by meiotic recombination. Importantly, the two sister kinetochores in each chromosome are also united into a functional unit and co-oriented during the first meiotic division [7–12]. This allows bipolar orientation of bivalents in the meiosis I spindle. Separase-mediated release of cohesin from chromosome arms during late metaphase I permits terminalization of chiasmata and chromosome segregation during anaphase I [13–17]. Importantly, however, release of pericentromeric cohesin is prevented during meiosis I [18–24]. This keeps sister kinetochores paired. Moreover, since sister kinetochores regain functional individuality, they become bioriented within meiosis II spindles and segregate to opposite spindle poles after separase-mediated destruction of pericentromeric cohesin during late metaphase II [25–28]. Faithful chromosome segregation during meiosis therefore relies on the unique functional unification of sister kinetochores during meiosis I, in combination with temporally controlled release of arm and pericentromeric cohesion during meiosis I and II, respectively.
Interestingly, the success of meiosis depends apparently not only on the two chromosomal cohesin populations in arm and pericentromeric regions, but also on yet another, functionally distinct cohesin pool acting within the centromere. Centromeric cohesin was proposed to be present and required exclusively before meiosis I in fission yeast for the functional unification and co-orientation of the two sister kinetochores [9, 26, 29, 30]. Analyses in mouse oocytes [31, 32] and plants [33] have provided further support for a functional unification of sister kinetochores by meiosis I-specific centromeric cohesin. The molecular mechanisms that establish and inactivate centromeric cohesin before meiosis I and II, respectively, are poorly understood. Spo13, Moa1 and Meikin, which appear to provide a similar function required for sister kinetochore co-orientation during meiosis I in budding yeast, fission yeast and vertebrates, respectively, might be involved in the generation of centromeric cohesion [11]. Moreover, destruction of centromeric cohesin during exit from meiosis I by separase seems to be an obvious, highly probable mechanism for sister kinetochore individualization before meiosis II.
Beyond unknown aspects of canonical meiosis, additional issues remain to be clarified in the context of derived meiotic variants. In Drosophila males, pairing and physical linkage of homologous chromosomes involves neither meiotic recombination nor synaptonemal complex formation [34]. Several components of an alternative homolog conjunction system have been identified genetically. Mutations in modifier of mdg4 in meiosis (mnm) and stromalin in meiosis/SA-2 (snm) result in random segregation of homologs in meiosis I [35]. Cytological analyses have revealed homolog conjunction defects in the mutants [35]. The proteins MNM and SNM accumulate exclusively in spermatocytes. At the start of meiosis, they are recruited prominently to a spot on the X-Y chromosome bivalent. Although the two sex chromosomes are strongly heteromorphic in Drosophila, they both contain a locus with an rDNA gene cluster and 240 bp repeats within the intergenic sequences. These repeats were shown to be required and sufficient for X-Y pairing [36]. Cytology has revealed co-localization of these repeats with MNM and SNM [35, 37]. In addition, weak MNM/SNM spots were also observed on autosomes. The molecular details of MNM and SNM recruitment to meiotic chromosomes remain to be analyzed. MNM contains a unique C-terminal FLYWCH Zn-finger domain and an N-terminal BTB/POZ domain that is also present on many other mod(mdg4) isoforms [35]. SNM is a distant relative of the stromalins which are known to be cohesin subunits [35]. Beyond mnm and snm, homolog conjunction in Drosophila males also requires teflon (tef) which codes for a protein with three C2H2-type zinc fingers and unknown localization during meiosis [38]. tef expression does not appear to be meiosis-specific and it is only required for autosome but not X-Y conjunction [38].
The fact that homologous chromosomes in Drosophila males are not linked by chiasmata and arm cohesion raises the question of how homologs are separated during meiosis I. In principle, separase activity might inactivate the alternative homolog conjunction system before anaphase I and thus cause homolog separation as in canonical meiosis. However, separase appears to target primarily cohesin which does not appear to be involved in alternative homolog conjunction. Hence, a separase-independent mechanism remains a possibility as well. Therefore, we evaluated the role of separase during Drosophila male meiosis. Moreover, the possibility of separase-independent chromosome segregation during Drosophila male meiosis I appeared to offer opportunities for confirmation that separase is actually required for sister kinetochore individualization during exit from meiosis I, since in this case normal chromosome segregation during meiosis I would be predicted to be followed by a failure of sister kinetochore biorientation during meiosis II in the absence of separase function. The apparent absence of a Drosophila rec8 homolog [39, 40] provided yet another reason for our interest in meiotic separase functions. A meiosis-specific Rec8 alpha-kleisin has been shown to be absolutely required for protection of pericentromeric cohesion from separase cleavage during meiosis I in a wide range of species [15, 21, 23, 24, 26, 33, 41]. Its putative absence in Drosophila further emphasizes the non-canonical nature of its meiosis, making meiotic separase functions highly unpredictable in this species where potentially not just chromosome separation during meiosis I but also during meiosis II might be achieved in a separase-independent fashion.
To evaluate meiotic functions of separase, we developed approaches for separase inhibition specifically during male meiosis. Thereby we were able to demonstrate that the inactivation of the alternative homolog conjunction system before anaphase I onset is entirely dependent on separase. Moreover, the fact that alternative conjunction in Drosophila males can be eliminated by specific mutations that do not affect sister chromatid cohesion also allowed us to address whether sister kinetochore individualization after metaphase I depends on separase. Interestingly, we find that sister kinetochore biorientation during meiosis II appears to be independent of separase.
Drosophila Separase (SSE) does not have a large N-terminal regulatory region as typically observed in other species [42]. However, it forms a complex with the product of the three rows (thr) gene which appears to have resulted from a separase gene split during Drosophila evolution [43, 44]. Moreover, the product of the pimples (pim) gene also needs to accumulate during interphase and join the SSE-THR complex for eventual separase function during mitosis [42, 45]. Therefore, embryos lacking pim function zygotically also display a separase loss-of-function phenotype. This mutant phenotype does not reveal that PIM actually also has an additional, separase-inhibitory role. However, PIM is known to be the Drosophila securin homolog that prevents premature separase activity until late metaphase [46, 47]. While PIM is required initially for separase complex formation, it has to be degraded again eventually during late metaphase via activation of the ubiquitin ligase APC/C. The meiotic function of Sse, thr and pim cannot be studied in zygotic null mutants. They do not develop to the developmental stages where progression through meiosis starts because of sister chromatid separation failure during the earlier mitotic divisions [42, 45, 48]. To investigate meiotic functions, we applied transgenic RNAi expressed specifically in spermatocytes using the bam-GAL4-VP16 (bG) driver. Knock down of thr was found to be most effective, causing male sterility (S1 Table).
To characterize the effects of THR depletion at a cellular level, we analyzed testis squash preparations after staining with a DNA stain and anti-tubulin. Anti-tubulin labeling facilitates identification of meiotic cells and discrimination of meiosis I and II also when chromosome segregation is abnormal. Cells during the meiotic divisions have spindles that change in a characteristic manner during progression through these divisions. During late anaphase and telophase for example, formation of a prominent central spindle occurs. Anti-tubulin labeling also revealed the overall cell size which is halved by cytokinesis first during meiosis I and once again during meiosis II. Conversely, cell number per cyst increases and also provided information whether cells are in meiosis I or II. Preparations with testes from bG males with and without UASt-thrRNAi transgene displayed normal prometaphase I figures (Fig 1A).
Up to four DNA masses could be resolved corresponding to the bivalents with the sex chromosomes (XY) and the three autosomes (2nd, 3rd, and small 4th chromosome). In the control preparations, telophase I figures and the figures from prometaphase II and telophase II were normal as expected (Fig 1A). Telophase cells contained two round daughter nuclei of comparable size close to the two spindle poles. In contrast, after THR depletion 86% of the telophase I figures were clearly abnormal (Fig 1A and 1B). Bi-lobed DNA staining with a connecting chromosome bridge was observed (67%), as well as cases with a single DNA mass that was no longer bi-lobed (19%). Subsequent meiotic stages were also abnormal after THR depletion (Fig 1A). Prometaphase II cells contained highly variable amounts of chromatin. Moreover, chromosome separation failure was again apparent during telophase II in 100% of the cysts (Fig 1C).
Time-lapse analysis of progression through the meiotic divisions without and with THR depletion (S1 Movie and S2 Movie) fully confirmed the findings from squash preparations. THR depletion did not affect meiosis I up to anaphase onset. However, chromosome separation during anaphase I did not succeed. Subsequent cytokinesis was irregular as well, producing a pair with a nucleate and an anucleate cell in some cases, or cutting through the undivided mass of chromatin in other cases. As a result, meiosis II spindles were often abnormal. But even in cases with normal meiosis II spindles, chromosome separation during anaphase II never occurred normally. Time-lapse imaging also revealed that THR depletion did not severely affect the dynamics of progression through meiosis I. However, the number of our movies that start before nuclear envelope breakdown in meiosis I is low (only two cysts from independent preparations) because time-lapse imaging was usually started only after finding a cyst within the testis preparation which already had initiated meiosis I. Starting time-lapse imaging earlier at a stage where cysts are still in the long premeiotic G2 phase allows analysis of progression through meiosis only in very rare fortuitous cases because cyst viability deteriorates in most of the cases before entry into meiosis. In the two completely tracked THR depleted meiotic cysts, the duration of prometaphase I, metaphase I, and anaphase I was 12/18, 12/16, and 11/12 minutes, respectively. The average duration of these meiosis I phases in controls was 12.3 ± 1.8, 13.3 ± 3.7, and 10.7 ± 1.2 minutes respectively (± s.d.; n = 6 cysts from independent preparations). Additional movies allowing determination of the duration of metaphase I (n = 2) and anaphase I (n = 4) after THR depletion provided further support for our conclusion that the temporal dynamics of congression of bivalents and onset of anaphase during meiosis I were not severely altered by THR depletion, although subtle effects cannot be excluded. As a result of the chromosome separation defects caused by THR depletion during meiosis I, meiosis II was severely affected in various variable ways and therefore, analysis of temporal dynamics during interkinesis and meiosis II after THR depletion was not attempted.
As bG-directed expression starts already during the final mitotic division cycles that generate the cysts of 16 interconnected spermatocytes, the meiotic defects described above might arise in principle as secondary consequences from earlier division abnormalities. However, several observations indicated that the cysts entering into meiosis after bG-directed THR depletion were normal. Cysts still comprised 16 cells as expected. Moreover, the number of centromeres detected during meiosis indicated that all spermatocytes were euploid. FISH with X and Y probes, as well as MNM/SNM immunolocalization also confirmed euploidy (see below). However, testes of bG males with UASt-thrRNAi were smaller than those of control males. Therefore, some germline stem cell or gonial cyst depletion might occur as a result of leaky expression during earlier stages.
To rule out off-target effects, we introduced into the males with bG and UASt-thrRNAi also an UASt-thrRr transgene predicted to be RNAi-resistant as a result of silent mutations. While the RNAi-resistant transgene was unable to restore a completely normal meiosis, it reduced the frequency of telophase I abnormalities fourfold (Fig 1B). Moreover, the residual chromosome bridges during telophase I were less massive and less stable when the RNAi-resistant transgene was present (S1 Fig).
For further confirmation that the abnormalities resulting after THR depletion reflect a loss of separase function, we depleted other separase complex subunits. Analogous bG directed expression of UASt-SseRNAi and UASt-pimRNAi resulted in no or milder abnormalities, respectively. Abnormalities after PIM depletion were also first observed during telophase I, as in case of THR depletion. Chromosome bridges were apparent in 56% of the telophase I figures (Fig 1B) and all cysts during telophase II also displayed chromosome bridges. To achieve SSE depletion, we applied an alternative method, deGradFP which allows regulated proteolytic degradation of GFP fusion proteins [49]. Therefore, we first generated a line in which the lethality associated with hemizygosity for an Sse null mutation (Sse13m/Df(3L)SseA) was prevented by a transgene driving expression of EGFP-SSE under control of the Sse cis-regulatory region (gEGFP-Sse). The resulting flies were also fertile (S1 Table). However, introduction of an additional transgene directing NSlmb-vhh-GFP4 expression by the bam regulatory region led to complete sterility (S1 Table). NSlmb-vhh-GFP4 has been shown to result in polyubiquitination and proteasomal degradation of GFP fusion proteins [49]. Analyses of testis squash preparations revealed that SSE depletion by deGradFP resulted in abnormalities that were indistinguishable from those caused by thr-RNAi (Figs 1B, 1C and S2).
In summary, spermatocyte-specific depletion of the three separase complex subunits, THR, PIM or SSE, with different strategies (transgenic RNAi and deGradFP) was found to cause the same specific defect during meiosis. Importantly, defects start already during anaphase I. Although the degree of interference was lower in case of PIM, presumably as a result of incomplete depletion, chromosome segregation during anaphase I was affected in each case. We conclude that separase function is required for chromosome segregation during both meiotic divisions in Drosophila males.
To determine whether separase is required for the removal of alternative homolog conjunction proteins before chromosome segregation during meiosis I in Drosophila males, we analyzed the effects of THR depletion on the subcellular localization of MNM and SNM with the help of a fully functional mnm-EGFP transgene [35] and anti-SNM antibodies [35]. As previously described [35], co-localized MNM-EGFP and anti-SNM signals were detected primarily in a highly prominent spot on the XY bivalent during prometaphase and metaphase of meiosis I (Fig 2A).
In contrast, no MNM-EGFP/anti-SNM spot was detected during telophase I (n = 47) (Fig 2A) and later meiotic stages in control testis where meiosis I is normal [35]. However, after THR depletion, MNM-EGFP/anti-SNM spots were not only present early during meiosis I, but also during telophase I (n = 209 spermatocytes) (Fig 2A) and all subsequent stages up to the postmeiotic stages of sperm tail elongation (S3 Fig). In telophase I cells with chromosome bridges, the MNM-EGFP/anti-SNM spots were usually associated with the bridge (94%). Telophase I cells without obvious chromosome bridges (6%) also had always an MNM-EGFP/anti-SNM spot in one of the two daughter nuclei. For confirmation, we also performed time-lapse imaging with testis expressing MNM-EGFP and a red fluorescent histone (Fig 2B, S3 Movie and S4 Movie). In control spermatocytes, the MNM-EGFP spot was observed to disappear during early anaphase I within 2–3 minutes (n = 18 spermatocytes). In contrast, after THR depletion chromosome separation failure was accompanied by persistence of the MNM-EGFP spot during exit from meiosis I (n = 76 spermatocytes).
By anti-SNM staining of testis squash preparations, we evaluated whether PIM depletion by RNAi or SSE depletion by deGradFP also resulted in SNM perdurance beyond anaphase I (S4 Fig). These stainings clearly revealed anti-SNM signals during telophase I and meiosis II in spermatocytes depleted for these other separase complex subunits. We conclude that separase function is required for the release of alternative homolog conjunction proteins from chromosomes during progression through meiosis I.
The observed failure of MNM and SNM release from chromosomes at anaphase I onset after THR depletion is likely responsible for the associated failure of homolog separation during meiosis I. Accordingly, when alternative homolog conjunction fails to be established, THR depletion is no longer expected to cause chromosome bridging during telophase I. To evaluate this prediction, we depleted THR in mnm and snm mutants. However, we first confirmed that mutations in mnm and snm do not cause chromosome bridges during telophase I independent of THR depletion. Consistent with earlier reports [35], instead of four bivalents as in wild type, we observed up to eight chromatin masses during prometaphase I in mnm and snm mutants (Figs 2C and S5), indicating the known defect in homolog conjunction. In addition, daughter nuclei were often unequal in size and DNA content (Figs 2C and S5), reflecting the known random distribution of chromosomes during meiosis I [35]. Chromosome bridges were very rare in telophase I (Fig 3B).
The few bridge-like structures in telophase I that were observed in mnm and snm mutants presumably represent univalents lagging in the division plane rather than stretched bivalents [35]. The low frequency of chromosome bridges during meiosis I in mnm and snm mutants also confirms that co-orientation of sister kinetochores during meiosis I does not depend on homolog conjunction during male meiosis [35]. THR depletion in mnm mutants did not induce any phenotypic change during meiosis I (Figs 2C and S5). Importantly, it did not cause chromosome bridges during telophase I (Figs 2C and 3B), in striking contrast to the effect of THR depletion in spermatocytes with functional alternative homolog conjunction (Fig 1). Moreover, THR depletion did also not result in chromosome bridges during telophase I in snm mutants (Figs 3B and S5). In addition, SSE depletion by deGradFP did also no longer cause chromosome bridges during telophase I when performed in mnm or snm mutants (Figs 3B and S5). In summary, the fact that mnm and snm mutations, which abolish the alternative homolog conjunction in male meiosis, suppress chromosome bridge formation during telophase I in spermatocytes progressing through meiosis in the absence of separase function indicates that separase is required for the resolution of homolog conjunction during meiosis I.
Our conclusion that absence of separase function during Drosophila male meiosis I results in a specific failure of homolog separation is further supported by analyses with an additional mutant. tef mutants were chosen because alternative homolog conjunction is also defective in these mutants but not completely. While autosomes fail to pair into bivalents, formation of the X-Y bivalent is not affected [38]. THR depletion in tef mutants, therefore, is predicted to result in a chromosome bridge during telophase I that always represents a stretched XY bivalent, if resolution of homolog conjunction during male meiosis I requires separase function. As in case of nmn and snm mutants, chromosome bridges during telophase I in tef mutants with undisturbed separase function (Figs 2C and 3B) were infrequent and presumably represent occasional lagging autosomes. THR depletion in tef mutants resulted in an increased frequency of chromosome bridges during telophase I (Figs 2C and 3B). However, these bridges were fewer and less massive compared to those induced by THR depletion in spermatocytes with entirely normal homolog conjunction (Figs 1, 2A and 2B), consistent with the notion that only the XY bivalent forms bridges in tef mutants, while all bivalents form bridges in case of normal homolog conjunction. To demonstrate directly that the bridges resulting from THR depletion in tef mutants represent stretched XY bivalents, we performed fluorescence in situ hybridization (FISH) with a red fluorescent probe for the X and a green fluorescent probe for the Y chromosome (Fig 2D). After THR depletion in tef mutants, the red and green signals were invariably observed on the bridge (Fig 2D). Moreover, anti-SNM staining also resulted in a dot on the non-segregating DNA mass resulting after THR depletion in tef mutants (Fig 2D).
FISH was also applied to analyze the segregation of the X and Y chromosome after THR and SSE depletion in spermatocytes with normal homolog conjunction, and also in mnm and snm mutants after SSE depletion (S6 Fig). These analyses confirmed that disjoining of X and Y was inhibited in the absence of separase function and that the suppression of chromosome bridges resulting from a lack of separase function by inactivation of the alternative homolog conjunction system was paralleled by random segregation of X and Y during meiosis I. Finally, live imaging of progression through meiosis in spermatocytes with green fluorescent centromeres and red fluorescent chromosomes indicated that THR depletion causes chromosome separation failure directly and not indirectly via impairment of spindle or kinetochore function or premature exit from meiosis I (S7 Fig and S5 and S6 Movies).
While mutations inactivating the alternative homolog conjunction system during Drosophila male meiosis very effectively suppressed chromosome bridges during telophase I after THR depletion (Figs 2C and 3B), they completely failed to do so during telophase II (Fig 3A and 3C). These telophase chromosome bridges presumably reflect a failure to separate sister centromeres during meiosis II in the absence of separase function. To confirm this notion, we performed experiments with mutations in sisters on the loose (solo). SOLO has no sequence homology to known proteins [50, 51]. However, analysis of the solo mutant phenotype revealed that it provides a function analogous to that of Rec8 in other eukaryotes. Rec8 is a meiosis-specific α-kleisin. Pericentric cohesin with Rec8 instead of the non-meiosis-specific Rad21 α-kleisin is protected from separase-dependent cleavage during meiosis I but no longer during meiosis II [6]. The Drosophila genome does not contain an obvious Rec8 homolog. But SOLO expression is also meiosis-specific [50, 51]. It interacts physically and functionally with the SMC1 core cohesin subunit and it is present at meiotic centromeres until anaphase II. Mutations in solo result in premature separation of sister chromatids. The chromosome bridges observed during telophase II after THR depletion in mnm mutants are therefore predicted to be abolished when the spermatocytes also lack solo function, if the bridges reflect stretched sister chromatids. Indeed, chromosome bridges during telophase II were not only absent in solo single mutants but also after THR depletion in solo mnm double mutants (Fig 3A and 3C). Chromosome bridges during telophase II were also missing after THR depletion in solo snm double mutants (Figs 3C and S8). These results demonstrate that Separase is required for resolution of sister chromatids in meiosis II.
We point out that mutations in solo, in contrast to mutations in mnm and snm, did not abolish the THR depletion-induced chromosome bridges during meiosis I (Figs 3B and S9), indicating that the alternative homolog conjunction system functions independent of sister chromatid cohesion. Consistent with this conclusion, earlier data reported as unpublished [50] indicated that MNM and SNM are present on bivalents in solo mutants where the large majority of bivalents also remain intact through metaphase I. Anti-SNM staining revealed that the prominent SNM dot on the XY bivalent cannot be detected any longer in telophase I and subsequent stages in solo mutants (S10 Fig), indicating that the inactivation of the alternative homolog conjunction during meiosis I occurs normally. However, THR depletion in solo mutants effectively prevented the disappearance of the SNM dot during exit from meiosis I (S10 Fig). Intense SNM dots were still present during meiosis II and during the early postmeiotic stages (S10 Fig). This perdurance of the alternative homolog conjunction presumably also explains the high frequency of chromosome bridges that was observed after THR depletion in solo mutants during meiosis II (Fig 3C).
The fact that separase is no longer required for chromosome separation during meiosis I in mutants with a non-functional alternative homolog conjunction system facilitates the analysis of a putative separase requirement for sister centromere individualization during progression through meiosis I. Since centromeric cohesin is thought to unite sister centromeres into a functional unit before meiosis I [12], destruction of centromeric cohesin by separase during meiosis I might enable sister centromere individualization and biorientation during meiosis II. After THR depletion in mnm mutants, univalents are randomly distributed into the two daughter cells. If separase function during meiosis I is required for sister centromere individualization, biorientation of these univalents in the meiosis II spindle is expected to fail and co-orientation of sister kinetochores as in meiosis I will also occur during meiosis II. To monitor sister centromere behavior we performed live imaging with spermatocytes expressing green fluorescent CID/Cenp-A as well as red fluorescent histone His2Av. Splitting of sister centromeres as well as sister kinetochore biorientation in meiosis II was clearly observed not only in control spermatocytes but also after THR depletion in mnm and snm mutants (Fig 4A and S7 and S9 Movies).
In fact, sister centromere splitting and biorientation could also be clearly detected after THR depletion in spermatocytes with functional homolog conjunction although meiosis II was often highly irregular as a result of the meiosis I chromosome separation failure (Fig 4A and S8 Movie). Importantly, sister centromere individualization and biorientation during meiosis II after THR depletion was detectable only transiently during metaphase II. During progression into metaphase II, the single centromere dots present in a secondary spermatocyte were split into a pair of dots along the spindle axis. However, the resolved sister kinetochores did not move apart towards opposite poles during anaphase II in the absence of separase function as a result of the failure in releasing sister chromatid cohesion.
While the sister chromatid separation failure indicated effective THR depletion, the presence of residual separase function still sufficient for sister centromere individualization during meiosis I cannot be ruled out definitively. To address this possibility, we carefully compared the effects of THR depletion in spermatocytes with either two or only one functional thr gene copy quantitatively. A reduction of the thr gene copy number is predicted to increase the efficiency of THR depletion and hence sister centromere splitting and kinetochore biorientation might be more compromised, if separase function during meiosis I is crucial for these processes. Therefore, we also performed THR depletion in spermatocytes heterozygous for either a P-element insertion within thr (thrk07805b) or a deficiency that deletes thr (Df(2R)BSC338). These THR depletion experiments were again performed in a background with mutations in mnm or snm, where meiosis II is not accompanied by spindle irregularities resulting from chromosome separation failure during meiosis I. We quantified the fraction of secondary spermatocytes that displayed centromere splitting during metaphase II (Fig 4B). Moreover, we also determined the inter-sister kinetochore distance during metaphase II (Fig 4C) because partial centromere individualization during meiosis I might result in a reduced inter-sister kinetochore distance during metaphase II. These measurements did not reveal a difference between metaphase II spermatocytes in controls and after THR depletion in spermatocytes with only one functional thr gene copy. Finally, by estimating the duration of the different phases (prometa-, meta-, ana-, telophase) in the time-lapse movies, we assessed whether the temporal dynamics of progression through meiosis II was affected by THR depletion in either mnm or snm mutants that had only one functional thr+ gene copy (Fig 4D). Compared to controls, the average time interval between nuclear envelope breakdown and anaphase onset was found to be slightly extended in the two THR depletion cases (28.4 ± 6.2 and 26. 1 ± 4.2 versus 22.5 ± 2.5 min; n = 8, 7 and 5, respectively) but not in a statistically significant manner. Similarly, prometaphase appeared to be extended in the two THR depletion cases (Fig 4D) although variablity in these two cases was considerable. Therefore, it is conceivable that separase makes a contribution to chromosome biorientation during meiosis II but it does not appear to be essential.
We demonstrate that the alternative system used for conjunction of homologous chromosomes before meiosis I that is used in Drosophila males instead of the canonical combination of chiasmata and sister chromatid cohesion, needs to be inactivated by separase during the transition from metaphase to anaphase for normal chromosome segregation during meiosis I. Moreover, we provide evidence that the individualization of sister kinetochores, which have been proposed to become united for co-orientation during meiosis I by centromeric cohesin, does not require separase function during exit from meiosis I. Our work also demonstrates that sister separation during meiosis II depends on separase.
As separase is required for development to the stages where male meiosis occurs, spermatocyte-specific depletion of separase complex subunits had to be developed for our analysis of separase function during male meiosis. In case of the THR subunit, transgenic RNAi expressed in early spermatocytes was found to be highly effective. Live imaging after THR depletion revealed a complete failure of homolog separation during meiosis I in all cells of all analyzed spermatocyte cysts. Similarly, cytological characterization of fixed testis squash preparations revealed chromosome bridges in late meiosis I figures (anaphase I, telophase I), consistent with the live imaging results. The 10% telophase I figures that did not display chromosome bridges in the fixed samples, at least in part, represent cases where the bridges did not persist long enough. We point out that chromosome bridges resulting from a failure of homolog separation during Drosophila male meiosis I are unlike those with continuous DNA throughout, resulting in mitosis after incomplete chromosome replication or in canonical meiosis after recombination defects. During Drosophila male meiosis I, homologs are thought to be conjoined by proteinaceous links that in exceptional cases might be severed eventually by spindle forces or other processes activated during exit from meiosis I. In principle, the few exceptional telophase I figures without chromosome bridges might also indicate residual THR function. To address the possibility of incomplete knockdown, we have carefully compared the effects of THR depletion in spermatocytes with either two or one functional copy of the endogenous thr+ gene. Since the effects of THR depletion were found to be entirely independent of the thr+ gene copy number, residual THR function does not appear to be present, although we cannot rule it out completely.
RNAi can have off-target effects. Several of our findings indicate that the consequences of THR depletion reported here do not reflect such off-target effects. Expression of the RNAi-resistant UASt-thrRr transgene resulted in significant although partial suppression of the THR depletion effects. Suppression simply as a result of Gal4 titration away from the UAS-V20thrshmiR9 transgene by the UASt-thrRr transgene can be ruled out, because experiments with two UASt-thrRr copies did not produce stronger suppression. We do not understand why UASt-thrRr expression does not result in complete suppression, but suspect problems caused by an inappropriately early temporal window of bG-mediated UASt-thrRr transgene expression, insufficient mRNA stability and translational control because of missing untranslated regions. Translational control is particularly pervasive in spermatocytes and is known to occur in case of Cyclin B and Twine/Cdc25 phosphatase, two well-studied cases of meiotic M phase regulators [52, 53]. We speculate that co-translation of all the separase complex subunits late during the four day spermatocyte growth phase might be required for the production of functional separase complexes for meiosis. UASt-thrRr transcripts are not present in late spermatocytes after expression using bG. Alternative GAL4 transgenes effectively driving expression in late spermatocytes do not exist.
The fact that SSE depletion in spermatocytes by deGradFP results in the same defects as THR depletion by RNAi provides further evidence against RNAi off-target effects. In case of SSE depletion by deGradFP, we were unable to achieve suppression by expression of an UASp-Sse transgene, presumably also for the reasons discussed above. These technical difficulties have precluded meaningful experiments addressing whether the effects of SSE depletion are suppressible by expression of an SSE variant predicted to be a catalytically inactive protease.
Our live imaging has revealed that the MNM-EGFP dot, which reflects XY chromosome conjunction, disappears very rapidly during the first 2–3 minutes of anaphase I in a THR-dependent manner. These observations and the corroborating analyses with anti-SNM on fixed samples strongly suggest that homolog separation during male meiosis I requires SSE protease activity, but they do not exclude a non-proteolytic role leading to MNM/SNM re-distribution throughout the cell that might not be detectable with our tools. A direct analysis of MNM/SNM protein levels during progression through meiosis I by immunoblotting would require the isolation of sufficient amounts of precisely staged meiotic cysts. The low abundance of meiosis I cysts has prevented such analyses so far. While the MNM-EGFP dot disappears during the cell cycle phase where separase is predicted to be active as a protease, a slightly earlier disappearance would have been expected. We suspect that proteolytic inactivation of the very high amounts of MNM/SNM present on the dot on the XY bivalent might be slower than the removal of the far less concentrated material from the autosome and that the XY bivalent therefore might be the last to separate during anaphase I.
Assuming that homolog separation during male meiosis I depends on protease activity of separase raises the question what the critical substrates might be. Drosophila melanogaster MNM and SNM contain sequences conforming to the consensus of separase cleavage sites [54], but they are poorly conserved within Drosophilid orthologs, and we have been unable to detect MNM and SNM cleavage during M phase after expression in mitotically proliferating cells (preliminary observations). As the molecular basis of homolog conjunction by MNM and SNM is far from being clear, it remains readily possible that they function together with additional unknown protein partners that might be cleaved by separase.
In principle, it appears conceivable that MNM and SNM co-operate with cohesin to bring about alternative homolog conjunction during male meiosis. The fact that SNM is a distant member of the SA/Stromalin family of cohesin subunits would appear to support this notion. Accordingly, separase might target α-kleisin as its critical substrate also during male meiosis I. Several observations argue against this. SNM as well as MNM do not co-localize with the SMC1 cohesin subunit [35] and therefore appear to function independent of cohesin. Moreover, the mitotic Rad21 α-kleisin does not appear to be involved during the meiotic divisions, at least in case of female meiosis [40]. The meiosis-specific α-kleisin family protein encoded in the Drosophila genome, C(2)M, is not required for male meiosis and does not function in a Rec8-like manner during female meiosis where it is required for normal synaptonemal complex formation but not sister chromatid cohesion [39, 55]. Genetically several genes have been identified that based on their mutant phenotype appear to provide a Rec8-like function during Drosophila meiosis [50, 51, 56–59]. Their protein products ORD, SOLO, and SUNN have no sequence similarity to α-kleisins. But they are mutually dependent on each other for their predominant localization around centromeres where they are co-occurring with SMC1 and SMC3. All evidence therefore suggests that these proteins might function in a Drosophila-specific variant cohesin complex providing sister chromatid cohesion during meiosis. While it is readily possible that one of these proteins is a separase substrate during meiotic divisions, this would not explain how separase brings about homolog separation during male meiosis I for reasons also provided by our experiments. We demonstrate that homolog conjunction by MNM and SNM during male meiosis does not depend on solo function, consistent with previous work [50]. This conclusion is suggested by our observations that THR depletion in solo mutants results in chromosome bridges during meiosis I which are no longer observed when THR is depleted in solo mnm or solo snm double mutants. In the absence of solo function, therefore, MNM/SNM establish chromosome conjunction that results in chromosome bridges during meiosis I in the absence of separase function. At present, a critical target that needs to be cleaved by separase for homolog separation during male meiosis I is not known. Similarly, the critical separase target for sister chromatid separation during meiosis II is also unknown.
The possibility to by-pass the separase requirement for chromosome separation during meiosis I in Drosophila males by mutational inactivation of the alternative homolog conjunction system, in combination with our ability to monitor progression through both meiotic divisions by time-lapse imaging, has also allowed us to address the role of separase for sister centromere individualization during exit from meiosis I. A series of extremely elegant experiments in fission yeast has provided strong evidence suggesting that the co-orientation of sister centromeres that is established specifically during meiosis I for biorientation of bivalents in meiosis I spindles depends on the presence of centromeric sister chromatid cohesion mediated by meiotic Rec8 cohesin complexes [9, 26, 29, 30]. Centromeric cohesion that keeps sister centromeres in close proximity might result in the assembly of a single kinetochore on each homolog at the onset of meiosis I. Importantly, to allow sister kinetochore biorientation within meiosis II spindles, centromeric cohesion would have to be resolved at some stage after co-orientation during meiosis I has been achieved. Our observations suggest that separase might not be required for sister centromere individualization during Drosophila male meiosis. We demonstrate that sister kinetochore biorientation is successful during meiosis II after THR depletion, while sister chromatid separation is completely abolished. Separase might be dispensable for the removal of centromeric cohesin during meiosis I because of an alternative cohesin removal mechanism. In principle, Wapl can open cohesin rings without separase although only when the rings have not yet been locked by SMC3 acetylation [60–64]. Alternatively, sister centromere co-orientation in Drosophila male meiosis might not involve centromeric cohesin. Finally, we acknowledge that our evidence cannot definitely rule out the possibility that some residual THR escaping depletion might still be sufficient for normal sister kinetochore biorientation but not for sister chromatid separation during meiosis II. Moreover, we also point out that our light microscopic analyses cannot resolve the aspects of sister centromere individualization during Drosophila male meiosis I that have been observed by serial sectioning and electron microscopic analysis [7]. At the ultrastructural level, a hemispherical kinetochore, where the two sister kinetochores cannot be resolved, is detected in early prometaphase I spermatocytes. By anaphase I, however, two closely associated but clearly distinct sister kinetochores in a side-by-side configuration were usually observed. For lack of spatial resolution, we cannot exclude that extent or dynamics of this sister kinetochore resolution process during meiosis I is abnormal after THR depletion. Similarly, our data can also not exclude that a possible separase contribution to sister kinetochore biorientation during meiosis II might eventually be compensated during prometaphase II in THR depleted spermatocytes by spindle forces for example. Our limited data on temporal dynamics of chromosome congression during meiosis II after THR depletion in mnm and snm mutants is consistent with the notion that separase makes some contribution to efficient chromosome biorientation during meiosis II but cannot prove it. The analysis of the role of separase for sister kinetochore biorientation during meiosis certainly deserves further attention, including studies in other organisms.
The following lines with previously characterized mutations or transgenes were used: Sse13m and Df(3L)SseA [42], thrk07805b [65], Df(2R)BSC338 [66], P{ry+, hsp70-mnm-EGFP}, mnmZ3-3298, mnmZ3-5578, snmZ3-0317, and snmZ3-2138 [35], soloZ2-0198 and soloZ2-0338 (Yan et al., 2010), tefZ2-4169 and tefZ2-3455 [67], P{w+, bamP-GAL4-VP16}III [68], P{w+, His2Av-mRFP}II.2 and P{w+, gcid-EGFP-cid}II.1 [69], P{w+, pUbi-EGFP-alphaTub84B}II [70], and P{w+, pUbi-EYFP-asl} [71].
Lines for transgenic RNA interference, UAS-V20thrshmiR9, UAS-V20thrshmiR10 and UAS-W20thrshmiR45, were generated by integrating the pVALIUM20 and pWALIUM20 constructs (see below) into the attP2 landing site. For production of P{w+, UASt-thrRr}attP40, allowing expression of a thr cDNA with silent mutations in the regions targeted by thrshmiR9, thrshmiR10 and thrshmiR45, a pUASt-attB construct (see below) was integrated into the attP40 landing site. For pim-RNAi, we used y w1118; P{w+, KK106514}VIE-260B (v100534). PBac{3xP3-ECFP, gEGFP-Sse}III.1 was generated by germline transformation with a PiggyBac construct. P{w+, bamP-NSlmb-vhh-GFP4}II.1 was isolated after P-element-mediated germline transformation with a pCaSpeR4 construct.
For all experiments, flies were cultured at 25°C. Detailed genotypes of the flies analyzed are provided in the supplemental material (S1 Table).
For the production of transgenic lines allowing GAL4-dependent expression of short hairpin microRNAs (shmiRs), we generated constructs using the vectors pVALIUM20 and pWALIUM20 [72]. Inserts were generated by annealing the following oligonucleotides: 5'-ctagcagt-CCCTTGGAAGCTACAAGTCAA-tagttatattcaagcata-TTGACTTGTAGCTTCCAAGGG-gcg-3' and 5'-aattcgc-CCCTTGGAAGCTACAAGTCAA-tatgcttgaatataacta-TTGACTTGTAGCTTCCAAGGG-actg-3' for thrshmir9, 5'-ctagcagt-AACGCTTCTAGTTCAACTAAA-tagttatattcaagcata-TTTAGTTGAACTAGAAGCGTT-gcg-3' and 5'-aattcgc-AACGCTTCTAGTTCAACTAAA-tatgcttgaatataacta-TTTAGTTGAACTAGAAGCGTT-actg-3') for thrshmir10, 5’-ctagcagt-AAGAAGTAGATCATTCTTCAA-tagttatattcaagcata-TTGAAGAATGATCTACTTCTT-gcg-3’ and 5’-aattcgc-AAGAAGTAGATCATTCTTCAA-tatgcttgaatataacta-TTGAAGAATGATCTACTTCTT-actg-3’ for thrshmir45. Capital letters indicate the regions corresponding to sequences within the thr coding sequence.
For the production of a transgenic line allowing expression of EGFP-Sse under control of the Sse cis-regulatory region, we generated pBac{3xP3-ECFP-gEGFP-Sse} using a modified version of the previously described gSse transgene construct [42]. During construction, an Sse cDNA fragment replacing the Sse genomic region containing the first two introns was introduced as well as the EGFP coding sequence fused at the N-terminus.
For the construction of pCaSpeR4-bamP-NSlmb-vhh-GFP4, the fragment coding for NSlmb-vhh-GFP4 was amplified from pUASt-NSlmb-vhh-GFP4 [49] using primers SCH1 (5’-GACTACCGGTATGATGAAAATGGAGACTGAC-3’) and SCH2 (5’-GACTGCGGCCGCTTAGCTGGAGACGGTGAC-3’). After digestion with AgeI and NotI, the insert was used to replace the GAL4-VP16 coding region released by the same restriction enzymes in pCaSpeR4-bamP-GAL4-VP16 [68].
A modified thr cDNA was inserted into pUASt-attB [73] for the production of pUASt-attB-thrRr. The thr cDNA [48] was modified by replacing the region containing shmiR target sequences with a synthetic variant (GenScript, Piscataway, NJ 08854, USA).
The fertility of 5–10 single males per genotype was assayed in parallel. Each single male was allowed to mate with three w virgin females for two days. Flies were transferred into a fresh vial and discarded after two more days. Flies eclosing from this vial were counted.
Testis squash preparations were done as described [74]. For immunolabeling, mouse monoclonal anti-α-tubulin DM1A (Sigma) was used at 1:10’000 and affinity-purified rabbit polyclonal anti-SNM [35] at 1:250. Secondary antibodies were Alexa488- or Alexa568-conjugated goat antibodies against mouse or rabbit IgG diluted 1:1000.
Dissection of testis, fixation with 4% PFA, permeabilization with PBST-DOC and anti- α-tubulin staining were done as described (protocol 3.2.2, steps 1–14) [74]. Cy5-conjugated goat anti-mouse IgG diluted 1:1000 was used as secondary antibody. Ethanol incubations and dehydration with a formamide series were also done as described (immuno-FISH protocol 3.2, steps 10–26) [75]. An oligonucleotide (5'-TTTTCCAAATTTCGGTCATCAAATAATCAT-3') with Atto-565 on 5’ and 3’ end (Integrated DNA Technologies, B-3001 Leuven Belgium) was used for detection of the X-specific 359 bp repeats at a concentration of ≈ 1 ng/μl in hybridization Buffer. An oligonucleotide (5'-AATACAATACAATACAATACAATACAATAC-3') with Alexa-488 fluorophore at the 3’ end (Sigma-Aldrich, 8107 Buchs, Switzerland) was used for detection of the Y-specific AATAC repeats at a concentration of ≈ 2 ng/μl in hybridization buffer. The denaturation step was performed at 98°C for 6 min, and hybridization over night at 16°C. Slides were washed twice in 50% formamide, 2x SSCT at 16°C for 1 hour each. Thereafter, additional washes were performed at room temperature in 25% formamide, 2x SSCT for 10 min and three times in 2x SSCT for 10 min each. DNA was stained with Hoechst 33258 (1 μg/ml) for 10 min and slides were washed twice in PBS for 5 min. Slides were mounted in 70% Glycerol, 50 mM Tris-HCl pH 8.5, 10 mg/ml propyl gallate, 0.5 mg/ml phenylendiamine.
Image stacks with 250 nm spacing between focal planes were acquired with a 63×/1.4 oil immersion objective on a Zeiss Cell Observer HS microscope. If not stated differently, the images displayed in the figures represent maximum intensity projections. The data used for statistical analyses of a particular genotype was obtained from multiple slides and each slide was prepared with about 14 dissected testes.
Testes from pupal or adult males were dissected in Schneider’s Drosophila Medium (Invitrogen, #21720), 10% fetal bovine serum (Invitrogen), 1% penicillin/streptomycin (Invitrogen, #15140). The dissected testes were transferred into 40 μl of medium in a 35 mm glass bottom dish (MatTek Corporation, #P35G-1.5-14-C), and opened with fine tungsten needles to release the cysts. To reduce movements within the sample, methylcellulose (Sigma) was added. A wet filter paper was placed inside along the dish wall before sealing the lid with parafilm. Time-lapse imaging was performed with a spinning disc confocal microscope (Visitron) with a 60×/1.4 oil immersion objective. Image stacks with 24–45 focal planes spaced by 0.5–1 μm were acquired with a time interval of 30–60 sec. Precise numbers are specified in the legends for each supplementary movie. The distance between sister centromeres during metaphase II was measured in 3D using Imaris software (Bitplane). To exclude that the measurements actually represent early anaphase time points, the last three time points before anaphase onset were excluded from consideration.
|
10.1371/journal.pcbi.1000970 | An Automated Phenotype-Driven Approach (GeneForce) for Refining Metabolic and Regulatory Models | Integrated constraint-based metabolic and regulatory models can accurately predict cellular growth phenotypes arising from genetic and environmental perturbations. Challenges in constructing such models involve the limited availability of information about transcription factor—gene target interactions and computational methods to quickly refine models based on additional datasets. In this study, we developed an algorithm, GeneForce, to identify incorrect regulatory rules and gene-protein-reaction associations in integrated metabolic and regulatory models. We applied the algorithm to refine integrated models of Escherichia coli and Salmonella typhimurium, and experimentally validated some of the algorithm's suggested refinements. The adjusted E. coli model showed improved accuracy (∼80.0%) for predicting growth phenotypes for 50,557 cases (knockout mutants tested for growth in different environmental conditions). In addition to identifying needed model corrections, the algorithm was used to identify native E. coli genes that, if over-expressed, would allow E. coli to grow in new environments. We envision that this approach will enable the rapid development and assessment of genome-scale metabolic and regulatory network models for less characterized organisms, as such models can be constructed from genome annotations and cis-regulatory network predictions.
| Computational models of biological networks are useful for explaining experimental observations and predicting phenotypic behaviors. The construction of genome-scale metabolic and regulatory models is still a labor-intensive process, even with the availability of genome sequences and high-throughput datasets. Since our knowledge about biological systems is incomplete, these models are iteratively refined and validated as we discover new connections in biological networks, and eliminate inconsistencies between model predictions and experimental observations. To enable researchers to quickly determine what causes discrepancies between observed phenotypes and model predictions, we developed a new approach (GeneForce) that automatically corrects integrated metabolic and transcriptional regulatory network models. To illustrate the utility of the approach, we applied the developed method to well-curated models of E. coli metabolism and regulation. We found that the approach significantly improved the accuracy of phenotype predictions and suggested changes needed to the metabolic and/or regulatory models. We also used the approach to identify rescue non-growth phenotypes and to evaluate the conservation of transcriptional regulatory interactions between E. coli and S. typhimurium. The developed approach helps reconcile discrepancies between model predictions and experimental data by hypothesizing required network changes, and helps facilitate the development of new genome-scale models.
| A current challenge in systems biology is reconstructing transcriptional regulatory networks from experimental data (e.g. gene expression, genome sequence, and DNA-protein interaction), due to the complexity of interactions in these networks and the limited information on network components and interactions for most organisms [1], [2]. Even for well-studied model organisms, such as E. coli and Saccharomyces cerevisiae, inferred or indirect regulatory interactions have to be included in genome-scale transcriptional regulatory reconstructions due to existing knowledge gaps in how genes are transcriptionally regulated [3], [4]. Reconstructed regulatory networks will be incomplete, reflecting incomplete knowledge about cis-regulatory networks, and may include incorrect interactions. As such, methods for iterative validation and refinement of regulatory reconstructions are needed in order to assess new experimental datasets as they emerge [5], [6]. Such approaches need to identify and eliminate inconsistencies between the reconstructed network and new experimental data, and to include newly discovered network interactions [3]. However, identifying the cause of inconsistencies in a highly interconnected network using manual efforts is not a trivial task, and can be labor intensive particularly for genome–scale transcriptional regulatory network models. Therefore, systematic approaches that automate this iterative procedure are useful for identifying new or incorrect connections in biological models; such approaches have been developed for analysis and correction of metabolic networks [7]–[9]. In this paper, we present an approach that allows for the automated adjustment of an integrated genome-scale metabolic and transcriptional regulatory network model, by comparing the emergent properties of the integrated networks to cellular growth phenotypes. These adjustments result in testable hypotheses about transcriptional regulation and metabolism in organisms.
While there are many types of regulatory modeling approaches (reviewed in [10]), Boolean modeling of regulatory interactions can be beneficial when modeling large-scale regulatory networks because (i) such formalism requires minimal parametric details to be incorporated [11], and (ii) these Boolean models can be integrated with constraint-based metabolic models [3], [4]. One of the commonly used constraint-based modeling approaches for metabolic models is flux balance analysis (FBA), which predicts an optimal steady-state flux distribution in a metabolic network [12]. This can be extended to integrated metabolic and regulatory models, referred to as regulated flux balance analysis (rFBA), which accounts for transcriptional regulation as well as the other governing physicochemical constraints [13], [14]. While the metabolic and regulatory models in rFBA are solved iteratively, newer approaches for integrating metabolic and regulatory models allow the models to be combined into a single model using an mixed-integer linear programming (MILP) formalism [15]. In this case, steady-state regulatory flux balance analysis (SR-FBA) is used to identify optimal flux distributions that satisfy both models simultaneously. We have recently developed an efficient SR-FBA formulation that systematically integrates transcriptional regulatory and metabolic networks [16] which was used here.
In this work, we developed an algorithm (called GeneForce) to reconcile integrated regulatory and metabolic model predictions with experimental data, by automatically identifying and overriding transcriptional regulatory rules that cause inconsistencies between model predictions and experimental observations. The approach can be used in cases where both the experimental data and an un-regulated metabolic model agree on a positive growth phenotype (cells can grow), but the integrated metabolic and regulatory model predicts a non-growth phenotype (cells cannot grow). In these cases, the GeneForce algorithm allows the integrated metabolic and regulatory model to achieve growth by violating regulatory rules as needed, while minimizing the total number of regulatory violations in order to maximally preserve the original regulatory interactions present in the regulatory network reconstruction. These rule violations indicate that regulatory rules describing gene expression are incorrect or that isozymes or alternative pathways are present in the metabolic network.
We first applied the GeneForce method to refine the genome-scale transcriptional regulatory network for E. coli, iMC1010v1 [3] which was updated here to include newly discovered Lrp regulatory interactions [17]. The algorithm was used to analyze a large collection of ∼50,000 E. coli knockout mutant growth phenotypes [18], [19], and the suggested regulatory corrections resulted in a ∼1–8% improvement in model accuracy over the original models, which had already been adjusted during their initial development to improve model accuracy. In addition to correcting regulatory rules, we applied the GeneForce algorithm to predict genes that, if overexpressed or constitutively expressed, could rescue non-growth phenotypes of E. coli strains (wild-type or mutants) in certain growth environments. Finally, we applied the GeneForce method to investigate the conservation of transcriptional regulatory interactions between E. coli and S. typhimurium. The E. coli transcriptional regulatory rules were integrated with a metabolic model for S. typhimurium that included metabolic genes and reactions from a recent metabolic reconstruction iRR1083 [20]. GeneForce suggested a small set of rule corrections for this hybrid network model were needed, based on analysis of S. typhimurium growth phenotyping data, suggesting that regulation may be highly conserved between these two organisms. While the approach has been used here to correct Boolean representations of transcriptional regulation, it could easily be extended to consider non-Boolean approaches to modeling transcriptional regulation as they are developed.
We developed an automated MILP approach, GeneForce, to identify problematic Boolean regulatory rules in an integrated metabolic and transcriptional regulatory model. The method identifies regulatory rules that prevent the models from predicting cellular growth in conditions, which are capable of supporting growth experimentally. The approach can be used when the integrated metabolic and regulatory model does not predict growth, but experimental data and metabolic model predictions (without any regulatory constraints) indicate growth occurs.
The basic idea of the GeneForce algorithm is to allow the integrated metabolic and regulatory model to violate a minimal set of transcriptional regulatory rules so that growth can occur in a particular condition. The algorithm therefore adds an additional constraint that the model must satisfy a minimal threshold growth rate. The algorithm uses a set of ‘rule-violation’ equations (see Supporting Information Text S1 for details) to relax certain regulatory constraints (by allowing for expression of un-expressed genes) thus allowing the model to readjust the metabolic and regulatory model solution space to include solutions with growth rates exceeding the minimum threshold. The ‘rule violation’ equations, invoked at the gene level, allow the regulatory rules for metabolic genes to be violated using additional surrogate gene expression indicators (y′g) that can differ in value from the gene expression indicators (yg), the latter of which are determined by Boolean regulatory rules. Normally in the integrated metabolic and regulatory model a flux is constrained to be zero if the necessary metabolic genes are determined to be un-expressed (yg = 0). In the GeneForce algorithm, the bounds on the metabolic fluxes are dependent on y′g instead of yg. The reaction dependence on y′g then allows the model to override a minimum number of gene expression indicators (where yg = 0 but y′g = 1) so that the threshold growth rate can be achieved.
The example in Figure 1 illustrates how the GeneForce algorithm uses the rule violation technique to achieve non-zero growth in an integrated metabolic and regulatory model in agreement with the metabolic model predictions. As shown in the Figure 1A the metabolic model predicts positive growth in the presence of Axt, whereas the integrated metabolic and regulatory model predicts no growth due to the regulatory interactions between gene G1 and transcription factor TF1 (Figure 1B). Expression of G1 is needed for growth since the corresponding enzyme catalyzes an essential reaction (B→C), but expression of G1 requires TF1 to be active, and the binding activity of TF1 is inhibited by metabolite A. This non-growth phenotype is overcome in the GeneForce algorithm by making the reaction availability dependent on the surrogate gene expression indicator y′G1, which is not dependent on the regulatory rules. In this case, yG1 = 0 and y′G1 = 1 since the associated reaction is essential for growth (Figure 1C). Since the flux ranges in GeneForce are directly dependent on y′g rather than yg, the reaction associated with G1 can carry flux through it allowing growth to occur, even though the gene is not expressed in the Boolean model. It should be noted that the other regulatory rule in the network (yG2 repressed by TF2) was not overruled (yG2′ = yG2 = 0), even though yG2′ could also take the value 1 instead of 0. This is because the algorithm minimizes the sum of the distances between the surrogate and the original gene expression indicators, (), and hence the number of regulatory rule violations by the algorithm. Minimization of this objective function forces the binary vector of y′g to remain as close as possible to that of yg, thus minimizing the number of rule violations. This ensures that the original, literature-derived regulatory rules are maximally conserved and reflected in the predicted behavior of the integrated metabolic and regulatory network.
We used the GeneForce algorithm to refine the regulatory rules in an updated metabolic and regulatory E. coli model based on iMC104 (the regulatory portion of the integrated iMC1010v1 model [3]), where we had revised the regulatory rules for genes in the Lrp regulon based on experimental data [17] (see methods for details). The Lrp-modified iMC104 model was combined with a metabolic model and the resulting integrated model was used to predict growth phenotypes that were compared to experimental growth phenotypes for a large number of knockout mutants tested for growth in various conditions [18], [19]. The model refinements were carried out in three successive steps. First, the updated regulatory rules from iMC104 were integrated with the metabolic model iJR904 [21] and rule corrections were made to give the first refined version of the regulatory model iMC105A. Second, the iMC105A regulatory model was integrated with an updated metabolic network iAF1260 [22] and adjusted to give the second refined regulatory model, iMC105AB. Finally, the iMC105AB regulatory model was further refined using phenotypic data generated in this study for three global transcription factor knockout mutants (ΔarcA, ΔpurR and Δlrp) to give the final version of the regulatory model iMC105ABC. Here we consider the regulatory models (iMC104, iMC105A, iMC105AB, and iMC105ABC) to be just the regulatory part of the integrated models (the number indicates the total number of transcription factors).
Integration of iJR904 with Lrp-modified iMC104 regulatory rules allowed comparison of 32,050 growth phenotype predictions to experimental data (Supporting Information Table S1). The GeneForce algorithm identified genes with possible problematic regulatory rules in 3,079 out of the 32,050 cases examined, where each case represents a mutant grown in a different condition. Alternative optimal solutions exist for only 298 of the 3,079 cases, where most (281 out of 298) were needed to correct predictions for growth on L-serine as a nitrogen source or the ΔsdaB mutant. These 3,079 cases correspond to cases where a zero growth prediction by the integrated metabolic and regulatory model contradicted both the experimental data and the metabolic model prediction (+/+/−; where + indicates growth and − indicates no growth, and the order corresponds to the results from experiments / metabolic model / integrated model). Not all regulatory rules identified by the algorithm in the 3,079 cases were adjusted, as they may cause new incorrect predictions in other conditions. Instead, corrections were made for regulatory rules that were frequently identified as problematic for a particular knockout mutant or growth environment (Table 1, refinement step A). In total, regulatory rules for ten genes (glmU, ilvY, ilvC, sdaC, cycA, gcvB, dsdX, rpiR, acnA, and ilvA) were corrected in the first regulatory model refinement, iMC105A. Two gene-protein-reaction (GPR) associations were also corrected in the metabolic model for two amino acid transport reactions (L-methionine and D-serine). These model adjustments led to an ∼8% improvement in the overall accuracy of the integrated model from 73.9% to 81.5% (Table 2 and Figure 2A). Due to the addition of the transcriptional regulator gcvB, this revised regulatory network iMC105A contained a total of 105 transcription factors.
The second set of refinements (refinement step B), occurred when the iJR904 metabolic network was replaced with the updated metabolic network iAF1260 [22]. The inclusion of the latest metabolic network allowed integrated model predictions to be compared against 50,327 growth phenotypes, since more genes and environments are represented in this larger metabolic network (Supporting Information Table S2). Using this extended set of growth phenotypes, the algorithm identified a new set of problematic regulatory rules in iMC105A, and corrections were made for eleven additional genes, argD, astCADBE, speA, metH, thrA, rhaS, and rhaR (Table 1, refinement step B) leading to a second revision of the regulatory model, iMC105AB. Initial correction of the regulatory rule for argD fixed 262 errors (+/+/− changed to +/+/+) associated with cases where arginine is the nitrogen source, but also introduced 297 new errors (−/+/− changed to −/+/+) for cases where arginine is the carbon source. To correct these new errors we subsequently refined the rules for the astCADBE operon and speA gene, in addition to argD, to reconcile the model with both arginine conditions. The prediction accuracy of the integrated metabolic and regulatory model (iAF1260+iMC105A) was 78.1% before all eleven rule corrections were made, and with this additional second set of regulatory refinements, the iMC105AB model could achieve a slightly higher accuracy, 79.9% (Table 2 and Figure 2A) and with significantly greater coverage of the available experimental data (50,327 cases versus 32,050 cases).
In the third set of refinements, the refined regulatory model, iMC105AB, was tested by comparing predictions to newly acquired experimental data for three transcription factor knockout mutants (ΔarcA, ΔpurR, and Δlrp). The five transcription factors in iMC105AB with the most metabolic gene targets are Crp, Fnr, ArcA, PurR and Lrp. Experimental data was already available for knockout mutants for two of these transcription factors (Crp and Fnr), however, growth phenotyping data for ΔarcA, ΔpurR, and Δlrp mutants was not available. Therefore, growth experiments on phenotype microarrays (Biolog, Hayward, CA) were conducted for the three mutant strains ΔarcA, ΔpurR, and Δlrp. The GeneForce algorithm identified an additional nine genes needing regulatory rule corrections (Table 1, refinement step C) which all caused incorrect phenotype predictions for the Δlrp mutant (Supporting Information Table S3). Incorporation of these Lrp specific corrections led to the final refined version iMC105ABC, which resulted in a small overall improvement (0.01%) in model accuracy over the previous version iMC105AB (Table 2 and Figure 2A) when evaluating all data, but a large improvement for the new phenotype measurements of the three transcription factor deletion mutants (81.7% accuracy using iMC105ABC vs. 66.1% using iMC105AB for Δlrp, ΔpurR, and ΔarcA phenotypes).
For each +/+/− case the minimum number of genes whose regulatory rules had been violated by the GeneForce algorithm was determined. The distribution of the number of rule corrections needed for the +/+/− cases is shown for the first two refinement steps in Figure 3 before and after the model adjustments were made (listed in Table 1). The results show that in most cases a single regulatory rule was preventing the integrated model from making the correct prediction. The first set of refinements eliminated most of the +/+/− cases (Figure 3A), leaving fewer genes needing rule corrections in the subsequent steps (Figure 3B), even though more experimental data (50,557 versus 32,050) could be compared to model predictions.
Alternative optimal solutions were generated for each refinement step by adding integer-cut constraints and re-solving the GeneForce problem. The number of +/+/− cases for which alternative optimal solutions exist can be found in Supporting Information Table S4. In most cases, they were specific to a particular knockout mutant or growth environment, and the alternative optimal solutions were two or three isozymes catalyzing an essential reaction. For some instances we were able to find enough information in the literature to determine the most likely isozymes involved (ilvA, metH, sdaC and thrA). For other cases described below (dctA, rpiB, and ilvBN), we performed additional growth phenotyping experiments to determine the final set of corrections. Overall application of the GeneForce algorithm to correct +/+/− cases led to (i) changes in the regulatory rules for metabolic genes (e.g. glmU, ilvBN, and dctA) (ii) changes in the rules for TF activities (e.g. RpiR), or (iii) changes in the gene-protein-reaction (GPR) associations in the metabolic network (e.g. dsdX). Some examples from the different types of changes are presented below, and in some cases additional mutant phenotypes were screened by experiments to confirm the necessary model changes identified by GeneForce.
In our analysis, the regulatory rule describing the regulation of glmU by the NagC transcription factor was identified as the most problematic rule, causing approximately one third of the total incorrect zero growth predictions by the Lrp-modified iMC104+iJR904 integrated model. GeneForce identified the regulatory rule for glmU gene as needing a correction for most mutants grown in conditions where any of the three amino sugars, N-acetylglucosamine (GlcNAc), N-acetylneuraminate, and N-acetylmannosamine were present. GlmU catalyzes two consecutive reactions producing an essential precursor UDP-N-acetyl-glucosamine (UDP-GlcNAc) for the cell wall of E. coli [23], [24]. This gene has been found to be essential in E. coli [25], supporting GeneForce's prediction that the regulatory rule for glmU is incorrect.
The following two regulatory rules in iMC104 precluded the gene from being expressed in the integrated model under certain conditions: ‘NagC is active if NOT (GlcNAc OR glucosamine-6-phosphate)’ and ‘glmU is expressed if NagC is active’. The first rule prohibited NagC from being active in the presence of any of the three amino sugars because glucosamine-6-phosphate is a common intermediate in their degradation pathways. The inactivity of NagC subsequently prohibited the expression of glmU in the integrated model, resulting in a non-growth phenotype prediction. GeneForce violated the glmU regulatory rule so that GlmU can carry out the two essential reactions. Although the regulatory rules shown above were in agreement with experimental data reported in the literature, the Boolean representation of the regulatory interaction was too stringent in the model. The glmU gene contains two upstream promoters P1 and P2, and the transcription factor NagC is shown to induce expression using the promoter P1 in the absence of any of the three amino sugars [26]. However, the second glmU promoter, P2, is weakly induced in presence of N-acetylglucosamine, suggesting that the activating role of NagC could be dispensable for this promoter [26]. This suggests that the expression of glmU is not completely abolished when NagC is inactive, and that the low level of induction at P2 is still sufficient to allow for the production of UDP-GlcNAc. Since glmU is required for growth in other environments as well, it is always expressed in the refined set of regulatory rules.
The integrated model made incorrect predictions for the Δlrp mutant in a few different conditions, including growth on glucose, gluconate, and L-malate as sole carbon sources. For the glucose and gluconate conditions, GeneForce found that either ilvHI or ilvBN needed to be expressed since these two isozymes are used for the synthesis of branched chain amino acids. To evaluate which of these isozymes is used by the cells, we screened a number of Lrp double mutants for growth on glucose and found that only ΔlrpΔilvB is unable to grow; however, ΔlrpΔilvN, ΔlrpΔilvH, and ΔlrpΔilvI were all capable of growing in glucose minimal media (Figure 4A). This is consistent with earlier reports that the catalytic subunits (ilvB and ilvI) are still active in the absence of the smaller regulatory subunits (ilvM, ilvN and ilvH) [27]. To reconcile the positive growth phenotype of a Δlrp mutant grown on malate, GeneForce needed to override the regulatory rule for one of the malate transporters in E. coli. We subsequently found that ΔdctA and ΔlrpΔdctA mutants did not grow on L-malate, while the Δlrp mutant grew (Figure 4B) , implying that the dctA rule needed correction.
Ribose-5-phosphate isomerase (RPI) catalyzes the reversible conversion of ribose-5-phosphate to ribulose-5-phosphate in the pentose phosphate pathway. Two RPIs have been identified in E. coli, RpiA and RpiB, which are genetically and biochemically distinct. RpiA is constitutively expressed and accounts for most of the RPI activity in wild-type cells [28]. RpiB also functions as an allose-6-phosphate isomerase, catalyzing the second step in the allose degradation pathway [29]. It has been shown that rpiB expression is repressed by a regulator, RpiR, which is located on the same operon [30]. We subsequently measured growth of ΔrpiA, ΔrpiB, and ΔrpiAΔrpiB mutants on D-ribose and D-allose, and found that only the double deletion exhibited a lethal phenotype on D-ribose (Figure 4C), while neither ΔrpiB nor ΔrpiAΔrpiB mutant grew on D-allose (Figure 4D). The original regulatory rule for RpiR had only D-ribose as an inducer; as a result no growth in the D-allose medium condition was incorrectly predicted by the integrated model. We subsequently changed the rule for rpiR to also include D-allose as an inducer based on the study of allose catabolism [31].
Another interesting case was the suggested regulatory refinements for cycA based on the utilization of D-alanine and D-serine. In this case the algorithm helped lead to improvements in GPR associations in the metabolic network as well as the regulatory rule for cycA. The integrated model incorrectly predicted that these compounds could not be used as carbon and nitrogen sources, which GeneForce attributed to the expression rule for the CycA transporter. We subsequently measured growth of ΔcycA, ΔdsdX, and ΔcycAΔdsdX mutants, and found that the ΔcycA and ΔcycAΔdsdX mutant were unable to grow with D-alanine as a carbon source (Figure 4E) and that only the ΔcycAΔdsdX double mutant was unable to grow with D-serine as a carbon source (Figure 4F). This indicates that cycA is expressed under both conditions and that dsdX is also expressed when D-serine is present as a carbon source. The transport of D-serine by DsdX has only been shown in an uropathogenic strain of E. coli [32] and based on our phenotyping results this protein appears to have the same function in BW25113 as well. As a result, the DsdX transporter was to be added to the metabolic model and the regulatory rule for cycA was modified. Altogether, these experimental results illustrate how GeneForce can help identify incorrect regulatory rules or missing metabolic functionality which cause model-data discrepancies.
To investigate the effects the model corrections have in other conditions we evaluated how many new false positives were introduced (i.e. −/+/− cases became −/+/+) for each refinement step (Supporting Information Table S4) and whether the predicted flux distributions would change using flux variability analysis [33]. In refinement steps A and B, the number of new false positives was only ∼7% of the total number of corrected errors. Although 17 new false positives were introduced in refinement step C to correct 53 model errors, the corrections were supported by experimental results. Aside from the argD case described above, we did not find any rule corrections that caused significantly more false positives for other knockout mutant or medium conditions. We further evaluated the effects the model changes had on predicted wildtype optimal metabolic flux distributions. Flux variability analysis was done before and after Refinement A with iJR904 and before and after Refinements B+C with iAF1260. This analysis was done for conditions in which the models predict non-zero growth rates before and after the refinements (84 media conditions for iJR904 and 112 media conditions for iAF1260), since the model changes were not intended to affect these conditions. We found that the model changes had no significant effect on the predicted wildtype fluxes for the 84 and 112 conditions examined (maximum and minimum predicted flux values changed by less than 0.004 mmol/gDW/hr, which corresponds to ∼0.04% of the carbon source uptake rates), except for the two conditions with L-malate and D,L-malate as carbon sources. In these two conditions, the regulatory rule change for dctA in Refinement C allows D-malate to be transported and L-malate to be transported with a more energetically efficient transporter. As a result higher growth rates can be achieved for these two conditions and the optimal flux distributions will change significantly.
In addition to identifying regulatory rules that cause inconsistencies between model predictions and experimental growth phenotypes, another utility of the GeneForce algorithm is to identify genes whose transcriptional regulation prevents cells from growing. In this case the integrated model and regulatory rules are correct, and the un-expressed state of certain metabolic genes prevents the cells from utilizing a particular carbon or nitrogen source. The algorithm functions in the same manner as before, with the difference being that it is used in −/+/− cases in which cells are incapable of growing experimentally, the metabolic model indicates that the genes necessary to support growth are present in the genome, but the integrated metabolic and regulatory model correctly predicts a non-growth phenotype because the necessary genes are not expressed. While the algorithm would falsely violate regulatory rules in order to allow the model to achieve a non-zero growth rate, such false violations are of interest since they indicate which genes if over-expressed would allow for growth. Experimentally, such results could be tested by increasing the expression of the identified genes.
In our analysis of ∼32,000 mutant phenotypes using iJR904+iMC105A, we identified ten medium conditions, where the GeneForce algorithm repeatedly identified genes whose over-expression could enable aerobic growth of mutant (and likely wild-type) E. coli strains. In each of these nutritional states: (i) the majority of the E. coli knockout mutants were unable to grow, (ii) the metabolic model incorrectly predicts growth, and (iii) the integrated metabolic and regulatory model correctly predicts no growth. In seven out of the ten aerobic conditions, either a single gene or a single operon was needed to be expressed in violation of the regulatory rules to allow for growth. This list included citT, xylA, allC, fucO, atoDAEB, ttdAB, and nirBD, which correspond to the different medium conditions listed in Table 3. The distribution of the number of genes needing overexpression to rescue these non-growth phenotypes (−/+/−) occurred in the first two refinement steps (Supporting Information Table S1 and Supporting Information Table S2) is shown in Figure 5. Similar to the case for rule corrections (Figure 3), most of the rescue non-growth cases required over-expressing a single gene. The refinement of the regulatory rules (listed in Table 1) slightly reduced the number of rescue non-growth cases by ∼1–10% as some cases changed from −/+/− to −/+/+ (Figure 5).
We subsequently looked for experimental evidence in the literature that would corroborate the algorithm's predictions of genes whose overexpression can rescue non-growth phenotypes. We found direct evidence in support of citT, fucO and atoDAEB rescuing the inability of wild-type E. coli to grow aerobically on citrate, 1,2-propanediol, and butyrate, respectively [34]–[38]. The citT gene encodes a citrate transporter, and Pos et al. have shown that plasmid mediated over-expression of citT allows for aerobic growth on citrate [34]. The 1,2-propanediol oxidoreductase (FucO), is required for growth on 1,2-propanediol anaerobically, but under aerobic conditions this gene is not expressed preventing utilization of this compound. Constitutive expression of fucO leads to an ability to grow on 1,2-propanediol aerobically [35], [37]. Wild-type E. coli is unable to utilize saturated short chain fatty acids, such as butyrate, and studies have shown that the constitutive expression of atoC, an activator of the atoDAEB operon, instills the ability to grow on butyrate [36], [38].
We were unable to find direct evidence in support of allC (allantoin) and ttrAB (L-tartrate) but these genes encode enzymes in the catabolic pathways for these substrates. Wild-type E. coli can utilize allantoin as a sole nitrogen source anaerobically, but not as a sole carbon source [39]. The inability to degrade allantoin aerobically is thought to be due to the oxygen mediated inhibition of the regulatory gene allS, which is an activator of the allantoin regulon containing allC [39]. It is possible that constitutive expression of allC would allow for utilization of allantoin in an oxic environment. Similar strategies may also be proposed for the ttrAB operon, which is also repressed in the presence of oxygen thereby preventing aerobic growth on L-tartrate, a substrate that can be used anaerobically [40].
The two bacterial strains, S. typhimurium LT2 and E. coli K-12 MG1655 are closely related and both organisms have been well studied experimentally and modeled. However, the transcriptional regulatory network of S. typhimurium is less characterized experimentally, than E. coli's, and a genome-scale transcriptional regulatory model for S. typhimurium is not available. Recently, a metabolic network model, (iRR1083) for S. typhimurium was published [20], and we investigated the effects of conserving the E. coli transcriptional regulatory interactions in S. typhimurium by superimposing the E. coli regulatory constraints on the Salmonella metabolic network. The regulatory model iMC105A was integrated with the metabolic model iRR1083, and we evaluated whether this chimeric model was consistent with growth phenotypes for S. typhimurium. The expectation was that if the transcriptional regulatory networks were highly conserved few regulatory rule violations would be needed to correctly predict growth.
We transferred the regulatory rules in iMC105A for E. coli genes to their orthologs in S. typhimurium, and used the GPR association in the S. typhimurium metabolic network to constrain fluxes through the metabolic reactions. Among the 1,083 S. typhimurium genes included in iRR1083, 782 genes had orthologs in E. coli that were included in iJR904 (which included a total of 904 metabolic genes). Additionally, among the 105 E. coli transcription factors in iMC105A, we found 86 had orthologs in S. typhimurium which were incorporated into the chimeric model. These differences in conservation of metabolic and regulatory genes allowed us to transfer approximately 83% of the regulatory rules in iMC105A, while discarding the remaining rules associated with metabolic genes not present in S. typhimurium or regulatory rules involving transcription factors present only in E. coli. Any metabolic ortholog present in S. typhimurium but regulated by transcription factors without orthologs in S. typhimurium were kept unregulated in the chimeric model.
We applied the GeneForce algorithm to this hybrid E. coli regulatory-S. typhimurium metabolic model and evaluated model predictions against wild-type S. typhimurium growth phenotypes in 196 medium conditions (Supporting Information Table S5), which resulted in a surprisingly small number of regulatory rule violations, suggesting a highly conserved transcriptional regulatory network between E. coli and S. typhimurium, at least for conserved orthologs. As seen in Table 4, only a total of 18 genes (out of 505 genes with regulatory rules) needed regulatory rule corrections, some of which (argD, rhaR and rhaS) also needed rule corrections in subsequent refinements of the E. coli integrated model as well (Table 2, correction list B). Thus, 15 out of the 18 genes suggested some regulatory differences between the two species (Table 4). For example, the prp operon was forced to be active by the algorithm because S. typhimurium is capable of utilizing 1, 2-propanediol aerobically while E. coli is not [41]. The prpBCDE operon of S. typhimurium encodes enzymes that are needed for utilization of 1,2 propanediol [42]. Regulatory rules in iMC105A state that the E. coli prp operon is induced by propionate, while for S. typhimurium, there is evidence that this operon is induced by the coordinated function of regulatory proteins PrpR, IHF, and RpoN, where activation of PrpR is induced by 2-methylcitrate, a reaction intermediate in the 1,2-propanediol utilization pathway [43], [44]. Therefore, the algorithm correctly identified the prp genes as having incorrect rules for S. typhimurium but not for E. coli.
The glnA gene (encoding glutamine synthetase) was also identified as requiring a rule correction in S. typhimurium but not in E. coli for growth on glucose and D-gluconate medium. This difference is primarily attributed to differences in the GPR association for glutamine synthetase between the two metabolic networks, where glnA encodes a sole enzyme for glutamine synthesis in S. typhimurium, whereas in the E. coli models an additional isozyme YcjK can catalyze the same reaction when glnA expression is suppressed. Recently, however the YcjK has been shown to be incapable of synthesizing glutamine from glutamate and ammonia so the regulatory rule for glnA and GPR association for glutamine synthetase needs to be updated in the E. coli model as well [45]. Similarly, the focA gene encodes the sole formate transporter in S. typhimurium but has an alternative gene focB in E. coli, which explains why the focA regulatory rule was not problematic in our analysis of E. coli growth phenotypes.
The initial integrated metabolic and regulatory model for S. typhimurium, when tested against the wild-type S. typhimurium growth phenotypic data in 196 medium conditions was 77% accurate. The unregulated metabolic model (unregulated iRR1083) was 82% accurate against the same growth phenotyping data. After introducing the refined rules (Table 4), the refined integrated metabolic and regulatory model was able to achieve 83% accuracy for this dataset, a value similar to those found in this study for integrated models of E. coli metabolism and regulation.
In this work we developed a new optimization-based approach, GeneForce, for systematically refining a genome-scale transcriptional regulatory model by comparing model predictions against high-throughput growth phenotypic data. The developed approach was used to (i) refine existing transcriptional regulatory and metabolic models of E. coli and suggest regulatory rule corrections, (ii) explain how transcriptional regulation prevents cellular growth in certain conditions and identify genes which can rescue non-growth phenotypes if expressed, and (iii) construct and refine a new integrated regulatory and metabolic model for S. typhimurium.
We showed that even well curated transcriptional regulatory and metabolic models for E. coli [3], [21], [22] can be further improved by using the developed approach. Here, cases where the integrated model under-predicted growth (cells grew experimentally and the metabolic model predicted growth, but the integrated model did not predict growth) were used to improve the integrated metabolic and regulatory model. A total of 42 model corrections (27 listed in Table 1 and an additional 15 described in the Materials and Methods section) were identified and when implemented they improved the accuracy of the models by 1–8%. The improved integrated metabolic and regulatory model predictions were found to better predict metabolic mutant phenotypes than other constraint-based methods using only metabolic models. When the iAF1260 metabolic model was used, flux balance analysis (FBA) was ∼76.5% accurate and minimization of metabolic adjustment (MOMA) [46] was ∼75.6% accurate (data not shown), while the integrated metabolic and regulatory model (iAF1260+iMC105ABC) was ∼79.6% accurate when predictions were made for the metabolic gene knockouts. The integration of metabolic and regulatory network models is thus important for being able to more accurately predict behavior of metabolic mutants, as well as, transcription factor mutants.
In addition to fixing incorrect model predictions, we showed that GeneForce can also be used to evaluate correct model predictions of non-growth conditions to explain how regulation prevents the use of particular nutrients since the needed enzymes are encoded in the genome. We used the approach to suggest a set of genes which if expressed can rescue non-growth phenotypes of mutant and wildtype strains. Experimental testing of these hypotheses would validate that particular metabolic transformations occur and could be used to engineer novel growth phenotypes in an organism.
In addition to applying the GeneForce approach to already developed and refined metabolic and regulatory E. coli models, we also applied it to a new integrated model for S. typhimurium. We constructed an initial transcriptional regulatory model for S. typhimurium, by transferring the regulatory network from a closely related organism. We then applied the approach to correct regulatory rules in the integrated metabolic and regulatory model of S. typhimurium [20]. The results showed that the transcriptional regulatory network in E. coli is highly consistent with the growth phenotypes of S. typhimurium, indicating that the regulatory networks in these two organisms may be highly conserved between these two organisms. A similar observation was found previously by Babu et al., where ∼90% of E. coli regulatory interactions were predicted to be conserved in S. typhimurium based on the presence of orthologs of transcription factors and their target genes in S. typhimurium [47]. While the resulting integrated model is still consistent with observed phenotypes for S. typhimurium, additional work is still needed to expand this initial regulatory model to include organism-specific regulatory interactions (such as altered regulons and regulons that are unique for S. typhimurium).
The number of available metabolic reconstructions is rapidly increasing, which is made possible by the increased number of genome sequences [48]. However, the development of genome-scale transcriptional regulatory models is currently limited by the lack of available data for most organisms. Different approaches have been developed to model transcriptional regulatory networks (reviewed in [10]), but a Boolean approach has been commonly used for building genome-scale transcriptional regulatory models due to its scalability. Integrated models of metabolism and transcriptional regulation have been developed using a Boolean approach for some model organisms, such as E. coli [3] and S. cerevisiae [4], but our current understanding of transcriptional regulation in microorganisms is still limited due to its complexity, interconnectivity, and intrinsic noise in these networks compared to metabolism. The GeneForce approach can be very useful for validating and refining transcriptional regulatory models against new experimental data, as well as for developing new regulatory models where initial models often yield a number of predictions that are inconsistent with experimental data. In the past, the identification of regulatory rules causing inconsistencies between model predictions and experimental observations was done through a time intensive, trial and error process [3].
Other types of non-Boolean methods are needed to integrate genome-scale metabolic and regulatory models, since Boolean approaches cannot capture all transcriptional regulatory interactions (e.g. regulation of essential genes) and gene expression and metabolic fluxes have variable levels that cannot be reflected using ‘On/Off’ variables. Modeling methods are available to predict gene expression levels [49], and these predictions could be used to constrain metabolic fluxes [50]–[52] at a genome-scale. The GeneForce algorithm could be easily extended to consider other types of integrated metabolic and regulatory models as they are developed, where the number of genes needing expression levels higher than those predicted by the regulatory models could be minimized. As such, the approach would still reconcile integrated metabolic and regulatory network models with observed growth phenotypes and suggest improvements of such models. Other approaches have been developed for metabolic models that use experimentally determined flux distributions as a means to refine metabolic models [53], and the GeneForce algorithm could be extended to compare more quantitative data including biomass yields (where the measured yields are used to determine the minimum growth rate threshold) and measured fluxes (where model fluxes are constrained to be a certain distance from the experimental values) as such quantitative data become available at a large-scale.
Although automated approaches for refining metabolic models have been developed [7], [9], [54], [55], such an approach has not been created for integrated models of metabolism and transcriptional regulation. The approach developed here finds a minimum set of refinements needed to correct one case at a time. While we did not find it to be a significant problem here, it is possible that making model refinements to correct one case may cause a significant number of new incorrect predictions for other cases. Approaches that consider multiple cases simultaneously could be advantageous, but they were not considered here because of the added computational burden for considering all conditions simultaneously. The approach described here can be used to improve transcriptional regulatory network models by accounting for how a hypothesized regulatory network will affect metabolism and thereby cellular behavior. We envision that predictions of cis-regulatory networks, based on genomic analysis and/or experimental data, can be translated into Boolean regulatory models that can be rapidly refined using our developed approach. The identified refinements can then suggest further experiments and lead to a re-evaluation of cis-regulatory networks. By integrating models of metabolism and regulation, phenotypic data can be evaluated against regulatory network predictions (which is difficult to do without a metabolic model), thereby expanding the types of datasets (e.g. gene expression, genome sequence, and DNA-protein interaction) that can be used to reconstruct transcriptional regulatory networks.
The Keio collection of in-frame single-gene deletion strains [25] and E. coli K-12 BW25113 (the parent strain of the Keio collection) were used to confirm the model changes identified by GeneForce. The kanamycin resistant gene (kan) was removed from the single-deletion strains before screening mutant phenotypes in the microplate reader (for methods see [56]). In addition, seven double mutants (lrp::kan ΔilvB, lrp::kan ΔilvN, lrp::kan ΔilvH, lrp::kan ΔilvI, lrp::kan ΔdctA, rpiA::kan ΔrpiB, and cycA::kan ΔdsdX) were generated using P1 transduction (for methods see [57]).
Phenotype microarray (PM) experiments were conducted for the arcA::kan, purR::kan, and lrp::kan strains from the Keio collection using PM1 and PM2 plates following manufacturer protocols (Biolog Inc., CA). Briefly, strains were grown on BUG+B agar plates and resuspended in inoculating fluid containing Dye A and loaded onto plates. Plates were incubated at 30°C and absorbance readings were taken at 600nm at 24 and 48 hours. Other strains were screened for growth in triplicate at 37°C in a Tecan Infinite 200 microplate reader (Tecan Group Ltd., Switzerland). Optical density measurements at 600 nm were taken by the microplate reader every 15 minutes. The Tecan OD measurements (ODTecan)were converted to an OD value in a spectrophotometer with a 1cm pathlength (OD600), using a predetermined linear relationship, OD600 = (2.566)ODTecan+0.0028. Strains were pre-cultured overnight in 2 g/liter glucose-supplemented M9 minimal medium, except for a few strains (listed in Figure 4A) that were evaluated for their ability to grow on glucose which were instead pre-cultured in LB medium. Pre-cultured cells were washed and resuspended in media containing a new carbon source so that the starting OD (at 600 nm) was around 0.05. All carbon sources were tested in M9 minimal medium (6.8 g of Na2HPO4, 3 g of KH2PO4, 0.5 g of NaCl, 1 g of NH4Cl, 2 ml of 1M MgSO4, and 100 µl of 1 M CaCl2 per liter) supplemented with 2 g/liter of carbon source.
High-throughput growth phenotyping (Biolog Inc., CA) data for E. coli from the ASAP database [18] were analyzed to assign “growth (+)” or “no growth (−)” for mutants grown in different conditions. In addition to the dataset (Mutant Biolog Data I) evaluated by Covert and colleagues [3], an additional dataset (Mutant Biolog Data II) was analyzed in this study. We considered the phenotype microarray (PM) data for carbon (PM1 and PM2) and nitrogen sources (PM3) that can be simulated by the computational models, which consisted of 223 mutants in 130 conditions or 303 mutants in 153 conditions, depending on which model was used (see below). For each PM plate, the negative control value was subtracted from each data point (OD600), and a cutoff parameter of 0.1 was applied to determine whether the cells could grow (+) or not grow (−). The cutoff parameter was obtained by separating a bimodal-like distribution of the data (Supporting Information Figure S1), and the results were not highly sensitive to this parameter. Another set of high-throughput phenotyping data for single gene knockout mutants of E. coli [19] was also used. This dataset includes growth phenotypes for 1,440 mutants in 95 environmental conditions using a GN2-MicroPlate (Biolog Inc., CA). However, the conditions that can be simulated by the models consist of only 102 mutants in 30 conditions or 128 mutants in 31 conditions depending on the model used, since the majority of evaluated mutants involved knockouts of genes with unknown function. The phenotypic data for three global transcription factor knockout mutants (ΔarcA, ΔpurR, and Δlrp) was generated in this study using the phenotype microarrays (Biolog Inc., CA) as described above. In this study, we have excluded the phenotypic data for cells grown on formate and L-serine as carbon sources, and xanthine and xanthosine as nitrogen sources, as they are likely false positives in the PM datasets (formate [3]; xanthine and xanthosine [7] and L-serine (tested in this study, data not shown) ).
The genome-scale models of metabolism (iJR904 [21], iAF1260 [22]) and regulation (iMC104v1 [3]) for E. coli were integrated and used in this study. First, regulatory interactions for the global transcription factor, Lrp, were updated in the regulatory rules represented in the iMC104 model based on the recent regulatory reconstruction from analysis of gene expression and ChIP-chip data [17]. The Lrp reconstruction categorized regulatory interactions into six different modes based on the gene expression responses of genes controlled by Lrp to exogenous leucine. We have converted each regulatory mode into Boolean logic rules, and updated the regulatory rules in conjunction with existing rules. Preliminary computational analysis was performed to identify essential genes for growth in glucose minimal media that were predicted to be un-expressed based on the updated Lrp rules; the regulatory rules for these seven essential genes were then changed back to the original ones before mutant phenotypes were evaluated. In addition, when the metabolic part of the integrated model was replaced with the recent metabolic reconstruction, iAF1260 instead of iJR904, another set of preliminary rule corrections were needed for the eight genes that are essential only in iAF1260, due primarily to changes in the biomass equation. These fifteen preliminary rule corrections were made before the integrated model was compared to mutant phenotypes, and thus they are not listed in Table 1 (see Supporting Information Table S6 for details).
Simulation conditions for the models were determined based on the available carbon or nitrogen sources in the media as previously described elsewhere [3] (see Supporting Information Table S7). When testing the growth on different carbon sources, ammonia was used as a nitrogen source and the maximum uptake rate for ammonia was constrained to be 10 mmol/gDW/hr. Pyruvate was used as a carbon source for testing growth on different nitrogen sources, and its uptake rate was constrained to be 11.3 mmol/gDW/hr. Oxygen uptake rate was constrained to be 10 mmol/gDW/hr for all cases, and uptake rates for other essential nutrients in each model were specified as listed in Supporting Information Table S7.
Flux balance analysis (FBA) [58] was performed to predict the maximum growth rate for mutants under different conditions using the metabolic models. In order to simulate gene deletions in the metabolic models, we have included GPR associations where reactions are constrained to have zero flux if an associated gene is deleted. For the integrated metabolic and regulatory models, we have systematically formulated an SR-FBA problem [15] with gene knockout and transcriptional regulatory constraints [16]. Predictions were made by maximizing growth rate for each mutant in each condition. If the maximum growth rate was positive then the model predicted growth is designated as (+), or otherwise designated as (−).
GeneForce identifies the minimal set of genes that are required for growth, but are unexpressed in a given condition due to transcriptional regulatory constraints. In the GeneForce formulation, unexpressed genes are allowed to violate the regulatory rules, and the number of violations is minimized to prevent unnecessary rule violations. A rule violation is implemented by introducing surrogate gene expression indicator variables (y′g) to allow flux through reactions whose associated genes are not expressed according to the Boolean regulatory rules. A minimum growth rate requirement is introduced by setting the lower bound for growth rate to a minimum threshold value, and the threshold value was set to 10% of the maximum growth rate predicted by the metabolic model in this study. The algorithm was relatively insensitive to threshold values between 5 and 50% (see Supporting Information Table S8), because most integrated model growth rate predictions were above 80% or below 5% of the metabolic model predicted growth rate (see Supporting Information Figure S2). Alternative optimal solutions were found by adding integer-cut constraints and re-solving the problem. See Supporting Information Text S1 for more details.
To identify possible regulatory rule corrections, we analyzed the cases where the metabolic model and experimental data agree that the mutant can grow, but the integrated model predicts no growth (+/+/−; corresponding to experimental data / metabolic model / integrated model). If a certain set of regulatory rules were repeatedly violated in the GeneForce solutions to allow for growth of a specific mutant or in a particular medium condition, the regulatory rules for those genes were corrected based on experimental evidence from the literature. When alternative optimal solutions were available, meaning that different sets of rule violations could correct the non-growth phenotype predictions, we examined each set of solutions and chose the most appropriate one for the specific case based on results from additional experiments and/or information in the literature. If a set of rule corrections caused inconsistencies in other mutant or medium condition, such corrections were not made unless there was strong experimental evidence for the rule correction.
|
10.1371/journal.pcbi.1003840 | Mathematical Modeling of Bacterial Kinetics to Predict the Impact of Antibiotic Colonic Exposure and Treatment Duration on the Amount of Resistant Enterobacteria Excreted | Fecal excretion of antibiotics and resistant bacteria in the environment are major public health threats associated with extensive farming and modern medical care. Innovative strategies that can reduce the intestinal antibiotic concentrations during treatments are in development. However, the effect of lower exposure on the amount of resistant enterobacteria excreted has not been quantified, making it difficult to anticipate the impact of these strategies. Here, we introduce a bacterial kinetic model to capture the complex relationships between drug exposure, loss of susceptible enterobacteria and growth of resistant strains in the feces of piglets receiving placebo, 1.5 or 15 mg/kg/day ciprofloxacin, a fluoroquinolone, for 5 days. The model could well describe the kinetics of drug susceptible and resistant enterobacteria observed during treatment, and up to 22 days after treatment cessation. Next, the model was used to predict the expected amount of resistant enterobacteria excreted over an average piglet's lifetime (150 days) when varying drug exposure and treatment duration. For the clinically relevant dose of 15 mg/kg/day for 5 days, the total amount of resistant enterobacteria excreted was predicted to be reduced by 75% and 98% when reducing treatment duration to 3 and 1 day treatment, respectively. Alternatively, for a fixed 5-days treatment, the level of resistance excreted could be reduced by 18%, 33%, 57.5% and 97% if 3, 5, 10 and 30 times lower levels of colonic drug concentrations were achieved, respectively. This characterization on in vivo data of the dynamics of resistance to antibiotics in the colonic flora could provide new insights into the mechanism of dissemination of resistance and can be used to design strategies aiming to reduce it.
| Fecal excretion of antibiotics and resistant bacteria in the environment are major public health threats associated with extensive farming. Innovative strategies that reduce the intestinal antibiotic concentrations during treatment are in development and could help prevent the dissemination of resistance. In order to anticipate the impact of these strategies, the effect of lower exposure on the amount of resistant enterobacteria excreted needs to be quantified precisely. Here, we introduce a bacterial kinetic model to capture the complex relationships between dosage regimen, antibiotic fecal concentrations, loss of susceptible enterobacteria and growth of resistant strains in the feces of piglets receiving different doses of ciprofloxacin for 5 days. We use this model to evaluate by simulation how much it would be necessary to reduce the antibiotic colonic concentration in order to prevent the expansion of antibiotic resistance. This approach provides new insights into the mechanism of dissemination of resistance during treatments and can be used to design strategies to reduce it.
| Antibiotics are widely used in animal farming for curative, prophylaxis and metaphylaxis purposes. This results in massive excretion of antibiotics [1], [2] and resistant bacteria with the feces of the animals during treatments [3]. It impacts the ecology of the environment and ultimately contributes to increase resistance in bacteria infecting humans [4], making resistance of human bacteria to antibiotics one of the major threats to public health in the next decade [5], [6].
In particular, fluroquinolones (FQ) are widely used in animals, including in pets and farm animals for respiratory, urinary tract and skin infections, and have also been categorized as critical for human use (see the WHO list of Critically Important Antimicrobials [7]). Unfortunately resistance to FQ has regularly increased over the last decades and has reached a level that jeopardizes the treatment of common human infections caused by members of the Enterobacteriaceae family (enterobacteria), in particular Escherichia coli and Klebsiella spp, such as gastrointestinal and urinary tract infections [8], [9]. Besides causing infections, enterobacteria are also naturally present in the intestinal commensal flora of humans and several animal species [10], [11]. When a subject is treated with FQ, either by the oral or the parenteral route, a fraction of the dose administered is eliminated in the intestine after biliary and intestinal excretion [12]. These residual concentrations may be sufficient to eliminate FQ-susceptible species but not to act against resistant enterobacteria [13], [14]. Consequently, resistant enterobacteria can multiply in these free niches and reach high concentrations before being excreted in the feces [15]. This set of events is believed to be a major driver of emergence and dissemination of bacterial resistance [16] and this is why innovative strategies, such as charcoal-based adsorbent, are now being developed to reduce intestinal antibiotic residues [17], [18]. However, the effect of lower antibiotic exposure on the amount of resistant enterobacteria excreted has not been characterized, making it difficult to anticipate the impact of these strategies.
We previously reported that intestinal excretion of ciprofloxacin (a FQ) resistant enterobacteria increased with the colonic exposure to ciprofloxacin in piglets [15]. Here, using a mechanistic model, we now aim to characterize the complex relationships between antibiotic dosage regimen, pharmacokinetics in feces, loss of susceptible enterobacteria and growth of resistant strains. This approach has mostly been used to fit in vitro data during antibiotic treatment [19]–[24]. To the best of our knowledge, only very few papers aimed to fit in vivo bacterial kinetic data (see [25] for instance) and it has never been used to characterize the dynamic of enterobacteria in the intestinal flora during treatment. This lack of data may be due to the difficulty to obtain and analyze such data, characterized by a high level of variability both in drug concentrations and in bacterial counts [15].
This difficulty can be in part circumvented by using nonlinear mixed-effect models (NLMEM), a statistical approach that optimally uses all the information available in a population sample, including the between subject variability, in order to increase the ability to estimate model parameters [26]. Here, we used this technique to fit a dynamic mathematical model to the kinetics of drug concentrations and the counts of total and resistant fecal enterobacteria. After the model parameters have been estimated, a large number of scenarios can be evaluated in silico and this model was then used to predict the effect of reduced intestinal antibiotic concentrations on the amounts of FQ resistant enterobacteria excreted.
The protocol was approved by Pharnimal (Eghezée, Belgium) and Animal Breeding Parteners facilities - Faculty of Veterinary Medicine (Liège, Belgium). Animal housing and care comply with the guidelines of the local ethical committee, in accordance with EU Guideline on Good Clinical Practice for the conduct of Clinical trials and Veterinary Medicinal Products (Eudralex. Volume 7A: 7AE1a) and VICH guideline on Good Clinical Practice (VICH GL9).
The data we used are from a prospective randomized study previously published [15]. Briefly, 29 piglets from a single farm were included 4 weeks after birth and were put in individual cages throughout the study and randomly assigned (9∶10∶10) to once-a-day oral treatment with placebo, ciprofloxacin 1.5 or 15 mg/kg/day for 5 days (D1 to D5).
Ciprofloxacin concentrations and counts of total and ciprofloxacin resistant enterobacteria were measured in fecal samples at pre-dose on D1, D3 and D5 and on D7, D9, D12, D16 and D27. A microbiological assay that measures the ability of the antibiotic to inhibit the growth of Bacillus subtilis strain ATCC6633 was used to determine fecal concentrations of free (active) ciprofloxacin. Total and fluoroquinolones resistant enterobacteria were counted by plating serial dilutions of the feces on Drigalski agar without or with 2 mg/L of ciprofloxacin, respectively [15]. The 2 mg/L concentration was chosen in agreement with the EUCAST clinical breakpoints (www.eucast.org). The limit of detection was 0.1 µg/g and 102 CFU/g of feces for antibiotic concentrations and bacterial counts, respectively. The data will be available upon request.
Because the sensitivity of resistant enterobacteria to treatment could not be precisely identified (not shown), we estimated only C50S and we fixed the ratio C50R/C50S to various putative values equal to 4, 16, 100 and ∞ (i.e., ciprofloxacin had no activity in resistant enterobacteria and δmax*C/(C+C50R) was fixed to 0).
For each case, ciprofloxacin concentrations and counts of resistant and total enterobacteria observed in the experimental animals were simultaneously fitted using nonlinear mixed effect models (NLMEM), an approach which borrows strength from the whole sample to precisely estimate the population parameters, such as the mean and the between-subject variability (BSV) [34]. In this approach, each individual parameter θi of piglet i is modeled as a fixed part θ, which represents the median value of the parameter in the population, and a random part ηi chosen from a Gaussian distribution with mean 0 and standard deviation ω that accounts for the BSV. Therefore, all parameters can be written as where . Parameter variability was fixed to 0 in case of low estimated value and high relative standard error (RSE). We assumed combined error model for fecal ciprofloxacin concentrations (parameterized in and for additive and proportional error, respectively), and constant error model for log10 of resistant and total counts of enterobacteria, noted and , respectively.
Data were analyzed using MONOLIX 4.2.0 (www.lixoft.eu), a software devoted to maximum likelihood estimation of parameters in NLMEM, based on the SAEM algorithm [35]. The code for the model implemented in MONOLIX can be found in the supporting information. Details on the fitting method and the likelihood expression in kinetic models defined by ordinary differential equations can be found in [36]. Of note, maximum likelihood estimation can take into account the information brought by data under the level of detection [37]. After the population parameters were determined, the values of the parameters for individual piglets were deduced using empirical Bayes estimates, and predicted medians at each timepoint could be obtained for the resistant and total enterobacteria as well as for the susceptible ones. Model evaluation was done by analyzing the distribution of the residuals and by comparing the observed and predicted median values. The data fitting obtained for the different putative values of the ratio C50R/C50S were compared using the Bayesian Information Criterion (BIC, the lower the better), a fitting criterion that accounts for the number of parameters.
Using the estimated distributions of the parameters with the different putative values of the ratio C50R/C50S, we performed a Monte Carlo simulation to predict the amounts of resistant enterobacteria that would be excreted with similar or lower fecal concentrations of ciprofloxacin than those observed in the experimental piglets and with various treatment duration of 1, 3, 5 and 10 days. For each scenario, parameters for 1000 piglets were generated. Because the colonic drug concentration rapidly reaches a dose-proportional plateau level, Css, the effects of a x-fold lower drug exposure were obtained by simulating the model with a x-fold lower administered dose. For instance, the effect of a strategy that could reduce colonic drug concentrations by 99% as compared to the therapeutic dose of 15 mg/kg/day were obtained by simulating the model with a dose of 0.15 mg/kg/day.
Of note, in order to facilitate comparison between these scenarios, we assumed constant value for the maximal bacteria density Nmax (i.e., a = b = 0 in equation 3) in the simulation study. We calculated at each timepoint the median value for the antibiotic residual concentrations, Cmed, as well as the median counts of susceptible, resistant and total enterobacteria, noted Smed, Rmed and Tmed, respectively.
Further, we assumed an average remaining lifespan of 150 days (5 months) after the initiation of the treatment [38] and an excretion of about 100 g of feces per day [39]. Therefore, for a given treatment duration and a given drug exposure, the median total amount of resistant bacteria excreted from the beginning of the treatment (t = 0) until death, AR, is given by . Let AR0 and AR15 be the corresponding amounts excreted if there was no drug exposure and the reference exposure (i.e., corresponding to the therapeutic dose of 15 mg/kg/day), respectively. Then the normalized reduction of the amounts of resistant enterobacteria excreted for a given dose and treatment duration is given by .
The PK model described well the rapid increase in fecal concentrations of ciprofloxacin followed by a plateau (Figure 2, Figure S1 and Figure S2). The mean elimination rate constant of intestinal ciprofloxacin concentrations, ke, was equal to 1.97 day−1, corresponding to a half-life of about 8 hours (Table 1). The plateau exposure, Css, was estimated at 8.7 µg/g in the animals treated with the dose 1.5 mg/kg/day and at 87 µg/g in those treated with 15 mg/kg/day.
We found that the best fit was obtained when assuming that C50R/C50S was equal to ∞, i.e., ciprofloxacin had no activity in resistant enterobacteria (Table 2). Consequently, we neglected the effect of ciprofloxacin on resistant bacteria in the final model. Interestingly almost all parameters could be estimated with a reasonable precision: lower than 30% for fixed parameters and 50% for variability terms (Table 1). gs and ωδmax were out of these ranges (RSE of 82% and 84%, respectively) and therefore their estimated value should be interpreted with caution.
The model could well characterize the kinetics of total and resistant enterobacterial counts observed in experimental animals during and after treatment (Figure 2, Figure S1 and Figure S2). C50S was equal to 4.91 µg/g and lower than Css of both dosing groups. Therefore susceptible enterobacteria decreased rapidly after initiation of treatment, with a maximal elimination rate, δmax, estimated to 27.1 day−1, corresponding to a half-life (ln(2)/δmax) of 37 minutes. In absence of treatment, both resistant and susceptible bacteria were eliminated at a much lower rate kT equal to 0.1 day−1. Replication rates of resistant and susceptible enterobacteria were estimated to 1.9 day−1 and 13.7 day−1, respectively, giving a relative fitness of the resistant enterobacteria [40] of αR/αS = 1.9/13.7 = 14% in absence of treatment.
First, we simulated the evolution of enterobacteria counts that would be obtained with similar or lower fecal concentrations of ciprofloxacin than those observed in the experimental piglets, i.e., Css = 87, 58, 29, 18, 8.7, 2.9, 1.8, 0.9 or 0 µg/g and with treatment duration of 1, 3, 5 or 10 days (see methods). The simulated curves Cmed, Rmed, Smed and Tmed are presented in Figure 3. The amount of resistant enterobacteria excreted increased with fecal concentrations of antibiotic or treatment duration. In all scenarios, including those with low exposures, susceptible strains were rapidly eliminated after treatment initiation and replaced by resistant enterobacteria within 5–10 days (Figure 3). After treatment end, resistant enterobacteria disappeared slowly and it took 2–15 days for susceptible enterobacteria to return to pre-treatment levels, consistent with observations from both dosing groups (Figure 2). As the elimination rate in absence of treatment, kT, was low, resistant bacteria could remain in high counts for several weeks after the end of treatment (see Figure S3).
We then used these results to calculate the total amount of resistant enterobacteria excreted over the remaining lifespan of a piglet (150 days) according to the treatment duration (Table 3). In animals exposed to the reference drug colonic concentration Css of 87 µg/g, i.e., the one obtained with the therapeutic dose of 15 mg/kg, the predicted amounts of resistant enterobacteria excreted were equal to 7.5, 8.6 and 9.2 log10CFU for 1, 3, and 5 days of treatment, respectively. In other words, the level of resistance excreted can be reduced by 75% and 98% when reducing treatment durations from 5 days to 3 or 1 day, respectively.
Next, we estimated the reduction in colonic drug exposure that needs to be achieved in order to excrete 50% less resistance than with the reference drug colonic concentration of 87 µg/g. Figure 4 shows that this would require reducing drug exposure by 67%, 80% and 90% for a 1, 3 and 5 day treatment. If we focus on a 5 day treatment, reducing drug exposure by 80% would reduce excretion by only 33% (Table 3).
Finally, we evaluated the sensitivity of these results to the assumption that ciprofloxacin had no effect on resistant bacteria and similar simulations were conduced assuming that the ratio C50R/C50S was equal to 4, 16, 100 (see methods). The results did not change substantially when varying the ratio C50R/C50S (Table S1). For instance, assuming no effect of ciprofloxacin on resistant bacteria, and a dosing regimen of 15 mg/kg/day for 5 days, we found in the main analysis that drug concentrations would need to be reduced by 90% (i.e., divided by 10) in order to reduce by 50% the amount of excreted resistance. Assuming C50R/C50S equal to 4, 16 and 100, the drug concentrations would need to be reduced by ∼75%, 80% and ∼85%, respectively, i.e., the same order of magnitude than in the main analysis.
To our knowledge, the bacterial kinetic model we proposed here is the first one developed on in vivo within-host data to characterize the relationship between antibiotic concentrations and resistance to fluoroquinolones in feces. This approach brings new insights on fundamental and clinical aspects of drug resistance.
First, we showed that the proposed mathematical model of resistant and susceptible enterobacteria could well describe the kinetics of both populations during and after treatment. Initiation of treatment led to the rapid elimination of susceptible enterobacteria, with a decline of about 3 log10CFU/g in the first 24 h of treatment observed in fecal counts, consistent with the kinetics of Salmonella in pigs treated with enrofloxacin [41]. Using a model to fit these data allowed us to estimate the half-life of sensitive enterobacteria during treatment to about 37 minutes. Interestingly, an even more rapid half-life of about 5 minutes was found in Escherichia coli exposed to ciprofloxacin [19] but this estimate was found in vitro where the metabolism of the bacteria may be different.
The rapid elimination of susceptible enterobacteria created a vast replication space that enabled the proliferation of resistant enterobacteria that remained at high levels in feces for more than 3 weeks after treatment end in both dosing groups, consistent with previous reports [14]. Though an antibiotic effect on resistant enterobacteria cannot be ruled out, our results suggest that treatment has only minimal activity on resistant enterobacteria (i.e., C50R≫C50S). Because this effect could not be precisely estimated we tested different putative values for the ratio C50R/C50S and we found that the best description of the data was obtained when assuming no effect of ciprofloxacin on resistant enterobacteria (i.e., C50R/C50S = ∞). Interestingly resistant bacteria were pre-existing to treatment in all piglets. The presence of resistant bacteria can result from de novo spontaneous mutations or from continuous ingestion of ciprofloxacin-resistant bacteria. Although the presence of resistance due to de novo mutations in the GT was unlikely (see Methods), both hypotheses were tested, and a better fit was obtained when a continuous ingestion of susceptible and resistant bacteria was assumed rather than mutations (see Supporting Information).
Next, we performed Monte-Carlo simulations in order to estimate the relationships between treatment duration, antibiotic colonic exposure and excretion of resistant enterobacteria. Interestingly we showed that even a one-day exposure to antibiotics, a practice recommended to treat some types of infections [42] or in antibioprophylaxis [43], could induce a 20-fold increase in resistance excretion compared to an untreated piglet. Not surprisingly, for a given level of drug exposure, longer treatment led to higher median total amounts of resistant enterobacteria excreted over an average piglet's lifetime, noted AR. However, this relationship was highly nonlinear. Indeed, for a colonic exposure such as that resulting from a treatment of 15 mg/kg/day (i.e., within the range of a therapeutic dose), AR was equal to 7.5, 8.6 and 9.2 log10CFU for 1, 3 and 5 days of exposure, respectively, as compared to 6.2 log10CFU in an untreated piglet. The fact that even a low dose of antibiotic can lead to high and sustained levels of resistance for long period of time had already been observed in vitro [44] and confirmed recently in the human feces in vivo of healthy volunteers receiving ciprofloxacin [14]. Thus this approach is highly relevant to predict the effect of reduced length of treatments on antimicrobial resistance and confirms that unnecessary use of antibiotics, even for short period of time, may lead to massive resistance excretion [45], [46]. Therefore by all means the best way to reduce antibiotic selective pressure on intestinal bacteria is to avoid all unnecessary use of antibiotics.
Beside limiting the use of antibiotics, one can also play on the intestinal antibiotic concentration to reduce resistance excretion. In order to achieve about 50% reduction in the quantities of resistant bacteria excreted in the environment, we found that colonic concentrations had to be reduced by 67%, 80% and 90% for 1, 3 and 5 days of treatment, respectively. Importantly, these results did not change substantially when we assumed that ciprofloxacin had also an effect on resistant enterobacteria or when we assumed that resistant enterobacteria were due to mutations (see Supporting Information).
Recently, it has been shown that a charcoal-based specifically chosen adsorbent, formulated to target late ileum and colon to avoid upper tract adsorption of orally administered drugs, could decrease fecal levofloxacin (a FQ) concentrations administered by infusion in dogs, to the extent of 85% [47]. Therefore our prediction that concentrations have to be reduced from 67% to 90% does seem a realistic objective and our results could be highly useful in order to anticipate the impact of these new strategies [17], [18], [48].
The two major limitations of this study were the lack of frequent measurements and the uncertainty associated with these measurements. The use of a sophisticated statistical approach based on non-linear mixed effect models made it possible to precisely estimate most of parameters in spite of a high residual error. More detailed data will be needed to estimate some important parameters such as the proportion of ingested resistant bacteria gr/gs. Likewise the lack of frequent fecal drug concentrations did not allow for a more physiological PK model. However, the model used here was sufficient to describe the most important feature of ciprofloxacin fecal pharmacokinetics, i.e., the increase in ciprofloxacin concentrations (in the first two 2 days) followed by a plateau. Importantly a similar feature was also reported with enrofloxacin (another fluoroquinolone) in a study where PK was collected twice daily in pigs treated once daily during 5 days [49]. Another important limitation of this study is that our simulation extrapolates to a context of treatment longer than 5 days for which no data was available. Overall future experiments with richer data on larger populations, with different FQ compounds and various dosage regimens will be useful to overcome these limitations and to address other interesting aspects. In particular these studies should include isolation of piglets, control of ingested food and frequent cleaning of the life-place in order to better characterize the origin of resistant enterobacteria and to confirm our result that resistant bacteria in piglets are coming from the environment rather than via direct mutation in the gastrointestinal tract. Lastly, it is yet unknown to what extent these results can be extrapolated to other animal species or to humans. Here, the rapid growth of resistant strains was largely due to the fact that almost all piglets (25/29) had detectable levels of preexisting resistance. This is probably not the case in humans [14] and therefore understanding the growth of resistance in humans will require expanding the model to account for stochastic events such as the spontaneous apparition of resistant enterobacteria [21].
In summary, we proposed here the first approach to model in vivo the kinetics of resistant enterobacteria in feces during antibiotic treatment. This approach could be particularly relevant to design and evaluate novel strategies that aim to reduce intestinal exposure to antibiotic residues in order to reduce resistance excretion and dissemination in the environment.
|
10.1371/journal.pcbi.1000319 | Quantifying Global Tolerance of Biochemical Systems: Design Implications for Moiety-Transfer Cycles | Robustness of organisms is widely observed although difficult to precisely characterize. Performance can remain nearly constant within some neighborhood of the normal operating regime, leading to homeostasis, but then abruptly break down with pathological consequences beyond this neighborhood. Currently, there is no generic approach to identifying boundaries where local performance deteriorates abruptly, and this has hampered understanding of the molecular basis of biological robustness. Here we introduce a generic approach for characterizing boundaries between operational regimes based on the piecewise power-law representation of the system's components. This conceptual framework allows us to define “global tolerance” as the ratio between the normal value of a parameter and the value at such a boundary. We illustrate the utility of this concept for a class of moiety-transfer cycles, which is a widespread module in biology. Our results show a region of “best” local performance surrounded by “poor” regions; also, selection for improved local performance often pushes the operating values away from regime boundaries, thus increasing global tolerance. These predictions agree with experimental data from the reduced nicotinamide adenine dinucleotide phosphate (NADPH) redox cycle of human erythrocytes.
| The ability of organisms to survive under a multitude of conditions is readily apparent. This robustness in performance is difficult to precisely characterize and quantify. At a biochemical level, it leads to physiological behavior when the parameters of the system remain within some neighborhood of their normal values. However, this behavior can change abruptly, often becoming pathological, as the boundary of the neighborhood is crossed. Currently, there is no generic approach to identifying and characterizing such boundaries. In this paper, we address the problem by introducing a method that involves quantitative concepts for boundaries between regions and “global tolerance”. To illustrate the power of these concepts, we analyzed a large class of biological modules called moiety-transfer cycles and characterized the specific case of the NADPH redox cycle in human erythrocytes, which is involved in conferring resistance to malaria. Our results show that the wild-type system operates well within a region of “best” local performance that is surrounded by “poor” regions.
| Robustness, the notion that biological systems must be able to withstand a variety of perturbations is becoming a cornerstone of research in systems biology. Indeed, several approaches have been developed to understand this concept. These approaches tend to focus on the levels of genotype, intermediate network architectures, or phenotypic expression. None actually provides any relation between these levels because the fundamental mappings between levels have not been solved.
At the level of the genotype, there are approaches dealing with neutral or near neutral mutations, which may be considered the result of a genetic code optimized by natural selection. These include nucleotide substitutions that leave the secondary structure of an RNA unchanged [1], that result in a synonymous codon that leaves the protein sequence unchanged, or that lead to the substitution of an aminoacid with similar physical-chemical properties [2]. The fraction of mutations that fall into these classes provides a measure of the organism's “mutational robustness”.
At the level of intermediate network architectures, there are approaches dealing with the number of redundant paths between points in the network. The number of such redundancies provides another measure of robustness. Perhaps the best example of such architectures is provided by networks at the metabolic level [3]. However, these approaches at the level of genotype and network architecture have little to say about any specific biological function.
At the level of specific phenotypic function, the concept of robustness deals with the relationship between the physiological behavior and the underlying parameters of mechanistic models identified or hypothesized. Most approaches at this level have dealt with the local behavior as characterized by small (infinitesimal) changes. Robustness according to these approaches corresponds to parameter insensitivity–linear sensitivities [4], logarithmic sensitivities [5],[6], or second-order sensitivities [7]–[9]. All of these approaches have shown what has been long known from experimental studies, that there is a spectrum of sensitivities with many parameters having very little influence and a smaller number having the major impact.
There are other approaches that attempt to deal with local changes in parameter values analytically, but only in terms of preserving system stability. For systems with a stable steady state, parameter variations that lead to the loss of stability will first violate one of the last two Routh criteria. The magnitudes of these two conditions can be considered a measure of the “distance” from the boundaries of instability. This distance is often referred to as the margin of stability. The margin in the case of the penultimate condition is the more difficult to evaluate; it involves both kinetic order and rate constant parameters [10]–[12]. The margin in the case of the last Routh criterion is determined more simply by the determinant of the matrix of kinetic orders for the dependent variables [10],[13], alternatively by a method based on singular value decomposition of this matrix [14]. For many systems both conditions are critical and must be evaluated. However, these local approaches have little to say about a system's response to larger changes in parameter values.
One approach to deal with large changes in parameter values involves random sampling of values to obtain an estimate for the volume of parameter space corresponding to physiological behavior [15], although volume alone is not a sufficient measure. The shape of the volume is critical, as pointed out by Morohashi et al. [16]. Sengupta et al. [17] and Chaves et al. [18] have proposed a measure of robustness, based on a random walk in parameter space, that reflects the shape of the robust region. These methods are limited by the computational expense of dense sampling and random walks in high-dimensional parameter spaces.
All of the existing methods have advantages as well as significant limitations. Thus, there is need of a generic approach for dealing with robustness to large changes in parameter values and identifying a variety of qualitatively distinct phenotypes, including but not limited to loss of stability. In this paper, we introduce such a method and illustrate its use in the context of a specific class of biochemical systems, moiety-transfer cycles. In such systems, the variables and parameters, which define its structure, must remain within a neighborhood of their nominal values so as to produce a physiological phenotype. When this neighborhood is exceeded the system exhibits a pathological phenotype.
Our generic approach involves the precise characterization of boundaries between phenotypically distinct regimes and defines “global tolerance” as the ratio (or its reciprocal, depending on which is greater) between the normal value of a parameter and the value at such a boundary where there is an abrupt change in system performance. Thus, systems whose performance remains nearly constant for large deviations from the normal operating point are considered to be “globally tolerant”. This is in contrast to the conventional notion of “local robustness”, defined by small values for the system's parameter sensitivities [5], which results in important aspects of system performance remaining almost constant near the normal operating point. As biochemical parameters might be subject to considerable variation, a small global tolerance might be disadvantageous even if system performance is locally robust.
The notion that large global tolerances may evolve as “safety factors” against fluctuations in parameter values and/or in the loads placed by the environment has been proposed as a possible explanation for large mismatches found between actual biological capacities and apparent physiological needs [19]–[22]. For example, the measured capacity ( value) of hexokinase exceeds the physiological flux in the cardiac muscle of exercising rainbow trout by over three orders of magnitude [21]. More recent studies [23],[24] of concrete systems suggest that large tolerances of pathway fluxes to changes in the activity of the participating enzymes are the side-effect of fulfilling local performance criteria. However, we can envision a situation in which effective local performance will not necessarily lead to large tolerances, and therefore the possibility of performance breakdown due to normal variation in parameter values becomes a major consideration mediating natural selection. A similar point is highlighted by Morohashi M, et al. [11], showing that various aspects of the design for a biochemical oscillator can be rationalized as attending to a requirement for both good local performance and large global tolerance. Therefore, local robustness and global tolerance are both important aspects for the evolutionary design of biochemical systems.
In illustrating our generic approach, we also will address the question: does design for robust local performance necessarily improve global tolerance? In moiety-transfer cycles, a moiety is transferred from a moiety-donor metabolite () to an acceptor metabolite () by way of a charged carrier () (Figure 1). For our example, and under the conditions of interest, we will assume that the sum () of the charged carrier () and the uncharged carrier () is held constant. This form of coupling between reactions is very prevalent in metabolism. Indeed, of all the enzyme-catalyzed reactions in the reconstructed metabolic networks of Escherichia coli [25] and Saccharomyces cerevisiae [26], 836 (75%) in the former organism and 561 (67%) in the latter participate in moiety-transfer cycles. These calculations exclude cycles involving the ubiquitous metabolites H2O and H+, and pairs of forward-reverse reactions. Redundant reactions catalyzed by distinct (iso)enzymes were counted as a single reaction.
The large majority of these cycles mediate the transfer of moieties from catabolic (i.e., nutrient-disassembling and energy-producing) to anabolic (biosynthetic) processes. In this context, they act as “moiety-supply” units, analogous to power-supply units in electric circuits: they must reliably supply a given moiety at the required rate (analogous to current intensity) while keeping the concentration of the charged carrier (analogous to electric potential) fairly constant. Here we address moiety-transfer cycles that play this specific role. Henceforth, when we use the term “moiety-transfer cycles” it should be understood that we are referring specifically to the class of moiety-transfer cycles that act as “moiety-supply” units. We also compare our analytical results to existing experimental results for the NADPH redox cycle of human erythrocytes.
We will assume that each enzyme involved in a moiety-transfer cycle (Figure 1) has two substrates and that the reactions are irreversible. For our particular example, we will use Eqn (1), which is valid for a wide range of two-substrate enzymatic mechanisms (random-order equilibrium, compulsory-order, Theorell-Chance and ping-pong mechanisms) [27]:(1)where: is the concentration of substrate ; is the concentration of substrate ; is the rate of catalysis by enzyme ; is the maximum rate of catalysis by enzyme ; is the Michaelis constant of enzyme with respect to substrate ; is the Michaelis constant of enzyme with respect to substrate ; is the equilibrium dissociation constant for the enzyme-substrate complex ; is 1 if the enzyme follows a random-order equilibrium or a compulsory-order mechanism in which binds first and is 0 if the enzyme follows a ping-pong mechanism.
For purposes of illustration, we will assume that the charging enzyme follows a compulsory order mechanism in which binds first to the enzyme () and the uncharging enzyme follows a ping-pong mechanism (). For simplicity, and without ambiguity since we are only considering two different enzymes, we are going to discontinue using the subscript referring to the enzyme. Hence the terminology that we are going to use throughout the text is as follows (see Figure 1):
The investigation of tolerance requires a mathematical framework that is able to address the effects of large perturbations while avoiding the mathematical complexities of unstructured nonlinear systems. The strategy for our analysis involves (i) decomposition of the system's design space into unique regions with boundaries precisely defined by the “breakpoints” in the piecewise power-law representation, (ii) determination of the system behavior in each region, (iii) evaluation of system behavior according to a set of quantitative criteria based on the function of the system, and (iv) determination of the global tolerance to changes in the values for the parameters and concentrations of the system.
Our approach is based on the idea that performance differs when there is a change in the dominant flux or concentration terms. For instance (Figure 2A), for enzymes that obey the Hill function, the characteristic concentration—typified by the —marks the breakpoint between two regimes in logarithmic space. One is characterized by most of the enzyme being in the free form (slope equal to the Hill coefficient) and the other by most of the enzyme being bound to the substrate (slope equal to zero). More complicated enzyme mechanisms, will involve more than one breakpoint. For instance, some enzymes exhibit substrate inhibition at elevated substrate concentrations (Figure 2B). For these enzymes, there will be three regimes separated by two breakpoints. At substrate concentrations much below the , most of the enzyme is in the free form (slope equal to one); at intermediate concentrations, above the and below the , the enzyme is mostly bound by a single molecule of substrate (slope equal to zero); at substrate concentrations much above the , the enzyme is mostly bound in an abortive or dead end complex between the substrate and one or several enzyme forms (slope equal to −1).
The essential feature of a system, and that any mathematical framework for the analysis of tolerance has to capture, is thus the breakpoints between regimes. These ideas lead us to estimate tolerances within the framework of the piecewise power-law representation of enzyme kinetics, which is one of the four different representations within the power-law formalism of Biochemical Systems Theory [28]. This representation retains the mathematical tractability of the local power-law representation [5], which provides a characterization of the system in terms of logarithmic gains, robustness (as measured by parameter sensitivities) and local stability, while extending the range of application to global considerations.
Formulation of our piecewise power-law representation is analogous to the classical method of Bode [29] and involves three steps ([10], pp 335–341):
Using this method, we derive the piecewise power-law representation:(4)and(5)Although the asymptotes in this example are straight lines in both Cartesian and Logarithmic coordinates, this is not the general case. In the general case, the asymptotes are straight lines only in the Logarithmic coordinates.
Under the condition (Figure 3A) there are three different regimes each with a different steady state. For very small values of , the steady state in Systemic Regime a is valid. In this steady state, the charging enzyme operates within its linear region and the uncharging enzyme operates on its plateau. As increases, there is a transition to the steady state in Systemic Regime c, in which both enzymes operate within their linear regions. Finally, as increases even further, there is a transition to the steady state in Systemic Regime b, in which the charging enzyme operates on its plateau and the uncharging enzyme functions within its linear region.
Under the condition (Figure 3B) there are two different regimes each with a different steady state. For values of less than one, the steady state in Systemic Regime a is valid; when equals one the system experiences a discontinuity and transitions to the steady state in Systemic Regime b for values of greater than one.
Through the analysis of these cases, and of the remaining ones (see Text S1), we are able to determine the design space available to the moiety-transfer cycle (see Figure 4).
Each systemic regime is given by a specific and readily solvable steady-state equation for the dependent variable, and applies only to a particular region of the design space (Table 1). Given this partitioning of the design space into distinct regions, one can define global tolerance as the ratio between the value of a parameter at the operating point (white point in Figure 4A) and the value of that same parameter at the boundary to the next neighboring region (black double headed arrows in Figure 4A).
The system representation within each regime is a simple but nonlinear S-system for which determination of local behavior, after appropriate transformation, reduces to conventional linear analysis [10]. Thus, the local behavior is completely determined and readily characterized by the evaluation of the following quantitative indices.
Logarithmic gains in concentration (e.g., the charged moiety ) or flux (e.g., the rate of charged-moiety supply ) in response to change in value for an independent variable (e.g., the concentration of the moiety-acceptor ) are defined by the relative derivative of the explicit steady-state solution. For example,(6)
Parameter sensitivities of such state variables in response to change in the value for one of the parameters that define the structure of the system (e.g., Michaelis constants or maximal velocities) are defined by the relative derivative of the explicit steady-state solution. For example,(7)
Response time is given by the inverse of the eigenvalue, which is determined by analytical integration of the differential equation that applies for each systemic regime.
What criteria must a moiety-transfer cycle fulfill in order to be considered a good one? This is a question that only now is being posed by biologists. However, this question is analogous to one that engineers have long had to deal with, and the lessons they have learned can now be used to further our understanding of how biological systems are designed through natural selection.
The performance of the moiety-transfer cycle, which is analogous to that of the power supply in an electrical circuit, can be evaluated in each systemic regime according to the following quantitative criteria:
The concentration of charged carrier (analogous to the voltage of the power supply) should be well buffered against:
The supply of charged carrier (analogous to the electrical current) should
The sensitivity of the supply of charged carrier to changes in the concentration of moiety-acceptor should
The response time should
The local performance in the three systemic regimes is determined by the above methods and evaluated according to the criteria defined in the previous section. Our aim is to ascertain which of the systemic regimes is better suited for effective performance of the moiety-transfer cycle as a moiety-supply unit. Note that if this same cycle were to fulfill a different role in the cell, then we would have to define different criteria and, hence, the results could be different. For instance, Golbdeter and Koshland [30] have studied a different type of moiety-conserved cycle that exhibits ultra-sensitivity and switch-like behavior.
Optimum local performance of systems with respect to each criterion and within each regime corresponds to the minimum value possible for the criterion (Optimum Value).
In Table 2, we summarize the results from the analysis of local performance in Systemic Regime a. (Details of these results are presented in Text S2) It is apparent from these results that the performance in Systemic Regime a fulfills all of the criteria defined above. Furthermore, if Condition 1, , is valid, the optimization of criteria 1 through 6 follows the same strategy: and should decrease while, and should increase. Note that there is one apparent conflict between optimizing Criterion 7 along with the previous criteria. In order to optimize criteria 1, 2 and 6, should tend to low values, whereas to optimize performance according to Criterion 7, should tend to high values. This apparent conflict can be readily resolved with appropriate values for , or (for which there are no trade-offs).
Contrary to the results for Systemic Regime a, the performance in Systemic Regimes b and c cannot fulfill criteria 4 and 5 because there is no response to changes in moiety-acceptor (detailed results in Text S2). In addition, even though the performance in Systemic Regimes b and c can have a fast response time (Criterion 6), it will not be with respect to changes in . Therefore, the importance of this responsiveness becomes questionable. Finally, the optimum value of Criterion 1 in Systemic Regime c is 1, whereas that in Systemic Regime b is 3. Since Systemic Regimes b and c share the same optimum values for the remaining criteria, we conclude that overall local performance in Systemic Regime c is better than that in Systemic Regime b.
From the analysis of local performance, it is clear that the only systems that can fulfill all criteria and do it efficiently operate in Systemic Regime a. Although systems that operate in systemic regimes b and c can fulfill some of the performance criteria, they fail in that their supply of charged carrier, , does not respond to changes in the concentration of moiety-acceptor . In analogy to electrical circuits, they resemble a power supply that will not provide additional current when there is an increased demand by the rest of the circuit. Hence, this is a poor design for a power supply unit.
If there had been no regime capable of simultaneously fulfilling all the performance criteria then one would have to evaluate the relative impact on fitness of the failure to satisfy a specific criterion. Regimes that violate performance criteria with a weak effect on fitness would clearly be preferable to those that violate more important performance criteria. If the results showed that all regimes violated important performance criteria, then one may attribute this to an inappropriate model or to incomplete/inaccurate knowledge about the function of the system under analysis.
In summary, we predict that in nature, under basal conditions, a moiety-transfer cycle should operate in Systemic Regime a. Moreover, natural selection should maintain the operating point far from the boundaries to the other regimes for the following two reasons. First, the circuit's local performance improves as the operating point moves away from the boundaries. Second, even where the intra-regime gradient in local performance is modest, excursions into neighboring regimes of poor performance are less likely when the operating point is farthest from the boundaries.
Systemic Regime a holds in the region of design space (Figure 4) defined by the following inequalities:Systems represented within these boundaries exhibit the best local performance and thus these boundaries provide the basis for a natural definition of global tolerance. Namely,
By the use of this definition it is possible to determine analytically the global tolerance to change for each kinetic parameter and independent variable of the system operating in Systemic Regime a. In general, each parameter or independent variable can have a global tolerance with respect to its lower value as well as its upper value. These tolerance values will be denoted “[Tlow,Thigh]”; since one of these is often infinite, we also will use the notation “[Tlow” or “Thigh]” with the other infinite tolerance implied.
There are two different boundaries for Systemic Regime a, and , so we present the tolerance expressions with respect to each in Text S3. When considering each kinetic parameter and independent variable individually, its critical tolerance will be given by the lowest of its tolerance values given in Text S3. Numerical values for these tolerances are given for a specific system in the following section.
We have selected this moiety-transfer cycle to provide a numerical illustration of our results because the kinetic parameters of the enzymes and concentrations of the metabolites for this system have been well characterized experimentally [31]–[34] in view of this cycle's importance in malaria [35]. These values, which are in Text S4, lead to the design space in Figure 5 depicting the steady-state concentration in the z-direction with a heat map. The physiological operating point for this system is found in Systemic Region a, as expected. The design space depicting the steady-state flux has a similar appearance (data not shown).
The local behavior of this system can be evaluated according to the seven criteria described earlier. In this case we have the numerical values for the various parameters and, thus, we can calculate the numerical values for the criteria and compare their values to the optimum values. As can be seen from the resulting data summarized in Table 3, natural selection results in a design that has nearly optimal local performance according to the seven criteria.
Given the numerical values that characterize the operating point for this system, and the boundaries surrounding Systemic Region a, we are able to determine the numerical value of global tolerance for each of the kinetic parameters and independent concentration variables. The values, summarized in Table 4, are tolerances involving movement from Systemic Region a into Systemic Region c. They range from the smallest tolerance of 59 fold to the largest of 362 fold. The smallest values are associated with , , and , whereas the largest are associated with , , and .
It should be emphasized that no change in the value of any single parameter or concentration is capable of moving the operating point of the system from Systemic Region a into Systemic Region b. In this sense, the largest tolerances (essentially infinite) are associated with the boundary between systemic regions a and b.
The organization of biochemical systems has traditionally been viewed as adhering to few general rules. Should it be real, this perceived lack of generally applicable organizing principles would reduce molecular biology to an accumulation of disparate facts with limited predictive value. However, research in molecular systems biology is revealing a number of design principles that associate function with design. For example, such design principles have been found in metabolic pathways [36]–[40], signal transduction cascades [41]–[45], mode of gene control [28], [46]–[49] and coupling of gene circuits [50]–[54]. This research provides an understanding of why some designs are highly prevalent in biochemical systems while other feasible designs are rare. It also prompts predictive inferences of (i) what interactions among biochemical components should occur given the function of a network, or (ii) what is the likely function of a network given its component interactions.
A high priority in the research program of biochemical systems theory is the characterization of design principles for the most common constituents of biochemical systems such as elementary gene circuits and simple metabolic networks. As noted in the Introduction, moiety-transfer cycles are among the most common functional units in metabolic networks. Hence, the material presented in this paper serves not only to introduce an important analytical framework within which to quantitatively characterize the design of biochemical systems, but also to provide insight regarding the design principles that govern one of the most common functional units in metabolic networks.
It must be emphasized that the piecewise power-law representation described in this paper is not an arbitrary fit to the kinetic rate laws. It is not simply a convenient curve-fitting exercise that attempts to minimize the error in the representation by using a sufficiently large number of arbitrary pieces. The number of pieces, their slopes and the location of the breakpoints are all uniquely determined by the rational function in conventional Bode-type analysis ([10], pp 335–341). Moreover, this representation is rigorously justified for the rational functions known to characterize the traditional rate laws of biochemical kinetics [55]. Thus, the method is highly constrained by the model and it produces a unique representation. The class of models can be quite general; for example, it includes generalized mass action models of chemical kinetics and rational function models of biochemical kinetics. Regardless of how one obtains a given model (detailed kinetic analysis, an empirical fit to a model using limited data or a hypothetical model based on general considerations), as long as it falls within this very general class of functions then our approach can be applied.
Differences between the steady-state solutions of the rational function and piecewise representations are greatest around the breakpoints, as is evident from Figure 5. The lack of accuracy at these points may be considered a disadvantage of the piecewise power-law representation. Nevertheless, the piecewise power-law representation suggests the formulation of the design space, provides precise boundaries between regions, and gives a method for defining global tolerances in a quantitative manner. These are all major advantages that would be hard to derive directly from the rational-function representation. Thus, it must be emphasized that in our example the formulation of the design space and the boundaries were first derived from the piecewise representation (depicted in Figure 5A) and then used to display the results from the rational-function representation (depicted in Figure 5B).
The system design space that is defined by our approach provides an important framework to characterize the behavior of the system. Within each region, system behavior is readily solved, often analytically, as for the cases analyzed in this paper. The results presented in this paper can be generalized to other moiety-transfer cycles, as will be documented in a subsequent publication (Coelho et al., manuscript in preparation).
The system design space also provides an important framework to represent and compare wild-type and mutant variants of these systems. The kinetic parameters of the systems can be measured and the resulting values plotted within the common design space. An example is provided in Figure 5 by making use of the data for the wild-type NADPH redox cycle in human erythrocytes [31]–[34].
The location of the operating point for mutants (where such mutants and their kinetic data are available), in relation to that for the wild type and in relation to the boundaries between good and poor regions, will provide a method to quantitatively characterize the physiological significance of mutant phenotypes.
There is a general theorem indicating that the robustness of feedback control systems is a conserved quantity, and thus increasing the robustness in one operating regime must cause it to decrease in another [56]. This suggests that trade-offs are inevitable in the design of a system. It is not yet clear how our results might be governed by this theorem. The differences may reside in the global dynamics of the system, since our analysis focuses on the steady-state behavior and only considers dynamics in the local sense.
As we have seen, an important consideration affecting the location of the operating point for the wild type relative to regime boundaries is the interplay between global tolerance and local performance. Selection for improved local performance often pushes the operating point away from regime boundaries, thus increasing global tolerance. But in some cases modifying the value of a parameter in the direction that improves local performance may bring the operating point closer to regime boundaries, thus decreasing global tolerance.
Our analysis identified two cases of potential trade-offs between specific criteria for local performance and global tolerance. Namely, increasing improves the buffering of the response time against fluctuations in the values of parameters and independent variables, but decreases global tolerances with respect to changes in the values of most parameters. Likewise, decreasing can in some conditions improve buffering against changes in the concentration of moiety-acceptor , but it can decrease global tolerances with respect to changes in the values of most parameters.
However, because these same changes in or would also worsen several other important aspects of local performance they do not entail a real trade-off between overall local performance and global tolerances. Furthermore, none of the trade-offs mentioned above prevent the simultaneous improvement of both local performance and global tolerance by suitably changing the value of a second parameter. Therefore, the simple design of moiety-transfer cycles that we addressed here does not have any irresolvable trade-offs between global tolerance and local performance for the set of performance criteria we considered. This is a desirable property that facilitates the evolutionary adaptation of the cycle to changing environmental demands.
|
10.1371/journal.pntd.0000590 | A CATT Negative Result after Treatment for Human African Trypanosomiasis Is No Indication for Cure | Cure after treatment for human African trypanosomiasis (HAT) is assessed by examination of the cerebrospinal fluid every 6 months, for a total period of 2 years. So far, no markers for cure or treatment failure have been identified in blood. Trypanosome-specific antibodies are detectable in blood by the Card Agglutination Test for Trypanosomiasis (CATT). We studied the value of a normalising, negative post-treatment CATT result in treated Trypanosoma brucei (T.b.) gambiense sleeping sickness patients as a marker of cure.
The CATT/T.b. gambiense was performed on serum of a cohort of 360 T.b. gambiense patients, consisting of 242 primary and 118 retreatment cases. The CATT results during 2 years of post-treatment follow-up were studied in function of cure or treatment failure. At inclusion, sensitivity of CATT was 98% (234/238) in primary cases and only 78% (91/117) in retreatment cases. After treatment, the CATT titre decreased both in cured patients and in patients experiencing treatment failure.
Though CATT is a good test to detect HAT in primary cases, a normalising or negative CATT result after treatment for HAT does not indicate cure, therefore CATT cannot be used to monitor treatment outcome.
| The 2 year follow-up period required after treatment of human African trypanosomiasis (HAT) patients is a major challenge for patients and control programmes alike. The patient should return every 6 months for lumbar puncture and cerebrospinal fluid examination since, so far, no markers for cure have been identified in blood. The Card Agglutination Test for Trypanosomiasis (CATT) is a simple, rapid test for trypanosome-specific antibody detection in blood that is extensively used in endemic areas to screen for HAT. We examined the value of a normalising CATT as a marker for treatment outcome. We observed that CATT titres decreased after treatment both in patients who experienced treatment failure as well as in cured patients. We conclude that CATT, though a good screening test, is unreliable for monitoring treatment outcome. We also showed that the sensitivity of CATT in relapse cases was as low as 78%, and as a consequence some relapse cases might be missed in screening programs if they have no clinical signs yet.
| Since none of the drugs for human African trypanosomiasis (HAT) is 100% efficacious, it is recommended to follow-up sleeping sickness patients every 6 months after treatment, for a period of 2 years. Parasites may be difficult to detect in blood of HAT patients experiencing treatment failure, therefore assessment at follow-up visits relies mainly on lumbar puncture and examination of the cerebrospinal fluid (CSF) for presence of trypanosomes and white blood cell count. A patient is declared cured when, within 2 years, no trypanosomes have been detected and the CSF white blood cell count returned to normal [1]. Complete follow-up is seldom achieved because, when patients feel well, they are reluctant to comply to the follow-up examinations [2]–[5]. So far, no markers for cure or treatment failure after HAT treatment have been identified in blood.
The card agglutination test for trypanosomiasis (CATT) is a fast and simple agglutination test for detection of trypanosome specific antibodies in blood of Trypanosoma brucei (T.b.) gambiense infected patients [6]. With sensitivities between 87 and 98% and specificities of around 95%, the CATT test is extensively used in almost all HAT endemic areas for population screening, and has contributed to the current success of HAT control programs [7],[8]. Given the fact that drugs for HAT are toxic, and the specificity of CATT is limited, a confirmation step by parasitological techniques is needed [7]. Trypanosome specific antibodies, detectable by CATT have been demonstrated even 24 months after successful treatment in no less than 47% of gambiense HAT patients [3],[9],[10]. A positive post-treatment CATT result is therefore not indicative of treatment failure, but the predictive value of a negative CATT after treatment has hitherto not been evaluated. We explored the hypothesis that a normalising, negative post-treatment CATT result indicates cure in gambiense HAT and rules out treatment failure. If such CATT-normalising patients could be released from further follow-up, this would lead to major clinical and public health benefits as less lumbar punctures would be required and less patients should be followed for up to 24 months.
We report here on the pre- and post-treatment CATT serum results in a cohort of primary and retreatment HAT cases infected with T.b. gambiense.
Sleeping sickness patients originate from a prospective observational study (THARSAT) [11]. The Commission for Medical Ethics of the Prince Leopold Institute of Tropical Medicine, Antwerp, Belgium and the Ethical Commission of the Ministry of Public Health, Democratic Republic of the Congo approved the study. Written informed consent was given by all study participants prior to enrolment.
The cohort consisted of 242 primary HAT cases that had never been treated for HAT and of 118 retreatment cases previously treated for HAT, but with trypanosomes detected in CSF at inclusion. All cases were parasitologically confirmed before enrolment and were (re)treated according to the national guidelines: primary cases in first stage (n = 41) were treated with pentamidine, primary cases in second stage were treated with melarsoprol (n = 192) or eflornithine (n = 9). Retreatment cases were treated with melarsoprol (n = 7), eflornithine (n = 52), melarsoprol nifurtimox combination therapy (n = 57), melarsoprol eflornithine combination therapy (n = 1) or eflornithine nifurtimox combination therapy (n = 1). Patients were monitored for treatment outcome during 2 years. The detailed description of the clinical outcomes in the cohort is given elsewhere [11]. In brief, out of 242 primary cases, the final outcome was cure in 90 (cure or probable cure) and treatment failure in 118 (relapse, probable relapse, or HAT related death during follow-up). 34 primary cases were excluded from the analyses of post-treatment results since they could not be classified as cured or treatment failure because they were lost to follow-up, died during treatment or died over the following 2 years from non-HAT related causes. Out of the 118 retreatment cases, 85 were cured and 16 experienced a new treatment failure. Seventeen retreatment cases were lost to follow-up, died during treatment or died over the following 2 years from non-HAT related causes and were also excluded from the analyses of post-treatment results.
CATT/T.b. gambiense was performed following the titration-method as described by the manufacturers [6] on serum taken before treatment and at 3, 6, 12, 18 and 24 months post-treatment. The end titre (highest dilution giving agglutination) was determined. Patients with end titres ≥1∶4 were considered CATT positive, end titres <1∶4 were considered CATT negative.
The Chi square test or Fisher exact test (when the number of observations in a cell was <5) was performed for comparison of proportions using a 95% confidence limit. Odds ratios (OR) with binomial 95% confidence intervals (CI) were computed. STATA version 10 was used for data analysis.
The distribution of CATT end titres in primary and retreatment cases at inclusion is presented in figure 1. The median end titre in primary cases was 1∶16 (interquartile range [IQR] 1∶8–1∶16, mean±standard deviation: 14±9), while it was 1∶4 (IQR 1∶4–1∶8, mean±standard deviation: 8±14) in the retreatment cases included in the cohort. Sensitivity of CATT was 98.3% in primary cases (234/238, CI 95.8–99.5%) and 77.8% (91/117, CI 69.2–84.9%) in retreatment cases. The median time between previous and current treatment in retreatment patients was 10 months (IQR 6–16 months, data available for 103/118 cases).
The CATT results after treatment - in function of cure or treatment failure- are shown in figure 2.
In the 90 cured primary HAT cases, the median end titre decreased to 1∶8 (IQR 1∶4–1∶8) and 1∶4 (IQR <1∶4–1∶8) after 3 and 6 months respectively, and became <1∶4 afterwards. As shown in figure 2, the proportion of CATT positives decreased in the cured group over time to 52% (45/87) and 37% (30/81) at 6 and 12 months and to 18% (15/83) at the final follow-up visit at 24 months (also called test of cure). In the 118 primary cases who experienced treatment failure within the 2 years of follow-up, the median end titre decreased to 1∶8 (IQR 1∶4–1∶8) after 3 months and 1∶4 (<1∶4–1∶8) after 6 and 12 months. The proportion of CATT positives also decreased in function of time to 67% (44/66) and 62% (13/21) at respectively 6 and 12 months after treatment.
No significant relationship between CATT positivity and occurrence of treatment failure (p>0.05) could be observed at 3, 6 and 18 months post-treatment. A significantly higher proportion of treatment failures cases tested positive with the CATT compared to the cases that were cured 12 months (Chi square test, p = 0.040) and 24 months (Fisher exact test, p = 0.027) after treatment. The odds of a treatment failure case being CATT positive are 2.76 (95% CI 1.03–7.4) and 13.6 (95% CI 1.32–140) times greater than the odds of a cured case being CATT positive 12 and 24 months after treatment. In 7/113 primary cases trypanosomes were detected in blood at time of relapse. Two of them relapsed at 3 months with CATT titres 1∶8 and 1∶16 ; three others showed a titre 1∶4 (relapses at 6 and 12 months) and two relapsed at 24 months with titres 1∶8 and 1∶16.
In the 85 retreatment cases who were cured after the current treatment, the median end titre was 1∶4 (IQR <1∶4–1∶8, IQR <1∶4–1∶4) after 3 and 6 months, and became <1∶4 afterwards. The proportion of CATT positives decreased over time to 56% (45/81) and 36% (28/77) at 6 and 12 months and 17% at the 24 months test of cure visit (figure 2). In 16 retreatment cases that experienced a repeated treatment failure after the current treatment, the median end titre was 1∶4 (IQR <1∶4–1∶4) and the proportion of CATT positives 57% (8/14) 6 months post-treatment. No significant relationship between CATT positivity and occurrence of treatment failure (p>0.05) could be observed during follow-up. In 3/16 of retreatment cases trypanosomes were detected in blood at time of relapse. Their CATT titres were <1∶4 (relapse at 3 months), and <1∶4 and 1∶16 (relapses at 6 months).
We demonstrate for the first time that CATT sensitivity is low in retreatment cases, and that CATT titres decrease after treatment both in patients who experience treatment failure as well as in cured patients.
Before treatment, the CATT sensitivity in primary cases falls within the sensitivities previously reported for CATT in the Democratic Republic of the Congo, and for HAT in general [7],[10]. The low sensitivity of 78% observed in retreatment cases is explained by the decrease in CATT titre after a previous treatment, and largely corresponds to the proportion of CATT positives observed 6 and 12 months post-treatment within the groups of treatment failures. The observed end titers are relatively low, being in respectively 39% and 83% of the primary and retreatment cases below 1∶16. Treating serological cases based on a CATT end titer ≥1∶16, without parasitological evidence, might miss some HAT cases.
Although it has been shown that trypanosome specific antibody concentrations in blood of cured patients may persist up to 2 years or longer after treatment [3],[9],[10],[12], reports about the concentrations of specific antibodies in serum of sleeping sickness patients who experience treatment failure are rare. In 22 relapsing patients, Frézil et al. [12] describe that the immunofluorescence test remains positive in the majority of relapsing cases, but doubtful/negative in only 1 case. In a small cohort of 32 relapse cases, Miézan et al. [3] describe a decreasing antibody concentration and a CATT positivity rate of 94% at the moment relapse is diagnosed.
A negative CATT after unsuccessful treatment might be explained by trypanosomes that are cleared from peripheral tissues, such as lymph and blood, but that survive in the brain and thus do not trigger specific antibody production in the blood.
Our study has a number of limitations. As a consequence of the diagnostic procedure used by the HAT control program to detect HAT, the observed sensitivity of CATT of 98% in our group of primary cases might be higher than in other patient cohorts. Indeed, the patients in our cohort were identified as follows: CATT on whole blood, alongside cervical lymph node palpation, was used as a screening test and only those persons with a CATT positive result on whole blood, or having enlarged cervical lymph nodes underwent parasitological examinations for case confirmation. Although part of the false negatives in CATT will be found by cervical lymph node palpation, the true sensitivity of CATT in the primary cases might be lower than 98%. The number of treatment failures detected after ≥12 months is low, which prevents us from making reliable estimates of a further de- or increase in CATT titres after that time point, nor of the proportion of CATT positives. Moreover, follow-up examinations in this cohort- as in routine clinical care- were focused on cerebrospinal fluid examination and the need for blood examinations may have been given less importance by the nursing staff. As a consequence the cohort does not allow us to check if relapsing patients with trypanosomes in the blood had higher CATT titres than those without, since in the majority, relapse was confirmed by finding of trypanosomes in the CSF and no further blood examinations were performed. Finally, the majority of primary cases in this study were treated with the trypanolytic drug melarsoprol in an area of high treatment failure rates. Although a similar trend was observed in retreatment cases, who were treated differently, we cannot exclude that results could differ in primary HAT patients treated with other drugs.
Our findings have 2 practical implications. First, the considerable proportion of CATT negative results in cases experiencing treatment failure, which increases over time, implies that a post-treatment CATT negative result does not necessarily indicate cure. Knowing, moreover that a post-treatment CATT positive result does not indicate treatment failure, makes us conclude that CATT is unreliable for monitoring treatment outcome. Secondly, screening programs for HAT should take into consideration that a careful history about past HAT episodes is paramount, as the sensitivity of CATT in relapse cases is not optimal. Cases experiencing treatment failure are more likely to be false negative in CATT than new cases and, as a consequence, might be missed (i.e. not offered parasitological investigations) if they show no clinical signs. These data might cast some doubt on the performance of CATT as a screening test in the detection process, given the fact that some relapse cases appear to be negative in the CATT. Molecular -or other- diagnostics might eventually be taken up in an improved algorithm for diagnosis or follow-up but further investigation of these tests is necessary.
|
10.1371/journal.ppat.1000849 | SARS-CoV Pathogenesis Is Regulated by a STAT1 Dependent but a Type I, II and III Interferon Receptor Independent Mechanism | Severe acute respiratory syndrome coronavirus (SARS-CoV) infection often caused severe end stage lung disease and organizing phase diffuse alveolar damage, especially in the elderly. The virus-host interactions that governed development of these acute end stage lung diseases and death are unknown. To address this question, we evaluated the role of innate immune signaling in protection from human (Urbani) and a recombinant mouse adapted SARS-CoV, designated rMA15. In contrast to most models of viral pathogenesis, infection of type I, type II or type III interferon knockout mice (129 background) with either Urbani or MA15 viruses resulted in clinical disease outcomes, including transient weight loss, denuding bronchiolitis and alveolar inflammation and recovery, identical to that seen in infection of wildtype mice. This suggests that type I, II and III interferon signaling play minor roles in regulating SARS pathogenesis in mouse models. In contrast, infection of STAT1−/− mice resulted in severe disease, high virus titer, extensive pulmonary lesions and 100% mortality by day 9 and 30 post-infection with rMA15 or Urbani viruses, respectively. Non-lethal in BALB/c mice, Urbani SARS-CoV infection in STAT1−/− mice caused disseminated infection involving the liver, spleen and other tissues after day 9. These findings demonstrated that SARS-CoV pathogenesis is regulated by a STAT1 dependent but type I, II and III interferon receptor independent, mechanism. In contrast to a well documented role in innate immunity, we propose that STAT1 also protects mice via its role as an antagonist of unrestrained cell proliferation.
| The SARS coronavirus is a highly pathogenic respiratory virus that caused the first epidemic of the 21st century. During the epidemic ∼10% of those infected died and the elderly were particularly vulnerable. Severe cases developed acute lung injury with pulmonary fibrosis and Acute Respiratory Distress Syndrome (ARDS). Little is known about the molecular mechanisms governing its virus pathogenesis and high lethality. Using a mouse model of infection with the epidemic strain of SARS-CoV (Urbani) as well as a recombinant mouse adapted strain of SARS-CoV (rMA15), we showed that a protein normally associated with the innate immune response, STAT1, plays an important role in the development of severe end stage lung injury. However, the lack of a normal innate immune type I, type II and type III interferon response did not enhance virus pathogenesis. Our work suggests that STAT1 may play a key role in development of acute lung injury and other chronic lung pathology, most likely by affecting cell proliferation and wound repair pathways.
| SARS Coronavirus (SARS-CoV) is a highly pathogenic respiratory virus that emerged in China during the winter of 2002 and infected about 8,000 people globally and resulted in ∼800 deaths, with greatly increased mortality rates in persons over 50 years of age (WHO). On initial isolation of SARS-CoV from infected patients, it was identified as a novel Group 2 Coronavirus and the genetic mechanisms governing the increased pathogenicity of the virus remain undefined [1],[2]. In severe cases, SARS-CoV infection rapidly progressed to acute respiratory distress syndrome (ARDS) during the acute phase of infection or to an organizing phase diffuse alveolar damage following virus clearance; two clinically devastating end stage lung diseases. The molecular mechanisms governing these severe end stage lung disease outcomes are unknown, although similar pathologies have been reported following H5N1 and 1918 influenza virus infection.
The innate immune response is a key first line of defense against invading pathogens and is dependent on various signaling pathways and sensors that ultimately induce hundreds of anti-viral proteins to establish a suboptimal environment for replication and spread of invading pathogens [3],[4]. During virus infection the type I interferon (IFN) induction and signaling machinery is key to the initiation of this response. IFN induced from either infected cells or dendritic cells can activate an antiviral state in neighboring cells to signal that a viral infection is under way[5]. Not surprisingly, virus infections (mouse hepatitis virus, influenza virus, RSV, alphaviruses, flaviviruses, etc.) of rodents that lack type I or type II IFN regulatory networks result in increased pathogenesis and mortality, documenting the key role IFNs play in regulating disease outcomes[6]–[12].
Given the importance of the IFN system in regulating virus growth, many highly pathogenic viruses encode proteins that antagonize components of the innate immune system. The Ebola virus encodes VP35 which blocks STAT1 signaling[13],[14], influenza NS1 blocks IRF3 activation[15],[16] and V proteins from the Nipah and Hendra viruses induce STAT1 degradation[17]. Several IFN antagonist genes are encoded in the SARS-CoV genome and target IFN sensing and signaling and NFKB induction[18]–[21]. However, the role of IFN in regulating SARS-CoV pathogenesis in vivo is less clear. From microarray studies using RNA from PBMC's for 50 SARS patients, Cameron et. al. observed a robust type I IFN response early in the disease and predicted that it was essential for viral clearance and clinical recovery[22]. In addition to a strong IFN response, genes such as CCL2 (MCP-1) and CXCL10 (IP-10) that are typically induced by IFN were also up-regulated in patients at early times in SARS-CoV disease. Interestingly, Cameron and colleagues suggested that the IFN response that served to protect recovered SARS patients could become a dysregulated cytokine storm in severe cases and contribute to increased disease and a suboptimal adaptive immune response. The innate immune response to SARS-CoV also changes with age of the host; 12 month old BALB/c mice respond to SARS-CoV with an exacerbated and faster innate immune induction than 6 to 8 week old mice, potentially explaining their increased susceptibility and lung pathology following infection[23],[24].
Several studies have demonstrated the importance of the innate response in viral clearance. Respiratory syncytial virus (RSV) and SARS-CoV require MyD88 to produce an effective immune response to control severe infection in the lung [25],[26]. The Mx gene product plays an important role in the susceptibility of rodents to influenza virus infection and could regulate severity of SARS-CoV disease in vivo as well[27]. Glass et al demonstrated that SARS-CoV was cleared from the lungs of wildtype C57BL/6 mice and strains that lacked that T, B and NK cells with similar kinetics, suggesting that the innate immune response alone is sufficient for viral clearance [28]. Hogan et. al. also demonstrated that STAT1, a key modulator of IFN α/β, λ, γ signaling, was required for the resolution of wildtype SARS-CoV infection, once again indicating the importance of the innate response in the clearance of SARS-CoV [29].
In this paper, we compare the pathogenesis of both the epidemic virus (Urbani) and an isogenic, highly pathogenic mouse-adapted SARS-CoV (rMA15) in different mouse strains, each deficient in different innate immune signaling components. Specifically, we tested the role of type I, type II, and type III IFN and STAT1 in protection from SARS-CoV infection. We have found that absence of the IFN α/β, IFNγ and IFNλ, receptors alone or in some instances together, had a limited impact on pathogenicity and clearance of the non-lethal and lethal strains of SARS-CoV in mice. However, STAT1 deficient mice show increased susceptibility, prolonged virus shedding and mortality following infection with either virus. Importantly, STAT1−/− animals developed an organizing phase DAD, similar to lesions noted in severe late stage human cases of SARS. Our data reveals a new mechanistic pathway by which STAT1 regulates the severity of viral pathogenesis in the lung. We show that SARS-CoV pathogenesis is STAT1 dependent but independent of type I, II and III IFN signaling, and we provide evidence consistent with an essential role for STAT1 in the control of SARS-CoV replication, cell proliferation, wound repair and progression to severe organizing phase DAD and lethal disease.
To understand the role of type I (IFNα/β) and type II IFN (IFNγ) in the response to SARS-CoV infection we infected IFN receptor −/− mice with either a late phase SARS-CoV (Urbani) virus or a recombinant isogenic mouse-adapted virus (rMA15) that contained six virulence modifying mutations [30], anticipating that the mouse-adapted virus may display more prominent pathogenicity than the Urbani virus. The Urbani virus and the equivalent recombinant icSARS viruses are not lethal in 10-wk old BALB/c, C57BL6 and 129 WT mice, typically replicating in the lungs with a peak titer at 2 days post-infection (dpi) before being cleared over the next 2–4 days without clinical disease or mortality [26],[30]. In contrast, the rMA15 virus is lethal in 10-wk old BALB/c mice causing greater than 20% weight loss by 4 days post-infection and death by 4–5 days post-infection. We infected 129 WT, Type I IFN receptor knockout mice (IFNAR1−/−) and Type I/II double IFN receptor knockout mice (IFNAGR−/−) with the Urbani virus (Figure S1) and 129 WT, IFNAR1−/− and IFNGR−/− with the rMA15 virus (Figure 1).
In 129 WT, IFNAR1−/− and IFNGR−/− mice, clinical findings including weight loss data and morbidity were identical after infection with rMA15. 129 WT, IFNAR1−/− and IFNGR−/− mice infected with rMA15 virus lost ∼15% of their weight by day 4 and then steadily recovered over the next 5 days (Figure 1). In contrast, Urbani virus infected mice continued to gain weight through the course of infection (Figure S1). In both cases, virus titers in the lungs peaked at day 2 post-infection, approaching titers of ∼5.0×107 pfu or TCID50/g (rMA15 or Urbani, respectively), and all were rapidly cleared by day 9 post-infection. Although virus titers were ∼5-fold higher in IFNAR1−/− mice than titers in 129 WT mice on day 5 post-infection, we found that IFNAR1−/− and IFNGR−/− mice showed no increase in susceptibility, pathogenesis or histological outcomes to rMA15 or Urbani virus infection. To examine the importance of cooperative IFN pathways in disease, IFN receptor double knockout (IFNAGR−/−) mice were infected with the Urbani virus; no increase in virus titer or weight loss was seen compared to 129 WT mice or the single IFN receptor knockout mice (Figure S1). These data suggest that neither type I nor type II IFN receptors are critical for the regulation of SARS-CoV infection and pathogenic outcomes in mice.
In contrast to the results from IFN receptor knockout mice, previous studies have suggested that STAT1 −/− mice do not clear the Urbani virus by day 22 post-infection[29]. We infected STAT1−/− mice to evaluate whether the absence of a downstream IFN signaling protein results in increased susceptibility to SARS-CoV infection, and to determine the course of infection and pathologic changes associated with the virulent mouse-adapted virus. Importantly, STAT1−/− mice infected with rMA15 virus were more susceptible to disease than 129 WT, IFNAR1−/− or IFNGR−/− mice. Following rMA15 virus infection, STAT1−/− mice lost 15% of their starting weight by day 4 and continued to lose weight through day 9 post-infection (Figure 1). As weight loss approached 30%, the mice were moribund and succumbed to lethal infection. In stark contrast and consonant with earlier reports, STAT1−/− mice infected with Urbani virus initially gained weight as 129 WT mice did, through day 12 post-infection (Figure S1). However, over the next 15 days they displayed worsening clinical disease and lost significant body weight. They did not recover by day 29 post-infection, and 30% of them died.
Lung virus titers in STAT1−/− mice were also higher at each timepoint tested compared to the titers seen in IFN receptor knockout and 129 WT mice, (Figure 1B). When infected with rMA15 virus, STAT1−/− mice showed higher peak virus titers in the lungs at day 2 (∼108 pfu/g) that remained as high as >106 pfu/g at day 9 post-infection, while 129 WT mice had cleared the virus by 9 days. Urbani virus infected STAT1−/− mice also showed increased and sustained virus replication in the lungs as late as 15 days post-infection while virus was detectable only through day 5 post-infection in the 129 WT mice (Figure S1E). Taken together, these data suggest that a STAT1-dependent, IFN type I and II receptor-independent pathway plays a key role in regulating viral clearance and survival following SARS-CoV infection.
Lungs from the various mouse strains infected with SARS-CoV were analyzed for severity of histopathology (Figure 2). In the 129 WT mice, rMA15 virus caused a denuding bronchiolitis at 2 days post-infection with significant apoptosis of airway epithelial cells (noted by multilobed nuclei, condensed chromatin and nuclear blebbing), accumulation of apoptotic debris within the airways and perivascular cuffing caused predominately by lymphocytes. Similar lesions have been noted in BALB/c and C57BL6 mice[26]. rMA15 virus infection was primarily localized in airway epithelium at day 2 post-infection (Figure 3) and did not disseminate to other areas of the lung or respiratory tract. Some perivascular cuffing was noted in the vasculature by day 2 post-infection as well. By day 5, the denuding bronchiolitis, obstruction of the conducting airways by apoptotic debris and apoptosis of the airway epithelium were rarely observed, although perivascular cuffing and a mixed inflammatory response with lymphocytic infiltration of eosinophils, neutrophils and macrophages was more prominent. By day 9 post-infection, the remaining inflammation caused by rMA15 virus infection was primarily found in peribronchiolar areas (Figure 2).
Histologic changes in the lungs of IFNAR1−/− and IFNGR−/− mice were similar to 129 WT mice; the inflammation found in lung sections was temporally related to viral titer. At day 2 post-infection, a denuding bronchiolitis characterized by significant apoptosis and cell death in the airway epithelium was observed and conducting airways were obstructed by apoptotic debris. A mixed infiltrate composed of lymphocytes, neutrophils and eosinophils was seen surrounding the bronchial epithelium. By day 5 post-infection, significant inflammation was evident throughout the lungs and alveoli, resulting in a mild to moderate pneumonitis. As noted in lungs from 129 WT mice, the inflammation was clearing by day 9 post-infection; minimal inflammatory infiltrates remained in the periphery of the lungs and minor peribronchial lymphocytes remained.
At day 2 post-infection with rMA15 virus, lung lesions in STAT1−/− mice were indistinguishable from those seen in 129 WT, IFNAR1−/− and IFNGR−/− infected animals. By day 5 post-infection, lung lesions were more severe in STAT1−/− mice, consistent with persistently high virus titers (Figure 2). Two features were notable; first, the extent of inflammation was more severe in STAT1−/− mice, with a much greater number of macrophages infiltrating into all areas of the lung, a finding that was not seen in the other mouse strains. Second, thickening of the alveolar septa was seen throughout the lungs with perivascular and peribronchial thickening. By 9 days post-infection, the inflammation continued to increase in STAT1−/− mice while lung inflammation was subsiding in 129 WT, IFNAR1−/− and IFNGR−/− infected mice. In STAT1−/− mice, inflammation was pervasive throughout the lungs with densely packed infiltrating cells especially prominent throughout the periphery of the tissue. Macrophages continued to infiltrate with the majority residing in the alveolar interstitium. Interestingly, large foci containing densely packed fibroblasts, macrophages and lymphocytes were seen throughout the lungs and scattered atypical large cells were found throughout the foci. A focal mixed inflammatory infiltrate was found with extensive fibrin deposition. Pleuritis signified by a breakdown of the pleura was seen in most samples as well. This pathology was consistent with proliferative and organizing phase diffuse alveolar damage (DAD). It is noteworthy that similar pathologic lesions were seen in many fatal SARS cases [31]–[34].
Lesions developing from days 2 to 9 in 129 WT, IFNAR1−/− and IFNAGR−/− mice inoculated with the Urbani virus were similar to those seen in rMA15-infected 129 WT mice. At days 2–3, there was diffuse bronchial and bronchiolar necrosis and migration of leukocytes (CD3+ T-cells, neutrophils and eosinophils) through vessel walls to peribronchiolar and perivascular areas but extension of lesions into the alveoli was not prominent (Figure S1F). Viral antigen was seen within the bronchial and bronchiolar epithelium (data not shown) as previously published[35]. By day 5, less necrosis was seen in the bronchiolar epithelium but peribronchiolar and perivascular cuffing remained. At day 9, necrosis was not observed and only minor perivascular cuffing was present. At days 15 and 22–29, the lungs were mostly normal except for minor perivascular and peribronchiolar cuffing of lymphocytes.
In STAT1−/− mice infected with the Urbani virus, there was a similar distribution of bronchial and bronchiolar lesions in the lung as seen in the 129 WT mice at day 3, but with more necrosis and inflammation. Edema was observed around peribronchiolar blood vessels and the larger inflammatory cell infiltrate contained many lymphocytes, and neutrophils with a few eosinophils and macrophages. Abundant viral antigen was seen in bronchiolar epithelium (data not shown). By day 5, the bronchiolar epithelium had foci of regeneration with little necrosis and the peribronchiolar and perivascular inflammation was less severe than at day 3. Plugs of cellular debris and fibrin filled some bronchioles. A blue tinge was noted in the perivascular edema fluid, suggesting early collagen deposition. Viral antigen was much less common in the epithelium but was abundant in the cellular debris in the airways. At day 9, inflammatory cells increased around residual inflammatory lesions, with abundant neutrophils and more macrophages in the lesions. Marked epithelial hyperplasia was also seen in these foci. Bronchiolitis obliterans was seen in a few airways and interstitial lesions developed around some airway lesions, one of which extended to the pleura and producing focal pleuritis. Much of the lung parenchyma was, however, histologically normal. Only a few cells or cell debris was seen expressing viral antigen. By day 15, a fibrinous pleuritis with pyogranulomatous lesions developed in 2 of 3 mice, with focal resolving parenchymal lesions including a few foci of chronic interstitial pneumonia but most of the lung parenchyma remained fairly normal. From day 15–24, a fibrinous peritonitis (Figure S2A), pleuritis and pyogranulomatous lesions in spleen, liver and omentum developed as the major lesions that likely contributed to illness and death in the STAT1−/− mice (Figure S2B-F). These lesions were characterized by a central area of necrosis with numerous neutrophils and an outer zone of macrophages. Viral antigen was found in some of the macrophages in these pyogranulomatous lesions (Figure S2D, E). Plasma cells became abundant in the splenic lesions by day 24 (Figure S2G). Abundant fibrosis (detected by Masson's trichrome stain) was seen in the splenic and liver lesions at day 24 (Figure S2F). The lungs of mice from days 15–24 were mostly normal with areas of residual chronic inflammation and a few pyogranulomas, some of which contained viral antigen. The pleura had nodules of pyogranulomatous and fibrinous inflammation.
We determined the localization of virus in infected lungs during the course of infection by in situ hybridization. We used a riboprobe complementary to the SARS-CoV nucleocapsid RNA that was labeled with radioactive nucleotides to visualize viral RNA in the tissue. In situ hybridization was performed on lungs from 129 WT, IFNAR1−/−, IFNGR−/− and STAT1−/− mice harvested at days 2, 5, and 9 post-infection. In all 4 strains of mice, the probe signal was predominantly localized in the airway epithelial cells at day 2 post-infection (Figure 3A). This correlated with the pathologic lesions in airway epithelial cells at day 2 post-infection. By day 5 post-infection, viral RNA was virtually eliminated from the control 129, and IFNAR1−/− mice and to a slightly lesser extent in IFNGR−/− mice. Occasionally, a few cells with viral RNA were noted in the periphery, consistent with the low titers of virus in these animals at day 5. By day 9, there was no viral RNA signal found in the lungs of 129 WT, IFNAR1−/− and IFNGR−/− mice, corresponding to the lack of viral replication at this time point. In contrast, lungs of STAT1−/− mice showed significant viral RNA signal throughout the lungs, including prominent distribution into the periphery of the lung at early and late times post-infection. Surprisingly, but consistent with the findings on histopathology, viral RNA signal was lost from the airway epithelial cells in the bronchioles by day 9 in STAT1−/− mice and was found throughout the periphery of the lungs, prominently focused in large focal compactions of cells. These foci correspond to the prominent focal lesions noted in H&E stained sections of the lung, which are predominantly composed of fibroblasts, macrophages and lymphocytes (Figure 3B). Interestingly, cellular debris can be seen in these foci that may represent lysed airway epithelial cells.
We observed marked differences in pathology of lungs from 129 WT mice compared to STAT1−/− mice after infection with the rMA15 virus. To investigate whether STAT1 affected the inflammatory infiltrate, we isolated leukocytes from enzymatically disrupted mouse lungs harvested on day 8 post-infection and used cell surface markers to quantify macrophage (CD11c−/CD11b+/GR-1int/MHCII+), neutrophil (F480−/CD11b+/CD11c−/GR-1+), and eosinophil (CD11c−/Siglec+/GR-1−) populations in 129 WT and STAT1−/− mice. A greater than 10-fold difference in leukocyte numbers was found between 129 WT and STAT1−/− infected mice (Table 1). As shown in Table 1, the eosinophil population increased from ∼2% in 129 WT mice to ∼30% in STAT1−/− mice and neutrophils increased from 3% in 129 WT mice to 35% in STAT1−/− mice. Additionally, the number of macrophages in STAT1−/− mice was more than double that detected in 129 WT mice. These numbers are concordant with the histological findings by H&E staining at 9 days post-infection.
We used real time RT-PCR and CBA analysis to quantify changes in mRNA and protein expression levels of several innate immune factors involved in pulmonary inflammation. Specifically, we compared the induction of several pro-inflammatory cytokines (TNFα, IL6, IFNγ and MCP1) for changes in expression during infection by analyzing lung homogenates by CBA (Figure 4A, B, C and D) or quantitative RT-PCR to measure levels of mRNAs in Urbani virus or rMA15 virus infected mice, respectively (Figure 4E, F, G and H).
In Urbani virus infected lungs, minimal changes were seen in 129 WT, IFNAR1−/−, IFNAGR−/− mice for all 4 cytokines (Figure 4A, B, C and D). However, in STAT1 −/− mice, significant increases in protein expression patterns were detected across different time points, peaking between 9 and 15 days post-infection. TNFα protein levels increased steadily from day 2 through day 29 post-infection while IL6 and IFNγ protein levels peaked at day 9, before reducing to levels seen in mock infected animals. MCP1 protein levels were increased between days 5 through 22 but diminished by day 29 post-infection.
Cytokine induction following rMA15 virus infection was compared with Urbani virus infection (Figure 4A–H). Comparing cytokine transcript levels on days 2, 5 and 9 post-infection, we noted some similarities and important differences. Following rMA15 virus infection, 129 WT mice showed a 10-fold induction of TNFα transcripts by day 2 post-infection, followed by progressively reduced levels of expression that returned to baseline by day 9. While IFNAR1−/− mice showed similar kinetics as 129 WT mice, infection in IFNGR−/− mice resulted in a continued rise in TNFα expression till day 9. In contrast, STAT1−/− mice had low baseline levels of TNFα expression through day 5, after which high levels of expression were noted at day 9, corresponding to the peak time of inflammation in the lungs, just prior to death.
IFNγ and MCP1 gene expression levels were similar in rMA15 virus-infected 129 WT, IFNAR1−/− and IFNGR-/mice. Transcript levels were induced by day 2 and then decreased through day 9. While STAT1−/− infected mice showed a similar trend of increased transcript levels at day 2, expression levels of IFNγ and MCP1 continued to increase through day 9 post-infection. Finally, IL6 expression showed similar kinetics in rMA15 virus infected 129 WT control and IFNAR1−/− mice with peaks at day 2 that decreased to baseline levels by day 5. IFNGR−/− mice showed similar kinetics, but with lower expression levels at day 2. Comparatively, induction of IL6 in STAT1−/− mice followed a different pattern, with high levels of expression at day 2 post-infection and persistent high level expression through day 9. IFNβ, IFNα4, IL28B, IL18, IRF1 and OAS1 were analyzed by Real-time PCR from lung RNA isolated during the rMA15 timecourse in WT, IFNAR−/− and STAT1−/− mice (Figure S3). IFNβ, IFNα4, IL28B, IRF1 and OAS1 show a peek of induction at 2dpi for all 3 strains with a reduction throughout the timecouse with a few exceptions. IFNβ, IFNα4 and OAS1 show either a sustained or late burst in expression of each in only the STAT1−/− mice. Interestingly, IL18 shows minimal induction (∼1.5 fold) in WT mice but even less in the IFNAR−/− and STAT1−/− mice. Thus, significant differences in cytokine expression patterns were noted between STAT1, WT and IFNAR deficient animals, where typically WT and IFNAR−/− mice had the same expression patters while STAT1−/− mice were typified by perceived loss of regulatory control and persistent high level expression.
STAT1 is a mediator of Type III IFN (IFN L, Lambda) signaling in addition to Type I and II IFN signaling. IFNL is minimally but significantly upregulated during infection in WT 129 mice at day 2 post infection which decreases through the course of 9 days post infection (Figure 5A). The receptor for IFNL is a heterodimer of IL10Rb and IL28Ra. IL28Ra−/− mice have been generated on the BALB/c mouse background but not on the 129 WT mouse background. Thus, although direct comparison with the other mouse strains is not possible, experiments with SARS-CoV in these mice are still informative. As in the C57B6 background, the Urbani virus does not cause disease or weight loss but the virus replicates in the lungs of BALB/c mice. Virus reaches peak titer by day 2 post-infection and infection is resolved by day 7[36]. In contrast, rMA15 virus infection of BALB/c mice causes death by day 4 or 5 post-infection[30]. We hypothesized that if IFNL was important for protection from SARS-CoV infection, the virus would be more virulent in mice lacking the IFNL receptor and IFNL receptor knockout mice would show evidence of disease while normal BALB/c mice would not. Further, rMA15 virus infection of IL28Ra−/− mice may result in enhanced pathology with more weight loss, higher virus titer, or increased lung pathology.
This hypothesis was tested by inoculating control BALB/c mice and IL28Ra−/− mice with the Urbani virus, rMA15 virus or PBS. There was no weight loss in either BALB/c or IL28Ra−/− mice infected with the Urbani virus (data not shown), thus the lack of IFNλ signaling did not potentiate clinical disease. Lungs from Urbani virus infected BALB/c and IL28Ra−/− mice were harvested at 2, 4, 7 and 21 days post-infection. We found no difference between the titers of virus in the lungs of BALB/c and IL28Ra−/− mice (Figure 5B). Lungs from infected mice were analyzed by H&E staining. Urbani virus infection of IL28Ra−/− mice produced mild inflammation that peaked at day 2 and resolved by day 7 post-infection; replicating the pattern seen in BALB/c mice (Figure S4A).
rMA15 virus infection caused >20% weight loss and/or death by day 4-5 post-infection in BALB/c and similar findings in IL28Ra−/− mice (data not shown); both lost ∼20% weight by day 4 post-infection. Lungs were harvested at days 2 and 4 post-infection and no differences in virus titers were observed (Figure 5B). Both mouse strains showed peak virus titers at day 2 with a 2-3 log decrease by day 4. Lung pathology was analyzed by H&E staining. Both BALB/c and IL28Ra−/− mice showed similar high levels of inflammation at 2 and 4 days post-infection, but remarkably similar pathologic outcomes (Figure S4B). Taken together, we found no role for IFNλ in protection from infection with either the Urbani or rMA15 strains of SARS-CoV.
BALB/c and 129 WT mice respond differently to rMA15 virus infection; rMA15 is lethal in BALB/c mice but only causes transient weight loss in 129/Sv mice. To analyze the effects of knocking out both Types I and III IFN signaling pathways, we treated IFNAR1−/− mice (on the 129 background) with neutralizing antibodies to IL28Ra, the receptor used by IFNλ. Mice were injected with 100 µg of anti-IL28Ra antibody at days -1, 1, 3, 5 and 7 days post-infection as described in the literature[37]. We found no difference in weight loss or pathogenesis of rMA15 in these mice compared to mice injected with PBS (Figure 6). Mice showed 15% weight loss by day 4 post-infection but recovered by day 9, ending at their starting weight. Lungs were analyzed at days 2, 4 and 9 post-infection and showed no difference in pathology compared to IFNAR1−/− mice (data not shown). Both groups of mice showed epithelial cell denudation at 2 day post-infection and repair and clearance by day 9. Virus titers were only slightly increased over PBS injected mice but the difference was not statistically significant. This suggests that the inhibition of both Type I and Type III signaling in mice does not increase the pathogenesis of the rMA15 virus and neither protect 129 mice from disease.
SARS-CoV infection in humans rapidly progressed from an atypical pneumonia, to acute phase diffuse alveolar damage and ARDS during the first 10 days of acute lung injury. In many patients this is followed by the development of an organizing phase DAD after virus clearance. Both pathologies were associated with severe clinical outcomes and death, especially prominent in the elderly. The molecular mechanisms and virus-host signaling networks that regulate these progressive end stage lung diseases are unknown but are of considerable importance, given the global disease burden associated with them. Previous studies in our laboratory have demonstrated that aged mice infected with recombinant viruses encoding S glycoproteins from early phase or zoonotic SARS-CoV strains developed ARDS, characterized by hyaline membranes and DAD. In this report, we show that STAT1 deficient mice are especially prone to the development of organizing phase DAD.
The innate immune system plays a central role in regulating early host responses to virus infection and promoting adaptive immune responses. The Type I, II and III IFNs are typically produced by different cell populations and use distinct membrane bound receptors to gain entry into cells; Type I uses IFNAR1, Type II uses IFNGR and Type III uses IL28Ra/IL10Rb. However, all three share a common cytoplasmic signaling protein, called STAT1, that is translocated to the nucleus and induces expression of multiple, overlapping IFN regulated genes (ISGs)[38]. Deletion of any or all of the signaling components involved in the STAT1 signaling pathway diminishes the innate immune response to pathogens and increases susceptibility to several bacterial and viral agents. Mice lacking IFN receptors show increased susceptibility to West Nile[9], influenza[12],[39], Ebola[11], Friend virus[40], RSV[7],[8] and Poliovirus[41] (reviewed in [42]). Moreover, the same phenotypes are noted in STAT1 −/− mice, demonstrating a key role of the entire innate immune pathway in regulating disease severity, viral titers, and pathology[10]. In stark contrast, we demonstrate the paradoxical finding that SARS-CoV Urbani and rMA15 viruses induce severe end stage lung disease by a STAT1 dependent mechanism that is independent of IFN receptor type I, II and III signaling. The data point to a novel mechanism by which STAT1 function regulates disease severity in the lung following SARS-CoV induced acute lung injury.
In contrast to the existing paradigm, SARS-CoV infection is successfully controlled and cleared in IFNAR1−/−, IFNGR−/−, IFNAGR−/− and IFNLR deficient mice while deletion of STAT1 leads to increased virus replication, morbidity and mortality following infection with either the human epidemic strain of SARS-CoV (Urbani) or a mouse adapted strain (rMA15). These results suggest a novel mechanism of STAT1 regulation of severe end stage lung disease following SARS-CoV infection that is independent of it's roles in IFN signaling and ISG expression. Although it is possible that cooperative combinations of several IFNs (IFNα/β, IFNγ, IFNλ) that signal through STAT1 are required to regulate SARS-CoV infection, we feel that the possibility exists that STAT1's role in controlling cell proliferation and wound healing may be the base cause of the increased disease seen in STAT1−/− mice. Nevertheless, the development of triple knockouts affecting all 3 IFN receptors would address this mechanistic possibility.
STAT1 functions a key gatekeeper in mediating IFNα/β, IFNγ and IFNλ signaling into the nucleus to induce overlapping but distinct ISGs. Less well appreciated in viral pathogenesis studies, STAT1 also plays key roles in cell cycle arrest and cell proliferation[43]–[45]. Thus, STAT1 defects may augment viral lung disease by several potential mechanisms. STAT1 was shown to be an important controller of tumor formation in several types of cancers including lung, colon, pancreatic and brain cancers[46]–[48] and its role in cell proliferation has been studied in the context of pulmonary fibrosis[43]. STAT1−/− mouse fibroblasts showed increased proliferation when exposed to growth factors compared to WT mouse fibroblasts. Additionally, STAT1−/− mice demonstrated a greater susceptibility to chemically induced pulmonary fibrosis via bleomycin treatment. These data suggested that outside of the innate immune response, STAT1 may function as a key regulator of cell proliferation and of the wound healing response[49]. Our data also suggest that STAT1 may regulate the wound healing response following acute lung injury associated with viral infection, similar to its cell cycle regulation seen in other models of disease.
Lungs of STAT1−/− mice infected with rMA15 virus progressed to an early stage pulmonary fibrosis-like disease in 9 days. The acute lung injury (ALI) seen in these lungs approximates the damage seen in ARDS that was often seen in severe cases of SARS in humans[50], especially in the elderly. We found perivascular cuffing, collapse of alveolar parenchyma, invasion and amplification of macrophages, neutrophils, eosinophils and fibroblasts and most importantly, extensive fibrin deposition throughout the lungs. These pathological findings mirror those seen in ALI, pulmonary fibrosis and ARDS.
SARS-CoV, like many highly pathogenic viruses, expresses several proteins that antagonize the IFN sensing and amplification network. SARS-CoV ORF6 blocks STAT1 nuclear import[18], PLP blocks IRF3 activation[51], NSP7[51], NSP15[51], ORF3b and N have been shown to be IFN antagonists as well[19]. Importantly, these antagonists only function in the context of SARS-CoV infection within discrete permissive cells. This suggests that in infected cells, the multiple pathways that inhibit IFN signaling may create essentially a STAT−/− environment which may contribute to the further pathology seen during disease. As has been described with influenza, it is likely that a cytokine storm significantly contributes to increased pathogenic outcomes by targeting non-infected cells. The loss of STAT1 in other cells of the knockout mouse may contribute as we have described here, to loss of wound healing control and induction of fibrosis, leading to the development of the lethal disease state after SARS-CoV infection.
The molecular mechanisms governing SARS-CoV pathogenesis have only recently begun to be evaluated using mice with targeted genetic defects challenged with the rMA15 virus. Sheahan et. al. identified a role for Myd88, an adapter protein of TLR signaling, and RAG1, necessary for antibody production, in protection from infection and disease. Myd88 −/− mice showed an enhanced susceptibility to rMA15 virus infection in C57/B6 mice[26] and RAG1−/− mice showed no increased morbidity and mortality from rMA15 virus infection, viral replication in the lungs was detectable through 28 days post-infection[26]. These data suggest that an intricate balance between the innate and adaptive immune response is necessary for protection from SARS-CoV infection. We are currently working on understanding how these two processes interact in the host. Using proteomic and microarray analyses Cameron et al[22] showed that individuals who survived SARS-CoV infection had controlled innate immune, ISG and cytokine responses while individuals who progressed to severe disease demonstrated an uncontrolled innate immune response, with high levels of ISG and immunoglobulin expression, increased cytokine responses and poor antibody responses to the spike protein. The cause of this lack of regulation is not understood.
Infection with the epidemic (Urbani) and mouse adapted strains of SARS-CoV caused increased pathology in the STAT1−/− mice. Thus, it seems likely that the mouse adapted mutations in rMA15 enhanced the intrinsic pathogenic properties of SARS-CoV to produce severe end stage lung disease in the mice. However, SARS-CoV Urbani spreads from the respiratory tract into the spleen and liver of STAT1−/− mice. This suggests that although the damage may have been induced by viral infection, the pathology results from the host's response to the infection, though it is not clear whether this represents a normal pattern of spread or whether mutations have evolved in the virus that promote spread and distribution to other organs. Similar findings were seen in autopsy samples from individuals who died from SARS. Airways were intact and regenerated while acute cases showed spread to the outer pulmonary parenchyma[52].
The similarities between STAT1−/− mice infected with SARS-CoV and elderly individuals infected with SARS-CoV during the epidemic are intriguing. Recent work has shown that in cells from aged hosts, STAT1 signaling cascades are less responsive to stimuli; STAT1 signaling in aged macrophages were hypo-responsive to IFNγ[49]. As observed in our study, IFNγ expression increases substantially in STAT1−/− mice. We hypothesize that this may be in response to a lack of negative feedback via the STAT1 signaling pathway. During the SARS epidemic, aged individuals were the cohort with most severe disease and highest mortality rates[53]. We propose that altered STAT1 signaling in aged individuals may have lead to increased susceptibility to severe disease.
We cannot rule out the possibility that all 3 types of IFN receptors are redundant for SARS-CoV mediated disease and that only by deleting all 3 types of receptors will be observe an increase in pathogenesis. Proof must await the availability of a mouse strain containing the deletion of all three receptors. However we find that the type of pathology produced in STAT1−/− mice by both WT and rMA15 SARS-CoV demonstrates a role for STAT1 in non-innate immune processes that resemble those produced in severely infected SARS-CoV patients.
We propose a new potential pathway by which STAT1 regulates end stage lung disease following viral infection. We hypothesize that the increased susceptibility of STAT1−/− mice results from different roles of STAT1 in the cell. First, loss of STAT1 results in a deficient IFN response that results in higher titers of virus in the lungs. Secondly, loss of STAT1 allows for unregulated cell proliferation in response to the innate immune response, causing enhanced damage to the lungs and death of the animals. Our findings also point to an increasing role of the cell damage response to viral infection that can be potential targets for therapy in highly pathogenic respiratory infections.
All mice in this study were treated following IACUC guidelines. For infection mice were pre-treated with Ketamine and Xylazine as an anesthetic. Mice were euthanized if their weight dropped below 20% of their starting weight or if clinical symptoms warranted it per our IACUC approvals. Animal housing and care and experimental protocols were in accordance with all UNC-Chapel Hill Institutional Animal Care and Use Committee guidelines or NIH guidelines, depending where the experiments were performed.
Vero E6 cells were grown in MEM (Invitrogen, Carlsbad, CA) supplemented with 10% FCII (Hyclone, South Logan, UT) and gentamicin/kanamycin (Gibco-BRL). Stocks of the biological SARS-CoV (Urbani), recombinant SARS-CoV (icSARS) and mouse-adapted SARS-CoV (rMA15) were propagated and titered on Vero or Vero E6 cells and cryopreserved at −80°C until use as described [30]. All experiments with infectious virus were performed in a Class II biological safety cabinet in a certified biosafety level 3 laboratory containing redundant exhaust fans with personnel wearing protective equipment including Tyvek suits, hoods, and HEPA-filtered powered air-purifying respirators (PAPRs) as described [30].
129S6/SvEv wildtype and STAT1−/− mice (catalog number 002045-M-F) were obtained from Taconic Farms (Germantown, NY). For the mouse adapted SARS-CoV infections, Type I IFN receptor deficient (IFN alpha/beta receptor) (IFNAR1−/−) mice were bred in at the UNC mouse facility (Chapel Hill, North Carolina). Type II IFN receptor deficient (IFN gamma receptor) (IFNGR−/−) mice (stock number 002702) were purchased from The Jackson Laboratories (Bar Harbor, ME). For the Urbani virus infections, IFNAR1−/− mice were obtained as a gift from Dr. Joan Durbin at Ohio State University and IFN alpha/beta/gamma receptor double knockout (IFNAGR−/−) mice were bred at the NIH animal facility (Bethesda, MD). IFN-lambda receptor knockout mice (IL28Ra−/−, Zymogen, Seattle, Washington) were bred at the UNC Chapel Hill animal facility. Animal housing and care and experimental protocols were in accordance with all UNC-Chapel Hill Institutional Animal Care and Use Committee guidelines or NIH guidelines, depending where the experiments were performed. All animal studies were conducted in Animal Biosafety Level 3 laboratories using SealSafe Hepa-filtered caging and personnel wore personal protective equipment, including Tyvek suits and hoods and positive pressure HEPA-filtered air respirators. 10 week old mice were anesthetized with a mixture of ketamine/xylazine or isoflurane and intranasally infected with either PBS alone or 105 pfu/50 µl rMA15 or the recombinant or biological epidemic virus, icSARS or Urbani, in PBS (Invitrogen, Carlsbad, CA). Mice were monitored at 24 h intervals for virus-induced morbidity and mortality. Subsets of mice were euthanized at days 2, 5, and 9 post-infection (dpi) for characterization of rMA15 infection, while the less pathogenic Urbani virus infected animals were sampled on days 2, 3, 5, 9, 15, 22 and 29 post-infection. All tissues were analyzed for histopathology changes and for viral titers.
To quantify the amount of infectious virus in tissues from rMA15 infection, each organ was weighed, placed in 0.5 ml DPBS, homogenized, and titered via plaque assay on Vero E6 cells as previously described [26]. For Urbani infection, supernatants of 10% (w/v) lung homogenates were prepared and titrated on Vero cell monolayers in 24- and 96-well plates as previously described [40]. Virus titers are expressed as TCID50 per g of tissue. The lower limit of detection was 101.5 TCID50/g.
Lung tissues were fixed in PBS/4% para-formaldehyde, pH 7.3, embedded in paraffin, and 5 µm sections were prepared by the UNC histopathology core facility. To determine the extent of inflammation, sections were stained with hematoxylin and eosin (H & E) and scored in a blinded manner.
Supernatants of 20% (weight/volume) lung homogenates were used for detection of cytokines and chemokines using BD CBA kits (BD Biosciences) according to the manufacturer's instructions. The lower limit of detection for each protein is included in the kit protocol.
35S-UTP-labeled riboprobes specific to the N gene of SARS-CoV (Urbani) or to the EBER2 gene from Epstein-Barr virus (negative control probe) were generated with an SP6-specific MAXIscript in vitro transcription kit (Ambion) and in situ hybridization was performed as described previously[26]. Briefly, deparaffinized tissue sections were hybridized with 5×104 cpm/µl of 35S-labeled riboprobes overnight. Tissues were washed, dehydrated through graded ethanol, coated in NTB autoradiography emulsion (Kodak), and incubated at −80°C for 7 days. Following development, sections were counterstained with hematoxylin and silver grain deposition was analyzed by light microscopy. rMA15-specific signal was determined by comparing silver grain deposition on parallel sections hybridized with the 35S-labeled riboprobe complementary for the EBER2 gene of Epstein-Barr virus.
Lungs from mock or SARS-CoV infected mice were removed and homogenized directly in 1 ml of Trizol reagent (Invitrogen) and total RNA was isolated following manufacturer's instructions. Complementary cDNA was generated from 1 µg of total RNA using 250 ng random primers (Invitrogen) and Superscript II reverse transcriptase (Invitrogen). Real-time PCR experiments were performed using Taqman gene expression assays and an ABI Prism 7300 (Applied Biosystems). 18S rRNA was used as an endogenous control to use for normalization in all assays. The relative fold induction of amplified mRNA were detected using the Ct method. Taqman primer sets used were 18S (#Hs03003631 Applied Biosystems), IFNγ (Mm00801778 Applied Biosystems), TNFa (Mm99999068 Applied Biosystems), MCP1 (Mm99999056 Applied Biosystems), IFNB (Mm00439552 Applied Biosystems), IFNA4 (Mm00833969 Applied Biosystems), IL28B (Mm00663660 Applied Biosystems), IL18 (Mm00434225 Applied Biosystems), IRF1 (Mm01288574 Applied Biosystems) and OAS1 (Mm00449297).
Mice were inoculated as described above, sacrificed by exsanguination at 2 and 4 days post-infection, and lungs were perfused via cardiac puncture with 1× PBS. Lungs were dissected, minced, and incubated for 2 hrs with vigorous shaking at 37°C in digestion buffer [RPMI, 10% FBS, 15 mM HEPES, 2.5 mg/ml collagenase A (Roche), 1.7 mg/ml DNase I (Sigma)]. Cells were passed through a 40 micron cell strainer, resuspended in RPMI media, layered on 5 ml lympholyte-M (Cedarlane), and centrifuged 30 min at 2500 rpm. Cells were collected, washed in wash buffer (1× HBSS, 15 mM HEPES), and total viable cells were determined by trypan blue exclusion. Isolated cells were incubated with anti-mouse FcγRII/III (2.4G2; BD Pharmingen) for 20 min on ice and then stained in FACS staining buffer (1× HBSS, 1% FBS, 2% normal rabbit serum) with the following antibodies from eBioscience: anti-F4/80-FITC, anti-Gr-1-PE, anti-CD11b-APC, anti-CD11c-PE, anti-Ly-6C-FITC, anti-CD3-FITC, anti-CD8-APC, anti-CD4-PerCP, and anti-NK1.1-PE. Cells were fixed overnight in 2% paraformaldehyde and analyzed on a Cyan cytometer (Dako) using Summit software.
Percent starting weights, viral titers and inflammatory cell numbers were evaluated for statistically significant differences by the non-parametric Mann-Whitney test within GraphPad Prism or unpaired t-tests using GraphPad InStat3 software. P values of ≤0.05 were considered significant.
|
10.1371/journal.pgen.1005624 | Oriented Cell Division in the C. elegans Embryo Is Coordinated by G-Protein Signaling Dependent on the Adhesion GPCR LAT-1 | Orientation of spindles and cell division planes during development of many species ensures that correct cell-cell contacts are established, which is vital for proper tissue formation. This is a tightly regulated process involving a complex interplay of various signals. The molecular mechanisms underlying several of these pathways are still incompletely understood. Here, we identify the signaling cascade of the C. elegans latrophilin homolog LAT-1, an essential player in the coordination of anterior-posterior spindle orientation during the fourth round of embryonic cell division. We show that the receptor mediates a G protein-signaling pathway revealing that G-protein signaling in oriented cell division is not solely GPCR-independent. Genetic analyses showed that through the interaction with a Gs protein LAT-1 elevates intracellular cyclic AMP (cAMP) levels in the C. elegans embryo. Stimulation of this G-protein cascade in lat-1 null mutant nematodes is sufficient to orient spindles and cell division planes in the embryo in the correct direction. Finally, we demonstrate that LAT-1 is activated by an intramolecular agonist to trigger this cascade. Our data support a model in which a novel, GPCR-dependent G protein-signaling cascade mediated by LAT-1 controls alignment of cell division planes in an anterior-posterior direction via a metabotropic Gs-protein/adenylyl cyclase pathway by regulating intracellular cAMP levels.
| During embryogenesis an entire organism develops from a single cell. This process is vital for the formation of life, thus cell division occurs with a very distinct orientation and pattern that is tightly controlled by several signaling pathways. The mechanisms underlying these pathways are complex and not yet fully understood. In the roundworm Caenorhabditis elegans, a common genetic model, the patterns and orientations in which cells divide in the embryo have been well characterized offering an ideal model to study the molecular mechanisms involved. Here, we show that the signal mediated by the adhesion G protein-coupled receptor LAT-1 is based on cAMP. This second messenger is essential for the orientation of distinct cell division planes in the early embryo. Studies based on a lat-1 knockout mutant reveal that LAT-1 signaling affects the levels of the second messenger cAMP in the cells via a specific G protein. Thereby the receptor is activated by an intrinsic sequence. This pathway is the first one clearly shown to involve a G protein-coupled receptor-dependent G-protein signal in orientation of embryonic cell division, offering a novel level of regulation of this process among other described pathways.
| Spindle and cleavage plane orientation play a central role in many aspects of development as well as homeostasis of organs and organisms. Alignment of spindles during cell division is achieved by interactions of the spindle apparatus with the cell cortex. The machinery linking tubulin to the cortex and supplying force to move the spindle is well characterized [1–3]. However, the molecular mechanisms balancing forces in a specific direction are complex and less well understood. In the early Caenorhabditis elegans embryo a large network of only partially characterized signaling pathways including Wnt proteins [4] and PAR proteins [5–7] is engaged in the control of spindle orientations during asymmetric and symmetric cell divisions. These pathways form the basis for the invariant embryonic cell lineage of the nematode with highly reproducible cleavage planes and axes in the early embryo. A sequence of directed cell divisions establishes the particular geometry of early blastomeres, eventually creating an 8-cell stage embryo which contains AB4 descendants after the third round of cell cleavage [8]. At this stage, the Wnt/Frizzled (Fz) is involved in spindle rotation of one of the dividing AB descendants to ensure the proper contact of its daughter cells to neighboring blastomeres [9–11]. In the subsequent round of cell division, the Notch pathway induces the left-right asymmetry of the embryo via specific contacts of two of the then eight AB descendants [12,13]. Recently, we have shown that in addition to MOM-2/Wnt and MOM-5/Fz, the latrophilin homolog LAT-1 is a novel receptor required for the coordination of spindles and thus, division plane orientation at the 8-cell stage in the C. elegans embryo [14]. Some evidence even indicates a function of LAT-1 in parallel to components of the Wnt pathway [14]. In lat-1(ok1465) homozygous nematodes the spindle of ABal, one of the AB4 cells, is turned almost perpendicular to the anterior-posterior axis in which the cell normally divides. This tilted orientation causes not only one but both daughter cells to touch the neighboring blastomere MS [14] resulting in embryonic lethality as a consequence of incorrect cell-cell contacts. Thus, LAT-1 is involved in orienting cleavage planes in an anterior-posterior direction in descendants of the ABal blastomere. However, it remained elusive how the latrophilin homolog exerts this function.
Latrophilins were initially identified as targets for latrotoxin, a component of the black widow spider´s (Latrodectus mactans) toxin [15] and are described as modulators of neurotransmitter release [16,17]. They belong to the class of adhesion G protein-coupled receptors (aGPCRs), a family of seven transmembrane (7TM) receptors with emerging roles in cell and tissue polarity. One of the best understood members of this class, flamingo/CELSR, has been shown to be involved in the planar cell polarity pathway in mice as well as Drosophila melanogaster [18,19]. aGPCRs form the second largest class of GPCRs. Despite their essential physiological functions, especially in development and neurobiology [18,20–22], the signaling mechanisms of the majority of these receptors remain poorly understood, and for most family members agonists and downstream effectors are unknown [23,24]. Our previous work showed that LAT-1 mediates two distinct signals, one transduced via the 7TM and the C terminus, while the other requires only the extracellular N terminus containing the GPCR-autoproteolysis inducing (GAIN) domain and its integral motif, the GPCR proteolytic site (GPS) [25], which are characteristic for many aGPCRs [26,27]. During embryonic development, LAT-1 conveys its signal cell autonomously via the 7TM and the C terminus raising essential questions regarding the molecular details and effectors of the signaling cascades triggered by LAT-1.
Here, we have identified the mechanisms of LAT-1 signaling required to mediate the coordination of anterior-posterior tissue polarity in the early C. elegans embryo. We show that a Gs protein-signaling cascade is the key pathway. Following activation by a tethered agonist, LAT-1 elevates levels of the second messenger cyclic AMP (cAMP) via interaction with a Gs protein. By increasing intracellular cAMP levels LAT-1 controls spindle orientation in the dividing cell. We discuss how polarity is realized by the respective cascade and the non-polar cAMP signal.
Structure-function analyses have previously shown that the correct alignment of cell division planes of some AB descendants in the early C. elegans embryo is mediated by the latrophilin homolog LAT-1 via the 7TM-dependent signaling mode of the receptor [14,25]. This could be indicative of a classical GPCR signal via heterotrimeric G proteins. To investigate this hypothesis, we heterologously expressed wild-type lat-1 in COS-7 cells (Fig 1A and S1 Fig) and tested for functional G-protein coupling of the receptor to the three major G proteins Gs, Gq and Gi, which are highly conserved between metazoan species [28]. As no interaction partner of LAT-1 has been described which is capable of triggering G protein-mediated signaling of the receptor, its coupling abilities were determined by measuring its basal activity. In the absence of agonists an equilibrium between inactive and active GPCR conformation exists, with only a few receptors residing in the active state [29]. GPCR overexpression increases the amount of receptors in general and thus, in each state. At a certain threshold basal activity of the GPCRs in the active state can be detected and signaling pathways measured, which are normally only activated upon agonist stimulation. The signaling abilities of orphan GPCRs have been frequently identified by taking advantage of this capacity [30,31]. Using this well-established system we first tested the involvement of the Gs protein/adenylyl cyclase pathway in LAT-1 signal transduction by measuring the formation of cAMP. Cells overexpressing lat-1 showed a receptor amount-dependent increase of cAMP levels with a maximum of 1.4-fold (Fig 1B), suggesting that LAT-1 transduces its signal via Gs proteins. For the characterization of the G protein-coupling abilities of LAT-1 to Gi and Gq proteins we performed inositol trisphosphate (IP) accumulation assays. No increase of cellular IP levels was detected (Fig 1C), indicating that LAT-1 does not couple to Gq proteins under basal conditions in this assay. It was reported earlier that a chimera, in which the C-terminal 4 amino acids of Gαi (Gαqi4) are substituted with the corresponding ones from a Gαq subunit, reroutes the intracellular signals towards a Gq pathway modulating IP concentration [32]. Co-expression of lat-1 with this chimera did not cause an increase of IP concentration, indicating that Gi-protein coupling appears also improbable (Fig 1C). Applying the same concept by co-expressing a Gαqs4 chimera with lat-1, which redirects the Gs protein signal towards a Gq-signaling cascade, we detected an increase in basal IP levels when compared to COS-7 cells transfected with empty vector (Fig 1C). By being able to reroute the Gs protein signal we verified that the LAT-1-mediated elevation of cAMP levels was due to Gs-protein coupling and not a result of a secondary effect or Gβγ signaling. Therefore, we conclude that LAT-1 very likely activates the Gs protein/adenylyl cyclase signaling pathway.
The question which of the 21 Gα subunits in C. elegans is activated by LAT-1 is difficult to address as for many of them no effectors or signals are described. One likely candidate is GSA-1, the closest homolog of Gαs in C. elegans displaying 66% identity to mammalian Gαs proteins [33], which is also expressed in embryos [34]. gsa-1(pk75) homozygous animals survive embryogenesis but arrest in larval development [34], which may be explained by a maternal contribution of the embryonic functions. To elucidate a possible interaction of lat-1 with gsa-1 in early embryonic development we knocked down maternally and zygotically contributed gsa-1 activity using RNA interference (RNAi).
In embryos homozygous for lat-1(ok1465), subsequently referred to as lat-1, the ABal division plane was tilted towards a position almost perpendicular to the anterior-posterior axis (90.3°±17.9°, means ± SD, n = 18, p < 0.001 Fig 2B and 2E) whereas the wild-type orientation was more oblique (128.2°± 8°, means ± SD, n = 14; angles measured towards the posterior; Fig 2A and 2E).
When knocking down gsa-1 in wild-type nematodes by feeding dsRNA to young adult hermaphrodites we observed no embryonic lethality in the offspring. However, the embryos still displayed a minor turning of the ABal spindle towards the direction typical for lat-1 (123.5° ± 7.7°, means ± SD, n = 11; Fig 2E). It appeared that RNAi affected the progeny (F1) more drastically by rendering the adult hermaphrodites almost sterile. Only few embryos could be recovered as most hermaphrodites did not contain any embryo, some, however, one to five. In these second generation (F2) embryos, the orientation of the ABal division plane was more similar to that of lat-1 mutants (115.3° ± 12.2°, means ± SD, n = 12; Fig 2C and 2E).
In contrast to F1 embryos, the cleavage direction of F2 embryos was significantly different from wild-type embryos (p < 0.001). Although some F2 embryos showed a cleavage angle of 87°, which is typical for lat-1 (Fig 2E), no entire similarity to lat-1 was reached. The fact that the RNAi effect is only pronounced in the F2 generation suggests that either inactivation of the gsa-1 mRNA is very inefficient and/or slow or that oocytes already contain a sufficient amount of maternally contributed GSA-1 to rescue embryos in the absence of endogenous mRNA. Analyses of mRNA levels by qPCR revealed a 3- to 6-fold reduction of gsa-1 transcript in all RNAi-treated samples compared to untreated levels (S2A Fig). Protein levels of GSA-1 are significantly reduced 36 hours after onset of RNAi treatment, albeit not fully depleted (S2B and S2C Fig) suggesting that GSA-1 protein is still available in the cells but at lower levels than in wild-type animals.
Although the cleavage direction of gsa-1 RNAi embryos is the same as in lat-1 mutants this is not a stringent indication that both proteins function in the same pathway, they could still function in parallel. As lat-1(ok1465) is a null mutant, one should expect that an additional inactivation of GSA-1 should not have an effect on the cleavage direction of ABal, if they act in the same pathway. However, a synergistic effect should be visible if they work in different pathways. To analyze whether LAT-1 and GSA-1 function sequentially or independently lat-1 mutants were fed with bacteria expressing dsRNA for the gsa-1 sequence, which yielded a cleavage direction of ABal in F2 embryos that was not significantly altered compared to lat-1 embryos (97.7° ± 8.8°, means ± SD, n = 13, p > 0.1; Fig 2D and 2E). This is consistent with the notion that GSA-1 is the G protein downstream of LAT-1.
To investigate the physiological relevance of the LAT-1-dependent Gs protein-mediated cAMP signal in C. elegans, we first measured cAMP levels in embryos. Interestingly, the cAMP level in a population of lat-1 embryos was significantly reduced (0.26 nM) compared with wild-type embryos (0.52 nM) (Fig 2F). Next, we tested if an elevation of cAMP rescues lethality of lat-1 worms, a consequence of the tilted ABal division plane. Consistent with the maternal and zygotic requirement of LAT-1 [14] we incubated L4 larvae and subsequently developing embryos with different compounds promoting or mimicking elevation of cAMP levels: the adenylyl cyclase activator forskolin, the phosphodiesterase inhibitor 3-isobutyl-1-methylxanthine (IBMX) and the stable cAMP analogue 8-bromoadenosine 3′,5′-cyclic monophosphate (8-Br-cAMP). Treatment with any of these compounds led to an amelioration of embryonic lethality with forskolin having the strongest effect by increasing the survival rate from 33% to 73% (Fig 2G). As forskolin activates adenylyl cyclases and thus, potentially exhibits toxicity we tested various concentrations of forskolin and found that 80 μM had an optimal effect whereas higher concentrations were detrimental on C. elegans. As shown in Fig 2H, forskolin elevated cAMP in wild-type and in lat-1 hermaphrodites. Interestingly, we did not observe any involvement of cAMP in the LAT-1 function depending exclusively on the N terminus. As loss of this function leads to reduced fertility, we investigated the effect of elevated cAMP levels on brood size of lat-1 mutants. We did not detect any rescue of reduced brood size upon treatment with forskolin (S3A Fig), independently of time and duration of drug application (S3B Fig), suggesting that the 7TM-independent function of LAT-1 involved in fertility does not rely on a Gs/adenylyl cyclase-mediated cAMP signal.
We next investigated if forskolin corrects the defective ABal spindle orientation observed in lat-1 mutant embryos (95°± 7°, means ± SD, n = 12, p < 0.0005; Fig 2J and 2N) into the wild-type direction (117° ± 9°, means ± SD, n = 13; angle measured towards the posterior; Fig 2I and 2N). Upon application of 80 μM forskolin ABal spindles in lat-1 mutants returned to the oblique position (114° ± 10°, means ± SD, n = 14, p > 0.5) (Fig 2K–2N).
As a result of faulty cleavage plane orientation in lat-1 mutant embryos both daughters, ABalp and ABala, retain equal contact to MS after division [14] (Fig 2J) whereas in wild-type embryos only the posterior daughter ABalp remains in contact with the MS blastomere. The anterior daughter ABala is displaced to the most anterior position in the embryo (Fig 2I). The arrangement of cells at the 12-cell stage does not solely depend on the cleavage direction of ABal but also on that of ABar, which under specific circumstances may push ABala away from MS, whose position also varies in embryos. Therefore, we investigated the contact lengths of the ABala and ABalp blastomeres with MS. The ratio of the two lengths was used as a measure for cell position. In wild-type embryos ABala normally does not contact MS (ratio of 0.00 ± 0.00), whereas in lat-1 mutant embryos it is shifted to 0.72 ± 0.03 (Fig 2O). Forskolin strongly reduced the contact of ABala to MS. Embryos in which the contact length ratio of ABala-MS to ABalp-MS was smaller than 0.45, mostly survive. However, due to potential independent roles of LAT-1 in later processes and detrimental effects of forskolin, occasionally embryos with ratios smaller than this cut-off still died. These observations suggest that suppression of lethality in lat-1 mutants generally occurs by correcting the aberrant cleavage direction of the ABal blastomere in the mutant. It appears surprising that forskolin, which acts on all adenylyl cyclases, has such a specific effect. Initially we suspected that the drug may universally randomize spindle orientation and thus, ABala occasionally would not touch MS. However, the standard deviation for the angle of ABal cleavage is ± 9° in untreated wild-type embryos and ± 10.2° after forskolin treatment, indicating that the drug does not randomize cleavage directions. Thus, the suppression of the lat-1 phenotype by forskolin is very specific, suggesting that by activating adenylyl cyclases and subsequently raising cellular cAMP concentrations, it specifically mimics the process normally controlled by LAT-1.
Further, cAMP being the central element in this process is also supported by the fact that decreasing its levels by treating very early wild-type embryos with the adenylyl cyclase inhibitor 2',5'-dideoxyadenosine (ddA) leads to a specific partial phenocopy of lat-1 mutants resulting in embryonic lethality (S4 Fig).
Taken together, these results indicate that LAT-1 implements signaling via cAMP in vivo. As the lat-1 phenotype is rescued by elevation of cAMP which is not restricted to a certain cellular compartment the division plane orientation in an anterior-posterior direction is likely to be molecularly realized by components downstream of this second messenger.
We next asked how LAT-1 is activated to trigger the signaling cascade to ensure correct cell division plane orientation. Previously, we have postulated that an interaction between the GPS and 7TM domain is essential for LAT-1 activation [25]. Very recently, Liebscher et al. have shown for the aGPCR GPR126 in zebrafish that the sequence immediately C-terminal of the cleavage site in the GPS (the Stachel sequence) acts as an internal tethered agonist to activate GPR126 [35]. We hypothesized that in LAT-1 a similar sequence would also have agonistic properties. To test this hypothesis, we analyzed peptides of varying lengths comprising the sequence C-terminal of the GPS cleavage site (Fig 3A) for their ability to activate LAT-1 in vitro by measuring cAMP levels. Peptides of 12 and 13 amino acids (CP12, CP13) increased cAMP levels in lat-1-transfected COS-7 cells above basal levels, with CP12 displaying the highest agonistic activity by increasing the concentration 1.6-fold (Fig 3B). In contrast, a CP12-derived peptide with positions +1 (T530 in the full length receptor) and +3 (F532 in the full length receptor) mutated to alanine did not display any activity (CMP12) demonstrating that agonistic activity is highly sequence-specific (Fig 3B). The agonistic peptides were able to activate a chimeric LAT-1 with the extracellular N terminus exchanged for that of the rat muscarinic M3 acetylcholine receptor (M3R ECD::LAT-1, S5 Fig) suggesting that the sequence directly interacts with the 7TM.
As our initial experiments tested G protein-coupling abilities of LAT-1 only under non-stimulated conditions (Fig 1C) we performed these assays in the presence of the agonistic peptide CP12 to elucidate coupling to Gi and Gq proteins. However, we were unable to obtain any results with these chimeric G proteins. Due to rather low LAT-1 expression levels compared to mammalian Gi or Gq-coupled receptors we cannot fully exclude the possibility of coupling to other G protein families additionally to Gs proteins.
We next tested whether activation by a tethered agonist is conserved between species. In mammals, three latrophilin homologs exist. LPHN1 (ADGRL1) shows the highest conservation to LAT-1 overall and within the potential agonistic sequence (Fig 3A). Like LAT-1, this receptor is expressed in COS-7 cells (Fig 3C) and slightly activates the Gs-signaling pathway, but not a Gq- or Gi-signaling cascade in the absence of an agonist (Fig 3D and 3E). Further analyses showed that, similar to C. elegans LAT-1, rat LPHN1 was activated by a peptide representing the sequence C-terminal of the cleavage site. A peptide of 13 amino acids length (RP13) displayed the most efficient agonistic properties and increased basal cAMP levels 2.3-fold (Fig 3F). However, we did not observe any cross-activation of LAT-1 with the rat Stachel sequence and vice versa (Fig 3B and 3F). We also tested if the agonistic peptides are able to trigger a Gq or Gi pathway via rat LPHN1. Upon stimulation with the agonistic peptide RP13 coupling of the receptor to a Gq protein was observed in an IP accumulation assay (S6 Fig). These data suggest that the activation mechanism involving the Stachel sequence is conserved between species, but the implementation of the mechanism is sequence-specific, at least between distantly related latrophilin orthologs.
To assess the agonistic properties of the Stachel sequence in vivo and its impact on LAT-1 signaling in embryonic development we employed rescue experiments utilizing lat-1 mutant animals expressing a chimeric receptor. In this receptor the LAT-1 GPS and thus, the entire Stachel sequence is exchanged for the GPS of C. elegans LAT-2 (Fig 4A). This receptor has been shown to retain activity to complement the fertilization defect in lat-1 mutants, but does not rescue the tissue polarity phenotype [25], ensuring that it is devoid of any potential activation by the Stachel sequence. In this assay, soaking of hermaphrodites with the agonistic peptides CP11, CP12 and CP13 efficiently rescued the tissue polarity phenotype in the early embryo, demonstrating that all three peptides are also able to activate LAT-1 in vivo (Fig 4B). CP13 had the strongest effect (46 ± 12%) compared to untreated controls (26 ± 7%). Peptides with no agonistic activity in vitro (CP9, CP10, CP14, RP13, CMP12) did also not display any activity in vivo (Fig 4B). To test whether this effect is specific to LAT-1 signaling, we also treated lat-1 mutants with the respective peptides but did not detect any rescue (Fig 4B). We did not observe any effect of these peptides on wild-type nematodes, which might be due to the fact that receptor signaling is tightly controlled in vivo or that an increase in activity levels is not reflected in a specific phenotype. To further control for specificity we introduced mutations in the Stachel sequence of the full length LAT-1 protein (T530A and F532A, similarly to the mutations described in CMP12). Consistent with the in vitro experiments no rescue of developmental lethality was detected when expressing this construct in lat-1 mutant nematodes (Fig 4C). To ensure that the mutant protein is functional we assessed the expression of the lat-1T530A/F532A::gfp construct, which is indistinguishable from a wild-type receptor (Fig 4D). Biochemical activity was confirmed by the rescue of the reduced fertility in lat-1 mutants, therefore confirming the receptor´s 7TM-independent functionality (Fig 4D) [25]. These data show that the sequence immediately C-terminal of the GPS cleavage site in LAT-1 is an agonistic region essential for receptor activation and thus, crucial for LAT-1 signaling in embryonic development.
Oriented cell division and spindle orientation in early C. elegans embryogenesis are controlled by complex signaling pathways involving GPCRs such as the four frizzled family Wnt receptors [9,37–39] and in which the role for heterotrimeric G proteins has been firmly demonstrated in several elegant studies [40,41]. Gotta and Ahringer showed that proper spindle directionality of the cleavages in the 4-cell C. elegans embryo depends on Gß/γ and that Gα signaling is required for spindle placement in the 1-cell embryo [41]. However, there is evidence that the activity of the G proteins is modulated mostly in a GPCR-independent manner via G-protein regulators [42] and GEF proteins [43]. A GPCR-dependent and G protein-mediated signaling pathway has not been unambiguously defined despite some clear indications [44,45]. In the present study, we demonstrate that a GPCR-dependent G protein-mediated signal based on the adhesion GPCR LAT-1, which is involved in orienting spindles in an anterior-posterior direction in ABal descendants [14], is an essential mechanism for controlling oriented cell division. We provide functional evidence that LAT-1 couples to a Gαs protein which activates an adenylyl cyclase, thereby elevating cellular concentrations of the second messenger cAMP in vitro (Figs 1B and 3B). Similar coupling abilities were observed for the latrophilin ortholog rat LPHN1 (Fig 3E and 3F). consistent with analyses showing that this receptor triggers elevated cAMP levels upon treatment with α-latrotoxin in vitro [46]. In accordance with this study we also observe that LAT-1 triggers a change in basal IP levels, albeit only upon stimulation with an agonistic peptide (S6B Fig).
In vivo analyses have revealed that the increase of cAMP levels is a key signal for the anterior-posterior orientation of cleavage planes in the ABal cell in the C. elegans embryo. Treatment of lat-1 mutant nematodes with the adenylyl cyclase activator forskolin raises cAMP independent of the receptor to a point where cell division plane orientation is sufficiently restored and subsequent lethality is rescued (Fig 2H–2O). The small contact lengths between ABala and MS cells observed in some cases does not seem to have a detrimental effect on developing embryos. The incomplete reduction of contacts lengths to wild-type levels might be due to limitations in accessibility of forskolin to the embryo. Limited drug uptake may explain the partial rescue of lethality by the different compounds affecting cAMP levels (Fig 2G). However, it is also possible that lethality cannot be rescued to wild-type levels as it might be a result of different LAT-1 functions. As neither drug accessibility nor uptake rate or half-life of each compound in the organism could be resolved, we were unable to determine the exact time point and duration at which the signal is required. However, consistent with previous work demonstrating maternal and zygotic requirement of the receptor [14], treatment of L4 hermaphrodites and subsequently embryos with the respective drug was sufficient for rescue (Fig 2G).
Consistently with the putative role of LAT-1 as a regulator of cAMP, lat-1 mutant embryos display decreased levels of the second messenger. We cannot exclude that this decrease is exclusively a consequence of absence of LAT-1 signals in the cells of the early embryo. As there is evidence for the receptor to mediate effects in other cells as well [14], it is conceivable that these are also mediated via cAMP.
Combined with the functional in vitro data our analyses on rescue of lat-1 mutant defects by stimulating a Gs-mediated signal suggest that a GPCR cascade via a cAMP regulation is one essential pathway for the coordination of anterior-posterior cell division plane orientation in embryogenesis.
Our data indicate that the G protein involved in this cascade is GSA-1 (Fig 2A–2E). The major pathways involved in polarity decisions in the early embryo such as Notch/Delta and Wnt/Fz have been shown to mediate signals via routes different from G proteins [4,13]. However, the Wnt/Fz pathway component APC in C. elegans feeds into Rac [45] which supports a scenario in which spindle directions are also regulated by differential G-protein signaling. The signals mediated by these G proteins remain widely elusive, but involvement of different intracellular signals warrants a precise regulation and avoids intersection of the multitude of signaling pathways required to ensure tightly controlled oriented cell division. It could be speculated that LAT-1 contributes an additional signal via a Gs protein/adenylyl cyclase introducing a new level of regulation.
Interestingly, our data suggest that the cAMP-based signal mediating this polarity is not polar and thus, LAT-1 is not required to be asymmetrically localized prior to ABal cell division. Consistently, no asymmetrical distribution of LAT-1 has been found previously [14]. This is in contrast to asymmetric protein localization that has been described for many components of pathways involved in planar cell polarity (PCP) or Wnt/Fz signaling [3] but in accordance with some anterior-posterior tissue polarity models [47]. However, it is well possible that the signal transduced by LAT-1 to mediate an effect on polarity is polarized further downstream through effectors of cAMP such as protein kinases A (PKA) or A kinase anchoring proteins (AKAP). This effect has been shown for a cAMP-dependent PKA in the establishment of neuronal polarity [48,49]. It is also conceivable that the polarized effect of LAT-1 signaling is promoted by a temporal cue which could not be investigated in this study.
In order to transduce signals LAT-1 is activated by a tethered agonist downstream of the GPS. In vitro and in vivo analyses identified a sequence of 12 amino acids exhibiting agonistic properties (Figs 3B and 4B). These findings are in accordance with recent data on GPR126 and GPR133 which revealed similar agonistic regions, termed Stachel sequences [24]. The biological implications of both studies are intriguing as they raise the hypothesis that several aGPCR share the same mechanism of activation.
Our data suggest that the activation mechanism is evolutionary conserved as latrophilin orthologs in rat and C. elegans both display an intrinsic agonistic sequence of a similar lengths. However, the tethered agonist of one latrophilin receptor is highly sequence-specific (Fig 3F), it is not able to activate its ortholog despite 59% identity/76% similarity of both agonistic motifs. As mutations at the conserved positions +1 and +3 within the Stachel sequence are sufficient to abolish LAT-1 function this specificity is likely to be conferred by structural properties. Consistently, an exchange of the GPS for the one of the paralog LAT-2 (69% identity/87% similarity within the Stachel sequence) results in a loss of LAT-1 function in development [25]. These data are also in concordance with the study identifying the Stachel sequence of GPR126 in which no cross-activation of Stachel sequence-derived agonistic peptides from GPR126 and another aGPCR, GPR133, is observed despite a high amino acid identity among the agonistic regions [24]. However, cross-activation between certain aGPCR orthologs or closely related aGPCRs via the Stachel sequence cannot be excluded and might have interesting implications.
Interestingly, the Stachel sequence identified in both latrophilin orthologs corresponds exactly to the respective C-terminal section of the GPS. These findings support our previous structure-function analyses which provided evidence for an interaction between LAT-1 GPS and 7TM domain [25]. Future studies need to focus on the details of the activation mechanism and clarify how the interaction is induced as well as which region of the 7TM is a potential interaction site. As we have previously shown that cleavage at the GPS is not essential for LAT-1 function, a model in which the tethered agonist functions after receptor cleavage is unlikely. Conformational changes within the GPS-containing GAIN domain upon binding to extracellular proteins could be the stimulus for exposure of the sequence within the GPS. The 7TM-independent function of LAT-1 is not based on a cAMP signal. However, the separation of this function from the 7TM-dependent function is likely to be conferred by the tethered agonist.
In summary, our results show that a GPCR-dependent G protein-signaling cascade based on LAT-1 is involved in oriented cell division in the early C. elegans embryo. LAT-1 activates a Gs protein/adenylyl cyclase signaling pathway, probably via GSA-1. By regulating cAMP levels, the receptor controls coordination of anterior-posterior cleavage plane orientation in the ABal cell. The data support a model in which LAT-1 resides in an inactive state while the Stachel sequence is not interacting with the 7TM domain (Fig 5). We hypothesize that an unknown extracellular cue causes the tethered agonist to contact the 7TM domain resulting in an increase of cAMP levels. This signal then promotes coordination of anterior-posterior cleavage plane orientation after the fourth round of cell divisions. Future studies need to focus on the effectors of the cAMP signal and how polar division plane orientation is coordinated on a molecular level by the identified non-polar signal.
C. elegans strains were cultured and manipulated according to standard protocols [50]. Wild-type worms were C. elegans variety Bristol, N2. The allele lat-1(ok1465) was generated by the C. elegans gene knockout consortium and provided by the Caenorhabditis Genetics Center (CGC), which is funded by NIH Office of Research Infrastructure Programs (P40 OD010440). The following transgenes have been previously described: aprEx157[lat-1(lat-2 GPS) (pSP75) rol-6(su1006) pBSK] [25], aprEx77[lat-1::gfp (pSP5) rol-6(su1006) pBSK] [14]. The transgene aprEx185[lat-1T530A/F532A(pSP94) rol-6(su1006) pBSK] was generated for this study (for details see Supporting Experimental Procedures).
For functional assays, COS-7 cells were transiently transfected. To determine total as well as cell surface expression of receptors carrying N-terminal HA and C-terminal FLAG tags, indirect cellular enzyme-linked immunosorbent assays (ELISA) were employed [51]. cAMP concentrations were measured using the ALPHAScreen cAMP assay kit (PerkinElmer Life Sciences) according to the manufacturer's protocol. IP formation was determined as previously described [31]. Assay data was analyzed with GraphPad Prism version 5.0 (GraphPad Software). Statistics were performed using a two-way ANOVA in combination with Bonferroni as post-hoc test. For details see Supporting Experimental Procedures.
All compounds were obtained from Sigma Aldrich. Forskolin was dissolved in DMSO to 10 mM, 3-isobutyl-1-methylxanthine (IBMX) in Dent´s buffer to 10 mM, 8-bromoadenosine 3′,5′-cyclic monophosphate (8-Br-cAMP) in 1 N ammonium hydroxide to 250 mM and 2',5'-dideoxyadenosine (ddA) was diluted in DMSO to 100 mM prior to final dilution in Dent´s buffer. Peptides were synthesized (see Supporting Experimental Procedures) and dissolved in DMSO to 100 mM prior to final dilution in M9. Worms were incubated in 80 μM forskolin, 10 mM IBMX, 0.25 mM 8-Br-cAMP, 1 mM ddA or 0.1 mM peptide solution, respectively, containing E. coli OP50.
RNAi of gsa-1 was carried out using a feeding clone. The open reading frame of gsa-1 was amplified from total cDNA using primers RNAi_1/RNAi_2 (for primer sequences see S1 Table) and cloned into pCR4 (Life Technologies). It was then cloned into the NotI/SpeI sites of L4440 [52]. Feeding by RNAi was performed as previously described using the E. coli strain HT115 [53]. Embryos of the F1 and the F2 generation fed with the RNAi clone described were analyzed for spindle orientation in dividing blastomeres using 4D microscopy.
Nematodes were collected and approximately 2,000 hermaphrodites were placed in TRIzol (Thermo Fisher Scientific). Total RNA was extracted following the manufacturer´s protocol. cDNA was obtained from 1 μg RNA using Omniscript RT kit (Qiagen) and random hexamer primers. qPCR analysis of gsa-1 was performed with primers lat1_1034F/lat1_1035R using a LightCycler PCR machine and GoTaq® qPCR Master Mix (Promega) according to manufacturer´s protocol. As internal reference genes the following were used: act-1 (primers SP1/SP2), cdc-42 (primers SP3/SP4), eif-3 (primers SP7/SP8), tba-1 (primers SP9/SP10). For primer sequences see S1 Table. Data analysis was performed utilizing MxPro QPCR Software (Agilent Technologies).
Approximately 2,000 hermaphrodites were placed in 100 μl M9 containing protease inhibitor (Roche) and sonicated with 15 30 s pulses in a Bioruptor Standard (Diagenode). Approximately 20 μl of sample were boiled in Laemmli buffer for 5 minutes. For the mammalian cell control, 3 × 105 HEK293 cells were lysed in Laemmli buffer for 5 minutes. Protein was subject to electrophoresis as described previously [54] using a 12.5% sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE) gel and transferred to a nitrocellulose membrane (Amersham). Blots were probed with rabbit anti-Gsα antibody (Merck Millipore) at 1:1,000 dilution overnight at 4°C and subsequently incubated for 2 hours at room temperature with a horseradish-peroxidase-conjugated goat anti-rabbit antibody (Sigma Aldrich) at 1:10,000 dilution. Western blots were developed by an enhanced chemiluminescence (ECL) detection system (Thermo Fisher Scientific). For detection of actin as loading control, membranes were stripped in Stripping buffer (1% SDS, 0.1 M Tris pH 6.8, 0.175% β-mercaptoethanol) for 30 minutes at 50°C, blocked and probed with mouse anti-actin (Merck Millipore) at 1:1,000 dilution, and then incubated with horseradish-peroxidase-conjugated rabbit anti-mouse (Sigma Aldrich) 1:10,000 and processed as described above. Antibody signals were quantified by densitometric analysis using ImageJ software [55].
4D DIC imaging and quantitative evaluation of division plane angles were performed as previously described using SIMI Biocell software (SIMI Reality Motion Systems) [56]. Embryos were dissected from young adult hermaphrodites incubated for 120 minutes in Dent`s buffer, 80 μM forskolin or respective solvents as control. Live images were taken with a Zeiss Axioplan 2e and a Zeiss Examiner. Z-stacks with spatial spacing of 1 μm were taken every 35 ms for 300 min. Confocal and fluorescent images were collected with Zeiss LSM5 and LSM510 Meta setups.
The lethality rescue assay was conducted as previously described [14]. Fifty L4 hermaphrodites were transferred into wells of a 72-well flat-bottom Terasaki plates (Greiner Bio-One) containing OP50 and allowed to lay eggs for 24 hours at 22°C. Five to ten eggs were transferred into fresh wells with corresponding solutions and incubated at 22°C. The number of dead/surviving embryos was scored 24 hours later, the number of adult animals 48 hours later on an inverted microscope. Data were examined with an unpaired two-tailed t test for each genotype and condition.
|
10.1371/journal.pcbi.1006402 | Boolean model of growth signaling, cell cycle and apoptosis predicts the molecular mechanism of aberrant cell cycle progression driven by hyperactive PI3K | The PI3K/AKT signaling pathway plays a role in most cellular functions linked to cancer progression, including cell growth, proliferation, cell survival, tissue invasion and angiogenesis. It is generally recognized that hyperactive PI3K/AKT1 are oncogenic due to their boost to cell survival, cell cycle entry and growth-promoting metabolism. That said, the dynamics of PI3K and AKT1 during cell cycle progression are highly nonlinear. In addition to negative feedback that curtails their activity, protein expression of PI3K subunits has been shown to oscillate in dividing cells. The low-PI3K/low-AKT1 phase of these oscillations is required for cytokinesis, indicating that oncogenic PI3K may directly contribute to genome duplication. To explore this, we construct a Boolean model of growth factor signaling that can reproduce PI3K oscillations and link them to cell cycle progression and apoptosis. The resulting modular model reproduces hyperactive PI3K-driven cytokinesis failure and genome duplication and predicts the molecular drivers responsible for these failures by linking hyperactive PI3K to mis-regulation of Polo-like kinase 1 (Plk1) expression late in G2. To do this, our model captures the role of Plk1 in cell cycle progression and accurately reproduces multiple effects of its loss: G2 arrest, mitotic catastrophe, chromosome mis-segregation / aneuploidy due to premature anaphase, and cytokinesis failure leading to genome duplication, depending on the timing of Plk1 inhibition along the cell cycle. Finally, we offer testable predictions on the molecular drivers of PI3K oscillations, the timing of these oscillations with respect to division, and the role of altered Plk1 and FoxO activity in genome-level defects caused by hyperactive PI3K. Our model is an important starting point for the predictive modeling of cell fate decisions that include AKT1-driven senescence, as well as the non-intuitive effects of drugs that interfere with mitosis.
| Complex diseases such as cancer often alter more than one facet of a cell’s function. In addition to breakdown in individual functions, cancer progression leads to unhealthy combinations of cellular behaviors. For example, cancer cells rely on non-physiological combinations of cell functions drawn from an arsenal that includes proliferation, resistance to apoptosis, migration, and blood vessel recruitment. These functions are all critical to health or development, often in a different tissue than that of the tumor. Building predictive models that reproduce this coordination of functions could greatly boost our ability to combat complex disease. Here, we develop a large network model of the processes that control a mammalian cell’s life and death. Our model reproduces a non-intuitive oscillation in a key cell division pathway (PI3K/AKT1), along with the cell-cycle altering effect of its oncogenic activation. To do this, we incorporate the role of Polo-like kinase 1 (mitotic driver, chemotherapy target) and model mitotic failure when Plk1 is blocked. Finally, we offer testable predictions on the unexplored drivers of PI3K oscillations, their timing with respect to division, and the mechanism by which hyperactive PI3K leads to genome-level defects. Thus, our work can aid development of powerful models that cover most processes that go awry when cells transition into malignancy.
| Mammalian cells require extracellular growth signals to divide and specific survival signals to avoid programmed cell death (apoptosis) [1]. The pathways leading to proliferation, quiescent survival or apoptosis are not fully independent; rather, they have a large degree of crosstalk. For example, most pathways activated by mitogenic signals such as PI3K → AKT1 and MAPK signaling also promote survival [2,3]. Moreover, regulatory proteins required for normal cell cycle progression such as E2F1, Myc and cyclin-dependent kinases (CDKs) can promote apoptosis as well [4,5]. Conversely, cell cycle inhibitors such as p16INK4a can enhance survival [6]. As several of our most intractable diseases—cancer, cardiovascular problems and cellular aging-related complications—all involve dysregulation of these processes [7,8], creating predictive models to characterize them has been an ongoing focus for computational and systems biology. Approaches that couple computational modeling with experimental validation have made impressive strides in deciphering the networks in charge of cell cycle progression [9–11] and apoptosis [12–15], as well as the mechanisms of cell cycle arrest in response to stressors such as DNA damage [16–20]. Building on these efforts, our collective focus is increasingly shifting from models that describe individual functions towards ones that successfully integrate several aspects of cellular behavior [21–28]. These integrated models aim to predict the context-dependent outcomes of the crosstalk between different subsystem of large signaling networks, along with the knock-on effects of perturbing one subsystem on others. Furthering this effort, here we offer a comprehensive model of the nonlinear dynamics of PI3K → AKT1 ⊣ FoxO signaling coupled to the cell cycle. Our model can reproduce non-intuitive phenotypic effects of oncogenic PI3K [29], and offer testable predictions about the molecular mechanism responsible for them.
The canonical PI3K → AKT1 pathway is a major relay for growth and survival signals (Fig 1A) [30], as phosphorylated AKT1 has more than a hundred known direct targets [31,32]. First, AKT1 promotes the cell growth required for division and tissue growth, primarily by activating the mTORC1 signaling complex. Provided that the cell is not experiencing amino acid or energy deprivation, mTORC1 aids cell cycle commitment (Fig 1A, box 1), and orchestrates changes in cellular growth metabolism by increasing protein synthesis, lipid and nucleotide metabolism, and mitochondrial biogenesis [33,34]. Second, AKT1 inhibits GSK3β, counteracting its destabilizing effects on cell cycle-promoting and anti-apoptotic genes (Fig 1A, box 2) [35]. Third, AKT1 aids cell cycle entry and survival by translocating the FoxO family of transcription factors out of the nucleus, thus decreasing cell cycle inhibitor and pro-apoptotic gene expression (Fig 1A, box 3) [36]. Fourth, AKT1 phosphorylates the pro-apoptotic BAD, blocking its mitochondrial localization (Fig 1A, box 4) [37].
The array of signaling events described above all point to a coherent role of AKT1 in promoting survival and proliferation. There is mounting evidence, however, that PI3K → AKT1 activity during cell cycle progression is more complex [31,32]. Overactive AKT1 in cancer cells has been associated with driving cells into senescence (an aging cell state characterized by permanent cell cycle arrest) [38,39]. More intriguing are studies showing that active FoxO3 and/or FoxO1 not only block cell cycle entry but are paradoxically required for its subsequent completion. A study by Alvarez et al has shown that the attenuation of the PI3K → AKT1 pathway after restriction point passage was required for FoxO3 activity in G2, which in turn aided the completion of cytokinesis [29]. To explain their observations, the authors showed that FoxO3 upregulates the expression of the mitotic cyclin B and polo-like kinase 1 (Plk1), thus promoting the G2/M transition and progression out of telophase. Furthermore, work by Yuan et al also implicated FoxO1 in G2-phase Plk1 regulation [40]. This effect, however, may be short lived, as both FoxO factors are inhibited by Plk1 phosphorylation [41,42]. To capture these subtleties in a model, we first need to understand the mechanisms that generate a short-lived AKT1 pulse.
There are several known feedback mechanisms that can explain the pulse-like spike and subsequent attenuation of AKT1 following growth factor stimulation (Fig 1B). Most of these involve mTORC1, and several are specific to insulin and IGF1 signaling [32]. For example, mTORC1 is known to mediate the degradation of insulin receptor substrates IRS1/2 required for insulin and IGF1 signaling (Fig 1B, red arrows) [43]. In addition, inhibition of FoxO transcription downstream of AKT1 leads to attenuated transcription of insulin and IGF1 receptors [44]. To further complicate the picture, activation of the mTORC1 target S6K sets in motion two growth receptor-independent negative feedback loops. First, S6K can attenuate mTORC2 activity required for full AKT1 activation (Fig 1B, orange arrows) [42,43]. Second, S6K can promote nuclear export of the PI3K inhibition PTEN (Fig 1B, purple arrows) [45]. Together, these mechanisms are thought to dampen AKT1 activation following an initial peak and regulate the homeostatic maintenance of AKT1 under ongoing growth stimulation.
In addition to feedback downstream of AKT1, a study by Yuan et al has demonstrated that even before AKT1 signaling downstream of growth receptors has a chance to engage, only a relatively small subpopulation of cells (~30%) are responsive to these signals to begin with [46]. The remaining cells do not suffer from lack of receptor activation or lack of AKT1 protein; rather, they have very low levels of the PI3K subunit p110 at the time of stimulation. More surprising was their finding that as AKT1 peaked in the responding population of cells, the initially high p110 underwent rapid degradation. This effect essentially co-occurred with AKT1 activation; a process too rapid for the feedback mechanisms downstream of AKT1 to mediate. Moreover, the feedback detailed above acts on AKT1 phosphorylation and/or PI3K activity, not on their protein expression. Yuan et al showed that the cycle of rapid p110 degradation and subsequent re-synthesis was mandatory for sustaining proliferation, as clonal populations with degradation-resistant p110 and sustained peak AKT1 activity entered senescence at a high rate. Importantly, their findings were not restricted to a single growth receptor, pointing to a general, yet unrecognized set of feedback loops driving the expression cycle of p110. Finally, their work indicated that p110 heterogeneity in quiescent cells is strongly influenced by local cell density. Another study confirms that the catalytic p110 subunit of PI3K is indeed rapidly degraded upon growth stimulation in two additional cell lines, and that it re-accumulates slowly (~2 hours) [47]. This work points to the NEDDL4 ubiquitin ligase as the driver of p110 degradation [47]. In addition, AKT1 phosphorylation has been shown to exhibit at least two clear peaks before the end of S-phase in cells entering the cell cycle from quiescence [29], an effect observable despite rapid de-synchronization of the cell culture. These studies, however, do not address the molecular mechanisms that trigger p110 degradation specifically in response to growth factor signaling, its subsequent re-synthesis, or the way it’s oscillations interface with cell cycle control.
Current computational models of the regulation of mammalian cell life and death do not account for dynamic p110 expression [16]. Models that incorporate feedback on AKT1 activity typically focus on the intricacies of the mTORC1 / mTORC2 crosstalk [48] or the effects of negative feedback on the strength of AKT1 signaling [49,50], but do not encompass the full cell cycle. Here, we put forth a large Boolean model of the regulatory interactions driving dynamic growth factor signaling, cell cycle progression and apoptosis. We built the model by bringing together several separately published, disconnected pieces of evidence regarding p110 protein and mRNA regulation [47,51,52]. We then linked the resulting growth signaling layer to an updated Boolean cell cycle model [11], as well as the molecular network responsible for survival vs. apoptosis. The resulting Boolean model reproduces the cell-cycle dependent role of PI3K, AKT1 and FoxO proteins [29,46] by linking them to Plk1 regulation in G2. As expected, it generates straightforward behaviors such as lack of cell cycle commitment in the absence of high p110 expression [46], or G1 shortening in the presence of hyperactive PI3K / AKT1. The novelty and value of our model, however, stems from its ability to reproduce more intricate, non-intuitive phenotypic outcomes. First, our model reproduces the path to apoptosis in the event of a mitotic catastrophe [53]. Second, our model generates four distinct cell fates in response to Plk1 inhibition, depending on the timing of Plk1 loss [54]: i) G2 arrest [55], ii) mitotic catastrophe [54,56–58], iii) premature anaphase and chromosome mis-segregation leading to aneuploidy [59], and iv) failure to complete cytokinesis following telophase [60–62], which can lead to genome duplication [59]. Third, our model can replicate failure of cytokinesis and accumulation of binucleate telophase cells driven by hyperactive PI3K / AKT1 or by FoxO3 inhibition [29].
Our model’s ability to accurately reproduce a range of cell fates triggered by altered PI3K, AKT1, FoxO3 or Plk1 activity leads to several experimentally testable predictions. Namely, we predict 1) the molecular mechanisms of p110 degradation in response to high PI3K activation, and the transcriptional driver of its re-synthesis; 2) that the degradation / re-synthesis cycle of p110 occurs at least twice per division cycle (along with the molecular mechanism for their phase-locking); 3) that cell cycle defects in response to PI3K / AKT1 over-activation or FoxO3 knockdown are generally due to a loss of Plk1 in telophase; 4) loss of strong growth signaling in p110-overexpressing cells allows for normal cell cycle completion; and finally, 5) that cells in which p110 is inhibited after the start of DNA synthesis can still pre-commit to another cell cycle in the presence of saturating growth stimulation.
In order to build a mechanistic model of the dynamics of growths signaling and its influence on cell cycle progression and apoptosis, we turned to Boolean modeling [63]. Using a modular approach proposed in [11], we first collected key growth signaling pathways driving cell cycle commitment in a Growth Signaling module responsible for the dynamics of PI3K, AKT1, MAPK and mTORC. Next, we identified key regulatory subsystems that control cell cycle progression, such as the Restriction Switch driving the initial commitment to DNA synthesis [11], the Phase Switch driving cell cycle progression from G2 to M and back to G1 [11] (expanded from [11] to account for the mitotic role of Plk1 [54]), and a regulatory switch that tracks the licensing and firing of replication origins. Finally, we synthesized several published models of the survival vs. apoptosis decision into an Apoptotic Switch. These modules are tied together into an 87-node network by direct regulatory crosstalk, as well as a few nodes that represent cellular processes we do not track in molecular detail (e.g., DNA Replication, mitotic spindle assembly or cytokinesis). Following a detailed description of our model, we show that it faithfully reproduced quiescent, apoptotic and dividing cell phenotypes, and that its behavior is robust under synchronous or asynchronous update. To understand the role of dynamic PI3K signaling in healthy cell cycle progression, we then explore the consequences of Plk1 inhibition at different points along the cell cycle and show that the non-intuitive consequences of PI3K hyperactivation are explained by mild Plk1 inhibition in G2/M.
In order to build a Growth Signaling module that incorporates the molecular drivers of p110 dynamics, we turned to the literature in search of mechanisms that can drive rapid p110 degradation and gradual re-synthesis (Fig 2A). Both the free and p85-bound versions of the PIK3CA (p110α) subunit of PI3K have been shown to undergo proteasome-dependent degradation triggered by the E3 ubiquitin ligase NEDDL4 [47]. The activity of NEDDL4, in turn, requires Ca2+ and inositol trisphosphate (IP3) [51]. This led us to hypothesize that the ability of NEDDL4 to ubiquitinate p110 spikes in response to sudden growth factor stimulation. Namely, growth receptors activate phospholipase C γ (PLCγ), an enzyme that generates IP3 from membrane-bound PIP2. IP3 diffuses to the endoplasmic reticulum, where it triggers Ca2+ release into the cytosol [64]. Thus, IP3 and Ca2+ are available to activate NEDDL4 within minutes of receptor activation, leading to rapid p110 degradation. As membrane tethering of PLCγ requires PIP3—a product of active PI3K [65,66], the cascade leading to the polyubiquitination of p110 can only occur in cells that express high levels of p110 when growth signals arrive (as observed by Yuan et al [46]). To summarize, we posit that strong PI3K activation initiates a negative feedback loop leading to its own degradation, independently of its effect on AKT1 (Fig 2A, red links).
Next, we turned to the mechanism of p110 re-synthesis. Studies of the p110α promoter indicate that this gene is positively regulated by FoxO3 [52]. We hypothesized that reactivation of FoxO3 in the G2 phase of the cell cycle, after the initial AKT1 activation subsides, is the driving force behind p110 re-synthesis (Fig 2A, orange link). To integrate these negative feedback loops with the canonical PI3K / AKT1 signaling cascade activated by growth receptors, we introduced separate Boolean nodes to track basal vs. peak PI3K and AKT1 activity (Fig 2B; Boolean gates: S1A Table). Our model can thus distinguish between survival signaling in a low growth factor environment (where basal PI3K and AKT1 are ON) and peak PI3K/AKT1 activation following the arrival of a strong mitogenic stimulus. Complemented by a linear MAPK cascade and mTORC1/2 signaling, this non-linear PI3K/AKT1 axis dominates the behavior of the resulting Boolean Growth Signaling module (Fig 2B).
Modeling the two feedback loops controlling p110 expression in isolation shows that they generate a sustained, robust oscillation (Fig 2C), even though our model does not account for the fact that p110 degradation is significantly faster than its re-synthesis. This oscillation is the only attractor state of the small module regardless of Boolean update. As Fig 2C indicates, the synchronous attractor cycle clearly maps onto the cyclic succession of complex attractor states of the general asynchronous model (Fig 2C, weighted, directed network in the middle). In addition to never leaving the complex attractor shown on Fig 2C, asynchronous time series repeatedly walk through cycles of states that resemble the synchronous limit cycle (Fig 2D). Within the context of the larger Growth Signaling module, this oscillation only occurs under ongoing high growth factor stimulation.
In order to investigate the downstream consequences of an oscillating Growth Signaling module, we next updated our previously published cell cycle model [11] and extended it with an apoptotic switch (described in detail in Methods & Models).
Linked together, the modules generate a dense 87-node Boolean model with 375 links (Fig 3B). The synchronous dynamics of the full model is heavily constrained by the switch-like behavior of its modules, as evidenced by the small number of tightly coordinated behaviors (phenotypes) it generates. Indeed, when the state space of the network is sampled extensively using noisy synchronous update (Methods & Model–Mapping the attractor landscape of large Boolean networks using synchronous update), every attractor corresponds to a distinct cellular phenotype. These attractors are characterized in detail in S2 Table, along with key molecular signatures that allow us to match them to specific phenotypes. Fig 3C summarizes them according to the extracellular environment each phenotype occurs in; namely, the absence / low abundance / high abundance of growth factors (x axis on Fig 3C) combined with the presence / absence of the apoptotic signal Trail (y axis). Table 1 matches cell phenotypes generated by our model to experimentally documented cell behaviors in multiple cell types. As expected, irreversible apoptosis is stable in every environment. Moreover, the ongoing presence of saturating Trail (i.e., Trail input is ON 100% of the time) destabilizes every other cell state, leaving apoptosis as the only stable option [79–82]. Similarly, the complete absence of growth / survival signals also leads to apoptosis [83–85]. In contrast, low levels of growth signaling support quiescent cell states, and our model identifies two distinct forms. First is a healthy cell state with 2N DNA content (Fig 3C, elongated cell with blue nucleus on). Second, our model also produces a G0-like state representing cells that have failed to complete mitosis or cytokinesis in the past, now stuck with a 4N DNA content (Fig 3C, elongated cell with yellow circle around nucleus). Finally, exposure to high levels of growth factor results in a cyclic attractor representing continuously cycling cells (Fig 3C, mitotic cell).
Our modular approach allows us to attribute discrete transitions cells undergo to the dynamics of isolated regulatory switches, apparent in the activation patter of the interlinked modules under synchronous update. For example, the sequence of molecular changes that occur within our modules transitioning from a quiescent state into the cell cycle reveals the higher-order logic by which regulatory switches toggle each other (Fig 4). First, cell cycle entry involves the activation of the Growth Signaling Module. While the basally active parts of this module remain on, we see a cascade leading to MAPK signaling (Fig 4, upstream PI3K cycle). This part of the module remains stably ON in a high (saturating) growth factor environment. In contrast, the part of the module responsible for cyclic PI3K / AKT1 activation enters an oscillating pattern, as expected from the limit cycle on Fig 2C. Thus, our integrated model of growth signaling and cell cycle progression can reproduce the experimentally documented but unexplained oscillations in PI3K expression and AKT1 activity [46]. Next, cyclic AKT1 activity triggers downstream oscillations in mTORC1 signaling and GSK3β. As these AKT1 targets are subject to feedback from the rest of the network, they do not directly mimic the dynamics of PI3K and AKT1 (see Methods & Model). Full activation of the Growth Signaling module then toggles the Restriction Switch into a state representing restriction point passage (later we observe this switch partially, but not fully reset between each cycle). Around the same time, we observe licensing of replication origins (Origin Licensing Switch), subsequently reset by the firing of replication origins in S-phase. Now committed, the cell toggles through replication, G2, mitosis and cytokinesis under the control of the Phase Switch (see Cell cycle processes). In contrast, the Apoptotic Switch only experiences minor perturbations, without being flipped.
To test whether the orderly progression through the cell cycle is robust to random fluctuations in signal arrival time as they propagate through the network, we tested the model’s behavior under random order asynchronous update (Methods & Model–Boolean Modeling Framework) [86]. As fixed-point attractors of a Boolean model remain the same regardless of update [87], we focused on the cell cycle. As S1 Fig shows, a fully random update order does not abolish the model’s capacity to execute a correct cell cycle sequence, but it does introduce several non-biological behaviors. First, the signals that couple successful DNA replication to the establishment of a G2 state are lost under a subset of update orders, leading to G2 → G1 reset followed by a new cell cycle (endoreduplication). Second, the signals that drive cytokinesis can also be disrupted by certain update orders. Third, the balance of pro- and anti-apoptotic signals during metaphase can tip in favor of apoptosis, as if the cell experienced mitotic catastrophe. Interestingly, all three cell cycle errors are observed in vitro in cells experiencing knockdown or overexpression of a variety of cell cycle regulators [16,54,88]. Thus, we conclude that the asynchronous model with random update order mimics the occasional short-term loss of regulators, rather than the robust cycling of healthy cells.
In order to create a restricted random order that forbids asynchronous state transitions resulting from these non-physiological breaks in signal transduction, we identified genes and processes that deviated from their expected activity every time a particular error occurred and created an asynchronous version of the model with biased random update (Methods & Model–Boolean Modeling Framework). To do this, we placed a small subset of nodes at the start or end of each update order, depending on their activation status (11 nodes; list and rationale in S3 Table). Using this biased update our model repeatedly and correctly executes the cell cycle, in spite of the asynchronous update (Fig 5). Our update bias did not completely eliminate endoreduplication from G2 and apoptosis (S2 Fig), but the incidence of these errors decreased drastically. As these errors do occasionally occur in wild-type cells [16,82], we choose not to further restrict our update order to eliminate them. Rather, we measured the rate at which the two update schemes produce normal cell cycle events vs. different errors via a series of simulations at varying levels of growth factor and Trail stimulation. We did this by setting GFH or Trail ON with probability p in each time-step, OFF otherwise. As Fig 5B indicates, the asynchronous model with biased update shows a similar response to growth factors and Trail as the synchronous model. Moreover, the incidence of apoptosis or endoreduplication after G2 is significantly lower than under random update, and lack of cytokinesis all but disappears.
The apoptotic fixed-point is reachable from cell cycle under both random-order and biased asynchronous update, indicating that the cell cycle is not, strictly speaking, a complex attractor [89]. Nevertheless, starting an asynchronous time series from any state along the synchronous cell cycle attractor results in long time-courses featuring repeated (if occasionally incorrect) cycles (S4 Fig). This indicates that the system’s state space has a metastable region that traps its dynamics in a way that resembles a complex attractor. To test whether this metastable collection of states is also a cycle, we sampled the state transition graph of the asynchronous model with both update schemes by starting 10 independent time courses of 1000 steps from each state along the synchronous cell cycle. In order to sample the metastable basin rather than the path to the apoptotic attractor, we prematurely interrupted each run if it reached a fixed point. We then overlayed all observed states and transitions, visualizing the largest strongly connected component (S5 Fig, left). To test whether these state transition graphs are consistent with robust execution of the cell cycle, we classified each state as representing G1, S, G2, metaphase, anaphase, telophase and cytokinesis depending on the ON/OFF state of key processes (S6 Fig). Instead of a cycle, however, the resulting network revealed distinct regions of state-space representing G1, S and G2, then a few highly restricted and often-visited paths through anaphase and cytokinesis. Thus, asynchronous update indicates that there may be widespread molecular heterogeneity in G1, S and G2, but most of the network we model locks into a few unique states during anaphase.
It is worth noting that our model features two internal oscillators, the core cell cycle and the PI3K degradation / re-synthesis cycle. As Fig 5 and S1 Fig indicate, these two cycles are not completely phase-locked under asynchronous update. As the cell cycle proceeds, the small PI3K oscillator and the downstream mTORC1 pathway can be found in nearly any state. The sole exception is anaphase, where the two cycles appear to sync up. To show that the heterogeneity is chiefly within the growth pathway, we projected the state transition graph of each asynchronous model onto a subspace where each network state represents a unique ON/OFF state within the core cell cycle modules (Restriction SW, Origin of Replicaton SW, Phase SW and Cell cycle processes), regardless of the state of all other nodes. This process collapsed the complex state transition graph of the biased model onto a clear cyclic flow of transitions, representing normal cell cycle progression (S5 Fig, bottom right). In contrast, the random asynchronous model’s dynamics has a loop corresponding to the cell cycle, but it is dominated by prominent “backward” transitions representing endoreduplication from G2 (S5 Fig, top right).
To further test our model against published experimental data, we compared its least intuitive dynamical behaviors to experimental observations (Table 2) and described them in detail in S1A–S1D Text. To summarize, both our synchronous and biased asynchronous model reproduces the cyclic degradation and re-synthesis of p110, leading to oscillating AKT1 signaling (Figs 4 and 5). In cells entering the cell cycle from quiescence, this oscillatory behavior generates two distinct phospho-AKT1 peaks before cells complete DNA synthesis (Fig 4, S7 Fig). Furthermore, cells that lack high p110 protein expression fail to enter the cell cycle in response to growth factors (S8 Fig). Our models also reproduce the bifurcation of fates in cells cycling in non-saturating growth environments. Namely, a large fraction of cells were shown to pass the restriction point before cytokinesis (in late G2/M of the previous cycle), while the remainder reset into a G0-like state and commits to the next cycle again after a highly variable time-window (S9 and S10 Figs) [90]. Finally, a comparison of 34 model knockout and 11 overexpression phenotypes to experimentally manipulated cell behaviors indicate that our model can faithfully reproduce in vitro cell behavior under a wide range of genetic manipulations (S4 Table).
Experimental data indicates that hyperactive PI3K and/or AKT1 in G2 leads to an enrichment of binucleated cells stuck in telophase [29,40]. Studies that document these errors point to the loss of FoxO3 and/or FoxO1 activity in G2 (a consequence of hyperactive AKT1), two transcription factors that positively regulate the expression of mitotic cyclin B, as well as polo-like kinase 1 (Plk1). Cyclin B accumulation is only required for metaphase entry (a process that appears normal in cells with hyperactive PI3K/AKT1); its activity is not required for cytokinesis. Plk1, in contrast, plays distinct roles at every phase of mitosis and cytokinesis [54]. Thus, we hypothesized that telophase enrichment in cells with hyperactive PI3K/AKT1 may be due to compromised Plk1 expression in G2 or early mitosis [29,40], and that partial knockdown of Plk1 in our model phenocopies this error.
To test this, our previously published Phase Switch [11] required a revision to incorporate the complex regulatory role of Plk1 (Fig 3A). Experimental evidence indicates that Plk1 is upregulated in G2 by the FoxM1 transcription factor (also newly added). While the combinatorial regulation of Plk1 by FoxM1, FoxO3 and FoxO1 has not been investigated, experiments clearly show that Plk1 remains active until late telophase [60,61]. That said, Plk1 protein level drop dramatically in anaphase due to proteasomal degradation by APC/CCdh1 [60,61]. It is the availability of the remaining Plk1 pool, responsible for the assembly of a contractile ring, that appears compromised in the absence of FoxO activity in G2 [29]. To capture this within a Boolean framework, we accounted for the role of FoxO factors in creating an increased pool of Plk1 by introducing two Boolean nodes to track Plk1 activity (S11 Fig, S1E Text). Thus, the Plk1 node represents the active enzyme required for mitotic entry, normal mitotic progression and anaphase completion. In contrast, Plk1H = ON represents the short-lived accumulation of a large enough Plk1 pool to survive APC/CCdh1 mediated degradation past anaphase, and aid the assembly of a contractile ring by recruiting the RhoA GEF protein Ect2 [60].
Next, we tested whether our model can accurately account for all known roles of Plk1 during cell cycle progression. To this end, we first modeled the inhibition of Plk1 at different points along the cell cycle using synchronous update [54]. As Fig 6 shows, Plk1 inhibition in our model reproduces four distinct, experimentally documented phenotypic outcomes, depending on the precise timing of Plk1 inhibition during the cell cycle (Table 3). First, loss of Plk1 before prometaphase (i.e., before robust Cdc25C & Cdk1 activation) results in G2 arrest (Fig 6A). Second, complete of loss Plk1 at the prometaphase /metaphase transition or early metaphase leads to prolonged arrest and mitotic catastrophe (Fig 6B). Third, our model predicts that Plk1 loss in late metaphase can trigger permute anaphase rather than mitotic catastrophe, leading to chromosome mis-segregation and aneuploidy (Fig 6C). This occurs when Plk1 and CyclinB / Cdk1 are both available to phosphorylate the APC/C subunit of the Anaphase Promoting Complex [92], leading to Cyclin A degradation [93]. The loss of this key APC/CCdh1 inhibitor, together with the subsequent loss of Cdk1 activity in the absence of Plk1, results in premature APC/CCdh1 activation. APC/CCdh1 disassembles the incomplete mitotic spindle, allowing a narrow escape from apoptosis and instead leading to chromosome mis-segregation and premature telophase. Lack of Plk1 past this point guarantees that cytokinesis does not follow. Making matters worse, our model shows that continued growth factor signaling can lead to a new round DNA synthesis (Fig 6C). Fourth, Plk1 inhibition a time-step later leads to normal anaphase mediated by APC/CCdc20 (Fig 6D). As long as Plk1 inhibition starts before APC/CCdh1 activation, however, cytokinesis still fails (Fig 6D, lime green line).
Given that our model adeptly captures four distinct effects of Plk1 inhibition, next we asked whether partial loss of Plk1 can phenocopy the effects of hyperactive PI3K/ AKT1. We modeled partial knockdown of Plk1 by running stochastic simulations in different growth conditions, where we forced the OFF state of Plk1 in every time-step with a fixed probability and allowed the node to obey its normal regulation when not forced (Fig 7; Methods & Model–Modeling non-saturating growth factor stimulation and partial knockdown / overexpression within a Boolean framework). Our simulations indicate that the dominant failure mode in a population of cells depends on the strength of Plk1 inhibition. When Plk1 inhibition is very strong (but not complete), cells often start mitosis but do not complete it. This leads to increased mitotic length, often followed by apoptosis (Figs 7 and S12A). In contrast, aberrant mitosis leading to aneuploidy is more common at moderate Plk1 inhibition (peak at 60%), though apoptosis is still more likely (Figs 7 and S12A). The most common cell fate at this point, however, is normal mitosis followed by prolonged telophase (S12B Fig) and failure to undergo cytokinesis. This remains the prominent failure mode at moderate-to-weak Plk1 inhibition levels (peak at 30%; Figs 7 and S12A). Performing the same series of in silico experiment using biased asynchronous update lead to qualitatively similar results (Fig 7), with the caveat that the asynchronous model occasionally skip mitosis altogether–an effect that does not change with Plk1 inhibition. In summary, weak Plk1 inhibition in our model phenocopies the experimentally documented effects of hyperactive PI3K and/or AKT1–in line with the hypothesis that the cause of weakened Plk1 expression is lack of some (i.e., FoxO3 and FoxO1) but not all transcriptional Plk1 activators in G2 (FoxM1 remains active).
To test whether our model accurately links non-degradable p110 to altered Plk1 expression leading to failure cytokinesis, we ran in silico experiments in which we kept the p110H node forcibly ON, starting at different points along the cell cycle (Fig 8). As expected, expression of a non-degradable p110 leads to high sustained PI3K and AKT1 activity. Loss of FoxO3/FoxO1 during G2 and M prevents Plk1 levels from accumulating enough to outlive APC/CCdh1-mediated depletion (Plk1H does not turn on; Fig 8A). The result is failure to undergo cytokinesis (Fig 8A, red box), matching the experimentally documented enrichment of telophase cells in the presence of overactive PI3K, AKT1, or inactive FoxO3 [29]. In addition to telophase enrichment, our model shows genome reduplication in the resulting bi-nucleated cells, also supported by experimental evidence (Table 3). Intriguingly, our model predicts that the loss of high growth factors during G2 or M allows these cells to compete cytokinesis (Fig 8A, second cycle). This occurs because high p110 protein expression alone is not sufficient for high AKT1 activation; it also requires ongoing growth signaling and active Ras [91]. Thus, loss of strong growth stimulation allows the re-entry of FoxO3 into the nucleus, leading to Plk1 expression and cell cycle completion.
Synchronous update allows us to track the molecular consequences of locking p110 into a high-expression state, but it has several drawbacks. Most importantly, it assumes the presence of a saturating growth factor environment and 100% FoxO inhibition downstream of PI3K/AKT1. To test whether our results hold in the presence of intrinsic or extrinsic fluctuations such as moderate growth factor availability and incomplete hyperactivation of p110H, we tracked the fate of cells in a variety of non-saturating growth factor environments. In each environment, we tested the effect of incomplete p110H, p110H + PI3KH, or AKT1H over-expression by stochastically forcing the ON state of these nodes with a fixed probability (Figs 8B and S13). These results also point to a high prevalence of cells that cannot exit telophase in near-saturating growth environments (blue bars on Fig 8B). As hyperactive AKT1H in the model is forced ON regardless of growth signaling, it drives both an increase in proliferation and the failure to exit telophase even in low growth factor environments (S13B Fig).
As the mediators of cell cycle progression errors in hyperactive PI3K/AKT1 are thought to be FoxO factors, we next showed that partial inhibition of FoxO3 phenocopies hyperactive PI3K and/or AKT1 in our synchronous model (S13E Fig, Table 3). That said, strong FoxO3 inhibition also slows / stops re-synthesis of p110, leading to a lengthened cell cycle (documented in cancer cells and tumors in vivo [94]; Table 3). The subsequent weakening of AKT1 signaling counterbalances the loss of FoxO3, weakening its effects. In line with this, our biased asynchronous update results do not show an increase in cytokinesis failure with FoxO3 knockdown (Figs 8C and S13E).
In addition to reproducing the effects of hyperactive PI3K/AKT1, our model offers several experimentally testable predictions: 1) We predict that the observed cycle of p110 degradation and re-synthesis is driven by the network in Fig 2A. As a result, knockdown of PLCγ, NEDDL4, or the chelation of intracellular Ca2+ is expected to lead to sustained high p110 protein expression in vitro. 2) In addition to an increase in cell cycle length, we predict PLCγ knockdown to enrich for telophase cells that fail to complete cytokinesis. 3) Continuously cycling cells execute at least two rounds of PI3K activation and destruction for each round of division (Figs 4 and 5). 4) Loss of saturating growth signals in G2 allows p110-overexpressing to complete a normal cell cycle (S10 Fig). 5) Once committed to a cell cycle, saturating growth stimulation allows cells to keep cycling even if p110 levels drop later in the cycle, and pre-commitment in p110-inhibited cells is driven by mitotic mTORC1 aiding the re-activation of Myc (S10 Fig).
In this study we developed a detailed modular Boolean model of the regulatory pathways driving growth factor signaling, cell cycle progression and apoptosis (Fig 3B). While there are several published models with a similar coverage of cellular behaviors [15,21,23,25,26], the focus of our study was to capture the dynamical behavior of the PI3K → AKT1 signaling axis driving cell growth. To this end, we proposed a mechanism capable of driving the experimentally documented oscillations of PI3K protein expression [46,47], explored the importance of high and low PI3K activity during different phases of the cell cycle, and showed that our model can offer mechanistic insight into the cellular effects of hyperactive PI3K (failure of cytokinesis). To do this, we identified two negative feedback loops potentially responsible for driving PI3K dynamics. The first loop is triggered by high growth factor signaling and high PI3K activity, and it involves PLCγ-mediated activation of the NEDDL4 ubiquitin ligase [51], known to target the p110 subunit of PI3K for degradation [47]. The second loop involves the loss of AKT1-mediated FoxO3 inhibition as PI3K activity drops, allowing FoxO3 to drive the re-expression of p110 [52]. As these two pathways were key to our model’s ability to reproduce the effects of hyperactive PI3K and AKT1 on cell cycle progression, they represent its two most significant predictions.
Linking PI3K oscillations to the rhythm of cell division required an update of our previously published Phase Switch [11] to include the multifunctional Plk1 protein required for all phases of mitosis and cytokinesis [54]. According to our model, during normal cell cycle progression the low-PI3K / low-AKT1 phase of the PI3K oscillations lead to nuclear re-entry of FoxO3 and FoxO1, which aid the accumulation of Plk1 and are required for cytokinesis. In addition, we predict that the inhibitory influence of Plk1 on FoxO3 [41] helps lock PI3K oscillations to the cell cycle (Fig 4). That said, a strict phase-locking is only enforced at the metaphase / anaphase transition, as evidenced by the behavior of the asynchronous version of our model (Figs 5 and S5). In addition to offering testable predictions of the mechanisms behind PI3K oscillation summarized in Fig 9, our model is the first to account for the multifaceted role of Plk1 in cell cycle progression (Fig 6). Namely, we were able to reproduce G2 arrest in the complete absence of Plk1 [55], mitotic catastrophe in response to Plk1 removal in metaphase [54,56–58], the potential for premature APC/CCdh1 activation and chromosome mis-segregation [59], as well as failure to carry out cytokinesis in the absence of a Plk1 pool that survives APC/CCdh1-mediated destruction [60–62].
A limitation of our current model stems from uncertainties in the experimental literature on the connection between Plk1 and FoxO factors. As we detailed in Results, the combinatorial regulation of Plk1 by FoxM1, FoxO3 and FoxO1 is not characterized. It is not clear whether these factors cooperate or independently augment Plk1 expression. Moreover, our assumption that either FoxO factor alone can boost Plk1 sufficiently to survive until telophase has not been tested in vitro. Thus, the logic gates connecting Plk1 and the FoxO factors may need a revision in light of additional data. That said, aspects of Plk1 regulation that guarantee its loss in telophase but not earlier in cells with hyperactive PI3K/AKT1 requires key elements of our regulatory logic to remain intact [29].
Throughout this work we used synchronous and asynchronous Boolean modeling in parallel, allowing us to leverage the advantages and mitigate the drawbacks of each update scheme. A key advantage of synchronous update is that the dynamics it generates is entirely deterministic [63]. This allowed us to probe the effects of inhibiting nodes at specific times along a dynamical trajectory such as the cell cycle, and predict distinct phenotypic outcomes depending on the precise timing of inhibition. For example, using synchronous update to model Plk1 inhibition along the cell cycle points to a time-sensitive sequence of failure modes: G2 arrest, mitotic catastrophe and aberrant anaphase, followed by normal anaphase but failed cytokinesis (Fig 6). The power of these simulations is that they reveal distinct ways in which the molecular balance of Plk1, Cdk1/Cyclin B, premature activation of APC/CCdh1, and pro-apoptotic factors accumulated by mitotic delay can be tipped (Fig 9). In the presence of molecular noise in vitro, however, we expect Plk1 knockdown to generate a mix of these cell cycle errors. Indeed, experiments indicate that mitotic death and aberrant anaphase co-occur in Plk1-inhibited cells [59]. To reproduce this, we simulated the partial stochastic inhibition of Plk1, resulting in a changing mix of errors with both synchronous and asynchronous update (Fig 7). Our success with the latter is especially helpful for showing that the four failure modes are not artifacts of non-biological synergies in signal arrival, a pitfall of synchronous update.
Comparing cell cycle progression with the two update schemes revealed that asynchronous update introduces a stochasticity in cell cycle entry, observed in several mammalian cell lines [90,97,98]. This is similar to the behavior of the synchronous model in non-saturating environments, and it is largely due to the fact that the PI3K/AKT1 cycle does not stay in sync with cell cycle progression for most of the cycle. As a result, the ability of AKT1 to relay growth signals to the Restriction Switch in late G2 / early metaphase, and thus pre-commit cells to another division, remains stochastic under asynchronous update even in saturating growth environments. Cycling cells in vitro are likely somewhere in between; less random in their ability to re-commit during G2/M than the asynchronous model, but not deterministic either. In addition to cell cycle commitment, overly noisy signal propagation is likely responsible for the asynchronous model’s results in cells with hyperactive AKTH and low FoxO3 (Figs 8C and S13E). In contrast to simulations with synchronous update, the fraction of cells that failed to complete cytokinesis under asynchronous update was small. Here we think that synchronous update overestimates, while asynchronous update underestimates the rate of this cell cycle error. In summary, our parallel use of the synchronous and asynchronous Boolean frameworks helped us uncover subtle inter-dependencies in the dynamics of our coupled regulatory modules, but also guaranteed that our results are robust with respect to noise in signal propagation and do not depend on non-biological synchrony of parallel signals with a variety of speeds.
An intriguing model prediction related to the coupling of the cell cycle and the PI3K cycle leverages our model’s ability to reproduce pre-commitment to another cycle at the G2/M transition (Fig 5) [90], a feature inherited from our previous cell cycle model [11]. Even though high p110 expression is required for cell cycle entry from quiescence (S8 Fig) [46], we predict that under saturating growth factor conditions high p110 / PI3K is not required for pre-commitment to another cycle (S10 Fig). This is surprising, as pre-commitment at the G2/M boundary normally coincides with the AKT1-high portion of the second PI3K cycle. Our model suggests that the stabilization of Myc in p110-low cells occurs in pro-metaphase in spite of low AKT1 and high GSK3β, owing to increased activity of mTORC1 driven by Cdk1/Cyclin B, specifically in the presence of GSK3β [96]. Experimental validation of these predictions is an important step toward understanding the complex interplay of factors that control Myc expression and pre-commitment to another cycle at the G2/M boundary. The same experimental setup that showed the existence of pre-committed cells could accomplish this [90] by probing the effect of simultaneous MEK and PI3K inhibition on pre-commitment. According to our prediction, this dual inhibition would reduce but not eliminate the fraction of cells that finish their current cycle following MEK and PI3K inhibition, then complete another.
Our modeling results on partial Plk1 inhibition offer a cautionary note on targeting Plk1 as a tumor-suppressive strategy. On one hand, Plk1 inhibition does significantly limit proliferation due to G2 arrest (Fig 6A) and promotes apoptosis in cells that escape from G2 via mitotic failure (Fig 6B). On the other hand, our model predicts that an ill-timed, short-lived pulse of Plk1 inhibition can lead to faulty anaphase, chromosome mis-segregation resulting in aneuploidy, and subsequent genome duplication (Fig 6C). According to our model, weak Plk1 inhibition in individual cells is especially dangerous in this regard (Fig 7). Thus, it is possible that cancer therapy based on Plk1 inhibition [99] could increase genomic instability in cells that survive it. A propensity for aneuploidy and genome duplication was indeed observed in Plk1-inhibited cells [59], but also in a mouse model harboring the oncogenic mutation in the alpha subunit of PI3-Kinase [100]. The molecular mechanisms behind the latter were never explained. Our model not only reproduces these outcomes (Table 3), but also points to the ill-timed loss of Plk1 as the likely culprit (Fig 6).
Looking ahead, our current model lays the groundwork for modeling the mechanisms of AKT1-induced senescence [38,39,100]. An extended version of our model with DNA damage-induced G2 arrest in cells with hyperactive AKT1 would express most key drivers of senescence (i.e., mTORC1, RB, p53 and p21). Thus, an important next goal is to complement our model with a DNA damage signaling module [16–18,101,102], then build the regulatory switch that locks in and maintains senescence [20,24,103–105]. The predictive power of our model could be further expanded by revising our Growth Signaling and Apoptosis modules to leverage more detailed computational models of MAPK signaling [25], as well as the apoptosis/necrosis decision [15]. Finally, building a contact inhibition module to capture the connection between cell-cell contacts and p110 expression could pave the way towards modeling interacting epithelial cell communities [106]. We thus see our current model as a seed for more powerful models of the processes that go awry when healthy cells transition into malignancy.
To capture the complex combinatorial logic by which the 87 molecular species and cellular processes in our model interact, we used a Boolean network modeling formalism [63]. Boolean models approximate the activity of regulatory molecular species as ON (expressed and active) or OFF (not expressed or inactive) [107], and focus on the combinatorial logic by which multiple regulatory inputs work together. This requires specifying the ON/OFF response of each node for every combination of the states of its inputs. The resulting Boolean functions (gates) can be represented as truth tables (input-output tables that specify every response of a node explicitly), or via the Boolean logic operators AND, OR, and NOT (S1 Table). Once the Boolean gate of each node is specified, the time-dependent dynamics of the whole network can be simulated from an arbitrary initial condition [63]. Since the expression / activity of the molecules is discrete, time also proceeds in discrete steps in which nodes can change their ON/OFF state.
In a Boolean representation, a regulator network can have 2N possible expression / activity profiles, where N represents the number of molecules in the model. Starting a time-series from most of these 2N states reveals that they are not stable, in that several regulatory nodes immediately change their ON/OFF state as dictated by the ON/OFF state of their inputs. Allowing the network’s dynamics to proceed from an unstable state will lead to a sequence of expression/activity changes that can cascade through the network. Eventually, every such cascade must end in two ways, regardless of update: 1) a stable state in which all Boolean rules are satisfied (called a point attractor), or 2) a more complex set of states that a) repeat in an exact cycle termed a limit cycle attractor under synchronous update, or b) repeat in a more stochastic sequence of states called a complex attractor that traps the dynamics under asynchronous update. The latter can also represent a rhythmic, repeating series of state-changes, but this is not guaranteed. The collection of sequential state-changes running from each of the 2N model states to the model’s attractor states or cycles can be represented as a directed network of states, termed the state transition graph.
Under synchronous update the model’s dynamics leads to a single attractor from each unstable state. The collection of all the paths leading to the same final state creates a subgraph of the state transition graph, and represents the attractor basin of the final attractor state [111]. Conceptually, this attractor basin can be thought of as a valley in the pseudo-energy landscape of the model [112]. As most network states are unstable and lead, in time, to an attractor, biologically relevant robust phenotypes of the model are expected to correspond to its attractor states [113]. Moreover, rhythmic biological behavior such as that of a continuously cycling cell is expected to map onto a limit cycle attractor.
Under asynchronous update, some unstable states can lead to different attractors with different probabilities depending on update order, while others can only lead to a single attractor—making the definition of attractor basins less straightforward. In addition, regions of the state transition graph can act as metastable “valleys” (S5 Fig). These represent state collections that trap the dynamics of the system for long periods of time, but it is not strictly speaking impossible for the system to escape to a proper attractor. Indeed, the cell cycle in our asynchronous models is such a metastable “pseudo-attractor”.
To simulate the dynamics of our Boolean model and work through key methods, see “SI_notebook.ipynb” in S2 File, a Jupyter Notebook in Python (https://jupyter.org). The code in this notebook uses BooleanNet [86] and NetworkX [67]. S3 File contains BooleanNet model files, including the full model (“PI3K_cell_cycle_apoptosis”). To convert these files to commonly used formats used by other packages, see (http://colomoto.org/biolqm/doc/formats.html).
For synchronous update, see S2 File -- 1.a for the PI3K oscillator and S2 File -- 3 for the full model. To run the PI3K oscillator module using general asynchronous update see S2 File -- 1.b; for the full model with random order asynchronous and biased asynchronous update see S2 File -- 4.a-c. To sample the full state space of the individual network modules, see S2 File -- 2; to sample and visualize the general asynchronous state transition graph of the PI3K oscillator, see S2 File -- 1.c; to map the cell cycle pseudo-attractor of the full model, see S2 File -- 4. Both files are available as a package at https://github.com/deriteidavid/cell_cycle_apoptosis_Sizek_etal_PloSCompBio_2019.
In order to generate a comprehensive picture of all the attractor basins of the model, we use a stochastic state space sampling procedure adapted from [114], as described in [11]. To this end, we first implemented a noisy version of synchronous Boolean dynamics, in which each regulatory node is affected by a small amount of noise in every time-step. The noise is implemented as a small probability pn = 0.02 that each node generates the incorrect output, rather than the one dictated by its inputs [112]. This noisy dynamics sets up a Markov process, guaranteeing that the system can spontaneously visits any state (not just the attractors) with non-zero long-term probability [112,115]. We used the noisy dynamics to aid our sampling procedure by starting the network from a random initial condition and simulating a time-course of Nseries = 20 noisy time-steps. As the model generates this noisy dynamical trajectory, the algorithm pauses at each state it visits to perform two checks. First, it finds the attractor basin this state would fall into if the dynamics were to continue in a deterministic fashion. Second, it scans the immediate neighborhood of this state by enumerating every state the system could reach from the current one via a single node-state flip and identifying their attractor membership (via deterministic dynamics). This allows the algorithm to access parts of the state space the noisy dynamics might never go near, and to find even small basins relatively fast. As a result, the algorithm is quite slow on random Boolean networks with large numbers of small basins. Our model’s robust phenotype-representing attractor basins, by contrast, are typically large and thus rapidly found. The full algorithm descried in [11] tracks the visitation probability of each state, basin and transition (not used here). The only update to the algorithm since [11] involves partitioning the full state space of the model into sub-spaces corresponding to each unique environmental node state-combination and sampling each subspace from Nrnd = 500 random initial conditions.
In order to automatically model the dynamical behavior of any isolated subgraph (Fig 3A), we have previously developed an algorithm that defines the Boolean gates of nodes when they lose some of their incoming connections [11]. The main goal of this algorithm was to optimally preserve the regulation of a node by its remaining inputs. Briefly, whenever a subset of inputs is removed from a Boolean gate, the algorithm assumes that they are frozen into either an ON or an OFF state. To best preserve the dynamical influence of the remaining nodes, it finds one of the 2k possible combinations of frozen inputs such that: a) all remaining input nodes are functional (i.e., they are able to impact the output in some way), and b) the entropy of the remaining Boolean gate fragment, HG = - p · log(p)- (1 - p) · log(1 - p), is as large as possible (p is the fraction of OFF-outputs). For easy reproducibility of our module networks, S3 File includes a BooleanNet model file for each module.
To generate model predictions in non-saturating growth factor conditions, we ran time courses of T = 50,000 or 500,000 time-steps in which the GFH input node was randomly toggled ON/OFF in each time-step with a tunable ON-probability pHigh_GF (Fig 5B) [11]. The ongoing simulations tracked the number of cell cycles completed without error (black cycle on S6 Fig), the number of genome duplication even from G2 (orange transition on S6 Fig), the number of premature metaphase-anaphase transitions that did not involve completion of the mitotic spindle followed by genome duplication (green transition on S6 Fig), the number of genome duplication events in the absence of a cytokinesis step between telophase and the next S-phase (red transition on S6 Fig), and the number of apoptotic events (purple transition on S6 Fig & other apoptotic events). Time courses that resulted in apoptosis before time T were restarted until a minimum of T steps of live-cell dynamics were sampled. In addition, the simulation tracked the average length of G1, S, G2, metaphase and telophase (the time cells spent with 2 nuclei, even if the cell cycle control network reset to G0/G1).
To generate model predictions with incomplete knockdown or overexpression of a target molecule, we combined the non-saturating stochastic growth factor inputs described above with a similar stochastic locking of the target molecule OFF or ON with a tunable probability pKD (knockdown) or pOE (over-expression), respectively. In time-steps where the molecule was not locked ON or OFF, it followed the internal Boolean regulatory influences of the rest of the network as if it was unperturbed. To run sample time courses, see S3 File -- 1.a; to sample cell cycle errors see S3 File -- 1.b.
|
10.1371/journal.ppat.1003637 | Genome-Wide Mouse Mutagenesis Reveals CD45-Mediated T Cell Function as Critical in Protective Immunity to HSV-1 | Herpes simplex encephalitis (HSE) is a lethal neurological disease resulting from infection with Herpes Simplex Virus 1 (HSV-1). Loss-of-function mutations in the UNC93B1, TLR3, TRIF, TRAF3, and TBK1 genes have been associated with a human genetic predisposition to HSE, demonstrating the UNC93B-TLR3-type I IFN pathway as critical in protective immunity to HSV-1. However, the TLR3, UNC93B1, and TRIF mutations exhibit incomplete penetrance and represent only a minority of HSE cases, perhaps reflecting the effects of additional host genetic factors. In order to identify new host genes, proteins and signaling pathways involved in HSV-1 and HSE susceptibility, we have implemented the first genome-wide mutagenesis screen in an in vivo HSV-1 infectious model. One pedigree (named P43) segregated a susceptible trait with a fully penetrant phenotype. Genetic mapping and whole exome sequencing led to the identification of the causative nonsense mutation L3X in the Receptor-type tyrosine-protein phosphatase C gene (PtprcL3X), which encodes for the tyrosine phosphatase CD45. Expression of MCP1, IL-6, MMP3, MMP8, and the ICP4 viral gene were significantly increased in the brain stems of infected PtprcL3X mice accounting for hyper-inflammation and pathological damages caused by viral replication. PtprcL3X mutation drastically affects the early stages of thymocytes development but also the final stage of B cell maturation. Transfer of total splenocytes from heterozygous littermates into PtprcL3X mice resulted in a complete HSV-1 protective effect. Furthermore, T cells were the only cell population to fully restore resistance to HSV-1 in the mutants, an effect that required both the CD4+ and CD8+ T cells and could be attributed to function of CD4+ T helper 1 (Th1) cells in CD8+ T cell recruitment to the site of infection. Altogether, these results revealed the CD45-mediated T cell function as potentially critical for infection and viral spread to the brain, and also for subsequent HSE development.
| Herpes simplex encephalitis (HSE) is a lethal neurological disease resulting from infection with Herpes Simplex Virus 1 (HSV-1). Previous studies have demonstrated a human genetic predisposition to HSE. However, the gene mutations that have been suggested as critical in protective immunity to HSV-1, exhibit incomplete penetrance and represent only a minority of HSE cases, perhaps reflecting the effects of additional host genetics factors. In order to identify new host genes involved in HSV-1 and HSE susceptibility, we have implemented the first genome-wide mutagenesis screen in an in vivo HSV-1 infectious model. Using this large-scale approach, we have identified a loss-of-function mutation in the Receptor-type tyrosine-protein phosphatase C (Ptprc) gene. Mice carrying this mutation were characterized by defects in thymic and B cell development. Following infection, these mutant mice exhibited hyper-inflammation in their brains stems caused by viral replication. Transfer of total lymphocytes from resistant into mutant mice resulted in a complete HSV-1 protective effect. Furthermore, T lymphocytes were the only cell population to fully restore resistance to HSV-1 in the mutants. These findings revealed the T cell function as potentially critical for infection and viral spread to the brain, as well as to subsequent HSE development.
| Herpes simplex virus type 1 (HSV-1) is a large, enveloped virus of the Herpesviridae family. Its 152 kilobase (kb), double-stranded DNA genome encodes more than 80 polypeptides [1]. HSV-1 is among the most prevalent and successful human pathogens [2] and is typically transmitted through intimate contact and exchange of bodily fluids, such as saliva. This virus causes a life long infection, which consists of two distinct phases: an initial lytic stage, followed by a shift to latency once it reaches sensory neurons. Periodically, reactivation from latency occurs and is associated with numerous diseases, ranging from the common cold sore to ocular herpetic stromal keratitis, a leading cause of infectious blindness [3] [4]. Reactivation events as well as primary infections are also associated with herpes simplex encephalitis (HSE), a rare but life threatening consequence of infection of the central nervous system (CNS) [5]. In most of cases, the virus reactivates in the olfactory bulb or trigeminal ganglia, enters the brain via a retrograde axonal transport, replicates into the CNS, causing acute inflammation and significant pathological damages (review, see [6]). HSE is caused by direct lytic effects of the virus on neurons and glial cells, but mostly, by collateral damage, such as the disruption of the blood-brain barrier (BBB), due to the accompanying inflammatory reaction and leukocytes homing to the brain [7] [8] [9]. If left untreated, HSE is lethal in nearly 70% of cases; despite treatment, debilitating sequelae frequently develop [5] [10] [11]. In developed countries, it remains among the most common causes of viral encephalitis [12].
Since HSE was discovered in 1941, it has remained unclear why only a small proportion of otherwise healthy individuals exposed to HSV-1 develop the disease. In 2003, an autosomal recessive mutation in the STAT1 gene was the first genetic etiology to HSE reported in HSV-1 seropositive patients [13]. More recently, loss-of-function mutations in the UNC93B1, TLR3, TRIF, TRAF3, and TBK1 genes have also been associated with HSE in otherwise healthy children [14] [15] [16] [17] [18] [19], demonstrating the critical role of the UNC93B-TLR3-type I IFN pathway in the outcome of childhood HSE. Indeed, the fibroblasts of these patients displayed an impaired production of IFN-α, IFN-β and IFN-λ following TLR3 stimulation. These fibroblasts are also highly susceptible to infection with HSV-1 and this phenotype has recently been shown to recapitulate the impairment of TLR3-dependent HSV-1 control in UNC-93B deficient neurons and oligodendrocytes [20]. However, the TLR3, UNC93B1, and TRIF mutations exhibit incomplete clinical penetrance [21], perhaps reflecting the effects of additional environmental or host genetics factors. Furthermore, these mutations affect only a small proportion of children with HSE. Altogether, these observations suggest that predisposition to HSE may result from a set of diverse single gene defects and indicate the likely existence of other anti-HSE pathways.
Mouse models have also provided an efficient way to identify host factors that contribute to susceptibility or resistance to HSE. In 1975, Lopez and colleagues were the first to demonstrate the contribution of host genetics to HSE pathogenesis using different inbred strains of mice with varying resistance to HSV-1 [22]. Subsequently, HSE mouse models have employed several viral and mouse strains, knock-out mice, as well as various routes of infections. It has been demonstrated that STAT1 knock-out mice (STAT1−/−) are susceptible to HSE [23] with an increased HSV-driven immune pathology in the cornea [24] and also in the brain stems [25]. Unc93b1-deficient mice have also been associated with HSE susceptibility despite their ability to control viral replication in the brain [26]. Altogether, these studies are consistent with the genetic etiologies of HSE reported in humans. Mice deleted for Myd88, which encodes for the adaptor protein of TLR7 and 9, were also highly susceptible to HSE [27]. In contrast to this finding, Honda et al. showed no increased susceptibility to HSE for Myd88−/− mice, whereas Irf7−/− had significantly increased mortality, correlating with reduced IFN-α level in the sera [28]. Like Myd88, TLR9−/− mice have been shown to be either more susceptible [29] or resistant [26] to HSE. The discrepancy in these studies may be due to different routes of infection or differences in viral strains. Nevertheless, these reports clearly demonstrate the key role of antiviral IFNs in mouse models of HSE. Several knock-out mice have also been used to study the immune control of HSE by NK, T or B cells. B cell-deficient mice were more susceptible to HSE [30] [31] whereas the role of T cells, in particular the role of CD8+ T cells, is still controversial [32] [33]. For example, athymic nude mice lacking T cells have been shown to be resistant to HSE similar to wild-type (WT) mice [34] while Metcalf et al. found them more susceptible to HSE [35]. Similar to that of T cells, the contribution of NK cells in HSE has been unclear [36] [37] [38]. Hence, the role of each of these immune cell populations remains elusive, as susceptibility to HSV-1 depends on both viral (viral strain, route of infection, and viral dose) and host factors (strain and age of the animal). These investigations have nonetheless paved the way for the study of the genetic etiology of HSE. However, no gene associated with HSE has thus far been identified by positional cloning in mice and host factors that contribute to susceptibility or resistance to this pathology also remain largely unknown.
N-Ethyl-N-Nitrosourea (ENU) mutagenesis constitutes an inherently unbiased, functional genomic strategy for the identification of genes, including those without known functions or biological precedents, regardless of their chromosomal location and expression pattern [39]. Treatment with this chemical mutagen introduces random point mutations in the mouse genome, with an average substitution rate of one nucleotide per megabase (Mb). In high throughput phenotypic screens, such mutations may lead to the identification of genes responsible for specific immune response defects and/or genes associated with diseases. Furthermore, positional cloning of mutant genes is now facilitated by recent progress in the field of genomics, such as exome sequencing and the availability of the mouse genome sequence. Thus, ENU mutagenesis has been used to successfully identify genes, proteins, and signaling pathways involved in a wide range of biological processes, including susceptibility to infection [40] [41], obesity [42], muscle development and function [43], cardiomyopathy [44], thrombocytopenia [45] and immunodeficiency [46]. This last report, describing a single-nucleotide mutation in the Unc93b1 gene that abrogates signaling via TLR3, 7 and 9, has largely contributed in the discovery of the human UNC93B1 mutations [14].
In this study, we took advantage of the large scale mouse ENU mutagenesis platform developed in our laboratory to systematically search for novel host genes that directly impact infection by HSV-1 and the development of HSE. This led the identification of the Ptprc gene as a novel host factor associated with HSV-1 susceptibility.
To identify novel host genes and fundamental pathways that directly impact HSV-1 infection and HSE pathology, ENU-mutagenized mice were screened for their susceptibility to HSV-1. Generation 0 (G0) C57BL/6J (B6) male mice were first treated with the mutagen ENU and then mated with C57BL/10J (B10, to facilitate subsequent genetic mapping) mice to produce G1 males, the founders of the colony which are heterozygous for ENU-induced mutations (Figure 1A). The G1 males were then backcrossed to B10 mice, to produce G2 animals. Two randomly chosen G2 daughters were backcrossed to their G1 father to produce G3 pedigrees, where homozygous ENU-mutations are expected to segregate in about 25% of animals. These G3 pedigrees were then infected i.p. with 1×104 pfu of HSV-1 strain 17. This dose led to lethal encephalitis in susceptible A/J mice, whereas B6 mice remained unaffected (Figure S1). Following infection, the ENU-mutagenized mice were monitored for two weeks; the pheno-deviant offspring that exhibited clinical signs or succumbed to the infection were considered susceptible. Over 1,000 G3 mice, corresponding to sixty-nine G1 males, were screened for their susceptibility to HSV-1 infection. As shown in figure 1B, most A/J and BALB/c mice succumb uniformly between five and eight days post-infection (p.i.). In contrast, WT B6 and B10 mice, as well as ENU-mutagenized G3 mice, are susceptible in less than 10% of cases, which is largely under the expected frequency of susceptibility in pedigrees segregating a recessive mutation (25%). Sixty-nine G1 pedigrees were screened for their susceptibility to HSV-1 infection. Of these, a pedigree called P43 appeared promising. Indeed, mating the G1 male with either of the two G2 daughters (G2a and G2b in Figure 1A) produced HSV-1 susceptible G3 offspring with over 20% of mortality upon infection (Figure 1B). These susceptible animals generally succumbed later than in A/J and BALB/c mice, between day 11 and 13 p.i. Moreover, no difference was observed between male and female offspring (data not shown).
To map the locus responsible for the HSV-1 susceptibility we carried out a genome-wide scan on 45 G3 mice, including 11 susceptible and 34 resistant animals. Linkage analysis identified a significant quantitative trait loci (QTL) signal on chromosome 1 (position 118.6 to 144.9 Mb) with a logarithm of odds (LOD) score of 6.7 (Figure 2A). The critical interval was associated with homozygous B6-derived alleles from the original mutated G0 male, except in the case of one animal that survived the infection (Figure 2B, see G3.45). In contrast, homozygosity for B10 alleles and B6/B10 heterozygosity correlated perfectly with resistance. These findings suggested that HSV-1 susceptibility in the P43 pedigree is due to a recessive mutation located between 118.6 and 144.9 Mb on chromosome 1.
This 26 Mb interval on chromosome 1 contains more than 250 genes. Thus, to identify the ENU induced mutation involved in our HSV-1 susceptibility phenotype, whole exome sequencing, which consists of selectively capture and sequencing the coding regions of the genome, was performed on genomic DNA from two P43-derived G3 mutants. The generated sequences were then compared to the WT B6 reference sequence, filtered against public and in house databases, and only genetic variations supported by a minimum coverage of 10 sequence reads were considered (Table 1). This represented a total of 116 mutations, of which 88 were homozygous and 13 were common to both sequenced P43 mutant mice. Two mutations were localized in the distal region of chromosome 1 (118.6–144.9 Mb) previously identified by linkage analysis. One of these two mutations was a non-synonymous mutation I543F in the Kcnt2 gene, which encodes for a protein belonging to the voltage-gated potassium channel complex. The structure of Kcnt2 is probably not affected by the ENU-induced mutation since I and F residues are both hydrophobic, neutral and non-polar amino acid. However, the other mutation is likely to have a significant impact at the protein level: a unique A to T transversion in exon 1 (chromosome 1∶140,071,651) of the Receptor-type tyrosine-protein phosphatase C (Ptprc) gene, which generates the premature stop codon L3X (Figure 2C). The Ptprc gene is expressed on all nucleated hematopoietic cells and encodes for the multiple forms of the tyrosine phosphatase receptor CD45, including the isoform B220 present on B cells. CD45, when analyzed with the Gene Ontology (GO) database, is statistically enriched for “response defense to virus” and thus represents the best candidate gene to explain the HSV-1 susceptibility observed in this pedigree. First, the L3X mutation in Ptprc was confirmed by Sanger sequencing of all affected (n = 11) and unaffected individuals (n = 34) of the P43 pedigree. As shown in Figure 2D, the susceptibility to HSV-1 infection was associated with PtprcL3X genotype in more than 90% of cases (11/12), whereas HSV-1 resistance segregated with PtprcL3X/+ and Ptprc+/+ littermates. Furthermore, previously described Ptprc null mice, that lacked expression of all endogenous CD45 isoforms due to a targeted deletion on exon 9 (Ptprcex9−/−) [47], were susceptible as homozygous PtprcL3X animals (Figure 2D). Most notably, Ptprcex9−/− mice succumbed within the same time-frame than our ENU-induced mutant mice, providing independent evidence that PtprcL3X mutation determines the vulnerability to HSV-1 observed in pedigree P43. FACS analysis demonstrated that about 70% of peripheral blood mononuclear cells (PBMCs) from Ptprc+/+ mice were stained with antibody against B220, whereas no signal was observed in PBMCs from mutant PtprcL3X littermates (Figure 2E). A similar absence of B220+ cell compartment was also observed in the spleen of PtprcL3X mice (data not shown). These results indicated that PtprcL3X is a null allele resulting in complete absence of protein. In the following sections, PtprcL3X/+ were used as controls since these heterozygous littermates phenocopy Ptprc+/+ mice in their response against HSV-1.
To determine a possible impact of the infection route on the survival phenotype, PtprcL3X mice and heterozygous littermates were infected intranasally (i.n.) with 5×103 pfu of HSV-1 strain 17. This dose led to lethal encephalitis in susceptible A/J mice by day 6–7 post-infection, whereas all B6 mice survived (Figure S2). In this infectious model, more than 60% of PtprcL3X mice succumbed around the same time (by day 11–13 p.i.) as those infected i.p., while heterozygous littermates survived in more than 85% of cases (p-value = 0.003) (Figure 3B). Furthermore, i.p. infection with the McIntyre or F strains of HSV-1 also resulted in an increased susceptibility in PtprcL3X mice only (Figure 3C and 3D, respectively), despite the fact that, relative to strain 17, the McIntyre and F strains appear less virulent in our experimental model (compare survival curves in Figure 2D, 3C and 3D). Thus, the association between our HSV-1 susceptibility phenotype and the PtprcL3X genotype is independent of the infection route and HSV-1 strain.
To better characterize the pathogenesis in our mouse model, we followed the kinetics of viral replication in the brain and periphery (spleen and liver) following i.p. infection. At days 3, 6, and 10 p.i., the spleen and liver from both PtprcL3X/+ and PtprcL3X mice had undetectable levels of virus (data not shown). In contrast, the brain tissue from two out of five PtprcL3X mice showed significant viral titers at day 10 p.i. (Figure 4A), which is consistent with results presented in Figure 2D, where a large number of these mice succumb at days 11–13 p.i. Otherwise, the brains from all PtprcL3X/+ mice had undetectable viral titers, correlating with their survival. To confirm this result, we carried a second set of experiments where mice were weighed twice daily and their brain stems were harvested if the mice lost at least 15% of their pre-infection weight. The brain stems were chosen instead of whole brains because STAT1−/− mice, which are also susceptible to HSV-1, have been shown to exhibit a high viral titer and upregulate a large number of inflammatory markers in their brain stems following HSV-1 infection [25]. As a measure of virus load, we determined the expression of the ICP4 viral gene in the brain stems collected from PtprcL3X/+ and PtprcL3X mice using real time quantitative PCR (qPCR). In these conditions, ICP4 expression in the brain stems from seven out of nine PtprcL3X mice was increased by more than 20-fold relative to those from PtprcL3X/+ mice (Figure 4B). The same samples were used to determine the transcript level of MCP1, IL-6 and two matrix metalloproteases (MMP-3 and -8) that are highly expressed during HSE and can also cause the disruption of the blood-brain barrier (BBB) [48] [49] [50] [51] [52] [53] [54] [25]. The expression of these targeted transcripts was significantly increased in PtprcL3X mice (Figure 4C, upper panels), accounting for hyper-inflammation and pathological damage in the brain of susceptible mice. Most importantly, there is a strong correlation between the presence of HSV-1 and the high expression of MCP1, IL-6, MMP3, and MMP8 in the brain stems of PtprcL3X mice (Figure 4C, lower panels). This experiment was also performed on IL-1β, TNF-α, CCL-3, CCL-4 and CCL-5 genes, and gave the same results (Figure S3A). However, the expression of the β-actin gene in PtprcL3X infected mice was equivalent to that of PtprcL3X/+ infected mice (Figure S3B), suggesting that the increased expression seen in the brain stems of PtprcL3X mice results from an overwhelming inflammatory response. Altogether, these findings demonstrate that the HSV-1 susceptibility observed within this pedigree is associated with profound CNS inflammation, suggesting BBB lesions, both caused by uncontrolled viral replication in the brain.
We have shown that HSV-1 susceptibility is associated with the presence of virus and profound inflammation in the CNS. This suggests that the immune response of susceptible mice could be affected, limiting their ability to control viral spread from the periphery into the CNS. To this end, the blood and spleen from PtprcL3X and heterozygous littermates were collected and specific cell populations were examined by FACS. Of note, there was no significant difference in the total number of cells between PtprcL3X and heterozygous littermates for these tissues (data not shown). Thus, FACS data are represented as a percentage of total cells. FACS analysis in PtprcL3X mice revealed a lack of the CD3+ T cell compartment in the spleen, while heterozygotes showed normal percentages (Figure 5A). A similar absence of the CD3+ T cell compartment was also observed in the blood of PtprcL3X mice (data not shown). Notably, DX5+/CD3− NK cells from PtprcL3X mice were increased two-fold relative to heterozygotes (Figure 5B), consistent with a study on mice defective for CD45 [55]. The percentage of CD19+ B cell and CD11c+/MHCII+ dendritic cell compartments from PtprcL3X mice was unaffected (Figure 5C and data not shown). However, the IgD/IgM staining showed a reduced proportion of the mature follicular B CD19+/IgDhi/IgMlo cells in mutants, suggesting defects in the final stage of B cell maturation (Figure 5D).
To gain insight into the absence of CD3+ T cells in the periphery, thymuses from PtprcL3X and heterozygous littermates were collected and thymocytes were first stained for CD4 and CD8. As above, thymocytes from both PtprcL3X and heterozygous littermates were present in equal numbers (data not shown). Consistent with results presented in Figures 5A, CD4+ and CD8+ single positive (SP) thymocytes were not present in PtprcL3X mice (Figure 5E). However, CD4−CD8− double negative cells (or DN) from PtprcL3X mice were 4-fold higher relative to heterozygous littermates, suggesting a block at the DN stage. Then, the DN thymocytes were analyzed on the expression profiles of CD25 and CD44, two receptors expressed during T cell development. Thymocytes from PtprcL3X mice displayed an accumulation of immature CD4−CD8− DN3 cells (CD25+CD44−) and a reduced number of DN4 (CD25−CD44−) cells relative to heterozygous littermates (Figure 5F); these last results revealed a partial block at the β selection step of TCR-β rearrangement. These findings demonstrate that the Ptprc L3X mutation drastically affects the early stages of T cell development and could explain the HSV-1 susceptibility observed in the P43 pedigree.
We have shown that the L3X mutation in Ptprc leads to a lack of the CD3+ T cell compartment and affects B cell maturation. However, CD45 encoded by Ptprc has also been shown to play an important role in NK cell function, even if the phenotypic and functional consequences of CD45 deficiency are less severe than that of T cells [56]. Thus, through in vivo experiments, we attempted to establish which of the affected immune cell populations might be involved in the susceptibility to HSV-1 observed in PtprcL3X mice. First, PtprcL3X and heterozygous littermates were infected with HSV-1 before total splenocytes were harvested at day 7 p.i. This step is needed to prime the immune response. These splenocytes were then transferred into either PtprcL3X or heterozygous littermates, before these mice were themselves infected with HSV-1 (Figure 6A). As expected, the transfer of total splenocytes from PtprcL3X into PtprcL3X mice did not rescue these mutant mice from lethal HSV-1 infection, while PtprcL3X/+ mice receiving splenocytes from either PtprcL3X/+ or PtprcL3X mice survived (Figure 6B). On the other hand, the same cell transfer from PtprcL3X/+ into PtprcL3X mice resulted in a complete HSV-1 protective effect. These results show that transfer of total spleen cells from PtprcL3X/+ mice to PtprcL3X mice provide full protection against HSV1 infection. Next, we aimed to establish more precisely which immune cell population(s) is critical for the control of HSV-1 infection. In order to determine the possible role of NK cells in this protection, PtprcL3X/+ mice were treated with anti-asialo GM1 antibody, which is known to completely deplete NK cells [57] (Figure S4), and then infected with HSV-1. As shown in Figure 6C, the treatment with anti-asialo GM1 antibody had no impact on the survival of PtprcL3X/+ mice, suggesting that NK cells do not play a protective role against HSV-1. To distinguish the involvement of B cells from that of T cells in the protection against HSV-1, cells from PtprcL3X/+ mice were magnetically sorted and the individual sub-populations transferred into PtprcL3X mice. While the transfer of B cells failed to rescue PtprcL3X mice from lethal infection, the transfer of CD3+ T cells led to total protection against HSV-1 in all PtprcL3X mice (Figure 6D). To distinguish the relative contribution of CD8+ from that of CD4+ T cells in the HSV-1 protective effect, we carried a third set of transfer experiments where PtprcL3X mice were either injected with CD8+ or CD4+ T cells. The transfer of CD8+ T cells rescued about 50% of PtprcL3X mice from lethal infection (Figure 6E). In contrast to CD8+ T cells, the transfer of CD4+ T cells led to a total protection against HSV-1 in almost all (9/10) PtprcL3X mice (Figure 6E). Altogether, these data suggest that both the CD8+ and CD4+ T cells are involved in the HSV-1 protective effect, even if the CD4+ T cells appear to be more important than CD8+ T cells. However, it should be noted that a small proportion of CD8+ T cells (∼5%) were usually still present after the CD4+ T cell enrichment (data not shown). Thus, to demonstrate the strict dependence of CD8+ T cells in the CD4-mediated protection against HSV-1, we performed further transfer experiments where CD4+ T cells were purified from B6.H2-DbKb knock-out mice, whose H2-DbKb make them depleted in CD8+ T cells. In this condition, transfer of CD4+ T cells failed to rescue PtprcL3X mice from lethal infection (Figure 6E). This last result demonstrates that CD4+, without CD8+ T cells, are incapable of establishing an efficient immune response against HSV-1. Moreover, B6.H2-DbKb knock-out mice were susceptible to lethal HSV-1 i.p. infection (p-value = 0.003) and succumbed around the same time (by day 11–13 p.i.) as PtprcL3X mice infected i.p. (Figure S5), thus confirming the crucial role of the CD8+ T cells in protective immunity to HSV-1.
To better characterize the contribution of the CD4+ T cells in CD8+ T cell function, we performed additional FACS experiments. PtprcL3X mice were again transferred with CD4+ T cells purified from PtprcL3X/+ mice (same proportion of CD8+ T cells, as mentioned above, were still present after MACS separation, data not shown), and were then infected i.p. with HSV-1. Peritoneal cells, which in this infectious model are the first targets of HSV-1, were collected at days 2 and 7 p.i., gated on CD45.2+CD3+, and then analyzed for their CD4 and CD8 expressions. Here, we took advantage of the fact that PtprcL3X mice do not express any CD45 receptors, including the isoform CD45.2 expressed by all leukocytes, to distinguish endogenous from transferred cells in the context of PtprcL3X mice (Figure 7A). PtprcL3X/+ infected mice (PBS injected, positive control) showed about 35.5% of CD45.2+CD3+ cells in the peritoneal zone at day 2 and 32.7% at day 7 p.i., whereas PtprcL3X infected mice (transferred with WT CD4+ T cells) exhibited 8.21% of CD45.2+CD3+ at day 2 and 6.6% at day 7 p.i. (Figure 7B, left panel). In contrast, PtprcL3X infected mice (PBS injected, negative control) as well as those transferred with WT CD4+ T cells, but non-infected, did not show any T cells, suggesting that the recruitment of T cells to the peritoneal zone is infection-dependent. At day 2 p.i., we observed a high proportion of CD4+ T cells (>95% of total CD3+ T cells) in both PtprcL3X/+ (PBS injected) and PtprcL3X (transferred with WT CD4+ T cells) infected mice. At day 7 p.i., however, the frequency of CD8+ T cells raised to 50% in equal proportions to CD4+ T cells, indicating CD8+ T cell recruitment to the site of infection (Figure 7B, right panel). Furthermore, this increased CD8+ T cell proportion perfectly correlated with the viral clearance (Figure 7B, right panel). Altogether, these results suggest that CD4+ help in mobilizing effector CD8+ T cells to the site of infection, which contributes to proper control of the dissemination of HSV-1 into the CNS. Recently, Nakanishi et al. demonstrated the role of CD4+ T cells, through their secretion of IFN-γ, in mobilizing effector CD8+ T cells to local HSV-2 infection [58]. Thus, we asked whether CD4+ T cells could mediate CD8+ T cell recruitment by the action of IFN-γ. To this end, IFN-γ knock-out mice (IFN-γ−/−) were infected, their peritoneal cells were collected at day 7 p.i. and the proportion of CD8+ relative to the CD4+ T cells was compared to infected PtprcL3X/+ mice. As shown in Figure 7C, IFN-γ−/− mice were significantly affected in their CD8+ T cell recruitment relative to the PtprcL3X/+ mice (p-value = 0.01 between PtprcL3X/+ and IFN-γ−/− CD8+ T cells). Furthermore, the CD4+ T cell transfer from infected IFN-γ−/− to PtprcL3X mice did not rescue all of these mutant mice from lethal HSV-1 infection, suggesting that IFN-γ production by the CD4+ T cells is important, however, not indispensable (Figure 7D).
HSE is a lethal neurological disease resulting from infection with HSV-1. Previous studies have demonstrated a human genetic predisposition to HSE, involving the UNC93B-TLR3-type I IFN pathway in protective immunity to HSV-1. However, these gene mutations exhibit incomplete penetrance and represent only a minority of HSE cases, perhaps reflecting the effects of additional host genetics factors. Therefore, several groups have used mouse forward genetics to identify loci associated with susceptibility/resistance to HSE [21]. For example, Resistance to Herpes Simplex virus type 1 (Rhs1), a NK complex-linked locus present on chromosome 6, has been shown to be essential for the control of both acute and latent HSV-1 infection [59]. Another locus on the same chromosome, Herpes Resistance Locus (Hrl), has also been identified as a factor influencing HSV-1 infection and HSE pathology in mice [60]. More recently, a genome-wide linkage study has been performed in rats leading to the identification of the calcitonin receptor as a candidate gene for regulation of susceptibility to HSV-1 [61]. Thus, such in vivo approaches have illuminated natural variation in host components that contribute to CNS pathology in response to HSV-1 infection, notably by identifying loci associated with susceptibility/resistance to HSE; but no gene controlling resistance, however, has yet been identified by forward genetics.
In order to identify new genetic and cellular mechanisms involved in HSV-1 and HSE susceptibility, we have implemented the first genome-wide mutagenesis screen in an in vivo HSV-1 infectious model. Our high throughput approach allowed for the characterization of sixty-nine mouse pedigrees screened for their susceptibility to HSV-1, leading to the identification of our first susceptible mutant. Whole exome sequencing revealed two mutations, which are localized in the distal region of chromosome 1 (118.6–144.9 Mb) previously identified by linkage analysis. One of these two mutations is a non-synonymous mutation I543F in the Kcnt2 gene, which encodes for a protein belonging to the voltage-gated potassium channel complex. Based on its high expression in the brain (http://biogps.gnf.org/), Kcnt2 appeared promising, notably in the context of the CNS and viral encephalitis. Moreover, a gain-of-function mutation in the Kcnt1 gene, a gene whose function is closely related to that of Kcnt2, has been recently associated with a childhood epileptic syndrome called malignant migrating partial seizures in infancy (MMPSI) [62]. However, and in contrast to the Ptprc L3X mutation, the Kcnt2 I543F mutation is predicted to be benign when analyzed with the PolyPhen-2 database. Nevertheless, additional functional assays would be required to determine if Kcnt2I543F has any role in the infectious process. Indubitably, the large number of immunological mechanisms associated with Ptprc (http://amigo.geneontology.org/) as well as the HSV-1 susceptibility phenotype of Ptprcex9−/− mice provide strong evidence that PtprcL3X is causative of HSV-1 susceptibility observed in the P43 pedigree.
PtprcL3X presents distinctive features compared to other reported mutations within the Ptprc gene. First, the PtprcL3X null mutation arose on a pure B6 chromosome. Other null mutations have been introduced in ES cells of 129 origin using a neomycine cassette targeting either exon 9 (Ptprcex9−/−) or exon 12 (Ptprcex12−/−) [47], [63]. Therefore, ptrpc mutations in mice Ptprcex9−/− and Ptprcex12−/− remain surrounded by 129-derived genetic material even after extensive backcrossing to B6. Ptprc is embedded in a polymorphic gene-rich chromosomal region that could influence immune, infectious or developmental phenotypes in knock-out mice. In comparable experiments, PtprcL3X presents defects very similar to knock-out mice, including severely reduced numbers of T cells due to a defect in the transition from stage of development DN3 to DN4. PtprcL3X is a null mutation and represents an excellent complement to two additional ENU-induced alleles, which were previously hypomorphic. PtprcLoc and PtprcLight were found in flow cytometric screens for recessive blood cell abnormalities. PtprcLoc truncates the cytoplasmic domain of most CD45 isoforms but retains about 4% of wild-type CD45 expression [64]. PtprcLight retains 15% of CD45 wild-type expression despite an F503S substitution in the transmembrane domain of the protein [65]. These allelic series in combination with Ptprc−/− mice has been used to demonstrate that the level of CD45 expression determines the dual role of CD45 in TCR signaling or T cell maturation. Now, the availability of PtprcL3X mutant mice should allow a more accurate analysis of the full spectrum of CD45 functions.
In this study, we demonstrated that the HSV-1 susceptibility observed in our pedigree was independent of the infection route and was not viral strain-specific, two factors that are the leading causes of disparities observed in different mouse studies for HSV-1 [21]. We also report that this phenotype is associated with HSE pathogenesis characterized by a significant viral replication and a dramatic increase in the expression of proinflammatory mediators (MCP1 and IL-6) and enzymes (MMP3 and MMP8) in the brain stems. MCP1 and IL-6 have known implications in leukocyte recruitment and viral clearance in the brain. However, the modulation in their production is crucial to properly control the dissemination of viruses in the CNS without sustaining inflammation, which could result in significant neurological damage including the disruption of the blood-brain barrier (BBB). The BBB is a physical barrier between the peripheral circulation and the CNS [66] which contributes to prevent the exacerbation of leukocytes homing into the brain. In experimental mouse models of HSE and human cases, early findings demonstrate that this pathology leads to vascular alterations with disruption of the BBB [67] [68]. Other viral pathogens, such as human immunodeficiency virus 1 (HIV-1), human T-cell leukemia virus 1 (HTLV-1), lymphocytic choriomeningitis virus (LCMV), West Nile virus (WNV), rabies virus and mouse adenovirus type 1 (MAV-1), are also known to cause BBB disruption [54]. In most cases, the BBB damage is caused by indirect effects of viral replication in the CNS with the exception of HIV-1, where Tat, gp120 and Nef viral proteins have been directly involved. When highly expressed, MCP1 alters the actin cytoskeleton and localization of tight junction proteins in the brain endothelium, disrupting the BBB [69] [49]. Thus, MCP1 has been suggested to be a major determinant of brain damage after infection in mice [70], [71] and its high expression has been also correlated with HSE development [48]. Likewise, MMP3 and MMP8 are keys mediators of tight junction protein alterations, which degrade extracellular matrix proteins at the BBB leading to its disruption. Although IL-6 is protective against lethal HSV-1 ocular infection [51], elevated levels may be neurotoxic and also increase the permeability of the BBB [51] [50] [52] [53]. The high expression of MCP1, IL-6, MMP3 and MMP8 observed in the brain stems of PtprcL3X mice are comparable to those reported by several studies evaluating the BBB integrity. While further investigations are still required to evaluate neurological damages, in particular the disruption of the BBB, in the brains of PtprcL3X mice, these gene signatures nevertheless highlight the CNS inflammation during HSV-1 infection. Moreover, this demonstrates that our infectious experimental conditions represent a robust model for the identification of host factors that contribute to susceptibility or resistance to HSE. We were also able to establish a significant correlation between the presence of HSV-1 in the brain stems of mutant mice and the expression of MCP1, IL-6, MMP3, and MMP8. These last findings suggest that the post-HSV-1 infection mortality observed in this pedigree can be explained by a profound CNS inflammation caused by viral replication in the brain stem.
The first role demonstrated for CD45 was the regulation of T and B cell antigen receptor signaling via dephosphorylation of the Src family kinases (SFKs) Lck and Lyn, respectively (for a review, see [56]). In this study, we showed that the PtprcL3X mutation drastically affects not only the early stages of thymocyte development but also the final stage of B cell maturation, consistent with previous studies that have used CD45 knock-out mice (for T cells, see [47], [63], [72], [73]; for B cells, see [74]–[76]). SFKs have been identified as its primary target, but CD45 is now known to also modulate the Janus kinases (Jaks) [77], as well as TLR signaling pathways [78] and NK receptor function [79]. CD45 functions have also been shown in mast cells, macrophages, and dendritic cells, as well as in leukocyte adhesion and migration [56]. While further experiments are still required to fully characterize our mutant in the several immune functions known for CD45, we have strong evidence that deficient T cell function largely contributes to HSV-1 susceptibility in mice. Indeed, we showed that the transfer of total splenocytes from PtprcL3X/+ into PtprcL3X mice resulted in a complete HSV-1 protective effect; T cells were the only cell population to fully restore resistance to HSV-1 in the mutants, an effect that requires both the CD4+ and CD8+ T cells.
We also demonstrate that the presence of CD4+ T cells is a prerequisite for the CD8-mediated full protection against HSV-1. How CD4+ T cells provide help to CD8+ T cells – during priming, memory and/or mobilizing to the site of infection – remains a pending question, which will require further analysis. The critical role of the CD4+ T helper cells in the activation of the CD8+ T cells during the primary response to HSV-1, particularly in cytotoxic T lymphocyte priming, has been already demonstrated [80], [81]. In this study, we showed that the absence of IFN-γ leads to a partial but not absolute block in CD8+ T cells recruitment to the site of infection, consistent with the survival curve obtained from the CD4+ T cell transfer from IFN-γ−/− to PtprcL3X mice. It should also be noted that the susceptibility of IFN-γ−/− mice to HSV-1 is still very controversial, as attested by different reports showing either susceptibility [82] or resistance [33], [83]. Using our infectious experimental conditions, IFN-γ−/− mice are resistant to HSV-1 as WT littermates (data not shown). Nevertheless, this finding is not necessarily in contradiction with our current data, but suggest that the only absence of IFN-γ is not sufficient to render resistant mice susceptible to lethal HSV-1 infection. Our findings also correlate with a study demonstrating the critical role of CD4+ Th1 cells after genital HSV-2 infection by timely recruitment of CD8+ T cells to the site of infection [58]. In this last publication, however, the critical role of CD4+ Th1 cells in CD8+ T cell mobilization to the site of infection for host survival was not established. Collectively, these results could also explain why the CD8+, without CD4+ T cells, only protected 50% of infected PtprcL3X mice. Further investigations are needed to determine if other CD4+ T cell subset(s) (T regulatory cells, Th2, Th17…), and also which of the Tc1, Tc2 and Tc17 CD8+ T cell subset(s), contribute to rescue of PtprcL3X mice from lethal HSV-1 infection. While we demonstrate here the critical role of both CD4+ Th1 and CD8+ T cells in protective immunity to HSV-1 using an unnatural inoculation route, we can reasonably propose that this adaptive immune mechanism also helps to control HSV-1 and maintain its latency in the CNS, as already been suggested by others [84], [85]. Thus, future studies to understand the key players in CD8+ T cell responses and their dependence on CD4+ T cell help should allow the generation of new prophylactic treatments against HSV-1 infection.
As mentioned above, CD45 plays a critical role in NK cell function and is required for protection from cytomegalovirus infection [79]. In contrast, CD45 deficiency protects mice from the lethal cardiomyopathy caused by Coxsackievirus B3 infection [77]. Likewise, others have demonstrated that intermediate levels, but not complete knock-out, of CD45 reduced apoptosis and protected mice from Ebola and B. anthracis infections [86], [87]. Altogether, while it is known that pathogens elicit divergent immune responses in mammals, our data are consistent with previous reports using phylogenetically distant pathogens whose control converges at CD45-mediated process. However, it remains to be determined if the varied effects of CD45 in the different infectious models reflect the multiple roles of CD45 in the establishment of an efficient immune response against pathogens, as it acts at different levels of both innate and adaptive immunity.
Since the main role of CD45 in lymphocytes is to promote cell activation, genetic alterations may lead to either lymphocyte hyper- or hypo-responsiveness. Thus, most studies performed in humans have focused on polymorphic variants that alter CD45 isoform expression, which are often associated with disease susceptibility [56], [88]. Absence of CD45 is routinely associated with severe combined immune deficiency (SCID). To this day, no mutation in Ptprc has been reported in a cohort of patients who suffer from HSE. Nevertheless, our findings do not necessary contradict current human data. One possibility is that Ptprc is not subject to specific genetic variation in human HSE, but remains mechanistically important for virus-host interactions. Alternatively, other alterations affecting T cell function could be associated with future human cases of HSE. For example, a strong correlation between decreasing CD4 count and increasing rates of HSV reactivation has been shown in individuals co-infected with HIV and HSV-2, suggesting that reactivation is linked to immunosuppression [89]. As mentioned in the introduction section, an autosomal recessive mutation (called 1757–1758delAG) in the STAT1 gene was the first genetic etiology for HSE reported in HSV-1 seropositive patients [13]. This mutation has been associated with combined deficiencies in IFN-α/β, -λ and -γ signaling pathways since they rely on either STAT1/2 heterodimers (IFN-α/β and -λ) or STAT1 homodimers (IFN-γ). While the cellular phenotype of impaired STAT1 activation by IFN-α/β has been related to HSE susceptibility, we cannot exclude the possibility that an impaired response to IFN-γ in the infants also contributed to this disease, as documented for other viral susceptibilities [90], [91].
In sum, our screen for HSE identified a new CD45 null-allele causing susceptibility to HSV-1 infection. ENU mutagenesis is thus an effective tool for identifying potential causes of HSE. Characterization of susceptibility mutations in additional pedigrees should provide important information to rationalize alternative therapeutic strategies for a devastating, yet potentially preventable, disease.
All mice were maintained under pathogen free conditions and handled according to the guidelines of the Canadian Council of Animal Care. The experimental protocol (Protocol number 4792) was approved by the McGill University ethics committee.
C57BL/6J (B6), C57BL/10J (B10), A/J, BALB/c and IFN-γ knock-out mice were purchased from the Jackson laboratories (Bar Harbor, Maine, USA). ENU-mutagenized mice were bred in the animal facility of the Goodman Cancer Centre, McGill University. Ptprcex9−/− and B6.H2-DbKb knock-out mice were kindly provided by Dr. Denis R. Alexander and Dr. Hidde L. Ploegh, respectively; B6.H2-DbKb knock-out mice possess targeted deletions in the H2-D and H2-K genes and are depleted in CD8+ T cells. HSV-1 strain 17 was originally from the laboratory of Dr. Subak-Sharpe whereas F and McIntyre strains were purchased from ATCC; these viral strains were amplified and titrated on Vero cells as previously described [92].
Eight week-old G0 B6 male mice were mutagenized by i.p. injection of a fractionated dose of 3×90 mg/kg of ENU (Sigma) at weekly intervals. Efficiently mutagenized G0 males transiently lost and then regained fertility after 13 weeks. They were then out-crossed with B10 female mice to produce G1 offspring: these mice were F1 hybrids carrying one full set of mutagenized chromosomes and one full set of WT chromosomes. Individual G1 males were bred again to B10 females to generate G2 animals. Each G2 daughter inherits 50% of the B6 sequence variants in the G1. Two G2 females were then backcrossed to their G1 fathers to generate G3 animals, which were screened for their susceptibility to HSV-1 strain 17 infection.
Mice 7 weeks of age or older were each infected intraperitonealy (i.p.) with 1×104 plaque forming units (pfu) of HSV-1. When indicated, 4 week-old mice were each infected intranasally (i.n.) with 5×103 pfu under anesthesia (5 mg/ml ketamine and 15 mg/ml xylazine). In this inoculation model, 20 µl of virus suspension was dripped directly into the nasal cavity in both nostrils of the mice. Infected mice were monitored 1–2 times daily over a 2-week period for survival. Mice that succumbed to infection within 14 days post-infection (p.i.) were considered susceptible to HSV-1. A/J and BALB/c mice were used as susceptible controls, while B6 and B10 mice were used as resistant controls. Organs of interest were collected from mice and then homogenized before being serially diluted 10-fold in non-supplemented DMEM and plated on Vero cells for plaque forming assays as previously described [92].
Genomic DNA was isolated from tail biopsy by a standard phenol/chloroform extraction, as previously described [93]. Genome scanning was performed at the McGill University and Genome Quebec Innovation Centre (Montreal, Québec, Canada). DNA samples from 11 resistant and 34 susceptible G3 mice from the P43 pedigree were analyzed with a panel of 255 B6/B10 polymorphic markers (SNPs) distributed across the genome [94] using massArray platform from Sequenom. Mapping data were analyzed using the R/qtl software, version 2.12.2. The binary model was used and LOD scores were calculated using survival as the phenotype.
Statistical analyses were conducted with the program R; linkage was performed with the package “R/qtl,” version R 2.12.2. The scanone function of the R/qtl library was used to perform maximum likelihood interval mapping (EM) of the phenotype on genetic markers. Significance values were evaluated with 10,000 permutations. The logarithm of the odds (LOD) support interval of QTL peaks was calculated using a 1.5 LOD drop by R/qtl. p-values are a result of two-tailed t-tests and R square (R) values are a result of linear regression tests.
All graphs represent the mean, and include error bars of the standard deviation.
Genomic DNA samples from two affected individuals were processed at the Centre National de Génotypage (Evry, France). Exome capture was performed using a SureSelect Mouse All Exon kit (Agilent Technologies, USA) and parallel sequencing on an Illumina HiSeq 2000 (100-bp paired end reads). This generated over 8 Gb of sequence. Reads were aligned to mouse genome assembly July 2007 (NCBI37/mm9) with Burrows-Wheeler Alignment (BWA) tool [95] and coverage was assessed with BEDTools, showing an average of 58.9 reads covering each base of the consensus coding sequence genes for the mouse genome [96]. Single nucleotide variants and short insertions and deletions (indels) were called using samtools pileup and varFilter [97] with the base alignment quality adjustment disabled, and were then quality filtered to require at least 20% of reads supporting the variant call. Variants were annotated using both Annovar [98] and custom scripts to identify whether they affected protein coding sequence, and whether they had previously been seen in mouse dbSNP128 or in any of 2 mouse exomes sequenced in parallel. In order to detect splice site mutations, the threshold of detection was increased to 6 bps instead of the standard 2 bps flanking exons.
Brain stems were harvested from sacrificed mice and immediately homogenized in 2 ml of trizol (Invitrogen). Total RNA was extracted using RNeasy columns (QIAGEN) and transcribed into cDNA using M-MLV with random hexamers (Invitrogen), according to the manufacturer's instructions. qPCR was performed using Platinum SYBR Green SuperMix-UDG (Invitrogen) together with 50 ng of reverse transcribed total RNA and experimental or control primers. Experimental primers targeting the MMP3 and MMP8 genes were designed to span exon junctions using primer3plus. For MMP3, the sequences used were 5′TGGAGATGCTCACTTTGACG3′ and 5′GCCTTGGCTGAGTGGTAGAG3′, while 5′TTTGATGGACCCAATGGAAT3′ and 5′GAGCAGCCACGAGAAATAGG3′ were used to quantify MMP8 expression, 5′CACCACTGCCCTTGCTGTTCT3′ and 5′ ACACCTGGCTGGGAGCAAAG3′ for CCL-3, 5′TCTGCGTGTCTGCCCTCTCTC3′ and 5′GGCTTGGAGCAAAGACTGCTG3′ for CCL-4, 5′GATCTCTGCAGCTGCCCTCAC3′ and 5′CACACACTTGGCGGTTCCTTC3′ for CCL-5, 5′CAGGCAGGCAGTATCACTCA3′ and 5′AGGTGCTCATGTCCTCATCC3′ for IL-1β, 5′AGGTCATCACTATTGGCAACG3′ and 5′ATCTCCTTCTGCATCCTGTCA3′ for β-actin. The primers used to determine the expression of MCP1, IL-6, TNF-α and the ICP4 viral gene have been previously described ([99] and [100], respectively). Target transcripts were normalized to the control housekeeping gene hprt. Reactions were performed in duplicate using the PTC200 Thermal Cycler with Chromo4 Continuous Fluorescence Detector (MJ Research). Expression was analyzed using Opticon Monitor 3 software (MJ Research). Relative mRNA expression levels were analyzed using the infected B6 samples as the reference group and calculated by subtracting the mean ΔCT of infected B6 samples from the ΔCT of the infected samples (ΔΔCT). The amount of target mRNA, normalized to the endogenous reference, was calculated as 2−Δ(ΔCT).
Cells were isolated from spleen, blood or thymus as previously described [101]. Peritoneal cells were collected by lavage with a PBS-EDTA 10 mM solution: 3 ml of cold PBS was injected i.p. into the peritoneum, and the peritoneal area was vigorously massaged before the lavage fluid was withdrawn with a 3 ml-syringe. For each tissue, a minimum of 1×106 cells were stained for 20 minutes at 4°C in the dark with the following antibodies (all from eBioscience): PerCP-Cy5.5 anti-CD19 (1D3), PerCP-Cy5.5 anti-CD3e (145-2C11), PE anti-DX5 (DX5), APC anti-B220 (RA3-6B2), FITC anti-CD8a (53-6.7), PE anti-CD4 (L3T4), PerCP-Cy5.5 anti-CD25 (PC61.5), APC anti-CD44 (IM7), PerCP-Cy5.5 anti CD45.2 (104), efluo-450 anti-CD8a (53-6.7), APC Cy7 anti-CD3e (17A2) and efluo-506 fixable viability dye. Live cells were gated using the forward scatter (FSC)-area versus viability plot, and then analyzed by FACS. Data analysis was performed using the FlowJo 9.3.1 software.
Mice 7 weeks of age or older were each infected i.p. with 1×104 pfu of HSV-1. One week later, spleens were collected and splenocytes were prepared using the aforementioned protocol (see Materials and Methods, Immunophenotyping). When indicated, 2×107 total splenocytes were transferred intravenously (i.v.) into mice. Otherwise, CD19+ B cells, CD3+, CD8+ or CD4+ T cells were purified from total splenocytes using magnetic cell sorting (MACS; Miltenyi) according to the manufacturer's instructions. Mice were then injected i.v. with either 12×106 CD19+ B cells, 5×106 CD3+ T cells, 2.5×106 CD8+ T cells or 2.5×106 CD4+ T cells. After two hours, reconstituted mice were each infected i.p. with 1×104 pfu of HSV-1 and were then monitored for survival. For NK cell depletion, mice 7 weeks of age or older were injected i.v. with 35 ul of anti-asialo GM1 antibody (Wako). After 24 hours, mice were each infected i.p. with 1×104 pfu of HSV-1. To maintain NK cell depletion, the mice were treated with anti-asialo GM1 antibody every three days until the experimental endpoint. Throughout mice were monitored for survival.
|
10.1371/journal.ppat.1004701 | Chitin Recognition via Chitotriosidase Promotes Pathologic Type-2 Helper T Cell Responses to Cryptococcal Infection | Pulmonary mycoses are often associated with type-2 helper T (Th2) cell responses. However, mechanisms of Th2 cell accumulation are multifactorial and incompletely known. To investigate Th2 cell responses to pulmonary fungal infection, we developed a peptide-MHCII tetramer to track antigen-specific CD4+ T cells produced in response to infection with the fungal pathogen Cryptococcus neoformans. We noted massive accruement of pathologic cryptococcal antigen-specific Th2 cells in the lungs following infection that was coordinated by lung-resident CD11b+ IRF4-dependent conventional dendritic cells. Other researchers have demonstrated that this dendritic cell subset is also capable of priming protective Th17 cell responses to another pulmonary fungal infection, Aspergillus fumigatus. Thus, higher order detection of specific features of fungal infection by these dendritic cells must direct Th2 cell lineage commitment. Since chitin-containing parasites commonly elicit Th2 responses, we hypothesized that recognition of fungal chitin is an important determinant of Th2 cell-mediated mycosis. Using C. neoformans mutants or purified chitin, we found that chitin abundance impacted Th2 cell accumulation and disease. Importantly, we determined Th2 cell induction depended on cleavage of chitin via the mammalian chitinase, chitotriosidase, an enzyme that was also prevalent in humans experiencing overt cryptococcosis. The data presented herein offers a new perspective on fungal disease susceptibility, whereby chitin recognition via chitotriosidase leads to the initiation of harmful Th2 cell differentiation by CD11b+ conventional dendritic cells in response to pulmonary fungal infection.
| Humans often inhale potentially pathogenic fungi in the environment. While CD4+ helper T (Th) cells are required for protection against invasive disease, a subset of Th cells, called Th2 cells, are associated with increased mortality and allergy/asthma morbidity. Our study aimed to unravel the cellular and molecular basis of pulmonary Th2 cell induction in response to lethal infection with Cryptococcus neoformans. Antigen-presenting cells coordinate naïve Th cell priming and differentiation, but the precise leukocyte responsible for Th2 cell expansion to pulmonary cryptococcal infection has not been determined. Using an experimental mouse model of pulmonary cryptococcosis, we show that a subset of lung-resident dendritic cells is uniquely required for Th2 cell induction. We additionally sought to identify the molecular signal received by the host that allows dendritic cells to selectively induce Th2 cells. Since parasites and fungi elicit Th2 cell responses and both produce chitin, a molecule not found in vertebrates, we hypothesized that recognition of fungal chitin is a determinant of fungal disease. Here, we demonstrate that C. neoformans chitin and the host-derived chitinase, chitotriosidase, promote Th2 cell accumulation and disease. These findings highlight a promising target of next generation therapies aimed at limiting immunopathology caused by pulmonary fungal infection.
| Pulmonary mycoses, ranging from invasive fungal infection to severe asthma with fungal sensitization, affect millions of people worldwide [1,2]. Fungi inhabit a multitude of ecological niches, and consequently, humans continuously encounter potentially pathogenic fungi in the environment. Subsequent disease is determined by the size of the innoculum, virulence of the microbe, and immune status of the host. In particular, CD4+ helper T (Th) cell subsets are critical mediators of the immune response to fungal exposure. Interferon-γ from Th1 cells and interleukin (IL)-17 from Th17 cells contribute to protective immunity via classical activation of macrophages and neutrophil recruitment, respectively [3]. Conversely, Th2 cell production of IL-4, IL-5, and IL-13 impels eosinophilia, alternative macrophage activation, mucus and IgE production, and airway obstruction [4]. These type-2 responses drive fungal-associated allergies and positively correlate with invasive fungal disease severity [4]. Although a fair amount is known about type-2 responses and their downstream consequences, the basis of Th2 cell induction associated with pulmonary mycosis is less well defined.
Antigen presentation by an immune cell bearing major histocompatibility II (MHCII) is required for naïve Th cell priming and differentiation. Thus, a cellular intermediate must coordinate Th2 cell induction. Professional antigen presenting cells direct Th cell fate, and inflamed lungs contain several ontologically distinct immune cells with this potential capability [5]. The precise leukocyte subset responsible for priming Th2 cells, as well as the location that this event occurs, whether at the site of infection or within secondary lymphoid tissue, has not been comprehensively investigated. Furthermore, specific features of the infection that lead to Th2 cell lineage commitment remain largely unexplored in the context of pulmonary fungal infection.
While some models attribute induction of type-2 responses to protease cleavage of host proteins and wound repair of lung injury [6], many microbes that elicit Th2 cell responses produce chitin [7]. Chitin is a polysaccharide composed of polymeric N-acetylglucosamine. The rigidity of chitin is utilized in the cell wall of fungi as well as the exoskeleton of arthropods and filarial sheath of parasitic worms. Higher organisms rely on keratin for similar structural purposes, and as a result, vertebrates do not synthesize or store chitin. These differences allow an opportunity for the vertebrate immune system to detect chitin-containing pathogens as foreign [8,9]. While chitin detection may prove beneficial to the host in the context of parasitic infection [10], we hypothesize that inappropriate or dysregulated Th2 responses instigated by recognition of chitin promotes fungal pathogenesis.
Chitinases are a pivotal component of the host response to chitinous organisms [11]. Chitin is positioned beneath layers of mannans and glucans in the fungal cell wall, thus secreted host chitinases are needed to penetrate the wall matrix and make chitin fragments available to host surveillance [12]. Mammals encode two functional chitinases, chitotriosidase (Chit1) and acidic mammalian chitinase (AMCase) [11]. A naturally occurring allele of CHIT1 renders the enzyme inactive [13], and these mutations have been associated with susceptibility to parasitic worm infection in humans [14]. Likewise, AMCase has been linked to eosinophilia [15] and alternative macrophage activation [16] in mouse models of pulmonary allergy. Consequently, we reasoned that mammalian chitinases could be necessary for efficient host recognition of fungal chitin and subsequent Th2 cell priming.
Using an inhalation model of Cryptococcus neoformans infection and novel reagents to detect Cryptococcus-specific Th cells, we unravel the basis for Th2 cell induction in response to pulmonary fungal infection. We report profound accumulation of detrimental Th2 cells in the lungs of infected mice. We additionally show that lung-resident CD11b+ interferon regulatory factor (IRF) 4-dependent conventional dendritic cells present antigen to Th cells and drive potent Th2 differentiation at the site of infection in the lungs. Surprisingly, our results demonstrate that an excess of fungal chitin, as well as digestion of chitin via Chit1, and not AMCase, lead to chitin detection, Th2 cell accumulation, and enhanced disease. Lastly, we observed increased Chit1 activity in humans with confirmed fungal infections, reinforcing the relevance of Chit1 in human disease. This study offers novel insights into the cellular source of antigen presentation and molecular basis of chitin recognition via Chit1 that underlies deleterious Th2 cell formation during pulmonary mycosis.
We utilized a murine model of pulmonary cryptococcosis to investigate Th cell priming during fungal infection. Upon inhalation, Cryptococcus neoformans establishes a robust lower respiratory tract infection that causes tissue damage and ultimately leads to mortality from pulmonary complications and dissemination resulting in meningoencephalitis. To distinguish Th cell responses to infection from non-specific wound healing Th2 cell responses, we generated a recombinant peptide-major histocompatibility class II (pMHCII) tetramer that enabled identification of C. neoformans antigen-specific Th cells. The pMHCII tetramer contains a 13 amino acid peptide from an immunodominant cryptococcal protein, chitin deacetylase 2 (Cda2) (Table 1) [17]. The Cda2-MHCII tetramer labeled a population of antigen-experienced (i.e. CD44+) Th cells, but it did not stain non-activated (i.e. CD44−) Th cells from C. neoformans infected mice or CD44+ Th cells from naive mice (Figs. 1A, S1 for flow cytometry gating). In addition, mice infected with a C. neoformans mutant (cda2Δ) that lacks Cda2 protein expression [18] had marked reductions in Cda2-MHCII tetramer binding cells (Fig. 1A). Though Cda2 contains a dominant CD4+ T cell epitope, cross-reactivity to other closely related cryptococcal proteins likely account for the remaining tetramer binding Th cells generated during infection with cda2Δ (Table 1). Taken together, these studies show the Cda2-MHCII tetramer reliably identified antigen-specific CD4+ T cells produced in response to C. neoformans infection.
We characterized the immune response in the lung and lung-draining mediastinal lymph node (MLN) to determine the relative contributions of each site to CD4+ T cell subset differentiation. Pulmonary cryptococcal infection resulted in a progressive accumulation of Cda2-MHCII-specific T cells in the lungs that predominately expressed the Th2 cytokines IL-5 and/or IL-13 (Fig. 1B-D). In addition, Th2 cytokines IL-5, IL-13, and CCL5 were among the most abundant cytokines present in infected lung homogenates (Fig. 1E), and eosinophils, a downstream correlate of type-2 cytokines, represented an overwhelming majority of the bulk leukocyte population in the lungs (S2 Fig. for flow cytometry gating, S3A Fig.). In contrast, the Cda2-MHCII-specific Th2 cell response within the MLN (Fig. 1F-H) was significantly lower than the response observed in the lungs, and eosinophils comprised an insubstantial component of the lymph node resident leukocytes (S3 Fig.). These findings collectively suggest the local inflammatory environment in the lung may shape the differentiation and/or promote the selective expansion of Th2 cells.
Due to the seemingly contradictory roles of Th2 cells in beneficial wound healing responses and harmful allergic disease [19], it is not entirely clear whether Th2 cells simply correlate with or cause disease associated with fungal infection. To test the causal relationship of Th2 cells with disease severity in this model, we augmented the endogenous Th2 cell response to fungal infection using IL-2 cytokine/antibody complex treatment. IL-2 can be targeted to the high affinity IL-2 receptor to enhance Th cell proliferation by conjugating IL-2 cytokine with anti-IL-2 antibody to form IL-2 cytokine/antibody complexes [20,21]. Since Th2 cells generated during pulmonary cryptococcal infection expressed high levels of the alpha chain of the high affinity IL-2 receptor (CD25) (Fig. 2A), we sought to use IL-2 complexes to boost the Th2 cell response.
The wildtype strain of C. neoformans, KN99α, induces an extremely aggressive infection that leaves little room to increase the Th2 cell response. Consequently, we used an attenuated strain of C. neoformans, gpr4Δgpr5Δ (attenuation explained below/Fig. 5; deficient in production of large, chitinous cells). Treatment with IL-2 complexes increased Th2 cell numbers compared with similarly infected mice receiving antibody or cytokine alone (Fig. 2B). Th2 cells and cytokines were also elevated in lung homogenates from infected mice treated with IL-2 complexes (Fig. 2C&D). In addition to Th2 cells, Regulatory T (Treg) cells can be expanded by IL-2 complex treatment [20,21] (S4A Fig.). This increase of Treg cells in mice treated with the IL-2 complex could theoretically suppress protective Th cell responses and allow Th2 cells to predominate the response. However, IL-2 complex treatment did not affect Th1 or Th17 cell numbers (Fig. 2B) and only minimal changes in IFNγ cytokine (Fig. 2C) and monocyte accumulation (Fig. 2D) were observed, showing IL-2 complex treatment did not eliminate the effector activity of protective Th1 cells (Fig. 2D). Instead, Schulze et al. [22] showed Treg cells suppress Th2 cells during cryptococcal infection (S4B–S4C Fig.), suggesting the increase in Treg cells due to IL-2 complex treatment would actually limit Th2 cell accumulation in this system. Hence, IL-2 complexes can be used to augment the Th2 cell response during pulmonary fungal infection and assess the relationship between Th2 cells and fungal disease.
If Th2 cells promote disease, we hypothesized that increasing the Th2 response should accelerate death during infection with the virulence-attenuated strain, gpr4Δgpr5Δ. IL-2 complex treatment increased the Th2 cell response to levels even higher than the fully virulent KN99α infection (Fig. 2E). Treatment with IL-2 complexes also greatly reduced the survival time of infected mice (Fig. 2F) without affecting pulmonary fungal burden (Fig. 2G). Uninfected mice treated with the same regimen of IL-2 complexes survived more than 30 days and remained healthy (Fig. 2F), indicating the IL-2 complex treatment targeted detrimental cells that were only present during infection. IL-2 treatment of infected mice also induced obvious lung pathology consistent with increased Th2 activity, noted by increased metaplasia of the bronchiolar epithelium and mucous obstruction of the airways (Fig. 2H). En masse, these data indicate that Th2 cells exacerbate pulmonary disease during fungal infection.
The diminished Th2 response in the MLN compared to the lung led us to question whether lymphoid priming was required for Th2 cell induction during pulmonary fungal infection. Fms-like tyrosine kinase 3 ligand (Flt3L) is a differentiation factor for several hematopoietic cell subsets, and genetic deletion of Flt3L causes defects in antigen presenting cell traffic between the site of infection and secondary lymphoid organs [23]. Flt3L deficient mice infected with C. neoformans neither experienced mediastinal lymphadenopathy (Fig. 3A) nor elicited a polyclonal Th2 response in the MLN (Fig. 3B). Surprisingly, the Th2 cell response in the lungs after C. neoformans infection was unaffected by Flt3L deficiency compared to wildtype animals (Fig. 3C), indicating lymphoid priming is not required for pulmonary Th2 cell accumulation.
To determine the immune cell intermediate that primes Th2 cells in the lungs, we relied on the fact that Th cells are MHCII restricted. Three leukocyte subsets that express MHCII exist in the lungs of mice infected with C. neoformans: monocytes, CD11c+ cells, and B cells (Fig. 4A). Of these, CD11c+ cells are the most abundant in the lungs during cryptococcal infection (Fig. 4A). Consequently, we interrupted the specific interaction between CD11c+ cells and Th cells by generating mice with conditional deletion of MHCII in cells that express CD11c (CD11c-cre MHCII fl/fl) (Fig. 4B). Unlike NOD/SCID/Rag mice that fail to generate mature Th cells, naïve CD11c-cre MHCII fl/fl mice produced an equivalent number of Th cells as naïve wildtype mice (Fig. 4C), showing the peripheral Th cell compartment remained intact in CD11c-cre MHII fl/fl mice. Thus, conditional deletion of MHCII on CD11c+ dendritic cells allowed specific disruption of the interaction between the dendritic cells and the Th cells in the periphery. Pulmonary Th cell expansion during cryptococcal infection was completely abolished in CD11c-cre MHCII fl/fl mice (Fig. 4C-D). Consequently, MHCII-bearing CD11c+ cells prime antigen-specific Th cells in the lungs of mice infected with C. neoformans.
CD11c+ cells are a heterogeneous group of macrophages and several dendritic cells (DC) subsets in the lungs [24]. Therefore, we sought to discern the specific lineage of the CD11c+ antigen presenting cell that is responsible for pulmonary Th2 cell induction using an unbiased forward genetic screen of mouse lines genetically deficient in various CD11c+ subsets or their ability to interact with Th cells via MHCII (S5A Fig.). Lysozyme M (LysM)-cre MHCII fl/fl (macrophages and granulocytes [25]), BATF3−/− (CD103+ conventional dendritic cells [5]), and CCR2−/− (monocytes and monocyte-derived dendritic cells [26]) mice generated robust antigen-specific Th2 responses during cryptococcal infection (S5–S6 Fig.). Only mice deficient in CD11b+ conventional dendritic cells, abrogated using CD11c-cre IRF4 fl/fl mice, experienced blunted Th2 cell accumulation with cryptococcal infection (Fig. 4E-H). Therefore, our exhaustive search revealed lung-resident CDllc+ CDllb+ IRF4-dependent conventional DC (referred to as CD11b+ conventional DC) are uniquely required for Th2 cell induction in response to pulmonary cryptococcal infection.
Existing evidence and our data show CD11b+ conventional dendritic cells are capable of inducing both Th17 and Th2 cell responses to pulmonary fungal infection [27]; therefore, these DC are not inherently programmed to specify a single Th cell lineage. Determination of Th2 cell fate by these lung resident DC must require higher order detection of specific features of the fungal infection. Many chitin-containing pathogens, as well as asthma/allergy models using purified chitin, evoke type-2 immunity [9,16,28]. Consequently, the striking Th2 cell response to pulmonary fungal infection prompted us to explore the role of chitin as a Th2 cell adjuvant.
Maintaining chitin homeostasis is critical for cell wall integrity and microbe vigor. Chitin synthases that regulate cell wall chitin deposition are often essential for fungal viability or are part of a redundant pathway [29]. As a result, studies that attempt to correlate loss-of-function mutations in chitin synthesis genes with modulation of the host response and attenuation of virulence would be challenging to interpret. Whether loss of Th2 cells and alterations in disease were due to a decrease in chitin, a loss of cryptococcal fitness/growth, or unmasking of other antigens due to modifications of the fungal cell wall or capsule would not be easily distinguishable [30].
In lieu of testing the requirement of chitin in Th2 cell induction, we exploited a natural property of C. neoformans to determine if increased fungal chitin was sufficient to expand Th2 cell formation. Approximately 20% of wildtype cryptococcal cells (KN99α) recovered from the lungs of infected mice increase in diameter from <10 μm to 15–100 μm [31–33]. Previous studies have shown these enlarged cells, known as titan cells, exhibit increased thickness of the fungal cell wall [32]. Using the fluorescent dye calcofluor white to measure chitin content in individual cells by epifluorescence microscopy (Fig. 5A-B), or at the population level with flow cytometry (S7A Fig.), we found the large C. neoformans titan cells contained more chitin, at a higher density, than the typical sized cells. C. neoformans produces several enzymes that deacetylate chitin to form chitosan [18,30]. Biochemical analyses additionally revealed that the amount of chitosan produced by titan cells and typical size cells did not differ, whereas chitin was significantly more abundant in the titan cells (Fig. 5C). Therefore, enhanced chitin content accompanied cell size increases during formation of cryptococcal titan cells.
To control for the relative effects of cell size and chitin content on Th2-mediated disease, we utilized several mutants of C. neoformans. A strain with targeted deletions in G protein-coupled receptor (gpr) 4 & gpr5 produces >95% typical sized cells in vivo [34], and these cells retain normal amounts of chitin (Fig. 5A). Deletion of the transcription factor Rim101 (rim101Δ) abolishes titan cell production [34,35], yet the typical sized cells have increased expression of chitin synthesis genes and elevated chitin content [36,37] (Fig. 5A). Using these mutants, we were able to dissociate cryptococcal cell size from cell wall chitin as well as manipulate the total amount of chitin present during infection (Table 2).
We examined the impact of alterations in chitin content in C. neoformans on Th2 cell accumulation in the lungs. Antigen-specific Th cell priming and Th2 cell differentiation were reduced in mice infected with the low chitin gpr4Δgpr5Δ strain compared to infections with both the high chitin KN99α (due to titan cell production) and rim101Δ strains (Fig. 5D-E). The Cda2-independent polyclonal Th2 cell response to gpr4Δgpr5Δ infection was also significantly lower than the responses to KN99α and rim101Δ infections (S7B Fig.), indicating the defect in antigen-specific Th2 cell accumulation to gpr4Δgpr5Δ infection was not due to differential expression of cda2 between the strains. Finally, the secreted Th2 cytokines IL-5, IL-13, and CCL5 present in lung homogenates from mice infected with gpr4Δgpr5Δ were significantly reduced compared to KN99α and rim101Δ infections (Fig. 5F). Of note, secreted Th1 and Th17 cytokines, interferon-γ and IL-17A, did not concomitantly increase in response to gpr4Δgpr5Δ infection (Fig. 5F). Thus, Th cells not receiving strongly polarizing Th2 signals in the gpr4Δgpr5Δ infection failed to acquire an alternate Th1 or Th17 cell differentiation fate. Since KN99α and rim101Δ strains contribute more total chitin to the infection than the gpr4Δgpr5Δ strain (Table 2, S7A Fig.), these data demonstrate that chitin abundance and not cell size positively correlated with Th2 cell response intensity.
Uncontrolled factors affected by the gpr4Δgpr5Δ or rim101Δ mutations could correlate with chitin levels and independently contribute to Th2 cell accumulation. To directly test if purified chitin can increase Th2 cell numbers, mice were infected with gpr4Δgpr5Δ, and <10 μm chitin particles were co-administered into the lungs. Chitin treatments partially rescued the attenuated Th2 cell response (Fig. 5G). Importantly, this effect was observed in the clonal population of antigen-specific cells, and therefore, chitin increased the potency of Th2 cell induction during C. neoformans infection.
We examined the correlation between chitin, Th2 cell accumulation, and disease severity in our experimental model of cryptococcal infection. Infections with the high chitin strains, KN99α and rim101Δ, significantly hastened the time to death relative to mice infected with gpr4Δgpr5Δ (Fig. 5H). Interestingly, all strains had equivalent pulmonary colony forming units at 14-days post-infection, indicating the differences in disease were not simply due to a failure to control the infection (Fig. 5I), but rather paralleled the total amount of chitin present during infection (Table 2). In summary, both chitin production by C. neoformans and Th2 cell accumulation directly correlated with exacerbation of lethal fungal disease.
We next investigated host intrinsic factors that could influence detrimental Th2 cell responses to chitin—specifically the mammalian chitinases, AMCase and Chit1. pH-sensitive differences in enzyme activity allow for an assessment of the contribution of each enzyme to chitin cleavage. Consistent with published reports, recombinant AMCase cleaved 4-methlylumbelliferone chitotriose across a broad pH range, whereas recombinant Chit1 was only active at less acidic pH (Fig. 6A) [38]. Chitinase activity in lung homogenates from infected mice was significantly elevated compared with uninfected animals at pH 2 and pH 5 (Fig. 6A). Furthermore, lung homogenates from infected mice genetically deficient in Chit1 or AMCase both showed decreased cleavage of the fluorescent chitin substrate (Fig. 6A). These data indicate both chitinase enzymes are active during pulmonary fungal infection, and genetic abolishment of these enzymes can be used to understand the effect of chitin degradation on Th2-mediated fungal pathogenesis.
To test the hypothesis that Th2 cell-associated disease depends on chitinases, we infected wildtype, Chit1−/−, and AMCase−/− mice with C. neoformans and quantified the Th2 cell response. Despite no differences in pulmonary fungal burden at 14-days post-infection (S8A Fig.), Th2 cells were 10-fold less abundant in the lungs of infected Chit1−/− mice compared with wildtype controls (Fig. 6B-C), and this trend was consistent with all the strains of C. neoformans tested (S8B Fig.). Conversely, AMCase deficiency did not impact Th2 cell quantities after cryptococcal infection (Fig. 6B-C). Furthermore, Chit1 deficiency and not AMCase deficiency also significantly extended the survival of mice infected with C. neoformans relative to age matched, wildtype animals (Fig. 6D-E). The loss of Th2 cell accumulation with Chit1 deficiency was responsible for attenuation of disease, because the use of IL-2 complexes to boost Th2 cell numbers (Fig. 6C) and associated cytokines (Fig. 6F) also hastened lethal disease in Chit1−/− mice compared with infection-matched, untreated Chit1−/− controls (Fig. 6D). Similar to our previous studies using the IL-2 complex, Th1 cytokine production in the Chit1−/− mice was only minimally affected by the IL-2 complex treatment (Fig. 6F). Thus, the presence of Chit1, and not AMCase, positively influences Th2 induction and subsequent disease.
Chitin receptors in plants bind chitin oligomers [39], and chitin polymer size also influences the mammalian immune response to chitin [40,41]. Since appropriately sized chitin fragments could result from chitin digestion by Chit1, we used heterogeneous-length chitin and highly purified chitin heptamers to understand the effect of Chit1-associated degradation of chitin on Th2 cell accumulation. Although treatment with heterogeneous-length chitin augmented Th2 cell induction in wildtype animals infected with gpr4Δgpr5Δ cells (Fig. 5F), it did not increase Th2 cell numbers in Chit1−/− mice (Fig. 6G). Conversely, inoculation with mass equivalent amounts of chitin heptamers boosted the Th2 cell response in C. neoformans-infected animals with Chit1 deficiency (Fig. 6G), revealing the requirement for Chit1 in chitin polymer recognition can be bypassed by providing exogenous chitin fragments. Therefore, our data demonstrate a role for Chit1 in the chitin cleavage pathway that leads to Th2 cell accumulation.
We next examined how chitin fragments influence the upstream pathway of Th2 cell induction. To test the hypothesis that DC interact directly with chitin fragments, we cultured primary pulmonary leukocytes from infected mice in the presence of R-phycoerithrin-fluorophore conjugated chitin heptamers (RPE-GN7) or unbound REP-streptavidin (RPE-SA). While RPE-GN7 labeled a subset of B cells that have been shown to bind chitin [9], RPE-GN7 did not adhere to conventional CD11b+ DC (S9A Fig.). Thus, we did not detect direct binding of chitin heptamers to the conventional DC. Furthermore, conventional CD11b+ DC stimulated ex vivo with PMA + ionomycin produced an important Th2 cell differentiation cytokine, IL-4, yet this DC subset did not express IL-4 upon stimulation with chitin heptamers (GN7) (S9B Fig.). Combined, these data suggest the CD11b+ conventional dendritic cells may not sense chitin levels directly.
Pulmonary epithelial cells respond to chitin [42] and secrete several Th2-inducing alarmins, thymic-stromal lymphopoietin (TSLP), IL-25, and IL-33 [28,43,44]. As a result, alarmin receptors on DC could potentially mediate the indirect recognition of chitin, leading to Th2 cell polarization during pulmonary fungal infection. We found CD11b+ conventional DC from the lungs of fungal infected mice expressed high levels of TSLP receptor, but not receptors for IL-25 or IL-33 (S9C–S9D Fig.), indicating this DC subset is capable of sensing TSLP generated by epithelial cells. Taken together, our data suggest Th2 cell induction by conventional CD11b+ DC appears to involve an indirect recognition of chitin oligomers.
The importance of Chit1 in promoting Th2 cell-mediated disease in an experimental model of cryptococcosis prompted us to investigate the relevance of chitotriosidase activity in human fungal disease. Blood samples were collected from human donors: 46 Ugandan patients presenting at the hospital for the first time with AIDS and 38 similar AIDS patients experiencing acute cryptococcal disease (S1 Table). We analyzed chitotriosidase and AMCase enzymatic activity in plasma from each group as described for the mouse lung homogenates. Chitin substrate cleavage at pH = 5 was significantly elevated in plasma from AIDS patients with cryptococcal infection when compared to AIDS patients without cryptococcal infection (Fig. 6H). Comparatively low levels of enzymatic activity were detected at pH = 2 for each group, indicating that chitotriosidase and not acidic mammalian chitinase is the predominate chitinase produced by humans with cryptococcal infection (Fig. 6H).
To determine if the difference in chitotriosidase activity was due to an inherent propensity or deficiency in chitotriosidase expression, we used cryptococcal lysate antigens to stimulate chitinase production in whole blood samples from the human donors. Stimulation of whole blood from all patients induced robust chitotriosidase activity relative to unstimulated samples, and chitinase activity did not differ between the groups (Fig. 6H), indicating all human donors had equivalent capacity to produce chitotriosidase. Chitinase activity was not detectable in pure cryptococcal culture supernatants or cryptococcal lysate antigens (S10 Fig.), and as a result, the chitinase activity detected in the assays with human samples was not due to Cryptococcus-derived chitinases. Taken together, we conclude that fungal antigen induces chitotriosidase activity in humans experiencing cryptococcosis.
An association between pulmonary fungal exposure and allergic Th2 inflammation is well established [1,3]. Fungal proteases [45] and fungal chitin [46] impact the innate immune responses underlying allergic inflammation, but elements of Th2 cell induction are enigmatic. Using an experimental model of pulmonary cryptococcosis, we demonstrated that inhalation of C. neoformans establishes a robust pulmonary infection, and the potent antigen-specific Th2 cell accumulation required lung resident CD11b+ conventional DC. Since these CD11b+ conventional DC can stimulate Th2 or Th17 cell differentiation in response to C. neoformans and Aspergillus fumigatus exposure respectively [27], these DC must interpret specific features of the infection to direct Th2 cell fate. To this end, we found cryptococcal chitin and exogenous administration of chitin particles correlated with increased Th2 cell accumulation. We further showed that chitotriosidase activity was highest in mice and humans infected with C. neoformans, and Chit1 was necessary for efficient Th2 cell induction and disease in our murine model of cryptococcosis. Taken together, these data indicate the host response to fungal chitin is an important factor that enhances Th2 cell production during pulmonary fungal infection.
Our findings narrow an important gap in the mechanism of pattern recognition of fungal chitin in the lungs (Fig. 7). We have shown Chit1 functions as a “gatekeeper” in making chitin fragments available to host surveillance, thereby promoting Th2 cell accumulation and disease. Additionally, the use of pharmaceutical grade chitin heptamers to augment Th2 cell responses establishes a new minimum component for chitin recognition in vertebrates. While we did not detect direct interactions between the CD11b+ conventional DC and labeled chitin ex vivo, other systems have shown chitin binding to the mannose receptor and subsequent recognition by TLR9 and/or NOD2 [47]. Alternatively, pulmonary allergy models demonstrate that lung epithelial cells recognize chitin fragments and produce the necessary alarmins for Th2 cell induction: TSLP, IL-25, and IL-33 [43,44,48]. In our model, CD11b+ conventional DC express high levels of TSLPR. Finally, natural IgM has been shown to bind fungal carbohydrates, including chitin, and facilitate the interaction of DC and fungal carbohydrates [9]. Signals, such as alarmins or antibody-chitin complexes, could be received by CD11b+ conventional DC to direct Th2 cell differentiation via IL-4 or another novel pathway.
A major impediment in understanding Th cell responses to pulmonary fungal infection has been the lack of reagents to detect antigen-specific Th cells. Antigen-specific reagents are particularly important when examining Th2 responses, because Th2 cells can be induced by the wounding that occurs during infection. Our ability to track endogenously derived, antigen-specific Th cells with pMHCII tetramers [49] allowed us to present for the first time that Th2 cells produced in response to fungal pathogens are not part of a generalized wound healing response but are fungal-antigen specific. Unlike T cell receptor transgenic approaches, pMHCII tetramers permitted us to monitor the population of infection-specific Th cells in the polyclonal repertoire, while maintaining physiologic precursor frequency and clonal expansion. Thus, we are able to examine the Th cell response during the natural course of infection and keep all other variables constant. The availability of these pMHCII tetramers will undoubtedly empower cryptococcal researchers and accelerate the field of fungal immunology.
The use of IL-2 complexes allowed us to conveniently and reliably augment the Th2 cell response to further understand unappreciated elements underlying Th2-mediated disease. This strategy is amenable to any host or microbial genetic model, which facilitates direct comparisons. Also, this gain-of-function approach permitted us to test the sufficiency of Th2 cells to exacerbate disease. These data, combined with loss-of-function studies by other groups [50,51], alter the longstanding paradigm that susceptibility to lethal fungal disease is traditionally viewed as a breakdown in protective immunity. This paradigm is supported by the higher prevalence of invasive fungal disease in immunosuppressed individuals, including people living with HIV/AIDS, cancer patients undergoing chemotherapy, and solid organ transplant recipients. However, we propose that in addition to the lack of a protective response, an independent development of a harmful Th2 cell response further exacerbates disease. This is particularly important in the the case of human cryptococcocosis were a compromised immune system not only lacks sufficient quantities of lymphocytes to resolve the fungal infection, but the residual Th cell repertoire is plastic and detrimentally influenced by the microbe [52–55].
A subset of innate lymphoid cells (ILC) produce the Th2 cytokines, IL-5 and IL-13 [56], and these so-called ILC2 have been shown to contribute to allergic airway disease [4]. While IL-2 complex treatment dramatically increases Th2 cell accumulation and enhances pulmonary disease, ILC numbers are not affected by IL-2 complex treatment in our model. Likewise, CD25+ ILC2 exist in the lungs under homeostatic conditions [57], yet uninfected mice exhibit no ill effects of IL-2 complex treatment. However, the developmental relationship of ILC2 to lymphocytes, combined with the lack of lineage markers expressed by ILC, make it challenging to separate the relative effects of Th2 cells and ILC2 in driving immunopathology. Thus, our conclusion that Th2 cells are vital mediators of disease does not categorically exclude the participation of ILC2 in this process.
CD11b+ conventional DC are an ontologically distinct mononuclear phagocyte subset that require IRF4 for maturation [5]. Although it has been suggested that this subset is also functionally unique in programing specific Th cell differentiation, building evidence seems to indicate a plastic role of these cells in Th cell induction. CD11b+ conventional IRF4-dependent DC have been shown to coordinate Th2 cell priming following protease inoculation and worm infection in the skin [58], as well as house dust mite extract installation in the lungs [59]. However, several research groups have found the same DC subset, using identical host genetic systems, can control Th17 cell differentiation in the gut under homeostatic conditions [60] and the lungs after fungal infection [27]. While these observations infer that CD11b+ conventional DC are plastic, these cells may have different functions in the skin and gut compared to the lungs that could explain their role in priming Th17 and Th2 cell responses. Our findings showing Th2 priming by C. neoformans, combined with work by Schlitzer et al. showing IRF4-dependent DC are capable of priming Th17 cells in response to A. fumigatus to pulmonary fungal infection, highlight a fatal flaw in the notion that DC subsets are inherently specialized to control Th cell lineage fate. As a result, we explored an alternative hypothesis that higher order signals (e.g. chitin recognition) are required for pulmonary CD11b+ conventional DC to promote Th2 cell priming to fungal infection.
Gene deletions in microbes can cause pleiotropic phenotypes, as seen in the rim101Δ strain (35) that can complicate interpretation of the effects of these mutations on host responses. However, the data obtained from the mutants utilized in this study support the conclusion that fungal chitin promotes Th2-mediated disease for several reasons. First, equivalent fungal burden in the lungs at 14 days post-infection indicates these mutations do not confer an inherent survival advantage or disadvantage for the fungus, and any effect the mutants have on leukocyte accumulation or disease is not simply driven by antigen load. Second, Rim101 transcriptionally controls many elements associated with cryptococcal virulence, including pH responses, encapsulation, cell enlargement, and iron sequestration [61]. While individual mutations in each of these pathways should result in a loss of virulence, the rim101Δ mutants paradoxically exhibit equivalent or accelerated disease [36]. Altered expression of cell wall synthesis genes combined with unmasking of the cell wall has been posited [36,37], and in particular, our data implicate cell wall chitin in enhancing fungal pathogenesis associated with rim101 deficiency. More importantly, we view rim101Δ as an essential control for the effect cryptococcal cell size has on Th2 cell formation more than an independent test of the hypothesis that fungal chitin drives Th2 cell priming. Thus, we needed a strain of C. neoformans that produces mostly small cells but retained elevated chitin density to offset the loss of elevated chitin density with titan cell deficiency. At a minimum, these data prove that large cryptococcal cell size is not driving robust Th2 cell accumulation. Finally, due to the structural similarities, as well as the interdependent synthetic pathways, it is extremely challenging to decouple how chitin and chitosan separately impact a complex biological system. Although we cannot rule out immunomodulatory effects of chitosan in our experiments with C. neoformans mutants, we have confirmed chitin alone, when provided as an adjuvant, is sufficient to augment Th2 cell accumulation. Interpretation of the Cryptococcus chitin mutant data in the context of our additional data regarding host recognition of exogenous chitin builds a compelling argument that fungal chitin promotes detrimental Th2 cell induction.
Acidic mammalian chitinase has been previously implicated in innate type-2 responses [62]. However, a head-to-head comparison of the effect of both mammalian chitinases, chitotriosidase and acidic mammalian chitinase, has never been performed in any model, much less in the context of Th2 cell responses to fungal infection. Surprisingly, chitotriosidase, and not acidic mammalian chitinase, influenced Th2 cell accumulation and disease during pulmonary fungal infection. While no direct explanation exists, the different patterns of expression of AMCase and Chit1 could explain our results. AMCase is produced by a number of cells including several leukocyte subsets and pulmonary epithelial cells, whereas Chit1 expression is restricted to mononuclear phagocytes, including DC [38,63]. Since these DC are the main coordinators of antigen presentation to CD4+ T cells, the close proximity of chitin degradation and recognition likely allows the DC to efficiently influence the fate of CD4+ T cells during pulmonary fungal infection. Furthermore, the varying pH-dependent enzymatic activity of Chit1 and AMCase suggest these enzymes may function in disparate anatomical and subcellular compartments that may bias their participation in response to pulmonary fungal infection.
Considering Th2 cell responses likely evolved to resist parasitic infection [10], the asymmetric global distribution of CHIT1 alleles [64] combined with the data presented herein offers a unique perspective on why individuals from tropical regions with endemic parasites tend to experience frequent and severe mycosis [65–69]. Ethnic groups historically residing in regions with highly prevalent parasites like Strongyloides tend to maintain functional CHIT1, whereas populations from more temperate climates with lower endemic parasitic burdens frequently possess enzymatically inactive CHIT1 alleles [64]. Perhaps continuous fungal exposure in the absence of parasitic encounters provides sufficient negative selection pressure to eliminate functional CHIT1 alleles from these populations (e.g. Europe). Conversely, ethnic groups historically residing in tropical areas (e.g. Africa) that maintain functional CHIT1 alleles may have enhanced protection against common parasitic exposures, while these same individuals experience exacerbation of Th2-mediated fungal disease [67–69].
In conclusion, we elucidated a novel mechanism of Th2 cell induction during fungal infection. CD11b+ DC, and importantly recognition of chitin via Chit1, drive the generation of deleterious Th2 cells responding to pulmonary fungal infection. Next generation anti-fungal treatments should not only block fungal growth, but should also target the host immune response. A recent trial used IFNγ in combination with traditional anti-fungal therapy to promote beneficial immune responses. This treatment improved cryptococcal clearance, yet it had no significant impact on patient survival [70]. Our study suggests treatments that additionally aim to suppress the pathologic Th2 response, perhaps through chitotriosidase inhibition, may be necessary to improve clinical outcomes. Ultimately, the coordinated efforts of microbiologists, immunologists and infectious disease physicians will enable personalized medicine approaches that effectively combat lethal fungal infections by inhibiting fungal growth, promoting beneficial host responses, and dampening pathologic inflammation.
This study was approved by the institutional review boards of the University of Minnesota, Makerere University, and the Uganda National Council of Science and Technology. Written informed consent was obtained from all human participants prior to inclusion in the sutdy, and all data were de-identified [71]. All animal experiments were done in concordance with the Animal Welfare Act, U.S. federal law, and NIH guidelines. Mice were handled in accordance with guidelines defined by the University of Minnesota Institutional Animal Care and Use Committee (IACUC) protocol numbers 1010A91133 and 1207A17286 and University of Massachuesets IACUC protocol number A-1802.
All mice used in this study were derived from a C57BL/6 background. C57BL/6J, LysM-cre (B6.129P2-Lyz2tm1(cre)Ifo/J), CD11c-cre (B6.Cg-Tg (Itgax-cre) 1-1Reiz/J), MHCII loxP (B6.129X1-H2-Ab1tm1Koni/J), Batf3 −/− (B6.129S(C)-Batf3tm1Kmm/J), CCR2 −/− (B6.129(Cg)-Ccr2tm2.1Ifc/J), B6.129(Cg)-Foxp3tm3(DTR/GFP)Ayr/J) mice were purchased from Jackson Laboratories (Bar Harbor, ME) and Flt3L−/− (C57BL/6-flt3Ltm1Imx) were purchased from Taconic (Hudson, NY). Crosses were performed when necessary to generate the mouse strains used in this study, as indicated in S2 Table. Chit1 −/− [72] mice were infected and processed in the laboratory of Kirsten Nielsen. AMCase−/− [73] mice were infected and processed in the laboratory of Stuart Levitz per MTA stipulations. All mice were housed in specific pathogen–free conditions.
Cryptococcus neoformans var. grubii strains were streaked on yeast peptone dextrose (YPD) agar plates and incubated for 2 days at 30°C. YPD broth was inoculated with colonies from the aforementioned plate and incubated for 16 hours at 30°C with gentle agitation. The inoculum was prepared by pelleting the culture, washing 3 times with phosphate buffered saline (PBS), and resuspending in PBS at a concentration of 2x106 cells/mL. All strains used in this study were on a KN99α genetic background, and the complemented strains have wildtype phenotypes [18,34,35].
A well established intranasal pulmonary aspiration model of cryptococcosis was used for this study [74]. 6–8 week old, sex-matched mice were anesthetized with pentobarbitol or isoflurane. 5x104 cryptococcal cells in 25 μL of PBS was placed on the nares of each mouse, and the mice aspirated the inoculum into the lower respiratory tract. Finally, the mice were suspended by their incisors for 5 minutes and subsequently placed upright in their cage until regaining consciousness. For survival studies, ten mice per group were infected as described above. Animals were monitored for morbidity and sacrificed when endpoint criteria were reached. Endpoint criteria were defined as 20% total body weight loss, loss of 2 grams of weight in 2 days, or symptoms of neurological disease.
Nine amino acid peptides from Cda2 were selected using a MHCII loading algorithm [75]. pMHCII tetramers were produced as previously described [76]. In short, biotinylated pMHCII monomers were expressed in Drosophila melanogaster S2 cells and isolated from culture supernatants by affinity chromatography. Streptavidin-Phycoerythrin (Prozyme) was added to pMHCII monomers at a 4:1 molar ratio. Finally, tetramer formation was assessed by western blot analysis.
Lung leukocytes were isolated as previously described [77]. Briefly, lungs were excised and minced to generate approximately 1 mm3 pieces. The lung mince was incubated in HBSS (Invitrogen, Grand Island, NY) + 1.3 mM EDTA solution for 30 min at 37°C with agitation, and then transferred to RPMI-1640 (Invitrogen) medium supplemented with 5% Fetal Bovine Serum (FBS) (Invitrogen) and 150 U/ml type I collagenase (Invitrogen) and incubated for 1 h at 37°C with agitation. The cells were passed through a 70 μm filter, pelleted, and resuspended in 44% Percoll-RPMI medium (GE Life Sciences, Pittsburgh, PA). A Percoll density gradient was created (44% top, 67% bottom), and the samples were centrifuged for 20 min at 650 x g. The leukocytes at the interface were removed, washed 2 times with RPMI medium, and resuspended in PBS + FBS at a concentration of 107 cells/ml. CD4+ T cells were enriched using a Dynabeads CD4+ T Cell Negative Isolation Kit (Life Technologies, Grand Island, NY) per manufacture’s instructions. After enrichment, ∼106 cells were suspended in 200 μL of restimulation buffer (RPMI + 10% FBS + 1% Penicillin/Streptomycin + 5 μg Brefeldin A) with (stimulated) or without (unstimulated) 10 ng phorbol myristate acetate (PMA) and 50 ng ionomycin. After 6 hours, the cells were washed and immediately prepared for flow cytometry.
5 μg of murine IL-2 cytokine (Biolegend) and 25 μg of clone JES6-1A12 anti-IL-2 antibody (Bio X Cell, West Lebanon, NH) were added to 100 μL of PBS at room temperature. Each mouse received intraperitoneal injections of IL-2 complexes every other day beginning at 5 days post-infection.
Samples were incubated for 5 minutes with CD16/32 antibody (Biolegend) to block the Fc receptor and prevent nonspecific antibody binding. 25nM Cda2-tetramer was added to the sample and incubated at 25°C for 1 hour in the dark. Samples were surface-stained at 4°C for 30 minutes with the following antibodies (see gating S1 Fig.): CD4 (RM4-5, BV605, Biolegend), CD8 (53-6.7, APC-eFluor 780, eBioscience, San Diego, CA), CD11b (M1/70, PE-Cy5, eBioscience), CD11c (N418, PE-Cy5, eBioscience), B220 (RA3-6B2, PE-Cy5, eBioscience), and CD44 (IM7, Alexa Fluor 700, Biolegend). The cells were then incubated in Foxp3 Transcription Factor Buffer (eBioscience) at 4°C for 30 minutes. The cells were labeled with antibodes against the following intracellular antigens: Foxp3 (FJK-16s, FITC, eBioscience), IL-5 (TRFK5, APC, Biolegend), IL-13 (eBio13A, eFluor 450, eBioscience), IL-17A (TC11-18H10.1, BV650, Biolegend), and IFNγ (XMG1.2, PE-Cy7, eBioscience). 1:200 antibody concentrations were used for surface staining, and 1:100 antibody concentrations were used for intracelluar staining. For data acquisition, events from the entire sample (500,000–1,000,000) were collected on a BD FACSCanto II flow cytometer (BD Biosciences, San Jose, CA), and the data were analyzed with FlowJo X (Tree Star Inc., Ashland, OR).
To account for cell loss during CD4+ T cell enrichment and mitogen restimulation, several calculations were performed. Total leukocyte numbers were determined by hemacytometer count after Percoll density gradient separation. 2.5% of the sample was stained with the following antibodies: Ly6G (RB6-8C5, APC-eFluor 780, eBioscience), Ly6C (HK1.4, eFluor 450, eBioscience), CD11b (M1/70, BV650, Biolegend), CD11c (N418, BV605, Biolegend), NK1.1 (PK136, AF700, Biolegend), CD3 (17A2, PE-Cy5, Biolegend), CD4 (RM4-5, FITC, Biolegend), CD19 (6D5, PE-Cy7, Biolegend), Sca1 (D7, APC, Biolegend), and Siglec F (E50-2440, PE, BD Biosciences). Cells were identified as the following (see gating S2 Fig.): Th cells = CD11b- CD3+ CD4+. Eosinophils = CD11b+ CD11c- Siglec F+. Innate lymphoid cells = lineage- Sca1+. Dendritic cells and macrophages = CD11b+ CD11c+ Siglec F-. B cells = Siglec F- CD11b- CD11c- CD19+. Natural killer cells = Siglec F- CD11c- NK1.1+. Neutrophils = CD11b+ CD11c- Ly6G+. Monocytes = CD11b+ Ly6C+ Siglec F-. CD8+ T cells = Siglec F- CD11b- CD3+ CD4-. The proportion of CD4+ T cells was determined by flow cytometry, and this percentage was multiplied by the total number of lung leukocytes to calculate the number of CD4+ T cells per pair of lungs. The number of CD4+ T cells in the unstimulated, CD4+ T cell enriched sample was calculated after flow cytometric analysis, as described in the previous paragraph. The number of CD4+ T cells recovered in the unstimulated, enriched sample was divided by the total number of CD4+ T cells to calculate the CD4+ T cell isolation efficiency. Cell death due to mitogen restimulation was calculated by dividing the number of CD4+ T cells recovered in the stimulated sample by the number of CD4+ T cells recovered in the unstimulated sample. The number of Cda2+ Th2 cells was determined by dividing the number of Cda2+ IL-5 and/or IL-13+ positive cells by the CD4+ enrichment and cell viability indices.
Dendritic cell subsets were determined using the following antibodies and gating strategy. CD3 (17A2, PE-Cy5, Biolegend), CD19 (6D5, PE-Cy5, Biolegend), Siglec F (E50-2440, APC, BD Biosciences), CD64 (X54-5/7.1, PE, Biolegend), MHCII (M5/114.15.2, AF700, Biolegend), CD11c (N418, BV605, Biolegend), CD11b (M1/70, BV650, Biolegend), CD103 (2E7, e450, eBioscience), FcεRI (MAR-1, PE-Cy7, Biolegend), TSLPR (PE, R&D Systems), IL-25R (9B10, PE, Biolegend), IL-33Rα (DIH9, PE, Biolegend), IL-4 (BVD6-24G2), PE, Biolegend), and GN7-PE [78]. See gating S5B Fig. Dump = CD3+, CD19+, Siglec F+. Monocyte-derived DC = Dump-, CD64+, CD11c+, MHCII+, CD11b+, CD103-, FcεRI+. CD103+ cDC = Dump-, CD64-, CD11c+, MHCII+, CD103+, CD11b-. CD11b+ cDC = Dump-, CD64-, CD11c+, MHCII+, CD11b+, CD103-. While CD103, CD11b, FcεRI, and CD64 surface markers allow convenient detection of each DC subset, it is unclear whether other developmentally and/or functionally unrelated cells can express similar markers. Consequently, genetic blockade in the developmental pathways of each DC subset were assumed to be completely penetrant, notwithstanding the persistence of cells expressing surrogate markers for each ontological subset.
Lungs from mice 14-days post-infection were excised, snap frozen in liquid nitrogen, and homogenized in 3 mL of T-PER (Thermo Fisher Scientific) with Complete Protease Inhibitor Cocktail (Roche, Indianapolis, IN). The lung homogenate was pelleted, and the supernatant was collected and stored at 80°C until analysis. Samples were diluted 1:4 in assay buffer immediately before processing. Cytokines were quantified using Luminex technology according to manufacturer instructions (Bio-Rad, Hercules, CA).
Lungs were removed from mice 14 days post-infection, perfused via the right ventricle with cold PBS, inflated with 10% formalin (Thermo Fisher Scientific, Rockford, IL), and placed in a container of 10% formalin. Tissues were dried with organic solvent, embedded in parafin, sectioned, and stained with hematoxylin and eosin, before images were captured.
Cryptococcal cells were isolated from the lungs after enzymatic digestion and density gradient separation, as described above. Cells were fixed with 3.7% formaldehyde and stored at 4°C until analyzed. The cells were standardized to 1x106 cells/mL and stained for 2 minutes at 25°C with 1 μg/mL of Calcofluor White (Sigma Aldrich, St. Louis, MO) [79]. Cells were washed and immediately processed with an epifluorescence microscope (Axio Imager M1, 40X/0.6 lens, Zeiss Filter Set 02, Axio Cam MRc5, Axiovision 4.8; Carl Zeiss, Inc., Munich, Germany). ImageJ software (NIH.gov) was used to calculate fluorescence intesity per pixel. For flow cytometry, large and typical sized cells were first gated by forward scatter properties to distinguish size. Chitin/chitosan content was then determined by 405 nm laser excitation and fluorescence detection at ∼450 nm.
Biochemical chitin/chitosan quantification was adapted from Banks et al. [30]. Purified titan cell (>15 μm) and typical sized cell (<15 μm) samples collected from infected mice were each divided into two aliquots: one treated with acetic anhydride to fully acetylate the chitin/chitosan polymer and the other was left untreated. 5 μl of purified Streptomyces griseus chitinase (5 mg/ml in PBS) was added to hydrolyze chitin to N-acetylglucosamine (GlcNAc) and samples were incubated for 3 days at 37°C. For colorimetric determination of GlcNAc, the Morgan-Elson method was adapted for microplate readers. Chitinase-treated samples were incubated with 0.27 M sodium borate (pH 9.0) and heated at 99.9°C for 10 minutes. Immediately upon cooling to room temperature, freshly diluted 10X DMAB solution (10 g p-dimethylaminobenzaldehyde in 12.5 ml concentrated HCl and 87.5 ml glacial acetic acid) was added, followed by incubation at 37°C for 20 minutes. Absorbance at 585 nm was recorded for each sample. Standard curves were prepared from stocks of 0.2 to 2.0 mM of GlcNAc (Sigma). The amount of GlcNAc was calculated as mol/g cells (dry weight). The acetylated samples contained chitin plus chitosan, and the untreated sample contained chitin. The difference between the two measurements estimated the amount of chitosan.
Chitin was prepared as previously published [42]. Chitin from shrimp shells (Sigma Aldrich) was pulverized with a mortar and pestle. 12.5N HCl was added, and the slurry was incubated at 40°C for 30 minutes. Chilled 10N NaOH was added until a neutral pH was attained. The sample was centrifuged at 2x104 g for 5 minutes, the supernatant was decanted, and the sample was suspended in deionized water (dH2O). This step was repeated 3 times followed by a wash in ethanol. The sample was pelleted, suspended in dH2O, and filtered through a 10μm membrane (EMD Millipore, Billerica, MA). The solution containing <10 μm chitin was dried with a SpeedVac (Thermo Scientific, West Palm Beach, FL). The powder was weighed and resuspended in PBS to make a concentration of 50 mg/mL (i.e. 10X). Endotoxin was measured by Limulus amebocyte lysate assay (Associates of Cape Cod, East Falmouth, MA) and was found to be less than 0.03 EU/mL. Chitin heptamers were purchased from Carbosynth (Berkshire, UK). Mice were anesthetized and allowed to aspirate 125 μg of chitin suspended in 25 μL of PBS into the lungs at 0, 5, and 10 d.p.i. Pulmonary leukocytes from wildtype mice 14 days post-infection with KN99α were cultured and stimulated ex vivo for 5 hours with PMA + ionomycin (as previously described for Th cells) + 125 μg of chitin heptamers + golgi stop, or golgi stop alone (unstimulated) before processing by flow cytometry.
CycLex Acidic Mammalian Chitinase (AMCase) and Chitotriosidase (Chit1) Fluorometric Assays (MBL International, Woburn, MA) were used to detect chitinase activity. In brief, each of the following were added to pH 2 and pH 5 buffers containing 4-Methylumbelliferyl Chitotriose: 25 ng of recombinant AMCase, 25 ng recombinant Chit1, 10 μL of mouse lung homogenate, 10 μL of human plasma, 10 μL of lysate antigen, or 10 μL of culture supernatant of KN99α grown in YPD. The samples were incubated at 37°C in a Synergy H1 Microplate Reader (Biotek, Winooski, VT) and 360 nm excitation/450 nm emission readings were obtained every 2 minutes. The relative fluorescent units (RFU) at 1 hour of incubation were compared to the RFU of serial dilutions of 4MU standard, and the molar concentration of cleaved chitin was calculated.
The assay was performed as previously described [55]. Cell wall antigens were prepared from Cryptococcus neoformans, strain KN99α. The cells were flash frozen in liquid nitrogen, combined with glass beads, and vortexed vigorously for 2 hours at 4°C to disrupt the cells. The insoluble fraction (i.e., cell wall) was analyzed for protein concentrations (bicinchoninic acid protein assay; Thermo Fisher Scientific, Rockford, IL). Endotoxin levels in all antigen preparations were undetectable (<0.06 U/ml) by Limulus amebocyte lysate assay (Associates of Cape Cod, East Falmouth, MA). Whole-blood samples were obtained from AIDS patients at screening for the Cryptococcus Optimal Timing of Anti-retroviral Therapy Trial in Sub-Saharan Africa [71]. Peripheral blood samples from each subject were drawn into lithium heparin tubes, diluted 2-fold with PBS, and dispensed into a tissue culture plate. Cell wall antigens containing 5 μg of protein were added to the wells, and PBS was used as the “unstimulated” control. The plates were incubated at 37°C in 5% CO2 for 20 hours. After incubation, the plasma was separated from the cells and stored at 4°C until chitinase activity analysis.
P-values for pairwise comparisions were by Mann-Whitney U with Bonferroni adjustments for multiple comparisons. Global tests were by Kruskal-Wallis ANOVA. Surival curves were compared with Mann-Whitney tests. Power calculations were performed to assess appropriate sample size for all experiments. P-values ≤ 0.05 were considered statistically significant. All statistics and graphs were processed with Prism 6 (GraphPad Software, La Jolla, CA).
|
10.1371/journal.pcbi.1003545 | Prediction and Prioritization of Rare Oncogenic Mutations in the Cancer Kinome Using Novel Features and Multiple Classifiers | Cancer is a genetic disease that develops through a series of somatic mutations, a subset of which drive cancer progression. Although cancer genome sequencing studies are beginning to reveal the mutational patterns of genes in various cancers, identifying the small subset of “causative” mutations from the large subset of “non-causative” mutations, which accumulate as a consequence of the disease, is a challenge. In this article, we present an effective machine learning approach for identifying cancer-associated mutations in human protein kinases, a class of signaling proteins known to be frequently mutated in human cancers. We evaluate the performance of 11 well known supervised learners and show that a multiple-classifier approach, which combines the performances of individual learners, significantly improves the classification of known cancer-associated mutations. We introduce several novel features related specifically to structural and functional characteristics of protein kinases and find that the level of conservation of the mutated residue at specific evolutionary depths is an important predictor of oncogenic effect. We consolidate the novel features and the multiple-classifier approach to prioritize and experimentally test a set of rare unconfirmed mutations in the epidermal growth factor receptor tyrosine kinase (EGFR). Our studies identify T725M and L861R as rare cancer-associated mutations inasmuch as these mutations increase EGFR activity in the absence of the activating EGF ligand in cell-based assays.
| Cancer progresses by accumulation of mutations in a subset of genes that confer growth advantage. The 518 protein kinase genes encoded in the human genome, collectively called the kinome, represent one of the largest families of oncogenes. Targeted sequencing studies of many different cancers have shown that the mutational landscape comprises both cancer-causing “driver” mutations and harmless “passenger” mutations. While the frequent recurrence of some driver mutations in human cancers helps distinguish them from the large number of passenger mutations, a significant challenge is to identify the rare “driver” mutations that are less frequently observed in patient samples and yet are causative. Here we combine computational and experimental approaches to identify rare cancer-associated mutations in Epidermal Growth Factor receptor kinase (EGFR), a signaling protein frequently mutated in cancers. Specifically, we evaluate a novel multiple-classifier approach and features specific to the protein kinase super-family in distinguishing known cancer-associated mutations from benign mutations. We then apply the multiple classifier to identify and test the functional impact of rare cancer-associated mutations in EGFR. We report, for the first time, that the EGFR mutations T725M and L861R, which are infrequently observed in cancers, constitutively activate EGFR in a manner analogous to the frequently observed driver mutations.
| Cancer is a complex disease in which healthy cells undergo a series of genetic changes, eventually becoming cancerous, growing uncontrollably and spreading throughout the body [1]. Identification of the specific genetic changes that promote cancer traits within a cell can yield clues into potential treatments for the disease. Large-scale cancer genome sequencing studies have thus been initiated in order to catalog the mutations observed in human cancers [2]–[6].
Not all mutations have equal influence on the disease state of a cell, however. Certain mutations, called “drivers,” are known to have a causative effect, driving the transformation of a cell from healthy to cancerous, often by promoting cell growth or inhibiting apoptosis (programmed cell death) [1]. In contrast, the majority of mutations do not significantly affect the cancer characteristics of a cell, and can be considered relatively benign “passengers” in a tumor cell [7]. Mutated driver genes are worthwhile targets for drug discovery, because counteracting the mutation's effects can potentially slow or reverse cancer progression in individual patients [8], [9]. To fully realize this potential, however, there is a need to develop computational approaches that can (i) distinguish causative from non-causative mutations, and (ii) identify key causative mutations for experimental studies and clinical targeting.
Indeed, several previous studies have proposed methods to predict causative mutations in cancer genomes (reviewed in [6]). These methods fall into three major categories: (i) frequency-based approaches, (ii) structure-based methods and (iii) statistical and machine learning methods. Frequency-based approaches are based on the assumption that the mutations that occur in multiple patient samples are likely to be those that are causative [10]. Although such assumptions are valid for some recurrent mutations such as the L858R mutation in epidermal growth factor receptor (EGFR) in lung cancer [11] and V600E in BRAF in melanoma [12], [13], there is emerging evidence that rare mutations can be drivers [7], [14]–[16]. Moreover, a comprehensive analysis of several breast and colorectal cancer genomes revealed that the genomic landscapes of these cancers are dominated by a large number of rare gene mutations rather than recurrent oncogene mutations [17].
Structure-based methods offer a powerful way of predicting the impact of mutations by taking into account the three-dimensional context of the mutated residues [18]–[21]. However, such approaches are not applicable on a genome-wide scale because of the lack of experimental structure information for several oncogenes.
Because a wide variety of factors play into the oncogenic effect of any given mutation, machine learning approaches have become a method of choice to predict causative mutations based on a variety of contextual information [18], [22]–[30]. In general, supervised machine learning approaches learn from the features of known cancer-associated and benign mutations to classify unknown mutations. (Note that the term “cancer-associated” refers to mutations that are predicted to have direct or indirect oncogenic effect, while the term “driver” refers to confirmed causative mutations.) Although several machine learning-based methods have been proposed previously, there still remains a need to improve the sensitivity and efficacy of existing methods [28]. For example, most existing approaches use standard features of mutated residues for training the classifier and do not take into account gene- or family-specific features that can improve prediction accuracy [28]. Furthermore, existing approaches use one or two common machine learning algorithms and do not consider the biases introduced by these algorithms. Finally, existing approaches typically provide a binary “yes” or “no” classification for disease association, which does not solve the problem of prioritizing candidate mutations for follow-up experimental studies [31].
Here we apply a novel machine learning approach to predict and prioritize cancer-associated mutations in protein kinases, a class of proto-oncogenes frequently mutated in human cancers. Our approach differs in three major ways from previous approaches. First, we introduce new kinase-specific features, beyond those used in previous methods [28],[32],[33], to improve prediction accuracy. Mutations in the kinase domain, and particularly those at functional sites, have been shown to be more likely to be oncogenic [34], typically through mechanisms that constitutively activate the kinase [35]. Second, we use a multiple classifier approach (ensemble method), which by combining multiple machine learning algorithms (individual classifiers), overcomes the biases introduced by each method. Finally, we use our combined classifier to produce a numerical ranking of cancer-associated mutations in EGFR and test the impact of predicted mutations on EGFR kinase activity using cell-based assays. Our studies identify T725M and L861R as rare cancer-associated mutations in that EGFR kinase harboring these mutations display constitutive kinase activity.
The data sources used for training and evaluating the classifiers consist of a “positive” set of known cancer-associated mutations, a “negative” set of known benign mutations, and several attributes of proteins and amino acids, which we drew from several databases. The parameter settings used and detailed results for construction and evaluation of each classifier are described in detail in Supplementary Text S1.
Mutations are uniquely identified by the gene name, protein sequence position, wild-type amino acid, and mutant amino acid type. We extract features related to biochemical, structural, functional and evolutionary properties, which in the end generated 29 features in total, as follows.
Of these 29 features, 23 were previously explored by others, including amino acid biochemical properties, sequence conservation, and kinase subdomain [25], [27], [44], [45]. Our novel features are the protein kinase classification terms (group and family) and the conservation levels of the wild type and consensus type within alignments of all, group– and family-specific kinase sequences.
Feature selection serves two purposes: to choose a smaller, more computationally tractable subset of meaningful features which can be used to effectively predict the target attribute (causative or non-causative), and to understand the relative usefulness or contribution of each feature toward predicting the target [52]. With emphasis on the latter, we independently applied five feature selection algorithms, namely OneR algorithm [53], relief-based selection [54], chi-square selection, a gain-ratio-based filter approach [55] and correlation-based selection [56], to evaluate our attributes.
Evaluation was performed on the combined positive and negative data sets, with the positive set also including COSMIC mutations that occur only once, in order to obtain a larger data set for this step. The detailed feature selection results obtained from the five selection methods with 10-fold cross-validation are given in Table S2.
We used the 10-fold cross-validation routine implemented in Weka [57] to select the most relevant features from all the attributes we considered. As with other parts of the learning cycle, 10-fold cross-validation randomly splits the data up into 10 disjoint subsets. However, in the feature selection evaluation routine only the training folds are used, and there is no testing as such [58]. Feature selection is run on each training fold (90% of the data) in turn and the results are summarized. In the case of single-attribute evaluators (in our case OneR, Chi-Square, Relief, and Gain-Ratio), the output shows the average merit and average rank of each attribute over the 10 folds along with their respective standard deviations. In the case of subset evaluators such as Correlation-based feature selection in our case, the output shows, for each attribute, in how many folds it was part of the final best subset selected. In both these cases the aim is to provide some measure of robustness as well as stability of the feature with respect to small changes in the distribution of input attribute values.
The attributes were ranked in terms of effectiveness as a predictor according to each selection method. Those attributes selected by at least 3 out of 5 (60%) of the selection methods were retained, yielding a final feature subset of 17 selected features (from the original 29) which we used as input for the training process in subsequent analysis (Table 1). We also determined a score indicating the usefulness of each selected feature by taking the arithmetic mean of the feature's ordinal ranking across all 5 selection algorithms. This “average rank” score enables a complete ranking of the selected features, with scores ranging from 1.4 to 13 (Table 1).
To classify point mutations in human protein kinase sequences as either cancer-associated or non-causative, we applied 11 machine learning methods to our dataset. The machine learning methods are J48 (Tree) [59], [60], Random Forest [61], NB Tree [62], Functional Tree [63], Decision Table [64], DTNB [65], LWL (J48+KNN) [66], Bayes Net [67], Naive Bayes [68], SVM [69], [70], and Neural Network [71].
The detailed results of evaluation of each classifier as well as several alternative approaches are described in Supplementary Text S1.
Having evaluated the 11 trained models that are described in the previous section, we selected the models trained on the combined well-performing positive sets — mutations that appear more than once in the COSMIC dataset — for further application. We focused on the gene EGFR, a protein kinase that is frequently mutated in lung cancer, and used the previously trained models to evaluate the EGFR mutations that appear only once in the COSMIC dataset, as these mutations were excluded from the initial training set. Since these mutations have not been replicated in other tumor samples, it is more likely that some of them are not significantly associated with cancer. Thus, we use the following approach to combine the predictions of the trained classifiers to sort these EGFR mutations by their likelihood to be cancer-associated.
For each of the non-synonymous point mutations in the kinase domain of EGFR that were observed only once in COSMIC, we calculated a numerical score for the likelihood of a given mutation to be cancer-associated using two different approaches: a simple majority voting approach with one “vote” per classifier, and a more sophisticated approach in which each classifier's “vote” is weighted by its accuracy as previously estimated by cross-validation. Mutations that have been classified as cancer-associated by more classifiers are considered more likely to be true positives, whereas fewer “votes” indicate a mutation is less likely to be cancer-associated.
We first sought to determine a subset of features that show high predictive value in distinguishing cancer-associated from benign mutations, and to evaluate the contribution of the kinase-specific features we introduced in this study, namely the hierarchical protein kinase classification levels (group, family) and the conservation levels at each evolutionary depth (all kinases, group and family).
We applied 5 different feature selection algorithms (see Methods), each of which selected a subset of the full feature set, to produce a ranking of the 29 features, 17 of which met our criteria for inclusion in the final feature subset used for training the classifiers (Table 1). All 5 selection algorithms selected the features “Protein kinase family” and “Protein kinase group”, and each individual algorithm ranked these two features at the top. The features “amino acid type (WT)”, “BLOSUM62 score”, and “side chain polarity (Mutant)” were also selected by all 5 algorithms and ranked highly by individual algorithms. Conservation scores of the wild-type residue among all kinases, and of the alignment consensus type among all kinases, among the major groups and among major families, were also ranked highly, indicating that they extensively contribute to the prediction of the target attribute, a result that supports the importance of our novel proposed features.
To further test the contribution of the new features, we re-ran our classification experiments after removing the novel kinase-specific features from the 17 features identified through the feature selection process. Notably, the performance (as indicated by accuracy values in Supplementary Dataset S1) reduces substantially when the kinase-specific features are removed. We then performed a chi-squared test of the number of correct and incorrect predictions made by each of these two classifiers on the 557 mutations in our final training set: 551 correct and 6 incorrect versus 498 and 59 for the full and reduced feature sets, respectively. This statistical test confirmed that the decrease in accuracy when the novel features are removed is significant at p<.001 (, , simulated based on 100,000 replicates).
It is also interesting to note that the performance of the combined classifier is much less degraded when the kinase-specific features were removed compared to the single classifier (Dataset S1), suggesting that the multiple classifier approach contributes to stability and robustness.
We used a weighted voting approach to combine 11 single classifiers to be a more robust ensemble classifier.
Table 2 and Table 3 present the in silico experimental results in terms of confusion matrix and several other measurement indexes which quantify the performance of the individual classifiers. All 11 classifiers performed fairly well, with recall rates at least 95.6% and False Positive (FP) rates at most 10.6% (Table 3). Of the individual classifiers, SVM performed the best on most metrics. However, the combined classifier performs better than the individual classifiers, reaching 98.7% for both precision and recall.
An alternative metric is the “F-Measure”, a harmonic mean of precision and recall, on which all single classifiers achieved a score of at least 0.919 (Table 3), a result consistent with previous studies [27]. The high F-measure score of 0.989 for the combined classifier also vouches for the stability of our feature set on the relatively small training dataset. Furthermore, the competitive performance of the 11 single classifiers suggests that they each contribute to the improved performance of the multiple (ensemble) classifier.
We used two separate 10-fold cross-validation loops, one for feature selection and another for training and testing. Using the cross-validation terminology described in [58], our approach is considered an OUT method (in which feature selection is done outside the training/testing loop) rather than an IN method (in which feature selection is in the same loop as training and testing). This may have caused a problem called “information leak” [58] due to the fact that the full data set was exposed to the feature selection methods before the training/testing cross-validation loop. However, the potential information leak is partially compensated for by the robust and comprehensive approach we used for feature selection, using multiple feature selection methods and cross-validation loops for each of them (see Methods). In the design of this study, it was necessary to use a single, fixed set of features for both supervised and unsupervised learning (discussed below). Furthermore, the OUT method does not affect the relative performance of the different classifiers [58], which is more important in this study than the absolute accuracy of each classifier. Nevertheless, the performance evaluation of the 11 single classifiers and the combined classifier shown in Table 3 should be interpreted with caution given the potential for bias due to information leak.
Since there exists a level of uncertainty in the labels (“cancer-associated” and “benign”) in our dataset, the predictive model that is trained by the supervised learning approach, resulting in the combined classifier, might be biased. In this section, we denote the prediction of the combined classifier as the Supervised score (S-Score). We introduce another unsupervised learning module to help reduce the label uncertainty. The unsupervised module performs clustering using Euclidean distance in the space of the 17 selected features, without considering the labels, and the labels are only used for the computation of the Unsupervised Score (U-Score), which measures cancer-association based on clustering in the feature space. We conducted further analysis on our dataset by combining and comparing both S-Score and U-Score (see section “Learning Methods” in Supplementary Text S1) because such comparisons can potentially reveal suspicious mutations labeled incorrectly. Mutations with both S-Score and U-Score above 0.5 are considered cancer-associated while mutations with both U and S-scores below 0.5 are considered benign. All other mutations are considered uncertain, or suspicious.
Our analysis reveals that majority of cancer-associated and benign mutations fall into “Expected” clusters. Specifically, 219 out of 226 instances (≈97%) labeled as cancer-associated fall into the “Expected” category, and 255 out of 331 instances (≈77%) labeled as benign fall into the “Expected” category (see section “Identifying Suspicious Mutations in COSMIC-FG1 v.57” in Supplementary Text S1).
We combined the 11 classifiers to effectively identify and prioritize rare EGFR mutations for experimental studies. We ranked the unconfirmed mutations in EGFR using the combined classifier (Table 4). The detailed results and log files of the computational experiments are given in Table S3.
EGFR point mutations T725M, E746K and L861R showed increased auto-phosphorylation compared to WT as shown in Figure 2. T725M and L861R resulted in hyper-phosphorylation at almost all the sites examined, i.e. Y1086, Y1045, Y845, Y1173 and Y1068. E746K, however, showed enhanced phosphorylation at Y1068, Y845, Y1173 and Y1068. The mutations G724S and L858Q, ranked 2 and 25 respectively (Table 4), did not show significant difference in C-terminal tail autophosphorylation compared to wild-type EGFR (Figure 2; Figure 3). However, G724S and L858Q showed elevated levels of AKT phosphorylation in the presence of EGF compared to WT and other mutants (Figure 2).
In this study, we reported novel features and a multiple classifier approach for identifying cancer-associated mutations in the cancer kinome. Our studies revealed that: (i) the depth of conservation of the mutated residue is a useful, novel feature for predicting cancer-associated mutations; (ii) combining multiple classifiers can improve prediction accuracies; and (iii) our novel features and multiple classifiers can be effectively applied in the identification of rare mutations in EGFR.
Mutational activation of EGFR is implicated in many cancers including lung, head and neck cancer, and clinical and cancer genome sequencing studies have identified hundreds of mutations in the protein kinase domain. However, much of the focus thus far has been on a handful of frequently observed mutations such as L858R and L861Q, while relatively little is known about the many rare mutations in EGFR such as T725M. The scoring scheme and multiple classifier approach we have introduced here help identify key rare mutations for follow-up experimental studies.
In particular, our studies suggest T725M as a likely cancer-associated mutation because it increases EGFR auto-phosphorylation activity in comparison to wild-type and other activating mutations such as L858R. The impact of the T725M mutation cannot be predicted by existing structural or functional information alone, and clinical samples do not currently highlight this as a highly recurrent mutation, as it appears only once in COSMIC v.50 and twice in COSMIC v.57. T725, however, is predicted to be a likely phosphorylation site [80], [81]. Thus, it is possible that the T725M mutation activates the kinase through loss of an inhibitory phosphorylation site. Indeed, mutational gain or loss of phosphorylation sites has been previously noted in cancer datasets [82], [83].
The frequently observed mutation L861Q is known to activate EGFR, but the impact of the rare L861R is not known. Here we showed that L861R activates EGFR in the absence of the activating EGF ligand, suggesting that it is also likely to be cancer-associated.
L858Q and G724S are two predicted mutants that do show appreciable change in EGFR autophosphorylation (Figure 2; Figure 3). This does not necessarily mean that these mutations are not causative, as these mutations can alter other aspects of EGFR signaling not considered in our studies. For example, recent studies showed that cancer mutations alter the temporal regulation and phosphorylation rates of the C-terminal tail tyrosines in EGFR [84]. Such changes in temporal regulation can contribute to abnormal downstream signaling without appreciable change in the level of C-terminal tail phosphorylation. Consistent with this view, the G724S and L858Q mutants increase phosphorylation of AKT despite little or no change in EGFR autophosphorylation (Figure 2). Although these observations must be further investigated through in vitro studies, the machine learning approach appears have used multiple correlative features to predict the causativeness of these mutants (see Table 5).
L858R is a recurrent lung cancer mutation which activates EGFR and also impacts drug binding [11]. This information perhaps contributed to the classification of L858Q as cancer-associated. However, our mutational experiments revealed that the L858Q mutation does not significantly alter the levels of EGFR autophosphorylation. However, as mentioned above, L858Q does alter downstream AKT phosphorylation (Figure 2). The context of L858Q suggests that the activation loop of EGFR is a frequent site of activating mutations; however, the L858Q mutation appears to alter downstream signaling in a manner distinct from L858R.
As our catalog of known drivers improves we can further improve our prediction system, using additional features such as protein dynamics and atomic details, and machine learning techniques such as semi-supervised learning [85] and clustering [86] to build a more sophisticated model to differentiate between causative and non-causative mutations in cancer. Moreover, our work could be extended to a prediction tool with clinical value, as well as provide a basis for further investigation into the relationship between protein evolution and disease.
|
10.1371/journal.pgen.1008295 | Age- and stress-associated C. elegans granulins impair lysosomal function and induce a compensatory HLH-30/TFEB transcriptional response | The progressive failure of protein homeostasis is a hallmark of aging and a common feature in neurodegenerative disease. As the enzymes executing the final stages of autophagy, lysosomal proteases are key contributors to the maintenance of protein homeostasis with age. We previously reported that expression of granulin peptides, the cleavage products of the neurodegenerative disease protein progranulin, enhance the accumulation and toxicity of TAR DNA binding protein 43 (TDP-43) in Caenorhabditis elegans (C. elegans). In this study we show that C. elegans granulins are produced in an age- and stress-dependent manner. Granulins localize to the endolysosomal compartment where they impair lysosomal protease expression and activity. Consequently, protein homeostasis is disrupted, promoting the nuclear translocation of the lysosomal transcription factor HLH-30/TFEB, and prompting cells to activate a compensatory transcriptional program. The three C. elegans granulin peptides exhibited distinct but overlapping functional effects in our assays, which may be due to amino acid composition that results in distinct electrostatic and hydrophobicity profiles. Our results support a model in which granulin production modulates a critical transition between the normal, physiological regulation of protease activity and the impairment of lysosomal function that can occur with age and disease.
| Progressive decline in maintenance of protein homeostasis clearly contributes to the development of neurodegenerative disorders, yet the molecular basis of this decline is poorly understood. Here, we take advantage of molecular genetic techniques available in the model organism C. elegans to investigate the mechanism underlying neurodegenerative disease due to mutations in the progranulin gene. We find that age, gene mutation and physiological stress lead to the accumulation of lysosomal granulins (the cleavage products of the progranulin protein) thereby disrupting cellular protein homeostasis. Granulin expression impairs animal fitness, resistance to stress and neuronal function, and stimulates a lysosomal stress response in an attempt to up-regulate lysosomal genes and restore normal function. Our findings are particularly important because they suggest a new, rational target—inhibition of progranulin cleavage into granulins—for neurodegenerative disease therapy.
| Aging and stress are thought to enhance neurodegenerative disease risk through the accumulation of misfolded and aggregated proteins [1–3]. The lysosome is the key degradative organelle within the cell [4], and therefore plays a pivotal role in the maintenance of protein homeostasis. It contains specialized enzymes, called cathepsins, which work optimally at the acidic pH in this compartment and have a crucial role in processing and degrading proteins [5]. The transcription factor EB (TFEB) controls the expression of genes involved in lysosomal biogenesis and function [6, 7]. TFEB dysregulation has been associated with neurodegenerative disease [8, 9] and its overexpression may help to promote the clearance of protein aggregates [10, 11]. Although, genetic and functional studies have implicated lysosomal dysfunction in the pathogenesis of multiple neurodegenerative diseases [12–14], understanding of the molecular basis of this phenomenon remains incomplete.
Heterozygous progranulin (PGRN) loss-of-function mutations lead to autosomal dominant transmission of the neurodegenerative disorder frontotemporal lobar degeneration (FTLD) with TAR DNA binding protein 43 (TDP-43) inclusions [15–17]. The molecular function of the progranulin protein (PGRN) remained elusive until it was indelibly linked to lysosomal function by the finding that loss of both gene alleles results in the lysosomal storage disease, neuronal ceroid lipofuscinosis [18]. Progranulin localizes to lysosomes [19–22] where it may act to promote lysosomal biogenesis and function [20, 23–25].
The progranulin (PGRN) protein can be proteolytically cleaved to liberate multiple cysteine-rich “granulin” peptides [26]. Granulins are highly conserved, disulfide-bonded miniproteins with unknown biological function [27–31]. Like progranulin, granulin peptides have been shown to localize to the endolysosomal compartment [32], and can be generated through the action of cysteine proteases on progranulin [32–34]. Owing to the twelve cysteines and six disulfide bonds found in each cleaved granulin, these peptides adopt a stacked β-sheet configuration that is compact, structurally stable and potentially protease resistant [35]. Several lines of evidence exist that cleaved granulin peptides oppose the function of the full-length protein. While progranulin has proliferative [35, 36] and anti-inflammatory [37, 38] properties, granulin peptides have been shown to inhibit cell growth [35] and stimulate inflammation [38]. In addition, we have previously demonstrated a role for C. elegans granulins in selectively promoting the accumulation of TDP-43, thereby exacerbating TDP-43 toxicity and potentially contributing to the pathogenesis of disease [39]. However, the mechanism by which granulins exert this specific regulation on TDP-43 metabolism remains unknown. C. elegans provides many advantages as a model system to study granulin function, including conservation of the progranulin gene, and the many available molecular and cell biology techniques.
In this study, we further investigate the molecular mechanisms of C. elegans granulins on lysosomal function and protein homeostasis. We show that C. elegans granulins localize to the endolysosomal fraction. Granulin production increases with age and stress, and granulin expression reduces animal fitness by impairing lysosomal protease expression and activity. This prompts cells to activate a compensatory transcriptional program involving HLH-30/TFEB nuclear translocation and up-regulation of the transcription of HLH-30/TFEB-related genes. Overall, our findings highlight granulins as critical regulators of proteolytic lysosomal function and potential drivers of neurodegenerative disease pathogenesis.
We have previously shown that C. elegans progranulin (pgrn-1) null mutants exhibit enhanced resistance to endoplasmic reticulum (ER) unfolded protein stress [40]. As a genetic null, pgrn-1(-) animals produce neither full-length progranulin nor cleaved granulins; therefore, absence of either the holoprotein or the cleavage fragments could be responsible for the ER stress resistance. Based on our earlier finding that granulins could exacerbate TDP-43 toxicity [39], we hypothesized that the bioactive granulins were responsible for inhibiting ER stress resistance. Hence, to isolate granulin activity, we expressed individual C. elegans granulins 1, 2 and 3 at comparable levels in a pgrn-1 null background [39]. Granulin expression in a progranulin null background completely abolished the ER stress resistance phenotype (Fig 1A). In contrast, animals over-expressing C. elegans full-length progranulin in a progranulin null background remained ER stress resistant (Fig 1B). Over-expressed full-length progranulin was not cleaved under ER stress (S1A–S1D Fig), and could promote ER stress resistance in the presence of granulin (S1E Fig). Furthermore, transgenic expression of human tau protein and TDP-43 in a progranulin null background did not abrogate ER stress resistance (S1F Fig). Taken together, these data suggest that it is the granulins, and not full-length progranulin, that specifically inhibit ER stress resistance.
Given that granulins impair ER stress resistance, we wondered if they might more broadly impact protein homeostasis. Thus, we measured endogenous levels of heat shock protein HSP-4, the nematode homolog of human BiP/Grp78 [41]. HSP-4/BiP expression is upregulated during the unfolded protein response (UPR) [42]. We found that granulin-expressing animals displayed a trend for increased basal expression of HSP-4/BiP on day 1 of adulthood, reaching significance in animals expressing granulin 2 and 3 (Fig 1C). Therefore, in the absence of the progranulin holoprotein, granulin expression upregulates HSP-4 and this is indicative of UPR induction and perturbed protein homeostasis.
While working with the granulin-expressing lines, we noted a decrease in overall animal fitness attributable to the granulins. Granulin production significantly reduced animal viability by lowering the number of eggs that hatched and slowing the development of animals to maturity (Fig 1D). Granulin-expressing animals that did reach adulthood were smaller in size (Fig 1E). Short-term associative learning can be assayed in C. elegans using a positive olfactory learning paradigm [43, 44]. When granulin-expressing animals were tested in this assay they underperformed compared to controls (Fig 1F), suggesting that granulin expression may result in neuronal dysfunction. These data, coupled with previous work by others on the function of progranulin [35–38], suggest that granulins impair animal fitness, resistance to stress and neuronal function, while progranulin promotes these qualities.
To establish the trafficking and localization of granulin peptides within a whole organism, we utilized microscopy and biochemistry techniques. First, we determined the sub-cellular localization of full-length progranulin using a translational progranulin reporter, PGRN-1::RFP, and organelle-specific markers. As expected, in cells that secrete progranulin, such as the intestine, the reporter co-localized with both a Golgi marker, mannosidase II (Fig 2A), and a lysosomal marker, lysosomal-associated membrane protein 1 (LMP-1) (Fig 2B). However, in coelomocytes, a cell type that takes up but does not produce progranulin [45], the progranulin reporter was only seen in the endolysosomal compartment (Fig 2C–2E), suggesting that extracellular progranulin is transported through endosomes to reach the lysosome.
Having established that progranulin can be trafficked from one tissue type to another, we next sought to better understand the subcellular localization of granulin peptides. To do so, we developed a protocol for subcellular fractionation of C. elegans. The purity of cytosolic, ER and endolysosomal fractions was confirmed with established markers (S2 Fig). Individual granulins that were transgenically expressed also demonstrated lysosomal localization (Fig 2F–2I). Therefore, C. elegans progranulin and granulins localize to the endolysosomal compartment.
In C. elegans and mammals, progranulin production increases with age [45, 46] and injury [47, 48]. However, the degree to which granulin peptides are liberated has not been measured. We first asked if progranulin cleavage into granulins increases with age. Using our PGRN-1::RFP translational reporter, we found that granulin production does indeed increase in an age-dependent fashion (Fig 3A and S3A and S3B Fig), suggesting that either an increase in expression and cleavage of progranulin, and/or an age-associated decline in granulin turnover, contributes to granulin accumulation. Granulin cleavage also increased in response to certain physiological stressors such as starvation (Fig 3B and S3C and S3D Fig). Thus, age and stressful stimuli, such as starvation, appear to promote the cleavage of full-length progranulin into granulins.
In order to determine the subcellular compartment in which cleaved granulin peptides are produced, we performed fractionation of fed or starved animals expressing the PGRN-1::RFP reporter. In fed animals, full-length progranulin was enriched in the endolysosomal fraction with very little lower molecular weight granulin observed in any fraction (Fig 3C). Upon starvation, the cleaved granulins increased primarily in the endolysosomal fraction, confirming that the majority of the age and stress-induced granulins are, in fact, endolysosomal (Fig 3C). Therefore, granulin peptides are produced in vivo in the endolysosomal compartment in a stress-responsive manner.
Given that granulins impair organismal fitness, localize to the endolysosomal fraction and impair stress resistance, we next investigated their impact on lysosomal morphology. In C. elegans, coelomocytes scavenge and detoxify the pseudocoelomic cavity and therefore have a well-developed endo-lysosomal system [49]. Although we could not image coelomocyte lysosomes in granulin 1-expressing animals due to the presence of a GFP co-expression marker, we found that both loss of progranulin and expression of granulins 2 and 3 grossly deformed these organelles (Fig 3D–3G). Lysosomes lost their spherical shape, more frequently exhibited membrane protrusions and tubular extensions (Fig 3D–3G), and became smaller in size, reaching significance for pgrn-1(-) animals and pgrn-1(-); granulin 3(+) animals (Fig 3H). Together, these data suggest that granulin peptides accumulate in endolysosomes with age and starvation, where they, as well as loss of progranulin, may disrupt lysosomal morphology.
As we observed that expressed granulins disrupt lysosomal morphology, we next assessed their effect on lysosomal function by measuring the expression level and enzymatic activity of lysosomal proteases in lysates from granulin-expressing C. elegans. Granulin expression resulted in decreased protein levels of ASP-3, the nematode ortholog of mammalian cathepsin D (CTSD), reaching significance in granulin 2-expressing animals (Fig 4A). Expression of all granulins significantly reduced CPL-1 expression, the nematode ortholog of mammalian cathepsin L (CTSL) (Fig 4B). This decrease in protease expression correlated with a decrease in protease activity (Fig 4C and 4D), reaching significance in granulin 2 and 3-expressing animals for ASP-3 activity and granulin 1 and 2-expressing animals for CPR/CPL-1 activity. Overall, our data suggest that granulin peptides disrupt C. elegans lysosomal protease activity in vivo.
As we observed differences between the three granulins in terms of the magnitude of their phenotypic effects within assays, we sought to determine whether these differences might be explained by variations in their amino acid sequence and physicochemical properties. C. elegans granulins 1, 2 and 3 share less than 50% sequence identity among themselves (Fig 5A), and less than 40% when excluding the highly conserved network of disulfide bonds. Electrostatic analysis (Fig 5B) shows that granulin 3, located at the C-terminus of C. elegans PGRN-1, is positively charged at neutral pH, while granulin 1, the N-terminal granulin domain, remains negatively charged at all analyzed pH values (pH = 4 to 8). The central granulin 2 domain has little to no overall net peptide charge at neutral pH. A further comparison of granulin hydrophobicity (Fig 5C) shows that the central region of granulin 2 (residues 202 to 221) and granulin 3 (residues 309 to 326) is predominantly hydrophobic, as measured by Kyte and Doolittle (K&D) hydrophobicity scores greater than zero. In contrast, the K&D score for the corresponding region of granulin 1 (res. 120 to 139) is slightly negative. While the functions of the individual C. elegans granulin domains remain to be further elucidated, these observed differences might suggest that each domain participates in unique protein-protein interactions (PPIs), and thus differing roles in the endolysosomal system.
We further compared the C. elegans granulin sequences with those of different species, including Homo sapiens (H. sapiens), Mus musculus (M. musculus) and Danio rerio (D. rerio) (Fig 5A). We found that C. elegans granulins share higher identity scores to certain granulins from other species than among themselves. Similar to C. elegans granulins, differences in pH-dependent electrostatics (Fig 5B) were noticeable for all species studied, with a recurring trend for the C-terminal granulin domains being the most positively charged. The low sequence identity and distinct physicochemical properties among the granulin domains were also observed for H. sapiens, M. musculus and D. rerio, contrasting with the highly conserved network of disulfide bonds. Taken together, these data highlight the importance of the amino acid residues situated outside of the well-conserved granulin sequence consensus for contributing to the charge and hydrophobicity profiles of each granulin domain. These may drive unique recognition patterns for PPIs that may ultimately be relevant in a disease context.
To determine if granulin-induced disruption of lysosomal morphology and function promoted a transcriptional response, we performed RNA-seq profiling of wild-type, pgrn-1(-) and granulin-expressing animals (S1–S5 Tables). Visual inspection of the RNA sequencing reads confirmed a high and comparable expression of granulin 1, 2 and 3 transgenes, as well as a read drop-out in progranulin null animals (S4A Fig). Wild-type animals had a low but detectable expression of endogenous progranulin transcript (S4A Fig). We first compared pgrn-1(-) or pgrn-1(-); granulin animals to wild-type animals. Compared to wildtype, a total of 7084 differentially expressed genes (DEGs) were identified across all strains (Fig 6A and S4B Fig). The majority of DEGs identified for pgrn-1(-) animals were down-regulated compared to wild-type animals. These DEGs were enriched for GO terms associated with growth, development, cation and sugar binding (S4C and S4D Fig). In contrast, the majority of DEGs for granulin-expressing animals were up-regulated compared to both wild-type and pgrn-1(-) animals (Fig 6A and S4B Fig). GO term analysis for DEGs in granulin-expressing animals showed a shared enrichment in genes associated with lysosomal function, including protein metabolic process and hydrolase activity acting on ester bonds (S4E–S4K and S5 Figs). Expression of granulin 2 resulted in the highest number of DEGs compared to both wild-type and pgrn-1(-) animals, followed by granulin 3 and then granulin 1 (Fig 6A and S4B Fig). The observed overlap in enriched GO terms on granulin 2 and 3 expression further suggests similarities between these two granulins compared to granulin 1, and also reflects the phenotype severity observed in development and behavioral assays. Interestingly, the upregulated DEGs identified in pgrn-1(-); granulin 3(+) animals were significantly enriched for genes whose promoters contained the putative TFEB binding site E-box sequence 5’-CACGTG-3’ (P = 0.011). This trend was also observed in the upregulated DEGs for pgrn-1(-); granulin 1(+) (P = 0.149) and pgrn-1(-); granulin 2(+) (P = 0.097) but did not reach statistical significance. TFEB is the master lysosomal transcription factor that regulates lysosomal biogenesis and autophagy [6, 7], and the C. elegans TFEB is HLH-30 [50].
In response to starvation, stressful stimuli and aging, HLH-30/TFEB translocates from the cytosol to the nucleus to activate its transcriptional targets [6, 7, 50, 51]. This program, known as the Coordinated Lysosomal Expression and Regulation (CLEAR) response induces expression of genes involved in lysosomal function and autophagy, including progranulin. We assessed HLH-30/TFEB cytoplasmic versus nuclear localization in control, pgrn-1(-) and granulin expressing animals. Granulin expression promoted nuclear localization of HLH-30/TFEB, reaching significance in granulin 3-expressing animals (Fig 6B and 6C). This effect was not seen in pgrn-1(-) animals where a much lower number of DEGs were identified, and was also not observed in pgrn-1(-) animals expressing human tau or TDP-43 protein (S6A Fig). These results suggest that the disruption of lysosomal morphology and protein homeostasis seen in granulin-expressing animals leads to a specific compensatory translocation of HLH-30/TFEB from the cytosol to the nucleus.
When granulin-expressing animals were crossed into a wildtype background, the presence of wildtype progranulin partially mitigated the negative effects of granulin-expression on development (S6B Fig), lysosome morphology (S6C and S6D Fig) and HLH-30/TFEB localization (S6E Fig). Interestingly, granulin-expression in a wildtype background resulted in higher ER stress sensitivity than granulin-expression in a progranulin null background (S6F Fig). We speculate that ER stress may promote the cleavage of endogenous PGRN, resulting in even higher levels of cleaved granulins (endogenous and transgenic granulins) and enhanced ER stress sensitivity. These data further suggest a reciprocal relationship between full-length progranulin and cleaved granulins, and highlights that their relative levels may be important for normal animal development and fitness.
To determine if the upregulation of TFEB target genes was a compensatory transcriptional response in granulin-expressing animals, we crossed these animals into an hlh-30(-) null background. When lacking hlh-30, granulin-expressing animals had further impairments in overall fitness, with fewer growing to adulthood (Fig 6D) and more arresting at early larval stages (Fig 6E). Together, these data demonstrate that granulin expression, even in the absence of stress or starvation, is sufficient to activate a compensatory CLEAR response and induce expression of genes containing TFEB binding sites. Overall, the ability of granulins to 1) impair a proteotoxic stress response, 2) disrupt lysosomal morphology, 3) direct TFEB to the nucleus and 4) induce a CLEAR response indicates that granulin-dependent impairment of lysosomal function negatively impacts cellular protein homeostasis (Fig 6F).
We have previously shown in C. elegans that expression of granulin peptides enhances TDP-43 toxicity and prevents its degradation [39]. In this study, we sought to understand the mechanism by which granulins exert their effects and determine if they more broadly impacted protein homeostasis. We found that granulins are produced in an age and stress-dependent manner, and consequently impair lysosomal protease expression and activity. Their expression negatively impacts cellular protein homeostasis and drives a compensatory lysosomal stress response in an attempt to up-regulate HLH-30/TFEB-regulated genes. These effects manifest as an overall decrease in animal fitness.
This study contributes a new dimension to our understanding of the regulation of lysosomal proteostasis via the identification of C. elegans granulins as age and stress-produced peptides that impair overall animal fitness by reducing lysosomal function. C. elegans granulins, similar to the human peptides, localize to the endolysosomal compartment [32]. Granulins are composed of evolutionarily conserved stacked beta hairpins stabilized by disulfide bonds, which are often found in natural protease inhibitors [52]. This highly compact and stable structure is thought to confer resistance to denaturation and protection against proteolytic cleavage in the lysosomal environment [53]. Indeed, a role for granulins in regulating protease maturation has previously been demonstrated in plant cysteine proteases that incorporate a granulin domain C-terminal to the catalytic domain, such as RD21 in A. thaliana [54]. In further support of granulins as regulators of protease activity, homozygous progranulin mutation carriers develop a progressive myoclonic epilepsy syndrome that phenocopies loss of function mutations in another lysosomal protease inhibitor, cystatin B [18, 55]. Recent studies have shown that human full-length progranulin and individual granulin domains may physically interact with CTSD and stimulate the enzymatic activity of the protease [25, 56–58]. However, in the absence of full-length protein, C. elegans granulins promote a distinct phenotype of impaired resistance to ER stress, delayed growth, decreased CTSD and CTSB/L activity and activation of the CLEAR transcriptional program.
Granulins likely play a normal physiological role in regulating protease expression and activity. Given their ability to promote the CLEAR program, granulins may serve as a signal for stress or impaired health that requires regulated checks on protease activity, perhaps to limit inflammation. This would be consistent with the role of progranulin in complement-mediated synaptic pruning by microglia [59]. We speculate that under conditions of progranulin haploinsufficiency, the normal balance between progranulin and granulins becomes skewed towards excessive granulins. In excess, the inhibitory effect of granulins upon protease activity impairs the function of lysosomes; with age, the natural compensatory mechanisms such as the CLEAR program become overwhelmed, resulting in cellular dysfunction. When this occurs in neurons and/or support cells such as microglia, the end result may be neurodegeneration. Because granulins increase with age, it remains possible that accumulation of granulins directly contribute to the proteostatic pressures associated with increasing age. Comprehensive measures of progranulin-to-granulin ratios with age and in progranulin mutation carriers are needed.
The lentiviral delivery of progranulin to degenerating brain regions protects against neurotoxicity and cognitive defects in mouse models of Parkinson’s disease [60] and Alzheimer’s disease [61]. As such, efforts to increase progranulin production in patients are underway [62–65]. However, a more recent study has suggested that progranulin delivery to brain promotes in T-cell infiltration and neuronal and glial degeneration [66]. Progranulin cleavage and granulin levels were not measured in these studies, and may account for differences in the observed results.
Progranulin is a highly conserved protein [27, 29, 30]. The number of granulin domains has increased through phylogeny from one in Dictyostelium discoideum and plants, three in nematodes to seven-and-a-half in humans [29, 54]. It is intriguing to speculate that this expansion in cleavage fragments could lead to regulation of additional proteases. In support of this, we find that the amino acid residues situated outside of the well-conserved granulin sequence consensus contribute to distinct charge and hydrophobic profiles for each granulin domain. These unique characteristics may be important for driving specific protein-protein interactions and thus different roles in the cellular environment. Indeed, the distinct effects of granulin 2 and 3 on protein homeostasis, lysosomal function and TDP-43 toxicity [39], as compared to granulin 1, may suggest functional differences between granulins.
Our results establish age-regulated granulins as modulators of lysosomal function, and suggest that a toxic gain of granulin function, rather than or in addition to simply loss of full-length progranulin, may contribute to FTLD disease pathogenesis. This could explain why progranulin loss-of-function mutations are transmitted in an autosomal dominant fashion. The presence of granulins only in the haploinsufficiency state could explain why TDP-43 pathology is not seen in the null state [18]. Several lysosomal proteases that cleave progranulin have recently been identified [32–34], although how those proteases decide when and where to cleave progranulin remains unknown. This study prompts several important follow up questions regarding the rate and order in which granulins are liberated from progranulin, how pH changes impact the predicted association of granulins with lysosomal proteases and whether increased granulin impact other neurodegenerative disorders such as Alzheimer’s disease. The current study also has implications for therapeutic progranulin repletion efforts, as care should be taken to determine whether replacement progranulin is processed into granulins. Finally, our findings suggest that in addition to progranulin repletion, prevention of progranulin cleavage into granulins could represent a rational therapeutic target in neurodegeneration.
C. elegans strains were cultured at 20 °C according to standard procedures [67]. Some strains were provided by the Mitani Laboratory (National Bioresource Project, Japan) at the Tokyo Women’s Medical University and the Caenorhabditis Genetics Center (CGC) at the University of Minnesota. Strain descriptions are at www.wormbase.org. The N2E control strain was used as the wild-type strain. The pgrn-1(tm985) strain has a 347 bp deletion in the pgrn-1 gene resulting in a null allele [45]. The following C. elegans strains were used in this study:
CF3050 pgrn-1(tm985) I
AWK33 pgrn-1(tm985) I; rocIs1[Ppgrn-1+SignalSequence::granulin1::FLAG::polycistronic mCherry + Punc-122::GFP]
AWK43 pgrn-1(tm985) I; rocEx14[Ppgrn-1+SignalSequence::granulin2::FLAG::polycistronic mCherry + Pmyo-2::GFP]
AWK107 pgrn-1(tm985) I; rocIs5[Ppgrn-1+SignalSequence::granulin3::FLAG::polycistronic mCherry + Pmyo-2::GFP]
AWK308 N2E; rocIs1[Ppgrn-1+SignalSequence::granulin1::FLAG::polycistronic mCherry + Punc-122::GFP]
AWK309 N2E; rocEx14[Ppgrn-1+SignalSequence::granulin2::FLAG::polycistronic mCherry + Pmyo-2::GFP]
AWK310 N2E; rocIs5[Ppgrn-1+SignalSequence::granulin3::FLAG::polycistronic mCherry + Pmyo-2::GFP]
AWK459 pgrn-1(tm985) I; muIs216[Paex-3::huMAPT 4R1N +Pmyo-3::RFP]
CF3588 pgrn-1(tm985) I; muIs206[Pegl-3::TDP-43::GFP]
AWK524 pgrn-1(tm985) I; muIs189[Ppgrn-1::pgrn-1::polycishronic mCherry +Podr-1::CFP]
AWK466 pgrn-1(tm985) I; muIs189[Ppgrn-1::pgrn-1::polycishronic mCherry +Podr-1::CFP]; rocEx14[Ppgrn-1+SignalSequence::granulin2::FLAG::polycistronic mCherry + Pmyo-2::GFP]
CF3778 pgrn-1(tm985) I; muIs213[Ppgrn-1::pgrn-1::RFP]
AWK181 pgrn-1(tm985) I; unc-119(ed3)III; pwIs503[vha6p::mans::GFP + Cb unc-119(+)]; muIs213[Ppgrn-1::pgrn-1::RFP]
AWK360 pgrn-1 (tm985) I; unc-119(ed3) III; pwIs50[Plmp-1::lmp-1::GFP + Cbr-unc-119(+)]; muIs213[Ppgrn-1::pgrn-1::RFP]
AWK395 pgrn-1 (tm985) I; unc-119(ed3) III; cdIs54[pcc1::MANS::GFP + unc-119(+) + myo-2::GFP]; muIs213[Ppgrn-1::pgrn-1::RFP]
AWK374 pgrn-1 (tm985) I; bIs34[rme-8::GFP + rol-6(su1006)]; muIs213[Ppgrn-1::pgrn-1::RFP]
MAH235 sqIs19[Phlh-30::hlh-30::gfp + rol-6(su1006)]
AWK403 pgrn-1(tm985) I; sqIs19[Phlh-30::hlh-30::gfp + rol-6(su1006)]
AWK404 pgrn-1(tm985) I; sqIs19[Phlh-30::hlh-30::gfp + rol-6(su1006)]; rocIs1[Ppgrn-1+SS::granulin1::FLAG::polycistronic mCherry]
AWK405 pgrn-1(tm985) I; sqIs19[Phlh-30::hlh-30::gfp + rol-6(su1006)]; rocEx14[Ppgrn-1+SS::granulin2::FLAG::polycistronic mCherry + Pmyo-2::GFP]
AWK406 pgrn-1(tm985) I; sqIs19[Phlh-30::hlh-30::gfp + rol-6(su1006)]; rocIs5[Ppgrn-1+SS::granulin3::FLAG::polycistronic mCherry + Pmyo-2::GFP]
AWK467 N2E; sqIs19[Phlh-30::hlh-30::gfp + rol-6(su1006)]; rocIs1[Ppgrn-1+SS::granulin1::FLAG::polycistronic mCherry]
AWK469 N2E; sqIs19[Phlh-30::hlh-30::gfp + rol-6(su1006)]; rocEx14[Ppgrn-1+SS::granulin2::FLAG::polycistronic mCherry + Pmyo-2::GFP]
AWK471 N2E; sqIs19[Phlh-30::hlh-30::gfp + rol-6(su1006)]; rocIs5[Ppgrn-1+SS::granulin3::FLAG::polycistronic mCherry + Pmyo-2::GFP]
AWK546 pgrn-1(tm985) I; sqIs19[Phlh-30::hlh-30::gfp + rol-6(su1006)]; muIs216[Paex-3::huMAPT 4R1N +Pmyo-3::RFP]
AWK547 pgrn-1(tm985) I; sqIs19[Phlh-30::hlh-30::gfp + rol-6(su1006)]; muIs206[Pegl-3::TDP-43::GFP]
JIN1375 hlh-30(tm1978) IV
AWK514 pgrn-1 (tm985) I; hlh-30(tm1978) IV
AWK516 pgrn-1 (tm985) I; hlh-30(tm1978) IV; rocIs1[Ppgrn-1+SS::granulin1::FLAG::polycis mCherry]
AWK518 pgrn-1 (tm985) I; hlh-30(tm1978) IV; rocEx14 [Ppgrn-1+SS::granulin2::FLAG::polycistronic mCherry + Pmyo-2::GFP]
AWK519 pgrn-1 (tm985) I; hlh-30(tm1978) IV; rocIs5 [Ppgrn-1+SS::granulin3::FLAG::polycistronic mCherry + Pmyo-2::GFP]
AWK521 pgrn-1 (tm985) I; hlh-30(tm1978) IV; muIs189[Ppgrn-1::pgrn-1::polycistronic mCherry +Podr-1::CFP]
AWK296 N2E; Ex[Pced-1::asp-3::mrfp + pRF4(rol-6)]; unc-119(ed3) III; pwIs50[Plmp-1::lmp-1::GFP + Cbr-unc-119(+)]
AWK333 pgrn-1(tm985) I; Ex[Pced-1::asp-3::mrfp + pRF4(rol-6)]; unc-119(ed3) III; pwIs50[Plmp-1::lmp-1::GFP + Cbr-unc-119(+)]
AWK247 pgrn-1(tm985) I; pwls50[lmp-1::GFP + Cbr-unc-119(+)];rocEx14 [Ppgrn-1+SS::granulin2::FLAG::polycis tronic mCherry + Pmyo-2::GFP]
AWK334 pgrn-1(tm985) I; Ex[Pced-1::asp-3::mrfp + pRF4(rol-6)]; unc-119(ed3) III; pwIs50[Plmp-1::lmp-1::GFP + Cbr-unc-119(+)]; rocIs5[Ppgrn-1+SS::granulin3::FLAG::polycis tronic mCherry + Pmyo-2::GFP]
AWK177 asp-3(tm4450) X
VM487 nmr-1(ak4)II
To generate strains expressing individual granulins, each granulin was amplified separately from wild-type C. elegans progranulin cDNA as previously described [39].
ER stress assays were performed as previously described [40].
L4 stage animals were allowed to lay eggs overnight. Fifty synchronized eggs were transferred to seeded plates. After three days, the fraction of animals that developed to the L4 stage was quantified.
L4 animals were staged, grown at 20 °C overnight and imaged the following day as day 1 adults. Animals were mounted on a 2% agarose pad with 25 mM sodium azide (Spectrum Chemical, #SO110) and imaged using a Zeiss AxioImager microscope at 10 x. Body size was measured in ImageJ software using the skeletonize function.
Short-term associative learning assays were performed as previously described [43, 44].
Sixty L4 stage animals were allowed to lay eggs overnight (~sixteen hours). Adult worms and hatched larvae were washed off the plates with M9 buffer. Eggs were collected with a cell scraper and transferred to a newly seeded plate by chunking. These eggs were allowed to develop to early L4 stage and 200 μl of 20 mM FUDR (Fisher Scientific, #AC227601000) was added to prevent development of progeny and overgrowth of plates. At each time point, animals were collected from plates with ice cold M9 and washed once to remove food. The worm pellet was resuspended 1:1 in freshly made ice cold RIPA buffer (50 mM Tris pH 7.4, 150 mM NaCl, 5 mM EDTA, 0.5% SDS, 0.5% SDO, 1% NP-40, 1 mM PMSF, cOmplete protease inhibitor (Roche, #04693124001) and PhosSTOP phosphatase inhibitor (Roche, #04906837001), 0.3 mM Pefabloc (Roche, #11429868001)). Worms were transferred to Eppendorf tubes and sonicated for 4 cycles of 1 minute on and 2 minutes off (BioRuptor, Diagenode). Lysates were centrifuge for 5 minutes at 13,000 rpm at 4 °C. Supernatant was transferred to a fresh Eppendorf tube and samples were boiled at 95 °C (with 4x LDS, 10% reducing agent) for 5 minutes and analyzed by SDS PAGE. 10–50 μg total protein was resolved on 4–12% gradient SDS-PAGE gels and transferred to PVDF.
Commercial antibodies used for Western blotting were the following:
Anti-HSP-4/BiP (Novus Biologicals, #NBP1-06274, 1:1000 dilution)
Anti-RFP (GenScript, #A00682, 1:1000 dilution)
Anti-FLAG (Sigma, #F3165, 1:1000 dilution)
Anti-LMP-1(Developmental Studies Hybridoma Bank, #LMP1, 1:100 dilution)
Anti-HSP-70/HSC-70 (Santa Cruz Biotechnology Inc., #sc-33575, 1:1000 dilution)
Anti-calnexin (Novus Biologicals, #NBP1-97476, 1:1000 dilution)
Anti-CPL-1 (Abcam, #ab58991, 1:500 dilution)
Anti-actin (EMD Millipore, #MAB1501R, 1:5000 dilution)
Goat anti-mouse (LI-COR IRDye 800CW, #925–32210, 1:10,000 dilution)
Goat anti-rabbit (LI-COR IRDye 800CW, #925–32211, 1:10,000 dilution)
Donkey anti-goat (LI-COR IRDye 800CW, #925–32214, 1:10,000 dilution)
Donkey anti-mouse (LI-COR IRDye 680RD, #925–68072, 1:10,000 dilution)
Antibodies made in-house and used for Western blotting were the following:
Anti-granulin 1(RB2481, Biomatik, epitope HQCDAETEC(acm)SDDET, 1:1000 dilution)
Anti-granulin 3 (RB2487, Biomatik, epitope CTVLMVESARSTLKL, 1:1000 dilution)
Anti-ASP-3 (Fred Hutchinson, epitope CTGPTDVIKKIQHKIG, 1:1000 dilution)
Imaging and quantification were performed on the LI-COR Odyssey Infrared System. Three independent blots were performed.
Animals were mounted on microscope slides with 2% agarose pads containing 30 mM levamisole hydrochloride (Fisher Scientific, #AC187870100) and imaged using a Zeiss LSM 700 laser-scanning confocal microscope using 488 nm and 561 nm lasers and 63x and 100x objectives. L1 animals were imaged 1–2 h after hatching. Z-stacks were taken every 0.7 μm. Image processing was carried out using ImageJ software. A maximum intensity projection of the z-stack for each animal was created. Images at 488 nm and 561 nm were overlaid and analyzed for co-localization.
Thirty L4 stage animals were picked to 60 x 10 cm plates per strain. Plates were confluent with mixed stage animals after four days growth at 20 °C. Progranulin cleavage was observed after starving animals for an additional seventy-two hours at 20 °C. A lysosomal fraction was isolated from a light mitochondrial-lysosomal fraction as previously described [68] with the following modifications. Animals were collected in 0.25 M sucrose (pH 7.2) and washed twice with 0.25 M sucrose. Lysosomes and mitochondria were separated using a discontinuous Nycodenz (Progen Biotechnik, Germany, #1002424) density gradient. Lysosomes were collected from the 19.8% / sucrose interface and the 26.3 / 19.8% interface and pooled. Lysosomes were diluted five times with 0.25 M sucrose, and pelleted at 37,000 × g for 15 minutes. Cytosolic, ER and lysosomal fractions were confirmed by immunoblotting for specific subcellular fraction markers (LAMP-1, HSC-70, calnexin).
Protease activity was measured using commercially available kits (BioVision Cathepsin D Activity Fluorometric Assay Kit, #K143-100 and BioVision Cathepsin L Activity Fluorometric Assay Kit, #K142-100). Animals were staged as for immunoblotting, but without the addition of 20 mM FUDR. At day 1 of adulthood, worms were collected from plates with ice cold M9 and washed twice to remove food. Worm pellets were resuspended in 1% NP-40 buffer (Fisher Scientific) without protease inhibitors and frozen at -80 °C overnight. Pellets were thawed and sonicated for 4 cycles of 1 min on and 2 min off (BioRuptor, Diagenode). Lysates were centrifuged for 5 minutes at 13,000 rpm at 4 °C and supernatant was transferred to a fresh tube. 0.25 μg total protein per sample was used per assay and samples from one strain were run in triplicate. Fluorescence measurements were taken every minute at 25 °C (Infinite M200, Tecan). As controls, 250 nM Pepstatin A (for pan-aspartyl protease inhibition in CTSD assay, BioVision) or 10 μM CA-074 (for Cathepsin B inhibition, EMD Millipore, #205530) and 10 μM CTSLiII (for Cathepsin L inhibition, EMD Millipore, #219426) were added to the lysate and pre-incubated for 10 minutes on the bench at room temperature. Linear regression was performed on at least 30 minutes of data to calculate the rate of enzyme activity.
Sequences for C. elegans (Q9U362), Homo sapiens (P28799) and Mus musculus (P28798) PGRN were extracted from Uniprot (The Uniprot Consortium, 2019), while Danio rerio PGRNb (AAH96854.1) sequence was obtained from National Center for Biotechnology Information (NCBI) Protein database (https://www.ncbi.nlm.nih.gov). Amino acid multiple sequence alignment was performed using the MAFFT online service (version 7, https://mafft.cbrc.jp/alignment/server/) [69]. The EMBOSS Needle server was used for pairwise sequence alignment between C. elegans granulin 1, granulin 2 and granulin 3 and individual granulin domains from H. sapiens, M. musculus and D. rerio (https://www.ebi.ac.uk/Tools/psa/emboss_needle/) [70]. Identification of granulin domains from the full-length sequences was based on sequence similarity to H. sapiens granulin A using the Basic Local Alignment Search Tool protein (BLASTp) server (https://blast.ncbi.nlm.nih.gov/Blast.cgi). Granulin A (PDB ID: 2JEY.A) was used as a reference for homology modeling of all granulin domains [71] using the Prime software. Electrostatic analysis ranging from pH 4 to 8 was performed on the in silico models with the software propKa 3.1 [72]. Kyte & Doolitle (K&D) hydrophobicity scales were obtained from the ExPASy Bioinformatics Resource Portal (https://web.expasy.org) for PGRN sequence of all species here studied. For the K&D per-residue score, a window size of 5 was used, i.e. the final score for a given residue i is the sum of the scale values for i and i-2, i-1, i+1 and i+2.
Total RNA was isolated from wild-type (N2E), pgrn-1(-), pgrn-1(-); granulin 1(+), pgrn-1(-); granulin 2(+) and pgrn-1(-); granulin 3(+) expressing animals synchronized at day 1 of adulthood. Animals were collected from plates with ice cold M9 and washed three times to remove OP50 food. After harvesting, the animals were resuspended in QIAzol (Qiagen #79306) and flash frozen in liquid nitrogen. RNA was extracted and purified using a Qiagen miRNeasy kit (Qiagen #217004). Samples were extracted in quadruplicate (four biological replicates for each strain), for a total of 20 samples. Total RNA was quantified using the RiboGreen assay (ThermoFisher, #R11490) and RNA quality was checked using an Agilent TapeStation 4200 (Agilent). RNA Integrity Numbers (eRINs) were >8 in all the samples. Libraries for RNA-seq were prepared using the Illumina TruSeq library preparation protocol (Illumina Inc), multiplexed into a single pool and sequenced using an Illumina HiSeq 4000 sequencer across 4 PE 2 x 75 lanes on a single flowcell. After demultiplexing, we obtained between 13 and 32 million reads per sample, each one 75 paired end bases long. Quality control was performed on base qualities and nucleotide composition of sequences. Alignment to the C. elegans genome (ce11) was performed using the STAR spliced read aligner [73] with default parameters. Additional QC was performed after the alignment to examine the following: level of mismatch rate, mapping rate to the whole genome, repeats, chromosomes, and key transcriptomic regions (exons, introns, UTRs, genes). Between 92 and 93% of the reads mapped uniquely to the worm genome. Total counts of read fragments aligned to candidate gene regions within the C. elegans reference gene annotation were derived using HTS-seq program and used as a basis for the quantification of gene expression. Only uniquely mapped reads were used for subsequent analyses. Following alignment and read quantification, we performed quality control using a variety of indices, including sample clustering, consistency of replicates, and average gene coverage. One sample for pgrn-1(-); granulin 1(+) was excluded from analysis as a quality control outlier. Differential expression analysis was performed using two parallel approaches, the EdgeR Bioconductor package [74], and voom [75]. Differentially expressed genes (DEGs) were selected based on False Discovery Rate (FDR, Benjamini-Hochberg adjusted p-values) estimated at ≤ 5%. There was a large overlap between DEGs identified by edgeR and voom (edgeR: 89.0% common DEGs with voom (6307/7084), voom: 93.9% common DEGs with edgeR (6307/6714)). Clustering and overlap analyses were performed using the Bioconductor packages within the statistical environment R (www.rproject.org/). Gene Ontology annotation was performed using DAVID (david.abcc.ncifcrf.gov/) and GOrilla [76, 77].
The promoter regions of all differentially regulated transcripts were analyzed for the presence of the C. elegans TFEB/HLH-30 binding site E-box sequence 5’-CACGTG-3’. Enrichment of TFEB binding sites was tested by comparison to the expected distribution based on 10,000 random permutations. A permutation test was used to calculate p-values.
Forty L4 animals were picked, grown at 20 °C overnight and imaged the following day as day 1 adults. The nuclear localization of HLH-30::GFP was imaged using a Zeiss AxioImager microscope at 10x. Animals were imaged within 5 minutes of mounting on a 2% agarose pad with 25mM sodium azide (Spectrum Chemical, #SO110). Data from three independent experiments were pooled.
|
10.1371/journal.ppat.1003251 | Caveolin-1 Protects B6129 Mice against Helicobacter pylori Gastritis | Caveolin-1 (Cav1) is a scaffold protein and pathogen receptor in the mucosa of the gastrointestinal tract. Chronic infection of gastric epithelial cells by Helicobacter pylori (H. pylori) is a major risk factor for human gastric cancer (GC) where Cav1 is frequently down-regulated. However, the function of Cav1 in H. pylori infection and pathogenesis of GC remained unknown. We show here that Cav1-deficient mice, infected for 11 months with the CagA-delivery deficient H. pylori strain SS1, developed more severe gastritis and tissue damage, including loss of parietal cells and foveolar hyperplasia, and displayed lower colonisation of the gastric mucosa than wild-type B6129 littermates. Cav1-null mice showed enhanced infiltration of macrophages and B-cells and secretion of chemokines (RANTES) but had reduced levels of CD25+ regulatory T-cells. Cav1-deficient human GC cells (AGS), infected with the CagA-delivery proficient H. pylori strain G27, were more sensitive to CagA-related cytoskeletal stress morphologies (“humming bird”) compared to AGS cells stably transfected with Cav1 (AGS/Cav1). Infection of AGS/Cav1 cells triggered the recruitment of p120 RhoGTPase-activating protein/deleted in liver cancer-1 (p120RhoGAP/DLC1) to Cav1 and counteracted CagA-induced cytoskeletal rearrangements. In human GC cell lines (MKN45, N87) and mouse stomach tissue, H. pylori down-regulated endogenous expression of Cav1 independently of CagA. Mechanistically, H. pylori activated sterol-responsive element-binding protein-1 (SREBP1) to repress transcription of the human Cav1 gene from sterol-responsive elements (SREs) in the proximal Cav1 promoter. These data suggested a protective role of Cav1 against H. pylori-induced inflammation and tissue damage. We propose that H. pylori exploits down-regulation of Cav1 to subvert the host's immune response and to promote signalling of its virulence factors in host cells.
| Infection with the bacterium Helicobacter pylori (H. pylori) mainly affects children in the developing countries who are at risk to progress to gastric cancer (GC) as adults after many years of persistent infection, especially with strains which are positive for the oncogenic virulence factor CagA. Eradication of H. pylori by antibiotics is a treatment of choice but may also alter the susceptibility to allergies and other tumor types. Thus, novel diagnostic or prognostic markers are needed which detect early molecular changes in the stomach mucosa during the transition of chronic inflammation to cancer. In our study, we found that the tumor suppressor caveolin-1 (Cav1) is reduced upon infection with H. pylori, and CagA was sufficient but not necessary for this down-regulation. Loss of Cav1 was caused by H. pylori-dependent activation of sterol-responsive element-binding protein-1 (SREBP1), and this event abolished the interaction of Cav1 with p120 RhoGTPase-activating protein/deleted in liver cancer-1 (p120RhoGAP/DLC1), a second bona fide tumor suppressor in gastric tissue. Conclusively, Cav1 and DLC1 may constitute novel molecular markers in the H. pylori-infected gastric mucosa before neoplastic transformation of the epithelium.
| Helicobacter pylori (H. pylori) is a Gram-negative bacterium which colonizes stomachs of approx. 50% of the world's population and increases the risk for development of chronic gastritis, peptic ulcer disease, gastric mucosa-associated lymphoid tissue (MALT) lymphoma, mucosal atrophy and gastric cancer (GC) [1], [2]. Based on this etiology, H. pylori has been classified as a class I carcinogen by the World Health Organisation (WHO) in 1994 [3].
The two major H. pylori toxins [4], CagA and VacA, are internalized into gastric epithelial cells by injection via the bacterial type IV secretion system (CagA) [5] or by direct insertion into lipid rafts (VacA) [6], [7]. Lipid rafts are cholesterol and sphingolipid-rich microdomains of the plasma membrane [8], [9] which are exploited by many pathogens, including viruses, parasites and bacteria, to facilitate uptake of whole organisms and/or internalisation of toxins into host cells [10], [11], [12]. For example, Neisseria spec. uses lipid rafts and Rho-mediated signaling of the actin cytoskeleton to gain access to the cytosol [13]. Pseudomonas aeruginosa exploits lipid raft-associated toll-like receptor 2 for infection of lung epithelial cells [14].
Caveolin-1 (Cav1) is the 21–24 kDa major and essential structural protein of caveolae, a specialized form of lipid raft microdomains. Caveolae are 50–100 nm flask/tube-shaped invaginations of the plasma membrane abundant in macrophages, endothelial and smooth muscle cells, type I pneumocytes and adipocytes, where they participate in cellular transport processes including endocytosis, cholesterol efflux and membrane traffic [15], [16]. In this context, Cav1 can also act as an inhibitor of clathrin-independent endocytosis and block pathogen/toxin uptake [17], [18]. Through binding to its scaffolding domain, Cav1 directly inhibits a plethora of receptors and enzymes including tyrosine kinases of the Src and Ras family, G-proteins and nitric oxide synthases [15]. In addition to a role in membrane traffic, Cav1 thus constitutes a control platform for regulation of cell proliferation and survival [19]. Cav1 also exerts an important function in cell motility and migration and, within epithelial, stromal and endothelial tissues, by enforcing cell-cell contacts, cell-matrix adhesion and immune responses [20], [21], [22], [23].
Cav1 directly binds cholesterol, and transcription of Cav1 is negatively regulated by the transcription factor sterol-responsive element-binding protein-1 (SREBP1) [24]. SREBP1 is bound to the endoplasmic reticulum (ER) as an inactive 125 kDa precursor and is activated under conditions of cholesterol deficiency by proteolytic cleavage in the Golgi apparatus. This cleavage is followed by translocation of the active 68 kDa SREBP1 into the nucleus where it binds to sterol-responsive elements (SREs) of target genes, including Cav1, involved in synthesis of cholesterol and fatty acids [25]. H. pylori has been shown to metabolize cholesterol from the host cell membrane, and host cholesterol alters the oncogenic properties of CagA [26], [27].
We therefore hypothesized that the cholesterol-binding proteins SREBP1 and Cav1 are targets of H. pylori infection and/or effector functions. Specifically, we asked whether (i) H. pylori exploits Cav1 to facilitate injection and down-stream signalling of CagA in gastric epithelial cells or (ii) Cav1 acts as a protective “barrier-enforcing” protein that counteracts disease evoked by H. pylori. To test this, the phenotypes which result from H. pylori infection were studied in Cav1-deficient mice and in human GC cell lines. Our data showed that Cav1 protected B6129 mice against H. pylori-related gastritis and tissue damage in vivo independently of CagA. H. pylori also activated SREBP1 and down-regulated expression of murine and human Cav1 independently of CagA. In addition, Cav1 counteracted CagA-dependent cytoskeletal rearrangements in vitro by recruitment of the tumor suppressor deleted in liver cancer-1 (DLC1).
Animal studies were conducted in agreement with the ethical guidelines of the Technische Universität München (German Animal Welfare Act, Deutsches Tierschutzgesetz) and had been approved (#55.2-1-54-2531-74-08) by the government of Bavaria (Regierung von Obb., Munich, Germany).
Homozygous Cav1 knockout (Cav1-KO) (strain Cav1tm1Mls/J; stock number 004585) and matched control wild-type (WT) (strain B6129SF2/J; stock number 101045) mice (8 weeks) were obtained from the Jackson Laboratory (Bar Harbor, Maine) and maintained on a mixed background in a pathogen-free mouse facility [28], [29]. Experimental gastric ulceration was performed with indomethacin as published before [30]. Infection of mice with the mouse-adapted CagA/VacA-delivery deficient H. pylori strain SS1 was performed by oral gavage as described [31]. The average time mice from different genetic backgrounds (C57BL/6, B6129, BALB/c) take to progress to chronic gastritis and beyond (gastric atrophy, hyperplasia, dysplasia) [32] ranges between 10 and 15 month upon infection with the standardized reference strain SS1 [28], [33], [34], [35]. We therefore decided to perform our analysis within this time frame.
Chemicals were from Merck (Darmstadt, Germany) or Sigma (Taufkirchen, Germany). Polyclonal antisera were SREBP1 (#PA1-46142, Thermofisher Scientific, Waltham, MA), Cav1 (N-20, sc-894), SREBP1 (C-20, sc-366), CagA (b-300, sc-25766), FAK (A-17, sc-557), phospho-FAK (Tyr-397, sc-11765), Hsp90 alpha/beta (H-114, sc-7947), Lamin A/C (H-110, sc-20681, all from Santa Cruz Biotech., CA), general and phospho ERK1/2 (p44/p42), p38, JNK (all from Cell Signaling, Danvers, MA) and Ki-67 (SP6, DCS GmbH, Hamburg, Germany). Mouse monoclonal antibodies were Cav1 (#610406) and phospho-Cav1 (pY14, #611338) (both from BD/Transduction Lab., San Jose, CA), DLC-1 (C-12, sc-271915) and beta-Actin (AC-15, sc-69879) (both from Santa Cruz Biotech.). The macrophage-specific rat anti-mouse F4/80 antibody (#MF48000) was obtained from Invitrogen (Life Technologies, Darmstadt, Germany). Chicken anti-H. pylori polyclonal Ab was used as described [36]. Serum cytokines were measured by ELISA (R&D Systems, Minneapolis, MN) according to the manufacturer's instructions. Pull-down assays for the small GTPases Rho/Rac/Cdc42 were purchased from Biocat (Heidelberg, Germany).
Human embryonic kidney (HEK293), Madin-Darby canine kidney (MDCK), parental human GC cell lines (AGS, MKN45, N87) (all from the American Type Culture Collection, Rockville, MD) and stably transfected clones generated thereof were maintained as described previously [37]. Infection of cells with the cell-adapted CagA-delivery proficient H. pylori strain G27 was performed as before [36].
The expression plasmid pEGFP-CagA was mentioned elsewhere [38]. The ∼800 bp fragment of the proximal human Cav1 promoter (AF019742, position 69 to 859) [24] was amplified by PCR from the genomic DNA of human normal liver and cloned into the KpnI/HindIII sites of pGL3-luc luciferase reporter plasmid (Promega GmbH, Mannheim, Germany). Isoform 4 of the human DLC1 mRNA [39] (DLC1v4, NM_001164271.1) was amplified from human hepatoma HepG2 cells and inserted in the BamHI/NotI sites of the expression vector pTarget (pT, Promega GmbH). Transient transfection and luciferase assays were performed as before [37].
H. pylori SS1 and G27 bacteria were recovered from −80°C glycerol stocks and grown on Wilkins-Chalgren (WC) blood agar plates under microaerobic conditions (10% CO2, 5% O2, 85% N2; 37°C) for 2–3 days. The mouse-adapted H. pylori SS1 was harvested from agar plates for in vivo infections as published previously [31]. The SS1 strain was PCR-positive for the cagA gene and mRNA but did not inject functional CagA protein [40] as evident by the absence of the “humming bird” phenotype in infected AGS cells (data not shown). The cell-adapted H. pylori bacteria CagA-delivery proficient G27 wt and the CagA-deletion mutant G27 Delta cagA were harvested from agar plates and subsequently grown in continuous co-culture with MDCK cells as described [36].
Whole stomachs were excised from mice, and colony formation was determined essentially as described [31]. An antral strip of the stomach was weighed, placed into 5 ml of Brucella broth and vortexed for 10 min. Dilutions of 1∶10, 1∶100 and 1∶1000 were prepared, and 100 µl of each dilution was plated onto H. pylori-selective WC blood agar plates. The number of bacterial colonies was determined after 5 days and normalised to the weight of the corresponding stomach pieces.
The remaining stomach was washed with sterile water. An antral strip was cut, frozen in liquid nitrogen and stored at −80°C until RNA extraction. The rest of the stomach was placed into 3 ml of 4% (w/v) paraformaldehyde (PFA) in phosphate buffered saline (PBS) and incubated for 24 h at 4°C. Then, the stomach was cut along the greater and small curvature into two halves, followed by dehydration and embedding into paraffin for histological analysis.
Cells were infected with the H. pylori G27 strain for 2 to 24 h at a multiplicity of infection (MOI) of 500∶1. Thereafter, cells were washed three times with PBS to remove residual bacteria and were additionally incubated for 2 h at 37°C in a humidified atmosphere in DMEM/F12 (10% FCS, 10% Brucella broth) supplemented with gentamycin (200 µg/ml), penicillin/streptomycin (100 µg/ml) and chloramphenicol (100 µg/ml). Absence of extracellular bacteria was confirmed under the microscope, and the cells were subsequently lysed for detection of intracellular CagA by Western blot (WB).
Detection of immunoprecipitated proteins by SDS-PAGE and WB was performed as before [41]. Matrix-assisted Laser Desorption/Ionization mass spectrometry (MALDI-MS) was described in detail in [29].
The staining was performed in triple-color mode visualizing 4,6-diamidino-2-phenylindole (DAPI), Alexa-488 and -594 using a digital camera-connected (Axiovision, release 4.4) fluorescence microscope (Axiovert 200M, Carl Zeiss MicroImaging GmbH, Hallbergmoos, Germany). Confocal microscopy (Axiovert 40, Zeiss) and 3D-reconstruction of H. pylori-infected cells with LSM510 (Zeiss) and Volocity (Improvision, Tübingen, Germany) was done as before [37].
Chronic active gastritis was defined by the simultaneous presence of both neutrophilic polymorphnuclear (PMN) and mononuclear cells (lymphocytes and plasma cells) within the gastric mucosa. Active (PMN) and chronic (mononuclear) infiltrate was assessed as follows: Paraffin-embedded gastric tissue was cut into 3 µm sections using a semi-automatic microtome (Leica Microsystems GmbH, Wetzlar, Germany). The sections were then stained using Hematoxylin & Eosin (H&E) solutions. The histopathological analysis was carried out by three pathologists (CR, SR, TK) blinded to the study setup. Morphological alterations in the gastric mucosa were classified according to the updated Sydney system [32], [42]. The grade of gastritis was scored based on the density of intramucosal inflammatory infiltrates from mononuclear and PMN cells as published before [43]: none (0), mild (1+), moderate (2+) and severe (3+). In addition, hyperplastic or regenerative epithelial alterations, loss of parietal cells and the frequency of lymphoid follicles or lymphoid aggregates were noted. The intensity of H. pylori colonization in the gastric mucosa was recorded as mild (few and single bacteria in a random distribution), moderate (single and clustered bacteria in a discontinuous distribution) and severe (dense bacterial clusters covering the gastric mucosa in continuous layers). Multiple scores of different regions of the stomach were determined. Immunohistochemistry (IHC) was performed on paraffin sections as described before [44].
ChIP (Kit from Upstate, Millipore GmbH, Schwalbach, Germany) and all other methods were performed as described previously [45]. Oligonucleotides are listed in Table S1.
Viability of adherent cells was measured by 1-(4,5-dimethylthiazol-2-yl)3,5-diphenyl-formazan (MTT) assay (Roche Diagnostics GmbH, Mannheim, Germany) as recommended by the manufacturer. To determine cell adhesion, 1×104 cells were seeded into 6 cm cell culture dishes for 1 to 6 h followed by repetitive washing with PBS. The remaining adherent cells were fixed with 4% (w/v) PFA in PBS, stained with crystal violet and subsequently counted using ImageJ (NIH, Bethesda, MD). Wound healing assays were performed essentially as described in [46]. Briefly, cells were grown to confluence in 6 cm dishes, and a 5 mm scratch was introduced into the monolayer using an inverted blue tip followed by incubation of the cell culture plates for additional 24, 48 and 72 h. Wound closure was monitored upon fixation and staining of cells with crystal violet using bright field microscopy (Axiovert 200M, Carl Zeiss MicroImaging GmbH).
Results are means ± S.E. from at least 5 animals per genotype or 3 independent experiments from different cell passages. The software GraphPad Prism (version 4.0, La Jolla, CA) was used to analyze the data. P-values (*p<0.05) were calculated using Student's t and Fisher Exact tests.
Human: Cav1: NM_001753.4, Q03135; b2M: NM_004048.2, P61769; IL8: NM_000584.3, P10145; DLC1 v1: NM_182643.2, Q96QB1; DLC1 v4: NM_001164271.1, Q96QB1; ACS: NM_018677.3, Q9NR19; HMGCoAS: NM_001098272.2, Q01581; HMGCoAR: NM_000859.2, P04035; LDLR: NM_000527.4, P01130; beta-Actin: P60709; Lamin A: P02545; Lamin C: P02545; Hsp90 alpha: P07900; Hsp90 beta: P08238; ERK1 (p44): P27361; ERK2 (p42): P28482; FAK: Q05397; JNK1: P45983; JNK2: P45984; p38: Q16539; Src: P12931; SREBP1: P36956; Ki-67: P46013; Mouse: Cav1: NM_007616.4, P49817; b2M: NM_009735.3, Q91XJ8; TNFalpha: NM_013693.2, P06804; IFNgamma: NM_008337.3, P01580; IL1beta: NM_008361.3, P10749; IL6: NM_031168.1, P08505; CD4: NM_013488.2, P06332; CD19: NM_009844.2, P25918; CD25: NM_000417.2, P01589; CD86: NM_019388.3, P42082; CCL5: NM_013653.3, P30882; CXCL1: NM_008176.3, P12850; PPARg2: NM_015869.4, P37231; TFF2: NM_009363.3, Q9QX97; Dog: b2M: NC_006612, XP_850148; H. pylori: CagA: YP_002266135.1, B5Z6S0; UreB: YP_626814.1, Q1CV82.
To assess the histological changes induced in gastric tissue upon H. pylori infection, B6129 WT and Cav1-KO mice were infected with the mouse-adapted and CagA-delivery deficient H. pylori strain SS1. The mice were euthanized 11 months later, and H. pylori was isolated from resected stomach tissue [31]. Cav1-KO mice showed less bacterial colonisation of the gastric mucosa than WT mice (7.3±2.4 WT versus 1.6±0.5 KO ×103 CFU/mg stomach tissue; *p = 0.0141; n = 15 per genotype) (Fig. 1A). Histopathological analysis revealed that both WT and Cav1-KO mice developed active chronic gastritis accompanied by infiltration of mononuclear and polymorphnuclear (PMN) cells into the gastric mucosa (Fig. 1B). In contrast, uninfected WT and Cav1-KO mice had no intramucosal inflammation (data not shown). Instead, the gastritis was markedly enhanced in H. pylori-infected Cav1-KO mice compared with infected WT mice (Fig. 1C). In Cav1-KO mice, the average score of gastritis (0.7±0.2 WT versus 1.7±0.1 KO; *p = 0.0002, n = 15 per genotype) was more severe (Table 1) than in WT mice, and the stomach mucosa exhibited intramucosal B-cell follicles, foveolar hyperplasia and loss of parietal cells. This data indicate that Cav1-deficiency is associated with an increased inflammatory response in the gastric mucosa and a less efficient colonisation by H. pylori.
To assess the identity of immune cells which contribute to H. pylori-related inflammation in Cav1-KO mice, RT-qPCR analysis of selected cytokines, surface markers and chemokines was performed (Fig. 2A). Consistent with the observed inflammation, H. pylori SS1 induced expression of TNFalpha and IFNgamma in the gastric mucosa of both WT and KO mice. In addition, we stated an increased mRNA expression of CD19 (B-cells) (1.6±0.3 WT versus 3.3±0.9 KO; p = 0.0512; n = 15 per genotype) and RANTES (CCL5) (1.3±0.2 WT versus 2.1±0.6 KO; p = 0.0449; n = 15 per genotype) in gastric tissue of H. pylori-infected Cav1-KO mice compared with infected WT mice. In contrast, mRNA levels of CD4 (T-helper cells), CD25 (T-regulatory cells) and CD86 (antigen-presenting cells) were suppressed by H. pylori independently of the Cav1 status. Immunohistochemistry (IHC) detected a marked increase of intramucosal F4/80-positive macrophages in gastric tissue of infected Cav1-KO mice compared with WT littermates (Fig. 2B). CD3-positive lymphocytes were located around and within intramucosal follicles (data not shown).
Similar results were obtained from experiments introducing rapid gastric injury in mice by injection of indomethacin [30] (Fig.S1). Consistent with the enhanced tissue damage in Cav1-KO stomachs (*p = 0.0161, WT versus KO, n = 9 per genotype), characterized by inflammation, erosion and ulceration, Cav1-deficient mice also expressed higher amounts of mRNAs encoding for the ulcer healing proteins trefoil factor-2 (TFF2) (0.8±0.3 WT versus 2.3±0.4 KO; *p = 0.0048; n = 9 per genotype) and peroxisome proliferator-activated receptor-gamma (PPARg) (0.6±0.2 WT versus 2.5±0.5 KO; *p = 0.0008; n = 9 per genotype). In sum, these data indicated that loss of Cav1 enhances the susceptibility of mice to gastric inflammation and tissue damage.
To assess the function of Cav1 during H. pylori infection in vitro, the human gastric epithelial cell line AGS was used which had been stably transfected with Cav1 expression plasmid (AGS/Cav1) or empty vector (AGS/EV) [37]. First, we examined whether Cav1 influences cell survival upon H. pylori infection (Fig. 3A). AGS clones with and without Cav1 were infected for 48 h with the cell-adapted CagA-delivery competent H. pylori strain G27 at different multiplicities of infection (MOI) ranging from 1∶100 to 1∶2000. Colorimetric MTT assays revealed that Cav1 had no effect on overall survival of AGS cells upon H. pylori infection. Similar results were obtained with CagA-delivery incompetent H. pylori SS1 and by Western blot (WB) analysis detecting the expression and phosphorylation of survival kinases (AKT/PKB, ERK1/2, p38MAPK) (data not shown). Since both H. pylori and Cav1 interact within lipid rafts, we asked whether adhesion of bacteria to cells depends on the presence of Cav1. AGS/Cav1 and AGS/EV cells were infected (MOI = 10) with G27 (Fig. 3B,C) or SS1 (data not shown) bacteria for 30 min, followed by washing and subsequent incubation in fresh medium for 2 h. Thereafter, cells were stained for immunofluorescence microscopy, and the number of bacteria which adhered to the Cav1-expressing or empty vector-transfected cells were counted (Fig. 3B,C). No differences in adhesion were observed between AGS/Cav1 and AGS/EV cells, suggesting that Cav1 does not influence adhesion of H. pylori bacteria to host cells.
The formation of needle-like projections (“humming bird”) is a typical morphological phenotype of AGS cells in response to infection with CagA-delivery proficient H. pylori strains and translocation of CagA into the cytosol [47]. To examine the role of Cav1 in this stress-induced rearrangement of the actin cytoskeleton, AGS/Cav1 and AGS/EV were infected for 16 h with H. pylori G27 wt or the isogenic mutant Delta cagA (MOI = 100). Infected cells were stained as described above, and the numbers of elongated AGS cells were determined (Fig. 4A,B). Cav1-deficient AGS cells showed considerably more elongated morphologies than Cav1-expressing cells (11±0.8% AGS/EV versus 4±0.8% AGS/Cav1; *p = 1.1×10−8; n = 3 per clone). As expected, no “humming bird” phenotype was obtained in cells infected with the CagA-delivery deficient SS1 or the CagA-deletion mutant G27 Delta cagA strains which are both unable to inject functional CagA protein into the host cells (data not shown). AGS/EV cells also produced more IL8 mRNA upon H. pylori G27 infection than AGS/Cav1 cells (64±19 EV versus 19±6 Cav1; *p = 0.0176; n = 3 per clone) (Fig. 4C). These data indicated that Cav1 protects against CagA-related cell stress.
In support of these findings, cell adhesion and wound closure rates were more pronounced in AGS/Cav1 compared with AGS/EV cells (Fig.S2). Consistent with its function as a target protein of CagA and component of focal adhesions [48], WB analyses (Fig. 4D) also detected higher levels (0.4±0.1 AGS/EV versus 1.4±0.1 AGS/Cav1, *p = 0.0012; n = 3 per clone) of phosphorylated focal adhesion kinase (FAK) in Cav1-expressing cells infected with H. pylori G27. These data corroborated that AGS/Cav1 cells infected with CagA-delivery competent H. pylori maintained their spread-out epithelial shape as compared with the stressed elongated phenotype of Cav1/EV cells.
Cav1 has been shown to be phosphorylated by cytosolic tyrosine kinases (Src, Abl) at tyrosine 14 [49], and phosphorylated Cav1 and Src both activate the small GTPases Rho/Rac/Cdc42 which regulate cytoskeletal functions [13], [50]. To identify the underlying molecular mechanism how Cav1 protects against CagA-related cell stress, we assessed the signalling pathways initiated by CagA-delivery proficient H. pylori G27. Infection of AGS cells evoked a rapid phosphorylation of Cav1 in AGS/Cav1 cells and of Src in both AGS/Cav1 and AGS/EV cells. This result indicated that Cav1 acts downstream of CagA-dependent Src activation but upstream of the activation of the small GTPases (Fig. 5A,B). Consistent with this conclusion, protein levels of phosphorylated JNK, which resides below of Src, were higher in AGS/EV cells compared with AGS/Cav1 cells.
We were unable to detect a direct interaction or quantitative colocalization of CagA protein or H. pylori G27 bacteria with Cav1 in CoIP or immunofluorescence experiments (Fig. 6A,B). Gentamycin protection assays revealed that the total amount of injected intracellular CagA was also independent of Cav1's presence (Fig. 6C). Thus, Cav1 neither inhibited adhesion of H. pylori bacteria to nor injection of CagA into the host cell, but rather reduced the down-stream effects of CagA on intracellular signalling.
To identify a candidate protein which confers protection against CagA in a Cav1-dependent manner, a protein interaction screen based on MALDI-MS was performed (Fig. 7A). AGS/Cav1 cells were infected for 16 h with H. pylori G27 (MOI = 100) followed by lysis of the cells at room temperature in MES-buffered 1% (v/v) Triton-X100. Protein bands precipitated by Cav1 antiserum were visualized by silver staining, and peptides were identified by MALDI-MS as published previously [29]. A protein fragment of ∼95 kDa contained peptides corresponding to variant 4 of p120 Rho GTPase-activating protein/deleted in liver cancer-1 (p120RhoGAP/DLC1) [51], [52], a tumor suppressor associated with focal adhesions and caveolae/lipid rafts [53]. DLC1 variant 4 (DLC1v4) has a predicted size of ∼110 kDa and was enriched in samples from cells that had been infected with H. pylori G27 compared to uninfected cells (Table S2). These results were confirmed by CoIP of Cav1 and endogenous DLC1 protein in AGS/Cav1 cells (Fig. 7B), indicating that H. pylori G27 evoked a specific recruitment of DLC1 to Cav1 in infected human gastric epithelial cells.
This result prompted us to amplify the cDNA of variant 4 of human DLC1 [39] from human hepatoma HepG2 cells (Fig. 7C). The cDNA was inserted into the expression vector pTarget (pT-DLC1v4) followed by transient transfection into parental AGS or HEK293 cells for 24 h. WB analyses detected expression of a ∼110 kDa protein, consistent with the predicted size of DLC1v4 [39]. Transiently transfected AGS cells were then infected with H. pylori G27 (MOI = 100) for additional 16 h. Immunofluorescence staining revealed that DLC1 per se did not inhibit formation of the CagA-induced “humming bird” phenotype (19±2% AGS/DLC1 versus 19±2% AGS/EV; n = 3 per clone) compared with empty vector-transfected cells (Fig. 8A,B). Instead, DLC1 promoted cell spreading (20±3% AGS/DLC1 versus 11±2% AGS/EV; *p = 0.0067; n = 3 per clone) consistent with its role in regulation of focal adhesions [54], [55], [56] (Fig. 8A,C). Pull-down assays which detected the activity of the small GTPases Rho/Rac/Cdc42 corroborated previous findings [56], [57], [58] that the CagA-proficient H. pylori G27 strain was a weak activator of these GTPases (data not shown). In sum, this data proposed that Cav1 inhibited CagA-induced cytoskeletal changes through alterations in the assembly or disassembly of focal adhesions via FAK rather than via the small GTPase pathways.
We showed previously that Cav1 is frequently down-regulated in human GC [37]. We therefore asked whether H. pylori infection contributes to repression of the Cav1 gene. RT-qPCR analyses of total RNA isolated from stomach tissue of uninfected mice and mice infected with CagA-delivery incompetent H. pylori SS1 (for 11 month) were performed. Infected WT mice showed a significantly reduced expression of Cav1 mRNA (13±4 WT+H. pylori versus 38±9 WT mock;*p = 0.0081; n = 15 per group) as compared to the uninfected WT mice (Fig. 9A).
Similar results were obtained from in vitro studies. Two different human GC cell lines with endogenous Cav1 expression, N87 and MKN45, and MDCK cells were infected for 3 days with CagA-delivery incompetent SS1 (Fig. 9B) or CagA-proficient G27 (Fig. 9C) H. pylori strains. In all cell lines, a robust reduction of Cav1 mRNA expression (by 62 to 85%; H. pylori versus mock; *p = 0.0001 to 0.0043; n = 3 per cell line) was observed compared with uninfected cells. Similar results were obtained for Cav1 protein by WB (Fig. 9C).
The mouse-adapted H. pylori SS1 strain, which had been used for our in vivo infections, contains the cagA gene, expresses cagA mRNA (data not shown) but does not exert CagA protein-dependent effector functions [40], [59], whereas the cell-adapted G27 strain delivers active CagA [56] into the host cells. We therefore assessed whether Cav1 down-regulation is CagA-dependent or not. The same three cell lines were infected with H. pylori G27 Delta cagA (Fig. 9C) for three days. The CagA-deleted strain also decreased the amounts of Cav1 mRNA compared with uninfected cells (by 67 to 89%; H. pylori versus mock; *p = 6.1×10−5 to 0.0249; n = 3 per cell line), emphasizing that the repression of the Cav1 gene was CagA-independent in vitro and in vivo.
To determine whether the down-regulation of the Cav1 mRNA was caused by inhibition of the Cav1 promoter, reporter assays were performed (Fig. 9D). MKN45 cells were transfected with a luciferase reporter plasmid pGL3 containing the human proximal Cav1 promoter (pGL3-CAV1p) followed by a 16 h infection with CagA-proficient H. pylori G27 (MOI = 100). As a positive control served the pGL3-SeRE plasmid which harboured a CagA/stress-responsive serum-response element (SeRE) [60]. H. pylori G27 infection significantly reduced the activity of the Cav1 promoter (to 53±1% H. pylori versus mock; *p = 2.7×10−6 to 0.0052; n = 3) compared with uninfected cells. Similar results were obtained from HEK293 cells (Fig. 9D) and with CagA-delivery incompetent H. pylori SS1 (data not shown). In contrast, the activity of the SeRE was increased in H. pylori G27 infected MKN45 cells but not of an unrelated control promoter from the human bile salt export pump (BSEP) (Fig. 9D). This data confirmed that Cav1 gene expression is down-regulated at the transcriptional level independently of CagA.
Next, we were interested to identify the H. pylori-responsive repressor of the Cav1 gene. H. pylori lowers cholesterol levels in the host [26], and SREBP1 is activated by sterol deficiency to negatively regulate Cav1 gene transcription [24]. We therefore examined whether there is a higher binding rate of active nuclear 68 kDa SREBP1 to the sterol-responsive elements (SREs) of the proximal human Cav1 promoter upon H. pylori infection. MKN45 cells were infected with H. pylori G27 for 24 h, and ChIP was performed using antisera against SREBP1 and H4-acetyl histone [45], a marker for transcriptionally active “open” chromatin. Immunoprecipitated DNA was amplified by an whole genome amplification approach [61] and used for genomic qPCR analysis (Fig. 10A). Upon infection, we observed an increased binding of SREBP1 to the sterol-responsive element-3 (SRE3) [24], [62] of the Cav1 promoter (2.5±0.5 H. pylori versus 0.4±0.2 mock;*p = 0.0093; n = 3). In contrast, the amount of H4-acetyl-histone protein at the SRE3 was reduced upon infection (0.5±0.5 H. pylori versus 3±0.6 mock;*p = 0.0478; n = 3). These results suggested that H. pylori inhibits transcription at this site by recruitment of SREBP1 as a repressor of the Cav1 gene.
We corroborated these results using EMSA (Fig. 10B) [45]. MKN45 cells were infected with H. pylori G27 (MOI = 100) for 24 h. We could detect binding of protein to the SRE3 oligonucleotide from the Cav1 promoter exclusively in nuclear extracts of infected cells. WB assays evinced that H. pylori G27 evoked accumulation of the active 68 kDa SREBP1 fragment in the nucleus. RT-qPCR analyses demonstrated that the expression of other bona fide SREBP1 target genes, which are positively regulated by SREBP1, was also affected by H. pylori. The mRNAs encoding 3-hydroxy-3-methyl-glutaryl-CoA synthase (HMGCOAS), HMGCOA reductase (HMGCOAR), low density lipoprotein receptor (LDLR) and acetyl-coenzyme A synthetase (ACS) were up-regulated by CagA-proficient G27 wt, CagA-deleted G27 Delta cagA and CagA-delivery deficient SS1 bacteria to a maximum of 21-fold (H. pylori versus mock; *p = 0.0137 to 0.0196; n = 3 per cell line) compared with uninfected cells (Fig. 10C). Conclusively, these data emphasized that the Cav1 promoter was inhibited by H. pylori-activated SREBP1 independently of CagA.
In this study, we describe a novel role for Cav1 in H. pylori-mediated gastritis and cell damage. Since many years, lipid rafts have been shown to mediate uptake of pathogens (virus, bacteria, parasites) and their toxins into host cells [8], [10], [11]. Internalization of the two major toxins of H. pylori, VacA and CagA, via clathrin-independent lipid raft-dependent endocytosis and the bacterial type IV secretion system have been thoroughly validated using in vitro systems, including the human gastric epithelial cell line AGS [4], [7]. However, the role of Cav1 in H. pylori infection in vitro and in vivo remained unknown.
Recent reports on Cav1-deficient mice revealed a general enhanced susceptibility to disease provoked by local or systemic infection through certain pathogens including bacteria (Salmonella typhimurium, Pseudomonas aeruginosa) or parasites (Trypanosoma cruzi) [12], [63], [64], [65], [66], [67], [68]. Cav1-KO mice succumb to systemic infection earlier and suffer from a more severe disease phenotype than WT littermates. This sensitivity is presumably caused by certain defects in either the innate or the adaptive immune system. Since the predominant cell types with Cav1 expression are macrophages [68] and endothelial cells [69], recruitment and maturation of leukocytes (e.g. of regulatory T-cells [70]) may be impaired in absence of Cav1. Loss of Cav1 in macrophages results in defective phagocytosis [68], [71] and altered release of nitric oxide [72] and pro-inflammatory cytokines (TNFalpha, IL1beta) [64], [65]. Bacterial lipopolysaccharide has been reported to up-regulate Cav1 expression in B-cells [23], and Cav1 was shown to be associated with molecules of the synapse between T-cells [21], [73] and antigen-presenting cells. Thus, systemic absence of Cav1 may impair immune responses to pathogens at multiple levels.
Consistent with these reports, we found that Cav1-deficient mice responded with an enhanced active chronic gastritis and tissue damage to infection with the CagA-delivery deficient H. pylori SS1 strain compared with WT littermates. This response was accompanied by loss of parietal cells and foveolar hyperplasia. A bias towards a T helper 1 immune response is expected to facilitate the elimination of H. pylori bacteria from infected stomachs, however, at the expense of a more severe gastritis in humans and mice [74], [75]. Consistent with this concept, we showed that Cav1-KO animals had a reduced bacterial burden but an augmented local infiltration of macrophages, marked formation of intramucosal lymph follicles and production of chemokines (e.g. RANTES/CCL5) in the infected gastric tissue. In line with the known immunomodulatory effects of H. pylori [76], [77], [78], the expression of CD-markers related to T helper (CD4/GATA4) and regulatory T cells (CD25/FOXP3) was suppressed upon an 11-month infection with H. pylori SS1, and this phenomenon was most pronounced in Cav1-KO mice. Although more detailed studies have to characterize the gastric milieu and the immune defects of Cav1-KO mice, one may conclude that loss of Cav1 enhances the susceptibility to pathogen-related disease by defects in the generation of a pathogen-directed adaptive T-cell immune response.
H. pylori has developed many strategies to evade the host's immune system [78]. It has been described that H. pylori is auxotrophic for cholesterol and extracts cholesterol from the host cell membrane to actively inhibit phagocytosis and modify the generation of adaptive T-cell responses [26], [79]. Depletion of host cell membranes from cholesterol inhibits CagA-dependent effects on cell elongation and IL8 production in vitro [27], [80]. These observations led us to the hypothesis that H. pylori may create a cholesterol-deficient microenvironment in/around infected cells which activates the cholesterol deficiency sensor SREBP1. Indeed, we demonstrated that SREBP1 was activated by H. pylori to down-regulate Cav1 gene expression in vitro and in vivo. This effect was independent of H. pylori's major oncoprotein CagA, but was strain-specific, because it was not observed upon infection with other Helicobacter species such as H. hepaticus (unpublished observation). In contrast to the CagA and VacA delivery-competent G27 strain, SS1 bacteria fail to exert bona fide effector functions of CagA and VacA proteins within host cells [40], [59], [81]. This defect may be attributed to the type IV secretion system or toxin delivery. Thus, further studies are necessary to identify the factors which are responsible for activation of SREBP1 by SS1.
To explore one of the in vitro mechanisms how Cav1 protects against H. pylori-related cell damage, we demonstrated that Cav1 neither interfered with adhesion of H. pylori SS1 or G27 bacteria to human gastric epithelial cells nor with injection of G27-derived CagA protein into the cytosol. Instead, Cav1 inhibited the down-stream effects of intracellular CagA on the rearrangement of the actin cytoskeleton and on the production of IL8. This in vitro phenomenon of spike-like cell elongation (“humming bird”) is supposed to resemble, at least in part, an in vivo event which facilitates access of live H. pylori into favourable niches of the gastric epithelium for successful persistence of the microorganism within the host organ [78]. We found that Cav1 did not directly interact with CagA. Instead, Src kinase was phosphorylated on specific tyrosine residues upon infection with H. pylori G27 independently of Cav1. Cav1 altered the activation status of two kinases downstream of Src, it reduced phosphorylation of JNK but enhanced that of FAK, proposing that Cav1 blunts stress-related CagA-signalling downstream of active Src but promotes cell adhesion.
Mechanistically, we evinced that the CagA-proficient H. pylori G27 evoked the recruitment of p120RhoGAP/DLC1 to Cav1. DLC1 has been initially described as an inhibitor of small GTPases which localizes to focal adhesions and lipid raft/caveolae membrane microdomains [53], [54], [55], [82]. Cav1 may thus promote the function/activity of DLC1 as a tumor suppressor via direct interaction through a bona fide Cav1-binding motif [53] identified in DLC1. Unexpectedly, DLC1 did not inhibit the formation of fiber-like elongations in cells infected with CagA-delivery competent H. pylori G27 which have been attributed to activation of the small GTPases RhoA/Rac1/Cdc42 by CagA. This result may be explained by previous reports [56], [57], [58] showing that the G27 strain is only a weak activator of those GTPases. Instead, DLC1 promoted cell adhesion and spreading. This phenotype was presumably caused by changes in the assembly and/or disassembly of focal adhesions, since FAK is a direct target of H. pylori's CagA [48] and, together with other components of focal adhesions, such as talin and tensins, directly interacts with DLC1 [53], [54], [55]. Conclusively, Cav1 seems to exert its protective effect against intracellular CagA effector functions on the actin cytoskeleton via DLC1. Cav1 did not require direct interaction with CagA to exert its pro-adhesive effects. Hence, the Cav1/DLC1 complex may also protect cells against CagA-delivery deficient H. pylori strains, including SS1, which have been used in our in vivo study. However, future experiments have to explore this assumption.
Cav1 is a ubiquitous adapter molecule in many immune receptor signalling pathways. One may thus speculate that H. pylori exploits down-regulation of Cav1 to subvert the host immune system or to enhance the signalling efficiency of its virulence factors in gastric epithelial cells. In case of clinical isolates from infected individuals, the pre-mouse Sydney strain-1 (PMSS1) or G27 [32], [77] those virulence factors may comprise CagA and VacA, in case of SS1, other yet unknown bacterial proteins could be involved.
Loss of Cav1, in its function as a tumor suppressor and inhibitor of growth factor receptor signalling which stabilizes cell-cell and cell-matrix contacts, is a hallmark of many human cancers including GC [19]. Absence of Cav1 in primary tumors promotes cell proliferation and enables clonal expansion [15], [19]. Similar to Cav1, DLC1 is a tumor suppressor silenced or deleted in many human cancer entities including GC, e.g. by gene methylation [39], [51]. Thus, down-regulation of Cav1 by H. pylori in stomach tissue in vivo may be part of an early molecular sequence of events in the transition of inflammation to GC also in humans.
We would like to emphasize that the aim of the current work was not the identification of novel virulence factors in the SS1 H. pylori strain, which are unknown since 15 years, but rather to clarify the role of Cav1 in H. pylori-induced gastric pathology by directly comparing the lesions obtained in the Cav1-KO mice to the results of other researchers who used SS1 H. pylori strain in other mouse genotypes or backgrounds (C57BL/6, B6129, BALB/c) [28], [33], [34], [83]. The Cav1-KO mice did not progress to gastric neoplasia with CagA-delivery incompetent SS1. We therefore additionally investigated the molecular mechanisms of Cav1 on CagA-delivery proficient G27 strain signalling in gastric epithelial cell lines, in order to strengthen the relevance of our findings to the situation in humans, where CagA-injection competent strains are associated with development of GC [1], [3], [5]. In the future, we shall infect Cav1-KO mice with PMSS1, an H. pylori strain which injects functionally active CagA protein into the host gastric epithelium [77], [84], [85].
Our study describes two different aspects of Cav1's role in stomach disease: (i) first, an in vivo protective role against H. pylori-induced inflammation which was independent of CagA/VacA, and (ii) second, an in vitro protective role against H. pylori-induced cytoskeletal rearrangement which was dependent on CagA and DLC1. These data comprise two separate aspects of H. pylori biology which are not easily reconciled. Nevertheless, our major objective was to present a first description of a beneficial role for Cav1 in H. pylori-related diseases, against gastritis in vivo and cytoskeletal stress in vitro, rather than to elaborate on the potential virulence mechanisms of SS1. To our knowledge, this novel role of Cav1 in H. pylori biology was previously unknown and may thus provide the initial basis for further detailed in vivo and in vitro studies
We have been well aware of the fact that H. pylori strain SS1 is incapable of exerting CagA and VacA-dependent effector functions [81], [83], [86], [87]. Over the years, a consensus has been reached that SS1 expresses CagA mRNA but does not inject functional CagA protein into the host cells via the type IV secretion system [40]. Similarly, SS1 is devoid of VacA-dependent vacuolating cytotoxicity and induction of IL8 [81]. Nevertheless, SS1 is still able to induce severe gastric pathology in vivo independently of these two important virulence factors, especially after chronic infection and persistent colonization [35], [59]. The SS1 virulence factors responsible for gastric inflammation, tissue damage and carcinogenesis have remained unknown since the introduction of the strain as a standardized reference model in 1997 [32].
Several alternative mechanisms of virulence have been described for SS1. For example, SS1 up-regulates matrix-metalloproteinases (MMPs) inducing inflammation and tissue damage independently of CagA [86]. Moreover, SS1 per se evokes mutations and genotoxic stress in mice, a pathology which may contribute to pre-neoplastic alterations in the gastric mucosa [33], [88]. Interestingly, the CagA-delivery competent PMSS1 strain promotes genotoxic stress as well independently of its common virulence factors CagA and VacA [89]. Instead, bacterial adhesion factors were necessary to achieve the mutagenic effect [89]. Nevertheless, SS1 bacteria deficient in certain adhesion factors still evoked severe gastric pathology in vivo (here in gerbils) [90], and the presence of CagA or VacA had no effect on the ability of HP strains to adhere or invade gastric epithelial cells in vitro [87]. Thus, the quest for pathogenic virulence mechanisms of SS1 is still ongoing.
These reports demonstrated, that single major virulence factors like CagA or VacA are not the exclusive responsible agents for the observed gastric histopathology induced by SS1, but rather a combination of so far unknown bacterial factors and last but not least the host immune response. Relating to the latter, H. pylori-induced changes in cholesterol content at host cell membranes and the well-described immunomodulating effect of H. pylori may be of higher importance for pathogenicity than the action of single cytotoxins/oncoproteins. Confirming this assumption, we showed that SS1 (CagA−/VacA−) and G27 (CagA+/VacA+) strains both down-regulated SREBP1-mediated Cav1 gene expression independently of CagA/VacA, constituting a potential novel pathogenic mechanism of H. pylori which acts independently of classical virulence factors.
|
10.1371/journal.ppat.1003663 | Sequestration by IFIT1 Impairs Translation of 2′O-unmethylated Capped RNA | Viruses that generate capped RNA lacking 2′O methylation on the first ribose are severely affected by the antiviral activity of Type I interferons. We used proteome-wide affinity purification coupled to mass spectrometry to identify human and mouse proteins specifically binding to capped RNA with different methylation states. This analysis, complemented with functional validation experiments, revealed that IFIT1 is the sole interferon-induced protein displaying higher affinity for unmethylated than for methylated capped RNA. IFIT1 tethers a species-specific protein complex consisting of other IFITs to RNA. Pulsed stable isotope labelling with amino acids in cell culture coupled to mass spectrometry as well as in vitro competition assays indicate that IFIT1 sequesters 2′O-unmethylated capped RNA and thereby impairs binding of eukaryotic translation initiation factors to 2′O-unmethylated RNA template, which results in inhibition of translation. The specificity of IFIT1 for 2′O-unmethylated RNA serves as potent antiviral mechanism against viruses lacking 2′O-methyltransferase activity and at the same time allows unperturbed progression of the antiviral program in infected cells.
| Cellular messenger RNAs of higher eukaryotes are capped with a methylated guanine and, in addition, methylated at the 2′O position of the first ribose. Viruses unable to methylate their RNA at the 2′O position of the cap and viruses generating uncapped RNA with 5′ triphosphate groups are inhibited by an antiviral complex of different IFIT proteins. How IFIT proteins restrict viruses lacking 2′O methylation at the RNA cap remained unclear. We used a mass spectrometry-based approach to identify proteins binding to capped RNA with different methylation states. We found that IFIT1 directly binds to capped RNA and that this binding was dependent on the methylation state of the cap. Having identified IFIT1 as being central for recognition of 2′O-unmethylated viral RNA we further examined the mode of action of IFITs in vitro and in vivo. Our experiments clearly show that the antiviral mechanism of IFIT1 is based on sequestration of viral RNA lacking cap 2′O methylation, thereby selectively preventing translation of viral RNA. Our data establish IFIT1 as a general sensor for RNA 5′ end structures and provide an important missing link in our understanding of the antiviral activity of IFIT proteins.
| Effective control of viral infection by host organisms requires sensing of pathogens and activation of appropriate defence mechanisms [1]–[3]. One component commonly sensed by the host is viral genetic material, whether DNA delivered to the cytoplasm through viral infection or viral RNA bearing motifs not commonly found on eukaryotic RNAs [4], [5]. Most cellular cytoplasmic RNAs are single-stranded, and bear a 5′monophosphate (rRNAs and tRNAs), or an N7 methylated guanosine cap (mRNAs) linked via a 5′-to-5′ triphosphate bridge to the first base. In higher eukaryotes, mRNA is further methylated at the 2′O position of the first ribose [6], [7]. Viruses, in contrast, can form long double-stranded RNA (dsRNA) and generate RNAs bearing 5′triphosphosphates (PPP-RNA) or RNAs lacking methylation [8]–[10]. All these distinct features of viral as opposed to cellular RNAs have been shown to activate the innate immune system and elicit synthesis of antiviral cytokines including Type I interferons (IFN-α/β), which ultimately restrict virus growth [11]–[14]. Among the proteins that sense viral RNA and are linked to IFN-α/β synthesis are retinoic acid-inducible gene I (RIG-I) and melanoma differentiation-associated gene 5 (Mda-5), which form the family of RIG-like receptors (RLRs) [5]. A further set of host proteins appears to bind virus-derived RNAs to directly inhibit virus production [8]. Several of these proteins are highly expressed upon stimulation of cells with cytokines like IFN-α/β and their antiviral effects become apparent only after binding to virus-derived nucleic acid. Prominent examples for such proteins are dsRNA binding proteins such as dsRNA-activated protein kinase R and 2′-5′ oligoadenylate synthetase, and proteins that bind PPP-RNA, like interferon-induced proteins with tetratricopeptide repeats (IFIT) 1 and -5 [3], [15], [16]. Little is known about the repertoire of cellular proteins that recognise unmethylated cap structures, although replication of viruses with inactive RNA 2′O methyltransferase is strongly inhibited by IFN-α/β in vitro and in vivo [11], [15]. Some of this antiviral activity has been genetically linked to Ifit1 and -2 in mice [17]–[19]. Here, we used an unbiased mass-spectrometry-based approach to identify cellular proteins that bind to 5′ unmethylated and methylated capped RNA, and explored their contribution to antiviral host responses.
To identify proteins that interact with 5′ capped RNA we used a proteomics approach based on affinity purification and mass spectrometry (AP-MS) [16]. RNA bearing terminal 5′ hydroxyl (OH-RNA), 5′ triphosphate (PPP-RNA), an unmethylated cap (CAP-RNA), a guanosine-N7 methylated cap (CAP0-RNA), or a guanosine-N7 methylated cap and a ribose-2′O methylated first nucleotide (CAP1-RNA) was coupled to agarose beads. The beads were then incubated with lysates of naïve HeLa cells or HeLa cells treated with IFN-α to increase the abundance of antiviral proteins (Fig. 1a, Fig. S1). By employing liquid-chromatography coupled to tandem mass spectrometry (LC-MS/MS) followed by quantitative interaction proteomics analysis, we identified 528 proteins that interacted with unmodified or RNA-coated beads (Fig. S2a, Table S1). While a large number of proteins were equally well represented in the bound fractions obtained with all RNAs (Fig. S2a), 68 proteins were found to be significantly enriched in samples recovered with 5′modified RNA compared to OH-RNA (Fig. S2b). As expected, the PPP-RNA binding proteins RIG-I (DDX-58), the IFIT1, -2, -3 complex and IFIT5 were enriched in PPP-RNA affinity purifications of IFN-α-treated HeLa cell lysate (Fig. 1b), validating the approach and confirming previous data [16]. Using unmethylated CAP-RNA as bait, we significantly enriched for proteins known to associate with cellular capped RNA (12 of 16 proteins) (Fig. 1c, Fig. S2b, Table S1). However, an important feature of cellular mRNAs is methylation on the N7 position of the guanosine cap and the ribose-2′O position of the first nucleotide (CAP1). N7 methylation is known to increase the affinity of the cap structure for proteins such as EIF4E and other cap-binding proteins [6], [7]. A methylation-dependent increase in protein binding was also evident in our AP-MS analysis when unmethylated CAP-RNA and methylated CAP1-RNA were used as baits (16 vs. 27 identified proteins), as the latter captured a higher number of significantly enriched proteins and, overall, these were enriched to a greater degree, as measured by label-free quantification (Fig. 1c–d, Table S1). Notably, we identified IFIT1, -2 and -3 among the uncharacterised CAP-RNA binding proteins, suggesting that the IFIT complex binds to RNA in a cap-dependent manner (Fig. 1c). IFIT5, which shows 57.2% aminoacid sequence identity and 75.6% similarity to IFIT1 and has recently been shown to form a tight binding pocket that specifically accommodates PPP-RNA [20], was not detected in fractions that bound capped RNA. When we compared our AP-MS dataset with transcriptome data of interferon-stimulated cells [21], IFITs were the only interferon-induced proteins found to be specifically enriched in CAP-RNA purifications, suggesting a predominant role of IFITs in innate immune responses directed against CAP-RNA (Fig. 1c, Fig. S2b). To analyse whether the set of proteins that binds to 5′modified RNA is conserved in other species, we performed the same AP-MS analysis on lysates of naïve and IFN-α-treated mouse embryonic fibroblasts (MEFs) (Fig. S3a, b, Table S2). Surprisingly, although PPP-RNA specifically enriched for Ifit1, the abundance of Ifit2 and Ifit3 was not increased (Fig. 1e). Instead we found enrichment of Ifit1c (also known as Gm14446), an uncharacterised IFIT protein that is strongly induced by IFN-α/β or virus infection (Fig. S4), suggesting that the architecture of the murine IFIT complex differs from that of its human counterpart. Significant enrichment for Ifit1 and Ifit1c could also be achieved with unmethylated CAP-RNA, but not with methylated CAP1-RNA, despite the fact that the latter bait captured more proteins with higher enrichment scores (Fig. 1f, g). We concluded from these analyses that, in both human and mouse, the IFIT complex is the only IFN-induced component that shows significant affinity for capped RNA.
Since human IFIT1, -2 and -3 associate with each other to form a multiprotein complex, we wished to determine which of them was responsible for tethering the IFIT complex to unmethylated CAP-RNA. We overexpressed each of the IFIT proteins, tagged with Renilla luciferase, in 293T cells and performed affinity purifications using OH-RNA, PPP-RNA and CAP-RNA. Remarkably, only human and murine IFIT1 were detected when CAP-RNA was used as bait (Fig. 2a, b), suggesting that IFIT1 mediates binding of the IFIT complex to CAP-RNA. Consistent with the MS analysis, IFIT5 exclusively bound to PPP-RNA but not to CAP-RNA. To exclude contribution of cellular factors to the interaction between IFIT1 and CAP-RNA we used recombinant human IFIT proteins for RNA precipitations which confirmed a direct interaction of IFIT1 with capped RNA (Fig. 2c). A structure-based modelling approach using IFIT5 [20] as template suggested that the RNA-binding cavity of IFIT1 is ∼700 Å3 larger than that of IFIT5 (Fig. S5) – implying that IFIT1 has slightly different RNA-binding properties. However, a lysine at position 151 and an arginine at position 255 of IFIT1, two residues involved in binding the terminal 5′ triphosphate group on PPP-RNA by IFIT5 and IFIT1 [20], were also required for binding of IFIT1 to capped RNA (Fig. 2d), indicating an overall similar mode of binding.
To provide additional evidence that binding of IFIT1 is indeed responsible for associating the IFIT complex to CAP-RNA, we performed AP-MS experiments on wild-type (Ifit1+/+) and mutant, Ifit1-deficient (Ifit1−/−) MEFs. The overall precipitation efficiency was comparable in both cell types, as evidenced by equal enrichment of the RNA-binding protein Syncrip and the cap-binding protein Ncbp1 (Fig. 2e and Fig. S4b). Ifit1c was not enriched in precipitates from Ifit1−/− MEFs, which is consistent with the notion that the murine Ifit complex binds to CAP-RNA through Ifit1. These results show that the specific binding properties of IFIT1 are essential for recruitment of the human and murine IFIT complexes to their RNA targets.
To identify proteins that bind capped RNA in a methylation-dependent manner we used unmethylated CAP-RNA and fully methylated CAP1-RNA as baits with IFN-treated HeLa cell lysates and quantified the captured proteins by LC-MS/MS. As expected [6], [7], most cellular proteins were significantly enriched in the CAP1-RNA bound fraction (Fig. 3a, Fig. S2c). The most notable exceptions were IFITs and the cellular 2′O-methyltransferase FTSJD2, both of which clearly favoured CAP-RNA (Fig. 3a, Fig. S2c and Fig. 1 c, d and f, g). We confirmed the MS data by a series of RNA precipitations followed by western blotting for endogenous proteins. Proteins associating to RNA in a 5′ independent manner, such as ILF3, precipitated similarly well regardless of the RNA used (Fig. 3b). Cap N7 methylation increased the association of EIF4E to RNA and methylation of the 2′O position did not impair precipitation efficiency. In accordance with the MS results, IFIT1 bound well to unmethylated CAP-RNA and CAP0-RNA (N7 methylated cap) but revealed reduced binding to CAP1-RNA (N7 methylated cap and 2′O methylated first ribos).
We next tested the contributions of individual cap methylation sites to IFIT1 binding. To this end, we measured binding of luciferase-tagged human and murine IFIT1 with either CAP-, CAP0- or CAP1-RNA. The unmethylated CAP-RNA bait captured more human or murine IFIT1 than either of the methylated RNAs (Fig. 3c). Furthermore, the analysis suggested that N7 methylation on the cap and 2′O methylation of the first ribose both contributed to the reduced binding of IFIT1 to RNA. Similarly, the precipitation efficiency of recombinant human and murine IFIT1 was reduced when capped in vitro transcribed RNAs were enzymatically methylated at the N7 and 2′O position (Fig. 3d) or when chemically synthesised RNAs with the same modifications were used (Fig. 3e). This was in contrast to EIF4E that showed prominent binding when CAP0- or CAP1-RNA was used (Fig. 3d, e). Collectively, these data suggest that human and murine IFIT1 have the capability to directly sense the methylation state of capped RNA.
Having established that IFIT1 binds directly to capped RNA and that methylation on the 2′O position of the first ribose markedly reduces binding, we tested the impact of IFIT1 on virus replication. Probably as a result of evolutionary pressure, most viruses that infect higher eukaryotes have evolved mechanisms to generate RNA that is methylated on both the N7 position of the guanosine cap and the 2′O position of the first ribose [9]. We therefore used wild-type human coronavirus (HCoV) 229E (229E-WT), which generates CAP1-RNA, and a mutant variant that has a single amino acid substitution (D129A) in the viral 2′O methyltransferase that is part of non-structural protein 16 (229E-DA), and consequently only produces CAP0-RNA [11]. IFN-α-treated HeLa cells infected with the 229E-DA mutant expressed significantly reduced levels of viral RNA and protein relative to those exposed to 229E-WT (Fig. 4a, b). Moreover, this effect was strictly dependent on IFIT1, since the two viruses replicated equally well in HeLa cells treated with siRNA against IFIT1 (Fig. 4a, b). Similar effects were observed in an analogous mouse model. Thus, when IFN-α treated macrophages (MΦs) from C57BL/6 (Ifit +/+) mice were infected with a wild-type murine coronavirus (mouse hepatitis virus strain A59; MHV-WT) and a mutant strain carrying the equivalent amino acid substitution (D130A) in its 2′O methyltransferase [11], [17] (MHV-DA), the latter produced 100-fold less viral RNA and comparably reduced levels of viral protein (Fig. 4c, d). In contrast, when Ifit1-deficient MΦs were infected, no significant virus-dependent differences were observed, again pointing to a critical role for Ifit1 in restricting replication of MHV-DA. Note that the presence of Ifit1 itself did not increase IFN-α/β production (Fig. S6), suggesting a direct antiviral effect of Ifit1. We next assessed the impact of Ifit1 on virus growth in vivo. MHV-WT grew to high titres in the spleens of infected Ifit1+/+ mice, whereas no viral replication could be detected upon infection with MHV-DA (Fig. 4e). In agreement with the in vitro data, growth of MHV-DA was partially restored in Ifit1-deficient animals. These data suggest that IFIT1 has a central role in restraining the growth of 2′O methyltransferase-deficient coronaviruses in vitro and in vivo, which is compatible with the greater affinity of IFIT1 for non-2′O-methylated RNA cap structures. The data further imply that this role is conserved in mouse and human.
RNA capping is essential for a variety of cellular functions. The presence of a 5′ cap regulates mRNA export from the nucleus, protects RNAs from degradation and is necessary for efficient translation [7], [22]. An involvement of IFIT1 in nuclear-cytoplasmic transport is unlikely, given the exclusively cytoplasmic localisation of IFIT proteins and their negative effect on coronaviruses, which replicate in the cytoplasm. We therefore measured the stability of the RNAs generated by MHV-WT or MHV-DA in MΦs that had been stimulated with IFN-α. Since MHV-WT replicates significantly better than the mutant virus, we blocked virus replication by adding cycloheximide (CHX) shortly after infecting MΦs with the two viruses (Fig. 5a). CHX inhibits de novo synthesis of the viral polymerase, a prerequisite for transcription of viral RNA and thereby allows to normalise for viral transcripts in coronavirus infected cells. The abundance of viral transcripts 4 h and 8 h after infection was indistinguishable in CHX treated cells infected with MHV-WT and MHV-DA (Fig. 5b), suggesting that 2′O methylation of the first ribose does not affect the stability of the viral RNA within the timeframe of this experiment.
Many cellular antiviral defence mechanisms generally block translation of mRNA, thereby also severely inhibiting virus growth. To assess the global impact of Ifit1 on the translation machinery, we used pulsed stable isotope labelling in cell culture (SILAC) [23]. In pulsed SILAC, unlabelled cells are transferred to SILAC growth medium containing 13C- and 15N-labelled arginine (Arg10) and lysine (Lys8). Newly synthesized proteins incorporate the heavy label and pre-existing proteins remain in the light form, which allows to measure relative changes in the translation of individual proteins, regardless of the absolute amount of RNA present. We pulsed Ifit1+/+ and Ifit1−/− MΦs infected for 5½ h with either MHV-WT or MHV-DA for 2 h with SILAC medium (Fig. 5c) and analysed infected cells by whole-proteome shotgun LC-MS/MS. We could reliably quantify 721 proteins in terms of heavy/light ratios in all three biological replicates tested. Heavy/light ratios of cellular proteins were comparable in Ifit1+/+ and Ifit1−/− MΦs, irrespective of the virus used for infection (Fig. 5d, boxes), suggesting that neither the presence of Ifit1 nor infection with MHV-DA affected the overall rate of translation in the cells. The expression profiles of individual proteins known to be important in innate immune responses against viruses, such as the pattern recognition receptor RIG-I (DDX58), signalling molecules (STAT1, -2, -3), interferon-induced proteins (Ifi205b, Ifi35, Gvin1) and components of the major histocompatibility complex (H2-K1, H2-D1, Cd74), were similar in both cell types infected with either virus (Fig. S7). However, translation of viral nucleocapsid and membrane proteins was selectively reduced in Ifit1+/+ MΦs infected with MHV-DA (Fig. 5d and Fig. S7). Variation in large datasets can be best evaluated by principal-component analysis, which computes the variable with the greatest effect in a given dataset. This analysis revealed that Ifit1+/+ MΦs infected with MHV-DA showed the highest variation (Component 1 accounting for 55.9% of variation) as compared to all other conditions tested (Fig. 5e), and among all identified proteins, MHV proteins were mainly responsible for this variation (Fig. 5f). Taken together, these data indicate that synthesis of proteins encoded by viral RNAs lacking 2′O methylation on the first ribose is specifically inhibited by IFIT1. Expression of proteins encoded by fully methylated RNA, such as cellular mRNA or 2′O methylated viral RNA, is not affected by the activity of IFIT1.
Translation of cellular capped mRNA requires binding of the cap-binding protein EIF4E, which has a high affinity for methylated cap structures [7], [22]. Therefore, we tested whether IFIT1 could compete with EIF4E for binding to RNA template. We coupled limiting amounts of unmethylated CAP-RNA, N7-methylated CAP0-RNA and fully methylated CAP1-RNA to beads and tested whether the binding ability of recombinant EIF4E is altered by the presence of recombinant IFIT1. When we used CAP-RNA or CAP0-RNA, EIF4E binding to the beads was reduced by addition of IFIT1, suggesting that the two proteins compete for the RNA target (Fig. 6a). In contrast, when methylated CAP1-RNA was used the amount of EIF4E recovered was not affected by the presence of IFIT1. Competition between Eif4e and Ifit1 for capped RNA was also seen when total lysates of IFN-α-stimulated MEFs were used as inputs for experiments. Unmethylated CAP-RNA captured considerably more Eif4e from lysates of IFN-α treated Ifit1−/− MEFs than from lysates of Ifit1+/+ MEFs (Fig. 6b). This difference disappeared when methylated CAP1-RNA was used as bait (Fig. 6b). We therefore conclude that IFIT1 competes with cellular translation initiation factors for mRNA, thereby selectively regulating translation based on the 5′ methylation status of the RNA templates present (Fig. 6c).
We previously identified IFIT1 as a nucleic acid-binding protein that recognises the 5′triphosphosphate present on genomes and transcripts of most negative-strand RNA viruses [16]. Here we show that, in addition, IFIT1 binds mRNAs that lack 2′O methylation on the first ribose, such as those produced by RNA viruses that replicate in the cytoplasm and are deficient in RNA cap-specific ribose-2′-O methyltransferase activity. This suggests that IFIT1 has a unique ability to recognize 5′ RNA modifications that are present on viral nucleic acids. Co-purification experiments with human IFIT proteins clearly show formation of a multiprotein complex comprising IFIT1, -2 and -3. Overexpression of single IFIT proteins, including IFIT1, only marginally affects viral growth [16], [17], suggesting that the cooperative action of IFIT proteins is required for full antiviral action. This is supported by loss-of-function experiments in cell culture and in vivo that show a requirement for Ifit2, which by itself does not bind CAP-RNA, to restrict viruses lacking 2′O methyltransferase activity [17], [18]. IFIT2 is known to bind to components of the cytoskeleton [24], which could allow intracellular trafficking of the IFIT complex to its sites of action. While some IFITs possess conserved biological activities in different species, e.g. human and murine IFIT1 which bind to PPP-RNA and unmethylated CAP-RNA, others appear to have evolved in a species-specific manner. We showed here that the yet uncharacterised murine interferon-induced Ifit1c binds to RNA-coated beads in an Ifit1-dependent manner, and we therefore propose that a corresponding Ifit complex with a different protein composition exists in mice.
Residues previously identified to be important for binding of the triphosphate moiety are also required for binding of unmethylated CAP-RNA by IFIT1, suggesting a conserved mechanism of RNA binding. In this context it is of interest to note that crystallographic analysis indicates that PPP-RNA binding to IFIT5, which shows high similarity to IFIT1, occurs in a fashion that is reminiscent of CAP-RNA binding by cap-binding proteins, in that the first two nucleotides are stacked by an aromatic phenylalanine [20]. However, the higher affinity of IFIT1 for unmethylated relative to fully methylated capped RNA is unusual among cellular proteins since 5′ methylation has so far been reported to increase the affinity of cellular proteins for RNA [7], a notion clearly supported by our RNA AP-MS data. Like its specific antiviral activity, this property of IFIT1 may only become apparent during infections with viruses that produce non-methylated RNA 5′ ends [25], [26]. We propose that IFIT1 acts as a molecular switch that allows selective translation based on the 5′ methylation state of the mRNA. The phenomenon of translational control by IFIT1 based on its differential affinity for the capped RNA is reminiscent of the 4E homologous protein (4EHP) in Drosophila and mice, which has been found to control translation by competing with EIF4E for the RNA cap structure, thereby regulating development-specific gene expression [27], [28]. Similarly, in our hands, IFIT1 does not associate directly with the translation machinery ([16] and data not shown), which again suggests that it perturbs translation through sequestration of viral RNA. Such a model is consistent with the high expression levels of IFIT proteins resulting from infections with viruses or treatment with IFN-α/β.
Rather than mediating general inhibition of translation, IFIT1 shows high selectivity for mRNAs that lack 5′ methylation. This is supported by pulsed SILAC experiments showing specific, IFIT1-dependent inhibition of translation of capped RNAs lacking 2′O methylation at the first ribose, such as those generated by MHV and HCoV mutants expressing inactive 2′O methyltransferase. Lower eukaryotes and viruses that infect them lack 2′O methylated CAP RNA [29]–[31], and the latter should be susceptible to the antiviral activity of IFITs. Consequently, the IFIT defence system is likely to contribute to a species barrier that puts selective pressure on viruses to generate 5′ methylated RNA. Our data provide a mechanistic rationale for why most viruses make considerable effort and dedicate part of their coding capacity to produce genomic and subgenomic RNAs with 5′-terminal ends that perfectly mimic those of cellular mRNAs, including fully methylated 5′-cap structures [9], [31]–[33]. Other viruses have evolved specific mechanisms to hide their uncapped/unmethylated 5′ ends, for example, by covalent binding of viral proteins to the 5′ end of viral RNAs and use of alternative strategies for translation initiation, thereby escaping IFIT1-based surveillance, which is centred on RNA 5′ end structures. Despite these viral strategies to generate host-like mRNAs, IFIT1 remains active against viruses that generate 5′ triphosphate RNA, most likely through translation-independent mechanisms. The ability of IFIT1 to target viral RNAs selectively allows the cell to specifically fight virus infections while pursuing an antiviral program aimed at destroying the intruding pathogen.
All animal experiments were performed in accordance with Swiss federal legislation on animal protection and with the approval of the Animal Studies Committee of the Cantonal Veterinary Office (St. Gallen, Switzerland), license nr. SG 11/03.
Interferon-α (IFN-α A/D) was a kind gift from Peter Stäheli. Expression constructs for human and murine IFIT proteins [16], [20] and the human aminopeptidase N (APN) were described previously. Products tagged with Renilla luciferase were expressed from constructs obtained by Gateway cloning into pCDNA-REN-NT-GW (a kind gift from Albrecht v. Brunn). For expression in bacteria, human EIF4E cDNA was cloned into pETG10A-GW [16]. Recombinant IFIT proteins and human EIF4E were expressed in E. coli and purified using HisPur Ni-NTA resin (Thermo Scientific). Streptavidin-agarose beads were obtained from Novagen. Polyclonal antibodies directed against human and mouse IFIT1 were described previously [16]. The antibody against MHV nucleoprotein (MHV-N556) was kindly donated by Stuart Siddell. Primary antibodies against ILF-3 (Sigma; HPA001897), the nucleoprotein of HCoV-229E (Ingenasa; mAb 1H11) and EIF4E (Cell Signaling; C46H6) were obtained from commercial sources. For western blot analysis we used horseradish peroxidase (HRP)-coupled antibodies specific for actin (Santa Cruz; sc-47778), the His-tag (Santa Cruz; sc-8036) or the c-Myc-tag (Roche; 1667149), and HRP-coupled secondary antibodies (Jackson ImmunoResearch). All cell lines used (293T, HeLa, Vero-E6, Huh7, L929, 17Clone1, and Ifit1+/+ and Ifit1−/− mouse embryonic fibroblasts) were described previously [11], [16], and were maintained in DMEM (PAA Laboratories) containing 10% fetal calf serum (PAA Laboratories) and antibiotics (100 U/ml penicillin, 100 µg/ml streptomycin). DMEM medium containing antibiotics, 10 mM L-glutamine, 10% dialyzed fetal calf serum (PAA Laboratories) and 84 mg/L 13C6 15N4 L-arginine and 146 mg/L 13C6 15N2 lysine (Cambridge Isotope Laboratories) was used for SILAC experiments. Murine bone marrow-derived macrophages were generated in vitro by cultivating bone marrow from mouse femur and tibia in DMEM supplemented with 10% (v/v) fetal calf serum, 5% (v/v) horse serum, 10 mM HEPES pH 7.4, 1 mM sodium pyruvate, 10 mM L-glutamine and 20% (v/v) L929 cell-conditioned medium (containing macrophage colony-stimulating factor) for 6 days. Reagents for transfection with plasmid DNA (Nanofectin) or siRNA duplexes (siRNA Prime) were obtained from PAA Laboratories. Wild-type and 2′-O-methyltransferase-deficient recombinant coronaviruses [mouse hepatitis virus strain A59 (MHV) and human coronavirus 229E (HCoV-229E) [11]], Sendai virus, RVFV Clone13 [34] and VSV-M2 (mutant VSV with the M51R substitution in the matrix protein) [35] have been described previously. Duplex siRNAs targeting human IFIT1 [sense#1: r(CAUGGGAGUUAUCCAUUGA)dTdT; antisense#1: r(UCAAUGGAUAACUCCCAUG)dTdA; sense#2: r(CCUUGGGUUCGUCUACAAA)dTdT, antisense#2: r(UUUGUAGACGAACCCAAGG)dAdG] and the green fluorescent protein [sense: 5′ r(AAGCAGCACGACUUCUUCAAGU)dT 3′; antisense 5′ r(CUUGAAGAAGUCGUGCUGCUUU)dT 3′] were synthesized by the Core Facility at the MPI of Biochemistry.
Triphosphorylated PPP-RNA was synthesized by in vitro transcription with SP6 or T7 polymerase (RiboMAX Large Scale RNA Production Systems; Promega), in the presence or absence of biotin-16-UTP (Enzo), from plasmids encoding antisense 7SK RNA (7SK-as) [13] or Renilla luciferase (pRL-SV40; Promega), and purified by ammonium-acetate isopropanol precipitation. Aliquots of PPP-RNA were then mock-treated, dephosphorylated with alkaline phosphatase (FastAP; Fermentas), or modified with different 5′ cap structures using the ScriptCap 2′-O-Methyltransferase and m7G Capping System (CellScript) according to the manufacturer's instructions. Briefly, 20-µg samples of RNA were heat-denatured at 65°C for 5 min, cooled on ice, then incubated with ScriptCap Buffer in the presence of 500 µM GTP, 100 µM SAM, 100 U 2′-O-methyltransferase (VP39), 10 U Vaccinia Capping Enzyme (VCE) and 40 U RNase inhibitor for 1 h at 37°C. Capped RNAs were further treated with FastAP to dephosphorylate any residual PPP-RNA, and then column-purified using the NucleoSpin RNA II kit (Macherey-Nagel). To add radioactively labelled methyl groups to in vitro transcribed RNA, 500 ng of each RNA was incubated with 100 U 2′-O-methyltransferase or 10 U of VCE in 0.5 µM S-adenosylmethionine and 1.4 µM S-[3H-methyl]-adenosylmethionine (78 Ci/mmol; Perkin-Elmer) for 1 h at 37°C. Reactions were purified on SigmaSpin Post-Reaction Clean-Up columns (Sigma) and eluates were mixed with 2 ml Ultima Gold scintillation fluid for measurement of 3H incorporation with a Packard Tri-Carb liquid scintillation counter (Perkin Elmer).
Capped m7Gppp-oligoribonucleotides matching the first 22 nucleotides of the 5′ untranslated region of Severe Acute Respiratory Syndrome Coronavirus HKU-39849 were prepared by adding N7-methylated cap structures to chemically synthesized RNA oligomers with a 3′-terminal C6 amino linker. A triphosphorylated RNA oligomer [PPP-r(AUAUUAGGUUUUUACCUACCC)-NH2) and a corresponding 2′O-ribose methylated RNA-oligomer [PPP-r(AmUAUUAGGUUUUUACCUACCC)-NH2] were ordered from ChemGenes Corporation (Wilmington, MA, USA) and capped as described above using the m7G Capping System (CellScript). Capped RNA oligomers were then HPLC-purified, biotinylated with biotin-N-hydroxysuccinimide ester (Epicentre) according to the manufacturer's instructions and again HPLC-purified. As control we used a corresponding 3′-terminal biotinylated and HPLC-purified oligoribonucleotide harbouring a 5′ hydroxyl group [OH-r(AUAUUAGGUUUUUACCUACCCU)-biotin].
For quantitative purification of RNA-binding proteins, streptavidin affinity resin was first incubated with 1-µg aliquots of biotin-labelled OH-RNA, PPP-RNA, CAP-RNA, CAP0-RNA or CAP1-RNA (all 7SK-antisense) in TAP buffer [50 mM Tris pH 7.5, 100 mM NaCl, 5% (v/v) glycerol, 0.2% (v/v) Nonidet-P40, 1.5 mM MgCl2 and protease inhibitor cocktail (EDTA-free, cOmplete; Roche)] in the presence of 40 U RNase inhibitor (Fermentas) for 60 min at 4°C on a rotary wheel. Control or RNA-coated beads were then incubated with 2-mg samples of HeLa cell lysate for 60 min, washed three times with TAP buffer, and twice with TAP buffer lacking Nonidet-P40 to remove residual detergent. Three independent affinity purifications were performed for each RNA. Bound proteins were dentatured by incubation in 6 M urea-2 M thiourea with 1 mM DTT (Sigma) for 30 min and alkylated with 5.5 mM iodoacetamide (Sigma) for 20 min. After digestion with 1 µg LysC (WAKO Chemicals USA) at room temperature for 3 h, the suspension was diluted in 50 mM ammonium bicarbonate buffer (pH 8). The beads were removed by filtration through 96-well multiscreen filter plates (Millipore, MSBVN1210), and the protein solution was digested with trypsin (Promega) overnight at room temperature. Peptides were purified on stage tips with three C18 Empore filter discs (3M) and analyzed by mass spectrometry as described previously [36]. Briefly, peptides were eluted from stage tips and separated on a C18 reversed-phase column (Reprosil-Pur 120 C18-AQ, 3 µM, 150×0.075 mm; Dr. Maisch) by applying a 5% to 30% acetonitrile gradient in 0.5% acetic acid at a flow rate of 250 nl/min over a period of 95 min, using an EASY-nanoLC system (Proxeon Biosystems). The nanoLC system was directly coupled to the electrospray ion source of an LTQ-Orbitrap XL mass spectrometer (Thermo Fisher Scientific) operated in a data dependent mode with a full scan in the Orbitrap cell at a resolution of 60,000 with concomitant isolation and fragmentation of the ten most abundant ions in the linear ion trap.
N-terminally Renilla luciferase-tagged proteins were transiently expressed in 293T cells. Three micrograms of each construct were transfected into 6×106 cells using 9.6 µl nanofectin (PAA Laboratories) in 10-cm dishes according to the manufacturer's instructions. After 24 h, the medium was removed, and cells were lysed in ice-cold TAP lysis buffer. An aliquot (10%) of the lysate was removed to determine input luciferase activity. The rest was added to streptavidin-agarose beads coated with 250 ng of RNA as described above, and incubated on a rotary wheel at 4°C for 60 min. Beads were washed three times and resuspended in 50 µl TAP buffer. Luciferase activities present in the suspension and in the input lysate were assayed in Renilla reaction buffer (100 mM K3PO4, 500 mM NaCl, 1 mM EDTA, 25 mM thiourea) containing 10 µM coelenterazine as substrate. The reactions were performed in triplicate and results were quantified using an Infinite 200 PRO series microplate reader (Tecan). For affinity purification of recombinant proteins with different RNAs, 50 to 250 ng of biotinylated RNA were coupled to streptavidin-agarose beads for 60 min at 4°C. Beads were washed three times with TAP buffer and incubated with recombinant His-tagged proteins for 60 min at 4°C. After three washes beads were boiled in Laemmli buffer for 10 min at 95°C and subjected to SDS-PAGE and Western Blot analysis.
Total RNA was isolated using the NucleoSpin RNA II kit (Macherey-Nagel), including on-column DNase digestion, and 200 to 500 ng of RNA was reverse transcribed with the RevertAid H Minus First Strand cDNA Synthesis Kit (Fermentas). RNA levels were then quantified by real-time RT-PCR using the QuantiTect SYBR Green RT-PCR kit (Qiagen) and a CFX96 Touch Real-Time PCR Detection System (BioRad). Each cycle consisted of 15 sec at 95°C, 30 sec at 50°C and 30 sec at 72°C, followed by melting curve analysis. Primer sequences were as follows: Renilla luciferase (5′-CGAAAGTTTATGATCCAGAAC-3′ and 5′-AATCATAATAATTAATAAATG-3′), hCycB (5′-CAGCAAGTTCCATCGTGTCATCAAGG-3′ and 5′-GGAAGCGCTCACCATAGATGCTC-3′), mTBP (5′-CCTTCACCAATGACTCCTATGAC-3′ and 5′- CAAGTTTACAGCCAAGATTCA-3′), mIFN-β (5′-ATGGTGGTCCGAGCAGAGAT-3′ and 5′-CCACCACTCATTCTGAGGCA-3′), MHV-N (5′-GCCTCGCCAAAAGAGGACT-3′ and 5′- GGGCCTCTCTTTCCAAAACAC-3′), 229E-N (5′-CAGTCAAATGGGCTGATGCA-3′ and 5′- AAAGGGCTATAAAGAGAATAAGGTATTCT-3′), mIfit1 (5′- CCATAGCGGAGGTGAATATC-3′ and 5′- GGCAGGACAATGTGCAAGAA-3′), mIfit1c (5′-AATCAGAAGAGGCAGCCATC-3′ and 5′-CATGGCTTCACTTGTGTTCC-3′), mIfit2 (5′-TCAGCACCTGCTTCATCCAA-3′ and 5′-CACCTTCGGTATGGCAACTT-3′), and mIfit3 (5′-GCTGCGAGGTCTTCAGACTT-3′ and 5′-TGGTCATGTGCCGTTACAGG-3′).
C57BL/6 mice were obtained from Charles River Laboratories (Sulzfeld, Germany), and Ifit1−/− mice have been described [16], [17]. Mice were maintained in individually ventilated cages and used at 6 to 9 weeks of age. All animal experiments were performed in accordance with Swiss federal legislation on animal protection and with the approval of the Animal Studies Committee of the Cantonal Veterinary Office (St. Gallen, Switzerland). Wild-type and Ifit1−/− mice (kindly provided by Michael Diamond) were injected intraperitoneally with 5,000 plaque-forming units of MHV. Virus titers in samples of spleens removed and frozen 48 h post infection were assessed by standard plaque assay on L929 cells. Bone marrow-derived macrophages or mouse embryo fibroblasts (1 to 5×105 cells) were treated or not with IFN-α and infected with the indicated viruses at a multiplicity of infection (MOI) of 5. For synchronised infection, cells were infected with virus on ice and pre-warmed DMEM growth medium was added 1 h later. To quantify the effects of siRNA-mediated knockdown of IFIT1, aliquots of 105 HeLa cells that had been transfected for 48 h with 15 pmol siRNA and 500 ng expression plasmid for human APN using the siRNA Prime reagent (PAA Laboratories) according to the manufacturer's instructions, were pretreated with IFN-α as indicated and infected with HCoV-229E at an MOI of 1 for 24 h.
For pulsed SILAC, mouse macrophages labelled with heavy isotopes (see above) were lysed in SDS lysis buffer (50 mM Tris pH 7.5, 4% sodium dodecyl sulfate). The lysate was then heated for 5 min at 95°C, sonicated for 15 min with a Bioruptor (Diagenode) and centrifuged for 5 min at 16,000× g at room temperature. Protein concentration was determined by Lowry assay (DC Protein Assay, BioRAD), and 50-µg aliquots were reduced with 10 mM DTT for 30 min, alkylated with 55 mM IAA for 20 min at room temperature, and precipitated with 80% acetone for 3 h at −20°C. After centrifugation for 15 min at 16,000× g at 4°C, pellets were washed with 80% acetone, dried for 30 min at room temperature and dissolved in 6 M urea-2 M thiourea. Proteins were digested with LysC and trypsin at room temperature and peptides were purified on stage tips and analysed by LC-MS/MS using a Easy nano LC system coupled to a Q Exactive mass spectrometer (Thermo Fisher Scientific). Peptide separation was achieved on a C18-reversed phase column (Reprosil-Pur 120 C18-AQ, 1.9 µM, 200×0.075 mm; Dr. Maisch) using a 95-min linear gradient of 2 to 30% acetonitrile in 0.1% formic acid. The mass spectrometer was set up to run a Top10 method, with a full scan followed by isolation, HCD fragmentation and detection of the ten most abundant ions per scan in the Orbitrap cell.
Raw mass-spectrometry data were processed with MaxQuant software versions 1.2.7.4 and version 1.3.0.5 [37] using the built-in Andromeda search engine to search against human and mouse proteomes (UniprotKB, release 2012_01) containing forward and reverse sequences, and the label-free quantitation algorithm as described previously [36], [38]. In MaxQuant, carbamidomethylation was set as fixed and methionine oxidation and N-acetylation as variable modifications, using an initial mass tolerance of 6 ppm for the precursor ion and 0.5 Da for the fragment ions. For SILAC samples, multiplicity was set to 2 and Arg10 and Lys8 were set as heavy label parameters. Search results were filtered with a false discovery rate (FDR) of 0.01 for peptide and protein identifications. Protein tables were filtered to eliminate the identifications from the reverse database and common contaminants.
In analyzing mass spectrometry data from RNA affinity purifications, only proteins identified on the basis of at least two peptides and a minimum of three quantitation events in at least one experimental group were considered. Label-free quantitation (LFQ) protein intensity values were log-transformed and missing values filled by imputation with random numbers drawn from a normal distribution, whose mean and standard deviation were chosen to best simulate low abundance values. Significant interactors of RNAs with different 5′ end structures were determined by multiple equal variance t-tests with permutation-based false discovery rate statistics [39]. We performed 250 permutations and the FDR threshold was set between 0.02 and 0.1. The parameter S0 was empirically set between 0.2 and 1, to separate background from specifically enriched interactors.
For data analysis from pulsed SILAC experiments, we used log-transformed heavy to light protein ratios. Only proteins with valid values were considered for analysis, and normalized by dividing by the row median. Profile plots were generated using LFQ intensities of log-transformed heavy-labelled protein intensities. We excluded proteins containing less than 10 valid values in all 12 measurements, and missing values were filled by imputation. LFQ intensities were then normalized by dividing by the row median.
Results were plotted using R (www.R-project.org) and GraphPad Prism version 5.02. Multiple sequence alignments were generated with ClustalW (http://www.ebi.ac.uk/Tools/msa/clustalw2/).
A homology model of human IFIT1 was obtained with MODELLER [40] using the X-ray structure of human IFIT5 (4HOQ) as a structural template [20]. A pairwise sequence alignment was generated with ClustalW (http://www.ebi.ac.uk/Tools/msa/clustalw2/) and further refined with MODELLERs align2d. Human IFIT1 and IFIT5 share approximately 75.6% sequence similarity, with 57.2% of all residues being identical. Cavity volumes in both structures were calculated in a two-step process with the rolling probe method using 3V [41]. First, a solvent-excluded volume was calculated for each structure using a probe radius of 1.5 Å (corresponding to water). A larger probe size of 5 Å was used to calculate so-called “shell volumes”. The solvent-accessible cavity volumes were obtained by subtraction of each solvent-excluded volume from the corresponding shell volume.
|
10.1371/journal.pntd.0004618 | Prevalence of Lymphatic Filariasis and Treatment Effectiveness of Albendazole/ Ivermectin in Individuals with HIV Co-infection in Southwest-Tanzania | Annual mass treatment with ivermectin and albendazole is used to treat lymphatic filariasis in many African countries, including Tanzania. In areas where both diseases occur, it is unclear whether HIV co-infection reduces treatment success.
In a general population study in Southwest Tanzania, individuals were tested for HIV and circulating filarial antigen, an indicator of Wuchereria bancrofti adult worm burden, before the first and after 2 consecutive rounds of anti-filarial mass drug administration.
Testing of 2104 individuals aged 0–94 years before anti-filarial treatment revealed a prevalence of 24.8% for lymphatic filariasis and an HIV-prevalence of 8.9%. Lymphatic filariasis was rare in children, but prevalence increased in individuals above 10 years, whereas a strong increase in HIV was only seen above 18 years of age. The prevalence of lymphatic filariasis in adults above 18 years was 42.6% and 41.7% (p = 0.834) in HIV-negatives and–positives, respectively. Similarly, the HIV prevalence in the lymphatic filariasis infected (16.6%) and uninfected adult population (17.1%) was nearly the same. Of the above 2104 individuals 798 were re-tested after 2 rounds of antifilarial treatment. A significant reduction in the prevalence of circulating filarial antigen from 21.6% to 19.7% was found after treatment (relative drop of 8.8%, McNemar´s exact p = 0.036). Furthermore, the post-treatment reduction of CFA positivity was (non-significantly) larger in HIV-positives than in HIV-negatives (univariable linear regression p = 0.154).
In an area with a high prevalence for both diseases, no difference was found between HIV-infected and uninfected individuals regarding the initial prevalence of lymphatic filariasis. A moderate but significant reduction in lymphatic filariasis prevalence and worm burden was demonstrated after two rounds of treatment with albendazole and ivermectin. Treatment effects were more pronounced in the HIV co-infected subgroup, indicating that the effectiveness of antifilarial treatment was not reduced by concomitant HIV-infection. Studies with longer follow-up time could validate the observed differences in treatment effectiveness.
| Parasite infections and HIV show large geographical overlap in sub-Saharan Africa and could hence potentially interact in co-infected individuals. In a general-population study conducted in Southwest Tanzania, we found high prevalence of both, lymphatic filariasis and HIV, with 42.5% of the adult population infected with Wuchereria bancrofti and 16.8% infected with HIV. Seven percent of the adults were infected with both pathogens. When adjusting for age, there was no statistically significant difference in initial prevalence or worm burden between HIV-positive and negative participants. For 798 individuals test results for both diseases were available in 2009, before and in 2011, after 2 rounds of treatment against lymphatic filariasis. Between 2009 and 2011, a significant drop of prevalence and worm burden in infected individuals were observed, which was more pronounced in the HIV co-infected subgroup. Hence, HIV co-infection does not seem to negatively affect lymphatic filariasis treatment programmes.
| Lymphatic Filariasis (LF) is a mosquito-borne disease caused either by Wuchereria bancrofti which is distributed throughout the tropics, or Brugia malayi and Brugia timori, both limited to Southeast-Asia. It is estimated that 120 million people world-wide are infected with one of these pathogens, and 1 billion are at risk to acquire LF during their lifetime [1]. Before larger treatment programmes started, LF was present in most of the 21 regions of Tanzania with up to 63.8% of individuals testing positive for circulating filarial antigen, a marker for LF infection [2]. Since the year 2000 the “Global Alliance to Eliminate Lymphatic Filariasis” uses annual mass drug administration (MDA), with the aim to control and ultimately eliminate the disease [3]. The campaign of the Tanzanian National Lymphatic Filariasis Elimination Programme (NLEFP) commenced in 2001 in the coastal regions of Tanzania. In the Mbeya district in Southwest-Tanzania the treatment programme started in October 2009 with the annual distribution of albendazole (400mg) and ivermectin (150–200μg/kg). Ivermectin is considered to be mainly microfilaricidal [4], for albendazole an effect on the release of intrauterine antigen components of the adult worm was described [5]. Some studies report on the treatment effectiveness of the combination of albendazole and ivermectin after 12-month: in Ghana a significant reduction in circulating filarial antigen (CFA) levels but no measurable reduction of CFA prevalence was described in 370 individuals receiving both drugs [6, 7]. A longitudinal study from Northern Tanzania showed only small reductions of CFA positivity after two annual drug distributions (from 53.3% to 51.4%), but a significant drop to 44.9% and 19.6% after four and seven years of treatment, respectively [8].
In South Western Tanzania, both LF and HIV are public health concerns. The HIV prevalence in the country has been documented in several national surveys [1, 9, 10]. The third population based Tanzanian HIV/AIDS and Malaria Indicator Survey in 20011/2012 (THMIS) revealed a country-wide HIV prevalence of 5.1% in Tanzanian adults between the age of 15 and 49 years, and a prevalence of 9.0% for this age-group in Mbeya Region [10]. Large scale distribution of antiretroviral (ART) drugs was initiated in Tanzania in 2005. At the time of our study, ART was not widely available in Southwest Tanzania. [10–12].
Local differences in initial prevalence, coverage of treatment programs, co-infection with other pathogens, etc. can all affect treatment success, thus careful surveillance of the programs is necessary to control the infection. [1, 13–17]. Only few manuscripts focus specifically on the possible interaction of HIV with LF and most of these use cross-sectional data [18–21]. Only one recently published study investigates the treatment effectiveness of MDA drugs in HIV/LF co-infected individuals [22], but focusses on changes in CD4 and HIV viral load after antifilarial treatment in selected HIV-positive individuals. No study concentrated on the antifilarial treatment effectiveness of MDA drugs in HIV/LF co-infected individuals.
Our study assesses LF prevalence in the Mbeya Region, before and after the governmental eradication program reached the area and examines the potential impact of HIV co-infection on LF treatment.
Data were collected during the SOLF cohort-study (Surveillance of Lymphatic Filariasis, http://www.mmrp.org/projects/basic-research/solf.html) in the Kyela district/Mbeya region in Southwest Tanzania which was conducted at the National Institute for Medical Research (NIMR)—Mbeya Medical Research Centre (MMRC) between 2009 and 2011. The study was embedded into the population based EMINI (Evaluation and Monitoring of the Impact of New Interventions, http://www.mmrp.org/projects/cohort-studies/emini.html) cohort study, which was carried out in 9 selected communities in the Mbeya region (Fig 1) from 2006 to 2011. More than 170,000 inhabitants from ~42,000 households of these communities were registered and 10% of households randomly selected to participate in the study. No additional households entered the surveillance, but some new participants entered through birth or marriage into included household.
The SOLF study was approved by the Mbeya Medical Research and Ethics Committee and the Tanzanian National Institute for Medical Research—Medical Research Coordinating Committee as an amendment to the EMINI cohort study. Prior to enrolment, each EMINI participant had provided written informed consent regarding study participation. Parents consented for their children below 18 years of age. In addition, children above the age of 12 years signed their own assent form.
Data and samples from participants in the Kyela site of the EMINI study were collected annually from 2007 until 2009. During the last two surveys (2010 and 2011) only half of the study households were visited in each year. During each visit, which took place between 8 am and 2 pm, blood, urine and stool samples were collected from each participant. Samples from 2,165 participants from March 2009 were used to estimate the prevalence of LF directly before the government treatment program commenced in Kyela in October 2009. In March 2011, 18 month after the first and 6 month after the second delivery of antifilarial treatment, samples from 1,010 participants were used to evaluate treatment impact.
From each study participant, 2.7 ml of blood was collected during morning hours in EDTA tubes and immediately stored at 4°C. Cells and plasma were separated within 24 hours and subsequently stored at -80°C. All laboratory tests were performed at NIMR-MMRC, Mbeya Tanzania.
HIV testing was performed using the SD-Bioline HIV-1/2 3.0 (Standard Diagnostics, Kyonggi-do, South Korea) rapid diagnostic test (RDT). Negative RDT results from one survey, followed by another negative RDT result in a subsequent survey, were regarded as confirmed negative and not further tested. All positive results were confirmed using an ELISA HIV test (Enzygnost Anti HIV 1/2 Plus, DADE-Behring, Marburg, Germany), and tested by Western blot (MPD HIV Blot 2.2, MP Biomedicals, Geneva, Switzerland) if discordant. For all HIV incident cases, the negative result of the previous year, as well as the new positive results was confirmed by the testing algorithm described above. For children below the age of two years, HIV testing was done by PCR. Further details are described elsewhere [23]. Because confidential disclosure of the HIV-status could not be ensured during household visits, we did not inform participants about their HIV status. Instead they were offered voluntary counseling and testing by an independent team, which was travelling with our study team, who provided referral to the local care and treatment center, to everyone who was tested positive.
A commercially available ELISA (TropBio Og4C3 serum ELISA, Townsville, Australia) was used to detect circulating filarial antigen (CFA) using 100 μl of the collected sera. The Og4C3 antibody detects Wuchereria bancrofti antigen with high specificity (98.5%) and no known cross-reaction to Onchocerca volvulus, Brugia spp., Mansonella, Dracunculus medinensis, Ascaris lumbricoides or Strongyloides stercoralis [24]. Sensitivity varies between 73% [25] and 100% [26], but was found 97.9% in individuals carrying microfilariae [24]. CFA is secreted by fully developed W. bancrofti adults and can be found at similar levels during day and night. Antigen levels thus reflect the W. bancrofti worm burden. The measurement of CFA with the Trop Bio ELISA is semi-quantitative; seven control tubes with standardized amounts of antigen are supplied and allow an estimation of the filarial antigen levels in the analysed plasma according to the measured optical density (OD). LF test results were considered negative, indeterminate or positive if the OD was <0.2, ≥0.2 and ≤0.3, or >0.3 respectively.
Statistical analyses were performed using Stata statistics software (version 14; Stata Corp., College Station, TX). Pearson´s chi-squared test was used to compare binominal outcomes between groups and to compare CFA positivity before and after treatment in all participants. McNemar´s exact test for paired data was used to compare CFA positivity before and after treatment in those individuals who participated in both surveys. The non-parametric Wilcoxon rank sum test was used to compare selected baseline characteristics of continuous variables, since none of these was normally distributed. In order to examine the association of LF infection with HIV status and other potentially important covariates we performed uni- and multi-variable log link binomial regression analyses with robust variance estimates.
In March 2009, before the first national MDA commenced, valid CFA results were obtained from 2,104 individuals (Table 1). Indeterminate results were found for the 61 of the tested 2,165 samples (2.8%). Their median age was 16.6 years (range 0–94, IQR: 8.8 to 34), and 51.0% were female. Only 4 (1.6%) of the 245 children below the age of 5 years were CFA-positive; LF prevalence started to rise in participants above 10 years and was 42.3% in adults above 18 years of age (Fig 2). When including all age groups, 24.8% of the study population were CFA-positive with a trend to higher prevalence in males (26.5%) than in females (23.1%, chi-squared p = 0.074). In the adult population above 18 years the difference in CFA-positivity between males (47.3%) and females (38.0%) was significant (chi-squared p = 0.003).
In March 2011, 18 months after the first MDA and six months after the second, ~50% of the initially included households were revisited for interviews and blood sample collection. Some scheduled participants were not found in 2011, and some new individuals had entered the visited households (see study population and design). In addition to an analysis where the data of all participants form each Survey (= open cohort) are evaluated, which reflects more a cross sectional design, a second analysis included only the 798 individuals who actively participated in both years of the surveillance longitudinally (= closed cohort). The numbers of participants is shown in Table 1. Of the 974 valid test results in 2011, 19.7% were CFA-positive, leading to a calculated prevalence reduction of 5.1% (24.8% vs. 19.7%, chi-squared p = 0.002) when including all subjects who participated in at least one survey (Table 1, open cohort). In the analysis of samples from 798 individuals who actively participated in both surveys (Table 1, closed cohort), a lower prevalence reduction (21.6 to 19.7%, McNemar´s exact p = 0.036) was measured (Fig 3).
At baseline the overall HIV prevalence in our study cohort was 8.9%, with a prevalence of only 2.1% in children and adolescents below the age of 18 years, and a prevalence of 16.9% in individuals ≥18 years of age (Fig 3).
HIV-infection was more prevalent in female (10.7%), compared to male participants (7.1%, chi-squared p = 0.003). Sixty-eight of the 968 adult individuals (7.0%) were infected with both pathogens and among the whole group of 2,104 individuals 69 co-infections (= 3.3%) were observed. The initial univariable analysis of the potential association of HIV with LF infection showed a higher prevalence of LF in HIV-positive (36.9%); compared to HIV-negative individuals (23.6%) (RR = 1.56, 95% CI = 1.26 to 1.94, p<0.001). But we already demonstrated that HIV and LF are both less common in children than in adults, which confounds this association. To further study the pattern of co-infection we analysed CFA positivity in HIV infected and uninfected individuals stratified by age (Fig 4); in adults (> = 18 years) only; and in log-link binomial multivariable regression adjusted for age and gender. None of these analyses showed a significant association of LF infection with HIV, neither within the single age strata nor overall in the multivariable regression model where the influence of age and gender were confirmed, but where the adjusted RR for HIV was only 1.04 (Table 2). When only analysing data from adults above 18, the CFA prevalence was 42.6% in the HIV-negative and 41.7% in the HIV-positive subgroup (univariable log-link regression RR = 0.98, 95% CI = 0.80 to 1.20; p = 0.84).
In order to compare antifilarial treatment success in the HIV-negative and positive subgroups we again performed two analysis: one for all tested individuals who participated in at least one survey (open cohort using chi-squared testing), and one only for the individuals who participated in both surveys before and after treatment (closed cohort, using McNemar´s exact test).
For the open cohort a CFA prevalence reduction from 23.6% to 18.9% (chi-squared p = 0.015, relative drop = 19.7%) was found in HIV-negative participants, and from 36.9% to 27.5% (chi-squared p = 0.023, relative drop = 25.4%) in HIV-positives. For the closed cohort we observed a drop in CFA positivity from 20.9% to 19.4% (McNemar´s exact p = 0.117, relative drop = 7.3%) in 723 HIV-negatives and from 27.5% to 21.7% (McNemar´s exact p = 0.125, relative drop = 21.1%) in 69 HIV-positive participants. The reason for this pronounced difference (7.3% vs. 21.1%) is a higher incidence of CFA positivity in the HIV negative participants where 15 (2.6%) of the 572 initially CFA negative participants turned CFA-positive, whereas none of the 50 HIV-positive participants who were initially CFA-negative turned CFA-positive (chi-squared p = 0.246). The proportion of initially CFA positives who turned CFA negative was very similar in HIV-negative (26/151 = 17.2%) and HIV-positive participants (4/19 = 21.1%, chi-squared p = 0.679). When combining this information about change in LF status in one outcome variable (-1 = turned CFA negative; 0 = no change in CFA status; 1 = turned CFA positive) univariable linear regression modelling resulted in a coefficient β for the HIV infected subgroup of -0.043 (95%CI = -0.102 to 0.016, p = 0.154).
Analysing the prevalence reduction in the closed cohort for adults > = 18 years only, a drop from 42.7% to 40.3% (McNemar´s exact p = 0.248, relative drop = 5.6%) was noted for HIV-negatives, and from 32.7% to 25.5% (McNemar´s exact p = 0.125, relative drop = 22.0%) in the HIV-positive subgroup. Summarizing our results, we found more pronounced drops in prevalence among the HIV positive subgroup, compared with the HIV negative, no matter, whether all participants or only adults are analysed and also with both possible ways of evaluating the data (open cohort or closed cohort).
The measurement of CFA with the Trop Bio ELISA is semi-quantitative; with the OD of the plasma samples reflecting the participant’s worm burden. Our findings for CFA intensities parallel those for CFA prevalence: geometric mean intensities before treatment were relatively similar between HIV-positives (157 units) and HIV-negatives (179 units, Wilcoxon rank sum p = 0.34), which is also true for the relative reduction of geometric mean intensity after treatment, which was 26% and 30% respectively (Wilcoxon rank sum p = 0.50)
In our study we found a significant decrease in LF-prevalence after only 2 years of MDA in an area with high HIV co-infection in South-West Tanzania. This was in contrast to our expectations and previously published manuscripts [2, 6, 7, 16, 27]. Furthermore we did not find any evidence that HIV co-infection impairs the effectiveness of antifilarial treatment. On the contrary, our data show a more pronounced decrease in prevalence and CFA intensity among HIV-positive compared to HIV-negative participants. We tried to consider several factors which could have affected our analysis. The age distribution of LF and HIV infection had to be taken into account, but also the composition of the study population and potential changes during follow-up. However, an almost three-fold relative drop in LF prevalence was seen for HIV-positive compared to HIV-negative participants in the most stringent analysis which only considered individuals for whom we have data both before and after treatment. We do not want to overstate this result since the overall numbers of participants with HIV/LF-co-infection was low, despite an initially large cohort and accordingly the differences in cure and incidence rate between HIV-infected and uninfected participants are not significant. Moreover, we are unable to present an explanation for this finding. Reports about the LF prevalence in HIV-negative and HIV-positive individuals have been rare and conflicting in the past. Nielsen et al. described a positive association of HIV and LF infection after adjusting for age and sex in a cross-sectional study from Northern Tanzania [20], even though a further evaluation of this group of individuals did not support an association between HIV and LF [19]. No difference regarding CFA levels was found in other cross sectional studies from India [18] and Malawi [28]. The latter findings are supported by the cross sectional analysis of our study participants in 2009. To date, no other longitudinal study has compared the effectiveness of antifilarial treatment in HIV-positive and negative subgroups.
One interesting finding is the higher pre-treatment CFA prevalence in the open compared to the closed cohort which demonstrates that LF-positive individuals were more likely to be lost to follow-up than LF-negative participants. This can at least partly be explained by the higher prevalence of LF in adults, who are more likely than younger participants to relocate (e.g. in search of a job), to harbour diseases that prevent them from further study participation (e.g. HIV-infection) and to die.
Most other studies analyse the treatment effectiveness after 6, 8 or even more years of treatment. Thus one limitation of our study is the short duration of follow-up, which is a consequence of funding restrictions and does not allow for a conclusive analysis of MDA effectiveness. Furthermore we are not able to identify individually who of our participants was treated and who was not: the local district medical officer and Neglected Tropical Diseases (NTD) coordinator, who supervised the drug distribution in Kyela district, reported coverages of 60.8% in October 2009 and 68.2% in 2010 (Mrs. Masawe, personal communication). However, very few participants of the SOLF study were aware of the treatment program against LF, when asked about their participation. Therefore, firm assumptions about efficacy, i.e. the effect of antifilarial treatment on CFA prevalence and intensity under ideal conditions cannot be made. Instead our data are better suited to describe the effectiveness of MDA under real-life conditions. Furthermore, information on CD4 count and antiretroviral treatment status of HIV infected individuals would have helped to refine our analysis, but this information was not collected.
The Og4C3 antibody is supposed to specifically recognise Wuchereria bancrofti antigen with no relevant cross-reaction to Onchocerca volvulus, Brugia spp., Mansonella, Dracunculus medinensis, Ascaris lumbricoides or Strongyloides stercoralis. In spite of this, cross reactivity with Loa microfilariae has been found in another test (Binax Now Filariasis Immunochromatographic Test, Alere, Scarborough, ME, USA), which detects the same antigen as the Trop Bio ELISA [29]. However, there is no significant reported disease burden of Loiasis in our study area [30].
In an area with high prevalence of and no previous treatment against LF we investigated the potential association of HIV and LF infection. When adjusting for age we found similar CFA prevalence and intensities in HIV-positive and negative participants. After two rounds of treatment a significant reduction in CFA prevalence and intensity was demonstrated, which was more pronounced in the HIV-positive compared to HIV-negative participants. Hence, HIV co-infection does not seem to negatively affect antifilarial treatment.
|
10.1371/journal.pntd.0002805 | Statistical Modeling Reveals the Effect of Absolute Humidity on Dengue in Singapore | Weather factors are widely studied for their effects on indicating dengue incidence trends. However, these studies have been limited due to the complex epidemiology of dengue, which involves dynamic interplay of multiple factors such as herd immunity within a population, distinct serotypes of the virus, environmental factors and intervention programs. In this study, we investigate the impact of weather factors on dengue in Singapore, considering the disease epidemiology and profile of virus serotypes. A Poisson regression combined with Distributed Lag Non-linear Model (DLNM) was used to evaluate and compare the impact of weekly Absolute Humidity (AH) and other weather factors (mean temperature, minimum temperature, maximum temperature, rainfall, relative humidity and wind speed) on dengue incidence from 2001 to 2009. The same analysis was also performed on three sub-periods, defined by predominant circulating serotypes. The performance of DLNM regression models were then evaluated through the Akaike's Information Criterion. From the correlation and DLNM regression modeling analyses of the studied period, AH was found to be a better predictor for modeling dengue incidence than the other unique weather variables. Whilst mean temperature (MeanT) also showed significant correlation with dengue incidence, the relationship between AH or MeanT and dengue incidence, however, varied in the three sub-periods. Our results showed that AH had a more stable impact on dengue incidence than temperature when virological factors were taken into consideration. AH appeared to be the most consistent factor in modeling dengue incidence in Singapore. Considering the changes in dominant serotypes, the improvements in vector control programs and the inconsistent weather patterns observed in the sub-periods, the impact of weather on dengue is modulated by these other factors. Future studies on the impact of climate change on dengue need to take all the other contributing factors into consideration in order to make meaningful public policy recommendations.
| As dengue virus transmission is through a human-to-mosquito-to-human cycle, the influence of meteorological factors on dengue is likely to be associated with their impact on mosquito populations and behavior. Other than the influence of weather factors, the shift of dominant serotypes and pre-emptive measures taken against dengue vectors may possibly affect the dengue transmission trend. In this study, we investigate the impact of weather factors on dengue in tropical Singapore, taking into consideration the disease epidemiology and profile of virus serotypes. We found that absolute humidity, as a composite index of mean temperature and relative humidity, is a more stable and better predictor for modeling dengue incidence than the other unique weather variables when virological factors are taken into consideration. This research suggests that absolute humidity needs to be considered together with all the other contributing factors in order to make meaningful public policy recommendations for dengue control.
| Dengue fever (DF) is the most common vector-borne viral disease in humans and is distributed worldwide, mainly in tropical and subtropical countries. In recent decades, dengue has been expanding globally possibly due to climate change [1] and highly intra and extra-country connectivity through traffic, commerce, and migration [2]. DF is caused by one of four distinct dengue virus serotypes (DEN 1–4). This viral infection has resulted in an estimated 50 million to 100 million annual cases of DF worldwide, with about 500,000 of these cases developing into life-threatening Dengue hemorrhagic fever (DHF)/Dengue shock syndrome (DSS) [2], .
In Singapore, which is a tropical island city state, DF is endemic, with year-round transmission observed. The integrated vector control program, implemented by the government, that started in the late 1960s resulted in a prolonged period of low dengue incidence [5]. The key strategy for dengue control in Singapore is to tackle the root of the problem, which is to deny Aedes mosquitoes the place to breed, i.e., source reduction [6], [7]. With a multi-pronged approach [6], [7], Singapore had adopted: 1) preventive surveillance and control, in which daily mosquito surveillance operations are conducted with the aid of the Geographical Information System; 2) public education and community involvement through working with construction sites, schools and community councils; 3) enforcement for carrying out intensive search and destroy operations at outdoor as well as indoor areas under legal laws upon notification of a dengue cluster; and 4) research for combating dengue disease including polymerase chain reaction, rapid antigen test kits, sequencing and bioinformatics, etc.
In addition to the preventive surveillance approaches, general practitioners and hospitals in Singapore are obliged to report probable dengue cases to the Ministry of Health and all reported dengue cases of DF/DHF are then confirmed by one or more laboratory tests including anti-dengue IgM antibody, enzyme linked immunosorbent assay (ELISA), and polymerase chain reactions (PCR). To our knowledge, there was no change in the notification process during the period studied in this work.
In Singapore, more than 80% of notified dengue cases were hospitalized [8]. Although under intensive dengue surveillance, we still experienced dengue hyperendemic in 2005 and in 2013 [9], with the number of laboratory confirmed cases reaching 14,209 cases (with 27 deaths) and 22101 cases (with 7 deaths) respectively. The re-emergence of hyperendemic may be due to low herd immunity, shift of dominant serotypes, high subclinical dengue infection and weather conditions etc. In an earlier report based on Singapore dengue data [10], it is estimated that only 1 out of 23 dengue cases are diagnosed and notified, which indicates a substantially high unreported dengue rate, i.e., a majority of dengue cases is either asymptomatic or subclinical but they are able to transmit dengue viruses to uninfected mosquitoes to trigger further infections. Other than the high subclinical cases possibly causing the dengue transmission to worsen, the tropical weather condition favors the year-round presence of Aedes mosquitoes, which is key in the dengue-human transmission chain. Thus, a better understanding on the association between weather and dengue incidence is important for a more proactive surveillance strategy of dengue control.
The impact of weather on dengue incidence has been widely studied [11], [12], [13], [14], [15], [16], [17], [18] as it is relatively easy to obtain basic meteorological data in dengue affected countries. Earlier studies have found many specific relationships between weather factors and dengue incidence. For example, the seasonality of dengue is well established for Thailand [19], [20], [21] and Vietnam [22], where dengue epidemic coincides with the rainy season. Malaysia also reported a strong seasonal pattern but its correlation to weather appears to be more complicated [23]. The number of dengue cases in Malaysia appears to be positively correlated with two to three month lag to the heavy rain in the first wet season of the year. For specific weather variables in Singapore, mean temperature and relative humidity were found to be the most important weather factors upon comparing models which considered long-term climate variability and linear lag effects of weather variables including temperature, humidity and rainfall [24]. In another study from Brazil [25], maximum temperature and minimum temperature were found to be the best predictors for the increased number of dengue cases.
In Singapore, a model consisting of lag effects of mean temperature and rainfall was built and applied to forecast the number of dengue cases over a 16 week period [15], [26], [27]. Mean temperature and relative humidity at a lag of 2 weeks and Niño Southern Oscillation Index at a lag of 5 weeks were found to have significant impact on dengue [24]. However, the effect of absolute humidity on dengue incidence, which reflects the combined impact of temperature and relative humidity, has not been well described. In addition to weather, the impact of the dynamics of circulation of dengue virus serotypes on dengue epidemiology has been well documented [28]. Infection with one serotype confers life-long immunity to that particular serotype [29], [30]. Some studies have also reported a time-lagged correlation between dengue virus serotype dynamics and disease incidence rates [31]. The variation of dominant serotypes needs to be taken into account in studies of environmental factors on dengue incidence.
In this study, we modeled and compared the effect of absolute humidity with the effect of temperatures (maximum, minimum, mean), relative humidity, rainfall and wind speed on dengue in Singapore from 2001 to 2009. The model used is a distributed lag non-linear model, i.e., an over-dispersed Poisson model with regressions on autocorrelation, lagged effect of weather factors, population sizes and dengue trends. The model is further refined by comparing the impact of weather variables in sub-periods divided based on the dominant circulating dengue serotypes. The model selection criterion applied in this study is the Quasi Akaike's Information Criterion.
Singapore is a tropical island city state with approximately 710.2 km2 land area. The average size of the total population over the years, from 2001 to 2009, is about 4.41 million (Department of Statistics, 2013). The mean temperature ranges from 25.2°C to 30.3°C, with the maximum daily temperature and maximum daily rainfall reaching up to 34.5°C and 479.7 mm respectively.
A vector control program in the 1960s to 1980s had successfully prevented dengue outbreaks for two decades since 1973, with less than 1,000 reported cases per year [5]. However, since 1989, Singapore has observed increased notifications of dengue infection despite a low Aedes house index of less than 1%. The factors contributing to the re-emergence includes an increase in human population and density, increases in cross border and in country travel and low herd immunity, resulting from low transmission in the prior decade [5]. The most recent large outbreaks occurred in 2005 [32] and 2013 raise more concern on dengue spread in Singapore.
Weekly notified DF/DHF cases in Singapore from 2001–2009 were retrieved from the Weekly Infectious Diseases Bulletin [9] of the Singapore Ministry of Health. The human population data used was based on the mid-year Singapore total population data obtained from the Singapore Department of Statistics [33].
Whilst all four dengue serotypes have mostly been detected in Singapore, typically there is one predominant circulating serotype, with switches in predominance associated with the outbreaks (Table 1). The dominant serotype was defined as one that causes more than 50% of cases sampled. The estimated proportion of each viral serotype was obtained from the Singapore Communicable Diseases Surveillance reports [34] of Singapore Ministry of Health. DEN-2 was the dominant circulating serotype in the years 2001–2003, DEN-1 in 2004–2006 and DEN-2 in 2007 to 2009.
Weather data including Mean temperature (MeanT, °C), Minimum temperature (MinT, °C), Maximum temperature (MaxT, °C), Rainfall (Rain, mm), Relative humidity (RH, %) and Wind speed (WindS, m/s) were obtained from the National Environment Agency, Singapore. Absolute humidity (AH, g/m3), which is the mass of water in a unit volume of air, was estimated through dry bulb temperature and relative humidity using the approximated equation, assuming standard atmospheric pressure [35]:(1)where Tc is the dry bulb temperature (in our studies, Tc is the daily mean temperature), andwhere Td is the dew point temperature. Td is approximated from the equation below, based on dry bulb temperature and relative humidity:where , and . Weekly weather data were calculated by averaging the daily weather values over each week. The relationship between AH, Tc and RH is presented in Figure 1.
Spearman rank correlation tests were then applied to assess the association between weekly dengue cases and weather factors for a range of time lags – from 0 to 20 weeks, over the whole study period (from 2001 to 2009) (see Figure 2). As the number of dengue incidence is a Poisson count data, it is thus not feasible to check how it is linearly related to weather factors. As such, Spearman rank correlation is usually chosen as it is designed to assess how well two variables are monotonically related even if their relationship is not linear [36]. As autocorrelation was detected in each time series, it would not be appropriate to calculate p-values of the correlation coefficients by traditional methods. Therefore, the p-values were calculated through Adaptive Wavelet-Based Bootstrapping [37] with a sample size of 5000. This was implemented in R software (version 3.0.2; package ‘wmtsa’). In this study, the p-value of the correlation coefficients between every two time series was calculated using this method.
Furthermore, the associations between each weather predictor and the risk of dengue were modeled. The number of observed dengue cases, , at week , was assumed to follow an over dispersed Poisson distribution [38] with mean . The effect of weather variable on was described by a Distributed Lag Non-linear Model (DLNM) [39], [40] given as follows:(2)where is the intercept, and are coefficients of the auto-regression terms, is a function to denote smoothed relationships between and a single weather factor (i.e., MinT, MeanT, MaxT, Wdsp, Rainf, RH or AH) with a maximum lag number of , which enables to include the lag effect of predictors into the model. The nonlinear effect of weather factor was described by a natural cubic spline (ns) smoothing function with degrees of freedom (df) and knots at equally spaced quantiles, while the lag effect of was described by an ns smoothing function with df of . is the corresponding coefficients vector. is an ns smoothing function with df of 1 per year applied to fit the long-term trend of dengue incidence. Here, the df, = 9 and is the corresponding coefficients vector. is the mid-year population size of Singapore and is the offset term. Besides the DLNM, the single lag effect of each weather factor was also investigated. When considering the effect of weather factor at lag , was replaced by in Eq. 1 with being the lag number, and being the coefficients vector, i.e., the effect of was modeled by an ns function with df of .
In order to reflect the goodness-of-fit, Quasi Akaike's Information Criterion (QAIC) was used with a smaller QAIC implying a better fit [40], [41]. QAIC is given by(3)where L is the log-likelihood of the fitted model with parameters (in Eq.2, ) and (i.e., the estimated overdispersion parameter), whereas k is the number of parameters. In (Eq.2), was selected from 0 to 20 weeks [15]. The df () of each was selected from 1 to 5, while the df () of lag was selected from 1 to 3. Higher df implies higher flexibility, but may introduce over-fitting. The selection criterion was QAIC and model flexibility. For the space of each weather variable, QAIC indicated = 4 or 5 for all weather variables; whilst for the lag dimension, QAIC indicated = 2 or 3. In this article, we adopted = 4 and = 3. The analyses were performed in R software (version 2.13.2; package ‘dlnm’; R Development Core Team, 2011) [42]. We first investigated the maximum lag considering the overall effect of each weather variable on dengue incidence for the whole period. Once the best model was established based on the smallest QAIC, the model was further studied and evaluated for both the entire studied period and the three distinct sub-periods based on the predominant circulating serotypes.
We found that Absolute humidity (AH) was positively correlated with Relative humidity (RH) and Temperature (see (Eq. 1 and Figure 1)). The correlation coefficient between AH and RH is 0.21, whilst the correlation between AH and mean temperature is 0.54. A higher RH or a higher temperature was associated with a higher AH. However, the correlation between MeanT and RH was negative (the correlation coefficient is −0.71). Therefore, as a composite index of MeanT and RH, the impact of AH on dengue incidence was studied further.
The Spearman rank correlation analysis, using time lagged weather data (0–20 weeks), showed that temperature (MeanT, MaxT, MinT), absolute humidity and rainfall exhibited significant association with dengue incidence. On the other hand, no significant relationship was observed between dengue and wind speed, and relative humidity. The correlation between AH and dengue incidence was the highest (its correlation coefficient was 0.234 with p-value<0.05 at a 7-week lag) among all the studied weather variables (see Figure 2). The second highest correlation was between MeanT and dengue, with the lag period of 12 weeks and a corresponding correlation coefficient of 0.211 with p-value<0.05. The correlation between rainfall and dengue incidence is, although significant, numerically quite small, about less than 0.15.
It was also observed that AH was associated with the smallest QAIC values, among all weather predictors in both single and distributed lag models (see Table 2). The best single lag effect of AH was 1 week, after adjustment for the impact of previous dengue incidence. When considering the cumulative lag effect of AH, a 0–16 weeks lag of AH showed the best fitting performance. Residual analysis is shown in Figure 3. The smaller the fitted number of dengue cases was, the less the variability of the residual values would be seen (Figure 3B). This supported our statement that overdispersion existed in the distribution of dengue. Autocorrelation function and partial autocorrelation function of residuals (Figure 3C & Figure 3D) demonstrated the independence of the residuals, implying that autocorrelation of the dengue cases has been explained by the DLNM-AH model.
Summing up each single lag effect from 0 to16 weeks, the 17-week overall effect of AH on relative risk of dengue incidence for the full period is shown in Figure 4A. It can be seen that a higher AH was associated with a higher dengue incidence. It is important to note that that the relative risk here is the ratio of the probability of dengue incidence occurring at a certain value of a weather variable to the probability of the event occurring at a reference value of the same weather variable. The change of reference points may affect the width of confidence interval, but it will not affect the RR curve itself. In some research work, mean was chosen as reference [43], while the point of overall minimum mortality was chosen as the reference in some other work [40]. Here, the reference value of AH is 22.4 g/m3, which is both mean and median of AH during the studied period.
The estimated weekly dengue incidence, using only the AH term (i.e., exp(), see Eq. 2) is shown in Figure 5A. The correlation coefficient between the estimated dengue and observed dengue cases is 0.374 (p-value<0.01), which shows a moderate positive relationship. It can be clearly seen that the peaks of AH and dengue incidence are very well synchronized.
As MeanT has been used as an indicator by National Environment Agency (NEA) of Singapore for dengue surveillance in recent years [44], we also modeled MeanT's impact on dengue incidence and compared it with the impact of AH. Based on our model analysis, the longest lag that best reflects the effect of MeanT on dengue is 9 weeks. Residual analysis is shown in Figure 6. Similar phenomena were detected in the residuals compared with the residuals of the DLNM-AH model. Nevertheless, slightly higher values were detected in autocorrelation function and partial autocorrelation function of residuals (Figure 6C & Figure 6D).
The effect of 0–9 weeks lag of MeanT for the full period is shown in Figure 4B. In general, it can be seen that a higher MeanT is associated with a higher risk of dengue incidence but this observed relationship does not hold true when the MeanT is higher than 27.8°C. The estimated number of weekly dengue cases using the MeanT term, described in Eq. 1, is shown in Figure 5B, which showed that the correlation coefficient between the estimated dengue and the observed dengue cases is only 0.150.
In addition to studying the pattern for the entire period (2001–2009), analyses were also carried out on the three distinct sub-periods, namely, 2001–2003 (sub-period 1, DENV2), 2004–2006 (sub-period 2, DENV1), and 2007–2009 (sub-period 3, DENV2). The aim is to evaluate the coupling effect of weather factors as well as the impact of the dominant serotypes in each period. The overall effects of AH on dengue incidence in each sub-period are presented in Figure 7(A1 to A3). In sub-period 1 and sub-period 2, the impact of AH on dengue incidence was found to be similar to that observed in the whole period, i.e. increasing the AH generally increased the risk of dengue incidence. However, in sub-period 3, it can be seen that the effect of AH on dengue was not significant.
The effect of 0–9 weeks lag of MeanT for each sub-period is shown in Figure 7(B1 to B3). It can be seen that the impact of MeanT on dengue incidence in the three sub-periods was not consistent across the three sub-periods or with the pattern observed during the whole period. In sub-period 1, the impact of MeanT on dengue was not significant when MeanT was less than 27.8°C; whilst in sub-period 2, this effect turned to be not significant when MeanT was higher than 27.8°C. Interestingly, the effect of MeanT in sub-period 3 was an inverse U curve, as shown in Figure 7(B3).
In general, rain, temperature and relative humidity had been the most common weather variables associated with dengue incidence and outbreaks [24], [45], [46]. The influence of these meteorological factors on dengue is likely to be associated with their impact on mosquito populations and behavior [47]. Rain provides more breeding habitats and opportunities for proliferation in the environment. There is also compelling evidence supporting the hypothesis that mosquito oviposition, development from mosquito larva to adult, biting rate and virus replication rate in mosquito are strongly enhanced at raised ambient temperatures [48], [49]. The hatch percentage for Ae aegypti eggs was also found to increase with the increase in relative humidity in Texas [50].
However, in our study, it was observed that there is no significant relationship between RH and dengue (see Figure 2). On the other hand, we found that temperature is positively correlated with the count of dengue cases, although temperature is negatively correlated with relative humidity. Hence, we further studied the relationship between AH and dengue in this work with the consideration that AH measures absolute moisture in the ambient air as a composite factor of mean temperature and relative humidity.
To reflect the influence of absolute moisture in the ambient air on dengue incidence, we explored the cross-correlation of dengue incidence with absolute humidity and found that it had the best correlation with dengue cases in Singapore among the major meteorological variables. Furthermore, as indicated by the DLNM-AH model, a moderate positive correlation between dengue and its estimation using only the AH term (correlation coefficient is 0.374, p<0.01) was obtained. This correlation coefficient is relatively high compared with other weather factors. Besides the significant correlation coefficient, it was also noted that the peaks of absolute humidity were well synchronized with dengue peaks. Although MeanT is being used for risk assessment of dengue by the authorities [44], our modeling results suggests that AH may be a better indicator to predict dengue incidence, as demonstrated by the RR curves and the higher correlation coefficient when compared to MeanT.
Interestingly, rainfall, which had been found to be associated with dengue in many places, did not seem to have much bearing on dengue cases in Singapore. This is perhaps consistent with the findings of the National Environment Agency which claimed that typically about 70% of breeding habitats of Ae aegypti were associated with homes and the most common breeding habitats were indoor ornamental containers and household items where the impact of rainfall is likely to be limited.
In our study, the effect of AH on dengue was found to have an optimal maximum lag of 16 weeks, an interval which is consistent with an earlier study [15], [26]. The non-linear lag effect of weather predictors on dengue incidence has also been reported in many studies [15], [26], [45]. The lagged effect of dengue incidence could account for the length of life cycle as well as the host-vector-pathogen transmission cycle of vectors [15].
MeanT is being used for dengue surveillance in recent years [44] in Singapore. Following our studies, when evaluating over the whole studied period and sub-period 2 and 3, no significant effect of MeanT on dengue was observed, i.e., higher MeanT corresponding to higher rate of dengue incidence was only found in sub-period 1 when MeanT>27.8°C. In comparison, the effect of AH on dengue was more significant.
We also highlighted that, in the 9-year studied period, the dominant serotype has shifted every 3 years: Firstly, serotype 2 was the dominant one (sub-period 1: 2001–2003); then the dominant serotype shifted to serotype 1 in sub-period 2 (2004–2006); then in sub-period 3, it shifted back to serotype 2. Three key differences were observed in these three sub-periods:
It is interesting to note that the impact of AH on the risk of dengue was prominent for the first two sub-periods but not significant in sub-period 3. Sub-period 3 was also markedly different when MeanT was studied showing a reverse correlation when compared with sub period 1. The inconsistent pattern observed in sub-period 3 for both AH and MeanT suggests that one or more of the observed differences described above, could have played a role in modulating the correlation between dengue trends and the weather parameters. This demonstrates the need for studies of the correlation of infectious diseases with environmental parameters to take into consideration changes in control programs, circulating viruses and other epidemiological parameters.
Although in our study we have highlighted, based on our results that AH is an important weather indicator which impacts dengue incidence significantly, it does not mean that AH is the only weather factor to be considered for predicting dengue incidence. We had also carried out preliminary multivariate analysis to make further evaluation. The selection of weather factors to be included in the multivariate model is carried out according to QAIC: AH was first selected due to its minimum QAIC value among all candidate weather variables. Then, under the QAIC criterion, among the other weather factors, MeanT was the second one added to the model. After MeanT was selected representing temperature effect, both minimum and maximum temperatures were excluded from the variables selection procedure. The selection procedure continued among wind speed, relative humidity and rainfall. Then, lastly rainfall was the third one included in the model based on the QAIC criterion, i.e., an AH-MeanT-Rainfall model was constructed following our simplified selection approach. However, it was observed that the impact of AH on dengue incidence was similar irrespective of whether other weather factors were included during the modeling evaluation. We also used data in 2001–2008 to fit the two models (AH-MeanT-Rainfall model and AH model) and used the 2009 data for dengue prediction. The results (Mean Average Error) showed that the performance of the multivariate model (AH-MeanT-Rainfall model) was just slightly better than the AH model. This showed that AH can be a very useful weather factor for indicating dengue incidence trends. Furthermore, the use of a simple model with fewer variables would provide reference more clearly for policy makers in dengue surveillance operations. As this work focused on studying AH's impact on dengue incidence using our model, we believe that a more extensive research needs to be carried out to study the prediction models considering all the combination of AH and other available weather factors.
Cross correlation analysis and DLNM modeling showed that AH was the best predictive weather factor among the weather factors studied. AH presented a more stable effect on indicating dengue incidence than MeanT did over the whole studied period as well as during sub-periods. A higher AH was associated with a higher dengue incidence. As such, AH could potentially be a better weather indicator for predicting dengue and assisting pro-active dengue prevention efforts in the future.
The shift of dominant serotypes and pre-emptive measures taken against dengue vectors since 2005 in Singapore may possibly explain the inconsistent weather-dengue patterns observed. As such, further studies are recommended to identify, evaluate and possibly include more diverse virological, immunological, entomological and public health factors into the dengue models.
|
10.1371/journal.pntd.0005965 | The Babesia bovis hap2 gene is not required for blood stage replication, but expressed upon in vitro sexual stage induction | Babesia bovis, is a tick borne apicomplexan parasite responsible for important cattle losses globally. Babesia parasites have a complex life cycle including asexual replication in the mammalian host and sexual reproduction in the tick vector. Novel control strategies aimed at limiting transmission of the parasite are needed, but transmission blocking vaccine candidates remain undefined. Expression of HAP2 has been recognized as critical for the fertilization of parasites in the Babesia-related Plasmodium, and is a leading candidate for a transmission blocking vaccine against malaria. Hereby we identified the B. bovis hap2 gene and demonstrated that it is widely conserved and differentially transcribed during development within the tick midgut, but not by blood stage parasites. The hap2 gene was disrupted by transfecting B. bovis with a plasmid containing the flanking regions of the hap2 gene and the GPF-BSD gene under the control of the ef-1α-B promoter. Comparison of in vitro growth between a hap2-KO B. bovis clonal line and its parental wild type strain showed that HAP2 is not required for the development of B. bovis in erythrocytes. However, xanthurenic acid-in vitro induction experiments of sexual stages of parasites recovered after tick transmission resulted in surface expression of HAP2 exclusively in sexual stage induced parasites. In addition, hap2-KO parasites were not able to develop such sexual stages as defined both by morphology and by expression of the B. bovis sexual marker genes 6-Cys A and B. Together, the data strongly suggests that tick midgut stage differential expression of hap2 is associated with the development of B. bovis sexual forms. Overall these studies are consistent with a role of HAP2 in tick stages of the parasite and suggest that HAP2 is a potential candidate for a transmission blocking vaccine against bovine babesiosis.
| Babesia bovis, is a tick borne apicomplexan parasite responsible for important cattle losses globally. Babesia parasites have a complex life cycle including asexual replication in the mammalian host and sexual reproduction in the tick vector. Novel control strategies aimed at limiting transmission of the parasite are needed, but transmission blocking vaccine candidates remain undefined. In this study we analyze the conservation and role of the hap2 gene in the erythrocyte stage of the life cycle of the parasite and found that expression of the gene is not required for the development of the parasite in erythrocytic stages, using a hap2 mutated parasite line. In addition, we developed an in vitro system for the induction of sexual forms of B. bovis and found expression of the hap2 gene and surface localization of the protein. However, hap2-KO parasites are unable to develop sexual stages. We concluded that HAP2 is a leading candidate for a transmission blocking vaccine against bovine babesiosis due of the high level of conservation, surface exposure, and specific expression in tick stage and in in vitro induced sexual stages parasites.
| Bovine babesiosis is a tick-borne disease that limits food production in tropical and subtropical regions worldwide. The disease is mainly caused by Babesia bovis, B. bigemina, and B. divergens and is endemic in large parts of Australia, Africa, Asia, Europe, and Latin America [1]. Parasites of the genera Babesia are transmitted by ixodid ticks including Rhipicephalus spp [2–4]. Animals that survive acute infection remain persistently infected and are reservoirs for tick transmission [5, 6]. Bovine babesiosis control strategies have been met with limited success in some countries. However, these strategies, including acaricide treatment and live, attenuated vaccines [1, 7–9], are restricted due to increasing acaricide-resistant tick populations and by practical constraints of the live Babesia vaccines, such as possible reversion to virulence and the risk of tick transmission [7, 10, 11]. Despite safety concerns, some countries in endemic regions still use live vaccines to mitigate acute infection and prevent mortality.
To complete its life cycle, Babesia may require strict regulation of gene expression to develop, invade, replicate and survive in distinct and diverse hosts and tick vectors. Babesia parasites have a complex life cycle including asexual replication of haploid stages in the mammalian hosts and sexual reproduction of diploid stages in the tick vector [12]. The initial phenotypic differentiation of Babesia into sexual stages that occurs in the tick midgut lumen may require the expression of a subset of proteins necessary for fusion and formation of diploid zygotes [13]. Zygotes selectively infect tick midgut epithelial cells and subsequently develop into kinetes [14]. Mature kinetes are released into the tick hemocoel and invade various tick organs including salivary glands and ovaries. Eventually, the parasite is vertically transmitted to the next tick generation where another morphological change occurs as the parasite transforms into sporozoites in larval salivary gland acinar cells [12].
In a closely related human pathogen, Plasmodium specific proteins have been identified with important functions for parasite development within mosquitos. Plasmodium expresses a protein known as HAPLESS2/GCS1 (HAP2) and it is exclusively expressed on the surface of microgametes that occur in the mosquito gut lumen [15]. This protein is critical for the fertilization of parasites prior to development of the stage that infects mosquito gut epithelial cells. In Plasmodium, HAP2 is a candidate for a transmission blocking vaccine. Similar to Plasmodium, fertilization of B. bovis gametes within the vector tick midgut lumen is an obligate step for the parasite to perpetuate its life cycle. Disruption of B. bovis fertilization during parasite development in tick midgut would prevent transmission via tick vectors. Recent research described additional members of the B. bovis 6-Cys genes and defined the 6-Cys A and B genes as markers for midgut stages [16]. However, little else is known regarding the expression of additional sexual stage proteins by B. bovis, or the events leading to sexual reproduction of the parasite during its development in the midgut. In silico analysis demonstrated the presence of a gene in B. bovis genome that is orthologous to the Plasmodium hap2 gene. The pattern of expression, localization and biological significance of HAP2 in B. bovis remains unknown. In this study, we demonstrate that hap2 is transcribed exclusively during B. bovis development within the tick midgut, and not by blood stage parasites. We also demonstrated that deletion of hap2 does not affect the growth of B. bovis blood stages in cultures, and that the expression of HAP2 is associated to sexual stage development in in vitro sexual stage induction experiments. Collectively, the data indicates that similar to Plasmodium [17], B. bovis HAP2 is a potential candidate antigen for developing transmission blocking vaccines that might elicit a host immune response able to disrupt the development of a B. bovis stage infectious for tick midgut epithelial cells.
To examine the pattern of hap2 expression in B. bovis infected tick midgut, approximately 20,000 Rhipicephalus microplus larvae, La Minita strain, were placed under a cloth patch on a splenectomized calf as previously described [18, 19]. When approximately 1% of the nymphs molted to adults, the calf was inoculated intravenously with B. bovis Texas strain stabilate contained 107 infected erythrocytes [20] to synchronize peak parasitemia with female tick repletion. Replete female ticks were collected, washed in tap water, dried and incubated at 26°C in 93% relative humidity. During development of B. bovis within tick midgut, five engorged ticks from the incubator were dissected daily for 6 consecutive days. Individual midgut was placed into 1 ml of Trizol (Thermo Fisher Scientific, Waltham, MA) and stored at -80°C. To evaluate hap2 expression in B. bovis blood stages, infected defibrinated blood was collected and the erythrocytes washed five times with Puck’s saline G to remove white blood cells. Parasites were pelleted by centrifugation of infected blood and suspended in Trizol. To extract RNA from in vitro sexual stages induced culture, parasites were isolated by differential centrifugation of 400 xg to 2,000 xg to pellet the extracellular stages. Total RNA extracted using the Trizol according to the manufacturer’s protocol. The samples were treated with DNase I, quantified by Nanodrop (Thermo Fisher Scientific), and 150 ng of total RNA utilized for synthesizing cDNA (Thermo Fisher Scientific). Primer sets for hap2 were designed to amplify a 165 base pair fragment (Table 1). PCR cycling conditions consisted of 95°C for 3 min followed by 35 cycles of 95°C for 30 sec, 55°C for 30 sec and 72°C for 30 sec, with a final extension of 72°C for 5 min. PCR products were visualized by 2% agarose gel electrophoresis. The PCR amplicon was cloned into PCR 2.1-TOPO (Thermo Fisher Scientific) and submitted for sequencing (Eurofins MWG Operon, Louisville, KY).
Full length B. bovis hap2 cDNA synthesized from infected female tick midgut RNA was used to compare to the complete annotated B. bovis genome sequence [5] and other apicomplexan genomes using Multiple Sequence Alignment by CLUSTALW (http://www.genome.jp/tools/clustalw/). Domain prediction of hap2 gene was performed using the Simple Modular Architecture Research Tool (SMART) (http://smart.embl-heidelberg.de). Trans-membrane domains and signal peptides in the HAP2 protein were predicted using the Transmembrane Hidden Markov Model package 2 (TMHMM2) (http://www.cbs.dtu.dk/services/TMHMM-2.0). The detection of glycosylphosphatidylinositol (GPI) anchor was predicted using an online GPI prediction server (http://mendel.imp.ac.at/gpi/gpi_server.html).
The complete gDNA sequence for hap2 gene was compared among four geographically distinct B. bovis strains including Texas strain, Mo7, Argentina L17 virulent and Argentina L17 attenuated. Strain-specific single nucleotide polymorphisms (SNPs) were then estimated in order to calculate the ratio of synonymous to non-synonymous changes. To estimate ω (dN/dS ratio), “SNAP” was used (http://hcv.lanl.gov/content/sequence/SNAP/SNAP.html). The parameters were set up as follows: ω >1 indicated positive selection, as the selection had caused some amino acid substitution; ω<1 indicated occurrence of purifying selection and a high degree of sequence conservation [21]. Nucleotide substitutions were calculated manually.
B. bovis Texas strain parasites were maintained in long-term microaerophilous stationary-phase (MASP) cultures as previously described [22, 23]. Cultured blood parasites were used as to knockout the hap2 gene. Briefly, a recombinant plasmid containing a fusion luciferase-GPF-BSD (LUC-GFP-BSD) gene under the control of the ef-1α-B promoter flanked by portions of the hap2 gene, 660 bp in 3’ and 950 bp in 5’, was constructed (Fig 1A and 1B, GenBank accession number: KX234096).
E. coli were transformed, 5 colonies selected, and grown overnight in 5 ml of LB broth. The phap2-luc-gfp-bsd plasmids were extracted and submitted for sequencing. Recombinant plasmids were purified using EndoFree Plasmid Maxi Kit (Qiagen, Santa Clarita, CA) according to the manufacturer’s protocol (EndoFree Plasmid Maxi Kit, cat # 12362). Twenty μg of plasmids phap2-luc-gfp-bsd or pBlueScript (pBS) as a control were diluted into 25 μl Cytomix and electroporated with 75 μl of 20% B. bovis infected erythrocytes as previously described [24]. Six hours after transfection, blasticidin was added to culture medium to a final concentration of 4 μg/ml for parasite selection. Parasite growth was determined by counting the parasitemia using Giemsa stained blood smears. One week after transfection, the expression of GFP was examined by fluorescent microscopy as previously described [25].
The B. bovis transfected-hap2KO-gfp-bsd (Tf-hap2KO-gfp-bsd) line was cloned by fluorescence activated cell sorting using 96-well plates [26].
Genomic DNA was isolated from Tf-hap2KO-gfp-bsd clonal line (cln) and B. bovis wildtype Texas strain. Briefly, B. bovis Tf-hap2KO-gfp-bsd was expanded to 25% parasitemia. The erythrocytes were pelleted and washed with phosphate-buffered saline. Erythrocytes were lysed with red blood cell lysis solution (Qiagen, Hilden, Germany) incubated for 5 min at room temperature. Parasites were lysed using cell lysis solution (Qiagen, Hilden, Germany) with 20 μg/ml of Proteinase K and incubated at 56°C for 30 min. Proteins were removed and DNA isopropanol precipitated, washed with 70% ethanol, and suspended in 100 μl of DNA hydration solution (Qiagen, Hilden, Germany).
A PacBio library was constructed using the SMRTbell Template Prep Kit v1.0. Genomic DNA was sheared using the Covaris G-Tube at 1350G for 20 min, cleaned and size selected using Ampure XP beads (Beckman Coulter, Indianapolis, IN). Standard sequencing was performed on the PacBio RSII using P6/C4 sequencing chemistry and MagBead loading. Genome sequences were assembled de novo with the Hierarchical Genome Assembly Process (HGAP) v 2.0 that is integrated into the SMRT analysis package. Single contig containing transfection specific sequences in the genome of the transfected clonal line were identified using BLAST utilizing all portions of the donor plasmid as queries.
Sexual forms were induced essentially as previously described [27], with few modifications. The parasites used in these experiments were derived either from stabilates generated from blood of an infected splenectomized calf, or from hap2-KO culture, and maintained in in vitro cultures for one week before induction. These in vitro cultured B. bovis infected erythrocytes were suspended in an induction medium consisting of 0.465 ml final volume of culture (from a 100 ml stock solution containing 58 ml HL-1 culture medium (Lonza, Rockland, ME, USA), 40 ml bovine serum, 0.01 M TAPSO, 1 ml of 100X antibiotic-antimytotic solution (Invitrogen, CA, USA), and 100 uM xanthurenic acid (Sigma, St. Louis, MO, USA), with 10% bovine red blood cells (0.0465 ml of packed blood). Cultures were incubated at 26°C in air for up to 20h. Cultures were also incubated with the induction medium at 37°C for 20h in a MASP as previously reported [28], or in induction medium without xanthurenic acid at 26°C for 20 h.
Three synthetic peptides from the extracellular region of HAP2 were manufactured by BioSynthesis, Inc. (Texas, USA). Peptide 1: DGPEKRFRQRKGFFVC (15-mer, amino acids 2 to 17), peptide 2: KTPKGGAKKKKQKLDSSEWEHK (21-mer, amino acids 454 to 475) and Peptide 3: ERKREQESRERQAEHER (17-mer, amino acids 726 to 743). The peptides were conjugated to keyhole limpet hemocyanin (KLH) and used to immunize rabbits (BioSynthesis, Lewisville, TX, USA). Rabbits were inoculated with 0.5ml of conjugated peptides (conc.at 1.43 mg/ml) mixed in complete Freund’s adjuvant for the initial inoculation and in incomplete Freund’s adjuvant for all the booster injections. The adjuvants were mixed with the antigens at a ratio of 1:1. Inoculations were performed subcutaneously along the back, and intramuscularly in the hind limbs. All injections (less than 0.2ml/site) were done at multiple sites regardless of the route. The resulting immune sera were titrated by ELISA and used in subsequent immunoblot assays. The specificity of the anti HAP2 polyclonal antibody was further tested in immunoblots using non-purified recombinant HAP2 expressed in prokaryotic expression system pBAD/thio-TOPO (Invitrogen, CA, USA) (S1 Fig). The full size hap2 gene was amplified from B. bovis cDNA by PCR amplification, using primers hap2 full length (Table 1). The resulting amplicons were cloned into the pBAD/thio-TOPO vector. Plasmid DNA extracted from E coli positive clones were sequenced to confirm their identity and the correct orientation of the hap2 insert. One positive clone was selected for expression of HAP2 in E. coli-transformed cultures (125 ml) using expression induction with 0.2% arabinose for 3 h at 37°C. Bacterial pellets were suspended and homogenized in lysis buffer-Nonidet-P40 (NP-40) (150 mM sodium chloride 1.0% NP-40–50 mMTris, pH 8.0) and protease inhibitor (1 μg/ml). Total protein from cell lysate used for the immunoblots. The polyclonal antibody for Bbo 6-Cys A was described and obtained in a previous study [16].
The antigens used in the immunoblots were prepared from B. bovis-infected erythrocyte culture or xanthurenic acid induced culture. Sexual stages were pelleted from induced culture by differential centrifugation of 400xg to 2,000xg to pellet the sexual stages. Parasites were suspended and homogenized in lysis buffer and 1 μg/ml protease inhibitor (Roche Diagnostics, Indianapolis, IN, USA). Total protein was quantified by Micro BCA Protein Assay (Thermo Fisher Scientific Inc., Waltham, MA, USA), 5 μg of total protein were mixed with 5x SDS-PAGE sample buffer (GenScript, Fl, USA), boiled for 5 min and then sonicated for 2 min with 20 sec intervals, and separated into 4–20% Mini-PROTEAN TGX Precast Gels (BioRad Laboratories, Hercules, CA, USA). Proteins were transferred to a nitrocellulose membrane (Whatman, Dassel, Germany) for 1 h at 100 V. The membranes were blocked with 5% skim milk in TBS (Tris-buffered saline: 50 mM Tris-HCl/ 150 mM NaCl, pH7.6) for 1 h at room temperature, washed three times in TBS and incubated for 1 h with 1:100 dilution of primary antibody against B. bovis 6-Cys A and HAP2. Monoclonal RAP-1 antibodies were used to detect RAP-1 protein during in vitro cultured B. bovis [29] as well as pre-immune rabbit serum as positive and negative controls, respectively. After three washes in TBS, The membranes were incubated for 30 min with 1:5000 dilution of HRP conjugated goat anti-rabbit IgG (H+L) or anti-mouse IgG (H+L) antibodies (KPL, Gaithersburg, Maryland, USA), and washed again three times with TBS. Antibody reactivity was visualized using chemiluminescent HRP antibody detection reagents (KPL, Gaithersburg, Maryland, USA).
Sexual stages were enriched from in vitro cultures induced by 20h using differential centrifugation as described above. Parasites were washed in 3% normal goat serum in PBS. A portion of the cells were then incubated for 1h with 1:50 anti-HAP2, or anti-6-Cys A primary antibodies diluted with 10% normal goat serum in PBS. The cells were then washed twice in the PBS by 400 xg centrifugation and incubated for 30 min with 1:1000 goat anti-rabbit Alexa Fluor 555 secondary antibodies (Thermo Fisher Scientific) diluted with 10% normal goat serum. The cells were again washed twice with PBS, and air dried on slides, and nuclei were stained with 4, 6-Diamidino-2-phenylindole dihydrochloride (Thermo Fisher Scientific). Identically produced negative controls were performed using pre-immune (PI) rabbit serum instead of the primary antibodies. All samples were independently visualized by fluorescent microscopy and images were processed as described below. Slides were viewed and digitally photographed using an Axio Imager, M1 microscope (Carl Zeiss Imaging, Inc., Phoenix, AZ, USA). The microscope is equipped with an X-Cite 120 Fl illuminating system (EXFO Photonic Solutions). Digital images were captured using an AxioCam MRm digital camera connected to a desktop computer running the AxioVision (version 4.8.1.0) program. Image stacks were obtained using optimal z-axis spacing [250 nm z-step, Plan-Apochromat 63x/1.4 oil M27 objective (Carl Zeiss Imaging, Inc., Phoenix, AZ, USA)]. Z-stack image files were imported for processing into the ImageJ-based open source processing package Fiji (version 1.48b; http://pacific.mpi-cbg.de/) [30]. Surface exposure of HAP2 in in vitro Xanthurenic acid induced parasites was confirmed by analyzing parasites in IFA after trypsinization [31] as follows. Sexual stages of B. bovis induced in in vitro cultures were washed twice in PBS by 400 xg centrifugation. Cells were trypsinized for 30 minutes at 37o C with 0.05% trypsin-EDTA (Gibco BRL/Invitrogen, Carlsbad, CA, USA), trypsinization was terminated with the addition of trypsin inhibitor (Sigma-Aldrich, St Louis, MO, USA) for 10 min at 37°C. Parasites were then washed in 3% normal goat serum in PBS. A portion of the cells were then incubated for 1h with 1:50 anti-HAP2 primary antibodies diluted with 10% normal goat serum in PBS, and washed twice in the PBS by 400 xg centrifugation and incubated for 30 min with 1:1000 goat anti-rabbit Alexa Fluor 555 secondary antibodies (Thermo Fisher Scientific, CA, USA) diluted with 10% normal goat serum. The cells were again washed twice with PBS. To estimate cell viability, cells were suspended in PBS and mixed with equal volume of 6-Carboxyfluorescein Diacetate (6-CFDA [31], final concentration in PBS. 10 μg/ml: Calbiochem-Behring, La Jolla, CA, USA), and Incubated at room temperature for 15 minutes. The cells were then washed once with PBS and incubated with nucleic acid stain Hoechst 33342 (Thermo Fisher Scientific, CA, USA) for 30 minutes. Finally, cells were washed twice with PBS, and air dried on slides. All samples were independently visualized by fluorescent microscopy as described above.
This study was approved by the Institutional Animal Care and Use Protocol Committees of the University of Idaho, Moscow, Idaho (protocol #2016–20) in accordance with institutional guidelines based on the U.S. National Institutes of Health (NIH) Guide for the Care and Use of Laboratory Animals. The rabbit antibodies were generated according to the approved animal care protocol D16-00398 (OLAW) by BioSynthesis, and to USDA Research license number 23-R-0089.
A single copy B. bovis hap2 gene is located in between 1,452,162 bp and 1,454,808 bp of chromosome 3, containing 8 introns and 9 exons. This multi-intron structure is usually conserved in the hap2 genes among apicomplexan parasites [32–35]. The annotated hap2 mRNA [GenBank XM_001611756] revealed an orf of 2,271 bp, coding for a 79.53 kDa protein containing 723 amino acids. B. bovis HAP2 protein contains a single HAP2 domain Similar to Plasmodium falciparum 7G8 HAP2 (XP_001347424).This domain is functionally involved in a highly conserved sperm protein that is essential for gamete fusion. The HAP2 domain is located between amino acids 348 and 394, suggesting a similar conserved function for this gene among these parasites. The HAP2 domain is predicted to be located in the extracellular region of the protein, which is likely exposed on the surface of the parasites.
Overall, B. bovis HAP2 deduced amino acid sequence appears relatively well conserved when compared with its homologues in other species. In silico predictions suggests that the B. bovis HAP2 protein lacks a glycosyl phosphatidylinositol (GPI) anchor. In addition, HAP2 is also predicted to contain a signal peptide between amino acids 1–33, a hydrophobic transmembrane domain located between amino acids 683–705, and a predicted coiled coil domain between amino acids 721–753, located near the C-terminus. The coiled coil domain is also present in an identical location in many viral fusion proteins, consistent with possible role in the membrane fusion reaction. Collectively, all these features are suggestive of the possible trafficking of HAP2 to the external surface of the parasite, and its possible role as a fusogenic protein.
We examined the occurrence of sequence variation and single nucleotide polymorphisms (SNPs) among the hap2 gene among distinct B. bovis strains. The hap2 gene is highly conserved among the distinct strains (99% to 99.9% aa identity). The analysis was performed using a sequence database including hap2 gene derived from B. bovis Texas strain, Mo7, Argentina L17 virulent and Argentina L17 attenuated. The calculated synonymous and non-synonymous S/N ratios (Table 2) with the parameter, ω, (ω = dN/dS), as an indicator of potential selection pressures. In all cases, ω of less than 1 was obtained and revealed no evidence for positive selection for the hap2 gene, suggesting that B. bovis hap2 gene is under no diversifying immune selection. The data also indicates a low likelihood of selective forces such as immune pressure of the host acting on the evolutionary history of this gene, consistent with low or lack of exposure of the protein to the immune system of the host during infection, which suggests no expression of hap2 in blood stages of the parasite.
The pattern of hap2 transcript expression was investigated by RT-PCR analysis performed on RNA extracted from B. bovis infected erythrocytes and tick midguts. Previous transcriptome and RNA sequence analysis using short-term cultured merozoites from strains differing in origin and virulence phenotypes show that the hap2 gene is transcribed at very low or undetectable levels compared to constitutively expressed rap1 in the blood stages (S2 Fig) [36]. Consistently, only rap-1, but not hap2 transcripts were detected in B. bovis blood stages by RT-PCR. In contrast, hap2 transcripts were transiently detected at days 2, 3 and 4, but not at days 0, 1, 5 and 6 post-repletion during the development of B. bovis in the tick midgut (Fig 1). Sequence analysis of the RT-PCR derived 165 bp amplicon (Fig 1) demonstrated identity to the hap2 sequence from the annotated B. bovis genome [5]. Overall, the results indicate that the hap2 gene is differentially transcribed during B. bovis development within tick midgut, but not during development of parasites within the mammalian host. Interestingly, the differential intensity of the RT-PCR bands (Fig 1) suggests that expression of hap2 peaks at day 2 post repletion, but this observation needs to be confirmed using a quantitative assay.
The hap2 gene was disrupted using the transfection plasmid phap2-lucgfpbsd. The structure of the B. bovis hap2 locus and the experimental design for the disruption of hap2 are represented in Fig 2.
Selection with blasticidin resulted in the emergence of the transfected and green fluorescent Tf-hap2KO-gfp-bsd cell line. In contrast, parasites electroporated with the control pTf-pBS plasmid did not survive upon blasticidin selection (Fig 3A). Clonal cell lines were generated from the mixed parasite line Tf-hap2KO-gfp-bsd by flow cell sorting [37, 38]. Clonal lines were evaluated by expression of GFP (Fig 3B). The clonal line termed Tf-hap2KO-gfp-bsd-cln was selected for further analysis. Growth curve analysis demonstrated that both B bovis Tf-hap2KO-gfp-bsd-cln and its wild type parental strain had similar replication kinetics (Fig 3C).
The clonal line Tf-hap2KO-gfp-bsd-cln was fully sequenced. Analysis of the full genomic DNA sequence of the Tf-hap2KO-gfp-bsd-cln line revealed an output of polished assembly of 609 contigs with the largest contig being 87kb. BLAST of the hap2 gene (BBOV_III006770) against assembly contigs revealed a single hit at contig 1592. The area covered by contig 1592 was roughly 1,438,762 to 1,455,743 of chromosome 3, which contained a ~6, 448 bp insertion (GenBank accession number: KX234097). Full sequencing of the genome of the Tf-hap2KO-gfp-bsd-cln confirmed replacement of the hap2 gene by the 5’ hap2 (935 bp), EF promoter (761 bp), luciferase (1,651 bp), gfp-bsd (1,100 bp), 3’ rap1 (1,288 bp), and 3’ hap2 (656 bp) fragments, present in the transfection plasmid. Thus, analysis of the structure of the hap2-KO gene in the clonal line Tf-hap2KO-gfp-bsd-cln indicates that these sequences were inserted by homologous recombination. It was the only foreign DNA insert detected by whole genome sequencing. The rest of the B. bovis Tf-hap2KO-gfp-bsd-cln genomic sequence was essentially identical to the wildtype B. bovis genome sequence [5]. Collectively, these data confirmed a successful insertion of the transfected genes disrupting the targeted hap2 locus of B. bovis, and suggests that disruption of the hap2 locus did not affect the pattern of growth of the parasite in in vitro cultures.
B. bovis sexual stages were induced in vitro by decreasing the temperature to 26°C and the addition of xanthurenic acid to the culture media. Microscopic inspection of Giemsa stained cells from induced cultures showed the presence of extra-erythrocytic parasites with long projections and large round parasite stages, indicative of parasite sexual stage development (Fig 4A). No such sexual stages forms were found upon similar microscopic inspection of Tf-hap2KO-gfp-bsd-cln parasites (Fig 4B), and fluorescent microscopy inspection of Tf-hap2KO-gfp-bsd-cln live parasites (S3 Fig) developing in in vitro cultures with induction medium xanthurenic acid (XA) at 26°C.
Previous comparative studies performed in B. bovis parasites from in vitro cultures and in tick midgut, defined the expression of the 6-Cys A and B genes as markers of sexual stage parasites [16]. A similar comparative transcript analysis using RT-PCR and sequencing confirmed expression of 6-Cys A and B and hap2 genes in parasites emerging upon in vitro sexual stage induction. In contrast non-induced parasites failed to produce hap2 and 6-cys A and B transcripts (Fig 5A). In addition, Tf-hap2KO-gfp-bsd-cln parasites, produced rap-1 transcript, but failed to produce 6-cys A and B transcripts (Fig 5B) upon xanthurenic acid induction.
In addition, anti-6-Cys A and anti-HAP2 antibodies react with wild type B. bovis antigens of ~60 kDa and~80 kDa respectively in induced cultures but did not recognize any native protein in non-induced B. bovis culture in immunoblots (Fig 6). In contrast, the control 60 kDa RAP-1 is recognized in lysates from both, induced and non-induced parasites with comparable signal intensities (Fig 6). Tf-hap2KO-gfp-bsd-cln Induced parasites didn’t show reactivity against anti-HAP2 antibodies (S4 Fig). The size of the antigens recognized by all antibodies matches the predicted sizes of the RAP-1, 6-Cys A and anti-HAP2 proteins. Thus, the data is consistent with co-expression of the sexual stage marker 6-Cys A and HAP2 proteins in induced wild type B. bovis cultures, but not in non-induced cultures.
In addition, live immunofluorescence assays (Live IFA) confirmed expression of the 6-Cys A and HAP2 proteins on the surface of sexual stage induced B. bovis T3B strain, but not in the non-induced parasites (Fig 7A–7D). Overall the data is consistent with the notion that expression of the hap2 gene is associated with the development of B. bovis sexual stage forms induced in vitro.
We confirmed surface exposure of HAP2 by performing live immunofluorescence analysis on trypsin-treated B. bovis induced cells (Fig 8). Parasites treated with trypsin are no longer recognized by anti-HAP2 antibodies in live immunofluorescence assays (Fig 8 AF 568). In addition, the trypsin treatment did not alter the membrane integrity and the viability of the treated parasites, as they are still stained with the vital 6-CFDA stain in a pattern that is similar to non-trypsin-treated parasites (Fig 8. AF 488).
The hap2 gene products in B. bovis related apicomplexan parasites have been consistently associated with differential expression and the formation of sexual forms. In this study, we demonstrated transcription of hap2 in parasites residing in the midgut of replete R. microplus female tick fed on a bovine infected with B. bovis, but not in in vitro cultured blood stages. A requirement of B. bovis to perpetuate its life cycle is the ability to develop sexual stages within the lumen of the Rhiphicephalus tick midgut. The fusion of gametes results in a stage infectious to tick midgut epithelial cells. Within midgut cells, B. bovis transforms into kinetes. This stage egresses from the midgut into the hemolymph to further infect ovaries [12]. Previous work indicated the differential expression of members of the 6-Cys family in tick stages [16], suggested that specific B. bovis proteins are necessary for parasite development within the tick vector. Interestingly, transcription of hap2 is limited to days 2 to 4 after dropping. This pattern of transcription may be required for synchronized sexual stages formation, or be related to the timing of gamete fusion inside the tick midgut. These observations are consistent with Plasmodium where HAP2 is expressed only in gametocytes and gametes [15]. Differential expression of HAP2 is indispensable for fertilization of Plasmodium parasites, with a demonstrated specific fusion function during gamete interaction [15].
Importantly, genome sequence analysis among B. bovis isolates demonstrated that HAP2 is highly conserved with an identity of 99% to 99.9%.The high degree of HAP2 sequence conservation among strains also supports the usefulness of HAP2 as a potential antigen for vaccine development aimed to block B. bovis transmission. The synonymous and non-synonymous ratios (Table 2) revealed no evidence for positive selection for the hap2 gene, consistent with a low frequency of single nucleotide polymorphisms in hap2 from different B. bovis isolates. These results suggest that B. bovis HAP2 is under no diversifying selection, a property shared with current transmission-blocking vaccine candidates in Plasmodium [39]. The data also suggest the occurrence of functional restrictions to sequence variations for this gene, which enhances its potential as a vaccine candidate.
We also examined if knocking out the hap2 gene affected the growth fitness of the parasite in in vitro cultures. The in vitro growth fitness of the hap2 KO parasites was similar to the wildtype strain indicating that the gene is not critical for B. bovis development within erythrocytes. Importantly, full sequence of hap2 knocked out parasites demonstrated the insertion of a single copy of the transfected selectable marker/reporter genes disrupting the hap2 locus, and thus such transfected parasites are ideally suited for exploring the functional significance of the hap2 gene in B. bovis. Furthermore, the remainder of the genome of the transfected B. bovis genome was unaltered, and no other insertions derived from the transfection plasmid were found in the genome of the KO strain confirming the high specificity and efficiency of the homologous recombination mechanisms operating in B. bovis. These data also confirm the usefulness of transfection as an approach to study gene function by disrupting gene expression in different B. bovis stages.
In this study we were able to induce B. bovis sexual forms using xanthurenic acid in in vitro cultured parasites for the first time. Xanthurenic acid is a metabolic intermediate derived from the metabolism of tryptophan which is present in the gut of the Anopheles mosquito where it is known to induce gametogenesis of Plasmodium falciparum [40, 41]. It remains unknown whether this metabolite is also present in the tick midguts, or if gametogenesis in Babesia parasites also requires a xanthurenic acid depending mechanism. However, similar to previous observations in B. bigemina [27], we were also able to induce changes in B. bovis morphology using particular culture incubation settings in the presence of xanthurenic acid. The induced parasites present several distinct morphology and shapes, consistent with previous similar inductions on B. bigemina and with forms found in the midgut of ticks engorged on Babesia infected cattle. Importantly, because expression of the 6-cys A and B genes are known to be markers of B. bovis sexual stages, the molecular data on expression of these two genes upon induction included in our study validated for the first time the observation that the addition of xanthurenic acid concomitant with decreased incubation temperatures, results in the induction of sexual stages, as visualized before just by changes in the morphology. In addition, and consistent with morphology changes, induction result in their progression into a life stage that is morphologically and molecularly different than the life stage forms that typical in blood parasites cultured under standard (non-induced) culture conditions. Indeed the detection of expression of the 6-cys A and B genes in these induced forms is fully consistent with the formation of sexual forms normally induced in the midgut of R. microplus ticks feeding in Babesia infected animals [16]. Interestingly, our data show a correlation among the inability of Tf-hap2KO-gfp-bsd-cln parasites to change morphology, and to generate sexual stage specific expression products such as the members of the 6-Cys gene family. In contrast, these mutant parasites Tf-hap2KO-gfp-bsd-cln are fully able to develop and grow in vitro in erythrocytes, supporting the concept that an intact copy of the hap2 gene is required for sexual stage induction but irrelevant for blood stage development. The results can be compared with similar previous findings in malaria parasites, where sexual stage fusion was found dependent on the expression of the hap2 gene [42–44]. Importantly, two distinct lines of evidence, direct live immunofluorescence and loss of recognition of surface exposed HAP2 upon trypsinization, supports that B. bovis induced parasites express HAP2 in their surface. Therefore, it is likely that the B. bovis HAP2 indeed also functions as an ancestral gamete fusogen in this parasite, since highly diverse eukaryotic gametes carrying loss-of-function mutations in HAP2 also fail to fuse [45].
In summary, HAP2 is differentially expressed by B. bovis during its development within R. microplus and the in vitro induction data suggests that surface exposed expression of this protein might be connected to the completion of the B. bovis life cycle during parasite development in the tick midgut. The absence of detectable hap2 transcripts by B. bovis blood stages suggested that hap2 is unnecessary for parasite development during infection of mammalian host. In contrast, the data supports that expression of HAP2 occurs in concurrence with the development of sexual stages upon induction with xanthurenic acid under in vitro culture conditions. Overall, these findings strongly suggest a role of hap2 during tick stages of the parasite, probably including sexual reproduction and supports HAP2 as a leading candidate for a transmission blocking vaccine against bovine babesiosis. Further in vivo studies are necessary to determine if disrupting hap2 interferes with the development of B. bovis within tick midgut and beyond.
|
10.1371/journal.pgen.1006484 | GW182-Free microRNA Silencing Complex Controls Post-transcriptional Gene Expression during Caenorhabditis elegans Embryogenesis | MicroRNAs and Argonaute form the microRNA induced silencing complex or miRISC that recruits GW182, causing mRNA degradation and/or translational repression. Despite the clear conservation and molecular significance, it is unknown if miRISC-GW182 interaction is essential for gene silencing during animal development. Using Caenorhabditis elegans to explore this question, we examined the relationship and effect on gene silencing between the GW182 orthologs, AIN-1 and AIN-2, and the microRNA-specific Argonaute, ALG-1. Homology modeling based on human Argonaute structures indicated that ALG-1 possesses conserved Tryptophan-binding Pockets required for GW182 binding. We show in vitro and in vivo that their mutations severely altered the association with AIN-1 and AIN-2. ALG-1 tryptophan-binding pockets mutant animals retained microRNA-binding and processing ability, but were deficient in reporter silencing activity. Interestingly, the ALG-1 tryptophan-binding pockets mutant phenocopied the loss of alg-1 in worms during larval stages, yet was sufficient to rescue embryonic lethality, indicating the dispensability of AINs association with the miRISC at this developmental stage. The dispensability of AINs in miRNA regulation is further demonstrated by the capacity of ALG-1 tryptophan-binding pockets mutant to regulate a target of the embryonic mir-35 microRNA family. Thus, our results demonstrate that the microRNA pathway can act independently of GW182 proteins during C. elegans embryogenesis.
| Animal cells possess different small RNA species capable of precisely controlling the gene expression. Among them, microRNAs form a silencing complex with an Argonaute protein (known as miRISC) that abrogates protein production by targeting specific messenger RNAs. While there is a consensus that miRISCs are effective to mediate gene silencing, it is still unclear if they exist in different types in animals. Here we report specific mutations in the C. elegans microRNA-specific Argonaute ALG-1, which alter its association with the orthologs of GW182 proteins, important factors for miRISC-mediated silencing. Our genetic characterization of this mutant shows that part of miRISCs function without the GW182 orthologs during the embryogenesis. These findings suggest the presence of distinctive miRISC that can regulate gene expression in different ways during animal development.
| MiRNAs are highly conserved small non-coding RNAs that orchestrate gene expression in a broad range of developmental processes. The production of miRNA implicates a successive two-step processing involving two RNase III enzymes, Drosha and Dicer, which cleave the primary and precursor miRNA molecules in the nucleus and cytoplasm, respectively. The 21–23 nucleotide RNA products are loaded onto Argonaute proteins to form the ribonucleoprotein complex referred to as microRNA induced silencing complex or miRISC (Reviewed in [1], [2]). Utilizing sequence complementarity, the miRNAs then guide the miRISC to the 3' untranslated region (3'UTR) of target mRNAs to silence their expression. In humans and Drosophila, the miRISC is associated with a key partner protein, GW182, which contains glycine-tryptophan (GW) repeats. The GW182 N-terminal domain uses these GW repeats to interact with Argonaute [3–5], while the C-terminal domain recruits the PAN2-PAN3 and CCR4-CAF1-NOT deadenylase complex [6–8]. As a result, the complex triggers mRNA deadenylation and/or translational repression. Despite the difference in their domain organization from that of Drosophila and human GW182 proteins, two related C. elegans proteins, AIN-1 and AIN-2 [9, 10], appear to be orthologs of GW182 in C. elegans (reviewed in [11]). Both AIN-1 and AIN-2 are known to interact with Argonautes proteins through their GW repeats, but only AIN-1 interacts with PAN and NOT proteins [12] indicating that AIN-1 is most likely the bona fide functional GW182 ortholog. Thus, the interaction between Argonaute and GW182 proteins is clearly important for miRNA-mediated gene silencing across species, although the domain architectures of GW182 proteins are varied among species.
Argonaute proteins have a bilobed structure, each composed of the N and PAZ domains or the MID and PIWI ones (Reviewed in [13]). The PAZ and MID domains are engaged in the guide-RNA recognition at the 3’ and 5’ ends, respectively, while the PIWI domain harbors an RNase H-like active site that catalyzes the endonucleolytic cleavage of nucleic acids. Besides conferring the “slicing activity” on some Argonautes, the PIWI domain has also been reported to be important for the recruitment of silencing factors such as GW182 [4, 14, 15]. The crystal structure of human Argonaute2 (hAgo2) identified two hydrophobic pockets on the surface of the PIWI domain that were occupied with free tryptophan residues, suggesting that GW182 proteins could be tethered to hAgo2 via these two pockets on the PIWI domain [16]. The physical interaction between hAgo2 and GW repeats was subsequently validated by NMR studies [17]. Furthermore, analogous binding pockets were also identified on the surface of human Argonaute 1 (hAgo1) [18], suggesting that these pockets could be a conserved feature for recruiting GW-proteins. Notably, although the molecular interactions between Argonaute and GW182 proteins have become clearer in recent years, the functional importance of this interaction in the context of miRISC-mediated silencing during animal development has yet to be determined.
We set out to address the necessity and function of GW182 proteins during miRISC-mediated gene silencing throughout animal development. To achieve these goals, we generated transgenic C. elegans strains expressing an ALG-1 mutant (ALG-1TPmut) whose Tryptophan-binding Pockets lost an interaction with AIN-1 and AIN-2. We have demonstrated that the tryptophan-binding pockets are required for its interaction with AIN-1 and AIN-2. Surprisingly, loss of the physical interaction between ALG-1 and AINs phenocopied the null allele of alg-1(0), whereas embryonic lethality due to lack of alg-1 and alg-2 was rescued by alg-1(TPmut) alone. These results indicate the existence of a type of miRISC that plays an essential role, without the aid of GW182 proteins, during embryogenesis in animal development.
Recent structural studies of hAgo1 and hAgo2 identified two tryptophan-binding pockets on the exterior of the PIWI domain, which were predicted to serve as the binding site of GW182 proteins [16–18]. Our homology model based on the hAgo structures indicated the presence of two tryptophan-binding pockets on ALG-1 (Fig 1A–1C; S1 Fig). These putative pockets consisted of residues K803 and E838 in the first pocket, and P733 and F802 in the second one, which could recognize tryptophan residues of the bound GW182 protein (Fig 1A and 1B). We tested whether mutations of these residues affected the interaction between ALG-1 and the orthologs of GW182 in C. elegans both in vitro and in vivo. First, we purified a recombinant AIN-1 fragment (recAIN-1) encompassing the previously mapped binding site of ALG-1 [12] along with glutathione-S-transferase (GST)-tagged full-length wild-type ALG-1 (recALG-1). We also created at GST-tagged ALG-1 in which the four key residues of the two Tryptophan-binding Pockets were mutated to alanine (hereafter recALG-1(TPmut); Fig 1C and S2 Fig). Consistent with our expectations, the interaction between recALG-1(TPmut) and recAIN-1 proteins in vitro was less than detectable by western blotting (Fig 1D), strongly suggesting that the tryptophan-binding pockets on ALG-1 are essential for the interaction with AIN-1.
To test whether tryptophan-binding pockets mutation also disrupts the interaction with AIN-1/AIN-2 in vivo, we generated an alg-1(0) mutant C. elegans strain expressing a single copy of either alg-1 wild-type (wt) or alg-1(TPmut) gene under the control of endogenous regulatory elements. We then immunoprecipitated ALG-1 from these animals using a specific antibody [19]. While we could recover both wild-type ALG-1 and ALG-1(TPmut) in our immunoprecipitations, we could only observe an association of AIN-1 and AIN-2 with wild-type ALG-1 (Fig 1E and S3 Fig). These results prove that ALG-1 binds to AINs through the tryptophan-binding pockets, in a manner consistent with the interaction between human Argonautes and GW182 [17].
It is well known that the alteration of miRNA-specific Argonautes causes significant effects on the levels of miRNAs [20]. In C. elegans, the loss of alg-1 gene leads to a dramatic decrease of mature miRNAs along with the accumulation of miRNA precursors, suggesting a role in miRNA processing [21, 22]. To assess whether ALG-1 tryptophan-binding pockets mutant retains its function in miRNA biogenesis, we investigated the levels of precursor and mature miRNAs in worms carrying null alleles of the alg-1 gene and expressing either wild-type ALG-1 or ALG-1(TPmut). Quantitative real-time PCR and Northern blotting analyses showed that the level of mature miRNA were reverted to that of wild-type by the expression of ALG-1(TPmut) (Fig 2A and S4A Fig). Accordingly, the levels of miRNA precursors also decreased to those of wild type (S4A Fig). These results indicate that the interaction between ALG-1 and AINs is not necessary for miRNA processing.
Next, we assessed whether the mutations of the tryptophan-binding pockets affect binding of ALG-1 to miRNAs. We used 2'-O-methyl RNA affinity columns to purify ALG-1(TPmut) that had been loaded in vivo with miRNAs complementary to the affinity matrix [23, 24]. Comparable amounts of wild-type ALG-1 or ALG-1(TPmut) were associated with lin-4 and let-7 miRNAs (Fig 2B). Conversely, we also observed a similar level of miRNAs bound to different ALG-1 immunoprecipitated complexes (S4B Fig), suggesting that GW182 proteins do not affect the interaction between the Argonaute and miRNAs. Thus, these results suggest that GW182 proteins are dispensable for miRISC assembly.
We next investigated whether AINs-free miRISC can still control gene expression in animals. To this end, we took advantage of developmental phenotypes caused by loss of specific miRNAs. It is known that loss of let-7 miRNA family in C. elegans causes a characteristic phenotype, in which the animal bursts from the vulval opening after L4 moult [25]. This lethal phenotype was observed in a fraction of the population of worms deficient for alg-1 (Fig 3A). The addition of extra chromosomal transgene arrays expressing wild-type alg-1 gene significantly reduced the number of animals that burst (Fig 3A). In contrast, the expression of ALG-1(TPmut) did not rescue alg-1(0) animals (Fig 3A). In both the strains, the expression levels of AIN proteins were comparable (S5 Fig).
To further characterize ALG-1(TPmut) in miRNA-mediated gene silencing, we monitored the developmental patterning of seam cells. These lateral rows of hypodermal cells undergo a postembryonic developmental program, consisting of patterned rounds of division during each larval stage (L1 to L4), and ended by terminal differentiation encompassing exit from the cell cycle, cell fusion and production of a cuticular structure (called alae) at the adult developmental transition (Fig 3B). This developmental program is controlled at different larval stages by lin-4 miRNA [26], the let-7 family miRNAs (miR-48, miR-84, miR-241 and let-7) [25, 27] and their targets lin-14, lin-28, hbl-1, daf-12 and lin-41 [25, 27–33]. In absence of alg-1, the symmetric seam cell division program that occurs once at the L2 stage is repeated, leading to an increase of seam cell numbers and structural defects in cuticular alae caused by inappropriate terminal differentiation (Fig 3B–3D; [21]). Consistent with the lethality caused by the loss of let-7 family regulation, the seam cell developmental phenotype in alg-1(0) mutant animals was rescued by the expression of wild-type ALG-1 but not of ALG-1(TPmut) (Fig 3D). Using transgenic animals expressing a GFP reporter under the control of the lin-41 3'UTR, a known target of let-7 miRNA regulated at the L4-Adult transition [27, 32], we observed that the repression of lin-41 by let-7 miRNA is altered in the ALG-1(TPmut)-expressing adult animals (Fig 3E). Interestingly, in all cases the phenotypes observed in ALG-1(TPmut) animals are more severe than the ones caused by a complete loss of ALG-1 proteins. These observations suggest that with its retained capacity of interacting with microRNAs, ALG-1(TPmut) sequester microRNA from ALG-2, the other functional microRNA-specific Argonaute in worms [19, 21].
To assess whether the deficiency of miRNA-mediated gene repression in the ALG-1(TPmut)-expressing animals might be resulting from a defect in binding of target mRNAs, we generated a LambdaN (λN)/Box-B tethering-based reporter that enables interaction of Argonaute with a target, independently of a miRNA-mRNA interaction. In cultured cells, this has been a conventional system to decipher the molecular basis for gene silencing by Argonaute and GW182 proteins [14, 34, 35]. To apply the system to animals, we made a GFP gene reporter where the well-characterized lsy-6 miRNA binding sites in the cog-1 3'UTR were replaced by six Box-B stem loop structures (Fig 4A). We then made a transgenic C. elegans strain co-expressing a single copy of this reporter along with either wild-type or TPmut λN::mCherry-tagged alg-1 gene, both of which were under the control of the alg-1 promoter and 3′UTR regulatory regions. We first confirmed that the presence of the N-terminal tag does not affect ALG-1 function by performing alg-1 mutant rescue. When wild-type λN::mCherry::ALG-1 was co-expressed with the Box-B reporter, a significant decrease of the GFP signal was measured in the pharynx of young adult animals (Fig 4B to 4G). The expression of λN::mCherry::ALG-1(TPmut) protein, however, failed to repress the GFP reporter, suggesting that the interaction with AIN-1 is essential to trigger the repression of the tethering reporter (Fig 4B, 4H). Taken all together, we conclude that the physical contact of ALG-1 to AINs through its tryptophan-binding pockets is important for miRNA-mediated gene silencing in animals.
The miRNA pathway is a regulatory mechanism that is essential for the control of various steps during animal development including embryogenesis (for reviews see [36–38]). In C. elegans, this phenomenon is exemplified by the fact that loss of both miRNA-specific Argonaute genes alg-1 and alg-2 leads to embryonic lethality [19, 21]. We therefore decided to use our ALG-1 (TPmut) to test whether the interaction between GW182 proteins and miRISC is essential during embryogenesis. To achieve this, we knocked down alg-2 in alg-1(0) or (TPmut) animals by feeding them with bacteria expressing dsRNA against alg-2. Consistent with the phenotype observed in simultaneous RNAi knockdown of alg-1 and alg-2 [21, 22], we observed that nearly 70% of the F1 alg-1(0) population exposed to alg-2 (RNAi) displayed embryonic lethality, while the remaining F1 animals arrested just after hatching (Fig 5A). Surprisingly, the expression of ALG-1(TPmut) in an alg-1(0); alg-2(RNAi) background is sufficient to rescue embryonic lethality at a level comparable to that of alg-1; alg-2 loss of function animals expressing a wild-type ALG-1 transgene (Fig 5A). Strikingly, nearly all F1 progeny arrest in early stages of larval development (Fig 5A) suggesting that GW182 proteins are required for miRISC function during larval development but not during embryogenesis.
To directly test the contribution of GW182 proteins for miRNA-mediated silencing during C. elegans embryogenesis, we constructed a balanced strain with loss-of-function alleles of both AIN genes, (ain-1(ku322); ain-2(tm2432)/dpy-5), exposed them to dsRNA-expressing bacteria targeting ain-2 (ain-2 (RNAi)) to remove all maternally loaded ain-2 mRNA, and scored the F1 progeny. Even though AINs were well expressed during embryogenesis (S3 Fig and S6 Fig), the alteration of both AIN genes did not cause significant embryonic lethality but rather led to severe larval developmental arrest of the F1 population as seen for alg-1(TPmut); alg-2 mutant animals (Fig 5B) demonstrating that the alteration of GW182 function in C. elegans embryos phenocopies the loss of interaction with the miRISC.
To further test the importance of AINs in the embryonic miRISC, we tested the capacity of ALG-1(TPmut) to control animal sex determination, an embryonic gene regulatory pathway controlled by the mir-35-41 microRNA family cluster [39]. We therefore utilized the her-1(n695gf) allele that causes a weak derepression of her-1 expression in C. elegans hermaphrodites leading to mild masculinization (with a low penetrance of intersex and pseudomale; Fig 6). While the loss of alg-1 in this sensitized background significantly increased the number of masculinized animals observed, the expression of ALG-1(TPmut) in her-1(gf)/alg-1(0) animals completely reestablished it to the level observed in her-1(gf) animals (Fig 6) demonstrating that the interaction with AINs is not required for the function of the mir-35 microRNA family in this embryonic decision. Taken all together, our findings support that the embryonic miRISC does not necessitate GW182 proteins to silence gene expression.
Most of the reported mutations in Argonaute proteins affect the binding of both miRNA and GW182 proteins [3, 40, 41]. Until recently, there have been only two Argonaute mutants reported deficient in the interaction with GW182 protein without affecting miRNA binding [3, 40]. Interestingly, based on our prediction, two point mutations reported in Drosophila AGO1 (R771 and F777) could also be involved in forming the same tryptophan-binding pockets 1 and 2 in C. elegans. In this study, we have generated such point mutations in ALG-1 and demonstrated the functional importance of these pockets to sustain the interaction even with a different type of GW182 proteins that possess a non-canonical domain architecture. Interestingly, a very recent study using culture cell systems reported that the mutation of these binding pockets in Drosophila AGO1 and human Ago2 abolished the interaction with their GW182 proteins without affecting microRNA binding [42]. These Argonaute variants have been and can continue to be useful to study the mechanism of miRNA-mediated gene silencing independent of GW182 proteins. Thus, application of Argonaute tryptophan-binding pockets mutant variants will provide novel strategies to uncover new types of gene regulation in animals.
GW182 proteins have been long thought to be essential for miRNA-mediated gene silencing in animals. Recent observations using Drosophila S2 cells as well as cell-free systems, however, suggest that GW182 is not always necessary for miRISC-mediated gene silencing. For example, Drosophila Ago1 and Ago2, the latter of which mainly associates with an siRNA-duplex, were both able to repress translational reporters in GW182-dependent and–independent manners [43]. Using miRNA-mediated reporter assays with or without polyA tails, Fukaya and Tomari demonstrated that Drosophila Ago1 could block translation independently of GW182 [44]. More recently, the Carthew’s group reported that in fly, miRISC retains the silencing activity under conditions lacking GW182 protein (i.e. when nutrients are removed from S2 cell cultured media)[45]. In these experimental conditions, the absence of GW182 still leads to gene silencing that results from the inhibition of either early translation or elongation. These data implied a possible GW182-independent miRNA repression in cell culture though it remained unclear whether this was the case in animal development.
Using C. elegans as a model, we show here that there are two miRNA-mediated gene-silencing pathways that appear to be necessary for specific time windows during development. Our in vivo approach demonstrates that the abrogation of miRISC interaction with GW182 proteins does not cause embryonic lethality as seen in animals lacking miRNA-specific Argonautes, ALG-1 and ALG-2. These data are in striking contrast with the severe developmental phenotypes observed in those animals after hatching. Given that GW182 proteins are not essential for miRISC-mediated regulation during animal embryogenesis, the miRNA-mediated gene silencing may preferentially block translation, instead of deadenylation and mRNA degradation that requires the recruitment of GW182 proteins to the target mRNAs. This model is reminiscent of the observation that gene silencing can occur independently of mRNA deadenylation during zebrafish embryogenesis [46]. Since GW182 proteins are essential and sufficient for mRNA deadenylation and translational repression, the silencing complex without GW182 must limit the turnover rate of the bound mRNAs during embryogenesis. In this case, the dwell time of miRISC to discriminate the proper target mRNAs would be extended, and could be tested by single molecule studies.
Our study discovered that ALG-1 functions, without the aid of AIN-1 and AIN-2, as an essential factor in early stages of development in C. elegans. This sheds light on the enigmatic miRNA-mediated gene silencing during embryogenesis in animals, which bypasses a regular gene silence pathway that requires GW182 proteins. However, we cannot exclude a possibility that such a GW182 protein-free miRISC plays another unidentified, but critical role during embryogenesis, in addition to gene silencing. We believe that this study lays a strong foundation and experimental context for future studies to understand how and why miRNA-mediated gene silencing pathways are varied at different developmental stages, and how each pathway involves GW182 proteins.
All C. elegans strains were cultured and handled using standard methods. The transgenic strains were generated by MosSCI single insertion method [47, 48] or by extrachromosomal non-integrated transgene expression [49]. The strategies to build plasmids as well as strains for this study are listed in S1 Text.
E.coli BL21 codon+ cells transformed with pGEX plasmids encoding a GST-fusion construct were grown at 37°C. After adding 0.1mM Isopropyl β-D-1-thiogalactopyranoside (IPTG) at OD600 = 0.8, the E. coli cells were grown at 15°C for 16h. Harvested cells were resuspended in STE buffer (10mM Tris pH8, 150mM NaCl, 1mM EDTA, 5mM DTT, 1mM PMSF) supplemented with 2.5% (w/v) of N-Lauryl-Sarkosyl and lysed at 27kPSi in a constant cell disruptor (One Shot Cell Disruptor, Constant System). 1.5% (v/v) final Triton X-100 was added into the cell lysis. GST-tagged ALG-1 proteins were purified under non-denaturing conditions by affinity chromatography using Glutathione Sefinose matrix and quantified on a 8% SDS-PAGE gel.
The pulldown assay was carried out by mixing 100ng of GST::ALG-1 resin coupled protein with 500ng of AIN-1 fragment in binding buffer (100mM Potassium Acetate, 30mM Hepes-KOH pH7, 2mM Magnesium Acetate, 1.5% Triton X-100, 1mM DTT, 1 tablet/10mL Complete Mini Protease Inhibitor without EDTA (Roche)). The mix was incubated for 1h at 4°C with gentle rotation and washed 2 times with high salt PBS buffer (300mM then 500mM NaCl) followed by a non-stringent final wash with PBS only. Beads were resuspended into 2X denaturing Laemmli buffer and loaded on a 4–15% SDS PAGE gel. The upper part containing GST-tagged ALG-1 was stained with Coomassie Brilliant Blue whereas the lower part was immunoblotted with a primary rabbit polyclonal AIN-1 antibody (dilution 1:1000).
Staged young adults worms were obtained by Alkaline Hypochlorite Solution treatment and plated onto NGM Agar plates seeded with OP50 bacteria. After 4 days at 15°C, animals were harvested in M9 solution and lysed by sonication into ice-cold lysis buffer (100mM KAc, 30mM Hepes-KOH pH7, 2mM Magnesium Acetate, 1mM DTT, 1.5% triton X-100, 1 tablet /10mL Complete Mini Protease Inhibitor without EDTA (Roche)). Immunoprecipitation and miRISC pull-down assays were carried out as described in [50, 51], respectively. The 2′-O-methyl oligonucleotides sequences have been previously described in [23]. Primary rabbit polyclonal ALG-1 and AIN-1 antibodies were used at 1:1000 dilution in PBST supplemented with 5% of milk with overnight incubation at 4°C. AIN-1 and AIN-2 antibodies were generated by injection of two rabbits with either AIN-1 peptide (EQRAPASTEDYHYS) or AIN-2 peptide (GPPDHYYDYSFLG) and affinity purified using the same epitope (Feldan).
RNA preparation and microRNA quantification by quantitative RT-PCR were performed as described in [50]. To quantify the level of microRNA bound to ALG-1, 4mg of total protein extract was used to immunoprecipitate ALG-1. A fraction of 10% was mixed to Laemmli denaturing buffer and loaded on 8% SDS-PAGE. 90% of the remaining beads were treated with 20μg of proteinase K and RNA was extracted using TriReagent (Sigma). Samples were spiked and normalized with 50fmol of human synthetic miR-20a as technical control.
The RNAi of alg-2 and ain-2 were carried out by feeding using cDNA fragment cloned into RNAi feeding vector L4440 and expressed into inducible IPTG HT115 (DE3) bacterial strain as described in [52]. The oligonucleotides used to generate the plasmids as well as the different plasmids are listed in S1 Table and S2 Table, respectively.
DIC Nomarski images and GFP, mCherry fluorescence expressions were collected in animals using a Zeiss AxioCam HRm digital camera mounted on a Zeiss Axio Imager M1 microscope using the same settings for each animal. Intensity of fluorescence in pharynx was measured with Axiovision 4.6 software.
|
10.1371/journal.pcbi.1003422 | The Brain Ages Optimally to Model Its Environment: Evidence from Sensory Learning over the Adult Lifespan | The aging brain shows a progressive loss of neuropil, which is accompanied by subtle changes in neuronal plasticity, sensory learning and memory. Neurophysiologically, aging attenuates evoked responses—including the mismatch negativity (MMN). This is accompanied by a shift in cortical responsivity from sensory (posterior) regions to executive (anterior) regions, which has been interpreted as a compensatory response for cognitive decline. Theoretical neurobiology offers a simpler explanation for all of these effects—from a Bayesian perspective, as the brain is progressively optimized to model its world, its complexity will decrease. A corollary of this complexity reduction is an attenuation of Bayesian updating or sensory learning. Here we confirmed this hypothesis using magnetoencephalographic recordings of the mismatch negativity elicited in a large cohort of human subjects, in their third to ninth decade. Employing dynamic causal modeling to assay the synaptic mechanisms underlying these non-invasive recordings, we found a selective age-related attenuation of synaptic connectivity changes that underpin rapid sensory learning. In contrast, baseline synaptic connectivity strengths were consistently strong over the decades. Our findings suggest that the lifetime accrual of sensory experience optimizes functional brain architectures to enable efficient and generalizable predictions of the world.
| While studies of aging are widely framed in terms of their demarcation of degenerative processes, the brain provides a unique opportunity to uncover the adaptive effects of getting older. Though intuitively reasonable, that life-experience and wisdom should reside somewhere in human cortex, these features have eluded neuroscientific explanation. The present study utilizes a “Bayesian Brain” framework to motivate an analysis of cortical circuit processing. From a Bayesian perspective, the brain represents a model of its environment and offers predictions about the world, while responding, through changing synaptic strengths to novel interactions and experiences. We hypothesized that these predictive and updating processes are modified as we age, representing an optimization of neuronal architecture. Using novel sensory stimuli we demonstrate that synaptic connections of older brains resist trial by trial learning to provide a robust model of their sensory environment. These older brains are capable of processing a wider range of sensory inputs – representing experienced generalists. We thus explain how, contrary to a singularly degenerative point-of-view, aging neurobiological effects may be understood, in sanguine terms, as adaptive and useful.
| Aging is generally thought to be accompanied by reduced neuronal plasticity and a loss of neuronal processes that accounts for a loss of grey matter, which progresses gently with age [1]–[3]. Many concomitants of physiological aging have been studied. In particular, studies of the mismatch negativity (MMN) speak to a decline in sensory learning or memory [4], [5]. For example, elderly subjects show a significant reduction in superior temporal gyrus responses, which has been interpreted as “an aging-related decline in auditory sensory memory and automatic change detection” [6]. In this work, we examine the physiological basis of attenuated mismatch responses using dynamic causal modeling in a large cohort of human subjects. However, we motivate the present study using an alternative – and slightly more optimistic – model of normal aging.
Our basic premise is that aging reflects a progressive refinement and optimization of generative models used by the brain to predict states of the world – and to facilitate an active exchange with it. Evidence that the brain learns to predict its environment has been demonstrated in the perceptual [7], motor [8] and cognitive domains [9]. These studies are motivated by formal theories – such as the free energy principle and predictive coding – that appeal to the Bayesian brain hypothesis [10]–[14]. In this theoretical framework [12], the quality of the brain's model is measured by Bayesian model evidence. Crucially, model evidence can be expressed as accuracy minus complexity. This means that as the brain gets older – and maintains an accurate prediction of the sensorium – it can progressively improve its performance by decreasing its complexity. This provides a normative account for the loss of synaptic connections and fits intuitively with the notion that as we get older we get wiser, more sanguine and ‘stuck in our ways’. Formally, under the Free Energy Principle, the brain supports active exchanges with the environment in order to minimize the surprise associated with sensory inputs. Over time, learning optimizes brain connectivity to support better predictions of the environment [15]. These ‘better’ models must conform to Occam's razor by providing accurate predictions with minimal complexity [16]. In Figure 1 we illustrate model qualities prescribed by the Free Energy Principle, potential age effects and their context or environmental sensitivity. This formulation of Free Energy minimization is based on hierarchical message passing and predictive coding. Neuronal implementations of predictive coding have been proposed as the mechanisms underlying the MMN [17], [18]. In the present study, we address the corollary of model complexity minimization; namely, less reliance on Bayesian updating through sensory learning and underlying neuronal plasticity. Mathematically, an attenuation of Bayesian learning precludes overfitting of sensory data; thereby minimizing complexity and ensuring that explanations for those data generalize. In other words, as we age, we converge on an accurate and parsimonious model of our particular world (Figure 1B) - whose constancy we actively strive to maintain (Figure 1B). Its neuronal implementation would be consistent with a large literature on synaptic mechanisms in aging and a progressive decline in neuromodulatory (e.g., dopaminergic [19], [20]) activity that underwrites changes in synaptic efficacy [21].
The implications for the neurobiology of aging are that – over the years – cortical message passing may become more efficient (providing accurate predictions with a less redundant or complex hierarchical model) and increasingly dominated by top-down predictions. This is consistent with reports of age-induced shifts in neuronal activation from sensory to prefrontal regions [22]. The hypothesis addressed in the present study was that the Bayesian updating implicit in the sensory learning of standard stimuli in the MMN paradigm would fall progressively with age. In particular, we predicted that changes in effective connectivity during the processing of repeated stimuli (namely, changes in forward connections to superior temporal cortex) would be attenuated as a function of age.
Here, we examined age-related attenuation of sensory learning by quantifying synaptic coupling or effective connectivity changes using the mismatch negativity (MMN) paradigm and dynamic causal modeling (DCM). There is a large literature on DCM and the MMN [23]–[25], where changes in coupling during repetition of standard stimuli are revealed by differential responses to oddball stimuli – producing the MMN (oddball minus standard) difference in event related potentials that peaks around 150 msec. These connectivity changes (plasticity) are expressed in both intrinsic connections within auditory sources and in an increase in the effective connectivity from auditory to superior temporal sources during the processing of oddball relative to (learned) standard stimuli [26]. These changes have been interpreted in terms of predictive coding, in which bottom-up or ascending prediction errors (under modulatory gain control) adjust representations at higher levels in the cortical hierarchy – that then reciprocate descending predictions to cancel prediction error at lower levels.
Recent studies of age-related changes in functional connectivity provide evidence for changes in long-range coupling with age [27]. Our hypothesis rests on changes in (directed) effective connectivity that produces the functional connectivity or dependencies in measured activity [28]. To quantify changes in effective connectivity we used DCM [29] to model magnetoencephalographic (MEG) recordings. DCM uses forward models of evoked responses based on neuronal mass formulations that account for the laminar specificity of forward and backward connections [29]. These models have been previously validated using animal [30] and human recordings [31], and provide subject-specific measures of intrinsic (within source) and extrinsic (between source) synaptic coupling.
We measured event-related MEG responses in 97 subjects, aged 20 to 83 and applied DCM to quantify the underlying synaptic coupling producing observed responses. We used an auditory oddball paradigm to elicit the mismatch negativity or MMN [23]. Our stimuli comprised pseudo-random tone sequences, with standard (frequent) tones interspersed with infrequent oddball tones (with a presentation frequency of 88% and 12% respectively). Consistent with previous studies of MMN generation [32], [33], source localization revealed hierarchical responses (Figure 2A), with large magnitude responses in auditory, temporal and inferior frontal sources (p<0.05 family-wise error corrected; Figure 2A). A prominent MMN (oddball – minus standard) was observed, as expected, around 150 msec post stimulus (Figure 2B).
Following previous DCM studies of the MMN, we used a six–source model to characterize age effects within the MMN network (Figure 2C). For each subject, we inverted the ensuing DCM to obtain subject-specific measures of (changes in) connectivity based on their evoked responses to standards and oddballs. In this DCM, auditory input enters bilaterally at Heschl's gyrus (HG), these primary auditory sources were connected via forward connections to superior temporal gyrus (STG) sources, which in turn sent forward connections to the inferior frontal gyrus (IFG). Reciprocal backward connections were included to allow signal propagation down the hierarchy from IFG to STG and from STG to HG (Figure 2C). Each source was modeled with a neural mass model comprising three neuronal populations, with distinct receptor types and intrinsic connectivity [31]. Specifically, the model contains synaptic parameters that encode the contribution of AMPA, NMDA and GABAa receptor mediated currents in three populations: comprising pyramidal cells, inhibitory interneurons and granular-layer spiny-stellate cells. These populations are connected intrinsically and receive extrinsic inputs according to their laminar disposition: forward connections drive spiny stellate cells and backward connections drive pyramidal cells and inhibitory interneurons [29]. Crucially, we included stimulus-specific parameters that changed the strength of extrinsic connections when responding to standard and oddball inputs. This enabled us to test our hypothesis of age-related differences in connectivity changes. Specifically, we hypothesized that the learning or repetition-dependent increase in sensitivity to extrinsic forward afferents – conveying prediction errors induced by the oddball events – would be attenuated in older subjects.
An analysis of model fits confirmed that DCM provided an accurate account of the evoked responses (193 data sets were inverted in total), accounting for 81%±12% (mean ± std) of the empirical variance (for a representative example see Figure 3A). We found no evidence for age-dependent differences in model fit (p>0.1, Pearson correlation of age and proportion of variance explained).
Having established the accuracy of the DCM, we then asked whether the subjects' age could be predicted by neuronal parameters that included: i) the strength of forward and backward extrinsic connections, ii) changes in these connections during oddball (compared to standard) tones, iii) the strength of intrinsic connections within each source, iv) parameters controlling synaptic adaptation; namely, time constants of AMPA, NMDA and GABAa receptors, membrane capacitance, subcortical input strength and axonal delays (37 parameters and a constant term see Table 1). Electromagnetic lead field parameters were optimized for each DCM but not included in this predictive analysis (see Methods).
Using a multiple linear regression, we found that the neuronal DCM parameters could predict age with a high degree of reliability (R2 = 0.56; F37,59 = 2.06; p = 0.006; Figure 3B). Post-hoc t-tests were used to identify the parameters with the greatest predictive ability. Across all regression coefficients, the largest and only significant regression coefficient (correcting for 38 tests) was associated with the learning dependent increase in forward connectivity from the right primary auditory cortex to the right superior temporal gyrus (β = −36.41; p<0.05 Bonferroni corrected, Figure 3C). This increase was attenuated over the lifespan, speaking to a reduced sensitivity of STG responses to ascending (prediction error) afferents from primary auditory cortex. This was in contradistinction to the latent connectivity strengths from right primary auditory to superior temporal gyrus - that do not reflect learning – which were consistent across the lifespan population (Figure 3D).
The Free Energy Principle [11] provides a description of neurobiological circuit processing that attributes specific computational roles to forward, backward, lateral (extrinsic) connections and intrinsic connections and their neuromodulation [34]. Each level of a processing hierarchy transmits predictions to the level below, which reciprocates with bottom-up prediction errors. Bayes optimal perception and action is achieved by maximising the Negative Free Energy (F):(1)Maximising this functional at every point in time ensures homeo/allostasis [12], by minimising the surprise (the negative log model evidence ) of incoming sensory signals s caused by states of the world , represented in the brain with their sufficient statistics μ. It renders the current prediction of states of the environment; , close to the true probability of those states; (where the distance measure is the Kullback-Leibler divergence KL). This process is dependent on the model the brain instantiates, m. Rearranging this equation, we see that the quality of this model can be decomposed into two components; representing accuracy and complexity.(2)In our connectivity analysis, the only consistent aging effect was manifest in trial-by-trial updates and revealed during the presentation of the oddball. This is represented mathematically as the KL-divergence from the approximate posterior to the prior, ie. the complexity penalty; which reduced over the lifespan (Figure 1). Over-learning of the standard tone by younger subjects is indicative of brittle models. These effects were significantly less pronounced as our cohort (cross-sectionally) aged. In contrast, accuracy was equivalent across the lifespan on a trial-average basis, since younger subjects learned the standard tone; indicating poor predictions to early standard and all deviant tones with better predictions to later standard tones; while age induced greater baseline predictions overall, that were generalizable to auditory deviants.
These results are interesting for two reasons. First, the ability of subject-specific DCM parameters to predict age in such a reliable way suggests that the coupling estimates have a high degree of predictive validity. Second, it is remarkable that the most predictive parameter encoded a sensory learning effect – as opposed to a connection engaged by the predictive coding of standard or oddball stimuli per se. Furthermore, the particular connection implicated – the forward primary auditory afferent to STG – has been found to increase in previous DCM studies of the MMN [26]. The present study is the largest DCM study reported to date and underscores a general point; namely, that biologically grounded models of evoked responses can disclose important associations between quantitative estimates of functional brain architectures and the behavioral or clinical phenotype. In particular, we used our data to estimate the underlying causes of evoked responses – and did not simply look for correlations between age and a particular data feature (e.g., the MMN magnitude). This means that we could account for a range of potentially age-related confounds (e.g., intersubject differences in lead fields) that would otherwise obscure structure-function relationships of interest.
In conclusion, our results suggest that effective connectivity in the human brain does not undergo indiscriminate age-related decline but shows a selective and specific attenuation of plasticity in the face of short-term sensory learning or memory. In other words, there were no systematic age-related changes in effective connectivity when processing auditory stimuli per se. This is consistent with the conjecture that older brains are more efficient (less complex) models of the sensorium and are less predisposed to short-term (Bayesian) updating.
The present study was motivated by recent perspectives provided by theoretical neurobiology [12] that offer a principled explanation for the reduction in connectivity (complexity) with progressive optimization of the generative models the brain uses for hierarchical Bayesian inference. A corollary of this complexity minimization is decreased Bayesian updating and neuroplasticity that we confirmed experimentally with a sensory learning (oddball) paradigm. Our results may call for a reinterpretation of aging neuroimaging studies; in particular, the compensation hypothesis that has been provided as explanation for age-related changes in the pattern of cortical activations [22], [35], [36]. Indeed, a reinterpretation has been offered from a cognitive perspective [37] where a shift from bottom-up to top-down processing has been proposed to explain better cognitive performance in older individuals [38]. These performance gains have been shown to accrue in unconventional (generalized) re-test circumstances; e.g. using distractors that should have been ignored in one task, to complete later tasks [39]. From the perspective of task performance, complexity reduction would similarly support reliability, as exhibited by older participants in a recent study of performance consistency across multiple cognitive domains [40]. The complexity minimization perspective may also account for de-differentiation in cortical specialization [41]–[43] and cognitive structure [44], [45] due to age - in the sense that simpler generative models require fewer degrees of freedom (functional specialization) to predict sensorimotor contingencies. While our results focus on functional connections, structural changes commensurate with complexity reduction have recently been demonstrated in a non-aged but practiced cohort of ballerinas. In their study [46], highly trained ballet dancers - who show improved stability in response to spinning - exhibited grey matter reductions in cerebellar grey matter compared to controls. Furthermore, controls showed enhanced vestibular perception that was positively correlated with cortical white-matter measures, an effect absent in the dancers, effects summarized by the authors as “training-related attenuation”.
Interestingly, the schema presented in Figure 1 was supported by learning effects in early sensory cortex. These were constrained to the right hemisphere, where classical MMN effects are most pronounced [47]. Complexity reduction could potentially evolve over the lifespan, providing a balance of metabolic cost [48] to allow for an elaboration of model components in multi-modal regions. It could also contribute directly to the poor discriminability of (unimodal) sensory inputs observed in older adults [49], which in turn may preface as a ‘common cause’, age-related cognitive disruption [50].
From a physiological perspective, predictive coding may provide a useful process theory for neuronal computations in aging. For example, simulations of the mismatch negativity paradigm predict a rapid trial-by-trial suppression of evoked responses that rests on the neuromodulation of superficial pyramidal cells reporting prediction error. Previously, we confirmed this prediction empirically using dynamic causal modeling and a placebo-controlled study of cholinesterase inhibition [18]. In a complementary simulation study of frequency-based MMN, NMDA mediated synaptic plasticity has been shown to underpin model reorganization at the predictive cell population [17]. Given the therapeutic benefit of cholinesterase inhibition [51], and the role of NMDA receptors [52] in dementia further modeling of non-invasive psychopharmacological studies may provide important insights into the synaptic basis of age-related changes in perceptual processing.
We studied 97 healthy volunteers, 55 female, who were cognitively normal with no neurological or psychiatric illness or serious medical history. Subjects were aged 20 to 83 and all completed the recording paradigm.
Subjects were paid for their participation and consented to all procedures, which were conducted in accordance with the Declaration of Helsinki (1991). Protocols were approved by the South-East Strategic Health Authority Regional NHS Ethics Committee.
MEG recordings were made in a magnetically shielded room using a 275-channel CTF system with SQUID-based axial gradiometers (VSM MedTech Ltd., Couquitlam, BC, Canada). Recordings were obtained during two sessions with a small rest period between scanning, during which time subjects remained in the MEG scanner. Head localisation was performed at the beginning of each session.
Auditory responses were elicited by stimuli comprising pure tones presented binaurally over headphones. Two stimuli, at 500 Hz and 800 Hz were presented in a pseudo-random sequence for 70 msec with 10 msec rise and fall times. The first tone served as the standard and was presented on 88% of trials, while the second, which served as the oddball, was presented on 12% of trials. The sequence ensured that the minimal interval between oddballs was 2 trials and the maximum was 25 trials. The ISI was fixed at 1100 msec. Loudness was adapted to each subject's auditory threshold to be clearly audible binaurally – as measured in a test run while in the scanner. We collected data over two sessions for 96 subjects. For one subject we recorded just one session. Sessions were 6 minutes in length.
MEG data were first filtered off-line (band-passed from 0.5–30 Hz), down-sampled (to 200 Hz), epoched (from −150 ms to 350 ms peri-stimulus time), baseline corrected to 0 ms peristimulus time, artefact corrected (peak-to-peak threshold 5pF) and averaged to obtain event related fields (ERFs). The analysis routines we used are available in the academic freeware SPM8 (http://www.fil.ion.ucl.ac.uk/spm/).
For source localization, multiple sparse priors were used to estimate the cortical sources of the sensor recordings, using standard settings [53]. Multiple sparse priors employs several hundred patches of activation that are iteratively reduced until an optimal number and location of active patches are found using a greedy Bayesian search. A tessellated cortical mesh set in canonical Montreal Neurological Institute (MNI) anatomical space – as implemented in SPM8 – served as a brain model [54]. This dipole mesh was used to calculate the forward solution using a spherical head model. Source activity measures were then interpolated into MNI voxel space and analysed using statistical parametric mapping – at the between subject level – using an F test: A contrast of standard vs deviant stimuli was computed at p<0.05 family-wise error corrected (Figure 2) based on the evoked power over frequencies from 0–30 Hz and from 60 to 300 msec peristimulus time.
For dynamic causal modeling, we used source location priors as described in previous DCM analyses of the mismatch negativity (MMN) paradigm [23], [25]. These included sources in Heschl's gyrus, superior temporal cortex and inferior frontal gyrus and were consistent with the source localisation analyses. The MNI coordinates were as follows: left HG: x = −42, y = −22, z = 7; right HG: x = 46, y = −14, z = 8; left STG: x = −61, y = −32, z = 8; right STG: x = 59, y = −25, z = 8; left IFG: x = −46, y = 20, z = 8; right IFG: x = 46, y = 20, z = 8. These prior locations were optimised at an individual level during DCM inversion using distributed dipoles and the forward solution from the above source localisation [55].
In DCM, event related fields are modelled as the response of a dynamic input–output system to exogenous (experimental) inputs [29]. The DCM generates a predicted ERF as the response of a network of coupled sources to sensory (thalamic) input – where each source corresponds to a neural mass model of three neuronal populations. Our dynamic causal models comprised hierarchical sources with prior locations as defined above, extrinsic input to primary sensory regions and extrinsic connections of forward and backward type [56]:
MEG sensor data were fitted over 0–300 msec peristimulus time, with the following model: auditory input (modelled as a Gaussian bump-function, with a prior onset of 64 msec) entered bilateral Heschl's gyrus, which provided forward connections to STG within each hemisphere. STG sent top-down backward connections to HG. STG also sent forward connections up to IFG and received backward type connections from IFG. To accommodate trial-dependent differences, stimulus specific parameters were included for all extrinsic connections. The neural mass model describing the activity of each source comprised three subpopulations, each assigned to three cortical layers – which determine how they receive external inputs [56]. Spiny stellate cells receive input via forward and thalamic inputs and are located in layer IV. Pyramidal cells and inhibitory interneurons are located outside of layer IV. These receive inputs from backward connections. Extrinsic output cells are the pyramidal cell subpopulation in each region.
The neuronal dynamics were based on a conductance based model with intrinsic AMPA receptors (at all cell populations), GABAa receptors (at pyramidal cell populations and inhibitory interneurons) and NMDA receptors (at pyramidal cell populations and inhibitory interneurons) [57] (specified as the “NMDA” model in the SPM interface). The DCM generates a predicted ERF as the response of the network of coupled sources to sensory input. This input takes the form of a narrow (16 msec) Gaussian impulse function, which accounts for some temporal smoothing in thalamic volleys.
For computational expediency, DCMs were computed following dimensionality reduction to eight channel mixtures or spatial modes. These were the eight principal modes of a singular value decomposition (SVD) of prior predictive covariance based upon the prior source locations. Note that data are normalized prior to model inversion and the forward model which accounts for source transmission to the MEG sensors is also parameterised and optimised during inversion.
Where data were collected over multiple trial runs (96 out of 97 subjects), DCMs were fitted for each run separately and post-hoc conditional parameter means were computed using Bayesian parameter averaging (BPA). These were used for the regression models and lifespan correlation. BPA involves a weighted average where each model's posterior mean (in DCM.Ep) is weighted with its relative precision, where precisions are obtained from the inverse of the posterior covariance.
|
10.1371/journal.ppat.1004043 | Concerted Spatio-Temporal Dynamics of Imported DNA and ComE DNA Uptake Protein during Gonococcal Transformation | Competence for transformation is widespread among bacterial species. In the case of Gram-negative systems, a key step to transformation is the import of DNA across the outer membrane. Although multiple factors are known to affect DNA transport, little is known about the dynamics of DNA import. Here, we characterized the spatio-temporal dynamics of DNA import into the periplasm of Neisseria gonorrhoeae. DNA was imported into the periplasm at random locations around the cell contour. Subsequently, it was recruited at the septum of diplococci at a time scale that increased with DNA length. We found using fluorescent DNA that the periplasm was saturable within minutes with ∼40 kbp DNA. The DNA-binding protein ComE quantitatively governed the carrying capacity of the periplasm in a gene-dosage-dependent fashion. As seen using a fluorescent-tagged derivative protein, ComE was homogeneously distributed in the periplasm in the absence of external DNA. Upon addition of external DNA, ComE was relocalized to form discrete foci colocalized with imported DNA. We conclude that the periplasm can act as a considerable reservoir for imported DNA with ComE governing the amount of DNA stored potentially for transport through the inner membrane.
| Bacterial transformation is the import and inheritable integration of external DNA. As such, it is believed to be a major evolutionary force. A key step is the import of DNA through the outer membrane. Here, we have characterized the spatio-temporal dynamics of DNA during import and residence in the periplasm of the Gram-negative pathogen Neisseria gonorrhoeae. We found that the periplasm can serve as a reservoir for imported DNA that can fill within five minutes by importing DNA from the environment. The amount of imported DNA roughly corresponds to the size of a phage genome. The periplasmic DNA-binding protein ComE is homogeneously distributed in the periplasm in the absence of extracellular DNA. It relocates rapidly to imported DNA when external DNA is added to competent gonococci. As ComE governs the carrying capacity of the periplasm, we propose that it might condense DNA, thus linking DNA uptake to its compaction. Although the import through the outer membrane was localized all around the cell contour, the major part of the imported DNA relocated to the septum at the center of diplococci. Our findings strongly support the idea that the periplasm masses DNA independently of transport through the inner membrane.
| Natural competence for transformation is widespread among different bacterial species [1]. Transformation is thought to speed up adaptive evolution but it is also discussed in the context of genome maintenance [2] [3]. The currently available data from Gram-positive species strongly supports the idea of a coordinated DNA transformation machine that binds DNA at the extracellular side, powers translocation of DNA through the cell envelope and hands the DNA over to the recombination machine at the intracellular side [4] [5]. With the only known exception of Helicobacter pylori, all characterized naturally competent species are associated with the type IV pilus (T4P) system for DNA import. At the extracellular side, T4P proteins are essential for DNA binding although it is unclear whether long pilus filaments are necessary [1]. Whereas DNA binding to the competence pilus has been demonstrated in Streptococcus pneumoniae [6], Neisseria gonorrhoeae that generate non-retractile T4P show strongly impaired binding efficiency [7]. In the following, the nomenclature of N. gonorrhoeae will be used to describe the proteins required for transformation. The major pilin subunit is essential for binding and import of DNA [8] [7] but replacing the gonococcal pilin PilE by the major subunit of Pseudomonas aeruginosa or of Fracisella tularensis supports DNA import and transformation as well [9], [10]. The transformation rate is modulated by the relative levels of the minor pilins ComP. PilV acts antagonistically at the level of DNA binding with ComP increasing transformability in a dose-dependent fashion and PilV decreasing it [7], [11]. PilV appears to exert its inhibitory effects by competing with ComP for access to the Tfp assembly machinery [11]. The presence of a 12 bp DNA Uptake Sequence (DUS) strongly enhances the probability for DNA-import by N. gonorrhoeae [12] [3]. ComP binds DNA in a sequence-specific manner, selecting for DNA containing the DUS [13] [7]. The outer membrane channel formed by PilQ is essential for T4P extrusion and DNA import into a DNase-resistant state and moreover shows DNA-binding potential [14], [15]. In the periplasm, three components are linked to transformation. The DNA-binding protein ComE has four identical gene-copies on the gonococcal genome [16]. Gradual deletion of these copies leads to gradual decrease in transformation rate by decreasing the probability for DNA import [16]. The DNA-binding peptidoglycan-linked lipoprotein ComL and the lipoprotein tetrapac (Tpc) which is associated with separation of dividing diplococci are not essential for DNA uptake but for transformation [17] [18]. ComA proteins form the channel through which DNA is transported from the periplasm to the cytoplasm [19]. In the Gram-positive species Bacillus subtilis it has been shown that incoming ssDNA is immediately coated by single-strand binding proteins [20]. Single strand binding proteins have been proposed to generate a reservoir of ssDNA in the cytoplasm and to direct the DNA to homologous recombination in B. subtilis and Streptococcus pneumoniae [21] [4]. For N. gonorrhoeae there is evidence that ssDNA forms transiently in the periplasm [22].
DNA import has been characterized at the single molecule level for B. subtilis and H. pylori [23] [24] [25] [26]. DNA uptake in B. subtilis proceeds at a rate of 80 bp/s at low external forces. Application of force using laser tweezers showed that the import was irreversible for forces up to 50 pN. The speed of DNA import was considerably larger in H. pylori with 1.3 kbp/s at low force. Application of 23 pN or more triggered extrusion of the previously imported DNA here, demonstrating the import into the periplasm is reversible. Transport through the outer and the inner membrane of H. pylori are not coupled in time. H. pylori imports fluorescently labeled Cy3-DNA into the periplasm at a somewhat decreased rate relative to unlabelled DNA [24]. Like the case of the ComEC channel based system in B. subtilis, the H. pylori ComEC-based system does not support transport of Cy3-DNA into the cytoplasm [24], suggesting that the processes for inner membrane transport are analogous between these competent species [27].
In rod-like B. subtilis, DNA uptake complexes form at the poles or at the growing septa [28] [20]. They are stable even when the cell wall is disrupted [29]. Accordingly, DNA uptake proceeds preferentially from the cell poles. The homologue of gonococcal ComE, ComEA, distributes homogeneously around the cell contour and is thought to enhance initial DNA binding to the cell surface [30] [31]. H. pylori accumulates imported Cy3-DNA primarily at the cell poles and at the septa [24]. Very recently, it has been shown that competent Gram-positive S. pneumoniae recruits Cy3-DNA and ComEA at midcell location [32]. It is unknown, however, where DNA uptake occurs in Gram-negative cocci.
Here, we visualized the spatio-temporal dynamics of DNA during import from the environment into the periplasm and within the periplasm of N. gonorrhoeae. We validated the approach of using fluorescent Cy3-DNA by characterizing a number of mutant backgrounds that had been shown to be impaired for DNA uptake. We found that DNA import through the outer membrane occurred all around the cell contour. The periplasm was saturable with DNA and held a considerable amount of Cy3-DNA. With short fragments, saturation occurred on a time scale of minutes. The periplasmic DNA-binding protein ComE strongly affected the carrying capacity of the periplasm, suggesting a role in either removing DNA from the uptake machine or in compacting DNA.
To investigate whether Cy3-DNA could be used to study DNA uptake in gonococci, we incubated gonococci with a 3 kbp fragment of Cy3-DNA containing one DNA uptake sequence (DUS) for 30 min and subsequently treated them with DNase. DNA was labeled randomly along its entire contour. Throughout this work, we used a recAind background without induction to prevent the cells from antigenic variation of pilE. The recAind strain will be labeled wt in the following, since it shows wt behavior in terms of DNA import through the cell envelope. Wt cells were associated with fluorescent foci that were clearly distinguishable from the fluorescence background (Fig. S1a), indicating that Cy3-DNA was imported into a DNase-resistant state.
The minor pilins ComP and PilV have been shown to strongly affect DNA uptake at the level of DNA binding [7], [11]. We investigated the effect of these minor pilins on the uptake of Cy3-DNA. We performed the DNA uptake assay in a ComP overproducing strain (PpilEcomP) and in a pilV deletion strain (ΔpilV) (Fig. 1a). As expected, both strains showed a strongly increased fluorescence signal (Fig. S1b–d), indicating that more Cy3-DNA was imported within 30 min. Since the high signal of the ΔpilV strain was very convenient for data analysis, we performed all following experiments in this background.
Up until now, DNA uptake by N. gonorrhoeae has been investigated at the population level. Therefore, it is unclear whether there is heterogeneity in this process at the level of single cells. To investigate any potential for heterogeneity, we measured the distribution of DNA uptake efficiencies of individual cells. We incubated gonococci with the 3 kbp Cy3 fragment for 30 min, treated the cells with DNase and subsequently quantified the fluorescence intensity of individual cells (Fig. S2). We found that the distribution of fluorescence intensities was very broad (Fig. 1e). A fraction of cells showed no import of DNA and this fraction was variable between different experiments. Most of the cells in our samples were diplococci. For our analysis we did not distinguish between monococci and diplococci, as the difference in the histograms was visible but small (Fig. S3).
To determine the background level of fluorescence, we repeated the experiment using a strain with a deletion in the retraction ATPase PilT (Fig. 1b) which is unable to import DNA [7]. Furthermore, we investigated a pilQ deletion strain that does not form the outer membrane pore, is deficient in DNA uptake, and shows reduced DNA binding (Fig. 1c). We found that both mutations strongly shifted the distribution of fluorescence intensities to lower values (Fig. 1e). Although the distributions were very broad we calculated the average fluorescence values (Fig. 1f). We found that in the pilT pilV background strain, the amount of DNase-resistant Cy3-DNA was reduced by a factor of 12 as compared to the ΔpilV strain and in the ΔpilQΔpilV strain the amount of DNase-resistant Cy3-DNA was reduced by a factor of 22. The strong decrease in fluorescence is consistent with previous experiments using radioactively labeled DNA although the factors are considerably lower [7], [11], suggesting that the Cy3-DNA methodology has reduced sensitivity.
Next, we investigated whether import of Cy3-DNA into a DNase-resistant state was dependent on the putative inner membrane channel formed by ComA (Fig. 1d). When repeating the DNA uptake experiment using a comA deletion strain (ΔcomAΔpilV) we found that the fluorescence distribution was not significantly different from the ΔpilV strain (Fig. 1e, f), indicating that the imported DNA accumulated in the periplasm. This result was analogous to what was observed previously for H. pylori [24].
Transformation of N. gonorrhoeae depends on the DNA uptake sequence (DUS). ComP has been shown to be a positive effector of sequence-specific DNA binding and that it is directly involved in binding of DUS-containing DNA [7] [13]. The DNA uptake assay with fragments lacking the DUS showed strongly reduced fluorescence (Fig. 1e, f). Thus, the uptake of Cy3-DNA was strongly enhanced by the DUS.
We conclude that Cy3-DNA is imported into the periplasm of gonococci and that DUS-related import is dependent on the outer membrane channel formed by PilQ and on the T4P retraction ATPase PilT, but not on the inner membrane channel ComA. Since these results are consistent with previous DNA uptake studies using radioactively labeled DNA, they validate our method for studying DNA uptake using N. gonorrhoeae Cy3-DNA. In contrast to the classic method using radioactively labeled DNA, our approach enables us to study DNA uptake at the single cell level and thus reveals strong cell-to-cell variability in terms of the total amount of DNA imported after 30 min.
We investigated whether the periplasm acts as a reservoir for imported DNA. ΔpilV cells were incubated with Cy3-DNA for 1 h, treated with DNase and subsequently single cell fluorescence was measured. The total fluorescence intensity per cell did not vary when incubated with Cy3-DNA with fragment lengths of 0.3 kbp, 1 kbp, and 10 kbp at equimolar concentrations (Fig. 2), indicating that the total amount of imported DNA was independent of fragment length. To convert fluorescence intensity into amount of DNA, we quantified the fluorescence intensity of individual 6 kbp fragments (Fig. S4) [33] and compared them to the total fluorescence of individual cells that were incubated with Cy3-DNA from the same labeling reaction for 1 h. We found that the periplasm contained ∼40 kbp of Cy3-DNA. We note that this value might be biased slightly by binding of proteins to the DNA in the periplasm [34]. Furthermore, the Cy3-labeling efficiency is somewhat variable and therefore the fluorescence intensity cannot be directly compared to other experiments using different Cy3-DNA stocks.
Next, we addressed the question whether the periplasm can be saturated with taken up DNA. To this end, we incubated gonococci for 1 h with unlabeled genomic DNA (gDNA) or 3 kbp fragments containing DUS, washed the cells, subsequently incubated them with 3 kbp Cy3-DNA for 30 min and finally treated them with DNase. We found that the distribution of fluorescence intensities was clearly shifted towards lower values when cells were pre-incubated with DNA (Fig. 3a), indicating that ample amounts of DNA remain within the periplasm during at least 30 min. Deletion of the inner membrane channel ComA did not affect saturation strongly (Fig. 3b). For averaging over the fluctuations in the histogram, we integrated over the distribution of single cell fluorescence (Fig. 3 c,d). This cumulative histogram shows the fraction of cells with fluorescence intensity (FU) up to a given value. Furthermore, we examined the stability and integrity of the imported DNA. Cells were incubated with unlabeled 10 kbp fragments for 1 h, subsequently treated with DNase, and further incubated for various periods of time. Exploiting a protocol recently developed for Vibrio cholerae [35], we used duplex PCR with primer pairs against the newly imported 10 kbp fragment and against gDNA of gonococci. We found that the 3 kbp DNA fragments amplified from the imported DNA were clearly detectable even after 60 min after DNase treatment in ΔpilV and ΔpilV ΔcomA backgrounds (Fig. S5). Using wt, the 3 kbp DNA fragments were still detectable after 30 min. Since the signal was lower for wt cells, we did not attempt to amplify DNA at later time points.
DNA turnover in the periplasm may occur as a consequence of dilution due to cell division, degradation, or export. To address turnover, we incubated cells with Cy3-DNA for 1 h, washed them, and subsequently incubated them for a period of 1 h either in the presence or in the absence of unlabeled DNA. We found that the distribution of fluorescence intensities shifted towards lower values independent of ComA after incubation (Fig. S6 a, b), indicating that turnover occurred and was independent of transport through the inner membrane.
Together, these data show that DNA import through the outer membrane occurs independently of inner membrane transport and that large amounts of DNA can be amassed in the periplasm over a time scale of hours.
We next investigated the temporal dynamics of Cy3-DNA during DNA uptake by monitoring single ΔpilV cells during incubation with 300 bp Cy3-DNA in real-time. The fluorescence intensity per cell showed a saturating kinetics with FU(t) = FUmax(1−exp(−t/τ)) with τ = (4.5±0.6) min (Fig. 4 a, b, Movie S1). FUmax is a measure for the total amount of Cy3-DNA that can be imported into the periplasm. Cy3-DNA tended to accumulate at the septa of diplococci. To test whether the kinetics characterized binding or import of Cy3-DNA, we repeated the experiment in ΔpilTΔpilV and ΔpilQΔpilV backgrounds. During 30 min, we did not detect an increase in fluorescence intensity (Fig. 4 c), demonstrating that Cy3-DNA was imported with a characteristic time of 4.5 min in the ΔpilV strain.
In conclusion, the periplasm saturates with short Cy3-DNA fragments within minutes.
It has been shown previously that ComE is necessary for DNA uptake into a DNAse-resistant state and transformation [16]. We investigated whether ComE acted by increasing the carrying capacity of the periplasm or by speeding up DNA import. To this end, we generated isogenic backgrounds varying in comE copy number. Since the amount of imported Cy3-DNA was similar for the backgrounds with four and three copies and for one or no comE copies (Fig. S7), we concentrated on strains with two (ΔcomE34) versus no (ΔcomE1234) alleles in the following. In a first set of experiments, we incubated ΔpilV cells with 3 kbp DNA for variable amounts of time before treating them with DNase (Fig. 5 a,b). Comparing the patterns formed in ΔpilV and ΔcomE34ΔpilV at various time points did not reveal a striking difference. For example, at 1 h both mutants showed multiple foci (Fig. 5 c, d). The kinetics could be well described by a single exponential function FU(t) = FUmax(1−exp(−t/τ)) (Fig. 5e). In the ΔcomE34ΔpilV strain, that carries two copies of the comE gene, the capacity was decreased by a factor of ∼3 as compared to the ΔpilV strain (Fig. 5g). The complete comE deletion strain ΔcomE1234ΔpilV showed a decrease of fluorescence intensity by a factor of ∼24. This reduction is similar to the reduction in a pilQ deletion strain in agreement with ComE being necessary for DNA uptake. We note, however, that residual fluorescence was observed in some cells. The characteristic time to saturation was τ = (100±17) min in ΔpilV. If ComE would enhance the speed of DNA import, then we would expect that reduction of the ComE concentration leads to an increase in the characteristic time. Instead, we observed a decrease to τ = (62±12) min in ΔcomE34ΔpilV (Fig. 5f).
Since we found that ComE increased the carrying capacity of the periplasm, we tested whether ComE was necessary for importing very short DNA fragments of 100 nm. Since Cy3-DNA import showed saturation within minutes, we quantified single cell fluorescence in real-time during incubation with 300 bp Cy3-DNA (Fig. 6). We found that Cy3-DNA import of short fragments was dependent on ComE. A full comE null mutant did not show any increase in fluorescence, indicating that ComE was essential for Cy3-DNA import. The saturating fluorescence intensity Fmax was strongly decreased in the ΔcomE34 strain, confirming that ComE quantitatively controls the carrying capacity of the periplasm.
As Cy3-DNA was not homogeneously distributed within the periplasm (Fig. 1), we investigated whether the periplasmic DNA-binding protein ComE co-localized with DNA. To this end, we generated a strain in which one of the comE genes was fused to an mcherry ORF. In the absence of transforming DNA, ComE-mCherry showed a mostly ring-like distribution, indicating that it was homogeneously distributed within the periplasm (Fig. S8a). Some cells showed pronounced foci which were most often located at the septa between the cocci of diplococci. To test whether these foci arouse from DNA that was present due to lysed cells, we incubated cells with DNase for 30 min and subsequently let the bacteria grow for three generations in liquid culture. Upon this treatment, the foci disappeared almost completely revealing homogeneous distribution of ComE-mCherry in the periplasm (Fig. S8b, Fig. 7a).
In the next step, we incubated the comE-mcherry ΔpilV cells with 3 kbp Cy5-DNA for 15 min. The distribution of mCherry-fluorescence became spotty, often revealing distinct foci (Fig. 7b, c), reminiscent of the fluorescence pattern generated by imported Cy5-DNA. The patterns of Cy5-DNA and ComE-mCherry were highly correlated, indicating co-localization between ComE and imported DNA (Fig. 7 b, c). Most cells that had little or no Cy5-DNA signal retained their ring-like ComE-mCherry fluorescence (e.g. Fig. 7 b inset).
In summary, ComE-mCherry is homogenously distributed in the periplasm in the absence of transforming DNA. Imported DNA forms foci in the periplasm and co-localization of ComE indicates that the periplasmic DNA interacts with ComE.
We investigated the spatio-temporal dynamics of the foci/spotty pattern of Cy3-DNA in the periplasm. To determine the initial location of Cy3-DNA after import, we acquired a time lapse of 10 kbp Cy3-DNA import. Individual 6 kbp Cy3-fragments are clearly visible (Fig. S4). Indeed, we observed the successive appearance of well-defined fluorescent spots correlated with a step-wise increase of fluorescence intensity (Fig. 8 a–d). These foci showed little movement over a time scale of 60 min (Movies S2, S3). To investigate whether there was a preferred location of DNA import (as reported for B. subtilis), we projected the location of initial occurrence to the major axis of a diplococcus and normalized this distance to the cell length (1 min time resolution). Only diplococci were analysed in order to find the location relative to the septum. We found no preferred location of DNA uptake (Fig. 8e). Using DNA that was labeled with the reversibly intercalating dye YOYO, we repeated the experiment. Again, no preferential location of YOYO-DNA foci was observed (Fig. 8e).
Next, we investigated the spatio-temporal dynamics of ComE-mCherry during exposure to unlabeled 10 kbp transforming DNA. These experiments were performed at 37°C and therefore the image quality was not comparable to Fig. 7. Initially, the fluorescence intensity was homogeneous or ring-like (Fig. 9, Movie S4). Spontaneously, the fluorescence accumulated to form a focus whereas the fluorescence in the reminder of the cell decayed. The intensity of the focus increased for ∼10 min and was then stable. At 18 min a second focus appeared again taking ∼15 min to reach its maximum intensity. This second focus was stable for another 30 min. Monococci also formed stable multiple foci (Movie S5). This experiment indicates that the formation of stable foci is not caused by fluorescence labeling of DNA.
To characterize the distribution of focus location, we optimized the imaging conditions using a camera with higher pixel density and we imaged at room temperature. ΔpilV cells were incubated with Cy3-DNA at 37°C for 15 min, treated with DNase and subsequently imaged. For different Cy3-DNA fragments, we found foci around the cell contour (Fig. 10a). We wrote an algorithm that detected diplococci, defined the cell contour, aligned them, and normalized the cell sizes. With this algorithm, the two-dimensional probability distribution of focus location was plotted (Fig. 10b). We found that foci were located all around the cell contour with a slight bias towards the −0.5/0.5 position for long fragments. To assess whether particularly bright foci had a different distribution, we plotted the distribution of the 10% brightest foci (Fig. 10c). Very clearly, the 300 bp fragments accumulated at the septa between the cocci, in agreement with Fig. 4a. Finally, we assessed the distribution of 3 kbp Cy3-DNA in wt cells. Again we found that foci formed (Fig. 10 d,e). Although the total number of foci was low, we found that foci were distributed all around the cell contour.
To test whether there is re-organization of long Cy3-DNA over longer periods of time, we incubated cells for 1 h with 3 kbp Cy3-DNA fragments, washed them and subsequently incubated with unlabeled DNA. We found that the fraction of cells showing central foci increased strongly with time (Fig. 11a, b, c). This behavior was independent of ComA. Time lapse microscopy revealed formation of central foci (Fig.11d, Movie S6) and movement of central foci (Fig. 11d). When cells were incubated for one hour without unlabeled DNA, the pattern did not change significantly (Fig. 11b, c).
In summary, our data strongly suggests the following spatio-temporal dynamics. Transforming DNA is imported at random locations. DNA fragments form foci with associated ComE that are recruited at the septa. The time scale of recruitment depends on DNA length.
We have shown that ComE-mCherry is homogeneously distributed within the periplasm in the absence of transforming DNA, i.e. after DNase treatment and prolonged incubation in liquid culture. When harvested without DNase treatment, a fraction of cells showed foci that were mostly at the septa between cocci. This finding strongly suggests that gonococci accumulate DNA presumably released by lysis of siblings or by secretion in the periplasm while growing in microcolonies. Most importantly, upon addition of Cy5-DNA, ComE-mCherry localized to the Cy5-DNA foci. These experiments demonstrate that ComE is free to move within the periplasm and is captured by periplasmic DNA. While this manuscript was in revision, it has been reported that the homolog ComEA-mCherry diffuses rapidly in the periplasm of V. cholerae and co-localizes with YOYO-DNA [36]. Together with experiments showing that expression of N. gonorrhoeae-comE can restore functionality in a V. cholerae-comEA deletion mutation, ComE-dependent DNA-import into the periplasm can be considered a general mechanism for Gram-negative competent species. Our data further show that association of ComE-mCherry with Cy5-DNA is not a transient phenomenon but that ComE remains associated with periplasmic DNA. Interestingly, the homologue of ComE in B. subtilis, ComEA, is the only known competence protein that does not localize to the cell pole [28]. It has been suggested, that ComEA molecules sequester DNA all around the cell shuffling the DNA to the site of DNA uptake where the rate-limiting step of DNA import occurs [29]. Similarly, in gonococci ComE is likely to compact DNA and to increase the capacity of the periplasmic reservoir.
Indeed, we found that ComE quantitatively governed the carrying capacity of the periplasm. Since the length of the imported DNA fragments did not influence the carrying capacity, we conclude that the ComE concentration is a limiting factor for the amount of DNA that the periplasm can hold. With a height of ∼20 nm [37] and a cell radius of ∼0.4 µm, the periplasm has a volume of ∼3·10−5 µm3. Considering a 10 kbp DNA fragment and assuming no excluded volume interactions, the volume of the statistical coil with a radius of gyration of 235 nm assumes ∼5·10−2 µm3, indicating that energy must be spent for packaging the DNA. To assess whether the function of ComE is to package the DNA, we incubated gonococci with DNA fragments of the order of the Kuhn segment length of 100 nm (300 bp). The Kuhn segment length is a measure for the stiffness of the polymer, in other words it describes over which length the polymer can be bent by Brownian motion. We found, however, that ComE was essential for the import of 300 bp fragments into the periplasm. Mechanistically, we propose that during DNA import, DNA binds weakly to the machine that pulls the DNA through the outer membrane. To render the import irreversible ComE binds to the incoming DNA and hinders its backward movement, generating a translocation ratchet [26]. Additionally, ComE might compact the DNA beyond its Kuhn segment length of 100 nm.
Using DNA fragments with a length of 10 kbp, we found that Cy3-DNA foci occurred at random locations around the cell contour. As the time resolution of our time-lapse acquisition was 1 min, this location most likely coincides with the location of import through the outer membrane. The 10 kbp DNA fragments did not show strong mobility over the period of 30 min. The stochastic distribution of DNA import complexes correlates with the distribution of T4P in gonococci [38], supporting the idea that DNA directly interacts with components of the T4P system. However, we did not observe T4P-like structures coated with Cy3-DNA during real-time experiments. Therefore, we cannot draw conclusions about the potential role of T4P fibers directly in DNA uptake. This result strongly suggests that multiple DNA import complexes exist and are distributed around the cell contour. In B. subtilis and H. pylori, DNA import was found to be localized either to the cell pole or to the septum [28] [16] [24]. In B. subtilis, one or two DNA import complexes or accumulations of the latter were stable even when the cell wall was dissolved resulting in cellular deformation to a round morphology [29]. Mostly single T4P pili were found in V. cholerae that were randomly located around the cell contour [39]. Here, we found that DNA was imported from multiple sites in a single bacterium. We speculate that bacteria that use T4P uniquely for DNA import may form a single pilus, whereas N. gonorrhoeae is a peritrichously piliated bacterium in agreement with multiple DNA import sites reported here.
Interestingly, when cells were incubated with 3 kbp Cy3-DNA and subsequently with unlabeled DNA, Cy3-DNA foci tended to move to septa of diplococci. These dynamics were observed only in the presence of unlabeled transforming DNA. Similar central accumulations were observed with 300 bp Cy3-DNA after several minutes. One possible explanation would be that newly incoming DNA pushes the previously imported DNA away from the sites of DNA uptake. The local accumulation of periplasmic DNA is somewhat reminiscent of DNA-filled membrane blebs that have been termed “transformasome” in Haemophilus influenzae [40], [41]. In TEM-images these membrane-blebs were found at a 10× increased frequency in competent versus non-competent H. influenzae. These sites were proposed to be the sites of DNA-import.
The lipoprotein Tetrapac (Tpc) is essential for transformation and deletion of this putative murein hydrolase shows a severe defect in cell separation generating tetrapacs instead of diplococci [18]. It is reasonable to assume that Tpc acts at the septum by hydrolyzing the cell wall. Periplasmic DNA would then be trapped at the septum location where Tpc degrades the cell wall. Furthermore, DNA was shown to bind to FtsE which is believed to be involved in cell division through interaction with FtsZ which localizes to the septum [42]. Very recently, it has been shown that the pneumococcal EndA nuclease localizes to the midcell upon induction of competence [32]. Interestingly, fluorescently labeled Cy3-DNA also localized to the midcell of competent pneumococci, suggesting that active uptake occurs at this location. Furthermore, they found that midcell localization was independent of comEC expression in agreement with our finding that midcell localization was independent of comA. It is tempting to speculate that recruitment of DNA to the site of future cell division is a general property of competent cocci although it is currently unclear whether this accumulation is linked to DNA transport through the inner membrane.
Here we have introduced a novel approach for studying the spatio-temporal dynamics of DNA import in Gram-negative species at the single cell level. So far, DNA import has been mostly studied at the population level. We have verified that the static properties agree with previous reports using radioactively labeled DNA, including the effect of the outer membrane pore [7], the inner membrane channel [17], the minor pilins [7], [11], and the periplasmic proteins [16]. We have previously quantified the effect of Cy3-labels on the speed of DNA-import in H. pylori and found that the import speed was decreased by a factor of ∼2 [24]. Considering that DNA import depends on the pore formed by PilQ and that its inner opening is wide enough to allow translocation of the type IV pilus, we do not expect that the speed of DNA import is strongly influenced by the Cy3-label. One clear limitation of our approach is the fact that Cy3-DNA is not transported into the cytoplasm and as a consequence, transport through the inner membrane cannot be characterized. However, we performed saturation experiments using unlabeled DNA, to verify that accumulation of DNA in the periplasm is not caused by Cy3-labeling. When cells were pre-incubated with unlabeled genomic DNA or with PCR-fragments for 1 h, the fluorescence signal obtained after subsequent incubation with Cy3-DNA was strongly reduced. This experiment verifies that massing of DNA within the periplasm occurs with unlabeled DNA and is not caused by Cy3. Moreover, ComE-mCherry foci formed after treatment with unlabeled transforming DNA and were stable for up to 30 min. This experiment further supports the claim that stable periplasmic DNA foci are not caused by Cy3-labeling. Previous experiments with H. pylori showed that YOYO-DNA was imported into the periplasm [24]. Here, we found YOYO-DNA binding to the gonococcal surface, but the signal was lost upon DNase treatment (data not shown). This finding might be caused either by rapid formation of ssDNA in the periplasm, by expulsion of the dye from dsDNA due to DNA compaction in the periplasm, or by quenching.
Most experiments were performed in a pilV deletion background to facilitate imaging. This strain has been reported to show a strong increase in the level of both DUS-specific DNA uptake and in the rate of transformation [11]. Thus, transport through the inner membrane is not the only bottleneck for transformation and the accumulation of DNA in the periplasm is not likely to be caused by pilV deletion. Our experiments show that the gonococcal periplasm is saturable with DNA in a ComE-dependent manner, but it remains unclear, whether the wt reaches saturation. However, within biofilms, gonococci are surrounded by a large amount of genomic DNA and a high abundance of DUS [43]. Being continuously exposed to large amounts of DUS containing DNA and potentially dividing at a much lower rate, it is conceivable, that saturation is achieved by the wild type in biofilms. In our experiments, the Cy3-DNA fragments contained only a single DUS, however the deletion of pilV increases DUS-dependent binding and uptake, so using the pilV mutant in our experimental setup is most likely mechanistically not very different to the wildtype situation in biofilms. Concerning the molecular mechanism of DNA import, the pilV strain is not different to that seen in wt backgrounds as DNA import depends on pilE, pilT, DUS, and on assembled pili [11]. In wt gonocoocci, the distribution of DNA-foci was similar to the ΔpilV background and imported DNA was stable as well. All things considered, we propose that Cy3-labeling of DNA is a useful tool for studying the spatio-temporal dynamics of DNA import into the periplasm.
Our results support the following spatio-temporal dynamics of DNA uptake. DNA transport through the outer membrane is powered by DNA uptake complexes that are randomly distributed over the cellular contour. The periplasmic DNA-binding protein ComE is homogeneously distributed in the periplasm in the absence of DNA. Upon contact with periplasmic DNA, it relocalizes to foci formed by DNA. The carrying capacity of the gonococcal periplasm for DNA depends on ComE in a gene-dosage-dependent fashion. Nanomanipulation experiments will be necessary to investigate whether ComE has an additional role in directly driving DNA-import through the outer membrane. When external DNA is present, the periplasmic DNA is relocated to septa of diplococci. It is tempting to speculate that midcell/septum location of transport through the cytoplasmic membrane is conserved between Gram-positive and Gram-negative cocci. It will be interesting to assess whether the location of inner membrane transport is at the septum.
N. gonorrrhoeae (Table S1) was grown overnight at 37°C and 5% CO2 on agar plates containing gonococcal base agar (10 g/l Bacto agar (BD Biosciences, Bedford, MA, USA), 5 g/l NaCl (Roth, Darmstadt, Germany), 4 g/l K2HPO4 (Roth), 1 g/l KH2PO4 (Roth), 15 g/l Proteose Peptone No. 3 (BD), 0.5 g/l soluble starch (Sigma-Aldrich, St. Louis, MO, USA)) and the following supplements: 1 g/l D-Glucose (Roth), 0.1 g/l L-glutamine (Roth), 0.289 g/l L-cysteine-HCL×H20 (Roth), 1 mg/l thiamine pyrophosphate (Sigma-Aldrich), 0.2 mg/l Fe(NO3)3 (Sigma-Aldrich), 0.03 mg/l thiamine HCl (Roth), 0.13 mg/l 4-aminobenzoic acid (Sigma-Aldrich), 2.5 mg/l β-nicotinamide adenine dinucleotide (Roth) and 0.1 mg/l vitamin B12 (Sigma-Aldrich). Before each experiment gonococcal colonies were resuspended in GC-medium.
ΔpilQ ΔpilV, ΔpilT ΔpilV, and ΔcomA ΔpilV were constructed by transforming genomic DNA from existing deletion mutants in the N400 background into GV1 (ΔpilV) [44] (Table S1). The genomic DNA was isolated from GQ21 [45], GT17 [46] and ΔcomA (derived by transformation of N400 with the ΔcomA allele originally detailed in Facius and Meyer 1993 [19]). The constructions of the ΔcomE strains are described in the Supplementary Methods (Text S1). Antibiotics and IPTG were used at the following concentrations: 50 µg/ml kanamycin (Roth), 50 µg/ml apramycin (Sigma-Aldrich), 2.5 µg/ml erythromycin (Sigma-Aldrich), 10 µg/ml chloramphenicol (Roth), 40 µg/ml spectinomycin (Sigma-Aldrich), 2 µg/ml tetracyclin (Roth), 1 mM IPTG (Roth).
The covalent attachment of Cy3 and Cy5 dyes to DNA was achieved with the help of the Label IT Nucleic Acid Labeling Kits (Mirus). According to the manufacturer, the Label IT reagent is bound by a reactive alkylating group to any reactive heteroatom of the DNA without altering the structure of the nucleic acid. The labeling reagent, labeling buffer and 5 µg of DNA were mixed in Milli-Q-H2O to a total volume of 50 µl according to the manufacturer's protocol with a 1∶1 (v/w) ratio of Label IT reagent to DNA. The incubation time at 37°C was elongated to 2 h to improve the labeling density. The samples were purified subsequently by using the provided microspin columns. For comparative quantifications, only labeled DNA from the same labelling reaction was used. YOYO-1-iodide (Invitrogen) is a dimeric cyanine with a highly specific affinity to bind to DNA. Due to its structure, it can easily intercalate between single base pairs, modifying the DNA less than covalently attached dyes. We used a dye∶bp ratio of 1∶50. For this, 1 µg of DNA in 10 µl Milli-Q-H2O was mixed with 0.3 µl of a freshly diluted 0.1 mM YOYO solution in Milli-Q-H2O.
Several bacterial colonies of 16 h–20 h old cultures grown on GC-agar were resuspended with a 10 µl inoculation syringe in 100 µl DNA-uptake-medium (GC-medium supplemented with Isovitalex and 7 mM MgCl2) to an OD600 of 0.1. Cy3- or YOYO-labeled PCR-fragments were added to the cell suspension to a final concentration of 1 ng/µl. The cells were incubated with DNA at 37°C with 5% CO2 and subsequently treated with 10 U DNAse I (recombinant, Fermentas) for further 15 min at 37°C. 50 µl of this dilution were applied to cover slides for microscopic analysis. For each comparative experiment, the different strains or conditions were characterized on the same day using the same stock of labeled DNA. Each condition was characterized independently on at least three different days.
All fluorescence quantification and real-time experiments were conducted at an inverted microscope (Eclipse TE2000 Nikon) in epi-fluorescence mode at 37°. A 120 W metal halogenide fluorescence lamp (X-Cite, EXFO) served as illumination source. Images were taken with an EMCCD camera (Cascade II:512, Photometrics). The 100× oil immersion CFI Plan Fluor objective (NA 1.3, Nikon) was used in all applications. Day to day variations in the brightness of the fluorescence lamp were detected by using the test beads on the Focal Check Fluorescence Microscope Test Slide #3 (Invitrogen). The intensities from day to day stayed relatively stable with no more deviations than ±10%. A correction factor to this reference intensity was calculated for every day and was applied to the respective data sets. The analysis is described in the Supplementary Methods (Text S1) and Fig. S2.
For localization experiments an inverted microscope (TI-E Nikon) was used at room temperature. A 120 W metal halogenide fluorescence lamp (Intensilight Nikon) served as illumination source. Images were taken with a CCD camera (Orca 2.3 Hamamatsu) that had a high pixel density for Cy3-DNA focus distribution. For ComE-mCherry/Cy5-DNA co-localization, an EMCCD camera (IXON X3897 Andor) was used.
Cells selected for piliation were grown on a GC agar plate for <20 h, harvested, resuspended in DNA-uptake-medium and adjusted to OD600 = 0.01–0.02. 150 µl of the suspension were applied to a polystyrene-coated coverslip inside a microscopy chamber and incubated 10 min on the microscope stage. Subsequently, 1.4 µg of 300 bp Cy3-DNA diluted in 100 µl DNA-uptake-medium was added and DNA uptake was recorded for 60 min. To control for DNA bound to the exterior of the cells, 10 U/ml of DNaseI (Fermentas) was added and fluorescence intensity was recorded for another 15 min.
|
10.1371/journal.pcbi.1004870 | Maximum-Entropy Models of Sequenced Immune Repertoires Predict Antigen-Antibody Affinity | The immune system has developed a number of distinct complex mechanisms to shape and control the antibody repertoire. One of these mechanisms, the affinity maturation process, works in an evolutionary-like fashion: after binding to a foreign molecule, the antibody-producing B-cells exhibit a high-frequency mutation rate in the genome region that codes for the antibody active site. Eventually, cells that produce antibodies with higher affinity for their cognate antigen are selected and clonally expanded. Here, we propose a new statistical approach based on maximum entropy modeling in which a scoring function related to the binding affinity of antibodies against a specific antigen is inferred from a sample of sequences of the immune repertoire of an individual. We use our inference strategy to infer a statistical model on a data set obtained by sequencing a fairly large portion of the immune repertoire of an HIV-1 infected patient. The Pearson correlation coefficient between our scoring function and the IC50 neutralization titer measured on 30 different antibodies of known sequence is as high as 0.77 (p-value 10−6), outperforming other sequence- and structure-based models.
| Affinity maturation is a very complex biological process which enables activated B-cells to produce antibodies with increased affinity for a given antigen. Once B-cells begin to proliferate, each of the progeny cells introduces mutations in the antigen binding region in order to explore different affinities for the antigen. Selection rounds occurring in the so-called germinal centers in lymph nodes and spleen prune out poorly binding receptors and clonally expand good binders. Thanks to high-throughput sequencing techniques it is now possible to have access to a fairly representative sample (of the order of 105 to 106 sequences) of the immune repertoire of a given individual. Our approach is to first exploit this large amount of sequence data to infer a statistical model for the sequenced portion of the immune repertoire, and then to use the inferred probability of this model as a score when predicting the neutralization power of a given antibody sequence for the antigen of interest. The results we obtained on a specific data set of sequences of an HIV-1 patient show that our score correlates very well with experimentally assessed neutralization power of specific antibodies of known sequence. The performance of the method crucially relies on the ability of our model to account for long-range intragenic epistatic interactions between residues along the whole antibody chain.
| The prediction of antibody (Abs, or immunoglobulins, Igs) affinity for antigens is among the most interesting open challenges across bioinformatics and structural immunology. Most of the current methods rely on the structures (either experimentally resolved or modeled) of both antibodies and their cognate antigens to predict their binding affinity. Currently, available methods are time demanding and, more importantly, their predictions are hard to assess [2, 3]. On the other hand, because of the scarcity of available data-sets for which both Abs sequences and their affinity for an antigen are known, there is still no method that can model the affinity as a function of the sequence of the antibody variable region. Also, it is still not clear if and how it would be possible to set up a coherent fitting procedure to estimate the (possibly) huge number of parameters of a generic mapping from the space of Abs sequences to the affinity for the antigen.
Thanks to the recent developments of sequencing techniques (e.g. Deep Sequencing, and Next Generation Sequencing), Repertoire Sequencing (Rep-Seq) experiments (see [4] for a review of the argument) start to be routinely performed. Recently, the complete Ig repertoires of several simple organisms such as the zebra-fish, whose immune system has only ∼300.000 Abs producing B cells, have been sequenced [5]. Higher organisms, such as humans, show a remarkably more complex immune system and it is widely accepted that the typical human Ab repertoire amounts to ∼109−10 different molecules. In this case, a large sample of the entire repertoire can be extracted (see for example [6] for Rep-Seq experiment on Igs in human).
Rep-Seq data allow for a detailed description of the sequences distribution based on Maximum Entropy (MaxEnt) modeling of repertoires, as it has been proven in the case of zebra-fish Abs [7] and human T cell receptors [8, 9]. While these studies focus on a model-based description of the initial repertoire of the adaptive immune system arising mainly from the V(D)J genetic rearrangement, here we focus on the affinity maturation process.
A number of statistical mechanics inspired methodologies have been recently successfully devised to analyze evolutionarily related proteins for inferring structural properties and, in particular, residue-residue contacts [10]. In particular, homologous proteins can be characterized in terms of multiple sequence alignments (MSAs). In spite of the considerable sequence heterogeneity (up to only 40% sequence identity) in families of homologous proteins, their folded structures are often almost completely conserved [11]. A MaxEnt modeling technique developed more than a decade ago, could detect signals of the evolutionary pressure beyond the sequence variability in MSAs of homologous proteins [12]. Maintaining the same underlying idea that co-evolution of residue pairs is related to their spatial proximity in the folded protein structure, a large number of works successfully reconsidered MaxEnt in different flavors: (i) the application of mean-field approximations known as Direct-Coupling Analysis (DCA) [13–15], (ii) pseudo-likelihood maximization (PlmDCA), [16–18], (iii) Multivariate Gaussian Modeling (MGM), [19, 20]. All these methods rely on the inference of a generative probabilistic model for sequences in the presence of selective pressure. This feature makes this kind of analytic techniques particularly suited for the study of Ab affinity maturation. In fact, this process closely resembles a Darwinian evolutionary framework where B-cell clones compete for the antigen in the germinal centers, and it is now widely accepted that the affinity for the target antigen represents the main contribution to the fitness in this evolutionary scenario. Thus, as qualitatively sketched in Fig 1, for every antigen, the evolutionary dynamics explores the space of Ab sequences searching for the global optimum of the fitness function, i.e. the best affinity for the related antigen.
Here we exploit the evolutionary nature of the affinity maturation process by applying a MaxEnt inference techniques originally developed for the analysis of homologous protein families. The above mentioned plethora of model inference methods aim at reconstructing a reliable contact map from the space of homologous protein sequences through an analysis of residues coevolution that disentangle indirect correlations, but in our context, they provide little information on Abs internal structure. However, the inference procedure provides a natural and reliable scoring function (see Section “Inference Methods”) from the space sequences to that of binding affinity for the target antibody related to the probability for a sequence to appear in the data set that we can use as a proxy to the binding affinity to the antigen, in the spirit of series of recent publications [21–23] where deep sequencing of the immune repertoire was used to predict binding vs. non-binding Abs with different therapeutic applications.
Finally, we report that very recently maximum entropy modeling has been also used in [24] to predict the fitness landscape of the HIV-1 protein from the relative abundance of the virus strains, and in [25] to predict in silico the effect of mutations related to disease and antibiotic drug resistance.
In the present work, we apply MaxEnt methods to study the affinity maturation process on publicly available data from an HIV-1 infected donor [1].
The immune system of this patient had developed over the years the so-called broadly neutralizing antibodies (bNAbs), which can bind with high affinity to the virus capsid protein gp120 and impair the viral ability to infect new cells. The broadness of Abs neutralization entails their capability of neutralizing multiple HIV-1 strains, as opposite to non-bNAbs which are specific for individual viral strains. The following data from Wu et al., all derived from the antibody repertoire of the patient, have been used in the present work: (i) a X-ray crystallographic structure of gp120 in complex with VRC-PG04, a broadly neutralizing Ab identified through cell sorting; (ii) a Rep-Seq data of the donor’s immunoglobulins heavy chains (IGH) variable region repertoire (see Section “Deep sequencing data”); (iii) half maximal inhibitory concentration measurements (IC50) of chimeric Abs against some isolates of the antigen gp120. IC50 will be considered hereafter as a proxy for the IGH contribution to the antigen-Ab complex binding affinity (see Section “Neutralization measurements” for details).
Our study is based on two main working hypotheses: (i) the Ab sequences that are similar to the highly responding Ab VRC-PG04 are informative about their binding energy [1]; (ii) This specific subset of Abs has evolved through affinity maturation, i.e. developing somatic mutations in gp120-binding sequences to enhance their binding energy toward the antigen.
As summarized in Fig 2, we have developed a bioinformatics pipeline to select a subset of aligned Ab amino acid sequences from the whole Rep-Seq data set. We claim that the selected sequences have performed affinity maturation to achieve a high and broad power against gp120. In the “Clustering analysis” section we explain how the choice of the gp120-responding ensemble (which we call from now on hypermutated cluster) is done, while in the “Multiple sequence alignments” section we describe how we constructed the custom Hidden Markov model to align sequences.
From these premises we used MGM [20] (see Section “Multivariate Gaussian Modeling”), a particular version of MaxEnt modeling, to infer an accurate statistical model for the ensemble of Abs in the data set clonally expanded for their affinity against antigen gp120, as schematically shown in Fig 3. The MGM model allows taking into account in a probabilistic sense long range intragenic epistatic interactions across the whole heavy-chain variable region of the Ab. Furthermore, the inferred model naturally defines a statistical scoring function (MGM-score) for Ab sequences. In Section “Affinity predictions” we show that the MGM-score correlates significantly (Pearson correlation coefficient up to 0.77) with the IC50 assay performed on a large set of Abs of known sequence. We stress that: (i) the MGM-score is inferred on the hypermutated set of sequences for which IC50 measurements are not available; (ii) the set of artificial chimeric Abs of VRC01 origin (a human immunoglobulin that neutralizes about 90% of HIV-1 isolates) for which the IC50 measures are available were not part of the data set from which the MGM was inferred.
We further investigated whether the intragenic epistatic signal captured by the MGM is related to the structural properties of the gp120-Ab complex. In Section “Structural predictions” we discuss our findings: even if the DCA score [20] is poorly correlated with the internal structure of the Ab (as shown in Section “Contact map predictions”), we find a weak signal that can be used in combination with IC50 measurements to predict residues that are part of the interaction surface (as shown in Section “Prediction of binding sites”).
Wu and coworkers [1] used 70 sequenced heavy chain variable regions, which originated mostly from immunoglobulins using the IGHV1-2 gene, for constructing chimeric antibodies by combining them with the light chain of VRC-PG04. Among these, 45 have been tested for their neutralization power against 20 HIV-1 mutations.
When included in the sequencing data set and used as input for the clustering procedure, 30 of these 45 tested Abs are found to belong to the hypermutated cluster. The remaining 15 (none of which was found to be neutralizing) belong to the germline cluster. Although in general the neutralization power depends on both the light and the heavy chain sequences (cf. Fig. 4A in [1]), the light chain plays only a minor role in the interaction (most notably steric contacts with its CDR1 and CDR3 regions) here, as visible from the solved structure of VCR-PG04 (PDB code 3SE9). We therefore will make the simplifying assumption that the neutralization measurements on chimeric Abs depend on the heavy chain contribution alone.
Under the assumption that the hypermutated cluster is a statistically representative sample of the Abs that underwent affinity maturation against gp120, we can use the statistical properties of this set of sequences to construct a predictor for the Abs neutralization power. We thus inferred an MGM on the MSA of this cluster and used the MGM-score of the inferred model as a proxy for the neutralization power of the related Abs. Although the inference step is completely blind to the binding affinity of the Abs (the binding affinities of sequences belonging to the hypermutated cluster were not measured in [1]), nonetheless the capability of predicting binding energies is not unexpected. Indeed, the aim of a maximum entropy model of the hypermutated set, is to provide an accurate statistical description of the set of Abs responding to gp120, and so it is not completely surprising that, according to the model, sequences with low probability are more likely to have a low binding affinity for the antigen compared to sequences of high probability.
To test the predictive power of the method, we used the panel of 30 sequences (not included in the hypermutated cluster) tested for HIV neutralization power and compared the IC50 neutralization titer with the MGM-score of the same sequence. Note that values of IC50 that are reported in [1] as greater than 50 μg/ml (not-neutralizing) are considered here to be equal to this value. The two quantities are compared by means of the Pearson correlation coefficient. We consider as measures of the neutralization power the average IC50 over the different neutralized viruses. A scheme of the model inference and testing procedure is shown in Fig 3.
The result of the model inference procedure depends on the choice of the regularization parameter π defined in the “Inference methods” section. We therefore repeated the test procedure for different values of π. In Fig 4 the Pearson correlation coefficient between the MGM-score and the average IC50 over the neutralized viral isolates is shown for different values of π. The two panels refer to the two score proposed: the original inferred MGM-score and the MGM-score with gap correction (see Section “Score with gap correction” for details). We thus argue that the MGM-score inferred on a representative Rep-Seq data set provides a remarkably good proxy for the neutralization power of the analyzed sequence. We also display the details of our best performance on a per-virus base in Fig 5.
We also assessed the performance of the MGM-score to discriminate binding vs. non-binding sequences. The dataset in this somehow simpler task reduces in a set of 21 non-binding and 24 binding sequences. The performance of the MGM-score are displayed in terms of the ROC curve shown in Fig. F in S1 Text (red curve): the (normalized) area under the ROC (AUROC) turns out to be 0.97. We also compared this value against a much simpler scoring strategy defined in terms of the Hamming distance from the consensus sequence of the hypermutated cluster. As shown in Fig. F in S1 Text (blue curve), the AUROC turns out to be 0.86.
We also inferred the model using PlmDCA [17] rather than MGM. The results are shown in Fig. G in S1 Text: The best Pearson correlation coefficient obtained with this method of inference is slightly worse than the one obtained with MGM. This result is non-trivial since PlmDCA is known to perform better than MGM in terms of protein contact prediction. We also note that in a recent publication [25], a variant of DCA (mean-field DCA) that is essentially equivalent to MGM was used to successfully predict the ΔΔG between mutants and wild type sequences for the beta-lactamase TEM-1.
A natural question is whether simpler inference strategies might achieve equally good results, and in particular whether it is necessary to use the second order statistics (i.e. multivariate vs univariate statistics) to infer Abs neutralization power. To this end, we tested a simpler version of the model, factorized over the different residues of the MSA. In this model the non-diagonal J terms are set to zero so that the residues are statistically independent (see Section “Multivariate Gaussian Modeling” and [20]). As shown in Fig 4 (squares and dashed lines), the Pearson correlation coefficient is dramatically reduced, dropping from a maximum of 0.77 for the full MGM to a maximum of 0.49 for the factorized model.
Our neutralization power predictor was compared with another sequence based method, the HMM-score (see Section “Using Hidden Markov Models to predict binding affinities”). This score takes only correlations between nearest neighbors in the sequence into account. Interestingly, as displayed in Fig 4, the prediction quality of this method is between the one obtained using the factorized MGM-score and the one obtained using the full MGM-score. This supports the observation that long range intragenic epistatic signals are crucial to reproduce neutralization power.
An important step in the procedure is to correctly identify the set of sequences that underwent affinity maturation towards the same epitope. Indeed, MGM models trained on different sets (for example the entire set of sequences coming from the germline of interest) display no significant correlation with neutralization measurement.
Some portions of the MSA are observed to be more important than others in reproducing the affinity function: The correlation between the inferred likelihood and the neutralization titers is essentially the same when only the ∼60 more variable residues of the hypermutated cluster MSA are used to construct the MGM, dismissing ∼3/4 of the columns of the MSA. Data of this MSA reduction analysis are reported in S1 Text (see Section “Affinity predictions”).
Our predictor was also compared with a structure-based method: we produced structural models for all the 45 antibody/antigen complexes for which the IC50 was measured and predicted their binding affinity using FoldX (see Methods for details). The results of this structural method show no significant correlation (r = −0.23, p-value = 0.13) with the experimental data.
Taken together, our findings indicate that: (i) MGM inferred on the proper set of clonally expanded sequences contains enough information to predict the neutralization power of Ab sequences. This suggests that the procedure can be used as a tool to generate new and highly neutralizing Abs; (ii) taking into account (pairwise) intragenic epistatic effects in the model improves remarkably the accuracy of the affinity prediction.
The identity/divergence analysis performed in [1] on the whole deep sequencing data set indicates that sequences with inferred IGHV1-2 germline gene (the same of VRC-PG04) are characterized by: (i) the presence of a cluster of highly mutated sequences that is well separated from the cluster of typically mutated sequences; (ii) Abs with a different IGHV inferred germline gene display a more uniform (i.e. less clustered) structure.
We performed an independent identity/divergence analysis on the data set resulting from our bioinformatics analysis in which we retain only productive sequences of IGHV1-2 origin. Our results are in complete agreement with [1], as shown in Fig 7. There we compare the identity to VRC-PG04 and the divergence from IGHV1-2*02 germline gene at a nucleotide level for each sequence in the data set.
Identity/divergence analysis gives a glimpse of the structure of the sample in the space of sequences. Nevertheless, a less biased analysis is required in order to test the cluster structure. We thus performed a sequence-based clustering analysis. Among the different clustering algorithms available, we chose the shallow tree clustering algorithm [32] since it provides a criterion of robustness against noise (see S1 Text Section “Sequence clustering analysis”). The clustering algorithm is based on the Hamming distance between sequences.
The most robust solution (see S1 Text for an explanation of what robust means in this context) found by the algorithm is a partition of the sequences into two clusters: a germline cluster composed of 2878 sequences (1634 unique) centered on the IGHV1-2*02 and IGHJ2*02 germline genes (with an average sequence divergence of ∼5% from the germline), and a hypermutated cluster composed of 3896 sequences (1578 unique) more similar to the broadly neutralizing antibody VRC-PG04 (with an average sequence divergence of ∼35% from the germline, see Section “Clustering analysis” in S1 Text for details). These results are confirmed by a test with the k-means clustering algorithm (run with k = 2). Information about the two clusters and their MSA characteristics are resumed in Table 2.
In the present work, we assume the hypermutated cluster to be a representative sample of the Abs that underwent affinity maturation for neutralizing HIV-1 gp120.
The structure of VRC-PG04 in complex with gp120 (PDB-id 3SE9) has been subjected to both visual inspection and quantitative predictions to assess the importance of each somatic mutation observed in the antibody to the binding affinity towards the antigen. Somatic mutations were retrieved using the IMGT database [34]. We used the FoldX software [35] to predict the difference in binding energy (ΔΔG) of the actual antibody with all the mutants obtained reverting each single somatic mutation to the original residue observed in the germline gene IGHV1-2.
In the present study, we proposed a sequence based maximum entropy model to analyze Ab affinity for the antigen. The predictive validity of the model has been tested using Rep-Seq data and neutralization power measurements from an HIV-1 infected donor [1]. The interplay between the HIV-1 virus and the immune response provides an interesting framework for our purpose: the affinity maturation of the Abs of interest (those whose epitope is the gp120 CD4 binding site) causes a dramatic increase of their neutralization power and a pronounced mutation ratio in comparison with the germline genes. This high density of mutations allows us to easily select sequences in the immune repertoire that respond to the antigen.
A maximum entropy model constructed on this set of hypermutated sequences has been successfully used as a predictor of the neutralization power of Abs. This predictor has been successfully assessed against experimental neutralization measurements of different viral isolates. These positive results suggest that the procedure could be used as a tool for generating new and highly neutralizing Abs.
In analogy with the application to protein families [12–20], the MaxEnt model has been used for predicting residue-residue contacts in the Rep-Seq sample without obtaining positive results. This is not surprising since the time-scale involved in the affinity maturation process (years) is not comparable to the typical evolutionary time-scale in protein families (millions of years).
The structure of the inferred statistical interactions is probably mostly driven by the interaction with the epitope and further investigations in this sense represent an interesting development of this work. Nevertheless, the joint analysis of the sequencing data statistics and neutralization measurements has been shown to provide some consistent structural information on antigen recognition mode.
In conclusion, the use of maximum entropy models can unveil relevant features of the protein fitness function. These features are related to the affinity maturation process and in particular to the evolutionary dynamics of the B cell population. This could be of interest for a statistical population genetics analysis of the affinity maturation process (for example in the spirit of [39] and [40]). The present case study shows how MaxEnt methods can be a useful tool for tackling immunological questions in a time when Rep-Seq data are becoming increasingly popular in immunology (see for instance [41], where T receptor repertoires are studied).
|
10.1371/journal.pntd.0007045 | NMR metabolomics of cerebrospinal fluid differentiates inflammatory diseases of the central nervous system | Myriad infectious and noninfectious causes of encephalomyelitis (EM) have similar clinical manifestations, presenting serious challenges to diagnosis and treatment. Metabolomics of cerebrospinal fluid (CSF) was explored as a method of differentiating among neurological diseases causing EM using a single CSF sample.
1H NMR metabolomics was applied to CSF samples from 27 patients with a laboratory-confirmed disease, including Lyme disease or West Nile Virus meningoencephalitis, multiple sclerosis, rabies, or Histoplasma meningitis, and 25 controls. Cluster analyses distinguished samples by infection status and moderately by pathogen, with shared and differentiating metabolite patterns observed among diseases. CART analysis predicted infection status with 100% sensitivity and 93% specificity.
These preliminary results suggest the potential utility of CSF metabolomics as a rapid screening test to enhance diagnostic accuracies and improve patient outcomes.
| Inflammation of the brain and spinal cord, known as encephalomyelitis, is a dangerous condition that can be caused by a wide range of pathogens, such as viruses and bacteria, and other medical conditions including autoimmunity or drug intoxications. Given the many possible causes, it is often difficult for clinicians treating patients with encephalomyelitis to identify the underlying cause, which in turn determines the appropriate treatment. Infections and other diseases causing neurological inflammation work by distinct biological mechanisms and, consequently, can cause unique biochemical changes that can be observed in cerebrospinal fluid of affected individuals. The researchers used a metabolomics technique to measure a range of small molecules in cerebrospinal fluid and examine biochemical differences in patients with encephalomyelitis caused by Lyme disease, West Nile Virus, multiple sclerosis, rabies, or fungal infection. The researchers found distinct differences in the biochemical profiles of patients whose encephalomyelitis was caused by infections versus patients with no infection, and also identified different patterns among the individual diseases. This study showed that metabolomics may be useful in improving diagnosis and treatment of diseases affecting the central nervous system by enhancing understanding of their unique effects on metabolism.
| Encephalomyelitis (EM) is a condition characterized by inflammation of the brain (encephalitis) and spinal cord (myelitis) that frequently causes permanent disability. There are myriad causes of EM syndromes, which are in aggregate relatively common [1–4] and include viral, bacterial, fungal, protozoal and prion infections, autoimmune encephalitis, intoxications, and metabolic encephalopathies, while other EM cases have unknown causes [5]. Clinicians face significant challenges to the rapid and accurate diagnosis and treatment of EM. Due to the rarity of a definitive diagnosis, many arbovirus and other viral causes of EM, including rabies, have limited evidence-based therapies; this may change with newer broad-spectrum antivirals currently in clinical trials [6]. Treatment of autoimmune EM relies on corticosteroids, immunoglobulin, plasmapheresis, cytotoxic agents and biologicals [7, 8], which are typically contra-indicated until infections can be excluded. Physicians are often forced to treat empirically for infections and delay appropriate therapy for autoimmune EM, thereby worsening patient outcomes. Moreover, for many causes of EM, no rapid diagnostic testing exists, and long delays pending laboratory test results commonly occur before definitive treatment may be initiated; however, superior outcomes depend on early intervention. Because there are numerous causes of EM, including multiple infectious agents that overlap or coincide in geographic distribution, diagnosis reliant on single-target testing is unsatisfactory as it requires quantities of tests that are not only prohibitive in cost but also involve collecting unsafe volumes of blood or cerebrospinal fluid (CSF) from patients. Improved diagnostics and proxy markers of therapeutic efficacy are sorely needed, especially as new treatment regimens develop.
In recent years, the development and expansion of omics technologies have presented opportunities for discovering disease mechanisms and biomarkers of clinical significance [9–11]. Metabolomics, the comprehensive study of small-molecule metabolites in a biofluid or tissue, offers a set of clues to the biochemical workings of a body system, organ, or compartment in a given physiological state, and has diverse applications in improving clinical diagnosis and treatment of central nervous system (CNS) diseases and intoxications [9, 12–17]. Metabolomics panels may also provide information about a broad spectrum of metabolic processes involved in a disease presentation compared to traditional single-molecule assays. Metabolites present in CSF may originate from brain metabolic processes, including intermediate and end products of energy metabolism, neurotransmission, inflammation and oxidative stress responses; thus, their analysis provides insights into metabolic disturbances occurring in CNS diseases. Among the methodological approaches taken in metabolomics studies of CSF, 1H-NMR spectroscopy carries advantages for exploratory studies both in the scope of metabolite detection and its quantitative ability [18]. An additional advantage of this method is the lack of sample consumption, given practical limitations on the volume of CSF usually available. Further, many CNS diseases and intoxications are prevalent in countries where advanced imaging facilities, reference laboratories and therapeutics are in short supply. Recent studies have applied 1H NMR-based metabolomics of CSF to identify single-molecule biomarkers and panels of metabolites associated with a range of neurological diseases such as infectious meningitis [14], multiple sclerosis (MS) [13, 19, 20], Alzheimer’s [21, 22] Parkinson’s [23] and Huntington’s diseases [24]. Further, this method has detected metabolic changes characterizing different stages of disease progression in rabies and MS [12, 25]. Proxy markers of disease progression or response to therapy may also accelerate therapeutic trials while lowering their cost.
Despite significant advances in the application of NMR metabolomics in the investigation of certain CNS diseases, such as multiple sclerosis, its potential to describe metabolic changes occurring in many infectious neurological diseases has been less studied. Lyme disease and West Nile Virus (WNV) are vastly under-studied in this sense, despite being the most common causes of vector-borne bacterial and viral disease, respectively, in the United States [26, 27]. Rabies is an important global zoonosis but may be underdiagnosed in some contexts due to challenges in distinguishing it clinically from other CNS infections, such as cerebral malaria, in areas where these are endemic [28]. Infectious diseases that invade the CNS have distinct molecular mechanisms driving their respective pathologies [29, 30]. Further, pathogen strategies to replicate while evading host immune responses can involve the disruption of a range of endogenous metabolic processes [31], many of which have yet to be illuminated for specific diseases; thus, explorative studies of the CSF metabolome in different disease states can provide an important window for examining potential pathogen effects on metabolism within the CNS to lay the groundwork for future targeted diagnostics or therapeutic interventions. In the present study CSF samples from patients representing diverse infectious and non-infectious diseases of the CNS were analyzed by 1H NMR-spectroscopy to determine if metabolomics profiling could distinguish diseases. We find preliminary evidence of the existence of discriminating metabolic features.
Twenty-seven patients were diagnosed with CNS Lyme disease (n = 5, all ages, at the New York State Department of Health), WNV meningoencephalitis (n = 5, all ages, New York State Department of Health), Clinically Isolated Syndrome (CIS) of multiple sclerosis (MS, n = 4, adults, Intermountain Healthcare), rabies (n = 10, all ages, at Canadian Food Inspection Agency, Centers for Disease Control and Prevention, Kimron Veterinary Institute, National Institutes of Health-Colombia, and New York State Department of Health), or Histoplasma meningitis (n = 3, anonymous, at Indiana University School of Medicine). Due to ethical concerns surrounding the collection of CSF from healthy individuals, healthy controls were not available for this study. Specimens obtained as discard material from 25 anonymous children aged 5–20 years at the Children’s Hospital of Wisconsin with no concurrent microbiological testing and no known encephalopathy or encephalitis served as a control group. This population includes mostly children with cancer in remission or children being treated for pseudotumor cerebri, a common non-inflammatory condition. Given patient samples were anonymous discard material, the study was ruled to not be human research requiring informed consent by the Children’s Hospital of Wisconsin IRB (protocol CHW 10/24). For rabies patients, for whom multiple specimens were available, the specimen taken closest to the fourth day of hospital admission was selected to minimize the influence of hypoglycemia, ketosis or renal insufficiency on presentation to the CSF metabolome. While the CSF was collected for diagnostic purposes, precise timing is uncertain other than for rabies patients. Initially, four specimens from patients with Histoplasma meningitis were analyzed, but one specimen had a metabolite profile inconsistent with CSF and was excluded on the basis of containing implausible values. Three Histoplasma specimens remained after this exclusion.
After collection, specimens were stored refrigerated and/or frozen until transport on dry ice to the site of analysis, where they were stored at -80°C until sample preparation. Once defrosted, samples were filtered using washed Amicon Ultra-0.5 mL centrifugal filters with a cut-off of 3000 MW (Millipore, Billerica, MA) to remove lipids and proteins. When needed, filtrate volume was adjusted to 207 μL when preparing for 3mm NMR tubes or 585 μL when preparing for 5mm NMR tubes with Type I ultrapure water from Millipore Synergy UV system (Millipore, Billerica, MI). Samples were prepared for analysis by the addition of 23 μL or 65 μL of internal standard containing approximately 5 mmol/L of DSS-d6 [3-(trimethylsilyl)-1-propanesulfonic acid-d6], 0.2% NaN3, in 99.8% D2O to 207 μL or 585 μL of CSF filtrate, respectively. The pH of each sample was adjusted to 6.8 ± 0.1 by adding small amounts of NaOH or HCl. A 180 or 600 μL aliquot was subsequently transferred to 3 mm or 5mm Bruker NMR tubes, respectively, and stored at 4 oC until NMR acquisition (within 24 hours of sample preparation). NMR spectra were acquired as previously described [12] on a Bruker Avance 600-MHz NMR equipped with a SampleJet autosampler using a NOESY-presaturation pulse sequence (noesypr) at 25°C.
NMR spectra were manually phased and baseline-corrected using NMR Suite v6.1 Processor (Chenomx Inc., Edmonton, Canada), and Chenomx NMR Suite v.8.1 Profiler (Chenomx Inc., Edmonton, Canada) was used for quantification of metabolites. Selected NMR spectral data from a previous rabies study in this lab [12] were compared to additional samples acquired from Lyme, WNV, histoplasmosis and MS patients.
After correcting metabolite concentrations for dilution, data were cluster-analyzed 2 ways for comparison using RStudio software (RStudio Version 1.0.136, Boston, MA, USA) or Stata software (SE 14, College Station, TX, USA). First taking a data-driven approach, concentrations were log10-transformed before principal component analysis (PCA) was carried out on the covariance matrix of the centered data as an unsupervised search for trends. Alternatively, to provide clinical context, data were normalized to z-scores using published reference ranges in CSF (www.hmdb.ca). In instances when published norms were discrepant, those that encompassed the range of our control population were selected. In rare instances when normal ranges were unavailable, means and standard deviations were constructed using our 25 controls. Normalization by z-scores constructed from population norms generated more skewed data than log10-transformation across the entire spectrum of diseases and controls. Factor analysis better tolerates skewed data than PCA and was applied to the z-scores.
Based on the separation found by PCA and factor analysis, differences in metabolite concentrations by infection status and by individual disease diagnoses were assessed on the untransformed data using Mann-Whitney U tests and Kruskal-Wallis tests, respectively. P-values were adjusted for multiple comparisons using false discovery rates. Homogeneity of variance between groups was tested using the Levene test to inform interpretation of the rank sum test results. For metabolites with significant differences by Kruskal-Wallis testing, Dunn’s multiple comparisons tests were performed between each pair of groups to determine which diseases were different from each other. For these tests, p-values were Bonferroni-adjusted within the 15 multiple comparisons carried out for each metabolite. After adjustment, p-values of less than 0.05 were considered significant. Cliff’s Delta statistics [32] were calculated to assess the degree of overlap in metabolite concentrations by infection status and between diseases that were found to have significant differences by the Dunn’s test.
Untransformed data were also analyzed by predictive analysis [33, 34]. Classification and regression trees (CART) and Random Forests were performed using Salford Predictive Modeler software suite CART and suite Random Forests (Salford Systems, San Diego, CA, USA). For CART, parent node and terminal node were 10 and 5, respectively. 10% leave-out samples were used for cross-validation. Random Forests are collections of decision trees, and each tree was grown on a random (~2/3) subsample of the data. The remaining data were used to determine the performance of the trees. The number of trees to build was 1000. The number of predictors considered for each node was the square root of the number of potential predictors, and the parent node minimum cases was 2. The variable importance was assessed using the GINI method. Target variable and predictors were the same as for CART.
CSF samples obtained from 25 controls and 27 patients with different neurological diseases were analyzed by 1H-NMR spectroscopy. Table 1 summarizes clinical characteristics of patients included in this study. A total of 57 compounds were identified and quantified in CSF samples; rabies spectra from a prior study [12] were repeat-profiled. Quantification for 13 metabolites present at very low concentrations in a majority of samples was considered not to be exact (S1 and S2 Tables) but still useful in detecting differences between groups. To further minimize the reversible behavioral effects of starvation and dehydration in the analysis normalizing by z-scores, we excluded 3 ketone bodies (3-hydroxybutyrate, acetoacetate, and acetone) and creatinine from the dataset.
A major clinical challenge is determining whether infection exists as a contraindication to immunosuppression. Unsupervised PCA was performed on metabolite data from patients diagnosed with a neurological disease and controls. Six compounds (acetaminophen, ethanol, ethylene glycol, glycerol, propylene glycol, and valproate) of likely exogenous origin were excluded from cluster analysis models. The first two principal components (PC) in this model accounted for 37.8 percent of the variation in metabolite concentrations. Prominent overlap was apparent between controls and MS, which separated distinctly from infectious diseases along PC 1 (Fig 1). In a scores plot of the first two components, PC 2 identified an apparent outlier in the WNV group, which upon closer examination was observed to have extremely low levels of citrate, lactate, and amino acids coupled with markedly high glutamate, pyruvate, acetate and 2-oxoglutarate compared to the rest of the samples. Since the general patterns generated by PCA did not change when this individual was removed from the dataset, the results shown in Fig 1 reflect this exclusion in order to better visualize clusters in the data.
When overlaid with loadings vectors, the scores plot of the first two PCs revealed two patterns of metabolites among infectious diseases, one characterized by higher levels of ketone bodies and the other by higher levels of pyruvate, glutamate, 2-oxoglutarate, carnitine, and glycine (Fig 1). Pearson correlation coefficients reflect moderate to high correlation among the metabolites in each pattern, with correlation coefficients ranging from 0.77 to 0.93 among ketone bodies and from 0.23 to 0.58 among metabolites in the second pattern. In contrast, metabolites including acetate, isobutyrate, myo-inositol, threonine, and glutamine appeared to characterize controls and MS using loadings vectors.
The contribution of ketone bodies to the PCA analysis prompted a second, clinically applicable analysis using z-scores of normal human values for each metabolite while excluding the potentially non-specific markers of dehydration and starvation, which yielded similar results. Unsupervised factor analysis discriminated CNS disease from controls, with 2 factors accounting for 35.6 percent of the variation. The WNV sample that appeared as an outlier by PCA was not influential in this analysis. Factor analysis excluding ketones and creatinine did not discriminate infections from normal as well as did the PCA analysis.
Given the graphical separation by infection status shown by PCA and factor analysis, Mann-Whitney U tests were performed to test for differences in metabolite concentrations between patients with an infectious CNS disease and those with no CNS infection (MS and controls). All metabolites were included. These results are summarized in Table 2. After correcting for multiple comparisons, significant univariate differences were detected in the concentrations of 29 compounds; these included several metabolites that appeared to drive separation in the PCA (ketones, pyruvate, carnitine, and glycine). Median concentrations of glutamate and 2-oxoglutarate were significantly higher in infectious diseases than patients with no infectious disease, and there was a trend towards higher citrate concentrations in the infectious disease group (p = 0.07). Also, in agreement with the PCA results, median concentrations of isobutyrate, fructose, N-acetylneuraminate, and serine were higher in the noninfectious disease group, and acetate exhibited different distributions between the groups. In a similar univariate analysis on z-scores for 43 variables, nine metabolites were identified (Table 2, among bolded metabolites), all of which were also identified using the previous method.
While CNS infections overlap as a syndrome, they are caused by viruses, bacteria, fungi, protozoa and prions that require different therapies. We therefore evaluated PCA discrimination within CNS diseases without the influence of controls. In the resulting model, PC1 and PC2 cumulatively accounted for 38.9 percent of the variation, and when loadings vectors were overlaid with PC scores, the resulting Gabriel’s biplot revealed the most important metabolites to be ketone bodies, glutamine, glutamate, and threonine. In a scores plot of the first two PCs, moderate separation by disease diagnosis pointed to differential as well as overlapping metabolic patterns among diseases (Fig 2), which were further dissected in additional analyses and are summarized in Tables 3 and 4. After removing ketones and creatinine, factor analysis of z-scores did not separate cleanly between disease groups.
After correcting for multiple comparisons, Kruskal-Wallis tests on untransformed data detected significant differences among diseases and controls in the concentrations of 31 metabolites. Metabolites and diseases for which concentrations were significantly different from control samples according to Dunn’s multiple comparisons tests are shown in Table 4. In particular, the CSF of WNV patients had markedly higher concentrations of pyruvate (p = 0.0008) and formate (p = 0.0005), and Lyme disease and WNV patients shared higher levels of formate and glycine compared to controls. Rabies patients had significantly different concentrations of energy-related metabolites including ketone bodies, lactate and 2-hydroxybutyrate, some of which were also elevated in WNV but not in histoplasmosis or Lyme disease.
CART analysis differentiated infection status with 100% sensitivity and 93% specificity (Table 5). High pyroglutamate alone discriminated WNV, Lyme and histoplasmosis from controls. MS or rabies could be identified from controls with 100% sensitivity and 76% specificity by high 2-hydroxybutyrate or low 2-hydroxybutyrate and high carnitine. Random Forest analyses confirmed the importance of the majority of metabolites identified by CART.
NMR metabolomics distinguished infectious and inflammatory disorders using laboratory-confirmed samples of 5 disorders using 2 approaches to normalization of the data, and 2 unsupervised cluster analytical approaches. CART decision analysis easily differentiated bacterial (Lyme), fungal (Histoplasma) and viral (WNV) causes of encephalomyelitis from controls. Decision analysis also differentiated rabies and the prodromal form of MS from controls, while separation by cluster analyses was incomplete between MS and controls. Notably, the greatest source of variation in metabolomics data found by PCA was the presence or absence of an infectious pathogen. If replicated, this finding is of paramount clinical impact because treatments for infections require almost polar opposite therapeutics than those for autoimmune diseases. There was also substantial agreement in the identification of influential metabolites between different approaches to data normalization and reduction and predictive approaches, including CART and random forest analysis. Metabolites driving separation in PCA (pyruvate, glutamate, quinolinate, 2-oxoglutarate, carnitine, and glycine) potentially suggest alterations in energy metabolism, excitotoxicity and antioxidant response. Patterns of these metabolites were not uniform. Rather, overlapping as well as distinguishing metabolic features were seen, highlighting the potential utility of measuring a suite of metabolites rather than searching for individual metabolic biomarkers for diseases, which may not exist. Overlap of profiles makes strong clinical sense given that EM syndromes overlap in signs and symptoms. The overlap also supports a clinical rationale for syndromic metabolic therapies across a range of infectious or autoimmune causes of EM. Distinguishing features provide promise of rapid, relatively specific diagnoses that enable prompt pathogen or process-directed therapies.
Significant differences by disease group were found in the CSF concentrations of several metabolites known to be involved in the synthesis of the antioxidant glutathione (GSH) and related pathways, including glycine, formate, pyroglutamate, and 2-hydroxybutyrate. The transsulfuration pathway links the methylation cycle of one carbon metabolism to GSH synthesis and produces 2-hydroxybutyrate as a secondary byproduct during the conversion of cystathionine to cysteine [35, 36]. Formate, an endogenous and bacterial metabolite that along with glycine was found at significantly higher levels in WNV and Lyme disease patients compared to controls in this study, is formed as a byproduct in several pathways including the tryptophan kynurenine pathway [37], pterin metabolism [38] and protein demethylation (following hypermethylation by S-adenosyl-L-methionine [39]), while it is also consumed in the folate cycle during the conversion of tetrahydrofolate (THF) to 10-formyl-THF [40]. An end product of purine catabolism, neopterin, has been found to be elevated in patients with rabies [41], Lyme disease, and other neuroinfections, while remaining low in MS and other neuroinflammatory conditions [42]. Pyroglutamate, which converts to glutamate before being incorporated into GSH and also activates amino acid transport systems at the blood brain barrier [43], was higher in histoplasmosis, Lyme disease and WNV and was an important predictor distinguishing these conditions from control samples. Given individual metabolites can participate in a number of biochemical pathways, further studies are required to parse out the mechanisms at play in the diseases studied here. A likely interpretation is that infection or inflammation in the CNS is associated with redox imbalances including glutathione metabolism and NADH/NAD+ ratios. It is of particular interest that these metabolites may profile mechanisms leading to insulin resistance and vascular disease [36], given that low dose insulin therapy was added to the Milwaukee protocol, version 4, with statistical improvements in survival [44].
Our analytical design sought to minimize the effects of starvation/ketosis and dehydration/uremia on the metabolic profile of rabies by prioritizing rabies samples taken four days after admission. Nevertheless, PCA analysis identified the importance of ketone bodies in identifying rabies. Factor analysis that deliberately excluded primary ketones, urea and creatinine from analysis still identified isopropanol and methanol (Table 3), both downstream metabolites of ketones, as discriminators of rabies. RF and CART analyses also identified ketones and carnitine (fatty acid oxidation) as predictors of rabies but not other infections (Table 5). Despite our experimental design, CNS ketosis may be a valid indicator of rabies encephalitis.
This study was originally intended to further explore the specificity of NMR metabolomics for the diagnosis of rabies, which is often confused with Guillain-Barre syndrome, acute psychosis and N-methyl-D-aspartate receptor (NMDAR) encephalitis and currently requires multiple tests for diagnosis at remote reference laboratories. Our findings suggest that the utility of the approach may instead lie in excluding competing diagnoses, many of which are more treatable. NMR metabolomics performed on a par with current rabies diagnostics (100% sensitivity, 76% specificity) and is likely complementary (particularly after 5 days). When restricted to the first week of hospitalization with rabies (when most patients die), NMR metabolomics did not perform as well as for other infections; gene expression studies of rabies CSF and detection of rabies-specific antibodies also performed poorly in the first week. Rabies can clearly be delineated from controls by NMR at later time points, and NMR of CSF also measures recovery [12]. The promise of an NMR metabolomics profile as a proxy marker for therapeutic response would be welcome for rabies, WNV, NMDAR encephalitis or acute disseminated encephalomyelitis for which efficacious treatments remain undefined.
This study is exploratory and is limited by the number of samples available for CNS diseases of rare incidence. The possibility of confounding effects of age, sex, disease stage, or other acute variations in metabolic processes should be considered in interpreting these results. Our control group was aged 5–20 years, while ages in the disease group ranged from 4 to 83 years. However, we confirmed that the distribution of metabolites of our controls overlapped with adult norms reported by the international Human Metabolomics Database (www.hmdb.ca). Further, clear inter-disease differences within groups of adult diseases (MS, WNV) were evident in PCA (Fig 2), suggesting disease was much more influential in driving variation than was age. Sensitivity analyses in rabies in a larger dataset [12] did not identify meaningful age differences, although we cannot exclude the possibility that this might occur for other inflammatory diseases of the CNS. Another potential source of confounding is the timing of sample collection, which was not precisely known for samples other than rabies. All forms of encephalitis are treated empirically upon hospitalization, so early diagnostic samples such as those analyzed here may reflect early empirical therapies that often overlap (e.g., rehydration, provision of glucose, use of antibacterials, sedation) but may also differ between diseases. Our choice of rabies samples centered on the fourth day of hospitalization was intended to minimize effects of dehydration and malnutrition, but may have biased rabies samples toward normality. Finally, differences in some metabolites should be interpreted with caution, since low concentrations in some specimens precluded exact quantification (carnitine and glycine), which may have artificially led to statistical differences. Other metabolites (glutamine and pyroglutamate) are potentially affected by protein removal [45], although this has not been shown in CSF.
This study provides justification for further analysis of samples from these and other causes of encephalomyelitis. Several prominent and as of yet unidentified peaks observed in the spectra of some patients may indicate the presence of important metabolites involved in disease pathogenesis that have not yet been elucidated. While further studies with larger sample sizes will be needed to determine the clinical utility of NMR in the diagnosis of EM, NMR or other ‘omics technologies may in the future serve as a rapid initial screening test that would allow medical practitioners to initiate treatment with antivirals or biological immune modifiers, while patient samples can then be triaged to appropriate reference laboratories for confirmation without delaying treatment. Rabies and many arbovirus reference laboratories require specialized containment facilities, immunization of laboratory workers, and highly trained personnel who perform subjective assays such as immunofluorescence. Reference laboratories for rabies, arboviruses, bacteria and fungi are often dispersed geographically, leading to substantial requirements in volume, delay, and cost for diagnosis of encephalomyelitis when all are considered. NMR and MS instruments, on the other hand, exist at most research universities, i.e. at a state or provincial rather than national level. NMR analytical procedures are easily standardized and permit detection of multiple diseases using a single experiment, as illustrated here. NMR spectra can be transmitted electronically for analysis, which can be automated [46]. Decision analytical approaches such as CART and RF offer diagnostic flow charts that are easily implemented once validated, with quantifiable diagnostic probabilities. Considering current challenges, its relative ease of use makes NMR metabolomics of CSF a potentially important tool for emergent diseases and distinguishing between autoimmune and infectious EM.
|
10.1371/journal.pcbi.1005458 | Achieving global perfect homeostasis through transporter regulation | Nutrient homeostasis—the maintenance of relatively constant internal nutrient concentrations in fluctuating external environments—is essential to the survival of most organisms. Transcriptional regulation of plasma membrane transporters by internal nutrient concentrations is typically assumed to be the main mechanism by which homeostasis is achieved. While this mechanism is homeostatic we show that it does not achieve global perfect homeostasis—a condition where internal nutrient concentrations are completely independent of external nutrient concentrations for all external nutrient concentrations. We show that the criterion for global perfect homeostasis is that transporter levels must be inversely proportional to net nutrient flux into the cell and that downregulation of active transporters (activity-dependent regulation) is a simple and biologically plausible mechanism that meets this criterion. Activity-dependent transporter regulation creates a trade-off between robustness and efficiency, i.e., the system's ability to withstand perturbation in external nutrients and the transporter production rate needed to maintain homeostasis. Additionally, we show that a system that utilizes both activity-dependent transporter downregulation and regulation of transporter synthesis by internal nutrient levels can create a system that mitigates the shortcomings of each of the individual mechanisms. This analysis highlights the utility of activity-dependent regulation in achieving homeostasis and calls for a re-examination of the mechanisms of regulation of other homeostatic systems.
| Homeostasis, the ability to maintain relatively constant internal conditions in the face of fluctuating environments, is fundamental to many biological processes. In nutrient homeostasis, a model homeostatic system, homeostasis is typically thought to be achieved through negative feedback regulation of the plasma membrane transporters synthesis by intracellular nutrient levels. Here, we first derive the general conditions that can achieve global perfect homeostasis in a simple uptake system. We found that this condition can be satisfied by the ubiquitous but less studied mechanism of activity-dependent transporter downregulation. If transporter downregulation is dependent on nutrient uptake rates, i.e., activity-dependent downregulation, the system in principle can achieve homeostasis in any external environment. Activity-dependent and internal regulation can synergize to achieve homeostasis across a wide set of conditions at minimal energetic cost. Activity-dependent downregulation is likely to play a role in many diverse homeostatic systems.
| Cells must maintain relatively constant internal concentrations of nutrients even though the supply of nutrients from the environment can fluctuate wildly, a process called nutrient homeostasis [1,2]. In microorganisms, errors in nutrient homeostasis can have dramatic effects on growth, since low internal nutrient concentrations limit growth, while excessive internal nutrient concentrations can be toxic [3,4]. In mammalian cells, nutrient uptake, cell growth, and proliferation are controlled by the overlapping signaling pathways [5,6] and defects in nutrient regulation play a role in the pathogenesis of diseases such as cancer and diabetes [1,7–9]. Nutrient homeostasis is a major determinant of both organismal and cellular fitness.
There are two axes that are important for homeostasis. The first axis is the robustness of homeostasis: the more robust the homeostasis, the smaller the change in internal nutrient concentration for a given change in external nutrient concentration. The limit of robust homeostasis is when the internal nutrient concentration is completely insensitive to the external nutrient concentration, a condition we refer to as 'perfect' homeostasis. The second axis is the range of homeostasis. The wider the range of homeostasis, the large the range of external nutrient concentrations over which the system achieves a given robustness of homeostasis. Global homeostasis occurs when the system is homeostatic regardless of the external nutrient concentration. In this work, we solve for conditions that achieve global perfect homeostasis.
There are many examples of biological systems that exhibit homeostasis [10–14]. In general, homeostasis can arise from fine tuning kinetic parameters or from structural properties of the regulating network [1,2,12,14]. Achieving nutrient homeostasis requires cellular circuitry that is able to sense nutrient levels and then regulate uptake and/or usage accordingly. All nutrient homeostatic systems need a plasma membrane transporter that allows passage of the nutrient through the plasma membrane. The majority of nutrient homeostatic systems share a common architecture where the synthesis of this plasma membrane transporter is under the regulation of nutrient levels (Fig 1) [3,4,11]. This regulation is a negative feedback system such that when nutrient concentrations are low, transporter synthesis is increased and when nutrient synthesis is high transporters, synthesis is decreased [1,5,6,11]. In eukaryotes, this type of regulation has been demonstrated for metal ions [10,15–18], sugars [1,2,19,20], phosphate [3,4,21], and amino acid transport [6,22–24]. While the mechanistic details of this design can vary, e.g. regulation of synthesis through transcription [1,7–9,25] or trafficking [10,26], regulation of transporter synthesis is typically assumed to be the critical factor in nutrient homeostasis. It has been shown that this negative feedback regulation makes nutrient homeostasis more robust and this robustness depends on the sensitivity of the transporter synthesis rate to nutrient levels [1,2,12–14].
A second and equally widespread motif in homeostatic systems is post-translational downregulation of transporters [28–32]. As with regulation of synthesis, the mechanistic details of this design can vary; e.g., plasma membrane transporters can be inactivated by modification, sequestrations, or degradation. In yeast, this mode of regulation has been demonstrated for many different homeostatic systems including ion transporters such as zinc, copper, and iron [33–35]; sugar transporters such as glucose and maltose transporters [25,36]; a phosphate transporter [37]; and amino acid transporters [25,38]. This mechanism is usually considered a stress response to an extreme change in nutrients [39]. Indeed, the response to extreme changes is a homeostatic process. Transporter downregulation has been less well studied than transcriptional regulation of transporter synthesis. In the best-characterized systems, transporter downregulation is mediated by ubiquitination [25,40–42] but the theoretical cost and benefits of using transporter downregulation to achieve homeostasis are not understood.
Mathematically, regulation by transporter synthesis and transporter downregulation are interconvertible at steady-state; i.e., adding a term to the transporter synthesis flux (thin green arrow in Fig 1) and adding 1 over that term to the transporter downregulation flux (thin red arrow in Fig 1) yields the same steady state solution. Yet, biological realizations of some mathematical terms are not easily achieved. Hence, these two forms of regulation may not be biologically interconvertible. We determined that if transporter levels are inversely proportional to flux through the plasma membrane transporters, global perfect homeostasis is achieved. Flux sensing is distinct from the internal and external nutrient sensing that is typically considered to regulate nutrient homeostatic systems. Work from the Heinemann lab [43,44] has shown that flux sensing is used for the regulation of some intracellular metabolism. We showed that flux sensing based homeostasis can be easily achieved if active transporters are downregulated. We will refer to this as activity-dependent downregulation of transporters. Given that both transporter synthesis and downregulation are regulated in most nutrient homeostatic systems, we sought to determine the potential trade-offs of each architecture [45–48]. The combination of activity-dependent downregulation and regulation through synthesis makes a more efficient system than either mechanism alone.
To define homeostasis we consider a dynamical system that has an output, e.g. the internal nutrient concentration (Sint), which depends on other variables such as external nutrient concentration (Sext) and transporter concentration (T). The output of the system achieves global perfect homeostasis with respect to a specific variable if the stationary concentration of the output is invariant to perturbations in this variable. Formally, for the example of nutrient transport, Sint would achieve global perfect homeostasis with respect to Sext if Sint has a steady state such, Sintss, such that ∂Sintss(Sext,Tss(Sext))∂Sext=0 for every value of Sext. Some systems may exhibit perfect homeostasis for a range of Sext values; we refer to this as local homeostasis. Additionally, to quantify the dependence of the SintSS on Sext, we defined a unitless parameter that is the steady state value of Sint for a given Sext normalized by the maximal steady state value of Sint over the range of relevant Sext, r(Sext)=Sintss(Sext,Tss(Sext))maxSext=[0,∞)Sintss(Sext,Tss(Sext)). We will refer to, r, as the robustness of homeostasis. A system achieves global perfect homeostasis when r = 1 for every Sext. When the system is not perfectly homeostatic, r can be any value between zero and one and this value can change as a function of Sext. Biological systems that approach this limit of perfect homeostasis, i.e. r is close to 1, are often still considered homeostatic although there is no standard value for r which segments between whether or not a system is considered homeostatic [10]. In this work, we look for the biological circuitries that achieve global perfect homeostasis. Note that homeostasis as we define it is a property of the steady state concentration; during transitions between different steady states, the homeostatic output can transiently change.
We first we sought to determine all conditions that could lead to global perfect homeostasis in a general uptake system (Fig 1). Our system is composed of an external nutrient (Sext), an internal nutrient (Sint), and a plasma membrane transporter (T) (Fig 1). Transporters allow nutrient to pass into the cell through the plasma membrane and can be both synthesized and destroyed. While there are more molecular players, e.g., mRNA and translocation machinery, this system encapsulates the key biological variables while subsuming the rest of the players into the parameters. We use the following notation convention: Greek symbols denote fluxes (with units of concentration/time), the symbol k denotes rate constants (with units of 1/time or 1/concentration/time depending on the reaction order), and u denotes nutrient flux per transporter. This system (Fig 1) can be described by two ordinary differential equations,
S˙int=T⋅u(Sext,Sint)−γS(Sext,Sint)T˙=αT(Sext,Sint,T)−kγT(Sext,Sint)⋅T,
(1)
where u is the rate of nutrient uptake per transporter, γs is the nutrient usage flux, αT is the transporter synthesis flux, and kγT is the transporter downregulation rate. We made the standard simplifying assumption that transporter downregulation flux is linearly proportional to transporter levels, γT=kγT(Sext,Sint)⋅T [49–51]. In theory, the system could be further generalized by making u, γs, and kγT arbitrary functions of T, but this is not supported by the biology of any of the commonly studied nutrient uptake systems. Simplified versions of this system, such as Eq (1), have been used to show that internal nutrient-dependent regulation of transporter synthesis can be homeostatic and thereby has provided a rationale for the ubiquity of this architecture [10].
To define the necessary and sufficient conditions for global perfect homeostasis we applied the method of Steuer et al. [14]. The conditions that are necessary and sufficient for homeostasis of this system are (S1 Appendix, sections I, II):
∂Sext(kγT(Sext,Sint)/αT(Sext,Sint,T))kγT(Sext,Sint)/αT(Sext,Sint,T)⋅(1−T⋅∂TαT(Sext,Sint,T)αT(Sext,Sint,T))=∂Sextu(Sext,Sint)u(Sext,Sint)−∂Sextγs(Sext,Sint)γs(Sext,Sint).
(2)
This relationship is complicated and it is hard to imagine the regulatory interactions that would allow a biological instantiation of this general system. But Eq (2) does yield the insight that global perfect homeostasis is achieved by regulating transporters levels (T) such that they compensate for the change in usage rate (γs) or uptake per transporter (u). In the following sections, we will constrain this general system; this will reduce Eq (2) to a condition that has a clear biological interpretation.
We sought to determine special cases of the general uptake system described in Eq (1) where the resulting homeostatic criterion is achievable by biologically plausible mechanisms. We started with the following biologically reasonable and standard assumptions: 1) there is little or no evidence for transporter levels directly affecting transporter synthesis, αT(Sext, Sint); 2) Sext negligibly affect nutrient usage, γS(Sint); and 3) internal nutrients negligibly affect nutrient uptake per transporter, u(Sext). Under these simplifying assumptions, Eq (1) reduces to:
S˙int=T⋅u(Sext)−γS(Sint)T˙=αT(Sext,Sint)−kγT(Sext,Sint)⋅T.
(3)
The criterion for homeostasis, Eq (2) (S1 Appendix, section II), reduces to:
kγT(Sext,Sint)αT(Sext,Sint)∝u(Sext)⋅func(Sint)
(4)
where func(Sint) is a general function that depends solely on Sint (and could also be constant) (Fig 2). This condition states that any system in which the regulation of the transporter does not explicitly depend on Sext cannot provide global perfect homeostasis. Moreover, as transporter levels at steady state are given by αT/kγT, this criterion is satisfied when the transporter level, at steady state, is inversely proportional to the nutrient uptake rate. When this condition is met, the solution for the steady state value of Sint is given by solving
0=func(Sint)−γs(Sint).
(5)
While, this condition does not guarantee a steady state solution for Sint, when a solution exists, it is independent of Sext.
Is there a biological mechanism that can satisfy Eq (4) by making transporter downregulation proportional to nutrient uptake per transporter, i.e. kγT(Sext,Sint)∝u(Sext)? This condition corresponds to the requirement that only transporters that are actively transporting have the potential to be downregulated. We postulated that a simple biological scheme could achieve activity-dependent downregulation and thereby satisfy Eq (4). This scheme is composed of a standard transporter cycle, where: 1) external nutrient binds to a transporter, 2) the transporter undergoes a conformational change allowing the nutrient to be released on the opposite side of the membrane, and 3) the transporter returns to its original conformation. In addition to this core system, we added an enzyme that recognizes and modifies only the nutrient bound conformation of the transporter (Fig 3). The modified transporter is downregulated. As long as the process is irreversible, direct and indirect inactivation are equivalent. Indeed, this system almost trivially couples the uptake and downregulation rate making them directly proportional (Fig 3 and S1 Appendix, section IV).
Two studies directly link uptake and downregulation [52,53]. In both cases, nutrient uptake leads to a conformational change in the transporter that is then ubiquitinylated in a manner analogous to our scheme in Fig 3. This basic scheme is also supported by a series of crystal structures that show the conformational changes that occur upon nutrient binding [54,55].
While we only know of two proven examples for transporters, there are many examples of activity-dependent downregulated receptors. For example, ligand-dependent modification and downregulation is a core feature of G protein-coupled receptors and is required for ligand-mediated desensitization [56–62]. In bacteria, methyl-accepting chemotaxis receptors are modified in a ligand-dependent manner and this modification affects their sensing [63,64]. This activity-dependent methylation was demonstrated to be critical for robust adaptation [65] a process that is similar to homeostasis. While these examples involve receptors, transports share many features with receptors. Nutrient sensors and transporters are high related [66–71]. Multiple transporters have both transporting and signaling functions [72–78] and point mutations can interconvert receptors and transporters [79]. In some cases, receptor-mediated endocytosis [80–82] can even be considered a hybrid of transport and downregulation. Furthermore, transporters are modified and internalized in a manner that is very similar to receptors. Many transporters undergo internalization and degradation in a ubiquitin-dependent manner.
Together, these examples argue that the proposed activity-dependent mechanism is not just mathematically possible but likely ubiquitous. The paucity of examples is likely due to a lack of experiments that have been performed in a manner such that activity-dependent regulation would have been observed. Indeed, changing the external nutrient concentration can stimulate transporter degradation [21,28,30,32–34] or inactivation [30,37,52,83,84] consistent with a ubiquitous role of activity-dependent regulation.
When activity-dependent downregulation is the only form of transporter downregulation, global perfect homeostasis can be achieved. But, in real systems, dilution from cell growth and basal protein degradation will always contribute to transporter downregulation. To isolate the impact of dilution and protein degradation on global perfect homeostasis we considered the following minimal system:
S˙int=kcat⋅T⋅SextSext+Kext−kγS⋅SintT˙=αT−(kγTc+kγTa⋅SextSext+Kext)⋅T.
(6)
In this case, αT is constant, and the uptake per transporter has a standard Michaelian form, u=kcatSextSext+Kext. kγTa is maximal activity-dependent downregulation rate constant and kγTc is the combined rate constant for all other downregulation processes (Fig 4A). We additionally assumed that nutrient uptake is Michaelian, while this was not essential, it is the standard assumption for nutrient uptake kinetics.
Given that basal degradation and dilution are not homeostatic, the robustness of homeostasis, as quantified by our robustness metric, r, depends on the relative magnitude of kγTa and kγTc (Fig 4B). As kγTa/kγTc increases, the system, Eq (6), becomes more robust to changes in Sext (Fig 4B and 4C). In the limit where activity-dependent downregulation dominates, kγTa >> kγTc, homeostasis is achieved (S1 Appendix, section V); in the limit where the degradation and dilution dominates, kγTa << kγTc, Sint tracks Sext. While Sint is robust to changes in Sext when kγTa >> kγTc, transporter levels are not. Instead, T tracks Sext and serves as the latent variable that allows the system to be robust; T adapts to keep the uptake rate, kcat⋅T⋅SextSext+Kext, constant (Fig 4D). The decrease in robustness when basal degradation dominates is mirrored by a decrease in the sensitivity of the T to Sext (Fig 4D).
We wished to explore whether other common forms of regulation could achieve global perfect homeostasis. The criterion of Eq (4) could also be satisfied by a sensor that directly measures the nutrient flux (Fig 5A) or a sensor with identical binding kinetics as the transporter (Fig 5B). This sensor could then regulate transporter synthesis or downregulation. Recent theoretical and experimental works from Kotte et al. and Kochanowski et al. have described the existence and role that flux sensing can play in metabolic regulation [43,44] and some nutrient systems contain external nutrient sensors [67]. If either of these sensors led to the downregulation of nutrient transporters it would be functionally equivalent to activity-dependent downregulation. In fact, the enzyme that modifies the nutrient bound transporter in Fig 3 is effectively acting as a flux sensor. But, both of these other mechanisms would require an extra level of regulation to normalize for the number of sensor molecules. In the case of an internal flux sensor, the activity of the sensor depends on the total nutrient flux, T⋅u(Sext). Simple molecular interactions between the transporter and sensor would lead the downregulation rate of the transporter to depend on the activity of the transporters multiply the number of transporters, i.e. square of the transporter concentration, u(Sext)⋅T2.
In the case of an external sensor, the activity of the sensor depends on binding the external nutrients, so the transporter downregulation will have the form kγT(Sext,Sint)∝TSextSext+Ksensor. Since uptake has the form of kγT(Sext,Sint)∝u(Sext), this mode of regulation will only achieve homeostasis when the sensor's Michaelis constant is close to transporter's Michaelis constant (Fig 5C; S1 Appendix, section VI). As these constants deviate, the system loses the ability to be robust to changes in external nutrient concentration (Fig 5C). Therefore, while all three mechanisms are biologically feasible, we believe activity-dependent downregulation will be the most common.
Above we described how activity-dependent downregulation can achieve global perfect homeostasis. This system is distinct from transcriptional regulation of transporter synthesis by high gain feedback of internal nutrient concentrations (e.g. negative feedback with high cooperativity) which is considered to be a homeostatic mechanism [10]. In this system, it is typically assumed that the synthesis rate of the transporter is decreasing as a function of the internal nutrient concentration, e.g. αT(Sext,Sint)=kαT(KMnKMn+Sintn) and that the transporter downregulation rate is constant,kγT(Sext,Sint)=γT. It is easy to see that in this case the condition in Eq (4) is not met, i.e. kγT/αT does not depend on Sext, and thus this system cannot achieve global perfect homeostasis for any finite n (Fig 6; S1 Appendix, section VII).
Under real biological conditions, neither internal nutrient sensing nor activity-dependent sensing can achieve global perfect homeostasis. The two regulatory systems can be compared based on the parameters required to achieve the same robustness, r (Fig 7). Internal nutrient sensing approaches global perfect homeostasis when n is large. But, high cooperativity is mechanistically difficult to achieve. Activity-dependent downregulation approaches global perfect homeostasis when kγTa >> kγTc. High levels of activity-dependent downregulation are easy to achieve mechanistically but might come at a high cost due to increased protein turnover.
To explore the potential trade-off due to cost that comes with activity-dependent regulation, we used two metrics: efficiency and robustness. We defined efficiency as the total nutrient uptake of a single transporter over its average lifetime. Robustness is defined above. When there is no activity-dependent downregulation, robustness takes on its minimal value and efficiency its maximal value (Fig 8A; S1 Appendix, section VIII). As kγTa/kγTc is increased, robustness increases but efficiency goes down.
As most nutrient homeostatic systems likely utilize both internal nutrient and activity-dependent regulation, we tested whether the combination improved the robustness-efficiency trade-off. We first asked whether the combined system is still able to achieve the global perfect homeostasis that activity-dependent downregulation is able to achieve alone. Indeed, the combined system still satisfies Eq (4) and thus can still achieve global perfect homeostasis (Fig 2). But this combined system might have additional desirable properties over a system with just activity-dependent regulation. Indeed, transcriptional repression by internal nutrient levels improves the efficiency of the system for any given robustness (Fig 8B). As n is increased, the minimal robustness of the system is also increased (Fig 8B). In addition to the minimal robustness increasing, the trade-off between robustness and efficiency also has a higher slope such that for the same efficiency, higher robustness is achieved and vice versa (Fig 8B; S1 Appendix, section VIII). This result is intuitive—when nutrient levels are high, instead of just degrading transporter, the combined systems allows fewer transporters to be made.
We analyzed a general nutrient uptake system and derived the conditions for global perfect homeostasis. For a large family of scenarios, the internal nutrient can be made independent of the external nutrient if the ratio of transporter synthesis and depletion rates explicitly depends on the net uptake rate of the nutrient from the environment. A simple way to achieve this regulation is for active transporters to be preferentially downregulated, that is, have activity-dependent downregulation.
While the transcriptional regulation of transporter gene synthesis has been relatively well characterized, the mechanisms that lead to the downregulation of plasma membrane transporters are less clear. Interestingly, a recent study of the yeast uracil transporter, Fur4p, suggests it exhibits properties that cause its degradation to depend on its activity. Specifically, there is evidence that binding of nutrients to the transporter causes it to adopt a conformation that marks it for ubiquitinylation, followed by endocytosis and degradation [52]. While this detailed analysis has not been performed on many transporters, it’s known that many transporters are endocytosed when nutrient levels increase. While previous work had suggested that this likely serves to protect cells from toxicity during acute increases in nutrient levels [39], we propose that this mechanism could also play a crucial role in homeostasis at all external nutrient concentrations.
Homeostasis can be regulated by internal and external nutrient concentrations. Both forms of regulation have limitations that affect their utility over all external concentrations, but both are useful in a range of biologically relevant regimes. Systems that integrate both internal and external nutrient concentrations can be robust over a wider range of concentrations (with lower total energy input) than systems that sense only internal or only external nutrient concentration. This complementarity may explain why these two modes of regulation are common in homeostatic systems.
While out of the scope of this work, in future work it will be interesting to expand on this simple system in several ways. While here we focused on steady-state behaviors, kinetic differences between different modes of regulation are likely important. For example, different mechanisms of action can accentuate kinetic difference, e.g. transcriptional versus post-translation control. In addition, many homeostatic systems utilize internal nutrient storage or nutrient recycling which in principal can affect the homeostatic response. Furthermore, analysis of addition constraints that can be placed on Eq (1) could identify other biologically achievable systems that satisfy the homeostatic relationship in Eq (2).
The results we derived here are relevant for any multi-compartment biological system that implements homeostasis, where there is flux among the compartments. This work thus should be useful as a guide in studying homeostasis in any biological system and well as in the design of synthetic ones.
Detailed derivations can be found in S1 Appendix.
To determine the requirements for homeostasis for the general uptake system described in Eq (1), we used the method developed in [14]. In brief, we required the rule
rank(P|I)=rank(I)
(7)
where P is a column of elements pi=∂log(σi)∂log(Sext) with σi = αS, γS, αT, γT, which are the uptake and usage fluxes of the internal nutrient and uptake and usage fluxes for transporter respectively; each of those could be a function of Sext, Sint, and T. I is the matrix representation of largest parameter-independent subspace spanned by the columns of (M | K) where M is a column of elements mi=∂log(σi)∂log(T) with σi = αS, γS, αT, γT, and K=(αS00γS000αT00γT0) where αS0,γS0,αT0,γT0 are the steady state solutions of fluxes.
We constructed P|I and I matrixes and analytically found all the conditions that satisfy Eq (7). The homeostasis condition was analytically derived using the described above criterion. Criterion necessary for global perfect homeostasis for special cases of Eq (1) were analytically derived by substituting into the resulting general solution with simplified forms of αS, γS, αT, γT.
We considered several simplified conditions. If transporter production is independent of transporter concentration the following homeostasis condition was obtained
γT(Sext,Sint,T)αT(Sext,Sint)∝αS(Sext,Sint,T)γS(Sext,Sint)⋅func(Sint).
(8)
where func(Sint) is any function of internal nutrient concentration.
To evaluate homeostasis we used two metrics: efficiency: e = αS / γT - the total nutrient uptake of a single transporter over its average lifetime and robustness: r = Sint / Sint,max where Sint,max is the maximum Sint across all Sext.
To compare activity-dependent transporter downregulation with the internal nutrient based transcriptional regulation we compared the following two systems:
S˙int=kcat⋅T⋅SextSext+Kext−kγS⋅SintT˙=αTKsynnKsynn+Sintn−kγTcT
(9)
and
S˙int=kcat⋅T⋅SextSext+Kext−kγS⋅SintT˙=αT−kγTSintnKdegn+Sintn⋅T
(10)
where Ksyn and Kdeg are saturation constants and n is the Hill coefficient.
To compare a system with internal nutrient based regulation of synthesis to a system with activity-dependent transporter downregulation we analytically calculated the parameters required to achieve a threshold robustness of r = 0.95 and r = 0.8 (Fig 7). In addition, we analytically compared the efficiency-robustness dependence in a system with both internal nutrient based regulation of synthesis and activity-dependent transporter downregulation (n = 0, 1 in Fig 8; n = 0, 1, 3, 10 in S2 Fig).
All analytical solutions were found in Mathematica version 10.2 (Wolfram Research) and were visualized in MATLAB R2015a (MathWorks).
|
10.1371/journal.pbio.2005653 | Structural basis for overhang excision and terminal unwinding of DNA duplexes by TREX1 | Three prime repair exonuclease 1 (TREX1) is an essential exonuclease in mammalian cells, and numerous in vivo and in vitro data evidenced its participation in immunity regulation and in genotoxicity remediation. In these very complicated cellular functions, the molecular mechanisms by which duplex DNA substrates are processed are mostly elusive because of the lack of structure information. Here, we report multiple crystal structures of TREX1 complexed with various substrates to provide the structure basis for overhang excision and terminal unwinding of DNA duplexes. The substrates were designed to mimic the intermediate structural DNAs involved in various repair pathways. The results showed that the Leu24-Pro25-Ser26 cluster of TREX1 served to cap the nonscissile 5′-end of the DNA for precise removal of the short 3′-overhang in L- and Y-structural DNA or to wedge into the double-stranded region for further digestion along the duplex. Biochemical assays were also conducted to demonstrate that TREX1 can indeed degrade double-stranded DNA (dsDNA) to a full extent. Overall, this study provided unprecedented knowledge at the molecular level on the enzymatic substrate processing involved in prevention of immune activation and in responses to genotoxic stresses. For example, Arg128, whose mutation in TREX1 was linked to a disease state, were shown to exhibit consistent interaction patterns with the nonscissile strand in all of the structures we solved. Such structure basis is expected to play an indispensable role in elucidating the functional activities of TREX1 at the cellular level and in vivo.
| Three prime repair exonuclease 1 (TREX1) was shown to participate in various cellular events such as DNA repair, immunity regulation, and viral infection. In addition to relating to autoimmune diseases, this exonuclease also acts as a potential protein target for anticancer or antiviral therapies. A key for such broad attendance of TREX1 is the activities of precise trimming of the 3′-overhang in a double-stranded (dsDNA) and breaking of the terminal base pairing of the duplex. Here, we designed a series of structural DNA substrates and activity assays to delineate the underlying mechanisms. The structures newly resolved in this work indicated that the Leu24-Pro25-Ser26 cluster of TREX1 is essential for the enzyme to carry out the aforementioned activities. Together, our results established an integrated structure view into the versatile exonuclease functions of TREX1 and illuminated the molecular origin for the unique catalytic properties of TREX1 in processing various DNA intermediates in DNA repair and in cytosolic immunity regulation.
| Three prime repair exonuclease 1 (TREX1) is a member of the DEDDh family of exonucleases and accounts for most of the 3′–5′ exonuclease activity in mammalian cells [1,2]. Anchored in the plasma membrane of the endoplasmic reticulum (ER) through the C-terminal domain, TREX1 degrades a variety of substrates to prevent initiation of autoimmunity [3–8]. The targeted nucleic acids in this activity include single-stranded DNA (ssDNA) [3,5] and double-stranded DNA (dsDNA) [9–11]. It was also suggested that DNA/RNA hybrids are potential targets of TREX1, since the deficiencies of TREX1 demonstrate similar features of autoimmune diseases as those of RNase H2 that was known for processing the hybrid substrates [12–15]. Such genetic materials in the cytoplasm are mostly originated from replication of aberrant DNA intermediates and possibly also due to unrestrained endogenous retroelements [8,13–17].
Malfunctioning of TREX1 has thus been shown to lead to inflammation and autoimmune diseases such as inflammatory myocarditis in Trex1-/- mice [18] and systemic lupus erythematosus (SLE), Aicardi-Goutières syndrome (AGS), retinal vasculopathy, cerebral leukodystrophy, and familial chilblain lupus (FCL) in TREX1-deficient humans [14,19,20]. The ability of TREX1 in modulating immune responses is also exploited by type 1 human immunodeficiency retrovirus (HIV-1) that enslaves cytosolic TREX1 to degrade its cDNA, thereby escaping detection by the nucleic acid sensors of the infected host and hence the concomitant antiviral responses [21]. Recent studies indeed showed that TREX1 knockdown in human tissues and humanized mice delayed HIV infection and suppressed local viral replication with increased production of type 1 interferons [22].
Moreover, TREX1 was shown to respond to the events of DNA damage or Granzyme A (GzmA)-mediated apoptosis with concomitant translocation to the nucleus [3,23,24]. TREX1 was indeed characterized as an exonuclease that involves in DNA repair and in DNA proofreading via working with other nuclear enzymes such as Poly [ADP-ribose] polymerase 1 (PARP-1) and DNA polymerase β [1,24,25]. Additionally, TREX1 was established to play a role in GzmA-mediated cytotoxicity [23]. Cytotoxic T lymphocytes and natural killer cells secrete GzmA proteases to induce cell death of the targeted cancer cells or virus-infected cells by releasing TREX1 and nonmetastatic protein 23 homolog 1 (NM23-H1) from ER-bound patient SE translocation protein (SET) complex, and their translocation to the nucleus promoted chromosomal DNA fragmentation [23]. With such diverse activities toward a variety of nucleic acid substrates—including ssDNA, dsDNA, DNA/RNA duplexes, structural DNAs, and DNA hybrids with an interject nick—the linkage of TREX1 to apoptotic DNA degradation in dying cells [26] and in chromosomal fragmentation during telomere crisis [27] was hence not unexpected. Regulation of the exonuclease activity of TREX1 was indeed recognized as an influential factor in cancer therapy [28–30].
The broad attendance of TREX1 in the activities of immune silencing and responding to genotoxic stresses highlights the elementary role of properly processing the nucleic acids that are involved in these vital yet very complicated cellular functions. Mechanistic understanding of substrate handling is hence key to comprehending the connections between different enzymes in the pathways. However, such knowledge is mostly elusive as informative molecular details are lacking. For example, precise excision of short 3′-overhangs in generation of blunt-end dsDNAs and the terminal unwinding of these duplexes for further degradation are two events in urgent need of structure basis. Although X-ray structures of TREX1 with ssDNA [7,31] and Y-structural DNA are available [10], the 3′- and 5′-overhang of the substrates therein were rather long (>4 nt). Therefore, the structure information is currently insufficient to deduce the mechanism by which double-stranded substrates are processed (see Discussion).
TREX1 is, in fact, unique in the DEDDh family for the activity of dsDNA degradation, which enables its participation in preventing autoimmune responses and in restricting retrotransposons [9–11]. The DEDDh family, also named DnaQ-like exonuclease family, or RNase T superfamily in the protein families (Pfam) database (https://pfam.xfam.org/), contains over 17,000 members across more than 3,000 species [32,33]. Toward ssDNA substrates, the catalytic properties of TREX1 are similar to those of the Escherichia coli homolog, RNase T. When first reported, both enzymes were shown to exhibit activities involved in DNA repair, namely trimming the single-stranded segments of the mispaired ends in a DNA duplex, such as the single-stranded 3′-overhangs in duplex, Y-, or flap structural DNAs [1,34,35]. The analogy in this activity is in line with the in vivo and in vitro experiments that concluded that the two nucleases likely play similar roles in various pathways of DNA repair [24,33]. Furthermore, TREX1 can digest single-stranded RNA (ssRNA) as RNase T does, and this RNase activity of TREX1 is potentially responsible for degrading DNA/RNA hybrids in the cytosol [13,14] and/or during tRNA maturation [36]. The main difference between TREX1 and RNase T, though, is the catalytic ability of TREX1 in degrading dsDNA, for which RNase T lacks [35,37].
In this work, we solved 4 X-ray structures of TREX1 with representative DNA substrates to delineate the molecular details of the multifaceted catalytic properties of TREX1. On one hand, TREX1 allows precise trimming of structural DNAs to generate blunt-end products in preparation for specific pathways of DNA repair. On the other hand, the enzyme enables digestion of dsDNA for immunity regulation. The 4 new structures we presented here reveal how TREX1 removes the single-stranded 3′-overhang in producing DNA duplexes with blunt ends and, for the first time, illustrate how the enzyme overcomes the resistance of the double-strand structure for degradation of dsDNA. The structural DNA substrates in our analysis were also specifically designed to mimic the DNA intermediates in various DNA repair pathways. Therefore, the new structures presented in this work provide unprecedented insights into the functional roles of TREX1 at the molecular level. Together with measurements of nuclease activities, the structure-based mechanism was also employed to discuss the cellular functions of TREX1 in DNA repair and in immune silencing.
We determined the structures of TREX1 complexed with 4 different substrates: (i) the TREX1-deoxyinosine (dI)-ssDNA structure at the 2.3 Å resolution is with a ssDNA containing a dI (dI-ssDNA); (ii) the TREX1-dI-T-dsDNA structure at the 3.4 Å resolution is with a dsDNA with a scissile strand containing a dI (dI-T-dsDNA); (iii) the TREX1-L-structural dsDNA structure is with a duplex DNA that the two termini of 1-nt- and 4-nt-long 3′-overhang form an S-Tetromino shape (referred to as L-structural dsDNA based on the shape of terminals); and (iv) the TREX1-Y-structural dsDNA structure is with a Y-structural dsDNA that the 3′- and 5′-ends of noncomplementary sequences form a Y-shape structure (Y-structural dsDNA). The crystallization conditions and the diffraction and refinement statistics of these structures are listed in Fig 1 and S1 Table. The crystals of these structures were all grown using mouse TREX1 with the C-terminal transmembrane domain truncated.
Nuclease activity assays showed that the exonuclease activity of TREX1 in the crystallization conditions of TREX1-dI-ssDNA, TREX1-dI-T-dsDNA, and TREX1-Y-structural dsDNA was inhibited, and the nucleic acid substrate in these structures was thus intact. The dsDNA of the TREX1-dI-T-dsDNA structure was further labeled with γ-32P and analyzed using 20% denaturing polyacrylamide gels to confirm that dI-T-dsDNA was indeed intact in the crystal (S1A and S1B Fig). The nuclease activity of TREX1 was enhanced in the crystallization conditions of the TREX1-L-structural dsDNA structure. In this case, the input DNA substrate was digested by TREX1 into 2 small ssDNA fragments, including 6-nt- and 9-nt-long ssDNA (6-nt-long ssDNA: 5′-GGCCCT-3′ and 9-nt-long ssDNA: 5′-GGCCCTCTT-3′). The 2 ssDNA strands then hybridized to form a duplex DNA with 1-nt- and 4-nt-long 3′-overhang in the crystal. Two magnesium ions (Mg2+) were observed at the TREX1 active site in all of these structures, even though that magnesium was not additionally added in preparing the crystallization conditions. These Mg2+ were likely from the expression hosts as similar phenomena had been observed in other structures of DEDDh exonucleases [35,37]. In the following, we first present the results of activity assays to characterize the catalytic properties of TREX1 against various substrates. Next, we discuss the 4 structures that collectively reveal the molecular principles for handling such diverse substrates.
Two members of DEDDh exonucleases, Thermus thermophiles TTHB178 and E. coli RNase T, were identified as DNA repair–related exonucleases [33,38]. Both enzymes can remove deaminated bases like dI (also named hypoxanthine) and uracil in ssDNA with similar efficiencies as in digesting regular nitrogenous bases in vitro [33,38]. In addition, in vivo studies showed that TTHB178 and RNase T exhibited similar responses to the genotoxic stresses of H2O2 and UV irradiation [33,38].
Since TREX1 was shown to respond to such genotoxic stress by translocating to the nucleus [24], it would be of interest to characterize whether TREX1 can digest DNA substrates damaged by H2O2 and UV irradiation as TTHB178 and RNase T do. Therefore, we incubated TREX1 with ssDNA substrates that each contain a methylated base, a deaminated base, an oxidized base, or an abasic site at the 3′-terminal end (5′-GAGTCCTATAX-3′) and measured the activities of TREX1 in ssDNA degradation. The results presented in S2A Fig clearly showed that TREX1 exhibited similar activities in digesting ssDNAs containing a methylated base (O4-methylthymine [O4-mT] and O6-methylguanine [O6-mG]) or a deaminated base (uracil and hypoxanthine) as in digesting ssDNAs with regular bases (adenine). The ssDNA with an abasic site or an oxidized base (8-oxoguanine [8-oxoG]), though, was more resistant to TREX1. Therefore, the base preference of TREX1 to ssDNA substrates is similar to that of TTHB178 and RNase T in the DEDDh family.
RNase T showed in in vitro experiments that it can serve as a downstream exonuclease of Endonuclease V (Endo V)–mediated alternative base excision repair (AER) [33]. Endo V–mediated AER repairs DNA lesions originated from deamination of the adenine base (hypoxanthine), frame shift mutations, or replication errors [39–41]. The initiation step is making a nick at the 3′ side 1 base away from the damaged site. Expression of eukaryotic Endo V in DNA repair–deficient E. coli cells reduced the mutation frequency in the host, suggesting that the DNA repair function of Endo V may commonly display in E. coli and mammalian cells [40,41]. Therefore, TREX1 may also serve as a downstream exonuclease of Endo V, as RNase T does. In addition to DNA substrates, recent studies showed that both Endo V and TREX1 are also ribonucleases, as they exhibited considerable activities in digesting RNA substrates [28,32]. Despite the common ground at the level of enzyme activity, the involvement of TREX1 in this DNA repair pathway and its specific roles would require further investigation to establish.
A prominent question, therefore, is whether TREX1 can process the structural DNAs generated by Endo V. In this regard, we designed and synthesized a bubbled dsDNA containing a dI with or without the 5′-overhang to mimic the end product of the enzyme (Fig 2A and 2B). This double-stranded substrate is referred to as dI-dsDNA. The Tris-borate-EDTA (TBE) gel electrophoresis was used to check the structure property of dI-dsDNA. With identical numbers of bases, the moving speed of dI-dsDNA in the gel was in between that of a Y-structural dsDNA and that of a regular dsDNA, indicating an intermediate, bubbled structure (S2B Fig). The results of our nuclease activity assay showed that TREX1 indeed broke the terminal base pairing and removed the last base pair near the penultimate hypoxanthine base at the 3′-end of dI-dsDNA (the 2 site labeled in Fig 2B). In addition, TREX1 also removed the penultimate hypoxanthine base and antepenultimate adenine base (the 1 and 0 sites labeled in Fig 2B). For the dI-dsDNA substrate with 5′-overhang, similar results were observed, indicating that the dangling 5′-ends did not affect TREX1 in processing the dsDNA.
In summary, our activity measurements established that TREX1 not only digested the dI in ssDNA; the enzyme can also extend into the duplex region and removed the last 3 bases at the 3′-end of a bubbled dI-dsDNA that mimics the product of Endo V. These results also highlighted that TREX1 is able to unwind the terminal base pairs in a DNA duplex to conduct digestion.
To provide the structural basis for TREX1 in complexing an ssDNA substrate containing dI, we determined the TREX1-dI-ssDNA structure. Fig 2C showed that the 3′-end of dI-ssDNA is inserted into the active site of TREX1 in each of the dimer in the asymmetric unit. The last nucleotide at the 3′-end is stacked by Leu24 and Ile84 (Fig 2C). Each active site contained 2 Mg2+, MgA and MgB. MgA coordinates with Asp18, Glu20, and Asp200 of TREX1 and was in contact with the phosphate oxygens of the scissile DNA strand. The His195 general base in the active site lacked the nucleophilic water that would bind MgA, a commonly observed configuration among the structures of DEDDh members [10,33,37]. MgB in the structure coordinated with 6 partners in the octahedral geometry, including Asp18, 2 phosphate oxygens in the scissile DNA stand, and 3 water molecules (Fig 2C). Both active sites of TREX1 in the asymmetric unit showed an active conformation as in the structures of other classical DEDDh exonucleases (S3A Fig).
In the TREX1-dI-ssDNA structure, the hypoxanthine base was designed to locate in the penultimate position at the 3′-end of dI-ssDNA as the products generated by Endo V (Fig 2A and 2C). The omitted electron density map of the hypoxanthine base in ssDNA fits very well to the three-dimensional structure (Fig 2C). Superposition of the structure of the TREX1-dI-ssDNA structure with that of TREX1 bound with a regular ssDNA showed that TREX1 bound both substrates with high structure similarity (S3B Fig). Therefore, the chemical modification into a hypoxanthine base at the penultimate position of the 3′-end did not affect the binding mode of TREX1 with ssDNA and its active site structure.
In the TREX1-dI-T-dsDNA structure we solved, both protomers bound with a dI-dsDNA, as shown in Fig 3A. Both active sites of the enzyme dimer displayed the same active conformation as in the TREX1-dI-ssDNA structure. The omitted electron density maps in the double-strand region of both dI-dsDNA strands, though, were well defined for structure determination. In contrast, the electron densities of the 2 short 5′-overhang were not as continuous and weak, suggesting that the 3-nt-long 5′-overhangs were disordered (S4A Fig).
The structure of the TREX1-dI-T-dsDNA complex vividly illustrated the mode by which TREX1 broke the last base pairing in the duplex region. In particular, the Leu24-Pro25-Ser26 cluster acted as a wedge to chisel the duplex end formed by the last G-C pair and the dI-T wobble pair, and the DNA adopted a Y-like structure (Fig 3A and 3B). The complex of TREX1-dI-T-dsDNA hence provided a first structural basis for the ability of TREX1 in unwinding the duplex region of dsDNA. In addition to the Leu24-Pro25-Ser26 cluster, the structure indicated that the narrow pocket of the TREX1 active site was also involved in breaking the double-stranded structure as Leu24 and Ile84 stacked with the last nucleotide at the 3′-end (Fig 3A). Furthermore, the last nucleotide at the 3′-end of the substrate formed several hydrogen bonds with Glu20, Ala21, and Tyr129 in the narrow pocket (S4B Fig). These couplings to the 3′-end would thus facilitate unwinding of the terminal base pair. In summary, TREX1 exhibited the delicate machinery that the Leu24-Pro25-Ser26 cluster partnering with the narrow active site pocket to unwind the duplex end for digestion of dsDNA.
The mechanistic insights provided by the TREX1-dI-T-dsDNA complex structure were in line with the biochemical assay measurements presented earlier that TREX1 removed the 2 to −1 bases at the 3′-end of the bubbled dI-dsDNA (Fig 2B). Our results thus illustrated that the unique catalytic properties of TREX1 in digesting dsDNA and in processing the hypoxanthine base can serve to digest the product generated by Endo V. This finding represented an evidence that a DEDDh member like TREX1 and RNase T may act as a downstream exonuclease for Endo V–mediated AER in DNA repair.
Our activity measurements showed that TREX1 can remove the single-stranded regions in structural DNAs, and blunt-end duplexes are a main form of product of TREX1 (Figs 4A and 5A). To uncover the molecular origin of such precise removal of the 3′-overhang, we determined the structures of TREX1 complexed with the substrates that we devised to mimic the intermediates in UV-induced DNA repair. The structures of TREX1-L-structural dsDNA and TREX1-Y-structural dsDNA were solved at 1.7 Å and 2.0 Å resolution, respectively. The protein residues interacting with the DNA substrates in these two structures are shown in S5 Fig. The active sites in both structures also adopted the active form of conformation. Both structures contain regions of short 3′-overhang (1-nt- or 2-nt-long) to illustrate how the 3′-overhang was removed.
In the TREX1-L-structural dsDNA structure, 1 asymmetric unit contained 2 TREX1 monomers and 2 ssDNAs of 6 and 9 nt in length (Fig 4B). The 2 strands paired to their symmetric ssDNA in the crystal via GC-rich regions. The annealed DNA strands formed 2 duplexes of 1-nt- and 4-nt-long 3′-overhang, respectively. Each 3′-terminus of the DNAs displayed an L-like conformation, and for both duplexes, the 3′-overhang was inserted into the narrow active site of a TREX1 monomer. We designate molecule A as the TREX1 monomer that bound with the duplex DNA with a 1-nt-long 3′-overhang, 1-nt-L-DNA and molecule B as the TREX1 monomer that bound with the DNA duplex with a 4-nt-long 3′-overhang, 4-nt-L-DNA (Fig 4B).
Superposition of 1-nt-L-DNA and 4-nt-L-DNA revealed 2 distinct modes of binding duplex DNA for molecule A and molecule B (Fig 4D and S6A Fig). It can be seen clearly that the duplex segments of 1-nt-L-DNA and 4-nt-L-DNA sit at 2 different loci with respect to the protein. The Leu24-Pro25-Ser26 cluster of molecule A capped the 5′-end of 1-nt-L-DNA via contact with the last base of the nonscissile strand (Fig 4C). Leu24 and Ile84 in the narrow pocket of the active site in molecule A also stacked with the last nucleotide at the 3′-end of the scissile strand as in the TREX1-dI-T-dsDNA structure, indicating joint actions of the two structure motifs for removal of the 3′-overhang. Molecule A also formed 4 hydrogen bonds with the nonscissile strand of 1-nt-L-DNA and via Ser26, Arg128, and Lys160. In molecule B, however, it was a groove composed of Ala161, Leu162, Ala214, Gln217, and Trp218 on the other side of the Leu24-Pro25-Ser26 cluster that contacted with the 5′-end of 4-nt-L-DNA (S6B Fig). Moreover, molecule B did not form any hydrogen bonds or stacking interactions with the nonscissile strand of 4-nt-L-DNA.
Therefore, the TREX1-L-structural dsDNA structure showed that, in trimming a duplex DNA to generate duplexes with blunt ends, TREX1 can cap the nonscissile 5′-end via the Leu24-Pro25-Ser26 cluster and form hydrogen bonds and stacking interactions with the nonscissile strand. Such binding mode was neither observed in the case of molecule B bound with 4-nt-L-DNA (Fig 4D and S6B Fig) nor in earlier structures of TREX1 with substrates that contain a longer 3′-overhang [10].
The binding mode of Leu24-Pro25-Ser26 capping the nonscissile 5′-end was also observed in the TREX1-Y-structural dsDNA structure. In this case, the crystal also contained 2 TREX1 molecules in the asymmetry unit, and both bound with a Y-structural DNA of 2-nt-long 3′- and 5′-overhang. The sequence and structure of the substrate are shown in Fig 5B. The scissile strand of the Y-structural DNA was trapped by TREX1 via more than 10 amino acids, and the last nucleotide of the 3′-end was inserted into the active site and stacked by Leu24 and Ile84 in the narrow active site pocket, as observed in the structures of TREX1-dI-T-dsDNA and TREX1-L-structural dsDNA. The last base (G1) at the 5′-overhang of the substrate sits in the gap between TREX1 and the last base at the duplex region of Y-structural DNA and was flanked by the Leu24-Pro25-Ser26 cluster of TREX1 and the T3 base on the nonscissile strand (Fig 5B and 5C). With the nonpairing T2 base of the nonscissile strand flipped out, this binding mode is similar to that of molecule A bound with 1-nt-L-DNA in the TREX1-L-structural dsDNA structure (Fig 5C).
Therefore, both structures that contain duplex DNAs with a short 3′-overhang revealed the binding mode of Leu24-Pro25-Ser26 cluster in TREX1 capping the nonscissile 5′-end. Signature of this mode also highlighted the joint actions of the Leu24-Pro25-Ser26 cluster and the active site pocket in coupling to the 3′-end of the scissile strand, and specific interactions with the nonscissile strand were observed as well. These results provide the structural basis for duplex DNAs without a 3′-overhang being a major product form of TREX1 in activity assays (Figs 4A and 5A).
The Leu24-Pro25-Ser26 cluster that wedged the nonscissile 5′-end in the TREX1-dI-T-dsDNA complex structure located at the N-terminal of a short helix of Pro25 to Ser27, and the helix was linked to the β strands of β1 and β2 by 2 loops. Although β1 and β2 are highly conserved, bioinformatics analysis indicated that the sequence and structure of the region in between exhibits significant variation in the DEDDh exonuclease family (Fig 6A and S7A Fig). Only TREX1 and TREX2 in the family, the corresponding region of Pro25 to Ser27, adopted a wedge-liked structure involving a small helix, whereas the corresponding region in other family members adopted a loop form of structure (Fig 6A). In targeting a single-stranded substrate, the X-ray structures of TREX1 resolved in this work (the TREX1-dI-ssDNA structure, Fig 2C) and those of RNase T [35] indicated that this region stacked the substrate with residues in the active site pocket in either a helical or loop form of secondary structure, and the nuclease activities of TREX1 and RNase T toward ssDNAs were similar.
In TREX1, the helical form of the Leu24-Pro25-Ser26 cluster in between β1 and β2 shaped a wedge-like structure, which can cap the 5′-end for producing blunt-end duplexes (the TREX1-L-structural dsDNA structure, Fig 4B and the TREX1-Y-structural dsDNA structure, Fig 5B) and even overcome the hindrance of duplex regions for processing dsDNAs (the TREX1-dI-T-dsDNA structure, Fig 3). Such unique catalytic powers compared to other members in the DEDDh family echoed the distinct helical structure form of the Leu24-Pro25-Ser26 cluster in TREX1 and the concomitant higher mechanical strength. With a loop form of structure in the corresponding region, the nuclease assay of E. coli RNase T did not reveal any activity toward dsDNA, even at an enzyme concentration as high as 10 μM under the same reaction conditions (Fig 6B and S7B Fig). The capability of digesting duplex structures was only observed for the closely related homolog TREX2 that also contained a wedge helix (Fig 6B and S7B Fig).
Indeed, over the incubation time of our nuclease activity assays, TREX1 was shown to fully consume ssDNA as well as dsDNA, albeit the enzyme concentration for the double-stranded substrate was higher (Fig 6C). To further examine the roles of Leu24-Pro25-Ser26 cluster in the activity of TREX1, we generated 4 single-site mutants (L24G, L24A, L24W and S26W) and 2 triple mutants (L24G/P25G/S26G and L24W/P25W/S26W) of the enzyme and measured the activities of these TREX1 mutants against ssDNA and dsDNA substrates. In comparison to the wild type, all mutants showed reduced activities toward both dsDNA and ssDNA substrates (Fig 6D, 6E and 6F and S7C, S7D and S7E Fig). Therefore, it can be inferred that perturbations to the Leu24-Pro25-Ser26 cluster due to these mutations affected not only the ability of stacking the last base at the 3′-end (with Ile84) to properly position the substrate in the active site for catalysis (Figs 2C, 3A, 4C and 5B); they also eliminated the power of breaking the terminal base pairing. Our mutagenesis and activity characterizations highlighted the importance of the Leu24-Pro25-Ser26 cluster in both terminal unwinding of dsDNA and 3′-ended nucleotide stacking for TREX1.
As the enzyme concentration was further raised to 500 nM, the measured activity for digesting the duplex regions was also augmented to accomplish full digestion of dsDNA as long as 708 bp (Fig 6C). At an enzyme concentration of 50 nM, on the other hand, TREX1 processed the ssDNA region of 3′-overhang in double-stranded substrates, leaving a blunt-end dsDNA (Figs 4A and 5A); the enzyme also trimmed the weaker pairing of a damaged base at a duplex terminal, as in the case of a nicked DNA generated by Endo V (Fig 2B). These results showed that the dsDNA degradation activity is concentration dependent, and under the in vivo scenario, other factors could also come to modulate the activity. For example, it is highly likely that the activity of TREX1 is coupled with partner proteins such as high mobility group box 2 (HMGB-2). HMGB-2 is also a component of ER-associated SET complex and is related to cytosolic nucleic acid–mediated innate immune responses [42–44]. The activity of HMGB-2 in bending the dsDNA structure potentially would generate distorted structures to facilitate dsDNA degradation by TREX1. Our preliminary results of activity measurements showed raised ability of TREX1 in digesting PCR products in the presence of mouse HMGB-2 (mHMGB-2) (S9 Fig). With mHMGB-2, TREX1 can fully digest PCR product at a lower enzyme concentration of 0.2 μM, evidencing the interplay between mHMGB-2 and TREX1 in modulating DNase activities.
To date, the reported structures of duplex-bound TREX1 contained a substrate with a long 3′-overhang (≥4-nt-long) [10]. As a result, consensus features in such structures of binding were only observed for the last 2 nucleotides at the end of the 3′-overhang, while the farther-away regions of the scissile and nonscissile strands exhibited different directions and orientations (Fig 7A and S8A Fig). Therefore, using dsDNA substrates with a long 3′-overhang experienced significant difficulties in revealing the molecular mechanism by which TREX1 conducts overhang trimming and terminal unwinding.
On the contrary, a consistent mode of binding was observed in the structures resolved in this work that TREX1 complexed a dsDNA with a short 3′-overhang. The 3′-end of the scissile strand and the 5′-end of the nonscissile strand (Fig 7B) exhibited reproducible patterns in the TREX1-L-structural dsDNA, TREX1-Y-structural dsDNA, and TREX1-dI-T-dsDNA structures. For example, stacking between the nonscissile strand and the Leu24-Pro25-Ser26 cluster as well as the hydrogen bonding between Arg128 and the nonscissile strand (S4 and S5 Figs) are both prominent features that were commonly observed in these structures.
Based on the 4 structures newly resolved in this work and literature data, we summarize the binding modes of TREX1 with various structural DNAs in Fig 7C and 7D. The first mode is for ssDNAs longer than 4 nt and, in this case, TREX1 bound to the last 4 nucleotides at the 3′-end with the 5′-end of the strand dangling. For damaged ssDNAs with hypoxanthine bases, we showed that TREX1 digested such substrates with a similar activity as degrading DNAs with normal nitrogenous bases. The ability of TREX1 in processing dI-ssDNA also pointed to a possible role in Endo V–mediated AER. TREX1, though, lacked the activity of cutting abasic site and 8-oxoG sites as the extra oxygen could hinder the enzyme to stack bases in the substrate. The resistance of 8-oxoG to TREX1 appears as an intrinsic design for transferring the signal of oxidative DNA damage by triggering stimulator of interferon genes complex (STING)–dependent immune sensing in cytoplasm [45]. Two members in DEDDh exonuclease family, T. thermophiles TTHB178 and E. coli RNase T, are homologous to TREX1 and demonstrated similar base preference, suggesting that these proteins perform similar functions in oxidative DNA damage signaling.
The second mode is for the class of duplex L- and Y-structural DNA substrates with a long 3′-overhang (≥4 nt), and most of the hydrogen bonds formed between TREX1 and the substrate were in the last 3 nucleotides at the 3′-end, i.e., in the 3′-overhang region. The nonscissile strand, on the other hand, did not land on consensus positions on the protein surface (Fig 7A and S8 Fig). Therefore, in binding DNA duplexes with a longer 3′-overhang, TREX1 more tightly coupled to the last 3 bases of the 3′-overhang region, and the rest of the substrate exhibited different conformations (Fig 7D).
For structural DNAs containing a short (<4 nt) 3′-overhang, the structures resolved in this work revealed the third class of TREX1-DNA binding. The Leu24-Pro25-Ser26 cluster in TREX1 was identified as a wedge to stack and interact with the 5′-end of the nonscissile strand, particularly at the last base in the double-stranded segment. Such specific couplings of capping the nonscissile 5′-end were not observed for TREX1 structures in binding with substrates that contained a longer 3′-overhang and provided a clear structural basis for the generation of blunt-end duplexes as the main product form after removal of the 3′-overhang (Fig 7D).
Our structure analysis also showed that the Leu24-Pro25-Ser26 cluster plays a critical role for TREX1 to break the terminal base pairing in dsDNAs. In the activity assays of TREX1 with duplex DNAs, minor products that lacked the last 1 or 2 nucleotides at 3′-end in the double-strand region were also observed (Figs 2B, 4A and 5A). For the structure with a wobble base pair shown in Fig 3, the last base pair was indeed separated for digestion. Together, these results indicate that the Leu24-Pro25-Ser26 cluster and other specific interactions enable TREX1 to break the base pairing around the nonscissile 5′-end and hence allow further nucleotide removal in the scissile strand. The terminal unwinding activity, in fact, allowed TREX1 to process the wobble-paired dsDNA generated by Endo V and to conduct dsDNA degradation as shown in Figs 2B and 6C.
Another key residue that was identified through the TREX1 structures resolved in this work is Arg128, which makes several contacts with the nonscissile strand in a duplex. As stated earlier, for all of the dsDNA substrates that we designed to have a short (<4 nt) 3′-overhang, the specific interactions of Arg128 are with the nonscissile strand. On the contrary, in the structures of TREX1 with a dsDNA containing a long 3′-overhang, Arg128 instead made contact with the single-stranded region of the scissile strand, and the mode of interactions is similar to that seen in the TREX1 structures with an ssDNA [10,31,46]. Mutation of the satellite residue Arg128 in TREX1 has been shown to result in an approximately 8-fold reduction in the activity of digesting dsDNA, but only about 2-fold of activity reduction was observed for ssDNA degradation [47]. This result highlighted the greater importance of Arg128 in dsDNA degradation. Our structures thus provide the necessary molecular details for understanding this result as the binding mode of TREX1 with a DNA duplex that contains a short 3′-overhang is different from that of binding an ssDNA. Since dsDNAs are primary substrates for immune silencing, our results also provide structure basis for the in vivo observation that Arg128 mutation is related to an autoimmune disease of SLE [20].
A mystery of TREX1 in DNA repair is that the rate of spontaneous mutation in TREX1-/- mice did not increase [18], despite the fact that genotoxic stress did lead to translocation of TREX1 to the nucleus and elevated expression levels of the enzyme [3,24]. A hypothesis that offers a potential explanation is participation of TREX1 in DNA repair processes like Endo V–mediated AER, and in this case, other pathways such as base excision repair (BER) for repairing DNAs with a hypoxanthine base would functionally overlap with those of TREX1 [48–50]. Although this scenario is subject to further studies to firmly establish, it is in line with the substrate compatibility of TREX1 with Endo V products, as observed in our activity characterization. As such, a single-gene knockout in either of the two overlapping pathways might not result in a significant increase in the rate of spontaneous mutation. An analogue example was that double mutations of DNA glycosylase and Endo V (nfi) were shown as necessary for an increased rate of spontaneous mutation to be observed, and transforming only intact gene 1 of 2 into the doubly mutated E. coli was sufficient to rescue the phenotype of a higher rate of spontaneous mutation [40,41].
Entanglement of multiple pathways would thus complicate the characterization of physiological impact for TREX1, and it is thus essential to consider that this enzyme is likely coupled to other partners in DNA repair–related processes. For example, TREX1 is also expected to respond to UV light–induced DNA stress since translocation of the enzyme was observed upon exposure to UV illumination [24]. The consequent repairing processes include removal of L- or Y-structural dsDNA in the replication restart pathway, gap-filling homologous recombination, and RecA-dependent homologous recombination that the TREX1 homolog E. coli RNase T is also known to perform [33]. In this regard, the activity of TREX1 in trimming Y-structural dsDNA probably overlaps with that of the endonuclease Mus81-Mms4 (Slx2-Slx3), which also targets similar structures [51,52]. Along a similar line, the proofreading function of TREX1 in BER would overlap with that of Ape1, which bears a 3′–5′ exonuclease activity and interacts with DNA polymerase β in editing mismatched nucleotides [53–55]. Furthermore, TREX1 may also involve in other DNA repair pathways, such as by interacting with and regulating PARP-1, a DNA repair enzyme for which the molecular details await further studies to resolve [25]. The overlapping model thus provides a rationale for the unaffected rates of spontaneous mutation in TREX1 knockout mice, and for the proficiency of double-strand break (DSB) repair and BER in TREX1-deficient fibroblasts [5,18].
It is important to highlight that the DNA repair functions require TREX1 to be present in the nucleus. Translocation of TREX1 to the nucleus was observed during GzmA-mediated apoptosis or under genotoxic stresses in several works via immunoprecipitation and/or immunofluorescence microscopy [3,23,24]. TREX1 was also found in the nucleus and was shown to interact with a well-known DNA repair enzyme, PARP-1, via western blotting and coimmunoprecipitation [25]. However, TREX1 translocation to the nucleus was not observed in the immunofluorescence microscopy data of a recent work [5], suggesting the need of further studies to characterize the DNA repair function of TREX1.
TREX1 provides the major 3′–5′ exonuclease activity in mammalian cells [1,2] as numerous in vivo studies showed that TREX1 broadly displayed important roles in immune silencing [3,21], genotoxicity responses [3,24], apoptotic DNA degradation in dying cells [26], and chromosomal fragmentation during telomere crisis [27]. In this work, we aim to provide the structure basis for the molecular origin of TREX1 that enables such diverse activities toward a wide range of nucleic acid substrates, including ssDNA, dsDNA, DNA/RNA duplexes, and structural DNAs. The endeavor of crystallizing TREX1 with various duplex DNA substrates with a short 3′-overhang revealed the structure details for the unique catalytic powers of TREX1. We identified that the Leu24-Pro25-Ser26 cluster at the N-terminal of a short helix can cap the nonscissile 5′-end for trimming the 3′-overhang for producing a blunt-end duplex and can wedge into the duplex end to unwind the terminal base pairing in a dsDNA. These results also provide sophisticated molecular pictures for rationalizing the cellular functions of TREX1 in DNA repair, DNA proofreading, and immune silencing.
The trex1 gene from Mus musculus, with a length of 1–242 amino acids, was subcloned into BamHI/XhoI site in plasmid pET28a. The trex2 gene from M. musculus was subcloned into NdeI/XhoI site in plasmid pET28a. The expression vector was transformed into E. coli BL21-CodonPlus(DE3)-RIPL or Rosetta2(DE3)pLysS strain (Stratagene, United States) cultured in LB medium supplemented with 50 μg/mL Kanamycin. Cells were grown to an OD600 of 0.5–0.6 and induced by 1 mM IPTG at 18 °C for 20 h. The harvested cells were disrupted by sonication in 50 mM Tris-HCl (pH 7.5) containing 300 mM NaCl for 20 min. TREX1 and TREX2 were further purified by Ni-NTA resin affinity column (QIAGEN), HiTrap Heparin (GE Healthcare), and a Superdex 200 gel filtration column (GE Healthcare). The purified proteins in 50 mM Tris-HCl, pH 7.0 and 300 mM NaCl were concentrated to suitable concentrations and stored at −20 °C until use. All of the TREX1 mutants were generated by QuickChange site-directed mutagenesis kits (Stratagene) and purified by the same procedures as for wild-type TREX1. His-tagged TREX1 were treated with thrombin to generate non-His-tagged TREX1 for crystallization experiments. RNase T was purified as previous described [33].
The sequences of DNA substrates are listed in S2 Table. DNA substrates were synthesized by BEX Co., Tokyo, Japan, or MDBio, Inc., Taiwan. Substrates were labeled at the 5′-end with γ-32P or FAM. The γ-32P was labeled at the 5′-end of substrate DNA by T4 polynucleotide kinase (New England Biolabs) and then purified by a Microspin G-25 column (GE Healthcare) to remove the nonincorporated nucleotides. Labeled substrates (0.5 μM) were then incubated with the exonuclease mixture in 120 mM NaCl and 20 mM Tris-HCl at pH 7.0 and 37 °C. The reaction was stopped by adding DNA-loading dye at 95 °C for 5 min. The digestion patterns were resolved on 20% denaturing polyacrylamide gels and visualized by autoradiography (Fujifilm, FLA-5000) or UV light. When the substrate was a PCR product, we added protease K into the reaction mixture at 37 °C for 1 h to stop the reaction. The DNA digestion patterns were resolved on 1% agarose gel and visualized by UV light.
His-tagged or non-His-tagged TREX1 (15–25 mg/mL) in 300 mM NaCl and 50 mM Tris-HCl, pH 7.0 were mixed with different DNA substrates in the molar ratio of 1:1.5. Detailed information regarding DNA sequences and crystallization conditions of the 4 structures are given in S1 Table. All crystals were cryoprotected by Paraton-N (Hampton Research, USA) for data collection at BL13C1, BL13B1, and BL15A1 in NSRRC, Taiwan, or at the BL44XU beamline at SPring-8, Japan. All diffraction data were processed by HKL2000, and diffraction statistics are listed in Fig 1. Structures were solved by the molecular replacement method and using the crystal structure of M. musculus TREX1 (PDB: 3MXM) as the search model by MOLREP of CCP4 [56]. The models were built by Coot [57] and refined by Phenix [58]. Diffraction structure factors and structural coordinates have been deposited in the RCSB PDB with the PDB ID code of 5YWV for the TREX1-dI ssDNA complex, 5YWU for the TREX1-dI-T-dsDNA complex, 5YWT for the TREX1-L-structural dsDNA complex, and 5YWS for the TREX1-Y-structural dsDNA complex.
|
10.1371/journal.pntd.0007312 | Implications of environmental and pathogen-specific determinants on clinical presentations and disease outcome in melioidosis patients | Melioidosis is gaining recognition as an emerging infectious disease with diverse clinical manifestations and high-case fatality rates worldwide. However, the molecular epidemiology of the disease outside the endemic regions such as northeast part of Thailand and northern Australia remains unclear.
Clinical data and B. pseudomallei isolates obtained from 199 culture-confirmed cases of melioidosis diagnosed during 2006–2016 in South India were used to elucidate the host and pathogen specific variable virulence determinants associated with clinical presentations and disease outcome. Further, we determined the temporal variations and the influence of ecological factors on B.pseudomallei Lipopolysaccharide (LPS) genotypes causing infections. Severe forms of the disease were observed amongst 169 (85%) patients. Renal dysfunction and infection due to B.pseudomallei harboring BimABm variant had significant associations with severe forms of the disease. Diabetes mellitus, septicemic melioidosis and infection due to LPSB genotype were independent risk factors for mortality. LPSB (74%) and LPSA (20.6%) were the prevalent genotypes causing infections. Both genotypes demonstrated temporal variations and had significant correlations with rainfall and humidity.
Our study findings suggest that the pathogen specific virulence traits under the influence of ecological factors are the key drivers for geographical variations in the molecular epidemiology of melioidosis.
| Amidst the ambiguity of its true incidence, melioidosis is gaining importance as an emerging infection in the Indian subcontinent. Variable virulence genes contributing to varied clinical presentation and the association of LPS genotypes provides key insights regarding the disease dynamics, host and pathogen specific determinants for disease presentations and outcomes among melioidosis patients in India. The study shows the divergence of Indian B.pseudomallei strains in context to LPS genotypes compared to isolates reported worldwide. The association of LPS B with mortality foreshadows the need for understanding the cascade of immune response triggered in the presence of different LPS genotypes.
| Melioidosis is a fatal infectious disease caused by soil saprophytic bacterium, Burkholderia pseudomallei. Infection occurs mostly through the inhalation or percutaneous inoculation of the bacteria from contaminated soil or surface water. The disease manifests with diverse clinical presentations ranging from mild localized infection to fulminant sepsis. B. pseudomallei being a soil saprophyte, uses horizontal gene transfers as a mechanism for its persistence both in the environment and the host. It is possible that the virulence attributes of B.pseudomallei can significantly vary under the influence of regional environmental/ecological conditions, in turn leading to the occurrence of distinct clinical manifestations. Lipopolysaccharide (LPS) of B. pseudomallei is a well-known virulence factor that confers serum resistance and helps in evading host immune defenses during the early stages of infection. In this context, LPS is gaining recognition as a potential candidate for vaccine and diagnostic assay development [1,2]. Three genotypes of LPS namely A, B and B2 were reported previously with variations in their geographic distribution and ability to induce immune responses in animal models [3]. Burkholderia intracellular motility (BimA) and filamentous hemagglutinin gene (fhaB3) were reported previously as the significant variable virulence factors based on their geographic distribution and associations with clinical presentations [4].
Amidst the ambiguity of its true incidence, melioidosis is gaining importance as an emerging infection in the Indian subcontinent [5–7]. Sporadic cases reported from different parts of the country have shown assorted clinical presentations [7–10]. Distinct/novel sequence types of B. pseudomallei using multi-locus sequence typing(MLST) were previously reported from the south-western coastal part of India [11].This geographic region might presumably be one of the endemic hotspots of melioidosis in India reporting several cases [6,7,10]. Clinical isolates of B. pseudomallei from this region were genetically diverse from the Australian and Southeast Asian isolates and the prevalent sequence type (ST 1368) lacked significant association with any particular clinical presentation of the disease [11]. With this background, the present study documents the frequencies of variable virulence factors of B. pseudomallei and their association with clinical presentations of melioidosis and the temporal variations and influence of ecological factors on commonly occurring LPS genotypes of B. pseudomallei. Moreover, the host and pathogen-specific determinants for mortality are elucidated.
The present study was carried out at a tertiary care hospital with 2030 inpatient capacity that caters residents of southwestern coastal parts of Karnataka, India covering nearly 150–200 km radius of geographical area. This part of the country experiences tropical climatic condition with an annual rainfall of >4000 mm during June-October. Microbiological culture confirmed cases of melioidosis over a decade (2006–2016) were included in the study.
The study was approved by the Institutional Ethical Committee of Kasturba Hospital, Manipal. All the isolates were obtained from the archived collection and the identity of the patients was kept confidential.
Isolates from blood culture in 97 bacteremic cases with (n = 50) or without (n = 47) other organ involvement and from other sites in 102 non-bacteremic cases were included in the study. For the extraction of bacterial DNA, QIAamp DNA mini kit (Qiagen, Hilden, Germany) was used as per the manufacturer’s instructions. Before inclusion, all the study isolates were confirmed as B. pseudomallei using a species-specific PCR targeting the TTSS1 gene cluster as described previously [12].
Collectively, we aimed at detection of LPS genotypes (A, B and B2), BimA gene variants (BimABp and BimABm) and fhaB3.
Multiplex PCR assay simultaneously detected three different LPS genotypes LPS A, B, and B2 [3]. The PCR reaction was set at a final volume of 25 μl using JumpStartTaq Ready Mix (Sigma-Aldrich).Amplification was carried out in a Master cycler gradient (Eppendorf, Hamburg, Germany) with an initial denaturation at 95°C for 10 min, followed by 35 cycles of 95°C for 30sec, 59°C for 30 sec, 72°C for 30secand a final extension step of 72°C for 7 min. The oligonucleotide primers used in the present study and the expected amplicon sizes are tabulated in Table 1.
Both variants of Bim A, BimABm and BimABp, were detected using previously reported PCR primers [4]. PCR reaction volumes and cycling conditions for the detection of BimA genes were similar to that of the LPS genotypes, except for a change in the annealing temperature to 56°C for 1 min. Presence of BimABm and BimABp were considered when amplicons sized 104 bp and 60 bp respectively were positive on 2.5% Agarose gel stained with 0.5% ethidium bromide.
fhaB3 gene was detected using 0.3 uM of each forward and reverse primers to generate a 58 bp product [4]. Amplification was carried using similar cycling conditions as mentioned above for LPS genotypes detection. However, PCR for the detection of fhaB3 was performed separately considering the similar size of the amplicons for both BimABp and fhaB3.
Clinical and epidemiological data were documented in structured study forms. Monthly rainfall and relative humidity data for a period of six years (2010–2016) were obtained from the Indian Meteorological Department, Pune, India.
Microbiological culture of blood and/ or other clinically relevant specimens was the mainstay for laboratory diagnosis of melioidosis. For analysis and reporting purposes the following case definitions were used in the present study:
Descriptive statistical tools were used to determine the frequencies of categorical study variables. Pearson’s Chi-square test and Fisher’s exact test was used to check for the presence of any significant association of host and pathogen characteristics with individual clinical presentations. Risk factors for various clinical presentations and outcomes (in hospital mortality & Discharge Against Medical Advice) among the study population were determined using univariate analysis and step-wise multivariate logistic regression model (Backward LR) (SPSS, version 16). Time series analysis was used to decompose the trend, seasonal and residual components for the climate data and the LPS genotypes. Correlations of the climate data with the LPS genotypes was obtained using Pearson’s correlation coefficient. Generalized additive model was used to predict the LPS genotypes using the climate data. Analysis was carried out using R version 3.3.3. All values were considered significant with p≤0.05.
Mean age of our patients was 47.7±15.4 years, with an age range between 7–86 years. Of the 199 patients, 187 (94%) were from Karnataka and 12 patients (8 from Kerala and 4 from Goa) were from other states on the southwestern coastal part of India (Fig 1). Majority of the patients were males (153, 76.9%) and had the disease episode during monsoon (148, 74.3%).
A large number of patients (169, 85%) had either one or more of the severe forms of the disease, like bacteremic (97, 48.7%), pulmonary (71, 35.6%), septicemic (50, 25.1%) neurological (22, 11%) and others like osteoarticular and deep seated abscess (36, 18%). Amongst bacteremic patients, 47 (48.4%) had no other focus of infection as diagnosed clinically or radiologically. In septicemic melioidosis, 38 (76%) patients had bacteremia with or without evident foci and the rest 12 patients had osteoarticular or pulmonary forms of the disease. Localized form of the disease was observed amongst 30 (15%) patients.
Diabetes mellitus (DM) (124, 62.3%) and renal dysfunction (27, 13.5%) were the common co-morbid illnesses. Three patients had malignancy, one had thalassemia, and none had HIV infection. Using univariate analysis, we observed that patients with renal dysfunction had 8.75 (Crude OR:8.75; 95% CI: 3.60–21.25; p<0.001) and 3.52 (Crude OR: 3.52; 95% CI: 1.41–8.77; p = 0.005) times more odds for developing septicemic and bacteremic forms of the disease respectively(Table 2).
Majority of our study isolates belonged to LPS B (n = 147, 73.8%) followed by A (n = 41, 20.6%) and B2 (n = 11, 5.5%) genotypes. Amongst the variants of BimA gene, BimABp and BimABm were observed among 190 (95.4%) and 9 (4.5%) of the isolates respectively. Majority of the isolates were positive for fhaB3(190; 95.4%) and 181 (90.4%) isolates harbored both BimABp and fhaB3genes. None of the three LPS genotypes had a significant association with clinical forms of the disease in our study population, as it was observed withBimA gene variants with neurological form of the disease(Crude OR: 12.72; 95% CI: 3.11–51.89; p>0.001)(Table 3).
Majority (73.8%) of the B. pseudomallei clinical isolates including all the 12 isolates from patients in the adjacent states belonged to the LPSB genotype (Fig 1). The prevalence of LPS A and LPS B genotypes were consistent throughout the study period (2006–2016), and LPS B2 genotype was observed only during the last three years (2014–2016). We noticed a steady decline of LPSB during the years 2014–2016 (79% in 2014, 56.6% in 2015 and 50% in 2016) in our settings. At the same time, there was a steady increase ofLPSA (16.6%, 26.6% and 33.3%) and B2 (4.1%, 16.6% and 16.6%) genotypes during the same duration (Fig 2).
Monthly rainfall and humidity data for the years 2010–2016 was plotted against time and decomposed data for trend, seasonality and residual component were compared with LPSA and B genotypes. LPSA genotype showed a reversal peak for trend and seasonal components of rainfall, whereas rainfall had a significant increasing effect on LPSB genotype. There was a rise in LPSA genotypes observed over time, as compared to the LPSB genotypes. On omitting the seasonal trends, the residual component showed a positive effect on LPSB (p<0.001). The trend (p = 0.04) and seasonality(p = 0.01) for rainfall also showed a positive effect on LPSB genotype, whereas seasonality only had a positive correlation (p<0.001) for LPSA (Figs 3 & 4). Comparing the patterns of LPS genotypes with humidity, a positive correlation was observed between the seasonal component of humidity with LPSA (p<0.001) and LPSB (p = 0.003). A lag period was observed in the peaks of both LPSA and B genotypes compared to trend and seasonal component of rainfall.
Among the 199 cases, 173 (87%) patients received melioidosis specific therapy. Of the 26 septicemic patients who did not receive pathogen specific therapy, 17 and 9 cases were with and without bacteremia respectively. Mortality due to melioidosis was observed among 51 (25.1%) patients. None of the patients with localized form of the disease had adverse clinical outcomes. Among the 50 patients with septicemic melioidosis, 23 (46%) succumbed to death. Case fatality rates were 14.5% (n = 25) among patients who received pathogen-specific therapy in comparison to 100% (n = 26) among those who did not receive the specific therapy.
Renal dysfunction was an independent risk factor (after considering all the host and pathogen characteristics) for both bacteremic (Adjusted OR: 3.52 (1.41–8.87), p = 0.007) and septicemic (Adjusted OR: 9.70 (3.88–24.22), p<0.001) forms of the disease (Tables 2 & 3). Infection due to B. pseudomallei having BimABm variant (Adj OR: 14.41 (3.16–65.58), p = 0.001) was an independent risk factor for neurological melioidosis in our study population. DM(26/51; 51%)[Adjusted OR: 2.37 (95%CI: 1.19–4.74), p = 0.014], septicemic melioidosis(32/51; 63%)[Adjusted OR: 2.74 (95%CI: 1.31–5.85), p = 0.007] and infection due to B. pseudomallei LPSB genotype (46/51; 90.1%) [Adjusted OR: 4.47 (95%CI: 1.16–12.20), p = 0.003] were independent risk factors for mortality in our study population.
B. pseudomallei, the etiological agent of melioidosis, demands no further negligence in view of its increasing geographical distribution, possession of numerous virulence traits and intrinsic resistance mechanisms to several antimicrobial agents. Severity of the infection and disease outcomes among patients significantly depend on the underlying host-factors, time for diagnosis/detection and befitting medical management. Few hostfactors such as DM, renal dysfunction and chronic alcoholism are well associated with poor disease outcomes among patients with melioidosis. B.pseudomallei possesses numerous proteins that play a pivotal role in the pathogenesis of the disease [4].Some genes that encode these proteins (conferring virulence) are known to be ubiquitously present among all the B. pseudomallei isolates [4].More recently, a study on the global evolution of B. pseudomallei reported geographically distinct genes/variants, conferring virulence among Australasian and Southeast Asian isolates [13]. However, there is a visible scant of whole genome sequencing data and the virulence attributes of B. pseudomallei isolates from regions outside northern Australia and northeast Thailand. Numerous studies have reported the influence of ecological factors on B. pseudomallei positivity in environmental niches [14, 15]. However, it is currently unknown if there is any influence of the ecological factors on the variable virulence genes of B. pseudomallei and whether these variations in the virulence gene profiles can lead to distinct clinical presentations and outcomes. Given this context, we report here the correlation of ecological factors in a geographical locality on infecting LPS genotypes of B. pseudomallei amongst patients.
Lipopolysaccharide of B. pseudomallei is an important virulence factor that facilitates the evasion of human immune responses during the early stages of infection. Monoclonal antibodies against the LPS of B. pseudomallei were found to reduce the severity of disease in animal models, thus implying the role of LPS as a potential vaccine candidate [1].Most intriguing finding from our study is that the majority (74%) of our patients were infected by the LPSB genotype of B. pseudomallei. This observation is in contrast with the findings from Thailand (2.3%) and Australia (13.8%), where LPSA was reported to be the prevalent infecting genotype [4]. Immunological responses and disease outcomes among animal models administered with LPS A and B types of B. pseudomallei were reported previously. While LPSA is known to confer serum resistance and grow in the presence of 10–30% of normal human serum [16], evidence from recent experimental and animal model studies suggest that LPSB is a more potent inducer of the pro-inflammatory cytokines and septic-shock [17]. In our previous study ST 1368 was the most common sequence type observed among 32 B.pseudomallei isolates [11]. Out of 14 ST 1368, 12 (85.7%) were LPSB,2 (14.2%) were LPSA and none were LPSB2. This finding suggests that the LPS genotypes can vary among isolates belonging to the same sequence type and thus making it difficult to ‘brand’ a particular ST as either a more or a less virulent one. In the present study, we did not observe a significant association of any of the three genotypes (A, B and B2) with any particular clinical presentation. We observed that infection due to LPSB genotype was an independent risk factor for mortality among our study population. However, we foresee the need for further validating this finding amongst patient populations from other geographic locations.
Expression of Burkholderia intracellular motility (BimA) protein is crucial in the pathogenesis of the disease. Among the two variants of Bim A known, BimABp was the only variant reported among isolates from Thailand and other South Asian countries. On the contrary, isolates from Australia were reported to have both BimABp and BimABm variants [4]. Among our study isolates BimABp variants were more commonly observed, but with no association with any particular clinical form of the disease. Presence of BimABm was also observed in few (n = 9) of our isolates and had a significant association with neurological presentations. Similar association of BimABm variant with neurological melioidosis was reported in Australian patients and more recently in a study using mice model [4, 18]. Filamentous hemagglutinin (FHA) is a surface protein of B. pseudomallei involved in adhesion to the host epithelial cells and formation of multinucleated giant cells [19].fhaB3 is one of the three variable genes responsible for encoding the FHA protein, which was reported previously among 100% and 83% of the isolates from Thailand and Australia respectively [4]. Further, presence of fhaB3 gene was reported in all the B. pseudomallei isolates obtained from Thai patients with bacteremic form of the disease and absence of fhaB3 gene was reported to have a significant association with cutaneous melioidosis among Australian patients [4]. Presence of fhaB3 gene was observed in almost 95% of our study isolates with no significant association with any form of the disease.
The epidemiology of melioidosis is characterized by environmental factors where rainfall plays a key role in transmission of the disease [20]. Majority of the cases in other endemic nations are known to occur during the monsoon when patients acquire the disease via inhalation or inoculation of the bacteria from soil and water. Occurrence of cases during the dry season, in many instances, is considered as a consequence of long latency and activation of the pathogen from latent foci [21]. In the present study, we observed that the LPSA genotypes had a reverse correlation with rainfall. This finding suggests the possibility of presence of LPSA genotype in dry environmental conditions, unlike the LPSB genotypes, which had a positive correlation with rainfall and humidity. However, we did not observe a significant increase in the occurrence of infections due to the LPSA genotypes during the dry season to support the assumption. Further analyzing the data, a lag period was observed between the occurrence of the cases and rainfall. The observed lag period could not be determined due to lack of enough data points and unavailability of weekly rainfall data, which remains as one of the limitations of our study. Temperature did not show any correlation with the infecting genotypes in our study, which can be attributed to the absence of striking variations in temperature through the year across the western coastal part of the country. Upon dismissing the seasonal trends, LPSB showed a positive correlation with the residual component, suggesting an influence of other environmental factors along with rainfall and humidity which needs further investigations.
Among the co-morbid illnesses observed in the present study population, renal dysfunction was found to be an independent risk factor for septicemic and bacteremic forms of the disease. DM was found to be an independent risk factor for mortality due to melioidosis in our settings, as reported in patients from Thailand and northern parts of Australia [22,23]. Strong association of DM with mortality was not surprising since 62% of the 169 patients with severe form of the disease were diabetic in our study cohort. However, DM did not have any significant association with any one particular form of the disease. Considering the high prevalence (16%) of DM amongst adult population residing in our settings [24], we foresee the need of more focused clinical studies to understand the disease outcome amongst patients with controlled and uncontrolled DM.
In our study cohort, though not statistically significant, we observed localized melioidosis amongst a higher proportion of patients with age <18 years (28.6%) in comparison with those with age >51 years (10.5%). Skin/soft tissue infections were the common clinical presentations among cases of localized melioidosis in our settings. These infections are more likely to occur in children due to their exposure to the bacteria while playing in water lodged fields during monsoon season. Higher frequency of localized, non-bacteremic cases were previously reported among Australian patients belonging to age group of <16 years [25]. Increased occurrence of localized form of the disease among patients of < 18 years of age can also be attributed to the lack of other predisposing factors responsible for the dissemination.
Put together, the present study reports few important host and pathogen-specific virulence determinants that have significant associations with clinical presentations and disease outcomes among Indian patients infected with B.pseudomallei. These findings can help in identifying high-risk cohort of patients for future studies aiming to understand the molecular pathogenesis mechanisms of the disease using integrated omics based approaches.
|
10.1371/journal.pgen.1003691 | Generation of Tandem Direct Duplications by Reversed-Ends Transposition of Maize Ac Elements | Tandem direct duplications are a common feature of the genomes of eukaryotes ranging from yeast to human, where they comprise a significant fraction of copy number variations. The prevailing model for the formation of tandem direct duplications is non-allelic homologous recombination (NAHR). Here we report the isolation of a series of duplications and reciprocal deletions isolated de novo from a maize allele containing two Class II Ac/Ds transposons. The duplication/deletion structures suggest that they were generated by alternative transposition reactions involving the termini of two nearby transposable elements. The deletion/duplication breakpoint junctions contain 8 bp target site duplications characteristic of Ac/Ds transposition events, confirming their formation directly by an alternative transposition mechanism. Tandem direct duplications and reciprocal deletions were generated at a relatively high frequency (∼0.5 to 1%) in the materials examined here in which transposons are positioned nearby each other in appropriate orientation; frequencies would likely be much lower in other genotypes. To test whether this mechanism may have contributed to maize genome evolution, we analyzed sequences flanking Ac/Ds and other hAT family transposons and identified three small tandem direct duplications with the structural features predicted by the alternative transposition mechanism. Together these results show that some class II transposons are capable of directly inducing tandem sequence duplications, and that this activity has contributed to the evolution of the maize genome.
| The recent explosion of genome sequence data has greatly increased the need to understand the forces that shape eukaryotic genomes. A common feature of higher plant genomes is the presence of large numbers of duplications, often occurring as tandem repeats of thousands of base pairs. Despite the importance of gene duplications in evolution and disease, the precise mechanism(s) that generate tandem duplications are still unclear. In this study we identified nine new spontaneous duplications that arose flanking elements of the Ac transposon system. These duplications range in size from 8 kbp to >5,000 kbp, and all cases exhibit features characteristic of Ac transposition. Using similar criteria in a bioinformatics search, we identified three smaller duplications adjacent to other hAT family transposons in the maize B73 reference genome sequence. Our results show that transposable elements can directly generate tandem duplications via alternative transposition, and that this mechanism is responsible for at least some of the duplications present in the maize B73 genome. This work extends the significance of Barbara McClintock's discovery of transposable elements by demonstrating how they can act as agents of genome expansion.
| In addition to generating additional copies of coding sequences that can be used as substrates for gene evolution [1], gene duplication may also cause immediate phenotypic impacts such as human disease [2]. Segmental duplications (SD)–two or more chromosomal segments with high homology–are common in higher plant and animal genomes. In humans and mice, ∼5% of the genome is composed of segmental duplications (≥90% in identity and ≥1 kb in length); tandem duplications (direct and inverted) account for 35.2% and 21.6% of the total duplications in the mice and human genomes, respectively [3], [4]. Many plants contain an even higher percentage of duplicated sequences. In rice, segmental duplications comprise 15–62% of the genome, depending on the sequences compared and classification criteria employed [5]–[8]. Moreover, ca. 29% of rice genes are arranged in tandem repeats [9]. Recent studies by others have also confirmed the presence of numerous duplicated sequences in the maize genome [10]–[14]. Comparison of genome sequences from different individuals of the same species revealed that copy number variation (CNV) is widespread, and that tandem duplications account for a significant proportion of the observed CNV. In Arabidopsis and maize, more than 50% of CNV segments contain tandem duplications [15]–[17]. In cattle and mice, copy number “gain” CNVs are predominantly associated with tandem local duplications, rather than interspersed duplications [18]. These observations indicate that CNVs and associated tandem duplications are contributing to rapid genome evolution.
There are several mechanisms proposed to generate tandem duplications, including 1) non-allelic homologous recombination (NAHR) between short repeats flanking a DNA segment [19], [20]; 2) break-induced replication (BIR) [21], [22] which can be mediated by short microhomology regions [19], [23]; and 3) fork stalling and template switching (FoSTeS) [24]. Here, we investigated the potential role of Class II transposable elements in directly generating tandem sequence duplications via aberrant transposition reactions.
The standard model for transposition of DNA elements involves excision of the termini of a single transposon from a donor locus and reinsertion into a target site; the net effect is the movement of the element, without any other changes to the genome. In contrast, Alternative Transposition (AT) events involve the termini of two separate, usually nearby elements. AT reactions can generate a variety of genome rearrangements; for example, the Drosophila P element system can undergo Hybrid Element Insertion (HEI) events that produce a wide array of flanking rearrangements [25]–[27] In maize, the Ac/Ds transposable element system is known to undergo at least two types of AT events that lead to genome rearrangements. First, Sister Chromatid Transposition (SCT) involves the directly-oriented 5′ and 3′ termini of closely-linked elements located on sister chromatids. Depending on the location of the transposition target site, SCT can generate chromatid bridges and breaks [28], [29], as well as flanking inverted duplications and deletions [30]. Second, Reversed Ends Transposition (RET) involves the reversely-oriented 5′ and 3′ termini of two elements located nearby each other on the same chromatid. In addition to bridges and breaks [29], RET can generate flanking inversions, deletions, permutations, and reciprocal translocations [31], [32]. An additional type of AT event termed Single Chromatid Transposition (SLCT) which involves the directly-oriented 5′ and 3′ termini of nearby elements on the same chromatid has been observed in transgenic rice containing maize Ac/Ds elements, but this reaction was not detected in maize [33].
We predicted that RET may also generate tandem direct duplications. Here we show that a single pair of reversed Ac termini induced a series of nine flanking tandem duplications ranging in size from 8157 bp to ∼5.3 Mbp. The structures of these tandem duplications and their associated deletions strongly indicate that they were indeed generated by reversed Ac ends transposition. Moreover, we identified three tandem duplications flanking other hAT transposons with the features predicted by RET in the maize B73 reference genome sequence.
To detect newly-formed duplications, we screened maize materials that contain elements of the Ac/Ds transposon system inserted into the p1 gene that controls kernel pericarp (seed coat) pigmentation. We initiated the screen with the progenitor allele P1-ovov454, which carries a pair of reversely-oriented Ac termini in the p1 gene intron 2 (Figure 1A). If transposition of the reversed Ac ends occurs during DNA replication and the excised termini insert into the sister chromatid, two unequal chromatids can be generated: one chromatid contains a tandem direct duplication, and the other contains a corresponding deletion (Figure 1D, lower and upper chromatids, respectively; for animated version please see Movie S1). These two chromatids will segregate into two adjacent daughter cells at mitosis; further mitotic divisions could generate a visible twinned sector. The new mutant chromosomes can be transmitted through meiosis to the kernels within the sectors and subsequently propagated as heritable alleles. Because the P1-ovov454 allele specifies orange variegated pericarp and orange variegated cob, both gains and losses of p1 expression can be recognized. The sector containing the deletion chromosome (white twin, p1-ww-Twin) would have white (colorless) pericarp due to loss of p1 gene exons 1 and 2, while the sector with the duplication chromosome (red twin, P1-rr-Twin) would contain two copies of Ac and exhibit fewer red and white stripes due to the negative Ac dosage effect [34], [35] (see Methods for details). We screened ∼2000 P1-ovov454/p1-ww ears and identified six ears with this type of twinned sector. Two such ears which gave rise to duplication alleles P1-rr-T1 and P1-rr-T481 are shown in Figure 2; the remaining four twin sector ears gave rise to more complex rearrangements which are still under investigation.
The RET model (Figure 1) predicts that the breakpoints of the duplication alleles (sequence a in Figure 1D) should be adjacent to Ac and p1 sequences. Therefore we used Ac casting [36], [37] and inverse PCR to isolate the sequences at the junction of the two duplication segments (Text S1). Comparison with the maize B73 genome sequence (Release 5b.60) [14] indicates that the breakpoints in P1-rr-T1 and P1-rr-T481 are located ∼460 kb and ∼5.3 Mb proximal to p1, respectively. For each allele we designed two new primers (1 and 2, Figure 1) flanking the predicted insertion sites and used these in PCR together with Ac-specific primer Ac5. Primers 1+2 amplified products containing the intact insertion sites, and primers 1+Ac5 amplified the duplication junctions of sequence a with 5′ Ac (Figure 3); the results indicate that the breakpoint sequence is duplicated in both P1-rr-T1 and P1-rr-T481. Previous semi-quantitative PCR analysis indicated that the p1 sequence proximal to Ac is duplicated; hence these alleles carry duplications. To determine the relative orientations of the duplicated segments, we performed PCR with primers 1+3 which flank the duplication junction of each allele. As shown in Figure 1D, primers 1 and 3 are separated by a 4565 bp Ac element at the duplication. By use of short PCR cycle times we could preferentially amplify products derived from somatic excision of Ac. PCR bands with sizes expected from Ac excision were amplified from both P1-rr-T1 and P1-rr-T481; sequencing of the PCR products shows that the sequence a of each breakpoint allele is linked to p1 gene sequences via a short footprint sequence typical of an Ac excision (Figure S1), and that the duplicated segments are in direct orientation as shown in Figure 1D. Together these results confirm the conclusion that P1-rr-T1 and P1-rr-T481 each carry a large segmental duplication of the sequence proximal to p1, in direct orientation.
Another prediction of the RET model (Figure 1) is that the white twin alleles (p1-ww-T1 and p1-ww-T481) should each carry a deletion as the reciprocal product of their corresponding red duplication twins. To test this, PCR analysis was performed with primer pairs 2+Ac3 and 1+Ac5 which are specific for the predicted deletion and duplication junctions, respectively (Figure 4). Products of the expected sizes were amplified from p1-ww-T1 and P1-RR-T1 (Figure 4B). Importantly, sequencing of the PCR products showed that the 8 bp sequences immediately flanking the fAc 3′ end in p1-ww-T1 and the Ac 5′ end in P1-rr-T1 are identical, indicating their origin as a target site duplication (Figure 4C), the hallmark of Ac/Ds transposition. This result confirms that the twinned duplication/deletion alleles P1-rr-T1 and p1-ww-T1 originated as reciprocal products of a single reversed Ac ends transposition event.
We attempted to isolate the p1-ww-T481 allele, but none of the plants grown from the seven kernels within the white twin sector carried the expected deletion; all carried a standard p1-ww allele derived from the normal homologous chromosome. Because the duplication in the corresponding red twin is 5.3 Mb, and a deletion of this size is most likely gametophyte lethal, we suspect that female gametophytes that received the deletion chromosome in meiosis had aborted and thus were not represented in the mature sector. This idea is consistent with the fact that the white sector contained fewer kernels than its red co-twin (P1-rr-T481; Figure 2)
DNA gel blotting was conducted to further test the structures of the candidate duplication alleles (Figure 5). Genomic DNAs were digested with SalI, and the blot was hybridized with p1-specific probe 15. The progenitor allele P1-ovov454 shows three probe 15-hybridizing bands: a 5451 bp band containing fAc, a 2693 bp band located proximal to Ac, and a 1269 bp band which is present on both sides flanking p1 and hence has a two-fold intensity on the blot. In the P1-rr-T1 and P1-rr-T481 samples, the 2693 bp band is twice the intensity of the 5451 bp band, consistent with a duplication of this proximal segment. In the p1-ww-T1 lane the 2693 band is deleted, and the 5451 bp band is absent and has shifted to a new band of ∼12 kb due to the deletion. An additional band of 1075 bp present in the P1-ovov454 and p1-ww-T1 lanes is derived from the p1-ww allele that is present in heterozygous condition in these samples (Figure 5).
As described above, the P1-rr-T1 and P1-rr-T481 duplication alleles were isolated from twin sectors with a pericarp phenotype distinct from the parental allele. Multikernel twin sectors are produced by transpositions that occur during a narrow window of early ear development and thus are relatively rare. Therefore we asked whether additional duplication alleles could be isolated from whole ears that exhibited a similar phenotype as that of the red co-twins (i.e. less red/white pericarp variegation). These whole-ear cases could have originated from reversed-ends transposition events that occurred either earlier in embryo development (such that the red twin sector encompassed the entire ear), or as pre-meiotic events. Approximately ∼80 ears of this type were identified among the ∼2000 p1-ovov454/p1-ww ears screened. Plants grown from these whole-ear cases were analyzed by semi-quantitative PCR (Figure S2) to detect changes in copy number of the p1-proximal sequences. In this way we identified 13 additional candidate duplication alleles. The breakpoints of 11 duplication candidates were cloned via Ac casting or inverse PCR (iPCR); sequencing the PCR products revealed that the breakpoints were located at various sites up to 3.3 Mb proximal to the p1 gene on chromosome 1 (Text S1). Based on the breakpoint sequences and the maize genome sequences, new primers 1 and 2 specific for each candidate allele were designed and used in PCR together with Ac primer Ac5. The results of PCR using primers 1+2+Ac5 (Figure S3) confirmed that seven of the 11 candidates carried tandem direct duplications ranging in size from 8157 bp to 3.3 Mb (Table 1). PCR using primers 1+3 flanking the presumed duplication breakpoint confirm that all of the seven alleles derived from whole ears contain tandem direct duplications. The structures of the other four alleles are more complex and are under further investigation.
These seven candidate duplication alleles were also subject to DNA gel blot analysis (Figure S4); the results show a higher relative intensity of the 2693 bp fragment in all of the candidate alleles except for P1-rr-E20, whose 8157 bp duplication does not extend into the 2693 bp fragment detected by the probe. Together the DNA gel blot results confirm the allele structures predicted from the duplication breakpoint sequences. The DNA gel blot results and semi-quantitative PCR indicated that P1-rr-E301 and P1-rr-E336 also contain duplications, but their breakpoints are not yet cloned.
The experiments described above identified nine tandem direct duplication alleles apparently generated de novo by RET of Ac/Ds elements. If this mechanism has contributed to genome evolution, one would expect to find evidence of transposon-induced duplications in the maize genome sequence. Therefore we conducted a bioinformatics search of the maize B73 reference genome for duplications with the structural features predicted by the RET model. First we identified sequences flanking known hAT family transposons and compared the flanking sequences to detect duplications; we then analyzed these candidate duplications for the sequence features predicted by the RET model. In total, 26 known maize hAT family transposons, including Ac/Ds element and 25 dhAT family elements identified in the lab of Dr. Jinsheng Lai, China Agricultural University (personal communication), were used to search for associated duplications in maize B73 reference genome (ZmB73_RefGen_V2). In this way, we identified three small duplicated segments (Figure 6) that have the sequence features predicted by the RET model (Figure 1). These three tandem duplications are associated with 3 different dhAT family elements, dhAT-Zm1, dhAT-Zm13 and dhAT-Zm24. The first duplication is located on chromosome 1 and contains two tandem direct repeats of 147 bp and 148 bp that are 93% identical. The duplicated segments are initiated by two dhAT-Zm1 elements with 95% sequence identity (Figure 6). The second duplication is located on chromosome 7 and contains two tandem direct repeats of 1262 bp and 1257 bp that are 96% identical. The duplicated segments are initiated by two dhAT-Zm13 elements with 95% sequence identity; one is intact (568 bp) and the other has a deletion of 12 bp from the 5′ TIR sequence (Figure 6). In both duplications, the first dhAT element is flanked by 8 bp direct repeats that represent the Target Site Duplications (TSDs) generated by hAT element insertion. Whereas, the second hAT element is flanked on one side by the same TSD as the first element, but the other terminus does not have a matching TSD. This is exactly the structure predicted by the RET model (Figure 1) and observed in the Ac-induced duplications (Figure 4): the first transposon has TSDs derived from the original insertion of the transposon (pre-duplication); the second transposon copy has the same TSD on one end, but the other end has a non-matching flanking sequence because it represents the subsequent RET event that generated the duplication. The third duplication (on chromosome 6) has a somewhat different structure, but is still consistent with the predictions of the RET model. This case contains direct repeats of 116 bp and 118 bp that are 99% identical; these repeats are initiated by two fractured dhAT-Zm24 elements with 96% identity. The intact dhAT-Zm24 element is 904 bp long, whereas these fractured elements contain only 288 bp and 289 bp from the 3′ end. A duplication with these structural features could also be formed by a mechanism of RET as shown in Figure S5 (Movie S2).
By taking advantage of a visual screen to identify chromosome rearrangements associated with Ac transposition events, we have isolated and characterized nine tandem duplications that arose de novo from a single progenitor allele. The endpoints of all nine duplications coincide precisely with Ac termini. Two duplications were isolated from phenotypic twinned sectors, and in one case we were able to recover and characterize a complementary deletion allele. Importantly, the endpoints of the twinned duplication/deletion alleles share a matching 8 bp TSD which is a hallmark of Ac transposition. These results indicate that the duplications originated through reversed Ac ends transpositions (RET) that occurred during or shortly after DNA replication; the excised Ac/fAc ends inserted into sites in the sister chromatid, resulting in reciprocal chromatids, one containing a tandem direct duplication, and the other bearing a corresponding deletion (Figure 1). These structures are not consistent with origin via other mechanisms. BIR and FoSTeS generally do not produce a deletion and a reciprocal duplication in the same event [19]. NAHR can generate a deletion and a reciprocal duplication. However, if these duplications were generated by NAHR between non-allelic Ac elements, then they should contain three copies of Ac (one Ac flanking the proximal and distal duplication endpoints, and one between the duplicated segments). All of the duplications we isolated lack an Ac element at one breakpoint. Although it is formally possible that one Ac element excised after the formation of the duplication, this can be excluded because the sequences at the junctions do not contain any evidence of an Ac excision footprint. Moreover, duplications generated via NAHR are recurrent; independent NAHR events between the same repeats generate duplications of the same size. However, our duplications share only one breakpoint in intron 2 of the p1 gene; the second breakpoint is different for each of the duplications, resulting in a set of nine overlapping duplications ranging in size from 8157 bp to ∼5.3 Mb.
The Drosophila P element transposon can undergo various types of alternative transposition events that can produce a multitude of rearrangement structures, depending on which transposon termini are involved in the transposition reactions, and the location of the target site (see [25] for review). In the case of the maize Ac/Ds system, fewer types of alternative transposition can occur because the transposition competence of each Ac/Ds end is dependent on strand-specific hemi-methylation of the transposon TIR. The tandem duplications described here are entirely consistent with the RET model shown in Figure 1, and with the known restriction on transposition competence of Ac/Ds elements [38], [39].
NAHR is reported to occur at a frequency of 10−5 to 10−6 in human [40]–[42]; in Arabidopsis, a frequency of 10−4 to 10−6 was observed for NAHR between two ∼1.2 kb repeats separated by ∼4.0 kb unique DNA sequence [43]. Rates of NAHR have not, to our knowledge, been reported for maize. Our results indicate that transposition-induced duplications can occur at a relatively high frequency, depending on the presence of an active transposon system with appropriately positioned elements. From a population of approximately 2000 plants, we identified seven whole ears and two twinned-sector ears with transposition-generated tandem direct duplications. DNA gel blotting and semi-quantitative PCR results indicate that two additional cases (P1-rr-E301 and P1-rr-E336; Figure S4) also carry duplications, although we could not clone their breakpoints. The calculated duplication frequency (∼0.5%) is very likely an underestimate for two reasons. First, the visual phenotype used to detect duplications (darker red pericarp and fewer purple aleurone spots) is somewhat subtle and some events may have been overlooked. Second, the screen would not have detected distal duplications because these would not alter the p1 gene or Ac dose. Distal duplications would result from insertion of the excised Ac/fAc termini into a site between the p1 gene and telomere (Figure S5; Movie S2), and these would be expected to occur as frequently as proximal duplications. Thus the real frequency of duplications derived from the P1-ovov454 allele may be closer to 1%. Given this high frequency, we asked whether Ac/Ds-induced tandem duplications could be detected in the maize B73 genome, which contains ∼50 Ac/Ds elements [44]. However, we failed to find any Ac/Ds copies adjacent to a tandem duplication, possibly because the Ac/Ds elements in the B73 genome are widely separated, and efficient reversed-ends Ac/Ds transposition requires two elements in close proximity and correct orientation [29].
In addition to the Drosophila P element and Ac/Ds systems, the Antirhinnum Tam3 element, a founding member of the hAT transposon superfamily, is known to induce flanking genome rearrangements [45]–[47], possibly via alternative transposition mechanism(s). This suggested that other transposons, in particular hAT family elements, may be capable of undergoing alternative transposition to mediate genomic rearrangements. Therefore we extended our bioinformatics searches for transposon-associated tandem duplications to a set of 25 other hAT family elements previously identified in the maize B73 reference genome (personal communication). These searches returned a total of 7611 hAT element insertions, and among these we identified three tandem direct duplications with the key structural features predicted by the RET model: First, they have exactly two repeated copies, and each repeat is initiated precisely by the transposon. Moreover, in two of the duplications the first hAT element is flanked by 8 bp TSDs, while the second (middle) element is flanked by only one of these 8 bp sequences. These features are not expected from other duplication mechanisms such as NAHR, BIR and FoSTeS, but they are perfectly predicted by the RET model. Although the duplications observed are relatively short and their frequency is low, it is possible that some examples may not have been detected for various reasons. First, the maize B73 reference genome sequence still has numerous gaps and uncertainties in the order and orientations of many sequence contigs, and these ambiguities will interfere with the identification of duplications, especially larger ones. Second, those more recent and therefore nearly identical duplications may be under-represented in the reference sequence due to collapse during sequence assembly [48], [49]. Third, those duplications in which either one of the TEs excised after formation of the duplication would not be detected by our search criteria. Nevertheless, we conclude from these results that RET-induced tandem duplication has occurred in maize evolutionary history. Given the high frequency and diversity of Class II transposons present in many eukaryotic species [50], [51], the impact of this mechanism in eukaryotic genome evolution may be significant. The RET model described here provides the conceptual basis for additional bioinformatics searches that will be necessary to assess the actual impact of this mechanism in different species.
The maize p1 gene encodes a Myb-like transcription factor controlling the pigmentation of floral tissues, including kernel pericarp (seed coat) and cob. The suffix of a p1 allele indicates its expression pattern in pericarp and cob, e.g., P1-rr specifies red pericarp and red cob, p1-ww specifies white (colorless) pericarp white (colorless) cob, and P1-ovov specifies orange variegated pericarp (seed coat) and orange variegated cob. The numeral following the suffix indicates the origin of the allele; alleles with the same phenotype but different numeral may have different structures. The P1-ovov454 allele conditions a high frequency of colorless sectors, presumably resulting from alternative transposition events which interrupt or delete the p1 gene [52]. The p1-ww-[4Co63] allele is from the maize inbred line 4Co63 [53]; heterozygous plants of genotype P1-ovov454/p1-ww-[4Co63] were fertilized with pollen from plants of genotype C1, r1-m3::Ds [4Co63]. Ac induces excision of Ds from r1-m3::Ds, resulting in restoration of r1 gene function and purple aleurone sectors. Ac/Ds transposition is subject to the negative Ac dosage effect [34], [35], in which increases in Ac copy number result in a developmental delay in Ac/Ds transposition. If reversed Ac ends transposition occurs as shown in Figure 1, two different sister chromatids would be produced: one carrying a tandem direct duplication, and the other a reciprocal deletion (Figure 1D). These chromatids will separate into two adjacent daughter cells at mitosis, forming a twinned sector after successive rounds of cell division. The sector with the deletion chromosome has lost Ac and exons 1 and 2 of the p1 gene, and thus should have colorless pericarp with no purple aleurone sectors. The sector with the duplication chromosome retains a functional P1-ovov454 gene and two copies of Ac, and thus should exhibit fewer colorless pericarp sectors, and smaller kernel aleurone sectors.
Total genomic DNA was extracted using a modified cetyltrimethylammonium bromide (CTAB) extraction protocol [54]. Agarose gel electrophoresis and Southern hybridizations were performed according to Sambrook et al [55] , except hybridization buffers contained 250 mM NaHPO4, pH 7.2, 7% SDS, and wash buffers contained 20 mM NaHPO4, pH 7.2, 1% SDS.
Sequences of oligonucleotide primers used in PCR reactions are given in Table 2; note that primers 1 and 2 are specific to each allele. PCR was performed using HotMaster Taq polymerase from 5 PRIME (Hamburg, Germany). Reactions were heated at 94°C for 2 min, and then cycled 35 times at 94°C for 20 s, 60°C for 10 s, and 65°C for 1 min per 1 kb length of expected PCR product, then 65°C for 8 min. For difficult templates, 0.5–1 M betaine and 4%–8% DMSO were added. The band amplified was purified from an agarose gel and sequenced directly. Sequencing was done by the DNA Synthesis and Sequencing Facility, Iowa State University, Ames, Iowa, United States. Ac casting and inverse PCR were performed as described previously [36].
The sequences of 26 hAT family transposable elements were used as queries to search for homologous elements in the maize B73 reference genome (ZmB73_RefGen_V2) via local BLASTN with default parameters. Two types of homologous sequences were identified: 1) intact elements, which contained both 5′ and 3′ termini; 2) fractured elements, which contained one terminal end (either 5′ or 3′) but having lengths greater than 100 bp. A PERL script was developed to extract two 100 bp segments flanking each transposon, one 5′ adjacent and one 3′ adjacent. Pair-wise comparisons were performed among the segments flanking the same terminal end within each individual hAT family. Two hAT family members with the same orientation, less than 100 kb apart, and with homologous sequences flanking one terminal end but not the other end were retained for further structural analysis. Such cases were examined manually for the following features: 1) the duplication comprises the complete sequence between the two hAT elements, and 2) the duplication is initiated by the transposable element insertion. Sequences that met the above criteria were considered as putative duplications generated by alternative transposition and were examined further for the presence of TSDs as described in the text.
|
10.1371/journal.pgen.1006884 | An ancient neurotrophin receptor code; a single Runx/Cbfβ complex determines somatosensory neuron fate specification in zebrafish | In terrestrial vertebrates such as birds and mammals, neurotrophin receptor expression is considered fundamental for the specification of distinct somatosensory neuron types where TrkA, TrkB and TrkC specify nociceptors, mechanoceptors and proprioceptors/mechanoceptors, respectively. In turn, Runx transcription factors promote neuronal fate specification by regulating neurotrophin receptor and sensory receptor expression where Runx1 mediates TrkA+ nociceptor diversification while Runx3 promotes a TrkC+ proprioceptive/mechanoceptive fate. Here, we report in zebrafish larvae that orthologs of the neurotrophin receptors in contrast to terrestrial vertebrates mark overlapping and distinct subsets of nociceptors suggesting that TrkA, TrkB and TrkC do not intrinsically promote nociceptor, mechanoceptor and proprioceptor/mechanoceptor neuronal fates, respectively. While we find that zebrafish Runx3 regulates nociceptors in contrast to terrestrial vertebrates, it shares a conserved regulatory mechanism found in terrestrial vertebrate proprioceptors/mechanoceptors in which it promotes TrkC expression and suppresses TrkB expression. We find that Cbfβ, which enhances Runx protein stability and affinity for DNA, serves as an obligate cofactor for Runx in neuronal fate determination. High levels of Runx can compensate for the loss of Cbfβ, indicating that in this context Cbfβ serves solely as a signal amplifier of Runx activity. Our data suggests an alteration/expansion of the neurotrophin receptor code of sensory neurons between larval teleost fish and terrestrial vertebrates, while the essential roles of Runx/Cbfβ in sensory neuron cell fate determination while also expanded are conserved.
| Our perception of the external world comes from our senses. Often overlooked the skin is our largest sensory organ. Specialized neurons located in the dorsal root ganglion (DRG), which innervate the body, and trigeminal ganglion (TG), which innervate the face, sense the somatosensory perceptions: light touch, temperature, pain (nociceptors) and muscle/limb position (proprioception) via nerve endings that project to the skin. These neurons receive and relay information from these diverse stimuli through distinct subclasses of neurons. Since these neurons arise from common lineages, they provide an excellent system to study how neurons develop and diversify into different subtypes. Runx transcription factors have been shown in terrestrial vertebrates (birds and mammals) to be instrumental in specifying nociceptor and proprioceptor populations by regulating the expression of a class of genes that code for the neurotrophin receptors, which are thought to be essential for specifying these neuronal fates. In our study we show that mechanisms by which Runx transcription factors regulate neurotrophin receptor expression are conserved between zebrafish and terrestrial vertebrates, yet the type of neuron specified by these genes are different such that in zebrafish the neurotrophin receptor TrkC is expressed in a nociceptor lineage instead of the proprioceptor/mechanoreceptor lineage as in terrestrial vertebrates. These data demonstrate that the specification of neuronal lineages is not fundamental to a given neurotrophin receptor but has adapted and evolved from the time fish and terrestrial vertebrates diverged 350 million years ago. Furthermore we show in fish that zebrafish Runx3 has properties that are divided between Runx1 and Runx3 in terrestrial vertebrates. Finally we show that the Runx co-factor Cbfβ is essential for its function, but the high level of Runx3 expression can overcome the loss of Cbfβ, demonstrating that Cbfβ in this context serves solely as a signal amplifier of Runx3 activity.
| Sensory neurons of the dorsal root ganglia (DRG) and trigeminal ganglia (TG) of terrestrial vertebrates convey somatosensory information from the body and face, respectively. Distinct but overlapping sensations, including touch, proprioception (body position), and nociception (pain) are perceived by different sensory neuron populations [1]. How these distinct neurons acquire their functional properties during development is an important question in understanding the assembly of the somatosensory circuits and may shed light on how these properties change in pathogenic states such as chronic pain.
A consensus model has emerged for how distinct sensory neuron populations develop in mammalian embryos. After their initial establishment from placodal and neural crest origins, sensory neurons become specialized into distinct populations. The development and survival of these populations are controlled by tropomyosin-receptor kinase (Trk) proteins that act as receptors for the neurotrophin family of growth factors. In the DRG, the three primary classes of sensory neurons are marked by distinct expression of neurotrophin receptors: nociceptors including thermoceptors and pruriceptors express TrkA, mechanoreceptors express TrkB and proprioceptors express TrkC [1]. In the TG, TrkC is primarily associated with mechanoreceptors that do not coexpress TrkB [2]. However, it appears that in early zebrafish development neurotrophin receptors may label different subset of somatosensory neurons [3].
The specification of sensory neuron subtypes is under transcription factor (TF) control. In particular, the Runt domain (Runx) TFs have been shown to be key regulators of somatosensory cell fate in terrestrial vertebrates. In the DRG, Runx3 is critical for the development of proprioceptors where it acts to suppress TrkB expression and promote TrkC expression, while in the TG Runx3 is required for the specification of TrkC-expressing mechanoceptors that innervate Merkel cells [2,4–6]. Runx1, which is expressed in a largely distinct population of somatosensory neurons from Runx3, is required for the specification of nonpeptidergic nociceptors from TrkA+ precursors [7,8]. Runx1 is also required for the expression of numerous somatosensory receptors including TRPV1, TRPA1 and Piezo2 [7,8]. The transcription cofactor Core binding factor-beta (CBFβ) is required for Runx-mediated signaling where it functions to enhance Runx binding to DNA 5 to-10-fold and protect Runx from ubiquitin-mediated degradation [9–11]. Whether or not CBFβ plays a modular or obligate role in Runx-mediated somatosensory neuron specification is unknown.
In this study we characterize the Trk receptor code in developing zebrafish and examine the roles of Runx TFs in regulating Trk expression and sensory neuron specification. While it is difficult to make comparisons across species that develop along different timescales and environmental constraints/pressures (terrestrial vertebrates develop in utero or in ovo, while zebrafish develop as free-swimming larvae that are capable of perceiving and responding to external stimuli), we found striking conservation of function as well as substantial differences between these animals. In contrast to terrestrial vertebrates, we find that in larval zebrafish, TrkA, TrkB and TrkC label distinct and overlapping classes of nociceptors. By generating loss of function mutations in zebrafish, we find that Runx3 supports a hybrid of the independent functions Runx1 and Runx3 play in terrestrial vertebrates. We also illuminate the role Cbfβ plays in Runx mediated neuronal specification, showing that Cbfβ acts as a signal amplifier of Runx3 signaling whose loss can be overcome by excess Runx3 expression. These observations argue that while fish diverged from terrestrial vertebrates over 350 million years ago, and may have substantially different somatosensory requirements, core aspects of the molecular program regulating somatosensory neuron specification even in early larval zebrafish remain largely intact.
In the TG of larval zebrafish, we characterized the overlap of the neurotrophin receptor populations with each other and with the trpv1 and trpa1b nociceptive ion channels, which respectively are required for larval zebrafish responses to noxious heat and the noxious chemicals, such as allyl isothiocyanate (AITC) [12]. We have previously shown as in mammals that neurons that express trpa1b are completely within the set that expresses trpv1 [13]. In zebrafish at 3dpf, trkC1 expression shows little overlap with trkA or trkB1, while trkA is expressed in a subset of trkB1 neurons (Fig 1A–1C and 1H, Table 1). Expression of trkB1 coincides with trpv1 while trkC1 partially overlaps with trpv1 (Fig 1D–1E, Table 1). This expression pattern is different than that described for the mouse DRG, where during early development trpv1 is broadly coexpressed in TrkA neurons which specify nociceptors, but not in TrkB- or TrkC-expressing neurons which give rise mechanoreceptors and proprioceptors [14,15]. To confirm that in zebrafish trkB1 and trkC1 are indeed expressed in nociceptors, we compared their expression to trpa1. Expression of the zebrafish trpa1b:GFP transgene completely coincides with trkC1 expression and is largely independent of trkB1 expression. We also observed a small population of neurons that coexpressed a combination of trkC1, trkB1, trpv1 and trpa1b:GFP. As in situ hybridization labeling has imperfect cellular resolution, we sought to confirm this small population expressing multiple markers by scatter labeling using CRSPR mediated insertion of the fluorescent reporter mRuby3 into the trkB1 promoter region. Our in situ hybridization results predicted that ~19% (~6/~31) of trkB1-expressing TG neurons should coexpress trpa1b:GFP at 3dpf (Table 1). Similarly we found that 20% (15/74, n = 11 animals) of trkB1:mRuby3-expressing TG neurons co-expressed trpa1b:GFP (S1 Fig). Taken together these results demonstrate that trkA, trkB1 or trkC1 are largely expressed by all or the majority of nociceptive neurons as determined by trpa1b:GFP or trpv1 expression during early larval development. This suggests that while the roles of neurotrophin receptors in somatosensory neuron survival and specification may be conserved, the distinct neuronal populations that these receptors represent are different between zebrafish and tetrapods.
To determine if transcriptional regulators of Trk expression and somatosensory neuron diversification are conserved between fish and tetrapods, we investigated the role of zebrafish Runx TFs and their cofactor Cbfβ in neuronal fate specification. We first characterized the expression of runx1, runx3, and cbfb using colormetric RNA in situ hybridization across multiple timepoints. runx1 is only expressed in Rohon-Beard (RB) sensory neurons, which innervate the body, while runx3 and cbfb are expressed in all three somatosensory neuron populations: RB, DRG, and TG neurons (Fig 2). All three genes are also expressed in other cranial ganglion, such as the facial (VII), glossopharyngeal (IX), and vagal (X) ganglia (Fig 2). In RB neurons, runx1 is transiently expressed with expression observed at the 4 somite (s) stage, but absent by 24hpf (hours post fertilization) (Fig 2A–2D, 2S, 2T and 2Y). In contrast, runx3 and cbfb RB expression begins later at 14s and continues to at least 5dpf (days post fertilization) (Fig 2G–2R and 2U–2Y). In the TG, cbfb expression was detected as early as 4s, while definitive runx3 expression was observed at 10s. Both continued to be expressed till at least 5dpf, while we found no evidence of runx1-expression even at the earliest timepoints measured (Fig 2A-X and 2EE). These data suggest that Runx TFs are in a position to regulate zebrafish somatosensory neuron diversification. However the temporal expression of runx1 in RB neurons and the lack of expression of runx1 in the TG suggest that individual Runx TFs may have specialized roles in tetrapods that are not required in larval teleost fish.
To examine the effect of impaired Runx function in somatosensory neuron development, we obtained a runx1W84X truncation mutant and generated mutations in runx3 and cbfb using CRISPRs and TALENs, respectively [16–18]. We created two nonsense mutations, runx3w144 (1bp del) and cbfbw128 (4bp del); all three mutations predict an early truncation (S2 Fig). In situ hybridization verified the lack of the gene expression in homozygous mutants, most likely caused by nonsense-mediated decay of the RNA transcript (Figs 3B, 3G and 3L and 4B and 4F). All mutants were viable and showed no obvious deformities during the time course of the experiments conducted. However all mutations induced lethality 8-10dpf. In the RBs of wild-type (WT) fish, we found that cbfb was expressed in a large proportion (~76%) of RB cells, while runx3 and runx1 expression was more tightly restricted (Fig 3A, 3E, 3I and 3M; Table 2). Mutations in runx3 or cbfb resulted in a reduction to about half of the runx1+ cells found in WT at 18s stage (Fig 3A, 3C, 3D and 3N; Table 2). Similarly, mutations of runx1 or cbfb resulted in an almost complete loss of runx3 expression in RBs at 24hpf and complete absence by 3dpf (Fig 3F, 3H and 3N; Table 2). As a result in relation to RB neurons the runx1 mutants could be considered a de facto double null mutant for runx1 and runx3. By contrast there was no change in cbfb expression in RB cells at 24hpf or 3dpf in cbfb, runx1 or runx3 mutants (Fig 3J, 3K and 3N; Table 2). Animals heterozygous for mutations in both runx1 and runx3 had significantly fewer runx1- and runx3-expressing RB neurons suggesting that in this neuronal population, Runx1 and Runx3 are functioning in the same pathway (S3 Fig). Animals heterozygous for mutations in cbfb and either runx1 or runx3 however had no effect on runx1, runx3 or cbfb expression (S3 Fig). We can therefore conclude that in the RB population, Runx and Cbfβ help maintain runx1 and runx3 expression and that a cumulative loss of Runx1 and/or Runx3 inhibits the ability of Runx proteins to facilitate runx1 and runx3 expression. Loss of either runx gene, however, does not impact cbfb expression, indicating that cbfb expression is regulated by a Runx independent mechanism.
We found different regulatory patterns in the TG, where runx3, but not runx1 is expressed. TG somatosensory neurons were identified based on their location and their expression of Elavl, a pan-neuronal marker [19]. In 3dpf wild-type fish, cbfb was expressed in all TG neurons while runx3+ neurons comprised of less than half of the TG population (Fig 4G, 4I and 4K, Table 3). Although at 24hpf, there is a transient reduction in runx3 expression in the cbfb mutant, by 3dpf the number of runx3-expressing neurons is no different than WT (Fig 4A–4C, 4G, 4H and 4K; Table 3). cbfb expression also appeared unchanged in the runx3 mutant at 24hpf and 3dpf in the TG and the total number of cbfb+ cells at 3dpf was unchanged (~96% of WT) (Fig 4D–F, 4I and 4J; Table 3). Consistent with our findings that runx1 is not expressed in the TG, we saw no gross effect on runx3 expression in runx1 mutants at 3 dpf (S4A and S4B Fig). These data suggest that cbfb is not required for the survival of runx3-expressing TG neurons and vice versa. These results are similar to what has been reported in mouse DRG, where runx expression is independent of Runx activity [2].
In some contexts, loss of mammalian Runx function results in sensory neuron death [4,20,21]. We tested whether mutation of zebrafish runx or cbfb genes affected neuronal number in the TG and the RBs. None of the mutants showed any change in the number of RB neurons at 24hpf (Table 4; S5A–S5D and S5H Fig). There was also no change in the number of cells in the TG at 24 hpf or 3dpf (Table 4; S5E–S5G and S5I Fig). We next investigated if loss of Runx activity affected caspase-3 dependent apoptosis. Wildtype, runx1, runx3 and cbfb mutants were stained for activated caspase-3 at 24, 36, 48 and 72 hpf. We observed little activated caspase-3+ apoptotic cells in the TG at any of the timepoints and no difference among the genotypes (Table 5; S6 Fig). Within the time frame observed, these data demonstrate that loss of Runx signaling in zebrafish does not lead to an increase in neuronal cell death or changes in neuron number, though we cannot rule out changes at later timepoints.
Although there is some discrepancy between different studies, it is clear that Runx regulates Trk receptor expression in mammals. Conditionally knocking out Runx1 in the DRG neurons of mice results in the expansion of TrkA-expressing neurons [7]. In Runx3 knockouts, some studies found an increase in TrkB-expressing neurons in the DRG and TG although other studies reported no change in the DRG [4,6]. Runx3 knockouts also showed either a loss of TrkC+ neurons in the DRG or a decrease in TrkC expression in the DRG and TG [4,6,22]. We performed in situ hybridization for trk receptors in each zebrafish mutant to determine whether Runx could play similar roles in regulating trk receptor expression. Analysis of the TG is simplified as only runx3 but not runx1 is expressed. trkA expression was unchanged in the TG at 24hpf and 3dpf in the runx3 or cbfb mutant (Fig 5A–5C). trkB1 expression increased at 24hpf and 3dpf in the runx3 and cbfb mutant while trkC1 expression was reduced in the TG at 24hpf and absent at 3dpf (Fig 5D–5I). To quantify these changes, we performed fluorescent in situ hybridization and counterstained for Elavl. In WT, we found that ~55% of the TG expressed trkB1 and ~42% of the TG expressed trkC1 (Fig 5P). In both runx3 and cbfb mutants we observed a significant increase in the number of trkB1+ neurons to ~89% of total and a complete loss in the number of trkC1+ neurons (Fig 5J–5P; Table 6). Surprisingly, there was no apparent change in trkC1 expression in other cranial ganglia, such as the facial and vagal ganglia that also express runx and cbfb, suggesting that neurotrophin receptor expression is regulated by an alternative mechanism in these structures.
trkA expression was also unchanged in the RBs at 24hpf in the runx1, runx3 and cbfb mutants (Fig 6A–6D; Table 7); we were unable to detect trkA expression in RBs in 3dpf larvae. The number of trkB1+ RB cells was increased at 24hpf and 3dpf in all three mutants while trkC1 expression was reduced at 24hpf and absent at 3dpf (Fig 6E–6N; Table 7). At 24hpf, trkB1+ neurons were about 70% and trkC1+ neurons were about 30% of the total RB neurons (Fig 6M; Table 7). In the runx1 and cbfb mutant, there was a significant increase of trkB1 expression, where almost all of the RB neurons expressed trkB1, and a corresponding decrease such that almost no RB neurons expressed trkC1 (Fig 6M; Table 7). The runx3 mutant showed a less dramatic, but still significant increase in trkB1+ neurons and decrease trkC1+ neurons at 24hpf, potentially due to compensation by early runx1+ expression; however by 3dpf, the distribution of expression in the runx3 mutant was indistinguishable from the other mutants (Fig 6M; Table 7). Furthermore runx1/runx3 heterozygous animals showed a significant increase in trkB1+ neurons and decrease trkC1+ neurons at 24hpf (S3 Fig).
The RB and TG data are very similar and support a role for runx and cbfb in suppressing trkB1 expression and promoting trkC1 expression. Since the increase in trkB1+ neurons is mirrored with a decrease in trkC1+ neurons and there is no change in neuronal number or apoptosis, we suggest that Runx/Cbfβ activity suppresses trkB1 expression and promotes trkC1 expression within the same subset of neurons. While apparently having no role in regulating trkA expression, a role Runx1 plays in mammals, the ability of Runx3 to regulate trkB1 and trkC1 expression is conserved between larval fish and terrestrial vertebrates even though these receptors label different neuronal subtypes in these species.
In mammals, loss of Runx1 leads to decreased or complete loss of expression of sensory ion channels including the noxious heat receptor TRPV1, the noxious chemosensor TRPA1, the light touch receptor Piezo2 and nociceptor specific voltage gated sodium channel NaV1.9 in the DRG, with corresponding somatosensory behavioral deficits [7,8]. In the TG, we examined expression of the zebrafish orthologs trpv1, trpa1b, piezo2b, and scn1α (NaV1.7 ortholog) in runx3 and cbfb mutants. The phenotypes of the runx3 and cbfb mutants were identical. trpv1 and scn1a expression appeared normal at 24hpf and 3dpf (Fig 7A–7C). By contrast, trpa1b expression by in situ and by GFP expression in the trpa1b:GFP transgenic reporter line was present at 24hpf, but absent at 3dpf (Fig 7D–7F and 7J–7M; Table 8). Loss of trpa1b expression was highly specific to the TG as expression was normal in other cranial ganglia, again suggesting that the effects of Runx/Cbfβ signaling are specific to somatosensory neurons. piezo2b expression was also present at 24hpf, but was completely gone from the TG by 3dpf (Fig 7G–7I). Consistent with Runx1 not having a role in TG neuronal specification, we observed no change in the number of GFP expressing neurons in runx1 mutants with the trpa1b:GFP transgenic reporter (Table 8; S4C and S4D Fig)
The phenotype in the RB neurons mirrored the TG phenotype. The expression of trpv1 and scn1a was unchanged at 24hpf and 3dpf (Fig 8A–8D and 8Q–8R; Table 9; S7P–S7T Fig). The expression of trpa1b as measured by in situ and GFP expression in the trpa1b:GFP transgenic was significantly reduced at 24hpf and absent at 3dpf in all three mutant lines (Fig 8E–8H and 8M–8P; Table 9). piezo2b expression was normal at 24hpf, but was absent from RB neurons at 3dpf (Fig 8I–8L). Similar to trkC1 expression, Runx/Cbfβ is required for maintenance of trpa1 and piezo2b expression but not for initiation. These data suggests that in larval zebrafish only some aspects of mammalian Runx function in regulating the expression of different sensory ion channels and other nociceptor markers were present. However, these functions are dependent on Runx3 and not Runx1 signaling.
We also asked whether loss of zebrafish Runx function changed expression of markers that define nociceptive neuron subtypes in the mouse, such as CGRP, Ret and the ATP receptor P2X3 [7,8]. We examined mRNA expression of cgrp and ret, as well as GFP expression in transgenic reporter lines Isl1SS:Kaede, which labels a subpopulation of peptidergic nociceptors, and P2x3b:eGFP in the runx3 and cbfb mutants [3,23]. In the TG, we saw no change in any of these markers at any of the timepoints we examined (Table 8; S7A–S7F and S7J–S7O Fig). Unfortunately we were unable to document clear and consistent expression of cgrp and ret in RB neurons. We can conclude however that in zebrafish TG, Runx3/Cbfβ signaling is not involved in regulating the expression of markers of peptidergic and nonpeptidergic nociceptors. These functions were likely acquired in terrestrial vertebrates alongside Runx1 specialization in specification of nociceptor cell fate.
To determine whether changes in trpa1b ion channel expression had functional consequences, we used a locomotor assay at 5 dpf to test whether Runx mutants showed defects in their responses to heat and AITC [13]. runx1, runx3 and cbfb mutants showed no behavioral change in response to heat compared to WT fish (Fig 9A–9C), reflecting our earlier findings showing no change in trpv1 ion channel expression. In contrast, the locomotor responses of 5dpf runx3 and cbfb mutants to AITC were abolished, which reflects the loss of trpa1b expression specifically in the TG and RB of these mutants (Fig 9E and 9F). However, runx1 mutants, which have normal trpa1b expression in the TG but a loss of trpa1b expression in RBs, responded normally to AITC (Fig 9D). To determine if the loss of Trpa1b in RB neurons could affect localized behavioral responses in runx1 mutants, we performed a tail-deflection assay in which AITC was puffed onto the tails of head immobilized 5dpf larvae and tail flick responses were monitored. Tail-deflection to AITC was almost completely abolished in runx1 mutants compared to WT controls while responses to a mechanical stimulus were unaffected (Fig 9G; S1–S6 Videos). This result indicates that while trpa1b expression in RB neurons is required for localized sensitivity to AITC in the body/tail, expression in the TG is sufficient to evoke WT levels of locomotion in response to bath applied AITC.
We next examined whether overexpression of runx and/or cbfb could change the fates of somatosensory neurons resulting in ectopic Runx dependent gene expression in normally Runx-negative neurons. We monitored trpa1b:GFP expression in WT, cbfb or runx3 mutant embryos after introduction of runx1, runx3 or cbfb mRNA. In either the TG or in RBs, injection of runx1, runx3, or cbfb into WT embryos resulted in little change the number of neurons expressing GFP (Fig 10A and 10C; Table 10). Chromatin inaccessibility may prevent exogenous Runx/Cbfβ complex from promoting GFP expression in neurons that do not normally express it. To test this idea, we incubated embryos with the histone deacetylase inhibitor valproic acid in combination with runx3 overexpression. Again we saw no increase in GFP-expressing neurons over WT levels (Table 10). This data indicates that Runx/Cbfβ signaling alone is not sufficient to drive trpa1:GFP expression in TG neurons.
We also tested whether Runx1 could compensate for loss of Runx3 in the TG, and whether overexpression of runx3/runx1 could overcome the loss of cbfb. Each mRNA was able to partially rescue the loss of trpa1:GFP expression in the TG and RBs in corresponding mutants (Table 10). We saw little difference in the ability of runx1 or runx3 to rescue the runx3 null mutant (Table 10), suggesting that they are functionally interchangeable and that differences in phenotype are due to spatial and temporal differences in expression. High concentrations (250pg) of cbfb RNA were unable to rescue trpa1:GFP expression in the TG of the runx3 mutant (Table 10). Excess Cbfβ was able to partially rescue GFP expression in the RB population, likely due to its ability to interact with remaining Runx1 activity in these neurons (Table 10). We next sought to test whether Cbfβ is an obligate cofactor for Runx activity, given the similarities in phenotypes after cbfb and runx loss of function, by injecting runx mRNA into cbfb null mutants. High (400pg) but not low (50pg) concentrations of runx3 RNA rescued GFP expression in cbfb null mutants in both TG and RB neurons (Fig 10B and 10D; Table 10). In cbfb null mutants that showed rescue of trpa1:GFP TG expression after injection of runx3 RNA (400pg), we found rescue of trkC1 expression in the TG of 78% of larvae (7/9) and suppression of trkB1 expression in the TG of 38% of larvae (3/8) (Fig 10E–10L). Together these data indicate that Cbfβ is acting solely through its complex with Runx TFs to facilitate sensory neuron differentiation and that excess Runx can compensate for the loss of Cbfβ enhancement of Runx DNA binding affinity and/or Runx protein stability.
We hypothesized that trkC1-expressing neurons adopted a trkB1+ cell fate in the absence of Runx function. Since expression of runx3 is unchanged in cbfb mutants, we could examine the fates of runx3+ cells in these animals. In WT fish, we found that the runx3 population contained almost all cells that expressed the trpa1b:GFP transgene and trkC1 mRNA (~96% of trpa1b:GFP+ cells express runx3; ~93% of trkC1+ cells express runx3; Fig 11C and 11G; Table 11). By contrast the trkB1 population is largely separate from the runx3+/trkC1+/trpA1b:GFP+ population (~16% and ~21% of trkB1+ cells express trkC1 and runx3 respectively; Fig 11A and 11C; Table 11). In cbfb mutants, we found in nearly all runx3+ TG cells (86%) co-expressed trkB1 (Fig 11A–11F; Table 11). These results, together with our earlier observations that showed no overall changes in TG neuron number or neuronal cell death, suggest that loss of Runx signaling results in trkC1-expressing neurons assuming a trkB1-expressing fate.
To test directly whether cells changed fate, we labeled cells with the photoconvertible protein Eos. We first examined the time course of expression loss of trpa1b:GFP using high-resolution confocal microscopy. With this methodology, we observed GFP expression in cbfb mutants at 3dpf, when trkC1 and trpa1b expression are absent as measured by in situ hybridization (see Figs 5 and 6) However, GFP was absent by 6dpf (S8 Fig). The ability to visualize these neurons at 3dpf indicated that these neurons were still present and had not been replaced in cbfb mutants with a new runx3+ neuronal subtype. We then performed scatter-labeling of TG neurons by CRISPR-mediated insertion of nls-Eos into the trkB1 or runx3 promoter region of WT and cbfb mutants transgenic for trpa1b:GFP. At 3dpf we photoconverted nls-Eos and imaged the TG to observe residual GFP to identify double positive neurons (Fig 12). We then re-imaged the same animals at 6 dpf when GFP was absent. In all animals imaged, all photoconverted runx3:nls-Eos (WT, 15/15 TG neurons, n = 4 animals; cbfb mutants, 12/12 TG neurons, n = 4 animals) and trkB:nls-Eos (WT, 7/8 TG neurons, n = 2 animals; cbfb mutants, 11/11 TG neurons, n = 2 animals) neurons were still present at 6dpf including those that were formally double positive neurons in cbfb mutants. While these results are consistent with survival at the time points measured, we cannot rule out that these neurons may be lost at later stages. In sum, these data support our conclusion that runx3+ neurons that would normally express trkC1 in WT animals assume a trkB1-expressing cell fate in Runx signaling mutants.
In this study, we set out to define the roles that zebrafish Runx/Cbfβ complexes play in refining early larval somatosensory cell fate specification. We first examined expression of neurotrophin receptors that in mammals define distinct somatosensory populations. It is inherently difficult to compare potentially dynamic developmental expression patterns across species due to the temporal differences in development and differences in the time points examined. In most terrestrial vertebrates the neurotrophin receptors TrkA, TrkB and TrkC are required respectively for the survival and specification of distinct classes of nociceptive, mechanoceptive and proprioceptive (DRG) or mechanoceptive (TG) somatosenosory neurons. In contrast, we found in larval zebrafish that the neurotrophin receptor orthologs trkA, trkB1, and trkC1 were expressed in distinct and overlapping patterns in nociceptive neurons as determined by the coexpression of the nociceptive ion channels trpv1 and trpa1b. Collectively these neurons account for nearly 90% of the TG at 3dpf with the remaining neurons not expressing any of these markers. These findings are in line with our previous work that showed that all early born TG neurons expressed nociceptive markers [13]. It is possible that zebrafish neurotrophin receptors segregate into different populations of somatosensory neurons or take on additional roles similar to terrestrial vertebrates at later developmental stages.
To test the roles of Runx TFs in regulating somatosensory neuron specification, we obtained a runx1 mutant and generated loss of function mutations in runx3 and cbfb by genome editing. In contrast to mammals in which Runx1 and Runx3 have distinct functions regulating nociceptors and proprioceptors, we found that zebrafish runx1 and runx3 both regulate nociceptors development. Both runx1 and runx3 play an overlapping role to influence RB nociceptor development in the body, while in the TG, only runx3 has this function. Analogous to mammalian Runx1, zebrafish Runx3 controls the expression of the sensory receptors trpA1b and piezo2b. However while mammalian Runx1 additionally acts to suppress trkA and cgrp expression and promote ret, p2x3 and trpv1 expression, loss of zebrafish runx3 in the TG has no apparent effect on these markers at the time points examined.
As for mammals, we found that loss of zebrafish runx3 results in loss of trkC1 expression and an increase in trkB1 expression, despite the fact that all these genes are expressed in nociceptors in early larval zebrafish and not mechanoreceptors/proprioceptors as in mammals. Taken together these results demonstrate that the core Runx regulatory program is conserved amongst larval zebrafish and mammals. Future experiments examining loss of function of zebrafish neurotrophin genes will be necessary to determine to what degree the roles of neurotrophin receptors in the specification of somatosensory cell fate have diverged.
Regulation of expression of runx genes differed depending on what population of somatosensory neurons they were expressed. Runx/Cbfβ signaling in RB neurons was required for maintaining runx expression as runx1 and runx3 expression was lost in cbfb null mutants and loss of runx1 function led to loss of runx3 expression and to a lesser extent vice versa. This suggests that a primary role of Runx1 in RB neurons is to facilitate runx3 expression. These results are consistent with findings that Runx proteins have been shown to regulate their own promoters [24]. In the TG, loss of cbfb did not alter runx3 expression at the timepoints examined, suggesting that either Runx3 can maintain its own expression in the absence of cbfb or that an idependent mechanism controls runx3 expression. Loss of runx had no effect on cbfb expression in either the TG or RB neurons. This result is consistent with studies in mouse DRG where Runx and Cbfb gene expression were shown to be regulated independently [25].
Cbfβ can act as an obligate cofactor for Runx activity in some tissues while acting to enhance Runx function in others. Cbfβ enhances Runx binding to DNA 5 to-10-fold as well as promotes Runx protein stability. For example, in mouse, a conditional deletion of Runx1 and a conditional rescue of Cbfβ indicated that the formation of hematopoietic stem cells (HSCs) and erythroid/myeloid progenitors required both Runx1 and Cbfβ [26,27]. By contrast in zebrafish, Cbfβ mutants are able to form Runx1-dependent HSCs [28]. In addition, Runx2-dependent intramembranous and endochondral ossification can develop further in the absence of Cbfβ function than they do in Runx2-/- mice [29]. This suggests that in some cases Cbfβ may act to refine Runx function but is not essential for all Runx-dependent activity. A recent study in mice found that conditional knockouts of Runx1 and Cbfβ in the DRG, had identical somatosensory neuron cell fate specification defects, suggesting that in this context Cbfβ serves as an obligate cofactor of Runx1 [25]. Howerver, loss of Cbfβ in the DRG resulted in the loss of Runx1 protein but not runx1 mRNA, so it was not possible to measure the result of Runx transcriptional activity in the absence of Cbfβ [25].
We found that in zebrafish Cbfβ serves as an obligate cofactor of Runx3 such that all somatosensory defects caused by runx3 deletion were phenocopied by cbfb deletion. However expression of high (but not low) levels of runx3 mRNA was able to rescue trpa1b:GFP expression and trkC1 expression while suppressing trkB1 expression in cbfb mutants, suggesting that Runx can function in the absence of Cbfβ. In wild-type fish, neither runx3 and/or cbfb RNA could drive etopic expression of trpa1b:GFP. Additionally, although runx and cbfb are also expressed in other cranial ganglia in addition to the TG, they do not affect the expression of trkC1 or trpa1b in these structures. These results suggest that the role of the Runx/Cbfβ interaction in managing the expression of these genes is highly specific to somatosensory neurons and may require the coexpression of other factors to mediate its effect.
Our data supports the investigation of the genetic determinants of somatososensory neuron development/function in early born larval zebrafish. We have revealed a neurotrophin receptor code in larval zebrafish that is substantially divergent from that of terrestrial vertebrates, yet shown that the developmental program that gives rise to somatosensory neuron diversity remains largely intact.
Our results suggest that in terrestrial vertebrates Runx1 and Runx3 gained additional distinct roles by subfunctionalization, with Runx1 taking over most functions in nociceptors and Runx3 acquiring additional prominent roles in proprioceptors (DRG) or mechanoreceptor (TG) specification. Terrestrial vertebrates have a much larger and well-defined set of proprioceptive neurons, presumably gained with the need for postural somatosensory feedback that occurred as a consequence of the acquisition of tetrapod limbs and the move to land. It is not clear however if teleost fish require the light touch mechanoreceptors and proprioceptors associated respectively with TrkB and TrkC in terrestrial vertebrates [30]. The need for these sensations in terrestrial vertebrates may have necessitated the ceding of nociceptor specification to TrkA and the repurposing of TrkB and TrkC to promote mechanosensory and proprioceptive neuronal fates. The dominance of nociceptive markers in larval zebrafish suggests that early nociceptive development is critical for survival of the free-swimming larvae and may therefore take precedence over other somatosensory modalities.
Experiments using zebrafish were performed under the University of Washington Institutional Animal Care and Use Committee protocols #4216–02 (approved on 9/16/2016). The University of Washington Institutional Animal Care and Use Committee (IACUC) follow the guidelines of the Office of Laboratory Animal Welfare and set its policies according to The Guide for the Care and Use of Laboratory Animals. The University of Washington maintains full accreditation from the Association for Assessment and Accreditation of Laboratory Animal Care (AAALAC) and has letters of assurance on file with OLAW. The IACUC routinely evaluates the University of Washington animal facilities and programs to assure compliance with federal, state, local, and institution laws, regulations, and policies. The OLAW Assurance number is DL16-00292.
Zebrafish were maintained at 28.5°C on a 14h/10h light/dark cycle following established methods. Embryos were maintained in E2 medium, and staged according to the standard manual [31]. runx1W84X mutants were obtained from the Liu laboratory (National Institutes of Health, Bethesda, MD, USA) [16]. A subset of the trigeminal was identified using TgBAC(trpa1b:EGFP)a128TG and Tg(isl1:Gal4-VP16,14xUAS:Kaede)a128, referred to as trpa1b:GFP and Isl1SS:Kaede, from the Schier laboratory (Harvard University, Cambridge, MA, USA) [3], and Tg(p2rx3b:EGFP)sl1Tg, referred to as p2x3b:eGFP, from the Voigt laboratory (Saint Louis University School of Medicine, St. Louis, MO, USA) [23].
All mutations were generated the AB strain of zebrafish and prior to analysis each genetic mutant was backcrossed for at least two generations into the AB background.
Transcription activator-like effector nucleases (TALENs) were used to generate a mutation in zebrafish cbfb. TALENs were assembled using the Golden Gate assembly protocol and library [17]. TALE binding sites in exon 2,of the cbfb genomic sequence, 5′- TTAAATACACCGGTTTCCGC-3′ and 5′-TTCTGGAAGCGCGCCTGCCT-3′, were identified using TALE-NT 2.0 [32].
sgRNAs were designed using http://crispr.mit.edu. We used a two-oligo PCR method to make the template DNA [33]. A scaffold oligo containing the Cas9 recognition loop and an oligo with a T7 binding site, the runx3 sgRNA sequence, and homology to scaffold oligo were synthesized. The scaffold oligo sequence was 5′-GATCCGCACCGACTCGGTGCCACTTTTTCAAGTTGATAACGGACTAGCCTTAT TTTAACTTGCTATTTCTAGCTCTAAAAC-3′. The runx3 sgRNA oligo sequence was 5′-AATTAATACGACTCACTATA(GTGCAACAAAACCCTTCCCG)GTTTTAGAGCTAGAAATAGC-3′; the target in exon 3 is indicated in parentheses.
TALEN expression vectors were linearized with SmaI and transcribed in vitro using the mMessage mMachine T7 and Poly(A) Tailing Kit (ThermoFisher). The pT3TS-nCas9n plasmid (Addgene), linearized using XbaI and purified [18], was used in an in vitro transcription reaction (T3 mMessage mMachine, ThermoFisher). The runx3 guide RNA was synthesized using the MegaScript T7 Kit (ThermoFisher). RNA products were cleaned by phenol-chloroform and isopropanol precipitation. The TALEN mixture containing equal amounts of each mRNA (400 pg each) was injected into one-cell stage AB strain zebrafish embryos. The CRISPR/Cas9 injection contained 300pg of Cas9 mRNA and 50ng of runx3 gRNA and was also injected into one-cell stage zebrafish embryos. Injected embryos were raised to adulthood, outcrossed, and gDNA isolated from F1 embryos to identify mutants.
Individual embryos were processed as previously described [34]. Embryos were incubated in 1x base solution from a 50x stock (1.25 M NaOH, 10mM EDTA pH12) at 95°C for 30 min in 25μl and then a 2x neutralization solution from a 50x solution (2M Tris-HCl pH5) was added.
Primers for identifying the cbfb mutation were 5′- AACACTCTTCTGTGCCTTTTTCATCC -3′ and 5′- TGAGGTGCGTGTACTCACTATCTCTG -3′. Primers for the runx3 mutation were 5′- CCAAACTTTCTCTGCTCGGTCCT -3′ and 5′- GAGCGCGAGTTCTGTTTGTAGC -3′. HRM mix contained 400μl 5x Gotaq buffer (Promega), 40μl 10mM dNTPs, 120μl 25mM Mgcl2, 100μl DMSO (Sigma), 100μl 20x EvaGreen (Biotium), 60μl Taq, and up to 1000μl water. PCR reactions contained 0.5μl of each primer (10μM), 10μl of HRM mix, 1μl of gDNA, and water up to 20μl. PCR was performed in a BioRad CFX Connect, using 96 well plates (BioRad cat. No. HSP9601). PCR reaction protocol was 95°C for 2 min, then 40 cycles of 95°C for 45 sec, 60°C for 30 sec, and 72°C for 30 sec, followed by 95°C for 30 sec and 60°C for 1 min. Melting curves were generated over a 65–95°C range. Curves were analyzed with the BioRad Precision Melt Analysis Software to identify mutations. Sequencing identified cbfbw128 as a 4bp deletion (nt109-112 (CACG)) and runx3w144 as a 1bp deletion (nt247 (C)), which both predict an early truncation (S1 Fig).
Full length cbfb, runx1, and runx3 were cloned from total RNA extracted from 72hpf zebrafish embryos. cbfb, runx1, and runx3 cDNA were amplified by performing reverse transcription PCR with Superscript II (Invitrogen) using primers (F: 5’- GATAGAATTCATGCCTCGGGTGGTCC -3’; R: 5’-GATAGTCGACCTAGCGCATCTTGTGATCATCAGT-3’), (F: 5’- taatacgactcactatagggATGGTTTTTCTTTGGGACGCC-3’; R: 5’- TCAGTATGGCCTCCAGACGG -3’), and (F: 5’-GCTAGGTACC ATGCATATTCCCGTAGACC-3’; R: 5’-GCGCGAATTCtttctaaatcttagtacggc-3’) respectively. cbfb and runx3 coding sequences were TA cloned into pCR2.1 (ThermoFisher), linearized with KpnI and transcribed in vitro using the mMessage mMachine T7 and Poly(A) Tailing Kit to generate capped, polyadenylated mRNA. The runx1 coding sequence was purified and transcribed in vitro using the mMessage mMachine T7 and Poly(A) Tailing Kit to generate capped, polyadenylated mRNA. A range of mRNA concentrations was injected into embryos derived from a cross between trpa1b:GFP/runx3+/w144 and Tg(Trpa1b:GFP)/cbfb+/w128 parents. Data were analyzed using two-factor ANOVA.
Digoxigenin (DIG) labeled riboprobes for runx1, cbfb [28], cgrp [3], trkA, trkB1, trkC1 [35], trpv1, and trpa1b [13] and fluorescein (FLR) labeled riboprobes for trkB1 and trkC1 were generated as previously described. Full length (2.1kb) trkA was amplified using primers (F: 5’- ATGGCTGACCATAGGGTGGCC-3’ and R: 5’- TAATACGACTCACTATAGGG CTACTCCAGGATGTCCAGGTAGAC-3’) and transcribed with T7 polymerase (ThermoFisher) to generate DIG-labeled riboprobe. An 870bp (colormetric) or 450bp (fluorescence) runx3 fragment was amplified using primers (F: 5’- gcatattcccgtagacccga-3’ and R: 5’- gatcTAATACGACTCACTATAGGGCGGA GTATGTGAAGTG-3’; F: 5’-agccacttcacatactccgc-3’ and R: 5’-gatcTAATACGACTCACTAT AGGGttagtacggcctccag-3’) respectively and transcribed with T7 polymerase. ret was linearized with NotI and transcribed with T3 polymerase (ThermoFisher). scn1a and piezo2b were linearized with EcoRI and transcribed with T7 polymerase.
In situ hybridization was carried out as previously described [36,37]. In brief, embryos were hybridized with DIG-labeled or FLR-labeled RNA probes overnight at 55°C followed by stringent washes. Samples were incubated with anti-DIG Biotin-conjugated Fab fragments (Roche, 1:1000 DIG, 1:500 FLR) and then incubated with Cy3- or FITC-tyramide (PerkinElmer). Embryos were stained with mouse anti-Elavl antibody (ThermoFisher HuC+HuD antibody 16A11, 1:1000) to identify trigeminal sensory neurons and/or rabbit anti-GFP (Invitrogen, 1:1000) to identify a subset of trigeminal sensory neurons in trpa1b:GFP fish as previously described and imaged by confocal microscopy [13]. Isl1SS:Kaede fish were stained with rabbit anti-Kaede antibody (MBL International, 1:1000). Active caspase-3 staining was used to identify TG neurons undergoing apoptosis with rabbit anti-active Caspase 3 (ThermoFisher bdb559565). In double in situ hybridization experiments, double positive cells were identified as consistently as possible by shape. Colocalization of Elavl, GFP, or an in situ probe with an in situ probe was analyzed by confocal imaging in single optical planes. Data were analyzed using ANOVA.
Larvae were raised on a 14/10 h light/dark cycle at 28.5°C. At 5dpf, individual larvae were placed in single wells on a 100 M 96-well mesh plate (MANM 100 10; Millipore). The base plate, into which the mesh plate was inserted, was constructed from .002” aluminum Shim in a can (ASTM–B– 209; Shopaid), which had been scrubbed. Base plates were thoroughly washed and soaked in distilled water before use. Two base plates filled with embryo medium were placed on each side of a dual solid-state heat/cool plate (AHP-12000CP; Teca) set to control and test temperatures, with an intervening film of water to facilitate temperature transfer. Larvae were loaded into each well of the mesh plate placed on the control side with a cut pipette. The mesh plate was then transferred from control to test temperature and larval movement videotaped for 4 minutes. To assay chemical responses, larval movement was recorded for 4 min after the mesh plate was placed into the lid of a 24 multiwell tissue culture plate (353047; BD Labware) containing allyl isothiocyanate (AITC; mustard oil; Sigma-Aldrich 377430; diluted in 1% DMSO). Behavioral responses were recorded using a Canon high definition digital video camcorder with a frame rate of 60 fps. The locomotor activity of each larva was analyzed with EthoVision XT locomotion tracking software (Noldus Information Technology, Inc.). Data were analyzed using ANOVA.
At 3dpf, the larval zebrafish offspring of a runx1+/W84X cross were imbedded in 1.5% agarose at the bottom of individual wells filled with 10mL of E2 medium. The agarose surrounding the tail of each fish was cut away allowing the tail to float freely. These fish were then exposed to either 150μM AITC (Sigma-Aldrich) in 1% DMSO (Sigma-Aldrich), 1% DMSO vehicle, or touch with 0.018cm diameter fishing line (Maxima). All solutions contained phenol red (Sigma-Aldrich) for visualization and were applied in a 50ms pulse by a Picosprizter II microinjection apparatus (General Valve Corporation). The behavioral response of each fish was recorded using a Canon high definition digital video camcorder with a frame rate of 60 fps and were manually scored based on the number of tail deflections. Following behavioral analysis animals were genotyped as described. Statistical significances was determined using ANOVA.
Transient transgenic labeling of trigeminal neurons was performed as previously described [38]. In short, single cell embryos are injected with Cas9 protein, a reporter containing plasmid, a guide RNA (gRNA) targeting the endogenous promoter of trkB1 or runx3, and a gRNA targeting the injected plasmid. mRuby3 (pKanCMV-mClover3-mRuby3 was a gift from Michael Lin (Addgene plasmid # 74252)) and nls-Eos reporter plasmids were generated by cloning into the XbaI/BamHI sites of pbsk-Mbait-Hsp-GFP (gift from Shin-ichi Higashijima)[39]. The following gRNA sequences were used: mBait (GGCTGCTGCGGTTCCAGAGG), runx3 (GGGTTTAAGCGACCAATCAG), and trkB1 (GGTGTGTTTGCTGCTTCGTG).
At 3dpf injected embryos were photoconverted using a DAPI filter and screened for nls-Eos expression. Embryos expressing nls-Eos in the TG were anesthetized with Mesab, mounted in 2% agarose, and imaged on a Zeiss LSM 880 confocal microscope. They were then returned to standard rearing conditions, before imaging at 6dpf.
|
10.1371/journal.pcbi.1007209 | Ensemble of decision tree reveals potential miRNA-disease associations | In recent years, increasing associations between microRNAs (miRNAs) and human diseases have been identified. Based on accumulating biological data, many computational models for potential miRNA-disease associations inference have been developed, which saves time and expenditure on experimental studies, making great contributions to researching molecular mechanism of human diseases and developing new drugs for disease treatment. In this paper, we proposed a novel computational method named Ensemble of Decision Tree based MiRNA-Disease Association prediction (EDTMDA), which innovatively built a computational framework integrating ensemble learning and dimensionality reduction. For each miRNA-disease pair, the feature vector was extracted by calculating the statistical measures, graph theoretical measures, and matrix factorization results for the miRNA and disease, respectively. Then multiple base learnings were built to yield many decision trees (DTs) based on random selection of negative samples and miRNA/disease features. Particularly, Principal Components Analysis was applied to each base learning to reduce feature dimensionality and hence remove the noise or redundancy. Average strategy was adopted for these DTs to get final association scores between miRNAs and diseases. In model performance evaluation, EDTMDA showed AUC of 0.9309 in global leave-one-out cross validation (LOOCV) and AUC of 0.8524 in local LOOCV. Additionally, AUC of 0.9192+/-0.0009 in 5-fold cross validation proved the model’s reliability and stability. Furthermore, three types of case studies for four human diseases were implemented. As a result, 94% (Esophageal Neoplasms), 86% (Kidney Neoplasms), 96% (Breast Neoplasms) and 88% (Carcinoma Hepatocellular) of top 50 predicted miRNAs were confirmed by experimental evidences in literature.
| MiRNAs are known as gene regulators and play critical roles in various biological processes. Many associations between miRNAs and human diseases have been identified, which promotes the understanding towards the molecular mechanisms of diseases and contributes to prevention and treatment of diseases. Computational methods of predicting potential miRNA-disease associations make the discovery more efficient and experiments more productive. We developed EDTMDA by constructing a computational framework integrating ensemble learning and dimensionality reduction. We performed global LOOCV, local LOOCV and 5-fold cross validation to evaluate performance of EDTMDA, which outperformed many classic methods. In addition, we carried out three types of case studies on important diseases, which were used to evaluate performance of model based on known associations in HMDD V2.0, for new diseases without known associations and based on known associations in HMDD V1.0. As a result, most predicted miRNAs in top 50 predictions were confirmed by experimental evidences in literature. So, we believe that EDTMDA can make reliable predictions and guide experiments to uncover more miRNA-disease associations.
| MicroRNAs (miRNAs) are a kind of endogenous non-coding RNA with the length of about 22 nucleotides, regulating the expression of genes by base paring with target messenger RNA (mRNA) [1]. Since the first two miRNAs, lin-14 and let-7, both showing positive regulation for gene expression, were found [1], increasing new miRNAs have entered into researchers’ horizons. According to latest miRbase (Release 22), a miRNA database [2], 38589 entries representing hairpin precursor miRNAs and 48885 mature miRNA products in 271 species are collected. Accumulative evidences have revealed that miRNAs usually negatively regulate gene expression and they play critical roles in various biological processes such as cell proliferation, differentiation, aging and death [3–7]. In addition, mounting close relations between miRNAs and human diseases were confirmed by abundant experimental reports. For example, the existing study has validated that the expression of mir-140 was reduced in osteoarthritic cartilage [8]. Another example is that down-regulation of mir-145 was related to the increased expression of ERG, over-expression of which was the distinct characteristic of prostate cancer [9]. Besides, deregulation of a set of miRNAs including mir-150, mir-550, mir-124a, mir-518b and mir-539 was shown to be associated with transformation of gastritis into extranodal marginal zone lymphoma [10]. It is believed that uncovering more miRNA-disease associations gives an insight into molecular mechanisms of diseases and is favorable to diagnosis, prognosis and treatment of human complex diseases [11,12]. However, the existing knowledge of miRNA-disease associations is not enough and known associations were mostly obtained from previous biological experiments that were time-consuming and costly. Therefore, increasing studies were devoted to developing computational models to predict potential miRNA-disease associations [13]. These computational models could infer miRNAs that were more likely to be related to the given disease. Based on the prediction results, biological experiments were preferentially conducted for those miRNAs to improve experimental efficiency and save time as well as expenditure.
Base on the known miRNA-disease associations in some well-known biological databases such as HMDD V2.0 [14], dbDEMC [15] and miR2Disease [16], many computational models were proposed to predict associations between miRNAs and diseases, most of which were under the assumption that functionally similar miRNAs are likely to be associated with phenotypically similar diseases [17–19]. These methods cover two main categories, network algorithm and machine learning. For example, by integrating miRNA functional similarity network, the disease phenotype similarity network and the known disease-miRNA associations network, Jiang et al. [20] proposed initial computational model to uncover potential miRNA-disease associations. For an investigated disease d, each miRNA in the miRNA network was scored by the scoring function based on cumulative hypergeometric distribution. However, the model only considered local neighbor similarity information of each miRNA so that it did not show excellent prediction results. Xuan et al. [21] developed a model of Human Disease-MiRNA association Prediction (HDMP) to predict disease-related miRNAs. In this model, miRNA functional similarity was calculated and for miRNAs in the same family or cluster, their similarity scores were given higher weight because they tend to be associated with the same disease. For investigated disease d, relevance score of each miRNA candidate was calculated based on its most weighted k similar neighbors and then ranked to attained potential d-related miRNAs. Nevertheless, HDMP were unable to work for new disease without any known associated miRNAs. In addition, HDMP was also a local network similarity-based model that only considered miRNAs’ partial similarity information, such as neighbor information. In order to make full use of global network similarity information, Chen et al. [22] first adopted global network similarity measures and proposed a method of Random Walk with Restart for MiRNA-Disease Association prediction (RWRMDA) in which random walk was implemented on miRNA functional similarity network. Although the model achieved satisfactory prediction performance, it could not deal with new disease without any known associated miRNAs. Another model named MIDP was proposed by Xuan et al. [23] based on random walk on miRNA functional similarity network. Furthermore, MIDPE that was extended from MIDP could predict potential related miRNAs for new disease without any known related miRNAs. Chen et al. [24] proposed the model of Within and Between Score for MiRNA-Disease Association prediction (WBSMDA) to predict potential miRNA-disease associations, which specially calculated Gaussian interaction profile kernel similarity for diseases and miRNAs in addition to using the miRNA functional similarity and the disease semantic similarity. In this model, both of the Within-Score and Between-Score were defined from the view of miRNAs and diseases and the final association score for miRNA-disease pair was calculated by combining Within-Score and Between-Score. WBSMDA could also be effectively applied for new diseases and new miRNAs without any known associations. Chen et al. [25] further developed the model of Heterogeneous Graph Inference for MiRNA-Disease Association prediction (HGIMDA) in which the heterogeneous graph was constructed with the same inputs as WBSMDA. An iteration process was adopted based on the graph to infer potential miRNA-disease associations. For a further improvement of prediction accuracy, Chen et al. [26] proposed another method named Matrix Decomposition and Heterogeneous Graph Inference for miRNA-disease association prediction (MDHGI), which fully utilized matrix decomposition technique for known miRNA-disease associations before constructing the heterogeneous graph that was same as HGIMDA. In addition, method of Super-Disease and MiRNA for potential MiRNA–Disease Association prediction (SDMMDA) was proposed by Chen et al [27]. In order to improve the similarity measures of diseases and miRNAs, the model introduced ‘super-miRNA’ and ‘super-disease’ that were obtained by clustering as many as possible similar miRNAs or diseases. In addition, You et al. [18] proposed the prediction model of Path-Based MiRNA-Disease Association prediction (PBMDA) that integrated various biological datasets that was same as MDHGI into the heterogeneous graph. In the graph, the association possibility was calculated by summing all path scores between a miRNA and a disease. Specially, the model penalized long paths by a decay function because these paths were considered to make less contribution to the association score for the miRNA-disease pair. However, the distance-decay function in this model was relatively simple and could be further optimized. Yu et al. [28] proposed the prediction method, MaxFlow, which constructed a miRNAome-phenome network graph where a source node and a sink node were introduced. For the given disease, the maximum information flow from the source over all links to the sink were calculated and flow quantity leaving a miRNA node was used as the association score between the miRNA and the given disease. Furthermore, Chen et al. [29] developed another prediction model of Bipartite Network Projection for MiRNA–Disease Association prediction (BNPMDA). This model first constructed the bias ratings for miRNAs and diseases based on three networks, including the known miRNA–disease association network, the disease similarity network and the miRNA similarity network. Then bipartite network recommendation algorithm was implemented to reveal potential miRNA-disease associations.
In fact, many previous computational models were established based on other types of interaction networks, such as protein-protein interaction (PPI) network, miRNA-target interaction network and so on. For example, Shi et al. [30] developed prediction model by mapping disease genes and miRNA targets on PPI networks. For a given miRNA and disease, random walk was performed on the network using the disease genes and the miRNA targets as seeds simultaneously to obtain enrichment scores as association scores of the miRNA-disease pairs. Additionally, Mork et al. [31] proposed a model of miRNA-Protein-Disease (miRPD) association prediction with proteins as the mediators, which integrated miRNA–protein associations and protein–disease associations to predict novel associations between miRNAs and diseases. However, performance of miRPD was strongly limited by miRNA-target interactions with the high false positive rate. In addition, Pasquier et al. [32] established MiRAI model that represented distributional information of miRNAs and diseases in a high-dimensional vector space and predicted novel miRNA-disease associations in terms of vector similarities.
Nowadays, machine learning has been widely applied in biomedical research [33,34], such as drug target prediction [35], transcription factor binding prediction [36], functional variant annotation [37], synergistic drug combination prediction [38], small molecule-miRNA interaction prediction [39], association prediction between long non-coding RNAs and diseases [40], and disease related RNA methylation prediction [41]. Many machine learning-based methods have been proposed to infer potential miRNA-disease associations [13]. Unlike many previous models, the model of Matrix Completion for MiRNA-Disease Association prediction (MCMDA) developed by Li et al. [17] was only depended on known miRNA-disease associations where singular value thresholding (SVT) algorithm was used to conduct matrix completion procedure and predict new miRNA-disease association. The drawback of MCMDA was that it could not predict miRNAs for new diseases without any associations. Chen et al. [42] proposed a model named Restricted Boltzmann Machine for Multiple types of MiRNA-Disease Association prediction (RBMMMDA) to predicted not only novel miRNA-disease associations but also types of association. In RBMMMDA, a two-layer undirected graphical model of Restricted Boltzmann Machine (RBM) was constructed and trained to implement prediction. RBMMMDA also could not predict miRNAs for new diseases without any known association information. Xu et al. [43] proposed a method based on a heterogeneous MiRNA-Target Dysregulated Network (MTDN). A classifier named Support Vector Machine (SVM) was built to separate positive miRNA-disease associations from negative ones based on features extracted from MTDN. Nevertheless, it was difficult to select accurate negative samples because of unavailable validation for the negative ones. Another model named Regularized Least Squares for MiRNA-Disease Association prediction (RLSMDA) that did not need negative samples was developed by Chen et al. [44]. Under the framework of Regularized Least Squares (RLS), cost functions were defined and minimized to yield optimal classifiers from miRNA and disease sides, respectively. Then the weighted average strategy was adopted to combine two optimal classifiers to obtain final prediction results. Furthermore, Chen et al. [27] introduced the model of Ranking-based K-Nearest-Neighbors for MiRNA-Disease Association prediction (RKNNMDA) to infer potential associations between miRNAs and diseases. Based on k-nearest-neighbors for miRNAs and diseases, the model calculated Hamming loss to rank these neighbors with SVM and utilized weighted voting to each predicted miRNA-disease association. In addition, Chen et al. [45] proposed another model called Laplacian Regularized Sparse Subspace Learning for MiRNA-Disease Association prediction (LRSSLMDA) which achieved prediction scores from miRNA and disease side, respectively. The model’s inputs were miRNA/disease statistical features and graph theoretic features that were extracted from the miRNA/disease similarity. Then objective functions were built in miRNA/disease side with L1-norm constraint and Laplacian regularization terms. Final predictive results were attained by combining optimization results for objective functions. Furthermore, Chen et al. [46] developed the model of Predicting MiRNA–Disease Association based on Inductive Matrix Completion (IMCMDA), which was a matrix completion-based model. MiRNA-disease association matrix was a sparse matrix and missing association values of miRNA-disease pairs could be completed by means of miRNA similarity and disease similarity feature vectors.
Considering different limitations of previous models and improvement room for prediction accuracy, we developed the model of Ensemble of Decision Tree based MiRNA-Disease Association prediction (EDTMDA) to infer novel miRNA-disease associations. The inputs of the model were features which were extracted from integrated miRNA similarity, disease similarity and known miRNA-disease associations. The model adopted ensemble learning strategy that integrated multiple classifiers (base learners) to get final prediction results, which reflected association probability for candidate miRNA-disease pairs. Three cross validation methods, including global leave-one-out cross validation (LOOCV), local LOOCV and 5-fold cross validation (5-fold CV) were implemented to evaluate performance of EDTMDA. As a result, AUC of 0.9309 for global LOOCV, 0.8524 for local LOOCV and 0.9192+/-0.0009 for 5-fold CV were obtained. To our knowledge, the AUCs of EDTMDA are higher than almost all previous models. In addition, three types of case studies for important human diseases were further carried out to evaluate the ability to predict miRNAs related with the investigated disease. There were 47 (Esophageal Neoplasms), 43 (Kidney Neoplasms), 48 (Breast Neoplasms) and 44 (Carcinoma Hepatocellular) of top 50 predictions confirmed by previously published literature. These aforementioned validation experiments proved that EDTMDA is a reliable and excellent model to predict potential miRNA-disease associations.
In our work, known human miRNA-disease associations verified by experimental evidences in literature were obtained from HMDD V2.0 which included 5430 associations between 495 miRNAs and 383 diseases [14]. Here, Y ∈ Rnm×nd was used to denote an adjacency matrix, where nm and nd represented the number of miRNAs and diseases, respectively. If miRNA m(i) and disease d(j) had association according to HMDD V2.0, the element Y(m(i), d(j)) equaled to 1, otherwise 0.
MiRNA functional similarity scores could be computed based on the MISIM method proposed by Wang et al. [47] and downloaded from the website: http://www.cuilab.cn/files/images/cuilab/misim.zip. We denoted FS as the score matrix of miRNA functional similarity and the element FS(m(i), m(j)) represented the functional similarity scores between miRNA m(i) and m(j).
Disease semantic similarity was computed according to the literature [47]. we download MeSH descriptors from the National Library of Medicine (http://www.nlm.nih.gov/), from which the relationship of various diseases could be obtained based on disease Directed Acyclic Graph (DAG). For example, a DAG(D) = (D,T(D), E(D)) was used to represent disease D, where T(D) was the node set including all parent nodes of disease D and disease D itself, and E(D) was defined as the set of edges pointing to child nodes from parent notes. In DAG(D), we defined the semantic value of disease D to DV1(D) as follows:
DV1(D)=∑d∈T(D)D1D(d)
(1)
{D1D(d)=1ifd=DD1D(d)=max{Δ*D1D(d′)|d′∈childrenofd}ifd≠D
(2)
where D1D(d) represented the contribution of disease d to the semantic value of disease D in DAG(D). As shown in Eq 2, disease D was the most specific disease in DAG(D) and its contribution to the semantic value of itself was set to 1. Those parent nodes locating farther from node D are more general denominations, having fewer contribution to the semantic value of disease D. To realize that, semantic contribution factor Δ was introduced (0< Δ <1) and we set Δ = 0.5 in this study, referring to the literature [47]. Based on the assumption that two diseases sharing larger parts in their DAGs tend to have higher semantic similarity, the semantic similarity between disease d(i) and d(j) could be defined as follows:
SS1(d(i),d(j))=∑t∈T(d(i))∩T(d(j))(D1d(i)(t)+D1d(j)(t))DV1(d(i))+DV1(d(j))
(3)
In order to obtain more comprehensive and accurate disease semantic similarity assessment, we needed to measure the similarity from different perspectives. Therefore, another model of measuring disease semantic similarity was adopted according to the literature [21]. We considered that the number of disease DAGs that a disease term may appear in are not always the same and for disease terms in the same layer of DAG(D), the disease term appearing in fewer DAGs should be more informative. i.e., the disease term should have larger semantic contribution to disease D. In this model, semantic contribution of disease d to disease D in DAG(D) was defined as follows:
D2D(d)=−log[thenumberofDAGsincludingdthenumberofdiseases]
(4)
Similar to disease semantic similarity model 1, the semantic value of disease D and semantic similarity between disease d(i) and d(j) was respectively given as follows:
DV2(D)=∑d∈T(D)D2D(d)
(5)
SS2(d(i),d(j))=∑t∈T(d(i))∩T(d(j))D2d(i)(t)+D2d(j)(t)DV2(d(i))+DV2(d(j))
(6)
Two disease semantic similarity models defined semantic contributions of the disease d to disease D in DAG(D) in different ways. We defined it based on the theory that those parent nodes locating farther from node D are more general denominations, having fewer contribution to the semantic value of disease D in model 1, while in model 2, we defined it by considering that the disease appearing in fewer DAGs should be more special and have larger semantic contribution to disease D.
According to the literature [48], we could calculate Gaussian interaction profile kernel similarity for miRNAs (diseases), which constructed Gaussian kernel with the adjacency matrix Y. Taking miRNA as an example, the Gaussian interaction profile kernel similarity between miRNA m(i) and m(j) was calculated as follows:
GM(m(i),m(j))=exp(−γd‖Y(m(i),*)−Y(m(j),*)‖2)
(7)
Here, Y(m(i), *) and Y(m(j), *) are the ith and jth row of adjacency matrix Y, respectively, representing interaction information between corresponding miRNA and all diseases. Parameter γd controlled the bandwidth and was set as follows:
γd=γ′d/(1nm∑i=1nm‖Y(mi,*)‖2)
(8)
Analogically, according to the literature [48], Gaussian interaction profile kernel similarity for diseases could be calculated as follows:
GD(d(i),d(j))=exp(−γd‖Y(*,d(i))−Y(*,d(j))‖2)
(9)
γd=γ′d/(1nd∑i=1nd‖Y(*,d(i))‖2)
(10)
where Y(*, d(i)) and Y(*, d(j)) are the ith and jth column of adjacency matrix Y, respectively, meaning interaction information between corresponding disease and all miRNAs.
We computed disease semantic similarity based on DAGs of diseases, but we could not get DAGs for all diseases. That is, for the specific disease without DAG, the semantic similarity score between the disease and other diseases could not be computed in both disease semantic similarity models. In order to obtain all disease similarity information, we integrated disease semantic similarity with Gaussian interaction profile kernel similarity according to [24] as follows:
SD(d(i),d(j))={SS1(d(i),d(j))+SS2(d(i),d(j))2d(i)andd(j)hassemanticsimilarityGD(d(i),d(j))otherwise
(11)
where the average of two disease semantic similarity models was used as disease semantic similarity. Similarly, integrated miRNA similarity was given according to [24] as follows.
EDTMDA was implemented based on integrated miRNA similarity matrix SM, integrated disease similarity matrix SD and known miRNA-disease associations matrix Y. At first, according to literature [49], three types of miRNA (disease) features were extracted based on the above matrixes SM (SD) and Y and used to form the feature vectors, represented by FM (FD). Type 1 features covered the statistical measures summarized for each individual miRNA (disease) in Y and SM (SD) (including sum, mean, histogram distributions of miRNA/disease similarity scores); type 2 features included graph theoretical measures for network constructed by SM (SD) (including some neighbors’ attributes, betweenness, closeness, eigenvector centrality and Page-Rank scores of miRNA/disease similarity network); type 3 features focused on each miRNA-disease pair in Y based on matrix factorization of Y and graph theory-related statistics for network constructed by Y. Then, ensemble learning strategy was introduced based on random selection of negative samples and features, which included many base learnings and each base learning yield a base classifier, DT. Particularly, PCA was employed to reduce feature dimensionality during each base learning. The final association scores were obtained by computing the average of all prediction results from these DTs (motivated by the study of Ezzat et al. [50]). The base learning contained following steps (see Fig 1).
Firstly, construction of training sample set was operated. Because there were minority positive samples, accounting for about 2.9% of all possible samples in HMDD V2.0 used by our method, we chose all positive samples and some negative samples which were randomly singled out from the samples without known associations to construct the training set of our model. Particularly, negative samples were guaranteed to have the same number with positive samples. Here, P = {(m(i), d(j))|Y(m(i), d(j)) = 1} and U = {(m(i), d(j))|Y(m(i), d(j)) = 0} represented the set of positive samples and samples with unknown associations, respectively. The set N (N∈U) represented negative samples selected from U and |N| = |P| (|N| and |P| meant the number of elements in N and P, respectively). The set of T = P ⋃ N was training set in base learning. In addition, FM∈Rnm×d and FD∈Rnd×d (d represented the number of extracted miRNA/disease features) represented feature matrix of miRNAs and diseases in training set T, respectively. We constructed feature subsets of miRNAs and diseases by randomly selecting miRNA/disease features and used parameter r (0 < r ≤ 1) to control the size of feature subset. That is, ⌊r×d⌋ features were randomly sampled to construct feature subset. FM(1)∈Rnm×d1 and FD(1)∈Rnd×d1 represented feature subset of miRNAs and diseases, respectively (where d1 = ⌊r×d⌋).
Secondly, feature dimensionality reduction was applied to miRNA/disease feature subset. In our model, ensemble learning strategy was adopted to yield a large number of base learners, which brought much noise or redundant information to degrade prediction performance. To address this issue, PCA, an unsupervised dimensionality reduction algorithm [51], was employed to reduce miRNA/disease feature dimensionality of feature subset. Here, we saved top 10 miRNA (disease) features after dimensionality reduction, keeping almost all feature information. Here, FM(2) and FD(2) represented feature matrix of miRNAs and diseases after dimensionality reduction.
Thirdly, the DT, a base classifier, was trained with training set. For the sample in training set T, feature principle components of miRNA and disease, i.e., miRNA feature vector and disease feature vector in FM(2) and FD(2), were spliced as the feature vector of the sample, which was used as input vector of the DT. Our training set could also be denoted with T′ = {(x1, y1), (x2, y2), ⋯,(xn, yn)}, where xi=(xi(1),xi(2)⋯,xi(d2)) was the d2-dimensional input vector (d2 = 20) and yi represented the observed value of the ith sample in adjacency matrix Y, and n was the number of samples in training set. For the DT, we constructed the regression tree model with the arithmetic of CART, which was on the basis of squared error minimum criterion [52]. Yielding the regression tree could be described as a progress of building a binary decision tree recursively. If we selected the feature value xi(j) to partition feature space R, j and s (xi(j)=s) were the splitting variable and splitting point, respectively, and two subspaces were defined as follows:
R1(j,s)={x|x(j)≤s}andR2(j,s)={x|x(j)>s}
(13)
Then regression tree could be described as:
f(x)=ckx∈Rk,k=1,2
(14)
where ck denoted output value of subspace Rk and its optimal value was calculated by minimizing squared error ∑xi∈Rk(yi−f(xi))2. The solution was given as follows:
c^k=1Nk∑xi∈Rk(j,s)yix∈Rm,m=1,2
(15)
where Nk was the number of input vectors in subspace Rk. In order to choose the optimal splitting variable and splitting point, variable j and s were traversed to solve the following equation:
minj,s[∑xi∈R1(j,s)(yi−c1)2+∑xi∈R2(j,s)(yi−c2)2]
(16)
The optimal splitting variable j′ and splitting point s′ was obtained. The pair (j′, s′) was used to partition the feature space according to the formula (13) and the output was calculated based on the formula (14) and (15). Then new optimal splitting variable and splitting point were sought in subspace R1 and R2, respectively. Then new output c^k (k = 1,2,3,4) was calculated in 4 subspaces, respectively. This procedure was repeated until the subspace could not be partitioned. At last, the feature space was divided into K subspaces and the final regression tree was described as follows:
f(x)=ckx∈Rk,k=1,2,⋯,K
(17)
Based on random selection of negative samples and miRNA/disease features, M base learnings including above three steps were implemented to yield M DTs. The simple average strategy was adopted for these DTs to obtain final prediction scores. Fig 2 shows the pseudocode of EDTMDA. The code and data of EDTMDA is freely available at https://github.com/chi-young1/EDTMDA.
Based on known miRNA-disease associations in HMDD V2.0, we implemented LOOCV and 5-fold CV to evaluate the performance of EDTMDA. Receiver operating characteristic (ROC) curves are widely used to evaluate model performance in previous literature of predicting miRNA-disease associations and in order to more conveniently implement performance comparison, we also employed it in our study. Moreover, ROC curves are insensitive to class imbalance, which is suitable for assessing our model’s ability to recover hidden known associations from mass candidates (unknown associations).
LOOCV, including global LOOCV and local LOOCV, were implemented to evaluate the performance of EDTMDA. Global LOOCV was used to evaluate model’s global prediction ability for all disease simultaneously, which evaluated recover ability for a hidden miRNA-disease association from all unknown associations. Local LOOCV was used to evaluate model’s local prediction ability for a specific disease, which assessed the recover ability for a hidden miRNA-disease association from unknown associations of the investigated disease. Therefore, there is big difference for these two types of LOOCV. In global LOOCV, each known miRNA-disease association was singled out as test sample in turn and other known associations were treated as training samples for model training. Note that we recalculated Gaussian interaction profile kernel similarity of miRNAs and diseases when a known miRNA-disease association was removed, changing miRNA-disease adjacency matrix. Prediction scores of the test sample and all candidate samples (That is, those miRNA-disease pairs without association evidences) could be obtained after implementing EDTMDA. Then the test sample was ranked with all candidate samples based on their scores, and if the rank was higher than the specific threshold, the test sample was successfully predicted. Different from global LOOCV considering all diseases simultaneously, the test sample was only ranked with candidate samples containing the same disease as the test sample. In model performance evaluation, true positive rate (TPR, sensitivity) and false positive rate (FPR, 1-specificity) are usually calculated based on given threshold. Sensitivity indicates the percentage of the test samples ranked higher than the specific threshold; specificity means the percentage of negative samples ranked below the threshold. When different thresholds were given, we can obtain corresponding TPR and FPR to plot the ROC curve with the TPR as the vertical axis and FPR as the horizontal axis. ROC curve could be used to vividly show predictive performance of the model, and a ROC curve closer to the upper left corner of the figure represents more accurate performance. Furthermore, area under the ROC curve (AUC) was calculated to quantitatively evaluate model performance. AUC = 1 represents that the model has perfect prediction performance and AUC = 0.5 refers to random performance.
We compared the performance of EDTMDA with other classical models in terms of AUC under cross validation. The details of compared models were provided as follows: HGIMDA [25]: The model constructed a heterogeneous graph by integrating multiple biological data, where all paths with the length equal to three were summarized to infer potential miRNA-disease associations (The parameter used for comparison was α = 0.4). MDHGI [26]: The model employed matrix decomposition for miRNA-disease association matrix before implementing the heterogeneous graph inference that was same as HGIMDA (The parameters used for comparison were α = 0.1, μ = 10−4, maxμ = 1010, ρ = 1.1, ε = 10−6 and α = 0.4). RLSMDA [44]: The method combined two classifiers trained from the miRNA space and disease space respectively based on the framework of regularized least squares algorithm (The parameters used for comparison were ηM = 1, ηD = 1 and ω = 0.9). HDMP [21]: The relevance scores of unlabeled miRNAs were computed based on functional similarity of miRNAs’ k nearest neighbors. Besides, the members in the same miRNA family or cluster are assigned higher weight (The parameters used for comparison were α = 4, β = 4 and k = 20). WBSMDA [24]: The model defined the Within-Score and Between-Score from the miRNA side and disease side, then combined these score to infer potential miRNA-disease associations. RWRMDA [22]: Random walk was implemented on the miRNA-miRNA functional similarity network (The parameters used for comparison were r = 0.2 and threshold = 10−6). MCMDA [17]: The model utilized the matrix completion algorithm to update the adjacency matrix of known miRNA-disease associations (The parameters used for comparison were ε = 10-4 and max_iter = 500). MiRAI [32]: The model represented distributional information on miRNAs and diseases in a high-dimensional vector space and defined associations between miRNAs and diseases in terms of their vector similarity (The parameter used for comparison was r = 400). MaxFlow [28]: A combinatorial prioritization algorithm was designed for miRNA-disease association prediction by modifying the existing maximizing information flow method (The parameters used for comparison were α = 0.1, β = 0.6, γ = 100, η = 6 and σ = 10). PBMDA [18]: The model constructed a heterogeneous graph consisting of three interlinked sub-graphs and computed the accumulative contributions from all paths between a miRNA-disease pair as the association score, which specially set decay factor to cut down the contributions of longer paths to miRNA-disease association scores (The parameters used for comparison were T = 0.5, L = 3 and α = 2.26). LRSSLMDA [45]: A common subspace for the miRNA/disease profiles, a L1-norm constraint and Laplacian regularization terms were joint to construct the prediction model (The parameters used for comparison were γ = 2, μ = 1 and λ = 1). MIDP [23]: A novel random walk with different transition weight for labeled nodes and unlabeled nodes was implemented on miRNA functional similarity network to predict miRNAs related to the disease with some known related miRNAs and for the new disease without any known related miRNAs, the model extend the walking on a miRNA-disease bilayer network (The parameters used for comparison were rQ = 0.4 and rU = 0.1.).
Fig 3 showed the performance comparisons between EDTMDA and other several models in the framework of global and local LOOCV. EDTMDA, LRSSLMDA, PBMDA, MDHGI, HGIMDA, MCMDA, MaxFlow, RLSMDA, HDMP and WBSMDA obtained AUC of 0.9309, 0.9178, 0.9169, 0.8945, 0.8781, 0.8749, 0.8624, 0.8426, 0.8366 and 0.8030 in global LOOCV, respectively; they obtained 0.8524, 0.8418, 0.8341, 0.8240, 0.8077, 0.7718, 0,7774 0.6953, 0.7702 and 0.8031 in global LOOCV, respectively. RWRMDA and MIDP did not have an AUC value in global LOOCV because they could not simultaneously make predictions for all diseases. Additionally, global LOOCV also could not be implemented for MiRAI because the association scores yielded from the model were highly related with the number of known associated miRNAs of a disease. For a disease with more related miRNAs, the association scores for its candidate miRNAs were more likely to be higher. Therefore, it was not objective to simultaneously consider association scores of all diseases in global LOOCV. AUCs of 0.7891 for RWRMDA, 0.8196 for MIDP and 0.6299 for MiRAI were obtained in local LOOCV. Higher AUC values of EDTMDA in LOOCV indicated that our model had more accurate prediction than most previous models.
We implemented 5-fold CV to further evaluate the prediction performance of EDTMDA. In 5-fold CV, all positive samples (That is, those miRNA-disease pairs with known associations) were randomly divided into five equal-sized groups, four of which, along with same size of selected randomly negative samples, used to training the classifier. The omitted group (hidden positive samples) was added to all unknown associations to construct all candidates. Specially, we recalculated the Gaussian interaction profile kernel similarity of miRNAs and diseases when each group of miRNA-disease associations were removed. Then similar to global LOOCV, the association scores of candidates were calculated and then ranked by their scores. The higher the hidden positive samples were ranked, the better the performance was. That is, we removed some known associations and assessed ability to recover these hide associations to evaluate performance of model. This procedure was repeated 100 times because sample division was random in 5-fold CV. As a result, EDTMDA obtained average AUC with standard deviation of 0.9192+/-0.0009, surpassing all other methods compared (See Table 1), which further shows the superior performance of EDTMDA.
In our method, multiple base learnings were constructed to generate many base classifiers (DTs) base on random selection of negative samples and miRNA/disease features, which also brought some noise or redundancy to influence final prediction results. To address this issue, we used PCA to implement dimensionality reduction for miRNA/disease feature subset. To evaluate the effect of dimensionality reduction to our model, we assessed performance of the method after removing dimensionality reduction step in each base learning. That is, we spliced miRNA and disease features of feature subset as the input of base classifiers. The AUC comparison results between EDTMDA with dimensionality reduction and EDTMDA without dimensionality reduction were shown in Table 2, which indicated that dimensionality reduction in base learning contributed to improve prediction performance of the model.
We conducted comparison of prediction performance between EDTMDA and RF which is also an ensemble learning method with DT as base classifier. Extracted miRNA features and disease features were spliced as the input vector of RF and RF was implemented using RandomForestRegressor that is an algorithm package of RF in Python, where default parameter values were used other than n_estimators (It was set as 50, meaning that the number of trees in RF is same as in EDTMDA). As shown in Table 3, EDTMDA is notably outperformed RF under three cross validations. There are two main differences between EDTMDA and RF. First, EDTMDA randomly selected a different negative sample set for each base classifier while RF performed bagging on the same negative set. That is, EDTMDA used more negative samples for model training than RF. Second, EDTMDA included all positive samples in training set for each base classifier, but RF performed bagging on the positive samples so that each DT in RF used only a subset of all positive samples. We concluded that prediction performance of the model was sensitive to positive samples and the best strategy was to include all positive samples for each base classifier in ensemble learning. Moreover, EDTMDA incorporated more data for model training, obtaining better prediction performance than RF.
To further access the prediction ability of EDTMDA, three types of case studies were carried out. For the sake of brevity, we selected several important human diseases to analyze in detail. The first type of case study was concerned with Esophageal Neoplasms and Kidney Neoplasms, and known miRNA-disease associations in HMDD V2.0 were used as training samples. All candidate miRNAs that were unassociated with the investigated disease in HMDD V2.0 were ranked according to their predicted association scores. Top 50 of candidate miRNAs were validated in two other miRNA-disease association databases, dbDEMC [15] and miR2Disease [16].
Esophageal Neoplasms is a serious malignancy with high mortality rate, ranking sixth among all cancer in mortality [53]. Squamous cell carcinoma (SCC) is the most common type of Esophageal Neoplasms and the black with SCC was three times higher than the white [54]. There will be 17190 new cases in Esophageal Neoplasms and 15850 people dying of the Esophageal Neoplasms in 2018 according to the study [55]. Many previous studies have confirmed the associations between the Esophageal Neoplasms and various miRNAs. For example, the higher expression of miRNA-506 was found in squamous cell carcinoma (ESCC) patients than in heathy people [56]. Moreover, according to the study [57], the expression of miRNA-382-5p notably increased and miRNA-133a-3p notably decreased in esophageal adenocarcinoma (EAC). In case study of Esophageal Neoplasms, 10 out of top 10 and 47 out of top 50 predicted miRNAs related to Esophageal Neoplasms were confirmed by dbDEMC or miR2Disease (See Table 4).
Kidney Neoplasms, also known as Renal cell carcinoma (RCC), accounts for 2–3% of all the adult cancers [58]. It has been estimated that 65340 Americans will be diagnosed with Kidney Neoplasms and 14970 will die of the disease in 2018 [55]. Some studies have confirmed that dysregulation of miRNAs is closely related to Kidney Neoplasms. For example, Arai et al. [59] found that low expression of mir-10a-5p had association with overall survival in Kidney Neoplasms patients because downregulation of mir-10a-5p inhibited cancer cell migration and invasion. Another study showed that mir-21 played an important role in Kidney Neoplasms progression and could resist chemotherapeutic drugs used for treatment of Kidney Neoplasms [60]. As a result of case study for Kidney Neoplasms, 9 out of the top 10 and 43 out of the top 50 miRNAs were validated to have associations with Kidney Neoplasms by dbDEMC and miR2Disease (See Table 5).
We exhibited complete prediction results inferring potential disease-associated miRNAs that were ranked based on their predicted association scores, which we expect to be beneficial for experimental studies in the future (See S1 Table).
The second type of case study for Breast Neoplasms was implemented to prove the applicability of EDTMDA to new diseases without known related miRNAs. We removed all known Breast Neoplasms-miRNA associations in HMDD V2.0 so Breast Neoplasms could be regarded as new disease. After implementing EDTMDA to predict and rank potential Breast Neoplasms-related miRNAs based on other known disease-miRNA associations, we confirmed that 10 out of top 10 and 48 out of top 50 predicted Breast Neoplasms-related miRNAs were validated by HMDD V2.0, dbDEMC and miR2Disease (See Table 6). Hsa-mir-210, ranking first in our prediction result list, had the greatest possibility associating with Breast Neoplasms. The study of Zehentmayr et al. [61] has revealed the association that hsa-mir-210 was overexpressed in contralateral unaffected breasts (CUB) of patients with breast cancer. This case study showed that our model was also reliable when applied to predict miRNAs related with new diseases.
Finally, to test robustness of our model, we carried out the third case study for Carcinoma Hepatocellular based on known associations in HMDD V1.0 including 1395 associations between 271 miRNAs and 137 diseases. In this case study, we ranked candidate miRNAs for Carcinoma Hepatocellular and validated top 50 predictions with experimental evidences. As has been defined, a candidate miRNA was a miRNA unassociated with the Carcinoma Hepatocellular according to HMDD v1.0, which guaranteed that validation of the predictions was completely independent of training database HMDD V1.0. As a result, 10 out of top 10 and 44 out of top 50 potential miRNAs associated with Carcinoma Hepatocellular were validated by HMDD V2.0, dbDEMC and miR2Disease (See Table 7). For example, hsa-mir-146b (1st in the prediction list) was down-regulated in Carcinoma Hepatocellular and could inhibit tumor growth and metastasis of Carcinoma Hepatocellular [62]. Aforementioned results indicate that EDTMDA has good robustness, showing satisfactory performance in different dataset.
We randomly shuffled ‘1’ and ‘0’ elements and kept their respective numbers unchanged in adjacency matrix, which was used to test whether our model suffered from overfitting. The AUC of three cross validations including global LOOCV, local LOOCV and 5-fold CV were 0.4939, 0.4413 and 0.5005+/-0.0029 respectively, which indicated that EDTMDA effectively avoided overfitting. Furthermore, label randomization test was implemented in three case studies by randomly shuffling ‘1’ and ‘0’ elements and keeping their respective numbers unchanged in adjacency matrix. The results were shown in Table 8, compared with the results under true labels. From the comparison results, we could draw the conclusion that EDTMDA is an effective tool to unveil more potential miRNAs related to diseases.
In our model, we randomly selected some miRNA-disease pairs without known associations as negative samples. Moreover, considering that different diseases with different numbers of associated miRNAs, we designed a new way to select negative samples, which reflected the contribution of each disease to the positive sample set. For the new way, negative samples were sampled randomly for each disease to have the same size as the positive samples of the disease. That is, more negative samples were sampled for the disease with more known associated miRNAs. This new way to select negative samples was named local random and the previous way to select negative samples from all the negative was named global random. For the model using local random to select negative samples, we implemented model evaluation under three cross validations (global LOOCV, local LOOCV and 5-fold CV), and the AUCs were 0.8224, 0.7871 and 0.8180+/-0.0019 respectively, which was significantly inferior to AUCs of 0.9309, 0.8524 and 0.9192+/-0.0009 in our model using global random to select negative samples. For the local random to select negative samples, the poor performance of model could be that more false negative samples (miRNA-disease pairs with potential associations) were selected. It is apparently observed that miRNAs prefer to relate to some specific diseases in our dataset and we think that there should be more potential miRNA-disease associations for these specific diseases. But in local random to select negative samples, more selected negative samples were derived from the negative of those specific diseases with more related miRNAs, i.e., more false negative samples were selected. In global random to select negative samples, we avoided selecting more false negative samples for model training and obtained better model performance.
Increasing researchers are devoted to developing computational methods to infer potential miRNA-disease associations as these methods can be valuable complements to experiments. In this study, we proposed a computational method called EDTMDA under the framework of ensemble learning and dimensionality reduction. The Gaussian interaction profile kernel similarity scores for miRNAs and diseases were first calculated from known miRNA-disease associations. Then integrated miRNA (disease) similarity could be obtained via integrating miRNA functional similarity (disease semantic similarity) and Gaussian interaction profile kernel similarity of miRNAs (diseases). In addition, the feature vectors for the miRNA-disease pair was constructed by conducting feature extraction on integrated similarity and known miRNA-disease associations. Multiple base learnings were built based on random selection of negative samples and miRNA/disease features so that many decision trees (DTs, base classifiers) were attained. Particularly, in order to remove the noise or redundancy, PCA was utilized to reduce feature dimensionality during each base learning. Final prediction results were given by adopting simple average strategy for these DTs.
The success of this model is mainly due to the following points. First, comprehensive statistical features, graph theoretic features and matrix factorization results were extracted from similarity information and known associations so that informative input features for the model could be obtained. Furthermore, because feature profiles made the most of similarity and known associations, EDTMDA could work for new diseases without known association information. Second, ensemble learning was designed to integrate multiple basic classifiers for more accurate prediction. In addition, feature dimensionality reduction with PCA could remove noise or redundancy to further improve prediction performance. Third, for the base classifier, the regression tree model with the arithmetic of Classification and Regression Tree (CART) was selected in our model, which was the binary tree with simple structure and could avoid the data fragmentation existing in multi-branching tree.
However, there were several limitations in our prediction model. To begin with, known miRNA-disease associations were inadequate (with only 2.86% of 189,585 miRNA-disease pairs being labeled) and increasing associations confirmed by experiments in the future would further improve model performance. Additionally, similarity calculation of miRNA and disease in this study may not be perfect and we expect more biological information would be incorporated into similarity measurement. Moreover, EDTMDA might cause bias to miRNAs which have more associated disease records. Finally, negative samples (miRNA-disease pairs without associations) were needed in our model. We randomly sampled some pairs without known associations as negative samples for model training. In order to reduce bias and improve prediction performance, multiple base classifiers were trained and integrated. Moreover, dimensionality reduction was employed for each base classifier to reduce noise and redundant information, which further improve performance of model. Actually, it is still difficult to obtain true negative samples (That is, miRNA-disease pairs show no evidence of association), because these true negative samples are scarcely reported in literature. We will make efforts to develop the new approach to identify reliable negative samples in the future.
|
10.1371/journal.pgen.1006373 | Amplification of TLO Mediator Subunit Genes Facilitate Filamentous Growth in Candida Spp. | Filamentous growth is a hallmark of C. albicans pathogenicity compared to less-virulent ascomycetes. A multitude of transcription factors regulate filamentous growth in response to specific environmental cues. Our work, however, suggests the evolutionary history of C. albicans that resulted in its filamentous growth plasticity may be tied to a change in the general transcription machinery rather than transcription factors and their specific targets. A key genomic difference between C. albicans and its less-virulent relatives, including its closest relative C. dubliniensis, is the unique expansion of the TLO (TeLOmere-associated) gene family in C. albicans. Individual Tlo proteins are fungal-specific subunits of Mediator, a large multi-subunit eukaryotic transcriptional co-activator complex. This amplification results in a large pool of ‘free,’ non-Mediator associated, Tlo protein present in C. albicans, but not in C. dubliniensis or other ascomycetes with attenuated virulence. We show that engineering a large ‘free’ pool of the C. dubliniensis Tlo2 (CdTlo2) protein in C. dubliniensis, through overexpression, results in a number of filamentation phenotypes typically associated only with C. albicans. The amplitude of these phenotypes is proportional to the amount of overexpressed CdTlo2 protein. Overexpression of other C. dubliniensis and C. albicans Tlo proteins do result in these phenotypes. Tlo proteins and their orthologs contain a Mediator interaction domain, and a potent transcriptional activation domain. Nuclear localization of the CdTlo2 activation domain, facilitated naturally by the Tlo Mediator binding domain or artificially through an appended nuclear localization signal, is sufficient for the CdTlo2 overexpression phenotypes. A C. albicans med3 null mutant causes multiple defects including the inability to localize Tlo proteins to the nucleus and reduced virulence in a murine systemic infection model. Our data supports a model in which the activation domain of ‘free’ Tlo protein competes with DNA bound transcription factors for targets that regulate key aspects of C. albicans cell physiology.
| The ascomycete fungus Candida albicans is a leading cause of hospital-acquired bloodstream infections in the United States. Due to limited anti-fungal drug options, there is an approximately 40% mortality rate and over 10,000 deaths per year associated with systemic C. albicans infections. It is unknown why C. albicans is the primary cause of systemic Candidiasis, versus related ascomycetes such as Candida dubliniensis. The genomes of C. albicans and C. dubliniensis are remarkably similar, yet C. dubliniensis has reduced virulence and exhibits less phenotypic plasticity. A striking genomic difference between the fungi is the amplification of the TLO (TeLOmere-associated) genes in C. albicans, which encode a fungal-specific subunit of the Mediator co-activator complex. Amplification results in a large pool of ‘free’ (non-Mediator associated) Tlo protein in C. albicans that is absent in C. dubliniensis. Engineering a large ‘free’ pool of Tlo protein in C. dubliniensis, through overexpression, results in phenotypes common in C. albicans, yet typically absent in C. dubliniensis. Tlo proteins contain a potent transcriptional activation domain. Nuclear localization of the Tlo activation domain is necessary and sufficient for the TLO overexpression phenotypes. This study provides a mechanistic explanation for how TLO amplification in C. albicans may enhance its virulence.
| It is estimated that there are between 2 to 5 million species of fungi on earth, of which only a small fraction cause infections in humans [1]. An even smaller fraction of these are capable of causing life-threatening infections. Nonetheless, opportunistic fungal infections have emerged as a major cause of human disease [2,3,4]. Cryptococcus, Candida, Aspergillus, and Pneumocystis species account for more than 90% of fungal-related deaths [2]. Candida albicans is the primary agent of invasive candidiasis [3]. C. albicans can switch from growth as a natural human commensal to a pathogen and, particularly if the host is immuno-compromised, cause life-threatening systemic infections that have limited treatment options [5,6,7,8,9,10]. It is not well understood, from an evolutionary standpoint, why C. albicans is more potent pathogen in humans than closely related fungi that have attenuated virulence. The genomes of the ascomycetes C. albicans and Candida dubliniensis are remarkably similar, with 96.3% of genes exhibiting >80% identity and 98% of genes being syntenic. This close phylogenetic relationship is contrasted by the observation that C. dubliniensis is far less pathogenic in a range of infection models and is a far less prevalent cause of systemic infections [11]. Consistent with this observation, the ability to change morphology or adapt to stress in response to environmental cues, which is critical to C. albicans virulence, is compromised in C. dubliniensis [11].
The adaptive transitions that underlie C. albicans virulence are driven by diverse transcriptional programs, which require the coordination of multiple sequence-specific DNA-bound transcription factors. There are two primary morphological transitions, relevant to virulence, which are interrelated and regulated in response to a range of conditions. First, C. albicans transitions between yeast and filamentous (pseudohyphal and hyphal) forms—a trait that is strongly associated with virulence [10]. Second, C. albicans switches from planktonic growth to the formation of highly recalcitrant surface-associated biofilms. Biofilm formation requires the ability to switch between yeast and filamentous growth, and the induction of other pathways involved in matrix production and drug resistance [12,13]. C. albicans is one of the fungi most commonly isolated from catheter-based biofilms [14,15,16,17]. In addition to morphological transitions, there are multiple coordinated responses to stress that help C. albicans adapt to host niches and cope with the immune response [18]. In addition to impacting virulence, C. albicans’ transcriptional plasticity also impacts its commensal lifestyle [19], which has most likely exerted significant selective pressure on its evolution. The origin of C. albicans’ morphological and transcriptional plasticity isn’t clear, but comparative biochemical and genetic studies utilizing C. albicans and C. dubliniensis may be able to shed light on this question.
Considerable effort has been spent on identifying C. albicans genes critical for morphological transitions [10,20], adhesion/biofilm formation [21,13] and stress responses [18], as well as characterizing the array of individual transcription factors that positively and negatively regulate these genes. Most of these transcription factors are important virulence factors [10,13,18], yet these transcription factors are conserved between C. albicans, and C. dubliniensis, as well as many other related fungi. Therefore, what property of the C. albicans regulatory machinery enables the transcriptional plasticity that underlies its virulence? The answer to this question is most likely multi-fold and may include the particular wiring of gene expression networks [20] and the functional properties of the genes (adhesins, etc.) regulated in C. albicans [22]. Rather than a particular gene or network, however, we hypothesize that amplification of a single component of the C. albicans general transcription machinery, the TLO (MED2) subunit of the Mediator complex, can affect transcriptional plasticity.
Mediator, a conserved eukaryotic ~25 subunit co-activator complex, is an intermediary between DNA-bound transcription factors and the general transcription machinery in all eukaryotes [23,24]. Based on work in C. albicans [25–27], C. dubliniensis [28], and S. cerevisiae, it is clear that some subunits modulate the transcription of specific subsets of genes [29,30], while others facilitate the transcription of virtually all genes [31]. About half of fungal Mediator subunits are encoded by essential genes, which typically have amino acid identity levels as great as 20–30% with their mammalian counterparts. The Tlo (named Med2 in the formal Mediator nomenclature) and Med3 subunits, however, are widely divergent among Ascomycetes, and have no metazoan orthologs that combine similar sequence, structure and function. Mediator subunits Tlo/Med2, Med3 and Med15 are encoded by non-essential genes and are held in the Tail module through mutual interactions [32,33,26,28]. Typically the protein subunits of Mediator, and the genes encoding them are present in a one to one ratio [34]. However, there are as many as 15 TLO paralogs encoded by the C. albicans genome, compared to two in C. dubliniensis and one in all other sequenced fungi [35,36].
Sequence analysis of the C. albicans TLOs (CaTLOs) divided the paralogs into α, β and γ clades [37], of which the α and β clades were expressed at vastly higher mRNA [37] and protein levels [26] than γ. The C. albicans strain used in our studies has one β clade Tlo and a roughly equally number of α and γ clade paralogs [37]. The two C. dubliniensis TLO (CdTLO) paralogs, CdTLO1 and CdTLO2, diverge from the C. albicans clades, and each other, primarily in their C-termini [28]. CdTLO1 is the primary TLO in C. dubliniensis, and is expressed at least 30X higher at the protein level [28] and ~50x higher at the mRNA level [35] than CdTLO2. The CaTlo α and β proteins are incorporated, with comparable affinity, into Mediator. Hence, there are several different pools of Mediator complex in C. albicans, each with a different Tlo protein [26]. Since each of the CaTlo α and β proteins are expressed in excess of other CaMediator subunits, there is also a large pool of ‘free’ CaTlo protein that is nuclear localized [37], not associated with Mediator [26]. There is no published data to support the idea that the ‘free’ Tlo protein is a stoichiometric member of an alternative complex. In contrast, just enough CdTlo1 is expressed in C. dubliniensis for it to be a stoichiometric component of CdMediator, leaving no detectable ‘free’ Tlo pool [28]. Null mutants of C. albicans, or C. dubliniensis, MED3, and C. dubliniensis TLO1/2 exhibit several phenotypes including decreased resistance to oxidative stress, inability to utilize galactose as a carbon source, and varied cellular and colony morphology phenotypes [26–28,35]. The highly acidic C-termini of the CaTlo α and β proteins, and the C-termini of the CdTlo proteins encode transcriptional activation domains (TADs) [38]. Our one-hybrid experiments showed that the Tlo C-termini were potent TADs that functioned independently of incorporation into the Mediator complex, a property encoded exclusively by the Tlo N-terminus [38]. Even though the sequence of CaTlo and CdTlo C-termini diverge substantially from the S. cerevisiae ortholog (Med2), the potent TAD is also present in the model yeast [38]. Removing these C-terminal TADs in C. dubliniensis and S. cerevisiae left a Mediator complex that was deficient in its transcriptional response to alternative carbon sources and certain stresses [38]. We hypothesized that the ‘free’ Tlo population impacts morphological plasticity in C. albicans through a mechanism that is dependent on its TAD.
Here we show that engineering a large ‘free’ pool of CdTlo2 protein in C. dubliniensis, through overexpression, results in a number of virulence associated filamentation phenotypes that are primarily associated with C. albicans. These phenotypes are specific to CdTLO2, versus other CaTLOs and CdTLO1, and their amplitude is proportional to the amount of overexpressed Tlo protein. Nuclear localization of the CdTlo2 TAD, facilitated naturally by the Mediator association domain or artificially through a nuclear localization signal, is sufficient for the C. dubliniensis Tlo overexpression phenotypes. Mutations that inhibit the nuclear localization of Tlo protein in C. albicans strongly reduce its virulence in a murine systemic infection model. This data suggests that the ‘free’ pool of Tlo protein may be competing with DNA-bound activators for binding sites on co-activators and co-repressors.
Two primary approaches were available to investigate the impact of a ‘free’, non-Mediator associated Tlo pool on virulence related phenotypes. The first approach, depletion of the ‘free’ Tlo pool in C. albicans, is the most challenging technically. Knocking out both copies of all, or any, of the 14 CaTLO paralogs using conventional gene replacement methods has proven to be difficult given the high rates of recombination between TLOs at these loci [39]. The second approach, which we have successfully utilized here, was to create a ‘free’ Tlo population in C. dubliniensis, where it doesn’t normally exist. CdTlo1 and C. albicans α-clade CaTlo proteins, the predominantly expressed Tlos in their respective organisms, were initially chosen for overexpression in C. dubliniensis. The strong TDH3 promoter (PTDH3) was used to give high levels of transcription [40] of the TLOs in C. dubliniensis. Overexpressed proteins were HA tagged on the C-terminus and integrated into the native CdTLO1 locus (S1 Fig) unless otherwise stated. Immunoblotting revealed that integration of either one or two copies of the CdTLO1 gene under the control of the PTDH3 promoter yielded only modest increases in CdTlo1p protein versus endogenous HA-tagged CdTlo1p, while integration of increasing copies of CaTLOα12 led to increasing amounts of the protein that were in excess of the endogenous CdTlo1p (Fig 1A). The lack of increase in CdTlo1p was not related to mRNA levels, as RT-qPCR revealed that the constructs led to substantial increases in mRNA as expected (Fig 1B). Plotting the ratio of fold change in protein versus fold change in mRNA (Fig 1C) showed that the increased amount of CdTlo1 protein plateaued at a level ~2-fold higher than endogenous, while CaTloα12 protein continued to increase with increasing mRNA levels. Co-overexpression of CdTLO1 and CaTLOα12 led to a decrease in CdTlo1p (S2 Fig) that suggested CaTloα12p was competing for a factor that stabilized CdTlo1p. Our previous work had shown that the α-clade CaTlo proteins were stable in a non-Mediator associated form in C. albicans [26]. We speculated that CdTlo1 protein might only be stable when associated with Mediator complex, and that CdTlo1 protein in excess of that amount was rapidly degraded. This idea was supported by the finding that deletion of MED3, which releases the Tlo subunit from the Tail module of Mediator [26,28], led to a large decrease of endogenous CdTlo1p in C. dubliniensis (Fig 1D) and no decrease in an endogenous α-clade TLO, CaTloα34p (Fig 1E), in C. albicans. RT-qPCR showed that the decreased CdTlo1p levels in the med3 null strain are not a result of decreased mRNA amounts (S3 Fig). The compromised stability of ‘free’ CdTlo1p is intrinsic to its sequence, rather than species specific, as overexpression of CdTLO1 under the ACT1 promoter (PACT1) in C. albicans shows the same low protein to mRNA ratio (S4 Fig) observed in C. dubliniensis (Fig 1). Treatment with cycloheximide led to the rapid degradation of non-Mediator associated CdTlo1p in C. dubliniensis, created by deleting MED3 in an endogenous TLO1-HA tagged strain, while ‘free’ CaTlo protein was stable in C. albicans (Fig 2 and S5 Fig). A similar difference in stability was observed when HA-tagged CdTLO1 and CaTLOα12 were overexpressed in C. albicans (S6 Fig). Treating the C. dubliniensis endogenous CdTLO1-HA med3 null strain with a proteasome inhibitor (MG132) prior to treatment with cycloheximide prevented the degradation of non-Mediator associated CdTlo1 protein (Fig 2). MG132 had a similar effect on the degradation of overexpressed CdTlo1-HA in C. albicans (S7 Fig). We conclude from this set of experiments that there is an intrinsic property of the CdTlo1 sequence that makes the non-Mediator associated protein subject to rapid proteasome-dependent degradation, hence preventing the accumulation of a large ‘free’ pool of the protein. This result was consistent with the inability of CdTLO1 overexpression in C. dubliniensis to confer any notable phenotypes (S1 Table).
Despite its relative stability in the ‘free’ form, overexpression of CaTLOα12 in C. dubliniensis also did not confer any notable phenotypes (S1 Table). We speculated that divergent sequences in the CaTlo proteins, compared to CdTlo1, might prevent them from impacting phenotypes in C. dubliniensis. Since we were unable to isolate fully intact Mediator complex from a Cdtlo null strain that expressed CaTLOα12 (S8 Fig), we speculated that inefficient Mediator interactions could prevent CaTloα12p from impacting phenotype in C. dubliniensis. Hence, to engineer a version of CdTlo1p that could be stably expressed in a ‘free’ form, and to gain insight into native CdTlo1p instability relative to CaTloα12p, we engineered a hybrid overexpression construct that combined sequences from CaTLOα12 and CdTLO1. To determine what C-terminal sequence of CaTloα12p was critical for its differing stability from CdTlo1 in the non-Mediator associated form, HA-tagged overexpression constructs were made in which increasing lengths of CaTloα12p C-terminal sequence was replaced by the corresponding C-terminal sequence from CdTlo1p (S9 Fig). Evaluating these hybrid proteins by comparing protein to mRNA ratios showed that ‘unstructured’ C-terminal CaTloα12p TAD was not required, but the predicted ‘Helix 3’ (PSIPRED[41]) in the N-terminus of CaTloα12p was required for its relative stability in the ‘free’ form in C. dubliniensis (S10 Fig) and C. albicans (S11 Fig). A similar approach with N-terminal sequences revealed that the predicted ‘Helix 1’ of CaTloα12 also was important to its relative stability when overexpressed in C. albicans (S12 and S13 Figs). Purifying the N- and C-terminal hybrid Tlo proteins with a 6His-3Flag tag from a C. dubliniensis overexpression strain showed that retention of predicted ‘Helix 2’ from CdTlo1 was necessary to ensure incorporation into Mediator (S14 Fig). Using this information, a hybrid protein, HyNT1Cp (Hybrid N (HyN)-CdTlo1 C-terminus (T1C)), was created that had the CdTlo1 TAD, was more stable in its ‘free’ form and, by virtue of the CdTlo1 ‘Helix 2,’ was able to be readily incorporated into C. dubliniensis Mediator (Fig 3). The N-terminus of HyNT1Cp is identical to the N-terminus of 12TH2p, which is stably incorporated into CdMediator (Fig 3A and 3B). The N-terminus of the Tlo proteins contains all the Mediator association properties of the Tlo proteins [38]. Increasing copy numbers of the HyNT1Cp overexpression cassette leads to increases in mRNA and protein equivalent to CaTloα12 (Fig 3C and 3D). These properties were further confirmed by demonstrating that co-overexpression of HyNT1Cp and CdTlo1 in C. dubliniensis led to decreased amounts of CdTlo1p (S15 Fig). HyNT1Cp was also able to complement the tlo null gal- phenotype [28] in C. dubliniensis (Fig 3E). Despite these ‘optimized’ properties, overexpression of HyNT1Cp did not confer any notable phenotypes in C. dubliniensis (S1 Table). Although the other C. dubliniensis TLO gene, CdTLO2, was not typically expressed at high levels in C. dubliniensis, its sequence diverged from CdTLO1 and the CaTLOs. Hence, we tested whether overexpression of CdTLO2 might impact C. dubliniensis phenotype.
Overexpression of CdTlo2 results in a stability profile that is increased compared to CdTlo1, but slightly less stable than CaTlos and HyNT1Cp (Fig 4). Most importantly, it shows that the typically weakly expressed CdTlo can accumulate in the ‘free’ form (Fig 4). Unlike what we had observed with CdTLO1 or CaTLOs, over expression of the 1X and 2X CdTLO2 constructs (S1 Fig) in an otherwise wild type C. dubliniensis background led to dramatic increases in filamentous growth in embedded agar, and agar invasion from colonies (Fig 5). These 1X and 2X CdTLO2 strains, when grown in liquid YPD media, do not exhibit any apparent morphological differences from the wild type C. dubliniensis strain (Fig 5). When PCR positive transformants were isolated from the integration of the 1X and 2X CdTLO2 OE constructs, two colony morphologies were observed–‘smooth’ and ‘super wrinkled’ (SW) (Fig 5). These SW colonies constituted ~25% of the 1X CdTLO2 OE and ~50% of the 2X CdTLO2 OE integration positive transformants. The wrinkled SW colony appearance could vary between transformants (Fig 5). In addition, both 1X and 2X CdTLO2 OE SW strains grew in a filamentous form in liquid YPD media, and demonstrated a greater degree of agar invasion and embedded filamentation than their non-SW counterparts (Fig 5). We have never observed conversion of the ‘smooth’ form to the ‘SW’ form once a 1X or 2X CdTLO2 OE strain adopted the ‘smooth’ morphology. We did, however, infrequently observe that the 1X or 2X CdTLO2 OE ‘SW’ strains adopt the smooth morphology, an event initially manifested as a sectored colony. qPCR of CdTLO2 in genomic DNA from 1X and 2X CdTLO2 OE ‘smooth’ and ‘SW’ colonies showed that there were multiple copies of CdTLO2 over expression cassettes in the ‘SW’ strain compared to the ‘smooth’ strains (S16 Fig). It appears that 4 total copies of HA-tagged CdTLO2 are sufficient to induce a SW phenotype and that additional copies amplify the phenotype (S16 Fig). Since there is no ploidy increase in Chromosome 7 (on which CdTLO1 is encoded) relative to other chromosomes in these strains (S17 Fig), we believe that the overexpression construct has been inserted in additional genomic locations as opposed to the amplification of Chromosome 7. The insertion location(s) for these extra CdTLO2 copies in the ‘SW’ strains is currently unknown, but does not appear to be the second CdTLO1 locus since we continue to detect CdTLO1 genomic DNA in these strains. We observed that increases in CdTLO2 OE copy number led to increased amounts of CdTLO2 mRNA and protein product, which correlated with an increasing amount of filamentation (Figs 5 and 6). A majority of the overexpressed CdTlo2 protein is in a non-Mediator associated form as shown by the ratio of CdTlo2 to Med1 protein in a cell extract versus the ratio of Tlo to Med1 in a purified Mediator sample (Fig 6D). The existence of this ‘free’ CdTlo2p pool is also supported by the observation that an anti-Med1 antibody is able to pull-down almost all of the Tlo subunit in a purified Mediator sample, but only a small fraction of the overexpressed CdTlo2 protein in an extract (Fig 6D). From these studies, it appears that a CdTLO2 OE cassette copy number of approximately four is sufficient to induce the super wrinkled phenotype (S16C Fig). The increased filamentation, as well as the emergence of the SW phenotype, were not specific to using the TDH3 promoter for CdTLO2 overexpression or to the Wü284 strain background. Using ACT1 and ENO1 promoters to overexpress CdTLO2 in Wü284 and two other C. dubliniensis strain backgrounds resulted in similar morphologies to the TDH3 promoter driven overexpression in Wü284 (S18 Fig). The filamentation phenotypes were specific to CdTLO2 overexpression, as similar overexpression of CdTLO1, CaTLOα12 or the stabilized hybrid construct, HyNT1C, did not result in agar invasion or filamentation in embedded agar (S19 Fig). We had previously shown that C. dubliniensis med3 and tlo1 null strains exhibited growth defects under a number of stress and alternative carbon source conditions [28]. Testing CdTLO2 OE strains for an effect on these phenotypes using plate-based assays, however, revealed only minor growth differences compared to WT (S20 Fig). Interestingly, colonies from the CdTLO2 and CdTLO2 (SW) overexpression strains grown with galactose as the sole carbon source showed increased filamentation compared to control strains (S20 Fig).
In addition to the stability of the ‘free’ form of CdTlo2p, the C-terminal CdTlo2p TAD was also required for overexpression phenotypes. Replacing the C-terminal activation domain of CdTlo2p with the CdTlo1p activation domain in our overexpression constructs resulted in the loss of the filamentation phenotype (Fig 7). Fusing the C-terminal CdTLO2 TAD to the N-terminus of the stabilized CdTlo1 hybrid protein, HyNT2Cp (Hybrid N (HyN)-CdTlo2 C-terminus (T2C)), in an overexpression construct resulted in similar, albeit weaker filamentation phenotypes to the CdTLO2 overexpression construct, while fusing this same activation domain to the intrinsically unstable CdTLO1 N-terminal domain did not (Fig 7). We have previously shown that the N-terminus of a Tlo protein was necessary and sufficient for incorporation of the subunit into Mediator through interactions with the Med3 subunit [38]. To test whether association with Mediator was important for the CdTLO2 overexpression phenotypes, we integrated the CdTLO2 overexpression construct into a med3 null C. dubliniensis background. The phenotypes associated with CdTLO2 overexpression, such as filamentation in embedded agar (Fig 8), were not observed upon CdTLO2 overexpression in a med3 null C. dubliniensis strain. Further insight into why Med3, and Mediator association, were critical for CdTLO2 overexpression phenotypes resulted from evaluating the nuclear localization of the Tlo proteins.
Fluorescence microscopy revealed that overexpressed GFP-tagged versions of CdTlo1p, CaTloα12p, and CdTlo2p were all localized to the nucleus in a wild type C. dubliniensis background (Fig 9). CdTlo1p, which is quickly degraded in the ‘free’ form (Fig 2), is cleanly localized to the nucleus, while CaTloα12p and CdTlo2p were focused in the nucleus with some remaining signal in the cytoplasm (Fig 9). Performing this same experiment in a med3Δ/Δ C. dubliniensis strain revealed that MED3 was required for nuclear localization of overexpressed CdTlo2p-GFP in C. dubliniensis (Fig 9). In the case of CdTlo1p-GFP and CaTloα12-GFP overexpression in C. dubliniensis, the absence of Med3 appears to decrease, but not completely abrogate, its nuclear localization. The absence of both CdTlo2p nuclear localization (Fig 9) and CdTLO2 overexpression phenotypes (Fig 8) in the med3Δ/Δ C. dubliniensis strain suggests that these two phenomena may be linked. From a structural standpoint, it is interesting to note that the nuclear localization of Tlo proteins and Med3 is mutually dependent as CdMed3 fails to localize to the nucleus in a tlo1Δ/Δ C. dubliniensis strain (S21 Fig). Earlier work has shown that the endogenous α and β clade CaTlos are nuclear localized [37]. Here, we again used fluorescence microscopy to determine whether this nuclear localization of the endogenous free population of CaTlos in C. albicans is dependent on subunits required for Mediator association. We have previously shown that the Med3 and Med15 subunits, but not the Med16 subunit, of the C. albicans Mediator Tail module are required for the CaTlo subunit to be incorporated into the complex [26]. The nuclear localization of the endogenous CaTlo proteins was dependent on Med3 and Med15, but not Med16 (Fig 10). In both C. dubliniensis and C. albicans there is a direct correlation between Mediator association and nuclear localization for the ‘free’ Tlo population. Deletion of C. albicans MED3, which is required for nuclear localization, also leads to a strain that has highly attenuated virulence in the murine model for disseminated candidiasis (Fig 11). We currently don’t know, however, whether it is the lack of CaTlo nuclear localization, the multiple morphological defects present in the C. albicans med3Δ/Δ strain [26], or the combination of the two that result in attenuated virulence in the med3 null strain.
Based on the CdTlo2 TAD requirement for CdTLO2 overexpression phenotypes (Fig 7), and the correlation between nuclear localization and these phenotypes (Figs 8 and 9), we hypothesized that appending a nuclear localization signal (NLS) to the CdTlo2 TAD would be sufficient to confer CdTLO2 overexpression phenotypes. Fusion of an NLS-GFP sequence to various Tlo ORF fragments resulted in a highly focused nuclear localization (Fig 12A), even in the absence of the Med3 subunit (Fig 13A). The NLS driven nuclear localization is far more efficient than Mediator-dependent localization, as these strains exhibited almost none of the residual cytoplasmic accumulation observed in CdTLO2-GFP overexpression strains (Fig 9). NLS driven nuclear localization of 1X overexpression constructs, expressing GFP-CdTLO2 and GFP-CdTLO2 TAD, gave potent filamentation phenotypes (Fig 12B) that resembled the CdTLO2 SW phenotypes (Fig 5). These phenotypes included the constitutive filamentous growth in liquid media. Unlike the CdTLO2 SW phenotypes observed earlier, the ‘SW-like’ phenotypes in the NLS-GFP-CdTLO2 strains were not accompanied by increased copy numbers of the NLS-GFP-CdTLO2 overexpression construct (S22 Fig). This finding suggests that although a larger amount of total CdTlo2 protein in the cell can amplify filamentation phenotypes, the most important factor is the amount in the nucleus. Nuclear localization of the GFP-CdTLO1 TAD gave a weak agar invasion phenotype, but still had no effect on filamentation in embedded agar (Fig 12). NLS driven nuclear localization of GFP alone, GFP-CaTloα12p, GFP-CaTloα12TAD, GFP-CdTlo1p had no effect on agar invasion or filamentation in embedded agar (Fig 12). Both the NLS and ‘non-NLS’ driven phenotypes were documented in strains that had a C-terminal 3HA epitope tag, which allowed for monitoring protein expression and stability. Repetition of the NLS and ‘non-NLS’ driven phenotypes in strains which overexpress CdTLO2 constructs lacking the 3HA tag, demonstrated that the epitope tag was not required for the effect (S23 Fig). We speculated that the previously observed deficiency of CdTLO2 overexpression phenotypes in a med3Δ/Δ C. dubliniensis strain (Fig 8) was caused by the lack of nuclear localization (Fig 9), rather than by an indirect effect of med3 deletion on transcription. We tested this idea using the NLS fusions to drive nuclear localization of GFP-CdTLO2 in a med3Δ/Δ C. dubliniensis strain. The embedded agar filamentation and agar invasion phenotypes that resulted from NLS-GFP-TLO overexpression in the wild type background (Fig 12) were recapitulated in the absence of MED3 (Fig 13). The dispensability of the Tail module for the CdTLO2-TAD driven overexpression phenotypes, under conditions in which the activation domain was localized to the nucleus via a dedicated NLS, was further demonstrated by the presence of these phenotypes in a C. dubliniensis strain that overexpressed NLS-GFP-CdTLO2-TAD in a tlo1Δ/Δ background (S24 Fig). Considered as a whole, these observations support a paradigm in which the Mediator Tail module is a vehicle to bring CdTlo2 into the nucleus so its TAD can promote greater sensitivity to induction of filamentation.
Comparative genomics of the closely related fungi C. albicans and C. dubliniensis [35] provided the foundation for mechanistic studies to elucidate the origin of C. albicans’ enhanced virulence and phenotypic plasticity [11] versus C. dubliniensis. The presence of a large ‘free’ Tlo population, and the presence of a mixture of Mediator complexes with different Tlo subunits [26] were both potential explanations for how TLO amplification in C. albicans, versus C. dubliniensis, might impact virulence gene expression. Two central considerations led us to pursue the large ‘free’ population as the primary mechanism by which TLO amplification could influence the virulence properties of C. albicans at the transcriptional level. First, the observation that the CaTLOs are themselves transcriptionally regulated by pathways that impact virulence [43,44] suggested that modulation of the size of this free pool could play a role in these processes. Second, our recent discovery that the Tlo proteins contained a potent TAD [38] suggested that the large nuclear pool of Tlo protein, which possessed no recognizable DNA binding domain, could impact gene expression by competing for the targets of sequence-specific DNA bound transcriptional activators. The sequestration of activator targets off chromatin by overexpression of a TAD, or ‘squelching’ as this phenomenon has been called, has been demonstrated in artificial [45,46], and a limited number of physiological [47] systems. The ‘free’ pool of Tlo TAD could impact the interactions between DNA-bound transcriptional activators and their co-activators targets [24,48]. They could also impact interactions with co-repressors that negatively regulate transcription by ‘masking’ the TAD of certain DNA-bound transcriptional activators [49] and preventing co-activator interactions. Our finding that engineering a large ‘free’ population of Tlo protein in C. dubliniensis, through overexpression of CdTLO2, enables filamentous growth provides a mechanistic explanation for how genomic differences between C. dubliniensis and C. albicans could manifest themselves at the phenotypic level.
Several key findings reported here, support the Tlo TAD competition model and provide the first direct evidence that the ‘free’ population of C. albicans Tlo proteins resulting from gene amplification could impact virulence related phenotypes in the pathogen. Although we have focused on filamentation in this report, the Tlo TAD competition model supported by our studies could provide for a general plasticity in gene expression that is typically associated with C. albicans virulence. This study has shown that there are three key properties for an overexpressed TLO gene to impact filamentous growth in C. dubliniensis. First, it must be stable in a non-Mediator associated form. Second, it must possess a transcriptional activation domain that has, presumably, evolved to interact with specific co-activator and co-repressor targets in its native species. And third, it must possess a mechanism to efficiently transport the TAD to the nucleus where it can interact with these targets.
The N-terminal sequence of the Tlos is highly conserved, yet we have found that the primary expressed Tlo protein in C. dubliniensis, CdTlo1p, possesses instability in the ‘free’ form relative to CaTloα12 and CdTlo2. We have narrowed this property down to sequences in the first and third Helix in the predicted structure of the CdTlo1 N-terminus. The finding that CdTlo1 degradation is proteasome dependent suggests it might be a target for ubiquitylation. We have not, however, been able to find a particular amino acid in the CdTlo1 N-terminus that confers this property or direct evidence for ubiquitylation. It is possible that this intrinsic instability of CdTlo1 may have prevented the amplification of the Tlo genes in C. dubliniensis. If there were a selective advantage for the accumulation of a ‘free’ Tlo pool, amplification of CdTLO1 would never have conferred this advantage. Since CdTlo2 expression is inherently very low [28,35], amplification of this copy would also not have conferred any selective advantage. The failure of our stabilized hybrid Tlo1 protein (HyNT1C), however, showed that stability alone was not enough to confer the overexpression phenotypes observed with CdTLO2.
Our earlier studies had shown that the C-termini of CaTlo α and β, the CdTlo, and the S. cerevisiae Med2 (Tlo ortholog) proteins all contain potent transcriptional activation domains [38]. Of the CaTloα12, CdTlo1 and CdTlo2 TADs, however, we found only the CdTlo2 TAD can drive filamentation phenotypes in C. dubliniensis. TADs are typically unstructured and highly modular, in that appending them to any variety of DNA binding domains will lead to the increased expression of a reporter gene placed next to a binding site for the particular DNA binding domain [50,51]. The inability of the CaTlo proteins, and CdTlo1 TAD (appended to either the N-termini of the stabilized hybrid Tlo1 protein (HyNT1C) or the inherently stable CdTlo2) to confer the overexpression phenotypes indicates that possessing potent activity in our one-hybrid assay in C. albicans and S. cerevisiae is not sufficient for promoting filamentation. The activation domains of CaTloα12, CdTlo1 and CdTlo2 all have activation potential that equals or exceeds the prototypical VP16 activation domain when measured in S. cerevisiae [38]. Of these, the CdTlo2p activation domain potential was highest. The activation domains of CaTloα12 and CdTlo1 showed a similar pattern in a C. albicans one-hybrid reporter system. These findings suggest that there are particular C. dubliniensis co-activators, or co-repressors, that strongly interact with the CdTlo2 TAD compared to the TADs of CdTlo1 and CaTloα12. We infer from this result, that a converse situation may take place in C. albicans in which there are specific targets of the CaTlo TADs that do not interact as strongly with the CdTlo2 TAD. This specificity will facilitate future efforts to identify these targets.
Our model for the competition of the ‘free’ Tlo population with DNA-bound transcription factors for co-activator and co-repressor targets requires that these interactions take place in the nucleus. Prior to this study, however, very little was known about how individual or groups of Mediator subunits are imported into the nucleus. Studies in S. cerevisiae have shown that the fungal Tail module subunits Med2 (Tlo), Med3 and Med15 can exist as a stable trimer, both within the context of Mediator [32] and independently [52]. Our studies show that there is substantial co-dependence among these subunits for their nuclear import as well. Previous studies, which have tried to interpret the consequence of deletion of Mediator subunits on the phenotype, work under the tacit assumption that the nuclear localization of the remaining subunits persists unchanged in these deletion mutants [30]. Our results show that, at least in some instances, this is clearly not the case and localization of additional subunits should be taken into account when interpreting these results. The inability of Tlo protein to get into the nucleus in the absence of interactions with Med3 (and Med15) subunits allowed us to perform incisive experiments on the importance of the Tlo TAD. Appending an NLS to the CdTlo2 TAD, in the absence of its Mediator associating N-terminal domain, showed that the CdTlo2 TAD was sufficient to drive the filamentation phenotypes, while the CaTloα12 TAD was not. The ability of NLS-GFP-CdTLO2, but not CdTLO2, overexpression to drive filamentation phenotypes in a med3Δ/Δ C. dubliniensis strain demonstrates nuclear localization of the Tlo protein is necessary for these phenotypes. The CdTlo1 TAD appeared to have a weak agar invasion phenotype when fused to the NLS-GFP construct, even though the filamentation phenotypes were absent when the CdTlo1 TAD was fused to the CdTlo2 or HyNT1C N-termini. This result, the fluorescence microscopy studies, as well as the fact that the NLS-GFP-CdTLO2(TAD) overexpression phenotypes mimicked the phenotypes achieved at the highest levels of CdTlo2 overexpression (the SW phenotypes), suggests that high levels of nuclear localization are the key driver of overexpression phenotypes. The molecular origin of the CdTLO2 ‘SW’ phenotypes appears to originate from gene amplification without increase in chromosome 7 ploidy, but it does not appear to include increased nuclear localization of CdTlo2 per se (S25 Fig). The finding that the NLS-GFP-CdTLO2 strains generate a phenotype similar to the ‘SW’ strains suggests that this phenomenon is connected most directly to the increase in CdTLO2 expression rather than the location of the second insertion. Although both NLS-GFP-CdTLO2 and NLS-GFP-CdTLO2TAD overexpression resulted in filamentation phenotypes, the phenotypes differed from each other. Our earlier studies generally showed that full length Tlo/Med2 constructs had a weaker activation potential than their TAD by itself [38]. It is consistent with our model that variation of the activation potential could lead to varied phenotypes.
We speculate that the mechanism used by the ‘free’ Tlo protein in C. albicans evolved from properties that were inherently important for the Tlo/Med2 subunit in the context of the intact Mediator complex. Our previous studies of S. cerevisiae and C. dubliniensis Mediator showed that the presence of transcriptional activation domains in Tail module subunits of Mediator were important for the Tail module to facilitate the Mediator dependent induction of certain genes [38]. When the gene encoding this subunit became amplified in C. albicans, Mediator membership conferred two critical properties, a TAD and nuclear localization, necessary to impact the transcriptional programing of the cell. The mechanism by which Mediator membership facilitates the accumulation of a pool of ‘free’ Tlo protein is still a black box. Exchange of Mediator associated Tlo subunits is likely to be required for one molecule of Mediator to bring in multiple molecules of Tlo protein and accumulate a ‘free’ Tlo population. Since the N-termini of the C. albicans paralogs are nearly identical, it is likely that the ancestral C. albicans Tlo protein also possessed the property of being stable outside the context of association with Mediator prior to amplification. The ancestral C. albicans Tlo Mediator subunit was primed to take on its new role in its amplified form. It is unclear, however, what evolutionary history led to C. dubliniensis having two Tlo paralogs versus one in all other sequenced fungi besides C. albicans.
The results of our study beg the question whether TLO amplification represents a special case or could overexpression/amplification of Mediator subunits and/or TAD containing proteins represent a broader regulatory mechanism in health and disease. Amplification dependent and independent overexpression of the human Mediator Cdk8 kinase subunit, or other members of the Cdk8 module, occurs in a large percentage (>50%) of colon, breast, and gastric cancers and is proposed to act as an oncogene [53]. An additional Mediator subunit, S. cerevisiae Med3 [38], contains a TAD. Although no amplification dependent overexpression of Tlos, or other TAD containing proteins, had been identified in other fungal pathogens, it is still possible that future studies could find that amplification independent overexpression of Tlo/Med2 or Med3 orthologs could contribute to their virulence.
The C. dubliniensis and C. albicans strains used in this study are listed in the S2 and S3 Tables respectively. Each C. dubliniensis strain over-expressing TLO variants, listed in the S2 Table, was generated by transforming the parental strain with the indicated integrating plasmid by electroporation. The plasmids and the restriction enzymes used to release the integrative DNA cassettes from each plasmid are listed in the S4 Table. The sequences of the primers used to verify the correct insertion of the cassette at the TLO1 genomic locus are listed in the S5 Table. Each C. albicans strain overexpressing a TLO variant, listed in the S3 Table, was generated by transforming the parental strain with the indicated integrating plasmid by electroporation. The plasmids and the restriction enzymes used to release the integrative DNA cassettes from each plasmid are listed in the S4 Table. Their appropriate insertion at the RPS10 locus was confirmed by the primers pairs, whose sequences are listed in the S5 Table. Additional strains in the S2 and S3 Tables were created by introducing certain PCR products into the genomes of C. dubliniensis or C. albicans. Specifically, DNA cassettes amplified by ZL422/ZL423 from pFA-3HA-SAT1 [26] was used to tag endogenous TLO1 in Cdmed3Δ/Δ to generate yLM308. C-terminal 3XHA tagging of endogenous TLOα12 in C. albicans SN152 (resulting in yLM388), TLOα34 in C. albicans SN152 (resulting in yLM391) and C. albicans med3Δ/Δ (resulting in yLM392), were performed by transforming the strains with PCR product amplified from pFA-3HA-SAT1 by KPP035/KPP037 (targeting TLOα12) or KPP035/KPP041 (targeting TLOα34) accordingly. Deleting one or both alleles of MED3, and complementing the med3 null mutant in SN152 was achieved as described [26].
If not otherwise specified, C. dubliniensis and C. albicans strains were grown in YPD (2% Bacto Peptone (BD), 1% Bacto Yeast Extract (BD) and 2% Glucose) liquid media or maintained on YPD agar plate at 30°C. All the YP-based media used in this study were supplemented with 100 μM uridine. Steady-state protein levels of overexpressed TLO variants were compared using mid-log phase cells grown in YPD liquid culture.
For confocal microscopy, mid-log phase cells grown in YPD media were used for Blankophor staining (0.2% (wt/vol) Blankophor; 5 minutes at room temperature). To monitor GFP-tagged protein localization, mid-log phase cells, grown in SC media (6.7 g/L Difco YNB without amino acids (BD), 2 g/L Drop-out Mix Synthetic without uracil (US Biological), 2% Glucose, 200 μM uridine), was stained with 100 μg/mL of Hoechst 33342 with ~30 minutes incubation at room temperature in the dark without agitation before imaged by confocal microscopy with a Nikon A1 microscope.
To assess protein stability, overnight cultures were diluted to OD600 = 0.2 in fresh YPD media and grown for 2–3 cell divisions. Cycloheximide (10 mg/mL freshly dissolved in YPD) was added to a final concentration of 2 mg/mL [40] after saving an aliquot as the ‘0 min’ sample. Samples were collected at each indicated time point, washed and flash-frozen in liquid nitrogen.
MG132 inhibition of the proteasome degradation pathway in C. dubliniensis and C. albicans was performed following the previously described protocol established in S. cerevisiae [54]. Specifically, the indicated Candida strains were grown overnight in synthetic media (1.7 g/L YNB without ammonium sulfate, 1 g/L L-proline, 2% glucose, 200 μM uridine, 20 mg/L L-Histidine, 20 mg/L L-Arginine, 100 mg/L L-Leucine), diluted in the same media to OD600 = 0.2, and further grown for ~2 cell divisions. Cultures were supplemented with SDS to a final concentration of 0.003%, grown for 4 hours, and split into two for addition of MG132 (42 mM in DMSO, APExBIO) to a final concentration of 40 μM, or an equal volume of DMSO. After treatment for 45–60 minutes, cells were pelleted by centrifugation and directly flash-frozen in liquid nitrogen without any washing step, or further treated with cycloheximide (dissolved in the specified synthetic media instead of in YPD) as described above to monitor protein half life.
C. dubliniensis strains were grown overnight in YPD. After washing, the cells were diluted to 3X106, 3X105 3X104 and 3X103 cells/mL and spotted on YP (2% peptone, 1% yeast extract, 100μM uridine)-based 2% agar plates. Plates were incubated at 30°C if not otherwise specified.
To assess agar invasive growth 200 μL (~1 cell/μL) diluted overnight culture of each strain was spread on an YPD agar plate. After incubation at 30°C for 4 days, representative colony morphologies were imaged under a Nikon SMZ1500 stereoscope. Surface cells were removed by washing the plates under distilled water, and the agar-invasive growth was imaged directly or after cross-sectioning an agar slice.
Embedded filamentation assays were performed as described [55] with modifications. Specifically, cells from a fresh colony were grown overnight in liquid YPD media, diluted into 3 mL fresh media (~1000 cells/mL), and further grown for 4 hours at 30°C. 100 μL of this culture was mixed with ~25 mL of lukewarm (~45°C) YPS (2% Bacto Peptone (BD), 1% Bacto Yeast Extract (BD), 2% sucrose, 2% Bacto Agar (BD), 0.1 mM uridine) media and poured into a Petri dish. Plates were incubated at 30°C and the embedded colonies with representative morphology were imaged by Nikon SMZ1500 stereoscope at 48h and 72h. Images representing the embedded hyphae development at different time points are not necessarily taken from the same colony within the agar.
Affinity purification of C-terminal 6His3FLAG tagged Tlo protein variants to test for incorporation into intact CdMediator complex was performed as described previously [26], except the TALON step was omitted. For pull down experiments, a crude cell lysate of yLM416 was fractioned by Heparin Sepharose and the fraction which enriched with CdTlo2 and Med1 was incubated with Protein G Dynabeads which have been pre-coated either with anti-Med1 antibody or a non-related anti-HA antibody (Santa Cruz, F-7) for 2 hours at 4°C. Mediator complex purified though 6His3FLAG-tagged Tlo1 (as prepared [38]) was used as the standard for subunit stoichiometry and treated by the same depletion process.
To monitor the levels of Tlo variants in whole cell lysates, samples were prepared from frozen cell pellets (~4–5 OD of cells) following the ESB method [56], resolved by 10% SDS-PAGE and then either stained with coomassie blue or blotted by the indicated antibody [anti-HA (Roche, 3F10) or anti-FLAG (Sigma, F7425)] as described previously [57]. Western blot signals of given samples were measured by ImageQuant (Molecular Dynamics) and normalized to their relative total protein concentration, which was quantified by the total coomassie staining signals in the same gel area cross the samples (measured by UN-SCAN-IT (Silk Scientific)). Ca/CdMed1p antibody was the same rabbit polyclonal antibody generated against recombinant C. albicans Med1p in [26]. This antibody showed cross-reactivity to C. dubliniensis Med1p was used to monitor CdMed1 in semi-purified or purified CdMediator samples.
RNA samples were prepared from frozen cell pellets collected from mid-log phase cultures in YPD, and reverse-transcribed as described previously [27]. qPCR was performed using ‘Relative Standard Curve’ method (StepOne, Life Technologies). Relative Ca/CdACT1 level, quantified by AZq026/AZq027, was used as the internal reference between samples. TEF1 level (measured by ZL386/ZL387), instead of ACT1, was used as the reference when TLO2-3HA mRNA abundance was compared between C. dubliniensis strains with ‘super-wrinkled’ morphology versus their smooth counterparts, because ACT1 is reported to be differentially expressed in hyphae and yeast state [58,59]. mRNA levels of C-terminal 3XHA tagged TLO variants driven by an overexpression promoter, or the endogenous promoter, were measured by using ZL313/ZL201. When TLO variants were over-expressed in the non-tagged form, gene specific primers were used to evaluate their expression level: ZL190/191 for CdTLO1, ZL315/ZL316 for CaTLOα12, ZL190/ZL318 for HyNT1C and ZL426/ZL427 for CdTLO2.
A qPCR-based ploidy analysis for C. dubliniensis strains was modified from the method developed in C. albicans [60]. Genomic DNA was extracted from strains (yLM339, yLM343, yLM344, yLM345, yLM347 and yLM367) grown in 5 mL YPD liquid media over-night, 25 ng of which was quantified by qPCR using specific primer pairs to determine the copy number of nine representative genomic loci with the parental wild type strain Wü284 as the reference. Details are described in Supplemental Material and Methods section.
TLO2 copy number in a given strain was determined by qPCR using yLM339, yLM344, and Wü284 as reference strains as described in details in Supplemental Material and Methods.
The three day murine intravenous challenge model was used to assay fungal virulence as previously described [61]. Immunocompetent female BALB/c mice (6–8 weeks; Harlan, UK) were randomly assigned to groups of 6 and housed separately in individually ventilated cages. Group size was determined from power analyses using data obtained from previous experiments using the same infection model and parental fungal strain. Group size (n = 6) is the minimum groups size required to determine statistically significant differences in the parameters measured, where P ≤0.05, Power = 0.8. Food and water were provided ad libitum. C. albicans was grown overnight in NGY medium at 30°C with shaking. Fungal cells were harvested, washed, resuspended in sterile saline, and cells were counted. Approximately 1x106 cells were injected into each mouse via the lateral tail vein. The mice were monitored and weighed daily. At 72 h post-infection, mice were weighed, killed humanely by cervical dislocation, and their kidneys removed aseptically for determination of fungal burdens. Virulence outcome scores were determined by assessing renal fungal burden and percentage weight change at 72 h using the formula: outcome score = log (renal CFU g-1)—(0.5 × percentage weight change) [61]. Results were compared using Kruskal-Wallis comparison across all data sets and Mann-Whitney U tests for pair-wise comparisons using IBM SPSS (version 23) * P ≤0.05; ** P ≤0.01; *** P ≤0.001.
Animal experimentation was performed under UK Home Office Project license 60/4135, which was approved by the UK Home Office and by the University of Aberdeen Animal Welfare and Ethical Review Body (AWERB). All work conformed to European Directive 2010/63/EU.
During the colonization and infection studies, animals were monitored carefully for signs of illness and distress. Suffering was minimized by expert handling. Animals were monitored for changes in condition at least twice per day, and were weighed once per day. If animals had shown signs of severe illness (e.g. ruffled coat, hunched posture, unwillingness to move and 20% loss of initial body weight) euthanasia would have been performed by cervical dislocation. During these studies there were no animals were euthanized prior to the 72 h sampling time point. Analgesia and anesthesia were not employed in this study.
|
10.1371/journal.ppat.1000604 | Lethal Influenza Virus Infection in Macaques Is Associated with Early Dysregulation of Inflammatory Related Genes | The enormous toll on human life during the 1918–1919 Spanish influenza pandemic is a constant reminder of the potential lethality of influenza viruses. With the declaration by the World Health Organization of a new H1N1 influenza virus pandemic, and with continued human cases of highly pathogenic H5N1 avian influenza virus infection, a better understanding of the host response to highly pathogenic influenza viruses is essential. To this end, we compared pathology and global gene expression profiles in bronchial tissue from macaques infected with either the reconstructed 1918 pandemic virus or the highly pathogenic avian H5N1 virus A/Vietnam/1203/04. Severe pathology was observed in respiratory tissues from 1918 virus-infected animals as early as 12 hours after infection, and pathology steadily increased at later time points. Although tissues from animals infected with A/Vietnam/1203/04 also showed clear signs of pathology early on, less pathology was observed at later time points, and there was evidence of tissue repair. Global transcriptional profiles revealed that specific groups of genes associated with inflammation and cell death were up-regulated in bronchial tissues from animals infected with the 1918 virus but down-regulated in animals infected with A/Vietnam/1203/04. Importantly, the 1918 virus up-regulated key components of the inflammasome, NLRP3 and IL-1β, whereas these genes were down-regulated by A/Vietnam/1203/04 early after infection. TUNEL assays revealed that both viruses elicited an apoptotic response in lungs and bronchi, although the response occurred earlier during 1918 virus infection. Our findings suggest that the severity of disease in 1918 virus-infected macaques is a consequence of the early up-regulation of cell death and inflammatory related genes, in which additive or synergistic effects likely dictate the severity of tissue damage.
| The world is currently experiencing a pandemic caused by a new type of H1N1 influenza virus that originated in swine. Although it is too early to tell how virulent this new virus may be, some influenza viruses can cause severe disease. The 1918 pandemic, also caused by an H1N1 virus, resulted in over 50 million deaths worldwide. Highly pathogenic avian H5N1 influenza viruses are circulating in several parts of the world, and although human infection has been rare, the virus often causes a lethal illness. To determine whether highly pathogenic influenza viruses cause disease by similar means, we used a macaque infection model to study the host response to the reconstructed 1918 virus and an avian H5N1 isolate known as VN/1203. We found that although both viruses caused severe disease, infection with the 1918 virus was more likely to result in death, whereas animals infected with VN/1203 recovered from infection. In animals infected with the 1918 virus, there was increased expression of genes associated with inflammation, and these genes were turned on as early as 12 hours after infection. Drugs that limit the early inflammatory response may therefore be of benefit in treating infections caused by highly pathogenic influenza viruses.
| Influenza virus causes over 36,000 deaths [1] and 1.68 million hospitalizations in the United States every year [2]. However, the potential of this virus to disrupt society is best evidenced by the 1918–1919 influenza pandemic, which resulted in over 50 million deaths worldwide. The circulation of highly pathogenic avian H5N1 strains in Asia, and reports of their sporadic human-to-human transmission [3],[4], has raised concern over the potential for a deadly new influenza pandemic. Presently, a novel swine-origin influenza A (H1N1) virus (S-OIV) is spreading rapidly among humans, presenting the greatest threat since the emergence of influenza A (H3N2) virus in 1968 [5]. Although the World Health Organization has recently declared the first pandemic of the new millennium caused by this new S-OIV, initial analyses suggest that the clinical severity associated with the H1N1 virus is less than that seen in 1918 [6]. Nevertheless, preparedness for the current as well as future pandemics will require a better understanding of the host response to influenza virus infection, including mechanisms of viral pathogenesis and interactions with the host machinery, which will facilitate efforts to develop safe and effective therapeutics and vaccines.
Traditionally, studies of influenza virus pathogenesis compare viruses of high and low pathogenicity [7],[8]; nonetheless, comparisons of highly pathogenic strains are necessary to determine commonalities and differences in the host response and to develop effective countermeasures. Results from independent studies using macaque infection models suggest there are differences in lethality and disease progression between the 1918 and H5N1 viruses [8],[9]. In particular, infection with the 1918 virus leads to severe disease and death, whereas animals infected with VN/1203/04, an H5N1 avian virus, typically recover from infection. We have previously shown that an atypical host innate immune response may contribute to the severity of disease and fatal outcome in 1918-virus infected macaques [8]. Additional studies have suggested that early events during infection may be critical for disease outcome [10]. Microarray studies of H1N1 (A/Texas/36/91) influenza virus-infected pigtailed macaques (Macaca nemestrina) [11],[12] indicate that activation of inflammatory cells and apoptotic pathways correlates with tissue damage during infection.
In the present study, we examined two highly pathogenic influenza viruses, the fully reconstructed human H1N1 “Spanish flu” 1918 virus (1918) [8] and the highly pathogenic avian H5N1 A/Vietnam/1203/04 virus (VN/1203) in macaques (Macaca fascicularis), an established model for influenza virus infection [13],[14]. The Influenza A genome consists of eight gene segments designated PB2, PB1, PA, HA, NP, NA, M, and NS. The identities of these genes between 1918 and VN/1203 virus are 84, 84, 86, 63, 86, 82, 90, and 89%, respectively. We found specific differences in gene expression in the bronchus of animals infected with the 1918 virus compared with those infected with VN/1203, with the largest differences in expression occurring at 24 h after infection. Biochemical assays showed that both viruses elicited an apoptotic response in lungs and bronchi, although the response occurred earlier during 1918 virus infection. These results suggest that the differential regulation of genes associated with specific biological functions, including cell death and inflammation, are associated with increased severity of disease caused by the 1918 virus.
All animal experiments were performed under an approved animal-use document and according to the guidelines of the Canadian Council on Animal Care.
The highly pathogenic H5N1 virus used in this study (A/Vietnam/1203/04) was kindly provided by the Centers for Disease Control and Prevention. Genes of the 1918 virus (GenBank DQ208309, DQ208310, DQ208311, AF117241, AY744935, AF250356, AY130766, AF233238) were constructed with the 5′ and 3′ non-coding sequences of A/WSN/33 (H1N1) and cloned into plasmid vector pPolI, as previously described [15]. The 1918 virus was generated by reverse genetics [16], and titered stocks were prepared [15], as previously described. Forty-eight hours post-transfection, viruses were harvested and used to inoculate MDCK cells for the production of stock viruses. Eight genes of each transfectant virus were partially sequenced to confirm the origin of the gene.
The macaques (Macaca fascicularis) used in this study were inoculated by intratracheal (4 ml), intranasal (0.5 ml/nostril), intraocular (0.5 ml/eye) and oral (1 ml) routes with a suspension containing 106 plaque-forming units (PFU) per ml for a total infectious dose by all routes of 7×106 PFU. Six animals received the 1918 virus and were euthanized at 12, 24, or 48 h post infection (2 animals per time point). Eleven animals received the A/Vietnam/1203/04 virus and were euthanized at 12, 24, or 48 h or at 3 or 6 days post-infection (2 animals per time point). A single animal infected with A/Vietnam/1203/04 was euthanized at day 16. Viral titers in bronchus were quantified by standard plaque assay in MDCK cells. Tissue samples were placed in RNA Later (Ambion) for subsequent RNA extraction (Qiagen RNA Later kit).
The tissues were fixed in 10% neutralized phosphate-buffered formalin. Fixed tissues were dehydrated, embedded in paraffin, and cut into 5-µm-thick sections and then stained with standard haematoxylin and eosin. For viral antigen detection, sections were processed for immunostaining by the two-step dextran polymer method (DAKO), with a rabbit polyclonal antibody to A/WSN/33 (H1N1) used as the primary antibody.
Tissue homogenates were prepared in minimal essential medium supplemented with 0.1% bovine serum albumin and antibiotics (MEM/BSA) by using a bead mill homogenizer (Qiagen Tissuelyser) at 30 Hz for 5 min. Tissue debris was pelleted by centrifugation (2000×g, 5 min) and virus titers were determined in 10-fold serial dilutions by standard plaque assay on Madin Darby canine kidney (MDCK) cells in the presence of 1 µg/ml tosyl phenylalanyl chloromethyl ketone (TPCK)-treated trypsin, in duplicate for each dilution. Virus in swabs was similarly determined in undiluted and serial 10-fold dilutions of the swab suspension medium.
Separate microarrays were run for each experimental sample (one sample per animal and two animals per time point). Equal masses of total RNA isolated from bronchi collected from infected macaques were amplified with a Low RNA Input Linear Amplification Kit (Agilent Technology) according to the manufacturer's instructions. Global gene expression in infected bronchi was compared to pooled RNA prepared from equal masses of total RNA from bronchi tissue of six uninfected macaques. Probe labeling and microarray slide hybridization were performed as described elsewhere [17] with custom rhesus macaque (Macaca mulatta) oligonucleotide microarrays containing 22,000 rhesus probes corresponding to 18,000 unique rhesus genes (designed in collaboration with Agilent Technologies). Raw microarray image files were processed using Feature Extraction 8.1 software (Agilent Technologies) and entered into a custom-designed relational database (Expression Array Manager). We processed and analyzed the expression data using the limma package for the R programming environment [18]. Background correction was performed by using limma's “normexp” method, which ensures that there were no missing or negative corrected intensities. An offset of 50 was used for both channels to damp down the variability of the log-ratios for low-intensity spots. The resulting log-ratios were normalized by using ‘loess” methods. Between –array normalization was achieved by using limma's “Aquantile” method, which does quantile normalization of mean probe intensity values. The normalized expression data were analyzed together by using linear model methods. A batch factor was included in the linear model to remove potential batch effects. Differential expression was assessed by using moderated t-statistics. Primary data is available at http://viromics.washington.edu in accordance with proposed MIAME standards.
Functional and network analysis of statistically significant gene expression changes was performed with Ingenuity Pathways Analysis 7.1 (Ingenuity Systems). Genes from the data set that met the twofold (P<0.01) change cutoff and were associated with biological functions in the Ingenuity Pathways Knowledge Base were considered for analysis. For all analyses, Fisher's exact test was used to calculate a P-value determining the probability that each biological function assigned to that data set was due to chance alone. In the functional network shown in this paper, genes are represented as nodes, and the biological relationship between two nodes is represented as an edge (line). All edges are supported by at least one published reference or from canonical information stored in the Ingenuity Pathways Knowledge Base. For these analyses, Fisher's exact test was used to calculate a P-value determining the probability that each biological function and/or disease assigned to that data set was due to chance alone.
RT-PCR was performed to validate cellular gene expression changes as detected with microarrays. The QuantiTect reverse transcription kit (Qiagen Inc., Valencia, CA) was used to generate cDNA. qRT-PCR was run on the ABI 7500 PCR system, using TaqMan chemistry (Applied Biosystems, Foster City, CA). Gene expression assays specific to Rhesus cellular genes were purchased from Applied Biosystems. Differences in gene expression are represented as Log10RQ relative to a calibrator and normalized to a reference, using the 2−ΔΔCt method [19].
Lung and bronchus tissues were fixed in 10% neutralized phosphate-buffered formalin. Fixed tissues were dehydrated, embedded in paraffin, cut into 5 µm-thick sections, and then processed for TUNEL assay (X2044K2. ApoMark™ DNA Fragmentation Detection Kit. The Exalpha Biologicals, Inc., Maymard, MA). Briefly, the sections were rehydrated and permeabilized using Proteinase K solution. Endogenous peroxidases were inactivated with 3% hydrogen peroxide in methanol. After terminal deoxynucleotidyl transferase (TdT) reaction, biotinylated dUTP were visualized using HRP-conjugated streptavidin and 3.3′-Diamino benzidine (DAB). The sections were counterstained with methyl green solution.
All in vitro and in vivo procedures with the 1918 and VN/1203 influenza viruses were performed in the biosafety level 4 facility of the National Microbiology Laboratory of the Public Health Agency, Canada. Prior to removal from BSL4, all specimens were inactivated using Standard Operating Protocols and verified to be noninfectious.
We previously used a macaque infection model to examine the host response to the reconstructed 1918 virus at 3, 6, and 8 days after infection [8]. That study revealed that the 1918 virus caused a severe respiratory infection that culminated in acute respiratory failure and a fatal outcome. In addition, our histopathology and gene expression analyses suggested that events occurring well before the 3-day time point may have a critical influence on the course of disease progression. Furthermore, Baskin et al. [9] compared the host response to VN/1203 and A/Texas/36/91 containing two (HA and NA) or three (HA, NA and NS) genes from the 1918 pandemic virus in a macaque model at 1, 2, 4 and 7 days after infection. That study indicated that tissue damage induced by VN/1203 was the result of multiple factors such as an excessive and sustained type I interferon response and the deregulation of adaptive responses. In the present study, we sought to evaluate the host response at earlier time points after infection and to directly compare the host transcriptional response to these two highly pathogenic influenza viruses.
Animals infected with the 1918 virus or with VN/1203 were euthanized at 12, 24, or 48 h post-infection (p.i.). An additional group of animals infected with VN/1203 were euthanized at 3, 6, or 16 days p.i. Histopathology revealed that at 12 h p.i., the 1918 and VN/1203 viruses had already begun to cause pathology, and viral antigen was detected along the terminal and respiratory bronchiole (Figure S1). By 24 h p.i., antigen detection and severity of disease were more robust in the 1918-virus infected macaques, and the peribronchiolar alveoli showed severe alveolitis, edema, and hemorrhage (Figure 1, A). Additionally, the virus was widely distributed in bronchi and alveoli (Figure 1, C, E). This phenotype was lacking in macaques infected with VN/1203 (Figure 1, B, D, F). By 48 h p.i., the histopathology associated with 1918 and VN/1203 infection was similar, with clear alveolitis and alveolar edema around the bronchiole. Although the number of cells positive for viral antigen decreased in the edematous lesion, positive cells were still detected at the edematous lesion boundary (Figure S2). At later time points (days 3, 6, and 16), cells positive for viral antigen decreased in number in VN/1203-infected animals and by day 16 the lung showed prominent peribronchiolar lymph follicle development and tissue regeneration (Figure 2).
Analysis of viral titer revealed that at 12 h p.i., the titer of the 1918 virus appeared somewhat higher than the VN/1203 virus; whereas at 24 h p.i., titers for the two viruses were more similar (Figure 3). Remarkably, there was a decrease in viral titer for both viruses at the 48-h time point. Whereas we previously observed that the titer of the 1918 virus increased steadily at later time points (3, 6, and 8 days post infection) [8], VN/1203 titers did not increase after the 24-h time point and remained essentially constant at 3 and 6 days post infection (Figure 3). Thus, although 1918 and VN/1203 titers were similar at the 24-h time point, the 1918 virus had already caused greater pathology, which continued to worsen over time. In contrast, VN/1203 titers began to decrease after 24 h, and pathology began to diminish. These results suggest that macaques were able to mount a successful initial nonspecific innate response to both viruses as demonstrated by the similar decrease in titers during the 24–48 hour period. However, the 1918 virus appears to be able to overcome the host response and replicate at high levels [8], whereas VN/1203 replication was limited by the host response.
We used microarray analysis to determine whether the host transcriptional response to infection would provide insight into the differences in pathology and disease outcome observed for the 1918 and VN/1203 viruses. Our analyses revealed both similarities and differences in the transcriptional response to infection. For example, we found that the expression of type I interferon stimulated genes was up-regulated in response to both viruses as early as the 12-h time point, and the expression of these genes remained elevated at 24 and 48 h p.i. (Figure 4A). However, the 1918 virus elicited a stronger activation of these genes at 12 h p.i., but by 24 and 48 h p.i. this activation was higher with the VN/1203 virus (Table S1). ISG15 was one of the genes highly activated during infection with both viruses. It has been shown that ISG15 has antiviral activities against HIV-1 and Ebola [20],[21],[22], Sindbis [23],[24] and Influenza [25] viruses. Furthermore, ISG15 targets a wide variety of cellular pathways [26],[27],[28],[29],[30]. Consequently depending on the conjugation of ISG15, this could alter the host antiviral response or may impair viral replication.
We also analyzed the expression of selected chemokine and cytokine genes and again found a similar pattern of expression with a small transient difference at 24 h p.i. (Figure 4B). During our analysis we found that CXCL10 and CXCL11 were among the most highly induced chemokine genes during infection with both viruses. This illustrates a scenario where T cells are tethered to sites of inflammation at very early stages of infection [31]. A detailed listing of these genes and their change in expression can be found in Table S2. These results are consistent with our previous findings that both viruses elicit transcriptional induction of interferon- and cytokine-associated gene expression [8],[9]. However, even though the present direct comparison of the host response to these viruses indicates that the 1918 virus elicits a somewhat stronger induction of interferon-stimulated gene expression, it is unlikely that differences in this response alone would have a major impact on disease outcome.
Our analyses also revealed groups of genes that were regulated differently by the two viruses. T-test analysis of individual time points indicated that there were 55, 428, and 68 genes that were differentially regulated by the two viruses at 12, 24, and 48 h p.i., respectively. Functional analysis of these genes indicated that the majority were grouped into the categories of inflammatory disease, immune response, and cell death (Table 1). Because the greatest differences in the host response occurred at 24 h p.i, we focused our analyses at this time point. Figure 5A shows a heat map of the 428 genes that were differentially regulated at 24 h p.i. Notably, many of these genes were up-regulated in the bronchus of 1918 virus infected animals but down-regulated in animals infected with VN/1203. This included numerous genes associated with the inflammatory response and cell death pathways. Expression of selected genes was verified by RT-PCR (Figure S4).
Because the largest number of gene expression differences elicited by the two viruses was associated with the inflammatory response, we used Ingenuity Pathways Analysis to evaluate this response in greater depth and to visualize the interconnections between individual genes. This software analyzes gene expression data in the context of known biological response and regulatory networks as well as other higher-order response pathways. A network of differentially expressed genes associated with the inflammatory response is shown in Figure 5B. In this Figure, genes depicted in blue were up-regulated by the 1918 virus but down-regulated by VN/1203 virus, whereas genes depicted in green were up-regulated by both viruses. Of particular note was the differential regulation of NLRP3 (nucleotide-binding domain and leucine-rich-repeat-containing protein 3) and IL-1β, key components of the inflammasome, a recently described component of the innate immune response to influenza A virus [32],[33],[34]. Verification of IL1β and NLRP3 expression are provided in Figure S5. By 24 h p.i. the 1918 virus elicited the up-regulation of these genes; however they were both down-regulated during VN/1203 infection. In addition, only the 1918 virus elicited the increased expression of TNF-α, a key mediator of the inflammatory response. This clear difference in the transcriptional regulation of this network of inflammatory genes may have a significant impact on eventual disease outcome.
Our gene expression analyses also revealed that the 1918 and VN/1203 viruses perturbed the expression of genes associated with cell death, and that the two viruses elicited different patterns of gene expression associated with this response. We therefore sought to verify the presence of apoptotic cells by using the TUNEL assay in infected lung and bronchus. This assay revealed desquamated and apoptotic cells at the sites of inflammatory lesions as well as phagocytes containing cells undergoing apoptosis (Figure 6 and S3). In tissues from animals infected with the 1918 virus, there was evidence of apoptosis at 12 h p.i., but the level of TUNEL staining diminished markedly at later time points. In contrast, in tissues from animals infected with VN/1203, there was little evidence of apoptosis at 12 h p.i, but greater evidence of apoptosis was observed at the 24- and 48-h time points. These results indicate that both viruses elicited an apoptotic response in the lung and bronchus, but that the 1918 virus elicited an earlier response that later diminished, whereas VN/1203 elicited a comparatively delayed apoptotic response that was then maintained through the 48-h time point. Therefore, the timing and extent of the apoptotic response elicited by these two viruses may also have a significant impact on disease outcome.
The 1918 Spanish Flu pandemic was the deadliest infectious disease outbreak on record, and avian H5N1 influenza viruses have proven to be highly virulent in cases of human infection. To determine if there are differences in the host response to these two highly pathogenic influenza viruses, we performed extensive genomics analysis of bronchus tissue isolated from macaques infected with either the 1918 or VN/1203 viruses. We investigated the global transcriptional changes in bronchi, a lower respiratory tissue, rather than lungs because of the overwhelming cell infiltration that occurs in the lungs during influenza virus infection. Additionally, there was substantial replication of both viruses in this tissue. Our analyses revealed that these viruses elicited similar transcriptional profiles for interferon and cytokine genes, but markedly different transcriptional profiles for genes associated with inflammation and apoptosis. Our results depict a scenario where early during infection there was an accelerated up-regulation of the inflammasome by the 1918 virus in addition to the numerous cell death related genes that were up-regulated during infection with the 1918 virus but down-regulated during VN/1203 infection. These events represent important differences in the host response to these viruses and likely contribute to the severity and lethality of disease associated with 1918 virus infection.
Our previous study comparing macaque host responses to 1918 and A/Kawasaki/173/01 viruses showed that late after infection (3, 6 and 8 days) severity of tissue damage and fatal outcome was due to dysregulation of the antiviral response [8]. That study suggested that critical decisions influencing the outcome of the infection may occur very early. In this study, our results suggest that events taking place as early as 24 h after infection likely contribute to severity of tissue damage and lethal outcome during 1918 virus infection. For example, the 1918 virus elicited the increased expression of genes encoding NLRP3 and IL-1β, key components of the inflammasome, while these genes were being down-regulated by the VN/1203 virus. This finding suggests that an accelerated or excessive activation of the inflammasome is detrimental rather than protective in macaques. It is possible that early activation of NLRP3 and IL-1β production results in the infiltration of neutrophils and monocytes to the sites of infection where these cells continue to secrete cytokines resulting in a “cytokine storm”. Furthermore, Baskin et al. showed that VN/1203 infection of macaques was characterized by extensive tissue damage, an excessive and sustained type I interferon response, and innate immune induction [9]. Our present results also showed that VN/1203 caused significant pathology, but that 1918 infection generated more severe and sustained tissue damage. The study by Baskin et al. also compared the host response of VN/1203 infection to recombinant viruses containing two or three genes of the 1918 virus, whereas the present comparisons were performed using the fully reconstructed 1918 pandemic virus.
It is interesting to note that the virulence of these two viruses in macaques is somewhat different than in mouse models of infection. In the macaque model, the 1918 virus causes a lethal infection, whereas VN/1203 causes severe but typically non-lethal disease. In contrast, in the mouse model the avian VN/1203 virus is more pathogenic than the 1918 virus and is capable of causing disease and death at a much lower infectious dose. However, global transcriptional analysis of lung tissue from mice infected with the 1918 virus also revealed enhanced inflammatory and cell death responses that remained unabated until death [7]. Therefore, data from the mouse infection model also indicates a clear difference in the host response to 1918 and VN/1203 virus infection.
In addition to discovering significant differences in the inflammatory response elicited by the 1918 and VN/1203 viruses, we also identified notable differences in the apoptotic response induced by these viruses. Even though many studies have shown that influenza virus infection induces apoptosis, there are still contradicting reports about the induction and biological consequences of this response. For example, ectopic expression of the viral NS1 protein induces apoptosis [35]. In contrast, a NS1 deletion mutant has been shown to be a stronger inducer of apoptosis than the wild-type virus [36]. Regarding the consequences of the response, in vitro studies have demonstrated that expression of the anti-apoptotic protein Bcl2 results in impaired influenza virus production [37]. Caspase inhibitors also impair the propagation of influenza virus, possibly due to the nuclear retention of viral ribonucleoprotein (RNP) complexes [38]. Caspases may regulate the export of RNPs by increasing the diffusion limit of nuclear pores, thus allowing the passive diffusion of larger proteins [39]. Consequently, early during infection RNPs are transported out of the nucleus via an active export mechanism. As viral infection progresses and caspase activity increases cellular proteins are destroyed, thereby compromising the export of viral proteins. However, widening of nuclear pores may allow viral RNPs to use an alternative mode of exit from the nucleus thereby supporting viral replication. It has been suggested that this may allow influenza virus to take advantage of host cell antiviral responses to support viral replication [40].
Despite these apparent contradictions, differences in the activation of cell death pathways could help to explain the patterns of virus titers observed in our study. For example, the increase in apoptosis observed in the VN/1203 infected tissues (Figure 6, panels E, F) at the 48-h time point could result in less viral progeny as a consequence of cell death. In contrast, in the tissues infected with the 1918 virus, apoptosis decreased over time (Figure 6, panels A–C), which may ultimately provide a more favorable environment for the generation of additional viral progeny. The timing and extent of the apoptotic response elicited by these two viruses may therefore have a significant impact on eventual disease outcome.
We also found that a number of genes differentially regulated during 1918 virus infection were associated with cellular growth and proliferation and cell-cycle regulation (Table 1). The 1918 virus may take advantage of the altered expression of these host factors for utilization during its life cycle. In vitro experiments have demonstrated that nuclear extracts from uninfected cells increase influenza RNA synthesis as well as that of ectopically expressed virion RNPs, suggesting that cellular factors are involved in the switch between mRNA transcription and genome replication [41]. For example, RAF1 (RNA polymerase activating factor 1; identical to HSP90) appears to associate with the viral PB2 protein. This interaction may facilitate the association of unbound polymerase with RNA template, which could affect the interaction between PB1 and PB2 and therefore modulate the polymerase complex [42],[43],[44]. The involvement of host factors during the various stages of the viral life cycle is poorly understood and is the subject of major efforts to decipher the complex virus-host interactions that determine the outcome of infection.
In summary, our findings suggest that differential changes in the expression of inflammatory and cell death related genes as early as 12 to 24 h after infection likely contribute to the severity and lethal outcome of 1918 virus infection. Specifically, the accelerated up-regulation of the inflammasome components NLRP3 and IL-1β, as well as TNF-α, elicited by the 1918 virus may enhance neutrophil and monocyte infiltration and generate a strong antiviral response that will be detrimental rather than protective to the host. These findings may have therapeutic implications, suggesting that drugs that limit the inflammatory response may help to reduce the lung pathology associated with severe influenza virus infection.
The feasibility of such approach has been recently demonstrated by Aouadi et al. [45], who showed that silencing of the Map4k4 kinase in macrophages by oral delivery of β1,3-D-glucan-encapsulated siRNA particles protected mice from lipopolysaccharide-induced lethality by inhibiting TNF-α and IL-1β production. In addition, Marsolais et al. [46] showed that local administration of the sphingosine analog AAL-R to mouse airways significantly decreased the release of a variety of cytokines and chemokines known to contribute to the cytokine storm effect, resulting in less cytopathology of alveolar cells and inflammatory driven congestion of air spaces. Finally, Aldridge et al. [47] reported that decreased trafficking of a specific subset of dendritic cells by treatment with the peroxisome proliferator-activated receptor-agonist pioglitazone reduced morbidity and mortality associated with highly pathogenic influenza A virus infection. Thus, drugs that limit the early inflammatory response may be of particular benefit in treating infections caused by highly pathogenic influenza viruses.
|
10.1371/journal.pntd.0003946 | Mapping and Modelling the Geographical Distribution and Environmental Limits of Podoconiosis in Ethiopia | Ethiopia is assumed to have the highest burden of podoconiosis globally, but the geographical distribution and environmental limits and correlates are yet to be fully investigated. In this paper we use data from a nationwide survey to address these issues.
Our analyses are based on data arising from the integrated mapping of podoconiosis and lymphatic filariasis (LF) conducted in 2013, supplemented by data from an earlier mapping of LF in western Ethiopia in 2008–2010. The integrated mapping used woreda (district) health offices’ reports of podoconiosis and LF to guide selection of survey sites. A suite of environmental and climatic data and boosted regression tree (BRT) modelling was used to investigate environmental limits and predict the probability of podoconiosis occurrence.
Data were available for 141,238 individuals from 1,442 communities in 775 districts from all nine regional states and two city administrations of Ethiopia. In 41.9% of surveyed districts no cases of podoconiosis were identified, with all districts in Affar, Dire Dawa, Somali and Gambella regional states lacking the disease. The disease was most common, with lymphoedema positivity rate exceeding 5%, in the central highlands of Ethiopia, in Amhara, Oromia and Southern Nations, Nationalities and Peoples regional states. BRT modelling indicated that the probability of podoconiosis occurrence increased with increasing altitude, precipitation and silt fraction of soil and decreased with population density and clay content. Based on the BRT model, we estimate that in 2010, 34.9 (95% confidence interval [CI]: 20.2–51.7) million people (i.e. 43.8%; 95% CI: 25.3–64.8% of Ethiopia’s national population) lived in areas environmentally suitable for the occurrence of podoconiosis.
Podoconiosis is more widespread in Ethiopia than previously estimated, but occurs in distinct geographical regions that are tied to identifiable environmental factors. The resultant maps can be used to guide programme planning and implementation and estimate disease burden in Ethiopia. This work provides a framework with which the geographical limits of podoconiosis could be delineated at a continental scale.
| Podoconiosis is a neglected tropical disease that results in swelling of the lower legs and feet. It is common among barefoot individuals with prolonged contact with irritant soils of volcanic origin. The disease causes significant social and economic burden. The disease can be prevented by consistent shoe wearing and regular foot hygiene. A pre-requisite for implementation of prevention and morbidity management is information on where the disease is endemic and the identification of priority areas. We undertook nationwide mapping of podoconiosis in Ethiopia covering 1442 communities in 775 districts all over Ethiopia. During the survey, individuals underwent a rapid-format antigen test for diagnosis of lymphatic filariasis and clinical history and physical examination for podoconiosis. A suite of environmental and climatic data and a method called boosted regression tree modelling was used to predict the occurrence of podoconiosis. Our survey results indicated that podoconiosis is more widespread in Ethiopia than previously estimated. The modelling indicated that the probability of podoconiosis occurrence increased with increasing altitude, precipitation and silt fraction of soil and decreased with more clay content and population density. The map showed that in 2010, 34.9 million people lived in areas environmentally suitable for the occurrence of podoconiosis in Ethiopia.
| Podoconiosis is a form of elephantiasis that predominantly affects barefoot subsistence farmers in areas with red volcanic soil. It is characterized by bilateral swelling of the lower legs with mossy and nodular changes to the skin, and causes considerable disability. The aetiology is not fully understood; however, the current evidence suggests that mineral particles from irritant volcanic soils have a role, with some families having an additional genetic susceptibility to the condition [1,2]. In the last five years, there has been increased recognition of the disease and its importance. The World Health Organization (WHO) included podoconiosis in the list of neglected tropical diseases (NTDs) in 2011 [3]. The greatest burden of podoconiosis globally is assumed to occur in Ethiopia, and in 2013 Ethiopia included podoconiosis in its national NTD master plan [4]. Control of the disease is focused on early and consistent indoor and outdoor shoe wearing and regular foot hygiene for prevention, as well as simple lymphoedema management including foot hygiene, bandaging, massage, shoe and sock wearing and, in extreme cases, minor surgery for morbidity management [2,5]. To guide the implementation of these measures it is essential to have a detailed understanding of the geographical distribution of podoconiosis.
The first attempt to map the distribution of podoconiosis was based on school and market surveys conducted by Price in 1974 [6,7]. Although this work provides an important contribution, it is limited by the inclusion of non-representative populations because it was based on market-based sampling and counted all lymphoedema cases without excluding other potential causes. Moreover, Ethiopia has undergone economic and social transformation since the 1970s, and these economic changes will have affected shoe wearing habits, foot hygiene and housing conditions, which, in turn, may influence the risk of developing podoconiosis [8]. The more recently conducted studies [9–13] have typically been conducted in areas known to be endemic for the disease and at local scales [14].
In order to guide the Ethiopia NTD master plan, we conducted the first nationwide integrated mapping of podoconiosis and lymphatic filariasis (LF) between June and October 2013. Previous work described the methodology of the integrated mapping [15], and investigated the epidemiology and individual and household risk factors [8]. Building on this work, the aim of the present paper is to (i) describe the geographical distribution of podoconiosis across Ethiopia, (ii) identify environmental factors associated with the occurrence and prevalence of podoconiosis, (iii) define the spatial limits of disease occurrence, and (iv) estimate the population living in areas at risk from the disease.
Ethical approval for the study was obtained from the Institutional Review Board of the Medical Faculty, Addis Ababa University, the Research Governance and Ethics Committee of Brighton & Sussex Medical School (BSMS), and ethics committees at the Ethiopian Public Health Institute (EPHI) and Liverpool School of Tropical Medicine. Individual written informed consent was obtained from each participant ≥18 years of age. For those individuals <18 years old, consent was obtained from their parents or guardian and the participant themselves provided informed assent. Confirmed W. bancrofti infection was treated using albendazole (400 mg) and ivermectin (200 μg/kg body weight or as indicated by a dose-pole) according to WHO recommendations. For those with lymphoedema, health education was given about morbidity management.
Ethiopia is located in the Horn of Africa. The total population in 2013 is estimated to be 86.6 million [16,17], with the majority of the population living in rural areas. Ethiopia has a federal system of administration with nine regional states and two city administration councils (Fig 1A) [18]. The country has three broad ecologic zones, based on topography: the “kola” or hot lowlands, the “weyna dega” or midland and the “dega” or the cool temperate highlands[19]. Altitudinal variation in temperature gives rise to a variety of vegetation types and suitability of land for agriculture [16].
The data originated from two sources: the nationwide integrated LF and podoconiosis mapping in 2013 and a LF mapping survey in western Ethiopia, 2008–2010. The details of each survey are provided elsewhere [8,15,20]. In brief, the 2013 survey was conducted in 659 districts (woredas) and included 1,315 villages. During the survey, individuals underwent a rapid-format antigen test for diagnosis of LF (immunochromatographic card test [ICT]) and clinical history and physical examination for podoconiosis. Further details are given elsewhere [15]. The 2008–2010 survey included 116 districts located in five regional states in western Ethiopia, conducted by a team from Addis Ababa University. Thirty-seven of the 116 districts were found to be endemic for LF. Cases of podoconiosis were extracted from this data set, based on expert opinion. Presence of lymphoedema cases in districts not endemic for LF, without sign or symptoms of other potential causes were considered podoconiosis cases (see Supporting Information S1). All 37 districts endemic for LF were excluded from data extraction to avoid misclassification of cases, while podoconiosis data were extracted from the remaining 79 [20]. Combined, the two surveys contributed 1,442 clusters from 775 districts of Ethiopia. The aggregation of the data was conducted by combining the point data in each administrative unit and calculating the prevalence at district level: total number in district with disease divided by total number examined in the district.
The elevation data at 90 m resolution were derived from a gridded digital elevation model produced by the Shuttle Radar Topography Mission (SRTM)[21], and these data were processed to calculate slope in degrees. The mean atmospheric temperature and annual mean precipitation at 30-arcsecond (approx. 1 km) resolution were downloaded from the WorldClim database for the period 1950–2000 [22]. A suite of raster surfaces containing values of Enhanced Vegetation Index (EVI) were obtained from the African Soil Information System (AfSIS) project [23].
Soil data including silt, clay and sand content, dominant soil type and soil-pH at 1 km2 resolution were downloaded from the ISRIC-World Soil Information project[24]. A gridded map of soil texture included in the Harmonized Soil Map of the World at 1 km2 resolution was obtained from the Africa Soil Information Service (AfSIS), which is developing continent-wide digital soil maps for sub-Saharan African[24]. Straight line distance to water bodies was calculated using the data layers of water bodies produced by the SRTM at 250 m resolution[21]. Land cover type, according to the United Nations (UN) land cover classification system, was extracted from the qualitative global land cover map, produced at 300 m resolution from data collected by the environmental satellite (ENVISAT) mission’s Medium Resolution Imaging Spectrometer (MERIS) sensor[25]. Gridded maps of both population density and rural-urban classification for 2010 were obtained from the WorldPop project [26,27] and the Global Rural-Urban Mapping project (GRUMP), respectively[28,29]. Finally, Aridity Index data were extracted from the Global-Aridity datasets (CGIARCSI)[30,31]. Survey and covariate data were linked in ArcGIS 10.1 (Environmental Systems Research Institute Inc. [ESRI] Inc., Redlands CA, USA) based on the WGS-1984 Web Mercator projection at 1 km2 resolution. Bilinear interpolation was applied to resample numeric (continuous) raster data sets, whereas nearest neighbor interpolation was used with ordinal raster layers. Input grids were either extended or clipped to match the geographic extent of a land mask template of Ethiopia, and eventually aligned to it.
The data were entered using a Microsoft Excel 2007 (Microsoft Corporation, Redmond, WA) spreadsheet and exported into STATA 11.0 for analysis (Stata Corporation, College Station, TX, USA). Point prevalence maps were developed in ArcGIS 10 (ESRI, Redlands, CA) and covariate data extracted for each data point. Multicollinearity between the covariates was initially explored using cross-correlations and where correlation coefficients were >0.7 only non-linearly related covariates were included in the analysis (S1 Text).
Boosted Regression Tree (BRT) modelling[32,33] was used to identify the environmental factors associated with the occurrence of podoconiosis in Ethiopia. This approach has been effectively used in global mapping of dengue, LF, leishmaniasis and malaria vector mosquitos [34–37] and has superior predictive accuracy compared to other distribution models[38]. In brief, BRT modelling combines regression or decision trees and boosting in a number of sequential steps [32,33]. First, the threshold of each input variable that results in either the presence or the absence of podoconiosis is identified, allowing for both continuous and categorical variables and different scales of measurement amongst predictors [32]. Second, boosting is a machine-learning method that increases a model’s accuracy iteratively, based on the idea that it is easier to find and average many rough ‘rules of thumb’, than to find a single, highly accurate prediction rule.
Boosted Regression Tree utilizes data on both presence and absence of podoconiosis. Presence was defined as an area with at least one case in the two surveys and absence as an area with no cases in either survey. A selection of 16 environmental and climate covariates were included in a single BRT model in order to explore the relative importance of each covariate in explaining the occurrence of podoconiosis in Ethiopia. Four covariates (land cover, soil type, soil texture, urban rural classification) were excluded that showed little explanatory power (<1% of regression trees used the covariate) on the occurrence of podoconiosis. The retained covariates were used to build the final model included annual precipitation, elevation, population density, enhanced vegetation index, terrain slope, distance to water bodies, silt fraction and clay fraction. In order to obtain a measurement of uncertainty for the generated model, we fitted an ensemble of 120 BRT submodels to predict sets of different risk maps (each at 1km x 1km resolution) and these were subsequently combined to produce a single mean ensemble map and the relative importance of predictor variables was quantified. These contributions are scaled to sum 100, with a higher number indicating a greater effect on the response. Marginal effect curves were plotted to visualize dependencies between the probability of podoconiosis occurrence and each of the covariates. To assess the association of covariates and high prevalence podoconiosis, the prevalence estimates were plotted against each environmental variable. This will help to identify the areas with very high prevalence and to prioritize interventions. BRT modelling and model visualization was carried out in R version 3.1.1 [39] using the packages raster [40]and dismo[41].
The resulting predictive map depicts environmental suitability for the occurrence of podoconiosis. In order to convert this continuous map into a binary map outlining the limits of podoconiosis occurrence, a threshold value of suitability was determined, above which the occurrence was assumed to be possible. Using the receiver operating characteristic (ROC) curve, a threshold value of environmental suitability was chosen such that sensitivity, specificity and proportion correctly classified (PCC) values were maximized. Finally, we estimated the number of individuals at risk by overlaying the binary raster dataset displaying the potential suitability for podoconiosis occurrence on a gridded population density map[26,27] and calculating the population in cells considered to be within the limits of podoconiosis occurrence. The 95% CI of the population at risk were calculated based on binary maps of the lower (2.5%) and upper (97.5%) bounds of the predicted probability of occurrence.
The performance of each sub-model was evaluated using different statistics, including: proportion correctly classified [PCC], sensitivity, specificity, Kappa [κ] and area under the receiver operator characteristics curve (AUC). The mean and confidence intervals for each statistic were used to evaluate the predictive performance of the ensemble BRT model. In addition to ensemble approach to validation, an external validation was performed using data from 96 independent surveys conducted between 1969 and 2012 [6,7,9–12,42–44] which we previously identified through structured searches of the published and unpublished literature [14]. The AUC was used to assess the discriminatory performance of the predictive model, comparing the observed and predicted occurrence of podoconiosis at each historical survey. AUC values of <0.7 indicate poor discriminatory performance, 0.7–0.8 acceptable, 0.8–0.9 excellent and >0.9 outstanding discriminatory performance [45].
Data were available for 141,238 individuals from 1,442 communities in 775 districts (woredas) from all regional states of Ethiopia. The mean number of individuals sampled per community was 97.6; the majority of communities (1,350, 93.6%) had more than 90 examined individuals, while 47 communities (3.3%) had less than 70 individuals.
Overall, 5,712 (4.0% lymphoedema positivity) podoconiosis cases were identified in 713 communities, with lymphoedema positivity rates ranging from 0.9 to 54.6% by community. Fig 1B and 1C display the distribution of podoconiosis at community and woreda level, respectively, and highlight marked regional variation. No cases of podoconiosis were found in Addis Ababa, Affar, Dire Dawa and Gambella regional states, whereas few cases were found in Tigray, Somali, Benishangul Gumuz and Harari regions (Table 1). Disease lymphoedema positivity rate was highest in the central highlands of Ethiopia, in Amhara, Oromia and Southern Nations, Nationalities and Peoples (SNNP) regional state (Table 2). A further four districts in Benishangul Gumuz and Tigray and 1 district in Somali were found to be endemic (Fig 1D).
Fig 2 shows the marginal effect of each covariate on the predicted suitability of occurrence for podoconiosis, averaging across the effects of all other variables, and its relative contribution to the final BRT model. Major predictors of the occurrence of podoconiosis were annual precipitation (accounting for 30.7% of the variation explained by the model), elevation (22.6%), EVI (15.4%) and population density (12.7%). Slope only contributed 8.2% to the predicted occurrence. Annual precipitation causes an increase in probability of occurrence starting from precipitation values of around 1,000 millimeters (mm) per year. High suitability for podoconiosis is also positively associated with elevation, increasing between 1,000–2,000 m asl and then sharply declining after 2,000 m asl. EVI is linearly correlated to the risk of podoconiosis occurrence up to 0.5 and declines sharply thereafter. Population density is negatively correlated with the probability of podoconiosis occurrence, with population density greater than 10,000 population/ km2 causing no effect on the probability of occurrence of podoconiosis. Although silt fraction and clay fraction contributed little to the final BRT model, the occurrence of podoconiosis was found to be associated with decreasing clay fraction and increasing silt fraction.
Previous studies have indicated a relationship between the prevalence of podoconiosis and climate and environmental covariates (including rainfall, altitude, temperature and soil type), and have characterized high prevalence areas using certain environmental variables [46]. In order to assess this relationship in Ethiopia, Fig 3 depicts the relationship between the environmental variables and the prevalence of podoconiosis. Thus, the distribution of podoconiosis is clearly bounded within an altitude range of 1,000–2,800 m asl EVI > 0.2 and annual precipitation >1,000 mm.
Fig 4A presents the map of environmental suitability for podoconiosis and suggests that suitability is greatest in the central highlands of Amhara, Oromia and SNNP regional states. Absence of podoconiosis is predicted in Affar, Gambella and Somali regional states. A suitability cut-off of 0.49 with a sensitivity of 0.77 and specificity 0.86 provided the best discrimination between presence and absence records in the training data, and therefore this threshold value was used to reclassify the predictive risk map into a binary map outlining the potential environmental limits of occurrence (Fig 5). Uncertainty was calculated as the range of the 95% confidence interval in predicted probability of occurrence for each pixel (Fig 4B) indicating high uncertainty in the eastern part of Somali regional state. Cross-validation in the BRT ensemble model indicated high predictive performance of the BRT ensemble model with an AUC value of 0.84 (95% confidence interval (CI): 0.84–0.85; standard deviation (sd): 0.016). External validation against historical data showed an excellent performance of the final fitted model to classify at-risk areas, with an AUC value of 0.89 (CI 95%: 0.81–0.97).
The national population living in areas environmentally suitable for podoconiosis is estimated to be over 34.9 (95% CI: 20.2–51.7) million, which corresponds to 43.8% of Ethiopia’s population in 2010. The largest portions of the population at risk were found in SNNP (68.1%) Oromia (.48.0%) and Amhara (49.6%) (Table 3). We conducted a sensitivity analysis to determine the effect of the optimal suitability threshold (0.496) on the estimates of at-risk population. For that, we applied both a lower (0.3) and a higher (0.6) cut-off to dichotomize the final BRT model, and estimated the population living in suitable areas for podoconiosis based on these extreme thresholds. The total estimated population at risk would be 44.6 million (95%CI: 27.8–59.4) and 29.9 million (95%CI: 16.7–46.8) for the 0.3 and 0.6 cut-offs respectively.
Despite the growing global awareness of podoconiosis [3,47], national scale epidemiological data about the distribution of podoconiosis are lacking in all endemic countries. Understanding the geographical distribution and estimating the population at risk are important first steps to optimally use the resources allocated to podoconiosis [48,49]. To our knowledge, this is the first nationwide mapping of podoconiosis using a predefined clinical algorithm to diagnose podoconiosis. It is also the first attempt to develop a risk model of podoconiosis based on remotely sensed environmental data and robust statistical techniques. Our study showed that podoconiosis is widely distributed in Ethiopia and covers substantial parts of the country. Besides, our results indicate that 43.8% of the Ethiopian population lives at risk of podoconiosis and a quarter of the landmass is conducive to podoconiosis occurrence. Our mapping largely indicated high (close to 1) or low (close to 0) probability of occurrence of podoconiosis. This indicates the degree of certainty from the maps is very high for both presence and absence. Therefore the findings here will help guide interventions and resource allocation and estimate the disease burden caused by podoconiosis.
In the current analysis we identified specific environmental factors associated with the occurrence and prevalence of podoconiosis and used BRT modelling to delineate the environmental limits of podoconiosis in Ethiopia. Our results show that the probability of podoconiosis occurrence and its prevalence increase with annual precipitation and elevation, and decrease with population density. Previously, it had been observed that altitude governs temperature and other climatic conditions conducive to generation of soil suitable for podoconiosis occurrence [46,50]. Rainfall is also important in the pathway of soil formation, and may also increase exposure to the soil components [46,50,51]. Studies have indicated that soils associated with podoconiosis are slippery and adhesive if allowed to dry [46,50,51].
Our risk map, developed using BRT modelling, shows that the environmental conditions conducive to the occurrence of podoconiosis are found throughout the central highlands of Ethiopia, located in Amhara, Oromia and SNNP regional states. This distribution corresponds well with the historical distribution of podoconiosis in Ethiopia [14]. Furthermore, we were able to clearly identify environmental limits for the distribution and intensity of podoconiosis occurrence in Ethiopia. Podoconiosis occurred in areas where annul precipitation is >1000 mm, and elevation was between 1,000 and 2,800 m asl. In general, the high lymphoedema positivity rate (≥5%) districts were characterized by mean annual precipitation of 1,665 mm and altitude of 1,892 m asl.
Moreover, this work provides interesting insight into the regional distribution of podoconiosis in Ethiopia. Both the observed distribution (Fig 1C) and the map of environmental limits (Fig 4A) indicate a heterogeneous distribution within those regions most at risk of podoconiosis, namely Amhara, Oromia and SNNP. In Amhara, the highest environmental suitability is predicted in East Gojjam and West Gojjam, South Gondar and Awi zones, and similarly in the western part of Oromia including East Wollega, West Wollega and Kellem Wollega, Illubabor, Jimma, North and West Shoa zones. In SNNP, most of the zones are at a high risk of podoconiosis except Bench Maji and South Omo zones where LF is prevalent. These findings are in concordance with previous studies conducted in small areas in these three regional states [10–13], which almost exclusively cover the central highlands of Ethiopia where agrarian communities reside. Given the agriculture-led economy of the country, the findings here have several implications. First, podoconiosis is not only a health problem but may also be a constraint for economic development. To have a healthy and productive agrarian community, the government should prioritize prevention and control of podoconiosis in the most at-risk regions. The inclusion of podoconiosis into the national integrated NTD master plan was an important first step [52], but implementing this master plan will require resource mobilization and allocation.
Our results show that podoconiosis is more widely distributed in Ethiopia than previously thought. The population at risk and the landmass suitable for the occurrence of podoconiosis is considerably beyond previous estimates of 11–15 million people (or one fifth of the country’s landmass) [48,53]. There are several reasons for these differences. First, previous estimates were limited to rural areas and zones historically known to be endemic for podoconiosis. Second, they relied on school and market surveys, which might have underestimated the geographical distribution of the disease [6,7,54], for these counts were only localized to areas in which markets or schools were present. For instance, these studies were conducted some 40 years ago when the school coverage in Ethiopia was fairly limited. In addition, population movement and settlement schemes may have contributed to the current increase in at-risk population [43].
Globally this is the first comprehensive countrywide mapping of podoconiosis. We have included almost every district in Ethiopia and followed WHO recommendations for mapping LF [55]. We have used data from LF mapping in southwest Ethiopia [20], but only analysed data from non-endemic districts. The diagnostic criteria and sampling methods employed make both data sets comparable. Although the study has several strengths it is not without limitations. First, we used information from district offices to select study sites (mostly suspected endemic areas) within districts, which might have led to overestimation of prevalence. Second, although adult individuals were mobilized to central places for random selection, self-selection bias might have affected our findings, potentially overestimating occurrence. To minimize this, we mobilized the entire community prior to the survey using house to house visits by Health Extension Workers without mentioning the disease surveyed. Third, there is no definitive diagnostic test for podoconiosis to date, so we developed a clinical algorithm to diagnose podoconiosis[15]. We excluded all other potential causes of lymphoedema using stringent criteria that might—if anything—result in underestimation rather than overestimation of podoconiosis. Fourth, no mapping approach for podoconiosis has yet been defined, consequently we adopted the mapping approach for LF. The assumptions valid for LF might not hold true for podoconiosis: for example, the prevalence estimates from two villages per district might not reflect the actual distribution in the district [55]. However, from previous observational studies, podoconiosis distribution has shown to be less focal than that of LF [46,56]. Fifth, lack of perfect temporal overlap of the outcome and covariate is another limitation of the data. Nonetheless we used the long term averages of environmental data for our analysis for a number of reasons. The weathering of rock to soil takes place over extended periods of time. Podoconiosis is a chronic disease and requires several years of exposure to irritant soil. The prevalent cases seen today may have been exposed for more than a decade to the putative causes. The various covariate data are available for differing time periods; we have however sought to use the available data which covers the largest time period. Finally, an important issue concerning the use of remote sensing data to identify ecological association between environment and podoconiosis is spatial scale[57,58]. The variables which affect the distribution of podoconiosis at small area and large area might differ. Although previous studies identified several soil characteristics to affect the risk of podoconiosis at small area[59], such association was not maintained in the current analysis.
Studies have identified different risk factors for podoconiosis at different spatial scales[60,61]. At a household level, the risk of podoconiosis is influenced by individual shoe wearing[62], hygiene practices[8] and genetics [63], factors which were not captured in the our large area model due to lack of standardized data on such factors. At large geographical levels, previous studies report high levels of podoconiosis in areas with high red clay soils which adhere when dry on the skin[50]. In the current study, although there is a slight increase in the prevalence of podoconiosis with increasing silt content and decrease with increasing clay content, podoconiosis was most common in areas where the silt content is 30% and the clay content is 25–50% (Fig 3), attributes characteristics of clay soil[64]. The soil data used in our analysis are available at 1km2 resolution and only measure top soil (0–5 cm), and as such may belie small area variation and does not provide information on sub-surface soil.
This work makes three important contributions to increasing the understanding of podoconiosis. First, we have defined the environmental limits of podoconiosis in Ethiopia, enabling estimation of population at risk. With further validation, this may lead to delineation of the global limits of podoconiosis occurrence. Second, we have identified environmental factors which are associated with the occurrence of podoconiosis in Ethiopia. If these environmental factors are found to be associated with disease in other settings, a continental risk map of podoconiosis can be generated. Third, by narrowing the environmental limits of podoconiosis, the findings here will guide the identification of the exact mineral particles in the soil responsible for podoconiosis.
In addition to providing a predictive map of the risk of podoconiosis, we also provide a map of uncertainty in these predictions, and an illustration of how that uncertainty relates to environmental variables in the marginal effect plots. By providing a map of where risk of occurrence is less predictable using the environmental variables considered here, we hope to better inform policy makers and researchers about where the main prediction map is likely to be most reliable. This map may also be used when deciding where to target future surveillance for the disease and where further studies could help elucidate its main drivers.
The geographical distribution and burden of podoconiosis in Ethiopia is formidable and represents an important challenge to program planners and policy makers. Success in tackling this national problem is, in part, contingent on strengthening the evidence base on which control planning decisions and their impacts are evaluated. It is hoped that this mapping of contemporary distribution of podoconiosis will help to advance that goal. Empirical evidence has shown that podoconiosis management is effective in the early stages of the disease and improves clinical measures and the quality of life of patients [5]. If this management is found to be effective and cost-effective using more robust assessment, the next step will be scaling up interventions in all endemic districts. Prioritizing those districts with high prevalence would be a cost-effective approach. Scaling up prevention of podoconiosis through consistent shoe wearing is also vital. Studies in southern Ethiopia have identified cultural, financial and logistic barriers to shoe wearing [65,66], and have enabled to develop a community messaging intervention to enhance prevention of podoconiosis. This intervention requires testing and adaptation to other endemic districts, possibly in combination with the hygiene promotion package of the 16-package Health Extension Program.
In conclusion, our results provide a detailed description of the geographical distribution and environmental limits of podoconiosis in Ethiopia. This will enable optimal allocation of the limited resources available for podoconiosis control, permit evaluation of the impact of interventions in the future, and guide mapping of other potentially endemic countries and contribute to the global mapping of podoconiosis.
|
10.1371/journal.pgen.1004555 | A Genetic Strategy to Measure Circulating Drosophila Insulin Reveals Genes Regulating Insulin Production and Secretion | Insulin is a major regulator of metabolism in metazoans, including the fruit fly Drosophila melanogaster. Genome-wide association studies (GWAS) suggest a genetic basis for reductions of both insulin sensitivity and insulin secretion, phenotypes commonly observed in humans with type 2 diabetes mellitus (T2DM). To identify molecular functions of genes linked to T2DM risk, we developed a genetic tool to measure insulin-like peptide 2 (Ilp2) levels in Drosophila, a model organism with superb experimental genetics. Our system permitted sensitive quantification of circulating Ilp2, including measures of Ilp2 dynamics during fasting and re-feeding, and demonstration of adaptive Ilp2 secretion in response to insulin receptor haploinsufficiency. Tissue specific dissection of this reduced insulin signaling phenotype revealed a critical role for insulin signaling in specific peripheral tissues. Knockdown of the Drosophila orthologues of human T2DM risk genes, including GLIS3 and BCL11A, revealed roles of these Drosophila genes in Ilp2 production or secretion. Discovery of Drosophila mechanisms and regulators controlling in vivo insulin dynamics should accelerate functional dissection of diabetes genetics.
| Genome-wide association studies in patients with type 2 diabetes mellitus have identified more than 65 loci, encoding up to 500 candidate susceptibility genes. Thus, investigators are fundamentally challenged to (i) screen and identify relevant candidates in vivo, (ii) determine if loss- or gain-of-function underlies the association, (iii) link perturbed gene function to hallmark type 2 diabetes mellitus physiological phenotypes like insulin production or secretion, and (iv) identify relevant tissue(s) where the biological function of a specific regulator is required. Here we exploit Drosophila genetics to reveal the molecular functions of evolutionally conserved regulators that are associated with human type 2 diabetes mellitus. Targeted knockdown of Drosophila orthologues of diabetes risk genes revealed tissue-specific roles for these genes in regulating insulin production and secretion. These findings should accelerate use of Drosophila and other genetically-tractable systems to discover conserved mechanisms and regulators controlling in vivo insulin dynamics relevant to diabetes and other human diseases.
| Insulin is a major regulator of metabolism, growth and development in metazoans, including the fruit fly Drosophila melanogaster. Insulin resistance in the liver and other human tissues can lead to compensatory increases in insulin production and secretion by pancreatic β cells, a facultative response that fails during pathogenesis of type 2 diabetes mellitus (T2DM) [1]. The decline of both insulin sensitivity and insulin secretion may have a genetic basis in humans [2]. Drosophila could emerge as a powerful system for dissecting the genetics of insulin resistance and secretion if appropriate physiological assays, like quantification of circulating insulin, could be used to assess insulin dynamics.
Drosophila Insulin-like peptide 2, 3 and 5 (Ilp2, 3, and 5) are synthesized and secreted by insulin producing cells (IPCs), median neurosecretory cells located in the pars intercerebralis, and are crucial hormonal regulators of development, growth and metabolism [3], [4]. Ilp2 is a principal circulating insulin in flies, and is essential for maintaining normoglycemia [5]. Structural and biochemical studies of Drosophila insulin-like peptide association with its receptor suggest that Ilps might circulate at picomolar levels, similar to mammals [6]. Other than in mammals, however, no methods for determining the absolute concentration of circulating insulin with picomolar sensitivity exist, to our knowledge. The most widely-used method for assessing insulin secretion by Drosophila IPCs involves estimating the immuno-reactive signal for Ilp2 in IPCs [7]. By comparing the relative intensity of signal between experimental and control conditions, increased signal has been interpreted to indicate reduced or impaired secretion of insulin from IPCs. However, in using intracellular Ilp2 immunoreactivity as a surrogate for secretion, this method does not differentiate between changes in insulin production and secretion. To overcome these challenges, focus has shifted to the use of enzyme-linked immunosorbent assay (ELISA) as a potential method for Ilp2 quantification. An immunoepitope tagged Ilp2 was used to measure circulating Ilp2 by ELISA [8], but the tagged Ilp2 was overexpressed in IPCs, making it difficult to assess physiological regulation of Ilp2 production and secretion. A recent study used polyclonal antibodies to measure circulating Ilp2 and Ilp5 in adult hemolymph by ELISA [9], but only relative changes were reported. Moreover, the specialized nature of required reagents, like synthetic Ilp standards, have limited widespread adoption of this assay. Polypeptide-based immunoepitope tags can facilitate ELISA construction, but prior attempts over several decades to epitope-tag insulin, which undergoes extensive post-translational modification [6], have led invariably to loss or elimination of bioactivity [10], [11]. Maintenance of bioactivity in an epitope-tagged insulin would ensure that native mechanisms controlling crucial elements of insulin biology, like processing, storage, secretion and clearance, are being assayed. However, the bioactivity of prior epitope-tagged forms of Ilp2 has not been demonstrated or quantified [12].
Here we report the successful labeling of Ilp2, a crucial regulator of glucose metabolism in Drosophila, with two immuno-epitopes at specific positions that preserved Ilp2 bioactivity. Using unique fly strains expressing double-tagged Ilp2 (Ilp2HF), we developed robust and sensitive ELISA methods for quantifying circulating Ilp2 at picomolar concentration in Drosophila, and show that only a small fraction of total Ilp2 is secreted from IPCs in vivo and in vitro, like in mammals. Our studies reveal changes in either Ilp2 expression or secretion resulting from IPC-specific knockdown of Drosophila cognates of human T2DM genome-wide association study (GWAS) candidate loci, demonstrating genetic and molecular mechanisms linking these risk genes to insulin regulation. In addition, we uncovered previously undetected forms of genetic insulin receptor haploinsufficiency accompanied by adaptive insulin hypersecretion. Tissue specific dissection of this phenotype revealed a critical role for insulin signaling in the fat body in feedback regulation of systemic insulin levels. Thus, we provide the community with a potent new Drosophila tool for studies of insulin biology, integrative physiology and the genetic basis of human metabolic diseases.
To measure a circulating Drosophila insulin directly, we sought to tag Drosophila Insulin-like peptide 2 (Ilp2) with immuno-detectable epitopes while preserving its in vivo bioactivity. Broad misexpression of a transgene encoding Ilp2 from GAL4-responsive upstream activation sequences (UAS) is lethal [13], providing an assay of Ilp2 activity in vivo. To screen for permissive epitope-insertion sites that preserved the bioactivity of Ilp2, we misexpressed transgenes encoding variant forms of hemagglutinin- (HA) and FLAG-epitope-tagged Ilp2 (Figure S1), and scored the resulting lethality. Like mammalian insulins, Drosophila Ilp2 is comprised of ‘B-chain’ and ‘A-chain’ polypeptides linked by disulfide bonds (Figure 1A). Systematic variation of epitope position in the B-chain and A-chain of Ilp2 led to identification of tagged forms that remained lethal when broadly expressed, including a variant with HA-epitope fused to the B-chain carboxy-terminus and FLAG-epitope fused to the A-chain amino terminus (hereafter called “Ilp2HF”; Figure 1A). In contrast, a previously described FLAG-epitope labeled Ilp2 [12] achieved only 38% lethality (Figure S1). Substitution of a conserved A-chain cysteine by tyrosine (from a missense mutation called ‘Akita’) impairs insulin processing and activity in rodents [14], [15], and the orthologous substitution (C119Y) prevented Ilp2HF-induced lethality (Figure S1). Thus, we identified epitope-tags and positions in Ilp2 that preserved in vivo bioactivity in a synthetic lethality screen.
To assess if Ilp2HF also retained native Ilp2 function and activity in regulating development, growth, and metabolism, we generated a 2.4 kilobase pair genomic fragment (Figure 1B) containing the native Ilp2 or Ilp2HF gene under the control of the endogenous Ilp2 regulatory sequence (gd2 and gd2HF, respectively). We next assessed if developmental, growth, and metabolic defects observed in flies lacking Ilp2, Ilp3, and Ilp5 (Ilp2–3, 5) [5] could be rescued by introducing gd2 or gd2HF into Ilp2–3, 5 mutants. Development from egg to adult eclosion in Ilp2–3, 5 mutant females requires an average of 16 days, compared to 10 days for control flies (Figure S2). The delay is shortened to 11 days in mutants harboring a genomic Ilp2 rescue construct (Ilp2–3, 5 gd2) and 12 days in mutants harboring a genomic Ilp2HF rescue construct (Ilp2–3, 5 gd2HF; Figure S2). Thus, both gd2 and gd2HF substantially rescued the developmental delay observed in insulin deficient flies, although the eclosion time was delayed by 1–2 days with gd2HF rescue. In addition, both gd2 and gd2HF rescued the reduced wing length of Ilp2–3, 5 mutant adult flies to the same degree (Figure 1C). In Ilp2–3, 5 mutant flies, elevated levels of trehalose, the major circulating form of sugar in flies, were also rescued by either gd2 or gd2HF (Figure 1D). However, the C119Y missense mutant form of gd2HF failed to rescue developmental delay, wing length or trehalose phenotypes in Ilp2–3, 5 flies (Figures 1C and 1D, Figure S2). Thus, Ilp2HF rescues severe insulin-deficiency to an extent comparable to native Ilp2, providing a unique example of a dual epitope-tagged insulin that retains in vivo biological activity that is nearly indistinguishable from native insulin.
To investigate the physiological regulation of in vivo Ilp2 levels, we introduced a single copy of the gd2HF genomic rescue fragment by site-directed insertion into Ilp2 null mutants (hereafter Ilp21 gd2HF), thereby replacing endogenous Ilp2 with Ilp2HF in the genome. Immunostaining revealed Ilp2HF protein was restricted to adult IPCs of the pars intercerebralis without detectable ectopic expression in Ilp21 gd2HF brains (Figure 1E). Circulating trehalose levels were indistinguishable in Ilp21 gd2HF adults and controls (Figure 2A). Quantitative reverse-transcriptase polymerase chain reaction (qPCR) revealed that Ilp2 mRNA levels in Ilp21 gd2HF adults and controls were significantly reduced in 3 day-old flies compared to 1 day-old flies (Figure 2B). Thus, in vivo regulation of the gd2HF genomic rescue fragment recapitulates that of native Ilp2 and produces the same physiological responses. To quantify total and circulating Ilp2HF in adult flies we developed an ELISA based on commercially available monoclonal antibodies and peptide standards harboring both HA- and FLAG- epitope tags (Figure 2C). This assay detected signal in sample volumes of 1 µl in a standard range of 40 pM to 4 nM (Figure 2C; Materials and Methods). In contrast to mRNA levels, total Ilp2HF content of 1 and 3 day-old homozygous Ilp21 gd2HF adults did not change (Figure 2D), demonstrating that changes of Ilp2 mRNA levels do not strictly correlate with changes in total protein levels. The average circulating Ilp2HF concentration in hemolymph from 1 day-old homozygous Ilp21 gd2HF adults was 100 pM and increased to 350 pM in 3 day-old adults (Figure 2E), demonstrating further that levels of secreted Ilp2HF protein in hemolymph from adult flies are regulated independently of total Ilp2HF content. Based on an estimated adult hemolymph volume of 80 nanoliter [16], we determined that the total circulating Ilp2HF rises from 0.05 pg in 1 day-old adult flies to 0.22 pg in 3 day-old adult flies. Thus, we calculate that only 0.1% of total Ilp2HF circulates in the hemolymph of 1 day-old flies, increasing to 0.35% of total Ilp2HF content in 3 day-old flies (Figure 2E). Together these results indicate that a small fraction of total Ilp2HF in IPCs is secreted into the hemolymph in vivo, and demonstrate that our ELISA method permits assessment of physiological regulation of insulin production and secretion in flies.
To further assess the fraction of Ilp2 secreted upon stimulation of IPCs, we isolated heads from 3 day-old homozygous Ilp21 gd2HF flies, and stimulated them with 100 mM KCl, as previously reported [7]. Under physiological conditions in 3 mM KCl adult hemolymph-like solution (AHLS), 0.2 pg of Ilp2HF (per head) was secreted. This increased 7-fold to 1.4 pg of Ilp2HF (per head) upon stimulation with 100 mM KCl AHLS (Figure 2F). Similar to results in vivo, only 0.6% of the total Ilp2 content in heads was secreted in physiological AHLS and the fractional secretion increased to 2.8% of the total when heads were stimulated with 100 mM KCl, comparable to the fraction of insulin secreted from stimulated rat islets [17]. These results emphasize that only a small fraction of total Ilp2 content of IPCs is secreted, even after maximal depolarization induced by 100 mM KCl.
To assess if nutrient availability acutely regulates insulin secretion and circulating levels in hemolymph, as indicated by prior studies [7], we also measured Ilp2HF in fasted and re-fed adult flies. In 3-day old flies fasted for 24 hours then re-fed for 30 minutes, circulating Ilp2HF concentration peaked then declined (Figure 2G), a pattern and timing strikingly similar to that observed in fasted and re-fed humans [18]. Thus, our assays provided measures of systemic insulin levels in flies on a time scale comparable to in vivo measures of mammalian insulin dynamics.
IPCs are neurons whose stimulation by glucose evokes electrical responses [19], [20]. Modulating IPC activity is thought to affect insulin release, but changes in circulating insulin levels have not been demonstrated. Kir2.1 encodes an inward-rectifying potassium channel that has been used to silence electrical activity of Drosophila neurons and neuroendocrine cells [21]. We used ‘Geneswitch’ GAL4 to express Kir2.1 in adult IPCs [22], which permits mifepristone-dependent conditional gene expression, and minimizes the effects of insulin perturbation during animal growth and development. In control Ilp21 gd2HF heterozygous flies, circulating Ilp2HF levels were not affected by mifepristone feeding, and were maintained near 200 pM (Figure 3A), about half of the circulating Ilp2HF level in 3 day-old Ilp21 gd2HF homozygous flies, as expected. In subsequent experiments, Ilp2HF levels in Ilp21 gd2HF heterozygous flies were measured. We found that Kir2.1 expression induced by mifepristone feeding in adult IPCs significantly reduced hemolymph Ilp2HF concentration without affecting cell number (Figure 3A and B). These results support the postulated role of ion channel activity in regulating insulin secretion [7], and provide direct evidence that ion channel function may couple IPC activation to circulating insulin levels.
Unlike larval IPCs, adult IPCs are glucose responsive [20]. In humans, GLUT1 is a major glucose transporter of pancreatic β-cells. To further assess carbohydrate sensing in adult IPCs, we used RNAi in the IPCs to knockdown expression of the type-1 glucose transporter Glut1 [23], a gene not previously shown to regulate IPC function. In control experiments, we observed that RNAi knockdown of Ilp2 significantly reduced Ilp2 mRNA expression, total Ilp2HF content in flies, and circulating Ilp2HF levels (Figures 3C–E). Knockdown of Glut1 in IPCs severely reduced circulating Ilp2HF levels, but had no detectable effect on Ilp2 mRNA levels or total Ilp2HF content in flies (Figures 3C–E). Together, these data suggest that Glut1 in Drosophila IPCs is a conserved regulator of in vivo insulin secretion.
To identify uncharacterized regulators of insulin expression, production, and secretion in Drosophila IPCs, we performed loss-of-function analysis of fly genes corresponding to GWAS candidate genes for T2DM [2], [24]. Glis3 was recently shown to be required for insulin expression in mouse islet β-cells [25]. Knockdown of lmd (orthologue of Glis3) in IPCs severely reduced circulating Ilp2HF levels, total Ilp2HF content, and Ilp2 mRNA expression (Figures 3C–E), suggesting that lmd may regulate Ilp2 expression, similar to the role of rodent Glis3 in regulating Ins expression. BCL11A has been associated with type 2 diabetes mellitus [24], but prior work has not linked BCL11A to insulin regulation in mammals. In contrast to lmd, knockdown of CG9650 (orthologue of human BCL11A) increased circulating Ilp2HF levels without affecting Ilp2 mRNA levels or total Ilp2HF content in flies (Figures 3C–E), suggesting that CG9650 may regulate Ilp2HF levels post-translationally.
In mammals, genetic or acquired pathological insulin resistance provokes adaptive responses in β-cells, including enhanced insulin secretion [26], but it was not known if such facultative responses were conserved in Drosophila. Adult flies heterozygous for the InR05545 mutant allele do not have detectable growth [27] or trehalose phenotypes (Figure 4A). Remarkably, we found that circulating Ilp2HF concentration was doubled in InR05545 heterozygotes (Figure 4B). We observed similar phenotypes in flies heterozygous for a loss-of-function mutation in Akt1 (Figures 4A and 4B), which encodes an essential conserved regulator of insulin signaling [28]. To test whether elevated hemolymph Ilp2HF levels in InR05545 heterozygous flies derived from increased production or increased secretion, we measured total Ilp2HF content. Ilp2HF content was identical in InR05545 heterozygous flies and controls (Figure 4C), indicating that hyperinsulimia in InR05545 heterozygotes results from enhanced insulin secretion, not from enhanced insulin production. These results are reminiscent of adaptive phenotypes noted in mice harboring heterozygous mutations in genes encoding insulin receptor or other insulin signaling regulators [29], [30]. The two-fold increase in circulating Ilp2HF in InR05545 heterozygotes represents only a minor fraction of the total Ilp2HF content; thus we asked whether this subtle difference could be detected using previously established methods [7]. We could not detect differences in Ilp2 accumulation in IPCs from InR05545 heterozygotes and control flies by immunofluorescence (Figure 4D), suggesting that quantification of Drosophila insulin by the Ilp2HF system permits the discovery and characterization of phenotypes not detected by semi-quantitative assays of Ilp2 secretion.
To identify the tissue-specific basis of the enhanced insulin secretion phenotypes in heterozygous InR mutants, we systematically knocked down InR expression using RNAi in adult IPCs, muscle, or in fat body, an organ with functions orthologous to the liver. InR knockdown in muscle using Mef2-GAL4 did not detectably alter Ilp2HF levels (Figure 4E), reminiscent of normal serum insulin levels observed in muscle-specific insulin receptor knockout (MIRKO) mice [31]. InR knockdown in IPCs using Ilp215-1-GAL4 decreased circulating Ilp2HF levels (Figure 4E), reminiscent of insulin defects found in pancreatic β cell-specific insulin receptor knockout (BIRKO) mice [32]. In contrast, increased circulating Ilp2HF levels were evoked by RNAi-mediated InR knockdown in the adult fat body using Lk6-GAL4 or ppl-GAL4 (Figure 4E) without detectable effects on total Ilp2HF content in fat body-specific InR knockdown using ppl-GAL4 (Figure 4F). RNAi-mediated InR knockdown in fat body was confirmed by qPCR of fat body cDNAs (Figure S3). These results suggest that insulin secretion, not production, from IPCs is regulated by impaired insulin signaling in fat body. Thus, similar to mice with conditional insulin receptor loss in liver (LIRKO) [33], targeted impairment of insulin signaling in Drosophila fat body produced enhanced insulin secretion from IPCs.
In vivo measures of circulating insulin and other peptide hormones in organisms with powerful experimental advantages, like Drosophila, could transform the scope of physiological and genetic approaches possible in these systems, and advance their use for metabolic and genomic studies. Active insulin is produced from multiple post-translational processing steps, including proteolytic cleavage and extensive disulfide bonding, and modification of a single amino acid in insulin protein can significantly alter or eliminate its hormone activity. Thus, despite intensive efforts, labeling of insulin with useful peptide epitopes while preserving in vivo hormone function has remained a challenge. We exploited quantitative synthetic lethality tests in flies to screen multiple modifications in the Ilp2 protein, and found that Ilp2 tolerated epitope insertions only at specific locations while preserving bioactivity, specifically the HA-epitope at the B-chain carboxy-terminus and the FLAG-epitope at the A-chain amino-terminus. While structural analysis for Ilp2 is not available, to our knowledge, the structure of the related insulin-like peptide Ilp5 has been reported [6], revealing a disordered B-chain carboxy-terminus adjacent to the A-chain amino-terminus. To the extent that similar features may be found in Ilp2, we speculate that this structural feature may be permissive for Ilp2 epitope tagging while preserving function. If so, epitope-tagging methods described here may be used to quantify and investigate function of other processed circulating peptide hormones in Drosophila, or other species. We also found that Ilp2HF bioactivity is impaired by introduction of an “Akita” missense mutation, analogous to mutations previously shown to disrupt post-translational insulin processing in rodents, and in humans with dominant mutant proinsulin syndrome [14], [15]. This raises the likelihood that conserved mechanisms may underlie prepro-Ilp2 processing and folding in Drosophila IPCs, and that the Ilp2–3,5 gd2HF.C119Y line may provide a useful model for studies of protein-folding in Drosophila.
Our studies revealed that Drosophila insulin expression, production, and secretion are dynamic and independently regulated in IPCs. By contrast, intracellular immunofluorescence methods that infer IPC secretion responses do not discriminate between insulin expression, production, and secretion. Moreover we also found that, upon IPC stimulation, only a small fraction of the total Ilp2 in IPCs is secreted in vivo and in vitro. Based on synthesized peptide standards, we found the circulating Ilp2HF concentration increases from 100 pM in 1 day-old flies to 350 pM in 3 day-old flies without a change of total Ilp2 content during this period. Although Ilp2 affinity for the Drosophila insulin receptor has not been reported, competitive binding studies of purified Ilp5 revealed a Kd of 350–760 pM [6], consistent with our in vivo findings. Since distinct Ilps produced in IPCs may be co-released, Ilp2 levels may indirectly reflect release of Ilp3 and Ilp5 from IPCs. While the basis for enhanced Ilp2HF secretion in 3 day old flies is not yet known, feeding behavior may change over this period and underlie this effect. Alternatively, stimulus-secretion coupling mechanisms in IPCs may mature in the first 3 days. Both possibilities have been previously observed during the postnatal weaning and maturation period in mammals. Thus, a scalable and highly sensitive method of measuring insulin content and secretion should enable a new class of physiological studies in Drosophila, permitting genetic dissection of feeding behaviors and diet effects on insulin signaling.
Our system revealed molecular and cellular mechanisms for two fly orthologues of T2DM risk genes in regulating systemic insulin levels. Glis3 was recently shown to be required for insulin expression in mouse islet β-cells [25]. Consistent with this finding, we found that IPC knockdown of lmd, a fly orthologue of human GLIS3, reduced Ilp2 mRNA and total Ilp2HF protein levels, suggesting conserved mechanisms regulating insulin expression. In contrast, we found that CG9650 knockdown in IPCs increased circulating Ilp2HF levels, without affecting Ilp2 expression or production. Thus, the product of CG9650 likely regulates circulating insulin levels at a post-translational step. Prior studies suggested that CG9650 encodes a transcription factor with roles in axon guidance, Notch signaling and oxidative stress responses [34]–[36], but did not identify roles in insulin processing or secretion. Likewise BCL11A, a human orthologue of CG9650, has been associated with type 2 diabetes mellitus, but prior work has not linked BCL11A to insulin regulation in mammals. In addition, our system now permits further studies of circulating signals or neurotransmitters thought to regulate insulin secretion by IPCs, including Ilp6, Unpaired 2, and serotonin [9], [37], [38]. Thus, the ability of our system to measure insulin production and secretion permits mechanistic evaluation and linkage of candidate human diabetes susceptibility genes to roles in insulin expression, post-translational processing, or secretion.
Using our system, we also detected adaptive enhancement of insulin secretion in flies with heterozygous InR or Akt1 mutations. These phenotypes are similar to those reported in mice with IR or IRS deficiency [29], [30], in which impaired insulin signaling in peripheral tissues promotes a ‘pre-diabetic’ condition with adaptive hyperinsulinemia compensating for systemic insulin resistance while maintaining normoglycemia. The detection and quantification of haploinsufficiency phenotypes in heterozygous InR or Akt1 mutants suggests that genetic screens using deficiency lines could identify novel regulators of insulin production and secretion. We also observed changes in circulating Ilp2 levels after specific knockdown of the insulin receptor in Drosophila fat body or IPCs, but not in muscle. These results are consistent with prior reports that fat body signals might regulate IPCs [7], and suggest a role for fat body insulin-signaling in feedback regulation of systemic insulin levels in Drosophila. The changes in circulating Ilp2HF levels after InR knockdown in fly fat body, IPCs or muscle, were remarkably similar to changes in serum insulin observed after tissue-specific knock-out of insulin receptor in mouse liver, pancreatic β-cells or muscle [31]–[33], the so-called LIRKO, BIRKO and MIRKO mice. Thus, integrated analyses permitted by our assays revealed that mechanisms governing facultative adaptation to pathological states like impaired insulin signaling in multiple target organs are maintained from insects to mammals. We speculate that in vivo Ilp2HF quantification in Drosophila should be useful to identify conserved regulators of insulin expression, secretion and responsiveness relevant to human health and diseases.
y1 w1118 (Bloomington stock ID #6598), Ilp21 (#30881), Df(3L)Ilp2–3,Ilp53/TM3 (#30889), InR05545/TM3 (#1161), Act5C-GAL4/CyO (#4414), Mef2-GAL4 (#27390), Lk6-GAL4 (#8614), UAS-CD4-tdTomato (#35841), UAS-Kir2.1-eGFP (#6596), UAS-mCherry.RNAi (#35785; used as a control RNAi), UAS-InR.RNAi (#31594), UAS-Ilp2.RNAi (#31068), UAS-lmd.RNAi (#42871), UAS-CG9650.RNAi (#26713), and UAS-Glut1.RNAi (#40904) used in this study were obtained from Bloomington Stock Center. Drosophila orthologues for human genes were identified by Ensemble release 73. Ilp215-1-GAL4 used in this study is made from pIlp215-1-GAL4 construct (See below), and its adult expression is restricted in IPCs. Akt11/TM3 was provided by Dr. Clive Wilson (University of Oxford). UAS-FLAG-dilp2 [12] was provided by Dr. Matt Scott (Stanford University). ppl-GAL4 [39] was provided by Dr. Michael Pankratz (Universität Bonn). Ilp2-GeneSwitch was provided by Dr. Yih-Woei C. Fridell (University of Connecticut). dilp215-1-HStinger was previously described [40]. In all experiments, animals were either fed on cornmeal/dextrose/yeast food ad libitum, fasted on 1% agar only food, or re-fed on 2M glucose in 1% agar with 0.05% bromophenol blue for oral glucose-stimulated insulin secretion experiments at 22°C. Standard Drosophila cornmeal/dextrose/yeast food was prepared with the recipe: 1% (w/v) agar, 5% (w/v) cornmeal, 10% (w/v) dextrose, and 2.5% (w/v) baker's yeast. Please note that 10% (w/v) dextrose is about 555 mM. Mifeprestone (Sigma-Aldrich M8046) was added at 0.2 mM when needed.
pIlp215-1-GAL4 was generated by subcloning the 541 bp sequence upstream of the Ilp2 transcription start site [4] into pPTGAL. pUAST-Ilp2 was generated by subcloning 705 bp EcoR1-Xho1 fragment from DGC clone GH11579. pUAST-Ilp2HF was generated by PCR-based site-directed mutagenesis to add 5′-TAT CCA TAT GAT GTT CCT GAC TAT GCT-3′ (encoding the amino acids YPYDVPDYA) sequence after the end of Ilp2 B-chain and 5′-GAT TAT AAG GAC GAC GAT GAC AAG-3′ (encoding the amino acids DYKDDDDK) sequence before the beginning of Ilp2 A-chain (See Figure 1A). P-element mediated germline transformations were carried out to generate Ilp215-1-GAL4, UAS-Ilp2, and UAS-Ilp2HF transgenic lines. 2413 bp genomic Ilp2 region was amplified from y1 w1118 genomic DNA using 5′-CCGAGAATTCACACTTGGCCAACACACACACATTCATTA-3′ and 5′-ACTGTCTAGAATTGGCCAACTTGATTGGTAATGAAACGG-3′ primers and subcloned to EcoR1 and Xba1 sites on pBDP2 (a modified version of pBDP with EcoR1, Xba1, and Not1 cloning sites) [41] to generate pBDP2-gd2. pBDP2-gd2HF was generated by replacing Ilp2 coding region in pBPD2-gd2 with Ilp2HF ORF. pBDP2-gd2HF.C119Y was generated by PCR-based site-directed mutagenesis to change from 5′-TGCTGCAA-3′ to 5′-TGTTATAA-3′. phiC31 integrase-mediated germline transformations were carried out to generate gd2(attP2), gd2HF(attP2), and gd2HF.C119Y(attP2) transgenic lines using Bloomington stock #25710. gd2(attP2), gd2HF(attP2), or gd2HF.C119Y(attP2) transgene was recombined to Df(3L)Ilp2–3, Ilp53 mutant backgrounds to assess phenotypic rescue of Ilp2–3, 5 deficiency mutant. To replace endogenous Ilp2 gene with gd2HF(attP2) in the genome, the gd2HF(attP2) transgene was recombined into Ilp21 mutant chromosome to generate the y1 w1118; Ilp21 gd2HF(attP2) strain which was used to measure the circulating Ilp2HF in hemolymph. Please note that the y1 w1118; Ilp21 gd2HF(attP2) strain is homozygous for gd2HF, and their circulating llp2HF levels are 300–400 pM (Figure 2E). To express Kir2.1 in insulin producing cells conditionally, the Ilp2-GeneSwitch/CyO; Ilp21 gd2HF(attP2) dilp215-1-HStinger strain was crossed to flies harboring transgene encoding UAS-Kir2.1-eGFP, and appropriate progeny were fed 200 µM Mifeprestone or vehicle (ethanol) in cornmeal/dextrose/yeast food for 48 hours. The progeny carry only one copy of gd2HF, and their circulating Ilp2HF levels are 100–200 pM (Figure 3A). To knockdown genes in adult IPCs and measure Ilp2HF in hemolymph, TRiP RNAi lines were crossed to the UAS-Dcr-2.D; Ilp21 gd2HF(attP2) Ilp215-1-GAL4 strain. To knockdown genes in adult muscles and measure circulating Ilp2HF in hemolymph, TRiP RNAi lines were crossed to the UAS-Dcr-2.D; Ilp21 gd2HF(attP2) Mef2-GAL4 strain. To knockdown genes in adult fat body tissues and measure circulating Ilp2HF in hemolymph, TRiP RNAi lines were crossed to either the UAS-Dcr-2.D; Ilp21 gd2HF(attP2) Lk6-GAL4 or the ppl-GAL4 UAS-Dcr-2.D; Ilp21 gd2HF(attP2) strain. Progeny from these crosses carry only one copy of gd2HF, and their circulating Ilp2HF level are 100–200 pM (Figure 3E and Figure 4B and E).
One wing per female fly was dissected in isopropanol, mounted in Canada balsam:Methyl salicylate (4∶1) on a slide, and heated on 65°C hot plate for 1 hour to harden the mounting media. The distance between the distal end of the L3 wing vein and the posterior end of wing hinge was measured using AxioVision software. 5 wing spans were measured per genotype, and statistical differences between genotypes were determined with a two-tailed Student's t-test. The results are presented as the mean ± standard deviation.
Immunostaining of adult brains was performed as described [40] with modifications: Affinity purified rabbit polyclonal anti-Dilp2 antibody (0.5 µg/ml), mouse monoclonal anti-FLAG M2 antibody (1 µg/ml; Sigma-Aldrich F1804), Rat monoclonal anti-HA 3F10 antibody (0.1 µg/ml; Roche 1867423), and Alexa Fluor 488, 547, and 647 secondary antibodies (2 µg/ml; Life Technologies) were diluted and incubated in PBS with 0.3% Triton-X100. Confocal laser scanning microscope images were obtained using Leica TCS SP5 or SP8. Accumulation Ilp2 in IPCs was quantified as described previously [7]. Adult brains were dissected and stained for immunofluorescence as described above, and mounted with the IPCs oriented towards the coverslip. Confocal imaging parameters were optimized such that images of all samples could be acquired within the dynamic range of constant laser and scan settings. Confocal Z stacks of the IPCs were acquired with a step size of 1 µm. For quantification, image stacks were summed and the mean-pixel intensity of a region of interest (ROI) containing the entire IPC cluster was measured and subjected to background subtraction using an ROI drawn adjacent to the cell cluster. Average mean pixel intensity of IPCs across biological replicate brains for each condition is expressed in arbitrary units (a.u.).
In all assays, three separate samples per specific fly group or condition were collected. Unless otherwise noted, 3-day-old male flies fed ad libitum were used in all experiments. We avoided using female flies due to possible feeding behavior changes in virgin and mated females [42], and larger observed standard deviations in all metabolic assays we used. All flies were transferred to vials with fresh food 24 hours prior to hemolymph collection to ensure similar nutritional conditions except fasted groups which were maintained on vials with 1% agar for 24 hours prior to the collection. In acute re-feeding and re-fasting experiments, fasted flies were placed on 2M glucose in 1% agar with 0.05% bromophenol blue for 30 minutes, then re-fasted flies. Because not all 24-hour fasted flies commence feeding on 2M glucose, only flies with visibly ingested blue food coloring in their gut were selected for hemolymph sampling. To elute hemolymph, sixty male flies per group were placed in a modified Zymo-Spin IIIC column (Zymo Research Corporation C1006) in which DNA-binding filter were removed and thoroughly washed with water. The column containing male flies was centrifuged twice at 9,000 g for 5 minutes at 4°C. This yielded approximately 1.5 µl hemolymph, which was used for either trehalose assays or ELISA.
A single fly was placed in a 1.5 mL centrifuge tube with 100 µl of PBS containing 1% Triton X-100. Four samples were prepared for each genotype. The samples were ground using a pestle and cordless motor (VWR 47747-370), and lysed at room temperature for 30 minutes on a rotary shaker. The lysed samples were centrifuged at 21,000 g for 5 minutes at room temperature and 50 µl supernatant from the centrifuged samples were used for ELISA.
Adult hemolyph-like saline (AHLS) was prepared with the recipe: 2 mM CaCl2, 3 mM KCl, 8.2 mM MgCl2, 108 mM NaCl, 4 mM NaHCO3, 1 mM NaH2PO4, 3 mM Glucose, and 2% bovine serum albumin. Heads of 3 day-old Ilp21 gd2HF males were carefully separated from bodies in AHLS to maintain foregut and crop. 15 isolated heads were transferred to a centrifuge tube containing 100 µl of AHLS, and allowed to recover for 1 hour at room temperature. Three samples per condition were prepared. The samples were washed 3 times with AHLS, and incubated in 100 µl of AHLS containing either 3 mM or 100 mM KCl for 30 minutes. 100 µl of the incubated AHLS from the samples was saved and 50 µl was used for ELISA. To measure total content, 150 µl of PBS containing 1% Triton X-100 was added to the remaining heads in the tubes, and the samples were ground using a pestle and cordless motor. After 30 minutes lysis at room temperature, the samples were centrifuged at 21,000 g for 5 minutes. 10 µl supernatant from the centrifuged samples was diluted in 90 µl PBS, and 50 µl of the diluted sample were used for ELISA.
We coated wells in Nunc-Immuno modules (Thermo Scientific 468667) with 100 µl of anti-FLAG antibody (Sigma-Aldrich F1804) diluted in 0.2 M sodium carbonate/bicarbonate buffer (pH9.4) to a final concentration of 2.5 µg/ml, then incubated for 16 hours at 4°C. The plate was washed twice with PBS containing 0.2% Tween 20 (PBTw0.2), then coated with 350 µl of PBS containing 2% bovine serum albumin (Fisher Scientific BP1600) for 16 hours at 4°C. The plate was washed three times with PBTw0.2. For circulating Ilp2HF measurement, 1 µl of hemolymph or 1 µl of FLAG(GS)HA peptide standards (DYKDDDDKGGGGSYPYDVPDYAamide, 2412 daltons: LifeTein LLC) at 0.1–10 pg/µl was mixed with 50 µl of PBS containing 1% Triton X-100 and 5 ng/ml anti-HA-Peroxidase 3F10 antibody (Roche 12013819001), vortexed, centrifuged briefly, and transferred to wells on the plate. For total Ilp2HF content measurement, 50 µl of supernatant from single fly lysate or 50 µl of FLAG(GS)HA peptide standards at 5–500 fg/µl was mixed with 50 µl of PBS containing 1% Triton X-100 and 5 ng/ml anti-HA-Peroxidase 3F10 antibody, vortexed, centrifuged briefly, and transferred to wells on the plate. For samples derived from in vitro head assays, 50 µl of collected media or diluted head lysate was used. The wells were sealed with an adhesive film (Thermo Scientific 232698), and incubated in a humidity chamber for 16 hours at 4°C. We removed the samples by aspirating, and washed the wells six times with PBTw0.2. 100 µl of 1-Step Ultra TMB ELISA Substrate (Thermo Scientific 34029) was added to each well and incubated on a rotary shaker for 30 minutes at room temperature. The reaction was stopped by adding 100 µl of 2 M sulfuric acid, and the absorbance at 450 nm (A450) was immediately measured on a SpectraMax M5 (Molecular Devices). To convert concentration to mass in a given volume, we used a molecular weight of 7829 daltons for mature Ilp2HF protein.
1 µl eluted sample from the centrifuged flies was diluted in 9 µl PBS, vortexed, centrifuged briefly, and heated at 70°C for 5 minutes to inactivate endogenous Trehalase. 2 µl of the heated sample was added to 200 µl of Glucose Hexokinase Reagent (Thermo Scientific TR15421) with or without Porcine Kidney Trehalase (1∶1000; Sigma-Aldrich T8778-5UN), incubated at 37°C for 16 hours, and the absorbance at 340 nm (A340) was measured on a SpectraMax M5. The trehalose concentration in the sample was determined by subtracting the glucose concentration from the total sugar concentration.
Four female flies per group with three biological replicates were homogenized in 600 µl of TRIzol Reagent (Life Technologies 15596-018), and total RNA was isolated according to the manufacturer's protocol. To isolate total RNA from larval fat body, we dissected fat body tissues from 6 larva per group. Three biological replicates were homogenized in 600 µl of TRIzol Reagent. Total RNA pellet was resuspended in 30 µl of water. 1 µg of total RNA was treated with DNAse I, heat-inactivated, and reverse transcribed in 10 µl reaction using High Capacity cDNA Reverse Transcription Kit (Applied Biosystems 4368814). 1.5 µl of cDNA was used in a final volume of 15 µl for quantitative PCR reaction (Solaris qPCR Low ROX Master Mix, Thermo Scientific AB-4352/C), and PCR amplification was detected by 7500 Real Time PCR system (Applied Biosystems). Relative expression levels of Ilp2 or Ilp2HF were determined by Applied Biosystems Taqman probe for Ilp2 (Dm01822534_g1). Relative expression levels of InR were determined by Applied Biosystems Taqman probe for InR (Dm02136224_g1). Applied Biosystems Taqman probe for Rpl32 (Dm02151827_g1) was used as the internal control to determine relative expression of Ilp2 and InR.
|
10.1371/journal.pgen.1000159 | Context Differences Reveal Insulator and Activator Functions of a Su(Hw) Binding Region | Insulators are DNA elements that divide chromosomes into independent transcriptional domains. The Drosophila genome contains hundreds of binding sites for the Suppressor of Hairy-wing [Su(Hw)] insulator protein, corresponding to locations of the retroviral gypsy insulator and non-gypsy binding regions (BRs). The first non-gypsy BR identified, 1A-2, resides in cytological region 1A. Using a quantitative transgene system, we show that 1A-2 is a composite insulator containing enhancer blocking and facilitator elements. We discovered that 1A-2 separates the yellow (y) gene from a previously unannotated, non-coding RNA gene, named yar for y-achaete (ac) intergenic RNA. The role of 1A-2 was elucidated using homologous recombination to excise these sequences from the natural location, representing the first deletion of any Su(Hw) BR in the genome. Loss of 1A-2 reduced yar RNA accumulation, without affecting mRNA levels from the neighboring y and ac genes. These data indicate that within the 1A region, 1A-2 acts an activator of yar transcription. Taken together, these studies reveal that the properties of 1A-2 are context-dependent, as this element has both insulator and enhancer activities. These findings imply that the function of non-gypsy Su(Hw) BRs depends on the genomic environment, predicting that Su(Hw) BRs represent a diverse collection of genomic regulatory elements.
| Insulators are conserved genomic elements that define domains of independent transcription. One class of insulators in the Drosophila genome are defined by the binding of the Su(Hw) protein, with the gypsy insulator representing the classic Su(Hw)-dependent insulator. Su(Hw) associates with hundreds of non-gypsy regions distributed throughout the genome that differ in sequence and organization from the gypsy insulator. To gain insights into the role of Su(Hw) in genome organization, we defined the properties of the first non-gypsy Su(Hw) binding region identified, 1A-2. Our studies reveal differences in 1A-2 activity, depending on the context tested. We show that 1A-2 is an insulator in enhancer blocking studies but functions as a transcriptional activator within the natural genomic location. Our findings are reminiscent of properties of binding regions that associate with the vertebrate CTCF protein, which have defined insulator, activator, and repressor functions. Finally, our studies indicate that a noncoding RNA gene may contribute to independent transcriptional regulation in the genome.
| In eukaryotic genomes, neighboring genes often display distinct spatial and temporal patterns of transcription, even though intergenic distances are within the range of enhancer and silencer action. These observations suggest that constraints exist that limit promiscuous interactions between long distance regulatory elements and non-target promoters. Chromatin insulators represent one class of genomic elements that restrict enhancer and silencer action [1]–[5].
Insulators have been identified based on two functional properties. First, insulators prevent enhancer and silencer modulation of a promoter in a position-dependent manner, such that an enhancer or silencer is blocked only when the insulator is located between these elements and a promoter. Second, insulators protect gene expression from positive and negative chromosomal position effects associated with ectopic placement of genes within genomes, an activity referred to as barrier function. Sequences with one or both of these properties have been identified in most eukaryotic genomes and have been implicated in the regulation of diverse cellular processes, ranging from centromere function in yeast to imprinting in mammals [6],[7]. These observations imply that insulators are fundamental components of eukaryotic genomes.
One of the best-characterized insulators resides in the 5′ untranslated region of the Drosophila gypsy retrovirus. This versatile gypsy insulator blocks over twenty enhancers active in different tissues and developmental stages, prevents repressive effects caused by Polycomb group complexes and heterochromatin and protects an origin of DNA replication from chromosomal position effects [2],[5]. The gypsy insulator consists of a cluster of twelve repeats that are bound by the zinc finger Suppressor of Hairy-wing [Su(Hw)] protein [8]. At least three additional proteins are associated with the gypsy insulator, including Modifier of (mdg4) 67.2 (Mod67.2), Centrosomal Protein of 190 kD (CP190) and Enhancer of y2 [E(y)2]. In general, Mod67.2 and CP190 are required for enhancer and silencer blocking by the gypsy insulator, while E(y)2 has been shown to be required only for barrier function [9]–[13].
The Su(Hw) protein associates with hundreds of non-gypsy regions in the Drosophila genome that have a largely unknown function. The extensive co-localization of the four gypsy insulator proteins at non-gypsy regions has led to the proposal that these represent chromatin insulators. Yet, non-gypsy Su(Hw) binding regions are different in sequence and organization from the gypsy insulator, with the majority of BRs containing single Su(Hw) binding sites (BSs) [14]–[18]. This observation is striking, as at least four tightly spaced Su(Hw) sites from the gypsy insulator were required for robust enhancer blocking [19]–[21]. Direct tests of the non-gypsy BRs in transgene assays show that most, but not all, interfere with enhancer-activated transcription [15]–[18]. These findings imply that non-gypsy regions contain additional elements that assist the insulator function of Su(Hw).
The first non-gypsy Su(Hw) BR identified, named 1A-2, is a cluster of two Su(Hw) BSs located in cytological region 1A (Figure 1). Here we investigated the properties of 1A-2, using two strategies. First, we employed a quantitative transgene system to define the 1A-2 sequences required for enhancer blocking. Second, we performed homologous recombination to establish lines carrying a deletion of 1A-2 at the natural genomic location, representing the first deletion of a non-gypsy Su(Hw) BR in the Drosophila genome. Effects of the loss of these sequences on gene expression in the 1A region were determined, leading to the discovery that 1A-2 contributes to transcriptional activation of a novel, non-coding RNA gene. Taken together, our studies demonstrate that 1A-2 has both activator and insulators properties, depending on the context tested. These findings imply that properties of non-gypsy Su(Hw) BRs are influenced by the genomic environment, predicting that Su(Hw) BRs represent a diverse collection of elements with distinct regulatory functions.
The Su(Hw) BR 1A-2 is a 520 bp element that contains two Su(Hw) BSs [18] and a CP190 BS [9]. Previous studies using qualitative transgene assays demonstrated that 1A-2 blocked enhancer-activated transcription in a position-dependent manner, a key feature of insulator activity [17],[18]. We employed the quantitative Fat Body Enhancer (FBE)1-yolk protein (yp)2 -LacZ transgene to define DNA sequences required for 1A-2 enhancer blocking (Figure 1), a system previously used to characterize properties of the gypsy insulator [20],[22]. A reporter transgene was constructed wherein full length 1A-2(520) was inserted between FBE1 and the yp2 promoter. Multiple P[F-1A-2(520)-yp2] transgenic lines with single insertions were established. Quantitative β-galactosidase activity assays were completed to define the level of yp2 promoter activity. Protein extracts were isolated from adult females representing several independent lines, and multiple assays were undertaken to establish an average activity unit (aau) for each transgene (Figure 1). We found that transgenic P[F-1A-2(520)-yp2] females had low levels of yp2 expression (aau 0.86), similar to levels in P[F-gyp-yp2] females (aau 0.75) and significantly lower than levels found in the control P[FBE1-yp2] females (aau 5.97). We conclude that 1A-2 blocks FBE1, extending the enhancer blocking effects of 1A-2 to a new enhancer-promoter pair.
The minimal sequences required for 1A-2 insulator function were determined by generation of transgenic lines carrying transposons with insertion of subregions of 1A-2 between FBE1 and yp2-LacZ (Figure 2). P[F-1A-2(157)-yp2] females showed a strong enhancer block (aau 0.62). As this subregion lacks the CP190 BS [9], these findings indicate that direct CP190 binding is not required for insulator function. 1A-2(157) was further divided into two parts, one containing the two Su(Hw) BSs, 1A-2(79), and one containing the remaining sequences, 1A-2(78). Transgenic P[F-1A-2(79)-yp2] females showed a two-fold weaker enhancer block than 1A-2(157) (aau 1.29, P = 0.02), whereas P[F-1A-2(78)-yp2] females showed high yp2 activity levels, close to those obtained for the control P[F-yp2] females (aau 5.9 versus 5.97). These data suggest that 1A-2(78) contributes to the blocking effectiveness of the 1A-2 Su(Hw) BSs, but cannot itself block enhancer-promoter interactions.
We considered two possibilities to account for the contributions made by 1A-2(78). First, these sequences might contain a binding site(s) for a second insulator protein that cooperates with the Su(Hw) BSs for insulator function. Second, 1A-2(78) might improve the activity of the Su(Hw) BSs, perhaps by increasing in vivo association. We reasoned that if 1A-2(78) contained a binding site for a novel insulator protein, then insulator effects might require a reiterated element, as observed previously when individual binding sites for other insulator proteins were tested [23],[24]. To this end, we generated P[F-1A-2 (78×4)-yp2] that carried four copies of 1A-2(78) inserted between FBE1 and the yp2 promoter. Surprisingly, these transgenic females had higher yp2 activity than the control P[F-yp2] females (aau 18.78 versus 5.97 aau, P = 6.3×10−8). Transgenic P[F-1A-2(78×4)-yp2] males showed no yp2 activity (data not shown). Based on the retained transcriptional specificity of the P[F-1A-2 (78×4)-yp2] transgene, we conclude that 1A-2(78) is not a general transcriptional enhancer but improves the activity of FBE1. These data imply that 1A-2(78) may possess a general activity that facilitates factor association. To test this postulate, we determined whether 1A-2(78) restored enhancer blocking to a synthetic Su(Hw) BR containing three reiterated gypsy BSs (3R:3) that was previously shown to be inactive in this transgene system [20]. Supporting a facilitator function of 1A-2(78) we found that transgenic P[F- 3R:3-1A-2(78)-yp2] females had low yp2 activity (aau 0.22). These studies show that in the presence of 1A-2(78), 3R:3 became a strong insulator. As previous findings suggest that the effectiveness of enhancer blocking by the Su(Hw) protein is limited by the in vivo accessibility of Su(Hw), we conclude 1A-2(78) is a facilitator that may improve transcription factor binding to chromosomes.
As a first step in defining the role of 1A-2 within the y-ac region, we evaluated whether the existing annotation reflected the transcriptional potential of this region. These analyses were motivated by the recent studies showing widespread transcription in intergenic regions of the Drosophila genome [25]. A search of the NCBI databases uncovered a small, novel, processed EST of ∼400 nt that was transcribed from the y-ac intergenic sequences. Sequences corresponding to this EST are located ∼1.4 kb downstream of the y termination signal and transcribed in the same direction as the y and ac genes. Northern analyses of embryonic polyA+ RNA using a radiolabeled probe representing the intergenic EST identified a family of RNAs, with the most abundant species sized at ∼1.6 kb (Figure 3). Accumulation of these RNAs began ∼7 hours after the start of embryogenesis, in agreement with the expression profile obtained using tiling array studies of embryonic RNAs [25]. These data suggest that the y-ac intergenic region contains a previously uncharacterized gene, which we call yar, for y-ac intergenic RNA. Activation of genes in the 1A locus is temporally in an order following chromosomal position, such that ac, then yar and then y is transcribed.
The structure of the yar RNAs was defined using rapid amplification of cDNA ends (RACE, Figure 4). Sequence analysis of the 5′ RACE products revealed three discrete transcription start sites within an ∼200 bp region, with the most distal RNA starting ∼1.2 kb downstream of the y gene. Each putative start site showed weak homology to Drosophila transcriptional control elements [26], with two having a partial match to the TATA consensus sequence located 17 to 35 bp upstream of the start site. Sequence analysis of the 3′ RACE products identified multiple splice variants, each ending in a common exon that contained an unconventional polyadenylation signal sequence AAATACA, previously estimated to be present in ∼3% of Drosophila genes [27], that was located 12 bp upstream of the string of As in the RACE products. Predicted translation of the yar RNAs indicated that no transcript would encode a protein of more than 75 amino acids, implying that yar is a non-coding RNA gene.
Ends out gene targeting was used to delete 1A-2 from the y-ac region (Figures 4, 5). Gene targeting is a two step processes that requires establishment of transgenic flies that carry a transposon with the replacement gene, followed by the introduction of endonucleases to stimulate homologous recombination between the replacement gene and its endogenous homologue. To delete 1A-2, we constructed P[yΔ1A-2 target]. This transposon carried a modified y gene, wherein 1A-2 was replaced by the hypomorphic whs gene that was flanked by loxP sites (Figure 5). Transgenic lines were established in a y1 w1118 background, where the endogenous y gene carried a mutation of the translation start codon, and the endogenous w gene carried a deletion of the promoter. P[yΔ1A-2 target] flies had orange eyes and dark pigmentation of all cuticle structures except the wing, as the y gene lacked the wing enhancer. To stimulate recombination, transgenic y1 w1118; P[yΔ1A-2 target] males were crossed to females carrying the heat shock (hs)-FLP recombinase and the hs-I-SceI endonuclease transgenes and progeny of this cross were heat shocked to produce the endonucleases. Over 100 resulting females were crossed to y1 w1118 males and homologous recombinants were identified among the offspring of this cross in two ways. First, flies were screened for dark wings, as recombination at the endogenous y1 gene would reconstitute a wild type y transcription unit with all enhancers, whereas progeny with ectopic insertions of the replacement y gene would produce flies with lightly colored wings due to the absent wing enhancer. Second, we conducted genetic analyses to determine whether the w+ phenotype was linked to the X chromosome. Five putative homologous recombination lines were established based on dark wing pigmentation. Further genetic analyses showed that in one line, XGL339-23-38, the w marker mapped to the X chromosome, suggesting a correct targeting event. Southern analyses confirmed the structure of the y gene in these flies (Figure S1). This targeted allele was named, yΔ1A-2w.
We reasoned that if 1A-2 was an insulator in the y-ac locus, then deletion of 1A-2 would release constraints on the y and ac enhancers, causing changes in gene expression that would alter cuticle pigmentation and bristle number in yΔ1A-2w relative to wild type flies [28],[29]. However, we found that adult phenotypes of yΔ1A-2w flies were indistinguishable from wild type flies. In yΔ1A-2w, the whs gene replaced 1A-2. To rule out the possibility that this gene served as a surrogate insulator by carrying a promoter that captured the y and ac enhancers, yΔ1A-2w flies were crossed to flies carrying a source of Cre recombinase to remove the whs gene. Southern and PCR analyses confirmed the structure of y gene in yΔ1A-2 flies (Figure S1). Again, the cuticle and bristle phenotypes of the yΔ1A-2 flies were indistinguishable from wild type. Taken together, these data imply that 1A-2 is not an insulator at the endogenous genomic location.
Within the y-ac intergenic region, we identified a second cluster of Su(Hw) binding sites, which we called 1A-2′. These sites differ from the Su(Hw) consensus sequence at multiple highly conserved positions (Figure 4). Electrophoretic mobility shift assays demonstrated that 1A-2′ had ∼3-fold lower affinity for Su(Hw) than 1A-2 (data not shown). Even so, we considered it possible that weaker 1A-2′ Su(Hw) BR might provide a redundant function with 1A-2 to define regulatory interactions in the y-ac region. For this reason, we generated a second targeting vector, P[yΔ1A-2/Δ1A-2′ target], wherein the whs gene replaced an ∼1.0 kb deletion that encompassed both 1A-2 and 1A-2′. Following the procedure described above, six putative homologous recombinant lines were identified based on dark wing pigmentation. Further genetic analyses showed that one of these lines, XGL426-41-4, had marker linkage to the X chromosome. This allele was named yΔ1A-2/Δ1A-2′w. Flies from this line were used to obtain a derivative line lacking the whs gene, producing yΔ1A-2/Δ1A-2′. Southern and PCR analyses confirmed the structure of the y gene resulting from these targeting events (Figure S1). Comparison of adult phenotypes in yΔ1A-2/Δ1A-2′ and wild type flies showed that the cuticle color and bristle number were indistinguishable, suggesting that 1A-2′ did not compensate for 1A-2.
Prevailing models of gypsy insulator function predict that the gypsy insulator establishes independent transcriptional domains through cooperation with genomic insulators defined by non-gypsy Su(Hw) BRs. Recent findings indicate that the sequence and organization of non-gypsy BSs differ from the Su(Hw) BR in the gypsy retrovirus [14]–[16]. These observations imply that properties of non-gypsy BRs may be distinct from those of the gypsy insulator. We defined the properties of 1A-2, to gain insights into mechanisms of Su(Hw) insulator action.
We used the quantitative FBE1-yp2-LacZ reporter system to define the sequence requirements for enhancer blocking by 1A-2(520). Prior application of this system demonstrated that at least four gypsy Su(Hw) sites were needed for robust blocking [20]. Here, we show that 1A-2(157) provided as strong an enhancer block as the gypsy insulator (Figures 1, 2). A fragment containing only the Su(Hw) BRs [1A-2(79)] reconstituted a weaker enhancer block than 1A-2(157), but had a greater blocking capacity than the synthetic insulators made from reiterated copies of BS3 of the gypsy insulator [20]. While we do not know the reason for the more robust blocking, we note that these regions differ in sequence and distance of separation from Su(Hw) sites (Figure 4). Blocking effectiveness does not appear to be due to differences in DNA recognition, as the in vitro binding constants for Su(Hw) for the 1A-2 and gypsy BSs are similar [16]. Our experiments revealed that 1A-2 contains a second regulatory element located in 1A-2(78). When these sequences were positioned next to the inactive, synthetic Su(Hw) BR (3R:3), a functional insulator was reconstituted (Figure 2B). These data are consistent with previous findings that Su(Hw) chromosome association is limited [32]. Taken together, we propose that 1A-2 is a composite insulator that contains an enhancer blocking and a facilitator function that may improve Su(Hw) chromosome association. Further, we predict that in vivo effectiveness of enhancer blocking by the Su(Hw) protein is related to the accessibility of Su(Hw) BSs. If single or small clusters of Su(Hw) BSs are located in genomic regions of open chromatin, then these regions will demonstrate enhancer blocking, as defined in transgene assays. This proposal implies that genomic context greatly influences the properties of non-gypsy Su(Hw) BRs.
1A-2 is located between the independently regulated y and ac genes. Chromatin immunoprecipitation studies demonstrated that 1A-2 is associated with Su(Hw), Mod67.2 and E(y)2 in vivo [12],[16],[18], suggesting that this element binds a complex competent for establishing a genomic insulator. Based on these properties, we postulated that 1A-2 was responsible for the regulatory independence of the y and ac genes in the 1A locus [16]. As a first step in testing this proposal, we investigated transcription in the y-ac region to evaluate the current accuracy of the genomic annotation of this region. These studies identified a previously unannotated gene, yar, located ∼1.2 kb downstream of the y gene and ∼3.0 kb upstream of ac. Multiple, differentially spliced, polyA+ RNAs are encoded by yar, with the largest translation product predicted to be 75 amino acids, indicating that this is a non-coding RNA gene. Emerging data suggest that non-coding RNAs are abundant in eukaryotes and have a wide repertoire of biological functions, ranging from structural components in protein complexes to regulatory molecules involved in transcription and translation [33]–[35]. It is unknown whether yar has a function. As flies carrying a large genomic deletion that removes sequences upstream of y and extends downstream of ac (y− ac−) are viable and fertile, yar is a non-essential gene.
Having re-defined the transcriptional profile in the 1A locus, we tested the function of 1A-2 and a second, weaker Su(Hw) BR, 1A-2′, on gene regulation, using gene targeting to delete these elements. Our studies represent the first deletional analysis of any non-gypsy Su(Hw) BR in the Drosophila genome. Two targeted deletion lines, yΔ1A-2 and yΔ1A-2/Δ1A-2′ were established (Figure 4). Levels of y, ac, sc and yar RNA accumulation during development were studied using quantitative PCR. We find that loss of 1A-2 and 1A2′ has no effect on the timing and level of y, ac or sc RNAs relative to the wild type control (Figure S3), but strongly reduced yar RNA (Figure 6). These data suggest that the effects of loss of 1A-2 are limited to local changes of gene expression, implying that these sequences are not a chromatin insulator at the endogenous location. Instead, our data indicate that 1A-2 may be an activator of yar expression, consistent with other studies that have suggested a role for Su(Hw) in gene activation [36]–[38]. These data, coupled with genetic studies on the effects of the loss of Su(Hw) on expression of genes adjacent to Su(Hw) BRs [16], demonstrate that Su(Hw) BRs have diverse functions in the genome.
The complexity of the transcriptional effects associated with Su(Hw) BRs is reminiscent of regions in mammalian genomes that bind the versatile regulatory protein CTCF. High throughput genomic analyses have identified hundreds of CTCF binding sites within the mouse and human genomes [7], [39]–[41]. Although many of these sequences possess enhancer blocking activity [39],[42],[43], CTCF has been implicated in transcriptional activation [44]–[46], repression [47]–[50], and chromosome pairing [44],[51],[52]. These observations suggest that, similar to the non-gypsy Su(Hw) BRs, genomic context will have an important influence on the properties of CTCF BSs within a given region.
The mechanism(s) used to maintain transcriptional autonomy in the 1A locus are unclear. The discovery of yar provides an alternative explanation to the need for a chromatin insulator. Based on the developmental timing displayed by the 1A genes, we postulate that activation of yar transcription may cause inactivation of ac through transcriptional interference. Similarly, activation of y may repress yar transcription. Although yΔ1A-2 and yΔ1A-2/Δ1A-2′ flies show reduced yar expression, transcription is not abolished, suggesting that the remaining yar activity may be sufficient to turn off ac. Alternatively, other mechanisms can be considered that might influence enhancer preference, including selectivity of enhancers for certain classes of promoters [53],[54], the presence of promoter targeting sequences that direct enhancer action [55],[56], or promoter tethering elements that capture enhancers [57]. Further experiments to define the properties of DNA elements within the 1A locus will resolve how transcriptional independence is achieved.
Flies were raised at 25°C, 70% humidity on standard corn meal/agar medium. Description of the alleles used can be found at http://flybase.bio.indiana.edu.
The FBE1-yp2 -LacZ fusion gene [20] carried a BglII site, positioned at −335 relative to the transcription start site (TSS) that was used for insertion of tested 1A-2 fragments. Resulting transgenes were inserted into a P element transformation vector, generating P[F-1A-2(520)-yp2] with the full length 1A-2, P[F-1A-2(157)-yp2] with a 157 bp region of 1A-2, P[F-1A-2(79)-yp2] with two 1A-2 Su(Hw) binding sites, P[F-1A-2(78)-yp2] with the 78 bp 3′ region, P[F-1A-2(78×4)-yp2] with four tandem repeats of the 1A-2 78 bp element and P[F-3R:3(78)-yp2] with a hybrid insertion between a cluster of three tandem repeats of the gypsy Su(Hw) binding sites [nucleotides 732–759 [58]], as described in [20] and the 78 bp element. P transposons were injected into the host y1w67c23 strain or w1118 (Genetic Services, Inc, Cambridge, MA). Transgenic lines were analyzed by Southern and PCR analyses to determine the number and integrity of the transposons. Lines with single transposon insertions were used in subsequent analyses.
The yp2 promoter activity was assessed using quantitative β-galactosidase assays, performed essentially as previously described [20]. Each transgenic line was assayed using extracts isolated from three different matings. Each extract was assayed in duplicate, and the error between these samples was less than 10%. Average promoter activity and standard deviation were determined using the statistical analysis feature of the Microsoft Excel program.
Two targeting transposons were constructed for gene targeting, using pW25 [59]–[61]. This vector has multi-cloning site, NotI-SphI-Acc65I-Stop-lox-whs-lox-Stop-AscI-BsiWI. The lox sites are in direct orientation, permitting removal of the whs transformation marker by Cre recombinase. P[yΔ1A-2 target] (XGL339) was used to target an ∼0.43 kb deletion encompassing 1A-2 alone, whereas P[yΔ1A-2/Δ1A-2′ target] (XGL426) was used to target an ∼1.03 kb deletion that included 1A-2 and 1A-2′. These targeting transposons were generated in a two-step procedure. First, a 6.6 kb yellow fragment (−1842 to +4796 relative to the yTSS) was PCR amplified, using primers carrying the BsiWI and AscI sites and cloned into pW25 to make XGL235. This fragment contains the yellow transcription unit and the body enhancer, but lacks the wing enhancer. Second, PCR primers containing NotI sites generated a 3 kb fragment (y+5234 to y+8184 relative to the yTSS) to make P[yΔ1A-2 target] or a 3.5 kb fragment (y+5826 to y+9318 relative to the yTSS) to make P[yΔ1A-2/Δ1A-2′ target]. In all cases, PCR fragments were sequenced to confirm appropriate amplification. For targeting, we generated transgenic lines in a y1 w1118 mutant background. Gene targeting followed the procedure outlined in [59]. A combination of Southern and PCR analyses identified correctly targeted events. To remove the whs gene, red-eyed males carrying a targeted deletion event were crossed to females carrying Cre recombinase, as described in [62]. The white-eyed flies were collected and used to establish homozygous stocks. Deletion events were confirmed by PCR amplification and sequence analyses.
The structures of the yar RNAs were determined using RACE of total RNA isolated from 6–12 hour CS embryos. In the 3′-RACE experiments, 5 µg of RNA were reverse transcribed using the adaptor oligo-dT primer (3′-RACE kit, Invitrogen), and cDNA was amplified using a yar specific primer (1 µM) and the abridged universal primer (80 nM, Invitrogen). Several products were identified by agarose gel electrophoresis, gel purified and cloned into the TOPO vector (Invitrogen). Sequencing and BLAST search identified three yar splice variants that shared a common distal exon and poly-A signal. In the 5′-RACE experiments, 5 µg of RNA were reverse transcribed with a yar specific primer (100 nM), purified over a S.N.A.P column (Invitrogen) to remove unincorporated nucleotides and primers, and C-tailed at 4° for 2 hours, using terminal deoxynucleotidyl transferase. Tailed cDNAs were amplified with nested yar specific primers (400 nM) and an abridged anchor primer (400 nM, Invitrogen). PCR products were directly cloned into the TOPO vector. Forty-eight clones were analyzed by restriction digestion, revealing nine classes of insert. At least one representative of each class was sequenced. BLAST analyses of these data identified ten alternative splice variants and three alternative start sites. Both the 3′-RACE and 5′-RACE were performed on two independent RNA isolations. Gene-specific primer sequences are available upon request.
RNA was isolated from staged embryos collected from cages of wild type (CS) flies, using the NaDodSO4/phenol technique [63]. Five µg of oligo-dT selected polyA+ RNA was used in northern analyses and hybridized with radiolabeled fragments corresponding to y (a ClaI-BglII fragment, representing +2466 to +4815 relative to the yTSS), yar (EST DN154052, 418 bp ) and ac (a PCR fragment representing +115 to +531 relative to the acTSS). Hybridization with sequences corresponding to the ribosomal gene, RpL32, served as a loading control. For real-time PCR experiments, RNA was isolated from embryos and pupae from three lines: CS, yΔ1A-2 line XGL339-23-38, yΔ1A-2/Δ1A-2′ line XGL426-41-4. RNA isolation and real-time PCR analyses were performed as described in [16]. PCR primers amplified 100–200 bp fragments. y primers flanked the intron. yar primers were in the invariant fourth exon, to ensure quantification of all transcripts. Primer sequences are available upon request. Duplicate or triplicate reactions were performed and averaged, with the difference among the replicates no greater than 0.5 cycle threshold (CT). At least three independent experiments were performed for each primer set from two independent RNA samples. The expression level of each gene was determined using Ras64B as an internal control (ΔCT). The fold change in expression of each gene relative to the wild type (CS) value was determined with the ΔΔCT method.
|
10.1371/journal.pbio.1001376 | The Germinal Center Kinase TNIK Is Required for Canonical NF-κB and JNK Signaling in B-Cells by the EBV Oncoprotein LMP1 and the CD40 Receptor | The tumor necrosis factor-receptor-associated factor 2 (TRAF2)- and Nck-interacting kinase (TNIK) is a ubiquitously expressed member of the germinal center kinase family. The TNIK functions in hematopoietic cells and the role of TNIK-TRAF interaction remain largely unknown. By functional proteomics we identified TNIK as interaction partner of the latent membrane protein 1 (LMP1) signalosome in primary human B-cells infected with the Epstein-Barr tumor virus (EBV). RNAi-mediated knockdown proved a critical role for TNIK in canonical NF-κB and c-Jun N-terminal kinase (JNK) activation by the major EBV oncoprotein LMP1 and its cellular counterpart, the B-cell co-stimulatory receptor CD40. Accordingly, TNIK is mandatory for proliferation and survival of EBV-transformed B-cells. TNIK forms an activation-induced complex with the critical signaling mediators TRAF6, TAK1/TAB2, and IKKβ, and mediates signalosome formation at LMP1. TNIK directly binds TRAF6, which bridges TNIK's interaction with the C-terminus of LMP1. Separate TNIK domains are involved in NF-κB and JNK signaling, the N-terminal TNIK kinase domain being essential for IKKβ/NF-κB and the C-terminus for JNK activation. We therefore suggest that TNIK orchestrates the bifurcation of both pathways at the level of the TRAF6-TAK1/TAB2-IKK complex. Our data establish TNIK as a novel key player in TRAF6-dependent JNK and NF-κB signaling and a transducer of activating and transforming signals in human B-cells.
| The germinal center kinase family member TNIK was discovered in a yeast-two-hybrid screen for interaction partners of the adapter proteins TRAF2 and Nck, and here we show it is one of the missing molecular players in two key signaling pathways in B-lymphocytes. We found that TNIK is crucial for the activities of the CD40 receptor on Bcells and its viral mimic, the latent membrane protein 1 (LMP1) of Epstein-Barr virus (EBV). EBV is a human DNA tumor virus that is associated with various malignancies. It targets and transforms B-cells by hijacking the cellular signaling machinery via its oncogene LMP1. In normal Bcell physiology, the CD40 receptor is central to the immune response by mediating B-cell activation and proliferation. TNIK turns out to be an organizer of the LMP1- and CD40-induced signaling complexes by interacting with the TRAF6 adapter protein, well known for its role in linking distinct signaling pathways. Through this mechanism the two receptors depend on TNIK to activate the canonical NF-κB and JNK signal transduction pathways, which are important for the physiological activation of B-cells (a process that enables antibody production), as well as for their transformation into tumor cells. TNIK thus constitutes a key player in the transmission of physiological and pathological signals in human B-cells that might serve as a future therapeutic target against B-cell malignancies.
| TNIK was discovered in a yeast-two-hybrid screen for interaction partners of the adapter proteins TRAF2 and Nck [1]. The serine/threonine kinase TNIK is a member of the germinal center kinase (GCK) family, which belongs to the Ste20 group of kinases [2]. GCKs share high sequence homology in their N-terminal kinase and C-terminal germinal center kinase homology (GCKH) domains, while the intermediate domain is less conserved [2]. Current knowledge about the molecular and biological functions of TNIK is very limited. TNIK overexpression modulates the actin cytoskeleton and activates the JNK pathway, which is induced through the GCKH domain by a yet undefined mechanism [1],[3]. The molecular function of TNIK's interaction with TRAF molecules is unclear. A recent study suggested that TRAF2 and TNIK might be located within one signaling pathway that leads to Wnt pathway induction in chronic myelogenous leukemia stem cells [4]. TNIK also mediates proliferative Wnt signals in crypts of the small intestine and colorectal cancer cells by nuclear translocation and subsequent phosphorylation of the transcription factor TCF4 [5],[6]. In neurons, TNIK is involved in the regulation of neurite growth and neuronal structure [7],[8]. However, a physiological role for TNIK in hematopoietic cells has not been described.
The latent membrane protein 1 (LMP1) of Epstein-Barr virus (EBV) serves as proto-type of a viral receptor-like oncoprotein that usurps cellular signal transduction pathways for cell transformation. The gamma-herpesvirus EBV, classified as a human DNA tumor virus by the WHO, establishes a chronic latent infection in B-cells and is associated with various malignancies, such as Hodgkin's and Burkitt's lymphoma, life-threatening post-transplant lymphoproliferative disorders, or nasopharyngeal carcinoma [9]. LMP1 is found expressed in most EBV-associated tumors and it is crucial for viral cell transformation and continued in vitro proliferation of latently EBV-infected B-cells, so-called lymphoblastoid cell lines (LCLs) [9]. LMP1 resembles a constitutively active cellular receptor whose ligand-independent signaling activity is attributable to spontaneous homo-oligomerization of LMP1 molecules within the membrane [10]. By the recruitment of TRAF molecules, LMP1 mimics molecular functions of the CD40 receptor in B-cell activation and proliferation. However, compared to CD40, LMP1 assembles a unique and more efficient signaling complex [11]–[14]. Constitutive expression of LMP1 in the B-cell compartment of transgenic mice induces lymphomas, whereas timely activation of LMP1 signaling largely substitutes for CD40 deficiency in B-cell development and function [15]–[17].
LMP1 consists of a short N-terminal domain (amino acids 1–24), six transmembrane helices, and a C-terminal cytoplasmic signaling domain (amino acids 187–386) (Figure 1A). The signaling domain harbors the two functionally distinct C-terminal activating regions (CTAR) 1 and 2, which activate the NF-κB, c-Jun N-terminal kinase (JNK), MAPK, PI3-kinase, and IRF7 signaling cascades [18]–[20]. The consensus TRAF binding motif P(204)xQxT is essential for CTAR1 function and directly binds TRAF1, 2, 3, and 5 [13],[21]–[23]. CTAR1 triggers the non-canonical NF-κB pathway involving IκB kinase α (IKKα)-induced processing of the NF-κB precursor p100 to p52 [24]–[27].
CTAR2 (amino acids 351–386) activates JNK and IκB-dependent canonical NF-κB, which contribute critical anti-apoptotic and proliferative signals for survival, proliferation, and tumorigenicity of EBV-transformed B-cells [25],. The TNFR1-associated death domain protein (TRADD) interacts with the 16 C-terminal amino acids of CTAR2 and is involved in NF-κB signaling by facilitating IKKβ recruitment to CTAR2 [34]–[36]. TRAF6 is essential for canonical NF-κB and JNK activation by CTAR2, although direct binding of TRAF6 to LMP1 has not been demonstrated [33],[37],[38]. Interaction of both proteins might thus be indirect involving the transcription factor BS69 as a mediator and/or stabilizer [39]. Downstream of TRAF6, the E2 ubiquitin-conjugating enzyme Ubc13, the TGFβ-receptor-associated kinase 1 (TAK1), and the TAK1-binding protein 2 (TAB2) as well as IKKβ and IKKγ play important roles in CTAR2 signaling [33],[37],[40]–[42].
Apart from LMP1, TRAF6 also mediates canonical NF-κB and JNK signaling by cellular receptors such as CD40 or Toll-like receptors [43]. Current concepts of TAK1 and IKKβ activation by TRAF6 have been reviewed [43]–[45]. In brief, activated and K63-autoubiquitinated TRAF6 binds TAB2, which then mediates the recruitment of the MAP3kinase TAK1 to TRAF6. TRAF6-derived unanchored ubiquitin chains bind TAB2 and help to induce TAK1 [46]. Activated TAK1 phosphorylates MKK6 to upregulate the JNK pathway. TAK1 also phosphorylates IKKβ within its activation loop and IKKβ activation is further facilitated by interaction of its regulatory component IKKγ with TRAF6. However, IKKβ is also induced by a TAK1-independent mechanism [44],[46]. IKKβ phosphorylates IκB, which results in IκB degradation and the release of active p65/p50 NF-κB dimers to the nucleus. It is tempting to speculate that yet unknown factors might serve as additional organizers or scaffolding proteins for TRAF-TAK-IKK complexes within the cell to orchestrate NF-κB and JNK signaling.
It is still not fully understood how the signaling complex at CTAR2 of LMP1 is assembled and how activation of transforming downstream signals is conveyed. We hypothesized the existence of still undefined molecular players and set out to identify novel LMP1 interaction partners by a functional proteomics approach. We report the characterization of TNIK as a component of the LMP1 signaling complex in EBV-transformed human B-cells. TNIK has a critical role in LMP1-induced JNK and canonical NF-κB signaling by the formation of an activation-induced complex at LMP1 with TRAF6, TAK1/TAB2, and IKKβ. Accordingly, TNIK is required for proliferation and survival of lymphoblastoid cells. TNIK is also of critical importance for physiological activation of the two pathways in B-cells by the CD40 receptor. Taken together, we identified TNIK as a novel key player in TRAF6-dependent JNK and NF-κB activation by two members of the TNF receptor family.
We set out to identify novel interaction partners of the LMP1 signaling complex in its native context, the EBV-transformed primary human B-cell. To this end, HA-LMP1-liTEV-CT, an LMP1 variant optimized for proteomics studies, was expressed from a recombinant maxi-EBV genome in lymphoblastoid cells. To generate HA-LMP1-liTEV-CT, an N-terminal hemagglutinin (HA)-tag was added and a tobacco etch virus protease cleavage site coupled to a flexible linker (liTEV) was inserted between the transmembrane domain and the C-terminal (CT) signaling domain of LMP1 (Figure 1A). TEV protease cleavage after immunoprecipitation of the HA-LMP1-liTEV-CT complex allowed the release of the LMP1 signaling domain and its interaction partners from the beads for further analysis by mass spectrometry. By this approach the background of proteins was reduced which either interacted with the LMP1 N-terminus and/or the transmembrane domain or which unspecifically bound to the beads or antibodies.
Recombinant EBV expressing HA-LMP1-liTEV-CT from the viral LMP1 promoter was used to infect primary B-cells isolated from human adenoids. The recombinant virus efficiently transformed B-cells into lymphoblastoid cells, which showed typical clumpy LCL growth and green fluorescence due to the expression of a green fluorescence protein (GFP) marker gene located on the recombinant virus genome (Figure 1B). The clone LCL-TEV.5 was used for proteomics studies. The outgrowth of LCL-TEV cells further proved that HA-LMP1-liTEV-CT was fully functional because an intact LMP1 is mandatory for B-cell transformation by EBV [11]. Moreover, HA-LMP1-liTEV-CT was able to induce signaling as wildtype LMP1 in HEK293 cells (Figure S1).
HA-LMP1-liTEV-CT was immunoprecipitated from lysates of LCL-TEV.5 cells (Figure 1C). Parallel precipitations were performed with the lymphoblastoid cell lines LCL 721 expressing wildtype LMP1 and LCL3 expressing HA-tagged LMP1 [35]. Expression levels of the LMP1 proteins were comparable in all three cell types (Figure 1C). HA-LMP1-liTEV-CT and HA-LMP1 were efficiently immunoprecipitated by anti-HA antibodies. TEV protease cleavage released the signaling domain of HA-LMP1-liTEV-CT but not that of HA-LMP1 (Figure 1C). The known CTAR1 interaction partner TRAF3 verified functionality of the experimental system. As expected, TRAF3 specifically co-precipitated with both HA-tagged LMP1 variants but was only detected in the TEV eluate of LCL-TEV.5 immunoprecipitations (Figure 1C). TEV eluates of LCL-TEV.5 immunoprecipitations were analyzed by mass spectrometry as described in Materials and Methods. The identified candidate LMP1 interaction partners included signaling proteins, proteins involved in ubiquitinylation processes, cytoskeletal proteins, and proteins with other or unknown functions. Two peptides identifying the TRAF2- and Nck-interacting kinase (TNIK) were detected in the TEV eluate of LCL-TEV.5, but not of control cells, which indicated that TNIK interacts with the signaling domain of HA-LMP1-liTEV-CT and is thus a novel component of the LMP1 signaling complex (Tables S1 and S2).
To confirm the interaction between TNIK and LMP1 in lymhoblastoid cells, endogenous TNIK was immunoprecipitated from lysates of LCL 721 cells and analyzed for LMP1 binding (Figure 2A). Indeed, endogenous LMP1 specifically co-precipitated with TNIK. Vice versa, immunoprecipitation of LMP1 brought down TNIK (Figure 2A). These experiments verified the results that were previously obtained in the functional proteomics experiment and showed that TNIK is in fact part of the LMP1 signalosome in EBV-transformed B-cells.
Next we asked whether one of the two signaling-active subdomains of LMP1, CTAR1 or CTAR2, mediates the interaction between TNIK and LMP1. Wildtype LMP1 as well as LMP1(AAA) harboring a mutated PxQxT motif within CTAR1, the CTAR2 deletion mutant LMP1Δ371–386, and the CTAR1/CTAR2 double mutants LMP1(AAA, Δ371–386) and LMP1(AAA, Y384G) were transiently expressed in HEK293 cells and endogenous TNIK was immunoprecipitated from cell lysates. Immunoblot analysis of the precipitations revealed that wildtype LMP1 and the LMP1(AAA) mutant bound to TNIK equally well, excluding a critical role of CTAR1 for TNIK binding. In contrast, mutation of CTAR2 completely abolished interaction of LMP1 and TNIK, the exchange of tyrosine 384 to glycine being equally effective as the deletion of the 16 C-terminal amino acids of CTAR2 (Figure 2B). These experiments indicated but did not definitely prove that CTAR2 is the critical domain for LMP1's interaction with TNIK. Therefore, we repeated the experiment with the HA-LMP1-TNFR1-CTAR2 chimera, which consists of the LMP1 transmembrane domain fused to the TNFR1 signaling domain that carries amino acids 371 to 386 of CTAR2 replacing the TNFR1 death domain [35]. Except for CTAR2 residues 371–386, no other sequences of the LMP1 signaling domain are present within the chimera. TNIK readily bound to HA-LMP1-TNFR1-CTAR2 but not the control construct lacking the CTAR2 sequences (Figure 2C). In summary, these experiments demonstrated that CTAR2 is both critical and sufficient for TNIK recruitment to LMP1, whereas CTAR1 has no apparent role in mediating this interaction.
Having identified TNIK as a novel CTAR2 interaction partner, we asked whether TNIK has a role in LMP1 signal transduction. The JNK pathway initiates at CTAR2 and TNIK was shown to induce JNK signaling upon overexpression [1],[30]. Therefore, we investigated a potential role for TNIK as mediator of LMP1-induced JNK signal transduction. HEK293 cells were transfected with TNIK-specific siRNA or non-targeting control siRNA. Subsequently, wildtype LMP1 or the null control LMP1Δ194–386 were expressed, and HA-JNK kinase assays were performed to monitor LMP1-induced JNK1 activity. The knockdown of TNIK caused a drastic reduction of JNK activation by LMP1 (Figure 3A). We confirmed this result in the human lymphoblastoid cell line EREB2-5. Upon knockdown of TNIK with siRNA a robust reduction of endogenous JNK phosphorylation, a measure of JNK activity, was detected in EREB2-5 cells (Figure 3B). Notably, JNK activity in LCLs depends on LMP1 [11],[30],[47]. We have thus identified TNIK as a novel signaling mediator of LMP1 that is crucial for the induction of the JNK pathway.
Canonical NF-κB constitutes the second important signaling pathway that is initiated at the CTAR2 domain of LMP1. CTAR2, but not CTAR1, induces IKKβ activity, which is essential for CTAR2-mediated NF-κB signaling [25],[35],[42]. To test if TNIK is involved in this pathway as well, Flag-IKKβ kinase assays were performed in HEK293 cells (Figure 4A). Endogenous TNIK was depleted by TNIK siRNA, and LMP1 wildtype or the inactive null mutant LMP1(AAA, Δ371–386) was expressed and tested for their ability to activate IKKβ. LMP1 expression in cells treated with control siRNA caused a 2.6-fold induction of IKKβ activity, monitored as in vitro GST-IκBα substrate phosphorylation by the immunoprecipitated Flag-IKKβ. The knockdown of TNIK almost entirely abolished the activation of IKKβ by LMP1, demonstrating the critical importance of TNIK in the canonical NF-κB pathway (Figure 4A). In order to exclude a role for TNIK in non-canonical NF-κB signaling by LMP1, the effect of a TNIK knockdown on NF-κB p52 was examined. NF-κB p52 activation is a hallmark for CTAR1 signaling [24]–[27]. Downregulation of TNIK by a shRNA vector in HEK293 cells did not affect the LMP1-induced p52 translocation to the nucleus, whereas the nuclear shift of canonical p65 was largely inhibited (Figure 4B).
NF-κB reporter assays were performed in HEK293 cells to test the role of TNIK also at the level of NF-κB-dependent transcription. The siRNA-mediated knockdown of TNIK caused a nearly 50% reduction in NF-κB activation by LMP1 as compared to cells treated with control siRNA (Figure 4C). Given that a substantial proportion of total LMP1-induced NF-κB activity detected in reporter assays constitutes CTAR1-induced non-canonical NF-κB [28],[29], we concluded that knockdown of TNIK largely blocked CTAR2 signaling in the reporter assay. This conclusion was later corroborated by the use of a dominant-negative TNIK mutant that inhibited CTAR2, but not CTAR1, activation of the NF-κB reporter (see Figure 6F).
We confirmed our findings by siRNA experiments in EBV-transformed EREB2-5 cells. Knockdown of TNIK in these cells resulted in a marked reduction of phosphorylated IκBα and a concomitant stabilization of IκBα showing that canonical NF-κB signaling is defective upon depletion of TNIK in LCLs (Figure 4D). TNIK is thus an important signaling mediator of the canonical NF-κB pathway.
The LMP1-induced IκB-dependent NF-κB pathway and the JNK pathway are essential for lymphoblastoid cell survival and proliferation [31],[32]. Given the important role of TNIK in both pathways, its knockdown should interfere with LCL physiology. To test this hypothesis, TNIK expression was downregulated in EREB2-5 lymphoblastoid cells by siRNA and proliferation was monitored. In fact, TNIK deficiency strongly retarded proliferation of the cells and apoptosis was induced concomitantly (Figure 5A and 5B, respectively). The spontaneous apoptosis rate in EREB2-5 cells increased by a factor of 3.8 on average after the knockdown of TNIK (Figure 5B). Accordingly, many dead cells were visible in disintegrating LCL clumps in the siTNIK-treated EREB2-5 culture, whereas the siCTR-treated cells displayed normal LCL morphology (Figure S2). Thus, TNIK has a critical function in mediating proliferation and survival of LCLs, which is in line with its central role in LMP1 signal transduction.
As TNIK is critically involved in both JNK and canonical NF-κB signal transduction downstream of LMP1, we next asked whether these two pathways might bifurcate at the level of TNIK by dissecting the contribution of individual TNIK domains to the activation of JNK and NF-κB signaling. A set of HA-tagged TNIK constructs was generated that comprise full-length TNIK, the kinase domain (KD), the germinal center kinase homology domain (GCKH), as well as the ΔKD and ΔGCKH deletion mutants (Figure 6A). Additionally, a kinase-negative mutant (KM) of TNIK was used, which carries a mutation of the conserved lysine 54 residue in the ATP-binding pocket of the kinase domain [1]. We then tested for the ability of the individual TNIK constructs to induce canonical NF-κB signaling in IKKβ kinase activity assays. Wildtype TNIK activated IKKβ-dependent phosphorylation of GST-IκBα by a factor of 4.7-fold (Figure 6B). Notably, expression of the TNIK kinase domain alone was sufficient to fully induce IKKβ as TNIK-KD caused an 11-fold activation of IKKβ. Vice versa, mutation or deletion of the kinase domain completely abolished TNIK's potential to activate IKKβ. In contrast, neither deletion of the GCKH domain nor its overexpression had any effect on IKKβ activation. In line with these results, the exogenous expression of TNIK wildtype or TNIK-KD was sufficient to also induce the nuclear translocation of canonical NF-κB p65, whereas non-canonical NF-κB p52 remained unaffected (Figure 6C). This finding further corroborated our previous observations that TNIK has no function in non-canonical NF-κB signaling (see above). As expected, TNIK-KM was unable to shift any of the two NF-κB proteins to the nucleus (unpublished data). In summary, we concluded that the TNIK kinase domain and in particular its kinase activity is critical for canonical NF-κB induction by TNIK, while the GCKH domain is dispensable for this pathway.
Notably, JNK activation maps to a region of TNIK different from the NF-κB-activating kinase domain. The GCKH domain alone activates JNK as efficiently as full-length TNIK, whereas mutation of the kinase domain had no effect on TNIK's ability to induce JNK as determined by kinase assays in HEK293 cells (Figure 6D). This finding is consistent with previous results showing that the GCKH domain alone can induce the JNK pathway whereas the TNIK kinase domain is dispensable [1]. Taken together, JNK and IKKβ induction map to different TNIK domains, suggesting that TNIK constitutes the point of bifurcation of these two pathways.
Next we asked about the functional role of the TNIK kinase domain in IKKβ/NF-κB activation. One straightforward scenario would be that TNIK phosphorylates IKKβ for its activation. However, we did not detect direct IKKβ phosphorylation by TNIK in our experimental systems, for instance in TNIK kinase assays using IKKβ as a substrate (unpublished data). Previous studies demonstrated that TNIK phosphorylates itself [1],[3]. Therefore, we investigated if LMP1 expression affects TNIK autophosphorylation. In fact, LMP1 enhanced the phosphorylating activity of TNIK versus itself by a factor of 4.2-fold, demonstrating a role for TNIK autophosphorylation in LMP1 signaling (Figure 6E). The vast majority of the about 40 Ser/Thr phosphorylation sites of TNIK detected so far in vivo by mass spectrometry are located within the intermediate domain (databank: www.phosphosite.org; search term: TNIK). If the TNIK kinase domain phosphorylates TNIK within its intermediate domain and TNIK autophosphorylation is critical for NF-κB signaling, the exogenously expressed TNIK kinase domain alone would be non-functional but depend on endogenous wildtype TNIK to activate IKKβ. To test this possibility, HEK293 cells were depleted of endogenous wildtype TNIK by siRNA. Subsequently, the construct HA-TNIK-KDwob was transfected, which expresses the wildtype TNIK kinase domain, and IKKβ kinase assays were performed. As the HA-TNIK-KDwob construct carries silent wobble mutations at the nucleotide level, it is not targeted by TNIK-specific siRNA. The knockdown of endogenous TNIK abolished the potential of the exogenous TNIK kinase domain to activate IKKβ (Figure S3). A similar mechanism for JNK activation can be excluded because the kinase domain is dispensable for JNK signaling (see Figure 6D) and TNIK-KD overexpression does not induce JNK in HEK293 cells [1]. Taken together, these findings are in line with the concept of a role for TNIK autophosphorylation in NF-κB signaling by LMP1.
To further validate the importance of the TNIK kinase domain for canonical NF-κB signaling, we tested if overexpression of the kinase-negative mutant TNIK-KM had a dominant-negative effect on LMP1-induced NF-κB signaling in reporter assays (Figure 6F). In fact, TNIK-KM expression reduced NF-κB activation by wildtype LMP1 to almost 50%, a factor that was comparable to the effect of TNIK knockdown on LMP1-induced NF-κB (see Figure 4C). Moreover, NF-κB signaling of LMP1(AAA), which only harbors functional CTAR2, was affected by TNIK-KM but not that of LMP1Δ371–386, which solely signals via CTAR1 (Figure 6F). Thus, TNIK-KM exerted its dominant-negative effect on CTAR2-induced NF-κB signaling, confirming that the kinase activity of TNIK is critical for activation of canonical NF-κB by LMP1-CTAR2.
To better understand TNIK's molecular functions in JNK and NF-κB activation and its role as bifurcation point of the two pathways, it was necessary to identify TNIK interaction partners in LMP1 signaling. The first step was to investigate how TNIK interacts with LMP1 and to characterize potential mediators of this interaction. TNIK has been shown to bind TRAF2 via its intermediate domain [1]. This finding suggested that TRAF molecules might physically couple TNIK to upstream inducers/receptors. CTAR2 signaling to JNK and IKKβ/NF-κB essentially requires TRAF6 but not TRAF2 [18],[33],[34],[38],[48]. Despite the fact that an interaction of TRAF6 with TNIK has not been described so far, we tested if TRAF6 binds to TNIK in LMP1 signaling by immunoprecipitation experiments in HEK293 cells (Figure 7A). In the absence of LMP1 a weak co-precipitation of HA-TNIK and Flag-TRAF6 was detected. Strikingly, LMP1 induced a very strong interaction of both proteins, demonstrating (i) that TRAF6 is a novel binding partner of TNIK and (ii) that interaction of both proteins is greatly enhanced upon activation (Figure 7A). The effects of LMP1 on TNIK-TRAF interaction were, however, not restricted to TRAF6. CTAR2, but not CTAR1, induced a weak but detectable interaction of TNIK with TRAF2 (Figure S4). Because studies in TRAF2-deficient cells have clearly excluded a critical function for TRAF2 in CTAR2 signaling [18],[33],[48], we concentrated our further studies on the newly identified and CTAR2-critical TNIK interaction partner TRAF6.
The TNIK intermediate domain directly binds TRAF2, as has been shown by yeast-two-hybrid assays and immunoprecipitations [1]. To determine whether TRAF6 and TNIK are also direct interaction partners, in vitro binding assays using recombinant proteins purified from bacteria were performed (Figure 7B). Indeed, the C-terminal TRAF domain of TRAF6 (amino acids 310–522) specifically bound to the immobilized GST-tagged TNIK intermediate domain. Purified TRAF2 (amino acids 311–501) was included into the experiment as a control, which also interacted with the intermediate domain of TNIK. No interaction of the two TRAFs with the TNIK kinase domain, the GCKH domain, or the GST control beads was observed. Thus, the C-terminal TRAF domain of TRAF6 directly binds to the TNIK intermediate domain.
In order to investigate whether TRAF6 acts as mediator of the interaction between TNIK and LMP1 we analyzed the subcellular localization of transiently expressed HA-TNIK and LMP1 in TRAF6-deficient and wildtype mouse embryonic fibroblasts. Confocal immunofluorescence microscopy revealed a high degree of co-localization of TNIK and LMP1 in the TRAF6+/+ cells (Figure 7C). LMP1 did not induce translocation of TNIK into the nucleus as it has been shown for Wnt signaling in intestinal cells [5]. There was no significant co-localization of LMP1 and TNIK in TRAF6−/− cells. This finding was substantiated by a grey scale line scan analysis of the microscopic images confirming that the distribution of TNIK and LMP1 displays a high degree of co-localization in wildtype cells. In contrast, the absence of TRAF6 caused a more dispersed localization of TNIK and prevented its recruitment to LMP1 (Figure 7C). This result showed that TNIK and LMP1 interact in an indirect manner and that TRAF6 is crucial for this interaction. To verify this finding by a biochemical approach we performed a rescue experiment in TRAF6−/− cells. LMP1 and Flag-TNIK were expressed in TRAF6−/− cells in the absence or presence of exogenously expressed TRAF6 (Figure 7D). LMP1 co-precipitated with Flag-TNIK only when TRAF6 was transfected. Exogenous TRAF6 expression was thus able to rescue the interaction between TNIK and LMP1 in TRAF6-deficient cells. Taken together we revealed TRAF6 as a novel direct interaction partner of the TNIK intermediate domain and as critical mediator of the interaction between TNIK and LMP1.
TAK1 interacts via TAB2 with TRAF6 to activate JNK and IKKβ/NF-κB signaling (see Introduction). Previous studies have shown that TAK1 mediates JNK signaling by LMP1, whereas the role of TAK1 in NF-κB activation is controversial [37],[40],[42]. Having defined a role for TNIK as an interaction partner of TRAF6 in JNK and canonical NF-κB signaling by LMP1, we asked whether TAK1 and TAB2 interact with TNIK as well. Indeed, TNIK and TAK1 readily interacted in HEK293 cells (Figure 8A). As the presence or absence of LMP1 had no striking effect on the affinity of both proteins we concluded that TNIK and TAK1 bind to each other constitutively. We next analyzed this interaction with regard to the TNIK domains that mediate TAK1 binding by using TNIK deletion constructs for immunoprecipitations (Figure 8B). Whereas the GCKH domain alone bound to TAK1, no interaction was detectable with the TNIK kinase domain. Deletion of the GCKH domain (HA-TNIK-ΔGCKH construct) strongly diminished the interaction with TAK1. The main TAK1 interaction site of TNIK is thus the GCKH domain and the intermediate domain contributes some binding activity as well, possibly by an indirect mechanism. It is important to note at this point that the GCKH domain of the MAP4K TNIK induces JNK signaling (see Figure 6D) and at the same time binds the critical MAP3K for this pathway, TAK1.
Co-immunoprecipitation experiments in HEK293 cells showed that TAB2 also specifically co-precipitates with TNIK (Figure 8C). However, this interaction is activation-dependent, as TAB2 did only very weakly bind to TNIK unless LMP1 was present. LMP1 co-expression induced a strong interaction of TNIK with TAB2.
Notably, TNIK is required for the interaction of TAK1/TAB2 with the LMP1 complex. The knockdown of endogenous TNIK by expression of shRNA abolished binding of TAK1 and TAB2 to LMP1 in co-immunoprecpitation experiments (Figure 8D and 8E, respectively). Thus, TNIK has an important role in the assembly of the LMP1 signalosome by acting as an interaction mediator of critical components of the complex.
We have shown that LMP1 activates IKKβ via TNIK. Therefore, we asked whether IKKβ is also a component of the TNIK signaling complex. Indeed, IKKβ also bound to TNIK, albeit only in the presence of LMP1 (Figure 8F). The interaction of TNIK with IKKβ appeared to be weaker as compared to TRAF6, TAK1, or TAB2, potentially indicating an indirect recruitment of IKKβ to TNIK. In summary, we found that TNIK forms a dynamic complex incorporating critical components of TRAF6-dependent JNK and NF-κB signaling, namely TRAF6, TAK1/TAB2, and IKKβ. TAK1 seems to be constitutively associated with TNIK, whereas the other components enter the complex after activation.
We sought to verify the existence of an endogenous TNIK signaling complex in lymphoblastoid cells that endogenously express LMP1. TNIK was immunoprecipitated from LCL 721 cell lysates and components of the signaling complex were analyzed by immunoblotting (Figure 8G). We found that LMP1, TRAF6, TAK1, TAB2, and IKKβ specifically bind to TNIK in LCLs, thus proving the existence of the LMP1-induced TNIK signaling complex in its native context. Taken together, our results show that the TNIK complex, which is composed of TRAF6 and LMP1 as upstream components and of TAK1/TAB2 and IKKβ as downstream mediators, is essential for JNK and canonical NF-κB activation by LMP1 in EBV-transformed human B-cells.
Having characterized TNIK as a mediator of signal transduction by the viral pseudoreceptor LMP1, we tested a general requirement for TNIK in JNK and canonical NF-κB activation by a cellular receptor in B-cells. Because LMP1 is a functional mimic of CD40 and TRAF6 plays a pivotal role as an adapter protein for both LMP1 and CD40, we tested whether CD40 engages TNIK for signal transduction.
First we analyzed the effect of TNIK knockdown on JNK1 and IKKβ activation by CD40 in HEK293 cells. Overexpression of CD40 was sufficient to activate CD40 signaling in HEK293 cells without the need to further stimulate with CD40L (CD40 ligand). TNIK was downregulated by siRNA and cells were co-transfected with either HA-JNK or Flag-IKKβ and CD40 expression vectors. HA-JNK and Flag-IKKβ kinase assays proved that the downregulation of TNIK in fact blocked CD40-induced JNK and IKKβ activation (Figure 9A and 9B, respectively).
In order to confirm these results in human B-cells, BL41 cells were depleted of endogenous TNIK by siRNA and stimulated with recombinant soluble CD40L (Figure 9C). The knockdown of TNIK resulted in a nearly complete inhibition of CD40-induced JNK phosphorylation, demonstrating an important role of TNIK in JNK activation by CD40 also in B-cells. IκBα degradation after 10 to 20 min of CD40 stimulation indicated activation of the NF-κB pathway when cells were treated with non-targeting control siRNA. In contrast, after TNIK downregulation by siRNA the NF-κB pathway did not respond to CD40 stimulation as IκBα levels did not decrease over time (Figure 9C). These data demonstrated that TNIK is a novel and critical intermediate of endogenous CD40 signaling in human B-cells on the JNK and NF-κB axes.
CD40 stimulation activates BL41 cells, detectable as upregulation of activation markers at the cell surface such as CD54, an adhesion molecule also known as ICAM-1 and hallmark of B-cell activation [49]. CD54 upregulation by CD40 is dependent on canonical NF-κB in BL cells [50]. We tested if the knockdown of TNIK affected CD54 surface upregulation by CD40 ligand stimulation of BL41 cells. TNIK dowregulation resulted in a marked decrease of CD40-induced CD54 surface levels, demonstrating an important role for TNIK also in B-cell activation (Figure 9D).
TRAF6 is an essential signaling mediator of both LMP1 and CD40, and we have demonstrated recruitment of TRAF6 to TNIK in the context of LMP1 signaling. Therefore we asked whether CD40 stimulation can also induce an interaction between TNIK and TRAF6 in B-cells. BL41 cells were stimulated with CD40L for 0, 5, and 15 min and TNIK was immunoprecipitated and tested for TRAF6 co-precipitation. We observed that CD40 induced an interaction between endogenous TNIK and endogenous TRAF6 already 5 min after stimulation (Figure 9E). Ten minutes later the majority of TRAF6 had already dissociated from TNIK. The prompt interaction between TNIK and TRAF6 in response to CD40 stimulation demonstrates a role for the TNIK–TRAF6 complex in the context of CD40 signaling, suggesting that interaction of both molecules is a key step in signaling by LMP1 and CD40. Taken together we have identified TNIK as an important mediator of JNK and also canonical NF-κB in physiological CD40 stimulation.
In this study we have identified and characterized the germinal center kinase family member TNIK as a novel component of the TRAF6/TAK1/TAB2/IKKβ complex. TNIK is required for JNK and canonical NF-κB signaling by the EBV oncoprotein LMP1 and its cellular counterpart CD40. According to this critical function in signaling, TNIK has an important role in mediating proliferation and survival of EBV-transformed B-cells and in physiological B-cell activation by CD40. In an unbiased functional proteomics screen TNIK was isolated as an interaction partner of the LMP1 complex in EBV-infected primary human B-cells. TNIK binding to the CTAR2 domain of LMP1 is mediated by TRAF6, a newly described direct interaction partner of TNIK. We thus report the first molecular function for TNIK's interaction with TRAF molecules. The existence of a CTAR2-induced signaling complex was revealed involving activation-dependent binding of TRAF6, TAB2, and IKKβ to TNIK. Importantly, CD40 stimulation also induces association of TNIK with TRAF6. Because TNIK's activities in JNK1 and NF-κB signaling map to different TNIK domains, we propose a model in which TNIK orchestrates bifurcation and signal transmission of both pathways at the level of the TRAF6/TAK1/TAB2/IKKβ complex (Figure 10). Our discovery that TNIK is a new key player in TRAF6-dependent JNK and canonical NF-κB signaling significantly extends the current concept of molecular regulation of these pathways.
LMP1 is constitutively active and closely mimics the TNFR family member CD40 in B-cell activation [11]. Despite differences in the molecular composition and efficiency of their signaling complexes, LMP1 and CD40 share similarities with regard to the engagement of TRAF molecules and the pattern of activated signal transduction pathways [14]. TRAF6 plays a pivotal role in canonical NF-κB and JNK signaling by both receptors [33],[37],[38],[51]. Both LMP1 and CD40 induce association of TNIK with TRAF6, whose interaction is direct and involves the TRAF domain of TRAF6 and the intermediate domain of TNIK. This interaction couples TNIK to the upstream receptor. Apart from TRAF6, TNIK can also interact with TRAF2. However, TRAF2 is dispensable for JNK activation, IκB-dependent NF-κB signaling, and p65 nuclear translocation by LMP1 and can thus be excluded as an essential mediator of TNIK interaction with CTAR2 [33],[34],[48]. The CTAR2-induced association of TNIK and TRAF2 detected here might have a non-essential accessory role in CTAR2 signaling. Also the CTAR2-interacting factor TRADD is unlikely to play a central role in TNIK recruitment because it is exclusively involved in IKKβ/NF-κB activation by CTAR2 but not in JNK signaling, whereas TNIK is required for both signaling pathways [35]. Upon activation by LMP1, TAB2 and IKKβ are recruited to TNIK. This observation of activation-induced complex formation is in line with findings that the dynamic association of TAB2 with TRAF6 and TAK1 also occurs in other pathways, for instance in interleukin-1 signaling [52]. TAK1, in contrast, interacts constitutively with TNIK. We therefore propose that upon activation the TNIK-TAK1 complex is recruited to the CTAR2 domain via TRAF6 and recruits additional downstream signaling mediators such as TAB2 and IKKβ. The signaling complex is likely further stabilized by TRADD, which is involved in the recruitment of IKKβ to the LMP1 complex [35]. The TRADD-dependent stabilization of the complex at CTAR2 might in part explain the more efficient signaling complex of LMP1 as compared to CD40. In contrast to LMP1, CD40 induction of JNK and IκB phosphorylation involves TRAF2 in B-cells [53],[54]. Moreover, TRAF2 is involved in TRAF6 recruitment to the distal TRAF binding site of CD40 that induces JNK and canonical NF-κB signaling [45]. For these reasons, a more pronounced role of TRAF2 in TNIK interaction with CD40 seems feasible. Future studies will have to dissect the precise role of TRAF family members in coupling TNIK to CD40. However, because CD40 induces a rapid interaction of TNIK with TRAF6, we suggest a critical role of TRAF6 in this process, which could involve additional members of the TRAF family.
The TNIK-TRAF6-TAK1/TAB2-IKKβ complex mediates activation of the canonical NF-κB and JNK pathways. TNIK is, to our knowledge, the only known protein within this complex whose activities on the NF-κB and JNK axes are clearly allocated to separate domains of the same protein. NF-κB activation depends on the kinase and intermediate domains of TNIK, whereas signaling to JNK only involves the GCKH domain. Thus, TNIK seems to constitute the molecular organizer of JNK and NF-κB bifurcation. It has been shown that purified TAK1 together with TAB1/TAB2 is sufficient to phosphorylate and thus activate IKKβ in a test tube. This reaction further depends on TRAF6 and Ubc13/Uev1A [46],[55]. Our results demonstrate that TNIK is additionally required to assemble, organize, and activate the holocomplex in vivo and to recruit the complex to the receptor (here: LMP1) by acting as an adapter and scaffolding protein. Due to its interaction with TRAF molecules TNIK is likely involved in the specific coupling of the TAK1-TAB2 and IKK modules to distinct receptors.
The GCKH domain of the MAP4kinase TNIK mediates JNK activation and is also the main interaction site of the JNK-inducing MAP3kinase TAK1, suggesting that TNIK acts directly upstream of TAK1 in the signaling cascade. However, TNIK's kinase activity is dispensable for JNK activation. Similar to other germinal center kinases, TNIK may facilitate MAP3kinase activation by inducing conformational changes that induce MAP3K autophosphorylation and thus activation of the MAP3K [2],[56].
In contrast to the JNK pathway, activation of IKKβ by TNIK critically depends on the kinase activity of TNIK. IKKβ itself does not appear to be a TNIK substrate because we could not detect direct IKKβ phosphorylation by TNIK (unpublished data). However, LMP1 induces TNIK's activity to phosphorylate itself, supporting a role for TNIK autophosphorylation in signaling. The fact that exogenous TNIK-KD is incapable of activating IKKβ in the absence of endogenous TNIK suggests that phosphorylation is important for NF-κB activation but localizes to a domain other than the kinase domain of TNIK. A similar mechanism for JNK signaling is not likely because TNIK-KD is dispensable for JNK activation and TNIK-KD overexpression does not induce JNK (see Figure 6D) [1]. Multiple phosphorylation sites cluster within TNIK's intermediate domain (www.phosphosite.org) [1],[3], which is the likely target of TNIK's autophosphorylation. Phosphorylation of TNIK seems to have different effects. One study reported that autophosphorylated TNIK is found in the cytoskeletal fraction, where it mediates disassembly of F-actin [3]. Wnt/β-catenin signaling leads to the phosphorylation of TNIK at serine 764 and translocation of TNIK into the nucleus, where it interacts with TCF4 to mediate activation of Wnt target genes [5],[6]. Using a phophosite-specific antibody we observed that LMP1 does not induce phosphorylation of TNIK at serine 764 (unpublished data), which is consistent with the fact that LMP1 does not induce Wnt signaling [57]. We consider the relevance of a different phosphorylation site within TNIK in the context of canonical NF-κB signaling. It will be the focus of future studies to identify TNIK autophosphorylation sites as well as possible other TNIK substrates in the NF-κB pathway, for instance by phosphoproteomics.
Due to its universal expression pattern we envision that TNIK functions as a mediator of TRAF-dependent JNK and NF-κB activation in various tissues and cell types. So far, TNIK has been described to regulate neurite growth and neuronal morphology in the brain and to be involved in the activation of Wnt target genes in intestinal crypt cells [5],[6],[8],[58]. Here we extend the functions of TNIK to lymphocytes. Our data indicate important roles for this kinase in B-cell function, immunity, and cancer. JNK and NF-κB have pivotal roles in physiological activation and oncogenic transformation of B-cells [45],[59],[60]. LMP1 and CD40 are involved in various malignant diseases of the hematopoietic system, such as Hodgkin's and non-Hodgkin's lymphoma, post-transplant lymphoproliferative disease, or chronic lymphocytic leukemia, as well as in non-hematopoietic cancers such as nasopharyngeal carcinoma or renal carcinoma [60]–[62]. Notably, the LMP1-induced canonical NF-κB and JNK pathways are known to be essential for LMP1-mediated B-cell transformation by activation of anti-apoptotic and cell cycle-promoting signals [31],[32],[63]. Accordingly, we showed that TNIK is essential for lymphoblastoid proliferation and survival. CD40-induced NF-κB, which also involves TNIK, protects cells from apoptosis in some low-grade B-cell malignancies and promotes cell transformation of epithelial cells, for instance in breast cancer [62],[64]. Thus, our data implicate TNIK in LMP1- and CD40-induced cancer and indicate the potential of TNIK as a future target for therapy of EBV and CD40-associated malignancies.
pSV-LMP1, pSV-LMP1Δ194–386 lacking the LMP1 signaling domain, pCMV-HA-LMP1, pCMV-HA-LMP1Δ194–386, pCMV-HA-LMP1(AAA) harboring a P(204)xQxT to AxAxA mutation within CTAR1, pCMV-HA-LMP1Δ371–386 lacking the 16 C-terminal amino acids of CTAR2, the double mutants pCMV-HA-LMP1(AAA, Δ371–386) and pCMV-HA-LMP1(AAA, Y384G), as well as the fusion constructs pCMV-HA-LMP1-TNFR1ΔDD and pCMV-HA-LMP1-TNFR1-CTAR2 (alternative name: pCMV-HA-LMP1-TNFR1-LTB) have been described [34],[35],[38]. pESBOS-CD40, pCMV-HA-TAB2, pRK-TRAF2, pRK5-HA-JNK1, and pcDNA3-Flag-IKKβ have been described [30],[35],[40]. pCMV-HA-LMP1-liTEV-CT was cloned by a PCR approach on the basis of pCMV-HA-LMP1. A flexible linker sequence (AGASGGAGASGG) and a TEV cleavage site (ENLYFQG) were inserted between amino acids Y186 and H187 of LMP1. To generate pRK5-HA-TNIK, pRK5-HA-TNIK-KD, pRK5-HA-TNIK-GCKH, pRK5-HA-TNIKΔKD, and pRK5-HA-TNIKΔGCKH, TNIK sequences were amplified from human TNIK cDNA [1] and HA-tagged by PCR, and subsequently cloned into pRK5. The vector pRK5-HA-TNIK(KM) harboring a K54R mutation within the TNIK kinase domain was subcloned from pYCI-TNIK(KM) [1]. pRK5-Flag-TNIK was generated by PCR on the basis of pRK5-HA-TNIK. pRK5-HA-TNIK-KDwob was cloned by a PCR approach on the basis of pRK5-HA-TNIK-KD. pRK5-HA-TNIK-KDwob harbors silent wobble mutations at the nucleotide level to eliminate the targeting sequence of human Dharmacon TNIK ON TARGETplus SMARTpool siRNA J-004542-10 (targeting sequence: GAACATACGGGCAAGTTTA). pRK5-HA-TRAF6 and pRK5-Flag-TRAF6 were generated by PCR approaches based upon human TRAF6 cDNA [38]. pRK5-Flag-TAK1 was cloned by a PCR approach using a TAK1 cDNA [40]. Bacterial expression vectors for glutathione-S-transferase (GST)-fused TNIK domains were generated by subcloning TNIK-KD(KM), TNIK-IMD, and TNIK-GCKH sequences from pRK5 background into pGEX2T (GE Healthcare). The C-terminal TRAF domains of human TRAF2 (amino acids 311–501) and TRAF6 (amino acids 310–522) were cloned by PCR approaches from cDNAs [34],[38] into the pET17b vector (Novagen) with an N-terminal His-tag. All constructs were verified by sequencing. Detailed cloning strategies and PCR primer sequences can be made available upon request.
The EBV-positive lymphoblastoid B-cell lines LCL 721, EREB2-5, and LCL3 have been described [35],[65],[66]. The generation of LCL-TEV.5 cells is described herein. HEK293 human embryonic kidney cells, the human EBV-negative Burkitt's lymphoma B-cell line BL41 [67], and all lymphoblastoid cell lines were grown in RPMI full medium (Invitrogen) supplemented with 10% fetal calf serum (Biochrom AG). EREB2-5 cells were additionally kept in the presence of 1 µM β-estradiol to activate the conditional ER-EBNA2 transcription factor that drives EBV latent genes required for proliferation of EREB2-5 cells [66]. Wildtype and TRAF6−/− mouse embryonic fibroblasts [51] were grown in DMEM (Invitrogen) supplemented with 10% fetal calf serum. BL41 cells were stimulated with the indicated amounts of human recombinant soluble CD40 ligand (Source BioScience).
HEK293 cells were seeded in 6-well plates and transfected twice within 24 h with 100 nM of human ON TARGETplus SMARTpool TNIK siRNA (pool of four siRNAs J-004542-10 to 13, Dharmacon) or corresponding ON TARGETplus non-targeting control siRNA using the Dharmafect transfection reagent according to the manufacturer's protocol. The cells were transfected 24 h later with the indicated plasmids using Polyfect transfection (Qiagen) and analyzed 24 h after the last transfection. To achieve TNIK knockdown in larger cell culture dishes, HEK293 cells were co-transfected with pSM2-shTNIK (RHS1764-949310, Open Biosystems), an expression vector for short hairpin RNA targeting TNIK, or the non-targeting control vector pSM2-shControl (RHS1707-OB, Open Biosystems) as indicated in the figure legends. LCL 721, EREB2-5, and BL41 cells were incubated with 5 µM Accell SMARTpool TNIK siRNA (pool of four Accell siRNAs J-004542-18 to 21, Dharmacon) or Accell non-targeting pool siRNA for 72–96 h in serum-free Accell delivery medium (Dharmacon). Subsequently, B-cells were lysed in Laemmli-DTT buffer (25 mM Tris-HCl pH 6.8, 1% SDS, 5% glycerine, 25 mM DTT) for immunoblotting or analyzed as indicated.
Recombinant maxi-EBV p2089-HA-LMP1-liTEV-CT was generated as previously described [11],[68]. In brief, HA-LMP1-liTEV-CT sequences were subcloned from pCMV-HA-LMP1-liTEV-CT into the shuttle vector p2167.1 to transfer HA-LMP1-liTEV-CT into the context of the viral LMP1 locus, HA-LMP1-liTEV-CT replacing the wildtype LMP1 gene. Homologous recombination of the shuttle vector with the p2089 wildtype maxi-EBV bacterial artificial chromosome (BAC) in E. coli DH10B resulted in p2089-HA-LMP1-liTEV-CT. The packaging cell line TR-2/293 was transfected with p2089-HA-LMP1-liTEV-CT DNA and virus production was induced by transfection with expression vectors for the EBV genes BALF4 and BZLF1 as described [68]. For B-cell infection, primary B-cells were prepared from human adenoids and plated together with 2089-HA-LMP1-liTEV-CT virus supernatant on a feeder layer of γ-irradiated WI38 cells in 96-well plates as described [68]. Outgrowing lymphoblastoid LCL-TEV clones were further propagated and analyzed. Expression of HA-LMP1-liTEV-CT and absence of wildtype LMP1 were confirmed by RT-PCR and immunoblotting.
5×108 lymphoblastoid cells per sample were lysed in 15 ml of IP-lysis buffer (150 mM NaCl, 50 mM HEPES pH 7.5, 5 mM EDTA, 0.1% NP-40, 0.5 mM sodium orthovanadate, 0.5 mM NaF, 0.5 mM sodium molybdate, Roche complete proteinase inhibitor) and cleared by centrifugation at 16,000 g. To immunoprecipitate HA-LMP1-liTEV-CT the lysates were incubated with the anti-HA (12CA5) antibody (mouse, Roche) covalently coupled to protein-A-sepharose beads (Roche) by treatment with 20 mM dimethyl pimelimidate (Fluka). Immunoprecipitations were washed three times with IP-lysis buffer, and TEV protease cleavage was performed in TEV buffer (50 mM Tris-HCl pH 8.0, 0.5 mM EDTA, 1 mM dithiothreitol, 0.1% NP-40) for 4 h at 16°C. Subsequently the beads were removed by centrifugation at 500 g for 5 min to elute the released LMP1 signaling domain together with interacting proteins. Small aliquots were analyzed by immunoblotting and the remaining samples were further processed for mass spectrometry analysis. To reduce complexity of the samples, the eluate was separated on a 12.5% SDS gel and subdivided into three parts by excising coomassie-stained gel slices containing proteins in the range of <30 kDa, 30–70 kDa, or >70 kDa, respectively. After incubation of the gel slices in ABC buffer (50 mM ammonium carbonate, 30% acetonitrile), proteins were reduced by dithiothreitol and alkylated by iodoacetamide within the gel. After tryptic digestion the peptides were eluted from the gel in elution buffer (80% acetonitrile, 1% trifluoroacetic acid) and dried under vacuum. For LC-MALDI analysis of the complex mixture, peptides were dissolved in 3% acetonitrile and 0.5% trifluoroacetic acid, desalted, and subsequently separated by nano HPLC (Ultimate II HPLC, manufactured by Dionex/LC-Packings) on a C18 column using an acetonitrile gradient (5% to 80% acetonitrile, 0.08% trifluoroacetic acid) for elution. Eluted peptides were spotted with a Probot LC-MALDI spotting system (Dionex) onto a matrix-assisted laser desorption/ionization (MALDI) target with cyano-4-hydroxycinamonic acid as matrix. MALDI-TOF-TOF mass spectrometry of the peptides was performed using the Applied Biosystems (ABI) proteomics analyzer 4700. MS spectra were analyzed by the GPS 3.5 explorer software (Applied Biosystems). SwissProt databank search was performed using the Mascot algorithm.
HEK293 cells were seeded in cell culture dishes and transfected at 70% confluence with the indicated plasmids using the Polyfect transfection reagent according to the manufacturer's protocol (Qiagen). Twenty-four hours post-transfection, cells were lysed in IP-lysis buffer (see above). Lymphoblastoid or BL41 cells were lysed in IP-lysis buffer at a total protein concentration of 1 mg/ml. Immunoprecipitations from B-cells were performed with 3 to 5 mg of total protein per sample. Proteins were precipitated using antibodies that had been covalently coupled to protein-G-sepharose beads (GE Healthcare). The following antibodies were used for immunoprecipitation: Flag (6F7) (Sigma), HA (12CA5) (Roche), LMP1 (1G6-3) (provided by Elisabeth Kremmer) [69], and TNIK (BD Biosciences). After immunoprecipitation beads were washed four times with IP-lysis buffer and precipitated proteins were analyzed by SDS-PAGE and immunoblotting using standard protocols. The following primary antibodies were used for immunoblotting: Flag (6F7), Flag (M2) (Sigma); HA (12CA5), HA (3F10) (Roche); LMP1 (1G6-3); LMP1 (CS1-4) (Dianova); TNIK (BD Biosciences); CD40 (C-20), IKKβ (H-470), JNK1 (C-17), SAM68 (C-20), TAB2 (H-300), TAK1 (M-579), TRAF2 (C-20), TRAF3 (C-20), TRAF6 (H-274), TRAF6 (C-20), and tubulin (B-5-1-2) (Santa Cruz Biotechnology); and phospho-SAPK/JNK (Thr183/Tyr185), phospho-IκBα (Ser32) (14D4), p52/p100 (18D10), and p65 (C22B4) (New England Biolabs). Horseradish peroxidase-coupled secondary antibodies were purchased from New England Biolabs.
HEK293 cells were lysed 24 h post-transfection for overexpression studies and 48 h post-transfection for RNAi experiments. Cells were recovered from 10 cm cell culture dishes and washed twice in PBS at 4°C. Cells were lysed in 100 µl of swelling buffer (10 mM HEPES pH 7.7, 10 mM KCl, 2 mM MgCl2, 0.1 mM EDTA) on ice for 10 min. Subsequently, 0.65% NP40 was added, and the samples were incubated on ice for 1 min and centrifuged for 1 min at 16,000 g. The supernatant representing the cytosolic fraction was collected, and the pellet was washed once with swelling buffer. To retrieve the nuclear fraction the pellet was lysed in 50 µl of nuclear extraction buffer (50 mM HEPES pH 7.7, 50 mM KCl, 300 mM NaCl, 0.1 mM EDTA, 10% glycerol) and insoluble debris was removed by centrifugation. The samples were further analyzed for NF-κB proteins and the marker proteins tubulin (cytoplasm) and SAM68 (nucleus) by immunoblotting.
Immunocomplex kinase assays were essentially performed as described [38]. In brief, HEK293 cells were seeded into 6-well plates and transiently cotransfected with 2 µg each of the indicated constructs and 1 µg of pRK5-HA-JNK1 for JNK1 kinase assays or pcDNA3-Flag-IKKβ for IKKβ kinase assays using Polyfect transfection. Twenty-four hours post-transfection, cells were lysed in IP-lysis buffer and HA-JNK1 or Flag-IKKβ was immunoprecipitated overnight at 4°C using immobilized anti-HA (3F10) (Roche) or anti-Flag (6F7) (Sigma) antibodies. Beads were washed twice with IP-lysis buffer and twice with kinase reaction buffer (20 mM Tris-HCl pH 7.4, 20 mM NaCl, 10 mM MgCl2, 1 µM DTT, 2 µM ATP). In vitro kinase assays were subsequently performed in the presence of 10 µCi γ-32P-ATP and 2 µg of the recombinant purified substrates GST-c-Jun or GST-IκBα, respectively, for 25 min at 26°C. Substrates were included in the reaction buffer mix and their concentrations were not rate limiting for the phosphorylation reaction at the given reaction conditions [70]. For TNIK autophosphorylation assays, HA-TNIK was immunoprecipitated using the anti-HA (12CA5) antibody, beads with precipitated HA-TNIK were washed thoroughly, and the kinase reaction was performed in kinase reaction buffer lacking any other substrate. Kinase reactions were terminated by denaturating samples in Laemmli-DTT buffer. Subsequently samples were subjected to SDS-PAGE and autoradiography. Radioactive signals were quantified using the Fuji FLA-5100 phosphoimager.
For cell proliferation assays, 2×104 EREB2-5 cells per well of a 96-well plate were seeded at day zero in triplicates in Accell delivery medium supplemented with 1 µM β-estradiol and 5 µM Accell SMARTpool TNIK or Accell non-targeting siRNA. Proliferation was monitored at the indicated times by 3-(4,5-dimethylthiazol-2-yl)-2,2,5-diphenyl tetrazolium bromide (MTT) conversion as described [11]. Apoptosis was assayed by staining of the cells with propidium iodide (PI) and Cy5-labeled Annexin V using the Apoptosis Detection Kit (Biocat) and subsequent flow cytometry analysis with the Becton Dickinson FACSCalibur flow cytometer. For detection of CD40-induced CD54 surface expression, 5×104 BL41 cells were seeded per well of a 24-well plate in Accell delivery medium supplemented with 5 µM Accell siRNA. At day 1 and 2, the cells were stimulated with 1 µg/ml CD40 ligand or left untreated in the presence of 2% fetal calf serum. At day 3, surface CD54 was stained with an APC-conjugated anti-CD54 antibody (ImmunoTools) and detected by flow cytometry. Flow cytometry data were analyzed with the FlowJo software (TreeStar).
1.5×106 mouse embryonic fibroblasts were electroporated using a BioRad Gene Pulser II at 240 V and 950 µF with 2 µg each of pSV-LMP1 and pRK5-Flag-TNIK. Total transfected DNA was adjusted to 20 µg with empty vector. After transfection cells were seeded onto glass coverslips and cultivated overnight. Cells were subsequently fixed with 2% paraformaldehyde (Merck) for 15 min, permeabilized with 0.15% Trtion X-100 (Sigma) in PBS three times for 5 min, and then blocked three times for 10 min with blocking solution (PBS, 1% bovine serum albumin, 0.15% glycine). Cells were then incubated with the primary antibody in blocking solution for 2 h at room temperature. After washing once with PBS and twice with PBS containing 0.15% Triton X-100 cells were blocked for 7 min in blocking solution. Subsequently cells were incubated with secondary antibody diluted 1∶200 in blocking solution for 45 min at room temperature. The following primary antibodies were used: TNIK (mouse, BD Biosciences) and LMP1 (rat, 1G6-3). The following secondary antibodies were used: CY3-conjugated goat-anti-mouse IgG (H+L) and FITC-conjugated goat-anti-rat IgG (H+L) (both: Dianova). Images were acquired with a Leica TCS SP2 confocal laser scanning microscope fitted with a 63×1.4 HCX Plan Apo blue objective. The acquired digital images were deconvoluted and evaluated with Huygens Essential Suite 3.2 software (Scientific Volume Imaging). Colocalization events were further analyzed with grey scale signal intensity line scans.
HEK293 cells were transfected in 6-well plates with the indicated constructs and 5 ng of the NF-κB luciferase reporter 3xκBLuc [28] together with 50 ng of a pPGK-Renilla housekeeping gene reporter construct using Polyfect transfection (Quiagen). Twenty-four hours post-transfection, cells were lysed in reporter lysis buffer and firefly and renilla luciferase activities were measured using the Dual-Luciferase reporter assay kit (Promega). Luciferase activities were normalized for renilla activities to standardize for transfection efficiency.
His-tagged TRAF domains of TRAF2 (amino acids 311–501) and TRAF6 (amino acids 310–522) were expressed in E. coli BL21 Codon Plus RIPL cells (Stratagene) from pET17b vectors. Protein expression was induced by induction at an OD600 of 0.8 with 0.1 mM isopropyl-β-D-1-thiogalactopyranoside at 20°C overnight. Bacteria were lysed by sonication in 50 mM phosphate buffer, pH 8.0, supplemented with 10 mM imidazole, 300 mM NaCl, 1 mg/ml lysozyme, and Roche complete proteinase inhibitor cocktail. Cleared lysates were incubated with Ni2+-NTA agarose (Qiagen) to bind His-tagged TRAF proteins. Subsequently, His-tagged proteins were washed with 50 mM phosphate buffer, pH 8.0, 300 mM NaCl and increasing imidazole concentrations (20 to 100 mM), and eluted from Ni2+-NTA agarose with 50 mM phosphate buffer, pH 7.4, 300 mM NaCl, and 500 mM imidazole. Eluted TRAF proteins were further purified by gel filtration on a DextraSEC PRO10 column (Applichem) and the buffer was exchanged to TBS, pH 7.4, 20% glycerol. Proteins were either directly used for experiments or stored at −20°C for up to 4 wk. For in vitro protein binding assays, 1 µg of His-TRAF2(311–501) or His-TRAF6(310–522) were incubated for 1 h at 4°C in 500 µl TBS, pH 7.4, 0.1% (w/v) BSA, 0.5% Tween 20, with immobilized GST-TNIK-KD(KM), GST-TNIK-IMD, GST-TNIK-GCKH or GST, purified from E. coli and coupled to glutathione sepharose (GE Healthcare). Beads were washed 3 times with TBS containing 0.1% Tween 20 and bound His-TRAF proteins were analyzed by immunoblotting.
|
10.1371/journal.pcbi.1005051 | SARAH Domain-Mediated MST2-RASSF Dimeric Interactions | RASSF enzymes act as key apoptosis activators and tumor suppressors, being downregulated in many human cancers, although their exact regulatory roles remain unknown. A key downstream event in the RASSF pathway is the regulation of MST kinases, which are main effectors of RASSF-induced apoptosis. The regulation of MST1/2 includes both homo- and heterodimerization, mediated by helical SARAH domains, though the underlying molecular interaction mechanism is unclear. Here, we study the interactions between RASSF1A, RASSF5, and MST2 SARAH domains by using both atomistic molecular simulation techniques and experiments. We construct and study models of MST2 homodimers and MST2-RASSF SARAH heterodimers, and we identify the factors that control their high molecular stability. In addition, we also analyze both computationally and experimentally the interactions of MST2 SARAH domains with a series of synthetic peptides particularly designed to bind to it, and hope that our approach can be used to address some of the challenging problems in designing new anti-cancer drugs.
| We model the conformational changes and protein-protein interactions of enzymes involved in signaling along the Hippo pathway—a key molecular mechanism that controls the process of programmed cell death in eukaryotic cells, including cells affected by cancer. Combining modern computational modeling techniques with experimental information from X-ray crystallography and systems biology studies, can unveil detailed molecular interactions and lead to novel drugs. Here, we study the atomistic mechanisms and interactions between MST2 and RASSF-type kinases, through their respective SARAH domains—highly conserved, long, terminal α-helices, which play essential roles in the activation of MST kinases and, therefore, in modulating apoptosis. In spite of their key roles in mediating cell signaling pathways, there is little structural information available for the RASSF SARAH domains and their dimerization with the MST2 SARAH domains. In particular, the RASSF1A crystal structure is not available yet. Here, we model, refine and validate atomistic structural models of dimers of the RASSF1A and MST2 SARAH domains, studying the interaction and the dynamic behavior of these molecular complexes using homology modeling, docking and full atomistic molecular dynamics simulations. Experimentally, we validate our approach by designing a novel peptide that can disrupt effectively MST2 homo and hetero SARAH dimers.
| There is an acute need for novel drug targets and strategies in the fight against cancer. New directions could emerge from exploring the tumor-suppressive RASSF signaling pathway and its downstream effectors, the MST1/2 kinases, which control tissue homeostasis by balancing cell proliferation and cell death through apoptosis [1–4]. The activation of MST1/2 kinase activity is regulated by either homo-dimerization or by interactions with scaffold proteins such as WW45 and different members of the RASSF family. The regulation of MST1/2 by RASSF scaffolds is a key event in this pathway, but remains poorly understood [3, 5]. The evidence we have so far indicates that the RASSF family members RASSF1A and RASSF5 (also known as NORE1 or RALP) are tumor suppressors that mediate apoptosis through different effectors including MST1/2 kinases, but their exact regulation by RASSF proteins is incompletely understood [6]. RASSF1A and RASSF5 regulate MST1/2 kinase activity by direct protein-protein interaction through their respective SARAH domains [7]. The SARAH domain is a long, conserved α-helix at the C-terminal end, known to be a key protein-protein interaction domain [8]. A comparative analysis of the RASSF family SARAH domains has been previously published by Chan et al. [9] and discussed also in Ref. [6]. We showed that other proteins that do not have a SARAH domain themselves, such as RAF1, could nevertheless also regulate MST1/2 kinase activity through direct binding to their SARAH domain [1, 10], confirming the importance of protein-protein interactions via the SARAH domain in the regulation of these kinases. In addition, RASSF proteins were shown to be able to activate or inhibit MST1/2 kinase activity upon heterodimerization [5].
Given the importance that dimerization of MST1/2 and the RASSF proteins have on the regulation of MST1/2-dependent apoptosis, several studies have focused on the description of the interaction between RASSF5 and MST proteins through their SARAH domains, as summarized recently in Ref. [6]. Accordingly, crystal structures are available for the MST-RASSF5 SARAH domain dimers [11, 12]. The MST2-RASSF5 SARAH domain hetero-dimer (Fig 1) crystal structure was recently determined [11, 13], and further analysis of the MST2-RASSF5 interactions from the crystal structure was carried out from an experimental point of view [11]. However, only few studies considered the structure of the RASSF1A SARAH domain and its dimerization with the MST2 SARAH domain [14]. Importantly, the RASSF1A loss of expression is arguably one of the most frequent events in human solid tumors, and the characterization of RASSF1A-MST2 heterodimers could help to understand the important role of RASSF1A as a tumor suppressor [6].
In this study, we analyze the specific dimeric interactions between the helical SARAH domains of MST2 and RASSF enzymes. As indicated above, SARAH domains have been previously characterized in experimental in vivo studies of MST2, for example by using protein arrays to demonstrate their specific binding [1, 4]. However, this paper is our first modeling study that thoroughly examines the interactions between SARAH domains and SARAH—RASSFx interactions. Based on recently solved crystal structures of MST2-RASSF5 SARAH heterodimers, we use a combination of homology modeling, docking, molecular modeling and atomistic molecular dynamics (MD) methods, to construct and study a variety of models of MST2-MST2 homodimers, as well as of MST2-RASSF5 and MST2-RASSF1A SARAH heterodimers, and we identify the factors that control their high molecular stability. We also study the interaction of the MST2 SARAH domain with a de novo designed peptide, and demonstrate both via in silico modeling and experimentally that this peptide disrupts the MST2-RASSF1A interactions, as predicted. A summary, schematic representation of the principal monomeric and dimeric systems modeled in this study is illustrated in Fig 1A.
Fig 1B illustrates the MST2-RASSF5 complex from the only available crystal structure (PDB ID: 4LGD) [13] showing the direct interaction between the RASSF5 SARAH domain (red) and the MST2 SARAH domain (blue). The MST2 kinase domain (blue) is also resolved in the 4LGD crystal structure. To model the structure of RASSF1A SARAH domains, we performed multiple sequence analysis to infer similarities with structurally resolved homologues (e.g., RASSF5) from the same family. We performed a detailed sequence comparison of RASSF1-to-6 SARAH domains, using multiple sequence alignment obtained with the Clustal Omega software [15–17], which allowed us to infer its similarity with structurally resolved homologues (e.g., RASSF5) from the same family.
Our analysis showed clearly that the RASSF SARAH domains contain multiple conserved sites (Fig 2A). Importantly, the pairwise alignment between RASSF1A and RASSF5 (Fig 2B) showed that their respective SARAH domains have 54.1% sequence identity and 89.4% sequence similarity. These values are significantly large (i.e., larger than in other recent studies using homology modeling in conjunction with MD simulations [18]), and this observation is both a strong motivation and a justification for using the crystal structure of the RASSF5 SARAH domain as a template for the homology modeling of the RASSF1A SARAH domain structure [6].
We continued our study by generating an atomistic model of the structure for the RASSF1A monomer built by homology modeling, using the RASSF5 crystal structure (4LGD: Chain G) as a template (Fig 1). Subsequently, the RASSF1A SARAH structure was docked onto the MST2 SARAH domain. A large number (approx. 2000) of possible dimer structures were generated using the Zdock program [19–21] (Fig 3). We note that coarse-grained modeling approaches, including docking, have been very successful in other recent computational approaches for studies of protein-protein interactions [22–25]. The Zdock scoring function values for MST2-MST2 (blue), MST2-RASSF5 (red), and MST2-RASSF1A (green) dimers were shown to be effective in identifying only a few most favorable binding modes, indicating that our strategy was valid for this study. Importantly, our analysis assigned high Zdock scores for cases when crystal structures are available (Fig 3 red, yellow and purple arrows), further validating our approach. Thus, we decided to use the docking-generated structures with the highest scoring function as initial models to study the MST2-RASSF1A dimer in the subsequent atomistic molecular dynamics study. However, we note that other recent docking approaches, including additional factors such as side chain flexibility, may further improve structural models before being refined with atomistic MD approaches [23, 24].
Interestingly, the MST2-RASSF1A structure with the highest scoring function corresponded to that in which both protomers are aligned in an anti-parallel topology, similar to those found in the crystal structure of MST2-RASSF5 [11, 13].
In addition to the analysis of the dimerization of different SARAH domains we also analyzed the structure of the interaction between the SARAH domain and a peptide that has been shown experimentally to bind to the MST2 SARAH domain [4]. In this case, similarly to MST2-RASSF1A, our docking study identifies only one structure standing out from the rest with the highest scoring function (Fig 3). Moreover, the top scoring 100 docking structures (out of a total set of 2000) appear to be clustered near the very same interface area, and no additional high-scoring contact area between MST2 and the designed peptide was found using Zdock.
To validate and refine the results of our Zdock study we studied homo- and heterodimers of MST2-MST2, MST2-RASSF5 and MST2-RASSF1A using atomistic molecular dynamics (MD) with explicit water molecules.
A summary of the simulation types performed and analyzed here is given in Table 1.
Fig 4 shows the three dimers in our MD study: MST2-RASSF5 from crystal structure (4LGD), MST2-MST2, and MST2-RASSF1A. The structure of the MST2-MST2 homodimer modeled corresponds to a parallel alignment of both protomers (Fig 4A). The crystal structure of the MST2-RASSF5 dimer was refined with MD (Fig 4B). The structure of the MST2-RASSF1A dimer selected was the one with the highest scoring function, which additionally presented a parallel alignment of both SARAH domains (Fig 4C).
Cα root-mean-square deviation (RMSD) values for all four atomistic systems, both homo- and hetero-dimers, are shown in Fig S1 in S1 File. We observed that the RMSD values in both systems are converged and remain stable along our entire trajectories. Additionally, solvent accessible surface area (SASA) was estimated in order to assess the solvation properties of hetero-dimers. Fig S1 in S1 File shows that in all cases, SASA values also converged, remaining almost constant and not showing any significant changes along the trajectories. Thus, the buried surface at the interface between protomers upon dimer formation is almost constant along the trajectory.
Root-mean-square fluctuation (RMSF) values for all four dimer types were calculated in order to gain insight in the flexibility of the SARAH domains upon complex formation (Fig 4, lower panel r.h.s.). In general, all the residues fluctuate very little with respect to the initial structure (RMSF depicted as blue), with the exception of the C-terminal loop (white-red) which presents more flexibility. In the case of the MST2-MST2 dimer, the residues located in the centers of the helices presented more mobility as compared with the corresponding residues in the RASSF1-6 dimers.
We carried out additional MD simulation at different temperatures over the MST2-RASSF5 crystal structure dimer, at 400 K, 450 K and 500 K. Those trajectories showed that even at 400 K, MST2-RASSF5 dimer was stable at least during 160 ns (see Fig S2 in S1 File), while in the 450 K and 500 K, both SARAH domains became unstable after 30–50 ns.
We also calculated the interaction energy between protomers within the dimers for all the systems studied. The total interaction energy (see Fig S3A in S1 File, black line) was defined to include electrostatic (green) and van der Waals (vdW, blue) interactions. This analysis showed that the dominant term is electrostatic (Fig S3A in S1 File). The vdW term remains constant along the trajectory in all the cases, and it has a relatively minor contribution. The electrostatic interaction term accounts for -527.5 ± 69.6, -483.4 ± 77.1, -426.9 ± 71.5 kcal/mol for MST2-MST2, MST2-RASSF5, and MST2-RASSF1A dimers respectively, while the vdW term accounts for only -149.1 ± 7.1, 118.4 ± 7.6, 114.6 ± 7.2 kcal/mol.
These results based on MD simulations also show that the total interaction energy of the MST2-MST2 dimer is only marginally more stable than the MST2-RASSF5 and MST2-RASSF1A dimers (Fig S3 in S1 File). The average total interaction energy for MST2-MST2 is -676.5 ± 67.8 kcal/mol, while in the case of MST2-RASSF5 and MST2-RASSF1A, it accounts for -601.7 ± 75.0 and -541.5 ± 70.4 kcal/mol, respectively. Additionally, the total interaction energy distributions, calculated for each dimer along its trajectory (Fig S3B in S1 File), show that all three dimers present similar populations within the same range of energies.
To further probe the interaction preferences between the SARAH domains of MST2 and RASF1A and RASSF5, we have also used six different contact potentials to estimate the normalized contact energies based on structures obtained from MD trajectories in each case. While contact potentials offer only rough approximations for the relative stability of molecular structures, have been remarkably successful in other studies of protein-protein interactions [22, 24]. In Fig S11 in S1 File we illustrate a comparison of histograms of residue-residue contact potential values from 20 x 20 contact potential matrices [26] developed by Hinds and Levitt (HL [27]), Betancourt and Thirumalai (BT [28]), Miyazawa and Jernigan (MJ-99 [29]), Skolnick et al. (SJKG [30], and SKO from Ref. [31]), and Tobi et al.[32] (TSLE from Ref. [32]). These normalized contact potentials are further used in Fig S12 in S1 File to calculate relative contact potentials values for structures of MST2-RASSF1 (notation MR1 on the horizontal axis), MST2-MST2 (MM) and MST2-RASSF5 (MR5) dimers. The calculations were performed for the six popular contact potentials represented Fig S11 in S1 File for three cases: (A) dimer structures before MD, (B) the dimer structures from our MD simulations corresponding to frames with the smallest RMSD values compared to the average over the respective trajectory (RMSDave), and (C) the dimer structures from the same MD trajectories but corresponding to frames with the largest RMSDave (to illustrate that even in this case the relative values for MR1 are still smaller than for MM and MR5 dimers). In Fig S12 in S1 File, a residue-residue contact cut-off distance of 5.5 Å between side-chain atoms was used, though we obtained similar results when using cut-off distances of 5.0 Å and 6.0 Å on the same structures.
Interestingly, as shown in Fig S12 in S1 File, there is a remarkable agreement between results obtained for the six different contact potentials used here, in spite of their diversity and well-known approximate accuracy due to their intrinsic coarse-grained character. Nevertheless, in agreement with the atomistic MD energies, the results suggest that all three dimer types have similar stabilities, though this time MST2-RASSF1A dimers appear to be marginally more stable. This is in agreement with experimental observations that both RASSF1A and RASSF5 SARAH domains could disturb competitively MST2 SARAH homodimers.
As an additional probe of the relative stability of the various dimeric systems for which we have MD simulations available, we have also calculated potential of mean force (PMF) profiles as presented in Figs S14 and S15 in S1 File. In Fig S14 in S1 File, results are presented for the MST2-MST2, MST2-RASSF1 and MST2-RASSF5 systems calculated from the corresponding dimer all-atom MD simulations using the recent dynamic histogram analysis method (DHAM) method [33]. The profiles were calculated for the distance between carbon alpha (CA) atoms of the two monomers that has the smallest average value along the corresponding MD trajectory. To probe convergence, in Fig S14 in S1 File the PMF profiles are presented for (A) the full trajectory, (B) the first third of the data, (C) the second third, and (D) the final third of the data. The first 20 ns (i.e., ~10%) of data from each trajectory were not included in this analysis. Though free energy calculations for interactions between large molecular complexes are notoriously difficult, the calculated PMF profiles suggest that the RASSF1 SARAH domains could bind better to monomeric MST2 SARAH domains than MST2 itself (e.g., the MST2-RASSF1 PMF values in Fig S14 in S1 File, red curves, appear to have narrower profiles when compared to MST2-MST2, blue curves, though this effect is weaker in the MST2-RASSF5 case, yellow curves).
Similarly, PMF profiles for the MST2-PEPA, MST2-PEPL, MST2-PEPs, MST2-SCR, RASSF1-PEP, and RASSF1-SCR systems calculated from the corresponding dimer all-atom MD simulations using the DHAM method are also presented in Fig S15 in S1 File [33]. Here, the PMF profiles illustrate clearly the trends discussed in detail above (e.g., most notably that scrambled peptides have a lower dissociation barrier than their corresponding counterparts).
In order to provide further information on the MST2-MST2 homodimer, we have carried out a docking study (see yellow plot in the histogram in Fig 3) of the MST2-MST2 dimer using the crystal structure available (4HO9). We have evaluated the structural differences between the 4HO9 structure and the results from docking using 4HO9 as template. Our results (see in Fig S13 in S1 File upper panel) indicate that the differences between both docked and crystal structures are remarkably small, corresponding to alpha carbon (CA) RMSD of 2.208Å. Furthermore, we have repeated the experiment, comparing the dimer obtained from docking using 4LGD (MST2 monomer) and the available 4HO9 crystal structure (see in Fig S13 in S1 File lower panel). Once more, the docking results are in very high agreement with the crystal structure, with a CA RMSD of 2.096 Å.
In order to gain additional insight of the different interactions that occur upon complexation between the protomers, salt bridges and hydrogen bonds were analyzed. A deeper analysis of the MD structures revealed several salt bridges present in these dimers (see Fig 5 and Table 2).
The MST2-MST2 dimer presents symmetrical interactions between K441-D479, D456-R474 and E462-R469, with the last two pairs being dominant in terms of shorter distance along the trajectory (see Fig S4 in S1 File). This is because the first salt bridge (K441-D479) belongs to the terminal ends of both helixes and it is very flexible, making this salt bridge less stable. The MST2-RASSF5 dimer contains a large number of salt bridges, and a total of 12 interactions were found, while in MST2-RASSF1A dimer 9 salt bridges were located. In the MST2-MST2 dimer the electrostatic interactions are mainly between the two pairs R469-E463 (Pc = 0.745) and R474-D465 (Pc = 0.906), where Pc is the probability of formation of the corresponding contact as estimated from our MD trajectories. In the case of MST2-RASSF5, more salt bridges come into play accounting for the highest contact propensity along the MD simulation: R467-E388 (Pc = 0.877), K473-E366 (Pc = 0.154), R474-E385 (Pc = 0.289) which are competing with the formation of R474-E388 (Pc = 0.800) and E463-K398 (Pc = 0.751). Furthermore, the MST2-RASSF1A dimer showed similar interactions with the MST2-RASSF5 dimer, R467-E316 (Pc = 0.911) and R474-E316 (Pc = 0.819), but also some similarities with the MST2 homodimer, D465-R331 (Pc = 0.981) and E462-K326 (Pc = 0.609). In view of these results, we concluded that R474 (MST2) and E462 play an important role in all these dimers, based on the Pc values observed for those amino acids. R467 and D456 are also highlighted as main “anchoring” contacts between dimers, especially between RASSF1-6 and MST2.
Our structural analysis identified the charged residues and salt bridges involved in the formation of the three models (Fig 5): (A) MST2-MST2 (blue-blue), (B) MST2-RASSF5 (blue-red), (C) MST2-RASSF1A (blue-green). As observed, electrostatic interactions play a primary role in controlling the assembly and stability of MST2 homo- and heterodimers. Notably, in MST2-RASSF5 and MST2-RASSF1A interactions the two antiparallel helices present a significant charge complementarity between their N-terminal and C-terminal regions. The time-dependent dynamics of several representative salt bridges is illustrated in the Fig S4 in S1 File.
Recent studies have pointed out that hydrophobic isoleucine-leucine (ILE-LEU) pair interactions mediating packing between α-helixes also can contribute to the stabilization of such dimers [34]. Therefore, we analyzed the ILE-LEU pairs within the dimers, and several were found in MST2-MST2 and MST2-RASSF dimers (Fig 6). The number of those pairs and the particularly strategic location at the N-terminal, C-terminal and in the center of the dimer suggest a large stabilization. Thus, we conclude that these ILE-LEU pair interactions together with the salt bridges are the main elements responsible for dimer stability and, possibly, formation in the MST2-RASSF1-6 and MST2-MST2 SARAH domains.
In addition to the salt bridges discussed above, several hydrogen bonds (HBs) were identified between protomers, which also contribute to the stability of dimers. The total number of hydrogen bonds (NHB, calculated using VMD [35]), that occur along the MD simulation within the dimers, is 16, 31 and 24 for MST2-MST2, MST2-RASSF5 and MST2-RASSF1A. Table 3 shows the total occupancy of those HBs along the whole MD trajectory, including the donor and acceptor motif. Furthermore, the time-dependent dynamics of the hydrogen bonds is depicted also in Fig S5 in S1 File. We observed that MST2 key hydrogen donors correspond to Y470 and R474 amino acids that are involved in HB within all the dimers studied here. Other main important contributions arise from the residue R467, which presents hydrogen bonds with Q389 (12.5%) and E388 (52.0%) in RASSF5, and with E316 (48.3%) of RASSF1A, and from the residue R469 that presents hydrogen bonds with E462 (98.7%) in MST2-MST2 dimer.
Additionally, we calculated the hydrophobic (SASAH) and hydrophilic (SASAP) fractions of the solvent accessible surface area (SASA) for the representative structure of each system in order to gain insight on the hydrophobic interactions (see Fig S6 in S1 File). From this calculation, it was not clear whether hydrophilic (pink) or hydrophobic (green) exposed areas are predominant. To clarify this point we quantified the total exposed area in these dimers, together with the SASAP and SASAH values of the representative structures using the GetArea program [36]. These values, given in Table 4, show that the MST2-MST2 dimer has the largest exposed hydrophobic surface, while MST2-RASSF5 presents the smallest one.
In previous work, the interaction between RAF1 and MST2 protein was studied from experimental and computational point of view showing that MST2 coordinates crosstalk between the mitogenic Raf and pro-apoptotic MST2 pathway [4]. This study showed that a 17-mer peptide designed based on the binding site of RAF1 to the MST2 SARAH domain was able to disrupt RAF1-MST2 dimerization. Understanding how such peptides bind and disrupt the dimerization process is key for future development of anti-cancer drugs that can activate MST2 by releasing it from the inhibitory interaction with RAF1. In addition, it will help to design peptides that either improve or do not affect RASSF1A or MST2 interactions, as desired, or could also simultaneously disrupt RASSF1 and MST2 dimerization. For that reason, we carried out a computational study of the possible interactions between the so-called disruptor peptide and the MST2 and RASSF1A SARAH domains. In order to perform this study, we used homology modeling and docking studies to obtain the initial structures and, in a subsequent step, full atomistic MD simulations to validate and analyze these interactions.
Four different systems were tested. We first used the disruptor peptide (PEPS) with the sequence “RYTAKRQPILDAMDAK” corresponding to the minimal sequence of MST2 known to interact with RAF1 [1]. A longer peptide (PEPL) “IEELRQRYTAKRQPILDAMDAK”, including flanking sequences of MST2-RAF1 interaction domain, was also tested in order to see the influence of the length of the peptide on the interaction. Both initial dimeric structures (MST2-PEPS and MST2-PEPL) were obtained from the highest scoring structure in the docking study. In addition, PEPL was aligned to the original MST2-MST2 SARAH domains (PEPA). Finally, we also studied a control peptide, where the PEPL sequence was scrambled, “TDKRALDQLRMQEIKARYPFQA”. As the RAF1 binding domain overlaps with the RASSF1A binding in the MST2 SARAH domain we also investigated the structures of RASSF1A-PEPA and RASSF1A-SCR dimers (see Fig S9 in S1 File for sample structures of RASSF1A-SCR dimers along the 200 ns MD trajectory).
Fig S7 in S1 File shows the RMSD of the four MST2-Peptide systems studied. When we compared the interactions between PEPS and PEPL and the MST2 SARAH domain we observed that in the latter, the dimer was more stable than in the former. In fact, the fluctuation of PEPS around the MST2 SARAH domain is larger than PEPL. Once the peptide is aligned (PEPA) with the MST2 SARAH domain in the MST2-MST2 SARAH domain dimer, the peptide remains almost constant and no significant deviation from the initial structure can be found after 200 ns. This was shown by the stable RMSD along the MD trajectory.
Next, we compared MST2-PEPL with the MST2-SCR system, both from the best scoring docking structures, to gain insight in the structural stability of both systems. We observed that the RMSDs with respect to the initial structure (black) converge to the same values, between 5–7 Å, while with respect to the average structure (red) MST2-PEPL converges to a more stable structure (2–3 Å), and MST2-SCR seems to drift towards RMSD of 4 Å. This may be indicative of the decreased stability of the SCR peptide versus the PEPL. Furthermore, the highly structural stability of MST2-PEPA corroborates the experimental observation that the PEP but not the SCR can disrupt RAF1-MST2 interaction by blocking the RAF1 interaction domain of the MST2 SARAH domain.
Our simulations showed that the interaction between RASSF1A and PEPA dimer, the structure remains almost constant along the MD trajectory with a slight variation at 170 ns due to the fluctuation of the C-terminal part of the RASSF1A SARAH domain.
For RASSF1A-SCR dimer the picture was quite different, and RMSDave in respect to the average structure keeps fluctuating (see Fig S7 in S1 File). This is due to the unfolding of the scrambled peptide along the trajectory, which destabilizes this dimer (see Fig S8 in S1 File). This is also revealed by the large SASA variation for the RASSF1A-SCR dimer, indicating that PEP which was designed using the MST2 SARAH domain sequence could also bind to the homolog sequence of RASSF1A SARAH domain, since both SARAH domain sequences have 31.4% identity and 64.6% similarity (Fig 2A).
Root mean square fluctuation (RMSF) values for all six dimer types were calculated in order to gain insight into the flexibility of the SARAH domains upon complex formation (Fig 7). The differences between aligned structures for PEPA and the scramble SCR one are notable. While the aligned peptide, PEPA, is quite rigid along the trajectory, the SCR peptide shows more flexibility and visits less stable conformations.
In order to provide further insight in the relative stability of dimers containing PEP and SCR peptides, we analyzed the interaction energy along their corresponding MD trajectories (Fig 8 and Fig S10 in S1 File). Only the dimeric systems (MST2-PEPL, MST2-PEPA, and MST2-SCR) and (RASSF1A-PEPA, and RASSF1-SCR) have the same overall sequence composition and thus can be compared exactly to each other, though the estimated energy corrections for different residue composition are very small in this case for all systems.
In Fig S10 in S1 File is shown the total interaction energy (black) for the systems considered. Clearly, the MST2-PEPA system presents the strongest interaction energy. Again, once the peptide is aligned with the MST2 SARAH domain, the structure is very stable. In the case of MST2-SCR, the interaction energy drifts and decreases along the trajectory, showing a destabilization of the dimer. Interestingly, a similar behavior is found in the RASSF1A-PEP complex, in which this dimer is very stable. This further indicates that the peptide can also interact favorably with the RASSF1A SARAH domain and hence can disrupt the MST2-RASSF1A dimerization. Furthermore, the interaction energy distribution along the trajectory (see Fig S10 in S1 File) clearly shows that the MST2-SCR system presents a broad distribution profile with only a small population at more negative energies. However, in the MST2-PEPA distribution a peak of the energy values between -350 and -250 kcal/mol suggests a higher stability of the MST2-PEPA system as compared to MST2-SCR. Similar relative interaction strengths have been found for the RASSF1A-PEP and RASSF1A-SCR systems confirming the higher relative stability of dimers containing PEP peptides.
Since the computational simulations suggest the existence of stable dimers between MST2 and the disruptor peptide, we decided to test it experimentally in order to validate our atomistic models and prove the appropriateness of this approach for in silico testing of small molecules that target SARAH domain binding. The disruptor peptide was designed to include the minimal binding interface between RAF1 and MST2 as mapped by peptides arrays, and we showed that it disrupts this interaction very effectively [4]. In order to test our prediction that this peptide could also disrupt the RASSF1A-MST2 dimer, we performed co-immunoprecipitation experiments in MCF7 cells treated with the PEP or the SCR. Our experiments clearly demonstrated that this peptide disrupts the MST2-RASSF1A dimerization (Fig 9A). Moreover, when we tested the effects of this peptide on the MST2-MST2 interaction we also saw a disruption of the MST2 homodimerization (Fig 9B) that could have important effects in the activation of the kinase activity of this protein. All together, these experiments confirm the accuracy of our atomistic models and validate the molecular modeling and simulation methods used in this work.
SARAH domains are highly conserved throughout evolution and mediate the protein-protein interaction between the MST1/2 kinases and members of the RASSF family. RASSF1A and RASSF5 are key bona fide tumor suppressor genes, whose protein products have been shown to regulate MST1/2 kinase activity [4, 37]. The effect of RASSF1A and RASSF5 over MST1/2 activation is mediated by heterodimerization through the SARAH domain of these proteins. Thus, for a complete description of how RASSF1A and RASSF5 regulate MST2 we need to understand first how the SARAH domains mediate the formation of these complexes [6]. Previous studies have focused mainly on the RASSF5-MST2 dimer using crystal structures, NMR spectroscopy and performing limited computational analysis based only on structure minimization rather than using extensive atomistic MD. This approach has significant limitations since it does not allow the study of the dynamical behavior of protein dimers under realistic conditions including explicit solvent molecules. In order to get a better understanding of the structure of the SARAH dimers we used a combination of homology modeling, docking, molecular modeling and atomistic molecular dynamics (MD) methods, to construct and study a variety of models of MST2 homodimers, and of MST2-RASSF SARAH heterodimers, including the important case of RASSF1A-MST2 dimers. In the first step, based on the high sequence identity and similarity between RASSF1A and RASSF5, we used the MST2-RASSF5 dimer structure (Fig 1) based on the available crystal structure (PDB ID: 4LGD) [13] as a template for building new structural models for the RASSF1A SARAH domain interacting with MST2, as well as for MST2 homodimers. In the second step, to validate our models and also to search for alternative solutions, we used molecular docking to generate a broad variety of dimeric homo and heterodimer structures. Subsequently, we used atomistic MD simulations including explicit water representation to test the stability of our dimer models that have been previously ranked as best in the docking stage. With this approach, we showed that our models present a significant stability when probed with atomistic MD simulations, justifying additional experimental tests.
Importantly, in addition to the study of SARAH dimers we also simulated the structure of a synthetic peptide that was purposely designed to be a strong MST2 binding partner, and was used in our previous experiments. This peptide was designed to disrupt the RAF1-MST2 dimer, and to potentially activate MST2 kinase activity by binding to a short sequence of the MST2 SARAH domain where RAF1 interacts with this protein. Interestingly, our new simulations presented here show a different scenario for the effect of this peptide on the formation of RASSF1A-MST2 and MST2-MST2 dimers. Essentially, our new simulations indicate that the peptide would also prevent the formation of the RASSF1A-MST2 dimers, in addition to inhibiting the formation of MST2 homodimers. Importantly, we have validated experimentally these predictions of our in silico results, indicating strongly that we have correctly characterized the structure of the SARAH domain dimers, and confirming the relevance of our approach to the study of the structure of SARAH domains in particular, and of other protein-protein interactions in general. The molecular interaction mechanisms revealed here also shed new light on how RASSF1A regulates the MST2 kinase activity via dimer formation.
Significantly, our specific findings regarding SARAH domain interactions, together with the general methods used in this work, can help to design more effective strategies to target human cancer tumors (e.g., by deregulation of their RASSF1A/MST2 signaling networks). We also hope that this study is a first step towards integrating atomistic-level mechanistic information about the structures and conformational dynamics of proteins interacting through SARAH domains, with information available on their system-level functions in cellular signaling.
The initial structure of an MST2-RASSF5 SARAH domain dimer was constructed based on the crystal structure PDB ID: 4LGD, chain C and G for MST2 and RASSF5, respectively [13].
To our knowledge, there is no crystal structure of RASSF1A. Therefore we have used the sequence that is available from SwissProt [38] (Uniprot Q9NS23). Searches for homologous protein were carried out using UniProt [39] and the RCSB PDB data banks. The Clustal Omega program [15–17] was used for sequence alignment. A homology model of the RASSF1A SARAH domain was built using the comparative modeling environment, SWISS-MODEL [40–43]. The PDB structure 4LGD (chain G) for the RASSF5 SARAH domain was used as a template. In order to get an optimal packing structure between RASSF1A and MST2 SARAH domains, a docking study was carried out, which used the most probable structures (i.e, with the highest scoring function) generated by the Zdock server [19–21].
Four systems were prepared for atomistic MD simulations: one based on the crystal structure for the MST2-RASSF5 SARAH domain dimer, one with homology model structures for the MST2-RASSF1A SARAH domain dimer, one corresponding to the homodimer MST2-MST2 using the MST2 protomer structure from crystal structure and docking another copy of the MST2 protomer, and finally one with a homology model for the designed disrupting peptide.
Each system was solvated with explicit TIP3P water molecules [44] prior to minimization, heating and equilibration. The total number of atoms for each system including water molecules is reported in Table 1.
MD simulations were performed using the NAMD software [45] with the CHARMM36 force field [46]. All the atomistic MD simulations were performed in the NPT ensemble (i.e. constant number of atoms, pressure and temperature), using periodic boundary conditions, as in our similar recent MD studies [47–49]. We used the modified Nosé-Hoover Langevin piston method implemented in NAMD [50, 51] with damping time of 0.1 ps, while maintaining a pressure of 1.01325 bar. The temperature was set to 310 K and controlled using a Langevin thermostat with a 1 ps-1 damping coefficient. Ions were added using the automatic script provided in VMD [35] to achieve a neutral pH. The Particle Mesh Ewald method was used to include electrostatic effects [52]. The switching distance for non-bonded electrostatic and van der Waals interactions was 9.5 Å with a cut-off distance of 12 Å. The integration time step was 1 fs.
A summary of the simulations performed and analyzed here is given in Table 1. To address convergence, errors were estimated by block averaging and all the MD simulations were performed at least twice longer than needed to obtain the average values reported in each case. In addition, the PMF profiles presented in Figs S14 and S15 in S1 File were calculated and presented for different trajectory segments (e.g., the first, second and last third of each trajectory shown in Fig S14B, S14C, and S14D in S1 File, respectively, and the first and second half of each corresponding trajectory of Fig S15B and S15C in S1 File), as well as for the entire trajectory data (see Fig S14A and S14B in S1 File).
|
10.1371/journal.pntd.0004482 | Early Indicators of Fatal Leptospirosis during the 2010 Epidemic in Puerto Rico | Leptospirosis is a potentially fatal bacterial zoonosis that is endemic throughout the tropics and may be misdiagnosed as dengue. Delayed hospital admission of leptospirosis patients is associated with increased mortality.
During a concurrent dengue/leptospirosis epidemic in Puerto Rico in 2010, suspected dengue patients that tested dengue-negative were tested for leptospirosis. Fatal and non-fatal hospitalized leptospirosis patients were matched 1:1–3 by age. Records from all medical visits were evaluated for factors associated with fatal outcome. Among 175 leptospirosis patients identified (4.7 per 100,000 residents), 26 (15%) were fatal. Most patients were older males and had illness onset during the rainy season. Fatal case patients first sought medical care earlier than non-fatal control patients (2.5 vs. 5 days post-illness onset [DPO], p < 0.01), but less frequently first sought care at a hospital (52.4% vs. 92.2%, p < 0.01). Although fatal cases were more often diagnosed with leptospirosis at first medical visit (43.9% vs. 9.6%, p = 0.01), they were admitted to the hospital no earlier than non-fatal controls (4.5 vs. 6 DPO, p = 0.31). Cases less often developed fever (p = 0.03), but more often developed jaundice, edema, leg pain, hemoptysis, and had a seizure (p ≤ 0.03). Multivariable analysis of laboratory values from first medical visit associated with fatal outcome included increased white blood cell (WBC) count with increased creatinine (p = 0.001), and decreased bicarbonate with either increased WBC count, increased creatinine, or decreased platelet count (p < 0.001).
Patients with fatal leptospirosis sought care earlier, but were not admitted for care any earlier than non-fatal patients. Combinations of routine laboratory values predictive of fatal outcome should be considered in admission decision-making for patients with suspected leptospirosis.
| Leptospirosis is a common tropical illness that results from exposure to the urine of animals infected with Leptospira bacteria. Because leptospirosis shares signs and symptoms with other common tropical illnesses such as dengue, identification of patients with leptospirosis can be challenging. Early identification of patients with leptospirosis is necessary to initiate antibiotic therapy and in some cases provide in-hospital management. During an epidemic of leptospirosis in Puerto Rico that occurred during a concomitant dengue epidemic, we identified leptospirosis patients by screening specimens from suspected dengue patients. Of 175 leptospirosis patients identified, 26 (15%) died. After comparing leptospirosis patients that died to patients of a similar age that were hospitalized but survived, we observed that fatal cases were more often sent home after their first medical visit. We next identified several routinely available laboratory values from patients’ first medical visit that were associated with patients that died. Clinicians can use such laboratory values to diagnose and hospitalize leptospirosis patients at increased risk for fatal outcome.
| Leptospirosis is an emerging zoonosis caused by infection with bacterial spirochetes of the genus Leptospira, and is endemic throughout the tropics where >1 million cases and ~60,000 deaths occur annually [1, 2]. Human infection typically occurs through direct or indirect contact with the urine of infected animals [1]. Leptospirosis is typically a mild acute febrile illness (AFI); however, ~10% of patients progress to severe leptospirosis with acute kidney failure, jaundice, and/or pulmonary hemorrhage [1, 3]. The case-fatality rate for patients with severe leptospirosis ranges from 5–20% [4–6].
Due to similar clinical presentations, leptospirosis may be misdiagnosed as dengue [7–9]. Delayed or misdiagnosis of leptospirosis patients has been associated with increased mortality, potentially due to delayed administration of antibiotics [10–15]. Therefore, identification of early clinical markers of patients at risk for severe disease to thereby enable earlier patient admission may result in improved outcome. Severe thrombocytopenia, increased serum creatinine or BUN, hemoptysis, dyspnea, and jaundice have been associated with severe or fatal outcome in leptospirosis patients [5, 12, 14–18]; however, few studies have captured data from patients’ entire clinical course to identify demographic characteristics, clinical findings, or missed opportunities in clinical management associated with poor outcome [12, 14]. Consequently, early clinical indicators of patients that have or will develop severe disease have not been well elucidated.
During 1990–2014, a total of 729 leptospirosis cases were reported to Puerto Rico Department of Health (PRDH), of which 78 (10.7%) were fatal (S1 Fig). Such surveillance enabled documentation of leptospirosis epidemics in 2006, 2007, and 2010. However, because of underreporting of leptospirosis [19], which is attributable in part to misdiagnosis as dengue [20–22], it is unclear if these data represent the true epidemiologic trends of leptospirosis. Factors associated with severe or fatal outcome in leptospirosis patients have not previously been investigated in Puerto Rico.
To better understand the epidemiology of leptospirosis during the 2010 dengue epidemic in Puerto Rico [23], we conducted enhanced surveillance by performing leptospirosis diagnostic testing on specimens from suspected dengue patients. We also reviewed medical records from all health care visits of identified leptospirosis patients to identify demographic characteristics, clinical signs and symptoms, laboratory values, and clinical practices associated with fatal outcome.
This study was approved by the Institutional Review Board at the Centers for Disease Control and Prevention (CDC) (protocol # 6285).
Leptospirosis cases in Puerto Rico in 2010 were identified from four sources. First, suspected dengue cases reported via the Passive Dengue Surveillance System (PDSS) [24] that had no evidence of dengue virus (DENV) infection by rRT-PCR or anti-DENV IgM ELISA [23] (N = 2,519) were eligible to be tested for evidence of Leptospira spp. infection by microscopic agglutination test (MAT) [25] and polymerase chain reaction (PCR) with primers specific for Leptospira spp. LipL32 [26]. Specimens selected for leptospirosis testing (n = 1,133) came from cases for which either: a) paired acute and convalescent specimens were available (n = 654); or b) only a convalescent specimen was available and the case had reported fever, body pain or headache, and jaundice, hemorrhage, or pleural effusion (n = 479). Second, fatal leptospirosis cases were identified via the Enhanced Fatal AFI Surveillance System (EFASS) in which: a) serum or tissue specimens collected during autopsy were tested by MAT, PCR, or immunohistochemistry (IHC) [27]; and b) death certificates were reviewed for use of “leptospirosis” or “Weil’s disease”. Third, all leptospirosis cases reported to PRDH along with a positive diagnostic test result via the Notifiable Diseases Surveillance System (NDSS) were included. Last, two commercial laboratories were queried for leptospirosis cases that tested positive by IgM dot blot. Cases identified through more than one data source with matching names and dates of birth were considered a single case.
A laboratory-positive leptospirosis patient was defined as a person that had evidence of infection with Leptospira spp. by detection of: i) antigen in a tissue specimen by IHC; ii) nucleic acid in a serum or tissue specimen by PCR; iii) ≥4-fold rise in MAT titer in paired serum specimens; iv) MAT titer ≥800 in a single serum specimen; v) anti-Leptospira IgM antibody at a private laboratory; or vi) MAT titer ≥100 but <800 in a single serum specimen. A confirmed leptospirosis patient was defined by any of criteria i–v; a probable leptospirosis patient was defined by criteria vi. A suspected fatal leptospirosis patient was a person who died in Puerto Rico in 2010, had the word “leptospirosis” written on the death certificate, and had either: a) no leptospirosis diagnostic testing performed; or b) negative diagnostic testing performed at a commercial laboratory on a specimen collected within five days of illness onset.
Each fatal, laboratory-positive leptospirosis patient (i.e., cases) was matched by age within five years with up to three non-fatal, hospitalized, laboratory-positive leptospirosis patients (i.e., controls). All available medical records–including private office, out-patient clinic, emergency department, and inpatient hospitalizations–during the episode of illness were reviewed. Controls that left the hospital against medical advice or had incomplete medical records were replaced.
The frequencies of clinical, demographic and laboratory data were calculated by performing descriptive analyses of all leptospirosis patients identified in 2010 and compared using Student’s t-test or Chi squared test. Rates of leptospirosis by age group and municipality of residence were calculated using data from the 2010 United States Census [28]. Statistical differences and modeling of matched case-control data were performed using exact conditional logistic regression. Due to a limited number of matched pairs, several combinations of clinical lab results were considered for independent predictors of fatal outcome. Normal limits of laboratory values were defined by accepted standards [29].
All data analyses were conducted using SAS version 9.3 (SAS Institute Inc., Cary, NC), graphs were produced in SAS and Microsoft Excel (Microsoft Corp., Redmond, WA), and maps were created using ArcView (ESRI, Redlands, CA). Specimens were not anonymized prior to diagnostic testing to enable reporting of results to requesting physicians. Data included in the case-control study were anonymized prior to analysis.
Among 1,133 suspected but laboratory-negative dengue cases that were selected for leptospirosis diagnostic testing, 105 (9.3%) were laboratory-positive (S1 Table). Among 802 specimens from patients tested for leptospirosis at a private laboratory, 56 (7.0%) were positive. A total of 57 non-fatal leptospirosis patients were reported via NDSS in 2010, and laboratory diagnostic evidence was provided for 15 (26%). After consolidating individual patients identified by multiple systems, a total of 149 non-fatal, laboratory-positive leptospirosis patients were identified in Puerto Rico in 2010 (4.0 non-fatal patients per 100,000 residents), of which 91 (61%) were confirmed and 58 (39%) were probable leptospirosis patients. Dengue was ruled out for 134 (90%) non-fatal leptospirosis patients by rRT-PCR and/or IgM ELISA [23]; one apparent co-infection was identified in which DENV-1 was detected by RT-PCR and anti-Leptospira spp. IgM antibody was detected at a private laboratory.
A total of 26 fatal leptospirosis patients were identified (0.7 fatal patients per 100,000 residents), of which 21 were confirmed and five were suspected leptospirosis patients; only two (7.7%) had been reported to PRDH. Fifteen fatal, laboratory-positive leptospirosis patients had available kidney and liver tissue specimens, and Leptospira antigen was detected by IHC in all 15. Dengue was ruled out in 18 (86%) of the fatal, laboratory-positive leptospirosis patients and in two (40%) of the fatal, suspected leptospirosis patients. Two patients with fatal DENV/Leptospira spp. co-infection were identified [30]. Among all 26 fatal leptospirosis patients, the most common reported causes of death included respiratory, cardiac, or renal failure, and septic shock (S2 Table).
MAT-positive specimens (n = 130) from laboratory-positive leptospirosis patients showed strongest reactivity to serogroups including Icterohaemorrhagiae (57%), Australis (11%), Mini (5%), Bataviae (4%), Canicola (4%), Cynopteri (2%), Pyrogenes (2%), Pomona (1%), Djasiman (1%), and Autumnalis (1%); 12% had strongest reactivity against more than one serogroup. Of four PCR-positive serum specimens from one fatal and three non-fatal patients, multi-locus sequence typing [31] identified six of seven alleles suggestive of L. interrogans serovar Icterohaemorrhagiae/Copenhageni in the specimen from the fatal patient; MLST was not successful for the other specimens.
Leptospirosis patients had illness onset in all months of the year (Fig 1). Peak incidence of identified fatal and non-fatal leptospirosis patients occurred in October, in association with the rainy season. Most (79%) fatal and non-fatal laboratory-positive leptospirosis patients were male. Leptospirosis patients were identified in all age groups (Fig 2). Incidence was highest in individuals aged 40–69 years and lowest in individuals aged >80 years. Fatal patients were significantly older than non-fatal patients (mean of 50 vs. 41 years; p = 0.02). Confirmed and probable non-fatal leptospirosis patients were not significantly different by age (p = 0.34) or month of illness onset (p = 0.35); however, more confirmed than probable non-fatal patients were male (85% vs. 68%; p = 0.02). Most non-fatal (59%) and fatal (92%) leptospirosis cases were reported to have been hospitalized. Mortality by age group was highest in those aged 60–69 years (1.8 per 100,000 residents).
Fatal and non-fatal leptospirosis cases resided in both urban and rural municipalities across Puerto Rico (Fig 3). In the 59 (76%) municipalities for which cases were detected, incidence was highest in Patillas in the rainy southeast–where enhanced dengue surveillance was conducted at a community health center in 2010 [23]–and in the mountainous, agricultural center of the island. Incidence was lowest in Cabo Rojo in the arid southwest.
A case-control study was conducted in which data from medical records were compared between 21 laboratory-positive fatal cases and 52 age-matched, laboratory-positive, hospitalized but non-fatal leptospirosis controls. Cases and controls did not differ significantly by sex, occupation, or animal or environmental exposure history, nor by reported co-morbidities or chronic medical conditions (S3 Table).
Fatal cases first sought medical care sooner after illness onset than non-fatal controls, and more often sought care at a private or out-patient clinic (Table 1). Although controls first sought medical care at a hospital more frequently than cases, cases were more often admitted or referred for admission at the first visit. Cases and controls did not differ by day post-illness onset (DPO) of hospitalization or duration of hospital stay. Cases were more often admitted to the intensive care unit, intubated, and received hemodialysis (p ≤ 0.02). Blood products were administered to more than half of cases and controls.
Cases more often had leptospirosis included in the differential diagnosis at first medical visit (p = 0.01), whereas controls more often had “dengue” ever mentioned in any medical record (p < 0.01). The timing with which “leptospirosis” and “dengue” were mentioned post-illness onset and post-hospitalization did not differ between cases and controls. Antibiotics were administered to >70% of cases and controls. Corticosteroids were administered to roughly half of cases and controls, most frequently on the day of admission. The frequency, clinical setting (e.g., out-patient clinic vs. hospital), and timing of administration of both antibiotics and corticosteroids did not significantly differ between cases and controls.
Cases presented to first medical visit with either fever or cough less often than controls (Table 2). Similarly, cases less often developed fever throughout hospitalization. Most cases developed jaundice, edema, leg pain, hemoptysis, and altered mental status, while fewer than half of controls had these findings. Developing cyanosis and having a seizure were also associated with fatal outcome.
DPO of first laboratory values did not differ significantly between cases and controls. As compared to controls, at first medical visit cases had significantly elevated white blood cell (WBC) count, proportion of neutrophils, BUN, creatinine, and total bilirubin, and decreased bicarbonate and albumin (Fig 4, S4 Table). For cases, these values were also more frequently outside of normal ranges. Throughout the clinical course, cases had significantly elevated WBC count, proportion of neutrophils, BUN, and creatinine, and decreased hematocrit, bicarbonate, albumin, prothrombin time (PT), and partial thromboplastin time (PTT).
Because fever and cough were the only early clinical signs and symptoms that were associated with fatal outcome and may be spurious findings (see Discussion), only laboratory values were included as parameters in the model. BUN and PTT were removed from the model due to higher specificity of creatinine for kidney injury as opposed to dehydration and infrequency of the test being requested at initial patient presentation, respectively. Clinical laboratory values significantly associated with fatal outcome at first presentation as compared to controls included: decreased serum bicarbonate with elevated serum creatinine, elevated WBC count, or decreased platelet count; and elevated WBC count with elevated serum creatinine (Table 3).
Enhanced surveillance demonstrated a high rate of morbidity and mortality due to leptospirosis in Puerto Rico in 2010 (4.7 and 0.7 cases per 100,000 residents, respectively). Comparable incidences have been observed in other regions of the Caribbean that have conducted enhanced surveillance [32–36], which also demonstrated highest burden in older male agricultural workers and the unemployed [2, 36]. Although the patients identified in Puerto Rico reflected the expected clinical characteristics of severe leptospirosis (i.e., pulmonary hemorrhage, acute kidney injury, and/or septic shock with multi-organ failure), under recognition and underreporting of leptospirosis cases was prominent, as one-third of patients were never diagnosed with leptospirosis and two-thirds were not reported to public health authorities. These findings together demonstrate that leptospirosis remains a neglected tropical disease in Puerto Rico.
Several missed opportunities for early clinical intervention were identified in this study. First, although fatal cases sought care earlier and were more often diagnosed with leptospirosis at first medical visit; however, fatal patients less often first sought care at a hospital, and were not admitted to the hospital any sooner than non-fatal patients. Thus, delayed hospital admission may have contributed to fatal outcome, as has been previously reported [12, 13]. However, we saw no evidence that this delay was associated with the timing of initiation of antibiotic therapy, which did not differ between cases and controls. Although prospective clinical trials of antibiotics have not demonstrated a clear benefit to leptospirosis patient outcome [37], this should not preclude administration of antibiotics to patients with suspected leptospirosis [10]. Last, roughly half of all leptospirosis patients were given corticosteroids, which may result in increased risk of hemorrhage and immunosuppression. A recent systematic review demonstrated no clear benefit to leptospirosis patient outcome by administering corticosteroids [38]; however, prospective clinical trials have yet to be conducted.
To improve recognition of leptospirosis and thereby mediate earlier admission for care, clinicians should be aware of patient characteristics and clinical indicators associated with severe leptospirosis. Most previous studies that identified risk factors associated with death due to leptospirosis relied on data collected during the final medical visit, which may be suboptimal for identification of early indicators of fatal outcome. After matching for age and status of hospitalization, no patient characteristics, including gender and history of smoking [15, 39], were significantly associated with fatal outcome in this study. Similar to previously studies [5, 8, 12, 16–18], we observed that jaundice, hemoptysis, acute kidney injury, and dyspnea or respiratory insufficiency were significantly associated with fatal outcome in this study, though not at initial medical visit. Therefore, the utility of these signs and symptoms may be limited in early identification of leptospirosis patients at risk for fatal outcome. Unexpected risk factors associated with fatal leptospirosis in this study were absence of cough and fever at first health care visit and lack of development of fever throughout hospitalization. Cough at initial presentation has been previously associated with protection from fatal outcome [12], though for unclear reasons. Potential explanations for lack of fever being associated with fatal outcome include incomplete capture of fever history, self-administration of antipyretics, or earlier entry into decompensated shock. Further studies should address the association of these signs and symptoms with fatal leptospirosis.
A prominent utility of this study was the association of common clinical laboratory values with fatal leptospirosis, specifically decreased bicarbonate with decreased platelet count and increased WBC count with elevated creatinine, all of which have been previously associated with severe leptospirosis [5, 8, 16–18, 40]. However, we did not observe that elevated serum potassium either at first presentation or at any point during hospitalization was associated with fatal outcome, as has previously been reported [40–43]. Nonetheless, the values of the laboratory markers of fatal outcome identified in this study tended to be farther outside of normal ranges at first presentation in fatal as compared to non-fatal patients, suggesting that patients with fatal leptospirosis may have progressed to severe disease more rapidly. In line with this, fatal patients were more likely to be diagnosed with leptospirosis earlier than were non-fatal patients, who were more likely to ever be diagnosed with dengue. Because previous studies associated elevated WBC count and elevated serum creatinine with leptospirosis as compared to dengue [20, 44–46], these clinical laboratory values may have utility in not only differentiating leptospirosis patients from dengue patients, but also in identifying leptospirosis patients at risk for poor outcome. Future studies should evaluate the prospective benefit of using such combinations of laboratory values to improve patient outcome through early identification and admission.
Compared to previous studies that have identified risk factors associated with severe or fatal outcome in leptospirosis patients, a major strength of this study was the design of the case-control study. By reviewing medical records from each health care visits made by patients included in the case-control study, and not solely those from the patients’ hospitalization, we avoided biasing results towards points in patients’ illness in which they were likely to be more clinically severe (i.e., at point of hospitalization). This also enabled identification of clinical indicators that would be of clinical utility before patients were hospitalized, which could thereby mediate more rapid diagnosis and/or hospitalization of patients at-risk for fatal outcome. Moreover, by closely matching patients by age we avoided identification of risk factors that may be associated with older populations. These aspects of study design together may account for some differences in factors associated with fatal outcome identified by this study as compared to previous studies that did not control for age [3, 8, 12, 16, 17, 47]. Additional strengths of this study include: conducting surveillance for fatal leptospirosis cases by testing specimens collected during autopsy of patients that died following an AFI, without which many fatal cases would not have been diagnosed; and utilizing multiple surveillance systems to identify fatal and non-fatal leptospirosis patients and subsequently comparing them using a standardized instrument for chart abstraction.
Conversely, one limitation of this study is potential misclassification of some probable leptospirosis patients due to the presence of pre-existing neutralizing antibody. However, because several thousand suspected but dengue-negative cases reported to PRDH in 2010 were not tested for evidence of leptospirosis, the incidence of leptospirosis identified herein is likely an underestimate. Also, although previous studies have demonstrated that predictors of fatal leptospirosis include oliguria [8, 17, 18, 41, 48, 49] and anuria [12], we were unable to explore these factors due to the unavailability of routine clinical data on urine output. Moreover, due to limited sample size, we were also unable to identify specific cut-offs of clinical laboratory values associated with fatal outcome. Last, we were unable to evaluate DENV/Leptospira spp. co-infection as a risk factor for death since most leptospirosis cases were identified by screening suspected dengue cases that tested laboratory-negative for dengue.
Clinical trainings to improve early recognition of leptospirosis patients, interpretation of diagnostic test results, need for case reporting, and clinical management should be conducted among clinicians working in both out-patient and in-patient settings in Puerto Rico. Since improvements in case surveillance and clinical awareness have been associated with decreases in patient mortality due to leptospirosis [6], such trainings may also be needed in other areas of the tropics where clinical under recognition of leptospirosis may be high [2]. Population-based serosurveys should be conducted to accurately quantitate the burden of leptospirosis and identify modifiable risk factors associated with infection, including identification of the animal reservoirs that transmit Leptospira spp. to humans. Such findings can be used to develop educational campaigns to inform the public of population-specific strategies that can be employed to reduce their risk of leptospirosis.
|
10.1371/journal.pcbi.1004264 | Experimental and Computational Analysis of a Large Protein Network That Controls Fat Storage Reveals the Design Principles of a Signaling Network | An approach combining genetic, proteomic, computational, and physiological analysis was used to define a protein network that regulates fat storage in budding yeast (Saccharomyces cerevisiae). A computational analysis of this network shows that it is not scale-free, and is best approximated by the Watts-Strogatz model, which generates “small-world” networks with high clustering and short path lengths. The network is also modular, containing energy level sensing proteins that connect to four output processes: autophagy, fatty acid synthesis, mRNA processing, and MAP kinase signaling. The importance of each protein to network function is dependent on its Katz centrality score, which is related both to the protein’s position within a module and to the module’s relationship to the network as a whole. The network is also divisible into subnetworks that span modular boundaries and regulate different aspects of fat metabolism. We used a combination of genetics and pharmacology to simultaneously block output from multiple network nodes. The phenotypic results of this blockage define patterns of communication among distant network nodes, and these patterns are consistent with the Watts-Strogatz model.
| We discovered a large protein network that regulates fat storage in budding yeast. This network contains 94 proteins, almost all of which bind to other proteins in the network. To understand the functions of large protein collections such as these, it will be necessary to move away from one-by-one analysis of individual proteins and create computational models of entire networks. This will allow classification of networks into categories and permit researchers to identify key network proteins on theoretical grounds. We show here that the fat regulation network fits a Watts-Strogatz small-world model. This model was devised to explain the clustering phenomena often observed in real networks, but has not been previously applied to signaling networks within cells. The short path length and high clustering coefficients characteristic of the Watts-Strogatz topology allow for rapid communication between distant nodes and for division of the network into modules that perform different functions. The fat regulation network has modules, and it is divisible into subnetworks that span modular boundaries and regulate different aspects of fat metabolism. We experimentally examined communication between nodes within the network using a combination of genetics and pharmacology, and showed that the communication patterns are consistent with the Watts-Strogatz topology.
| Systems biology explores the emergence of patterns, structures, and properties in biological systems that cannot be understood by examining individual components. One of the goals of this discipline is to discover how the topological arrangements of proteins within signaling networks endow the networks with features that are important for their cellular functions.
Due to the availability of extensive proteomic data and the existence of strain collections bearing deletion mutations for most of its genes, the budding yeast Saccharomyces cerevisiae is an excellent system in which to discover topological principles governing the design of signaling networks [1,2]. Some network analyses in yeast have examined all of the proteins identified by genome-wide proteomic methods [3–15], while others have focused on essential genes that encode highly connected proteins, referred to as hubs, that are characterized by a lethal phenotype when removed [16–18]. There are disadvantages associated with using either of these approaches to analyze the relationships between network topology and function. First, although proteomic data define connections among proteins, not all connections made by a given protein are relevant when that protein performs its functions in a specific cellular process. Second, lethality can be produced through many different mechanisms, so genes and proteins required for viability do not necessarily have related functions. Third, the contributions of essential genes to survival can only be scored as viability or lethality. Most biological processes, however, exhibit variations in output strength, and incorporation of this information can add value to network models. Fourth, due to the lethal phenotype of these genes, networks of essential genes usually do not provide information about their relationships to the products of interacting nonessential genes.
Here we show that molecular mechanisms used for regulation of fat storage in yeast provide an excellent system for network analysis. First, the mutant phenotype, an alteration in fat levels, is specific enough to suggest that there should be molecular relationships among many of the proteins in the network. Second, the severity of the fat storage defect when a fat level-regulating protein is removed can be quantitatively assessed, and this can be used to determine the protein’s importance to network function. Third, since the loss of a fat storage-regulating gene usually does not cause lethality, mutants selected for quantitative changes in fat content can also be assayed for alterations in other aspects of fat metabolism, such as lipid droplet (LD) morphology and the ability to use different carbon sources for fat synthesis. By using a system-wide approach that combines genetic, proteomic, pharmacological, mathematical, and physiological analysis, we have identified and characterized a physically interconnected network of 94 proteins that regulates fat storage in budding yeast.
The fat regulation network is not scale-free, and is best approximated by the Watts-Strogatz model [19], which generates “small-world” networks with high clustering and short path-lengths. Such networks have many features that are useful for biological control. The importance of a protein to network function is dependent on a particular kind of topological centrality, and the use of this centrality measure may provide a guideline for future analysis of proteins in other biological networks. We were also able to validate the network model by experimentally blocking function of multiple network nodes and showing that the patterns of internode communication predicted by this analysis are consistent with the small-world architecture of the network.
We developed a quantitative 96-well plate assay to screen the viable Saccharomyces cerevisiae deletion collection for alterations in fat content. In this assay, stored fat levels in fixed yeast cells were assessed by staining with the lipid dye Nile Red together with the nuclear dye DAPI and measuring the Nile Red/DAPI fluorescence ratio. Positive mutants were confirmed using a thin layer chromatography (TLC) assay to measure triglycerides, as described by [20](Fig 1A) and by histological staining of fixed cells with another fat-specific dye, BODIPY 493/503. Mutations in 86 genes caused statistically significant increases in fat content (Fig 1 and S1 Table).
54 of the 86 genes identified in this screen have metazoan orthologs or relatives. Of these, brahma (a chromatin remodeling protein orthologous to SNF2), histone H4 (HHF1 ortholog), cdk12 (a Cdk family kinase orthologous to CTK1), and me31b (an RNA helicase orthologous to DHH1), were identified in one or both of the published screens of cultured Drosophila cells for LD morphology phenotypes[21,22].
Extensive proteomic data exist for budding yeast (see [10–13,23]). These data were obtained by a variety of methods, including the two-hybrid system[3,9], the protein fragment complementation assay[4], affinity purification and co-precipitation[7,8], and analysis of global protein phosphorylation patterns[5,6]. We assembled current data on physical interactions for proteins encoded by these genes from the Saccharomyces Genome Database (www.yeastgenome.org) and BioGRID Database (biogrid.org). Remarkably, superimposition of our genetic data on these proteomic data showed that 91% of the mutations we identified affect genes encoding proteins that have physical connections to one another, forming a densely interconnected network (Fig 1B). Most of the proteomic interactions shown in the figure were defined by affinity purification studies, indicating that they define stable and abundant complexes that exist in vivo.
The network defined by screening the viable deletion collection necessarily contains only nonessential genes. However, proteins encoded by essential genes are also known to be involved in regulation of fat synthesis and storage, or are components of pathways identified by the screen of nonessential genes (see next section for discussion of network pathways). We thus added eight proteins encoded by essential genes to the network. Fas1 and Fas2 are subunits of fatty acid synthetase, and Acc1 is acetyl-coA carboxylase. These enzymes are required for de novo synthesis of long-chain fatty acids. Kog1 and Lst8 are components of the TOR complex 1 (TORC1), which we identified as part of the network through the identification of two other TORC1 subunits, Tor1 and Tco89. Ste12 is a transcription factor activated by a MAPK signaling cascade, and Cdc42 is a small G protein that regulates MAPK signaling. Many of the genes in the network encode MAPK pathway components. Cdc39 is a component of the CCR4-NOT core complex, and we identified many other components of this complex in our screen [24–28]. These essential proteins were added in order to create a more complete and biologically meaningful network, not to artificially increase its connectedness by adding “hubs”. In fact, only two of these eight proteins, Cdc39 and Cdc42, have more than four connections to other proteins in the network. Moreover, as described below, we have shown that removal of all of these essential proteins from the network does not significantly affect its topological properties. There are 94 proteins (nodes) and 203 total connections (edges) in the complete network (S2 Table).
To evaluate the significance of this network, we first wished to determine whether the extensive interconnections among its proteins reflect a common biological function and are likely to have been selected by evolution. To address this issue, we asked whether randomly selected collections of similar numbers of proteins from the yeast proteome would display similar connection densities. We generated 200 random collections of approximately 98 proteins (nodes) each and annotated the interactions within each collection that had been identified in published experiments. Of the 200 networks thus defined, most had fewer than 30 total connections (edges), with an average of 24.8 and standard deviation of 11.5. Only one of the 200 randomly selected networks has more than 50 connections. This network contains a ubiquitin protein, UBI4, which forms a hub that makes 43 of the 73 total connections in that network (S1 Fig). In summary, then, the 203 edges in our network make the connectivity in our network ~15.5 standard deviations from randomly selected networks of yeast proteins with a similar number of nodes. This corresponds to a p-value of 10–54.
The network contains extra- and intracellular energy/glucose detection mechanisms controlled by Mth1 and the AMP-activated protein kinase complex (AMPK) encoded by the SNF1, SNF4, and SIP1 genes and their regulators TOS3, ELM1, and SAK1. High levels of extra- or intracellular glucose cause degradation of Mth1[29,30], while high AMP/ATP ratios cause activation of AMPK [31]. A reduction in signaling through either system is likely to produce a perceived surplus of energy that can stimulate fat storage. Mth1 and AMPK are connected to a set of transcriptional regulators and to four processes that may represent outputs. These are: 1) autophagy, which involves the TORC1 and vacuolar H+-ATPase (Vma) pathways [32]; 2) de novo fatty acid and sterol synthesis pathways; 3) MAP kinase (MAPK) pathways; 4) mRNA degradation, elongation, and initiation pathways involving the CCR4-NOT complex and its associated proteins. The MAP kinase pathways are involved in sugar and amino acid starvation responses, while the CCR4-NOT complex is known to repress the translation of mRNAs encoding proteins necessary for utilization of non-fermentable carbon sources and for glucose neogenesis [24,26].
We next examined if proteins that participate in the same process or pathway exhibit a higher tendency to have connections with each other than with other network proteins using a walk-trap algorithm. This function finds densely connected communities via random walks from one node to another [33]. The idea is that short random walks will tend to stay in the same community. This analysis showed that the network contains seven different communities (modules) composed of three or more proteins, and that each module is enriched for proteins that are involved in the same process (Fig 1C).
We performed an analysis of the Gene Ontology (GO) terms associated with each of the proteins in the network. GO includes three categories: Biological Process, Molecular Function, and Cellular Component. S3 Table shows a list of the GO terms that are enriched for network proteins relative to random yeast proteins, with p-values for the statistical significances of the enrichments. Many of these highly enriched categories are very general, with the top three being “growth”, “biological regulation”, and “protein phosphorylation”. Others overlap with the output processes described above (Fig 1), and are properties of the signaling pathways, biological processes, and cellular components represented in the network. Four of the modules within the network defined by the walk-trap algorithm (Fig 1C) are highly enriched for specific GO terms. These are terms related to MAPK signaling (green circles), vacuolar acidification and proton transport (dark blue circles), mRNA catabolic processes (light blue circles), and positive regulation of transcription (red circles). It has been previously shown that GO categories can correspond to clusters within the global yeast protein-protein interaction map[34].
The fact that proteins required for filamentous growth are overrepresented in the network (p = 1.8 x 10–8) is interesting, In limited nutritional conditions, yeast can adopt a filamentous growth pattern that permits a non-motile colony to explore its surroundings for additional nutrients. The induction of this state requires the action of the AMPK, MAP, and Tor kinase pathways [25], which form a considerable portion of the fat regulation network. We hypothesize that some yeast mutants that store excess fat do so because they are in a state of perceived energy excess, which is the opposite of the conditions in which filamentous growth would be favored.
In yeast, MAP kinase pathways are organized as cassettes composed of specific combinations of MAPKKKs, MAPKKs, and MAPKs, and these cassettes can cross-regulate one another [28]. Because we found mutations affecting multiple MAP kinase pathways, there may be redundancy between the cassettes with regard to control of fat storage. Consistent with this, we observed that a double mutant (fus3 kss1) lacking two MAPKs accumulates more fat than either single mutant (S2 Fig). We also note that mutations in TORC1 units and not TORC2 units cause an increase in fat storage, thus implicating only TORC1 in yeast fat storage regulation. This was confirmed by the fact that rapamycin (Rap), a selective inhibitor of the TORC1 complex[27], causes an increase in fat storage in wild-type yeast (S2 Fig).
We did not isolate any ‘lean’ mutants in our screen. This could be due to the fact that our growth media does not contain fatty acids that could be converted to phospholipids necessary for membrane synthesis during cell division, or, alternatively, to genetic redundancy. Indeed, in cases where a low fat storage yeast strain has been reported, a phenotype was only detected when more than one fatty acid synthesis gene is removed [35,36].
Most of the genes we identified have not been previously implicated in fat storage. Earlier screens were done for LD morphology defects, but only ~20% of the 171 genes previously reported to affect LDs showed an increase in fat storage in our assays [37–39]. This is not surprising, because these genes were identified in screens by altered LD morphologies in live mutant cells, not by measuring fat content. Because the goal of our screen was to find mutants with altered fat storage levels, not to study LD dynamics, we fixed cells in order to deactivate pumps that can affect dye uptake and block vesicular traffic to increase the specificity of the dyes to lipid droplets, and used quantitative assays to evaluate fat content [20,40,41].
The fact that the interconnection density of the fat storage regulation network is much greater than that of any randomly generated network of yeast proteins (S1 Fig) implies that the majority of the connections within the network have biological relevance. This makes yeast fat regulation an ideal system in which to examine whether a biological network conforms to a mathematical model for network design. See S1 Text for detailed information on methods for mathematical analysis used in this paper.
Many real-world networks, such as the Internet, power grids, and social networks, have been studied, and they tend to have certain features in common. A social network, for example, contains many different local clusters of people that are linked to each other by mutual acquaintances, so that any person within the cluster can be reached from another person by a small number of steps. Social networks also contain celebrities (“hubs”), who have exponentially more followers than does an average member of the network. Similarly, biological signaling networks tend to contain modules and have hub-like proteins that make many more connections than other proteins.
These features of real-world networks have been simulated using graph theory, which is the study of systems of objects, referred to as nodes, and their relationships, referred to as edges or connections. The properties of networks generated by graph generation models can be characterized by a variety of different parameters (for review see [42]), including degree distribution P(k), global clustering coefficient Cg, modularity M, and path length L. P(k) is the probability that a given network node has a certain number of connections, while Cg is a measure of the degree to which nodes in a network tend to cluster together [43]. M is a measure of the division of a network into communities or modules, produced by the tendency of some nodes to form connections primarily within the community to which they belong. Finally, L is the average number of steps along the shortest paths that connect all possible pairs of network nodes.
The simplest model used to generate networks is the Erdos-Renyi model [44], characterized by an equal probability of forming connections between any two nodes (Fig 2A). Erdos-Renyi networks have a degree distribution similar to a Poisson distribution (Fig 2B), and have low Cg, M, and L values. Most real-world networks are not approximated well by this model, because Erdos-Renyi graphs lack local clustering and hubs.
The Watts-Strogatz model was designed to produce better models of real-world networks by remedying the lack of clustering in Erdos-Renyi networks. Watts-Strogatz graphs are produced by randomly moving the edges of a regular ring lattice, thus producing “criss-cross” connections across the ring[19]. This does not imply that a real network approximated by the Watts-Strogatz model contains a structure resembling a ring lattice. Moving edges on a ring lattice is simply the algorithm by which these graphs are generated. Similarly, Erdos-Renyi and scale-free graphs (see below) are also generated by specific algorithms, but this does not imply that real networks with similar topologies are built using these algorithms. Watts-Strogatz networks are expected to have a degree distribution that depends on the rewiring probability β, which is related to the number of new connections between nodes that are introduced into the ring lattice. β varies between 0 (a regular ring lattice) and 1 (a lattice where so many connections are changed that it approximates an Erdos-Renyi graph). For moderate to high rewiring probabilities, the degree distribution of the network is a binomial distribution, which in the limit (β = 1) approximates a Poisson distribution. For a low rewiring probability (β near 0), the distribution resembles the P(k) of a regular ring lattice, which is a delta function. In all cases, graphs of P(k) as a function of k for such networks have a distinct peak. Networks generated by the Watts-Strogatz model have local clustering and small-world properties, with high Cg and M values relative to those of Erdos-Renyi networks[19] (Fig 2A).
Scale-free networks have a P(k) distribution that follows a power law, meaning that a few hub nodes make exponentially more connections than other nodes in the network. A graph of P(k) as a function of k for a scale-free network resembles an exponential decay curve (Fig 2B). The best-known model for generating such networks is the Barabási–Albert model, where the network evolves via the addition of new nodes that preferentially form connections to highly-connected nodes[45]. This model can account for the hubs found in real-world networks, but not for local clustering. Hierarchical scale-free networks are composed of modular units of nodes and connections that are combined in an iterative manner, while non-hierarchical scale-free networks have hubs but are not arranged in an organized pattern (Fig 2A). Some have argued that protein networks within cells are likely to be non-hierarchical scale-free networks[46], but others have found that real biological networks cannot be made to fit a power-law distribution[47,48].
The P(k) distribution of the experimental network has a distinct peak, and resembles a Poisson distribution more closely than an exponential decay curve, which indicates that the network is not scale-free (Fig 2B). To quantify this statement, we considered the SSE (standard sum of squares due to error) between a binomial distribution and the observed P(k) distribution of the experimental network. The minimal SSE obtained between a binomial distribution and the degree distribution for the experimental network is 0.012. The mean SSE on a selection of 200,000 simulated random (Erdos-Renyi) networks with the same expected distribution was 0.0099, with a standard deviation of 0.0057. This tells us that the SSE value of our network is less than half a standard deviation from the expected value. By contrast, the SSE obtained by fitting the experimental network to a power-law distribution was over two standard deviations from this expected mean. This provides strong evidence that the fat storage regulation network is much more likely to be approximated by an Erdos-Renyi model or a Wattz-Strogatz model than by a scale-free model (Fig 2B and S4 Table).
The experimental network has a relatively small L value (3.18), which is essentially the same as the expected L values for simulated random (Erdos-Renyi) networks (3.19). This does not distinguish between the Watts-Strogatz and Erdos-Renyi models, since both produce networks with low L values. However, the Cg and M values for the experimental network (0.22 and 0.567, respectively) are higher than the expected values for simulated Erdos-Renyi networks (0.048 and 0.44), but are in agreement with the Watts-Strogatz model for a moderate rewiring probability (β = 0.238). Its high Cg value and low L value characterize the experimental network as small-world. This small-world topology is a distinctive feature of the Watts-Strogatz model.
In a scale-free network, loss of peripheral nodes will have little effect on network parameters, but loss of hub nodes can produce major changes in path length L. An analogy to this is that shutdown or dysfunction of a hub airport (e.g., O’Hare in Chicago) due to bad weather will cause many travelers to have to take additional flights to reach their destinations. However, shutdown of a small regional airport will have no effect on the global pattern of air travel. To evaluate whether our network is vulnerable to deletion of any nodes, we determined the contribution made by each protein to the topological parameters of the entire network. This was done by calculating L, Cg, and M for the network and then recalculating those parameters (L-P, Cg-P, and M-P) when each network protein and its connections were removed. Only minor changes to these values were observed in response to removal of any one protein (<10%; S2 Table). This is of interest, because it argues that the Watts-Strogatz topology of the network may contribute to the robustness of its topological parameters to deletion of any node. Unlike scale-free networks, it lacks essential hub nodes for which elimination of function by mutation or damage would dramatically alter the entire network. We also simultaneously removed all eight of the proteins encoded by essential genes that were added to the network (see above), to ensure that these proteins were not required for network properties. This produced no significant changes in L or M, and only a 10% decrease in Cg.
To evaluate the relationships between the experimentally determined fat storage regulation network and manually curated GO categories, we generated networks of protein-protein connections for the proteins in yeast Biological Process categories, since fat storage regulation is more like a Biological Process than it is like a Molecular Function or Cellular Component. These categories range in size from a few proteins to more than 250. We then calculated edge density (the number of edges/number of possible edges), Cg, and M for these networks and compared these numbers to the values for the fat storage regulation network. S3 Fig shows that the edge density for our network is around the median for the GO categories. Cg is below the median, and M is above the median. These data indicate that the experimental fat storage regulation network has similar properties to those of networks formed from the proteins within GO Biological Process categories. Proteins in a ‘typical’ Biological Process category are extensively interconnected, and exhibit more clustering but less modularity than the proteins in the fat storage regulation network.
Because the fat regulation network was defined by a quantitative genetic screen, it has the useful property that the importance of a given protein to network output can be defined by the severity of the fat storage phenotype (the amount of fat added above wild-type levels) for a deletion mutation in the gene encoding that protein. This is graphically depicted in Figs 1 and 2, where the sizes of the circles in the network diagrams are proportional to the strength of the phenotype produced by loss of the corresponding protein. This allows us to evaluate the relationship between a node’s position in the network and its importance for network function. If such a relationship can be established, it can be used by future investigators to find the essential elements in other biological networks that are found to fit the Watts-Strogatz model, but for which quantitative information on phenotype is not available.
In network analysis, the centrality of a node, C(v), refers to indicators that identify the most important nodes [49]. For example, centrality analysis has been used to find the most influential person in a social network. There are many types of centrality, each of which emphasizes a certain quantifiable attribute of a given node. We examined the five standard centrality types (Degree, Betweenness, Closeness, Eigenvector, and Katz centralities) [49–51] to measure these attributes. For degree centrality, the value is evaluated by counting the number of directly connected nodes. Betweenness centrality is evaluated by counting the number of shortest paths that pass through a given node. Closeness centrality is a sum of the inverses of the shortest path lengths. Eigenvector centrality is an implicitly defined measure evaluated by asserting that the centrality of a single node is proportional to the sum of the centralities of all nodes it is connected to. For Katz centrality, the importance of a node is determined by how many nodes it is path-connected to, with a penalty that increases exponentially with the path length between those nodes. The colored heat-map diagrams in Fig 2C–2G show the values of each type of centrality for each node in a small toy network, with red indicating the highest centrality. The diagrams below superimpose centrality values, indicated by the sizes of the red circles, on the grey circles that indicate phenotypic severity.
We assessed the relationships between the centrality of a given node and the severity of the fat storage defect exhibited when the gene corresponding to that node is removed, using both independence and correlation testing (Fig 2C–2H and S4 Table). Dependence is any statistical relationship between two random variables, while correlation is a special type of dependence that can be used to predict the magnitude of the change that will occur in one variable in response to changes in a linked variable. Two random variables that are independent necessarily have low correlation, but variables with low correlation can be dependent. For example, driving while intoxicated and fatal car accidents are clearly not independent of each other, but the correlation between the two is only moderate, since only 32% of such accidents involved an intoxicated driver and most intoxicated drivers do not have an accident. Dependence is a more general and arguably more useful measure, as it is able to detect nonlinear relationships, whereas correlation assumes a simple linear dependence.
The only centrality measure that passed five separate independence tests (Blomqvist β, Goodman-Kruskal, Hoeffding D, Kendall tau, and Spearman Rank) was Katz centrality (Fig 2E), for which the p-value (probability of independence) was <0.027 for all 5 tests. There are good reasons that Katz centrality gives the best dependence score, which are based on how the centrality measures are defined. First, a good measure of the importance of a protein node should be global, and not just depend on proteins that are directly connected to it. Second, those nodes related by longer paths should have a smaller effect on each other’s function than those related by shorter paths. These attributes are held by closeness, eigenvector and Katz centrality measures. The effect of adding a node on the closeness centrality depends on the reciprocal of path lengths, for eigenvector centrality it is loosely related to degrees of the node, and for Katz centrality, the effect will become exponentially smaller the as the path length increases. In many models, it is reasonable to assume that only a fraction of a signal passes through a node to each of the other nodes to which it connects, and this assumption predicts an exponential drop-off.
Although the statistical analysis shows a clear relationship between the severity of the fat storage phenotype and Katz centrality, the moderate correlation between these parameters (0.24) suggests that this is not necessarily a linear relationship. Fig 2E displays this graphically. It shows that many of the larger grey circles have corresponding large red circles, but there are some large grey circles with very small red circles, indicating low Katz centrality. This shows that the quantitative impact on fat levels caused by removal of a network protein cannot always be predicted from its Katz centrality score in a linear way. In particular, mutations eliminating transcriptional regulators that have low Katz centrality scores often have large impacts on fat levels (e.g., Spt10). This may be due to the fact that the importance of a transcriptional regulator to the network is more likely to be related to the number of gene targets whose expression it controls than to the number of proteomic connections it makes. Most of the transcription factors that regulate fat storage levels bind to the promoter regions of genes that are themselves part of the network, creating potential feedback loops [52] (Fig 2H). Ino2, Srb5, and Ctk1 bind to the promoter regions of up to 40% of network genes. In total, 67% of network genes have network transcription factors other than the global regulators Hhf1 and Spt10 that bind to their promoter regions. The number of network genes that a given network transcriptional regulator binds to is an excellent predictor (correlation = 0.787; probability of independence <0.01; S4 Table) for the severity of the fat storage defect when the gene that encodes it is removed.
Another useful feature of yeast fat regulation is that mutants selected for increased fat content can also be examined for related phenotypes, such as LD morphology and metabolic alterations. We can then ask whether the genes and proteins that share these “sub-phenotypes” also form networks, and examine the relationships between these subnetworks and the larger fat regulation network.
Yeast LDs are composed of triglycerides and sterol esters surrounded by a monolayer of phospholipids. LDs may increase in size and/or number when trigylceride levels increase. A wild-type fixed yeast cell usually has three to eight LDs that are around 0.4 μm in diameter (Fig 3A). We examined all the mutants for LD morphology phenotypes, grouped them into three phenotypic classes, and created subnetworks encompassing all the proteins for each class. Class I contained mutants with small but numerous LDs, class II contained mutants with mixed populations of normal and small LDs, and class III contained mutants that have giant LDs (Fig 3). Each of the three subnetworks was highly interconnected, indicating that members of the same morphological class tend to have mutations affecting proteins that are connected to each other (S5 Table).
Fat levels are influenced by the rate of fat storage utilization, the rate of de novo fatty acid synthesis, and the level of caloric intake. We examined fat storage utilization by subjecting all of the mutants to glucose starvation for three days. ~85% of the mutants showed no reduction in fat storage or a reduced rate of fat storage depletion compared to wild-type when starved, indicating defects in the ability to use stored fat to meet energy demands (Fig 4A and S6 Table).
To be used as an energy source, fats have to be broken down to glycerol and fatty acids, which are used by mitochondria for ATP production [53,54]. We evaluated the overall status of mitochondria in all mutants with reduced fat utilization rate by growing them on glycerol. ~60% of mutants showed either slowed or no growth on glycerol, suggesting defects in mitochondrial function. All of these mutants also failed to grow on palmitic acid and lard. Another 26 mutants did grow on glycerol, but failed to grow on either palmitic acid or lard or both. Mutations that produce a similar growth defect on a given carbon energy source tend to affect proteins that are connected to one another, forming subnetworks (S6 Table and Fig 4B and 4C). These results show that a significant portion of the network is dedicated to maintaining normal mitochondrial function. Indeed, nearly half of the genes we identified were previously implicated in mitochondrial function[53].
In our growth conditions, cells were provided with glucose and a mixture of amino acids that can be used for de novo fatty acid synthesis. To examine if any of the mutants have alterations in this process, we grew them on media containing either 14C-labeled D-glucose or 14C-labeled L-aspartic acid. About 30% of the mutants showed an increase in the conversion of either D-glucose or L-aspartic acid to fat. Mutants that exhibited increased conversion of either nutrient to fat tend to encode proteins that are part of a connected subnetwork (Fig 5 and S7 Table).
To test the hypothesis that genes for which mutants exhibit similar physiological profiles would encode proteins that formed subnetworks, we examined all of the potential subnetworks encoded by subsets of genes with similar detailed sub-phenotypes, in the same way that we had examined the complete network. We found that these subnetworks also exhibited Watts-Strogatz small-world topology, with relatively high clustering coefficients (0.179–0.28) and short path lengths (3.09–3.85). To determine if the subnetworks represented functional subsets of the larger network, we then compared each subnetwork to randomized simulations of 104 networks with identical degree distribution and vertex counts. In all cases, the experimentally observed subnetworks had much higher clustering coefficients than their simulated equivalents (red and green numbers on the right sides of the panels of S5–S7 Tables and Figs 3–5), indicating that the subnetworks were selected by evolution, and do not represent randomly chosen subsets of the complete network. We also observed that in all cases, these subnetworks are not confined to a single module, but span modular boundaries. These results suggest that the larger network that controls fat storage contains within it smaller networks that govern different aspects of physiology related to a cell’s decision whether to store fat or metabolize it for energy.
Having established that the fat regulation network fits the Watts-Strogatz model, we then conducted experiments in which we perturbed the functions of multiple network nodes and measured the effects of these perturbations on fat content. This allowed us to evaluate whether the patterns of communication among distant nodes in the real network are consistent with the Watts-Strogatz topology. We call these patterns “signal propagation”, because, although they are not measured dynamically, they reflect the movement of information through the network. They thus represent signal flow along signal transduction pathways and crosstalk between these pathways.
We conducted “chemogenomic” experiments in which we simultaneously perturbed the functions of multiple network nodes by treating yeast bearing a deletion mutation in each network gene with a set of drugs that block the functions of specific network pathways, followed by measuring the effects of these perturbations on fat content. The approach of adding drugs to single mutants to block multiple nodes was chosen in order to avoid the slow growth or lethality frequently observed for double mutants (see [55]), as well as the necessity to construct all possible double mutants. The chemogenomic approach has often been employed in the yeast system to map synergistic and antagonistic relationships between drug targets and other genes (for recent reviews see [56,57]).
We selected five drugs that blocked signal propagation through specific network pathways. The first drug was U0126, an inhibitor of mammalian MAPKKs [58] that has been shown to block reporter expression controlled by the MAPK mating factor response pathway [59]. The second drug was Rap, an inhibitor of the mammalian and yeast TORC1 complex[27]. The third and fourth drugs were chloroquine (ChQ) and concanamycin A (Conc. A); these are both known to block the acidification of vacuoles by the Vma pump [60–63]. The fifth drug was cerulenin, which inhibits both yeast and mammalian fatty acid synthase [64,65] and thereby eliminates fat synthesis; cerulenin also served as a control to ensure that the effects of the mutations were dependent on network output. The specificity of the drugs we chose was confirmed by the fact that they did not produce increases in fat levels in mutants missing their potential targets (since in those mutants signals from the drug targets are already absent), and by the observation that they produce fat levels similar to mutations that remove these targets (see Supplemental Materials and Methods). Regardless of whether the drugs were completely specific for particular targets, they clearly caused perturbations in network function, as indicated by their ability to alter fat levels in wild type yeast (Fig 6B). The effects of these perturbations on mutant networks, each of which lacks a single node, can therefore be used to analyze internode communication within the network.
We defined three types of signaling relationships between network proteins that are removed by mutation and those whose activities are blocked by a given drug. First, the proteins could be in independent signaling pathways. In such cases, the application of the drug would produce an additive effect, such that the amount of additional fat would equal the sum of the amount of fat added by the mutation and the amount added to wild-type yeast in response to drug treatment. This type of interaction is like the negative interactions between deletion mutations (or between deletion mutations and drugs) that are often observed in double mutant or chemogenomic analyses of yeast growth phenotypes (reviewed by[56]). The strongest form of this type of negative genetic interaction is synthetic lethality, in which double mutant cells (or mutant cells treated with drug) die, while single mutants or drug-treated wild-type cells are viable. Second, the proteins could be part of the same pathway. In such cases, treating the mutant with drug would produce fat levels that are no higher than those in wild-type yeast treated with drug, since pathway signaling had already been eliminated by the mutation. Third, the proteins affected by the drug and by the mutation could be components of two pathways that relay part, but not all, of their signals through each other. We refer to these as synergistic pathways. In such cases, addition of a drug affecting pathway 1 to a mutant affecting pathway 2 would produce an increase in fat levels that is less than the sum of the amount of fat added by the mutant and the amount added by the drug, since the portion of the signal relayed through the action of both pathways was already blocked by the mutation, and the drug could therefore only affect the fraction of the signal that was still active and available for inhibition. In other words, part of the fat level increase observed for the mutation alone would be due to partial blockage of the drug-affected pathway (Fig 6A).
During the course of our analyses, we noticed that none of the drug-mutant combinations produced fat levels that approach the levels of our “fattest” mutant, spt10∆, indicating that our analysis was not limited by cells reaching their maximum fat storage capacity. We also never observed a situation in which the amount of fat in given drug-mutant combination was greater than the sum of the amount added by the drug alone and the mutation alone. These observations indicated that the behavior of the network could be understood using our method.
First, we found that, for each of the four different drugs that increase fat, 12–15% of network proteins behave as if they are in pathways that are independent of the pathway affected by the drug (e.g., they have strong synthetic negative interactions with the drug target). Second, 58% of network proteins had either “same pathway” or “synergistic pathway” relationships with all four drug target pathways (S8 Table and Figs 6 and S4), indicating that there is extensive communication across the network and that network outputs reflect integration among multiple signaling pathways. The fact that proteins involved in such relationships can be in different regions of the network is consistent with the idea that the short path length characteristic of small-world networks facilitates signal propagation between distant parts of the network.
Third, proteins that had a “same pathway”-type relationship with a given drug target sometimes were in different network communities from the drug target. Some of these relationships suggested hitherto unknown interactions between pathways. For example, components of the CCR4-NOT complex involved in mRNA processing had “same pathway” type relationships to the MAPK node affected by U0126, showing that the output of CCR4-NOT was relevant to regulation of fat by the MAPK pathway (Fig 6D). Proteins with “same pathway” type relationships to a given drug could be grouped into subnetworks that had Cg values that were much higher than those of simulated random networks of the same size and connectivity (0.22–0.35), indicating that they have been selected by evolution. These subnetworks did show significant overlap with one another (sharing some of the same proteins), but each one had a unique combination of proteins that was specific for a given drug (S9 Table and Figs 6 and S4).
Fourth, there is a “sub-additive” response (that is, an increase in fat content that is less than the sum of the increase produced by the drug and that produced by the mutation) to loss of function of two nodes for most node pairs. Fifth, eight proteins had a “same-pathway” type relationship with all four drugs, qualifying them as points of convergence for network signals (Fig 6E). This might suggest that these are hub proteins within a scale-free network. However, the hypothetical power-law distribution shown in Fig 2B predicts that in order for our network to be truly scale-free, the number of hubs should not have exceeded two. Furthermore, of these eight proteins, only Ste5 had more than four connections, and hubs should make many more connections than other proteins in the network.
We compared our results to some recent analyses of double mutant and drug-mutant growth and metabolism phenotypes[55,66,67]. These are genome-wide studies, so only a small percentage of mutants (~5%) exhibited a double mutant interaction or an interaction with a given drug. In our study, we investigated a subset of genes that we had defined as critical for fat storage regulation, and examined interactions with drugs that we had selected due to their ability to affect fat content, so we observed that most mutants displayed interactions with all of the drugs. In the genome-wide studies, the frequency of synthetic (negative) interactions was much higher for genes within the same GO Biological Process category than for genes in different categories. For most categories, this frequency was between 10 and 18%[55], which is in the same range as the frequency of genes within the fat regulation network that have strong negative interactions with a given drug target (12–15%; “independent pathways” category). Thus, the experimentally defined fat storage regulation network, which has interconnection density and modularity values that are similar to those of some Biological Process categories (S3 Fig), also behaves somewhat like a Biological Process category with respect to chemogenomic interactions. Finally, about 1/3 of interactions observed in genome-wide studies were positive, but we did not detect any positive interactions. This is due to the fact that we only isolated genes for which mutation increases fat contes, so all of our phenotypes have the same sign. Cellular signaling pathways, however, contain both positive and negative regulators, and mutation of regulators with opposite sign can produce phenotypes of opposite sign.
We also compared the results of Rap treatment of fat storage mutants with a study that examined interactions between Rap and all nonessential genes for a different TORC1-dependent phenotype, expression of a DAL80 reporter that is induced by TORC1 inactivation by Rap or starvation[68]. Of the 63 Rap-specific genes they identified, only 6 (Bck1, Las21, Slt2, Srb5, Swi4, Swi6) were identified in our screen as mutations that increase fat content, suggesting that fat storage regulation and DAL80 induction are not closely related processes. However, two of these common genes (Srb5 and Swi6), increase induction of the DAL80 reporter by Rap[68] and also synergize with Rap to increase fat content (Fig 6 and S8 Table).
By screening the yeast (Saccharomyces cerevisiae) viable deletion collection for mutations affecting fat content, we discovered a densely interconnected network of 94 proteins that regulates fat storage (Fig 1). From a computational analysis of this network, we derived three major conclusions. First, the network is not scale-free, and can be best approximated by the Watts-Strogatz model, which has not been previously applied to biological signaling networks (Fig 2). This model can account for the high degree of clustering observed in the experimental network. Second, the importance of an individual protein to network function is dependent on a particular measure of centrality, Katz centrality, which is influenced by the topological relationships between the test protein and the other proteins within its network community or module (Fig 2). Physiological analysis showed that the fat regulation network is divisible into connected subnetworks which span community boundaries and affect specific aspects of lipid metabolism (Figs 3–5). Finally, by combining drug perturbations with genetics (chemogenomics), we showed that there is extensive cross-talk between the different signaling pathways represented in the network (Fig 6).
The small-world topology endows the network with useful features. Its short path length property allows distant nodes to communicate with each other through a small number of steps. Its topological parameters are robust with respect to removal of any one protein. In scale-free networks, the removal of highly connected hubs will cause substantial changes in network topology and may fragment them into unconnected subnetworks [69–71], while the absence of modular structure in the Erdos-Renyi model makes it inconsistent with the formation of biological networks that must receive different types of inputs and couple them to unique outputs.
Our results also show that the importance of a protein to network function is dependent on its Katz centrality score (Fig 2). The Katz centrality of a node is determined by the number of shortest paths that pass through it to all other nodes in the network, with penalties assigned to connections to distant nodes (nodes that are connected to the node of interest only through a proximal node). The reason that this type of centrality gives the best prediction of a protein’s importance to the function of the fat regulation network may be due to the fact that it combines the local attributes of a node within its community with the position of that community within the network as a whole. As such, it is suited to modular/community based networks, and we predict that future analysis of other such networks will reveal that some of the most important proteins will be identifiable by their Katz centrality scores. In the fat regulation network, the importance of some of the proteins with high Katz centrality can be explained based on their known roles in the module to which they belong. Two examples are Bem1, which is a MAP kinase pathway scaffolding protein [72,73], and Snf1, which is the catalytic subunit of the AMPK complex [29,31].
By analyzing other phenotypes associated with alterations in fat storage, such as LD morphology and utilization of carbon sources, we further demonstrate that the network can be divided into subnetworks that span molecular categories and modules, but affect specific aspects of lipid metabolism (Figs 3–5). All subnetworks contain proteins that have connections that span modular boundaries.
To examine whether the topological features described above are associated with biologically relevant properties, we examined signal propagation within the network by combining genetics and pharmacology (Fig 6). The results indicate that the network has a sub-additive response to perturbation, because the blockage of two pathways or proteins by a drug and a mutation usually produces effects that are smaller than the sum of those caused by the drug and the mutation individually. Of course, the network is not immune to alteration; the removal of a protein from the network does produce an increase in fat levels, since that is how the network was defined. Nevertheless, the attenuated response of the network to blockage of multiple nodes is consistent with the idea that network structure buffers it against external or internal perturbations. The fact that proteins in distant parts of the network often interact with each other, as indicated by the drug-mutant experiments, are in agreement with the short path length and small-world properties of the Watts-Strogatz model.
Unlike Erdos-Renyi networks, Watts-Strogatz networks are modular. In neural networks, modularity is known to facilitate multifunctionality, which can divide large tasks into smaller compartmentalized subtasks that can be executed efficiently [74–76]. Multifunctionality also allows neural networks to integrate different inputs and generate diverse output responses. The yeast fat regulation network has properties that are analogous to multifunctionality in modular neural networks, because it takes in multiple inputs, processes them, and generates multiple outputs in response. The integrated input of the network represents an evaluation of the available nutrients and energy stores. It is provided by the glucose level detection function of Mth1 and the AMP and starvation detection mechanisms of MAP kinase and AMPK pathways. The network then generates anabolic and catabolic outputs that are appropriate to those inputs. Our analysis of mutant phenotypes indicates that these outputs affect many aspects of cell physiology, such as LD morphology, mitochondrial function, and fat store utilization.
Biological signaling networks often contain hub-like elements, and some researchers have proposed that biological networks with hubs are best described by scale-free models[46]. However, real interaction network datasets that have been tested do not fit power-law P(k) distributions, which are diagnostic of scale-free networks [47,48]. The P(k) distributions of proteins in yeast proteomic networks generated by the two-hybrid method have been subjected to mathematical analysis, and a variety of models were tested, including Poisson and binomial distributions (characteristic of Erdos-Renyi and Watts-Strogatz networks) and power-law distributions (characteristic of scale-free networks) were tested. None of the models could be definitively proven or ruled out [18,77].
Almost all of the proteomic connections used to define the fat storage regulation network were identified by affinity purification/coprecipitation methods, which provide a reliable means to identify abundant and stable protein complexes that exist in vivo [7,8,10,12,78,79]. Our computational analysis clearly shows that this network’s topology is best approximated by the Watts-Strogatz small-world model (Fig 2B). This topology has significant regulatory advantages that are correlated with biologically relevant features, and we suggest that mathematical analysis of other proteomic networks defined by a combination of genetics and affinity purification methods may reveal that many signaling networks within cells are of this type.
A collection of haploid MATa nonessential yeast deletion strains was purchased from Thermo Scientific. The strains were transferred into 150 μl YPD media in a 96 well-plate and grown for three days at 30°C. 5 μl of the resulting culture was respectively transferred into 150 μl synthetic complete dextrose media [80] in a 96 well assay plates (black wall with transparent bottom) and grown for two days at 30° C. After the growth period, formaldehyde was added to each well to a final concentration of 4%, followed by incubation at room temperature for 20 minutes. The plates were then spun down at 3000 rpm for two minutes and the supernatant was discarded. The pellets were then resuspended in 150 μl PBS containing 0.125 μg/ml Nile Red and 0.003 μM DAPI, followed by incubation at room temperature for 20 minutes. Fluorescence was measured using a spectrophotometer for both Nile Red (Ex485/Em590) and DAPI (Ex 358/Em440). Results were plotted as a ratio between the Nile Red and DAPI signals. Any positive lines were regrown in eight replicas, of which 4 were stained as above; the others were processed in the same manner, except that Nile Red and DAPI were not added. The latter group was used to measure autofluorescence, the value of which was later subtracted from the final reading. Only lines that remained positive were taken to the second round of selection.
Lines that passed both the 96-well plate assay and the histological examination were subjected to the TLC assay. In this assay cultures were grown in complete synthetic 2% dextrose media for two days at 30° C, and 20 ml normalized culture of OD600 = 1 was obtained. Protein from 1ml of OD600 = 1 culture was extracted and measured and reading was used for additional normalization step, and we rarely observe a situation in which cultures with the same OD600 ending up having a different protein measurements. The twice-normalized culture was then spun down, and the resulting pellets were suspended in 200 μl 2:1 chloroform: methanol mixture, and three glass beads (2 mm) were added to each tube. The samples were subjected to continuous agitation for one hour, and vortexed three times (1 minute each) during that period. The samples were then spun down for 2 min. at maximum speed and the lower phase was isolated and transferred to a fresh tube. The isolated lower phase was spun down again and run on a TLC plate as described [20].
For confocal microscopy, yeast were grown on synthetic complete 2% dextrose media at 30° C for two days. The samples were then spun down and processed for microscopy [81].
For additional information see S1 Text (Supplementary Materials and Methods), which contains methods for starvation studies, growth on different carbon sources, drug treatment, and mathematical analysis.
|
10.1371/journal.pbio.1001589 | NKT Cell-TCR Expression Activates Conventional T Cells in Vivo, but Is Largely Dispensable for Mature NKT Cell Biology | Natural killer T (NKT) cell development depends on recognition of self-glycolipids via their semi-invariant Vα14i-TCR. However, to what extent TCR-mediated signals determine identity and function of mature NKT cells remains incompletely understood. To address this issue, we developed a mouse strain allowing conditional Vα14i-TCR expression from within the endogenous Tcrα locus. We demonstrate that naïve T cells are activated upon replacement of their endogenous TCR repertoire with Vα14i-restricted TCRs, but they do not differentiate into NKT cells. On the other hand, induced TCR ablation on mature NKT cells did not affect their lineage identity, homeostasis, or innate rapid cytokine secretion abilities. We therefore propose that peripheral NKT cells become unresponsive to and thus are independent of their autoreactive TCR.
| Immune system natural killer T (NKT) cells help to protect against certain strains of bacteria and viruses, and suppress the development of autoimmune diseases and cancer. However, NKT cells are also central mediators of allergic responses. The recognition of one's own glycolipid antigens (self-glycolipids) in the thymus via the unique Vα14i T cell receptor, Vα14i-TCR, triggers the NKT cell developmental program, which differs considerably from that of conventional T cells. We generated a mouse model to investigate whether the Vα14i-TCR on mature NKT cells constantly recognizes self-glycolipids and to assess whether this TCR is required for survival and continued NKT cell identity. Switching the peptide-recognizing TCR of a mature conventional T cell to a glycolipid-recognizing Vα14i-TCR led to activation of the T cells, indicating that this TCR is also autoreactive on peripheral T cells or can signal autonomously. But TCR ablation did not affect the half-life, characteristic gene expression or innate functions of mature NKT cells. Therefore, the inherently autoreactive Vα14i-TCR is dispensable for the functions of mature peripheral NKT cells after instructing thymic NKT cell development. Thus the Vα14i-TCR serves a similar function to pattern-recognition receptors, in mediating immune recognition of foreign invasion or diseased cells.
| Natural Killer T (NKT) cells represent a subset of T cells in mice and humans that express NK cell markers and recognize a small class of glycolipid (auto-) antigens [1],[2]. Most mouse NKT cells express an invariant Vα14-Jα18 (Vα14i) TCRα rearrangement (Vα24-Jα18 in humans). In principle, all TCRβ-chains are able to pair with this Vα14i-TCR chain [3]. However, the selection of NKT cells by endogenous glycolipids presented by the monomorphic MHC class I-like CD1d induces a strong bias towards TCRs containing Vβ8, Vβ7, or Vβ2 [1],[3], which is abrogated in the absence of selection [3],[4]. Recently, crystallographic analysis demonstrated a conserved binding mode of the NKT cell TCR to various glycolipids, where only germline-encoded residues were in direct antigen contact, reminiscent of innate pattern-recognition receptors [5]. Moreover, several observations suggest that this receptor is inherently auto-reactive [1],[2] and thereby determines NKT cell identity and influences their function. The expression of several inhibitory NK cell receptors on NKT cells was suggested to control their self-reactivity and avoid autoimmune activation [6],[7].
During development in the thymus, the few T cells expressing a Vα14i-TCR are selected upon recognition of self-lipids on double-positive thymocytes. Although several good candidates have been put forward [8]–[10], the exact nature of the selecting glycolipids remains controversial. Homotypic interactions involving the SLAM family (SLAMf) receptors 1 and 6 are additionally required for NKT cell differentiation [11]. Auto-reactive activation during thymic selection is thought to induce a substantially stronger TCR stimulus in comparison to that during the development of conventional T cells [12],[13]. As a consequence, expression of the transcription factors Egr1 and Egr2 is strongly increased [13], which in turn directly induce PLZF, the key transcription factor controlling NKT cell differentiation, migration, and functions [13].
Interestingly, the homeostatic proliferation of NKT cells after adoptive transfer was similar in CD1d-deficient and wild-type mice, indicating that this process is mostly cytokine-driven and does not depend on continued TCR-mediated self-lipid-recognition [14],[15]. However, as the transferred cells contained CD1d, a role for antigen could not be completely excluded. In addition, tonic antigen-independent TCR signals might contribute to NKT cell maintenance and phenotype. During immune responses, NKT cell activation depends mostly on two parameters: engagement of the TCR and the presence of proinflammatory cytokines released from antigen-presenting cells activated by innate immune pathways such as toll-like receptor (TLR) signals. Lipids derived from different bacteria [16]–[19] were shown to directly activate mouse and human NKT cells in a TLR- and IL-12-independent manner, and NKT cells are required for productive immune responses against these pathogens. NKT cells can also be activated indirectly through cytokines such as IL-12, IL-18, or type I interferons (IFNs) [20]. However, it remains controversial whether, depending on the strength of the cytokine signal, weak responses to self-antigens presented by CD1d are an additional obligate requirement. In one study, CD1d-dependent signals were found to be necessary for full NKT cell activation in response to all tested pathogens [20]. In contrast, others reported that IL-12-dependent NKT cell activation after LPS injection [21] or MCMV infection [22] is independent of either foreign or self-glycolipid antigen presentation by CD1d.
Upon activation, the most distinguishing feature of NKT cells is their ability to rapidly produce and secrete large amounts of cytokines (Th1 and Th2 cytokines, among others). Their fast, effector-like response could be based on steady-state expression of cytokine mRNA in mice [23],[24] that was suggested to be a consequence of tonic self-reactive activation [2]. Recently, it was reported that human NKT cells do not constitutively express cytokine mRNAs. Instead, rapid cytokine-induced innate IFNγ production by NKT cells was suggested to rely on obligate continuous recognition of self-lipids, which retains histone acetylation patterns at the IFNG locus that favor transcription [25]. Another characteristic feature of NKT cells, their surface marker expression reminiscent of memory or recently activated T cells, was also connected to their inherent autoreactivity [2].
To thoroughly address the open questions regarding the nature and importance of TCR signaling for NKT cells, we generated a novel mouse model that allowed us to study the extent of Vα14i-TCR-mediated auto-antigen recognition in the periphery and its relevance for NKT cell identity. Furthermore, we monitored the fate of NKT cells after TCR ablation. Our results prove the inherent self-reactivity of the NKT cell TCR and demonstrate that although essential for positive selection, tonic TCR signaling is not required for NKT cell homeostasis, lineage identity, and rapid cytokine secretion.
In order to produce large numbers of NKT cells in a physiological manner and to manipulate the expression of the semi-invariant Vα14i-TCR in a conditional fashion, we generated Vα14iStopF knock-in mice. To this end we cloned a productive Vα14-Jα18 rearrangement, including the Vα14 leader exon, intron and 1.8 kb of upstream regulatory sequence, and 0.2 kb intronic sequence downstream of Jα18. These elements were inserted by homologous recombination 3′ of Jα1 upstream of the Cα constant region of the Tcrα locus (Figure 1A). Expression of putative upstream rearrangements is aborted by four SV40 polyA sites at the 5′ end of the construct, and expression of Vα14i is rendered conditional through a loxP-flanked STOP cassette. We obtained over 80% (271 of 325) homologous recombinant ES cell clones during gene targeting, indicating an unusually high targeting efficiency of our construct (Figure S1A). The development of conventional T and NKT cells, identified by staining with mouse CD1d-PBS57-tetramers (tetramer+), occurs unperturbed in Vα14iStopF/wt heterozygous mice. In homozygous Vα14iStopF/F mice, T cell development is abolished due to transcriptional termination of TCRα expression before the Cα exons (Figure 1B). We bred Vα14iStopF to CD4-Cre mice, in order to express the inserted Vα14i-chain in double-positive thymocytes, mimicking the physiological timing of TCRα-chain rearrangement and expression [26],[27]. On average 23 times more thymic and 43 times more splenic NKT cells were generated in these, compared to wild-type mice (Figures 1B and 2A–E). Around 9% of the tetramer+ T cells in CD4-Cre Vα14iStopF/wt mice expressed the CD8 co-receptor (over 80% as CD8αβ heterodimer; Figures 1C and S1B,C), which is also expressed by some human NKT cells, but normally not in mice [28]. The proportions of CD4− CD8− double negative (DN) and CD4+ cells were comparable between transgenic (tg) and wild-type NKT cells (Figure 1C). Furthermore, the tgNKT cells were largely comparable to wild-type NKT cells with respect to Vβ-chain bias (Figure 1D) and surface phenotype (Figure 1E). Finally, we found that NKT cells from CD4-Cre Vα14iStopF/wt animals expressed the critical transcription factors promyelocytic leukemia zinc finger (PLZF), GATA binding protein 3 (GATA-3), and T-helper-inducing POZ/Krüppel-like factor (Th-POK) (Figure 1F) [28],[29]. Interestingly, we also detected a substantial proportion of the recently described NKT17 subset in the transgenic animals. These DN NK1.1− NKT cells express the transcription factor ROR-γt and were shown to produce the cytokine IL-17 upon activation (Figure 1F) [29],[30].
Premature TCRα expression leads to aberrant T cell development in transgenic mouse models [26],[27]. To directly compare the consequence of premature to CD4-Cre-mediated timely Vα14i-TCRα-chain expression in our knock-in approach, we bred our mice to a germline Cre-deleter strain (Nestin-Cre) [31]. Compared to CD4-Cre-induced Vα14i-TCRα-chain expression, premature expression in Cre-deleter Vα14iStopF/wt led to significantly reduced numbers of NKT cells in thymus and spleen, especially of CD4+ NKT cells (Figure 2A–C). In addition, we found reduced thymocyte counts and a significant increase of most likely lineage-“confused” DN (CD4− CD8−) tetramer-negative T cells (Figure 2D,E). In fact Cre-deleter Vα14iStopF/wt mice strongly resemble the “first generation” Vα11 promoter-driven (Vα11p) Vα14i transgenic mice in these respects (Table S1) [32]. Moreover, in Cre-deleter Vα14iStopF/wt mice, we observed increased proportions of Vβ9-, Vβ10-, and Vβ14-containing Vα14i-TCRs, which can recognize α-GalCer-loaded tetramers, but most likely not endogenous self-glycolipids [3],[4], pointing to perturbed positive selection (Figure 2F). CD4-Cre Vα14iStopF/wt mice produce more NKT cells than any of the previously reported models, including mice with a Vα14i allele derived from a NKT cell nuclear transplantation experiment [11],[32]–[35]. A comparison of different Vα14i-transgenic models demonstrates that both the correct timing and endogenous control of TCR expression control favor NKT cell development (Table S1). Our analyses therefore showed that physiological timing of Vα14i-TCRα-expression at endogenous levels in CD4-Cre Vα14iStopF/wt mice contributes to the production of large numbers of correctly selected, bona fide NKT cells.
To test the functionality of our transgenic NKT cells, we injected CD4-Cre Vα14iStopF/wt mice with the NKT cell ligand α-Galactosylceramide (α-GalCer) and determined their cytokine production directly ex vivo. The transgenic NKT cells were able to mount a rapid and robust cytokine response. Although a reduced proportion of transgenic NKT cells responded, in absolute cell numbers there was a 6–10-fold increase compared to wild-type NKT cells (Figure 3A). We did not observe significant steady-state cytokine production by transgenic or control NKT cells, and we detected only minor increases in cytokine levels in the serum of some of these mice (Figure S1D). Since cytokine production also varies with NKT cell maturation, we analyzed NKT cell development in CD4-Cre Vα14iStopF/wt mice in more detail. This revealed a strong bias toward immature fractions in the thymus, due to the dramatic increase in NKT cell progenitors. In the periphery, 20% of NKT cells fully matured, as judged by the expression of NK1.1 and other NK cell markers (Figure 3B,C). This view is further supported by the reduced proportion of CD69 and T-bet-expressing NKT cells in CD4-Cre Vα14iStopF/wt compared to wild-type mice (Figure 3D). The expression of both CD69 and T-bet strongly correlated with NK1.1 surface levels (Figure S1E,F). This also explains the higher intracellular PLZF expression in CD4+ and DN NKT cells of CD4-Cre Vα14iStopF/wt animals in comparison to control animals (Figure 1F), as it was shown that PLZF expression is downregulated during NKT cell development [36]. Reduced maturation seems to be a common feature in mice with overabundance of NKT cells (Figure S1G and Table S1) [33]. Indeed, a comparison of different Vα14i-tg mice suggests that independently of the total number of NKT cells generated, the size of the homeostatic niche for mature NKT cells appears to be around two million cells (Table S1).
IL-15 is critical for the final maturation of NKT cells [37] and together with IL-7 required for their peripheral maintenance [14],[38]. NKT cells compete with NK cells for these resources [38]. The halved number of NK cells in CD4-Cre Vα14iStopF/wt mice (Figure 3E) suggests that the availability of these and maybe other cytokines might be insufficient due to the dramatically increased NKT cell numbers. The fact that a similar effect was observed in Vα11p-Vα14itg mice (Figure 3E) underscores this notion. These results let us conclude that while large amounts of NKT cells can be produced in mice, depending on the mode of Vα14i expression, the number of fully mature NKT cells is restricted by homeostatic constraints, some of which are shared with NK cells.
The strong self-lipid-induced TCR stimulus that early NKT cell progenitors receive in the thymus can be visualized through high GFP expression under the control of the Nur77 gene locus, reporting TCR signal strength [12]. However, the subsequent loss of GFP in mature NKT cells suggests that these cells are either not exposed to or not responsive to self-antigens. In order to answer this question and to study NKT cell TCR-autoreactivity in the periphery, we investigated the consequences of Vα14i-TCR signals for conventional naïve T cells. We wondered whether Vα14i-TCR expression on naïve T cells, lacking inhibitory receptors and generally a NKT cell “identity”, would lead to activation upon (self-)lipid recognition and what cellular fate(s) are elicited by such activation.
To this end, we generated mice enabling us to exchange the endogenous TCR-repertoire present on naïve peripheral T cells for a Vα14i-restricted TCR repertoire. The induction of Cre expression in Mx-Cre CαF/Vα14iStopF mice inactivates the CαF allele and simultaneously turns on the Vα14iStopF allele, leading to substitution of endogenous TCRα-chains with the Vα14i TCRα-chain (Figure 4A). As mentioned above, the Vα14i-chain can pair with all TCRβ-chains [3], although only Vβ2-, Vβ7-, and Vβ8-containing Vα14i-TCRs can recognize endogenous lipids such as iGb3 [3],[4]. Since TCRs containing one of these Vβ-chains constitute approximately 30% of the CD4+ and CD8+ peripheral T cell pool (Figure 1D and unpublished data), we predicted that our genetic switch experiment should generate sufficient numbers of T cells able to recognize self-lipids.
In Mx-Cre transgenic mice, Cre expression can be induced through injection of dsRNA, such as poly(I:C) [39]. However, low-level “leaky” recombination occurs also in absence of an inducer [39],[40], leading to increased numbers of tetramer+ T cells in naive Mx-Cre CαF/Vα14iStopF mice (Figure S2A). Therefore, splenocytes were depleted of tetramer+ T cells by magnetic cell separation (MACS, Figure S2A), and 20×106 purified cells were injected intravenously (i.v.) into recipient animals lacking conventional αβ T cells and NKT cells (Cα−/− or Vα14iStopF/F). After cells were allowed to engraft for 2 wk, the TCR switch was induced by poly(I:C) injection. Importantly, except for a short-term activation of the immune system, poly(I:C) injection in Mx-Cre mice per se has no significant long-lasting effect on peripheral conventional T cells [40],[41] or on the number and phenotype of NKT cells (unpublished data). To definitely exclude any effect of poly(I:C) injection on our results, we waited 2–4 mo before analyzing the animals after the induced TCR switch.
We found significant numbers of tetramer+ CD4+ and CD8+ T cells as a result of this switch experiment (Figure 4B–E). “Unloaded” tetramers did not stain these cells, demonstrating that they were not reactive against CD1d itself (Figure S2B). The TCR-switched tetramer+ T cells were predominantly enriched in cells expressing Vβ-chains that are associated with high avidity auto-antigen binding: Vβ2, Vβ8.1/8.2, and Vβ7 (Figure 4D,E) [3],[4],[42]. The exceptions were CD8+ TCR-switched tetramer+ T cells, in which Vβ7-expressing cells were not enriched. The bias toward tetramer+ CD8+ T cells (Figure 4C) is most likely due to more efficient Mx-Cre-mediated recombination in these cells [40].
Animals containing TCR-switched tetramer+ T cells, but not controls, displayed splenomegaly (Figure 5A,B), characterized by increased numbers of macrophages/monocytes, neutrophils, and Ter119+ erythroid progenitor cells, suggesting an inflammatory state (Figure 5C–E). In line with these findings, we could detect elevated serum TNF in more than half of these mice (Figure 5F). Elevated levels of other cytokines, such as IL-2, IL-4, IL-5, IL-6, IL-10, IL-17, and IFN-γ, were not found in the sera of these mice (unpublished data). Interestingly, we found that 6 (highlighted in red throughout the figure) of 17 spleens containing TCR-switched T cells were almost completely devoid of B cells (Figure 5G) as well as dendritic cells (DCs, Figure 5H), which present lipid antigens to NKT cells via CD1d [1]. Furthermore, tetramer- “conventional” T cells were also strongly reduced in these animals (unpublished data). Together, these results suggest that induced expression of the Vα14i-TCR on conventional naïve T cells causes sterile inflammation, possibly due to autoimmune activation.
The appearance of tetramer+ cells displaying a Vβ bias similar to antigen-selected NKT cells, together with signs of inflammation upon TCR switch and the absence of CD1d-expressing B cell and DCs in some cases, suggested auto-antigen-mediated activation of TCR-switched cells. To verify that the newly assembled Vα14i-TCR on conventional T cells is functional, we injected recipients of Mx-Cre CαF/Vα14iStopF and control cells with α-GalCer or PBS 2 mo after switch induction. Ninety minutes after α-GalCer, but not PBS, injection, CD4+ and CD8+ tetramer+ T cells produced IFN-γ and TNF (Figure 6A), demonstrating the functionality of the newly assembled Vα14i-TCR. In comparison to NKT cells from wild-type or CD4-Cre Vα14iStopF/wt animals, a smaller proportion of tetramer+ T cells produced cytokines (Figures 6A and S2C). Tetramer+ TCR-switched T cells could also be activated in vitro through α-GalCer-pulsed A20 cells overexpressing CD1d (unpublished data) [43].
To study the consequences of Vα14i-TCR expression on tetramer+ TCR-switched T cells in more detail, we analyzed their surface phenotype and transcription factor expression. Absence of NK cell markers (Figures 6B and S2D) and PLZF expression (Figure 6C) indicated that the Vα14i-TCR signals are not sufficient to induce NKT cell differentiation of mature conventional T cells. However, the TCR-switched tetramer+ T cells expressed significantly higher levels of Egr2 in comparison to tetramer− T cells in the same animals (Figure 6D), suggesting that the switched cells receive stronger TCR signals [13]. TCR-switched T cells showed further signs of cellular activation, as they expressed elevated levels of CD69 (Figure 6E). Interestingly, these T cells displayed also significantly increased surface levels of PD-1, LAG-3, and less frequently, BTLA and TIM-3, which is typical of exhausted/anergic cells (Figure 6F–H and unpublished data) [44],[45].
To test whether exhaustion/anergy of tetramer+ TCR-switched T cells prevented a more dramatic form of autoimmune inflammation, we injected mice with PD-L1 and PD-L2 blocking or control antibodies twice a week for 4 consecutive weeks, starting 2 d before switch induction. The administration of these blocking antibodies has previously been shown to efficiently prevent anergy induction of conventional T as well as NKT cells, and to partially reverse the exhaustion of CD8+ T cells [44],[45]. However, we did not observe any dramatic differences in spleen weight or cellularity, or signs of increased inflammation, between animals receiving PD-L blocking or control antibodies (unpublished data). In response to PD-1 blockade, other inhibitory receptors such as LAG-3, BTLA, or TIM-3 might control the TCR-switched T cells.
Taken together, our results showed that expression of the Vα14i-TCR on mature conventional T cells is not sufficient to induce a NKT cell differentiation program. Still, it is likely that Vα14i-TCR signals induce auto-antigen-mediated activation, possibly to the point of exhaustion. We therefore present strong evidence that the Vα14i-TCR can constitutively recognize self-lipids in the naïve steady state situation in vivo.
The evidence for autoreactivity of the Vα14i-TCR on mature peripheral T cells raised the old but still not completely resolved question whether and to what extent interactions with self-lipid-presenting APCs are required for NKT cell maintenance, cellular identity, and function. In order to evaluate the importance of constitutive TCR expression and signaling for NKT cells directly in vivo and for long periods of time, we ablated the TCR on mature T cells using poly(I:C) injection of Mx-Cre CαF/F mice [40].
Two weeks after induced Cre-mediated recombination, around 30% of CD4 and 65% of CD8 T cells had lost functional TCR expression in these mice (Figure 7A and [40]). To unambiguously identify TCR-deficient NKT cells, we developed a robust staining strategy based on CD4, NK1.1, CD5, and CD62L expression (Figure S3A). This limited us to CD4+ NKT cells, but our staining identified over 50% of the total NKT cell populations in thymus and spleen (Figure S3B). Around 65% of the thus identified NKT cells had lost TCR surface expression 2 wk after Cre induction (Figure 7A,B).
Due to complete Cre-mediated recombination in lymphoid progenitors, T cell development is blocked at the double positive stage in Mx-Cre CαF/F mice after induction of Cre [40]. This allowed us to study the T cell decay in the absence of cellular efflux from the thymus. In agreement with previous studies [40],[46], we found that loss of the TCR leads to decay of naïve CD4+ CD44low and memory/effector-like CD4+ CD44high T cells with a half-life of 40 d and 297 d, respectively (Figure 7C,D). Interestingly, we observed essentially no decay of receptor-less NKT cells, with a calculated half-life of 322 d (Figure 7E), and could find significant numbers of TCR-deficient NKT cells even 45 wk after TCR deletion (unpublished data). To evaluate the role of TCR signals during in situ homeostatic proliferation, we administered BrdU for 4 wk via the drinking water, starting 2 wk after induced TCR ablation. Naïve CD4+ CD44low as well as CD4+ CD44high memory/effector-like T cells showed significantly decreased BrdU incorporation in TCR-deficient compared to TCR-expressing cells (Figure 7F,G). In contrast, TCR ablation did not affect NKT cell proliferation (Figure 7F,G). Interestingly, the BrdU incorporation was identical in TCR-deficient CD4+ CD44high T and NKT cells, indicating that in the absence of TCR signals the cytokine-driven expansion of CD4+ CD44high memory/effector-like T and NKT cells is similar (Figure 7F,G). Our results therefore indicate that long-term in situ NKT cell homeostasis is completely independent of TCR-induced signals.
In absence of de novo T cell generation, we found elevated Egr2 expression in mature thymic, but not splenic, NKT cells compared to DP thymocytes and CD4+ T cells, respectively (Figure 8A). This indicates that NKT cells receive stronger TCR signals in the thymus, which is supported by the decreased Egr2 expression of mature thymic TCR-deficient NKT cells (Figure 8A). Surprisingly, in mature NKT cells in thymus and spleen, expression of the TCR-signal-induced key transcription factor PLZF is completely unaffected by TCR ablation (Figure 8B).
In order to more generally evaluate to what extent NKT cell TCR-expression is required for the maintenance of characteristic lineage-specific gene expression (resembling recently activated T cells), we extensively analyzed the cell-surface phenotype of NKT cells 6 wk after TCR ablation. Of all the analyzed markers, the only significant changes that we observed on splenic NKT cells upon TCR ablation were downregulation of NK1.1, CD4, CD5, and ICOS (Figures 8C,D and S3C–E). NK1.1 expression was also reduced in thymic TCR-deficient NKT cells, in addition to CXCR6 expression (unpublished data). CD5 and ICOS expression were also reduced in TCR-deficient splenic naïve as well as CD62Llow CD4+ T cells (Figure S3C,D). CD4 was upregulated on TCR-deficient CD4+ naïve, but downregulated on NKT and CD4+ CD44high T cells (Figure S3E). Strikingly, all other cell surface markers characteristic for the NKT cell lineage, among them the transcription factors PLZF, GATA-3, T-bet, and Th-POK, as well as many cell surface markers whose expression is also induced upon TCR engagement, remained largely unaffected by loss of the NKT cell TCR (Figure 8D).
Treatment of mice with LPS, a cell wall component of gram-negative bacteria, leads to release of IFN-γ by NKT cells via stimulation with IL-12 and IL-18 produced by innate immune cells. This does not require acute TCR engagement [21]. However, it has been proposed that the ability of NKT cells to rapidly release IFN-γ in this context critically requires continuous weak TCR activation in the steady state [25]. We therefore analyzed IFN-γ release of TCR+ and TCR- NKT cells after in vivo injection of LPS, α-GalCer, and PBS (Figure 9A,B). As expected, Egr2 expression could only be detected in NKT cells that were activated through their TCR (Figure 9A). Accordingly, 90 min after α-GalCer injection, the majority of TCR+ NKT cells, but virtually none of the TCR- NKT cells or the CD4+ conventional T cells, produced IFN-γ protein (Figure 9B). Interestingly, NKT cell activation through LPS injection in vivo was able to induce similar IFN-γ production by TCR- NKT cells in comparison to their TCR+ counterparts (Figure 9B). Our results thus clearly demonstrate that homeostasis and key features defining the nature of NKT cells, namely the unique activated cell-surface phenotype and the innate capacity for instant production of IFN-γ, do not require continuous auto-antigen recognition in the mouse.
The elucidation of NKT cell function and their intriguing semi-invariant TCR benefited enormously from Vα14i-TCR transgenic mouse models [11],[32],[47],[48]. Over the last years, it became increasingly clear that premature expression of transgenic TCRα chains, including Vα14i [11],[32], leads to various unwanted side-effects such as impaired β-selection and the generation of large numbers of DN T cells both in the periphery and in the thymus [26],[27]. This drawback affects even TCR alleles generated through nuclear transfer of mature NKT cells [33]. For that reason, Baldwin et al. developed a system in which a transgenic CAGGS-promoter-driven TCRα-chain is expressed upon CD4-Cre-mediated excision of a loxP-flanked STOP cassette, mimicking the physiologic expression time point [26]. Likewise, Griewank and colleagues expressed the Vα14i-TCR under direct control of CD4 promoter and enhancer sequences [11]. These are clear improvements, but carry the inbuilt caveats of the respective heterologous expression construct. For example, it has been shown that a large proportion of activated mature T cells loses expression from such transgenic CD4 promoter enhancer constructs [49].
Here, we present a novel approach, in which the expression of the transgenic Vα14i-TCRα-chain, and in the future any other TCRα-chain of interest, can be initiated via CD4-Cre at the DP stage in the thymus, and is under endogenous control of the Tcrα locus throughout the lifespan of the cell. In these mice, large numbers of bona fide CD4+ and DN NKT cells were generated. The reduced proportions of fully mature stage 3 NKT cells (NK1.1+, CD69high, T-bet+), as well as the reduced numbers of NK cells, are most likely a consequence of limiting amounts of common differentiation and maintenance factors, such as IL-15 [14],[37],[50]. In addition, attenuated TCR-signaling due to increased competition for self-antigen/CD1d-complexes might delay the full maturation of NKT cells in the transgenic animals. TCR signals have been proposed to play a role in the initiation of CD69 expression on NKT cells, as well as in the induction of IL-2Rβ, the β-chain of the IL-2 and IL-15 receptors [13].
Moreover, we observed the generation of tetramer+ CD8+ T cells. CD8+ NKT cells are found in the human, but not in wild-type mice. CD8 expression on Vα14i NKT cells does not interfere with negative selection, avidity for antigen presented by CD1d, or NKT cell function [28]. Instead, it was proposed that the absence of CD8+ NKT cells in the mouse is due to the constitutive expression of the transcription factor Th-Pok in all CD4+ as well as DN NKT cells [28]. Th-Pok has been shown to be crucial for the maturation and function of NKT cells, and directly represses CD8 expression [28]. This scenario fits well with the fact that the CD8+ tetramer+ T cells in the CD4-Cre Vα14iStopF/wt (as well as in the Vα11p-Vα14itg animals) did not express Th-Pok. These cells also lack many other characteristic features of NKT cells, including PLZF expression. Therefore, we refer to them as tetramer+ CD8+ T cells.
Given the faithful recapitulation of endogenous TCRα-chain expression timing and strength in our knock-in mice, combined with the extremely high homologous recombination efficiency, we believe that our strategy should prove useful for the generation of further novel TCR-transgenic mouse models. By replacing RAG-mediated Vα14 to Jα18 recombination with Cre-mediated activation of Vα14i expression in CD4-Cre Vα14iStopF/wt mice, we can directly couple conditional gain or loss of gene function with Vα14i-TCR expression in NKT cells. NKT cell-specific gene targeting in mice with physiological NKT cell numbers could be achieved through the generation of mixed bone marrow chimeras with Jα18−/− bone marrow, which cannot give rise to Vα14i-NKT cells.
Our studies were designed to elucidate whether or to what extent the expression of the autoreactive semi-invariant TCR would activate a peripheral mature naïve conventional T cell, convert it into an NKT cell, or induce gene expression typical of NKT cells. We took advantage of the conditional nature of the Vα14i-TCR knock-in transgene for a TCR switch experiment on conventional peripheral T cells. Naïve CD4+ T cells inherit a high plasticity [51]. Depending on TCR signaling strength and cytokine environment, they can differentiate in various subsets in periphery. This differentiation includes the induction of specific transcription factors, namely T-bet (Th1), GATA-3 (Th2), ROR-γt (Th17), and FoxP3 (peripherally derived regulatory T cells). For NKT cells, it is believed that strong TCR signaling, together with homotypic interactions involving the SLAM family (SLAMf) receptors 1 and 6, ultimately leads to PLZF induction during thymic development [11],[13]. DP thymocytes, presenting auto-antigen via CD1d and also expressing SLAMf members, are crucial for thymic NKT cell selection [11]. These SLAMf receptors are expressed on peripheral lymphocytes in comparable levels to double positive thymocytes (www.immgen.org). Therefore, lymphocytes, especially marginal zone B cells, which express CD1d to a similar level as DP thymocytes, should be able to present antigen and SLAMf-mediated co-stimulation, to naïve conventional T cells with a newly expressed Vα14i-TCR on their surface. The elevated levels of the TCR-induced transcription factor Egr2 in switched tetramer+ T cells suggest that they receive an (auto-)antigenic signal. This finding is in principle in agreement with our finding that tetramer+ TCR-switched T cells are enriched in cells that express Vβ2- and Vβ8.1-/8.2-containing Vα14i-TCRs. These TCRs were shown to have the highest avidity for NKT cell antigens [3]. Furthermore, Vβ7-containing Vα14i-TCRs were shown to be favored when endogenous ligand concentration are suboptimal in CD1d+/− mice [42]. In fact, in CD4+ tetramer+ TCR-switched T cells the relative enrichment for Vβ7-expressing cells was slightly higher than for Vβ2- and Vβ8.1-/8.2-expressing cells (unpublished data). However, the interpretation that this advantage is due to antigenic selection is at odds with the fact that Vβ7-expressing cells are not enriched in tetramer+ TCR-switched CD8+ T cells. We currently have no satisfactory explanation for this discrepancy. Both CD4+ and CD8+ Vα14i-TCR-expressing conventional T cells show features of activation and exhaustion/anergy, but do not develop into NKT cells, judged by absent PLZF and NK cell marker expression. This indicates that either mature T cells have lost the ability to enter the NKT cell lineage, the peripheral Vα14i-TCR signal is not strong enough, or as yet unidentified components of the thymic microenvironment are required to induce an NKT cell fate. Indeed, the high Egr2 expression of mature NKT cells that matured in the periphery and migrated back to the thymus (Figure 8A) suggests that stronger self-antigens are presented at this location. Interestingly, unlike TCR-switched tetramer+ T cells, Egr2 expression in mature splenic NKT cells was similar to that of conventional mature CD4+ T cells. Our data therefore suggest that in the periphery, the Vα14i-TCR can recognize self-lipids, but maturing NKT cells undergo a developmental program that prevents an auto-reactive inflammatory response. At this point, we cannot exclude the possibility that the observed cellular activation was antigen-independent. The fact that the internal control cells, the co-transferred tetramer− T cells, show no or significantly less signs of activation strongly argues for an involvement of antigen recognition or tonic signaling by the Vα14i-TCR. It also remains possible that the transient immune activation caused by the poly(I:C) administration contributes to the observed phenotypes. In all likelihood, this contribution is small, as we never observed any significant immune activation, not to mention loss of CD1d-expressing antigen-presenting B cells and dendritic cells, in Mx-Cre CαF/wt control mice that received poly(I:C). Despite these caveats, our results clearly show that under our experimental conditions, Vα14i-TCR expression on conventional naïve T cells leads to their activation and general immune deregulation.
These findings seemed to support notions that NKT cell maintenance [52], their activated surface phenotype, and especially their rapid cytokine expression abilities might depend on constant antigen recognition [25]. However, by ablating the TCR on mature NKT cells in situ, we unequivocally demonstrated that long-term mouse NKT cell homeostasis and gene expression are nearly completely independent of TCR signals. In this regard, they are similar to memory T and B cells, which can maintain their numbers, identity, and functional capabilities in the absence of antigen [53],[54]. Our results are hard to reconcile with a recent report suggesting that NKT cell maintenance requires lipid presentation by B cells [52]. While there might be some differences between mouse and man, a more likely scenario is that the observations of Bosma et al. reflect rather acute local activation than true homeostatic requirements. Most of the known functions of NKT cells critically depend on their ability to rapidly secrete large amounts of many different immune-modulatory cytokines shortly after their activation. Still, it is not fully understood how NKT cell activation is triggered in different disease settings, and especially to what extent signaling in response to TCR-mediated recognition of antigens versus activation by proinflammatory cytokines contributes to this. Various studies reported that CD1d-dependent signals were required for full NKT activation in vitro [19],[20],[55], although most of them contained the caveat of potentially incomplete blockade of CD1d function by blocking antibodies. Our experiments, in line with a recent report [21], show that even in the complete absence of TCR signaling for 4 wk, NKT cells can be robustly activated in vivo to produce IFN-γ upon LPS injection in similar amounts as their TCR+ counterparts. Thus, we demonstrate that in mouse NKT cells continuous steady-state TCR-signaling is not required to maintain the Ifng locus in a transcriptionally active state, as recently proposed for human NKT cells [25]. Therefore, our results clearly demonstrate that cellular identity and critical functional abilities of mature NKT cells, such as steady-state proliferation and innate cytokine secretion ability, although initially instructed by strong TCR signals, do not require further antigen recognition through their TCR.
Collectively, our data strongly support the view that Vα14i-TCR expression on developing NKT cells triggers a program that makes them unresponsive to peripheral self-antigens, which can continuously be recognized by their auto-reactive TCR. NKT cells are extremely potent immune-modulatory cells that upon activation can instantly secrete a large array of cytokines. Although they are selected by high affinity to auto-antigens, similar to regulatory T cells, they are not mainly suppressive cells. Therefore, it seems plausible that NKT cells are rendered “blind” to peripheral auto-antigens, rather than depend on continuous stimulation by self-lipids to maintain their cellular identity and innate functions. By keeping their activated state independent of self-antigen recognition, NKT cells can stay poised to secrete immune-activating cytokines while minimizing the risk of causing damage to self during normal physiology. On the other hand, the presence of the auto-reactive Vα14i-TCR serves to detect pathogenic states when a stronger signal is generated by the enhanced presentation of potentially more potent self-antigens or foreign lipids.
To generate Vα14iStopF mice, B6 ES cells (Artemis) were transfected, cultured, and selected as previously described for Bruce 4 ES cells [56]. Mx-Cre [39], CαF [40], CD4-Cre [57], Nestin-Cre [31], Vα11p-Vα14i-tg [32], and Vα14iStopF mice were kept on a C57BL/6 genetic background. As we did not observe any differences between CD4-Cre and Vα14iStopF/wt mice in NKT cell biology, they were sometimes grouped together as controls. Mice were housed in the specific pathogen-free animal facility of the MPIB. All animal procedures were approved by the Regierung of Oberbayern.
At the age of 6–8 wk (or 2 wk after cell transfer for the TCR switch experiment), animals were given a single i.p. injection (400 µg) of poly(I:C) (Amersham). All mice were analyzed 6–8 wk after injection, unless otherwise indicated.
Single-cell suspensions were prepared and stained with monoclonal antibodies: B220 (clone RA3-6B2), BTLA (8F4), CD11c (N418), CD122 (TM-b1), CD127 (A7R34), CD160 (eBioCNX46-3), CD25 (PC61.5), CD28 (37.51), CD38 (90), CD39 (24DMS1), CD4 (RM4-5), CD44 (IM7), CD45RB (C363.16A), CD5 (53-7.3), CD62L (MEL-14), CD69 (H1.2-F3), CD8α (53-6.7), CD8β (H35-17.2), CD95 (15A7), DX5 (DX5), Egr2 (erongr2), GATA-3 (TWAJ), Gr1 (RB6-8C5), ICOS (7E.17G9), IL-4 (11B11), IL-13 (eBio13A), IL-17A (eBio17B7), IFN-γ (XMG1.2), LAG-3 (eBioC9B7W), LFA-1 (M17/4), Ly49A/D (eBio12A8), Ly49C/I (14B11), Ly49G2 (eBio4D11), Mac1 (M1/70), NKG2A (16A11), NKG2D (CX5), NK1.1 (PK136), PD-1 (J43), ROR-γt (AFKJS-9), T-bet (eBio4B10), TCRβ (H57-597), Ter119 (TER-119), Th-POK (2POK), and TNF (MP6-XT22) (all from eBioscience). SiglecF (E50-2440) was from BD. TCRβ chains were stained with the mouse Vβ TCR screening panel (BD). PLZF antibody and the CXCL16-Fc fusion were generous gifts from Derek Sant'Angelo and Mehrdad Matloubian, respectively. mCD1d-tetramers were provided by the NIH tetramer core facility. For intracellular transcription factor stainings, cells were fixed and permeabilized with the FoxP3 staining kit (eBioscience). For intracellular cytokine stainings, mice were injected i.v. in the tail vein with 40 µg of LPS (Sigma) or 2 µg αGalCer (Funakoshi) in a total volume of 200 µl PBS. Afterwards, cells were treated according to manufacturer's instructions with the Cytofix/Cytoperm kit (BD). For multiplex measurement of cytokines in the serum, we used the mouse Th1/Th2 10plex Cytomix kit according to manufacturer's instructions (eBioscience). Samples were acquired on a FACSCanto2 (BD) machine, and analyzed with FlowJo software (Treestar). The heat map was generated using perseus (part of the MaxQuant software [58]).
Mice were fed with 0.5 mg/ml BrdU (Sigma) in the drinking water for 4 consecutive weeks. Directly afterwards, BrdU incorporation was analyzed with a BrdU Flow Kit (BD).
Serum TNF levels were determined by ELISA as recommended by the manufacturer (BD).
RNA was isolated (QIAGEN RNeasy Micro Kit) and reverse transcribed (Promega) for quantitative real-time polymerase chain reaction (PCR) using probes and primers from the Universal Probe Library (Roche Diagnostics) according to the manufacturer's instructions.
Statistical analysis of the results was performed by one-way ANOVA followed by Tukey's test, or by student t test, in Prism software (GraphPad). The p values are presented in figure legends where a statistically significant difference was found.
|
10.1371/journal.ppat.1006706 | Attaching and effacing (A/E) lesion formation by enteropathogenic E. coli on human intestinal mucosa is dependent on non-LEE effectors | Enteropathogenic E. coli (EPEC) is a human pathogen that causes acute and chronic pediatric diarrhea. The hallmark of EPEC infection is the formation of attaching and effacing (A/E) lesions in the intestinal epithelium. Formation of A/E lesions is mediated by genes located on the pathogenicity island locus of enterocyte effacement (LEE), which encode the adhesin intimin, a type III secretion system (T3SS) and six effectors, including the essential translocated intimin receptor (Tir). Seventeen additional effectors are encoded by genes located outside the LEE, in insertion elements and prophages. Here, using a stepwise approach, we generated an EPEC mutant lacking the entire effector genes (EPEC0) and intermediate mutants. We show that EPEC0 contains a functional T3SS. An EPEC mutant expressing intimin but lacking all the LEE effectors but Tir (EPEC1) was able to trigger robust actin polymerization in HeLa cells and mucin-producing intestinal LS174T cells. However, EPEC1 was unable to form A/E lesions on human intestinal in vitro organ cultures (IVOC). Screening the intermediate mutants for genes involved in A/E lesion formation on IVOC revealed that strains lacking non-LEE effector/s have a marginal ability to form A/E lesions. Furthermore, we found that Efa1/LifA proteins are important for A/E lesion formation efficiency in EPEC strains lacking multiple effectors. Taken together, these results demonstrate the intricate relationships between T3SS effectors and the essential role non-LEE effectors play in A/E lesion formation on mucosal surfaces.
| Enteropathogenic E. coli (EPEC) causes diarrhea and generates the attaching and effacing (A/E) lesion in human gut epithelium. A/E lesion formation requires the locus of enterocyte effacement (LEE) in the bacterial genome, which encodes a protein injection system delivering the translocated intimin receptor (Tir), which binds to intimin on the bacterial surface. Intimin-Tir interaction is sufficient for bacterial attachment to epithelial cells in vitro but additional effectors may be needed for A/E lesion formation in the human gut. By generating deletion mutants lacking combinations or the whole repertoire of protein effectors encoded by EPEC, we show that intimin-Tir interaction is not sufficient and reveal an additive role of non-LEE effectors for A/E lesion formation in human intestinal tissue.
| The gastrointestinal epithelium is an important defense barrier against infections [1]. Enteric pathogens have acquired virulence traits that enable them to colonize and break this barrier, by adhering to the epithelium, delivering toxins and invading intestinal epithelial cells. To this end, several important human and animal pathogens employ type III secretion systems (T3SS) to inject virulence factors into infected eukaryotic cells, where they take control of cell signaling [2].
Enteropathogenic E. coli (EPEC) and enterohemorrahgic E. coli (EHEC) are important human pathogens that colonize the gut mucosa through attaching and effacing (A/E) lesions [3], characterized by intimate bacterial attachment to the apical plasma membrane, localized accumulation of F-actin and effacement of the brush border microvilli [4]. The ability to induce A/E lesions requires the pathogenicity island the locus of enterocyte effacement (LEE) [5, 6]. The LEE encodes gene regulators, the adhesin intimin, chaperones, a filamentous T3SS composed of the translocators proteins (EspA, EspB and EspD), and six effectors (Tir, EspF, Map, EspG, EspH, and EspZ) [7]. In HeLa cells, clustering of intimin with its receptor Tir [8] triggers robust actin polymerization leading to formation of pedestal-like structures [4, 9]. On mucosal surfaces, intimin–Tir interaction is necessary for A/E lesion formation, but it is not currently known if this binding is sufficient [10].
Most LEE effectors, except EspZ, are strong inducers of cytotoxicity, cytoskeleton reorganization, and electrolyte imbalance leading to diarrhea [11, 12]. Map functions as a Cdc42 GEF (Guanine nucleotide exchange factor), leading to filopodia formation on HeLa cells within minutes after infection [13, 14]; EspH inhibits the activity of endogenous DH-PH RhoGEFs causing disassembly of focal adhesions (FAs) and cell detachment [15, 16] and EspG interferes with recycling endosomes [17, 18]. EspZ, which like Tir integrates into the plasma membrane, regulates effector translocation, thus protecting infected cells form cytotoxicity [19].
The prototype EPEC strain E2348/69 also contains 17 effector genes located in integrative elements (IEs) and prophages (PPs). These effectors are frequently found in gene clusters, with some effectors having duplicated gene copies and/or paralogs in different clusters [20]. A large proportion of the non-LEE effectors (e.g. NleB, C, D, E, F and H) inhibits host inflammation ([e.g. nuclear factor kappa B (NF-κB); mitogen-activated protein kinase (MAPK) and the non-canonical inflammasome] [12, 21, 22] and apoptosis (e.g. NleB, D and H) [23]. In particular, NleC is a zinc metalloprotease that degrades the p65 subunit of NF-κB [24].
Deng et al. reported two additional non-LEE effectors, NleJ and LifA/Efa1 [25]. While the function of NleJ is not known, LifA/Efa1, also called lymphostatin [26], has a putative glycosyltransferase activity and an important role in intestinal colonization of cattle by EHEC serogroup O5, O111, and O26 strains [27–29], as efa1 mutations dramatically reduced the number of mucosal associated bacteria and fecal shedding. The reason for this apparent attenuation is not known.
EPEC is a human restricted pathogen; for this reason human intestinal in vitro organ cultures (IVOC) have been used to study early interactions of EPEC with mucosal surfaces [30–33]. Following IVOC infection EPEC triggers A/E lesions that are indistinguishable from those observed in intestinal biopsies of patients with EPEC diarrhea. Using this model it has been shown that while intimin and Tir are essential for colonization, Tir tyrosine phosphorylation is dispensable for A/E lesion formation [10]. However, this infection model has not yet been used to investigate if intimin-Tir interaction is sufficient for A/E lesion formation. The aim of our study is to determine whehter intimin-Tir interaction is sufficient for A/E lesion formation in human IVOC identify further effector(s) required for their formation. To this end, we generated an effector-less mutant of E2348/69 strain and a library of intermediate deletion mutants lacking effectors and preserving the correct assembly and function of the T3SS injectisome. This unveiled that EPEC mutants expressing only Tir were unable to produce A/E lesions on IVOC, while able to produce typical actin pedestals on epithelial cells in vitro. In addition, we found that an EPEC mutant lacking all the non-LEE effector genes shows a marginal ability to form A/E lesions on human intestinal IVOC.
We employed a marker-less deletion/replacement strategy to generate a library of EPEC effector mutants, which allow multiple deletions and/or integrations while leaving neither an antibiotic gene cassette nor short heterologous DNA sequences ("scars") in the chromosome [34]. The mutant alleles were designed to delete the coding sequences of effector genes from the start to the stop codon, or in the case of gene clusters and operons, from the start codon of the first open reading frame (ORF) to the stop codon of the last ORF, maintaining upstream and downstream sequences containing endogenous regulatory elements (e.g. promoters, transcriptional terminators) intact (S1 Fig). We first tested this marker-less deletion strategy by generating an EPEC mutant in escN, encoding the ATPase of the T3SS, whose deletion abrogated secretion of EspA, EspB and EspD in DMEM, but not of EspC autotransporter (S2 Fig). Next, we generated a set of suicide vectors (pGE and pGETS derivatives) for the deletion of all the known effectors in E2348/69 (Table A of S1 Text) [20, 25].
We sequentially deleted LEE effector genes map, espG, espF and espH (Fig 1A) to obtain the mutant strain called EPEC9 (Table 1). The LEE effectors espZ and tir were not deleted at this stage, as EspZ, by regulating effector translocation, protects cells from cytotoxicity [19] and Tir, by mediating intimate attachment, enhances protein translocation [35].
Next, we deleted the genes encoding the non-LEE effectors (Fig 1B). The order of deletion followed was: IE5 (espG2 and espC), IE6 (espL, nleB1, nleE1, efa1/lifA) and IE2 (espL*, nleB*, nleE2, efa1/lifA-like). Although EspC is not a T3SS effector, we deleted espC together with espG2 in the IE5 because EspC has been reported to be internalized into the host cell in a T3SS-dependent manner [36, 37], can interact with translocon proteins [38], and is known to induce severe cytopathic effects and cell death on epithelial cells [39, 40]. The resulting effector mutant strains were called EPEC8, EPEC7 and EPEC6, respectively (Table 1).
We continued by sequential deletion of the effector genes in PPs: PP2 (nleH1, cif*, espJ), PP3 (nleJ), PP4 (nleG, nleB, nleC, nleH*, nleD) and PP6 (nleA/espI, nleH2, nleF, espO*) (Fig 1B), resulting in a strain we named EPEC2, which contains EspZ and Tir as the only effectors. We then proceed with the deletion of espZ. However, we found that deletion of the coding sequence of espZ (ΔespZ-1, S3A Fig), which is the first gene of the LEE2 operon, reduced the secretion of the translocators (EspA, EspB and EspD) of the T3SS (S3B Fig). We speculated that abortive translation initiation induced by the RBS of espZ could potentially affect translation of downstream genes in the LEE2 operon. Then, we generated a second mutant allele of espZ that included deletion of its RBS, called ΔespZ-2 (S3A Fig), which did not affect secretion of the translocators (S3B Fig). Using this mutant allele on EPEC2, we generated the EPEC1 strain that only carries tir. Lastly, we deleted tir in EPEC1 generating the effector-less strain EPEC0. The steps followed to delete T3 effectors in WT EPEC are summarized in Table B of S1 Text.
During generation of each mutant strain, we confirmed the expected deletion by PCR using specific primers (Table C). Confirmation of all deletions in EPEC0 is shown in S4 Fig. In addition, we performed whole-genome sequencing of the parental WT EPEC and EPEC1. Sequencing reads were assembled both using the reference genomes of EPEC E2349/69 and the in silico designed sequence of EPEC1, as well as fully assembled de novo from the sequencing reads. Genome comparison between WT EPEC and EPEC1 showed that the only differences between both strains were the designed deletions (Table D in S1 Text).
WT EPEC and the effector mutants EPEC2, EPEC1 and EPEC0 showed identical growth and viability at 37°C in LB and DMEM media (S5A and S5B Fig, respectively). In addition, microscopic analysis of bacteria from these cultures did not show changes in bacterial size or morphology (S5C Fig). To test the functionality of the T3SS, we analyzed the proteins secreted by WT EPEC, EPECΔescN (negative control), and the effector mutant EPEC strains, after 4 h growth in DMEM at 37°C. We found that the translocators EspA, EspB and EspD, which are secreted by the T3SS [41], accumulated at roughly similar levels in the extracellular media of cultures of WT EPEC and the effector mutant strains (Fig 2A, top panel), but not in the ΔescN negative control. As expected, the autotransporter EspC was absent in the media of strains with deletion of IE5 (from EPEC8 to EPEC0). The expression of the structural proteins EscC, EscJ, EscD, and the translocator protein EspB, was evaluated by Western blotting in protein extracts of whole bacteria from these cultures. All the effectors mutant strains showed equal expression of the analyzed injectisome proteins compared to WT EPEC (Fig 2A, bottom panels). Detection of cytoplasmic E. coli chaperonin GroEL was used as an internal loading control. Altogether, these experiments demonstrate that the effector mutant EPEC strains are not affected in bacterial growth and express normal levels of T3SS injectisomes able to secrete the translocators.
We investigated whether the effector mutants were able to translocate Tir and trigger actin-pedestal formation upon infection of cultured mammalian cells. HeLa cells were infected with WT EPEC and the effector mutants for 1.5 h, fixed and stained for immunofluorescence microscopy. All the effector mutant strains, but EPEC0, triggered actin polymerization upon infection (Fig 2B) and form typical microcolonies, indicating the correct expression of bundle forming pili (BFP) in these strains [42]. Quantification of the number of cells with actin pedestals in these infections shows similar values for WT, EPEC2 and EPEC1, with no actin pedestals found in EPEC0 (Fig 2C).
To confirm that actin accumulation induced by EPEC2 and EPEC1 was due to intimin-Tir interaction we generated eae (encoding intimin) deletion mutants in both strains. Whereas EPEC2Δeae and EPEC1Δeae secreted normal levels of T3SS translocators (S6A Fig), they did not induce actin-pedestals in HeLa cells (S6B Fig). This demonstrates that the actin accumulations observed in HeLa cells infected by EPEC2 and EPEC1 are actual actin-pedestals caused by the specific intimin-mediated clustering of translocated Tir.
We quantified the protein translocation levels of the EPEC effector mutants in HeLa cells using β-lactamase (Bla) fusions [43]. WT EPEC, EPEC2, EPEC1, EPEC0, and EPECΔescN as negative control, were transformed with plasmid pEspF1-20-Bla, which encodes a fusion between Bla and the N-terminal 20 amino acid signal of the EspF to drive its T3SS-dependent translocation [43, 44]. WT EPEC harboring pCX340, encoding Bla without T3 signal, was used as an additional negative control. Whereas no translocation was observed with the control strains, no significant difference in the level of protein translocation was observed from WT and EPEC-2 (Fig 2D). However, EPEC1, which is devoid of espZ, translocated higher levels of EspF1-20-Bla than WT EPEC and EPEC2. Conversely, EPEC0, which lacks intimate adhesion, translocate lower levels of Bla (Fig 2D). These observations are consistent with the reported activities of EspZ and Tir [19, 35].
We investigated the phenotypes following translocation of selected effectors from the different EPEC mutants. In order to maintain physiological expression levels, we integrated a single copy of the effector gene of interest in its native chromosomal location. We followed the marker-less strategy for gene integration, using suicide vectors with the effector gene and flanking homology regions that preserve genome context and native regulatory elements (i.e., promoters, RBS, terminators). We integrated individually the effector genes map and nleC, into the chromosome of EPEC2, EPEC1 and EPEC0 (Table 1).
We tested whether EPEC2, EPEC1 and EPEC0 expressing Map could produce filopodia early during infection. Swiss 3T3 cells were infected for 10 min with EPEC2, EPEC1 and EPEC0 and isogenic strains with an integrated copy of map. Actin staining of infected cells revealed the induction of filopodia by the effector mutant EPEC strains carrying map in the vast majority of infected cells, but not in cells infected by their parental strains (Fig 3). EPEC1+map showed the strongest phenotype of filopodia formation, whereas EPEC0+map induced the weakest phenotype.
We next tested whether EPEC2, EPEC1 and EPEC0 expressing NleC could degrade p65. HeLa cells were infected for 4h with WT EPEC and effector mutant strains, with or without reintegrated nleC. Western blots of the cell lysates with anti-p65 (N-terminal) antibodies revealed that p65 was proteolysed in cells infected with all the EPEC strains expressing nleC (Fig 4). Proteolysis of p65 in cells infected with the effector mutant strains carrying nleC was higher than that induced by WT EPEC, likely caused by the presence of other effectors in EPEC (e.g. NleE, NleB) that inhibit NF-kB activation and the release of free p65 subunit, which is the preferential substrate of NleC [45–47]. Taken together, these results show that the different strains in the effector mutant library contain a functional T3SS, thus allowing us to employ them for infection of human intestinal IVOC.
We aimed to determine if intimin–Tir interaction is sufficient for A/E lesion formation on mucosal surfaces. With this in mind, we infected human duodenal biopsies with EPEC1 or EPEC2. WT EPEC and EPEC0 were used as positive and negative controls, respectively. After 7 h of infection, biopsies were washed, fixed and analyzed by scanning electron microscopy (SEM). Inspection of the mucosal surface revealed A/E lesions in ca. 77% of the biopsies infected with WT EPEC, whereas no A/E lesions were seen in IVOC infected with EPEC2, EPEC1 or EPEC0 (Fig 5 and Table 2).
In order to control that mucus does not affect the interaction of EPEC2 and EPEC1 with the cells and therefore the formation of A/E lesions, mucus-producing human colonic cells LS174T (S7 Fig) were infected with EPEC2, EPEC1 and EPEC0. WT EPEC was used as a positive control. This revealed similar adhesion of bacterial microcolonies and formation of actin pedestals by WT, EPEC2 and EPEC1 (S8 Fig). Hence, despite inducing actin pedestals in cultured intestinal epithelial cells in vitro, EPEC1 and EPEC2 could not induce A/E lesions in human intestinal tissue ex vivo, indicating that intimin-Tir interaction is necessary but not sufficient and that additional effectors are needed.
As the LEE is universally conserved in clinical EPEC isolates we investigated if effectors encoded on the LEE, other than Tir and EspZ, were required for A/E lesions in human intestinal biopsies. For this, we infected IVOC with a derivative strain of EPEC2, called EPEC2LEE, in which the genes espG, map, espF, and espH were reintegrated into their original locus on the LEE. Infection with WT EPEC was used as a control. This revealed that EPEC2LEE was severely impaired in its ability to form A/E lesions (Fig 5 and Table 2), as a single A/E lesion was found in only one of the eleven biopsies infected by EPEC2LEE (Table 2). No adherent bacteria were seen in all the other 10 IVOCs infected with EPEC2LEE (Fig 5). In addition, we tested whether EPEC9, which expresses all non-LEE effectors and misses all LEE effectors except EspZ and Tir, can trigger A/E lesions in IVOC. This revealed that EPEC9 induced A/E lesions in biopsies at a comparable efficiency to WT EPEC (Fig 5 and Table 2). Taken together, these results indicate that non-LEE effectors are important for A/E lesion formation by EPEC in human intestinal tissue.
To investigate the contribution of the non-LEE effectors to A/E lesion formation, we first analyzed the outcome of IVOC infection with EPEC8, EPEC7 and EPEC6. These strains are derivatives of EPEC9 having sequential deletions of effectors genes present in IE5 (espG2, espC), IE6 (espL, nleB, nleE, efa1/lifA) and IE2 (espL*, nleB*, nleE2, efa1/lifA-like) (Table 1; Fig 1). EPEC8 formed A/E lesions at the same frequency as WT and EPEC9 (Tables 2 and 3). Deletion of IE6 alone (EPEC7) caused a reduction in the frequency of A/E lesions (54%), however this did not reach significance. Moreover the proportion of biopsies with A/E lesions decreased significantly 23% following infections with EPEC6 (Table 3). Together this suggest that IE6 and IE2 contribute to A/E lesion formation. Therefore, we investigated the contribution of individual effectors found within IE2. IE2 carries the pseudogenes espL* and nleB*, as well as the effector genes nleE2 and efa1/lifA-like. Therefore, we generated individual deletions of efa1/lifA-like and nleE2 in EPEC7 (Table 1 and S9A Fig). We confirmed by RT-PCR that deletion of nleE2 has no polar effects on the expression of efa1/lifA-like in IE2, and viceversa (S9B Fig). Infection of IVOC revealed that EPEC7ΔnleE2 (carrying a functional copy of efa1/lifA-like) induced A/E lesion in 64% of the infected biopsies, similar to the parental EPEC7 strain (Table 3). In contrast, EPEC7Δefa1/lifA-like triggered A/E lesions in 33% of the infected biopsies (Table 3), similar to EPEC6. These results show that deletion of efa1/lifA-like in EPEC7 has a significant impact on A/E lesion formation.
To further investigate the potential role of LifA-like and LifA in EPEC A/E lesion formation, we generated single (EPECΔlifA-like and EPECΔlifA) and double (EPECΔlifA-like ΔlifA) deletion mutants in WT EPEC (Table 1). These mutants secreted normal levels of EspA, EspB and EspD (S10A Fig) and produced microcolonies and actin pedestals in HeLa cells similar to the WT strain (S10B Fig). Interestingly, infection of human biopsies showed that A/E lesions were formed at efficiencies similar to the WT strain by ΔlifA-like and ΔlifA single and double mutant strains (Table 3 and S11 Fig). Collectively, these results indicate that non-LEE effectors play a major role for A/E lesion formation on human intestinal tissue ex vivo, and suggest an accessory role of LifA-like and LifA proteins in this process, which is masked in the presence of the entire repertoire of T3SS effectors.
EPEC is a major etiological agent of infant diarrhea [48, 49]. With the aim of defining the T3SS effectors implicated in A/E lesion formation, we generated a library of mutants missing part or the whole arsenal of effectors present in the prototypical strain E2348/69. We have demonstrated that the marker-less genome edition strategy generated precise deletions and gene integrations in EPEC. We have built the effector-less EPEC strain (EPEC0) devoid of all known T3SS effectors through 13 deletions, 326 bp was the smallest deletion (espZ) and 18260 bp the largest deletion (IE6). The effector genes were deleted from the start to the stop codon, maintaining their original transcriptional promoters and terminator signals. The only exception was the deletion of espZ, in which deletion of its RBS was necessary to maintain correct expression of the T3SS apparatus.
The LEE effector genes espZ and tir were deleted last as they are important to control protein translocation and bacterial attachment to host cells [19, 35, 50]. Infection of cultured epithelial cells with the WT EPEC and the effector mutant strains demonstrated the functionality of the T3SS. Infection of HeLa and mucin-producing LS174T cells with EPEC2 (espZ and tir) and EPEC1 (tir) showed accumulation of F-actin underneath the attached bacteria, confirming that EPEC only needs the effector Tir to induce the actin-pedestals during infection of epithelial cells in vitro. As expected, no pedestals were seen in cells infected with EPEC0.
Protein translocation assays indicated that all effector mutant strains, including EPEC0, translocate EspF1-20-Bla fusion into HeLa cells, albeit at different efficiency. EPEC1 showed the highest protein translocation level, likely due to absence of EspZ, which limits protein translocation [19]. In contrast, EPEC0 showed the lowest level of protein translocation owing to the absence of intimate adhesion [35, 50]. We have demonstrated that the EPEC effector mutants provide an excellent tool to study the function of individual effectors, under physiological expression levels, in an infection context as chromosomal single-copy integrations reproduce phenotypes previously reported for Map (filopodia formation) and NleC (NF-kB) degradation.
Importantly, using IVOC our study revealed that intimin–Tir interaction is not sufficient for A/E lesion formation and that other effector(s) are needed, as no A/E lesions were observed in biopsies infected with EPEC2 and EPEC1. Furthermore, infections of IVOC with EPEC2LEE (lacking all non-LEE effectors) and EPEC9 (expressing the whole non-LEE repertoire of effectors plus EspZ and Tir), showed that A/E lesion formation requires Tir and EspZ and non-LEE effectors. The difference in effector requirement for intimate adhesion of bacteria in cultured cells (Tir) and A/E lesion in intestinal tissue (Tir+non-LEE) might be due to a more stringent requirement for a productive interaction of bacteria with a complex tissue surface and/or for the hijack of cellular functions in intestinal tissue.
We further characterized the contribution of specific non-LEE effectors to A/E lesion formation by performing IVOC with mutant strains having sequential deletion of non-LEE effector genes. These experiments showed a dramatic reduction of A/E lesion formation when IE6 and, especially, IE2 are deleted (EPEC6). IE6 (espL, nleB1, nleE1 and efa1/lifA) and IE2 (espL*, nleB*, nleE2 and efa/lifA-like) encode a similar set of effectors. We reasoned that either nleE2 or efa1/lifA-like effectors of IE2 should play a role in A/E lesion formation. Albeit NleE2 in IE2 has an internal deletion of 56 residues that could impede its translocation or function [51], we generated EPEC7ΔlifA-like and EPEC7ΔnleE2 mutants. We found that EPEC7ΔlifA-like strain, but not the EPEC7ΔnleE2 strain, exhibited a reduced efficiency of A/E lesion formation to values close to those of EPEC6, suggesting that Efa1/LifA-like protein plays a role in A/E lesion formation ex vivo in the effector mutants.
Efa-1/LifA-like protein was identified in the genome of EPEC E2348/69 as a homolog with aprox. 30% amino acid identity with Lymphostatin (LifA), encoded in IE6 [20]. The lifA gene, for lymphocyte inhibitory factor A, was first described in EPEC as a chromosomally encoded protein of 365 kDa that inhibits proliferation of lymphocytes and the synthesis of proinflammatory cytokines [26, 29]. LifA was later shown to be secreted and translocated into mammalian cells in a T3SS-dependent manner [25]. Efa-1/LifA-like homolog is also secreted in a T3-dependent manner by EPEC, but there is no evidence of its translocation into mammalian cells [25]. LifA homologs are found exclusively in the genomes of A/E pathogens[27–29]. Interestingly, efa1/lifA has been found physically linked to the LEE in some EHEC and EPEC strains [52]. In EHEC, EPEC, and C. rodentium, LifA/Efa-1 has been associated to cell adhesion and tissue colonization [28, 53–55]. In addition, LifA/Efa-1 proteins have been implicated in the induction of intestinal barrier disruption by manipulation of cellular Rho GTPases [56]. While playing a role in A/E lesion formation efficiency, our data show that these proteins are not essential for this process. EPEC6 and EPEC7ΔlifA-like strains still induce A/E lesion formation in 23–33% of infected biopsies (Table 3). Moreover, EPECΔlifA-likeΔlifA behaves as the WT strain forming actin-pedestals on epithelial cells in vitro and A/E lesions on human intestinal tissue ex vivo. Thus, the efa1/lifA-like proteins have an accessory role in A/E lesion formation, which is masked by other T3SS effectors found in the repertoire of the WT strain. These evidences suggest that Ea1/lifA-like protein could act in the subversion of some cellular functions needed for the establishment of the A/E lesion, but its activity can be exerted by alternative EPEC effectors found in the wild type repertoire. This fact also strengthens our experimental approach in which the role of effectors should be better analyzed in the context of infection with strains expressing a reduced and defined set of effectors, since the WT EPEC strain may have multiple effectors with overlapping, synergistic and/or antagonistic effects. The role and molecular mechanism of Efa/LifA homologs in A/E lesion formation requires further investigation.
In summary, our study shows that intimin–Tir is not sufficient for A/E lesion formation in human intestinal mucosal tissue and other effectors are needed. EPEC expressing only the LEE effectors rarely produces A/E lesions, indicating that non-LEE effectors play a major role in this process, having an additive role the effectors encoded in the IE2, IE6 and PPs.
The EPEC strains used in this work are listed in Table 1. E. coli K-12 strains used for cloning are listed in Table A of S1 Text. Bacteria were grown in Luria-Bertani (LB) liquid medium and agar-plates (1.5% w/v) or in Dulbecco's Modified Eagle Medium (DMEM), at 37 oC, unless otherwise indicated. When needed for plasmid or strain selection, antibiotics were added at the following concentrations: ampicillin (Amp) at 150 μg/ml for plasmid selection, and at 75 μg/ml for selection of Amp resistance cassette in the chromosome; chloramphenicol (Cm) 30 μg/ml; kanamycin (Km) 50 μg/ml; tetracycline (Tc) 10 μg/ml; spectinomycin (Sp) 50 μg/ml. See S1 Text for details.
The plasmids employed in this study are listed in Table A of S1 Text. PCRs were performed with the Taq DNA polymerase (Roche, NZyTech) for standard amplifications in screenings or with the proof-reading DNA polymerases Herculase II Fusion (Agilent Technologies) or Vent DNA polymerase (NEB) for cloning purposes. When indicated, DNA was synthesized by GeneArt (Life Technologies). All DNA constructs were confirmed by DNA sequencing (Secugen and Macrogen). Oligonucleotides used in this work were obtained from Sigma and are described in Table C of S1 Text.
A summary of genome modifications and construction of EPEC strains are listed in Table B of S1 Text. Site-specific deletions and insertions in the chromosome of EPEC were originated using a marker-less genome edition strategy with I-SceI [34]. The EPEC strain to be modified was initially transformed with a plasmid pACBSR (CmR) or its SpR-variant pACBSR-Sp [57], both expressing the I-SceI and λ-Red proteins under the control of the PBAD promoter. Subsequently, these bacteria were electroporated with the corresponding pGE-based or pGETS- vector (KmR) and plated on LB-Km-(Cm or Sp). Selection of individual KmR-cointegrants and their resolution upon induction with L-arabinose for isolation of the strains with mutant alleles are described in detail in the S1 Text. All EPEC strains generated were cured of pACBSR before their analysis by serial passages on LB media lacking antibiotics and selection of Cm- or Sp-sensitive colonies. All EPEC strains were confirmed by PCR with specific primers (Tables B and C of S1 Text).
The genomes of EPEC1 and the parental EPEC WT strains were sequenced on an Illumina Miseq platform. Average reads length between 150 and 174 bases and the global coverage was >100X. Genomes were assembled de novo and using reference-guided assemblies with the genome sequence of EPEC O127:H6 strain E2348/69 (nc_011601) and the in silico-designed reference sequence of EPEC1, as described in the S1 Text. The accession number of the genome sequence of EPEC1 effector mutant strain is <PRJEB18717>, and that of the parental EPEC WT strain E2348/69 is <PRJEB18716>. These genome sequences are available at the 'European Nucleotide Archive' http://www.ebi.ac.uk/ena/data/view/<ACCESSION.NUMBERS>.
Sodium Dodecyl Sulfate–Polyacrylamide gel electrophoresis (SDS-PAGE) and Western blot were performed as reported previously [58]. Preparation of EPEC protein extracts are described in the S1 Text. For detection of EPEC proteins by Western blotting, membranes were incubated with primary rabbit antibodies anti-EspB (1:2000), anti-EscC (1:1000), anti-EscJ (1:5000), anti-EscD (1:1000) and anti-Intimin280 (1:5000). Use of polyclonal rabbit sera against EPEC Intimin-280, EscC and EscD were described previously [57, 59]. Rabbit polyclonal serum against EscJ and EspB was a kind gift of Dr. Bertha González-Pedrajo (UNAM, Mexico). Bound rabbit antibodies were detected with secondary Protein A-peroxidase (POD) conjugate (Life Technologies, 1:5000). GroEL was detected with mAb anti-GroEL-POD conjugate (1:5000; Sigma). Membranes were developed by chemiluminiscence using the Clarity Western ECL Substrate kit (Bio-Rad). The membranes were then developed by exposure to X-ray films (Agfa) or with a Fuji LAS 3000 image when the signal was quantified.
Complete description of infection conditions and microscopy is described in the S1 Text. Human HeLa cervix carcinoma cells (ATCC, CCL-2) were grown in DMEM supplemented with 10% heat-inactivated fetal bovine serum (FBS; Sigma) and 2 mM glutamine, at 37 oC with 5% CO2. HeLa cells were washed once with pre-heated serum-free DMEM 2 h before the infection, and infected with EPEC strains for 90 min using a multiplicity of infection (MOI) of 200:1, unless indicated otherwise. Infections were stopped by three washes of sterile PBS (sigma), fixed with 4% (w/v) paraformaldehyde (in PBS, 20 min, RT) and washed again with PBS. Cells were permeabilized by incubation in a solution of 0.1% (v/v) of saponin (Sigma) in PBS for 10 min and washed with PBS. To stain EPEC strains, bacteria were incubated with polyclonal rabbit anti-intimin280 (1:500), or anti-O127 (1:100) for Δeae mutants, and goat anti-rabbit secondary antibodies conjugated to Alexa488 (1:500, Life technologies) in PBS with 10% goat serum; along with Phalloidin TRITC (1:500; Sigma) and 4',6-diamidino-2-phenylindole DAPI (1:1000; Sigma) to label F-actin and DNA and mounted with 4 μl of ProLong Gold anti-fade reagent (Life technologies). To analyze translocation of NleC, HeLa cells were infected 1 h and then washed three times with PBS and incubated for additional 3 h with 200 μg/ml of gentamicin in DMEM. Cells were washed with PBS to remove unbound bacteria and cellular protein extracts were prepared analyzed by Western blot.
To analyze filopodia formation, infection with EPEC strain was done in Swiss 3T3 mouse fibroblasts (ATCC; CCL-92) cells grown in DMEM-high glucose (D5671; Sigma) supplemented with 10% of heat-inactivated fetal calf serum (FCS; Sigma), 2 mM glutamine and 1X of MEM non-essential amino acid solution 100X (Sigma). Swiss 3T3 cells were washed three times with sterile pre-warmed PBS (Sigma) and serum-free DMEM 2 h previous to the infection. Infections were done with 500 μl of EPEC cultures grown in DMEM during 3 h (aprox. MOI 500:1). The plates were centrifuged to synchronize the infection (500xg, 5 min, in a rotor pre-warmed at 37 oC) and the infection was continued for additional 5 min. Infections were stopped by three washes with sterile PBS (Sigma), fixed with 4% (w/v) paraformaldehyde (in PBS, 20 min, RT) and washed with PBS. Fixed monolayers were incubated with polyclonal rabbit anti-O127 (1:100) and secondary donkey anti-rabbit-Alexa488 (Jackson ImmunoResearch, 1:100), together with Oregon-green Phalloidin (1:100, Invitrogen) to label bacteria and actin respectively. Coverslips were washed 3 times with PBS after incubation and mounted with ProLong Gold anti-fade reagent (Life technologies).
LS174T colon adenocarcinoma cells (ECACC 87060401) were grown in DMEM supplemented with 10% heat-inactivated fetal bovine serum (FBS; Sigma), 2 mM glutamine and 1X of non-essential amino acids (Sigma) at 37 oC with 5% CO2. LS174T cells were washed three times with pre-heated PBS (Sigma) 2 h before the infection. The cells were infected with 200 μl (MOI ca. 200:1) for 90 min with EPEC grown in DMEM at 37 oC as described for HeLa cells infections. Following washes with PBS (Sigma), the cells were fixed with 4% (w/v) paraformaldehyde (in PBS, 20 min, RT), washed again with PBS and permeabilized with 0.1% of Triton X-100 (Sigma) in PBS for 10 min. The immunofluorescence staining of intimin, F-actin and cell nuclei was done described previously for infections of HeLa cells. Mucin produced by LS174T cells was stained with an anti-MUC2 rabbit-polyclonal antibody (1:250 Santa Cruz biotechnology) and goat anti-rabbit IgG conjugated to Alexa488 (1:500, Life technologies) as secondary Ab.
β-lactamase (Bla) translocation was quantified as reported previously [43, 44] using LiveBLAzer FRET-B/G Loading Kit with CCF2-AM (ThermoFisher Scientific). Plates were read in a SpectraMax M2 fluorometer (Molecular Devices) with a filter set 450/520 nm. See S1 Text for details.
This study was performed with approval from the University of East Anglia Faculty of Medicine and Health Ethics Committee (ref 2010/11-030). All samples were registered with the Norwich Biorepository (NRES ref 08/h0304/85+5). Biopsy samples from the second part of the duodenum were obtained with informed consent during upper endoscopy of adult patients at the Norfolk and Norwich University Hospital. All samples were anonymized.
Up to 6 biopsy samples per donor were taken from macroscopically normal areas. Samples were cut in half and infected with EPEC wildtype and mutant strains in duplicate. Each bacterial strain was examined in human IVOC on at least three occasions using tissues from different donors. IVOC was performed as described previously [30, 60]. Briefly, biopsies were mounted on foam supports in 12 well plates and incubated with 25 μl standing overnight culture (approximately 107 bacteria). Samples were incubated for 7 h on a rocking platform at 37°C in a 5% CO2 atmosphere. At the end of the experiment, tissues were fixed in 2.5% glutaraldehyde in PBS, dehydrated through a graded acetone series, and dried using tetramethylsilane (Sigma). Samples were blinded and examined in a scanning electron microscope (Jeol JSM-6390). Biopsies showing at least one A/E lesion were scored as positive.
RNA was extracted from the EPEC strains and reversed transcribed by RT-PCR as described previously [61]. The primers used for the RT-PCR of lifA-like, nleE2 and tir are listed in Table C of S1 Text as 108 to 113.
Mean and standard errors of experimental values were calculated with using Prism 5.0 (GraphPad software Inc). Statistical analyses comparing the mean of paired experimental groups were conducted with Student's t-test using Prism 5.0 (GraphPad software Inc). Statistical analyses comparing the number of A/E-positive and negative biopsies after infection with the indicated EPEC strains were conducted with Fisher's exact test to determine two-tailed P values using Prism 5.0 (GraphPad software Inc). Data were considered significantly different when p-values <0.05.
|
10.1371/journal.pgen.1006073 | miR-190 Enhances HIF-Dependent Responses to Hypoxia in Drosophila by Inhibiting the Prolyl-4-hydroxylase Fatiga | Cellular and systemic responses to low oxygen levels are principally mediated by Hypoxia Inducible Factors (HIFs), a family of evolutionary conserved heterodimeric transcription factors, whose alpha- and beta-subunits belong to the bHLH-PAS family. In normoxia, HIFα is hydroxylated by specific prolyl-4-hydroxylases, targeting it for proteasomal degradation, while in hypoxia the activity of these hydroxylases decreases due to low oxygen availability, leading to HIFα accumulation and expression of HIF target genes. To identify microRNAs required for maximal HIF activity, we conducted an overexpression screen in Drosophila melanogaster, evaluating the induction of a HIF transcriptional reporter. miR-190 overexpression enhanced HIF-dependent biological responses, including terminal sprouting of the tracheal system, while in miR-190 loss of function embryos the hypoxic response was impaired. In hypoxic conditions, miR-190 expression was upregulated and required for induction of HIF target genes by directly inhibiting the HIF prolyl-4-hydroxylase Fatiga. Thus, miR-190 is a novel regulator of the hypoxia response that represses the oxygen sensor Fatiga, leading to HIFα stabilization and enhancement of hypoxic responses.
| Sufficient oxygen supply is essential for animal survival. When cells or organisms are exposed to low oxygen levels (hypoxia), a complex molecular response is triggered, enabling adaptation to this stressful condition. A key mediator of this response is HIF, a transcription factor that induces the expression of a set of genes that mediate the adaptive response to hypoxia. The most important regulation of HIF is exerted by a family of prolyl-4-hydroxylases (PHDs), which prevent HIF accumulation under normal oxygen levels and lift this inhibition of HIF only in hypoxia. This pathway is highly conserved among metazoans, including humans and the fruit fly Drosophila melanogaster. microRNAs (miRNAs), which are small (~22 nucleotides long), non-coding RNAs that control gene expression post-transcriptionally, play central roles in stress responses. In the present study, we have performed a screen in Drosophila and identified miRNAs that regulate HIF-dependent adaptations to hypoxia. We found one miRNA, miR-190, that is induced in hypoxia and in turn enhances HIF-dependent biological responses, as well as the expression of HIF-inducible genes. The mechanism of action of miR-190 involves the inhibition of the Drosophila PHD, thereby positively regulating HIF-dependent responses to hypoxia at the molecular and organismal level.
| Cells and organisms exposed to environmental stress mount complex adaptive responses in order to maintain homeostasis. In mammals, hypoxic stress triggers cellular and systemic modifications, such as metabolic switches [1,2], erythropoiesis [3,4], angiogenesis and vasodilation [5,6], resulting in reduced oxygen consumption and increased oxygen transport to hypoxic tissues. Responses to hypoxia are principally mediated by a family of transcription factors named Hypoxia Inducible Factors (HIFs) [7–12], that are heterodimers composed of an oxygen regulated α-subunit (HIFα) and a constitutive β-subunit (HIFβ) [13,14]. HIFα activity is controlled by different mechanisms [15], the most prevalent being oxygen-dependent regulation of protein stability. In normoxia, HIFα is hydroxylated on two specific prolyl residues within the oxygen-dependent degradation (ODD) domain, enabling binding to the von Hippel-Lindau (VHL) tumor suppressor protein, a component of the elongin BC/cullin-2/VHL ubiquitin-protein ligase complex, which targets HIFα for degradation at the 26S proteasome [16–18]. HIFα hydroxylation is catalyzed by specific prolyl-4-hydroxylases (PHD1-PHD3) that are 2-oxoglutarate and Fe(II)-dependent dioxygenases [19,20]. Since PHDs use molecular oxygen as a co-substrate of the reaction, in hypoxia their activity is inhibited. Consequently, in hypoxia HIFα is not hydroxylated, accumulates, translocates to the nucleus, dimerizes with HIFβ and binds to HIF-responsive elements (HREs), thus promoting transcription of target genes [21–23].
We and others have demonstrated that Drosophila melanogaster has a hypoxia-inducible transcriptional response that is homologous to that of mammals [24], with Similar (Sima) [25] and Tango (Tgo) [26] being the homologs of HIFα and HIFβ, respectively [27,28], and Fatiga (Fga) the only Drosophila PHD enzyme [29,30].
We have previously shown that the microRNA (miRNA) machinery is required for full activation of the Sima-dependent transcriptional response to hypoxia, both in cell culture and in vivo [31]. Yet, the individual miRNAs involved in Sima regulation remained unrevealed. Here, we performed an overexpression screen in Drosophila embryos aimed at defining miRNAs that regulate the hypoxic response, and identified specific miRNAs whose overexpression enhances Sima-dependent transcription. One of these miRNAs, miR-190, is induced in hypoxia, is necessary for Sima-dependent gene expression and promotes terminal tracheal cell sprouting. Finally, we found that miR-190 directly targets the HIF prolyl hydroxylase fatiga transcript on its 3’UTR, thereby inhibiting its expression. We propose that miR-190 positively regulates Sima-dependent transcription by inhibiting the oxygen sensor Fatiga, which is the main negative regulator of the hypoxic response.
To identify miRNAs involved in the response to hypoxia in Drosophila, we performed an overexpression screen in stage 14–17 embryos. The rationale was that since suppression of the miRNA machinery inhibits the hypoxic response [31], overexpression of certain specific miRNAs could potentially enhance this response. For the screen, we utilized a collection of 93 fly lines (S1 Table) to overexpress individual miRNAs under control of a breathless-Gal4 (btl-Gal4) driver, and a HIF/Sima-dependent LacZ reporter (HRE-LacZ reporter) as a read out (Fig 1A; [27]). This transgenic reporter was not expressed in normoxic embryos, but induced at 5% O2 in a Sima-dependent manner (Fig 1A; [27]). In fatiga homozygous mutant embryos (fga9), Sima protein accumulates [29], and hence, expression of the reporter was strongly upregulated even in normoxia [29], being this induction suppressed by expression of sima RNAi (Fig 1A and S1 Fig). Given that the biological effect of Drosophila miRNAs is often mild, we sought to conduct the screen under sensitized conditions. To define an appropriate sensitized condition of the hypoxia response system, we used a UAS-fatiga RNAi line (fatigaRNAi; [32]) whose effect is modest. In normoxia, expression of fatigaRNAi had no effect on HRE-LacZ reporter induction, while at mild hypoxia (11% O2), β-galactosidase expression was readily detectable in these embryos (Fig 1A). In embryos bearing only the btl-Gal4 driver, no induction of the reporter was observed under these same conditions (Fig 1A). In strong hypoxia (5% O2), reporter expression was enhanced in the fatigaRNAi line in comparison to wild type controls (Fig 1A). Since mild hypoxia (11% O2) represented a sensitized condition for the hypoxia response machinery, in which potential effects of miRNAs regulating the system might become evident, we performed the screen by exposing the embryos that overexpressed miRNAs at 11% O2 for 4 h; isogenic embryos that did not overexpress any miRNA were used as negative controls (Fig 1B).
The screen was carried out in triplicate; overexpression of most miRNAs had no effect on HRE-LacZ reporter expression (Fig 2A and 2B), but 4 out of the 93 tested miRNAs, namely miR-190 (Fig 2C and 2G), miR-274 (Fig 2D and 2G), miR-280 (Fig 2E and 2G) and miR-985 (Fig 2F and 2G), scored as positives in the screen, inducing expression of the reporter. miR-970, one of the many miRNAs that had no effect on reporter expression, was randomly chosen as a negative miRNA control, and used in the rest of the experiments carried out in this work. We focused our studies on miR-190, whose occurrence in vivo has been experimentally validated by high-throughput sequencing of small RNA libraries generated from different tissues and developmental stages [33,34].
In order to confirm miR-190 participation in the Fatiga/Sima pathway, we began by studying biological responses characteristic of Sima accumulation. We previously reported that fatiga loss-of-function mutations provoke accumulation of high levels of Sima in normoxia, resulting in lethality at the pupal stage [29] and an increased number of terminal ramifications in 3rd instar larval tracheae [35]. Since our results suggested that miR-190 is a positive regulator of Sima (Fig 2), we tested whether overexpression of miR-190 can also induce similar developmental phenotypes, and to what extent they depend on Sima activity.
When overexpressed with an engrailed-Gal4 (en-Gal4) driver, miR-190, but not the control miRNA (miR-970), was associated with lethality at pupal or pharate adult stages (Fig 3A). Knock-down of sima by RNAi completely rescued the lethality caused by miR-190 overexpression, suggesting that lethality was indeed due to Sima accumulation (Fig 3A). In addition, coexpression of Fatiga B, one of the isoforms of the Drosophila HIF prolyl hydroxylase, also rescued the lethal phenotype (Fig 3A), further suggesting that over-accumulation of Sima was the causal factor. When expressed alone, neither sima RNAi nor Fatiga B overexpression had effects on viability (Fig 3A).
Tracheal terminal cells of Drosophila 3rd instar larvae are plastic and ramify in response to hypoxia (Fig 3D and 3E; [36]) in a Sima- and Fatiga-dependent manner [35,37]. As we previously reported, the number of terminal branches with more than 1 μm diameter (“thick terminal branches”, TTBs) of the dorsal branch of the 3rd segment of 3rd instar larvae is a sensitive parameter to quantify terminal tracheal branching after physiological or genetic interventions [35]. To investigate whether miR-190 can also modulate this process, we overexpressed miR-190 under control of the tracheal terminal cell-specific driver dSRF-Gal4. In normoxic larvae overexpressing this miRNA, we observed a significant increase in the number of TTBs (Fig 3C and 3F) in comparison to controls expressing the Gal4 driver only (Fig 3B and 3F), or larvae overexpressing an unrelated miRNA (miR-970) (Fig 3F). To investigate if this increase of ramification depends on Sima, we coexpressed miR-190 along with a UAS-simaRNAi, and observed complete reversion of the phenotype, attaining these larvae a normal number of TTBs (Fig 3F). Expression of the sima RNAi on itself did not induce changes in tracheal terminal sprouting. These results indicate that overexpression of miR-190 can induce Sima-dependent tracheal terminal sprouting, a typical physiological response to hypoxia.
To get additional evidence that miR-190 participates in the HIF pathway, we analyzed genetic interactions between miR-190, fatiga and sima, by assessing induction of the HRE-LacZ reporter as a read out. Overexpression of miR-190 with a btl-Gal4 driver in mild hypoxia enhanced expression of the HRE-LacZ reporter (Fig 2) in comparison with control individuals expressing an unrelated RNAi (Fig 4A); co-expression of this miRNA along with sima RNAi suppressed this enhancement (Fig 4A). Overexpression of miR-190 along with Fatiga B, a highly active isoform of the oxygen sensor Fatiga [30], sharply decreased induction of the reporter (Fig 4A). These results indicate that miR-190 enhances the HIF pathway, antagonizing the activity of the prolyl-4-hydroxylase Fatiga.
To analyze further these genetic interactions, we utilized miR-190 null mutant embryos (miR-190KO, [38]). Unlike the previous experiments in which the HRE-LacZ reporter was utilized in heterozygosis (Figs 1, 2 and 4A), the reporter was used in homozygosis to favor reporter induction in wild type embryos exposed to mild hypoxia (Fig 4B). Noteworthy, this induction was suppressed in miR-190KO mutants (Fig 4B), confirming that miR-190 contributes to Sima-dependent transcription. In fatiga homozygous mutant embryos (fga9), induction of the reporter occurs (Figs 1A and 4B; [29]), and interestingly, this expression was not altered in miR-190KO homozygotes (Fig 4B), indicating that miR-190 operates upstream of the fatiga gene. Taken together, our genetic interactions data are consistent with a model in which miR-190 inhibits Fatiga, resulting in an enhancement of the hypoxic response.
Having analyzed HRE-LacZ reporter induction upon miR-190 loss- and gain-of-function, we studied if miR-190 affects the expression of endogenous Sima target genes. We measured mRNA levels of two well-established Sima targets by real time RT-PCR, namely fatiga B (fgaB) and heat shock factor (hsf) [30,39] in embryos with gain- or loss-of-function of miR-190.
Ubiquitous overexpression of miR-190 with an actin-Gal4 (act-Gal4) driver in embryos maintained in normoxia or exposed to mild hypoxia (11% O2) for 4 h induced upregulation of fgaB and hsf transcripts in comparison to control embryos carrying only the act-Gal4 driver or overexpressing a control miRNA (Fig 5A and 5B). We confirmed these results in Drosophila S2R+ cells, where overexpression of miR-190 also resulted in upregulation of both fgaB and hsf mRNAs, in comparison with cells transfected with the empty vector (S2 Fig).
Next, we examined whether hypoxic induction of the HIF target genes fgaB and hsf is affected in miR-190 knock-out (miR-190KO) homozygous embryos or in embryos heterozygous for miR-190KO and the rhea79a microdeletion that covers the rhea locus [40]; miR-190 is encoded in an intron of the rhea gene [33,34] (S3 Fig). Hypoxic induction of both HIF target genes was severely impaired in miR-190 loss-of-function embryos (Fig 5C and 5D), indicating that miR-190 is necessary for HIF activation.
The results described so far demonstrate that miR-190 positively regulates Sima. Therefore, to investigate the mechanisms of Sima regulation by miR-190, we measured sima mRNA abundance following miR-190 overexpression. Using a ubiquitous act-Gal4 driver, we overexpressed miR-190 in embryos exposed to either normoxia or mild hypoxia (11% O2) for 4 h, and measured sima mRNA levels by quantitative real time RT-PCR. No differences were detectable in sima transcript levels, either in normoxia or in mild hypoxia (S4 Fig), indicating that the miR-190 regulatory mechanism is independent of sima transcription or mRNA stability.
To identify direct targets of miR-190, we searched for target genes related to HIF-dependent response to hypoxia using publicly available database. The miRNA target prediction database miRanda (www.microrna.org) [41–43] predicted two potential miR-190 binding sites within the 3’ UTR of the prolyl-4-hydroxylase fatiga, the main negative regulator of Sima. To determine whether miR-190 can regulate fatiga expression, we used a transgenic reporter construct that directly responds to Fatiga activity. This ubiquitously expressed reporter construct consists of a Green Fluorescent Protein (GFP) fused to the Sima oxygen-dependent degradation (ODD) domain, which is rapidly degraded when Fatiga is active. Conversely, the fusion protein accumulates when Fatiga activity diminishes (Tvisha Misra and Stefan Luschnig, personal communication). We overexpressed miR-190 or a control miRNA with an engrailed-Gal4 driver in the posterior compartment of wing imaginal discs, and analyzed the behavior of the GFP-ODD reporter by confocal microscopy. While expression of the control miRNA (miR-970) did not induce changes in GFP-ODD reporter levels, expression of miR-190 resulted in increased GFP signal in the posterior compartment of the discs (Fig 6A–6H), indicating a stabilization of the GFP-ODD reporter, and suggesting downregulation of Fatiga. A Red Fluorescent Protein (RFP) expressed under the same ubiquitous promoter was used as an expression reference construct. RFP labeling was homogenous throughout the disc and therefore unaffected by expression of the miRNAs (Fig 6C and 6D).
To investigate whether fatiga is a direct target of miR-190, we analyzed the expression of a luciferase reporter in which the firefly luciferase coding sequence is fused to the 3’UTR of fatiga (Fig 6I). The experiment was carried out in S2R+ cells transfected with a plasmid driving the expression of miR-190, in comparison to cells transfected with an empty vector; miR-12 and its specific luciferase reporter [31,44] were utilized as a positive control of the system (S5 Fig). Importantly, transfection of the plasmid expressing miR-190 strongly reduced luciferase activity of the reporter containing the fatiga 3’UTR, as compared to control cells transfected with the empty vector (Fig 6J). To assess binding specificity of miR-190, we mutagenized the strongest miR-190 recognition site within the fatiga 3’UTR (Fig 6I). The reporter bearing the mutant binding site became insensitive to the expression of miR-190 (Fig 6J), confirming specificity of the miRNA. Collectively, these data demonstrate that miR-190 directly targets and downregulates fatiga.
We next investigated if miR-190 expression is regulated by oxygen. RT-qPCR analysis revealed a significant increase of miR-190 expression in wild type embryos exposed to hypoxia (5% O2 for 4 h), in comparison to controls maintained in normoxia (Fig 7A). To determine if hypoxic induction of miR-190 depends on Sima, we analyzed miR-190 levels in embryos exposed to hypoxia and expressing sima RNAi. sima knock-down did not affect miR-190 hypoxic induction, suggesting that upregulation of miR-190 in hypoxia is independent of Sima (Fig 7A).
To investigate if miR-190 upregulation in hypoxia is regulated at a transcriptional level, we evaluated the expression of pre-miR-190. As depicted in Fig 7B, pre-miR-190 expression increased in hypoxia as compared to normoxia, and this induction was again unaffected after sima knock-down. These results suggest that hypoxic upregulation of miR-190 occurs at a transcriptional level, in a Sima-independent manner. Given that miR-190 is encoded in an intron of the rhea gene (S3 Fig), we investigated if rhea transcript levels are also upregulated in hypoxia. Similarly to miR-190, rhea was upregulated in hypoxia in a Sima-independent manner (Fig 7C). As a control of the effect of sima silencing, we assessed in the same embryos the expression of fatiga B, which is a well-known Sima target [30]. As shown in Fig 7D, fatiga B transcript levels were strongly increased in hypoxic wild type embryos, and this induction was reduced upon sima knock-down. Taken together, this set of experiments suggests that miR-190 is transcriptionally induced in hypoxia, as part of the rhea transcript, in a Sima-independent manner (S3 Fig).
Drosophila melanogaster has proved to be a useful model for studying the function of miRNAs as regulators of developmental programs, as well as in the maintenance of cellular homeostasis [38]. In the current work, we have carried out an in vivo screen, aimed at the identification of miRNAs involved in HIF-dependent hypoxic responses in Drosophila. Among 93 miRNAs tested, we identified miR-190, miR-274, miR-280 and miR-985 as positive regulators of Sima-dependent transcription. In mammalian cells, several miRNAs have been reported to participate in the response to hypoxia. Certain miRNAs, such as miR-20b, miR-199a, miR-155, miR-122, miR195, miR-335, miR-33a and miR-18a inhibit HIFα expression directly by binding its 3’UTR [45–52]. Other miRNAs, such as miR-424, miR-184, miR-210, miR-130, miR-494, miR-21 and miR-17 regulate HIFα expression positively through indirect mechanisms [53–60], which involve inhibition of negative regulators of this transcription factor. For example, miR-424 directly targets and reduces the expression of cullin2 (CUL2), a scaffold component of the ubiquitin ligase complex that targets HIFα for degradation in the 26S proteasome [53]. Likewise, miR-184 inhibits another cardinal regulator of HIFα: the factor inhibiting HIF-1 (FIH-1), an asparagine hydroxylase that hydroxylates HIFα, thereby inhibiting its association with the p300 transcriptional coactivator [54,61]. Another interesting example is the direct silencing of the succinate dehydrogenase complex subunit D (SDHD) by miR-210: inhibition of SDHD leads to accumulation of its substrate, succinate, which is in turn a product of HIFα prolyl hydroxylase (PHD) activity with inhibitory effects on the enzyme [62], which finally results in HIFαstabilization [55].
In this study, we have shown that miR-190 directly targets and downregulates the oxygen sensor fatiga, thereby exerting positive regulation on the hypoxia master transcription factor Sima (Fig 8). miR-190 is induced in hypoxia, a condition in which Fatiga activity is also inhibited due to low oxygen availability (Fig 8), providing a mechanism by which miR-190 enhances the strength of the hypoxic response. To our knowledge, this is the first report of a miRNA that directly downregulates an oxygen sensing prolyl-4-hydroxylase.
As documented in the miRNA database miRBase (www.mirbase.org), miR-190 is broadly conserved in evolution, not only within the Drosophilid lineage [34], but also in distant taxa, including mammals. In most mammalian species, two miR-190 family members occur, miR-190a and miR-190b. The miR-190a locus lies in an intron of talin2 (TLN2), which encodes a high molecular weight cytoskeletal protein. Remarkably, Drosophila melanogaster miR-190 is encoded in an intron of the gene rhea (S3 Fig), the homolog of talin2 (TLN2). Intron 53 of human TLN2-001 (which is 12,893 nucleotides long) and intron 14 of rhea-RB (which is 356 nucleotides long) only share sequence similarity within the miR-190 locus [33,34,63–70], reflecting the physiological relevance of this miRNA and perhaps some biological link with Rhea/Talin2. Interestingly, human PHD3 (also known as EGLN3), which is one of the three mammalian homologs of Drosophila Fatiga [19], has a predicted binding site for miR-190a, according to the miRNA target prediction databases TargetScan (www.targetscan.org) [71] and miRDB (mirdb.org) [72], even though with a relatively low score in both cases. Thus, it is possible that miR-190-dependent regulation of HIF-prolyl hydroxylases is conserved in evolution.
We found that Drosophila miR-190 is induced in hypoxia. Interestingly, mammalian miR-190 is upregulated in different types of cancer, including hepatocellular carcinoma [73,74], primary myelofibrosis [75], pancreatic [76], breast [77–79], rectal [80] and papillary thyroid cancer [81]. Hypoxic microenvironment is a common feature of many solid tumors [82,83], and most primary human cancers and their metastases exhibit increased levels of HIFα [84]. In addition to intratumoral hypoxia, genetic and epigenetic alterations can also stimulate HIF activity within tumors [82,84,85]. HIF promotes angiogenesis [82,86], metabolic switches [87], metastasis [88] and chemo/radio-resistance of cancer cells [89,90], and high levels of HIF are associated with poor patient prognosis and increased mortality [84,91]. On the other hand, many different miRNAs have been shown to play pivotal roles in cancer development, functioning as oncogenes or tumor suppressors [92–94]. Given that miR-190 is upregulated in diverse cancer types, our findings open the possibility that miR-190 contributes to HIFα stabilization in cancer cells, thereby enhancing tumor progression.
In line with this possibility, miR-190 directly inhibits the PH domain leucine-rich repeat protein phosphatase (PHLPP), a tumor suppressor protein that inactivates the kinase AKT through Ser437 dephosphorylation [95–97]. In human bronchial epithelial cells, trivalent arsenic (A3+) induces the expression of miR-190, which binds the 3’UTR of PHLPP transcript, decreasing PHLPP protein levels [95,96]. As a consequence, AKT phosphorylation and activation increase, finally resulting in vascular endothelial growth factor (VEGF) expression [95], which is induced following AKT activation [98]. Another bona fide miR-190 target is IGF-1, which is significantly reduced in serum of patients with hepatocellular carcinoma. Accordingly, miR-190b is upregulated in tumor tissues, contributing to insulin resistance through downregulation of IGF-1, which is associated with poor prognosis [73]. Thus, miR-190 favors carcinogenesis through distinct pathways.
Importantly, strengthening the notion of a possible involvement of miR-190 in mammalian responses to low oxygen, miR-190 is induced by hypoxia in a rat model of hypoxic pulmonary artery hypertension (PAH) [99–102]. miR-190 directly targets and represses the expression of Kcnq5, a member of the voltage-gated K+ channel family, resulting in augmented vasoconstriction of the pulmonary artery, a hallmark of hypoxic PAH [101].
In summary, the results reported here increase our understanding of the network controlling HIF-dependent responses to hypoxia, and open the possibility of analyzing the regulation exerted by additional miRNAs which may be part of this complex network.
The UAS-miRNA fly collection utilized in this study was previously described [103]. The following fly stocks were from the Bloomington Drosophila Stock Center (Indiana University, Bloomington, IN, USA): w1118, breathless-Gal4, engrailed-Gal4, dSRF-Gal4, actin-Gal4, UAS-GFP, UAS-white RNAi and miR-190KO. The following stocks were from the Vienna Drosophila RNAi Center: UAS-fatiga RNAi (VDRC 103382), UAS-sima RNAi (VDRC 106504). The HRE-LacZ reporter [27], UAS-Fatiga B [30] and fga9/TM3 [29] lines were generated in our laboratory and previously described. The rhea79a [40] mutant was kindly provided by Nicholas Brown.
Hypoxia was applied in a Forma Scientific 3131 incubator, by regulating the proportions of oxygen and nitrogen. To obtain synchronized individuals, embryos were collected on egg-laying agar plates for 4 h, and then incubated at 18°C or 25°C in normoxia until the desired stage. When necessary, embryos or first-instar larvae were sorted to obtain the desired genotypes using a fluorescent Olympus stereomicroscope MVX10.
For X-gal stainings, embryos were dechorionated in bleach for 1 min, incubated with heptane for 5 min, fixed with glutaraldehyde 0.5% in PBS for 20 min and then washed three times for 5 min in PT 0.3% (PBS containing 0.3% Triton-X 100). Samples were incubated 1 h with the staining solution (5 mM K4Fe(CN)6, 5 mM K3Fe(CN)6, 0.2% X-gal) at 37°C. After three washes with PT 0.3%, samples were analyzed using an Olympus stereomicroscope MVX10; and photographed after mounting in glycerol 80% with an Olympus BX60 microscope equipped with an Olympus DP71 digital camera.
The screen was performed using the Fly Condo (Flystuff, San Diego, CA, USA), which contains 24 independent chambers, allowing for high-throughput collection of Drosophila embryos. In each chamber, we placed adult males bearing the btl-Gal4 driver and the HRE-LacZ reporter, together with females of one miRNA line or a wild type line (w1118) as a negative control. Embryos from the offspring were collected in the 24-well stainless steels mesh plate provided with the condo and subjected to hypoxia (11% O2), for 4 h. Next, we evaluated the expression of the HRE-LacZ reporter performing X-gal stainings of the embryos within the mesh plate.
First-instar larvae were placed in fresh vials, at a density of 20 individuals per vial. When they reached the third-instar wandering stage, larvae were anesthetized with ether and ramifications of the terminal cell of the trachea in the dorsal branch of the third segment were counted and photographed using bright-field microscopy.
Embryos were incubated under hypoxia (11%, 8% or 5% O2) or normoxia for 4 h, at 25°C. Next, total RNA was isolated using Trizol reagent (Invitrogen, Carlsbad, CA, USA) from embryos of stages 14–17. Genomic DNA was removed from RNA samples using Ambion’s DNase (Ambion, Austin, TX, USA). Samples (1 μg) were reverse-transcribed with the M-MLV Reverse Transcriptase (Invitrogen, Carlsbad, CA, USA), following the manufacturer´s instructions, using oligo-dT as a primer. The concentration and integrity of RNA and cDNA were determined using Nanodrop ND-1000 spectrophotometry and gel electrophoresis.
The resulting cDNA was used for quantitative real time PCR, using a MX3005P instrument (Stratagene, La Jolla, CA, USA). The real time PCR reaction contained: 1 μL Sybr Green 1/1000, 0.3 μL ROX reference dye 1/10 (Invitrogen, Carlsbad, CA, USA), 0.2 μL of Taq DNA Polymerase Recombinant (Invitrogen, Carlsbad, CA, USA), 2.5 μL Buffer 10X, 1 μL MgCl2 50 mM, 0.5 μL dNTP mixture 10 mM (Invitrogen, Carlsbad, CA, USA), 1 μL of sense primer 10 μM, 1 μL of anti-sense primer 10 μM, 5 μL of template cDNA 1/30, 4.2 μL glycerol 30% and 8.3 μL of H2O. The thermal cycling conditions were the following: 95°C for 10 min, followed by 40 cycles at 95°C for 30 s, 60°C for 1 min and 72°C for 1 min, finishing with a cycle for the melting curve of 95°C for 1 min, 60°C for 30 s and 95°C for 30 s. Relative mRNA expression was normalized using rpl29, rpl32 or GAPDH as internal controls.
For quantification of miR-190 levels, the NCode VILO miRNA cDNA Synthesis Kit (Invitrogen, Carlsbad, CA, USA) was used, following manufacturer´s instructions. The 2S rRNA was used for normalization in quantitative real time PCR determinations of miR-190.
Larvae were dissected in PBS, fixed in 4% formaldehyde (Sigma, St. Louis, MO, USA) for 40 min at room temperature and then washed three times for 10 minutes in PT 0.3% (PBS containing 0.3% Triton-X 100). Samples were blocked with bovine serum albumin 5% in PT 0.3% (PBT) for 2 h and then incubated with the primary antibody in PBT overnight at 4°C. After washing three times for 15 min with PT 0.3%, tissues were incubated for 2 h at room temperature with the secondary antibody in normal goat serum 5% diluted in PT 0.3%. Next, samples were washed, imaginal discs were separated and mounted in glycerol 80%. Images were analyzed and captured using a Carl Zeiss LSM510 Meta Confocal Microscope.
We used mouse anti-Engrailed (1:100; Developmental Studies Hybridoma Bank, Iowa, IA, USA) primary antibody and donkey anti-mouse Cy5 (Jackson, ImmunoResearch Laboratories Inc., West Grove, PA, USA) secondary antibody. Fluorescence of GFP and RFP was analyzed without antibody staining.
The copper-inducible pMT/V5-His plasmid (Invitrogen, Carlsbad, CA, USA) was utilized as a backbone vector for generating reporter constructs. For generation of pMT-Luciferase renilla (pMT-Renilla), the coding sequence of renilla luciferase was subcloned from a pRL-SV40 vector (Promega, Madison, WI, USA) into HindIII/XbaI sites of pMT/V5-His. For the pMT-Luciferase firefly reporter construct (pMT-Firefly), the firefly luciferase coding sequence was subcloned from pGL3 vector (Promega, Madison, WI, USA) into EcoRI/XbaI sites of pMT/V5-His. fatiga 3’UTR sequence was generated by PCR from cDNA prepared from Drosophila yellow white embryos and cloned into XbaI/ApaI restriction sites of the pMT-Firefly plasmid. The primers utilized were:
Forward (Fw): 5’-GCTCTAGACCCAAGCCGACAGCGCAGCT-3’;
Reverse (Rv): 5’-GCCATTGGGCCCCATCAGCTCAGGCTTTTGTTTA-3’.
Point mutations in miR-190 binding site at the fatiga 3’UTR were introduced by nested PCR with the following primers:
Fw: 5’-CTGTAAATCATGAAGTATGTATATTTATGCCCTCGCTACATATTGTATG-3’;
Rv: 5’-CATACAATATGTAGCGAGGGCATAAATATACATACTTCATGATTTACAG-3’.
The Luc-CG10011 3’UTR reporter and pAc-miR-12 were a gift from E. Izaurralde [44]. The pAc-miR-190 overexpression plasmid was kindly provided by M. Milán [104]. The pAc-5.1/V5-His (Invitrogen, Carlsbad, CA, USA) was used as a negative control.
Semi-adherent Schneider (S2R+) Drosophila cells were maintained in Schneider Drosophila medium (Sigma, St. Louis, MO, USA) supplemented with Penicillin (50 U/ml, Invitrogen), Streptomycin (50 μg/ml, Invitrogen) and 10% fetal bovine serum (Invitrogen, Carlsbad, CA, USA) at 25°C in 25 cm2 T-flasks.
Cells were seeded in 24-well plates at a 35000 cells per well density and 0.3 μg of total DNA was transfected employing the Effectene transfection reagent (Qiagen, Valencia, CA, USA). All pMT-Firefly-3’UTR constructs were co-transfected at a 1:1 proportion with pMT-Renilla to normalize transfection efficiency. Expression of luciferase from pMT vectors was induced 24 h after transfection by addition of 0.7 mM CuSO4 for 7 h. Firefly and renilla luciferase activities were measured by the Dual-Glo Luciferase Assay System (Promega, Madison, WI, USA), following the instructions of the manufacturer, in a Veritas Microplate Luminometer (Turner BioSystems).
Data are expressed as mean ± standard deviation (SD). Infostat Statistical Software was used for statistical analysis. Comparisons were performed using one- or two-way analysis of variance (ANOVA) followed by Fisher's protected least significant difference (LSD) as post hoc test, or unpaired two-tailed Student's t-test. Data were tested for normality (Shapiro–Wilks test) and variance homogeneity (Levene test) to use parametric statistical analysis. If data did not fulfill these statistical criteria, Welch's correction or the Kruskal-Wallis one-way ANOVA non-parametric test were used. A p<0.05 was considered statistically significant.
|
10.1371/journal.pcbi.1004660 | Spontaneous Decoding of the Timing and Content of Human Object Perception from Cortical Surface Recordings Reveals Complementary Information in the Event-Related Potential and Broadband Spectral Change | The link between object perception and neural activity in visual cortical areas is a problem of fundamental importance in neuroscience. Here we show that electrical potentials from the ventral temporal cortical surface in humans contain sufficient information for spontaneous and near-instantaneous identification of a subject’s perceptual state. Electrocorticographic (ECoG) arrays were placed on the subtemporal cortical surface of seven epilepsy patients. Grayscale images of faces and houses were displayed rapidly in random sequence. We developed a template projection approach to decode the continuous ECoG data stream spontaneously, predicting the occurrence, timing and type of visual stimulus. In this setting, we evaluated the independent and joint use of two well-studied features of brain signals, broadband changes in the frequency power spectrum of the potential and deflections in the raw potential trace (event-related potential; ERP). Our ability to predict both the timing of stimulus onset and the type of image was best when we used a combination of both the broadband response and ERP, suggesting that they capture different and complementary aspects of the subject’s perceptual state. Specifically, we were able to predict the timing and type of 96% of all stimuli, with less than 5% false positive rate and a ~20ms error in timing.
| We describe a new technique for decoding perception from electrical potentials measured from the human brain surface. All previous attempts have focused on the identification of classes of stimuli or behavior where the timing of experimental parameters is known or pre- designated. However, real world experience is spontaneous, and to this end we describe an experiment predicting the occurrence, timing, and types of visual stimuli perceived by human subjects from the continuous brain signal. In this experiment, human patients with electrodes implanted on the underside of the temporal lobe were shown pictures of faces and houses in rapid sequence. We developed a novel template-projection method for analyzing the electrical potentials, where, for the first time, broadband spectral changes and raw potential changes could be contrasted as well as combined. Our analyses revealed that they carry different physiological information, and, when used together, allow for unprecedented accuracy and precision in decoding human perception.
| How does a two-dimensional pattern of pixels measured by our retina get transformed into the percept of a friend’s face or a famous landmark? It is known that the ventral temporal cortex represents different classes of complex visual stimuli within distinct regions. For example, category-selective areas have been established unambiguously at scale of several millimeters using functional imaging and macroscale field potentials [1–4]. Similar results have also been demonstrated at the single-unit level in epileptic human patients [5] and non-human primates [6]. More recently, high frequency electrocorticographic (ECoG) changes from these same ventral temporal regions have been shown to increase while viewing images of faces, places, and other objects [7–10]. However, rather than reflecting a discrete range of frequencies, >40Hz ECoG changes have been shown to instead be a reflection of broadband fluctuations across the entire frequency domain [11,12], and these broadband changes show robust increases across ventral temporal cortex during object perception [13].
Object-category specific responses in inferotemporal cortex were initially identified using event-related potentials (ERPs) in ECoG [14,15] or functional magnetic resonance imaging (fMRI) [1–4] although little spatial overlap was found between the ERP and the fMRI response [16]. In contrast, increases in high-frequency broadband power in cortical surface potentials recorded using ECoG matched well with the category-specific fMRI responses in the inferior temporal cortex [17,18]. The ERP and broadband signals show distinct, and partially overlapping, responses to faces [13,19] (Fig 1), but it is unclear whether the information content is itself distinct between the two. While both the ERP and the raw ECoG potential have previously been used to classify object categories [20–22], these studies required knowledge about the time of stimulus onset, rather than determining them spontaneously. Furthermore, the ability of the algorithms to establish object category from neural data was well below that of human performance (both in terms of accuracy and temporal fidelity).
A significant methodological obstacle to this type of macroscale physiology has been the difficulty interpreting heterogeneity in response morphologies. As illustrated in Fig 1, face-selective ERPs may have wide structural variation, with “peaks” and “troughs” that are very different in shape, latency, and duration, even when measured from brain sites separated by only 1cm. It remains unclear what the ERP shape actually corresponds to. Furthermore, methodology has not previously been developed to naively place morphologically-diverse ERPs in a common feature space. In contrast, broadband spectral changes in the ECoG signal have been shown to correlate with neuronal firing rate [23,24], although it has been unclear how ERPs relate to this, or what the best way to attempt such a comparison is [19]. Our work begins by describing a template-projection technique, where templates of averaged raw potentials (ERPs) and broadband changes (ERBB) from a training period are projected into the data from a testing period. This places ERP and ERBB features from different brain sites into a common feature space, where they can be directly compared with one another, and used together for decoding brain function.
To date, decoding of perceptual content has relied upon designated information about external stimuli, where the frequency of occurance and precise timing are known to the decoder. We propose that in addition to identifying the perceptual content (e.g. image type), decoding of the brain state should evolve to spontaneously identify whether a perceptual event has happened from the datastream, and, if so, predict the timing as accurately as possible. We denote this practice as “spontaneous decoding”.
Here we show that the ECoG signal contains sufficient information to allow near-instantaneous identification of object categories with an accuracy comparable to that of human behavioral performance. Our experiments measured ECoG recordings from several inferior temporal visual areas simultaneously while subjects viewed randomly interleaved images of faces or houses. We achieved the best results by combining broadband changes with raw potential changes (rather than with either independently), using a template projection approach. This shows that the two types of signals capture complementary aspects of the physiology reflecting a human subject’s perceptual state. With this combination, we were able to predict 96% of all stimuli correctly as face, house, or neither, with only ~20 ms error in timing.
All patients participated in a purely voluntary manner, after providing informed written consent, under experimental protocols approved by the Institutional Review Board of the University of Washington (#12193). All patient data was anonymized according to IRB protocol, in accordance with HIPAA mandate. A portion of this data appears in a different context in [13]. All data and analyses are publically available at http://purl.stanford.edu/xd109qh3109.
All 7 subjects in the study were epileptic patients (S1 Table) at Harborview Hospital in Seattle, WA. Subdural grids and strips of platinum electrodes (Ad-Tech, Racine, WI) were clinically placed over frontal, parietal, temporal, and occipital cortex for extended clinical monitoring and localization of seizure foci. Lateral frontoparietal electrode grids were discarded from analysis, and only strip electrodes were further considered. The electrodes had 4 mm diameter (2.3 mm exposed), 1 cm inter-electrode distance, and were embedded in silastic. Electrode locations relative to gyral surface anatomy were determined by projection of the post-implant CT to the pre-operative axial T1 using normalized mutual information in SPM, and the CTMR package, with Freesurfer-extracted cortical surface mesh reconstructions [25–28]. When the MRI or CT was of insufficient quality, hybrid techniques were used [29].
Experiments were performed at the bedside, using Synamps2 amplifiers (Neuroscan, El Paso, TX) in parallel with clinical recording. Stimuli were presented with a monitor at the bedside using the general-purpose BCI2000 stimulus and acquisition program [30]. The electrocorticographic potentials were measured with respect to a scalp reference and ground, subjected to an instrument-imposed bandpass filter from 0.15 to 200 Hz, and sampled at 1000 Hz.
To reduce common artifacts, the potential, Vn0(t), measured at time t in each electrode n, was re-referenced with respect to the common average of all N electrodes, Vn(t)=Vn0(t)−1N∑i=1NVi0(t). Electrodes with significant artifact or epileptiform activity were rejected prior to common averaging. There was no rejection of epochs of time within the data. Ambient line noise was rejected by notch filtering between 58–62 Hz using a 3rd-order Butterworth filter [31].
Subjects performed a basic face and house stimulus discrimination task. They were presented with grayscale pictures of faces and houses (luminance- and contrast-matched) that were displayed in random order for 400ms each, with 400ms blank screen inter-stimulus interval (ISI) between the pictures. The 10cm-wide pictures were displayed at ~1m from the patients while they were seated at the bedside (Fig 1). There were 3 experimental runs with each patient, with 50 house pictures and 50 face pictures in each run (for a total of 300 stimuli). In order to maintain fixation on the stimuli, patients were asked to verbally report a simple target (an upside-down house), which appeared once during each run (1/100 stimuli). There were few errors in reporting the upside-down target house in any run (approximately 2–3 across all 21 experimental runs).
Following previously described methodology [11,32,33], we perform discrete estimates of the windowed power spectrum, as well as a time-frequency approximation of the dynamic power spectrum from Vn(t). We then perform a “decoupling process” to identify underlying motifs in power-spectral change, isolating the timecourse of broadband spectral change, Bn(t). This process was originally described and illustrated in full detail for ECoG recordings from motor cortex [11], and later illustrated specifically for this face-house context [12]. Broadband changes have been shown to robustly characterize the magnitude and latency of cortical dynamics from ventral temporal cortex, in single trials, during this face and house viewing experiment [13]. Generically, the broadband power time course is meant to function as a time-varying estimate of changes in a multiplicative factor of the population firing rate [11,24].
ECoG signals were measured in seven subjects from electrodes implanted on the inferior temporal visual areas for the purpose of epilepsy monitoring. Subjects were presented with pictures of faces and houses (similar to those in Fig 1). We attempted to spontaneously identify the timing of face and house visual stimuli.
To test whether the ERBB and ERP provide useful information to decode whether, when and which class of stimulus was presented, we extracted the ERBB and ERP for all electrodes. Some electrodes show a classical face-specific N200 response [13–15]. Other electrodes show face-specific ERPs with very different shapes (Fig 1).
We first investigated whether the stimulus class could be decoded in single trials when the onset of the stimulus is given. We calculated template ERBB and ERP responses from training data, which consisted of 2/3 of the recorded data (two experimental runs). The test data (for the classifier) consisted of the other 1/3 (the remaining experimental run; i.e., 3-fold cross validation, or “leave-one-run-out” cross-validation). Fig 2 shows examples of the template ERBB responses for a face- and a house-specific site. Even in a two-dimensional subspace of the full feature space, a simple line serves as a good classification boundary between the two classes of stimuli (Fig 2G).
Using either the ERP or the ERBB feature, stimuli could be robustly and reliably categorized in all cases. The average prediction accuracy using the ERBB alone was 97% across all 7 subjects, while using the ERP alone, it was 90% (Fig 3). Using a combination of the two features, 97% of stimuli could accurately be classified as face or house. It is important to note that, in subjects 1 and 3, the addition of the ERP feature actually resulted in a decrease in classification accuracy, when compared with the ERBB alone, and subject 7 shows no change. This is because of what is known as the “bias-variance tradeoff” [42,43]. For a finite number of datapoints in a training set, the inclusion of features with higher amounts of noise (ERP features in this case) can hurt overall classification. The classifier overfits noise in the mediocre features (ERP), at the expense of a tight fit to high-yield (lower noise) features (e.g. ERBB), while simultaneously expanding the size of the feature space.
Figs 2 and 3 demonstrate that our analyses can accurately determine the stimulus class when given the timing of stimulus presentation. However, this type of decoding has been employed before in other experimental settings, albeit with less accuracy [20–22]. The more interesting technical question is: Can one spontaneously determine both the class and the onset of the stimuli from a continuous stream of ECoG signal features?
Our approach to the continuous decoding problem is illustrated in Figs 4–6, where template responses from a training period were applied to a period of testing data. The result of plotting the projection timeseries trajectory in a 2-dimensional subspace, ΓB(t), can be seen alongside training points Γn,SB(q) in Fig 4. Even in this 2-dimensional subspace projection, the furthest excursions of ΓB(t) into the face or house training clouds, Γn,SB(q), correlate with the times of predicted stimulus onset. Fig 5 shows an example similar to that in Fig 4, but for the ERP feature. Fig 6 shows an example of the synthesis between ERP and ERBB features when used together.
A combination between ERP and ERBB projections can be used to predict the onset timing and class of stimuli more accurately than either independently. The spontaneous classification of onset time and stimulus class was robust: 92% of stimuli were captured using the ERBB, 92% when using the ERP, and 96% of all stimuli were captured spontaneously when using a combination of both ERP and ERBB (Fig 7, top row). Furthermore, timing of stimulus onset could be predicted with approximately 20ms error when the ERP or a combination between the ERP and ERBB was used (Fig 7, middle row). The portion of incorrect predictions (e.g. false positive rate) was smallest (4%) when we used a combination of both the ERP and ERBB (i.e., predicted stimuli occured at >160ms from stimulus onset, or as the wrong class; Fig 7, bottom row).
In order to evaluate whether using both features together (ERP and ERBB) was significantly better than either independently, the labels of mean values (ERP vs ERBB vs ERP+ERBB) were randomly reshuffled (within each subject) 104 times to obtain a surrogate distribution of difference in means averaged across all subjects. The 96% of events captured using both features was significantly greater than the 92% when using either independently (p = 0.0015). The timing error for correct predictions was not significantly different for both features (19ms) vs ERP (20ms, p = 0.17), but was significantly better than ERBB alone (32ms, p<0.0001). The false positive rate using both features (0.04) was significantly less than either independently (ERP 0.11; ERBB 0.09; p = 0.0012). The fact that the overall best prediction performance was reached by a combination of ERBB and ERP suggests that these two cortical features convey complementary information about a subject’s perceptual state.
Note that our 20ms estimate of the temporal fidelity of the signals may actually be an underestimate. There may be instrumentation temporal error introduced due to frame-jitter on the refresh rate of the amplifiers, sample jitter during alignment to the stimulus, and/or the granularity of sample block size of the signals imported to BCI2000 program [30]. Furthermore, there are known variations in the magnitude and timing broadband responses that are related to semantic properties (such as novelty [13]), that are disregarded in this manuscript.
We designate this technique as “Spontaneous decoding” of the ECoG datastream. Our technique processes the data, without foreknowledge of the frequency of external stimuli, nor their timing, nor their content. It then produces predictions about the occurrence, timing, and content of external stimuli, based upon a simple set of internal rules. “Spontaneous” is defined as [44]: “performed or occurring as a result of a sudden inner impulse or inclination and without premeditation or external stimulus”, and so we feel that this term is the most specific way to describe this analysis approach. While “endogenous” or “intrinsic” decoding might also have been chosen, since these are used to describe internal brain states (which is an aspect of we are actually decoding), we chose not to use them–we feel that these terms convey assumptions about the role of the temporal lobe which have yet to be proven.
In human experience, environmental stimuli arrive continuously, producing a sequentially evolving perceptual state. It has remained unknown whether the brain surface electrical potential has sufficient spatiotemporal fidelity to capture this dynamically changing perceptual state. Our results demonstrate that a sparse sample of the cortical surface potential contains sufficient information to reliably predict whether and when a particular stimulus occurred, with approximately the fidelity of conscious perception. It has also remained unknown whether the mesoscale neurophysiologies of event-related potentials and broadband spectral changes reflect the same information.
Previous studies aimed at decoding perception have all pre-defined the onset time of each stimulus [6,20–22,45,46]. In the first-stage of our analysis, we performed this type of classification using pre-defined onset time, with 97% accuracy (Figs 2 and 3). Similar prior studies attained representative peak accuracies of 72% with MEG/fMRI [22], 89% with EEG [20], and 94% with MEG [21]. However, real-world perception rarely occurs at pre-defined times, and approaches to decoding perceptual experience should be extracted spontaneously from continuous cortical recordings.
We have developed a technique to do just this, applying a novel template projection technique that enabled us to capture some aspects of the neural response that have previously been difficult or impossible to capture. First, the ERP in face-selective regions in the fusiform gyrus is classically associated with a negative peak at ~200ms (“N200”). Our data show that the actual shape of face-selective fusiform ERPs can vary widely, even at fusiform sites 1 cm from one another (Fig 1). The template projection technique captures these diverse response patterns, allowing them to be exploited for classification of perceptual state. Second, broadband responses show variability in the pattern of response in every individual trial. The template projection method relies on a superposition of the single trial characteristic shape and a probability density function for modeling different shapes, offering a robust prediction of perceptual state in spite of the variability across single trials. Examination of the features separately demonstrated that broadband changes are more robust and reliable reflections of perceptual content than raw-voltage changes, but that projection of ERP into raw voltage changes produces sharper temporal precision. Together, these two measures complement one another, providing independent information that results in more accurate and temporally precise prediction of the perceptual state than either measure on its’ own.
Our decoding fidelity approaches that of conscious thought, correctly capturing 96% of all stimuli from a sparsely-sampled stream of cortical potentials. The missed 4% (as well as the <5% false positive rate) approaches what might be expected for rates of inattention by hospital patients viewing multiple stimuli each second (note that random guessing at the maximum rate in this spontaneous decoding would result in a 20% chance of each guess being correct, and 50% of stimuli deemed “captured”, with an 80% false positive rate). A temporal precision of ~20ms (Fig 7, middle row) is of the same order as the post-retinal temporal granularity of the visual system [47]. These ECoG measurements show that some electrodes in early visual cortex already display some stimulus-selective responses (e.g., Fig 5, purple site). This agrees with observations that fast eye movements can be made just based upon the Fourier spectrum of the images of different classes [48], and that people saccade towards a scene containing an animal or face within 140 ms [49,50]. By demonstrating that object categories can be decoded from a continuous image stream with accuracies matching expected human behavior (e.g. attentional lapses expected at a rate of approximately 5% in a task of this type [51]), our study lays the groundwork for capturing human perceptual states in a natural environment.
Although we applied this template-projection technique to prediction, the framework may be used in a wide variety of experimental settings. ERPs from adjacent cortical regions may be highly polymorphic, complicating cross-comparison of timing and magnitude effects. In this projection space, however, trial-to-trial ERP variations from different cortical sites may be compared directly, opening a new family of analyses that might be applied to cognitive settings, where image content and context are experimentally manipulated on single trials. Similarly, one might optimize the differential strengths of each feature, such as broadband for magnitude of response and ERP for timing of response, comparing these to stimulus properties to learn about subtleties of functional specialization in each brain region.
An important feature of this template projection approach is that it provides a robust, continuous, measure that is a summary statistic for how well the brain state at every point in time reflects the expected response (e.g. as if a perceptual event or action had occurred at that time–note that the shape of the expected physiological response, however idiosyncratic, is built into the method). This could be extremely useful in settings where the cortical dynamics and latency differ by region, yet a global behavior of a distributed visual [52], auditory [53,54], motor [55], or other network must be characterized. In emerging work, this technique is implemented in a different way, to generate broadband ECoG templates from a low-noise localizer task, and apply them to a visual discrimination task at the perceptual threshold, quantifying single trial variation in cortical physiology (neuronal response magnitude and timing) [56].
Our results beg the question: What is the underlying neural basis for the increased accuracies obtained by combining ERPs with broadband activity? A direct connection between neuronal population firing rate and broadband ECoG spectral change has been established with experimental and modeling work [11,23,24]. Each clinical ECoG electrode averages over approximately 5x106 neurons in the cortex beneath. Careful experimentation has shown that the broadband changes follow a power law in the power spectral density, implying that it reflects asynchronous spiking elements in the underlying population of neurons. The broadband measure may be loosely thought of as a real-time summation of this population’s firing raster (i.e., intrinsically averaged across the population of neurons). Increases in spike transmission within neurons in the population add in quadrature (e.g., proportional to the square root of the number of spikes), appearing as a “speeding up” of a random walk in the electrical potential time series, are difficult to see when looking at the raw potential, but apparent as broadband, P∼1fχ power-law, changes when inspecting in the frequency domain [24]. Recent work has shown that, in these data, the broadband timing is subtle enough to capture variational effects at the order of ~50ms due to context-dependent processing, such as sequential novelty [13].
Synchronized inputs, by contrast, add linearly and can be easily seen in the raw tracing of the electrical potential. Even if the synchronization is relatively weak, averaging across the neural population augments the synchronized portion, while the other aspects, such as broadband spectral change, are relatively diminished. Event-locked inputs, from subcortical nuclei, or other cortical regions, can trigger a synchronized physiologic cascade, evident at the macroscale as an ERP. It remains unclear whether the polyphasic ERP is a result of interplay between coordinated excitatory pyramidal neuron depolarization followed by interneuronal lateral inhibition, or whether it results from synaptic integration followed by characteristic depolarization and repolarization of cortical laminar dipoles [57]. The polymorphic nature of different ERPs from adjacent cortical regions may (perhaps) then relate to different pyramidal neuron morphologies, different milieus of neuronal subtypes, or different laminar organization; our projection technique unfolds these polymorphic ERPs into a common space for comparison. In this light, the improved decoding accuracy may be the result of multi-location timing information conveyed by ERP during the initial feed-forward wave of neural activation [58], complemented by the broadband response reflecting subsequent local recurrent and longer-range cortico-cortical processing of the visual stimulus.
|
10.1371/journal.pntd.0003169 | Efficacy and Safety of Amphotericin B Emulsion versus Liposomal Formulation in Indian Patients with Visceral Leishmaniasis: A Randomized, Open-Label Study | India is home to 60% of the total global visceral leishmaniasis (VL) population. Use of long-term oral (e.g. miltefosine) and parenteral drugs, considered the mainstay for treatment of VL, is now faced with increased resistance, decreased efficacy, low compliance and safety issues. The authors evaluated the efficacy and safety of an alternate treatment option, i.e. single infusion of preformed amphotericin B (AmB) lipid emulsion (ABLE) in comparison with that of liposomal formulation (LAmB).
In this multicentric, open-label study, 500 patients with VL were randomly assigned in a 3∶1 ratio to receive 15 mg/kg single infusion of either ABLE (N = 376) or LAmB (N = 124). Initial cure (Day 30/45), clinical improvement (Day 30) and long term definitive cure (Day 180) were assessed.
A total of 326 (86.7%) patients in the ABLE group and 122 (98.4%) patients in the LAmB group completed the study. Initial cure was achieved by 95.9% of patients in the ABLE group compared to 100% in the LAmB group (p = 0.028; 95% CI: −0.0663, −0.0150). Clinical improvement was comparable between treatments (ABLE: 98.9% vs. LAmB: 98.4%). Definitive cure was achieved in 85.9% with ABLE compared to 98.4% with LAmB. Infusion-related pyrexia (37.2% vs. 32.3%) and chills (18.4% vs. 18.5%) were comparable between ABLE and LAmB, respectively. Treatment-related serious adverse events were fewer in ABLE (0.3%) compared to LAmB (1.6%). Two deaths occurred in the ABLE group, of which one was probably related to the study drug. Nephrotoxicity and hepatotoxicity was not observed in either group.
ABLE 15 mg/kg single infusion had favorable efficacy and was well tolerated. Considering the demographic profile of the population in this region, a single dose treatment offers advantages in terms of compliance, cost and applicability.
www.clinicaltrials.gov NCT00876824
| Visceral leishmaniasis (VL) is highly prevalent in northeastern India, particularly the state of Bihar and its bordering areas with Bangladesh and Nepal. The current standards of treatment, namely, miltefosine (oral) and pentavalent antimonials (parenteral) have long treatment durations and are faced with increasing resistance, decreased efficacy, low compliance and safety issues. In this regard, lipid formulations of amphotericin B (AmB) have become an attractive treatment option due to their high efficacy, shorter treatment regimens and favorable safety profiles. This Phase III study evaluated the efficacy and safety of preformed AmB lipid emulsion (ABLE) versus liposomal AmB (LAmB) (both 15 mg/kg single dose infusions) in the treatment of VL. ABLE showed favorable efficacy measured in terms of initial cure at Day 30/45, and overall clinical improvement. ABLE was well tolerated and its adverse event profile was consistent with previously documented findings. Based on the favorable efficacy and safety profile of ABLE, and considering the demographic profile of the population in the endemic region, a single dose treatment may offer advantages in terms of compliance, cost and applicability.
| Visceral leishmaniasis (VL), also known as kala-azar, is a vector-borne disease transmitted to humans by the bite of an infected sandfly [1]. Globally, around 200,000–400,000 cases of VL occur each year of which 60% cases occur in India alone [2]. Kala-azar is a major public health problem in the areas of its prevalence, principally India and its neighbors Bangladesh and Nepal. In India, the disease is highly prevalent in Bihar, Jharkhand, West Bengal and pockets of eastern Uttar Pradesh. Among these, Bihar is the most affected with >90% of cases [3], of which 10% are fatal [2].
Contrary to the severity, few drugs are available for its treatment and are further limited by safety, reduced effectiveness and challenges in administration. Use of pentavalent antimonials, the mainstay of treatment for over 70 years, has been limited by its resistance and toxicities [4]. In India, almost 65% of previously untreated cases fail to respond promptly or relapse after treatment with antimonials [5]. Efficacy of the first-line oral treatment, miltefosine (MF) has declined rapidly over the past decade (final cure rate: 96.7% in 1999, 94% in 2002, 82% in 2007, and 72% in 2011) and is also associated with gastrointestinal side effects [6]–[9]. In addition, owing to its teratogenic effects, treatment with MF may require strict medical monitoring for treating women of child bearing age, which considering current demographic outlook of India is a significant factor [10], [11]. Paramomycin, an aminoglycoside, had shown 94% cure rate but is associated with systemic hepatic toxicity; the current regimen of 21 daily injections is also a major disadvantage for routine clinical use [12].
Amphotericin B (AmB), currently a second line drug used for treatment of VL, is highly effective with cure rates of 97%; however, the administration of 15 intravenous injections (i.v.) over 30 days of hospitalization, coupled with infusion- and drug-related adverse effects [13], has limited its wide-spread use. Liposomal formulations of AmB (LAmB) are better tolerated and thus preferable to conventional AmB [14], [15]. Despite the WHO-negotiated price of LAmB, treatment with it still remains limited and unaffordable in India [16]. Educational, social and economic background of patients in endemic areas entails therapy that does not bother patients with cost, undue compliance issues and long treatment duration, making a simplified treatment regimen a need of the hour. Thus, an affordable premixed AmB deoxycholate with lipid emulsion (ABLE) was developed (licensed in India) [17] and can be a potential candidate for treatment and elimination in endemic countries. Previous Phase II studies have reported safety and efficacy with a single infusion of 15 mg/kg of ABLE [17], [18] in the treatment of VL.
This Phase III study was conducted to evaluate the efficacy and safety of ABLE versus LAmB (both 15 mg/kg single dose infusions) in the treatment of VL.
The protocol was approved by an Independent Ethics Committee or Institutional Review Board at each study site and the study was conducted in accordance with the ethical principles originating in the Declaration of Helsinki and in accordance with ICH Good Clinical Practice guidelines, applicable regulatory requirements, and in compliance with the protocol. All participants including guardians in case of minors provided written informed consent to participate in the study. This study was registered at ClinicalTrials.gov (NCT00876824).
This was a prospective, multicentric, randomized, open-label, comparative Phase III study.
Patients were enrolled from 4 centers in Bihar, India, between August 2009 and January 2011. Male and female, aged 5–65 years (both inclusive) diagnosed with VL (fever >2 weeks duration and splenomegaly), who had amastigotes (Leishmania donovani bodies) at prescreening (detected by recombinant K39 protein [rK39] dipstick test) and confirmed VL by splenic or bone marrow aspirate smear examination were included in the study. Other inclusion criteria were hemoglobin (Hb) ≥5 g/dL, white blood cells count ≥1000/cmm, platelet count ≥50000/cmm, prothrombin time ≤4 seconds above the control, and alkaline transaminase, aspartate transaminase, and alkaline phosphatase ≤2.5 times the upper limit of normal. Patients with past history of treatment with AmB or any other drug for VL within 30 days prior to screening, major surgery within 2 weeks prior to screening, concurrent malaria, alcoholism or illicit drug use/abuse or any condition associated with poor compliance, hypersensitivity to AmB, inactive ingredients of ABLE and LAmB formulations were excluded from the study. Patients who received any of the prohibited medications (any other investigational drugs, antileishmanial drugs other than study drug, corticosteroids, skeletal muscle relaxants, cyclosporine, digoxin, vancomycin, aminoglycosides, antifungal, immunosuppresive agents, and all potentially nephrotoxic drugs), who were positive for human immunodeficiency virus, hepatitis C virus and hepatitis B surface antigen infections and immune-compromised, were also excluded from the study.
Eligible patients were randomized (3∶1) to receive either ABLE or LAmB, as 15 mg/kg single dose infusions (Figure 1). Prior to administration of full-dose, patients received initial test doses (ABLE and LAmB) of 1 mg in 5% dextrose as an infusion over ∼15–20 minutes for the ABLE treatment and over a period of 10 minutes for the LAmB treatment. Patients who experience any hypersensitivity or cardiopulmonary complications of hypersensitivity were withdrawn from the study. Full dose of ABLE and LAmB was diluted in 5% dextrose to a concentration of 1 mg/ml prior to administration. Patients received full doses of respective treatment in single intravenous infusion over 4–6 hours. Premedication was not allowed prior to the study drug administration. Patients were hospitalized for 7 days starting from day of first dose of the study drug for safety and efficacy evaluation.
To assess parasitological cure, splenic aspirate (or bone marrow aspirate in whom splenic aspirates was not feasible) was performed on Day 30 post infusion. Parasite density was graded by microscopy using a conventional logarithmic scale of 0 (no amastigotes/1000 oil-immersion fields) to +6 (>100 amastigotes/1000 oil-immersion field). Patients with +1 score on Day 30 were re-evaluated on Day 45. Patients were considered to achieve initial cure if the score was 0 on either Day 30 or 45. Patients with score >+1 on Day 30 and/or ≥1 on Day 45 were considered as treatment failures. They were withdrawn from the study and treated with rescue medication in appropriate doses as indicated in the protocol (LAmB 5 mg/kg i.v. on Days 1, 3, 5 and 7 or alternative antileishmanial drug in appropriate doses). Patients were further observed for clinical improvement presented as absence of fever and one or more of the following: increase in Hb concentration by ≥10%, weight gain, or decrease in spleen size by ≥33% (compared from baseline to Day 30). Patients who had achieved initial cure were followed-up for 6 months to study any sign/symptoms of relapse of VL. Patients with an initial cure and no signs or symptoms of VL at the last visit were considered to have achieved definitive cure.
All the patients were monitored for incidence of infusion related toxicities, nephrotoxicity, hepatotoxicity, number of adverse events (AEs), treatment-emergent AEs (TEAEs), serious AEs (SAEs), and laboratory values (normal; abnormal, not clinically significant; and abnormal, clinically significant) for different parameters. Any drug related Grade III or higher AEs recorded for abnormal clinically significant renal function tests or liver function tests were classified as nephrotoxicity or hepatotoxicity as per National Cancer Institute Common Terminology Criteria (NCI-CTC) AE, version 3.
A total of 500 patients in a 3∶1 ratio (375 in ABLE and 125 in LAmB) were planned to be enrolled assuming a dropout rate of 20% and non-inferiority margin fixed at −0.10. This was expected to provide an estimated difference in proportions of patients achieving definitive cure for ABLE vs. LAmB equals to zero, with at least 80% power for the non-inferiority test.
The permuted block randomization, with block size of 4, and ratio of 3∶1 in the two groups (ABLE and LAmB) were generated for each center. Eligible patients were sequentially allotted to unique subject ID and treatment (ABLE or LAmB) as per randomization schedule for that center. The screening and randomization log was maintained.
Data were expressed as means (±SD) for continuous variables and percentages for categorical variables. Proportion of patients achieving all three-efficacy (initial cure, clinical improvement and definitive cure) endpoints were to be compared across the two treatment groups.
For initial cure and clinical improvement, the data was to be analyzed using chi-square test at 5% level of significance. But as the expected number of patients achieving or non achieving initial cure in any of the treatment group was found to be <5, a Fisher's exact test was used. P-value<0.05 was considered as statistically significant. For definitive cure, non-inferiority was assessed by looking at the lower end of a two-sided 95% confidence interval (CI) of the difference Ptest - Pref (the difference in the proportions of patients achieving definitive cure in ABLE (Ptest) and LAmB (Pref). Non-inferiority was only accepted if the lower limit of the two-sided 95% CI was greater than the non-inferiority margin of −0.10. For the three efficacy parameters, the 95% CI was calculated by using Wald's confidence interval with Yate's continuity correction formula.
For safety, the number and percentage of patients experiencing toxicities and AEs (including laboratory abnormalities) across two treatment groups were summarized. Percentages were based on total number of patients in ITT population in each treatment groups.
The efficacy analysis was performed on modified intent-to-treat (mITT) population, which includes all patients who received study drug as per the protocol specified duration and had at least one efficacy assessment throughout the study. Safety analysis was performed on intent-to-treat (ITT) population, which includes all patients who received the treatment of study drug.
Of the 500 patients randomized, 376 patients received ABLE and 124 patients received LAmB. The percentage of patients who completed the study was lower in the ABLE group (86.7%) compared with the LAmB group (98.4%). A total of 50 (13.3%) patients discontinued the study in the ABLE group compared to 2 patients (1.6%) in the LAmB group (Figure 2). Patients were predominantly men (60.8%); mean age was 24.8 years (range: 5 to 62 years), and the rest of the baseline characteristics were similar for both groups (Table 1). In this study, all patients were qualified for treatment and included in the ITT population.
The proportion of patients with at least 1 AE was comparable in both the ABLE group and in the LAmB group (202 [53.7%] and 61 [49.2%]) (Table 3). The majority of AEs considered to be possibly related to the study drug was similar in both treatment groups (45.2%). Similarly, TEAEs in the ABLE (179 [47.6%]) and LAmB (56 [45.2%]) were comparable. The most common TEAEs in both ABLE and LAmB were chills (18.4% and 18.5%) and pyrexia (37.2% and 32.3%), respectively (Table 3). The majority (>35%) of the patients (152 [40.4%] vs. 46 [37.1%]) experienced AEs of mild intensity.
Two patients in each group had at least one SAE (ABLE 0.5% vs. LAmB 1.6%). Of these SAEs, one patient (0.3%) in the ABLE and 2 (1.6%) patients in the LAmB group was considered treatment-related. The SAEs that occurred in ABLE were anemia, diarrhea, vomiting and sudden death; while in LAmB, pancytopenia and diarrhea (in one patient each). In total, two deaths occurred in the ABLE group due to AEs. One death occurred 2 days after drug administration due to severe diarrhea and was considered probably related to the drug. The other death occurred on Day 157 and was not related to the study drug. In the LAmB group, one patient (0.8%) was discontinued from the study due to urticaria (Table 3).
The incidence of infusion related toxicities on Day 1 was comparable between the groups (43.6% in ABLE and 41.9% in LAmB group). None of the patients in both the treatment groups had signs and symptoms of nephrotoxicity and hepatotoxicity.
At present, VL remains one of the most neglected diseases globally [2]. To eliminate this endemic disease by 2015, a Tripartite Memorandum of Understanding agreement was signed in 2005 by the Governments of India, Nepal and Bangladesh wherein MF monotherapy was introduced as a first-line treatment [19]. However, of late, efficacy of MF has been declining steadily (96.7% to 72%) and its teratogenic potential remains a major concern in these areas where women are from low income groups and direct counseling is difficult, which limits its use in settings where the directly observed treatment is possible [10], [11], [20]. Other long-term treatment options, namely AmB (15 i.v injections over 30 days), pose a remarkable burden on the patient as well as health infrastructure [13]. Thus, short-course effective treatment regimens are greatly needed for the treatment of VL.
In this study, in the mITT population, efficacy of single day infusions of ABLE 15 mg/kg/day was satisfactory with an initial cure rate of 95.9% compared with 100% for LAmB. The difference in the initial cure rate was statistically significant between the groups (p = 0.028). However, this should be interpreted with caution, as in field settings LAmB is used as a single bolus dose of 10 mg/kg/day, compared to 15 mg/kg/day as was used in this study. This might have caused differences in the intended exposure to the treatment drug between groups. These results are in line with the results observed in a previous Phase II ABLE study [17].
Weight gain and decrease in spleen size were similar in both groups. The ABLE group showed greater increase in Hb concentration (79.4%) compared to the LAmB group (65.6%). Overall, clinical improvement (Day 30) was comparable (p = 0.6414) between the ABLE (98.9%) and LAmB groups (98.4%). Furthermore, the proportion of patients with no symptoms of relapse or showing no clinical signs of disease (definitive cure) was 85.9% in the ABLE group and 98.4% in the LAmB group.
The safety and tolerability of ABLE observed over 6 months duration was consistent with the earlier Phase II ABLE studies [17], [18]. Infusion related pyrexia and chills were the most common drug-related adverse events in both groups. These were mainly attributed to non-administration of premedication. However, in the field and in most studies, the patients are given premedication to prevent infusion related reactions [21]. No patient showed signs and symptoms of nephrotoxicity and hepatotoxicity, which was consistent with results from previous Phase II studies [17], [18]. Thus, the efficacy and safety results indicate that the treatment with ABLE is efficacious and safe.
Apart from efficacy and safety, which are important aspects of any drug or formulation, secondary aspects such as cost-effectiveness, affordability, better compliance and ease of availability must also be considered to assess the feasibility of its use. The necessity to assess secondary aspects of drug become even more important as VL mainly affects poor and neglected populations in East Africa and the Indian sub-continent [22]. It is estimated that illness due to VL may result in loss of income for up to 60% of the total household cost [23]. Thus, single-dose treatment regimens will not only reduce the hospital cost, but also drastically reduce the economic burden on the family [15]. Additionally, it has been proposed that extensive use of AmB and its formulations for treatment of VL may aid in decreasing the incidence of post-kala-azar dermal leishmaniasis, a potential reservoir of disease [24]. However, it is necessary to closely monitor and counsel patients about infrequent cases of post-kala-azar dermal leishmaniasis that tend to occur post-treatment with novel AmB formulations [25], [26].
In addition to poor economic conditions, co-infection with human immunodeficiency virus has changed the classical picture of VL in India, particularly in Bihar [27], [28]. Co-infected patients generally have poor response to the treatment, which leads to frequent relapses and high mortality [29]. In this context, it is imperative that such patients are diagnosed and treated appropriately through active case detection approaches as co-infected patients tend to transmit more virulent strains of VL.
In summary, novel formulations of AmB due to its high therapeutic index, short treatment courses and favorable safety profile have become an attractive treatment option [20]. However, in India, which bears the highest burden of VL, LAmB is not yet registered and is imported under special license, which can be a limiting factor in its availability. Also, India is not a recipient of the LAmB donation initiative, which may further impact its availability to poorer patients [16]. In addition, in India, LAmB is available at a preferential price only in limited quantities through specific treatment channels, thereby limiting its access to several patients infected with the disease. On the other hand, the ABLE formulation has been registered in India and can be supplied in liberal quantities, locally. Thus, favorable efficacy and safety profile, treatment compliance, affordability and availability of ABLE formulation make it a strong candidate for treatment of VL and for inclusion into the VL elimination program.
In this study, ABLE 15 mg/kg single bolus was found to be efficacious, safe and well tolerated in patients with VL. In addition, its ancillary properties such as favorable applicability and compliance (due to single dose administration), low cost and unrestricted supply, make it a suitable option for VL treatment in endemic countries.
|
10.1371/journal.ppat.1000775 | A New Nuclear Function of the Entamoeba histolytica Glycolytic Enzyme Enolase: The Metabolic Regulation of Cytosine-5 Methyltransferase 2 (Dnmt2) Activity | Cytosine-5 methyltransferases of the Dnmt2 family function as DNA and tRNA methyltransferases. Insight into the role and biological significance of Dnmt2 is greatly hampered by a lack of knowledge about its protein interactions. In this report, we address the subject of protein interaction by identifying enolase through a yeast two-hybrid screen as a Dnmt2-binding protein. Enolase, which is known to catalyze the conversion of 2-phosphoglycerate (2-PG) to phosphoenolpyruvate (PEP), was shown to have both a cytoplasmatic and a nuclear localization in the parasite Entamoeba histolytica. We discovered that enolase acts as a Dnmt2 inhibitor. This unexpected inhibitory activity was antagonized by 2-PG, which suggests that glucose metabolism controls the non-glycolytic function of enolase. Interestingly, glucose starvation drives enolase to accumulate within the nucleus, which in turn leads to the formation of additional enolase-E.histolytica DNMT2 homolog (Ehmeth) complex, and to a significant reduction of the tRNAAsp methylation in the parasite. The crucial role of enolase as a Dnmt2 inhibitor was also demonstrated in E.histolytica expressing a nuclear localization signal (NLS)-fused-enolase. These results establish enolase as the first Dnmt2 interacting protein, and highlight an unexpected role of a glycolytic enzyme in the modulation of Dnmt2 activity.
| Epigenetics refers to heritable changes in gene function that occur without alterations in the DNA sequence. The best characterized epigenetic modification is DNA methylation. In mammals, DNA methylation is associated with gene silencing and transposon control. We have previously established the presence of methyl cytosine in the genome of Entamoeba histolytica, an important unicellular human pathogen. Ehmeth, an enzyme that belongs to the DNA methyltransferase 2 (Dnmt2) family, catalyzes DNA methylation in the parasite. Recent evidence in support of the notion that human Dnmt2 is a tRNAAsp methyltransferase fuels the debate about the real function of the Dnmt2 family. Our results show that Ehmeth also catalyzes tRNAAsp methylation and indicates a dual function for this protein. In this study, we have also identified that enolase, a glycolytic enzyme, interacts with Ehmeth, and modulates its activity under conditions of glucose starvation. These data add to the emerging evidence that glycolytic enzymes have multifunctional roles, and emphasize the importance of energetic metabolism in the control of the epigenetic enzymatic machinery.
| The synthesis of 5-methylcytosine in both DNA and RNA is catalyzed by methyl 5-cytosine methyltransferases (m5C-MTase) with S-adenosylmethionine as a cofactor. The mammalian DNA methylation machinery consists of three active DNA m5C-MTases: Dnmt1, Dnmt3a and Dnmt3b. Dnmt1 has a high preference for hemi-methylated DNA as a substrate [1], whereas Dnmt3a and Dnmt3b are de novo DNA MTases that act on non-methylated DNA (for review, see Jeltsch [2]). A fourth DNA m5C-MTases, Dnmt2, belongs to a large family of proteins that are conserved in all species from Schizosaccharomyces pombe to humans. Dnmt2 stands apart from the three active DNA MTases because its length is relatively short when compared to that of Dnmt3a, Dnmt3b, or Dnmt1. Furthermore, this enzyme resembles prokaryotic DNA MTases because it does not have a large N-terminal regulatory domain [3].
Native tRNAAsp extracted from Dnmt2-deficient mice, Arabidopsis thaliana or Drosophila melanogaster were methylated in vitro by the human Dnmt2 (hDnmt2) protein. Accordingly, it was proposed that hDnmt2 is a tRNAAsp MTase rather than a DNA MTase [4], an idea that was further supported by the fact that it can also methylate transcribed tRNAs in vitro [5],[6]. On the other hand, the role of Dnmt2 seems to be not essential in higher eukaryotes because loss of function mutations of the Dnmt2 gene do not change genomic methylation patterns in the mouse [7]. In addition, depletion of D. melanogaster Dnmt2 (dDnmt2) by RNA interference has no detectable consequences on embryonic development [8]. However, a recent report has shown that loss of Dnmt2 in somatic cells eliminates H4K20 trimethylation at retrotransposons, and impairs maintenance of retrotransposon silencing [9]. Dnmt2 has been established as a genuine DNA methyltransferase in lower eukaryotes. Dnmt2 catalyzes DNA methylation in Dictyostelium discoideum [10],[11] and Entamoeba histolytica [12]. However, the weak DNA methyltransferase activity and the low expression level of Dnmt2 enzymes may explain the low methylation level that is found in these organisms [13]. Dnmt2 catalyzes cytosine methylation with a low preference for Cp(A/T) [8],[12],[14] or CC(A/T)GG [15], rather than the CpG motif. These results suggest that a dual specificity for DNA and RNA substrates emerged during the evolution of the Dnmt2 family [13]. Despite this dual specificity for DNA and RNA, the function of Dnmt2 as an RNA methyltransferase in lower eukaryotes has not yet been established.
The finding of interacting partners to members of the DNA/tRNA methyltransferase Dnmt2 is crucial for improving our existing understanding of its function. Until now, no interacting candidate has been reported for this family of proteins. In contrast, numerous proteins have been shown to interact with Dnmt1 and Dnmt3 thereby linking methylation to histone modifications and transcription regulation. For example both Dnmts were found to be associated with histone deacetylase [16],[17]. Dnmt1 was also found to interact with several chromatin- associated proteins, such as retinoblastoma protein, DNA methyltransferase 1 associated protein 1 and methyl CpG binding protein 2 [1], and Dnmt3 binds various transcription regulators, such as the transcriptional regulator RP58, the fusion protein of promyelocytic leukemia (PML) and the retinoic acid receptor-α (RARα) (PML-RAR) and heterochromatin protein 1 [18].
E.histolytica is an interesting model in which to study DNA methylation because Ehmeth, an enzyme that belongs to the Dnmt2 family, is the unique DNA methyltransferase that is present in this parasite [12]. The presence of methylated cytosine in E. histolytica ribosomal DNA [12] and the scaffold/matrix attachment region [19], together with the evidence that mutations can result from accelerated deamination of methylated cytosines in the reverse transcriptase of LINE retrotransposon (RT LINE) [20] support a role for Dnmt2 in the control of repetitive elements. This role has been confirmed in lower eukaryote Dictyostelium discoideum [10],[11] and in Drosophila [9]. Here, we establish that Ehmeth can catalyze the methylation of tRNAAsp. Moreover, we report, for the first time, that enolase, in addition to its involvement in the glycolytic pathway [21],[22], is an inhibitor of Dnmt2.
We carried out a yeast two-hybrid screen using a bait vector that expressed pAS1-Ehmeth that was fused to the GAL4 binding domain (GAL4BD) and an E.histolytica cDNA library that was fused to the GAL4 activation domain (GAL4AD) as prey. For this purpose, 106 clones were analyzed, and only two were selected based on their ability to grow on the selective medium (histidine, leucine, tryptophan and adenine) and results from the β-galactosidase complementation assays (data not shown). For each of the two positive clones, the recombinant plasmid that harbored the cDNA sequence that was fused to GAL4AD was isolated by transformation of E. coli cells, and then sequenced. These plasmids encode alcohol dehydrogenase (Accession number xp_653507.1) and enolase (Accession number xp_649161.1), respectively. Alcohol dehydrogenase was excluded from our analysis due to the presence of a frame shift mutation in its sequence.
In order to validate the interaction between enolase and Ehmeth, we carried out GST pull-down experiments. Ehmeth was first transcribed in vitro, and then translated in the presence of radioactive 35-S-methionine (TNT system) before incubating it with gluthatione beads that were coated with either GST-Ehenolase or GST. The result of this pull-down experiment shows that Ehmeth binds specifically to GST-Ehenolase, and not to GST (Fig. 1).
The existence of sequence homology between members of the Dnmt2 protein family and members of the enolase family suggests that the interaction between Ehmeth and enolase is conserved outside the Entamoeba genus. In order to test this hypothesis, Drosophila and human Dnmt2 proteins were transcribed in vitro, translated, and then incubated with GST-Ehenolase. Interestingly, both Dnmt2 proteins were able to bind to enolase (Fig. 1).
We previously reported that enolase is secreted by activated trophozoites [23]. In order to get further insights into the cellular localization of this protein, cytoplasmatic and nuclear trophozoite proteins that were prepared from HM-1∶MSS trophozoites were analyzed by western blotting with an antibody against enolase (Fig. 2A, 2C). The specificity of the enolase antibody that was raised against human enolase was confirmed against GST-Ehenolase using GST alone as the negative control (data not shown). The efficiency of the protein fractionation was examined by western blot analysis using antibodies against EhMLBP, a nuclear protein [24] and myosin II, a cytoplasmatic protein [25], as controls. As expected, EhMLBP was detected in the nuclear fraction and Myosin II in the cytoplasmatic fraction of the parasite (Fig. 2A). Enolase was detected as a 47 kDa protein present in the cytoplasmatic fraction of the parasite (Fig. 2A, 2C). Moreover, non-negligible amount of enolase were detected in the nuclear fraction of the parasite. To further validate these results, we examined the localization of enolase in the parasite using immunofluorescent microscopy (Fig. 2B). The result of this analysis showed that enolase is ubiquitously present in the parasite including its nucleus.
In order to test the binding of Ehmeth to enolase in the parasite, we conducted co-immunoprecipitation experiments using endogenous enolase with a calmodulin, histidine, hemagglutin (CHH)-tagged-Ehmeth in pJST4-Ehmeth transfected trophozoites nuclear lysate. We chose a tagged Ehmeth rather than the endogenous Ehmeth in these co-immunoprecipitation experiments because the antibody that we previously raised against Ehmeth [12] was unable to immunoprecipitate the protein (data not shown). A hemagglutin (HA) antibody was used to detect HA in the CHH tag. The expression of CHH-tagged Ehmeth in the nuclear fraction of pJST4-Ehmeth transfected trophozoites was confirmed by western blot analysis using an HA antibody (Fig. 2A).
We observed that enolase co-immunoprecipitated with CHH-tagged-Ehmeth (Fig. 3 left panel, Control). Ehmeth also co-immunoprecipitated with enolase (data not shown). In order to exclude the possibility that enolase interacts with the CHH tag and not with Ehmeth, enolase was immunoprecipitated from a nuclear lysate of trophozoites that expressed a CHH-KLP5 tagged protein [26] using the HA antibody. We observed that enolase does not co-immunoprecipitate with the CHH-KLP5 tagged protein, and this result indicates that no interaction occurred between enolase and the CHH tag (Fig. 3, right panel).
In order to delineate the enolase-interacting domains on Ehmeth, a series of deletion mutant proteins (Fig. 4, upper panel) were pulled down by either GST-Ehenolase or GST. We observed that N-terminal (from amino acid 1 to 103) and C-terminal (from amino acid 88 to 322) of Ehmeth were able to bind enolase in the same manner as full length Ehmeth (Fig. 4 lower panel). These results suggest that the specific region between amino acid 88 and 103, which is shared by the C-terminal and N-terminal Ehmeth mutant proteins is involved in the binding of Ehmeth to enolase. This region includes the catalytic site (domain IV) of Dnmt2 proteins [27]. In order to test this hypothesis, a mutant Ehmeth protein that lacks the amino acids 88 to 103 (EhmethΔ88–103) was generated, and its binding to GST-Ehenolase was examined. We found that the binding of EhmethΔ88–103 to enolase is impaired (Fig. 4 lower panel). It is important to emphasize that the input amount of the different Ehmeth deletion mutants proteins used in the GST-pull down assay were equivalent (data not shown). This result indicates that the domain IV contributes to the binding of Ehmeth to enolase. The catalytic domain of Dnmt2 proteins subsists as an exposed loop which is not part of the main structure [3]. According to this model, no significant conformational change in the structure of Ehmeth is expected , following the deletion of the amino acids 88 to 103.
We previously demonstrated that Ehmeth binds to EhMRS2, a DNA element, which contains the eukaryotic consensus scaffold/matrix attachment regions (S/MAR) bipartite recognition sequences [19]. We hypothesized that enolase regulate Ehmeth activity because it binds to its catalytic site. In order to test this hypothesis, GST-Ehmeth was incubated with P32 labeled EhMRS2 DNA in presence of various amount of GST-Ehenolase, and the denaturant-resistant DNA-Ehmeth complex [3] was analyzed by SDS-PAGE under denaturing conditions. In agreement with a previous report [19], GST-Ehmeth forms a complex with EhMRS2 DNA which is characterized by a retarded band in the SDS gel (Fig. 5A). No complex was observed when the labeled EhMRS2 DNA probe was incubated with either GST or GST-Ehenolase (Fig. 5A). The presence of Ehmeth in the retarded band was confirmed by mass spectrometry analysis (Fig. S1). Remarkably, the formation of Ehmeth-EhMRS2 complex was inhibited in the presence of Ehenolase (Fig. 5A). In order to confirm this result for hDnmt2, we tested its ability to bind EhMRS2 DNA. We found that hDnmt2 binds to EhMRS2 DNA (Fig. 5A). The formation of hDnmt2-EhMRS2 DNA complex was also strongly inhibited by Ehenolase. These results suggest that an identical inhibitory mechanism is used by enolase to inhibit the binding of Ehmeth and hDnmt2 to EhMRS2 DNA.
It has been reported that hDnmt2 catalyzes the methylation of tRNAAsp [4],[5],[6]. Therefore, we decided to examine this catalytic activity in E.histolytica because it has not yet been investigated in unicellular organisms. We found that the catalytic activity for Ehmeth was 9 U (Fig. 5B, left panel). This activity is substantially lower (about 100-fold) than that of hDnmt2 (Fig. 5B, right panel). GST has no detectable tRNAAsp MT activity. It has been reported that hDnmt2 methylates tRNAAsp using a DNA methyltransferase-like catalytic mechanism [6]. This last observation predicts that enolase will also inhibit the tRNAAsp MT activity of Ehmeth and hDnmt2. We confirmed this prediction by showing that the activity of Ehmeth and hDnmt2 tRNAAsp MT was strongly inhibited by enolase (approximately 60% and 90% inhibition, respectively) (Fig. 5B).
Enolase has been reported to undergo a conformational change following its binding to 2-PG [28],[29]. This observation prompted us to examine the effect of 2-PG on the inhibitory activity of enolase. For this purpose, the ability of enolase to inhibit the methylation of tRNAAsp by hDnmt2 was investigated in the presence of increasing concentrations of 2-PG. For this experiment, hDnmt2 was preferred to Ehmeth because its tRNA MT activity is significantly higher (see Fig. 5B). We observed that the inhibitory activity of enolase was reduced by 2-PG in a dose-dependent manner (Fig. 6A). This result may be explained by reduced enolase binding to hDnmt2 when 2-PG is present. In order to test this hypothesis, the binding of enolase and hDnmt2 was investigated in the presence of 2-PG (7 mM). Following the addition of 2-PG, we observed that the binding of enolase to hDnmt2 was strongly reduced (Fig. 6B). These results indicate that the inhibitory activity of enolase is regulated by its substrate, and suggest a link between the glycolytic pathway and Dnmt2 activity.
Our previous results indicated that 2-PG modulates the inhibitory activity of enolase. In order to assess the physiological relevance of this observation, we used glucose starvation as a means to reduce the level of 2-PG in the parasite. We chose to quantify intracellular pyruvate, the end product of glycolysis, as the method to monitor the effect of 12-hour glucose starvation instead of a direct measurement of 2-PG because its determination is easier than 2-PG. We observed that the level of pyruvate in glucose-starved trophozoites for12 hours was reduced by 50% when compared to non-starved control trophozoites (8×10−14 mol/ml vs 8×10−7 mol/ml). Longer glucose starvation (24 hours) resulted in significant death of the parasite (more than 50% of the original population, data not shown).
The localization of enolase during glucose starvation was followed by western blot analysis of cytoplasmatic and nuclear lysates. We consistently observed that at least three times more enolase was present in the nuclear lysate of 12-hour glucose-starved trophozoites than in non-starved control trophozoites (Fig. 7A, right panel). No accumulation of enolase in the nucleus was observed in trophozoites exposed to heat shock or oxidative stress (data not shown). The addition of glucose to the starved parasite restored the original distribution of enolase. This result emphasizes that the mechanism used to accumulate enolase in the nucleus is reversible. Moreover, immunoprecipitation analysis of the enolase-Ehmeth complex following glucose starvation for 12 hours showed that more enolase-Ehmeth complex was formed in the starved trophozoites than in the non-starved control trophozoites (Fig. 3, left panel).
In this study we showed that enolase inhibits Ehmeth. Accordingly, we hypothesized that the formation of Enolase-Ehmeth complex affects the level of DNA and tRNAAsp methylation following glucose starvation of the parasite. In order to test this hypothesis, the level of tRNA and DNA methylation in control and glucose starved trophozoites was determined. Accordingly, we observed, a significant decrease in tRNA methylation (38%) in glucose-starved trophozoites when compared to that determined in the non-starved trophozoites (Fig. 7B). Moreover, RT PCR analysis showed no significant difference in the amounts of tRNAAsp in glucose- starved and non-starved control trophozoites (Fig. 7C). In contrast, when we examined the level of DNA methylation in genomic DNA of control and glucose-starved parasites with an m5C antibody using dot blot analysis we could not detect any differences (Fig. 7D) [12]. This result indicates that DNA methylation is not affected by glucose starvation probably due to the short time (12 hours starvation). Therefore, to further examine the effect of enolase accumulation in the nucleus on DNA methylation we expressed enolase constitutively followed by a Nuclear Localization Signal (NLS) in the parasite.
The transfected trophozoites with NLS Enolase and trophozoites expressing a random 12 amino acids peptide followed by a NLS [30] which were used as control (NLS-Con transfectants) were cultured continuously in the presence of 24 µg mL−1 G418 for one month. The localization of enolase in NLS-Eno and NLS-Con transfectants was followed by western blot analysis of cytolasmic and nuclear lysates (Fig. 2C). We observed that 7 times more enolase was present in the nucleus of NLS-Eno transfectants than in NLS-con transfectants or non-transfected trophozoites (HM1∶MSS) (Fig. 2C, right panel). The level of DNA and tRNAAsp methylation in NLS-Con and NLS-Eno was determined (Fig. 7B and D). A significant decrease in both DNA and tRNAAsp methylation was observed in NLS-Eno transfectants when compared to that determined in NLS-Con transfectants. These results indicate that the continuous accumulation of enolase in the nucleus inhibit both Ehmeth DNA and tRNAAsp MT activity.
Of members of the Dnmt family of proteins, the roles of Dnmt1 and Dnmt3 are relatively well understood. In contrast, our knowledge about Dnmt2 is scanty. Furthermore there is no information about the molecules which interact with this protein. Therefore, the identification of such molecules would be a key step towards elucidating our understanding of Dnmt2 functions. Enolase, a glycolytic enzyme that catalyses the conversion of 2-PG to phosphoenolpyruvate, (PEP) is to the best of our knowledge the first Dnmt2-interacting protein to be described. For many years, glycolytic enzymes have been considered to be housekeeping cytoplasmatic proteins. Based on the results of studies on the function(s) of the glyceraldehyde-3-phosphate dehydrogenase, this concept has changed, and it is now well accepted that some of these enzymes that includes enolase, are multifunctional proteins which are involved in gene transcription, DNA replication, DNA repair, and nuclear RNA export (for review see [31]). The inability to select in complex growth media mutants of Bacillus subtilis [32], Escherichia coli [32] and E.histolytica enolase (data not shown) supports this multifunctional role. The catalytic activity of enolase in E. histolytica has been characterized [22], and it was found to be co-secreted with serpin and aldehyde alcohol dehydrogenase by activated trophozoites [23]. Indeed, antibodies against enolase have been detected in patients with amebiasis, and this suggests that enolase plays a role in the virulence of the parasite [33]. Such a role has been already reported in bacteria where enolase binds plasminogen [34]. The results of this investigation show that enolase is present in the cytoplasm and nucleus of E.histolytica. This ubiquitous localization is not unique to E. histolytica. In mammals, there are three isoforms of enolase (for review [35]), and each is characterized by its tissue distribution and expression. In HeLa cells, A. thaliana, and Plasmodium yoelii, enolase was found also in the nucleus. These observations raise the question about the significance of enolase presence in the nucleus. The results of our investigations on the nuclear role of enolase suggest that it is a Dnmt2 inhibitor.
The results from several recent studies have fuelled the debate on whether Dnmt2 is a DNA methyltransferase, a tRNA methyltransferase, or both. The results of our investigation support the notion that E.histolytica Dnmt2 (Ehmeth) is a DNA methyltransferase and a tRNA methyltransferase. Indeed, this is the first report of Dnmt2 being a tRNA methyltransferase in lower eukaryotes. Enolase has been reported to bind the bacteriophage-specific DNA adenine methyltransferase M.EcoT1. Interestingly, enolase binding to M.EcoT1 did not influence M.EcoT1 catalytic activity [36]. The domain IV of Ehmeth includes the catalytic sites, and is widely conserved among DNA-(cytosine-C 5)-methyltransferase. The binding of enolase to the domain IV of Ehmeth is probably the main mechanism of its inhibitory action. Dnmt2 methylates tRNA using a DNA methyltransferase-like catalytic mechanism [6]. Therefore, it is not surprising that the binding of enolase to Ehmeth interferes with both EhMRS2 DNA recognition and tRNAAsp MT activity. In S. cerevisiae, enolase interacts with cytosolic tRNALys in order to enable its translocation into the mitochondria, thereby displaying a function as a tRNA chaperone [37]. Our data showed that enolase does not interact with either DNA or tRNAAsp, thereby excluding competition as a mechanism to explain its Dnmt2 inhibitory activity. Only a few proteins have been reported to interact with the C-terminal domain, which contains the catalytic site for Dnmts. The P23 protein is a protein that is associated with steroid receptor complexes binds to the C-terminal of Dnmt1 [38]. However, its effect on Dnmt1 activity is still unclear. In contrast, p53 has been shown to stimulate Dnmt1 activity in vitro by binding to the C-terminal of Dnmt1 [39]. This last example together with our findings reinforce the notion that catalytic activity of Dnmt protein can be modulated by proteins that interact with their C-terminal.
The accumulation of enolase in the nucleus and the formation of an additional Ehmeth-enolase complex following glucose starvation support a central role for glucose metabolism in the regulation of Ehmeth activity. Glucose starvation was preferred to drugs in order to inhibit glycolysis because (i) one of the unwanted action of such drugs is the inhibition of proteasome activity [40], and (ii) the physiological relevancy of glucose starvation during Entamoeba differentiation [41]. Metabolites can act as sensors of the cell energy status. Therefore, they are convenient regulators of enzymes under conditions of physiological stress such as glucose starvation. For example, glucose starvation affects the activation or silencing of rRNA expression [42].
Glucose starvation led to significant TrnaAsp demethylation, but not to DNA demethylation. In contrast, forced expression of enolase in the nucleus led to both DNA and tRNAAsp demethylation. In mammals, active DNA demethylation is controversial [43]. Recently, a convincing mechanism of active DNA demethylation in which DNA glycosylase act as DNA demethylases through a base-excision-repair pathway has been proposed [44]. There is no evidence that active DNA demethylation occurs in E.histolytica. Passive demethylation occurs when DNA methylation is progressively reduced with cell division [45]. The generation time of the parasite is eight hours, and this would make it unlikely that DNA demethylation will occur following 12 hours of glucose starvation. However, this passive mechanism of DNA demethylation has probably occurred in the enolase-NLS strain during the numerous divisions of this strain. In contrast, the turnover of tRNA is much faster, and allows for rapid passive demethylation [46]. The physiological meaning of the Dnmt2-mediated methylation on tRNAAsp is still unknown. tRNA methylation has been involved in the control of tRNA stability [47],[48]. In S. cerevisiae, Trm9 mediated tRNA methylation is linked to the translation enhancement of genes related to stress response, DNA damage and other cellular functions [49],[50]. Mitochondrial tRNA methylation mediated by Trm 5 was shown to regulate mitochondrial protein synthesis [51]. These different functions for tRNA methylation represent an interesting starting point for further research on the role of tRNAAsp methylation in E.histolytica.
To conclude, the results of this investigation provide in vivo and in vitro evidence that establishes enolase as the first Dnmt2 interacting protein. Moreover, our results also provide strong evidence that link glucose metabolism and Dnmt2 activity. In addition, we have also shown that Dnmt2 is a tRNA methyltransferase in lower eukaryotes. The question of the significance of enolase-Dnmt2 interaction is higher eukaryotes needs further investigation.
Trophozoites of the E. histolytica strain HM-1∶IMSS were grown under axenic conditions in Diamond's TYI-S-33 medium (glucose concentration 750 mg/l) at 37°C. Trophozoites in the log phase of growth were used in all experiments. For the glucose starvation assays, trophozoites in the exponential phase of growth were washed three times and transferred to Diamond's TYI-S-33 medium that has been prepared without glucose (glucose concentration 31 mg/l). Recovery from glucose starvation was done by direct addition of 1% glucose to the culture of starved parasites.
Escherichia coli strain BL21 (DE3): F− ompT gal dcm lon hsdSB(rB− mB−) λ(DE3 [lacI lacUV5-T7 gene 1 ind1 sam7 nin5])
Saccharomyces cerevisiae strain Y190: MATa, gal4 gal180 his3 trp1–901 ade2–101 ura3–52 leu2–3, −112 + ura3::GAL→lacZ, LYS2: GAL(UAS)→HIS3 cyhr
S. cerevesiae Y190 was transformed with pGAL4-BD-Ehmeth (500 µg) using the LiAc transformation method [54].
The pGAL4-BD-Ehmeth strain was transformed with E.histolytica cDNA library (500 µg), and the transformants were then selected for their ability to grow on selective media that lacked leucine and tryptophan for four days at 30°C. After this first round of selection, the resistant clones were plated on a more selective media that lacked leucine, tryptophan, histidine, and adenine, and then grown for five days at 30°C. Fifteen resistant clones were then selected for further analysis. From these clones pACT2 vectors that contained cDNA inserts from E.histolytica library were isolated, and then transformed in the pGAL4-BD-Ehmeth strain. After the third round of selection, only two clones were able to grow on the selective media that lacked leucine, tryptophan, histidine, and adenine.
Coupled transcription and translation was carried out using a T7 TNT in vitro transcription/translation kit (Promega) in accordance with the manufacturer's instructions.
For the expression of the different GST-recombinant proteins, E. coli BL-21 that were transfected with the corresponding vectors were grown overnight in Luria Broth (LB) medium that contained 100 µg/ml ampicillin. The pre-cultures were inoculated (1∶100) with 2xYT medium that was supplemented with 100 µg/ml ampicillin, and grown for about two hours at 37°C until the OD600 reached 0.8. Induction of the fusion protein was initiated by adding isopropyl-beta-D-thiogalactopyranoside (IPTG) at a final concentration of 0.5 mM to the growing culture. After a four-hour incubation at 30°C, the bacteria were harvested in lysis buffer (100 mM KCl, 1 mM DTT, 1 mM PMSF, 100 µg/ml Lysozyme and Leupeptine 100 µg/ml in PBS), and then sonicated for five minutes with 30 seconds of pulses with 30 seconds between each pulsation session. The lysis was completed by addition of BugBuster protein extraction reagent (1∶100) (Novagen). The recombinant GST-proteins were purified under native conditions on a gluthatione-agarose resin (Sigma). Aliquots of GST fusion proteins that were bound to the glutathione-agarose beads were conserved at −70°C for the pull-down assay. The remaining recombinant proteins were then eluted with glutathione elution buffer (Tris HCl 50 mM pH 8.0, glutathione (Sigma) 10 mM), and their concentration was measured by Bradford's method [55].
Gluthatione sepharose beads that were coated with GST- Enolase, or GST alone (20–50 µg) were incubated with in vitro translated [35S]-methionine-labeled proteins (15 µl of the TNT reaction) in a final volume of 500 µl pull-down buffer (20 mM Hepes pH 7.9, 100 mM NaCl, 1 mM DTT, 6 mM MgCl2, 20% glycerol, 1% Nonidet P40 and 0.5 mM EDTA) for one hour at room temperature. The beads were then centrifuged at 3000 rpm for five minutes, washed three times with the pull-down buffer, and then incubated at 100°C in presence of 25 µl Laemmli sample buffer for five minutes. Interacting proteins were resolved on 12% SDS-polyacrylamide gel electrophoresis or 15% SDS-polyacrylamide gel when TNT-Ehmeth (1–88) protein was used. The resultant bands were visualized after staining with Coomassie blue, drying and autography exposure.
E.histolytica trophozoites nuclear and cytoplasmatic fractions were prepared in the identical manner as previously described [24]. Proteins were resolved on 12% SDS-polyacrylamide gel electrophoresis, and then transferred to nitrocellulose membranes. Blots were then blocked (3% skim milk powder), and then reacted with either 1∶500 enolase antibody (Santa Cruz Biotechnology) or with 1∶500 HA antibody (Santa Cruz Biotechnology). After incubation with the first antibody, the blots were incubated with 1∶5000 corresponding second antibody (Jackson ImmunoResearch), and then developed by enhanced chemoluminescence.
Trophozoites (106) that were grown in regular or glucose-deficient media were washed three times with PBS, and then resuspended in 2 ml of PBS. The trophozoites were lysed by freezing and then thawing to produce a total protein lysate. The pyruvic acid level in trophozoite lysates was determined according to a previously described method [56]. Briefly, 1 ml of 2,4- dinitrophenylhydrazine (DNPH) (0.0125% in 2 N HCL) was added to 1 ml of trophozoite lysate. After 15 minutes of incubation at 37°C in a water bath, the sample was removed from the water bath, and 5 ml 0.6 N NaOH was added. The absorbance of the sample was then measured in a spectrophotometer at 420 nm. A standard curve was generated using sodium pyruvate [56].
Male BALB/c mice were injected intraperitoneally with 100 mg of GST-Enolase recombinant protein that was emulsified in complete Freund's adjuvant. Two and four weeks later, the mice were injected with 100 mg of the recombinant protein in incomplete Freund's adjuvant. One week after the 4-week injection, about 0.8 ml of sera was obtained by retro-orbital puncture. Serum that was obtained from mice that were not injected with recombinant protein was used as the control.
Trophozoites in a logarithmic growth phase were harvested, transferred to 8 mm round wells on glass slides, and then incubated for 30 min at 37°C in order to allow them attach to the glass surface. An indirect immunofluorescence assay was performed. For this purpose, the amebae were fixed with cold methanol for 20 min at −20 C, and then incubated with 1∶400 enolase antibody for one hour at room temperature. After washing, the samples were then incubated with goat Cy3-conjugated anti-mouse (Jackson ImmunoResearch) 1∶1000 for one hour. Samples were then stained with 4,6-diamidino-2-phenylindole dihydrochloride (DAPI,Sigma) in order to visualize the nuclei. Fluorescent images were captured by a CCD camera attached to an Axioscop2 (Zeiss) epifluorescence microscope with a 100/1.30 Plan Neofluar oil immersion objective and a differential interference contrast filter. The images were analyzed with ImagePro@Plus software (Media Cyberneticx, USA).
Aliquots of nuclear protein fraction (50 µg) were diluted in 20 mM Hepes pH 7.5, 150 mM NaCl, 0.1% Triton, 10% glycerol (HNTG buffer) (300 µl), and then incubated with protein G beads (Sigma) (10 µl) for 30 minutes at 4°C. Non-specific interacting proteins were excluded by centrifugation (3000 rpm at 4°C for 5 minutes). The supernatant was incubated with either 1∶200-HA antibody or enolase antibody) for two hours at 4°C. Following incubation protein G-Sepharose beads (20 µl) were added to the samples which were then incubated for 16 hours at 4°C. Immunoprecipitated proteins were collected by centrifugation, washed three times with HNTG buffer, and then resolved by 12% SDS-polyacrylamide gel electrophoresis. The proteins were then transferred to nitrocellulose membranes by western blot analysis, and detected with the relevant antibody, mouse anti-enolase or rabbit anti-HA.
EhMRS2 was amplified from E.histolytica genomic DNA by PCR and the primers EhMRS2 5 and EhMRS2 3. EhMRS2 DNA (10 pmol) was end-labeled with T4-polynucleotide kinase (New England Biolabs) and γ-ATP in accordance with the manufacturer's recommendations. Unincorporated γ-ATP was removed with the ProbeQuant kit (Amersham).
Gluthatione sepharose beads that were coated with GST-Ehmeth, GST-hDnmt2, or GST alone (35 µg) were incubated in 100 µl blocking buffer (3% BSA and salmon sperm DNA 1 µg/ml in standard binding buffer (20 mM Tris-HCl pH 8, 50 mM NaCl, 1 mM EDTA in double distilled water) for 30 minutes at room temperature. Following blocking the beads were washed three times with standard binding buffer, and incubated with either 40 µg or 60 µg of GST- Enolase for one hour at room temperature (100 µl final reaction volume). The probe (0.3 µg) was then added, and binding was carried out at 4°C overnight. Subsequently, the beads were washed three times in standard binding buffer, boiled with 25 µl Laemmli sample buffer for 5 minutes; proteins were separated on 10% SDS-polyacrylamide gel electrophoresis. The signal of the proteins that were bound to the labeled DNA probe was detected directly from the polyacrylamide gel on X-ray film (Fuji).
The methylation assay of tRNAAsp with the DNMT2 variants was performed using a previously described method [6]. Briefly, the DNA template that encoded Drosophila tRNAAsp was amplified by PCR and the T7 primer and tRNAAsp primer. For in vitro transcription, 100 µl of the PCR reaction were incubated with 200 µl 2× transcription buffer (80 mM Tris-HCl at pH 8.1, 2 mM Spermidine, 10 mM DTT, 0.02% Triton-X-100, 60 mM MgCl2, 4 µg/ml BSA), 5 mM of each NTP (final concentration), and 10 µl of T7-Polymerase (200 units/µl; Fermentas) in a final volume of 400 µl for three hours at 37°C. Transcripts were purified over 12% denaturing PAGE, and bands of correct size were excised, eluted in 0.5 M ammonium acetate, and precipitated with two volumes of 100% ethanol. After centrifugation, RNA pellets were washed once with 80% ethanol, and then dissolved in double distilled water. The concentration of tRNA was measured with a nanodrop spectrophotometer.
Protein bands of interest were excised from the SDS-polyacrylamide gel and digested with trypsin using a previously published protocol [57], and then analyzed by MALDI-TOF mass spectrometry analysis that was done at the Institute of Biology, Technion, Israel. The peptide mass profiles that were produced by MALDI-TOF mass spectrometry were processed using PepMiner (this software is described at http://www.haifa.il.ibm.com/projects/verification/bioinformatics/). Peptides masses were compared with the theoretical masses that were derived from the sequences that were in the SWISS-PROT/TrEMBL (http://www.expasy.ch/sprot/), the NCBI (http://www.ncbi.nlm.nih.gov/), and the E. histolytica genome project databases (http://pathema.jcvi.org/cgi-bin/Entamoeba/GenomePage.cgi?org=eha2).
Aliquots (40 pmol) of Drosophila tRNAAsp were incubated with 0.4 nmol Ehmeth or 0.04 nmol GST-hDnmt2 for three hours at 37°C in 40 µl of methylation buffer (100 mM Tris/HCl at pH 7.5, 5% glycerol, 5 mM MgCl2, 1 mM DTT, and 100 mM NaCl) that contained 4.2 µM labeled [methyl-3H] AdoMet (NEN). When we examined the effect of enolase on Ehmeth activity, GST-Enolase (2 nmol) or GST as negative control (2 nmol) were incubated with Ehmeth for one hour at 37°C. When we examined the effect of enolase on hDnmt2 activity, GST-Enolase (0.4 nmol) or GST (0.4 nmol) were incubated respectively with hDnmt2 for one hour at 37°C. Samples (8 µl) were taken from reaction mix (40 µl) at different times, and loaded on Whatman filters. The filters were then washed with 10% Trichloroacetic Acid Solution (TCA) three times and finally with 100% ethanol. After washing the filters were air-dried and transferred into tubes following addition of 3 ml scintillation liquid (CytoScint). The incorporated radioactivity was measured in a scintillation counter (Counter Beta Tri-Carb 2100TR). tRNA methyltransferase activity (one unit (U)) was expressed as the incorporation of 1 pmol AdoMet per hour per nmol of protein. In vitro tRNA methylation assay in the presence of 2 phosphoglycerate (2-PG) (Fluka) was done in the identical manner with minor modifications. Increasing 2-PG concentrations (1–7 mM) were incubated with GST-Enolase (0.4 nmol) and with hDnmt2 (0.04 nmol). The activity of hDnmt2 in the presence of 7 mM 2-PG was used as control.
Total RNA was prepared with the TRI-Reagent kit (Sigma) from control or glucose-starved trophozoites and treated with DNase I to remove any contamination of DNA. Aliquots from the treated RNA (20 µg), were used as substrates for hDnmt2 in vitro tRNA methylation assay (see above protocol). The amount of methyl groups that was incorporated by hDnmt2 into the tRNA of each sample is proportional to the amount of unmethylated tRNA in the control sample.
E.histolytica enolase: XP_649161.1, Ehmlbp: XP_649236, Ehmeth: XP_655267.2, Myosin II: XM_651936.1, hDNMT2: NP_004403.1
|
10.1371/journal.pcbi.0030039 | Discovering Motifs in Ranked Lists of DNA Sequences | Computational methods for discovery of sequence elements that are enriched in a target set compared with a background set are fundamental in molecular biology research. One example is the discovery of transcription factor binding motifs that are inferred from ChIP–chip (chromatin immuno-precipitation on a microarray) measurements. Several major challenges in sequence motif discovery still require consideration: (i) the need for a principled approach to partitioning the data into target and background sets; (ii) the lack of rigorous models and of an exact p-value for measuring motif enrichment; (iii) the need for an appropriate framework for accounting for motif multiplicity; (iv) the tendency, in many of the existing methods, to report presumably significant motifs even when applied to randomly generated data. In this paper we present a statistical framework for discovering enriched sequence elements in ranked lists that resolves these four issues. We demonstrate the implementation of this framework in a software application, termed DRIM (discovery of rank imbalanced motifs), which identifies sequence motifs in lists of ranked DNA sequences. We applied DRIM to ChIP–chip and CpG methylation data and obtained the following results. (i) Identification of 50 novel putative transcription factor (TF) binding sites in yeast ChIP–chip data. The biological function of some of them was further investigated to gain new insights on transcription regulation networks in yeast. For example, our discoveries enable the elucidation of the network of the TF ARO80. Another finding concerns a systematic TF binding enhancement to sequences containing CA repeats. (ii) Discovery of novel motifs in human cancer CpG methylation data. Remarkably, most of these motifs are similar to DNA sequence elements bound by the Polycomb complex that promotes histone methylation. Our findings thus support a model in which histone methylation and CpG methylation are mechanistically linked. Overall, we demonstrate that the statistical framework embodied in the DRIM software tool is highly effective for identifying regulatory sequence elements in a variety of applications ranging from expression and ChIP–chip to CpG methylation data. DRIM is publicly available at http://bioinfo.cs.technion.ac.il/drim.
| A computational problem with many applications in molecular biology is to identify short DNA sequence patterns (motifs) that are significantly overrepresented in a target set of genomic sequences relative to a background set of genomic sequences. One example is a target set that contains DNA sequences to which a specific transcription factor protein was experimentally measured as bound while the background set contains sequences to which the same transcription factor was not bound. Overrepresented sequence motifs in the target set may represent a subsequence that is molecularly recognized by the transcription factor. An inherent limitation of the above formulation of the problem lies in the fact that in many cases data cannot be clearly partitioned into distinct target and background sets in a biologically justified manner. We describe a statistical framework for discovering motifs in a list of genomic sequences that are ranked according to a biological parameter or measurement (e.g., transcription factor to sequence binding measurements). Our approach circumvents the need to partition the data into target and background sets using arbitrarily set parameters. The framework is implemented in a software tool called DRIM. The application of DRIM led to the identification of novel putative transcription factor binding sites in yeast and to the discovery of previously unknown motifs in CpG methylation regions in human cancer cell lines.
| This paper examines the problem of discovering “interesting” sequence motifs in biological sequence data. A widely accepted and more formal definition of this task is: given a target set and a background set of sequences (or a background model), identify sequence motifs that are enriched in the target set compared with the background set.
The purpose of this paper is to extend this formulation and to make it more flexible so as to enable the determination of the target and background set in a data driven manner.
Discovery of sequences or attributes that are enriched in a target set compared with a background set (or model) has become increasingly useful in a wide range of applications in molecular biology research. For example, discovery of DNA sequence motifs that are overabundant in a set of promoter regions of co-expressed genes (determined by clustering of expression data) can suggest an explanation for this co-expression. Another example is the discovery of DNA sequences that are enriched in a set of promoter regions to which a certain transcription factor (TF) binds strongly, inferred from chromatin immuno-precipitation on a microarray (ChIP–chip) [1] measurements. The same principle may be extended to many other applications such as discovery of genomic elements enriched in a set of highly methylated CpG island sequences [2].
Due to its importance, this task of discovering enriched DNA subsequences and capturing their corresponding motif profile has gained much attention in the literature. Any approach to motif discovery must address several fundamental issues. The first issue is the way by which motifs are represented. There are several strategies for motif representation: using a k-mer of IUPAC symbols where each symbol represents a fixed set of possible nucleotides at a single position (examples of methods that use this representation include REDUCE [3], YMF [4,5], ANN-SPEC [6], and a hypergeometric-based method [7]) or using a position weight matrix (PWM), which specifies the probability of observing each nucleotide at each motif position (for example MEME [8], BioProspector [9], MotifBooster [10], DME-X [11], and AlignACE [12]). Both representations assume base position independence. Alternatively, higher order representations that capture positional dependencies have been proposed (e.g., HMM and Bayesian networks motif representations [13]). While these representations circumvent the position independence assumption, they are more vulnerable to overfitting and lack of data for determining model parameters. The method described in this paper uses the k-mer model with symbols above IUPAC.
The second issue is devising a motif scoring scheme. Many strategies for scoring motifs have been suggested in the literature. One simple yet powerful approach uses the hypergeometric distribution for identifying enriched motif kernels in a set of sequences and then expanding these motifs using an EM algorithm [7]. The framework described in this paper is a natural extension of the approach of [7]. YMF [4,5] is an exhaustive search algorithm which associates each motif with a z-score. AlignACE [12] uses a Gibbs sampling algorithm for finding global sequence alignments and produces a MAP score. This score is an internal metric used to determine the significance of an alignment. MEME [8] uses an expectation maximization strategy and outputs the log-likelihood and relative entropy associated with each motif.
Once a scoring scheme is devised, a defined motif search space is scanned (either heuristically or exhaustively) and motifs with significantly high scores are identified. To determine the statistical significance of the obtained scores, many methods resort to simulations or ad hoc thresholds. Several excellent reviews narrate the different strategies for motif detection and use quantitative benchmarking to compare their performance [14–18]. A related aspect of motif discovery, which is outside the scope of this paper, focuses on properties of clusters and modules of TF binding sites (TFBS). Examples of approaches that search for combinatorial patterns and modules underlying TF binding and gene expression include [19–23].
One issue of motif discovery that is often overlooked concerns the partition of the input set of sequences into target and background sets. Many methods rely on the user to provide these two sets and search for motifs that are overabundant in the former set compared with the latter. The question of how to partition the data into target and background sets is left to the user. However, the boundary between the sets is often unclear and the exact choice of sequences in each set arbitrary. For example, suppose that one wishes to identify motifs within promoter sequences that constitute putative TFBS. An obvious strategy would be to partition the set of promoter sequences into target and background sets according to the TF binding signal (as measured by ChIP–chip experiments). The two sets would contain the sequences to which the TF binds “strongly” and “weakly,” respectively. A motif detection algorithm could then be applied to find motifs that are overabundant in the target set compared with the background set. In this scenario, the positioning of the cutoff between the strong and weak binding signal is somewhat arbitrary. Obviously, the final outcome of the motif identification process can be highly dependent on this choice of cutoff. A stringent cutoff will result in the exclusion of informative sequences from the target set while a promiscuous cutoff will cause inclusion of nonrelevant sequences—both extremes hinder the accuracy of motif prediction. This example demonstrates a fundamental difficulty in partitioning most types of data. Several methods attempt to circumvent this hurdle. For example, REDUCE [3] uses a regression model on the entire set of sequences. However, it is difficult to justify this model in the context of multiple motif occurrence (as explained below). In other work, a variant of the Kolmogorov-Smirnov test was used for motif discovery [24]. This approach successfully circumvents arbitrary data partition. However, it has other limitations such as the failure to address multiple motif occurrences in a single promoter, and the lack of an exact characterization of the null distribution. Overall, the following four major challenges in motif discovery still require consideration: (c1) the cutoff used to partition data into a target set and background set of sequences is often chosen arbitrarily; (c2) lack of an exact statistical score and p-value for motif enrichment. Current methods typically use arbitrarily set thresholds or simulations, which are inherently limited in precision and costly in terms of running time; (c3) a need for an appropriate framework that accounts for multiple motif occurrences in a single promoter. For example, how should one quantify the significance of a single motif occurrence in a promoter against two motif occurrences in a promoter? Linear models [3] assume that the weight of the latter is double that of the former. However, it is difficult to justify this approach since biological systems do not necessarily operate in such a linear fashion. Another issue related to motif multiplicity is low complexity or repetitive regions. These regions often contain multiple copies of degenerate motifs (e.g., CA repeats). Since the nucleotide frequency underlying these regions substantially deviates from the standard background frequency, they often cause false-motif discoveries. Consequently, most methods mask these regions in the preprocessing stage and thereby lose vital information that might reside therein; (c4) criticism has been made over the fact that motif discovery methods tend to report presumably significant motifs even when applied on randomly generated data [25]. These motifs are clear cases of false positives and should be avoided.
In this paper we describe a novel method that attempts to solve the above-mentioned four challenges in a principled manner. It exploits the following observation: data often lends itself to ranking in a natural manner, e.g., ranking sequences according to TF binding signal: ranking according to CpG methylation signal, ranking according to distance in expression space from a set of co-expressed genes, ranking according to differential expression, etc. We exploit this inherent ranking property of biological data in order to circumvent the need for an arbitrary and difficult-to-justify data partition. Consequently, we propose the following formulation of the motif finding task: given a list of ranked sequences, identify motifs that are overabundant at either end of the list.
Our solution employs a statistical score termed mHG (minimal hypergeometric) [26]. It is related to the concept of rank-imbalanced motifs, which are sequence motifs that tend to appear at either end of a ranked sequence list. In previous work [26], the authors used mHG to identify sequence motifs in expression data. We use this simple yet powerful approach as the starting point for our study.
The rest of this paper is divided into two main parts, each of which is self-contained: in the Results we briefly outline our method and describe new biological findings that were obtained by applying this method to biological data. We address challenge (c4) by testing the algorithm on randomly ranked real genomic sequences. In the Methods, we describe the mHG probabilistic and algorithmic framework and explain how we deal with challenges (c1)–(c3).
Based on the mHG framework, we developed a software tool termed DRIM (discovery of rank imbalanced motifs) for motif identification in DNA sequences. A flow chart of DRIM is provided in Figure 1. The formal introduction and details of the mHG statistics are given in Methods. However, to facilitate the explanation and interpretation of our biological results, we begin with a brief description of the method.
Suppose we are given a set of DNA sequences and some measured signal associated with each sequence. We rank the sequences according to the signal. Now, given a sequence motif, we wish to assess whether that motif tends to appear more often at the “top” of a list compared with the “remainder” of the list. The mHG score captures this type of motif significance. More precisely, the mHG score reflects the surprise of seeing the observed density of motif occurrences at the top of the list compared with the rest of the list under the null assumption that all configurations of motif occurrences in the list are equiprobable. A unique feature of the mHG statistics is that the cutoff between the top and the rest of the list is chosen in a data-driven manner so as to maximize the motif enrichment. This is done by computing the motif enrichment over all possible set partitions and identifying the cutoff at which maximal statistical significance is observed.
The search for this optimal cutoff introduces a multiple testing problem. To solve this without resorting to multiple testing corrections, which diminish the score's sensitivity, we provide a novel algorithm for computing the exact p-value of mHG scores (see Methods, Calculating the p-value of the mHG score). This eliminates the need to resort to simulations or exhaustively calculated tables.
Our method also includes a new approach to modeling motif multiplicity by incorporating a multidimensional hypergeometric framework (see Methods, Multidimensional mHG score). Unlike some models, which assume linearity (e.g., that two binding motifs have twice the binding capacity as one motif), our model does not make such pre-assumptions. Instead, the degree of surprise is adjusted for each motif according to its own occurrence multiplicity distribution.
DRIM scans through a motif space, computes the mHG p-value of these motifs and reports the significant ones (see Methods, The DRIM software).
We begin by testing our method on synthetically generated clear-cut positive and negative control cases. We do this to verify that DRIM accurately identifies motifs in well-characterized and experimentally verified examples and at the same time avoids false identification of motifs in randomly ordered genomic sequences. The latter objective is of particular importance since the issue of false identification has been mentioned as one of the main shortcomings of motif discovery approaches. For example, in a previous study, six different motif discovery applications were used to search for TFBS motifs [25]. Each of the programs attempted to measure the significance of its results using one or more enrichment scores. The authors report that the applications outputted high-scoring motifs even when applied to random selections of intergenic regions. A different paper reports clusters of genes whose expression patterns correlate to the expression of a particular TF [27]. These clusters were then analyzed for enriched motifs. Again, the authors report that random sets, with sizes matching those of the real clusters, contained a large number of motifs with significant scores.
To test our method's false-prediction rate, we performed the following negative control experiment: five different random permutations of ChIP–chip data were generated by randomly selecting 400 promoters and randomly permuting their ranks. DRIM was then applied to these ranked lists and scanned more than 100,000 different motifs in each one. None of the motifs that were scanned had a significant corrected mHG p-value <10−3. Note that to get the corrected p-values, two levels of multiple test corrections are performed: correcting for the number motifs that are tested; and correcting for multiple cutoffs that are tested as part of the mHG optimization process.
How do the p-values of random motifs compare with those of true biological motifs? To test this, we chose five TFs (BAS1, GAL4, CBF1, INO2, and LEU3) whose motif binding sites are well-characterized and experimentally verified. We applied DRIM to the ChIP–chip data of these TFs as reported in [25]. In all instances, the true motifs were identified with corrected p-values of 10−6, 10−9, 10−76, 10−18, and 10−8, respectively. A comparison of the p-value distribution of the motifs in the randomly ordered sequences with that of the verified TFBS motifs is given in Figure S3. In all instances the true TFBS motifs were predicted with p-values that were several orders of magnitude more significant than the best p-value of a motif in the randomly permuted data. This indicates that the enrichment signals of true TFBS, as captured by the mHG p-value, are clearly distinct from the signals we expect to find in random rankings of genomic sequences.
To further test the effectiveness of our method, we used it for identification of TFBS in yeast by applying it to the Harbison and Lee–filtered ChIP–chip datasets [25,28], containing measurements of 207 TF binding experiments in several conditions (for details regarding dataset-filtering see Methods). Interestingly, we observed that in many of these datasets longer intergenic regions are biased toward stronger TF binding. We elaborate on this sequence length bias in the Methods section and in Figure S1.
In each of the ChIP–chip experiments, we ranked the intergenic regions according to the TF binding signal (we use the p-value of enrichment for the sequence represented on the array). This was used as input for DRIM, which then searched for motifs that tend to appear densely at the top of the ranked lists. If such a motif does exist, with a p-value less than 10−3, then we hypothesize that it is biologically significant and that it contributes to the TF's binding, either directly or indirectly.
The results on the Harbison filtered dataset are summarized in Table S2. A TF was assigned a motif if such was found in at least one condition. We compared the DRIM predictions with previously reported TFBS discoveries in ChIP–chip that incorporated predictions of six other motif discovery methods and conservation data [25]. The results of this comparison are summarized in Figure 2.
Overall, DRIM identified 50 motifs that were not picked up by the six other methods as reported in [25]. We further investigated these putative TFBS for additional evidence that they are biologically meaningful. First, we found that seven of them (ASH1, GCR1, HAP2, MET31, MIG1, RIM101, and RTG3) are in agreement with previously published results that are based on experimental techniques other than ChIP–chip. Second, we compared them with a list of conserved regulatory sites in yeast that was recently inferred using conservation-based algorithms [29]. Ten of our putative TFBS match these conserved sites (ARG81, ARO80, ASH1, CRZ1, DAL81, HAP2, IME1, MET31, MIG1, and RTG3). Taken together, these findings provide a strong indication that at least some of the new motifs identified by DRIM are true biological signals. In the following subsections, we focus on a few of these putative TFBS (see Figure 3) and present additional evidence that supports their biological role. We use these findings to discover new interactions in the yeast genetic regulatory network.
To examine our method's ability to predict sequence motifs that stem from data other than TF binding, DRIM was applied to a dataset containing the human cancer cell line–methylated CpG islands (for dataset details, see Methods) to seek for motifs that are enriched in hypermethylated regions. The promoters were ranked according to methylation signal, with hypermethylated promoters at the top. Note that different replicates of the same cell line may yield different ranking of the promoters.
DRIM identified significantly enriched motifs in each of the four cancer cell lines. Table 1 shows all the motifs that were independently discovered in at least two different replicates of the same experiment or that are in agreement with previous work [2]. Overall, DRIM discovered 13 motifs: ten novel motifs and three that have been previously predicted in hypermethylated CpG island promoters in the same cancer cell lines [2]. Some of these motifs have also been independently identified in methylated CpG regions of other cell lines [39,40].
Interestingly, nine of the novel ten motifs were independently identified in DNA regions to which the proteins of the Polycomb complex bind [41–43]. The Polycomb complex is involved in gene repression through epigenetic silencing and chromatin remodeling, a process that involves histone methylation. The fact that these two distinct key epigenetic repression systems, namely histone methylation and CpG methylation, bind to regions that share a similar set of sequence motifs suggests they are linked. To further establish this link we applied DRIM to Polycomb complex bound promoters in human embryonic fibroblasts [44]. We found four motifs that are similar to the CpG methylation motifs (Table 1). Our findings are consistent with a recent paper that showed that the EZH2 Polycomb protein binds methyltransferases via the Polycomb complex [45].
Most of the motifs we found are similar across more than one type of cancer cell line, e.g., variants of the GCTGCT motif appear in Caco-2, PC3, and Polyp1 cancer cell lines. This suggests that the same DNA binding factors are involved in CpG methylation of different types of cancers. It is also important to note that some of the motifs we discovered are G–C rich. The enrichment of these motifs may be partially attributed to the G–C content bias that is found in CpG methylation data.
The DRIM motif identification process can be used not only to identify novel motifs but also to partition the data in a biologically meaningful manner. In [2] the authors used a fixed threshold on the methylation signal (p-value < 0.001 ) to partition the dataset. Consequently, they identified 135 hypermethylated promoters. A data-driven partition would be to use the threshold that yielded the maximal motif enrichment. For example, in the Caco2 cell line, we identified the same motif as in the previous work [2]. However, the motif maximal enrichment was found in the top 209 promoters (an increase of 54% in target set size).
Human TFBS tend to be longer and “fuzzier” than TFBS of lower eukaryotes, and it is important to evaluate our method's performance on such motifs. To this end, we applied DRIM to the ChIP–chip experiments of HNF1α, HNF4α, HNF6 in liver and pancreas islets [46], as well as to that of CREB [47]. For each of the TFs, we generated a list of sequences containing 1,000 bases upstream and 300 downstream of the transcription start site (TSS). We ranked the list according to the TF ChIP–chip signal and used it as input to DRIM. DRIM successfully detected the TFBS of these TFs that are reported in TRANSFAC with extremely significant p-values: HNF1α liver—GTTAMWNATT (p = 10−8), HNF4α Islets—SCGGAAR (p = 10−53), HNF6 Liver—ATCRAT (p = 10−57), and HNF6 Islets—ATCRAT (p = 10−61). In the CREB experiments we identified the palindromic motif TGACGTCA (p = 10−16), which is known to bind CREB [47].
Three properties of the mHG enrichment score embodied in DRIM offer advantages over other motif discovery methods: the dynamic cutoff, the rigorous control over false positives, and the motif multiplicity model.
In this paper we examine the problem of discovering “interesting” motif sequences in biological sequence data. While this problem has often been regarded as tantamount to discovering enriched motifs in a target set versus a background set, we point out an inherent limitation to this formulation of the problem. Specifically, in most cases, biological measurement data does not lend itself to a single, well-substantiated partition into target and background sets. It does, however, lend itself to ranking in a natural manner. Our approach exploits this natural ranking and attempts to solve challenges (c1)–(c4) (see Introduction, Open challenges in motif discovery).
To address challenge (c1), instead of choosing an arbitrary cutoff for set partition, we search for a cutoff that partitions the data in a way that maximizes the motif enrichment. We present evidence that shows that the flexible mHG cutoff outperforms the rigid cutoff. One example of this is shown in Figure 5, where the flexible cutoff yields better results for all the tested TFs. Another example of the advantage of a flexible cutoff is the two motifs detected in three TFs involved in the sulfur amino acid pathway (Met4, Met31, and Met32). Figure 7 shows the number of motif occurrences in each of the top 59 promoters that were ranked according to Met32 binding signal (data from [25]). The motifs are highly frequent in the top 18 promoters, after which a strong drop in motif frequency is observed. DRIM identifies this, and partitions the set accordingly. In comparison, relying on the standard cutoff of 10−3 results in a target set of the top 48 promoters, most of which do not contain this motif. The signal-to-noise ratio is thus diminished, which may explain why these motifs were previously overlooked.
While the flexible cutoff is advantageous in many instances, it also introduces a multiple testing problem. To circumvent this (without resorting to strict multiple testing corrections that may mask the biological signal), we developed an efficient algorithm for computing the exact p-value of a given mHG score. This addresses challenge (c2). Another advantage of this exact statistical score is its straightforward biological interpretation: the mHG p-value reflects the probability of seeing the observed density of motif occurrences at the top of the ranked list under the null assumption that all configurations of motif occurrences are equiprobable.
Motif multiplicity is often indicative of biological function. It is therefore paramount to incorporate this type of information into the motif prediction model. We do so in a data-driven manner by developing the multi-mHG framework, thus addressing challenge (c3). The advantages of the multi-mHG model over the binary model are presented in Results, Binary versus multidimensional enrichment.
False prediction of motifs in randomly generated data is often mentioned as one of the drawbacks of computational motif discovery [25]. We report the testing of DRIM on random permutations of ranked sequences. When tested on more than 100,000 motifs, DRIM did not identify any significant motifs, thus addressing challenge (c4). The low false-positive prediction of our method is mainly attributed to the fact that it is based on rigorous statistics and relies on an exact p-value.
Another important issue that still requires consideration is the characterization of the motif search space. In this study we performed an exhaustive scanning of a restricted motif space (containing ∼105 motifs) followed by a heuristic search for larger motifs. However, the motif search space can be further extended to include motifs that are longer, “fuzzier,” or more complex. Additional considerations such as the distance of the motif from the transcription start site may be taken into account as well as logical relations between different motifs (e.g., “OR,” “AND” operations). It is clear that many of these features are required to correctly model complex regulation patterns that are observed in higher eukaryotes. Two inherent limitations need to be considered when extending the search space: first, as the size of the motif search space increases, the problem of efficiently searching the defined space becomes more acute in terms of running time. Second, since the size of the search space is virtually endless, the problem of multiple testing rapidly erodes the signal-to-noise ratio, requiring an appropriate refinement of the statistical models.
To test our method, we constructed a dataset containing ChIP–chip experiments of 203 putative TFs in Saccharomyces cerevisiae [25,28]. Surprisingly, we discovered a significant length bias in roughly one-third of these experiments. One possible explanation for this phenomenon is nonspecific binding between TFs and DNA, which causes longer sequences to bind more TFs. This explanation is also consistent with the “TF sliding hypothesis” [48]. Why only some TFs exhibit this length bias binding tendency remains an open question. To avoid false positives due to this phenomenon, we opted to filter out all ChIP–chip experiments that had significant length bias. Future work should address this point and focus on developing statistics that are insensitive to this type of bias.
We analyzed the filtered dataset using DRIM and report novel putative TFBS motifs. Additional evidence that indicates the newly discovered motifs are biologically functional was also presented. One interesting finding is that the Aro80 motif we identified, which exists only in seven copies throughout the entire yeast genome, resides in Aro80′s own promoter. This finding suggests that Aro80 regulates its own transcription by binding to its own promoter. Additionally, three GATA binding sites that reside in the Aro80 promoter adjacent to the motif occurrence lead us to speculate that Aro80′s putative self binding is inhibited by competing GATA binding factors (for details see Figure 4B).
Another interesting observation is the CA repeat motifs, which we identified in seven different yeast TFs as well as in human DNA methylation. This type of low complexity motifs have so far been mostly ignored or filtered out by other computational methods. By contrast there is no need to resort to this type of artificial filtering when using the mHG statistics. Our findings in yeast suggest that for certain TFs there is a significant correlation between a sequence's capacity to bind a TF and the presence of a CA repeat in the sequence. This supports a previous hypothesis that CA repeats alter the structure of DNA and thus contribute to TF binding [34]. Our findings constitute concrete evidence of this phenomenon and suggest it may be more frequent than previously appreciated.
We also applied DRIM to high-throughput measurements of methylated CpG islands [2] in human cancer cells, in order to try to identify motifs that are enriched in hypermethylated regions. Interestingly, we identified GA and CA repeat elements as highly enriched in methylated CpG regions of four different cancer cell lines. This is in agreement with previous studies of CpG methylated regions in other cell lines [39,40]. It is interesting to ask whether these repeat elements play some active role in CpG methylation. In [40] the authors give statistical argumentation against such a hypothesis. Instead, they hypothesize that CA (or TG) repeats are caused by an increased mutation rate of methylated CpGs that are deaminated into TpGs. Even if true, this still does not explain the enrichment of the GA repeats. Further experimental and bioinformatic interrogation of this point is therefore called upon.
Overall, DRIM discovered ten novel motifs in methylated CpG regions. Strikingly, nine of them are similar to DNA sequence elements that bind the Polycomb complex in Drosophila and/or human [41,42, 44]. The Polycomb complex is involved in epigenetic silencing via histone methylation. The suggested link between histone methylation and CpG methylation is in agreement with recent work that demonstrated the EZH2 protein interacts with DNA methyltransferases via the Polycomb complex [45]. We also note that the DNA sequence motifs of the two pathways were conserved in Drosophila and human, which is complementary to the observation that the Polycomb proteins are evolutionarily conserved [44,49]. Many of the motifs we found in the CpG methylation data are similar across different types of cancer cell lines. This may suggest that the CpG methylation mechanism is orchestrated by DNA binding factors that are similar in different types of cancer cell lines.
Perhaps the most important conclusion that can be drawn from this study is that looking at biological sequence data in a ranked manner rather than using an arbitrary fixed cutoff to partition the data enables the detection of biological signals that are otherwise overlooked. This suggests that other motif detection methods that rely on fixed cutoffs may benefit from dynamic partitioning. While the effectiveness of our approach was demonstrated on ChIP–chip and methylation data, it can also be applied to a wide range of other data types such as expression data or GO analysis. The DRIM application is publicly available at http://bioinfo.cs.technion.ac.il/drim.
In this subsection we introduce the basics of the mHG statistics, and demonstrate how it can be applied in a straightforward manner to eliminate the need for an arbitrary choice of threshold. To explain the biological motivation of mHG, consider the following scenario: suppose we have a set of promoter regions each associated with a measurement, e.g., a TF binding signal as measured by ChIP–chip [1]. We wish to determine whether a particular motif specified in IUPAC notation, say CASGTGW, is likely to be a TFBS motif. We rank the promoters according to their binding signals—strong binding at the top of the list and the weak at the bottom (Figure 1i). Next, we generate a binary occurrence vector with one or zero entries dependent on whether or not the respective promoter contains a copy of the motif (Figure 1ii). For simplicity we ignore cases where a promoter contains multiple copies of the motif (a refined model, which takes motif multiplicity into account, will be discussed later). Motifs that yield binary vectors with a high density of 1′s at the top of the list are good candidates for being TFBS.
Let us assume for the moment that we know the correct physical-based cutoff on the TF binding signal. The data could then be separated into “strong binding promoters” (i.e., the target set) and “weak binding promoters” (i.e., the background set). We are now interested to know whether there is a particular motif for which the target set contains significantly more motif occurrences than the background set. Let N be the total number of promoters B of which contain the motif, and n the size of the target set. Let X be a random variable describing the number of motif occurrences in the target set. Assuming a uniform distribution over all occurrence vectors with these characteristics, the probability of finding exactly b occurrences in the target set has a hypergeometric distribution, namely:
The tail probability of finding b or more occurrences in the target set is:
As we don't really always have a strict definition of the target set, we employ a strategy that seeks a partition for which the motif enrichment is the most significant, and compute the enrichment under that particular partition. Formally, consider a set of ranked elements and some binary labeling of the set λ = λ1,…,λN ∈ {0,1}N. The binary labels represent the attribute (e.g., motif occurrence). The mHG score is defined as:
where bn(λ) =
. In words, the mHG score reflects the surprise of seeing the observed density of 1's at the top of the list under the null assumption that all configurations of 1's in the vector are equiprobable. The cutoff between the top of the list and the rest of the list is chosen in a data-driven manner so as to maximize the enrichment (Figure 1iii). We discuss other variants of the mHG score in Texts S2 and S3.
The mHG flexible choice of cutoff introduces a multiple testing complication and therefore gives rise to the need for computing the exact p-value. In Text S1 and Figure S2 we demonstrate several bounds for mHG p-values. These bounds may be used for rapid assessment of the p-value of a given mHG score, which can be instrumental in improving algorithmic efficiency. In this section, we describe a novel dynamic programming algorithm for calculating the exact p-value of a given mHG score. This approach is related to a previously described approach for calculating exact p-values of other combinatorial scores ([50,51], with details in [52]).
As noted in the previous section, the mHG score depends solely on the content of the label vector λ. Set N and B, and consider the space of all binary label vectors with B 1′s and N−B 0′s: Λ = {0,1}(N−B,B). Assume that we are given a vector λ0∈Λ, for which we calculate the mHG score mHG(λ0) = p. We would like to determine pval(p) = Prob(mHG(λ) ≤ p) under a uniform distribution of vectors in Λ. Given an mHG score p, we do this by means of path counting. The space of all label vectors Λ = {0,1}(N−B,B) is represented as a two-dimensional grid ranging from (0,0) at the bottom left to (N,B) at the top right. Each specific label vector λ∈Λ is represented by a path (0,0) → (N,B) composed of N distinct steps. The ith step in the path describing a vector λ is (1,0) if λi = 0 and (1,1) if λi = 1 (see Figure 8). Each point (n,b) on the grid corresponds to a threshold (on ranks) n, and the respective value b = bn(1). It can therefore be associated with a specific HGT score: HGTn(λ) = HGT (bn(λ );N,B,n). A subset of the points on the grid can be characterized as those points (n,b) for which HGT (b;N,B,n) ≤ p. We denote this subset R = R(p) (see Figure 8).
The (0,0) → (N,B) path representing λ visits N distinct grid points (excluding the point (0,0)), representing the N different HGT scores that are considered when calculating its mHG score: mHG(λ) = min1≤n<NHGTn(λ). mHG(λ) ≤ p if the path representing λ visits R. Denote by Π(n,b) the total number of paths (0,0) → (n,b) and by ΠR(n,b) the number of paths (0,0) →(n,b) not visiting R. We then have:
We calculate ΠR(n,b) by means of dynamic programming. Initially, set ΠR (0,0) = 1 and ΠR(n,b) = 0 for b = −1 and along the diagonal b = n + 1, 0 ≤ n ≤ B. Then, for each 1 ≤ n ≤ N, and max(0,B − N + n) ≤ b ≤ min(B,n) calculate ΠR(n,b) using the formula:
In total, we perform a O(N2) routine in order to calculate ΠR(N,B) for a given score p. Trivially, we have Π(N, B) =
and pval(p) may be directly computed from Equation 4.
So far we have dealt with enrichment of binary attributes, in which a one or zero indicated whether or not the attribute appeared. There are cases where one would like to associate a number with an attribute. We revisit the scenario we described in previous sections in which we tried to determine whether a particular motif is likely to be a TFBS motif. The promoters were ranked according to their binding signals, and the corresponding binary occurrence vector was generated. Notice that some promoters may contain several copies of a particular motif. Clearly, this information is valuable and should be incorporated in the enrichment analysis. How exactly to incorporate this information is not clear. For example, consider two motif occurrence vectors generated for two different motifs, where the top ten entries of the vectors are all 1's and all 2's, respectively. Is the second motif more enriched than the first? Clearly, this depends on the rarity of double motif occurrences compared with single occurrences in the corresponding vectors. If the frequency of 2's is lower than that of 1's, then the second motif is more significant. However, if they are equally frequent (this is often the case for degenerate motifs such as poly A's) then both motifs are equally enriched.
To quantitatively capture this notion and address motif multiplicity in a data-driven manner, we propose a multidimensional hypergeometric model, which extends the previously defined framework for enrichment analysis to nonbinary label vectors. Formally, let λ be a uniformly drawn label vector λ = λ1,…,λN ∈ {0…k}N containing B1 1's, B2 2's … Bk k's and
. We would like to test for enrichment of 1's, 2's...k's at the top of λ. We define the multidimensional hypergeometric score (multiHG) for a set S of size N consisting of k + 1 subsets S0, S1, S2 … , Sk of respective sizes N – (B1 + B2 + …Bk), B1, B2…, Bk. Given a subset S′ ⊂ S of size n, the probability of finding exactly b1 elements of S1 and b2 elements of S2…, bk elements of Sk within S′ is:
Let X1,…Xk be random variables describing the number of 1′s, … ,k's, respectively, at the top n positions of λ. The multihypergeometric tail probability (multiHGT) of seeing at least b1 1's, at least b2 2's,…, and at least bk k's at the top n positions of the vector is:
The definition of the mHG score can now be extended to the minimum of the set of multiHGTs calculated on all prefixes of λ.
where bj(n, λ) =
. Exact p-values for the multidimensional mHG, under a uniform null distribution, can be computed in a k-dimensional space using a path enumeration strategy similar to the one we used in the binary case. The details on how to compute this p-value in a three-dimensional space are explained in Text S4.
The software tool DRIM implements the mHG framework for motif identification in ranked DNA sequences. A flow chart of DRIM is provided in Figure 1. In the rest of this section we describe the details of this implementation.
Exhaustive search of the restricted motif space. Ideally we would like to exhaustively search through the space of all biologically viable motifs and identify those that are significantly enriched at the top of the ranked list. However, this is infeasible in terms of running time (the space of viable TF binding sites includes motifs of size up to 20, i.e., 1520 k-mers). We therefore resort to a simple strategy where the motif search is broken into two stages: first an exhaustive search on a restricted motif space is performed. The “motif seeds” that are identified in the preliminary search are used as a starting point for a heuristic search of larger motifs in the entire motif space. The restricted motif space S used in this study is the union of two subspaces S1 and S2: S1 = {A,C,G,T,R,W,Y,S,N}7, where the IUPAC degenerate symbols (i.e., R,Y,W,S,N) are restricted to a maximum degeneracy of 2 and S2 = {A,C,G,T}3N3−25{A,C,G,T}3. The rationale behind the usage of the restricted IUPAC alphabet in S1 instead of the complete 15 symbol alphabet stems from DNA–TF physical interaction properties and TFBS database statistics as explained in previous work [53]. S2 captures motifs that contain a fixed gap (different motifs can have different gap sizes), which is characteristic of some TFs such as Zinc fingers).
mHG enrichment. For each of the motifs in S, we generate a ranked occurrence vector and compute the enrichment in terms of the multidimensional mHG. Due to running time considerations, we restrict the multidimensional mHG to three dimensions. This means that the model assumes each intergenic region contains either 0, 1, or ≥2 copies of a motif. To test whether this assumption is reasonable in the case of true TFBS motifs, we examined the occurrence distribution of TFBS motifs that were experimentally verified in S. cerevisiae (see Figure 9). It can be seen that the assumption holds for the five TFs that were tested since the majority of all intergenic regions contained either zero, one, or two copies of the TFBS. At the end of this stage, only motif seeds with mHG score <10−3 are kept. Similar motifs are filtered (as explained in Texts S5 and S6), and the remaining motif seeds are fed into the heuristic search module for expansion, Figure 1iii–1iv.
Motif expansion by heuristic search. The filtered motif seeds are used as starting points for identifying larger motifs that do not reside in the restricted motif space. This is done through an iterative heuristic process that employs simulated annealing. The objective function is to minimize the motif mHG p-value. We tested two different strategies for determining valid moves in the motif space. In the first, we defined a transition from motif M1 to M2 as valid if M1 and M2 are within a predefined Hamming distance D, with all valid moves being equiprobable. Additional bases can also be added to the motif flanks, thus enabling motif expansion. Note that the mHG adaptive cutoff is recalculated at each step. In the second strategy, all the motif occurrences in the target set that are within Hamming distance D are aligned. A consensus motif above IUPAC is extracted and the algorithm attempts a transition to that motif. While the second strategy converges much faster than the first, it is also more prone to converge to local minima (in the final application we use the second strategy with D = 1). At the end of the process, the exact p-value of each of the expanded motifs is computed. To correct for multiple motif testing, the p-value is then multiplied by the motif space size. Only motifs with corrected p-value <10−3 are reported.
Optimizations and running time. The DRIM application was implemented in C++. A “blind search” requires ∼100,000 motifs to be checked for enrichment in each run. It is therefore paramount to optimize the above-described procedures to enable a feasible running time. There are two bottlenecks in terms of running time: the motif occurrence vector generation and the mHG computation. We developed several optimization schemes to improve both. In the final configuration, the running time on a list of 6,000 sequences with an average size of 480 bases took ∼3 minutes on a Pentium IV, 2 GHz.
ChIP–chip dataset. A number of assays have been recently developed that use immunopercipitation-based enrichment of cellular DNA for the purpose of identifying binding or other chemical events and the genomic locations at which they occur. Location analysis, also known as ChIP–chip, is a technique that enables the mapping of transcription binding events to genomic locations at which they occur [1,54]. The output of the assay is a fluorescence dye ratio at each spot of the array. If spots are taken to represent genomic regions, then we can regard the ratio and p-value associated with each spot as an indication of TF binding in the corresponding genomic region. We applied DRIM to S. cerevisiae genome-wide location data reported in Harbison et al. [25] and Lee et al. [28]. The first consists of the genomic occupancy of 203 putative TFs in rich media conditions (YPD). In addition, the genomic occupancy of 84 of these TFs was measured in at least one other condition (OC). In each of the experiments, the genomic sequences were ranked according to the TF binding p-value. Surprisingly, we observed that 69 of the 203 ranked sequence lists of YPD had significantly longer sequences at the top of the list (first 300 sequences) compared with the rest of the list with t-test p-value ≤ 10−3. We observed a similar phenomenon in 76 of the 148 ranked sequence lists of OC experiments (see Figure S1). In other words, for some TFs, longer sequences are biased toward stronger binding signals. This observation is unexpected since, although longer probes hybridize more labeled material than shorter probes, the increase should be proportional in both channels. This type of length bias may cause spurious results under our model assumptions and hence the final dataset, termed “Harbison filtered dataset,” refers to the remaining 207 experiments (135 YPD, and 72 OC) of 162 unique TFs that did not have length bias (Table S1).
An additional ChIP–chip dataset was constructed using the data reported in Lee et al. [28] containing 113 experiments in rich media. The data is partially exclusive to the data of Harbison et al. [25]. The same filtering procedure was performed, resulting in a set of 65 experiments, termed “Lee filtered dataset.”
Methylated CpG dataset. Using a technique similar to ChIP–chip, termed methyl-DNA immunoprecipitation (mDIP), enables the measurement of methylated CpG island patterns [2,55]. The third dataset contains the CpG island methylation patterns of four different human cancer cell lines (Caco-2, Polyp, Carcinoma, PC3) where several replicate experiments were done for each of the cell lines. In each of these experiments, the CpG methylation signal was measured in ∼13,000 gene promoters as reported in [2].
Accession numbers for the genes discussed in the paper are given in Table S5. |
10.1371/journal.ppat.1006114 | Experimental Estimation of the Effects of All Amino-Acid Mutations to HIV’s Envelope Protein on Viral Replication in Cell Culture | HIV is notorious for its capacity to evade immunity and anti-viral drugs through rapid sequence evolution. Knowledge of the functional effects of mutations to HIV is critical for understanding this evolution. HIV’s most rapidly evolving protein is its envelope (Env). Here we use deep mutational scanning to experimentally estimate the effects of all amino-acid mutations to Env on viral replication in cell culture. Most mutations are under purifying selection in our experiments, although a few sites experience strong selection for mutations that enhance HIV’s replication in cell culture. We compare our experimental measurements of each site’s preference for each amino acid to the actual frequencies of these amino acids in naturally occurring HIV sequences. Our measured amino-acid preferences correlate with amino-acid frequencies in natural sequences for most sites. However, our measured preferences are less concordant with natural amino-acid frequencies at surface-exposed sites that are subject to pressures absent from our experiments such as antibody selection. Our data enable us to quantify the inherent mutational tolerance of each site in Env. We show that the epitopes of broadly neutralizing antibodies have a significantly reduced inherent capacity to tolerate mutations, rigorously validating a pervasive idea in the field. Overall, our results help disentangle the role of inherent functional constraints and external selection pressures in shaping Env’s evolution.
| HIV is infamous for the rapid evolution of its surface protein, Env. The ability to measure the effects of all mutations to Env under defined selection pressures in the lab would open the door to better understanding the factors that shape this evolution. However, this is a daunting experimental task since there are over 104 different single-amino acid mutations to Env. Here we leverage next-generation sequencing to perform a single massively parallel experiment that estimates the effects of all these mutations on viral replication in cell culture. Our measurements are largely consistent with existing knowledge about the effects of mutations at functionally important sites, and show that inherent mutational tolerance varies widely across Env. Our work provides new insight into Env’s evolution, and describes a powerful experimental approach for measuring the effects of mutations on HIV phenotypes that can be selected for in the lab.
| HIV evolves rapidly: the envelope (Env) proteins of two viral strains within a single infected host diverge as much in a year as the typical human and chimpanzee ortholog has diverged over ∼5-million years [1–4]. This rapid evolution is central to HIV’s biology. Most humans infected with HIV generate antibodies against Env that effectively neutralize viruses from early in the infection [5–7]. However, Env evolves so rapidly that HIV is able to stay ahead of this antibody response, with new viral variants escaping from antibodies that neutralized their predecessors just months before [5–7]. Env’s exceptional evolutionary capacity is therefore essential for the maintenance of HIV in the human population.
A protein’s evolutionary capacity depends on its ability to tolerate point mutations. Detailed knowledge of how mutations affect Env is therefore key to understanding its evolution. Many studies have estimated the effects of mutations to Env. One strategy is experimental: numerous studies have used site-directed mutagenesis or alanine scanning to measure how specific mutations affect various aspects of Env’s function [8–17]. However, these experiments have examined only a small fraction of the many possible mutations to Env. Another strategy is computational: under certain assumptions, the fitness effects of mutations can be estimated from their frequencies in global or intra-patient HIV sequences [18–22]. However, these computational strategies are of uncertain accuracy and cannot separate the contributions of inherent functional constraints from those of external selection pressures such as antibodies. Therefore, a more complete and direct delineation of how every mutation affects Env’s function would be of great value.
It is now possible to make massively parallel experimental measurements of the effects of protein mutations using deep mutational scanning [23–25]. These experiments involve creating large libraries of mutants of a gene, subjecting them to bulk functional selections, and quantifying the effect of each mutation by using deep sequencing to assess its frequency pre- and post-selection. Over the last few years, deep mutational scanning has been used to estimate the effects of all single amino-acid mutations to a variety of proteins or protein domains [26–39], as well as to estimate the effects of a fraction of the amino-acid mutations to many additional proteins (e.g., [40–42]). When these experiments examine all amino-acid mutations, they can be used to compute the mutational tolerance of each protein site, thereby shedding light on a protein’s inherent evolutionary capacity. Recently, deep mutational scanning has been used to examine the effects of amino-acid mutations on the binding of antibodies to Env protein displayed on mammalian or yeast cells [43, 44], or the effects of single-nucleotide mutations scattered across the HIV genome on viral replication in cell culture [45]. However, none of these studies comprehensively measure the effects of all Env amino-acid mutations on viral replication. Therefore, we currently lack comprehensive measurements of the site-specific mutational tolerance of Env.
Here we use deep mutational scanning to experimentally estimate how all amino-acid mutations to the ectodomain and transmembrane domain of Env affect viral replication in cell culture. At most sites, our measurements correlate with the frequencies of amino acids in natural HIV sequences. However, there are large deviations at sites where natural evolution is strongly shaped by factors (e.g., antibodies) that are absent from our experiments. Our results also show that site-to-site variation in Env’s inherent capacity to tolerate mutations helps explain why epitopes of broadly neutralizing antibodies are highly conserved in natural isolates. Overall, our work helps elucidate how inherent functional constraints and external selective pressures combine to shape Env’s evolution, and demonstrates a powerful experimental approach for comprehensively mapping how mutations affect HIV phenotypes that can be selected for in the lab.
We used the deep mutational scanning approach in Fig 1A to estimate the effects of all single amino-acid mutations to Env. We applied this approach to Env from the LAI strain of HIV [46]. LAI is a CXCR4-tropic subtype B virus isolated from a chronically infected individual and then passaged in human T-lymphocytes. We chose this strain because LAI and the closely related HXB2 strain have been widely used to study Env’s structure and function [8–11, 47–49], providing extensive biochemical data with which to benchmark our results. LAI’s Env is 861 amino acids in length. We mutagenized amino acids 31–702 (throughout this paper, we use the HXB2 numbering scheme [50]). We excluded the N-terminal signal peptide and the C-terminal cytoplasmic tail, since mutations in these regions can alter Env expression in ways that affect viral infectivity in cell culture [51–53]. The region of Env that we mutagenized spanned 677 residues, meaning that there are 677 × 63 = 42,651 possible codon mutations, corresponding to 677 × 19 = 12,863 possible amino-acid mutations.
To create plasmid libraries containing all these mutations, we used a previously described PCR mutagenesis technique [31] that creates multi-nucleotide (e.g, gca→CAT) as well as single-nucleotide (e.g, gca→gAa) codon mutations. We created three independent plasmid libraries, and carried each library through all subsequent steps independently, meaning that all our measurements were made in true biological triplicate (Fig 1B). We Sanger sequenced 26 clones to estimate the frequency of mutations in the plasmid mutant libraries (S1 Fig). There were an average of 1.4 codon mutations per clone, with the number of mutations per clone roughly following a Poisson distribution. The deep sequencing described in the next section found that at least 79% of the ≈104 possible amino-acid mutations were observed at least three times in each of the triplicate libraries, and that 98% of mutations were observed at least three times across all three libraries combined. The plasmid libraries therefore sampled most amino-acid mutations to Env.
We produced virus libraries by transfecting each plasmid library into 293T cells. The viruses in the resulting transfection supernatant lack a genotype-phenotype link, since each cell is transfected by many plasmids. We therefore passaged the transfection supernatants twice in SupT1 cells at an MOI of 0.005 to create a genotype-phenotype link and select for functional variants. Importantly, neither 293T nor SupT1 cells express detectable levels of APOBEC3G [54, 55], which can hypermutate HIV genomes [56, 57]. This is a crucial point: although HIV encodes a protein that counteracts APOBEC3G, a fraction of viruses will lack a functional version of this protein and so have their genomes hypermutated in APOBEC3G-expressing cells. For each library, we passaged 5 × 105 infectious particles in order to maintain library diversity. We used Illumina deep sequencing to quantify the frequency of each mutation before and after passaging. In order to increase the sequencing accuracy, we attached unique molecular barcodes or “Primer IDs” to each PCR amplicon [58–61]. We sequenced the plasmids to assess the initial mutation frequencies, and sequenced non-integrated viral DNA [62] from infected SupT1 cells to assess the mutation frequencies in the viruses. A concern is that errors from sequencing and viral replication (e.g., from viral reverse transcriptase) would introduce bias. To address this concern, we paired each mutant library with a control in which we generated wildtype virus from unmutated plasmid. Sequencing the control plasmids and viruses enabled us to estimate and statistically correct for the rates of these errors (S2 Fig). Overall, these procedures allowed us to implement the deep mutational scanning workflow in Fig 1.
Our deep mutational scanning experiments require that selection purge the virus libraries of non-functional variants. As an initial gene-wide measure of selection, we analyzed how different types of codon mutations (nonsynonymous, synonymous, and stop-codon mutations) changed in frequency after selection. In these analyses, we corrected for background errors from PCR, sequencing, and viral replication by subtracting the mutation frequencies measured in our wildtype controls from those measured in the mutant libraries (S2 Fig).
Stop-codon mutations are expected to be uniformly deleterious. Indeed, after correcting for background errors, stop codons were purged to <1% of their initial frequency in the twice-passaged viruses for each replicate, indicating strong purifying selection (see the data for “all sites” in Fig 2A). The second viral passage is important for complete selection, as stop codons remain at about ≈16% of their initial frequency in viruses that were only been passaged once (S3 Fig).
Interpreting the frequencies of nonsynonymous mutations is more nuanced, as different amino-acid mutations have different functional effects. However, a large fraction of amino-acid mutations are deleterious to any protein [63–65]. Therefore, one might expect that the frequency of nonsynonymous mutations would decrease substantially in the twice-passaged mutant viruses. But surprisingly, even after correcting for background errors, the average frequency of nonsynonymous mutations in the passaged viruses is ≈90% of its value in the mutant plasmids (see the data for “all sites” in Fig 2A). However, the average masks two disparate trends. In each library, a few sites exhibit large increases in the frequency of nonsynonymous mutations, whereas this frequency decreases by nearly two-fold for all other sites (see the data for the subgroups of sites in Fig 2A).
An obvious hypothesis is that at a few sites, amino-acid mutations are favored because they are adaptive for viral replication in cell culture. Consistent with this hypothesis, the sites that experienced large increases in mutation frequencies are similar among the three replicates (Fig 2B), suggestive of reproducible selection for mutations at these sites. Moreover, these sites are spatially clustered in Env’s crystal structure in regions where mutations are likely to enhance viral replication in cell culture (Fig 3 and S1 Table). One cluster of mutations disrupts potential glycosylation sites at the trimer apex (Fig 3A). This result suggests that some of the glycans that help shield Env from antibodies in nature [6, 66] actually decrease viral fitness in the absence of immune selection. This idea is consistent with previous studies showing that that loss of glycosylation sites can enhance viral infectivity in cell culture [67–69]. A second cluster overlaps sites where mutations influence Env’s conformational dynamics, which are commonly altered by cell-culture passage [70, 71]. It has been hypothesized that neutralization-resistant Envs primarily assume conformations that mask conserved antibody epitopes, while lab-adapted variants more efficiently sample different conformations associated with CD4 binding [72]. Thus, the adaptive mutations we observe may enable Env to more efficiently use CD4 in cell culture, but would not be selected in nature because they expose conserved epitopes. A third cluster is at the co-receptor binding interface (Fig 3B), where mutations may enhance viral entry in cell culture. Therefore, while most of Env is under purifying selection against changes to the protein sequence, a few sites are under selection for cell-culture adapting amino-acid mutations.
If our experiments are indeed identifying mutations to LAI that are beneficial in cell culture, then one expectation is that some of these mutations might fix after prolonged passage of LAI in cell culture. Interestingly, almost exactly such an experiment was performed in the early study of HIV. The LAI strain used in our study was initially isolated from a chronically infected individual and then passaged in cell culture for a short period of time before cloning [46, 77]. HXB2, another common lab strain, is derived from a variant of LAI that was repeatedly passaged in a variety of cell lines, initially as a contaminant of other viral stocks [78, 79]. There are 23 amino-acid differences between the Env proteins of LAI and HXB2. Although the predecessor for HXB2 was not passaged in the same SupT1 cell line that we used, if its passage in other cell lines led to mutations that were generally adaptive to cell culture, then we would expect them to introduce amino acids in HXB2 that are also selected in our deep mutational scan of LAI. Indeed, we found that most differences between LAI and HXB2 introduced mutations to amino acids that our experiments suggest are more preferred in cell culture than the wildtype LAI amino acid (S2 Table). Thus, our results are consistent with the expectation that HXB2 is more adapted to cell culture than LAI.
The average error-corrected frequency of synonymous mutations changes little after selection (an average decrease to 96% of the original frequency; see the data for “all sites” in Fig 2A). This overall trend is consistent with the fact that synonymous mutations usually have smaller functional effects than nonsynonymous mutations. However, synonymous mutations can sometimes have substantial effects [21, 80–82], particularly in viruses like HIV that are under strong selection for RNA secondary structure and codon usage [83, 84]. To assess selection on synonymous mutations on a more site-specific level, we examined the change in frequency of multi-nucleotide codon mutations across env’s primary sequence (Fig 4). The rationale behind examining only multi-nucleotide codon mutations is that they are not appreciably confounded by errors from PCR, deep sequencing, or de novo mutations from viral replication (S2 and S4 Figs). In a region roughly spanning codons 500 to 600, selection strongly purged both synonymous and nonsynonymous multi-nucleotide codon mutations (Fig 4). This region contains env’s Rev-response element (RRE) [85], a highly structured region of RNA that is bound by the Rev protein to control the temporal export of unspliced HIV transcripts from the nucleus [86, 87]. The finding of strong selection on the nucleotide as well as the amino-acid sequence of the RRE region of Env therefore agrees with our biological expectations.
The previous section examined broad trends in selection averaged across many sites. But our data also enable much more fine-grained estimates of the preference for every amino-acid at every position in Env. We define a site’s preference for an amino acid to be proportional to the enrichment or depletion of that amino acid after selection (correcting for the error rates determined using the wildtype controls), normalizing the preferences for each site so that they sum to one. We denote the preference of site r for amino acid a as πr,a, and compute the preferences from the deep-sequencing data as described in [88]. Since we mutagenized 677 residues in Env, there are 677 × 20 = 13,540 preferences. If selection in our experiments exactly parallels selection in nature and there are no shifts in mutational effects as Env evolves, then these preferences are the expected frequencies of each amino acid at each site in an alignment of Env sequences that have reached evolutionary equilibrium under a mutation process that introduces each amino acid with equal probability [31, 89].
Fig 5 shows Env’s site-specific amino-acid preferences after averaging across replicates and re-scaling to account for the stringency of selection in our experiments (details of this re-scaling are in the next section). As is immediately obvious from Fig 5, sites vary dramatically in their tolerance for mutations. Some sites strongly prefer a single amino acid, while other sites can tolerate many amino acids. For instance, site 457, an important receptor-binding residue [8], has a strong preference for aspartic acid. However, this site is adjacent to a variable loop (sites 460–469) where most sites tolerate many amino acids. Another general observation is that when sites tolerate multiple amino acids, they often prefer ones with similar chemical properties. For instance, sites 225 and 226 prefer hydrophobic amino acids, while sites 162 to 164 prefer positively charged amino acids.
To confirm that our experiments captured known constraints on Env’s function, we examined mutations that have been characterized to affect key functions of Env. Table 1 lists mutations known to disrupt an essential disulfide bond, binding to receptor or co-receptor, or protease cleavage. In almost all cases, the deleterious mutation introduces an amino-acid that our experiments report as having a markedly lower preference than the wildtype amino acid. Therefore, our measurements largely concord with existing knowledge about mutations that affect key aspects of Env’s function.
A crucial aspect of any high-throughput experiment is assessing the reproducibility of independent replicates. Fig 5 shows the average of the preferences measured in each replicate. Fig 6A shows the correlations among the 13,540 site-specific amino-acid preferences estimated from each of the three replicates. The correlations are modest, indicating substantial replicate-to-replicate noise. In principle, this noise could arise from differences in the initial plasmid mutant libraries, bottlenecks during the generation of viruses by transfection, bottlenecks during viral passaging, or bottlenecks during the sequencing of proviral DNA from infected cells. Analysis of technical replicates of the first or second round of viral passaging indicates that most of the noise arises from bottlenecks during the viral passaging or sequencing steps. Specifically, measurements from replicate 3 are no more correlated to those from replicates 3b-1 or 3b-2 (which are repeated passages of the same transfection supernatant, Fig 1B) than they are to those from totally independent replicates (compare Fig 6 and S6 Fig). However, replicates 3b-1 and 3b-2 (which shared the first of the two viral passages, Fig 1) do yield more correlated measurements than independent replicates (S6 Fig). The existence of bottlenecks during viral passage is also suggested by the data in S4 and S5 Figs. Therefore, the experimental reproducibility could probably be increased by passaging more infectious viruses at each step.
If bottlenecks cause each replicate to sample slightly different mutations, then perhaps the total number of tolerated mutations per site will be similar between replicates, even if the exact mutations differ. To test this hypothesis, we computed the effective number of amino acids tolerated at each site as the exponential of the Shannon entropy of the site’s amino-acid preferences. Fig 6B shows that the effective number of amino acids tolerated at each site is more correlated between replicates than the preferences themselves. We further reasoned that even if bottlenecking causes slight variations in the preferred amino acids between replicates, each site would still tend to prefer amino acids with similar chemical characteristics. To test this hypothesis, we quantified the extent that each site preferred hydrophobic or hydrophilic amino acids by computing a site-specific hydrophobicity score from the amino-acid preferences. Fig 6C shows that these preference-weighted hydrophobicities are more correlated between replicates than the preferences. Therefore, even though there is replicate-to-replicate noise in the exact amino acids preferred at a site, the effective number of tolerated amino acids and the chemical properties of these amino acids are similar among replicates.
In the previous section, we showed that our experimentally measured amino-acid preferences captured the constraints on Env’s biological function for sites with known mutational effects (Table 1). If this is true across the entire protein, then our measurements should correlate with the frequencies of amino acids in natural HIV sequences. Table 2 shows that there is a modest correlation (Pearson’s R ranging from 0.29 to 0.36) between the preferences from each experimental replicate and the frequencies in an alignment of HIV-1 group-M sequences (a phylogenetic tree of these sequences is in Fig 7A; sites in Env variable loops that can not be reliably aligned are excluded as described in the Methods). Since each replicate suffers from noise due to partial bottlenecking of the viral diversity, we hypothesized that averaging the preferences across replicates should make them more accurate. Indeed, averaging the replicates increased the correlation to R = 0.4 (Table 2).
The concordance between deep mutational scanning measurements and natural sequence variation is improved by accounting for differences in the stringency of selection in the experiments compared to natural selection [89, 91]. Specifically, if the measured preference is πr,a and the stringency parameter is β, then the re-scaled preference is (πr,a)β/[ ∑a′(πr,a′)β ]. A stringency parameter of β > 1 means that natural evolution favors the same amino acids as the experiments, but with greater stringency. Table 2 shows that for all replicates, the stringency parameter that maximizes the correlation is >1. Therefore, natural selection prefers the same amino acids as our experiments, but with greater stringency.
After averaging across replicates and re-scaling by the optimal stringency parameter, the Pearson correlation is 0.44 between our experimentally measured preferences and amino-acid frequencies in the alignment of naturally occurring HIV sequences (Fig 7B). Is this a good correlation? At first glance, a correlation of 0.44 seems unimpressive. But we do not expect a perfect correlation even if the experiments perfectly concord with selection on Env in nature. There are several factors that are expected to reduce the correlation between the experimentally measured preferences and amino-acid frequencies in natural sequences. First, our experiments examine the effects of mutations to Env from the LAI strain. However, it is well known that epistasis can cause the effects of mutations to differ among homologs of the same protein [92, 93], and many examples of this phenomenon have been documented in HIV Env [94–97]. Therefore, our measurements for the LAI Env are probably not completely generalizable to all other strains. In addition, natural HIV sequences are drawn from a phylogeny (Fig 7A), not an ideal ensemble of all possible Env sequences. The frequencies of amino acids in this phylogeny reflect evolutionary history as well as natural selection. For instance, if several amino acids are equally preferred at a site, one is likely to be more frequent in the alignment due to historical contingency. Additionally, natural evolution is influenced by the genetic code and mutation biases: a mutation from the tryptophan codon TGG to the valine codon GTT is extremely unlikely even if valine is more preferred than tryptophan. Mutation biases inherent in reverse transcription [98] or APOBEC3G-induced hypermutation [54] could also bias some evolutionary outcomes over others. Therefore, the correlation will be imperfect even if the preferences completely concord with natural selection—the question is how the actual correlation compares to what is expected given the phylogenetic history and mutation biases.
To determine the expected correlation if the experimentally measured amino-acid preferences reflect conserved constraints in Env, we simulated evolution along the phylogenetic tree in Fig 7A under the assumption that the experimentally measured preferences exactly match natural selection. Specifically, we used pyvolve [99] to simulate evolution using the experimentally informed site-specific codon substitution models described in [91], which define mutation-fixation probabilities in terms of the amino-acid preferences. In addition to the preferences and the stringency parameter β = 2.1 from Table 2, the substitution models in [91] require specification of parameters reflecting biases in the mutation process. We estimated nucleotide mutation bias parameters of ϕA = 0.55, ϕC = 0.15, ϕG = 0.11, and ϕT = 0.18 from the frequencies at the third-nucleotide codon position in sequences in the group-M alignment for sites where the most common amino acid had 4-fold codon degeneracy. We used the transition-transversion ratio of κ = 4.4 estimated in [100]. For these simulations, we scaled the branch lengths so that the average pairwise protein divergence was the same in the actual and simulated alignments.
The correlation between the preferences and amino-acid frequencies in a representative simulated alignment is shown in Fig 7C. As this plot illustrates, the expected correlation is only about 0.46 if the experimentally measured preferences exactly describe natural selection on Env under our model. The simulated frequencies in Fig 7C show the same pattern of bi-modality (most values near zero or one) as the actual frequencies in Fig 7B despite the fact that the preferences used in the simulations allow multiple amino acids at most sites (see Fig 5). This fact illustrates that bi-modality in the amino-acid frequencies can arise from the historical contingency inherent in a phylogenetic tree even if multiple amino acids are tolerated at most sites. As a control, we also simulated evolution using substitution models in which the preferences have been randomized among sites (Fig 7D); as should be the case, there is no correlation in these control simulations. So the actual correlation is nearly as high as expected if natural selection concords with the preferences measured in our experiment.
We next investigated if there are parts of Env for which there is an especially low correlation between our experimentally measured preferences and natural amino-acid frequencies. For instance, antibodies exert selection on the surface of Env in nature [6, 7, 101, 102]. We therefore examined the actual and simulated correlations between the preferences and frequencies as a function of solvent accessibility (Fig 7E and 7F). For all sites (right side of Fig 7E, left side of Fig 7F), the actual correlation is only slightly lower than the range of correlations in 100 simulations. For more buried sites, both the simulated and actual correlations increase (Fig 7E), presumably because sites in the core of Env tend to have stronger preferences for specific amino acids. But as sites become more surface-exposed, the actual correlation drops below the value expected from the simulations (Fig 7F). Therefore, our experiments provide a relatively worse description of natural selection on Env’s surface than its core—probably because the evolution of the protein’s core is shaped mostly by inherent functional constraints that are effectively captured by our experiments, whereas the surface is subject to selection pressures (e.g., antibodies) that are not modeled in our experiments.
Comparing disulfide-bonded cysteines and glycosylation sites vividly illustrates this dichotomy between inherent functional constraints and external selection pressures. Env has 10 highly conserved disulfide bonds, most of which are essential for the protein’s inherent function [49]. Env also has numerous N-linked glycosylation sites, many of which are also highly conserved in nature, where they help shield the protein from antibodies [6, 66]. In contrast to the disulfides, only some glycosylation sites are important for Env’s function in the absence of immune selection [67, 69]. Fig 8 shows that our experimentally measured preferences are highly correlated with natural amino-acid frequencies at the sites of the disulfides, but not at the glycosylation sites. This result can easily be rationalized: the disulfides are inherently necessary for Env’s function, whereas many glycosylation sites are important largely because of the external selection imposed by antibodies. Our experiments therefore accurately reflect the natural constraints on the former but not the latter.
The fact that we found well-tolerated mutations at all of Env’s glycosylation sites (S7A Fig) might seem surprising given that other studies have shown that some glycosylation sites are important for Env’s function in certain HIV strains [67, 69]. However, these studies were all performed in HIV strains substantially diverged from LAI. A study in HXB2 (which is closely related to LAI) found that individual mutations are at least partially tolerated at all glycosylation sites in Env’s gp120 subunit when assaying for viral infectivity in cell culture [103]. Therefore, glycosylation sites may be especially expendable in the LAI strain used in our study.
Different sites in Env evolve at different rates in natural HIV sequences. For instance, sites on the apical surface of Env evolve especially rapidly [104]. These differences in evolutionary rate arise from two factors. First, some sites are inherently better at tolerating mutations without disrupting Env’s essential functions. Second, some sites are under stronger immune selection for rapid sequence change. However, since Env in nature is under selection both to maintain its function and escape immunity, it is difficult to deconvolve these factors.
Our experiments estimate each site’s inherent tolerance for mutations under selection purely for Env’s function in cell culture, without the confounding effects of immune selection (for the remainder of this section, we define a site’s mutational tolerance as the Shannon entropy of its amino-acid preferences shown in Fig 5). We can therefore assess whether regions of Env that evolve rapidly or slowly in nature also have unusually high or low inherent tolerance to mutations.
We focused on two regions of Env. First, we analyzed portions of the protein classified as “variable loops” due to extensive variation in nature [105, 106]. These loops are frequently targeted by antibodies that drive rapid sequence evolution [102, 107]. Because these loops evolve rapidly, we hypothesized they would have a high inherent mutational tolerance. But an alternative hypothesis is that their rapid evolution more attributable to strong selection from antibodies than an unusually high mutational tolerance. Second, we focused on epitopes of antibodies that broadly neutralize many HIV strains. Because these epitopes are highly conserved in nature and often overlap with regions of known functional constraint [108–113], we hypothesized they would have a low mutational tolerance. However, an alternative hypothesis is that these epitopes evolve slowly not because they are mutationally intolerant but simply because they are under weaker immune selection. Indeed, broad immune responses targeting these epitopes only develop in 20% of infected individuals and generally only after multiple years of infection [114].
In testing these hypotheses, it is important to control for other properties known to affect mutational tolerance. This can be done by using multiple linear regression to simultaneously analyze how several independent variables affect the dependent variable of mutational tolerance. Relative solvent accessibility (RSA) is the strongest determinant of mutational tolerance in proteins [115], so we included RSA as a variable in the regression. The region of env that contains the RRE is under strong nucleotide-level constraint [85–87, Fig 4], so we also included being in the RRE as a binary variable in the regression. We defined the variable loops as indicated in Fig 5, and included being in one of these loops as a binary variable in the regression. Finally, we used crystal structures to delineate broadly neutralizing antibody epitopes. We focused on broadly neutralizing antibodies targeting the CD4 binding site, since most other broadly neutralizing antibodies target either glycans (which are subject to pressures that are not well-modeled in our experiments; Fig 8A) or a membrane-proximal region of gp41 that is not fully resolved in crystal structures of trimeric Env making it impossible to correct for RSA. Specifically, we analyzed the three antibodies with the greatest breadth from [116]: VRC01 (PDB 3NGB [117]), 12A21 (PDB 4JPW [118]), and 3BNC117 (PDB 4JPV [118]). We defined a site as part of an epitope if it was within a 4Å inter-atomic distance of the antibody, and included the number of epitopes in which a site is found as a discrete variable in the regression.
The results of the multiple linear regression are in Table 3. As expected, increased solvent accessibility is strongly associated with increased mutational tolerance, whereas presence in the RRE is strongly associated with decreased mutational tolerance. After correcting for these effects, sites in broadly neutralizing epitopes have significantly reduced mutational tolerance. In contrast, sites in the variable loops have higher mutational tolerance, but this effect is not statistically significant. Some of the loops are more variable in nature than others [119]. However, even when the loops are considered independently, none of these regions has a statistically significant association with mutational tolerance (S3 Table). Overall, this analysis provides statistical confirmation of something that is widely assumed in the study of HIV: broadly neutralizing antibodies are unique because they target regions of Env that are inherently intolerant of mutations. However, we fail to find strong statistical support for the hypothesis that variable loops are especially tolerant of mutations. Thus, the rapid evolution of these loops in nature is probably more attributable to strong immune selection than exceptionally high inherent mutational tolerance.
We have used deep mutational scanning to experimentally estimate the effects of all amino-acid mutations to most of HIV Env. Our experiments select for Env variants that enable HIV to undergo multi-cycle replication in a T-cell line. The broad trends in our data are consistent with what is expected from general considerations of how gene sequence maps to protein function: stop codons are efficiently purged by selection, many but not all nonsynonymous mutations are selected against, and synonymous mutations are less affected by selection except at regions where the nucleotide sequence itself is known to be biologically important. We also find a few sites where nonsynonymous mutations are strongly favored by selection in our experiments, probably because they adapt the virus to cell culture by affecting Env’s conformational dynamics, co-receptor binding, and glycosylation.
We use our experimental data to estimate the preference of each site in Env for each amino acid. We show that these preferences correlate with amino-acid frequencies in natural HIV sequences nearly as well as would be expected if the experimentally measured preferences capture the true selection on Env in nature. The strongest deviations between our measurements and amino-acid frequencies in HIV sequences occur at sites on the surface of the virus that in nature are targeted by pressures (such as antibodies) that are not present in our experiments.
The ability to identify deviations between our measurements and amino-acid frequencies in nature points to a powerful aspect of our approach: it can de-convolve the role of inherent functional constraints and external selection pressures in shaping Env’s evolution. For instance, it is known that some regions of Env are conserved in nature and thus are susceptible to broadly neutralizing antibodies. But other regions of Env such as the variable loops exhibit extensive variability and are generally targeted by more strain-specific antibodies. To what extent are these patterns of conservation shaped by Env’s inherent capacity to evolve versus the fact that immune selection tends to target the variable loops more readily than the broadly neutralizing antibody epitopes? By measuring Env’s mutational tolerance at each site under functional selection alone, we show that the epitopes of broadly neutralizing antibodies indeed have a reduced capacity to tolerate mutations irrespective of the action of immune selection. However, we do not find strong statistical support for the hypothesis that the variable loops are especially tolerant of mutations compared to the rest of the protein. Thus, the rapid evolution of these loops probably results more from strong immune selection than exceptionally high inherent mutational tolerance. In the future, our measurements could also be used to examine the role of Env’s mutational tolerance in shaping the evolution of epitopes targeted by cellular immunity [120].
More generally, our experiments provide high-throughput experimental data that can augment computational efforts to infer features of HIV’s fitness landscape [18–20, 22, 121]. Such data will aid in efforts to understand viral evolutionary dynamics both within and between patients. Our study examined the replication of the CXCR4-tropic LAI strain isolated from a chronically infected individual, and used a T-cell line that expresses high levels of receptor relative to many primary cells [122, 123]. This experimental setting is obviously a simplified representation of the actual environment in which HIV replicates. However, we anticipate that our approach could be extended to examine the effects of Env mutations in more complex experimental settings that may better mimic the selection on viruses in humans. For instance, comparing our measurements to those made on transmitted-founder viruses should help elucidate how selective constraints differ among HIV strains. Examining viral replication in cells with different receptor and co-receptor distributions should make it possible to isolate the role of cell-type specific selection in shaping HIV evolution [124, 125]. Adding factors such as antibodies should enable the comprehensive identification of how mutations affect HIV immune escape. Such experiments will augment the results described here with maps of how mutational effects shift under various biologically relevant scenarios, thereby further enhancing our ability to understand the internal and external forces driving HIV evolution.
The computer code to analyze the sequencing data and generate the figures is provided in a series of IPython notebooks in S3 File. Illumina sequencing data are available from the Sequence Read Archive (http://www.ncbi.nlm.nih.gov/sra) under the accession numbers in S10 File.
We use the HXB2 numbering system [50] unless otherwise noted. The “variable loop” definitions were taken from http://www.hiv.lanl.gov/, not including the flanking disulfide-bonded cysteines as part of the loops.
We created the codon mutant libraries in the context of the pro-viral genomic plasmid pLAI, which encodes the LAI strain of HIV [46]. This plasmid was obtained from the lab of Michael Emerman. The plasmid sequence is in S4 File.
We created codon mutant libraries of env using the PCR mutagenesis technique described in [31] (see also [33, 38]) except that we performed two total rounds of mutagenesis rather than the three rounds in [31]. The codon tiling mutagenic primers are in S5 File. The end primers were: 5’-ttggaatttctggcccagaccgtctcatgagagtgaaggagaaatatcagcacttg-3’ and 5’-catctgctgctggctcagc-3’. We created three replicate libraries by performing all the steps independently for each replicate starting with independent plasmid preps.
We cloned the PCR mutagenized env amplicons into the LAI plasmid with high efficiency to create plasmid mutant libraries. To seamlessly clone the PCR products into the proviral plasmid, we created a recipient version of the plasmid that had env replaced by GFP flanked by restriction sites for BsmBI, which cleaves outside its recognition sequence. We named this recipient plasmid pLAI-δ env-BsmBI; its sequence is in S6 File. We digested both this recipient plasmid and the gel-purified PCR amplicons with BsmBI (there are BsmBI sites at either end of the PCR amplicon), gel purified the digested PCR products, and ligated them into the plasmid using a T4 DNA ligase. We column purified the ligation products, electroporated them into competent cells (Invitrogen,12033-015), and plated the transformed cells on LB plates supplemented with 100 μg/mL ampicillin. For each of the three replicate libraries, we performed enough transformations to yield >1.4 million unique colonies as estimated by plating dilutions of each transformation on separate plates. Control ligations lacking an insert yielded at least 10-fold fewer colonies. The transformed cells were scraped from the plates, grown in liquid LB-ampicillin at 37°C for ∼4 hours, and mini-prepped to obtain the plasmid mutant libraries. For the wildtype controls, we prepped three independent cultures of the wildtype LAI proviral plasmid.
We generated the mutant virus libraries by transfecting the mutant plasmid libraries into 293T cells obtained from the American Type Culture Collection (ATCC). For each replicate, we transfected two 12-well tissue-culture plates to increase the diversity of the generated viruses. Specifically, we plated 293T cells at 2.4×105 cells/well in D10 media (DMEM supplemented with 10% FBS, 1% 200 mM L-glutamine, and 1% of a solution of 10,000 units/mL penicillin and 10,000 μg/mL streptomycin). The next day, we transfected each well with 1 μg plasmid using BioT (Bioland Scientific LLC, B01-01). For the three wildtype controls we used the same process but with only a single 12-well plate per replicate. At one day post-transfection, we aspirated the old media, replacing it with fresh D10. At ∼60 hours post-transfection, we filtered the transfection supernatants through 0.4 μm filters. To remove residual plasmid DNA from the transfection, we then treated the filtrate with DNase-I (Roche, 4716728001) at a final concentration of 100 U/mL in the presence of 10 mM magnesium chloride (Sigma, M8266) at 37°C for 20–30 minutes. We froze aliquots of the DNase-treated supernatant at -80°C. Aliquots were thawed and titered by TZM-bl and TCID-50 assays as described below.
We passaged the transfection supernatants in SupT1 cells obtained from the NIH AIDS Reagent Program [126]. SupT1 cells were maintained in a media identical to the D10 described above except that the DMEM was replaced with RPMI-1640 (GE Healthcare Life Sciences, SH30255.01). Before infecting cells, for replicates 1, 2, and 3 (but not replicate 3b), we first filtered thawed transfection supernatants through a 0.2 μm filter in an effort to remove any large viral aggregates. We then infected 108 SupT1 cells with 5 × 105 TZM-bl units of the mutant library transfection supernatant in a final volume of 100 mL SupT1 culture medium in a vented tissue-culture flask (Fisher Scientific, 14-826-80). In parallel, we passaged 105 TZM-bl units of transfection supernatant for each wildtype control in 20 million SupT1 cells in a final volume of 20 mL. At one day post-infection, we pelleted cells at 300×g for 4 minutes and resuspended in fresh media to the same volume as before. At two days post-infection, we added fresh media equal to the volume already in the flask to dilute the cells and provide fresh media. We harvested virus at three days post-infection (for replicates 1, 2, and 3) or four days post-infection (for replicate 3b) by pelleting cell debri at 300×g for 4 minutes and then collecting the viral supernatant for storage at -80°C. To remove residual culture media and plasmid DNA from the cell pellets, we washed pellets two times in PBS. The washed cells were resuspended in PBS to a final concentration of 107 cells/mL, and aliquots were frozen at -80°C for DNA purification.
We conducted a second passage by infecting new cells with the passage-1 viral supernatants. The second passage differed from the first passage in the following ways: Before infecting cells, we filtered passage-1 supernatant of replicate 3b-2 through a 0.2 μm filter but did not filter any of the other replicates. We also had to modify the passaging conditions for some replicates due to low titers of the passage-1 supernatants. For viruses in which the passage-1 supernatant was at too low a concentration to infect at an MOI of 0.005 in the volumes indicated above, we added additional passage-1 supernatant, and then reduced the volume to that indicated above during the day-one media change. As stated in the Results section, passaging more than 5 × 105 TZM-bl units of the mutant library at each step would probably help increase reproducibility between experimental replicates.
We measured viral titers using TZM-bl reporter cells obtained from the NIH AIDS Reagent Program [127]. Specifically, we added 2×104 cells in 0.5 mL D10 to each well of a 48-well plate. We made dilutions of viral inoculum and infected cells with 100 uL of each dilution. At 2 days post-infection, we fixed cells in a solution of 1% formaldehyde and 0.2% glutaraldehyde in PBS for 5 minutes at room temperature, washed with PBS to remove the fixing solution, and stained for beta-galactosidase activity with a solution of 4 mM potassium ferrocyanide, 4 mM potassium ferricyanide, and 0.4 mg/mL X-gal in PBS at 37°C for 50 minutes. After washing cells with PBS to remove the staining solution, we used a microscope to count the number of blue cells per well, computing the viral titer as the number of blue cells per mL of viral inoculum.
We were concerned that the infectious titer in SupT1 cells might differ from the TZM-bl titers. We therefore also performed TCID50 assay to directly measure infectious titers in SupT1 cells. To do this, we made dilutions of viral transfection supernatant in a 96-well tissue-culture plate and added SupT1 cells at a final concentration of 2.5×105 cells/mL in a final volume of 180 μL/well. At 4 and 8 days post-infection, we passaged supernatant 1:10 into fresh media to prevent cells from becoming over confluent. At 12 days post-infection, we measured the titer of culture supernatants using the TZM-bl assay to determine which SupT1 infections had led to the production of virus. Based on binary scoring from these TZM-bl assays, we calculated titers using the Reed-Muench formula [128] as implemented at https://github.com/jbloomlab/reedmuenchcalculator. At least for the LAI strain used in our experiments, the SupT1 TCID50 titers were approximately equal to the TZM-bl titers. Therefore, we used only the less time-consuming TZM-bl assay for all subsequent titering.
We purified non-integrated viral DNA from aliquots of frozen SupT1 cells using a mini-prep kit (Qiagen, 27104) with ∼107 cells per prep. In some cases, we then concentrated the purified DNA using Agencourt AMPure XP beads (Beckman Coulter, A63880) using a bead-to-sample ratio of 1.0 and eluting with half of the starting sample volume.
We next generated PCR amplicons of env to use as templates for Illumina sequencing. We created these amplicons from plasmid or mini-prepped non-integrated viral DNA by PCR using the primers 5’-agcgacgaagacctcctcaag-3’ and 5’-acagcactattctttagttcctgactcc-3’. PCRs were performed in 20 μl or 50 μl volumes using KOD Hot Start Master Mix (71842, EMD Millipore) with 0.3 μM of each primer and 3 ng/μl of mini-prepped DNA or 0.3 ng/μl of plasmid as template. The PCR program was:
For replicate 3b, there were a few modifications: the annealing temperature was 64.9°C, the extension time was 54 seconds, and we performed only 25 cycles. To quantify the number of unique template molecules amplified in each PCR, we performed standard curves using known amounts of template env in pro-viral plasmid, and ran the the bands on an agarose gel alongside our amplicons for visual quantification. We performed a sufficient number of PCR reactions to ensure that amplicons from plasmid were coming from > 106 unique template molecules, and amplicons from viral DNA were coming from ∼2 × 105 template molecules. All PCR products were purified with Agencourt beads (using a sample-to-bead ratio of 1.0) and quantified by Quant-iT PicoGreen dsDNA Assay Kit (Life Technologies, P7589).
We deep sequenced these amplicons using the strategy for barcoded-subamplicon sequencing in [38], dividing env into six subamplicons (this is a variation of the strategy originally described in [58–60]). The sequences of the primers used in the two rounds of PCR are in S9 File. Our first-round PCR conditions slightly differed from [38]: our 25 μL PCRs contained 12.5 μL KOD Hot Start Master Mix, 0.3 μM of each primer, and 5 ng of purified amplicon. For replicates 1, 2, and 3, the first-round PCR program was:
For replicate 3b, we used the same program, but with 9 PCR cycles instead of 11. Prior to the second round PCR, we bottlenecked each subamplicon by diluting it to a concentration that should have yielded between 3 and 5×105 unique single-stranded molecules per subamplicon per sample. We purified the second-round PCR products using Agencourt beads, quantified with PicoGreen, pooled in equimolar amounts, and purified by agarose gel electrophoresis, excising DNA corresponding to the expected ∼500 base pairs in length. We sequenced the purified DNA using multiple runs of an Illumina MiSeq with 2×275 bp paired-end reads.
We used dms_tools (http://jbloomlab.github.io/dms_tools/), version 1.1.dev13, to filter and align the deep-sequencing reads, count the number of times each codon mutation was observed both before and after selection, and infer Env’s site-specific amino-acid preferences using the algorithm described in [88]. The code that performs this analysis is in S3 File. Figures summarizing the results of the deep sequencing are also in this supplementary file.
We downloaded the 2014 filtered web alignment of env from http://www.hiv.lanl.gov/, including all subtypes for HIV-1/SIVcpz. We then curated this alignment in the following ways. First, we removed sequences differed in length from HXB2 (including gap characters) or contained a premature stop codon, ambiguous residue, or frame-shift mutation. Next, we removed columns in the alignment for which we lacked deep mutational scanning data, columns that had >5% gap characters, or columns in variable loops that appeared poorly aligned by eye. Finally, we randomly selected 30 sequences per subtype for group-M subtypes A, B, C, D, F, and G, for a total of 180 sequences. The resulting alignment is in S7 File. The phylogenetic tree in Fig 7 was inferred using RAxML [129] with the GTRCAT substitution model.
We computed absolute solvent accessibilities based on the PDB structure 4TVP (including all three Env monomers after removing antibody chains) using DSSP [130, 131]. We normalized absolute solvent accessibilities to relative ones using the maximum accessibilities provided in the first table of [132]. The relative solvent accessibilities are listed in S8 File.
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.