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.1000330 | RNA-Dependent Oligomerization of APOBEC3G Is Required for Restriction of HIV-1 | The human cytidine deaminase APOBEC3G (A3G) is a potent inhibitor of retroviruses and transposable elements and is able to deaminate cytidines to uridines in single-stranded DNA replication intermediates. A3G contains two canonical cytidine deaminase domains (CDAs), of which only the C-terminal one is known to mediate cytidine deamination. By exploiting the crystal structure of the related tetrameric APOBEC2 (A2) protein, we identified residues within A3G that have the potential to mediate oligomerization of the protein. Using yeast two-hybrid assays, co-immunoprecipitation, and chemical crosslinking, we show that tyrosine-124 and tryptophan-127 within the enzymatically inactive N-terminal CDA domain mediate A3G oligomerization, and this coincides with packaging into HIV-1 virions. In addition to the importance of specific residues in A3G, oligomerization is also shown to be RNA-dependent. Homology modelling of A3G onto the A2 template structure indicates an accumulation of positive charge in a pocket formed by a putative dimer interface. Substitution of arginine residues at positions 24, 30, and 136 within this pocket resulted in reduced virus inhibition, virion packaging, and oligomerization. Consistent with RNA serving a central role in all these activities, the oligomerization-deficient A3G proteins associated less efficiently with several cellular RNA molecules. Accordingly, we propose that occupation of the positively charged pocket by RNA promotes A3G oligomerization, packaging into virions and antiviral function.
| APOBEC3G is a human protein that inhibits the replication of HIV-1 in CD4+ T cells. It gains entry to the virus particles that are released from infected cells and subsequently interferes with viral genome replication, which in the case of HIV-1 is reverse transcription. APOBEC3G is a cytidine deaminase, and it catalyses the deamination of cytidines to uridines in viral single-stranded DNA replication intermediates, resulting in the generation of defective progeny viruses. In addition, APOBEC3G can inhibit reverse transcription by a poorly characterized deamination-independent mechanism. HIV-1 has evolved the viral Vif protein to counteract the antiviral properties of APOBEC3G. Vif associates with APOBEC3G and targets it for proteasomal degradation, such that intracellular levels of APOBEC3G are reduced and packaging into virions is averted. Based on the structure of a human homolog of APOBEC3G, APOBEC2, we performed a mutational analysis of amino acids that have the potential to mediate the assembly of APOBEC3G into multi-component complexes. We report that these amino acids affect the association of APOBEC3G with itself and cellular RNA, and that the same attributes are also required for packaging into virions and antiviral function. Thus, the processes of APOBEC3G self-association, RNA binding, and virion packaging are functionally linked and essential for virus inhibition.
| The human protein APOBEC3G (A3G) belongs to a family of cellular polynucleotide cytidine deaminases and is a potent inhibitor of HIV-1 in the absence of the viral protein Vif [1]. Vif-deficient HIV-1 (HIV-1/Δvif) is subject to A3G mediated cytidine to uridine deamination of single-stranded DNA that is generated during reverse transcription, a process also referred to as DNA editing or hypermutation [2]–[4]. In addition, A3G further suppresses infection by inhibiting reverse transcription in a poorly understood manner that seems to be independent of the deamination activity of the protein [5]–[9]. A3G is incorporated into progeny virions during particle assembly at the plasma membrane by associating with the NC domain of the viral Gag protein in an RNA dependent manner [10]–[15]. The viral Vif protein prevents the antiviral properties of A3G by targeting it for proteasomal degradation [16]. Specifically, Vif interacts with A3G and recruits the cullin5-elonginB/C-Rbx ubiquitin ligase complex, resulting in the polyubiquitylation and degradation of A3G [17]. This reduction of intracellular levels in A3G results in a substantial decrease in the packaging of A3G into virus particles and, therefore, suppresses its antiviral properties.
The recent reports of the crystal structure of APOBEC2 (A2) [18] and the NMR and crystal structures of the C-terminal cytidine deaminase (CDA) domain of A3G [19],[20] offer opportunities to investigate the structure-function organization of APOBEC proteins with greater incisiveness. Although the physiological function of A2 is as yet unknown, its structure shows all the hallmarks of a cytidine deaminase, being a five-stranded mixed β-sheet which presents on one face two α-helices containing the H/C-X-E-X23–28-P-C-X2-C catalytic centre that coordinates a zinc ion. Surprisingly, A2 associates into tetramers in a manner unprecedented among cytidine deaminases. Whereas tetrameric free-nucleotide cytidine deaminases of bacteria [21],[22], yeast [23] and vertebrates [24] adopt a globular structure in which each subunit interacts with the other three, A2 forms a linear tetramer in which monomers interact with either one or two of the other subunits [18]. An A2 monomer contains a single CDA domain and two of these form a dimer by joining the β-sheets present in each monomer in a side-by-side fashion such that one wide β-sheet is formed. The tetramer is assembled from two such dimers through head-to head interactions at one edge of the extended β-sheets.
In contrast to A2, A3G contains two CDA domains in a single polypeptide chain, which are termed the N- and C-terminal CDA domains (N-CDA and C-CDA, respectively). Indeed, the CDA fold observed in the A2 crystal structure closely matches the structure of the truncated A3G C-CDA domain as observed by NMR and crystallography [19],[20]. Differences arise mainly at the peripheral loops, which are generally longer in A3G than in A2. The A3G C-CDA fragment is exclusively monomeric, both in solution and in crystalline form [19],[20]. However, there is mounting evidence that A3G not only oligomerizes [12], [25]–[30], but can also assemble into large RNP complexes that accumulate in cellular microdomains that are associated with RNA regulation, such as P-bodies and stress granules [30]–[34]. We therefore asked whether the tetrameric structure of A2 may hold clues not only into how A3G oligomerizes, but also into its participation in other interactions. Here, we show that hydrophobic residues in A3G that are equivalent to those that mediate A2 oligomerization are required for RNA-dependent oligomerization, packaging into HIV-1 virions and the inhibition of HIV-1 infection. In addition, we present a homology model of an A3G dimer that reveals a positively charged pocket at the predicted dimer interface. Mutation of basic residues within this pocket also affects oligomerization, RNA interactions, virion packaging and virus inhibition.
The tetramer interface of A2 is formed of extensive hydrophobic, polar and electrostatic interactions, many of which are clustered in a loop termed L1 [18]. In particular, residues F155, M156 and W157 mediate key hydrophobic interactions (Figure 1A), whereas R153, E158 and E159 are involved in salt-bridges as well as in hydrogen bonding. Upon alignment of the A2 amino acid sequence with the N- or C-terminal CDA sequences of A3G, we identified a highly similar loop sequence in both CDA domains of A3G in which both charged and hydrophobic residues are conserved (Figure 1B). Arginines equivalent to R153 in A2 are present at positions 122 and 313 in A3G, whereas the tyrosines at positions 124 and 315 in A3G are at the equivalent position of F155 in A2. Although F155 does not make any direct interactions across the tetramer interface of the A2 crystal, it is involved in a cluster of hydrophobic packing interactions that sandwich M156 between Y61 and F155 at the tetramer interface. A tryptophan equivalent to W157 in A2 is present only in the N-CDA of A3G at position 127. W157 of A2 makes extensive hydrophobic interactions across the tetramer interface, notably with Y214 and W157 of the adjacent subunit.
We have previously reported a mutational analysis of residues 122–146 of A3G to define the site of interaction with Vif, which was mapped at positions 128–130 [35]. That analysis also revealed that substitutions at positions Y124 and W127 yield A3G proteins that are inefficiently packaged into virus particles and therefore lose their antiviral properties. Given the involvement of the conserved counterparts of these residues in A2 oligomerization, we sought to establish whether this region would have an analogous activity in A3G. To investigate this possibility, and to compare the contribution that the N- or C-CDAs of A3G may make to oligomerization, we introduced identical mutations (alanine, leucine and phenylalanine) at residues Y124 and Y315; these residues were chosen because they are present at equivalent positions in both the N- and C-CDA of A3G, and because the mutant proteins are expressed well. In contrast, the introduction of substitutions at the conserved arginine at position 122 resulted in poor expression [35], and mutants of R122 were therefore not examined further. The construction of mutations at position W127 has been described previously [35].
We first tested these mutant A3G proteins for their ability to inhibit HIV-1/Δvif infection (Figure 2A). The substitution of Y124 or Y315 to alanine or leucine caused marked losses of antiviral function, whereas substitutions to the chemically more similar phenylalanine resulted in less marked disruption. Determination of the A3G content of virus particles revealed that all mutations at position Y124 result in poor packaging, whereas packaging was maintained with mutations at position Y315 (Figure 2B). Interestingly, the Y124F mutation yielded low but clearly detectable levels of A3G in virions in comparison to mutants Y124A and Y124L, which likely explains why this protein showed a less severe loss of antiviral activity. We next determined the extent to which wild type or mutant A3G can act as a mutagen in a bacterial DNA editing assay (Figure 2C). In this analysis, we included two mutants of W127 (W127A and W127Y), which have previously been shown to have substantial packaging defects [35]. Editing activity was maintained following substitutions at positions Y124 and W127, but mutations at position Y315 caused a loss of editing. Together, these results indicate that the loss of antiviral activity imparted by mutations at residues Y124 and W127 corresponds to reduced packaging, whereas DNA editing activity is unaffected. Conversely, mutations at residue Y315 ablate DNA editing but not packaging into virions, which is consistent with the critical involvement of Y315 in substrate DNA binding at the catalytically active C-CDA domain [19],[20].
To begin to address the ability of A3G to oligomerize, we performed a yeast two-hybrid experiment (Figure 2D). Mutations were introduced into the prey-construct and assayed with a wild type A3G bait. Again, we observed a marked difference between the effects of substitutions in the N- and C-CDA domains of A3G. Mutations at Y124 and W127 resulted in a lack of reporter gene activity, whereas mutations at Y315 displayed wild type levels of activity and, hence, interaction. This result suggests that residues Y124 and W127 of the N-CDA domain play critical roles in A3G oligomerization, whereas Y315 of the C-CDA does not.
To investigate further the oligomerization of A3G, we next performed a series of immunoprecipitation and chemical crosslinking experiments. Wild type or mutant A3G was co-expressed with HA-tagged wild type A3G (A3G-HA) and then co-immunoprecipitated from cell lysates using a monoclonal anti-HA antibody. Aliquots were or were not treated with RNAse A and analyzed by immunoblotting (Figure 3A). Consistent with the results of the yeast two-hybrid experiments, wild type A3G and the Y315A mutant were efficiently co-precipitated with A3G-HA, whereas co-precipitation of the Y124A mutant was strongly reduced and a mere trace amount of W127A was detected in the immunoprecipitate. In all cases, co-precipitation with A3G-HA was substantially inhibited by the treatment with RNAse A, indicating that RNA is required for stable A3G oligomerization.
We then performed a chemical crosslinking experiment in which cell lysates from 293T cells expressing wild type A3G were treated with BM(PEO)3, a compound with two reactive maleimide moieties separated by an 18 Å linker that form irreversible covalent bonds with the sulfohydrils of cysteines [36]. After crosslinking and resolution by SDS-PAGE, A3G was detected as a band migrating at ∼80 kD as well as at ∼40 kD where the untreated, and presumably monomeric, A3G migrates (Figure 3B, lanes 1 and 2). Treatment with RNase A prior to crosslinking resulted in complete disappearance of the band at ∼80 kD; this was maintained, however, when the RNase treatment was performed after the crosslinking reaction (Figure 3B, lanes 3 and 4, respectively). To verify that the crosslinked A3G species migrating at ∼80 kD represents an A3G oligomer, we also perfomed an experiment in which myc-tagged A3G (A3G-myc) was co-expressed with A3G-HA, subjected to crosslinking and then immunoprecipitated with the anti-HA antibody (Figure 3C). Samples were split into two aliquots which were (or were not) subjected to RNase A treatment after crosslinking. Indeed, A3G-myc was detected in the immunoprecipitate as a species migrating at ∼80 kD after treatment with BM(PEO)3 and this was unaffected by treatment with RNase A. In the control sample without the crosslinker, detection of monomeric A3G-myc in the immunoprecipitate was abolished by the treatment with RNase A. This result indicates that the species migrating at ∼80 kD is indeed formed by intermolecular crosslinking between A3G-HA and A3G-myc.
To assess in greater detail the oligomerization characteristics of some of our mutant A3G proteins, we next performed a crosslinking experiment with untagged wild type A3G or A3G harbouring the Y124A, Y315A or W127A mutations. The ∼80 kD crosslinked species appeared at a low level with the Y124A mutant and was barely detectable with the W127A mutant, whereas it was efficiently generated with the Y315A mutant (Figure 3D). Crosslinking and co-immunopreciptitaton of A3G with the Y124F or W127Y mutations resulted in increased levels oligomerization in comparison to the respective alanine mutations (Figure S1), which correlates with the less disruptive effects on virus packaging of these substitutions relative to the alanine mutations (Figure 1B and [35]). We note that we were unable to detect these subtle differences with the yeast-two hybrid system (Figure 1D).
To assess the possibility that the ∼80 kD species may be due to two A3G monomers being bound in close proximity on the same RNA molecule, we performed crosslinking experiments in which A3G was co-expressed with the RNA helicase Mov10. Mov10 is known to associate with A3G in ribonucleoprotein complexes in an RNA-dependent manner, though it is unknown whether these proteins interact directly [34]. We were unable to detect any intermolecular crosslinks between A3G and Mov10 (results not shown), further suggesting that the ∼80 kD crosslinked species forms as a consequence of A3G intermolecular contacts. Together, these results indicate that residues Y124 and W127 play central roles in the N-CDA mediated oligomerization of A3G, and that this interaction is also dependent on the presence of RNA.
In a complementary approach for addressing the mode of A3G oligomerization, we constructed homology models of A3G dimers using the A2 crystal structure as a template (Figure 4A). In these models, the N- and C-CDA domains of one A3G polypeptide together form the extended β-sheet that is the equivalent of an A2 dimer. Models with either the N-CDA or C-CDA at the oligomer interface, which corresponds to the tetramer interface of A2, were then assembled and subjected to energy minimization. In convergence with the results of the experiments described above, models with N-CDA at the oligomer interface (Figure 4C) proved energetically more favourable than models with the C-CDA at the dimer interface by ∼2000 to ∼3000 kJ/mol, depending on interactions with the solvent (Table S1). N-terminal dimerization resulted in residues Y124 and W127 being buried within the oligomer interface, with Y124 predicted to be slightly more accessible to solvent than W127 (Table S2). We next assessed the effect of the Y124A and W127A mutations by determining the interactions that are lost upon introduction of these mutations into the A3G structure model with the N-terminal CDA at the dimer interface (Table S3). This analysis demonstrated that the Y124A mutation results in the loss of intra- and intersubunit interactions, while the W127A mutation affects mostly intersubunit interactions in this structure model.
Importantly, inspection of the charge distribution over the surface the model revealed a conspicuous clustering of positive charges that are located in a rather large pocket at the predicted dimer interface (Figure 4D): notably, residues Y124 and W127 are also located within this pocket (Figure 5A). In contrast, this positively charged surface is absent from the A2 structure (Figure 4B). To assess the accuracy of this modelling effort, the C-CDA domain from our model was superimposed with the NMR [19] and X-ray [20] structures for this domain (Figure S2). The overall agreement was good with both structures, as evidenced by a root mean square deviation (RMSD) of less than 5 Å, but our model displayed slightly more similarity to the X-ray structure (RMSD NMR: 4.920 Å and RMSD X-ray: 3.650 Å).
The presence of clustered charged and aromatic residues at the A3G oligomer interface is suggestive of a binding site for RNA. In particular, the basic residues R14, R24, R29, R30, K63, K99, R102, R122, R136, K141 and R142 all lie within the aforementioned pocket and we sought to test this feature of our model. All these arginines and lysines were mutated to alanine and their antiviral properties assessed in single-cycle infectivity assays; the R24A and R30A proteins had the most profound loss of antiviral function, whereas mutations at other positions had no or modest effects, as exemplified by R136A (Figure 6A, and data not shown).
As the basic residues may act cooperatively to bind RNA, we also produced a set of doubly mutated proteins in which the R24A or R30A mutation was combined with alanine substitutions at R63, R99, R102, R136, K141 or R142. Only the R24A+R136A and R30A+R136A composite mutants showed further reductions in virus inhibition, resulting in phenotypes similar in severity to the W127A mutation (Figure 6A, and data not shown). Analysis of the levels of A3G present in virus particles revealed that mutants R24A, R30A, R24A+R136A and R30A+R136A were each packaged less efficiently than wild type A3G, but that the R136A mutant was still packaged well (Figure 6B). None of the R24A, R30A and R136A mutants showed any loss of editing activity in bacteria, either as single or double mutants, indicating that these proteins were not misfolded (Figure 6C).
We next assessed the oligomerization properties of these mutant A3G proteins by co-immunoprecipitation and chemical crosslinking. In both assay systems, we consistently observed that the R24A and R30A mutants oligomerized less efficiently than the wild type protein, and this was accentuated further by the additional R136A substitution (Figure 6D and 6E). Together, these results demonstrate that the removal of basic residues from the predicted oligomer interface creates proteins with very similar phenotypes to the Y124A and W127A mutants. Specifically, these mutated proteins display limited antiviral activity, packaging, and oligomerization, and this is consistent with the close spatial proximity of these residues in our structure model (Figure 5).
To determine whether oligomerization-defective mutants of A3G are reciprocally deficient for associating with cellular RNA, we performed semi-quantitative reverse transcription coupled PCR on immunoprecipitates of wild type or mutant HA-tagged A3G to detect the presence of the Y1, Y4 and 7SL RNAs; these RNA molecules have each previously been shown to be present in A3G-associated RNPs [13],[32],[37]. Indeed, these RNAs were readily detected in association with wild type A3G and the oligomer-forming Y315A mutant (Figure 7A). In contrast, much less Y1, Y4 and 7SL RNA was detected in the immunoprecipitates of W127A, R24A, R24A+R136A and R30A+136A A3G, as well as in the A3F and luciferase negative controls. Modest exceptions were the Y124A and R30A mutants, for which low levels of Y4 as well as 7SL RNA, respectively, were detected. These differences were not due to different amounts of protein in the immunoprecipitate, as demonstrated by immunoblotting (Figure 7B).
Thus, mutations of hydrophobic and basic residues at the predicted A3G oligomer interface caused a loss in association with cellular RNA. These mutations do not, however, affect the ability of A3G to assemble into high molecular weight ribonucleoprotein complexes in 293T cells, as evidenced by velocity sedimentation of A3G-containing cell lysates through sucrose gradients (Figure S3). Moreover, all A3G-containing complexes maintained sensitivity to RNase treatment, suggesting that the mutations have not imparted pleiotropic defects in nucleic acid interactions or the capability to assemble into large RNP complexes.
We have performed a mutational study of residues in the N- and C-CDA domains of A3G whose counterparts in A2 are involved in key interactions that support oligomerization of A2. Our results show that mutations in the N-CDA, but not the C-CDA, are associated with reductions in A3G RNA-dependent oligomerization and packaging into virions. Upon modelling of an A3G dimer onto the template A2 crystal structure, we identified a positively charged pocket at the oligomer interface formed between two N-CDAs that bore the hallmarks of a nucleic acid binding site (Figure 5). Indeed, mutation of basic residues within this pocket also resulted in losses of antiviral function, packaging into virus particles, oligomerization and association with cellular RNA (Figures 6 and 7). Consistent with previous work showing that only the C-CDA of A3G is responsible for DNA editing [5], [38]–[40], this attribute was unaffected by disruption of this basic pocket. Thus, our findings demonstrate further segregation of functions between the N- and C-CDA domains of A3G.
Throughout our chemical crosslinking experiments, we consistently observed the generation of oligomeric A3G migrating at ∼80 kD, which is twice the relative molecular mass of untreated A3G, which migrates at ∼40 kD (Figures 3 and 6). This result suggests that A3G oligomerizes as a discrete dimer, an assertion that is further supported by the fact that we did not detect slower migrating species at ∼120 (trimer) or ∼160 kD (tetramer). We note, however, that we have not formally demonstrated dimerization of A3G, as attempts to perform analytical ultracentrifugation were unsuccessful owing to the poor solubility of purified full-length A3G at high concentrations (results not shown). Nonetheless, dimerization of A3G via the N-terminal CDA domain remains the simplest model to explain our current results. Although this conclusion is at odds with a recent study proposing oligomerization of A3G via the C-CDA domain [41], that study is also contradicted by the observations that the C-CDA of A3G appears as a monomer by both ultracentrifugation [19] and crystallography [20].
Importantly, we have furthermore demonstrated that the oligomerizaton of A3G is dependent on the presence of RNA, as evidenced by the disruption of oligomers upon treatment with RNase (Figure 3). These observations are explained by our combined modelling and structure-function analyses, which predict that the oligomer interface between the A3G N-terminal CDA domains produces a positively charged pocket that requires occupation by RNA to allow effective oligomerization. Thus, we propose that the formation of A3G oligomers requires hydrophopic and basic residues that mediate protein-protein interactions between the A3G subunits and/or protein-RNA interactions, in a manner similar to that proposed for PKR and RIG-I [42]–[44]. We acknowledge, however, that the precise contribution of these residues to RNA-dependent oligomerization of A3G must await advances in the biochemical characterization of this protein.
An additional piece of evidence supporting the interdependence between the oligomerization of A3G and the association with RNA comes from the analysis of Y1, Y4 and 7SL RNA in A3G RNPs (Figure 7). In general terms, we observed that oligomerization-impaired mutants of A3G exhibited much reduced co-immunoprecipitation of these RNA molecules. Indeed, correlations between oligomer formation and RNA interaction were excellent in that the R30A and Y124A mutants displayed partial A3G-A3G interactions as well as intermediate levels of RNA interactions (Figures 3, 5, and 7). An additional instructive observation was made upon velocity sedimentation of cell lysates with oligomerization-impaired A3G, which demonstrated that assembly into RNase-sensitive high molecular weight RNP complexes was not noticeably affected by these mutations (Figure S3). This demonstrates that A3G's assembly into at least two intermolecular complexes is RNA-dependent: the oligomerization of A3G and its recruitment into RNP complexes. Importantly, our mutational analysis shows that oligomerization can be disrupted selectively without grossly preventing RNP association. This suggests either that (1) there is a certain degree of specificity to the identity of RNAs that are required for A3G oligomerization, but not to the RNAs that promote RNP association, or that (2) recruitment of A3G to RNase-sensitive RNP complexes is driven predominantly by protein-protein interactions. Specifically, RNA-dependent A3G RNP formation through protein-protein interactions could be mediated by proteins that bind A3G directly and additionally bind RNA.
Throughout these studies, we have highlighted a tight correlation between the packaging into HIV-1 virions and the RNA-dependent oligomerization of wild type and mutant A3G proteins. Here, we have presented a structure model of an A3G dimer that readily accommodates these attributes. Indeed, the packaging of A3G into virus particles has been reported to require binding to RNA and this has been interpreted as reflective of an RNA-dependent interaction between the HIV-1 Gag protein and A3G [10]–[14],[45]. Although the identity of RNA required for packaging of A3G into HIV-1 virions remains debated [10],[13],[14],[37],[45], specificity with regards to the RNA molecules that mediate oligomerization of A3G may impart some of the selectivity for the establishment of an A3G-Gag interaction and virion packaging.
The structure of A3G is also of considerable interest with regard to the binding of the HIV-1 Vif protein and efforts to manipulate this interaction therapeutically. Previous analyses have shown that Vif interacts with a three amino-acid core motif in A3G at residues 128–130 [35], [46]–[49], which is directly adjacent to residues Y124 and W127. This would position the residues of A3G that interact with Vif in close proximity to the oligomer interface. In our previous study, we found that mutant proteins with substitutions at position Y124 or W127 remain responsive to regulation by the Vif protein [35], suggesting that oligomerization is not a prerequisite for binding of Vif. Indeed, the interaction of Vif with A3G in co-immunoprecipitation experiments is resistant to treatment with RNase [30],[34]. Similarly, mutations at residues 128–130 in A3G affect the interaction with Vif but not packaging into virus particles [35] or generation of the dimeric species by chemical crosslinking (results not shown). Thus, while the residues that mediate Vif-binding and RNA-dependent oligomerization are in close proximity, they appear to be functionally distinct.
We have presented evidence for the RNA-dependent oligomerization of A3G via its N-CDA domain. A structure model of an A3G dimer based on the A2 crystal structure readily rationalizes the RNA-dependency of oligomerization as it revealed a clustering of positive charge near the predicted dimer interface. Furthermore, the model proved consistent with the contribution of basic residues at the interface to RNA-dependent oligomerization and packaging of A3G into virus particles. We thus propose that this model can serve as a guide for the further dissection of the structure-function relationships of domains and motifs within A3G. Ultimately, this may help endeavours aimed at therapeutic intervention with the interaction between the HIV-1 Vif protein and A3G. In particular, such efforts should strive to preserve the antiviral functions of A3G by interrupting the interaction with Vif, while maintaining the interactions that mediate association with RNA, oligomerization and virion packaging.
Wild type and mutant A3G expression plasmids for infectivity studies, immunoprecitation, crosslinking and the bacterial editing assay were generated as described previously [35]. A3G expression plasmids for the yeast two-hybrid experiments were generated by cloning of the EcoRI fragment from the pCMV4-A3G plasmids into the EcoRI site of the pGBKT7 (bait) and pHB18 plasmids (prey) [50]. Proper orientation and sequence of the insert was confirmed by restriction digest or sequencing.
Stocks of HIV-1/Δvif [51] were prepared by cotransfection of 35-mm diameter monolayers of 293T cells with 0.5 µg of pA3G expression vector and 1.0 µg of pIIIB/Δvif using polyethylenimine (PEI). After 24 hr, the supernatants were harvested and volumes corresponding to 5 ng p24Gag used to infect 105 TZM-bl indicator cells. The producer cells were lysed in SDS-containing loading dye for the analysis of protein expression. The induced expression of β-galactosidase in whole cell lysates was measured 24 hr after the initiation of infection using the Galacto-Star system (Applied Biosystems).
Whole cell lysates prepared from virus producing cells, immunoprecipitates and purified HIV-1 virions were resolved by SDS-polyacrylamide gel electrophoresis (SDS-PAGE, 11% gel) and analysed by immunoblotting using primary antibodies specific for A3G [5], myc (ab9106; Abcam), HA (12CA5), Hsp90 (sc7947: Santa Cruz) and p24Gag [52]. Blots were resolved using either horseradish peroxidase-conjugated secondary antibodies and enhanced chemiluminescence (Pierce) or fluorescent secondary antibodies using the LI-COR infrared imaging technology (LI-COR UK LTD).
Virus stocks containing 20 ng p24Gag were spun in a benchtop centrifuge at 21000× g for 2 h at 4°C through a 20% w/v sucrose cushion (500 µl) in a 2 ml eppendorf tube. Viral pellets were resuspended in loading dye and analyzed by immunoblotting. Whole cell lysates from the corresponding producer cells were assessed for A3G and Hsp90 expression in parallel.
The KL16 strain of E. coli was transformed with pTrc99A-based, IPTG-inducible A3G expression vectors or the empty vector [53]. Individual colonies were picked and grown to saturation in LB medium containing 100 µg/ml ampicillin and 1 mM IPTG. Appropriate dilutions were spread onto agar plates containing either 100 µg/ml ampicillin or 100 µg/ml rifampicin and incubated overnight at 37°C. Mutation frequencies were recorded as the number of rifampicin-resistant colonies per 109 viable cells, which were enumerated using the ampicillin-containing plates. Colony counts were recorded in this manner on 12 rifampicin- and 12 ampicilin-containing plates for each construct, in sets of 4 of each at one time. To average the repeat experiments, the average colony count for wild type A3G was set at 100 and all other scores were normalized to this value.
Yeast Y190 cells were transformed with 1 µg of each of the pGBKT7 (bait) and pHB18 (prey) plasmids [50]. The Wild type A3G cDNA was inserted into the bait construct, and Wild type A3G and mutant derivative inserts thereof were cloned into the prey construct. Transformants were selected on medium lacking tryptophan and leucine for 3 days at 30°C. Pools of >20 transformed yeast colonies were scraped into β-Gal assay buffer (60 mM Na2HPO4, 40 mM NaH2PO4, 10 mM KCl, 1 mM MgSO4, 50 mM 2-mercaptoethanol, 0.01% SDS, pH 7.0) and normalized according to optical density in a final volume of 500 µl. Cells were lysed by addition of 25 µl of chloroform and vortexing. The β-Gal substrate chlorophenol red-β-D-galactopyranoside was added to a final concentration of 4 mM and samples were incubated at room temperature for 30 min. After centrifugation to remove cellular debris, absorbance was determined at 540 nm. Repeat experiments were normalized to the OD540 of samples with Tsg101 (bait) and Vps28 (prey) which was set at 4.0.
293T cells were transfected with 1 µg pA3G-HA and 1 µg pA3G (wild type or mutant) in 35-mm cultures. After 24 h, the cells were lysed in 600 µl lysis buffer (0.5% Triton X-100, 287 mM NaCl, 2.68 mM KCl, 1.47 mM KH2PO4, Na2HPO4, pH 7.2 and complete protease inhibitor cocktail from Roche). The lysates were cleared by centrifugation in a benchtop centrifuge at 21000× g for 10 min and 500 µl of each incubated with the 3F10 HA-specific antibody raised in rat (Roche) and protein G-agarose (Invitrogen) for 2 h at 4°C. 50 µl of the cleared lysate was kept to analyse protein expression levels. After binding to the beads the samples were washed twice with lysis buffer and split into two aliquots of 250 µl. To one aliquot of the samples 25 U of bovine pancreatic RNase A (Sigma) was added, and all samples were tumbled at room temperature for 30 min. The agarose beads were then washed three times with lysis buffer, and resuspended in 50 µl loading dye. 10 µl of the immunoprecipitated samples as well as 10 µl of the cleared lysate were resolved by SDS-PAGE (11% gel) and analyzed by immunoblotting using primary antibodies specific for HA, A3G or Hsp90.
293T cells were transfected with 2 µg pA3G (wild type or mutant) in 35-mm cultures. After 24 h, the cells lysed in 600 µl lysis buffer (0.5% Triton X-100, 287 mM NaCl, 2.68 mM KCl, 1.47 mM KH2PO4, Na2HPO4, pH 7.2 and complete protease inhibitor cocktail from Roche). The lysates were cleared by centrifugation in a benchtop centrifuge at 21000× g for 10 min. Samples were then split into aliquots of 100 µl to which 10 U of RNase A was, or was not, added either prior to or after addition of 1.25 µl of 20 mM BM(PEO)3 (Thermo Scientific) in DMSO. After incubation at 20°C for 1 h, 1 µl of 1 M DTT was added to quench the reaction. After the addition of 25 µl loading dye, samples were analysed by SDS-PAGE and immunoblotting. In the experiment describing crosslinking of A3G-myc to A3G-HA, 293T cells were transfected with 2 µg of each plasmid. After 24 h, cells were lysed in 600 µl lysis buffer and cleared by centrifugation. Samples were split into aliquots of 250 µl which were treated, or not, with 2.5 µl of 20 mM BM(PEO)3 in DMSO. After addition of 2 µl 1 M DTT samples were incubated with the 3F10 anti-HA antibody (Roche) and protein A-agarose beads. Subsequent immunoprecipitation was performed in the manner described above and the gel resolved samples analysed using a myc-specific antibody.
293T cells in a 10 cm dish were transfected with 12 µg of wild type or mutant A3G-HA expression vector and lysed after 24 h in lysis buffer (1% NP-40, 150 mM NaCl, 50 mM Tris-HCl pH 7.5 and complete protease inhibitor cocktail from Roche). The cell lysates were precleared overnight using an irrelevant monoclonal antibody and A3G ribonucleoprotein complexes were subjected to immunoprecipitation with the 3F10 rat anti-HA antibody using protein G-coupled agarose beads. Following immunoprecipitation, associated RNAs were recovered with the miRNAeasy mini kit (Qiagen). RNA was detected by semi-quantitative RT-PCR using the SuperScript III One-Step RT-PCR system with platinum Taq DNA polymerase (Invitrogen) (cDNA synthesis at 55°C for 30 min, denaturation at 95°C for 2 min, 15 amplification cycles of 95°C for 15 sec, 56°C for 30 sec and 68°C for 1 min, and a final extension step at 68°C for 5 min) using specific primers for Y and 7SL RNAs [37]. Products were resolved by electrophoresis on a 1.5% agarose gel and stained with ethidium bromide.
The structure of the dimer model of A3G was obtained by homology modelling using as a template the crystal structure of APOBEC2 (A2) (2NYT pdb entry). To generate the 3D-model, the alignment between A3G and APOBEC2 was submitted to the comparative structural modeling program MODELLER 8v2 [54]. 100 best solutions for the MODELLER objective function have been considered. Models were produced with either the N- or C-terminal CDA domain at the dimer interface.
293T cells were transfected with 2 µg pA3G (wild type or mutant) in 35-mm cultures. After 24 h, the cells were lysed in 250 µl lysis buffer (0.626% NP40, 100 mM NaCl, 50 mM KAc, 10 mM EDTA, 10 mM Tris pH 7.4 and complete protease inhibitor cocktail from Roche). The lysates were cleared by centrifugation in a benchtop centrifuge at 162× g for 10 min followed by 18000× g for 30 sec. Samples were then split into aliquots of 100 µl to which 10 U of RNase A (Sigma) was, or was not, added. Samples were then loaded on top of a 10–15–20–30–50% sucrose step gradient in lysis buffer and centrifuged for 45 min at 163000× g at 4°C. After centrifugation, samples of 78 µl were sequentially removed from the top of the gradient, added to 30 µl of loading dye and analysed by immunoblotting.
|
10.1371/journal.pgen.1004110 | NSUN4 Is a Dual Function Mitochondrial Protein Required for Both Methylation of 12S rRNA and Coordination of Mitoribosomal Assembly | Biogenesis of mammalian mitochondrial ribosomes requires a concerted maturation of both the small (SSU) and large subunit (LSU). We demonstrate here that the m5C methyltransferase NSUN4, which forms a complex with MTERF4, is essential in mitochondrial ribosomal biogenesis as mitochondrial translation is abolished in conditional Nsun4 mouse knockouts. Deep sequencing of bisulfite-treated RNA shows that NSUN4 methylates cytosine 911 in 12S rRNA (m5C911) of the SSU. Surprisingly, NSUN4 does not need MTERF4 to generate this modification. Instead, the NSUN4/MTERF4 complex is required to assemble the SSU and LSU to form a monosome. NSUN4 is thus a dual function protein, which on the one hand is needed for 12S rRNA methylation and, on the other hand interacts with MTERF4 to facilitate monosome assembly. The presented data suggest that NSUN4 has a key role in controlling a final step in ribosome biogenesis to ensure that only the mature SSU and LSU are assembled.
| Mitochondria perform a number of essential functions in the cell, including synthesis of ATP via the oxidative phosphorylation (OXPHOS) system. Normal mitochondrial function requires coordinated expression of two genomes: mitochondria's own genome (mtDNA), which encodes 13 respiratory chain subunits with essential structural and functional roles for the OXPHOS system, and the nuclear genome encoding the remaining ∼80 subunits. The mtDNA-encoded polypeptides are synthesized on mitochondrial ribosomes (mitoribosomes) located in the mitochondrial matrix. Biogenesis, maintenance and regulation of the complex mitochondrial translation apparatus are poorly understood despite its fundamental importance for cellular energy homeostasis. Here, we show that inactivation of the Nsun4 gene, encoding a mitochondrial m5C-methyltransferase, causes embryonic lethality, whereas tissue-specific disruption of Nsun4 in the heart causes cardiomyopathy with mitochondrial dysfunction. By performing sequencing of bisulfite-treated RNA we report that NSUN4 methylates C911 in 12S rRNA of the small ribosomal subunit. Surprisingly, NSUN4 can on its own perform this rRNA modification, whereas interaction with its partner protein MTERF4 is required for assembly of functional ribosomes. NSUN4 thus has dual roles in ribosome maturation and performs an important final quality control step to ensure that only mature mitoribosomal subunits are assembled into functional ribosomes.
| Expression of mtDNA is essential for production of ATP through oxidative phosphorylation in all eukaryotes. Mammalian mtDNA encodes 13 polypeptides that are translated on mitochondrial ribosomes (mitoribosomes) and are essential for oxidative phosphorylation. The mammalian mitoribosomes are composed of the mtDNA-encoded 12S and 16S rRNAs and more than a hundred different nucleus-encoded ribosomal proteins [1], [2]. In bacteria, rRNA modifications are established at precise points of ribosomal assembly and can participate in rRNA processing events as well as in folding and interactions with neighboring proteins [3]. Modifications of rRNA have also been reported in mammalian mitochondria and the best characterized example is the TFB1M-mediated dimethylation of two highly conserved adenosines at the 3′-end of mitochondrial 12S rRNA, which is necessary for assembly of the SSU [4]. Other putative methyltransferases, such as RNMTL1, MRM1 and MRM2, were recently proposed to play an important role for mitochondrial rRNA methylation, but the residues which they modify remain unknown [5]. Studies of hamster mitochondria have shown that the SSU rRNA, in addition to the above-mentioned two dimethylated adenosines (m62A), also contains two cytosine methylations (m4C and m5C) and one methylated uracil (m5U), whereas the LSU rRNA carries three ribose methylated nucleotides, one Um and two Gm [6]. The exact role of each of the modified residues is unclear, but it is worth noting that modified residues tend to concentrate in functionally important regions of ribosomes [7].
We have recently identified a novel mitochondrial rRNA methyltransferase, denoted NSUN4, which belongs to the same family of m5C-methyltransferases as the bacterial enzymes RsmB, RsmF and YccW [8]–[10]. In bacteria, RsmB establishes m5C-methylation on C967 of 16S rRNA, RsmF methylates C1400, C1404 and C1407 of 16S rRNA and YccW is responsible for m5C-methylation on C1962 in 23S rRNA [11]–[14]. The mitochondrial NSUN4 protein, unlike its bacterial counterparts, lacks PUA- or other types of known RNA-interaction domains. Instead, MTERF4 has been shown to form a stable complex with NSUN4 and target this complex to the LSU of the mitoribosome [8]. Inactivation of the Mterf4 gene leads to inhibition of mitochondrial translation and NSUN4 is no longer targeted to the LSU [8]. Moreover, in the absence of MTERF4, the SSU and LSU are present at increased levels but there is no formation of mature mitoribosomes [8]. The crystal structure of the NSUN4/MTERF4 heterodimeric complex has shown that the very stable interaction between both subunits occurs at the carboxy-terminus of MTERF4 and that both MTERF4 and NSUN4 likely are involved in RNA-binding [9], [10]. It is interesting to note that MTERF3, another member of the MTERF family, recently was reported to have an essential role in controlling the assembly of the LSU of the mitoribosome [15]. The remaining two members of the mammalian MTERF family, MTERF1 [16]–[19] and MTERF2 [20], do not seem to be directly involved in mitoribosomal biogenesis, but rather have roles in regulating mitochondrial transcription.
To gain novel molecular insights into the role of NSUN4 in mitoribosomal assembly, we inactivated the Nsun4 gene in the mouse. A germline knockout is embryonic lethal and conditional inactivation of Nsun4 in heart leads to respiratory chain deficiency due to impaired assembly of the mitoribosome with accompanying inhibition of mitochondrial translation. Mapping of m5C residues in 12S and 16S rRNA by sequencing of cDNA generated from bisulfite treated RNA showed that there is a single C5-methylated cytosine at relative position 911 in 12S rRNA in wild-type mice that is lost in Nsun4 knockout mice. Surprisingly, this methylated C911 modification is present in in Mterf4 mutant mice. Our findings thus show that NSUN4 modifies 12S rRNA at position 911 and this modification requires neither its interaction with MTERF4 nor its targeting to the LSU. In contrast, the NSUN4/MTERF4 complex plays an essential role in LSU assembly or maturation independent of the methylation activity of NSUN4. Indeed, using PAR-CLIP we identified two regions of 16S rRNA, which were crosslinked to MTERF4. The presence of a bi-functional NSUN4 protein involved in methylation of 12S rRNA of the SSU and assembly of the mature SSU and LSU establishes a novel mechanism for coordinated maturation of both ribosomal subunits during formation of translation competent mitochondrial ribosomes.
We generated mice carrying a conditional knockout allele of the Nsun4 gene to determine the function of NSUN4 in mitochondria (Figure 1A). Nsun4loxP/+ mice were mated with mice expressing cre-recombinase under the control of the β-actin promoter in order to obtain germline heterozygous knockout mice (Nsun4+/−). Intercrossing of Nsun4+/− mice did not result in any viable Nsun4−/− mice, whereas Nsun4+/+ and Nsun4+/− mice were recovered in mendelian proportions, consistent with embryonic lethality. Disruption of essential mitochondrial genes is frequently associated with embryonic lethality at ∼E8.5 [4], [8], [21], [22] and we therefore proceeded with analysis of embryos at this stage. Mutant embryos (Nsun4−/−) exhibited severely retarded growth with no clearly discernible anatomical structures at E8.5, whereas embryos of the other genotypes appeared normal (Figure 1B). These findings show that NSUN4 has an essential in vivo role.
We next proceeded with inactivating the Nsun4 gene in skeletal muscle and cardiomyocytes by crossing Nsun4loxP/loxP mice with mice expressing cre-recombinase under the control of a creatinine kinase promoter (Ckmm-cre). These conditional knockout mice were viable at birth, but had a much shorter life span than control mice with death before the age of 25 weeks (Figure 2A). We determined the heart to body weight ratio and found a gradual increase with age in the tissue-specific knockout mice (Nsun4loxP/loxP, +/Ckmm-Cre), but not in control mice (Nsun4loxP/loxP) (Figure 2B). Such a progressive cardiomyopathy is commonly seen in mice with heart- and muscle-specific disruption of genes that are essential for oxidative phosphorylation and precedes any detectable phenotypes in skeletal muscle [4], [8], [21], [22]. To assess oxidative phosphorylation capacity in heart, we analyzed the assembly of the five OXPHOS complexes using blue-native PAGE (BN-PAGE) of mitochondrial extracts from control and mutant mice (Figure 2C). This analysis showed that the steady-state levels of assembled OXPHOS complexes were unchanged in 5 weeks-old mice, but were drastically reduced in hearts from 20 weeks-old mutant mice. All complexes containing mtDNA-encoded polypeptides were reduced in abundance, whereas the steady state-levels of complex II, which does not contain any mitochondrially encoded subunits, were unchanged in knockout mice (Figure 2C). Thus, knockout of Nsun4 in the heart causes mitochondrial dysfunction due to impaired biogenesis of the respiratory chain complexes. Western blot analyzes showed drastically decreased levels of NSUN4 in extracts of heart mitochondria already in 5 weeks-old knockout mice (Figure 2D), consistent with efficient knockout of the Nsun4 gene.
Mitochondrial dysfunction in the heart triggers a compensatory increase in mitochondrial mass, which often is accompanied by an increase in the steady-state levels of mtDNA and an increase in mitochondrial transcription [4], [8], [15], [21]. We determined mtDNA levels with semi-quantitative TaqMan PCR analysis and found a moderate increase in the steady-state levels in 20 weeks old mutant mice compared to age-matched controls (Figure S1A). This observation was corroborated by the finding of an increase in the steady-state levels of mitochondrial transcription factor A (TFAM), which typically varies with the levels of mtDNA [4], on western blot analysis (Figure S2B).
Next, we examined the steady-state levels of mtDNA-encoded transcripts using northern blotting and autoradiography (Figure S1B, C). All tested mitochondrial transcripts (rRNAs, tRNAs and mRNAs) were significantly increased in mutant mice (Figure S1B, C) consistent with a general upregulation of mitochondrial transcription. Indeed, pulse-labeling of newly synthesized transcripts in isolated mitochondria, followed by autoradiographic analysis of synthesized transcripts, revealed a strong increase in mitochondrial de novo transcription in mutant mice compared to controls (Figure S2A). Consistent with these data, western blot analyses showed an increase in the steady-state levels of the mitochondrial transcription factor TFB2M and TFAM in mitochondrial extracts (Figure S2B). Additionally, mutant mitochondria had increased steady-state levels of LRPPRC (Figure S2C), a stability factor for mitochondrial mRNAs, whose steady-state levels correlate with the levels of mitochondrial mRNAs [22]. Thus, knockout of Nsun4 in the heart leads to increased de novo synthesis and increased steady-state levels of mitochondrial transcripts.
We have previously shown that NSUN4 is targeted to the LSU through its physical interaction with MTERF4 and that absence of MTERF4 leads to defective ribosomal assembly and inhibition of mitochondrial translation [8]. We therefore proceeded to determine if mitochondrial translation also was inhibited in Nsun4-knockout hearts. De novo labeling of mitochondrial translation products, followed by gel electrophoresis and autoradiography, revealed a strong inhibition of translation in mutant mitochondria (Figure 3A). Consistent with this, the steady-state levels of the mitochondrially encoded ATP8 and COXII proteins were decreased in knockout hearts (Figure 3B).
Next, we analyzed the assembly of the mitoribosome by ultracentrifugation of control and mutant mitochondrial extracts through linear-density sucrose gradients (Figure 3C). Sedimentation of the SSU and LSU was determined by immunological detection of the protein markers MRPS15 and MRPL13, respectively. In control extracts, we observed a typical sedimentation pattern of the separate SSU (28S) and LSU (39S) as well as of co-sedimentation of both subunits in fully assembled (55S) ribosomes [4], [8], [22]. In contrast, knockout mice exhibited an accumulation of assembled SSU and LSU without a corresponding increase in assembled ribosomes (Figure 3C). Furthermore, the steady-state levels of proteins from the LSU and SSU were increased in knockout mice (Figure 3D). We have previously reported that when ribosomal assembly and/or function are not inhibited, an accumulation of both the SSU and LSU coincides with accumulation of assembled ribosomes [22]. The data we present here therefore indicate that lack of NSUN4 inhibits the association between the SSU and LSU to form functional ribosomes.
Previous reports have shown that NSUN4 is an m5C-methyltransferase proposed to methylate an unknown residue in 16S rRNA [8]–[10]. However, studies on hamster mitochondria have shown that the 17S rRNA of the LSU does not contain any C5-methylated residues, unlike 13S rRNA of the SSU, which contains one m4C and one m5C at relative position 911 and 913, respectively (relative to tRNAPhe) [6]. We revisited these data by attempting to map all m5C residues in mouse 12S and 16S rRNA by using sequencing of cDNA generated from bisulfite treated RNA. The bisulfite method is based on the chemical conversion of non-methylated cytosine to uracil, whereas m5C residues are resistant to this treatment and remain unchanged in the treated RNA [23]. Other modifications, like m4C can also be detected, but with a lower frequency. Using this approach we probed the methylation status of heart mitochondrial rRNA from control mice (Figure 4A) and detected a single C5-methylated cytosine residue at position 911 in 12S rRNA (methylation rate 87.5±0.9%). Next, we assessed the methylation status of heart mitochondrial rRNA from Nsun4 knockout mice and found that methylation on C911 was now essentially absent (methylation rate 2.5±2.4%; Figure 4B). Alignment of the mouse and hamster sequences revealed that m5C911 in mouse 12S rRNA corresponds to the previously reported m5C913 in hamster 13S rRNA (Figure S3A). Thus, NSUN4 methylates C911 in 12S rRNA, and this modification is likely the only m5C-methylation present in mammalian mitochondrial rRNAs.
In addition to m5C911, there was also a cytosine at position 909 (C909) in mouse 12S rRNA that showed partial resistance to bisulfite treatment (methylation rate 23.6±1.7%; Figure 4A). The corresponding nucleotide in hamster 13S rRNA has previously been reported to harbor an m4-methylation [6]. The m4C909 modification in mouse 12S rRNA was not affected (methylation rate 21.1±3.4%) in the absence of NSUN4 (Figure 4B), thus showing that the m5C911 modification is specifically generated by NSUN4. Additionally, we tested if dimethylation of the highly conserved adenosines A1006 and A1007 (m62A1006 and m62A1007) in 12S rRNA is affected in the absence of NSUN4. The two m62-methylations are established by TFB1M during the maturation of the SSU [4]. Assembly defects of the mitoribosomal SSU, resulting from loss of m5C, could possibly affect these adenosine methylation modifications. However, primer extension assays on RNA from control and mutant hearts showed that methylation of these adenosine residues in Nsun4 knockout mice was indistinguishable from controls (Figure S3B, C).
We proceeded to test if presence of MTERF4 was necessary for methylation of 12S rRNA in vivo by deep sequencing of bisulfite treated RNA from mice with a tissue-specific knockout of Mterf4 at different ages [8]. In the absence of MTERF4, C911 exhibited methylation rates similar to those in control mice (Figure 4C) showing that interaction between NSUN4 and MTERF4 or targeting of the NSUN4/MTERF4 complex to LSU are dispensable for methylation of C911 in 12S rRNA. Moreover the monosome assembly defect observed in MTERF4 mutant mice cannot be explained by lack of m5C-methylation. Instead, the NSUN4/MTERF4 complex must play another role in LSU assembly.
In view of the finding that NSUN4 can methylate 12S rRNA in the absence of MTERF4, we investigated whether an assembled LSU was needed for methylation of C911 of 12S rRNA. To this end, we performed deep sequencing of bisulfite-treated RNA isolated from Mterf3-knockout mouse hearts, which were previously shown to lack an assembled LSU [15]. Methylation of C911 in Mterf3 mutant mice was only mildly reduced in comparison with controls, thus indicating that methylation of 12S rRNA can occur independently of the presence of an assembled LSU (Figure S4). Taken together our data indicate that methylation of 12S rRNA can occur independently of the targeting of NSUN4 to the LSU and that it represents a modification pathway for ensuring proper SSU maturation.
We performed photoactivatable ribonucleoside-enhanded cross linking and immunoprecipitation (PAR-CLIP) experiments in HeLa cells expressing either MTERF4-FLAG or NSUN4-FLAG to identify interacting regions on 16S and 12S rRNA. Using this approach, we were able to detect RNA fragments specifically cross-linked to regions of 12S and 16S rRNA. The number of interacting RNA fragments was small, which likely reflects the fact that only a small amount of NSUN4/MTERF4 is associated with the LSU under normal conditions [8]. This is expected as NSUN4/MTERF4 is involved in specific transient steps of ribosomal biogenesis and therefore is unlikely to be associated with every mature mitochondrial ribosome. Additionally, we performed CLIP experiments using cells expressing a trap-mutant of NSUN4 (NSUN4C258A-FLAG), which is able to form a covalent crosslink with the RNA substrate subjected to site-specific methylation [24], [25]. With this latter approach we identified 16 RNA fragments mapping to 12S rRNA (Figure 5A and Table S1). Five of these fragments of different lengths encompassed the region containing C911. The remaining sequences were distributed along 12S or 16S rRNA and likely represent additional specific or unspecific interactions with the fully assembled ribosome. Finally, MTERF4-FLAG was reproducibly found to be crosslinked to two 16S rRNA regions in two independent PAR-CLIP experiments (Figure 5B and Table S2). This finding is rather unexpected and suggests that the MTERF4/NSUN4 complex may interact with two different LSU regions simultaneously. Unfortunately, there is no atomic resolution structure the mitochondrial ribosome and we can therefore not exactly pinpoint the location of these regions. The RNA sequences identified by the CLIP experiments are predicted to form mixed double- and single-stranded structures [26] and we therefore tested whether the MTERF4/NSUN4 complex has any structural preferences when binding RNA. We performed gel shift experiments with short RNA fragments incubated with the recombinant NSUN4/MTERF4 complex or the recombinant NSUN4 protein. The NSUN4/MTERF4 complex binds RNA fragments with a pronounced double-stranded conformation (Figure S5A; right panels), whereas it poorly binds single-stranded RNA fragments derived from the same double-stranded RNA fragment (Figure S5A; left panels). These findings indicate that NSUN4/MTERF4 preferentially binds double-stranded RNA. In view of our finding that NSUN4 is able to methylate C911 independent of its interaction with MTERF4, we also tested if NSUN4 directly can bind an RNA fragment (ds12S: 878–949) containing the methylation site and indeed found such a specific interaction (Figure S5B), showing that NSUN4 can bind RNA on its own in the absence of MTERF4.
We demonstrate here that NSUN4 unexpectedly is a dual function protein. On the one hand, NSUN4 can alone methylate C911 in 12S rRNA and on the other hand NSUN4 in complex with MTERF4 is targeted to the LSU to regulate mitoribosomal assembly. The role for NSUN4 in rRNA modification can be uncoupled from its role in mitoribosomal assembly as Mterf4 knockout mice cannot assemble functional ribosomes despite the presence of the methylated C911 residue. Also other methyltransferases have been suggested to have dual functions. The yeast methyltransferase, Dim1p, which generates the highly conserved m62Am62A on cytoplasmic SSU rRNA, also has a role in rRNA processing [27]. With the use of different dim1 alleles both functions can be uncoupled, revealing that dimethylation is dispensable for growth whereas the role in rRNA processing is essential [28]. The bacterial ortholog of Dim1p, KsgA, has also been suggested to have dual roles in SSU biogenesis by establishing the m62Am62A modification of the SSU rRNA and by sterically blocking access for binding of the LSU and the initiation factor 3 to regulate the formation of assembled bacterial ribosomes [29], [30]. Moreover, conformational changes upon binding of the methyltransferase RsmC to its substrate, G1207, were proposed to protect rRNA against misfolding during SSU assembly [31]. Similar to the examples above, we hypothesize that the NSUN4/MTERF4 methylation complex plays a critical structural role during LSU assembly through its affinity for double-stranded rRNA. The NSUN4/MTERF4 complex, when bound to the LSU, may inhibit the formation of assembled ribosomes by two possible, but not necessarily mutually exclusive, mechanisms: i) by physically preventing interaction between both ribosomal subunits and ii) by occluding a putative 16S rRNA fragment required for subunit interaction or LSU activation. Ribosomal assembly in bacteria is characterized by continuous structural rearrangements of rRNA, which sometimes transit through misfolded states stabilized through interactions with ribosomal proteins or ribosomal assembly factors [32]. Later, these states are resolved through refolding of rRNA, release of the ribosomal assembly factors or both. It is possible that release of the NSUN4/MTERF4 complex results in conformational changes in 16S rRNA, which in turn can activate the LSU to enable and stabilize its interaction with the SSU. Thus the NSUN4/MTERF4 complex may also play an important structural role by preventing premature entry into translation.
The exact mechanistic function of m5-methylation on cytosine in RNA is generally unclear. An m5-methylation does not impair base pairing nor does it appear to induce conformational changes in the modified nucleotide. However, studies on tRNAPhe and tRNAVal(AAC) have suggested that m5C, likely in cooperation with other modified residues, is involved in the stabilization of tRNA folding under physiological conditions [33]–[35]. Similarly m5C911 may cooperate with the nearby m4C909 and other rRNA modifications in stabilization of 12S rRNA folding, thereby facilitating mitoribosomal assembly. Detailed crystallographic analyses on mitoribosomal structure are lacking, which hinders a detailed understanding of the importance of C911 for mitoribosomal assembly.
Based on the analyses of our mutant mouse strains, we have attempted to build a preliminary model for the late stages of mitoribosomal subunit assembly and the formation of translation competent assembled ribosomes (Figure 5C). According to this model, NSUN4 methylates C911 in 12S rRNA of SSU (Figure 5C: step 1). This step does not require interaction with MTERF4, suggesting that NSUN4 may be targeted by either interacting directly with the rRNA substrate or by interacting with an unknown SSU protein. Other modifications, like m62A1006 and m62A1007 by TFB1M, are established independently of m5C911 and stabilize the newly formed SSUs. NSUN4/MTERF4 complexes are incorporated into the LSU during its assembly (Figure 5C: step 2). LSU-bound NSUN4/MTERF4 complexes prevent partially assembled LSUs from forming abortive 55S complexes with available SSUs. Finally, LSU-bound NSUN4/MTERF4 complexes are dissociated from the LSU to enable its interaction with the SSU to form translation-competent ribosomes (Figure 5C: step 3).
In conclusion, we show here that NSUN4 is a dual function protein involved in coordinating ribosomal biogenesis in mammalian mitochondria by methylation C911 on the 12S rRNA and by interacting with MTERF4 to regulate the assembly of the two ribosomal subunits. We propose that NSUN4 functions as a quality control step late in ribosomal biogenesis to ensure that only mature SSUs and LSUs are assembled into functional mitoribosomes.
This study was performed in strict accordance with the recommendations and guidelines of the Federation of European Laboratory Animal Science Associations (FELASA). The protocol was approved by the “Landesamt für Natur, Umwelt und Verbraucherschutz Nordrhein-Westfalen”.
The Nsun4loxP/+ mice, that have exon II flanked by loxP-sites, were generated at TaconicArtemis GmbH (Cologne, Germany). The Nsun4loxP/+ mice were mated with mice ubiquitously expressing cre-recombinase to generate heterozygous knockout mice (Nsun4+/−). Heart- and skeletal muscle-specific knockout mice were generated as described previously [4], [8], [21], [22]. Nsun4loxP/loxP mice were crossed with transgenic mice expressing cre-recombinase under the control of the muscle creatinine kinase promoter (Ckmm-cre). The resulting double heterozygous mice (Nsun4loxP/+, +/Ckmm-cre) were mated with Nsun4loxP/loxP mice to generate tissue-specific knockout (Nsun4loxP/loxP, +/Ckmm-cre) and control (Nsun4loxP/loxP) mice.
RNA for bisulfite treatment was isolated using the miRNeasy Mini Kit (QIAGEN) and treated with TURBO DNase (Ambion) to remove mitochondrial DNA. Bisulfite treatment was performed as described previously [34]. Treated RNA was converted to cDNA and sequenced using the Illumina platform. Reads were submitted to fastqc for quality analysis. Fastq files were clipped with fastx_trimmer (fastx toolkit, http://hannonlab.cshl.edu/fastx_toolkit/) to remove the first 4 and last 35 bases, which were composed of lower quality bases at the start and end of the sequences. Clipped paired reads were aligned to the mouse reference genome with the Bismark software [36]. Bismark performs bisulfite mapping and allows calculation of methylation rates via perl scripts. Bisulfite reads are transformed into a C-to-T and G-to-A version (reverse strand) which are aligned to equivalently pre-converted forms of the reference genome using four parallel instances of the Bowtie aligner [37]. Duplicate sequences were removed and data were reevaluated. Methylation rates were calculated using Perl scripts from the Bismark website, (http://www.bioinformatics.babraham.ac.uk /projects/bismark/).
BN-PAGE experiments were performed using NativePAGE Novex Bis-Tris Gel System (Invitrogen) according to the manufacturer's recommendations. Heart mitochondria (40 µg) were solubilized with NativePAGE Sample Buffer containing 1% dodecylmaltoside (DDM). After 20 minutes on ice, samples were centrifuged (30 min, 16000×g, 4°C). Supernatants were supplemented with NativePAGE G250 Sample Additive and fractionated through 4–16% NativePAGE Novex Bis-Tris Gel. Respiratory chain complexes were transferred onto PVDF membrane and detected using specific antibodies. Pulse labeling of mitochondrial transcription products was performed in isolated mitochondria according to [38], [39]. In organello translation was performed as previously described [22], [40].
Assembly of 28S and 39S ribosomal subunits as well as 55S monosomes was assayed using ultracentrifugation through a 10–30% linear-density sucrose gradient as described previously [22].
For quantification of mtDNA, total DNA was isolated from heart tissue using the DNeasy Blood & Tissue Kit (QIAGEN). Semiquantitative RT-PCR was carried out on 4 ng of total DNA in a 7900HT Real Time PCR system (Applied Biosystems), using TaqMan probes specific for the CoxI and 18S genes (Applied Biosystems).
Purification of human recombinant NSUN4/MTERF4 complex for gel shift experiments was done according to [9]. RNA fragments were purchased from Thermo Scientific or Eurofins MWG Operon. Gel shift experiments were carried out using varying quantities of NSUN4/MTERF4, and in the presence of 40 ng RNA essentially as described in [15]. Fragment ds12S:878–949, which carries C911 in human 12S rRNA, contains three G and three C residues at the 5′- and 3′-end, respectively, to prevent formation of single-stranded termini. See Table S3 for sequences of the RNA fragments used for gel shift experiments.
Sequence alignments were performed using Clustal Omega [41] at default settings and were visualized using GeneDoc at shade level 1 [42]. Sequence accession numbers are given in figure legends.
RNA for northern blot analysis was isolated using Trizol Reagent (Invitrogen) and resuspended in formamide (Ambion). For detection of mitochondrial transcripts, 1–2 µg of total RNA was denatured in NorthernMax-Gly Sample Loading Dye (Ambion), separated in 1.2% agarose gels containing formaldehyde (SIGMA-Aldrich) and transferred to Hybond-N+ membranes (GE Healthcare). DNA probes, for the detection of mitochondrial mRNAs and rRNAs, were radiolabeled with α-32P-dCTP using the Prime-It II random primer labeling kit (Stratagene). For detection of tRNAs, oligonucleotides were labeled with γ-32P-ATP using T4-polynucleotide kinase (NEB).
Rabbit polyclonal antisera were used for the detection of TFAM, ATP8 and COXII [43]. A rabbit polyclonal antiserum was used for detection of LRPPRC [22]. Affinity purified rabbit polyclonal antibodies were used for the detection of TFB2M [4], MRPS15 [4] and MRPL13 [8]. Monoclonal antibodies against VDAC were purchased from Calbiochem. Immunodetection of NDUFA9, SDHA, UQCRC2, COX IV and ATP5A1 was performed with monoclonal antibodies from MitoSciences. Monoclonal antibodies against mouse NSUN4 were generated by AbD Serotec.
The HeLa cells stably transfected with pTreTight-hMterf4-FLAG DNA construct are described previously [8]. HeLa cell clones stably transfected with pTreTight-hNsun4-FLAG and pTreTight-hNSUN4C258A-FLAG were generated as described [8].
Photoactivatable-Ribonucleoside-Enhanced Crosslinking and Immunoprecipitation (PAR-CLIP) was performed by combining and adapting previously described methods [44], [45]. HeLa cells transfected with doxycycline-inducible MTERF4-FLAG, NSUN4-FLAG and NSUN4C258A-FLAG containing DNA constructs were induced with 1 µg/ml doxycycline and grown in the presence of 4-thiouridine (4-SU, final concentration of 100 µM) for 14 hours. The cells from 20 culture dishes (500 cm2 each) were washed with PBS, crosslinked at 365 nm, collected and frozen in liquid nitrogen. Next, cells were lysed, treated with RNase T1 and used for immunoprecipitation experiments with ANTI-FLAG M2 magnetic beads according to the manufacturer's recommendations (Sigma). The samples were further processed according to [44].
|
10.1371/journal.pgen.1001381 | Genomic Instability, Defective Spermatogenesis, Immunodeficiency, and Cancer in a Mouse Model of the RIDDLE Syndrome | Eukaryotic cells have evolved to use complex pathways for DNA damage signaling and repair to maintain genomic integrity. RNF168 is a novel E3 ligase that functions downstream of ATM,γ-H2A.X, MDC1, and RNF8. It has been shown to ubiquitylate histone H2A and to facilitate the recruitment of other DNA damage response proteins, including 53BP1, to sites of DNA break. In addition, RNF168 mutations have been causally linked to the human RIDDLE syndrome. In this study, we report that Rnf168−/− mice are immunodeficient and exhibit increased radiosensitivity. Rnf168−/− males suffer from impaired spermatogenesis in an age-dependent manner. Interestingly, in contrast to H2a.x−/−, Mdc1−/−, and Rnf8−/− cells, transient recruitment of 53bp1 to DNA double-strand breaks was abolished in Rnf168−/− cells. Remarkably, similar to 53bp1 inactivation, but different from H2a.x deficiency, inactivation of Rnf168 impairs long-range V(D)J recombination in thymocytes and results in long insertions at the class-switch junctions of B-cells. Loss of Rnf168 increases genomic instability and synergizes with p53 inactivation in promoting tumorigenesis. Our data reveal the important physiological functions of Rnf168 and support its role in both γ-H2a.x-Mdc1-Rnf8-dependent and -independent signaling pathways of DNA double-strand breaks. These results highlight a central role for RNF168 in the hierarchical network of DNA break signaling that maintains genomic integrity and suppresses cancer development in mammals.
| The repair of DNA damage is fundamental as illustrated by the many human syndromes, immunodeficiencies, and cancers associated with defects in DNA damage signaling and repair. RIDDLE syndrome, caused by mutations of the human RNF168, is a novel hereditary disease clinically characterized by radiosensitivity, immunodeficiency, dysmorphic features, and learning difficulties. RNF168 is an E3 ligase that modifies histones and chromatin structure at sites of DNA breaks. In this study, we show that Rnf168 deficiency in mice leads to increased radiosensitivity, immunodeficiency, and defective spermatogenesis. Additionally, dual inactivation of Rnf168 and p53 leads to increased cancer risk. Collectively these data demonstrate important and broad physiological functions for Rnf168.
| DNA damage checkpoint signaling and DNA repair pathways are key elements of the DNA damage response (DDR) and are critical for the maintenance of genomic integrity [1]–[3]. Mammalian cells constantly experience DNA damage as a result of exogenous exposure to ionizing radiation (IR), ultraviolet light (UV), chemical compounds, and radical oxygen species, as well as endogenous insults due to DNA replication errors. In addition, double-strand DNA breaks (DSBs) are also programmed to occur during immune-receptor rearrangements and meiosis.
Mutations of genes involved in DNA damage signaling or repair can lead to many diseases including neurodegenerative diseases, immunodeficiency and cancer, underlining the importance of these processes [1], [4], [5]. Among the various types of DNA damage, DSBs are the most serious and require elaborated networks of proteins to signal and repair the damage. In mammalian cells, DSBs are initially recognized by the Mre11/Rad50/NBS1 (MRN) complex that induces the activation and recruitment of the ataxia-telangiectasia-mutated (ATM) kinase to the break sites [6]–[8]. At the flanking sites of DSBs, H2A.X, a variant of the histone H2A, is rapidly phosphorylated at the serine 139 residue (γ-H2A.X). Phosphorylation of H2A.X is mediated by activated ATM, which itself is phosphorylated at serine 1981 (phospho-ATM-S1981), or alternatively by two other phosphoinositide 3-kinase like kinases (PIKKs), namely the ataxia telangiectasia and Rad3 related (ATR) and the DNA-dependent protein kinase catalytic subunit (DNA-PKcs). Active ATM also phosphorylates a number of other proteins including the structural maintenance of chromosomes 1 (SMC1), the Nijmegen breakage syndrome protein 1 (NBS1), the checkpoint kinase 2 (Chk2), the breast cancer 1- early onset (BRCA1) and the mediator of DNA damage checkpoint protein-1 (MDC1). Subsequent to its phosphorylation, MDC1 binds to γ-H2A.X via its tandem BRCA1 C-terminal (BRCT) domains, and recruits additional active ATM to sites flanking the DSBs [7], [9], [10]. MDC1 also associates with MRN complex through its Ser-Asp-Thr (SDT) repeats and the fork-head associated (FHA) domain of NBS1. Furthermore, through its C-terminus, NBS1 also interacts with ATM [11]–[15]. As such, MDC1 bridges the interaction of MRN to γ-H2A.X and ATM. Enrichment of ATM-MDC1 and ATM-MRN at the break sites further amplify the phosphorylation of H2A.X and triggers the recruitment of other DNA damage response proteins to the DSB flanking regions [16]. The three conserved T-Q-X-F clusters between 698 to 800 amino acids of MDC1 are also phosphorylated by ATM. The threonine-phosphorylated MDC1 has been shown to interact with the FHA domain of RING finger protein 8 (RNF8), thus recruiting this E3 ligase to the sites of damage. Subsequently, RNF8, along with the E2 ubiquitin-conjugating enzyme UBC13, mediate lysine 63 (K63)-linked ubiquitylation of the histone H2A at the flanking regions of DSBs [17]–[19]. Such ubiquitylated form of H2A interacts with the MIU2 (motif interacting with ubiquitin 2) domain of RNF168 and recruits this E3 ligase to DSB sites, allowing it to further ubiquitylate surrounding H2A [20]–[22]. This likely amplifies the modification of chromatin structure at regions adjacent to DSBs, and facilitates the recruitment of tumor suppressor p53 binding protein 1 (53BP1). In addition, RAP80 selectively binds to the K63-polyubiquitin chains on H2A via its tandem ubiquitin interaction motifs (UIMs) [23], and acts as a bridge to recruit BRCA1 to the regions of DSBs. In sum, DSBs signaling is a highly regulated process, in which RNF168 plays a major role through its contribution to the recruitment of various downstream DDR proteins.
RNF168 is mutated in RIDDLE syndrome, which is characterized by radiosensitivity, immunodeficiency, dysmorphic features and learning difficulties [24]. RNF168 contains a RING finger domain and two MIU domains. The RING finger domain of RNF168 is critical for its ubiquitin E3 ligase activity, whereas its MIU2 domain mediates its binding to ubiquitylated H2A [20]–[22]. Knockdown of RNF168 in human cells significantly impaired the formation of 53BP1, RAP80 and BRCA1 ionizing radiation induced foci (IRIF). While recent studies have revealed a role for RNF168 in ubiquitylating H2A during DSB signaling [20]–[22], the full physiological functions of RNF168 are not understood.
Here, we report the generation of a mouse model for RNF168 deficiency. Rnf168−/− mice are viable. Consistent with the clinical features of RIDDLE syndrome, Rnf168−/− mice are immunodeficient, and their cells show increased radiosensitivity. Similar to 53bp1 deficiency, long-range V(D)J recombination of T-cell receptor (TCR)δ is also impaired in Rnf168−/− thymocytes, while aberrant long insertions have been observed at class switch junctions of Rnf168−/− B-lymphocytes. Moreover, in contrast to the transient recruitment of 53bp1 to DSBs in H2a.x-, Mdc1- and Rnf8-deficient mouse embryonic fibroblasts (MEFs), 53bp1 recruitment to DSB sites is completely abolished in Rnf168-deficient MEFs. Our data also demonstrate novel roles of Rnf168 in spermatogenesis, maintenance of genomic integrity and cancer, and therefore, further highlight the other important physiological functions of this molecule.
To investigate the physiological role of Rnf168, we generated Rnf168 deficient mice using two different gene trap embryonic stem cell (ES) clones (156B6 and 405F11). Gene trap of Rnf168 in the two ES clones was confirmed by Southern blot, PCR and sequencing of the genomic site of the retrovirus integration (Figure S1A and S1B). RT-PCR revealed the disruption of full length Rnf168 transcript in the two Rnf168 gene trap ES clones (Figure S1C). The loss of Rnf168 protein was also confirmed in Rnf168−/− MEFs and splenocytes by immunoblotting (Figure S1D and S1E).
Intercrossing of Rnf168+/− mice obtained with 156B6 and 405F11 ES clones indicated that Rnf168−/− mice were viable and born at the expected Mendelian ratio (Table S1). Rnf168−/− mice did not exhibit any gross developmental defects and were indistinguishable from their wildtype (WT) littermates. Collectively, these data indicate that the E3 ligase Rnf168 is not required for embryonic and postnatal development.
Mutation of the RNF168 gene is associated with RIDDLE syndrome [22], [24]. To determine whether Rnf168 deficiency in mice confers increased sensitivity to DNA damage, primary lymphocytes isolated from peripheral lymph nodes (LN) of Rnf168−/− mice and WT littermates were subjected to various doses of IR or UV. Lymphocyte viability was examined using 7-aminoactinomycin D (7AAD) and flow cytometry. These studies demonstrate a significant increase in the sensitivity of Rnf168−/− lymphocytes to IR and UV treatment (Figure 1A).
DNA damage signaling plays essential roles in the activation of checkpoints and cell cycle progression [25]. Examination of the effect of Rnf168 inactivation on cell cycle progression using G1/S synchronized MEFs showed no difference between Rnf168−/− and WT MEFs (Figure S2A).
Atm−/−, H2a.x−/−, Mdc1−/−, Rnf8−/− and 53bp1−/− cells exhibit a defect in the early G2/M checkpoint [10], [17]–[19], [26]–[28]. Atm and 53bp1-deficient cells display late G2/M accumulation and inactivation of 53bp1 results in a mild impairment of the intra-S phase checkpoint [26], [28]. Therefore, we examined the effects of Rnf168 deficiency on cell cycle checkpoints. Immunostaining of irradiated Rnf168−/− and WT MEFs with anti-phospho-histone H3 (pHH3), and co-staining with propidium iodide (PI) demonstrated a higher proportion of irradiated Rnf168−/− MEFs (35.2±1.8%) have progressed to M phase compared to WT MEFs (10.7±4.3%, p<0.05) (Figure 1B). Next, the effect of irradiation on cell cycle progression of Rnf168−/− MEFs and WT controls was examined using the Bromodeoxiuridine (BrdU) and PI assay. The percentage of Rnf168−/− cells in S phase 3 hours post-IR (10 Gy) was slightly increased compared to WT cells (Figure 1C). These data suggest that similar to 53bp1−/− MEFs [24], [28], Rnf168−/− MEFs also have a mild defect in the intra-S phase checkpoint. In addition, cell cycle analyses of Rnf168−/− and WT MEFs 18 hours post-IR indicated that a higher proportion of the Rnf168−/− MEFs is found at the G2/M phase (Figure 1D). Therefore, similar to H2a.x, Mdc1, Rnf8 and 53bp1, Rnf168 is dispensable for the activation of the G1/S checkpoint, but is important for enforcing the G2/M DNA damage checkpoint.
Through ubiquitylation of histone H2A, RNF168 plays an important role in the recruitment of 53BP1 to the sites of DNA breaks downstream of H2A.X, MDC1 and RNF8 [20]–[22]. Although H2a.x, Mdc1 and Rnf8 are each essential for the stable ‘retention’ of 53bp1 foci at DSB sites, studies of H2a.x−/−, Mdc1−/− and Rnf8−/− MEFs demonstrate that these cells can still transiently recruit 53bp1 to form IRIF [29]. In addition, a partial accumulation of 53bp1 at DSBs has been also observed in activated Rnf8−/− B-cells post-irradiation [30]. To investigate the effect of Rnf168 deficiency on the recruitment of 53bp1 to sites of DSBs, Rnf168−/−, Rnf8−/− and WT MEFs were subjected to 5 Gy of IR, followed by a time course analysis of 53bp1 foci formation by immunofluorescence microscopy. In contrast to H2a.x−/−, Mdc1−/− and Rnf8−/− MEFs [5], no 53bp1 IRIF were observed in Rnf168−/− MEFs post-IR (Figure 2A and 2B and Figure S3A). Transfection of Rnf168−/− MEFs with GFP-tagged RNF168 expression constructs fully restored formation of 53bp1 IRIF (Figure 2C). In addition, radiosensitivity of Rnf168−/− MEFs was also rescued by expression of Rnf168 (Figure 2D). Rnf168 deficiency also significantly impaired Brca1 recruitment to sites of DSBs (Figure 2E and Figure S3B). These data indicate an important role for Rnf168 in the recruitment of both 53bp1 and Brca1 to the DNA break sites.
To examine whether Rnf168 deficiency affects the activation of Atm pathway, WT and Rnf168−/− thymocytes were irradiated ex vivo. One hour post-IR, the protein levels of total Atm, as well as the expression and phosphorylation levels of its downstream targets, were assessed by immunoblot (IB) analysis. The protein levels of Atm were similar between Rnf168−/− and WT thymocytes in response to IR (Figure 3A). Phosphorylation levels of the Atm substrates Smc1 (serine 966), Nbs1 (serine 343), Chk2 (threonine 68) and p53 (serine 15), were similar or slightly higher in Rnf168−/− thymocytes compared to WT controls (Figure 3A and 3B). IR also induced a slightly higher phosphorylation level of Chk2 (threonine 68) and p53 (serine 15) in Rnf168−/− MEFs compared to WT controls (Figure 3C). We also observed increased γ-H2a.x and Mdc1 foci formation in untreated and irradiated Rnf168−/− MEFs compared to WT controls (Figure 3D and 3E, Figure S3C and S3D). In addition, γ-H2a.x- and Mdc1- IRIF remained visible longer in irradiated Rnf168−/− MEFs compared to WT controls, suggesting defective DSB repair in the absence of Rnf168. As both Atm and DNA-PK can phosphorylate H2a.x at serine 139 in response to IR [31], we examined the effects of inhibition of DNA-PK on γ-H2a.x IRIF in WT and Rnf168−/− MEFs. The levels of γ-H2a.x IRIF in irradiated WT and Rnf168−/− MEFs were not affected by DNA-PK inhibitors (Figure 3F). Collectively, these data indicate that Rnf168 is dispensable for the activation of ATM-Chk2-p53 pathway and that Atm signaling pathway is unaffected in Rnf168−/− cells.
H2a.x, Mdc1 and Rnf8 are required for DDR upstream to Rnf168, and are essential for male fertility [10], [27], [30], [32]–[35], whereas 53bp1 homozygous mutant males are fertile [28]. We therefore investigated the fertility of Rnf168−/− mice. Rnf168−/− males and females were fertile at 8 weeks of age, albeit their intercrosses produced smaller litters (6.0 pups±0.6) compared to WT littermates (9.8 pups±0.7, p<0.05) (Figure 4A). In contrast, 12-month-old Rnf168−/− males showed either a significantly reduced fertility compared to WT littermates or were completely infertile (Figure 4B). Testicular sizes were comparable between 8-week-old Rnf168−/− and WT littermates, whereas at 12 months of age, Rnf168−/− males showed reduced testicular size compared to WT littermates (Figure 4C). Histological examination of testes from 8-week-old Rnf168−/− males indicated no abnormalities compared to WT littermates (Figure 4D). However, testes from aged Rnf168−/− mice displayed signs of testicular degeneration and atrophy as evidenced by the reduced number or lack of spermatids in seminiferous tubules, the increased vacuolization of germ cells and the prominent Leydig cells compared to WT littermates (Figure 4D). The amount of sperm was also drastically decreased in the epididymis of 12-month-old Rnf168−/− males compared to controls (Figure 4E). All together, these data indicate that Rnf168 is required for spermatogenesis in an age-dependent manner.
The proportion of B and T lymphocytes in one RIDDLE syndrome patient was reported to be within the normal range; however, this patient was immunodeficient and exhibited low levels of serum immunoglobulin G (IgG) [24]. To investigate the effect of Rnf168 deficiency on lymphocyte maturation, we first examined bone marrow (BM) from 6–8-week-old Rnf168−/− mice and their WT littermates. BM cellularity and the number of Pro- and Pre-B-cells were not affected by Rnf168 deletion (Figure S4A). Similarly, examination of the number of splenocytes, LN cells and their subpopulations also showed no significant difference between Rnf168−/− mice and WT controls (Figure S4B and S4C).
The repair of programmed DSBs is essential for class switch recombination (CSR) of immunoglobulins [36], [37]. Failure to initiate, signal, or repair these programmed DSBs will lead to defective CSR and immunodeficiency. To assess the role of Rnf168 in Ig heavy chain (Igh) class switching in vivo, we evaluated the concentrations of serum Ig isotypes in Rnf168−/− mice and WT littermates. Total serum IgG concentrations were significantly lower in 6–8-week-old Rnf168−/− mice and in 9–12-month-old Rnf168−/− mice compared to age-matched WT littermates (Figure 5A). To further examine the in vivo role of Rnf168 in CSR, we determined the portion of IgG1- and IgG3-expressing B-cells in Peyer's patches from Rnf168−/− mice and WT littermates. B-cells in the Peyer's patches constantly encounter the gut flora, and, therefore, they exhibit elevated levels of CSR [38]. Peyer's patches from Rnf168−/− mice showed decreased representation of IgG1- and IgG3-expressing B-cells compared to WT littermates (p<0.05; Figure 5B and 5C).
We further examined the role Rnf168 plays in CSR for various Igh isotypes using purified splenic B-cells from Rnf168−/− mice and WT littermates. When B-cells were activated in culture with LPS±IL4, Rnf168-deficient B-cells displayed a significantly reduced secretion of IgG1, IgG2a and IgG3 compared to WT controls (Figure 5D). We also stimulated in vitro purified splenic B-cells with anti-CD40 or LPS in combination with IL-4, and examined the levels of CSR to IgG1. To analyze the efficiency of CSR in each cell division, we used FACS analysis to assess the expression of surface IgG1 on CFSE-labeled stimulated B-cells (Figure 5E and 5F, Figure S4D and S4E). The portion of B-cells expressing surface IgG1 post activation was significantly reduced in Rnf168−/− mice (13.2±1.1%) compared to WT controls (28.3±0.8%, p<0.05) (Figure 5E). Examination of the CFSE dilution profiles indicated similar proliferation levels of mature B-cells from WT and Rnf168−/− mice (Figure 5F and Figure S4E). In addition, the efficiency of CSR in each cell division was decreased in activated Rnf168−/− B-cells compared to WT B-cells (Figure 5F). These data indicate that CSR defects in Rnf168−/− B-cells were directly due to the loss of Rnf168 function, rather than a result of proliferation defects of the mutant B-cells.
The RING finger domain of RNF168 is critical for its E3 ligase activity, while its MIU2 domain is important for its recruitment to DSB sites [20], [22]. To further explore the role of Rnf168 in CSR, we generated deletion or point mutant Rnf168 constructs using the pMSCV retroviral vector. Purified B-cells from WT or Rnf168−/− mice were activated with LPS and IL-4 and then retrovirally transduced with the various Rnf168-expression constructs. These cells were then examined for functional rescue of the impaired CSR by WT or mutants Rnf168 (Figure 5G, 5H and Figure S4F). Rnf168−/− B-cells, infected with pMSCV expressing either WT Rnf168 or Rnf168 with its MIU1 domain deleted (ΔMIU1), rescued IgG1 CSR to the level of WT B-cells. However, Rnf168−/− B-cells infected with retroviruses expressing Rnf168 with mutated RING finger domain (C21S; cysteine 21 residue to serine), deleted MIU2 domain (ΔMIU2) or deleted MIU1 and MIU2 domains (ΔMIU1/2), showed no rescue of their CSR defects.
We next examined the switch recombination directly by using a quantitative digestion-circularization PCR (DC-PCR) assay [39]. These analyses demonstrated lower frequency of Sμ-Sγ1 switch recombination in activated Rnf168−/− B-cells compared to WT B-cells (Figure 5I and Figure S4G).
The DNA sequences of CSR junctions can offer additional insights into the mechanistic defects of the joining process [40]. For instance, B-cells with homozygous mutations in Xrcc4 or DNA Ligase IV undergo impaired CSR and form CSR junctions with increased sequence microhomology, indicating the use of an alternative joining pathway in these mutant cells [41], [42]. Previous studies reported that B-cells deficient for H2a.x or 53bp1 show no significant differences in the extent of microhomology at the switch junctions [40], [43]–[45]. To examine the effects of Rnf168 deficiency on the nucleotide composition of the switch junction regions, we cloned and sequenced the Sμ-Sγ1 junctions from Rnf168−/− and WT B-cells stimulated in vitro with LPS and IL-4. Such analysis revealed no significant differences in the extent of donor/acceptor homology (Figure 6A and 6B), in the frequency of mutations (Figure 6C) nor in the average length of overlaps (Table S2). Interestingly, in contrast to WT controls, 5.1% of the examined CSR junctions in Rnf168−/− B-cells were found to harbor long insertions (Figure 6D). Similar long insertions of nucleotides were observed in 2 out of 40 (5%) CSR junctions of 53bp1−/− B-cells but not in Atm−/−, H2a.x−/−, Nbs1−/− or WT B-cells [40], [43], [44], [46]. These data therefore suggest that Rnf168 contributes to the 53bp1 function in the synapsis of DNA ends.
Collectively, these data demonstrate that Rnf168 is required for in vitro and in vivo CSR, and its inactivation in mutant mice leads to immunodeficiency, which parallels the symptoms of the RIDDLE syndrome. Furthermore, we demonstrate that the RING finger and MIU2 domains of Rnf168 are indispensable for efficient CSR. Finally, our data also indicate that, similar to 53bp1−/− B-cells, Rnf168−/− B-cells display increased frequency of long nucleotide insertions at the CSR junctions.
53bp1−/− mice exhibit impaired early thymocyte development, an accumulation of early CD4−CD8− double negative (DN) thymocytes at the DNIII stage (CD44−CD25+), reduced TCRβ expression, and lower frequency of TCRγδ+ cells compared to WT littermates [28], [47], [48]. In contrast to the pronounced decrease of TCRβ expression in 53bp1−/− thymocytes, expression of TCRβ was not significantly affected in thymocytes deficient for H2a.x, Mdc1 or Rnf8 [30], [34], [47].
FACS analysis of thymocytes from Rnf168−/− mice showed increased frequency of DN thymocytes compared to WT littermates (Figure S5A). Examination of DN subpopulations indicated a significantly impaired transition of Rnf168−/− early thymocytes at the DNIII stage (Figure 7A, left panel). In addition, total number of DNIII thymocytes in Rnf168−/− mice (1.5×106±2.8×105) was significantly increased compared to WT littermates (1.1×106±1.9×105, p = 0.025) (Figure 7A, right panel). Rnf168−/− thymocytes also displayed a mild decrease in their expression of TCRβ compared to WT controls (Figure 7B) and the frequency of TCRγδ+ thymocytes in Rnf168−/− mice was significantly reduced compared to WT controls (Figure S5B; WT 0.24±0.21%, Rnf168−/− 0.20±0.01%, p<0.005). Productive rearrangement of the TCRβ locus takes place during the DNIII to DNIV transition and leads to pre-TCR expression [49]–[51]. Only cells expressing functional pre-TCR undergo exponential expansion, a process referred to as β-selection [49]–[51]. To assess the proliferation level of each DN thymocyte subpopulations, we examined in vivo thymocyte BrdU uptake in WT and Rnf168−/− mice. The number of BrdU+ cells was not affected in Rnf168−/− mice compared to WT littermates (Figure 7C). Therefore, the increased number of DNIII thymocytes, the decreased expression of TCRβ and the reduced number of TCRγδ cells in Rnf168−/− thymocytes were not due to proliferative defects. However, it remains possible that developmental defects can also contribute to these differences.
Loss of function of 53bp1 results in impaired Tcr locus integrity due to dysfunctional long-range V(D)J rearrangement [47]. To examine whether Rnf168 facilitates joining of distal gene segments during V(D)J recombination, we performed quantitative PCR assays of the partial (Dδ2-Jδ1 and Dδ1-Dδ2) and complete (Vδ5-Dδ1 and Vδ4-Dδ1) rearrangements. We observed that short-range rearrangements were more abundant in Rnf168−/− thymocytes compared to WT controls (Figure 7D, left panel, Figure 7E, and Figure S5C). On the other hand, complete Vδ-to-DδJδ recombination was significantly reduced in Rnf168−/− thymocytes compared to WT controls (Figure 7D, right panel, Figure 7E and Figure S5C).
These data support a role for Rnf168 in early thymocyte development and indicate that long-range V(D)J recombination is impaired in the absence of Rnf168.
Defective signaling or repair of DSBs impairs genomic integrity [4], [52]. For instance, inactivation of Atm, H2a.x, Mdc1, Rnf8, 53bp1 or Brca1 results in increased genomic instability [10], [27], [28], [30], [48], [53]–[56]. To evaluate the effect of Rnf168 inactivation on genomic integrity, we examined for chromosomal aberrations in metaphase spreads of LPS activated Rnf168−/− and WT B-cells (Figure 8A and 8B). The level of spontaneous chromosomal aberrations was elevated in Rnf168−/− cells (25±9.0%) compared to WT cells (2.5±0%, p<0.05). Interestingly, Rnf168−/− cells demonstrated increased frequencies of DNA breaks (15.8±6.0%) and structural chromosomal aberrations (9.2±3.0%) compared to WT controls (DNA breaks: 2.5±0%, p<0.05, structural chromosomal aberrations: 0%, p<0.05). Increased spontaneous genomic instability in Rnf168−/− cells was consistent with the elevated frequency of spontaneous γ-H2a.x- and Mdc1-IRIF in Rnf168−/− cells (Figure 3D and 3E). In response to IR (2 Gy), the level of genomic instability was further elevated in Rnf168−/− cells (65±2.2%) compared to WT controls (35.8±1%, p<0.05). The frequencies of IR-induced breaks (50.8±1.7%) and structural chromosomal aberrations (17.5±1.4%) were greater in Rnf168−/− cells compared to WT cells (DNA breaks: 23.3±8.4%, p<0.05; structural chromosomal aberrations: 12.5±1.4%, p<0.05). Therefore, Rnf168 deficiency leads to increased spontaneous and IR-induced genomic instability.
Mice deficient for Mdc1, Rnf8, 53bp1 or Brca1 have increased cancer susceptibility likely due to their elevated levels of genomic instability [28], [30], [54]–[57]. To examine whether inactivation of Rnf168 predisposes for tumor development, we monitored cohorts of Rnf168+/+ (n = 56) and Rnf168−/− (n = 52) mice for a period of 12 months; however, none of the monitored Rnf168−/− mice developed tumors during the 1 year period (Figure 8C).
Activated Rnf168−/− B cells displayed increased genomic instability, but the activation of Atm-Chk2-p53 pathway remained intact in the absence of Rnf168. These findings prompted us to examine whether this pathway has any role in preventing tumorigenesis in Rnf168−/− mice. To do so, we generated mice lacking both Rnf168 and p53. Rnf168−/−p53−/− mice showed no gross overall developmental defects compared to their single mutant littermates. Cohorts of Rnf168−/−p53−/− mice (n = 11) and p53−/− mice (n = 18) were monitored for survival. Interestingly, we observed a significant decrease in the tumor free survival of Rnf168−/−p53−/− mice compared to p53−/− controls (p = 0.0096, log-rank test).
Consistent with previous studies [58]–[60], the majority of tumors developed by p53−/− mice were thymomas (Table S3). However, a different spectrum of tumors was observed in Rnf168−/−p53−/− mice, including thymomas (Figure 8D and 8E), B-cell lymphomas (Figure 8F and 8G), hemangiosarcoma (Figure S6A and S6B) and sarcoma (Figure S6C and S6D). Rnf168−/−p53−/− thymomas and B-cell lymphomas were found to infiltrate various non-lymphoid organs including lung, liver and salivary glands (Figure S6E–S6H).
To examine chromosomal translocations in these tumors, we performed multicolor fluorescence in situ hybridization (mFISH) experiments using three B-cell lymphomas and one thymoma from Rnf168−/−p53−/− mice. Whereas p53−/− thymic lymphomas have been reported to rarely harbor chromosomal translocations [61], [62], two of the three examined Rnf168−/−p53−/− B-cell lymphomas carried clonal reciprocal translocations between chromosomes 12 and 15 [t(12;15) and t(15;12)] (Table S4 and Figure 8H). A clonal non-reciprocal translocation, t(9;11), was also observed in one of the examined Rnf168−/−p53−/− B-cell lymphomas (Table S4). Finally, complex chromosomal abnormalities were observed in the examined Rnf168−/−p53−/− thymoma (Table S4).
Collectively, these data demonstrate that Rnf168 is important for maintaining genomic integrity, and it cooperates with p53 in suppressing tumorigenesis.
The RNF168 gene is located on human chromosome 3q29, and encodes a novel E3 ligase that plays an important role in the signaling of DSBs [20]–[22]. Interestingly, mutations of RNF168 have recently been identified as the genetic defects leading to RIDDLE syndrome in humans [22], [24]. In order to investigate the in vivo functions of this E3 ligase, we have generated a mouse model for Rnf168 mutation.
Similar to mice mutant for its upstream DSBs signaling proteins, including H2a.x, Mdc1 and Rnf8, Rnf168−/− mice are viable [10], [27], [30], [34], [35]. In contrast to Rnf8−/− mice that display reduced number of lymphocytes [30], [34], no defect in T- or B-cell numbers was observed in Rnf168−/− mice. The normal number of lymphocytes in Rnf168−/− mice is in accordance with the lack of lymphopenia in the patient with RIDDLE syndrome [24].
Initial recruitment of 53bp1 to DSB sites is not affected in the absence of H2a.x, Mdc1 or Rnf8 [10], [29], [30], [63]–[65]. However, inactivation of Rnf168 in MEFs completely abolishes both the transient and retained 53bp1 IRIFs. These data highlight the importance of Rnf168 for the initial recruitment of 53bp1 to DSB sites, and indicate that this function of Rnf168 is independent of H2a.x, Mdc1 and Rnf8.
Immunodeficiency, a hallmark for RIDDLE syndrome, is characterized by normal ratio of T and B cells, but low levels of serum IgG likely due to impaired CSR [24], [66]. Interestingly, defective CSR was also observed in B-cells deficient for Atm, H2a.x, Mdc1, Rnf8 or 53bp1 [10], [27], [30], [34], [39], [44], [45], [66], [67].
Rnf168−/− mice showed reduced serum immunoglobulin levels, although this defect was milder compared to Rnf8−/− mice [30]. The impaired B-cell development observed in Rnf8−/− mice [30], [34], but not Rnf168−/− mice, might contribute to the more pronounced reduction in non-IgM classes in Rnf8−/− mice compared to Rnf168−/− mice. Although Rnf168−/− mice showed a normal IgA levels in serum, they demonstrated impaired CSR for various IgG isotypes, thus reproducing the CSR defect associated with RIDDLE syndrome [24]. Defective CSR in Rnf168−/− mice is also consistent with impaired CSR following the knockdown of Rnf168 in the B-cell line CH12F3-2 [66].
Recent studies have demonstrated the functions of 53BP1 in DNA damage signaling and repair [40], [45], [47], [67]–[70]. 53BP1 promotes and maintains synapsis during V(D)J recombination [40], [47]. Prior to recombination, 53BP1 constitutively associates with chromatin, supports long-range tethering of recombination signal sequence synapsis, and prevents extensive DNA end resection [47], [69], [70]. 53BP1 is also important for early T-cell development and for CSR [47]. In this study, we show that inactivation of Rnf168 impairs long-range V(D)J recombination in both early T-cell development and in B-cell class switch junction, suggesting that Rnf168 is required for proper synapsis of distal DSBs.
Impaired signaling of DSBs often results in male infertility as observed in H2a.x−/− [32], Mdc1−/− [10] and Rnf8−/− mice [30], [34], [35]. Rnf8 deficient males exhibit complete or partial infertility, impaired ubiquitylation of H2A in the XY body [34], and failure of global nucleosome removal; however, they are proficient in meiotic sex chromosome inactivation [35]. In contrast to H2a.x−/−, Mdc1−/− and Rnf8−/− males, 53bp1−/− males are fertile, and 53bp1 is not required for meiotic recombination during spermatogenesis [28], [33]. While fertility of young Rnf168−/− males was normal, these males became infertile with age, and elderly mice displayed testicular degeneration and atrophy. These data indicate that Rnf168 plays important roles in spermatogenesis in an age-dependent manner.
Signaling of DSBs is critical for maintaining genomic integrity and suppressing cancer. Genomic instability and cancer susceptibility are increased in the absence of Mdc1, Rnf8, 53bp1 or Brca1 [10], [28], [30], [54]–[57]. Tumorigenesis in the absence of proteins involved in signaling of DSBs is further exacerbated in the absence of p53 as exemplified in 53bp1−/−p53−/− mice [71]. While Celeste et al. reported that H2a.x−/− mice exhibit increased genomic instability, these mice did not show increased tumor susceptibility [27], [72]. However, inactivation of p53 strongly promoted tumorigenesis of H2a.x−/− mice [72], [73].
In this study, we demonstrate that Rnf168 deficiency also leads to increased radiosensitivity and genomic instability. However, similar to H2a.x−/− mice, tumor susceptibility was not increased in Rnf168−/− mice. We also demonstrate that Rnf168 and p53 collaborate in suppressing cancer since Rnf168−/−p53−/− mice exhibited shorter life spans and tumor latency compared to p53−/− littermates. These data indicate that p53 is critical for preventing tumorigenesis in Rnf168−/− mice. Therefore, not only is Rnf168 important for maintaining genomic stability, but it also collaborates with p53 to suppress cancer development.
While facial dysmorphism and short stature that are associated with RIDDLE syndrome were not observed in Rnf168−/− mice, these mutants displayed increased radiosensitivity, impaired CSR and immunodeficiency similar to the clinical features of this disease. Importantly, our data further highlight novel roles of Rnf168 in spermatogenesis, genomic integrity and cancer.
For the generation of Rnf168-deficient mice, two gene-trap ES clones, 156B6 and 405F11, were obtained from The Center for Modeling Human Disease (Toronto, Canada). In both ES clones, Rnf168 gene was disrupted by the integration of the gene trap vectors between exons 5 and 6 of Rnf168. Both ES cell clones were successfully used to generate Rnf168 mutant mice. Southern blotting and PCR analysis confirmed the disruption of Rnf168 locus and were used to genotype the animals. The following primers were used for PCR genotyping: Rnf168 mutant allele (forward 5′-ATCGCCTTCTATCGCCTTCT-3′ and reverse 5′-GCAGAAGACTCCGAACCTTG-3′), Rnf168 WT allele (forward 5′-GCCCAAGTCTGGCTCATTTA-3′ and reverse 5′-GCAGAAGACTCCGAACCTTG-3′). Southern blot analysis was performed using standard procedures. Rnf168 probe was generated by PCR using the following primers: 5′-TATTCCTGCTGCTGCTGCTA-3′, and 5′-CTCAAACCTCTTGCCCTCAG-3′. Rnf168−/− mice were generated by intercrossing Rnf168+/− mice obtained from each gene-trap clone. All mice in this study were in a mixed 129/J×C57BL/6 genetic background, and were maintained in a specific-pathogen free environment.
MEFs were generated from day 12.5 embryos using standard procedures. MEFs were cultured in DMEM (Gibco Invitrogen Corporation) supplemented with 10% FCS. Splenocytes, thymocytes and lymphocytes were cultured under the same conditions in RPMI1640 (Gibco) with 10% FCS.
Mouse Rnf168 cDNA amplified by PCR from WT MEF total cDNA was subcloned into pBABE-puro retroviral vector. Virus supernatant was collected 48 hour post transfection of phoenix cells with pBABE-puro-Rnf168. 3T3 immortalized MEFs were infected with Mock or Rnf168 retroviruses and were subjected to puromycin selection. Infected cells (1×103) were seeded to 6 cm dishes, irradiated (2, 4 and 6 Gy) and cultured for 11 days. Number of colonies was counted with crystal violet staining.
Single cell suspensions were stained with antibodies at 4°C in PBS with 1% FCS (Wincent, Inc.). The following antibodies conjugated to allophycocyanin, PE, FITC or perifinin chlorophyll protein were used for staining: anti-CD4, anti-CD8, anti-TCRβ, anti-TCRγδ, anti-Thy1.2, anti-B220, anti-CD25, anti-CD44, anti-IgG1, anti-IgG3 and anti-IgM (BD Bioscience, eBioscience). Stained cells were analyzed by FACS (FACSCalibur, BD Biosciences) using the CellQuest software (BD Biosciences) or FlowJo analysis software (Tree star).
LN cells (2×105) were either treated with IR (0–4 Gy) or UV (0–80 J/m2). After 24 hours, cell death was examined using 7-aminoactinomycin D (Sigma).
MEFs (passage 2–5) were synchronized using aphidicolin (Merck-Calbiochem) and either left untreated or treated with 10 Gy of IR. Subsequently, BrdU (Roche) was added to the cultures at various time points. MEFs were harvested, fixed using 70% ethanol and stained with FITC conjugated anti-BrdU (eBioscience) and PI. G2/M checkpoint was assessed by phospho-histone-H3 staining. Primary MEFs were either left untreated or treated with 2 Gy of IR. MEFs were harvested 1 hour post IR, fixed using 70% ethanol and stained with phospho-histone-H3 antibody (Cell signaling Technology) and PI.
Total protein extracts from cells were prepared using modified RIPA buffer (2 mM Tris-HCl (pH 7.5), 5 mM EDTA, 150 mM NaCl, 1% NP-40, 1% deoxycholate, 0.025% SDS, 1 mM phenylmethylsulfonyl fluoride and protease inhibitor cocktail tablets (Roche, Branchburg)). Proteins were separated on 10% homemade or 4–20% Tris-Glysine gradient polyacrylamide gels (Novex, Invitrogen). The following antibodies were used in 5% powdered milk (Carnation, Nestle) in TBS-T: affinity purified anti-Rnf168 antibodies (raised against either a murine GST-Rnf168381–567 and a murine Rnf168 N-terminal or against a murine GST-Rnf168), anti-Chk2 antibody (raised against murine Chk230–47), anti-phospho Chk2 threonine 68 antibody (Cell Signaling Technology), anti-p53 antibody (FL393, SantaCruz), anti-phospho p53 serine 15 antibody (Cell Signaling Technology), anti-Nbs1 antibody (Novus Biologicals), anti-phospho Nbs1 serine 343 antibody (Novus Biologicals), anti-Smc1 antibody (Abcam), anti-phospho Smc1 (Abcam) and anti-Atm antibody (Cell Signaling Technology).
B-cells were purified from spleen by negative selection using Dynal Mouse B-cell Negative Isolation kit (Dynal, Invitrogen). All B-cell experiments were performed using B-cells with a purity of 95.7±0.31%. B-cells (1×106) were stimulated with mouse anti-CD40 antibody (10 µg/ml, eBioscience) or LPS (20 µg/ml, Sigma) plus recombinant mouse IL-4 (1000 U/ml, eBioscience) for 4 days, and switching of Ig to IgG1 and IgG2a isotypes was examined. LPS (15 µg/ml) was used to induce IgG2b and IgG3 switching.
Mutants for Rnf168 cDNA were generated by PCR. Wild-type and mutant Rnf168 cDNAs were cloned into the MSCV-IRES-GFP vector. The ecotropic retroviruses expressing WT or mutated Rnf168 were generated by transient transfection of Phoenix cells with the appropriate retrovirus constructs. Purified splenic B-cells were activated with LPS and infected with virus supernatants containing polybrene (Sigma). Infected B-cells were cultured with LPS plus IL-4 for 4 days, and stained with anti-IgG1 for FACS analysis.
Purified B-cells were labeled with CFSE (Molecular Probes, Invitrogen) and cultured to induce IgG1 class switching for 4 days. Cells were stained with anti-IgG1 antibody and analyzed by FACS using the CellQuest software (BD Biosciences) or FlowJo analysis software (Tree star).
Genomic DNA from B-cells stimulated with LPS plus IL-4 for 4 days was digested with EcoRI (New England Biolabs) overnight and ligated for 16 hour with T4 DNA ligase (New England Biolabs). Two rounds of PCR were performed using nested primer pairs for Sμ-Sγ1 and nAchR. Primer sequences for the first round of PCR are as follows: Sμ-Sγ1, 5′-GAGCAGCTACCAAGGATCAGGGA-3′ and 5′-CTTCACGCCACTGACTGACTGAG-3′; and AchR, 5′-GCAAACAGGGCTGGATGAGGCTG-3′ and 5′-GTCCCATACTTAGAACCCCAGCG-3′. Primer sequences for the second round of PCR are as follows: 5′-GGAGACCAATAATCAGAGGGAAG-3′ and 5′-GAGAGCAGGGTCTCCTGGGTAGG-3′; AchR, 5′-GGACTGCTGTGGGTTTCACCCAG-3′ and 5′-GCCTTGCTTGCTTAAGACCCTGG-3′.
Genomic DNA isolated from stimulated B-cells was amplified by PCR using Pfu ultra polymerase (Stratagene) and the following primers (5μ3, 5′-AATGGATACCTCAGTGGTTTTTAATGGTGGGTTTA-3′ and γ1-R, 5′-CAATTAGCTCCTGCTCTTCTGTGG-3′). PCR products (500–1000 bp) were cloned into pCR2.1 using the TOPO TA cloning kit (Invitrogen). Sequence analysis of the cloned PCR products was performed using Sequencher software (GeneCode) and NCBI-BLAST.
B-cells (1×106) were stimulated with mouse anti-CD40 antibody or LPS (20 µg/ml) plus IL-4 (1000 U/ml) for 4 days and the levels of IgG1, IgG2a, and IgM isotypes in the culture supernatants were determined. Isotype switching to IgG2b and IgG3 was similarly examined using supernatants from B-cells stimulated with LPS (15 µg/ml). The levels of immunoglobulin in cell culture supernatants and serum from young (6–8 weeks) and old (9–12 months) mice were determined using SBA Clonotyping System/HRP and Mouse Immunoglobulin Isotype Panel (SouthernBiotech).
Total RNA was isolated from MEFs using TRIzol reagent (Invitrogen) according to the manufacture's instructions. Total RNA (1 µg) was reverse transcribed using Superscript II (Invitrogen) and oligo(dT) primers according to the manufacturer's instructions. We performed RT-PCR using the following primers (Rnf168: forward 5′-GAATGTCAGTGCGGGATCTGTA-3′, reverse 5′-AGGGCTCTTCGTGTCACTCCTAT-3′, β-actin: forward 5′- TGTTACCAACTGGGACGACA-3′, reverse 5′-AAGGAAGGCTGGAAAAGAGC-3′).
Genomic DNA was extracted from total thymocytes and TCRδ rearrangement was assessed using the Applied Biosystems 7900HT Fast Real Time PCR System (Applied Biosystems), Power SYBR Green PCR Master Mix (Applied Biosystems) and specific primers as described previously [47].
Rnf168−/− and WT littermates received two intraperitoneal injections of BrdU (1 mg each) with 2 hours interval. Thymocytes collected from these mice, 2 hours after the second injection, were examined for BrdU incorporation using a BrdU Flow kit (Becton Dickinson) and cytofluorometry.
MEFs (passage 2–5) grown on glass slides were IR treated (5 Gy), and fixed with 2% paraformaldehide for 5 min at room temperature. Fixed MEFs were blocked with antibody dilution buffer (10% FCS and 0.05% Triton X-100 in PBS) and incubated with rabbit anti-53bp1 antibody (Bethyl Laboratories), rabbit anti-phospho-H2a.x (Ser139) antibody (Millipore), rabbit anti-Mdc1 antibody (Bethyl Laboratories), or a home made rabbit anti-Brca1 antibody (raised against a murine Brca1831–845) overnight at 4°C. Labeling was detected using Alexa Fluor 488-labeled goat-anti-rabbit immunoglobulin or Alexa Fluor 555-labeled goat-anti-mouse immunoglobulin secondary antibodies (Molecular Probes). Cells were stained with DAPI for 5 min and then mounted with Mowiol mount solution (Calbiochem). The slides were observed under a Leica DMIRB fluorescence microscope (Germany) equipped with digital camera (Leica DC 300RF). Images were acquired under 100× magnification using Leica Image Manager software. Foci-positive cells were quantified by manual counting.
Immortalized Rnf168−/− MEFs were either mock-transfected or transfected with wildtype RNF168 GFP-tagged expression vectors. 24 hours post transfection, cells were irradiated (5 Gy), fixed 1 hour later and stained with anti-53bp1 as described earlier.
Immortalized WT and Rnf168−/− MEFs were also treated with DMSO or DNA-PK inhibitors (NU7026 10 µM or NU7441 0.5 µM (TOCRIS)) prior to IR. Cells were irradiated (5 Gy), fixed 2 hours later, and stained with anti-γ-H2a.x as described earlier.
Paraffin sections of testes, epididymides, tumors and organs were stained with hematoxylin-eosin for histological analyses. The slides were observed under a Leica DM4000B microscope (Germany) equipped with digital camera (Leica DC 300RF). Images were acquired using Leica Image Manager software.
Purified B-cells were activated with LPS for 48 hours, and either left untreated or irradiated as indicated. Cells were collected following 4 hours of colcemid treatment and processed by standard cytogenetic procedures. Number of chromosomes and gross chromosomal rearrangements were determined in 60 metaphase cells per sample from each genotype. The slides were observed under a Leica DMIRB fluorescence microscope (Germany) equipped with digital camera (Leica DC 300RF). Images were acquired under 100× magnification using Leica Image Manager software.
Mouse mFISH probe kit was obtained from MetaSystems GmbH, Germany. In mFISH, all 21 chromosomes are each painted in a different color using combinatorial labeling. The stained metaphases were captured using the Axio Imager Z1 microscope (Carl Zeiss) with filter sets for FITC, Cy3.5, Texas Red, Cy5, Aqua, and DAPI. Images were captured, processed, and analyzed using ISIS mBAND/mFISH imaging software (MetaSystems). Detailed experimental procedures were outlined in earlier publications [74], [75].
Data are presented as the mean ± SEM. The statistical significance of experimental data (p-values; Values≤0.05) was determined using the Wilcoxon test. Log-rank (Mantel-Cox) test was used for comparisons of survival curves.
All experiments were performed in compliance with Ontario Cancer Institute animal care committee guidelines.
|
10.1371/journal.pbio.2000733 | The Machado–Joseph Disease Deubiquitinase Ataxin-3 Regulates the Stability and Apoptotic Function of p53 | As a deubiquitinating enzyme (DUB), the physiological substrates of ataxin-3 (ATX-3) remain elusive, which limits our understanding of its normal cellular function and that of pathogenic mechanism of spinocerebellar ataxia type 3 (SCA3). Here, we identify p53 to be a novel substrate of ATX-3. ATX-3 binds to native and polyubiquitinated p53 and deubiquitinates and stabilizes p53 by repressing its degradation through the ubiquitin (Ub)-proteasome pathway. ATX-3 deletion destabilizes p53, resulting in deficiency of p53 activity and functions, whereas ectopic expression of ATX-3 induces selective transcription/expression of p53 target genes and promotes p53-dependent apoptosis in both mammalian cells and the central nervous system of zebrafish. Furthermore, the polyglutamine (polyQ)-expanded ATX-3 retains enhanced interaction and deubiquitination catalytic activity to p53 and causes more severe p53-dependent neurodegeneration in zebrafish brains and in the substantia nigra pars compacta (SNpc) or striatum of a transgenic SCA3 mouse model. Our findings identify a novel molecular link between ATX-3 and p53-mediated cell death and provide an explanation for the direct involvement of p53 in SCA3 disease pathogenesis.
| Ataxin-3 (ATX-3) is a ubiquitously expressed protein that mutated in a neurodegenerative disease called spinocerebellar ataxia type 3 (SCA3). It contains a polyglutamine (polyQ) tract near its C-terminus, the expansion of which is known to be the causative factor for SCA3. It has been known for a long time that ATX-3 is a deubiquitinating enzyme (DUB). However, the substrates targeted by ATX-3 in the physiological context remain elusive, thus largely limiting our understanding of its cellular function and that of the pathogenic mechanism of SCA3. This study has identified p53 to be a novel substrate of ATX-3, and its function is tightly regulated by ATX-3. PolyQ expansion augments ATX-3’s cellular function in p53 regulation. Due to enhanced interaction to p53 and up-regulation of p53, polyQ-expanded ATX-3 led to an increased p53-dependent neuronal cell death in zebrafish and mouse models, thus providing clear in vivo evidences for the direct involvement of p53 in SCA3 pathology. This study not only establishes a basic function of ATX-3 but also provides an explanation of how the interplays between ATX-3 and p53 contribute to the SCA3 pathogenesis; thus, it is an important contribution for the future development of therapeutic approaches for this currently untreatable neurodegenerative disease.
| Spinocerebellar ataxia type 3 (SCA3), also known as Machado–Joseph disease (MJD), is an autosomal-dominantly inherited ataxia and one of at least nine polyglutamine (polyQ) neurodegenerative disorders described so far [1–3]. SCA3 is caused by an unstable cytosine-adenine-guanine (CAG) trinucleotide expansion mutation in the ATXN3 gene leading to an expanded polyQ tract within the ataxin-3 (ATX-3) protein [4]. As a deubiquitylase, ATX-3 is highly conserved and ubiquitously expressed in cells throughout the body [5]. ATX-3 knockout (KO) mice have no major abnormalities [6]. It is possible that, besides ATX-3, three other members of the MJD family of cysteine proteases, including ATX-3 Like, JosD1, and JosD2 [7], may exert similar functions to ATX-3 and compensate for its absence in KO models. ATX-3 has a structured N-terminal Josephin domain comprising the catalytic site, two ubiquitin (Ub)-binding sites, and an unstructured C-terminal, which contains two or three Ub-interacting motifs (UIMs) flanking a polyQ tract [8,9]. The expansion of the polyQ tract is thought to trigger a pathogenic cascade, leading to cellular dysfunction and selective neuronal cell death [10]. Expansion length is inversely correlated with age of disease onset and directly with disease severity. However, the precise pathogenic mechanism triggered by polyQ-expanded ATX-3 in SCA3 patients has remained elusive [11–16].
A number of works have been carried out to explore ATX-3’s biological and potential cellular roles, and identification of molecular partners interacting with ATX-3 is hoped to facilitate identification of its physiological functions. For example, as a highly specialized deubiquitinating enzyme (DUB), a function of ATX-3 has been shown to be involved in the cellular protein quality control system by interacting with p97/valosin-containing protein (VCP) [9,17–22] and several E3 Ub ligases [15,20,23–28]. Moreover, several lines of evidence have shown that ATX-3 can bind DNA and interact with transcription regulators, thus being involved in transcriptional regulation [29–31]. Thus, ATX-3 has been associated with a wide range of biological activities.
The absence of ATX-3 leads to an increase of total ubiquitinated protein levels in ATX-3 KO mice [6], whereas overexpression of ATX-3 results in significantly reduced cellular protein ubiquitination in HEK293 cells [32], suggesting that ATX-3 may regulate the ubiquitination status of many proteins. However, the substrates targeted by ATX-3 in the physiological context remain unclear, thus limiting our understanding of its cellular functions. Whether the polyQ expansion in ATX-3 may contribute to the neuropathology by affecting its molecular interactions with other proteins or endogenous functions of normal ATX-3 is unknown. To develop effective therapies for this incurable disorder, it is important to identify ATX-3’s preferred substrates and to determine how the polyQ expansion causes the protein’s dysfunction.
In the present study, we used immunoprecipitation coupled with mass spectrometry to search for the proteins that associate with ATX-3. We have found that ATX-3 interacts with p53 and functions as a novel p53 DUB. ATX-3 deubiquitinates, stabilizes p53, and further regulates the functions of p53 in transactivation and apoptosis both in vitro and in vivo. Whether and how the polyQ expansion in ATX-3 affects its functional regulation of p53 and further neurodegeneration have also been investigated.
We analyzed proteins co-immunoprecipitated with 3×Flag-tagged ATX-3 from H2O2-treated 293T cells using mass spectrometric analysis (Fig 1A). p53 was found to associate with ATX-3. The interaction between ATX-3 and p53 was confirmed under physiological condition (Fig 1B) and with in vitro purified forms (Fig 1C), indicating a direct association between these two proteins. Furthermore, the amino terminus of ATX-3 was found to be necessary for binding with p53 (Fig 1D). This result was corroborated in cells by immunoprecipitation (S1A Fig). In addition, glutathione S-transferase (GST)–pull-down assay showed that both the DNA-binding domain and the C-terminal regulatory domain of p53 were sufficient for its interaction with ATX-3 (Fig 1E).
The two or three UIMs of ATX-3, depending on the splice isoform, mediate its binding to poly-ubiquitinated substrates. We observed that full-length (FL) ATX-3 bound robustly to both the native and ubiquitinated form of p53 in vitro (S1B Fig) and in cells (Fig 1F). Mutating the active site cysteine 14 did not affect the Ub chain binding activity of ATX-3, whereas ΔC and ΔN deletion as well as the UIM mutations (S236/256A and A232/252G) resulted in either abolished or impaired Ub binding activity (S1D Fig). Consistently, as shown in Fig 1G, the cysteine 14 mutation did not affect the binding of ATX-3 to either native or ubiquitinated p53, whereas the ΔN mutant lost its binding to both forms of p53. The ΔC mutant was found to bind to the native p53 with decreased affinity, and the two UIM mutants showed significantly compromised binding to ubiquitinated p53. We constructed a catalytic inactive ΔC mutant to exclude the possibility that the N-terminal domain might be able to interact with ubiquitinated p53 but further be deubiquitinated before detection. Catalytic inactive ΔC mutant showed similar binding affinity to native p53 as the ΔC mutant did, confirming that N-terminal domain only bound to native p53 (S1C Fig). Taken together, these results indicated that the binding of ATX-3 to p53 was synergistically regulated by the Josephin domain and the UIMs, with the former being primarily responsible for the binding of ATX-3 to the native p53 and further facilitating the latter to bind to ubiquitinated p53.
The interaction between ATX-3 and p53 suggested that p53 might be a substrate of ATX-3. Therefore, we tested whether ATX-3 affected the levels of p53 ubiquitination in vivo. As the p53 levels may differ among different cell lines, to show the generality of the effect of ATX-3 on p53, we generated ATX-3 stably knockdown cell lines in HeLa, HCT116, and RKO cells using two non-overlapping short hairpin RNA (shRNA) constructs. For the following experiments using knockdown cells in the study, either two clones of ATX-3 shRNA stably knockdown cells were used or one clone of knockdown cells was used but with rescue experiments performed in parallel. As indicated in Fig 2A, knockdown of ATX-3 in HeLa cells significantly increased the level of ubiquitinated-p53 (lane 3), and transient expression of ATX-3 effectively eliminated the increased ubiquitination that resulted from ATX-3 knockdown (lane 4). Moreover, ATX-3 overexpression resulted in a suppression of p53 ubiquitination (lane 2). Similar results were observed in ATX-3+/+ and ATX-3-/- mouse embryonic fibroblast (MEF) cells (Fig 2B left and right panel). Furthermore, our results showed that ATX-3 affected p53 ubiquitination in cells, and both the N-terminal Josephin domain and the C-terminal UIM domains of ATX-3 are required for this activity (Fig 2C). These results suggest that ATX-3 may act as a p53-directed DUB.
To determine whether ATX-3 can deubiquitinate p53 directly in vitro, we performed in vitro deubiquitination assays. Our results showed that FL ATX-3 deubiquitinated the ubiquitinated-p53 directly in a time- and dose-dependent manner (Fig 2D), which could be repressed by a nonspecific deubiquitinating inhibitor N-ethylmaleimide (NEMi) (Fig 2E left and right panel), and the effective deubiquitination of p53 required both the DUB activity and the poly-Ub binding ability of ATX-3 (Fig 2F).
As ATX-3 interacts with p53 under physiological conditions and regulates the ubiquitination of p53 in cells, it is possible that ATX-3 may regulate the turnover of p53 via the Ub-proteasome pathway. We found that co-transfection of ATX-3 and p53 led to the accumulation of p53 protein compared with p53 alone-transfected cells (Fig 3A). In contrast, deletion of ATX-3 resulted in an overt reduction in p53 protein level (Fig 3B, left), with no appreciable change at the p53 mRNA level (Fig 3B, right), indicating that the regulation of p53 by ATX-3 is unlikely at the transcriptional level. p53 levels were also compared in ATX-3 wild-type (WT) and KO mice primary cultures by western blot (S1E Fig). The basal p53 levels in ATX-3 KO primary MEFs were significantly lower than those in WT primary MEF cells. ATX-3 regulates p53 posttranslationally, because the half-life of p53 was significantly shortened in the ATX-3 stably knocked-down HCT116 cells (Fig 3C upper). This result was further validated in another clone of ATX-3 shRNA-stable knockdown cell line (Fig 3C lower) as well as in ATX-3-/- MEF cells (Fig 3D). In HCT116 p53-/- cells, the ectopically expressed p53 showed significantly prolonged half-life in ATX-3 and p53 co-transfected cells compared to that in p53 alone-transfected cells (Fig 3E), thus confirming the positive regulatory effect of ATX-3 on p53 stability. Furthermore, we found that MG132, a proteasome inhibitor, but not NH4Cl, a lysosome inhibitor, or 3-methyladenine (3-MA), a well-characterized inhibitor of autophagy, could suppress p53 degradation after cycloheximide (CHX) treatment (Fig 3F). Pretreatment of cells with MG132 blocked the degradation of p53 in both ATX-3+/+ and ATX-3-/- MEF cells (Fig 3G). Taken together, these results indicate that ATX-3 stabilizes p53 in cells via the Ub-proteasome pathway.
To explore the functional consequences of ATX-3-modulated p53 stability, the normal functions of ATX-3 in regulating p53-dependent biological activities were tested. In dual luciferase reporter assay, the DNA binding ability of p53, which is measured by the fluorescence intensity of PG13-Luc reporter, was significantly decreased in both ATX-3 knockdown HeLa cells (Fig 4A) and ATX-3 KO MEF cell lines (Fig 4B). Furthermore, acetylation of p53 at K373/K382 is reported to be a marker for the stimulation of the p53 transactivation activity. We treated the MEF cells with doxycycline (DOX) and found that the levels of acetylated p53 and one of its target gene products, the cyclin-dependent kinase inhibitor p21cip1/waf proteins, were remarkably decreased in ATX-3-/- MEF cells compared to ATX-3+/+ controls, and this could be restored to normal levels by transient expression of ATX-3 (Fig 4C). These results suggest that ATX-3 deletion inhibits the stimulation of p53 transactivation activity. The expression of several p53 target genes, for example, CDNK1A, CCNB1, and BBC3, was affected at both mRNA (Fig 4D and S2A Fig) and protein levels (Fig 4E and S2B Fig) associating with ATX-3 levels in HCT116 and MEF cells. The mRNA levels of cyclin B1 were higher in ATX-3 knockdown cells than that in control cells. This is because p53 acts as a repressor for cyclin B1. ATX-3 knockdown leads to an inhibition of p53 transcriptional activities, thus relieving its suppression on cyclin B1, resulting in its increased expression. Notably, overexpression of ATX-3 induced up-regulations of CDNK1A and BBC3 but a down-regulation of CCNB1, which required both the DUB activity and the poly-Ub binding ability of ATX-3 (S2C–S2F Fig). In contrast, knockdown of ATX-3 significantly blocked their induction, and this could be restored upon ATX-3 ectopic expression (Fig 4D and 4E). The inductions of all three p53 target genes were not changed in HCT116 p53-/- cells, indicating that these inductions were p53 dependent (Fig 4D and 4E).
In addition, fluorescence-activated cell sorting (FACS) analysis of cell cycle showed that ATX-3 deletion resulted in an increased proportion of cells in G2/M phase (Fig 4F). Without p53 and ATX-3 double KO MEF cells in hand, we examined the p53-dependence of this effect by using ATX-3 stably knocked down HCT116 p53+/+ and HCT116 p53-/- cells. An increase of G2/M phase cells was observed in ATX-3 knockdown HCT116 p53+/+ cells but not in HCT116 p53-/- cells, suggesting that ATX-3 was involved in the regulation of cell cycle arrest in G2/M phase, which was also p53 dependent (S2G Fig). Therefore, our data demonstrated that ATX-3 was able to regulate p53-dependent gene expressions and cell cycle arrest.
As p53 is a well-established apoptosis-regulator, we next examined whether ATX-3 affected p53-dependent apoptosis. We found that hallmarks of apoptosis, including the cleaved caspase-3 and poly (ADP-ribose) polymerase (PARP1), were less in ATX-3-/- MEF cells, while overexpression of ATX-3 resulted in significant caspase-3 and PARP1 cleavage (Fig 5A), indicating that ATX-3 was involved in the regulation of apoptosis in cells. Flow cytometry analysis using Annexin V-FITC/propidium iodide (PI) staining in HCT116 cells showed that knockdown of ATX-3 led to a decrease of camptothecin (CPT)-induced apoptosis, while ectopic expression of ATX-3 but not the catalytic inactive mutant ATX-3-C14A resulted in a significant increase of apoptosis in HCT116 p53+/+ but not HCT116 p53-/- cells (Fig 5B), indicating that ATX-3 promoted p53-mediated apoptosis, which required its deubiquitinating enzymatic activity.
Using the zebrafish model system, we further examined whether ATX-3 induces p53-dependent apoptosis in vivo. As our cellular results showed that knockdown of ATX-3 led to a significant decrease of CPT-induced apoptosis, it is quite possible no apoptosis signal would be detected under unperturbed conditions when ATX-3 is knocked down or knocked out. Therefore, the apoptosis as well as neurodegeneration in zebrafish were performed under ectopic expression conditions instead of knockdown or KO conditions. p53 WT and mutant zebrafish embryos were injected with mRNA of FL and various ATX-3 mutants. Twenty-four h post injections, the embryos were harvested and TUNEL-positive cells were analyzed. When the FL ATX-3 mRNA was injected into WT but not the p53 mutant zebrafish embryos, significantly more apoptotic cells were observed in TUNEL-staining assays, indicating that ATX-3 caused p53-dependent apoptosis in vivo. In addition, injections of mRNA of the catalytic inactive mutant ATX-3-C14A (C/A), the C-terminal deletion (ΔC), and the two UIM mutants (S/A and A/G) ATX-3 in WT p53 zebrafish embryos exhibited significant reduction in apoptosis compared with that of FL ATX-3, and no apoptosis was observed in p53 mutant zebrafish, indicating that the induction of p53-dependent apoptosis was critically dependent on the catalytic activity and the UIM domain of ATX-3 (Fig 5C and 5D). Interestingly, we observed that most TUNEL-positive cells localized in the head area of the zebrafish, suggesting that the apoptosis may occur mainly in the nervous system. To confirm this, TUNEL assay was performed by using the Tg(HuC:EGFP) transgenic zebrafish embryos, in which GFP-positive cells represent the expression of zebrafish neuronal marker elavl3 (formerly known as HuC) [33]. We observed that the TUNEL-positive cells induced by the FL ATX-3 mRNA injection were localized in GFP-positive brain regions (telencephalon, S3B Fig; diencephalon/hindbrain, S3C Fig) of the zebrafish, indicating that the ATX-3 mRNA injection-induced apoptosis occurred mainly in the nervous system of zebrafish.
PolyQ-expanded ATX-3 (ATX-3exp (80Q)) is thought to undergo conformational changes and acquire toxic properties, leading to altered molecular interactions. We wonder whether the polyQ expansion affects the ATX-3/p53 interaction. GST—pull-down assay (Fig 6A) and coimmunoprecipitation experiments (Fig 6B) showed that p53, both native and ubiquitinated form, bound ATX-3exp (80Q) stronger than the normal ATX-3. Consistently, ATX-3exp (80Q) exhibited stronger DUB activity than the normal ATX-3 in vitro (Fig 6C) and in cells (Fig 6D). Besides, we found that the degradation of p53 in the ATX-3exp (80Q)-expressing cells was slower than that of normal ATX-3-expressing cells (S4C and S4D Fig), and ectopic expression of polyQ-expanded ATX-3 induced higher levels of p53 protein than the normal ATX-3 in RKO, 293T, and MEF cells (S4E Fig), indicating that polyQ-expanded ATX-3 possessed enhanced capability to stabilize p53. The expression levels of p53-responsive genes (such as p21 and Puma) were also higher in ATX-3exp (80Q) expressing RKO cells (S4F Fig), suggesting that p53 was functionally enhanced by polyQ expansion in ATX-3. For unknown reasons, HCT116 cells do not behave as significantly as other cell lines (such as RKO, 293T, and MEFs) in terms of p53 induction (Fig 6F and S4E Fig). To determine the p53 dependence issue, we used the HCT116 cell lines with both p53+/+ and p53-/- genotypes. In HCT116 p53+/+, but not HCT116 p53-/- cells, both the normal and polyQ-expanded ATX-3 increased the induction of p53-responsive genes at both mRNA (Fig 6E and S4A Fig) and protein levels (Fig 6F and S4B Fig) when compared to the empty vector control group, but the difference did not reach the statistical significance between the normal ATX-3 group and polyQ-expanded ATX-3 group. All together, these data indicated that the polyQ expansion did not disturb the binding and the DUB activity of the normal ATX-3 to p53 and appeared to augment p53 stabilization.
CPT-induced apoptosis was analyzed by flow cytometry using Annexin V/PI. Apoptotic cells are those positive for Annexin V staining (either positive or negative for PI staining) (S3A Fig). Our results showed that expression of the FL ATX-3exp (80Q) led to a similar increase of p53-dependent apoptosis (including early and late apoptosis/necrosis) in HCT116 p53+/+ (Fig 7A) and zebrafish (Fig 7C and 7D) comparing to the normal ATX-3. Early apoptotic cells are Annexin V+/PI- staining, whereas late apoptotic/necrotic cells are Annexin V+/PI+ staining (PI-positive staining is due to a loss of plasma membrane integrity). Interestingly, we found that the normal ATX-3 induced a significantly higher percentage of early apoptotic cells, whereas the ATX-3exp (80Q) led to more late apoptotic/necrotic cells in HCT116 p53+/+ but not HCT116 p53-/- cells (Fig 7A and S3A Fig). In support, we found that CPT treatment resulted in clearly different morphological nuclear changes in the ATX-3exp (80Q)- and the normal ATX-3-expressing HCT116 p53+/+ but not HCT116 p53-/- cells after staining the nuclear DNA by Hoechst-33342 (S5A Fig). Nuclei of the normal ATX-3-expressing cells were round shaped but without condensed and fragmented chromatin, which represent an early apoptotic event. In contrast, nuclei of ATX-3exp (80Q)-expressing cells looked more amorphous, without any defined surface outline, and the nuclear heterochromatin was extensively packaged, indicating necrotic cell death (S5A Fig). Furthermore, as shown in Fig 7B, the p53-dependent apoptosis program marker such as cleaved PARP-1 in the ATX-3exp (80Q)-expressing HCT116 p53+/+ cells was found to be higher than the empty vector control group but weaker when compared to the normal ATX-3 group, indicating that the mitochondrial apoptotic p53 program played a role in this process, but the extent of this influence may not be as profound as that of the normal ATX-3. More importantly, High Mobility Group Box 1 (HMGB1) protein that released into the culture medium (a classical biochemical hallmark specific for necrosis [34,35]) as well as the level of receptor-interacting serine/threonine protein kinases 1 (RIP1, an important mediator of necrosis) were found to be significantly enhanced in ATX-3exp (80Q)-expressing HCT116 p53+/+ cells compared to those of the empty vector control and the normal ATX-3 group after CPT treatment, indicating that CPT apparently triggered a necrotic program in addition to the mitochondrion-dependent apoptotic program in polyQ-expanded ATX-3-expressing cells. These markers were not significantly changed in HCT116 p53-/- cells (Fig 7B), demonstrating the changes of these markers were p53 dependent. Together, these results suggested that a continuum of apoptosis and necrosis existed in response to CPT insult in ATX-3exp (80Q)-expressing cells, both of which were mediated by p53.
Next, we set out to determine if polyQ-expanded ATX-3 causes more neuronal cell death in vivo by using zebrafish as a model vertebrate. We injected mRNA of the normal ATX-3 and ATX-3exp (80Q) into wild-type and p53 mutant fish one-cell embryos. In zebrafish, otx2 is expressed in the prospective forebrain/midbrain in mid/gastrulae [36–38], and neurogenin 1 (ngn1) is reported to be a determinant of zebrafish basal forebrain dopaminergic neurons [39,40]. We performed the whole-mount in situ hybridization using otx2 and ngn1 as two neural markers to evaluate the neural loss upon the normal ATX-3 and ATX-3exp(80Q) expression in zebrafish. We observed that expression of normal ATX-3 or ATX-3exp (80Q) resulted in decreased signals of otx2 (blue staining mainly in MB area, Fig 7E and S5B Fig) and ngn1 (blue staining including telencephalon (TE), midbrain (MB), and hindbrain (HB) areas, S5C Fig) in WT but not p53 mutant zebrafishes at 24 h post fertilization (hpf), with more profound reduced signals of otx2 and ngn1 in ATX-3exp (80Q) mRNA injection groups (Fig 7E, S5B and S5C Fig). By 48 and 72 hpf, brains of ATX-3exp (80Q)-injected zebrafishes showed even weaker levels of both otx2 and ngn1 (Fig 7E, S5B and S5C Fig). These results provided clear evidences that expression of ATX-3exp (80Q) led to more neuronal loss in brains of WT but not p53 mutant zebrafishes, suggesting that ATX-3exp (80Q) caused more severe neuronal degeneration in SCA3 in a p53-dependent manner.
Previous studies have shown that neurodegeneration affects particular brain regions in MJD pathology [41]. To further study the neurodegeneration in specific affected brain regions in MJD, we generated an in vivo MJD genetic mouse model by ectopic expression of WT and mutant ATX-3 (80Q and C14A) with an enhanced green fluorescent protein (EGFP) tag in the substantia nigra pars compacta (SNpc) or striatum of p53+/+ and p53-/- mouse brain using lentiviral vectors (LV). The expression of these LV was first validated in 293T cells using fluorescence microscopy (S6A Fig) and also by western blot with anti-ATX-3 antibody (S6B Fig). Immunostaining analysis for tyrosine hydroxylase (TH), a marker for dopaminergic neurons in the SNpc, and for dopamine- and cyclic AMP-regulated neuronal phosphoprotein (DARPP-32), a regulator of dopamine receptor signaling, was performed to evaluate the neurodegeneration induced by lentiviral transduction. Our results showed that, in p53+/+ mice, WT ATX-3 caused a nonsignificant reduction of 15% TH-positive neurons in the SNpc (Fig 8A and 8B) and a reduction of 31% DARPP-32–positive neurons in the striatum (S6C and S6D Fig), whereas mutant ATX-3exp (80Q) induced a more significant loss of neurons in both of these two brain areas when compared to the empty vector expressed brain section, with a reduction of 57% for TH-positive neurons in the SNpc (Fig 8A and 8B) and of 51% for DARPP-32–positive neurons in the striatum (S6C and S6D Fig). The loss of TH or DARPP-32 immunointensity mainly occurred in EGFP-positive neurons. No significant neuronal loss was detected in either the SNpc or the striatum of p53+/+ mice injected with the catalytic inactive mutant ATX-3-C14A (Fig 8A and 8B, S6C and S6D Fig). The evidences for p53 involvement in the mutant ATX-3exp (80Q)-induced neurodegeneration were provided by the fact that no significant decreases of both neuronal markers were detected in p53-/- mice after LV transduction (Fig 8C and 8D, S6E and S6F Fig). Further supports for p53 involvement were evidenced by the observation that WT and mutant ATX-3exp (80Q), but not the catalytic inactive mutant C14A, caused marked increases in p53 immunoreactivity in the SNpc (Fig 8A and 8B) and striatum (S6C and S6D Fig) of p53+/+ mice, whereas no p53 staining was detected in the p53-/- mice (Fig 8C and 8D, S6E and S6F Fig). These results provided consistent evidences for a role of p53 in the WT and mutant ATX-3exp (80Q) expression induced neuronal degeneration.
To determine whether loss of TH staining in the SNpc as well as DARPP-32 staining in the striatum, associated with increased expression of WT and mutant ATX-3exp (80Q), was due to neuronal death, we performed TUNEL analysis and monitored expression of activated caspase-3. The numbers of TUNEL-positive and activated caspase-3-positive cells were significantly increased in the EGFP-positive neurons in both the SNpc (Fig 8E and 8F) and striatum (S7A and S7B Fig) of p53+/+ but not p53-/- mice. These results confirmed the association of cell death with WT and mutant ATX-3exp (80Q) expression. It should be noted that WT ATX-3 induced significant higher TUNEL and activated caspase-3 staining compared to the mutant ATX-3exp (80Q). As signs of non-apoptotic cell death, including extracellular release of HMGB1 and higher expression of RIP1, were observed in mutant ATX-3exp (80Q)-expressed cells (Fig 7B), and it is generally accepted that activated caspase-3 are rarely detected in the case of necrosis [42], we hypothesized that necrosis might occur in these brain areas. To gain further insights into mechanism of the neuronal death, we evaluated the expression of RIP1, which is an important molecule mediating necrosis when caspases are inhibited [43]. We found that mutant ATX-3exp (80Q)-expressed brain sections of p53+/+ but not p53-/- mice had more intense staining for RIP1 than did the WT ATX-3 sections (Fig 8E and 8F, S7A–S7C Fig). A high magnification view of the anti-RIP1 and anti-TH immunostaining in the SNpc of p53+/+ mice showed that most TH-positive neurons showed intact nuclear membrane, whereas those neurons that had condensed chromatin structures and nuclei were highly RIP1 positive but negative for TH staining (S7C Fig). No obvious RIP1 immunostaining was observed in p53-/- mice (S7C Fig). These results indicated the involvement of RIP1 in polyQ ATX-3exp (80Q)-induced neuronal death in mouse brains. Together, our in vivo data have demonstrated that the polyQ ATX-3 caused p53-dependent neuronal death in both apoptotic and necrosis manner in mouse brains.
p53 activity is crucial in determining the cellular fate, keeping a delicate balance between cancer-suppressive and age-promoting functions [44–48]. Therefore, tight regulation of p53 is essential for maintaining normal cellular functions. It has been shown that p53 is mainly regulated at the level of protein stability, which occurs predominantly through the Ub-mediated proteasomal degradation. On the flip side of the regulation, deubiquitination, which is mediated by DUBs, provides a parallel important regulatory control of p53 stability. Previously, several DUBs from the ubiquitin-specific protease (USP) [49–53] and otubain (OTU) family members [54] have been shown to regulate the Mdm2-p53 pathway, each with different detailed mechanisms of action. For example, USP7 (also named HAUSP) was the first identified USP that stabilizes p53 [55]. Later, it was found to deubiquitinate Mdm2 and Mdmx as well [56], thus showing selective deubiquitination to regulate the homeostatic levels of p53, Mdm2, and Mdmx under both normal and stress conditions [55]. Unlike USP7, USP10 [52], a cytoplasmic DUB, had recently been shown to directly deubiquitinate p53, but not Mdm2 and Mdmx, and to regulate the subcellular localization and stability of p53 by opposing the effects of Mdm2. In the present study, we report that p53 is a novel substrate of ATX-3 under physiological conditions.
Previous studies have suggested a possible role of the tumor suppressor protein p53 in neurodegenerative diseases, although the evidences are indirect. p53 is mutated in approximately half of all human cancers, and accumulating evidences also supported a role of p53 in neurodegeneration [57,58]. For example, p53 was found to be highly elevated in brains affected by several neurodegenerative diseases, including Alzheimer’s disease (AD), Parkinson’s disease (PD), Huntington’s disease (HD), amyotrophic lateral sclerosis (ALS), HIV-associated neurocognitive disorders (HAND), etc. [59]. Furthermore, several epidemiological studies have found an inverse correlation between the risk of developing neurodegenerative disorders and cancer [60–62], suggesting that some common protein effectors might likely be involved between these two multifactorial chronic pathologies. Given the important role of p53 in neurodegenerative diseases and cancer, it is thus a likely possible candidate. Recent studies have reported that ATX-3 and ATX-3 like are involved in gastric cancer [63] and breast cancer [64], supporting the association of the Josephin family of DUBs with cancer. Importantly, aberrant activation of the p53 pathway has previously been reported in both MJD patient brain tissues and transgenic animal disease models [65–69], and elevated p53 level was observed in MJD transgenic mice [65]. Therefore, the overall above-cited data concur to suggest the possibility of a functional link between ATX-3 and p53. p53 has not come out as a potential issue in the ATX-3 KO mice from the literature. As we mentioned in the Introduction section, this may be because other members of the MJD DUBs may compensate for its absence in ATX-3 KO models under unstressed conditions. Here in our study, we discovered that ATX-3 interacts with p53 and functions as a DUB for p53.
We have observed that the Josephin domain of ATX-3 is sufficient for the direct binding of ATX-3 to native p53, whereas the ubiquitinated p53 interacts with ATX-3 primarily through the UIMs, indicating that UIMs function to help recruit and bind the ubiquitinated p53 (Fig 9A and 9B). Therefore, both the Josephin and UIM domain coordinately regulate the interaction between ATX-3 and p53 (Fig 9A and 9B). During the DUB process, in addition to the catalytic cysteine 14 site and the N-terminal Josephin domain, the UIMs of ATX-3 are also required for its DUB activity towards p53, which may function to position the polyubiquitinated p53 correctly relative to the catalytic site for subsequent cleavage (Fig 9B). Thus, ATX-3 deubiquitinates and stabilizes p53 (Fig 9C), which is an essential step for p53 function in cell cycle arrest and apoptosis (Fig 9D). The direct interaction between ATX-3 and p53 is primarily mediated by the Josephin domain, and the first two UIM domains function to enhance the interaction by trapping the Ub-chains on p53. The polyQ tract between UIM2 and UIM3 is expanded in MJD. Our results suggest that polyQ length enhances rather than disturbs the binding and deubiquitination of ATX-3exp to p53 (Fig 6A–6D), which, in turn, causes more p53-dependent apoptosis/necrosis (Figs 7A, 7B, 7E, 8E and 8F, S5B, S5C, S7A and S7B Figs).
Previously, mutant huntingtin with expanded polyQ was found to bind to p53 and cause more cell death than WT huntingtin in neuronal cultures [70]. Here, polyQ-expanded ATX-3 was found to cause an increased percentage of cells undergoing p53-dependent late apoptotic/necrotic cell death than the normal ATX-3 did in HCT116 cells (Fig 7A and 7B, S3A Fig) and in neurons (Fig 8E and 8F). In consistence with our result, Evert’s group also observed an increased necrotic cell death in a cellular model for SCA3 upon polyQ ATX-3 expression [71], and p53 have been recently reported to play important roles in activating necrotic cell death [35,72,73]. Whether polyQ-expanded ATX-3 has additional targets that work together with p53 in inducing necrosis is not known yet, but we did find that the level of RIP1, an important mediator in necrosis, increased significantly in p53+/+ cells upon polyQ-expanded ATX-3 expression when compared to the empty vector control and the normal ATX-3 group, but remained at a basal low level in p53-/- cells (Figs 7B, 8E and 8F, S7A and S7B Fig). However, we cannot rule out the possibility that some other signaling pathways or p53 status (such as modification or subcellular distribution) [74] might also been affected upon polyQ-expanded ATX-3 expression, which led to the change of the percentage of cells undergoing apoptosis and necrosis.
We first used zebrafish as a model vertebrate to test the influence of polyQ expansion on the neuronal cell death in vivo. It is intriguing to note that injection of ATX-3 mRNA into the zebrafish embryos led to p53-dependent apoptosis, which occurred mainly in the central nervous system of zebrafish at early development stage (24 hpf). However, the polyQ-expanded ATX-3 induced progressive severe p53-dependent neurodegeneration in the central nervous system of zebrafish, suggesting that it caused other kinds of p53-mediated neural cell death besides apoptosis, too.
By generating a lentiviral-based in vivo MJD genetic model in p53+/+ and p53-/- mice, our in situ detection of two apoptotic markers (TUNEL and active caspase-3) and one necrosis marker (RIP1) data have provided convincing evidences that both enhanced apoptotic-like and non-apoptotic cell death are observed in the SNpc and striatum of ATX-3exp (80Q)-expressed neurons in p53+/+ mice brains but not in p53-/- mice brains (Fig 8E and 8F, S7A–S7C Fig). Meanwhile, significantly more p53-positive cells were detected in ATX-3exp (80Q)-expressed mouse brains sections (Fig 8A and 8B, S6C and S6D Fig). These data supported an idea that, due to enhanced interaction to p53 and up-regulation of p53, polyQ-expanded ATX-3 led to an increased p53-dependent neuronal cell death (including both early apoptotic and late apoptotic/necrotic manner). All together, the aforementioned studies and our results provide consistent evidences for the involvement of p53 in SCA3 pathogenesis, and the activation of the p53 pathway likely triggers neuronal dysfunction and eventually neuronal cell death in SCA3.
In conclusion, by identifying p53 as a new substrate of ATX-3, our study not only reveals a physiological function of ATX-3 and a new mechanism of p53 regulation but also establishes a novel molecular link between disease mutant ATX-3 and p53-mediated neurodegeneration, which sheds light on the molecular pathogenic mechanisms in SCA3.
Our work involving zebrafish and mouse experiments was in full compliance with the Regulations for the Care and Use of Laboratory Animals by the Ministry of Science and Technology of China and with the Institute of Zoology's Guidelines for the Care and Use of Laboratory Animals. The experimental protocols was approved by the Animal Care and Use Committee at the Institute of Zoology, Chinese Academy of Sciences (Permission Number: IOZ-13048).
ATX-3+/+ and ATX-3-/- MEFs, HEK293T, A549, RKO, and HeLa cells were cultured in DMEM supplemented with 10% FBS. U2OS, HCT116 p53+/+, and HCT116 p53-/- cells were cultured in McCoy’s 5A supplemented with 10% FBS. WT zebrafish embryos were obtained from natural matings of zebrafish Tuebingen strain. Tg(HuC:EGFP) transgenic fish embryos express EGFP in the post-mitotic thalamic neurons. Homozygous p53(M214K) mutant fish line carrying a loss-of-function p53 point mutation was kindly provided by Prof. Jinrong Peng at College of Life Sciences, Zhejiang University. Embryos were raised in Holtfreter’s solution at 28.5°C and staged by morphology as described [75]. P53 +/- mice were obtained from Jackson Laboratory. P53+/- mice were intercrossed to get P53-/- mice and P53+/- mice. For genotyping the p53 locus, primers X7 (5′-TAT ACT CAG AGC CGG CCT-3′), NEO19 (5′-CAT TCA GGA CAT AGC GTT GG-3′), and X6.5 (5′-ACA GCG TGG TGG TAC CTT AT-3′) were used as described previously [76]. Four-wk-old mice were used. ATX-3 was cloned into pCS2-Flag, p3×Flag-CMV26-Myc, pGEX-4T-1, and pET-28a vectors. p53 was cloned into pCMV-HA, pCMV-Flag, pGEX-4T-1, and pET-28a vectors. ATX-3 mutants were generated by site-directed mutagenesis (Stratagene). Anti-ATX-3 (1H9) was purchased from Merck. Anti-p53 (DO-1, PAb1620, and pAb421) and anti-Bax were purchased from Calbiochem. Anti-p53 (FL393), anti-Ub, anti-p21, anti-Cyclin B1, and anti-PARP1 were purchased from SantaCruz. Anti-Flag (M2) was purchased from Sigma. Anti-His was purchased from Abmart. Anti-HA was purchased from Covance. Anti-Puma, anti-p17 specific caspase 3, anti-RIP1, and anti-HMGB1 were purchased from ProteinTech. Anti-GFP was purchased from Thermo Fisher Scientific Inc. Anti-DARPP-32 and anti-TH antibodies were purchased from Cell Signaling.
3×Flag-ATX-3 transiently expressed in HEK293T cells was immunoprecipitated with anti-Flag M2 affinity gel and eluted with 3×Flag peptide (Sigma). Eluted proteins were identified with a gel-based liquid chromatography-tandem mass spectrometry (Gel-LC-MS/MS) approach and ion trap mass spectrometry (LTQ; Thermo Electron). A Mascot database search and the Scaffold program (Proteome Software) were used to visualize and validate results.
Cells were lysed with NETN buffer (20 mM Tris-HCl, pH 8.0, 100 mM NaCl, 1 mM EDTA, 0.5% Nonidet P-40) containing 1 mM Na3VO4, 10 mM NaF, 10 mM NEMi, and a cocktail of protease inhibitors. Whole cell lysates obtained by centrifugation were incubated with primary antibody overnight at 4°C. Protein A/G PLUS-Agarose beads (Santa Cruz) were then added and incubated for 2 h at 4°C. The immunocomplexes were then washed with NETN buffer for six times and separated by SDS-PAGE. Immunoblotting was performed following standard procedures.
GST and His fusion proteins were expressed in Escherichia coli strain BL21 (DE3) and affinity-purified using Glutathione-Sepharose 4B beads (GE Healthcare) or Ni-NTA Agarose (Qiagen), respectively, according to the manufacturer's instructions. For in vitro binding assays, bead-immobilized GST proteins were incubated with purified His proteins or with cell lysates in assay buffer at 4°C for 3 h followed by extensive washing. The bound proteins were separated by SDS-PAGE and analyzed by western blot with indicated antibodies.
For the in vitro deubiquitination assays, the ubiquitinated p53 protein was incubated with recombinant GST-ATX-3 (100 ng) or the same amount of other indicated proteins in a deubiquitination buffer (50 mM Tris-HCl pH 8.0, 50 mM NaCl, 1 mM EDTA, 10 mM DTT, 5% glycerol) for 2 h at 37°C. Western blot was performed to detect p53 ubiquitination. GST controls were included in all DUB assays. The in vitro p53 ubiquitination assays were conducted in a total of 20 μl reaction buffer containing recombinant p53 (20 ng), Mdm2 (100 ng), UbE1 (0.025 μM, Boston Biochem), UbcH5 (0.4 μM, Biomol), Ub (40 μM, Boston Biochem), 50 mM Tris-HCl (pH 7.5), 5 mM MgCl2, 2 mM ATP, and 2 mM DTT in the absence or presence of varying amount of bacterial His-ATX-3 at 37°C for 2 h. The reactions were stopped by adding SDS sample buffer followed by immunoblotting with anti-p53 antibodies.
The cells were treated with proteasome inhibitor MG132 (20 μM) for 4 h and then lysed in NETN lysis buffer with mild sonication. p53 was immunoprecipitated from the cell extract and subsequently resolved by SDS-PAGE and analyzed by western blot. For the preparation of a large amount of ubiquitinated p53 as the substrate for the deubiquitination assay in vitro, HEK293T cells were transfected together with the Flag-p53, HA-Ub, and Mdm2 expression vectors. After treatment as described above, the ubiquitinated p53 was purified from the cell extracts with anti-Flag M2 beads. After extensive washing, the proteins were eluted with Flag peptides (Sigma).
For protein half-life assays, 20 μg/ml CHX was added to cell cultures to block protein synthesis. Cells were collected at indicated time points, and protein levels were measured by western blot. The relative intensities of the bands were determined by densitometry analyses using Photoshop 7.0 software (Adobe). The half-lives of proteins were calculated from three independent experiments. To determine which degradation pathway was involved, 20 μg/ml CHX was added for the indicated intervals in the presence of MG132 (20 μM), NH4Cl (20 mM), or 3-MA (5 mM).
ATX-3+/+ and ATX-3-/- MEF cells were co-transfected with the indicated reporter constructs PG13 and the internal control Renilla luciferase pRL-null (pRL-CMV, Promega) at a ratio of 8:1 using PEI. Luciferase assays were performed using a dual-luciferase reporter assay system (E1910, Promega) according to the instructions of the manufacturer. Data were normalized for activity of Renilla luciferase to account for transfection efficiency. The assays were performed in duplicate, and data represent the average of five independent experiments.
For the flow cytometric analysis of cell cycle with PI DNA staining, the cells were harvested and washed once with PBS, followed by fixation in cold 70% ethanol at 4°C overnight. Then the cells were washed twice with PBS and treated with ribonuclease. Two hundred μl of PI was added before flow cytometry analysis. Apoptosis was assessed by using Becton—Dickinson FACScan flow cytometer according to the manufacturer’s instructions. Cells were treated with or without 1 μM CPT (a topoisomerase I inhibitor) for 24 h. Cells were collected and washed once with PBS, followed by incubation in annexin V (A13201; Invitrogen) solution in dark at room temperature for 15 min and 10 μl of PI in annexin V binding buffer. Flow cytometry analysis was carried out within 1 h. Data analysis was performed with CellQuest software. The numbers of apoptotic cells that are positive for annexin V staining (positive and negative for PI staining) were counted as a proportion to the total number of gated cells and expressed as percent of apoptotic cells in a histogram. Early apoptotic cells are positive for annexin V staining and negative for PI staining, whereas late apoptotic/necrotic cells are positive for annexin V and PI staining due to a loss of plasma membrane integrity.
Total RNA was extracted from HCT116 or MEF cells using TRIzol (Invitrogen), and 1.5 μg of total RNA was used to prepare the first-strand cDNA using the SuperScript II polymerase (Invitrogen). Quantitative real-time PCR reactions were carried out in triplicate on a Thermal Cycler using SYBR Green dye to measure amplification. Relative mRNA levels of each gene shown were normalized to the expression of the housekeeping genes GAPDH.
Normal, necrotic, and apoptotic cells were observed under fluorescence microscopy. Cells were fixed by 4% paraformaldehyde in PBS, and their nuclear DNA was stained with Hoechst-33342 for detection of necrosis and apoptosis by morphological features, according to Gschwind and Huber [77].
mRNA of FL and various ATX-3 mutants were in vitro synthesized from corresponding linearized plasmids using mMESSAGE mMACHINE Kit (Ambion). Digoxigenin-UTP-labeled antisense RNA probes were transcribed in vitro using MEGAscript Kit (Ambion) according to the manufacturer’s instructions. Microinjection and whole-mount in situ hybridization were performed as before [78–81]. Apoptosis were determined by TUNEL assay [82]. WT and p53 mutant zebrafish embryos were injected with object mRNA at one-cell stage and then harvested at 24 hpf for TUNEL labeling using In Situ Cell Death Detection Kit, TMR red (12156792910, Roche) according to the manufacturer’s instruction. Tg(HuC:EGFP) zebrafish embryos were fixed and stained with anti-GFP antibody. TUNEL-positive cells were imaged by confocal microscopy. Images were analyzed with ImageJ software. Significance was analyzed using the unpaired t test.
Lentiviral-EGFP vectors encoding human WT ATX-3 or mutant ATX-3, including ATX-3-80Q and ATX-3-C14A, were produced in 293T cells with a three-plasmid system. The lentiviral particles were re-suspended in artificial cerebrospinal fluid. Viral stocks were stored at -80°C.
Mice were anesthetized with 1% pentobarbital sodium and then placed into a stereotactic frame. Concentrated viral stocks were thawed on ice. L were stereotaxically injected into the striatum at the following coordinates: anterior-posterior: + 0.6 mm from the bregma; mediolateral: -2.5 mm from midline; and dorsoventral: -3.2 mm below surface of the dura; tooth bar: 0. LV were stereotaxically injected into the SNpc at the following coordinates: anterior-posterior: -3.8 mm from the bregma; mediolateral: -2.1 mm from midline; and dorsoventral: -4.5 mm below surface of the dura; tooth bar: 0. P53+/+ and P53-/- mice received a single injection of lentivirus in each side: left hemisphere (LV-GFP) and right hemisphere (ATX-3-WT, ATX-3-80Q, or ATX-3-C14A). Mice were kept in their home cages for 8 wk before being killed for immunostaining analysis.
Mice were perfused transcardially with 0.1 M phosphate buffered saline followed by fixation with 4% paraformaldehyde. Serial coronal sections were cut through the entire striatum and SNpc at 25 μm. Free-floating cryosections from injected mice were blocked in PBS/0.3% TritonX-100 containing 10% normal donkey serum (Gibco) and then incubated overnight at 4°C with the following primary antibodies: a rabbit polyclonal anti- DARPP-32 antibody (Cell Signaling, 1:200); a rabbit polyclonal anti-TH antibody (Cell Signaling, 1:200); a mouse monoclonal anti-p53 antibody, clone PAb1620 (1:50; Millipore); a rabbit polyclonal anti-RIP1-specific antibody (1:50, Proteintech); and a rabbit polyclonal anti-caspase 3, p17-specific antibody (1:50, Proteintech). Sections were washed in PBS and then incubated with the corresponding secondary antibodies coupled to fluorophores (Molecular Probes) for 1 h at 37°C. TUNEL analysis was performed using cryosections with the ApopTag-fluorescein in situ apoptosis detection kit (Chemicon) according the manufacturer's instruction. Apoptotic cells were quantified, and apoptotic indices were calculated by computer-assisted image analysis following identification of apoptotic cells by morphological analysis, the TUNEL assay, or activated caspase-3 immunostaining. Imaging was done by total internal reflection fluorescence microscope and three-dimensional structured illumination microscopy. Cell counts were determined from anatomically matched sections from each of the animals, and three animals were used for cell counts.
|
10.1371/journal.pmed.1002745 | The impact of targeted malaria elimination with mass drug administrations on falciparum malaria in Southeast Asia: A cluster randomised trial | The emergence and spread of multidrug-resistant Plasmodium falciparum in the Greater Mekong Subregion (GMS) threatens global malaria elimination efforts. Mass drug administration (MDA), the presumptive antimalarial treatment of an entire population to clear the subclinical parasite reservoir, is a strategy to accelerate malaria elimination. We report a cluster randomised trial to assess the effectiveness of dihydroartemisinin-piperaquine (DP) MDA in reducing falciparum malaria incidence and prevalence in 16 remote village populations in Myanmar, Vietnam, Cambodia, and the Lao People’s Democratic Republic, where artemisinin resistance is prevalent.
After establishing vector control and community-based case management and following intensive community engagement, we used restricted randomisation within village pairs to select 8 villages to receive early DP MDA and 8 villages as controls for 12 months, after which the control villages received deferred DP MDA. The MDA comprised 3 monthly rounds of 3 daily doses of DP and, except in Cambodia, a single low dose of primaquine. We conducted exhaustive cross-sectional surveys of the entire population of each village at quarterly intervals using ultrasensitive quantitative PCR to detect Plasmodium infections. The study was conducted between May 2013 and July 2017. The investigators randomised 16 villages that had a total of 8,445 residents at the start of the study. Of these 8,445 residents, 4,135 (49%) residents living in 8 villages, plus an additional 288 newcomers to the villages, were randomised to receive early MDA; 3,790 out of the 4,423 (86%) participated in at least 1 MDA round, and 2,520 out of the 4,423 (57%) participated in all 3 rounds. The primary outcome, P. falciparum prevalence by month 3 (M3), fell by 92% (from 5.1% [171/3,340] to 0.4% [12/2,828]) in early MDA villages and by 29% (from 7.2% [246/3,405] to 5.1% [155/3,057]) in control villages. Over the following 9 months, the P. falciparum prevalence increased to 3.3% (96/2,881) in early MDA villages and to 6.1% (128/2,101) in control villages (adjusted incidence rate ratio 0.41 [95% CI 0.20 to 0.84]; p = 0.015). Individual protection was proportional to the number of completed MDA rounds. Of 221 participants with subclinical P. falciparum infections who participated in MDA and could be followed up, 207 (94%) cleared their infections, including 9 of 10 with artemisinin- and piperaquine-resistant infections. The DP MDAs were well tolerated; 6 severe adverse events were detected during the follow-up period, but none was attributable to the intervention.
Added to community-based basic malaria control measures, 3 monthly rounds of DP MDA reduced the incidence and prevalence of falciparum malaria over a 1-year period in areas affected by artemisinin resistance. P. falciparum infections returned during the follow-up period as the remaining infections spread and malaria was reintroduced from surrounding areas. Limitations of this study include a relatively small sample of villages, heterogeneity between villages, and mobility of villagers that may have limited the impact of the intervention. These results suggest that, if used as part of a comprehensive, well-organised, and well-resourced elimination programme, DP MDA can be a useful additional tool to accelerate malaria elimination.
ClinicalTrials.gov NCT01872702
| The emergence and spread of multidrug resistance in the Greater Mekong Subregion (GMS) threaten regional and global malaria control.
Mass drug administrations (MDAs) are controversial but could be useful in the control and elimination of malaria.
We wanted to know whether well-resourced MDAs can accelerate malaria elimination in the GMS.
We randomised 16 villages (clusters) to receive MDAs with antimalarial drugs (dihydroartemisinin-piperaquine [DP] plus low-dose primaquine) either in year 1 or year 2 of the study. The entire village population (except pregnant women and children under the age of 6 months) was invited to take 3 consecutive daily doses of antimalarial drugs 3 times at monthly intervals. Everyone was followed up for 1 year; all malaria cases were recorded, and quarterly malaria surveys were conducted using highly sensitive high-volume PCR detection.
Most (87%) of the villagers completed at least 1 round of the antimalarial drugs, which were well tolerated.
The intervention had a substantial impact on the prevalence of P. falciparum infections by month 3 after the start of the MDAs. Over the subsequent 9 months, P. falciparum infections returned but stayed below baseline levels.
MDAs might be a useful tool to accelerate falciparum malaria elimination in low-endemicity settings.
The effectiveness of MDAs depends on continued support for village health workers, adequate drug efficacy, high levels of community participation, and carefully planned roll out to minimise the risk of malaria reintroduction.
| Considerable advances in malaria control and elimination have been achieved globally over the last decade. Since 2010 several former malaria endemic countries have been certified malaria-free. These include Sri Lanka, which had a high malaria burden (>100,000 cases/annually) at the beginning of the century while suffering from the consequences of a 25-year civil war [1]. Such success stories show that a determined malaria control programme with widespread use of long-lasting insecticide-treated bednets, insecticide spraying where appropriate, early diagnosis, and effective treatment can control and eliminate malaria. However, these conventional control tools are failing in some areas. Susceptibility of malaria vectors to most insecticides has decreased, often markedly, over the last decade [2], while the first-line treatments for P. falciparum malaria, artemisinin combination therapies (ACTs), are losing their efficacy in the Greater Mekong Subregion (GMS), home to more than 300 million people [3–6]. This is particularly worrying as resistance against earlier first-line antimalarial treatments (chloroquine, sulphadoxine-pyrimethamine) started in the GMS, spread to India and then to Africa, and killed millions of children [7]. More recently, parasites with resistance to both artemisinin and piperaquine emerged in western Cambodia and then spread to neighbouring countries [8,9]. Mefloquine resistance has re-emerged on the Thailand–Myanmar border. The decline in the effectiveness of the current first-line malaria drugs leaves few treatment options for falciparum malaria in the GMS. The spread of ACT-resistant P. falciparum strains into sub-Saharan Africa could become a public health emergency. Stopping the spread of antimalarial resistance requires the interruption of P. falciparum transmission.
Mass drug administrations (MDAs) clear symptomatic infections and, critically, also asymptomatic infections, which otherwise escape detection. MDAs may be essential to stop transmission and speed up the elimination of malaria. MDAs have been a part of the malaria control armamentarium for more than 100 years [10]. Three major reviews of MDAs have been conducted [10–12], which found that MDAs could interrupt malaria transmission temporarily in several areas and were critical for the permanent elimination of malaria from islands in the Pacific Ocean [13]. The success of MDAs depends on the efficacy of the drug regimen, the coverage of the target population, and the local malaria epidemiology—in particular the sources of transmission and potential for re-importation of malaria. MDAs have generally provided only transient benefit in areas of higher transmission because of rapid reintroduction of malaria from surrounding areas, and their role has remained controversial [12,14–16]. Targeted malaria elimination (TME) combines MDA with ensuring access to long-lasting insecticide-treated bednets and provision of early diagnosis and appropriate treatment. How long TME can interrupt or reduce the transmission of P. falciparum infections in communities in the GMS with low but persistent malaria transmission is not known.
We performed a cluster randomised trial in Myanmar, Vietnam, Cambodia, and People’s Democratic Republic (Lao PDR) (Fig 1), where malaria transmission is generally low (entomological inoculation rates < 1 infective bite/person/year). Malaria transmission occurs all year but increases during the rainy season, which lasts from June to October in Myanmar, May to November in Vietnam, and May to October in Lao PDR and Cambodia [17–20]. This article describes the impact of MDA on P. falciparum infections; the impact on P. vivax infections will be the subject of a subsequent report.
The MDAs started on 27 May 2013 in the sites in Myanmar, followed by Vietnam on 11 November 2013, Cambodia on 21 July 2015, and Lao PDR on 21 April 2016, and the study ended with the completion of follow-up of the Lao PDR villages on 12 June 2017 (S2 Table). There was a 12-day delay between the start of the study on 27 May 2013 and the public registration of the trial on 7 June 2013 due to an administrative delay.
In each site study teams performed a baseline census. The study teams geo-localised all houses by GPS and assigned a unique identification number linked to the study number of each household member. The census recorded the de jure population in each village, which tallies people according to their regular or legal residence. During subsequent surveys, the study teams recorded the de facto population, i.e., the number of people sleeping in the village during the night before the survey.
At each site, formative research conducted in preparation for the study suggested that, in addition to the information content, the background of the person providing information was a critically important determinant of the effectiveness of the communications [38–43]. Establishing trust between community members and the study team was a key prerequisite for a functioning relationship with the study populations. Team members who resided in the village for extended periods were encouraged to participate in local life and community events, such as weddings, funerals, and festivals. Participation in daily community life allowed the team to develop an understanding of local hierarchies, the seasonality of village work, and the needs and preferences of community members [44,45].
To increase the uptake and thus impact of MDAs, the study teams conducted intensive community engagement at each study site, starting from the time of site selection. This included educational elements to ensure that every community member was informed about the purpose of the MDA and the need for the entire village to participate to maximise individual benefits, and education to allay fears about risks of the antimalarial drugs. Village leaders, village malaria workers, and volunteers formed committees that assisted the study teams in organising the survey and designing and implementing community engagement. The teams also organised community engagement meetings at different levels, from individual or household informal discussions to whole village assemblies. Children and young people were engaged through activities such as music festivals and theatre.
Study teams provided benefits and incentives for participation at all sites (S1 Table). They adapted incentives according to the wishes of the villagers, as discussed in community meetings, and the preferences of regulatory authorities. In some sites, the study teams provided individual cash or non-cash incentives combined with gifts and/or lottery tickets, whereas in other sites the study teams preferred community incentives, e.g., improved water supply for the entire village. The study teams provided basic primary healthcare in the study villages. This ancillary care was essential to establish trust between the study team and community members. Health education on topics unrelated to MDA, such as family planning, nutrition, and vaccinations, was provided to community members at their request. The research teams worked to ensure that all participating villages had uninterrupted access to early diagnosis, efficacious antimalarial treatment, and insecticide-treated bednets.
The primary target of the MDAs was falciparum malaria. Radical cure of latent P. vivax, which would consist of a 7- to 14-day course of the 8-aminoquinoline primaquine, was not provided. The study teams conducted DP MDAs at month 0 (M0), M1, and M2 in early (intervention) MDA villages, whereas deferred (control) MDAs were conducted at M12, M13, and M14 in all sites except for the study sites in Myanmar, where, for operational reasons—specifically the limited access to study sites during the anticipated heavy rains—the deferred MDAs were conducted at M9, M10, and M11 (S1 Fig). In each country we matched the village pairs by geographical proximity, population size, and parasite prevalence. In each pair we selected 1 village randomly to receive early MDA and the other village to receive deferred MDA [46]. The randomisation was based on computer-generated random numbers provided by the trial statistician. For each round of treatment, the study team set up a mobile clinic or specific malaria post. Participants received MDA at home if they were unable to come to the clinic. The drug regimen comprised 3 rounds of 3 daily doses of DP combined with a single low dose of primaquine (0.25 mg/kg) given at intervals of 1 month. For regulatory reasons, no primaquine was administered in Cambodia. One DP tablet contained 40 mg of dihydroartemisinin and 320 mg of piperaquine. A weight-based regimen aiming at a total dose of 7 mg/kg dihydroartemisinin and 55 mg/kg piperaquine phosphate was used (Eurartesim, Sigma Tau, Italy, or Guilin Pharmaceutical, Guilin, China). The study teams encouraged all residents in the study villages to take part in the drug administration except for children under the age of 6 months and pregnant women (first trimester pregnancies were excluded in Myanmar and Vietnam and all pregnant women were excluded in Cambodia and Lao PDR). All doses were administered under direct observation of the study teams. All drugs were provided with snacks to increase tolerability. For children unable to swallow the tablets, the staff crushed the tablets and administered them mixed with water. Newcomers to the village or returning residents arriving during the study period were contacted by village malaria workers and offered a single round of DP. No placebo was used in control villages.
The rationale for the treatment regimen has been described previously [47]. Briefly, a full 3-day treatment dose of DP was given in order to clear any P. falciparum infection. The rationale for 3 repeated monthly rounds (3× 3 doses) was to provide the 3 months of post-treatment prophylactic effect needed to interrupt malaria transmission. This period was considered necessary because infectious mosquitos can survive for 30 days, and some community members were absent during the drug administration (they were treated during subsequent rounds) [48–51]. A single low dose of primaquine is sufficient to sterilise and rapidly clear mature P. falciparum gametocytes, which are not susceptible to schizonticidal drugs [50].
The study teams observed participants for 1 hour after drug administration and in case of vomiting offered a full or half repeat dose depending on the interval since the primary administration. Study staff enquired about adverse events (AEs) using a structured questionnaire on days 2, 3, and 7, and again 1 month after drug administration. During the 3-month intervention, a mobile clinic staffed with a medical assistant was available in the village to provide free consultations to villagers and to assess participants presenting with AEs. All serious AEs (SAEs) including hospitalisations and deaths occurring within 3 months of the first drug administration were documented and investigated by the study team, and the association with treatment was assessed.
The surveillance period was 24 months in the study sites in Myanmar and Vietnam but stopped after 12 months in Cambodia and Lao PDR because of the delayed starts in Cambodia and Lao PDR (see also S1 Fig). Just before the MDA and then every 3 months after the MDA, the study teams invited all people residing in the study villages aged 6 months or older, including temporary residents and migrant workers arriving after the MDA, to participate in the cross-sectional prevalence surveys. The survey at M21 in Myanmar was omitted because access to the study sites became impossible for the study teams during heavy rains. At the quarterly surveys, the study teams recorded the presence or absence of each participant in the village during the preceding 3 months. Additional information was collected during home visits at 2-weekly intervals: a study staff member enquired about the presence of recorded household members. Study staff recorded all newcomers who intended to stay in the village for longer than 2 weeks. Demographic information was collected, and participants’ temperature (skin surface forehead or tympanic membrane), weight, and height were measured by the study team. The threshold for determining fever was 38°C for body surface temperature and 37.5°C for tympanic temperature. A rapid diagnostic test (RDT) was performed, and if the test was positive for malaria, patients were treated according to national guidelines. The study team collected venous blood (3 ml from all individuals aged ≥5 years, and 500 μl from children aged ≥6 months to 5 years) at each survey, and from any individuals who developed fever during the study period.
The study team stored blood samples in a cool box in the field and transported the samples within 12 hours to the local laboratory. The teams tested blood samples from all survey participants using standard microscopy and malaria RDTs (Myanmar, Lao PDR, and Vietnam: SD Bioline Malaria Ag P.f/Pan POCT, Standard Diagnostics, Yongin-si, Republic of Korea; Cambodia: Healgen Malaria P. falciparum/Pan 1-step RDT, Zhejiang Orient Biotech, China). Microscopists who had at least 5 years’ experience and/or were confirmed to be Level 2 or higher, as assessed by a standard WHO 55 slide set, performed the standard microscopy, counting the number of parasites per 500 white blood cells on Giemsa-stained peripheral blood thick films. After separation of plasma, buffy coat, and packed red blood cells, samples were frozen and stored at −80°C. The study teams transported frozen samples from Myanmar, Cambodia, and Lao PDR monthly on dry ice to the molecular laboratory in Bangkok, Thailand, and the samples from the Vietnam sites to Ho Chi Minh City, Vietnam, for DNA extraction and high-volume ultrasensitive quantitative PCR (uPCR).
We have previously reported a detailed description and evaluation of the uPCR methods [52]. In summary, we purified the DNA template for PCR detection and quantification of Plasmodium from the thawed packed red blood cell samples. The purified DNA was dried completely in a centrifugal vacuum concentrator and then suspended in a small volume of PCR grade water, resulting in a concentration factor defined by the original blood volume (100–2,000 μl) divided by the resuspended double distilled water volume (10–50 μl). We used 2 μl of resuspended DNA as template in the quantitative PCR reaction. We assessed the presence of malaria parasites and estimated the parasite load in each sample using an absolute quantitative real-time PCR method. The 18S rRNA–targeting primers and hydrolysis probes used in the assay have been validated and are highly specific for Plasmodium species [53]. The lower limit of accurate quantitation using this method is 22 parasites per millilitre of whole blood. We used a QuantiTect Multiplex PCR NoROX Kit (Qiagen, Hilden, Germany) in the Bangkok laboratory and an absolute quantitative real-time PCR (quantitative PCR) method (Roche, Basle, Switzerland) in the laboratory in Ho Chi Minh City. We determined the Plasmodium species in uPCR positive samples using nested PCR specific to P. falciparum (microsatellite marker Pk2), P. vivax (microsatellite marker 3.502), and P. malariae (18s rRNA) as described previously [53–55]. We reported positive samples for which there was insufficient DNA for species identification—or where no amplification was obtained in this step—as being of indeterminate species.
We assessed polymorphisms in the PfKelch13 gene by nested PCR amplification covering the full length of the gene (total 2,181 bp) and sequenced the gene by ABI Sequencer (Macrogen, Seoul, Republic of Korea) as described previously. We monitored cross-contamination by adding negative control samples in every run. Sequencing results were aligned against PfKelch13 of reference strain 3D7 (putative 9PF13_0238 NCBI Reference Sequence [3D7]: XM_001350122.1), using Bioedit software (Abbott, Santa Clara, CA, US). Two study technicians assessed polymorphic patterns blinded to the origin of the sample.
We quantified Pfplasmepsin2/3 gene copy number using relative quantitative real-time PCR based on Taqman probe on a Corbett Rotor-Gene Q (Corbett Research, Mortlake, NSW, Australia). Primers and probes have been described previously [56]. We performed amplification in triplicate on a total volume of 25μl as multiplex PCR using a QuantiTect Multiplex PCR NoROX Kit (Qiagen, Hilden, Germany). Every amplification run contained 9 replicates of calibrators and triplicates without template as negative controls. Plasmodium-specific beta-tubulin served as an internal standard for the amount of sample DNA added to the reactions.
We categorised each resident’s individual MDA exposure as (a) did not participate at all, (b) did not complete a single round (3 doses), (c) completed only 1 round, (d) completed only 2 rounds, or (e) completed all 3 rounds. For the estimation of MDA coverage, we defined the numerator as the number of participants during three MDA rounds and the denominator as the de facto population during the time of MDA rounds. We defined Plasmodium prevalence by the uPCR result, but in the absence of a uPCR result we considered a positive microscopy or RDT result as sufficient to classify an individual as infected.
A P. falciparum infection was defined by either a P. falciparum positive result or a mixed result of P. falciparum and P. vivax. The incidence was defined using the number of malaria infections as numerator and exposure time as denominator. The individual exposure time was defined as the number of days spent within the catchment area, i.e., the village and the surrounding farms. The exposure time was estimated in 3-month intervals. For example, if a resident was present during 2 sequential surveys, the exposure time was 90 days. If a resident was missing during a survey, we assumed he/she stayed in the village for 45 days after the last participation in a survey. Similarly, we assumed a new arrival had arrived 45 days before the first participation in a survey. We assumed both losses to follow-up and intermittent missing data were missing at random. Seasons were defined as wet or dry by country as described above.
The unit of randomisation and hence the unit of statistical inference was the village cluster. The primary approach to analysis was based on intention to treat (ITT). In order to assess the sensitivity of our assumptions in the ITT approach, we also performed a dose-related per protocol analysis. We compared changes in prevalence then in incidence of P. falciparum (including mixed P. vivax and P. falciparum) infections over 12 months between villages that received early MDA and those that received deferred MDA.
Before the data collection was completed we drafted a statistical analytic plan, which is included as S1 Text. We examined the impact of DP MDA on malaria using multilevel mixed-effects Poisson models to obtain incidence rate ratios (IRRs) of Plasmodium infections. For the multilevel models, level 1 was repeated measurements of villagers over the follow-up time, level 2 was participants from the same village, level 3 was the 16 randomised villages, level 4 was the 4 different countries.
First, we performed univariable analyses to obtain the unadjusted estimates of IRRs of the association between malaria infections and MDA status, followed by adjustment for variables prespecified in the statistical analytic plan that are predictive for outcome and potentially imbalanced, i.e., sex, age, fever, bednet use, season, and prevalence of P. falciparum infections in the village. In an alternative model we used MDA exposure, i.e., 0, 1, 2, or 3 completed rounds as the main independent categorical exposure variable. In the initial analysis, we included only the exposure time from the completion of the MDA at M3 to M12. In a secondary analysis, we included the period from M0 to M12, i.e., including the MDA implementation period. For the analyses reported here, focusing on P. falciparum, we excluded parasitaemias in which the species could not be identified (indeterminate species, Plasmodium spp.). As the lower limit of quantification is 22 parasites/ml, genome densities less than 22 parasites/ml (n = 95 participants at baseline) were not included in the analysis, but the infection status (infected or uninfected) was left unchanged. The control villages in Myanmar, which received MDA at M9 during the crossover period, were excluded from the analysis at M12. We provide the intra-cluster correlation coefficient (ICC) for the incidence of P. falciparum infections with village (cluster) as a unit of randomisation, accounting for the random effect of country, using the exact linearisation calculation approach [57].
The sample size, 4 village clusters per country, was chosen mainly for operational and practical reasons. A formal sample size calculation suggested that 16 villages would provide 80% power to detect a 95% fall in prevalence from a 10% initial prevalence, controlling for random changes in prevalence in the control groups, with a minimum of at least 152 individuals in each village recruited and followed up satisfactorily.
The study teams collected survey data on case record forms and entered the data on smartphones before exporting them into OpenClinica (OpenClinica, Waltham, MA, US). Graphical summaries have been presented to show prevalence and incidence patterns over time. Treatment and AE data were recorded on registers and then entered in Excel (Microsoft, Redmond, Washington, US). Analyses were performed in STATA 15.0 (StataCorp, College Station, Texas, US).
The studies were approved by the Cambodian National Ethics Committee for Health Research (0029 NECHR, dated 04 Mar 2013), the Institute of Malariology, Parasitology, and Entomology in Ho Chi Minh City (185/HDDD, dated 15 May 2013), the Institute of Malariology, Parasitology, and Entomology in Quy Nhon (dated 14 Oct 2013), the Lao National Ethics Committee for Health Research (Ref No 013-2015/NECHR), the Government of the Lao PDR, and the Oxford Tropical Research Ethics Committee (1015–13, dated 29 Apr 2013). Each participant, or parent/guardian in the case of minors, provided individual, signed, informed consent; illiterate participants provided a fingerprint countersigned by a literate witness (ClinicalTrials.gov Identifier: NCT01872702).
The de jure population in the 16 villages was 9,897 (4,738 in early MDA and 5,159 in deferred MDA villages). The de facto population at M0 was 8,445 (4,135 in early MDA and 4,310 in deferred MDA villages), with a median of 495 residents per village (Fig 2). The median age of participants was 20 years (interquartile range 9 to 36). The large majority reported using insecticide-treated bednets regularly (4,579/5,620; 82%; Table 1). At baseline (M0), uPCR detected an overall mean P. falciparum prevalence of 6.2% (95% CI 5.6% to 6.8%). The baseline P. falciparum prevalence was lower in early MDA villages (5.1%, 95% CI 4.4% to 5.9%) compared to villages with deferred MDA (7.2%, 95% CI 6.4% to 8.1%), as was the P. falciparum density: geometric mean 3,363 parasites/ml (95% CI 2,472 to 4,575) in early MDA villages compared with 10,607 parasites/ml (8,146 to 13,812) in villages assigned to deferred MDA.
Of the 4,423 people residing during M0, M1, and M2 in the 8 villages randomised to early MDA, 3,790 (86%) completed at least 1 round (3 doses) of DP MDA: 635 (14%) completed 1 round of antimalarials, 635 (14%) completed 2 rounds, and 2,520 (57%) completed all 3 rounds (Fig 2). Thus, 633 residents (14%) did not complete a single round or took no antimalarials at all. In all, 2,707 residents lived in the 4 control villages in Myanmar and Vietnam where 24-month follow-up was conducted. Of these, 2,185 (81%) completed at least 1 round of MDA: 530 (20%) completed 1 round of antimalarials, 618 (23%) 2 rounds, and 1,037 (38%) all 3 rounds. Thus, 522 residents (19%) did not complete a single round or took no antimalarials at all (S2 Fig). In total, of the 8,749 study participants present during any time of the study period, 5,848 (67%) participated in at least 3 of the 5 possible surveys, and 2,815 (32%) in all 5 surveys (Table S2). The start and end date of each MDA is listed in S2 Table. The participation in follow-up surveys is shown in S3 Table.
Three months (M3) after the first round of drug administrations, the P. falciparum prevalence in the early MDA villages had fallen by 92%, from 5.1% (171/3,340) to 0.4% (12/2,828), while in control villages the P. falciparum prevalence had decreased by 29%, from 7.2% (246/3,405) to 5.1% (155/3,057; difference in differences −2.5%, 95% CI −3.9 to −1.1%, p < 0.001; Fig 3). The P. falciparum prevalence rose steadily over the following 9 months in the villages that received early MDA, to 3.3% (96/2,881) at M12, a 7-fold increase. The P. falciparum prevalence in the control villages, which received deferred MDA, rose from 5.1% (155/3,057) to 6.1% (128/2,101) during the same period, a 20% increase. The P. falciparum incidence after early MDA was 18 infections per 1,000 person-years and in control villages (deferred MDA) was 217 per 1,000 person-years (p < 0.001). Over the following 9 months, the P. falciparum incidence rose in early MDA villages to 142 per 1,000 person-years and in villages with deferred MDA to 262 per 1,000 person-years. There were 10 cases of acute falciparum malaria in intervention villages and 12 in control villages during the surveillance period.
Following the crossover, MDA surveillance continued for a further 12 months in 8 villages in Myanmar and Vietnam (S3 Fig). Following the deferred MDAs in Myanmar and Vietnam, the P. falciparum prevalence fell by 90% over 6 months, from 6.1% (128/2,101) to 0.6% (10/1,575), while in villages that had received early MDA during the previous year (but none in the current year), P. falciparum prevalence nearly doubled over 3 months, from 3.3% (96/2,881) at M12 to 6.1% (90/1,483) at M15. It fell to 3.3% (44/1,350) at M18 and then stabilised between 3.3% (44/1,350) and 3.5% (54/1,549) between M18 and M24. The incidence of P. falciparum infections also fell by 90% in the 6 months after the deferred MDA, from 262 to 27 per 1,000 person-years, while in villages that had received early MDA during the previous year, the P. falciparum incidence rose from 142 to 261 per 1,000 person-years, then dropped to 77 per 1,000 person-years by M24. The overall impact of MDA in reducing the incidence of P. falciparum infections was highly significant. The adjusted IRR was 0.41 (95% CI 0.20 to 0.84) over the 9 months following implementation (Table 2).
The impact of MDA on falciparum malaria varied by country. The greatest impact was in Lao PDR, followed by Cambodia and Myanmar, and there was little effect in Vietnam. This resulted in a country effect variance of 6.82 (p = 0.009; Fig 4). The impact was lower in villages with a baseline P. falciparum prevalence ≤ 5% (adjusted IRR 0.71, 95% CI 0.51 to 0.99) compared to villages with a baseline prevalence > 5% (adjusted IRR 0.13, 95% CI 0.02 to 0.79). The P. falciparum prevalence from 3 to 12 months after the MDA as assessed by uPCR was 0 or close to 0 (<1%) in 4 of 8 villages receiving early MDA and also in 4 of 8 control villages receiving deferred MDA (S4 Fig). The ICC for the incidence of P. falciparum infections for villages (as clusters), accounting for the random effect of country, was in the range from 0.06 to 0.32, estimated at baseline and every 3 months up to M12. The weighted-average ICC was 0.27 over 1 year of follow-up.
In regression models that used village allocation of MDA as the main covariate, male sex, age, and presence of fever were each independently and significantly associated with P. falciparum infections (Table 2, model A). In models that replaced village allocation of MDA with MDA coverage (Table 2, model B), there was a highly significant dose–response relationship between protection and number of completed rounds (IRR 0.63, 95% CI 0.56 to 0.72, p < 0.001). Protection against P. falciparum infection was lowest in people who had not participated in the MDA and was reduced in people who took 1 or 2 doses but did not complete a single 3-dose round. Protection against P. falciparum infection was highest in participants who completed all 3 rounds of the 3-dose regimen. Models in which the observation period included M0 to M12 showed similar exposure–response relationships (S4 Table). In a model including respondents of all ages, the protection for P. falciparum infection increased significantly with regular bednet use. In a model that included only children under 12 years of age, no protection attributable to bednets could be detected (S5 Table). As data for bednet use was missing for 35% (7,801/22,239) of the total observations of participants, this variable was not included in the multivariable analyses.
Before early and deferred MDAs, we identified 269 individuals with P. falciparum infections, of whom 258 (96%) participated in at least 1 round of DP MDA, i.e., received DP on 3 consecutive days (Fig 5). A follow-up blood specimen was obtained from 221 (86%) participants 1 month after MDA. There were 14 (6%) participants whose infections persisted after treatment (13 in Vietnam and 1 in Cambodia), while the remaining 207 (94%) participants had cleared their infection.
Ten of the 269 (4%) P. falciparum infections had the PfPailin genotype [8] (a long haplotype containing PfKelch13 C580Y, conferring artemisinin resistance, and multiple copies of the Pfplasmepsin2/3 genotype conferring piperaquine resistance): 5 of 113 (4%) were in Vietnam and 5 of 8 (63%) were in Cambodia (p < 0.001; Fig 5). All but 1 of the 10 participants with PfPailin cleared their parasitaemia after receiving at least 1 round of MDA (i.e., clearance rate 90%, 95% CI 56% to 100%). One subclinical PfPailin genotype infection in Cambodia persisted after 3 rounds of MDA but by M6 had cleared without further drug treatment.
Following the administration of DP (59,375 individual doses), 121 (0.2%) participants vomited. Of these, 104 vomited shortly after taking the drugs and were offered a repeat dose (Table 3). Data on dizziness and itching were not recorded in Vietnam, but in Cambodia, Lao PDR, and Myanmar, where a total of 26,898 doses of DP were administered, 586 (2.2%) events of dizziness and 12 (0.04%) events of itching were reported following drug administration on day 1. Within 1 month of the MDAs, 1,535 of 8,112 (19%) MDA participants recalled 2,577 AEs, of which 911 (35%) were considered related to the antimalarials; 592 (23%) of the 2,577 AEs were dizziness, 199 (8%) nausea, 96 (4%) vomiting, and 39 (2%) itching, and 1,653 (64%) participants reported a range of other minor complaints. There were no cases of severe haemolysis.
Within 3 months of the MDA, 6 SAEs leading to death or requiring hospitalisation were reported in villages receiving early MDA (6/4,135 participants; 0.15%), whereas in control villages with deferred MDA, 1 SAE (1/2,596 participants; 0.04%) was reported in the same time span (p = 0.187; Table 4). The investigators found none of the SAEs to be related to the administration of study drugs.
This cluster randomised trial demonstrated that in a setting where early diagnosis, effective treatment, and insecticide-treated bednets have already been made available, mass antimalarial drug administration with DP can substantially reduce the transmission of P. falciparum infections over a 1-year period. This period of protection afforded by the MDA was much longer than the post-treatment prophylactic effect provided by piperaquine (the slowly eliminated and therefore longer acting component of the antimalarial regimen), suggesting that the reduction of the asymptomatic parasite reservoir made a lasting impact on transmission of P. falciparum infections. An analysis starting 1 month after the last drug dose (i.e., from M3 to M12), after the complete implementation of the intervention and after the prophylactic period, showed a substantial benefit. Twelve months after early DP MDA, the P. falciparum prevalence had become very low or reached 0 in 4 of 8 intervention villages. However, malaria was also reduced substantially in control villages that did not receive early DP MDA, which was probably related to uninterrupted access to basic malaria control measures in all villages. The intensive community engagement conducted alongside the study activities played a critical role in promoting uptake. Overall, 86% of the target population participated in at least 1 round of the early MDAs, and 81% participated in at least 1 round of the deferred MDAs. In Myanmar and Vietnam, where DP MDA was less effective than in Cambodia and Lao PDR, the proportions of residents participating in all 3 rounds of the MDA was only 57% during the early MDAs and 38% in the deferred MDAs. Completion of the 3-round regimen was significantly associated with a reduction in the risk of becoming infected with P. falciparum in a multivariable regression model, while completion of a single round was not. Despite widespread artemisinin resistance, we found an overall clearance of 94% of subclinical P. falciparum infections after 1 or more rounds of MDA. The DP MDA drug regimen with a single low dose of primaquine was safe and remarkably well tolerated. There were no drug-attributable SAEs.
Artemisinin resistance was first reported in western Cambodia in 2008 [27,28], followed 8 years later by the detection of concomitant piperaquine resistance [4,6]. A single co-lineage of parasites (PfPailin) has since spread in a broad sweep encompassing northeast Thailand and southern Lao PDR, into southern Vietnam [8]. As a result, the clinical efficacy of DP in symptomatic falciparum malaria in these areas has fallen—often to below 50% [26,58]. Yet in our study, 9 of the 10 individuals with subclinical multidrug-resistant PfPailin infections who participated in at least in 1 round of DP MDA cleared their infections. This emphasises the substantial contribution of immunity to drug efficacy in people with asymptomatic malaria and suggests that, at these levels of DP resistance, the drug may still be of value in MDA. However, if malaria transmission in this region continues to increase, it will likely lead to higher levels of resistance, rendering DP progressively less effective. Our findings support the hypothesis that once a large proportion of the subclinical P. falciparum reservoir has been removed, transmission is reduced or even interrupted completely. This hypothesis is supported by the strong increase in protection with increasing number of MDA rounds in our study, and by recent entomological studies [59].
Where measured directly, the malaria protective effect of long-lasting insecticide-treated bednets in this region has been limited [60]. It has been estimated that in the GMS two-thirds of infective mosquito bites occur outside the home between 5 AM and 9 PM, i.e., where and when bednets are unlikely to be used. This has been confirmed along the Thailand–Myanmar border by the use of serological biomarkers, which show no correlation between bednet use and the human antibody response to malaria vector bites (salivary antigens) or P. falciparum infections [61,62]. Important local vectors such as Anopheles maculatus, An. dirus, and An. minimus tend to be exophilic and exophagic [63]. But our study did show an overall significant residual benefit of regular bednet use. This observation could be due to confounding, because irregular bednet use may indicate that the villagers were sleeping unprotected in and around forest edges, which is recognised as a major risk factor for malaria in the region [63–65]. This hypothesis is supported by the observation that only participants aged 12 years and older appeared to be protected by the use of bednets, while bednet use showed no protection in younger children, who are unlikely to participate in forest work (S5 Table).
Our study has a number of limitations. The attrition of the beneficial effect of DP MDA over time in this exploratory study was expected and is related to the study design, where, in each country, a small number of villages located within a malaria endemic area were given DP MDA over 3 months. Residual untreated infections in non-participants, importation of malaria infection from neighbouring untreated villages, or exposure to new infections through travel of villagers to surrounding areas were likely sources of malaria reintroduction [14–16]. Further limitations of the study were the absence of regulatory approval to include a single low dose of primaquine in the drug regimen in Cambodia. In Myanmar, the deferred MDAs took place in 2 villages at M9 instead of M12 because of difficult access in the peak of the rainy season, and the survey at M21 had to be cancelled. Only Myanmar and Vietnam could participate in the surveillance for 1 year after deferred MDAs, due to the delayed start in Cambodia and Lao PDR. The results from the second year are therefore based on a comparison of only 4 versus 4 village populations. Some of the observed higher impact in people who adhered to the complete 3 rounds of MDA compared to people who took part in none or fewer than 3 rounds could be due to the people adhering to the 3 rounds being healthier than the people who did not adhere [66,67]. Furthermore, some study teams did not record reliably AEs/SAEs from control villages because their focus was on the implementation of the MDAs, which may help to explain some of the differences in AE rates between the intervention and control villages.
If MDA is rolled out at the same time in an entire region, the risk of reintroduction of P. falciparum infections should be much reduced and hence the benefits should be sustained much longer [68]. Timely, accurate diagnosis and the appropriate treatment of residual malaria episodes after completion of MDAs will be essential for the permanent interruption of malaria transmission. This will require the presence of well-supported village health workers who provide several healthcare interventions in order to sustain the motivation for good malaria control as the incidence of malaria illness falls [69]. Further work is still needed to assess the source of P. falciparum reintroduction after clearing the asymptomatic reservoir, the prevalence thresholds for use of MDA, and the optimum MDA deployment strategies [70,71]. Additional options to achieve and maintain elimination include use of endectocides (i.e., ivermectin) in MDAs to kill vector mosquitoes, and the addition of a malaria vaccine. Even an imperfect vaccine providing a relatively short period of protection could prevent the re-importation of infections during the critical elimination phase [72,73]. Until new antimalarial drugs become available, and while efficacy remains stable at its current level, DP MDA can safely be used in low-transmission zones to accelerate regional elimination of P. falciparum malaria. Finally, the observation that 3 MDA rounds provide significantly more protection than a single round has direct implications for implementation, suggesting that reducing the 3-round DP MDA regimen to fewer rounds for logistic convenience may be ill advised.
In conclusion, despite imperfect adherence and widespread artemisinin resistance, the DP MDAs in our study were associated with a significant and clinically important long-lasting reduction in P. falciparum infections. Both the prevalence and incidence of P. falciparum infections were reduced and became negligible in half of the studied villages. This study, like others, demonstrates the critical importance and challenges of mobilising the target populations to participate in MDAs. To be effective, MDA needs to be part of a comprehensive, well-organised, and well-resourced elimination programme. This requires political will. In the eastern GMS, it is now over 10 years since artemisinin resistance—and the threat it posed to global malaria control and elimination—was recognised. Despite high investment, malaria transmission is increasing, and the antimalarial drugs are failing. The window of opportunity to use DP MDA effectively in the GMS may be closing. Outside the areas where DP resistance has become established, DP MDA could accelerate elimination in malaria hotspots as part of a concerted elimination programme.
|
10.1371/journal.ppat.1003346 | Recruitment of EB1, a Master Regulator of Microtubule Dynamics, to the Surface of the Theileria annulata Schizont | The apicomplexan parasite Theileria annulata transforms infected host cells, inducing uncontrolled proliferation and clonal expansion of the parasitized cell population. Shortly after sporozoite entry into the target cell, the surrounding host cell membrane is dissolved and an array of host cell microtubules (MTs) surrounds the parasite, which develops into the transforming schizont. The latter does not egress to invade and transform other cells. Instead, it remains tethered to host cell MTs and, during mitosis and cytokinesis, engages the cell's astral and central spindle MTs to secure its distribution between the two daughter cells. The molecular mechanism by which the schizont recruits and stabilizes host cell MTs is not known. MT minus ends are mostly anchored in the MT organizing center, while the plus ends explore the cellular space, switching constantly between phases of growth and shrinkage (called dynamic instability). Assuming the plus ends of growing MTs provide the first point of contact with the parasite, we focused on the complex protein machinery associated with these structures. We now report how the schizont recruits end-binding protein 1 (EB1), a central component of the MT plus end protein interaction network and key regulator of host cell MT dynamics. Using a range of in vitro experiments, we demonstrate that T. annulata p104, a polymorphic antigen expressed on the schizont surface, functions as a genuine EB1-binding protein and can recruit EB1 in the absence of any other parasite proteins. Binding strictly depends on a consensus SxIP motif located in a highly disordered C-terminal region of p104. We further show that parasite interaction with host cell EB1 is cell cycle regulated. This is the first description of a pathogen-encoded protein to interact with EB1 via a bona-fide SxIP motif. Our findings provide important new insight into the mode of interaction between Theileria and the host cell cytoskeleton.
| The apicomplexan parasite Theileria can reprogram the cell it infects, inducing uncontrolled proliferation and clonal expansion. This is brought about by the schizont, which resides free in the host cell cytoplasm. As the schizont never leaves the cell to infect other cells, it can only persist provided it is distributed over the two daughter cells each time the host cell divides. This is achieved by interacting dynamically with microtubules (MTs) that form part of the host cell mitotic apparatus. How MTs are recruited to the schizont surface is not known. MTs are highly dynamic, undergoing continuous cycles of growth and shrinkage that is regulated to a large extent by an array of proteins, called +TIPs, that associate with the free plus-ends of MTs. End-binding protein 1 (EB1) is a master regulator and central adaptor that mediates MT plus-end tracking of potentially all other +TIPs. We established that a schizont surface protein, p104, provides a docking site for EB1, which critically depends on a consensus SxIP motif, present in p104. These finding provides important new insight into the complex interaction of the transforming schizont with host cell MTs. To our knowledge, p104 is the first pathogen-derived protein identified so far to join the SxIP family of EB1-binding proteins.
| The tick-borne Apicomplexan parasites, Theileria annulata and Theileria parva, are the causative agents of lymphoproliferative diseases of cattle, tropical Theileriosis and East Coast fever, that cause significant economic losses in large parts of Asia and Africa. T. annulata predominantly infects macrophages/monocytes and B-cells, while T. parva infects predominantly T-cells and B-cells. Both species possess the unique capacity of transforming their host cells, inducing uncontrolled proliferation and resistance to apoptosis (reviewed in [1], [2])
Like other Apicomplexan parasites (such as Toxoplasma. and Plasmodium spp.), the life cycle of Theileria is complex and involves several morphologically different stages. Sporozoite entry in the target cells is a rapid process and within 15–30 minutes of invasion the infective sporozoite eliminates the enclosing host cell membrane upon which it associates with host cell MTs. Free in the cytoplasm, the parasite then differentiates into a multinucleated syncytium called a schizont [3]. Not restricted by the confines of a parasitophorous vacuole, the transforming schizont is in a perfect position to interfere with host cell signaling pathways that regulate cell proliferation and survival [1]. The schizont is strictly intracellular and depends entirely on its interaction with host cell MTs to ensure its persistence within the host cell - and thus maintenance of the transformed phenotype. By associating with the mitotic apparatus during mitosis and cytokinesis, the parasite secures the equal distribution of the schizont between the two new daughter cells [4]. This process involves the recruitment and stable association of de novo synthesized astral and central spindle MTs with the schizont surface [5]. We found that the mitotic kinase polo-like kinase 1 (Plk1) is recruited to the schizont surface in a cell cycle-dependent manner, and that association of the schizont with the central spindle, but not astral MTs, depends upon Plk1 activity. The distribution of the schizont between the two daughter cells during host cell cytokinesis is dependent on this association and by preventing the interaction with both astral and central spindle MTs proper segregation between the two daughter cells is disrupted [5].
The molecular mechanisms by which MTs are recruited and stabilized at the parasite surface are largely unknown. Proteins expressed on the surface of the schizont, or those secreted into the cytoplasm, qualify as good candidates for mediating host-parasite interactions, and yet to date only a few schizont surface proteins have been characterized in detail. gp34, a GPI-anchored schizont surface protein has been implicated in host-parasite interactions during host cell division [6]. It was proposed that TaSP, a proline-rich immunodominant T. annulata surface protein, can bind host cell α-tubulin and interact with the host cell microtubular network, but the molecular basis for these interactions is unclear [7].
MTs are involved in many cellular processes including cell differentiation, vesicle transport and cell division, and interact with various intracellular structures, including the actin cytoskeleton, the cell cortex and kinetochores. MTs are polarized, possessing minus ends that are usually stabilized and embedded in MT-organizing centers (MTOCs), and plus ends that extend into the cytoplasm and switch between periods of growth and shrinkage in a process termed 'dynamic instability [8]. MT dynamics are regulated to a large extent by MT associated proteins (MAPs), an important subset of which is the plus end tracking proteins (+TIPs). +TIPs constitute a structurally diverse family of MAPs that localize to the growing plus ends of MTs and play important roles in the regulation of MT growth and stability (reviewed in [9]). Since the description of the first +TIP, cytoplasmic linker protein CLIP-170, more than 12 years ago [10], the +TIP family has continuously expanded. They contribute to the regulation of mitosis, maintenance of cell polarity and positioning of organelles.
A network of +TIPs regulates different aspects of MT dynamics; XMAP215 promotes MT growth by catalyzing the addition of tubulin subunits to the growing MT plus end [11], while end-binding proteins (EBs) promote MT growth by suppressing catastrophes [12]). CLIP-170 promotes MT rescue [13], CLASP promotes MT stability [14] and the kinesin MCAK (XKCM1) has MT depolymerizing activity [15], [16]. By forming dynamic interaction networks with one another and with MTs, +TIPs cooperate to tightly control the rates of MT growth and shrinkage and finely tune MT dynamics that is so crucial for many functions, including the faithful segregation of chromosomes and successful cell division (for reviews, see [17], [18], [19]).
EBs are evolutionarily conserved in mammals, plants, fungi and other lower eukaryotes, and take a central position at the hub of the +TIP interaction network. EB1 was originally identified as a binding partner of adenomatous polyposis coli (APC) tumor suppressor protein [20]. EBs can bind directly to growing MT plus ends, independently of any binding partners [21], and interact with most known +TIPs, thus directing them to MT plus ends (reviewed in [19]). Protein depletion and rescue experiments showed that EB1 and EB3, but not EB2, promote persistent MT growth by suppressing catastrophes [12]. EBs track growing plus ends via their N- terminal MT-binding domain [12], which is highly conserved and adopts a ‘calponin homology’ (CH) fold [22]. The C-terminus of EBs interact with a large range of structurally diverse +TIPs, and is composed of an alpha-helical coiled-coil domain that mediates EB dimer formation [23], [24], and an acidic C-terminal tail of low complexity that includes a C-terminal EEY/F motif, like that found in alpha-tubulin [25].
Interactions between EB1 and its partners can be broadly classified into two groups based on their modes of interaction. First, the cytoskeleton-associated protein glycine-rich (CAP-Gly) domains of CLIP-170 and the large subunit of the dynactin complex p150glued specifically recognize the EEY/F motif of EBs [26]. The second mechanism of EB-partner interaction is mediated via a unique EBH domain of EBs (Conserved domain Database pfam03271), which partly overlaps with the coiled-coil domain. A large family of +TIPs (reviewed in [19], [27]), interacts with the EBH domain of EBs via a conserved S-x-I-P motif (SxIP, where x is any amino acid) that is embedded within a region of low complexity that is rich in basic, serine and proline residues. The complexity of the +TIP interactome was recently confirmed in a proteome-wide screen for mammalian SxIP motif-containing +TIPs, in which biochemical and bioinformatics approaches were combined [28].
In the present work, we report that p104, a protein previously described as a Theileria sporozoite microneme-rhoptry protein, is a major schizont surface protein that functions as a bona fide EB1-binding protein, interacting with EB1 via a classic ‘SxIP’ motif. In this capacity, p104 is the first parasite-encoded EB1-binding protein described to date.
A bioinformatics search was performed to identify schizont surface proteins that have the potential to interact with EBs. As the presence of a GPI-anchor is a reliable parameter for surface expression, T. annulata GeneDB (old version) was queried for genes encoding proteins containing a predicted signal peptide and GPI-anchor sequence (for detailed information, see Table S2 and Figure S5, upper section). This resulted in 19 candidates that were screened for the presence of an SxIP motif and representation in a T. annulata schizont proteome database obtained by mass spectrometry [29]. We identified TA08425, annotated in Theileria annulata GeneDB as encoding ‘Theileria parva microneme-rhoptry antigen of 104 kDa’ (henceforth referred to as ‘p104’), as the only candidate. The protein encoded by TA08425 (accession number XM_948006) shows 51% identity and 65% similarity to T. parva p104 and possesses a predicted signal peptide and putative GPI anchor signal. Even though p104 was originally identified as a protein expressed by T. parva sporozoites [30], T. annulata p104 could also detected by mass spectrometry in Triton X-114 lysates of purified T. annulata schizonts enriched for membrane proteins [29]. Sequence analysis of p104 (as isolated from TaC12 cells; Figure S1) predicted a protein with a globular N-terminal domain and a highly disordered C-terminal region of low sequence complexity that is rich in basic, serine and proline residues. The N-terminal half of the protein contained four so-called ‘FAINT (Frequently Associated in Theileria) domains’ (InterPro domain DUF529, IPR007480), a highly polymorphic domain with an average length of 70 residues [31]. FAINT domains are expanded in a lineage-specific manner in Theileria and characteristic of proteins that represent the schizont secretome [32]. Embedded within the basic-S/P rich C-terminal region, we identified an ‘SxIP’ motif, characteristic of EB1-binding proteins [33]. The p104 SxIP motif also fulfilled the contextual conditions outlined by Jiang et al, which stipulates that in the sequence X1-X2-[ST]-X3-[IL]-P-X4-X5-X6, at least one of the residues X1 to X4 should be an R or K and none of the residues X1 to X6 should be a D or E [28]. An alignment of the corresponding domains for a number of EB1-binding proteins is shown in Figure 1A.
A recent screen of a SMART cDNA library derived form T. annulata (Ankara) schizont cDNA [31] (kind gift of Dr. Gordon Langsley) revealed that the mAb 1C12 recognizes p104. 1C12 is one of a panel of mAbs obtained after immunization of mice with T. annulata (Hissar)-infected leukocytes [34]. This mAb recognizes T. annulata schizonts, but not merozoites and piroplasms [34], [35], and serves as a useful diagnostic marker for the parasite [5], [6], [36]. The identity of this immunodominant surface antigen, however, was unknown until now. Immunofluorescence microscopy (IFM) using the mAb 1C12 showed that p104 is expressed abundantly on the surface of the schizont (Figure 1B).
p104 cloned by PCR from genomic DNA isolated from the T. annulata-transformed cell line TaC12 differed from the T. annulata Ankara sequencing strain [31]. A comparison of the predicted proteins (AGB56140.1 and XP_953099.1) yielded 92% identities. We performed all of our experiments using TaC12 cells, and all of the expression constructs used in this study were based on the TaC12 p104 sequence (Figure S1, submitted to GenBank JX965955). In order to ensure cytoplasmic expression in mammalian cells, the sequence encoding the predicted signal peptide or GPI anchor sequences were omitted in expression vectors. Western blot analysis of lysates of COS-7 cells expressing V5-tagged ‘full-length’ (amino acids 20-871), C-terminal (p104-CT, 470-871) and N-terminal (p104-NT, 20-469) p104 fragments revealed that 1C12 recognizes an epitope in the N-terminal portion of p104 (Figure1C).
To investigate if T. annulata p104 has the potential to interact with cytoskeletal structures of the mammalian cell, we transiently expressed soluble V5-tagged p104 fragments corresponding to residues 20-469 (p104-NT-V5) and residues 470–871, (p104-CT-V5) in COS-7 cells, and analyzed their localization by IFM. V5-tagged p104-CT co-localized clearly with endogenous EB1 at MT plus ends (Figure 2A), displaying the typical ‘comet-like’ dashes, characteristic of EB1 and other +TIPs [37]. This pattern of expression was also detected when the full-length p104 (20 – 871) was expressed (not shown). Expression of p104-NT-V5, on the other hand, resulted in a diffuse cytosolic localization (Figure 2B). Importantly, MT plus end localization of p104-CT-V5 was abolished upon mutation of the SKIP motif to SKNN (p104-CTSKNN) (Figure 2C), indicating that p104 association with MT plus ends depends on a functional SxIP-motif. Overexpression of EB1 is known to result in MT bundling [38]. Co-expression of p104-CT-V5 with EB1-GFP resulted in the co-localization of both proteins to bundled MTs (Figure 2D). A detailed genome-wide bioinformatics screen of the T. annulata predicted proteome revealed the presence of more than 200 proteins containing an SxIP motif (see lower section of Figure S5, Table S3 and discussion below). Only a minority of these possesses a predicted signal peptide. Three of these candidates were expressed as V5-tagged proteins in COS-7 cells. TA20980 contains two SxIP motifs; TA17545 is a member of the subtelomere-encoded variable secreted protein (SVSP) family [31] and TA17375 is known as p150 (polymorphic antigen precursor in T. parva [39]. Interestingly, in contrast to p104-CT-V5, none of the proteins colocalised with EB1 or showed the typical comet-like dashes, despite possessing credible SxIP motifs (Figure S2)
To determine whether EB1 can interact with the parasite surface, we transiently expressed GFP-tagged EB1 and EB3 (a kind gift from Anna Akhmanova) in TaC12 cells. EB1-GFP labeled the entire surface of the schizont (Figure 3A). The same pattern could be observed using plasmids encoding EB1 containing a C-terminal myc or V5 tag (not shown) confirming that the observed localization of ectopically expressed EB1 is not an artifact caused by the GFP tag. In live imaging experiments, EB1-GFP-labeled ‘comets’ were readily detected moving from the centrosome to the periphery as reported [37], along with a striking association of EB1-GFP with the parasite surface (Movie S1), ruling out the possibility that the association of over-expressed EB1 with the schizont was a fixation artifact. Transfection experiments using EB3-GFP revealed that EB3, which shows overlapping functions with EB1 [12], [25], [40], also associates with the schizont surface (Figure S3).
To verify whether EB1 recruitment to the schizont requires MTs, TaC12 cells transiently expressing EB1-myc were treated with nocodazole, a drug that inhibits MT polymerization. Cells exposed to the drug (0.1 µg/ml) for 16 h completely lacked MTs and were arrested in prometaphase (an image of a prometaphase cell that was not subjected to nocodazole treatment and possesses intact MTs is included for comparison). Under these conditions, EB1-myc could still be found in association with the schizont, suggesting that the interaction of EB1 with the schizont is independent of MTs (Figure 3B). To explore the mode of EB1 interaction with the parasite surface, we generated EB1 truncation mutants and monitored the association of transiently expressed GFP-tagged EB1 fragments with the schizont surface by live imaging and IFM (Figure 3C). The C-terminal tyrosine residue of EB1 was found to be dispensable for parasite-localization, suggesting that binding to the schizont did not involve a ligand containing a CAP-Gly domain. The C-terminal portion of EB1, containing the linker region, coiled-coil dimerization domain, and the EBH domain (EB1125–268–GFP), was sufficient to bind to the parasite surface, while the N-terminal MT binding domain (EB11–133–GFP) did not localize to the schizont. These data suggested that the EB1-schizont interaction is dependent on the EBH domain of EB1, and likely is mediated via an SxIP-containing protein. An intact dimerization domain of EB1 was also dispensable for schizont interaction, as EB1208–268, a fragment consisting of the EBH domain and the acidic C-terminal tail, was sufficient to bind to the schizont. The position of the copGFP tag influenced the extent to which EB1208–268 associated with the schizont. When EB1208–268 was fused to copGFP at its N-terminus, all transfected cells showed clear parasite labeling; when the tag was placed at the C-terminus, however, staining patterns were more variable, with only few transfected cells showing labeled schizonts. Finally, we determined that EB1208–251, defined as the shortest EB1 fragment capable of interacting with APC [41], failed to associate with the schizont.
We next carried out pull-down experiments to confirm that T. annulata p104 is a functional EB1 binding partner. Recombinant Halo-tagged p104 fragments containing a C-terminal V5 tag were produced in E. coli and used for pull-down analysis of lysates prepared from COS-7 cells. Recombinant p104-CT migrates more slowly in SDS-PAGE than predicted (70 kDa rather than the predicted 46 kDa), while the N-terminal fragment has an apparent molecular weight as expected of 55 kDa (Figure 4A and 1C). Endogenous EB1 was detected after pull-down with recombinant p104-CT, but not with p104-NT. Mutation of the SKIP domain to SKNN (p104-CTSKNN) abolished the interaction between EB1 and p104-CT, supporting our observations made by IFM that p104 can interact with EB1, and that this interaction is mediated via the SKIP motif. Conversely, in pull-down assays performed on lysates of T. annulata-infected TaC12 cells, GST-tagged EB1 produced in E. coli bound parasite-derived p104 (Figure 4B), but not TaSP, another major schizont surface protein [7].
Further biochemical evidence for a specific p104/EB1 interaction was obtained by subjecting lysates of COS-7 expressing EB1-GFP fragments to pull-down experiments with recombinant p104-CT (Figure 4C). While full-length EB1-GFP and EB11–133-GFP migrated at the predicted molecular weight of 60 kDa and 45 kDa respectively, EB1125–268-GFP, predicted to be 46 kDa, was resolved as two bands of approximately 40 and 50 kDa, likely due to the presence of a second in-frame translational start codon. In agreement with our observations made by IFM, full-length EB1-GFP and EB1125–268-GFP, but not EB11–133-GFP, bound to p104-CT. No binding occurred to p104 containing a mutated SKIP domain (p104-CTSKNN).
To identify the region of p104 required for EB1 interaction in more detail, we expressed a fragment of p104, (p104521–634) consisting of 114 aa and containing the SKIP motif, in mammalian cells. In transfected COS-7 cells, GFP-p104521–634 tracked MT plus ends, producing ‘comet-like structures’ characteristic of EB1 and many EB1-binding proteins (Movie S2) [12], [33], [42]. The plus end tracking behavior of GFP-p104521–634 was completely abolished upon mutation of the SKIP motif to SKNN (not shown). To further define the region of p104 required for plus end tracking, we generated recombinant GFP-tagged p104 fragments of 40 aa (GFP-p104554–593) for expression in mammalian cells, and also in E. coli for microinjection experiments. The p104554–593 aa sequence corresponds to that shown in the alignment in Figure 1A. GFP-p104554–593 tracked growing MT plus ends in transfected COS-7 cells (not shown), and strongly labeled the centrosome of microinjected TaC12 cells (Movie S3). GFP-p104 554–593 tracked growing MTs emanating from the centrosome (Movie S3) and also tracked MT plus ends along the surface of the schizont, in the same manner as EB1-GFP (compare with Movie S1). In mitotic cells, GFP-p104554–593 localized to centrosomes and to the tips of growing spindle MTs (not shown), as was reported for EB1-GFP [43].
Having shown that p104 and EB1 interact in an SxIP-dependent manner, and that ectopically expressed EB1-GFP associates with the schizont, we focused our attention on endogenous host cell EB1 in T. annulata-transformed cells. In IFM preparations fixed with methanol, EB1 was detectable as ‘comet-like’ fluorescent dashes at the plus ends of MTs and at the centrosome, as described in other cell-lines [37] [44]. While the extent to which EB1 association with the schizont varied within a mixed population of cells, we noticed a striking association of EB1 with the schizont in cells undergoing cell division (Figure 5A, and 5B last panel). In cells in metaphase, however, EB1 was largely undetectable at the parasite surface.
Because mitotic cells were detected only infrequently in an unsynchronized culture, we made use of various cell synchronization protocols to analyze in more detail the association of EB1 with the schizont. Cells were first synchronized in prometaphase by overnight nocodazole treatment and then released from nocodazole block into media containing the proteasomal inhibitor MG132 that blocks the degradation of cyclin B, thus synchronizing cells in a metaphase-like state. Images representative for the different cell cycle stages are presented in Figure 5B. In interphase cells, which have a more flattened morphology, the characteristic ‘comet-like’ distribution of EB1 was clearly noticeable and, in many cells, the outline of the schizont was labeled with EB1 (Figure 5B, first panel). EB1 also decorated the surface of the schizont during prometaphase. As cells progressed through metaphase, however, schizont staining was notably absent whereas the mitotic spindle was strongly labeled. Upon release from metaphase arrest, EB1 re-associated with the schizont surface and parasite-associated EB1 could easily be detected in cells in anaphase and telophase. EBs are very conserved and the T. annulata genome contains a gene (TA18025) encoding EB1, which shows up to 49% positivity when compared with bovine EB1. Recombinant TaEB1 subjected to immunoblot analysis failed to react with the anti-EB1 antibody (KT51), and no reactivity could be demonstrated with lysates of purified parasites, thus excluding cross-reactivity (Figure S4).
Endogenous T. annulata p104 has a predicted molecular weight of 101 kDa, but migrates in SDS-PAGE with an apparent molecular mass of approximately 130 kDa. As p104 contains multiple predicted phosphorylation sites (NetPhos 2.0), we investigated whether p104 is phosphorylated in TaC12 cells. Incubation of lysates from unsynchronized TaC12 cells with λ-phosphatase caused a significant downshift in the apparent molecular weight of p104, indicating that endogenous p104 is extensively phosphorylated (Figure 6A). While p104 appears to be highly phosphorylated throughout the duration of the cell cycle, we noticed a slight up-shift in endogenous p104 migration in lysates from mitotic cells (those arrested in metaphase), compared to lysates prepared from unsynchronized cells (Figure 6A). To obtain more detailed information on p104 phosphorylation in the region important for EB1 interaction, we subjected lysates from cells expressing p104521–634-V5 to λ-phosphatase treatment and subsequent western blot analysis. A significant downshift was observed upon phosphatase treatment, demonstrating that phosphorylation occurs within this region. Furthermore, the difference in protein migration pattern of p104521–634-V5 observed between unsynchronized and mitotic cells was more striking than for endogenous p104, suggesting that extensive cell-cycle dependent phosphorylation occurs in this region (Figure 6B). Within this 114 amino acid stretch, there are 6 ‘strong’ Cdk1 phosphorylation sites (S/T*-P-x-K/R), and three additional ‘weak’ Cdk1 phosphorylation sites (S/T*-P) [45]. Preliminary LC-MS/MS analysis revealed serine phosphorylation of at least 4 of these sites (serines 589, 601, 607 and 622). Treatment of TaC12 cells synchronized in mitosis with the Cdk1 inhibitor RO-3306 resulted in a partial, but pronounced, reduction in p104 phosphorylation (Fig. 6D). The state of p104 phosphorylation does not appear to affect its interaction with recombinant EB1, however, as, p104 present in lysates prepared from mitotic as well as unsynchronized or RO-3306-treated cells was capable of binding to GST-EB1 in pull-down experiments (Fig. 6E).
The association of EB1 with the schizont appeared to correlate inversely with the pattern of Cdk1 activity, with EB1 being undetectable at the schizont surface during metaphase, when Cdk1 activity peaks. To investigate this further, we analyzed the effect of Cdk1 inhibition on the localization of EB1 with the schizont. As mentioned before, in cells synchronized in metaphase, EB1 was found to localize to the mitotic spindle, but could not be detected at the schizont surface. Blocking Cdk1 by treatment with the inhibitor RO-3306 resulted in the induction of furrow ingression, and a pronounced recruitment of EB1 to the schizont surface. This indicates that EB1 interaction with the schizont is, at least to some degree, negatively regulated by Cdk1 (Figure 6C).
Next we aimed to establish whether membrane-anchored p104 is capable of recruiting endogenous EB1 independently of other parasite proteins. To that extent, p104-CT was fused to the the C-terminal tail of the vaccinia virus F1L protein which targets the outer membrane of mitochondria [46]. V5-tagged p104-CT was targeted to COS-7 cell mitochondria and EB1 localization monitored by IFM. In transfected cells in which p104-CT was clearly visible at the mitochondria, EB1 was also notably re-distributed to the mitochondria (Figure 7A) Interestingly, in most instances, EB1 mislocalization was accompanied by reduced MT plus end binding. Importantly, whereas p104-CT-SKNN was also targeted to the mitochondria, it failed to recruit EB1 and, in such cells, EB1 localized correctly to MTs (Figure 7A lower panel). Figure 7B shows an area containing two transfected cells and one untransfected cell for comparision, demonstrating that in untransfected cells, EB1 was normally distributed to the plus ends of MTs. These data show that membrane-bound p104 functions as an EB1-recruiting protein, independently of other parasite proteins.
In a process known as dynamic instability, MTs alternate between phases of growth and shrinkage, allowing them to ‘explore’ different regions of the cytoplasm, and contribute to intracellular organization, cellular function and polarity. The Theileria schizont taps into this network, and the interaction between the schizont and host cell MTs during mitosis and cytokinesis is crucial for the persistence of the parasite in its transformed host [5]. Considering the intricate interactions observed during mitosis and cytokinesis [5], the association between the schizont and host cell microtubular network can be expected to be complex involving different host and parasite factors. The +TIP interaction network is large and dynamic, and EB1 fulfills the role of ‘master regulator’. Using pull-down approaches as well as co-localization and MT plus end tracking assays we now demonstrate that p104, a major T. annulata schizont surface protein, is a genuine EB1-binding protein of the SxIP family. We posit that, by providing docking sites for EB1, p104 contributes to capturing growing MTs emanating from the centrosomes in interphase cells and from spindle poles during mitosis. Whereas our bioinformatics search (using http://old.genedb.org/genedb/annulata) did not predict the presence of other GPI-anchored schizont proteins containing a potential SxIP motif, our findings do not exclude the possibility that other parasite surface proteins, or proteins secreted by the schizont, participate in EB1 binding. SxIP motifs can be found in a wide range of proteins in different organisms including Theileria. For instance, a bioinformatics analysis of two transforming (T. annulata and T. parva) and one non-transforming Theileria (T. orientalis) yielded 217, 201 and 154 proteins with predicted SxIP motif, a number of which also contained a predicted signal peptide (see Figure S5 and Table S3). At this stage, it is not clear if the reduced number of SxIP motif-containing proteins in T. orientalis reflects different requirements for EB binding partners in the parasite's life cycle. Whether any of these many candidate proteins function as genuine EB1-binding partners that could potentially interact with host cell EB1 can only be established experimentally. This is underpinned by our observation that TA20980, TA17545 and TA17375 all failed to co-localize with EB1 at MT plus ends in COS-7 cells, despite the presence of convincing SxIP-motifs.
While EB1 might facilitate the initial interaction of MTs plus ends with the schizont, it can be assumed that additional molecules of parasite and/or host cell origin are required to stabilize MTs at the schizont surface. Preliminary observations suggest that host cell CLASP1, a protein with known MT-stabilizing activity [14], could contribute to MT stabilization at the parasite surface. In this context, we recently found CLASP1 to be constitutively recruited in abundant amounts to the schizont surface (unpublished). How CLASP1 associates with the parasite surface is presently under investigation. To what extent TaSP [7] or gp34 [6] are potentially involved in these processes is presently unclear.
The identification of p104 as a major schizont surface protein was surprising. In earlier work, p104 was described as a strongly antigenic protein localized in the microneme-rhoptry complexes of sporozoites [30]. Mass spectrometric analysis [29] and screening of a cDNA expression library prepared from schizonts, however, confirmed that p104 is also expressed by the transforming schizont. We also established that the mAb 1C12, used as a useful diagnostic tool and marker of the parasite [36], recognizes the N-terminal region of p104.
When expressed as a soluble protein in COS-7 cells, p104 shows MT plus end tracking, typical of +TIPs. The fact that the p104 EB1 binding domain encompassing the SKIP motif could be reduced to as little as 40 aa excludes a role in EB1 recruitment of the FAINT domains, contained in the N-terminal half of p104. Importantly, in all experiments, p104/EB1 interaction was critically dependent on an intact SKIP motif and mutation to SKNN abrogated EB1 binding in all experiments.
For a number of +TIPs, the affinity for EB1or MTs is regulated by phosphorylation in a spatio-temporal manner [47]. For example, MCAK binding to EB1 is negatively regulated by Aurora B phosphorylation [33], and Cdk1-dependent phosphorylation of SLAIN2, a protein that links MT plus end–tracking proteins and controls MT growth during interphase, disrupts its binding to EB1 during mitosis [42]. In the case of CLASP2, multisite Cdk- and GSK3-dependent phosphorylation was found to fine-tune the strength of CLASP2/EB1 interaction [48], [49]. Like many +TIPs, p104 is also phosphorylated. Phosphorylation is extensive and cell cycle-dependent changes could be observed. Upon treatment with the Cdk1 inhibitor RO-3306, pronounced alterations in the p104 phosphorylation pattern were observed. Phosphorylation was not completely abrogated, however, indicating that p104 is also a substrate for kinases other than Cdk1. The exact role of p104 phosphorylation in EB1/p104 interaction is still unresolved. Pull-down experiments revealed that different phosphorylated forms of p104 can bind GST-EB1. In this regard, p104 differs markedly from SLAIN2, which, in its hyperphosphorylated state, fails to bind to GST-EB1 [42].
While inhibition of the mitotic kinase Cdk1 clearly resulted in increased EB1 accumulation at the schizont surface, our observations indicate that the cell cycle-dependent interaction of EB1 with the schizont surface during mitosis is regulated in a more complex manner than merely through a selective, phosphorylation-dependent interaction of EB1 with p104.
Endogenous EB1 may differ substantially from recombinant EB1 used in pull-down experiments and display altered ligand-binding properties during the cell cycle. The spatiotemporal regulation by which EB1 is recruited to different host cell structures involves its interaction with specific proteins that are often modified in a cell cycle-dependent manner [47]. It is worth noting, however, that EB1 itself also undergoes cell cycle-specific modifications. This includes acetylation of K220 in the EBH domain, which has been shown to prevent EB1 interaction with various SxIP proteins [50]. Significantly, Xia and colleagues found that K220 acetylation is readily apparent in metaphase cells and decreased in anaphase cells. EB1 is not detected at the parasite surface during metaphase, but reappears at anaphase. Considering the striking coincidence, it is conceivable that EB1 interaction with the parasite is also regulated by acetylation. The dynamic acetylation of EB1 was found to help orchestrate accurate kinetochore/MT interactions and acetylation [50]. Thus, while it is required to contribute to important mitotic functions – such as the alignment of host cell chromosomes in preparation for exit from mitosis [43], [51] - EB1 might be transiently inaccessible for the parasite.
That EB1 dissociates from the parasite surface during metaphase could be considered counterintuitive. The transient lack of EB1 association with the schizont surface is not accompanied by a loss of MTs, however. On the contrary, we have shown that the parasite remains tightly associated with host cell MTs throughout mitosis [5] (see also the control panel in Fig. 3B). During anaphase, EB1 might contribute to the docking of central spindle MTs to the parasite surface, which we have shown to be essential for the distribution of the parasite over the two daughter cells [5].
Cell cycle-dependent recruitment of host cell proteins to the schizont surface is not without precedent. A similar phenomenon has been observed for host cell Plk1, another protein important for mitotic progression. Plk1 associates with the parasite surface during G2 and early mitosis, is released from the parasite surface during prometaphase, and promptly re-associates as the cell enters anaphase [5].
Although our observations strongly indicate that p104 acts as an EB1-binding protein, up till now, we have failed to interfere with EB1 binding to the schizont surface. Following transfection of TaC12 cells with mitochondria-targeted p104, very few (<0.1% of those transfected) surviving cells were recovered rendering a meaningful analysis impossible. Overexpression by transient transfection or microinjection of soluble p104 (GFP-p104554–593) failed to disrupt EB1/schizont interaction. GFP-p104554–593 plus end tracking in microinjected T. annulata-infected cells resembled that observed for EB1-GFP, and GFP-p104554–593-labeled MT plus ends could also be observed tracking along the parasite surface, suggesting that these short p104 fragments had neither a dominant negative effect on the association of EB1 with MT plus ends nor on EB1 binding to the parasite surface. One explanation could be that membrane-bound p104 binds EB1 with higher affinity than soluble p104. That membrane-bound p104 can indeed provide a strong docking site for EB1 is reflected by the fact that EB1 was extensively mislocalized in cells expressing p104 targeted to the outer membrane of mitochondria. Additionally, in a competitive binding assay, a three-fold molar excess of p104554–593 was required to obtain 50% inhibition of EB1 binding to recombinant p104 (data not shown), suggesting that short soluble fragments of p104 might not be good competitors for EB1-p104 binding. An alternative explanation could be that EB1 binding sites other than the one contained in p104 are available at the parasite surface. In this context, it is of interest that another +TIP, CLASP1 (CLIP-170 associating protein 1), which contains two SxIP motifs and acts as a verified EB1 interaction partner [27], was recently found decorating the entire surface of the schizont (unpublished observation from our laboratory). It is thus conceivable that SxIP motifs of parasite-associated CLASP1 provide additional EB1 docking sites. Considering the importance of MT recruitment for parasite persistence, the existence of redundant EB1 recruitment mechanisms at the parasite surface would not be surprising.
The schizont, as a cytoplasm-dwelling organism, shares to some extent properties with cellular organelles like the endoplasmic reticulum, the Golgi apparatus and the endosomes/lysosomes, which all depend on MT dynamics for their positioning and function in the cell (reviewed in [52]). p104 could fulfill a role similar to STIM1 (stromal-interaction molecule 1), one of the few well-characterized EB1-binding +TIPs described up till now that is not cytoplasmic. STIM1 is involved in ER growth and remodeling [53] and, by binding EB1, mediates interaction between MT ends and ER membranes, required for ER tubule extension and membrane remodeling [53].
To our knowledge, Theileria p104 provides the first example for a pathogen-derived protein capable of interacting with EB1 via a functional consensus SxIP motif. The movement protein of tobacco mosaic virus (TMV-MP), which facilitates cell-to-cell spread of infection, was also reported to interact with EB1 in vivo and in vitro, [54]. TMV-MP does not contain an SxIP motif, however, and also the mode of interaction with EB1 was not elucidated.
In summary, Theileria parasites have evolved mechanisms to ‘hijack’ proteins that function as ‘hubs’ (those involved in many interactions) or ‘bottlenecks’ (proteins central to many pathways) [55]. Plk1, an important regulator of mitosis [5] or IKK, a central regulator of NF-κB pathways [56], are two striking examples of important proteins recruited to the parasite surface. As a cytoplasmic parasite co-evolving with its mammalian host, the parasite has also acquired the ability to hijack EB1, the ‘master regulator’ of host cell MT dynamics, thereby gaining access to the regulatory mechanisms that control the microtubular cytoskeleton.
A SMART cDNA library that consists of T. annulata cDNA embedded in the multiple cloning site of the phage λTriplEx2 (Clontech) was provided by Gordon Langsley [31]. E. coli were transduced with the phage library, and approximately 2 x 105 plaques were screened with the mAb 1C12 [34] Phage DNA was converted into a plasmid (pTriplEx2) by site-specific recombination, and sequenced to identify TA08425.
T. annulata (TaC12)-infected macrophages, the SV40-transformed cell line of Theileria-uninfected bovine macrophages (BoMac) and COS-7 cells were cultured as described previously [5]. Schizonts were purified from TaC12 cells, basically as described previously [57]. Transfections were performed following manufacturer's instructions using Lipofectamine 2000 (Invitrogen) for COS-7, or Amaxa 4D Nucleofection (Lonza) (cell solution SF, program DS103) for TaC12 cells. For depolymerization of MTs and the induction of cell cycle arrest in prometaphase, cells were treated with 0.1 µg/ml nocodazole (Biotrend) for 16 h, and harvested by shake-off. To arrest cells in metaphase, cells were harvested by shake-off following nocodazole treatment, and transferred into medium containing 20 µM MG132 (Alexis) for 2 h. For inhibition of Cdk1 activity, metaphase-synchronized TaC12 cells were incubated with 10 µM RO-3306 (Alexis) in MG132-containing media for 30 min.
C-terminal GFP-tagged mouse EB1 and human EB3 were generous gifts from Anna Akhmanova (Utrecht University). The EB1-GFP plasmid was used as a template for subsequent cloning of mouse EB1 cDNA fragments into pmaxFP-Green N or Green C vectors (Lonza) by PCR and using Sac1 and Kpn1 restriction enzymes, or pEF6-Myc-His (invitrogen) with Kpn1 and Not1. For a list of primers, please see Table S1. For expression of V5-tagged proteins in COS-7 cells, coding regions of TA08425 (p104) were amplified from genomic DNA (gDNA) from Theileria annulata (TaC12)-infected macrophages, and cloned using EcoRV and Xho1 into a modified version of pmaxCloning vector (Lonza) that contains a C-terminal V5 tag. Fragments of TA17375 and TA20980 were amplified from TaC12 cDNA and gDNA respectively, and TA17545 was amplified from a cDNA clone (a kind gift from Gordon Langsley) prior to cloning into the modified pmax-V5 vector with EcoRV and Xho1. For mitochondrial targeting of p104, the 20 amino-acid C-terminal tail of the vaccinia virus F1L protein [46] was cloned downstream of the V5 tag in the pmaxCloning vector using Age1 and Sac1, prior to the subsequent insertion of p104 fragments. The SKIP motif of p104 was mutated to SKNN by PCR mutagenesis. For live cell imaging, p104 fragments were cloned into pmaxFP-Green C using HindIII and BamH1. For expression of recombinant protein, p104 fragments with a C-terminal V5 tag were cloned into the pFN18A HaloTag T7 Flexi vector (Promega), using Sgf1 and Nco1. For expression of recombinant mouse EB1 in E. coli, mouse EB1 cDNA was cloned with a C-terminal myc tag into the pFN18A HaloTag T7 Flexi vector using Sgf1 and Nco1 restriction enzymes, or the pGEX-6P2 vector with EcoR1 and Not1. T. annulata EB1 was amplified from TaC12 cDNA and cloned using Nco1 and Kpn1 into the expression vector pHIS-parallel1 [58]. For expression of recombinant GFP-p104 fragments, the copGFP coding region was amplified from pmaxFP-Green N vector and cloned into the pGEX-6P2 vector (GE healthcare) with BamH1 and EcoR1, before subsequent insertion of fragments of p104 with EcoR1 and Xho1.
TaC12, BoMAC, and COS-7 lysates were prepared for western blotting and pull-down assays in modified RIPA buffer (50 mM Tris-HCl pH7.4, 150 mM NaCl, 1 mM EDTA. 1% NP-40, 0.25% Na-deoxycholate, 2 mM Na-vanadate, 25 mM NaF, protease inhibitor cocktail, Roche, and phosphatase inhibitor cocktail, Sigma). For removal of phosphate groups prior to SDS-PAGE analysis, lysates were prepared in the absence of phosphatase inhibitors and treated with lambda protein phosphatase (NEB), following manufacturer's instructions. Recombinant p104-V5 and EB1-myc protein were expressed and purified from E. coli using the HaloTag Protein Purification System (Promega) following manufacturer's instructions. For pull-down assays with recombinant p104, approximately 500 µg COS-7 lysate were incubated with Halo-tagged proteins overnight, before being washed 5 times in wash buffer (containing 500 mM NaCl and 0.01% tween-20), and eluted by TEV protease cleavage. Recombinant mouse EB1 and GFP-p104554–593 were purified from E. coli using Glutathione Sepharose 4b (GE Healthcare) followed by cleavage with Precision protease (GE Healthcare), following manufacturer's instructions. For pull-down assays with GST tagged EB1, 1 mg TaC12 lysate was incubated with GST-EB1 coated Glutathione Sepharose 4b for 3 hours, before being washed 5 times in wash buffer (containing 500 mM NaCl and 0.01% tween-20), and eluted by precision protease cleavage. Primary antibodies used for western blotting and immunofluorescence analysis were mouse mAbs anti-p104 (clone 1C12), anti-V5 (Invitrogen), anti-α-tubulin (clone DM1A, Sigma), anti-cyclin B1 (clone GNS-11, Pharmingen), c-Myc (9E10, Santa Cruz), anti-His (GE Healthcare) rat monoclonal anti-EB1 (clone KT51 Absea), and rabbit polyclonal anti-Turbo GFP (AB514, Evrogen). Rabbit polyclonal anti-TaSP antibodies were generated by the laboratory of Isabel Roditi (Bern) using the procedure described by Schnittger et al. [59].
Interphase cells were grown on coverslips, and cells harvested by mitotic shake-off were seeded on poly-L-lysine coated coverslips (Sigma). Samples were fixed with methanol for 10 min at −20°C or with 4% paraformaldehyde in PBS for 10 min at room temperature followed by permeabilization in 0.2% Triton-X-100. Antibody incubations were performed in PBS containing 10% heat-inactivated FCS. DNA was stained with DAPI, and the mitochondria with MitoTracker (Molecular Probes), and cells were mounted using DAKO mounting media. Wide-field microscopy was performed with a Nikon Eclipse 80i microscope equipped with a Retiga 2000R CCD camera (Qimaging) using 60x and 100x Plan Apo objectives (Nikon) and Openlab 5 software (Improvision). Images were processed using Photoshop (Adobe). Time-lapse imaging was performed by recording fluorescence at 2 sec intervals for 2 min using a TE2000E-PFS microscope (Nikon) equipped with a Plan Fluor 60x objective (Nikon), Orca ER CCD camera (Hamamatsu), and incubation chamber (Life Imaging Services). Microinjection was performed using the FemtoJet microinjection system and Femtotip II microinjection capillaries (Eppendorf).
A schematic presentation of the two approaches that were followed is shown in Figure S5. For the initial identification of p104, T. annulata GeneDB (http://old.genedb.org/genedb/annulata) was queried using the option ‘Complex/Boolean Query’ for genes encoding ‘proteins containing a predicted signal peptide’ and ‘proteins containing a predicted GPI-anchor’ yielding 559 and 33 entries, respectively. Intersection of the selected queries resulted in 19 candidates. Downloaded sequences were screened for the presence of an SxIP motif (LIG_SxIP_EBH_1) using the ELM (Eukaryotic Linear Motif) resource (http://elm.eu.org) and for representation in the T. annulata schizont proteome database [29]. TA08425 (p104) emerged as the only candidate fulfilling all criteria: signal peptide, GPI-anchor, evidence of protein expression and an SxIP motif located in a disordered region and surrounded by the appropriate residues as defined [28], [33].
The following approach was used to search for SxIP motif-containing proteins in T. annulata, T. parva and the non-transforming T. orientalis: a genome-wide search at predicted proteome-level in all three publicly available Theileria genomes was performed. T. parva and T. annulata proteome sequences were obtained from PiroplasmaDB 2.0 release while the T. orientalis sequence was retrieved from the NCBI genome database. These consisted of 4129, 3799 and 3734 predicted proteins, respectively. We obtained the SxIP motif pattern (LIG_SxIP_EBH_1) from the Eukaryotic Linear Motif (ELM) database [60]. To screen the full proteome, we used a simple SxIP motif search using the EMBOSS (http://emboss.sourceforge.net/) ‘preg’ utility. The filtered proteins were submitted through a perl script to the ELM web server for the prediction of SxIP motif with disorderness and required surrounding residues. Signal peptides were predicted for SxIP proteins using standalone SignalP 3.0 [61] and SignalP 4.0 [62] tools with default settings. For GPI anchor prediction, the PredGPI tool [63] with default parameters was used. To establish the species specificity, we performed the ortholog analysis using OrthoMCL 2.0 [64]. For OrthoMCL analysis we included 41 species. For Apicomplexa we included Babesia, Plasmodium, Toxoplasma, Cryptosporidium, Eimeria and Neospora species. In addition to Theileria and other apicomplexan species, to determine phyletic distribution amongst other Eukaryotes, we included a major ciliate, Perkinsus, a couple of Chromerids, Green, Brown, Red algal genomes, Arabidopsis, Diatoms, etc (for details, please see ‘Taxonomy list’ in Table S3). The OrthoMCL clusters were generated on default settings. Based on the phyletic distribution of Ortholog clusters as determined by Ortho-MCL, we categorized gene clusters as restricted to Theileria (termed as ‘Theileria-specific’, or restricted to apicomplexan species (referred to as ‘Apicomplexa-specific’) or commonly present across Eukaryotes (referred to as ‘Eukaryotes’).
|
10.1371/journal.ppat.1005659 | Nipah Virus C Protein Recruits Tsg101 to Promote the Efficient Release of Virus in an ESCRT-Dependent Pathway | The budding of Nipah virus, a deadly member of the Henipavirus genus within the Paramyxoviridae, has been thought to be independent of the host ESCRT pathway, which is critical for the budding of many enveloped viruses. This conclusion was based on the budding properties of the virus matrix protein in the absence of other virus components. Here, we find that the virus C protein, which was previously investigated for its role in antagonism of innate immunity, recruits the ESCRT pathway to promote efficient virus release. Inhibition of ESCRT or depletion of the ESCRT factor Tsg101 abrogates the C enhancement of matrix budding and impairs live Nipah virus release. Further, despite the low sequence homology of the C proteins of known henipaviruses, they all enhance the budding of their cognate matrix proteins, suggesting a conserved and previously unknown function for the henipavirus C proteins.
| Nipah virus is a deadly pathogen (40–100% mortality) that has yearly outbreaks in Southeast Asia, resulting from spillover from its natural fruit bat reservoir. The viral C protein is one of only nine virus proteins, but its role in promoting virus replication is not fully understood. Here, we found that the C protein promotes the efficient release of budding Nipah virus from infected cells. It does so by recruiting an essential factor in the host ESCRT complex, Tsg101. The ESCRT complex has well-characterized functions in mediating membrane pinching off events that resemble virus budding. Further, we found that the C proteins of related viruses within the same genus (Henipavirus) also promote virus budding, suggesting that this previously unknown function of the henipavirus C proteins is conserved. This work illuminates the basic biology of henipaviruses with significant outbreak and public health concern, and opens the door to further lines of inquiry.
| The host ESCRT (endosomal sorting complex required for transport) pathway catalyzes cellular membrane scission events, such as multivesicular body formation and cellular abscission, in which the cytoplasm has access to the inside of the constricting membrane neck [1,2]. Many enveloped viruses have co-opted the ESCRT pathway to catalyze the membrane scission required to pinch off and release budding virions [3]. HIV-1, for example, recruits the ESCRT pathway via specific motifs on the virus Gag protein, and inhibition of ESCRT prevents HIV-1 release from the cell surface [4–6]. Influenza A virus (IAV) is a rare example of an enveloped virus with a delineated ESCRT-independent budding pathway, in which IAV encodes its own factor to catalyze membrane scission [7].
Nipah virus (NiV), a deadly zoonotic virus within the Henipavirus genus of the Paramyxoviridae family, has been thought to have an ESCRT-independent budding pathway. The Nipah matrix protein (NiV-M) has the central role in virus assembly and budding [8,9]. The budding of NiV-M, which is released from cells in virus-like particles (VLPs) when expressed on its own, was not affected by inhibition of ESCRT [10]. Although two motifs in NiV-M have been suggested as potential late domains (ESCRT-dependent motifs that trap budding at a “late” step when disrupted), no evidence that they act as late domains in the context of NiV-M was presented [10,11]. As this lack of evidence for a role for ESCRT in NiV budding was based on NiV-M VLP release in the absence of other viral components, we wondered if NiV budding might be ESCRT-dependent in the context of the full virus.
Inhibition of ESCRT resulted in a significant decrease in released NiV titers of several logs, indicating that in the context of live virus replication, NiV has an ESCRT-dependent budding pathway. We then investigated whether viral components other than NiV-M might be responsible for this reliance on ESCRT. To our surprise, we found that the NiV C protein, an accessory factor previously thought to solely play a role in counteracting innate immunity, enhanced NiV-M budding in an ESCRT-dependent manner. This finding was reminiscent of previous findings that the C protein of Sendai virus (SeV, a related paramyxovirus in the Respirovirus genus) enhanced SeV-M release in a manner dependent on the ESCRT factor Alix [12–14]. However, depletion of Alix or inhibition of ESCRT was not found to affect the budding of live SeV [15], making the functional significance of these findings regarding SeV-C unclear.
Interestingly, a large segment of NiV-C aligned with the essential host ESCRT factor Vps28. We found that like Vps28, NiV-C interacted directly with the ESCRT factor Tsg101, which was required for the NiV-C enhancement of budding as well as the efficient release of live NiV. Finally, the C protein of other known henipaviruses also enhanced the budding of their cognate matrix proteins, suggesting a conserved role for the henipavirus C protein in the budding process.
The classical test for ESCRT-dependence is sensitivity to a dominant-negative Vps4, the indispensable ATPase that drives the scission process [1,3]. We created 293 stable cell lines with inducible overexpression of either wild-type (WT) or dominant-negative (DN) Vps4A (S1 Fig). To interrogate the involvement of the ESCRT pathway in the budding of live NiV, we infected the Vps4A-inducible 293 cell lines with recombinant NiV expressing a secreted Gaussia luciferase reporter [16–18]. WT or DN Vps4A was induced post-infection, and not before, to avoid any potential effect of Vps4A overexpression on the infection event itself. While induction of WT Vps4A did not affect viral titers released into the supernatant at 12 and 24 hours post-infection (hpi), induction of DN Vps4A reduced such titers by over 3 logs at 24 hpi (Fig 1A). Notably, neither WT nor DN Vps4A affected the amount of Gaussia luciferase reporter secreted into the infected cell supernatant, indicating that the defect in released viral titers was not due to a deleterious effect on virus protein production per se (Fig 1B). Furthermore, when we infected the inducible cell lines with wild-type non-recombinant NiV and compared the amount of NiV-M found in pelleted virions by Western analysis, we found that the induction of DN but not WT Vps4A resulted in a marked reduction in NiV-M release (Fig 1C). Altogether, our results suggest that virion production and release in the context of live NiV replication is an ESCRT-dependent process that is sensitive to DN Vps4A inhibition.
NiV-M budding in the absence of other virus components is ESCRT-independent and insensitive to DN-Vps4A inhibition [10]. However, since live NiV budding appears to be ESCRT-dependent, and is sensitive to DN Vps4A inhibition, we wondered what viral component(s) might promote M-mediated virus budding in an ESCRT-dependent pathway. Previous studies suggested that co-expression of NiV nucleocapsid protein (N), fusion protein (F), or attachment protein (G) did not affect matrix release [9,11]; however, other NiV proteins such as the phosphoprotein (P), V protein (V), W protein (W), and C protein (C) (Fig 2A) were not tested for a potential role in budding, perhaps due to their previously characterized roles in virus transcription and/or antagonism of innate immunity.
We co-transfected NiV-N, -P, -V, -W, -C, -F, and -G with NiV-M in 293T cells in a VLP budding assay. NiV-M VLPs released in the supernatant were pelleted through a 20% sucrose cushion, and relative amounts of M in VLPs versus cell lysates were compared by quantitative Western blotting. Consistent with previous reports, co-expression of NiV-N, -F, and -G did not affect the efficiency of M VLP release (Fig 2B). Only co-expression of NiV-C significantly enhanced M VLP release (Fig 2B), suggesting a role for C protein in the NiV budding pathway. Importantly, NiV-C and NiV-M expression from our transient transfection studies was equal to or lower than the natural level of expression observed in live NiV infection when comparing equivalent conditions and time points (S2 Fig). These data underscore the physiological relevance of our findings.
We reasoned that if C protein had a role in budding, knockout of C protein from the live virus should result in decreased virus release. We mutated the Gluc-expressing NiV to delete expression of NiV-C without affecting expression of the overlapping NiV-P ORF, as done previously [19–22]. C-deficient NiV had decreased virus release as determined by infectious titers (Fig 2C) as well as Western analysis of physical virion production (Fig 2D), despite having equivalent virus protein production as determined by the Gluc reporter (Fig 2E).
Finally, to determine whether the C enhancement of NiV-M budding was dependent on ESCRT, we performed the M VLP budding assay in the Vps4A-inducible 293 cells. While NiV-C enhancement of M release was not affected by induction of WT Vps4A expression (Fig 2F, compare lanes 3 and 4), it was abrogated by DN Vps4A (Fig 2F, compare lanes 7 and 8). Altogether, our data suggest that NiV-C enhances NiV-M budding in an ESCRT-dependent manner.
The NiV C protein has mainly been studied for its role in antagonism of innate immunity [20–23], although there is evidence that NiV-C has other functions in NiV replication. For example, while C-deficient NiVs, created through reverse genetics, induce a more robust IFN-β response and are attenuated both in vitro and in vivo, C-deficient NiVs are also unexpectedly attenuated in IFN-α/β -deficient Vero cells (1–2 logs reduction in viral titers compared to wild-type NiV) (Fig 2C) [19,20,22]. The latter suggests that the C protein has additional proviral function(s) that are independent of its immune antagonism activities. Our data suggest that C protein enhancement of NiV-M budding may be one such function.
Thus, we asked whether and how the NiV-C protein might interact with NiV-M to promote budding. We found that NiV-C and NiV-M interact by co-immunoprecipitation (Fig 3A), that NiV-C does not bud by itself but is efficiently incorporated into M VLPs (Fig 3B), and that NiV-C and NiV-M are co-localized at the cell surface as revealed by confocal microscopy (Fig 3C). Our results are consistent with a previous report that C protein is present in NiV virions [24] and suggest that C protein may directly interface with NiV-M at the cell surface to promote M VLP release.
To gain further insight into the mechanism of NiV-C enhancement of M budding, we performed a structural homology modeling search of NiV-C using the Phyre2 program [25]. Interestingly, a large middle segment (MiD) of the C protein aligned to the C-terminal domain (CTD) of the host factor Vps28, an essential component of the ESCRT pathway (Fig 4A).
Vps28, a component of the ESCRT-I subcomplex, plays an adaptor role between ESCRT-I and downstream factors, particularly the ESCRT-II subcomplex [1,29]. In yeast, the N-terminal domain of Vps28 interacts directly with Vps23 (human Tsg101) in ESCRT-I, whereas the C-terminal domain interacts directly with Vps36 in ESCRT-II. Since the MiD of NiV-C aligned with the C-terminal domain of Vps28, we wondered if NiV-C might also analogously interact with human Vps36 (hVps36). However, we could not detect specific co-immunoprecipitation of endogenous or overexpressed hVps36 with either NiV-C or human Vps28 (hVps28). This is perhaps not surprising as the ESCRT components between yeast and humans, though highly conserved, are not completely orthologous. For example, while hVps28 and hVps36 can be found associated with other factors in a complex, the interaction of recombinant hVps28 and hVps36, unlike their yeast counterparts, could not be detected in the absence of these other factors [30]. Indeed, hVps28 and hVps36 likely interact via a different, yet to be characterized interface, as charged side chains in metazoan Vps28 not present in the homologous yeast Vps28 would disrupt the known interface between yeast Vps28 and Vps36. Furthermore, metazoan Vps36 also lacks the zinc-finger domain present in yeast that interacts with Vps28 [1,26].
The NiV-C-aligned C-terminal domain of Vps28 forms a conserved four α-helix bundle (Fig 4B) [26,28]. Despite the abovementioned differences, close inspection of the alignment between NiV-C and hVps28 revealed a striking stretch of nearly identical amino acids between NiV-C and hVps28 (Fig 4A, orange box) that is predicted to be in the 3rd α-helix, which in the homologous yeast (S. cerevisiae) structure lies at the interface between Vps28 and Vps36 [26] (Fig 4B). Since NiV-C appears to mimic this stretch of residues in Vps28, we hypothesized that this Vps28 interface is functionally conserved in the ESCRT recruitment machinery even if the specific ESCRT components recruited by yeast and human Vps28 might actually differ. Our hypothesis predicted that disruption of this conserved, strategically located stretch of residues in NiV-C might also disrupt the NiV-C function of enhancing the budding of NiV-M.
To interrogate our hypothesis, we mutated the central portion of this conserved stretch in NiV-C to alanines, and compared these mutants to alanine mutations in neighboring residues also conserved in the alignment between NiV-C and hVps28, but not predicted to be functional Vps28-Vps36 interface residues in their homologous positions in yeast Vps28 (Fig 4B and 4C). Mutagenesis of W103 and L104 within this central conserved stretch, but not the neighboring residues (P88 and M110), significantly reduced the NiV-C enhancement of NiV-M budding (Fig 4C). Although mutagenesis of W103 and L104 led to lower expression of NiV-C, the M110A mutant, which also had lower NiV-C expression, continued to enhance NiV-M budding as well as WT NiV-C. This result is consistent with the possibility that NiV-C-mediated recruitment of ESCRT-II is important for the budding role of C protein.
The alignment of NiV-C with Vps28 suggested that C protein might also interact with the other known interacting partner of Vps28, the ESCRT-I factor Tsg101. Indeed, like hVps28, NiV-C immunoprecipitated endogenous Tsg101 (Fig 5A). To define the portion of NiV-C responsible for this interaction, we performed truncations of NiV-C at the N- (ΔNTD) and C-terminal (ΔCTD) boundaries of the aligned middle domain (MiD) (Fig 5B). A double glycine immediately following NiV-C-MiD, perhaps representing a flexible linker between domains (Fig 4A), allowed for a natural demarcation for making the ΔCTD truncation mutant. Loss of the C-terminal domain (ΔCTD) abrogated the interaction of NiV-C with Tsg101 (Fig 5C). Furthermore, the CTD of NiV-C fused to EGFP, but not EGFP alone, also interacted with Tsg101, indicating that the CTD of NiV-C was both necessary and sufficient for the interaction (Fig 5D).
We then determined the corresponding NiV-C-interacting domain on Tsg101. Vps28 is known to interact with the C-terminal domain of Tsg101 (Tsg101-CTD) [31–33]. Following the hypothesis that NiV-C mimics the interactions of Vps28, we wondered whether NiV-C might also interact with Tsg101-CTD, analogous to Vps28. Truncation of Tsg101-CTD led to loss of interaction with NiV-C (Fig 5E and 5F). Conversely, Tsg101-CTD fused to EGFP, but not EGFP alone, was sufficient to bring down NiV-C (Fig 5G). These results indicated that Tsg101-CTD was both necessary and sufficient for the interaction.
Finally, to determine if the interaction between NiV-C and Tsg101 was direct, we purified recombinant full-length NiV-C, the EGFP-Tsg101-CTD fusion, and EGFP alone from E. coli (Fig 5H, left panel). As in Fig 5G, but with the purified recombinant proteins, we showed that NiV-C co-immunoprecipitated EGFP-Tsg101-CTD but not EGFP alone, suggesting that the interaction between NiV-C and Tsg101 was indeed direct (Fig 5H, right panel). Altogether, our results show that NiV-C-CTD and Tsg101-CTD, the C-terminal domains of NiV-C and Tsg101, respectively, interact directly without the need for any intermediary host factors.
Since hVps28 interaction with Tsg101 plays a critical role in the recruitment of ESCRT complexes that lead to membrane scission or budding events, we hypothesized that NiV-C-mediated recruitment of Tsg101 is also essential for its enhancement of NiV-M budding. To test this hypothesis, we identified the minimal truncation in NiV-C-CTD that would result in disruption of interaction with Tsg101, and we then determined whether this mutant would no longer enhance M release. Loss of only seven residues from the C-terminus (159tr) resulted in loss of interaction with Tsg101 (Fig 6A and 6B). This truncation mutant was not able to enhance M release (Fig 6C), consistent with the importance of the interaction with Tsg101. We therefore sought to determine whether depletion of Tsg101 would affect the C enhancement of M release as well.
Tsg101 is essential for cell growth and proliferation, and ablation of Tsg101 results in embryonic lethality in the peri-implantation period [34,35]. As such, it may be difficult to generate a stable Tsg101 knockout cell line. To overcome this barrier, we used a strategy that allowed maintenance of the essential function of Tsg101, and thus cell viability, until knockdown was desired. To do this, we leveraged CRISPR technology to knock-in a destabilization domain tag (DD) onto all endogenous copies of Tsg101 (Fig 6D), similar to a strategy we published previously for the essential gene TCOF1 [36]. In the presence of a stabilizing compound (in this case the antibiotic trimethoprim), the fusion protein is protected from proteasomal degradation, whereas in the absence of compound, the DD tag targets the fusion protein for degradation. Because removal of compound specifically targets only the DD-tagged protein for degradation, one can therefore test the effects of knockdown in an isogenic cell line with minimal off-target effects.
Genotyping PCR confirmed complete knock-in of the DD tag in all loci of single cell-cloned 293T cells (Fig 6E). We confirmed that removal of compound indeed destabilized DD-Tsg101 (Fig 6F) and resulted in impaired cellular growth (S3 Fig), as expected from the loss of the normal cellular functions of Tsg101, Removal of compound and thus depletion of Tsg101 abrogated the C enhancement in the knock-in DD-Tsg101 cells (Fig 6G). This result suggested that Tsg101 is required for the NiV-C enhancement of M release.
Next, we tested whether specific depletion of Tsg101 in the DD-Tsg101 knock-in cells would impair live NiV release. In experiments performed under BSL4 conditions, removal of the stabilizing compound resulted in a significant 1.5 log reduction in released titers in DD-Tsg101 cells (Fig 6H, left panel). In contrast, removal of the compound resulted in no reduction in the release of Rift Valley fever virus (Fig 6H, right panel), a bunyavirus without a matrix protein analog or any known role for ESCRT in its budding pathway [37].
Finally, we examined whether we could detect a defect in the late stages of budding by transmission electron microscopy (TEM). Although depletion of Tsg101 impaired NiV release, the virus was still able to bud and replicate, albeit at a significantly lower efficiency. We therefore expected that while TEM might not show clear signs of “trapped” virions as for HIV-1, which cannot bud without ESCRT, it might reveal a more disorganized, inefficient budding process upon depletion of Tsg101. Indeed, Tsg101 depletion resulted in accumulation of cell-associated virions that appeared in many cases to be stuck to each other and to the cell, suggestive of an inefficient budding process (Fig 6I). Paramyxovirus particles are much more pleiomorphic than HIV particles but identification is aided by the immunogold labeling (right panels, Fig 6I).
NiV, along with the closely related Hendra virus (HeV), are the founding members of the Henipavirus genus within the Paramyxoviridae family. Although sequencing of field samples from the natural bat reservoir suggests a wide diversity of henipaviruses, there are thus far only two additional complete or nearly complete sequenced henipavirus genomes: Cedar virus (CedPV) [38] and an African henipavirus isolated from bat in Ghana (Gh-M74a strain, GhV) [39] (Fig 7A). Another recently sequenced virus from rodents in China, Mojiang virus (MojPV) [40], has been suggested to be “henipa-like”, since it shares sequence similarity with the henipaviruses (Fig 7A) but has not yet been shown to share other conserved features, such as use of the ephrinB2 receptor.
Having identified a new role for the NiV C protein in budding, we wondered if this function might be conserved among the henipa- or henipa-like viruses. We co-transfected the C proteins of HeV, CedPV, GhV, and MojPV with their cognate matrix proteins (Fig 7B). The C proteins of HeV, CedPV, and GhV significantly enhanced M VLP release; by contrast, the C protein of MojPV did not enhance M VLP release. This result suggests that while C protein enhancement of budding is a feature conserved among known henipaviruses, it may not be conserved among more distantly related “henipa-like” viruses such as MojPV.
Although some enveloped viruses have been suggested to have budding mechanisms independent of the ESCRT pathway [3,41], such a mechanism has been described in molecular detail only for Influenza A virus, which encodes a factor capable of membrane scission [7]. Nipah virus was previously suggested to have an ESCRT-independent budding pathway, as NiV-M VLP budding in the absence of other virus components was insensitive to DN Vps4 inhibition [10].
In this work, we found that the efficient budding of live NiV is in fact sensitive to DN Vps4 inhibition (Fig 1), indicating an ESCRT-dependent budding pathway. This result also indicated that other viral component(s) besides NiV-M likely contribute to the budding process. Unexpectedly, we found that the viral C protein promotes budding by recruiting the ESCRT factor Tsg101, previously shown to play critical roles in the budding of other viruses [3,41]. The C-terminal domain of NiV-C interacted directly with the C-terminal domain of Tsg101 (Fig 5), and disruption of this interaction or specific depletion of Tsg101 resulted in abrogation of the budding enhancement function of NiV-C (Fig 6). We also found a suggestive sequence alignment between the middle domain of NiV-C and the C-terminal domain of the ESCRT factor hVps28 (Fig 4A). hVps28 plays an adaptor role in the ESCRT machinery, with one domain of hVps28 (the NTD) interacting with the ESCRT-I subcomplex via its essential component Tsg101, and the other domain of hVps28 (the CTD) interacting with a complex of downstream ESCRT factors including ESCRT-II components such as hVps36 [28,30]. The interesting parallel of one domain of NiV-C (the CTD), like hVps28-NTD, interacting with Tsg101-CTD while a different domain of NiV-C (the MiD) aligns to hVps28-CTD suggests that NiV-C-MiD, like hVps28-CTD, may also interact with downstream ESCRT factors (see model in Fig 8). Although we have not been able to detect hVps36 interaction with NiV-C, our finding that a cluster of residues conserved between NiV-C-MiD and hVps28-CTD was required for the NiV-C budding function (Fig 4C) supports the possibility that NiV-C may also act as an adaptor that interacts with multiple ESCRT components. The identity of potential downstream interacting factor(s) remains under investigation.
Our finding that NiV-C recruited ESCRT to mediate efficient budding was unexpected for several reasons. First, the C protein is the least conserved of the virus proteins in the Paramyxovirinae subfamily, with the C protein even being completely absent from the Rubulavirus and Avulavirus genera. Within the Henipavirus genus, only NiV-C and HeV-C are relatively similar (90% amino acid similarity) due to the high relatedness of NiV and HeV; the other pairwise comparisons between henipavirus C proteins show 39–52% similarity. That NiV-C should have such a critical function is therefore somewhat surprising. Second, the immune modulatory function of NiV-C has been well established, with multiple groups demonstrating clear effects of NiV-C on limiting type I interferon responses and proinflammatory cytokine upregulation [20–23]. NiV-C is the smallest of the viral proteins at 166 a.a., and if NiV-C embodied independent molecular mechanisms to both enhance budding and antagonize innate immunity, this would illuminate the principle that viruses have evolved to make the most of their limited coding capacity. It would be interesting to investigate, however, how much of the immune antagonism function of NiV-C is an indirect outcome of its recruitment of ESCRT. Viral recruitment of ESCRT factors may deplete their accessibility for normal cellular functions, and modulation of ESCRT would have pleiotropic effects due to the role of ESCRT in essential cell processes such as cytokinesis as well as multivesicular body formation, which is important for downregulation of transmembrane proteins via internalization into endosomes and subsequent degradation. Inhibition of ESCRT therefore leads to secondary effects including induction of autophagy, dysregulation of signaling pathways, and even modulation of RNAi activity, as RISC complexes are assembled on multivesicular bodies [42–44]. Third, the C protein is translated from an alternate reading frame within the NiV-P gene. There are thus strong coding constraints for the C protein resulting from the requirement to maintain the essential function of the NiV P protein [45]. That a virus protein under such coding constraints could mimic the sequence of a host factor, in this case the ESCRT factor Vps28, is remarkable. The extent to which the virus has used its limited coding capacity to recruit ESCRT for its budding pathway highlights its importance for virus replication and pathogenesis.
It is interesting to note that our results confirm that NiV-M can bud independently of ESCRT (Fig 2F), and it is possible that NiV-M possesses some intrinsic membrane deformation and scission capability, as reported for purified Newcastle disease virus matrix protein [46]. Nevertheless, recombinant C-deficient NiV, although replication-competent, is strongly attenuated in various cell lines in vitro as well as in vivo [19–22], including type I IFN-deficient Vero cells, indicating that there must be an unknown function of NiV-C independent of innate immune antagonism that is critical for virus replication. Our finding that NiV-C specifically recruits the ESCRT pathway to promote virus budding provides a mechanistic rationale for this heretofore unexplained phenomenon.
Finally, we found that the C proteins of other known henipaviruses also enhance matrix VLP release. A role in the budding pathway may therefore be a conserved feature of the henipavirus C proteins. We were interested to find that the C protein of Mojiang virus, which has thus far been considered “henipa-like” due to its greater sequence divergence (Fig 7A) [40,47], did not enhance budding. Further afield, the C proteins of some other paramyxoviruses such as certain members of Respirovirus have been thought to play a role in budding, although the data has been unclear or conflicting [12–15,48]. For example, for Sendai virus (SeV), the prototypic virus within Respirovirus, the C protein was suggested to enhance SeV-M-driven VLP budding through recruitment of the ESCRT component Alix [12,13]; in contrast, another study determined that knockdown of C protein expression or inhibition of ESCRT did not affect live SeV budding [15], thus leaving the significance of SeV-C for the budding process unclear. Other paramyxovirus C proteins have defined roles in viral RNA replication and antagonism of interferon signaling, and no known role in budding [45]. Within the henipaviruses, performing the same structural homology modeling search on HeV-C, CedPV-C, and GhV-C that was previously performed for NiV-C shows that while HeV-C also aligns to Vps28, due to the high sequence similarity between NiV-C and HeV-C, there appears to be no potential alignments of CedPV-C and GhV-C with factors associated with ESCRT. It will therefore be fascinating to determine whether the C-mediated enhancement of budding is ESCRT-dependent for the other henipaviruses, as it is for NiV.
Our finding that NiV-C recruits ESCRT to mediate efficient budding not only expands the universe of enveloped viruses that depend on this host pathway, but also defines a previously unknown, yet critical role for the henipavirus C proteins.
293T and Flp-In T-REx 293 cells (Invitrogen) were propagated in Dulbecco’s modified Eagle’s medium (Invitrogen) supplemented with 10% fetal bovine serum (Atlanta Biologicals) and penicillin/streptomycin. Flp-In T-REx 293 cells were additionally maintained in blasticidin and zeocin according to manufacturer protocol. To generate the Vps4A-inducible 293 cells, the Vps4A gene (the kind gift of Jiro Yasuda) was N-terminally tagged with HA and inserted into pcDNA5/FRT/TO (either wild-type Vps4A or containing the K173Q mutation that renders it dominant-negative), then transfected into the parental Flp-In T-REx 293 cells along with pOG44 (encoding the Flp-recombinase). Stable cell lines with doxycycline-inducible expression of wild-type or dominant-negative HA-Vps4A were generated by selection with hygromycin (replacing the zeocin) and blasticidin according to manufacturer recommendations. To generate the 293T cell line with destabilization domain (DD)-tagged endogenous Tsg101, the method was as described [36] but with several differences. The puromycin and blasticidin resistance donor constructs (as shown in Fig 6D) incorporated a left homology arm of 731 bp and a right homology arm of 536 bp specific to Tsg101. Further, a ecDHFR-based destabilization domain [49] sensitive to the compound trimethoprim (TMP) (Sigma, T7883) was used. The ecDHFR DD sequence was codon-optimized for human expression (Invitrogen) and synthesized as a GeneArt String (Invitrogen). Finally, a CRISPR construct with nickase (D10A) Cas9 was used, px335 (Addgene, cat# 42335, from Feng Zhang), with a double strand nick strategy for increased specificity. Upon single cell isolation and clonal expansion of the DD-Tsg101 cells, clones were genotyped with PCR using primers flanking (thus not contained within the donor constructs) the donor homology arms (as in Fig 6D and 6E). The sequence for the donor constructs is shown in S4 Fig, and sequences for CRISPR guide RNAs and primers are available on request. For virus-like particle budding and live virus infections performed with DD-Tsg101 cells, cells were incubated in either 10 uM TMP or DMSO vehicle for three days preceding the assay to ensure complete knockdown of DD-Tsg101. Removal of TMP was accomplished with three serial washes of trypsinized cells (pelleting, aspiration of supernatant, and resuspension) in media lacking TMP.
Nipah virus (199901924 Malaysia prototype strain) was kindly provided by the Special Pathogens Branch (Centers for Disease Control and Prevention, Atlanta, GA, USA). The recombinant NiV expressing Gaussia luciferase (Gluc) is as previously described [16–18] and incorporates a Gluc-P2A-EGFP reporter as an independent ORF between the N and P genes. The NiV-C knockout virus was generated from the Gluc-expressing recombinant NiV as done previously [19,21], by mutating the two initiating start codons for NiV-C (t2429c, t2432c) as well as introducing a downstream stop codon (c2438a), all without disturbing the overlapping NiV-P ORF. For the Vps4A-inducible 293 infection, cells in 12-well were infected at a multiplicity of infection (MOI) of 2 for 45 minutes at 37°C, washed in the well (with the final wash saved for determination of residual background titers), and incubated with media with or without 16 ng/mL doxycycline. Supernatants were collected at the indicated time points and stored at -80°C. Titers were determined by plaque assay on Vero cells, and Gluc activity was detected with the BioLux Gaussia Luciferase Assay (New England Biolabs). The infection of DD-Tsg101 293T cells was similar, except with the cells in the presence or absence of trimethoprim (as described above) both preceding and after the infection. All work with live virus was carried out under biosafety level-4 conditions in the Robert E. Shope and the Galveston National Laboratory at the University of Texas Medical Branch.
The codon-optimized NiV-M gene in pcDNA3 (pNiV-M) is as previously described [50]. The codon-optimized NiV-F gene and NiV-G genes were as described [51] in pcDNA3 with C-terminal HA tags. The native sequences of NiV-N, -P, -V, -W, and -C were HA-tagged and amplified from the previously described pTM1-based expression constructs [52] (with NiV-V, -W and -C amplified from the P gene construct) and inserted into pcDNA3. The human Vps28 gene was synthesized with a N-terminal HA tag as a GeneArt String (Invitrogen) and inserted into pcDNA3. Tsg101 was amplified from pcGNM2/TSG-F (NIH AIDS Reagent Program, Cat# 11483, from Eric Freed) along with the additional N-terminal 10 residues of Tsg101 not present in the construct (MAVSESQLKK) and inserted into pCMV-3Tag-1A (Agilent), which appends a N-terminal 3XFLAG tag. FLAG-EGFP-Tsg101-CTD represents residues 303–390 of Tsg101 fused to the C-terminus of EGFP via a GSG linker in pCMV-3Tag-1A. The codon-optimized HeV-M gene [18] was inserted into pcDNA3 with a N-terminal HA tag via a double glycine linker. Sequences for CedPV-M (Genbank JQ001776.1), GhV-M (Genbank HQ660129.1), and MojPV-M (Genbank KF278639.1) were codon-optimized (Invitrogen), HA-tagged, synthesized as GeneArt Strings (Invitrogen), and inserted in pcDNA3. Likewise, sequences for NiV-C (Genbank AF212302.2), HeV-C (Genbank AF017149.3), CedPV-C, GhV-C, and MojPV-C were codon-optimized with C-terminal HA tags via a glycine linker, synthesized as GeneArt Strings (Invitrogen), and inserted in pcDNA3. The codon-optimized NiV-C-HA expression construct (pNiV-C-opt) was used as the basis for all NiV-C mutant constructs. HA-EGFP-C-CTD represents residues 128–166 of NiV-C fused to the C-terminus of HA-EGFP via a GSG linker in pcDNA3.
Samples in SDS Laemmli buffer were run on reducing SDS PAGE, either 10% Tris-glycine or Any kD TGX gels (Bio-Rad). Upon transfer to low fluorescence PVDF (Immobilon-FL, Millipore; or Immun-Blot LF, Bio-Rad), the membranes were incubated in Odyssey blocking buffer (Li-Cor Biosciences), then incubated with primary antibodies, followed by fluorescent secondary antibodies. For imaging on a Li-Cor Odyssey imaging system, the secondary antibodies were goat IRDye 800CW or 680LT antibodies (Li-Cor Biosciences). For imaging on a Bio-Rad ChemiDoc MP imaging system, the secondary antibodies were goat Alexa Fluor 647 or 546 antibodies (Invitrogen). The following primary antibodies were used: rabbit anti-NiV-M [50], rabbit anti-NiV-C (produced by 21st Century Biochemicals from rabbits immunized with a purified peptide corresponding to amino acids 13–31 of NiV-C), rabbit anti-HA (Novus, NB600-363), rabbit anti-COX IV (Li-cor Biosciences, 926–42214), mouse anti-Tsg101 clone 4A10 (Genetex, GTX70255), and mouse anti-FLAG clone M2 (Sigma, F3165).
293T or Vps4A-inducible 293 cells in 6-wells were transfected with pNiV-M along with the indicated HA-tagged NiV protein expression plasmid, with pcDNA3 vector to 2 ug total per well, using BioT transfection reagent (Bioland) according to manufacturer protocol. The medium was changed at 4 hours post-transfection, and cell lysates and supernatants were collected at 24 hours post-transfection. Supernatants were clarified and ultracentrifuged through 20% sucrose at 145,000 x g for 1.5 hours. Pellets were resuspended directly in SDS Laemmli sample buffer. Serial dilutions of high-expressing sample were included on each gel during Western analysis to aid in quantification of relative amounts of NiV-M. The relative quantities were normalized to the NiV-M with vector only control, and the budding index was determined as (relative M in VLPs / relative M in cell lysates).
293T cells in 6-wells were transfected with 1 ug of expression constructs or empty pcDNA3 vector for a total of 2 ug per well, using BioT transfection reagent (Bioland) according to manufacturer protocol. At 24 hours post-transfection, cells were lysed in cold lysis buffer (1% NP-40, 100 mM Tris-HCl pH 7.5, 150 mM NaCl, 5% glycerol, 1 mM EDTA, 1X protease inhibitor cocktail (Roche)), clarified at 14k rpm for 5 minutes at 4°C, and incubated with 3 ug of rabbit anti-NiV-M [50] or rabbit anti-HA (Novus, NB600-363) at 4°C overnight. Following incubation with protein G agarose (Pierce) for 2 hours at 4°C, the bound protein was washed 4 times with cold wash buffer (same composition as lysis buffer but with 0.2% NP-40), then eluted in reducing Laemmli SDS sample buffer at 95°C.
pNiV-M and pNiV-C-opt were co-transfected into 293T cells plated on coverslips coated sequentially with poly-L-lysine and collagen. At 24 hours post-transfection, cells were fixed in 2% paraformaldehyde, washed 3 times with PBS, incubated in blocking buffer (0.5% saponin, 3% BSA in PBS), then incubated with 4.7 ug/mL rabbit anti-NiV-M [50] and 4 ug/mL chicken anti-HA (Novus, NB600-361) in blocking buffer. After 3 washes in 0.5% saponin in PBS, the coverslips were incubated with goat anti-rabbit Alexa Fluor 594 and goat anti-chicken Alexa Fluor 488 (1:1000, Invitrogen). After further washing and staining with DAPI to visualize nuclei, coverslips were mounted on slides. Confocal imaging was performed on a Leica SP5 confocal microscope in the microscopy core facility at the Icahn School of Medicine at Mount Sinai.
FLAG-EGFP and FLAG-EGFP-Tsg101-CTD (both with the A206K mutation in EGFP that renders it monomeric), as well as full length NiV-C (native sequence) with a N-terminal HA tag, were inserted into the pET-15b vector (Novagen), which appends an additional N-terminal 6XHis tag. The recombinant proteins were purified from BL21(DE3) E. coli as previously described [36] with minor modifications. Cells at OD600 0.5 were induced with 0.5 mM IPTG at 28°C for 4 hours, then collected and lysed in sodium phosphate buffer (pH 8) containing 10 mM imidazole. 6XHis-tagged proteins were purified on 5 mL HisTrap HP (GE Healthcare) columns with increasing imidazole concentrations up to 500 mM, with 0.1% triton X-100 present throughout to maintain protein solubility. Pure fractions were dialyzed overnight into PBS with 5% glycerol and 0.1% triton X-100.
DD-Tsg101 293T cells in the presence or absence of TMP were infected with wild-type NiV at a MOI of 2 as described above. At 24 hours post-infection, cells were fixed with PFGPA solution (2.5% formaldehyde, 0.1% glutaraldehyde in 0.05 M cocadylate buffer pH 7.3, 0.01% picric acid and 0.03% CaCl2) for 24 hours at 4°C. Cells were collected and post-fixed in 1% OsO4 in 0.1 M cocadylate buffer for 1 hour at room temperature. Cells were then en bloc stained with 2% aqueous uranyl acetate (UA) for 20 minutes at 60°C, and dehydrated through a graded series of ethanols. Samples were then processed for infiltration using mixtures of propylene oxide and epoxy (poly/bed 812; Polysciences Inc) and embedded in 100% poly/bed 812. After overnight polymerization at 60°C, ultrathin sections were cut on Leica EM UC7 ultramicrotome, stained with lead citrate and examined in a Philips 201 transmission electron microscope at 60 kV. For immunogold staining, cells were stained en bloc with 2% UA without prior post-fixation, dehydrated in ethanol, and embedded in LR White resin medium grade (Electron Microscopy Sciences). Ultrathin sections were stained with anti-NiV-N antibody (kindly provided by Christopher Broder, Uniformed Services University) followed by secondary goat anti-rabbit IgG conjugated to 15 nm colloidal gold particles (Electron Microscopy Sciences). Sections were fixed with 2% aqueous glutaraldehyde, and stained with 2% UA and lead citrate.
|
10.1371/journal.pntd.0001637 | Dengue Infection and Miscarriage: A Prospective Case Control Study | Dengue is the most prevalent mosquito borne infection worldwide. Vertical transmissions after maternal dengue infection to the fetus and pregnancy losses in relation to dengue illness have been reported. The relationship of dengue to miscarriage is not known.
We aimed to establish the relationship of recent dengue infection and miscarriage. Women who presented with miscarriage (up to 22 weeks gestation) to our hospital were approached to participate in the study. For each case of miscarriage, we recruited 3 controls with viable pregnancies at a similar gestation. A brief questionnaire on recent febrile illness and prior dengue infection was answered. Blood was drawn from participants, processed and the frozen serum was stored. Stored sera were thawed and then tested in batches with dengue specific IgM capture ELISA, dengue non-structural protein 1 (NS1) antigen and dengue specific IgG ELISA tests. Controls remained in the analysis if their pregnancies continued beyond 22 weeks gestation. Tests were run on 116 case and 341 control sera. One case (a misdiagnosed viable early pregnancy) plus 45 controls (39 lost to follow up and six subsequent late miscarriages) were excluded from analysis.
Dengue specific IgM or dengue NS1 antigen (indicating recent dengue infection) was positive in 6/115 (5·2%) cases and 5/296 (1·7%) controls RR 3·1 (95% CI 1·0–10) P = 0·047. Maternal age, gestational age, parity and ethnicity were dissimilar between cases and controls. After adjustments for these factors, recent dengue infection remained significantly more frequently detected in cases than controls (AOR 4·2 95% CI 1·2–14 P = 0·023).
Recent dengue infections were more frequently detected in women presenting with miscarriage than in controls whose pregnancies were viable. After adjustments for confounders, the positive association remained.
| Dengue is the most prevalent mosquito-borne infection with two billion of the world's population at risk and 100 million infections every year. Dengue is increasingly important due to expansion in the vector's range, increased population density in endemic areas from urbanisation, social and environment change. Miscarriage and stillbirth is associated with dengue when the illness is severe. Dengue can also be transmitted directly from the ill mother through the placenta to the fetus in later pregnancy with variable effect to the fetus. However, dengue infection is asymptomatic to mild only in almost 90% of cases and up to 20% of pregnancies miscarry. Little is known if dengue infection in early pregnancy particularly when it is asymptomatic or mild has an effect on miscarriage. Our study explored the relationship between dengue and miscarriage by looking at recent infection rates amongst women who had miscarried and those whose pregnancies were healthy in an area were dengue is common. Our study finds a positive association between recent dengue infection and miscarriage. This finding may be important in explaining some of the miscarriages in areas where dengue is common. It is also relevant to newly pregnant women from non-dengue travelling to dengue endemic areas.
| Each year there is an estimated 100 million dengue infections [1], 500 thousand hospitalisations due to severe dengue illness [2] and 25 thousand dengue deaths [1] worldwide. Dengue's geographical reach is expanding [3] due in part to the complex interplay of climate warming [4] coupled with travel and trade [5] impacting on the observed extension in the range of the responsible mosquito vector [6]. Dengue's intensity in endemic areas has also increased due to the spread of urbanisation [3].
Dengue illness is caused by any one of four serotypes. Infection by one serotype is thought to produce lifelong immunity to that serotype but confers only a few months immunity to the others [7]. Secondary infection by a different serotype increases the risk of severe illness [8]. Typically, dengue infection is asymptomatic or minimally symptomatic in 87% of cases [9]. Hence, obvious dengue illness is the tip of the iceberg in dengue infection.
In Malaysia where dengue is hyperendemic, all four serotype circulate concurrently [10]. In a recent report on a national sample of 1000 Malaysian adults aged 35–74, the dengue seroprevalence rate was 91.6% with a positive age related trend [11].
Severe dengue illness during pregnancy is associated with major adverse outcome of maternal deaths [12], [13], perinatal deaths [12]–[14], preterm births [12], [14] and haemorrhages in labour with much of the data from case reports and small case series [15]. In a large prospective study of recent dengue infection detected at delivery, the pregnancy outcome of 63 parturients who were dengue IgM positive is not different from 2468 IgM negative controls [16]. Vertical transmission to the fetus (particularly later in pregnancy) is established [15]. The vertical transmission rate of maternal dengue infection can vary from at least 1·6% [16] to 6·8% [17] based on prospective studies.
Miscarriages have been reported in association with dengue illness [13], [14], [18], [19] but it is not clear whether the miscarriages were secondary to profound systemic disturbance as a consequence of severe dengue illness or to vertically transmitted dengue infection. Little is known on the effect of dengue infection in early pregnancy with regard to the risk of miscarriage. According to a recent review, pregnancy does not appear to increase incidence or severity of dengue but the literature is very sparse with very few systematic studies of dengue infection on pregnancy available [20].
Dengue specific immunoglobulin M (IgM) can be detected as early as the second day of symptoms and for up to three months [21]. The IgM level typically peaks on the sixth day [22] reaching 100% detection by the eighth day [23]. The dengue IgM test cross-reacts across all four dengue serotypes [24] and is produced even in a secondary infection [25]. The dengue non structural protein 1 (NS1) antigen test is designed to diagnose all serotypes of dengue infection from Day 1 to Day 9 [22]. Dengue IgG antibodies is detected in 100% of cases by Day 15 of primary infection [23] and its presence without other evidence for acute infection is usually taken as distant exposure to any combination of the four dengue serotypes.
We hypothesise that miscarriage is associated with exposure to dengue during early pregnancy due either to maternal dengue illness or vertical transmission. We sought to evaluate the prevalence of recent dengue infection (using dengue IgM and NS1 tests) in women presenting with miscarriage and in similar gestation women with viable pregnancy and to estimate the relative risk of miscarriage in women recently exposed to dengue infection.
Ethics approval for the study was obtained from the University of Malaya Medical Centre Medical Ethics committee (approval no. 715.25 dated 22nd April 2009). The study was conducted in keeping with the Declaration of Helsinki (amended Seoul, Korea 2008) on human study.
We performed a prospective matched case control study on women who presented to our city-based university hospital with miscarriage and control women attending our hospital for pregnancy related care whose pregnancy was viable. Relevant care providers in our hospital were briefed about the study and requested to contact the research team when they diagnosed a miscarriage at up to 22 weeks gestation so that the patient can be approached and consented for study enrolment. All participants provided their written consent.
From 9 June 2009 to 16 March 2011, women who presented acutely to our hospital and diagnosed as a miscarriage were approached. For the purpose of the study, miscarriage is defined as the presence of non-viable product of conception on ultrasound or physical confirmation of expelled products and a positive pregnancy test. In the event of early miscarriage, a positive urine pregnancy test and a history of expelled products of conception is also acceptable if the uterus is empty at presentation. Gestation should not be more than 22 weeks. Gestation is calculated from the reported first day of the last menstrual period and corrected by ultrasound assessment when deemed clinically appropriate to do so. We excluded women with ectopic pregnancy or whose pregnancy was of unknown location. For every case with miscarriage, we recruited three control women with viable pregnancies matched for maternal age (within 3 years) and gestational age (within 3 weeks). The controls were recruited at the earliest opportunity after the index miscarriage case by a researcher (MZS) from amongst women attending for their pregnancy care at our hospital. Miscarriage cases and controls were derived from the same providers and hospital sources.
All participants answered a short questionnaire about whether they had a clinical diagnosis of dengue, made during the current pregnancy or in their lifetime prior to the current pregnancy or a febrile illness, in the last 10 days or at any juncture in their current pregnancy.
Venous blood (5 ml) was collected from each participant and drawn into plain blood bottles. The blood samples were sent directly to the laboratory for immediate processing or kept in a refrigerator at 4°C prior to transfer if the laboratory was closed. The laboratory is based at our university with extensive experience in dengue work.
Samples were discarded if after centrifugation, the supernatant had the appearance of lysed blood. The spun samples were aliquoted and then stored at −70°C. Sera were tested for dengue specific IgM, dengue NS1 antigen and dengue specific IgG. These tests cross react against all four dengue serotypes.
Dengue specific IgM was tested for using an in-house IgM capture-ELISA (enzyme linked immunosorbent assay) test [26]. Samples were considered positive if the ratio of optical density of the positive to negative control was at least two. Panbio Dengue Early ELISA Catalogue No. E-DEN02P (Inverness Medical Innovations Australia Pty Ltd, Queensland, Australia) kits were used as a dengue NS1 antigen capture ELISA test. For dengue IgG detection, we used Panbio Dengue IgG Indirect ELISA catalogue No. E-DEN01G (Inverness Medical Innovations Australia Pty Ltd, Queensland, Australia) kits. The Panbio kits were utilised as per manufacturer's instructions.
We define a recent dengue infection as the detection in sera of dengue specific IgM and/or NS1 antigen.
The miscarriage cases had their chart reviewed up to their discharge from hospital follow up. All included cases must have a confirmed diagnosis of miscarriage at hospital discharge. Cases who did not attend any hospital follow up subsequent to their recruitment into the study or whose latest follow up data were still inconclusive for a diagnosis of miscarriage were contacted by telephone at least three months after their recruitment to confirm the diagnosis of miscarriage with a view to excluding those with a misdiagnosis. Similarly, controls that did not have confirmation of pregnancy viability beyond 22 weeks gestation from chart review because they transferred care elsewhere were contacted by telephone to confirm pregnancy viability beyond 22 weeks gestation.
We had planned to exclude controls that could not be confirmed to have a viable pregnancy beyond 22 weeks as we could not exclude subsequent late miscarriages in them. Our study is designed to evaluate miscarriages up to 22 weeks gestation; it would be inappropriate to have as controls, women who subsequently had miscarriages. We excluded controls who had miscarried after recruitment as we could not exclude a new dengue infection around the time of their miscarriage as no dengue testing was done then.
In a recent study on parturients performed at our hospital, the maternal dengue infection rate as defined by a positive dengue IgM test was 2·5% [16]. Since pregnancy per se has no effect on the acquisition of dengue infection, we expect our control population to have a 2·5% recent dengue infection rate and for the purpose of the sample size calculation we postulate that miscarriage cases will have a 10% recent dengue infection rate. Taking alpha of 0·05, 80% power, 1 to 2 case to control ratio (using PS sample size calculator) [27], 112 cases and 224 controls are required, . As we expect significant attrition in the number of controls due to care transfer to other institutions and hence their drop out from follow up, we increased our planned recruitment ratio to 1 case to 3 controls i.e. at least 112 miscarriage cases and 336 viable pregnant controls.
Data was entered in to SPSS version 15 (SPSS Inc. San Diego CA USA). Comparison (between cases and controls) of the means of continuous variables was by the Student t test, ordinal variables by the Mann Whitney U test and categorical variables by the Chi square test. Fisher exact test is used in place of the Chi Square test for categorical variables if 2 or more cells contain <5 subjects. Multivariable logistic regression analysis, incorporating in the model all characteristics with P<0·05 on bivariate analyses between cases and controls was performed to adjust for these differences in order to establish their independent association with miscarriage. All tests were 2-sided and P<0·05 was taken as a level of significance.
Figure 1 depicted the flow of the recruits within the study and their eventual destination. We recruited and drew blood from 116 cases with a diagnosis of miscarriage and 348 similar gestational aged controls whose pregnancies were viable at recruitment, totalling 464 women for the study cohort. Blood samples from seven controls were found to be lysed after processing and discarded, leaving 457 samples to be tested for dengue IgM, NS1 antigen and IgG. On follow-up, one case of miscarriage was found to be a misdiagnosis and that pregnancy resulted in a term livebirth. Six controls miscarried before 22 weeks gestation and another 39 controls could not be confirmed to have a viable pregnancy beyond 22 weeks gestation despite attempted contact by telephone. These 46 women were excluded from the final analysis.
The characteristics of included cases and controls are shown in Table 1. Prevalence of recent dengue are 5·3% and 1·7%; RR 3·1 (95% 1·0–10) P = 0·047 in cases versus controls. Cases with miscarriages on bivariate analyses were also more likely to be older, of higher gestational age at recruitment and parity and less likely to belong to our major ethnicities of Malay, Chinese or Indian but not significantly more likely to have been ill with a fever during their pregnancy nor more likely to have had distant exposure to dengue (dengue IgG positive status). All the characteristics with P<0·05 on bivariate analysis were included in the model for multivariable logistic regression analysis. After adjustment, recent dengue infection remained an independent risk factor for miscarriage; adjusted odds ratio AOR 4·2 (95% Confidence Interval 1·2–14) P = 0·023.
As a post hoc analysis, if all participants with available dengue test results (n = 457) were analysed, excluding only the seven with lysed samples, the result is not altered: recent dengue infection is still a significant independent risk factor for miscarriage (AOR 3·8 95% CI 1·2–13 P = 0·026).
The 11 participants from both arms of the study group who had recent dengue infection were of similar age, gestational age, parity status, prior miscarriage status, ethnicity and personal awareness of dengue diagnosis in the past but as anticipated, more likely to report a recent febrile illness (as dengue illness is typically febrile) and to be dengue IgG positive (as IgG conversion follows IgM response within a week or two in primary infections) compared to the 400 without evidence of recent dengue – Table 2. These findings support the perception that in our environment with dengue hyperendemicity, dengue is an equal opportunity infection amongst pregnant women.
Four of the five controls with recent dengue infection reported a febrile illness during pregnancy compared to two such reports amongst the six miscarriage cases. All five controls with recent dengue infection and known outcome proceeded to livebirths at term. In the six miscarriage cases, there was one case of concurrent miscarriage and severe dengue illness. That case required intensive care therapy during her hospitalisation but her miscarriage just preceded the worst of the dengue illness. These findings are not helpful in differentiating whether dengue illness or vertical dengue infection contributed more to the abortion process.
The literature on dengue and miscarriage or spontaneous abortion is sparse. We performed a PubMed (http://www.ncbi.nlm.nih.gov/sites/entrez) search applying terms dengue and miscarriage or dengue and abortion without limitations on October 1, 2011 and retrieved only 11 reports. Of these 11 reports, only four publications involved the study of miscarriages in association with dengue, presenting anecdotal data on only five cases of miscarriage during dengue illness [13], [14], [18], [19]. A recent report on women in refugee camps at the Thai-Burmese border investigating antenatal febrile illness has found a dengue infection rate of 9.5%. One woman (of 20; 5%) who had dengue associated fever during pregnancy subsequently miscarried [28].
Recent dengue infection was found in 11/411 (2·6%) of our study participants, a very similar rate to a recent report from our hospital of 2·5% (63/2531) dengue IgM rate in unselected parturients at time of their delivery [16]; (Chi Square test, P = 0·82). This shows that as a whole, the participants of our current study closely reflect unselected parturients at our hospital with regard to the risk of recent dengue infection and supports the assertion that the selection of our study group is unbiased. The similar dengue IgG positive and reported recent febrile illness rates amongst cases and controls in our current study further bolster the above assertion.
Our dengue IgG positive rate of 213/411 (52%) indicating prior dengue exposure in participants whose mean age was 30 years old is similar to the 63% dengue IgG positive rate from a cross sectional community-based prevalence study on subjects aged 21 to 40 years [29] who were derived from the same geographical catchment area as our participants. This similarity lends support to our study group as a whole being a representative sample of the host population.
It is noteworthy that 6/11 (54·5%) of the women with recent dengue in our sample reported a febrile illness in the course of the pregnancy leading up to their recruitment at a mean gestation of only10·8 weeks whereas in an earlier study at our hospital of 63 dengue IgM positive parturients (delivered at a mean gestation of 39·4 week), only 7/63 (11·1%) reported a febrile illness over the entire course of pregnancy [16]; (Chi Square test P<0·001). Dengue is reported to be asymptomatic or minimally symptomatic in 87% of infections [9], an illness rate similar to that reported by the parturients but far lower than in our current study group. Mild exposures to hyperthermia or fever during the preimplantation period and more severe exposures during embryonic and fetal development often result in prenatal death and abortion [30]. The possible interpretation is that dengue infection in very early pregnancy may give rise to a more symptomatic (or severe) presentation. This more severe effect may contribute to its role in miscarriage whereas dengue infection late in pregnancy is far less likely to be severe and thus did not adversely affect on delivery outcome [16]. However, with only 11 cases of recent dengue found in this study, further corroboration of this hypothesis is required.
Although dengue illness can cause fatigue at two months [31] and other persistent symptoms at two years [32] beyond the acute phase, all five cases of recent dengue infection in our control group resulted in term livebirths despite the fact that four of these five had reported a febrile illness during early pregnancy. The interpretation might be that dengue infection very early on during pregnancy may have a severe impact sufficient to contribute to early miscarriage but does not have a prolonged adverse effect deeper into pregnancy.
There are limitations in our study design. As dengue IgM may persist for up to six months [1] after infection though rarely, our methodology was not capable of identifying only those dengue infections that were within pregnancy – some those identified with recent dengue infection might have been infected prior to pregnancy. Specific IgM might not be detectable in a small proportion of secondary infections [1], [25] and given the dengue IgG detection rate of just over 50% in our participants, probably half of the recent infection we detected were secondary infections. However, these limitations apply to both cases and controls, so should not affect the assessment of relative risk and if anything, would be biased against the finding of a relationship between recent dengue and miscarriage. Our IgM capture ELISA assay for dengue cross reacts with IgM against other flaviviruses notably Japanese encephalitis and Yellow fever viruses but IgM capture ELISA for dengue has demonstrated only a low cross reactivity with Japanese Encephalitis virus [33]. Both yellow fever and viral encephalitis were very rare in Malaysia with a reported rate in 2010 of 0 per 100000 and 0·2 per 100000 population respectively compared to dengue illness rate of 148·7 per 100000 population [34]. With only 11 recent dengue infections identified from a sample of 411 (2·6%), the recent dengue infection rate in our study is somewhat lower than that assumed in our sample size calculation of 2·5% recent dengue infection rate in controls and an assumed 10% in miscarriage cases. As a consequence, although analysis shows a significant association between dengue infection and miscarriage at the 5% level of significance, the relatively small number of infections means that the 95% confidence interval is wide.
In summary, in the context of our setting with dengue hyperendemic, 5·2% of the women with miscarriages displayed evidence of recent dengue infection compared to 1·7% in controls with viable pregnancies (AOR 4·2 95% CI 1·2–14: P = 0·023), a statistically significant and clinically important finding. This interaction between dengue infection in early pregnancy and miscarriage is a highly relevant public health issue due to high number of dengue infections worldwide annually and the typically higher fertility rates in the dengue endemic regions of the world. There is an implication also for travellers from dengue free regions in the early stages of their pregnancies who are journeying to dengue endemic areas. Further study to replicate our data and finding is urgently needed.
|
10.1371/journal.pntd.0002887 | Incidence of Human Taenia solium Larval Infections in an Ecuadorian Endemic Area: Implications for Disease Burden Assessment and Control | Human cysticercosis is a zoonotic disease causing severe health disorders and even death. While prevalence data become available worldwide, incidence rate and cumulative incidence figures are lacking, which limits the understanding of the Taenia solium epidemiology.
A seroepidemiological cohort study was conducted in a south-Ecuadorian community to estimate the incidence rate of infection with and the incidence rate of exposure to T. solium based on antigen and antibody detections, respectively. The incidence rate of infection was 333.6 per 100,000 person-years (95% CI: [8.4–1,858] per 100,000 person-years) contrasting with a higher incidence rate of exposure 13,370 per 100,000 person-years (95% CI: [8,730–19,591] per 100,000 person-years). The proportion of infected individuals remained low and stable during the whole study year while more than 25% of the population showed at least one antibody seroconversion/seroreversion during the same time period.
Understanding the transmission of T. solium is essential to develop ad hoc cost-effective prevention and control programs. The estimates generated here may now be incorporated in epidemiological models to simulate the temporal transmission of the parasite and the effects of control interventions on its life cycle. These estimates are also of high importance to assess the disease burden since incidence data are needed to make regional and global projections of morbidity and mortality related to cysticercosis.
| Human cysticercosis is a neglected zoonotic parasitic disease causing severe health disorders such as epilepsy and even death. Cysticercosis is related to poverty, inadequate hygiene conditions and traditional pig farming. The present study describes the dynamic nature of human Taenia solium larval infections in an Ecuadorian endemic community. In this study we report for the first time incidence rate and cumulative incidence figures of human T. solium larval infections in Latin America. The simultaneous use of antibody and antigen serological detections allowed estimating both parasite exposure and infection rates, respectively. While about 13% of the inhabitants were exposed to T. solium eggs, less than 1% of the population became yearly infected with the parasite. This contrast between exposure and infection may be linked to an effective resistance to the parasite acquired through long-term exposure of the population and differs from the African situation, where much higher levels of infection have been observed. These estimates are of high importance to understand the epidemiology of T. solium in order to develop ad hoc cost-effective prevention and control programs. They are also essential to assess the burden of T. solium cysticercosis since longitudinal data are needed to make regional and global projections of morbidity and mortality related to cysticercosis.
| Human cysticercosis (CC) is a parasitic disease caused by the development of the metacestode larval stage of Taenia solium (cysticercus) in the muscles, the central nervous system (causing neurocysticercosis (NCC)), the subcutaneous tissue and the eyes (causing subcutaneous and ocular cysticercosis, respectively) [1]. The life cycle of the parasite includes humans as sole definitive hosts and pigs as main intermediate hosts. Humans get infected by consumption of raw or undercooked pork infected with cysticerci, resulting in the development of an adult intestinal tapeworm (taeniosis). Pigs become infected by ingestion of T. solium eggs contained in infected human feces, through coprophagic behavior or via ingestion of contaminated water or food, and develop porcine CC. Man can also act as a dead-end intermediate host by accidental ingestion of T. solium eggs [2] and develop human CC. NCC may cause severe neurological disorders and even death [3], [4]. It is the most important parasitic disease of the central nervous system and the main cause of acquired epilepsy in T. solium endemic areas, where NCC is associated with 14.2 to 50% of the epilepsy cases [5], [6]. The maintenance of the parasite life cycle is associated with poor sanitation, lack of hygiene and traditional pig rearing systems allowing free roaming of the animals. Endemic areas have been identified in Asia, Africa and Latin America [7]–[10]. In Latin America the infection has been reported in at least 18 countries and is considered a major public health problem, especially in poor rural areas [7], [8]. The Andean region of Ecuador and neighboring countries is hyper-endemic for cysticercosis [11]. While reliable prevalence data become available worldwide, they may considerably vary depending on the diagnostic test used [12]–[14]. Several tools are available for the diagnosis of human CC, i.e. imaging and serological techniques. Serological antigen and antibody detections are valuable tools when conducting epidemiological studies, since they inform on infection with and exposure to the parasite, respectively. Taking the latter distinction into account, studies conducted in Ecuadorian endemic rural communities have shown an exposure to the parasite ranging from 25 to 40% and a proportion of infected individuals ranging from 2.25 to 4.99% [15]–[17]. However, prevalence figures do not inform on the evolution of the number of positive cases over time and estimates for human cysticercosis incidence rate and cumulative incidence are lacking, which limits the understanding of the transmission dynamics of T. solium and does not allow a precise estimation of its disease burden. García et al. (2001) [18] conducted longitudinal studies in endemic areas of Peru and Colombia and demonstrated the presence of transient antibody responses suggesting a high number of antibody seroconverted cases per year ranging from 8 to 25% of the population depending on the studied area. Through rule-based modeling, Praet et al. (2010) [16] simulated the annual antibody seroconversion rate in an endemic area of Ecuador. They estimated an annual incidence rate of exposure of people becoming seropositive of 14 per 100 person-years. On the other hand, studies estimating both incidence rate of infection and cumulative incidence are scarce [16], [18], [19]. Mwape et al. (2013) [20] reported an incidence rate of infection of 6,300 per 100,000 person-years in a rural community of eastern Zambia. Such estimates for Latin America are inexistent. For this reason, the present study aims at estimating the cumulative incidence and the incidence rate of human CC in an endemic area of Ecuador. A sero-epidemiological cohort study was conducted to investigate the transmission dynamics of T. solium among individuals living in a southern Ecuadorian rural community. This paper reports estimates of the incidence rates, cumulative incidences of active infection and exposure rates to T. solium and discusses the implications for the disease burden assessment and control.
The protocol used in this study was approved by the Ethical Committee of the Central University of Ecuador (IRB 00002438) and by the Ethical Committee of The University Hospital of Antwerp, Belgium. Written informed consent was obtained from each individual willing to participate in the study. For participants aged less than 18 years old written informed consent was also obtained from a parent or a legal adult representative. Individuals testing positive for T. solium cysticercosis antigens were referred to the local health center for follow-up.
The study was conducted in the rural parish of Sabanilla (4° 12′S, 80° 8′W) belonging to the Celica canton in the Southern Ecuadorian province of Loja. The parish has 1145 inhabitants; most of them are farmers involved in activities related to agriculture and animal husbandry. The climate is semi-arid, and the altitude is 700 meters above sea level. The region is endemic for T. solium cysticercosis and presents the risk factors for the transmission of the parasite [15], [21]. A sero-epidemiological community-based cohort study was performed. Three blood sampling rounds were organized in Sabanilla in a period of 13 months: the first sampling round took place in June 2009 (SR1), the second in November 2009 (SR2) and the third one in July 2010 (SR3). Based on the three sampling rounds, three periods of time were defined as follows: a six-month period from June 2009 to November 2009 (P1), a seven-month period from November 2009 to July 2010 (P2), and a total 13-month period from June 2009 to July 2010 (P3).
First, an informative meeting inviting the population to participate took place at the beginning of the study in collaboration with the local authorities. Then, a census of the population was conducted based on a door-to-door survey, including collection of information on age and sex of the inhabitants. After informed consent, all individuals older than one year willing to participate and present at the time were blood sampled.
At each sampling round, 10 ml of blood was collected in dry tubes. After coagulation and centrifugation, serum was collected and stored at −20°C until analysis. Two serological diagnostic tests were performed. (1) The Enzyme Linked Immunosorbent Assay for the detection of circulating antigens of the metacestode of T. solium (Ag-ELISA) [22]–[24]. The sensitivity and specificity of the Ag-ELISA for detecting active infection in humans are 90% (95% CI: [80–99%]) and 98% (95% CI: [97–99%]), respectively. No cross-reaction with other parasites has been reported [13], [23]. (2) The Enzyme-Linked Immunoelectrotransfer Blot (EITB) for the detection of antibodies directed against seven specific T. solium metacestode glycoproteins [25]. The sensitivity and specificity of the EITB for detecting exposure to the parasite range from 97% to 98% and from 97% to 100%, respectively [13], [25].
The antigen and antibody seroprevalence (Ag and Ab seroprevalence), as based on the results of the Ag-ELISA and of the EITB, respectively, were calculated for each sampling round for the whole population and by sex.
A multinomial Bayesian model adapted from Berkvens et al. (2006) [26] was used to estimate the true prevalence of T. solium larval infections for each sampling round based on the antigen seroprevalence data and on prior information on the test characteristics (sensitivity and specificity of the Ag-ELISA). Prior information was extracted from the available literature [13]. A uniform distribution with lower and upper limits of 0.80 and 1.00, and 0.97 and 1.00 were used to constrain the sensitivity and the specificity of the test, respectively. The analysis was conducted in WinBUGS and R [27], [28]. Three chains, 20,000 iterations, following a burn-in of 5,000 were used to assess the convergence of the results. Criteria assessing the fit between prior information and the seroprevalence data were evaluated, i.e. the Bayesian p-value (Bayesp), the Deviance Information Criterion (DIC) and the number of parameter effectively estimated by the model (pD) [26], [28].
First, proportion of change to antigen seropositivity/seronegativity (change to Ag seropositivity/seronegativity) and proportion of antibody seroconversion and seroreversion (Ab seroconversion and seroreversion) were calculated to characterize the transmission dynamics of the disease.
Seroconversion is defined as the change from a negative to a positive serological test result between 2 sampling rounds; the opposite is defined as seroreversion [29]. The proportion of Ab seroconversion and the proportion of change to Ag seropositivity reflect the cumulative incidence for a defined time period. They were calculated by dividing the number of new cases by the number of susceptible individuals (having a negative test result at the previous sampling round) during a given time. The proportion of Ab seroreversion and the proportion of change to Ag seronegativity were calculated by dividing the number of positive tests that turned negative by the number of positive tests at the previous sampling round.
The incidence rate of infection with the larval stage of T. solium and the incidence rate of exposure to T. solium eggs were also calculated based on the results of the antigen and antibody detection tests, respectively.
The incidence rate was calculated as the number of new (change from seronegativity to seropositivity) cases in a defined time period divided by the number of person-time units at risk during the time-period. Yearly incidence rates were multiplied by 100,000 to be expressed by 100,000 person-years [30], [31]. The person-time unit represents one person for a defined period of time. The latter was calculated as described in Ngowi et al. (2008) assuming that the infection occurs uniformly over time and considering halfway the period between two sampling rounds [32]. For example, if a person is followed up for six months and does not seroconvert during this time, this person will contribute 0.5 person-year to the person-time at risk. If a person that is followed up for the same period but seroconverts during that period, this person will contribute 0.25 person-year to the person-time at risk. Yearly incidence rates were calculated based on this calculation method. Ninety-five % exact Poisson confidence intervals were calculated using the epitools package in R for all incidence rates [33].
Data were entered in Excel 2010 (Microsoft Office 2010). Statistical analyses were performed in Stata (Stata Corp., College Station, TX) and in R: [34]
Fisher exact test was used to compare (1) Ag/Ab seroprevalence between sex within each sampling round and (2) Ag seroprevalence with Ab seroprevalence within each sampling rounds. Also, McNemar test was performed to compare the sero-Ag and sero-Ab prevalence between rounds. Multivariate logistic regression analysis was used to study the association between sero-Ag/Ab prevalence and age and sex, and this for the three samplings rounds. The significance level was set at 0.05.Fisher exact test was used to compare (1) the proportion of Ab seroconversion with the proportion of antibody seroreversion and the proportion of change to Ag seropositivity with the proportion of change to Ag seronegativity within periods, (2) the proportions of Ab seroconversion/seroreversion and the proportions of Ag change to seropositivity/seronegativity between sexes, (3) the proportions of Ab seroconversion/seroreversion and the proportions of change to Ag seropositivity/seronegativity between periods.
In addition, a change point analysis was used to compare the proportion of Ab seroconversion with the proportion of Ab seroreversion in function of age. The change point analysis classifies the population into 2 age groups at different age points (10, 20, 30, 40, 50, 60, 70, 80 years old). The Fisher exact test was then used on both age groups in order to identify any change of significance when comparing the proportions of Ab seroconversion and seroreversion [16], [20], [35]. The significance level was set at 0.05 for all statistical analyses.
A total of 967 (84.45%) individuals from the 1145 inhabitants listed in the census participated in the study. Depending on the willingness of the individuals to participate, their presence at the time of sampling and on the quantity of serum available, EITB was performed on 743 (64.9%), 538 (47%) and 518 (45.2%) sera for June, November and July, respectively. Ag-ELISA was performed on 744 (65%), 538 (47%) and 514 (44.9%) sera for the same time periods. Figure 1 describes in detail sera availability and individual participation during the three sampling rounds.
The Ag and Ab seroprevalence for each sampling round for the whole population and by sex are presented in Table 1. The prevalence adjusted for misclassification error of T. solium larval infections was 0.7% (95% Credibility Interval (CI): [0.03–1.75]), 0.7% (95% CI: [0.03–2.00]) and 1.1% (95% CI: [0.05–2.84]) for SR1, SR2 and SR3, respectively. All except one Ag positive individuals were also Ab positive in the 3 sampling rounds. Fisher exact test did not reveal any significant difference of Ag and Ab seroprevalence between sexes. McNemar test did not reveal any significant difference of Ag and Ab seroprevalence between rounds. Ab seroprevalence was significantly higher than Ag seroprevalence within each sampling round. Multivariate logistic regression analysis showed a significant positive correlation between Ab seroprevalence and age.
Table 2 shows the proportions of antigen and antibody seropositive and/or seronegative individuals who participated in all 3 sampling rounds and whose sera were available for both tests (n = 277). Only one individual changed to antigen seropositivity status throughout the entire study period. Eighteen percent of this restricted population remained antibody positive throughout the entire study period while about 20% of the individuals showed at least 1 change of antibody positivity status.
The overall incidence rate of human T. solium larval infection based on -antigen detection was 333.6 per 100,000 person-years (95% exact Poisson CI: [8.4–1,858] per 100,000 person-years). The overall incidence rate of exposure to T. solium based on antibody detection was 13,370 per 100,000 person-years (95% exact Poisson CI: [8,730–19,591] per 100,000 person-years). Ag proportion of changes to seropositivity/seronegativity and proportion of Ab seroconversion/seroreversion by period are represented in Table 3. Incidence rates estimates for individuals who participated in at least two of the sampling rounds are given in Table 4.
Fisher exact test did not show any difference of proportion of Ab seroconversion/seroreversion and proportions of change to Ag seropositivity/seronegativity between sexes, nor between periods. Proportion of Ab seroreversion was significantly higher than proportion of Ab seroconversion for each period (Figure 2). The change point analysis showed that the proportion of Ab seroreversion was significantly higher than proportion of Ab seroconversion until the age of 30 years. After this change point, the difference was not significant (Figure 3).
This is the first study reporting cumulative incidence and incidence rate figures of human T. solium larval infection in Latin America. The overall incidence rate of infection in the endemic rural community of Sabanilla, was 333.6 per 100,000 person-years (95% exact Poisson CI: [8.4–1,858] per 100,000 person-years), which suggests that less than 1% of the population becomes infected yearly with the parasite. In contrast, the incidence rate of exposure to T. solium is much higher: about 14% of the population has a yearly contact with the parasite. The latter estimates are in line with observed and simulated antibody seroconversion rates ranging from 8 to 25% in Peru, Colombia and Ecuador [16], [18]. Proportions of change to Ag seropositivity/seronegativity and Ab seroconversion and seroreversion were identical in males and females indicating that both genders get equally infected with/are equally exposed to the parasite. Moreover, these proportions did not significantly vary in time (one year period). On the other hand, proportion of Ab seroreversion was significantly higher than proportion of seroconversion Ab for each period and a change point analysis showed that proportion of Ab seroreversion was significantly higher than proportion of Ab seroconversion until the age of 30 years. After this change point, the difference was not significant. These results corroborate the findings of Praet et al. (2010) [16] suggesting a higher proportion of seroreversion before the age of 40 years due to a higher number of primary immune responses before this age. In other words, individuals will serorevert more rapidly before the age of 30–40 years because primary humoral response is shorter and weaker than secondary response. Thus, the proportion of seroreversions depends on the immunological status of the individuals.
The dynamics of infection and exposure in the population, represented by the proportions of antigen and antibody results from the individuals who participated in all 3 sampling rounds, showed that the proportion of infected individuals remains low and stable during the whole study year, while the proportion of exposed individuals is remarkably higher. Of note is the high level of serological status variation with more than 20% of the population showing at least one antibody seroconversion/seroreversion during the year. Together with the prevalence estimates presented by period, these longitudinal data corroborates the findings of other studies conducted in Latin America highlighting a high prevalence of exposure to the parasite but a low prevalence of active infections. This contrast between exposure and infection may be linked to an effective resistance to the parasite acquired through long-term exposure of the population. In addition, these results confirm the occurrence of transient antibody responses in individuals living in T. solium endemic areas and suggest exposure to the parasite without infection or mild infections that are aborted by the natural immunity of the individual [20]. Mwape et al. (2013) [20] conducted a similar community-based longitudinal study in the Eastern Province of Zambia. While a much higher incidence rate was observed in the African endemic area, similar higher proportion of Ab seroreversion than Ab seroconversion and the presence of transient antibody responses were described. Further studies are needed to unravel the difference of parasite transmission patterns in different epidemiological settings. Specifically, research should focus on identifying the causes for differences in infection levels. In this context, the identification of tapeworm carriers in a community should be based on improved methods, because sensitivity and specificity of conventional coprological methods are low [36].
Our study has limitations that are mainly due to the inhabitant proportion of participation. Although 967 individuals from a total of 1045 inhabitants provided at least one blood sample, 396 provided only one sample and another 283 provided two blood samples. Compliance in participating in all 3 sampling rounds was of 288 volunteers despite extensive information sessions prior to the sampling procedure. The main reason for irregular participation was the absence of the individuals for professional duties. Reduced participation can have an impact on the precision of the estimation of the incidence rate, however, the participation on the three sampling rounds is still representative of the total population and all the participants selected for the incidence rate estimation match all the selection criteria for this calculation (n = 288 (27.6% (95% CI: [24.9–30.4]))). Another limitation of the study is the limited number of samplings and the relatively long sampling intervals depending on logistic, economic and ethical constraints. In other words, much more information could have been produced if more sampling had been organized at shorter time intervals, i.e. the positivity status of the participants would have been more accurately monitored over time. The incidence rate estimation for both infection and exposure is likely to be lower with increasing intervals between samplings: a proportion of the new infections may be undetected and the time of occurrence of a new infection overestimated. A quicker detection of new infections will result in a decrease of the number of person-years at risk and consequently in higher estimates of the incidence rate. Finally, even though the tests used in this study have shown high sensitivity and specificity, false positive and negative individuals may bias the prevalence and the incidence rate estimates. Bayesian estimation of infection with T. solium larva prevalence has been used to estimate the true prevalence of infection with an exposure to T. solium. The Bayesian estimation corrects the apparent prevalence at, but does not allow to know the true infection status at the individual level. Consequently, it does not allow to estimating the true incidence rate.
In conclusion, the present study underlines the importance of conducting longitudinal serological follow-up allowing generating incidence rather than prevalence data to fully understand the transmission dynamics of the infection and to avoid under/overestimation of the occurrence of the parasite. Similar cohort studies assessing the effect of risk factors such as development of immunity and behavioral factors should be conducted to identify all the parameters that may influence parasite transmission. Understanding the transmission dynamics of T. solium is essential to develop ad hoc cost-effective prevention and control programs. The estimates generated here may now be incorporated in epidemiological models to simulate the temporal transmission of the parasite and the effects of control interventions on its life cycle [19]. These estimates are also of high importance to assess the burden of T. solium cysticercosis since incidence data are needed to make regional and global projections of morbidity and mortality related to cysticercosis. To this end, the link between the incidence rate of infection and health outcomes related to human cysticercosis, such as epilepsy and chronic headache, as well as the case-fatality ratio still need to be estimated.
|
10.1371/journal.pgen.1007335 | Linkage mapping of yeast cross protection connects gene expression variation to a higher-order organismal trait | Gene expression variation is extensive in nature, and is hypothesized to play a major role in shaping phenotypic diversity. However, connecting differences in gene expression across individuals to higher-order organismal traits is not trivial. In many cases, gene expression variation may be evolutionarily neutral, and in other cases expression variation may only affect phenotype under specific conditions. To understand connections between gene expression variation and stress defense phenotypes, we have been leveraging extensive natural variation in the gene expression response to acute ethanol in laboratory and wild Saccharomyces cerevisiae strains. Previous work found that the genetic architecture underlying these expression differences included dozens of “hotspot” loci that affected many transcripts in trans. In the present study, we provide new evidence that one of these expression QTL hotspot loci affects natural variation in one particular stress defense phenotype—ethanol-induced cross protection against severe doses of H2O2. A major causative polymorphism is in the heme-activated transcription factor Hap1p, which we show directly impacts cross protection, but not the basal H2O2 resistance of unstressed cells. This provides further support that distinct cellular mechanisms underlie basal and acquired stress resistance. We also show that Hap1p-dependent cross protection relies on novel regulation of cytosolic catalase T (Ctt1p) during ethanol stress in a wild oak strain. Because ethanol accumulation precedes aerobic respiration and accompanying reactive oxygen species formation, wild strains with the ability to anticipate impending oxidative stress would likely be at an advantage. This study highlights how strategically chosen traits that better correlate with gene expression changes can improve our power to identify novel connections between gene expression variation and higher-order organismal phenotypes.
| A major goal in genetics is to understand how individuals with different genetic makeups respond to their environment. Understanding these “gene-environment interactions” is important for the development of personalized medicine. For example, gene-environment interactions can explain why some people are more sensitive to certain drugs or are more likely to get certain cancers. While the underlying causes of gene-environment interactions are unclear, one possibility is that differences in gene expression across individuals are responsible. In this study, we examined that possibility using baker’s yeast as a model. We were interested in a phenomenon called acquired stress resistance, where cells exposed to a mild dose of one stress can become resistant to an otherwise lethal dose of severe stress. This response is observed in diverse organisms ranging from bacteria to humans, though the specific mechanisms governing acquisition of higher stress resistance are poorly understood. To understand the differences between yeast strains with and without the ability to acquire further stress resistance, we employed genetic mapping. We found that part of the variation in acquired stress resistance was due to sequence differences in a key regulatory protein, thus providing new insight into how different individuals respond to acute environmental change.
| A fundamental question in genetics is how individuals with extremely similar genetic makeups can have dramatically different characteristics. One hypothesis is that a small number of regulatory polymorphisms can have large effects on gene expression, leading to the extensive phenotypic variation we see across individuals. In fact, gene expression variation is hypothesized to underlie the extensive phenotypic differences we see between humans and chimpanzees despite >98% DNA sequence identity [1, 2]. This hypothesis is supported by numerous examples of gene expression variation affecting higher-order organismal traits.
For example, human genome-wide association studies (GWAS) have found that a substantial fraction of disease-associated variants are concentrated in non-coding regulatory DNA regions [3–8]. Further examples include gene expression variation being linked to differences in metabolism [9–11], physiology [12–16], morphology [17–23], and behavior [24–27].
While gene expression variation is pervasive, there is often a lack of obvious phenotypic change associated with differentially expressed genes. This can occur for a variety of reasons. First, a large fraction of expression variation has been postulated to be evolutionarily neutral with no effect on organismal fitness [28–30]. Second, co-regulation of genes that share the same upstream signaling network and transcription factors can lead to genes whose expression differences correlate with phenotype but are not truly causative. Finally, some gene expression differences may truly affect phenotype, but only under specific conditions. For example, the predictive power of expression quantitative trait loci (eQTL) mapping studies on higher-order phenotypes can be poor unless multiple environments are considered [31]. Similarly, tissue-restricted eQTLs are more likely to map to known disease-associated loci identified from GWAS than non-tissue-restricted eQTLs [32, 33].
Thus, a major challenge for connecting gene expression variation to downstream effects on higher-order traits is the choice of which conditions and traits to examine. To this end, we have been leveraging natural variation in the model eukaryote Saccharomyces cerevisiae, and a phenotype called acquired stress resistance. Many studies have shown a poor correlation between genes that respond to stress and their importance for surviving stress [34–43]. Thus, we and others have argued that the role of stress-activated gene expression is not to survive the initial insult, but instead protects cells from impending severe stress through a phenomenon called acquired stress resistance [44, 45]. Acquired stress resistance (sometimes referred to as “induced tolerance” or the “adaptive response”) occurs when cells pretreated with a mild dose of stress gain the ability to survive an otherwise lethal dose of severe stress. Notably, acquired stress resistance can occur when the mild and severe stresses are the same (same-stress protection) or across pairs of different stresses (cross protection). This phenomenon has been observed in diverse organisms ranging from bacteria to higher eukaryotes including humans [44–50]. The specific mechanisms governing acquisition of higher stress resistance are poorly understood, but there are wide reaching implications. In humans, ischemic preconditioning (transient ischemia followed by reperfusion—i.e. mild stress pretreatment followed by severe stress) may improve outcomes of cardiovascular surgery [51–54], while transient ischemic attacks (“mini-strokes”) may protect the brain during massive ischemic stroke [55–57]. Thus, understanding the genetic basis of acquired stress resistance in model organisms holds promise for mitigating the effects of stress in humans.
A previous study found that a commonly used S288c lab strain is unable to acquire further ethanol resistance when pretreated with a mild dose of ethanol [44]. We found this phenotype to be surprising, considering the unique role ethanol plays in the life history of Saccharomyces yeast, where the evolution of aerobic fermentation gave yeast an advantage over ethanol-sensitive competitors [58]. Because ethanol is a self-imposed stress that induces a robust stress response [59–63], we expected that ethanol should provoke acquired stress resistance in wild yeast strains. Indeed, this turned out to be the case, with the majority of tested wild strains acquiring resistance to severe ethanol following a mild ethanol treatment [45]. Furthermore, this phenotype correlated with extensive differences in the transcriptional response to acute ethanol stress in the lab strain when compared to a wild vineyard (M22) and wild oak (YPS163) strain (>28% of S288c genes were differentially expressed at an FDR of 0.01) [45, 64]. We performed linkage mapping of S288c crossed to a wild vineyard strain (M22) and wild oak strain (YPS163), and observed numerous “hotspots” where the same eQTL loci affect the expression of a large number of transcripts (anywhere from 10–500 transcripts per hotspot) [64].
In the present study, we provide new evidence that one of these eQTL hotspot loci affects natural variation in acquired stress resistance, namely the ability of ethanol to cross protect against oxidative stress in the form of hydrogen peroxide. The causative polymorphism is in the heme-activated transcription factor Hap1p, which we show directly impacts cross protection, but not the basal resistance of unstressed cells. Finally, we show that the Hap1p effect is mediated through novel regulation of cytosolic catalase T (Ctt1p) during ethanol stress in wild strains. This study highlights how strategically chosen traits that are better correlated with gene expression changes can improve our power to identify novel connections between gene expression variation and higher-order organismal phenotypes.
We previously found that an S288c-derived lab strain was unable to acquire further ethanol resistance when pretreated with a mild dose of ethanol, in contrast to the vast majority of ~50 diverse yeast strains [45]. In addition to the S288c strain’s acquired ethanol resistance defect, ethanol also failed to cross protect against other subsequent stresses [44, 65]. In nature, wild yeast cells ferment sugars to ethanol, and then shift to a respiratory metabolism that generates endogenous reactive oxygen species [66–68]. Thus, we hypothesized that ethanol might cross protect against oxidative stress in wild yeast strains. We tested this hypothesis by assessing whether mild ethanol treatment would protect a wild oak strain (YPS163) from severe oxidative stress in the form of hydrogen peroxide (H2O2). Cross protection assays were performed by exposing cells to a mild, sublethal dose of ethanol (5% v/v) for 60 min, followed by exposure to a panel of 11 increasingly severe doses of H2O2 (see Materials and Methods). Confirming the observations of Berry and Gasch [44], ethanol failed to cross protect against H2O2 in S288c, and in fact slightly exacerbated H2O2 toxicity (Fig 1). In contrast, ethanol strongly cross protected against H2O2 in YPS163 (Fig 1).
The inability of ethanol to induce acquired stress resistance in S288c correlates with thousands of differences in ethanol-dependent gene expression in comparison to wild strains that can acquire ethanol resistance [45, 64]. In light of this observation, and the known dependency of cross protection on stress-activated gene expression changes [44], we hypothesized that differences in cross protection against H2O2 by ethanol may be linked to differential gene expression. To test this, we performed quantitative trait loci (QTL) mapping using the same mapping population as our original eQTL study that mapped the genetic architecture of ethanol-responsive gene expression [64]. Specifically, we conducted QTL mapping of both basal and acquired H2O2 resistance in 43 F2 progeny of S288c crossed with YPS163 (see Materials and Methods). While we found no significant QTLs for basal H2O2 resistance, we did find a significant QTL peak on chromosome XII that explained 38% of the variation in cross protection (Fig 2). It is unlikely that our failure to detect a chromosome XII QTL for basal H2O2 resistance was due to a lack of statistical power, because two independent basal H2O2 resistance QTL studies using millions of S288c x YPS163 F2 segregants also found no significant associations at this locus [69, 70]. Additionally, we estimated the heritability of phenotypic variation in basal resistance to be 0.79, which is slightly above the median value estimated by Bloom and colleagues for 46 yeast traits [71], and is only moderately lower than the heritability for cross protection (0.92). Lastly, the shape of the distribution of phenotypes in the F2 were markedly different between basal and acquired H2O2 resistance, with basal resistance showing a transgressive segregation pattern and acquired resistance showing a continuous distribution (S1 Fig). Altogether, these results suggest that the genetic basis of natural variation in acquired stress resistance is distinct from the basal resistance of unstressed cells (see Discussion).
The significant QTL for cross protection was located near a known polymorphism in HAP1, a heme-dependent transcription factor that controls genes involved in aerobic respiration [72–74], sterol biosynthesis [75–77], and interestingly, oxidative stress [77, 78]. S288c harbors a known defect in HAP1, where a Ty1 transposon insertion in the 3’ end of the gene’s coding region has been shown to reduce its function [79]. In fact, we previously hypothesized that the defective HAP1 allele was responsible for the inability of S288c to acquire further resistance to ethanol. However, a YPS163 hap1Δ strain was still fully able to acquire ethanol resistance, despite notable differences in the gene expression response to ethanol in the mutant [45]. Likewise, despite previous studies implicating Hap1p as a regulator of oxidative stress defense genes [77, 78], HAP1 is apparently dispensable for same-stress acquired H2O2 resistance [47]. These observations suggest that the molecular mechanisms underlying various acquired stress resistance phenotypes can differ, even when the identity of the secondary stress is the same.
Because we previously implicated HAP1 as a major ethanol-responsive eQTL hotspot affecting over 100 genes, we hypothesized that ethanol-induced cross protection against H2O2 may depend upon Hap1p-regulated genes. However, it was formally possible that HAP1 was merely linked to the truly causal polymorphism. To distinguish between these possibilities, we generated deletion mutations in the YPS163 background for every non-essential gene within the 1.5-LOD support interval of the QTL peak (encompassing IFH1 –YCS4). Of the 36 mutants tested, two showed significantly and highly diminished acquired H2O2 resistance (Fig 3 and S2 Fig), hap1Δ and top3Δ (encoding DNA topoisomerase III). To determine whether different alleles of HAP1 and/or TOP3 were responsible for natural variation in acquired H2O2 resistance, we applied an approach called reciprocal hemizygosity analysis [80], where the TOP3 and HAP1 alleles were analyzed in an otherwise isogenic S288c-YPS163 hybrid background (see Fig 4A for a schematic). In each of the two reciprocal strains, one allele of the candidate gene was deleted, producing a hybrid strain containing either the S288c or YPS163 allele in single copy (i.e. hemizygous for TOP3 or HAP1). While we found only mild allelic effects for TOP3, the effects of different HAP1 alleles were striking (Fig 4B and 4C). The hybrid strain containing the HAP1YPS163 allele showed full cross protection, while the strain containing the HAP1S288c allele showed none. Thus, we examined the effects of HAP1 on acquired H2O2 resistance further. Intriguingly, we found that the YPS163 hap1Δ mutant was unaffected for acquired H2O2 resistance when mild H2O2 or mild NaCl were used as mild stress pretreatments (Fig 5), suggesting that Hap1p plays a distinct role in ethanol-induced cross protection (see Discussion).
Finally, we performed allele swap experiments to examine the effects of the different HAP1 alleles in the original parental backgrounds. We introduced only the Ty element from HAP1S288c into the YPS163 HAP1 gene, and observed a loss of acquired H2O2 resistance similar to the YPS163 hap1Δ strain (Fig 6). We next tested whether repair of the defective hap1 allele in S288c could restore cross protection. Surprisingly, S288c repaired with HAP1 YPS163 was largely unable to acquire further H2O2 resistance (Fig 6). This additional layer of genetic complexity suggests that S288c harbors additional polymorphisms that affect cross protection. To determine whether this was due to allelic variation in TOP3, the only other locus showing a difference in acquired H2O2 resistance, we genotyped each of the segregants at both the HAP1 and TOP3 loci. We identified two segregants with both the HAP1 YPS163 and TOP3YPS163 alleles that were nonetheless unable to acquire further resistance (S3 Fig, S1 Table). These data, along with the continuous distribution of F2 phenotypes (S1 Fig), is consistent with other loci outside of the chromosome XII QTL peak contributing to variation in acquired H2O2 resistance. Moreover, the causative alleles at these loci are apparently masked in YPS163-S288c hybrids that fully acquire H2O2 resistance, suggesting that they are recessive (see Discussion). We also noted during the genotyping that a small number of segregants contained the HAP1 S288c (or TOP3S288c) allele but were still able to acquire further H2O2 resistance (S3 Fig and S1 Table), suggesting that HAP1 function is conditionally necessary in certain genetic backgrounds. To determine whether this was due to a unique genetic background for YPS163, we deleted HAP1 in three additional wild strains. A wild oak (YPS1000) and wild vineyard (M22) strain showed defects in acquired H2O2 resistance similar to that of the YPS163 hap1Δ strain, while a wild coconut (Y10) strain showed a very slight defect (S4 Fig). Altogether, these results are consistent with HAP1 being necessary for ethanol-induced cross protection against H2O2 in some genetic backgrounds, including those of several wild strains, but not others (see Discussion).
Because Hap1p is a transcription factor, we hypothesized that acquired H2O2 resistance relied on Hap1p-dependent expression of a stress protectant protein. We reasoned that the putative stress protectant protein should have the following properties: i) a biological function consistent with H2O2 detoxification or damage repair, ii) reduced ethanol-responsive expression in S288c versus YPS163, iii) be a target gene of the HAP1 eQTL hotspot, and iv) possess evidence of regulation by Hap1p.
We first looked for overlap between our previously identified HAP1 eQTL hotspot (encompassing 376 genes) and genes with significantly reduced ethanol-responsive induction in S288c versus YPS163 (309 genes) [64]. Thirty-four genes overlapped for both criteria, including several that directly defend against reactive oxygen species (TSA2 encoding thioredoxin peroxidase, SOD2 encoding mitochondrial manganese superoxide dismutase, CTT1 encoding cytosolic catalase T, and GSH1 encoding γ-glutamylcysteine synthetase (Fig 7A and S1 Table)). Of those 34 genes, 8 also had direct evidence of Hap1p binding to their promoters [81] (Fig 7B and S1 Table), including CTT1 and GSH1 (though both TSA2 and SOD2 have indirect evidence of regulation by Hap1p [82, 83]).
We first focused on CTT1, since it is both necessary for NaCl-induced cross protection against H2O2 in S288c [84], and sufficient to increase H2O2 resistance when exogenously overexpressed in S288c [85]. We deleted CTT1 in the YPS163 background, and found that ethanol-induced cross protection against H2O2 was completely eliminated (Fig 8). The complete lack of cross protection in the ctt1Δ mutant suggests that other peroxidases cannot compensate for the lack of catalase activity under this condition. Next, because CTT1 was part of the HAP1 eQTL hotspot (Fig 7C, plotted using the data described in [64]), we tested whether the S288c HAP1 allele reduced CTT1 expression during ethanol stress. To do this, we performed qPCR to measure CTT1 mRNA induction following a 30-minute ethanol treatment (i.e. the peak ethanol response [45]). Consistent with our previous microarray data [45, 64], we saw lower induction of CTT1 by ethanol in S288c relative to YPS163 (Fig 9A). Moreover, we saw dramatically reduced induction of CTT1 in a YPS163 hap1Δ mutant compared to the wild-type YPS163 control (Fig 9A). Further support that HAP1 is causative for reduced CTT1 expression was provided by performing qPCR in the HAP1 reciprocal hemizygotes, where we found that the HAP1S288c allele resulted in significantly reduced CTT1 induction compared to the HAP1YPS163 allele (Fig 9A).
To determine whether the differences in CTT1 induction across strain backgrounds also manifested as differences in each strain’s ability to detoxify H2O2, we measured in vitro peroxidase activity in cell-free extracts. We compared in vitro peroxidase activity in extracts from unstressed cells and cells exposed to ethanol stress for 60 minutes (i.e. the same pre-treatment time that induces acquired H2O2 resistance (see Materials and Methods)). For wild-type YPS163, ethanol strongly induced peroxidase activity, and this induction was completely dependent upon CTT1 (Fig 9B). Mirroring CTT1 gene expression patterns, the induction of peroxidase activity was reduced in a YPS163 hap1Δ mutant. Additionally, reciprocal hemizygosity analysis provided further support that lack of HAP1 function results in decreased peroxidase activity, as the hybrid containing the HAP1S288c allele showed significantly reduced peroxidase activity following ethanol stress compared to the hybrid containing the HAP1YPS163 allele (Fig 9B). Notably, the hybrid containing the HAP1YPS163 allele had lower CTT1 induction and in vitro peroxidase activity following ethanol shock than wild-type YPS163, despite equivalent levels of acquired H2O2 resistance in the strains. These results suggest that HAP1 may play additional roles in acquired H2O2 resistance beyond H2O2 detoxification, depending upon the genetic background (see Discussion). Interestingly, S288c showed no induction of peroxidase activity upon ethanol treatment, despite modest induction of the CTT1 transcript. This result is reminiscent of Ctt1p regulation during heat shock in the S288c background, where mRNA levels increase without a concomitant increase in protein levels [84]. Thus, in addition to strain-specific differences in CTT1 regulation at the RNA level, there are likely differences in regulation at the level of translation and/or protein stability.
In this study, we leveraged extensive natural variation in the yeast ethanol response to understand potential connections between gene expression variation and higher-order organismal traits. Previous screens of gene deletion libraries have found surprisingly little overlap between the genes necessary for surviving stress and genes that are induced by stress. [34–43]. Instead, gene induction may be a better predictor of a gene’s requirement for acquired stress resistance [84]. Thus, we hypothesized that phenotypic variation in acquired stress resistance may be linked to natural variation in stress-activated gene expression. Our results provide a compelling case study in support of this notion—namely that a polymorphism in the HAP1 transcription factor affects natural variation in acquired H2O2 resistance, but not the basal H2O2 resistance of unstressed cells. Forward genetic screens have shown that the genes necessary for basal and acquired resistance are largely non-overlapping [34, 36, 84], suggesting that mechanisms underlying basal and acquired stress resistance are distinct. We provide further genetic evidence to support this model. YPS163 hap1Δ mutants and the hybrid carrying the HAP1S288c allele had strong acquired H2O2 defects, but no differences in their basal H2O2 resistance (Figs 4 and 6). Moreover, the YPS163 hap1Δ mutant was affected only when ethanol was the mild pretreatment, and was able to fully acquire H2O2 resistance following mild H2O2 or mild NaCl (Fig 5). These results suggest that the mechanisms underlying acquired resistance differ depending upon the mild stress that provokes the response. Further dissection of the mechanisms underlying acquired stress resistance will provide a more integrated view of eukaryotic stress biology.
Our results reveal a new role for Hap1p in cross protection against H2O2 that has been lost in the S288c lab strain. We propose that a major mechanism underlying ethanol-induced cross protection against H2O2 is the induction of cytosolic catalase T (Ctt1p), and that in the YPS163 background, Hap1p is necessary for proper induction of CTT1 during ethanol stress. We based this mechanism on the following observations. First, over-expression of CTT1 in S288c is sufficient to induce high H2O2 resistance [85]. Second, a YPS163 ctt1Δ mutant cannot acquire any further H2O2 resistance following ethanol pre-treatment (Fig 8), suggesting that no other antioxidant defenses are able to compensate under this condition. Lastly, the defect in cross protection for the YPS163 hap1Δ mutant correlates with reduced CTT1 expression and peroxidase activity during ethanol stress (compare Figs 6 and 9). How Hap1p is involved in the regulation of CTT1 during ethanol stress remains an open question, but we offer some possibilities. Hap1p is activated by heme, thus promoting transcription of genes involved in respiration, ergosterol biosynthesis, and oxidative stress defense including CTT1 [75, 76, 78, 82]. Because heme biosynthesis requires oxygen, Hap1p is an indirect oxygen sensor and regulator of aerobically expressed genes [74, 75, 86]. There is currently no evidence that heme levels are affected by ethanol stress, nor is there evidence that Hap1p is “super-activating” under certain conditions. Thus, we disfavor a mechanism of induction caused solely by Hap1p activation. Instead, we favor a mechanism where Hap1p interacts with other transcription factors at the CTT1 promoter during ethanol stress, leading to full CTT1 induction. One possibility that we favor is recruitment of the general stress transcription factor Msn2p, which plays a known role in acquired stress resistance [44, 45]. We previously showed that a YPS163 msn2Δ mutant had no induction of CTT1 mRNA during ethanol stress [45], suggesting that Msn2p was an essential activator for CTT1 under this condition. The CTT1 promoter region contains three Msn2p DNA-binding sites, two of which are ~100-bp away from the Hap1p binding site. Hap1p binding to the CTT1 promoter could help recruit Msn2p during ethanol stress, possibly through chromatin remodeling that increases accessibility of the Msn2p binding sites as proposed by Elfving and colleagues [87].
What is the physiological role of Hap1p-dependent induction of CTT1 during ethanol stress? One possibility is that regulation tied to the heme- and oxygen-sensing role of Hap1p ensures that CTT1 induction only occurs under environmental conditions where reactive oxygen species (ROS) are most likely to be encountered—namely stressful conditions that are also aerobic. In the context of ethanol stress, aerobic fermentation would lead to subsequent respiration of the produced ethanol and simultaneous ROS production. Under these conditions, CTT1 induction leading to ethanol-mediated cross protection against ROS would likely confer a fitness advantage. On the other hand, during stressful yet anoxic conditions, Ctt1p and other ROS-scavenging proteins are likely unnecessary. Furthermore, because heme is not synthesized during anoxic conditions [74], Hap1p would fail to induce CTT1 and other genes encoding non-essential heme-containing proteins. This may improve fitness by conserving energy used for biosynthesis and by redirecting limited heme to more essential heme-containing proteins.
The S288c lab strain has long been known to possess a defective HAP1 allele [79]. Apparently, the defective allele arose relatively recently, as only S288c contains a HAP1 Ty1 insertion out of over 100 sequenced strains [88, 89]. The lack of HAP1 function in S288c could be due to relaxation of selective constraint, though others have argued in favor of positive selection for reduced ergosterol biosynthetic gene expression [90, 91]. Regardless, the loss of ethanol-induced acquired H2O2 resistance is likely a secondary effect of the loss of Hap1p function. Intriguingly, we did find that two (non-S288c) domesticated yeast strains also lack ethanol-induced cross protection against H2O2 (S5 Fig), suggesting that phenotypic differences in acquired stress resistance may differentiate domesticated versus wild yeast. Because environmental stresses are likely encountered in combination or sequentially [92], acquired stress resistance is likely an important phenotype in certain natural ecological settings. Future studies directed at understanding differences in acquired stress resistance phenotypes in diverse wild yeast strains may provide unique insights into the ecology of yeast.
While our QTL mapping identified HAP1 as the major effector of cross protection, we note that additional complexity remains unexplained. Notably, despite the strong cross protection defect in the YPS163 hap1Δ mutant, some residual cross protection persists that is absent in S288c (Fig 6). Intriguingly, the residual cross protection is also absent in the hybrid carrying the HAP1S288c allele, suggesting the involvement of other genes depending upon the genetic background (Fig 4B and 4C). It is known that yeast strains with respiratory defects have increased ROS sensitivity [93, 94], potentially due to increased programmed cell death [95]. It is possible that reduced respiratory activity and concomitant ROS sensitivity in strains lacking HAP1 is exacerbated by genetic interactions with other alleles.
The lack of cross protection in S288c and the HAP1S288c hybrid correlates with the lack of inducible peroxidase activity following ethanol pretreatment in those strains. The lack of inducible peroxidase activity in S288c despite modest induction of CTT1 mRNA could be due to translational regulation, which is supported by the observation that while mild heat shock induces CTT1 mRNA, protein levels remain nearly undetectable [84]. Strikingly, the hybrid carrying the HAP1YPS163 allele still cross protects despite levels of CTT1 mRNA induction and peroxidase activity that are lower than in the YPS163 hap1Δ strain that is unable to acquire further resistance (Fig 9). These data suggest that HAP1 plays an additional role in ethanol-induced cross protection beyond H2O2 detoxification by Ctt1p. Moreover, the continuous distribution of the cross protection phenotype in the segregants (S1 Fig) and the results of allele swap experiments (Fig 6) strongly implicate other genes and processes in this complex trait. Specifically, the lack of complementation by the HAP1YPS163 allele in the S288c background suggests that additional loci in S288c render HAP1 necessary but not sufficient for cross protection in this background. Moreover, our genotyping of the segregants at HAP1 revealed a small number that still possessed cross protection in the absence of functional HAP1 (S3 Fig and S1 Table), suggesting that HAP1 is dispensable in certain genetic backgrounds. We examined the effects of hap1Δ mutations in other wild strain backgrounds and found two additional strains with a strong HAP1 requirement and a third strain with at most a mild HAP1 effect (S4 Fig). This result, as well as those from other recent studies [96–98], suggests that these types of genetic background effects are likely the rule rather than the exception. Future high resolution mapping experiments will be necessary to identify and characterize the source of these genetic background effects.
Gene expression variation is extensive in nature and is hypothesized to be a major driver of higher-order phenotypic variation. However, there are inherent challenges to connecting gene expression variation to higher-order organismal traits. Hundreds to thousands of genes are often differentially expressed across individuals, so identifying which particular transcripts exert effects on fitness is difficult. By studying acquired stress resistance—a phenotype better correlated with stress-activated gene expression changes—we were able to uncover a novel connection between gene expression variation and an organismal trait.
Strains and primers used in this study are listed in S2 and S3 Tables, respectively. The parental strains for QTL mapping were YPS163 (oak strain) and the S288c-derived DBY8268 (lab strain; referred to throughout the text as S288c). The construction of the S288c x YPS163 QTL mapping strain panel (44 F2 progeny) is described in [99] (kindly provided by Justin Fay). Genotypes for the strain panel are listed in S4 Table. During the course of analyzing HAP1 genotypes, we found one segregant (YS.15.2) to be a mixed population, so it was removed from subsequent analyses. Deletions in the BY4741 (S288c) background were obtained from Open Biosystems (now GE Dharmacon), with the exception of hap1 (whose construction is described in [45]). Deletions were moved into haploid MATa derivatives of DBY8268, M22, and YPS163 by homologous recombination with the deletion::KanMX cassette amplified from the appropriate yeast knockout strain [100]. Homozygous hap1Δ strains of YPS1000 and Y10 were generated by moving the hap1Δ::KanMX allele from the BY4741 background into the strains, followed by sporulation and tetrad dissection. All deletions were verified by diagnostic PCR. DBY8268 containing a wild-type HAP1 allele from YPS163 was constructed in two steps. First, the MX cassette from the hap1Δ::KanMX deletion was replaced with a URA3MX cassette, selecting for uracil prototrophy. Then, URA3 was replaced with wild-type HAP1 from YPS163 (amplified using primers 498-bp upstream and 1572-bp downstream of the HAP1 ORF), while selecting for loss of URA3 on 5-fluoroorotic acid (5-FOA) plates. Deletions and repair of HAP1 were confirmed by diagnostic PCR (see S3 Table for primer sequences). YPS163 containing a HAP1S288c allele was constructed by first inserting a KanMX cassette into S288c 117-bp downstream of the Ty element to create JL1032. We then amplified and transformed the Ty element into YPS163 using primers that annealed 103-bp upstream of the Ty element and 177-bp downstream of the KanMX cassette, generating JL1069. Diploid strains for HAP1 and TOP3 reciprocal hemizygosity analysis were generated as follows. The hemizygote containing the wild-type S228c HAP1 allele (JL580) was generated by mating JL140 (YPS163 MATa hoΔ::HygMX hap1Δ::KanMX) to JL506 (DBY8268 MATα ho ura3 hap1). The hemizygote containing the wild-type YPS163 allele (JL581) was generated by mating JL112 (YPS163 MATα hoΔ::HygMX HAP1) to JL533 (DBY8268 MATa ho ura3 hap1Δ::KanMX). The hemizygote containing the wild-type S288c TOP3 allele (JL1107) was created by mating JL1066 (YPS163 MATa hoΔ::HygMX top3Δ::KanMX) to BY4742 (MATα TOP3). The hemizygote containing the wild-type YPS163 allele (JL1106) was created by mating JL1121 (BY4741 MATa top3Δ::KanMX) to JL112 (YPS163 MATα hoΔ::HygMX TOP3). All strains were grown in batch culture in YPD (1% yeast extract, 2% peptone, 2% dextrose) at 30°C with orbital shaking (270 rpm).
To identify possible promoter polymorphisms, the HAP1 promoters of the DBY8268 (JL505), YPS163 (JL111), and S288c HAP1YPS163 (JL975) strains were amplified using primers that anneal 1091-bp upstream and 134-bp downstream of the HAP1 start codon. PCR products were purified with a PureLink PCR cleanup kit (Invitrogen) and sequenced by Sanger Sequencing (Eurofins Genomics) using a primer that anneals 498-bp upstream of the HAP1 start codon. Sequences were aligned to the S288c and YPS163 reference sequences using SnapGene v4.1 (GSL Biotech). This verified the presence of a 1-bp indel within a poly-A stretch that differs between S288c and YPS163. The S288c HAP1YPS163 (JL975) strain contains the YPS163 HAP1 promoter sequence. Additionally, the YPS163 strain containing the HAP1S288c was constructed to only contain the Ty element and not the S288c promoter polymorphism.
The HAP1 allele of each segregant for the QTL mapping panel was genotyped by differential PCR analysis where the same forward primer (HAP1 int 3’ F) was paired with two different reverse primers. One primer (Ty R) anneals specifically to the Ty element, yielding an 856-bp product when amplifying the S288c allele. The second primer (HAP1 3’ end R) anneals 3’ to the Ty element of HAP1S288c, yielding a 570-bp product for HAP1YPS163 and a 6.5-kb product for HAP1S288c. Each segregant was genotyped using both sets of primer pairs, and only one segregant (YS.15.2) appeared to contain both HAP1 alleles. Subsequent analysis of multiple colonies verified that YS.15.2 was a mixed population, and thus it was removed it from all subsequent analyses.
The TOP3 alleles of S288c and YPS163 contain two non-synonymous SNPs at nucleotide positions 1,398 and 1,422. Segregant genotypes at TOP3 were determined by analyzing restriction fragment length polymorphisms. TOP3 was amplified using primers (TOP3 up F and TOP3 down R) that anneal ~500-bp upstream and downstream of the open reading frame, generating a 2.9-kb product. PCR products were digested with either 1) PstI, which cuts at position 1,248 only within the TOP3YPS163 ORF allele yielding 1.7- and 1.2-kb products, or (2) KflI, which cuts at position 1,155 only within the TOP3S288c yielding 1.6- and 1.3-kb products. Genotypes for HAP1 and TOP3 are listed in S1 Table.
Cross-protection assays were performed as described in [44] with slight modifications. Briefly, 3–4 freshly streaked isolated colonies (<1 week old) were grown overnight to saturation, sub-cultured into 6 ml fresh media, and then grown for at least 8 generations (>12 h) to mid-exponential phase (OD600 of 0.3–0.6) to reset any cellular memory of acquired stress resistance [85]. Each culture was split into two cultures and pretreated with YPD media containing either a single mild “primary” dose or the same concentration of water as a mock-pretreatment control. Primary doses consisted of 5% v/v ethanol, 0.4 M NaCl, or 0.4 mM H2O2. Thereafter, mock and primary-treated cells were handled identically. Following 1-hour pretreatment at 30°C with orbital shaking (270 rpm), cells were collected by mild centrifugation at 1,500 x g for 3 min. Pelleted cells were resuspended in fresh medium to an OD600 of 0.6, then diluted 3-fold into a microtiter plate containing a panel of severe “secondary” H2O2 doses ranging from 0.5–5.5 mM (0.5 mM increments; 150 μl total volume). Microtiter plates were sealed with air-permeable Rayon films (VWR), and cells were exposed to secondary stress for 2 hours at 30°C with 800 rpm shaking in a VWR symphony Incubating Microplate Shaker. Four μl of a 50-fold dilution was spotted onto YPD agar plates and grown 48 h at 30°C. Viability at each dose was scored using a 4-point semi-quantitative scale to score survival compared to a no-secondary stress (YPD only) control: 100% = 3 pts, 50–90% = 2 pts, 10–50% = 1 pt, or 0% (3 or less colonies) = 0 pts. An overall H2O2 tolerance score was calculated as the sum of scores over the 11 doses of secondary stress. Raw phenotypes for all acquired stress resistance assays can be found in S1 Table. A fully detailed acquired stress protocol has been deposited to protocols.io under doi dx.doi.org/10.17504/protocols.io.g7sbzne. Statistical analyses were performed using Prism 7 (GraphPad Software).
Phenotyping of the QTL mapping strain panel for basal and acquired H2O2 resistance was performed in biological duplicate. Because cross-protection assays on the entire strain panel could not all be performed at the same time, we sought to minimize day-to-day variability. We found that minor differences in temperature and shaking speed affected H2O2 resistance; as a result, we used a digital thermometer and tachometer to ensure standardization across experiments. Moreover, we found that differences in handling time were a critical determinant of experimental variability. To minimize this source of variability, all cell dilutions were performed quickly using multichannel pipettes, and no more than two microtiter plates were assayed during a single experiment. To ensure that replicates on a given day were reproducible, we always included the YPS163 wild-type parent as a reference.
Single mapping scans were performed using Haley-Knott regression [101] implemented through the R/QTL software package [102]. Genotype probabilities were estimated at every cM across the genome using the calc.genoprob function. Significant LOD scores were determined by 100,000 permutations that randomly shuffled phenotype data (i.e. strain labels) relative to the genotype data. The maximum LOD scores for the permuted scans were sorted, and the 99th percentile was used to set the genome-wide FDR at 1%. This resulted in LOD cutoffs of 3.07 for QTL mapping of basal H2O2 resistance, and 4.24 for acquired H2O2 resistance.
Broad-sense heritability (H2) was estimated from the segregant data as described in [71] using a random-effects ANOVA model implemented through the lmer function in the lme4 R package [103]. H2 was estimated using the equation σG2(σG2+σE2), where σG2 represents the genetic variance due to the effects of segregrant, and σE2 represents the residual (error or environmental) variance. The proportion of variance explained by a QTL was estimated using the equation 1−10(−2n*LOD), where n represents the number of segregants.
Induction of CTT1 by ethanol was assessed by real-time quantitative PCR (qPCR) using the Maxima SYBR q-PCR Master Mix (Thermo Fisher Scientific) and a Bio-Rad CFX96 Touch Real-Time PCR Detection System, according to the manufacturers’ instructions. Cells were grown to mid-exponential phase (OD600 of 0.3–0.6) as described for the cross-protection assays. Cells were collected by centrifugation at 1,500 x g for 3 minutes immediately prior to the addition of 5% v/v ethanol (unstressed sample) and 30 minutes post-ethanol treatment, which encompasses the peak of global expression changes to acute ethanol stress [45]. Cell pellets were flash frozen in liquid nitrogen and stored at -80°C until processed. Total RNA was recovered by hot phenol extraction as previously described [104], and then purified with a Quick-RNA MiniPrep Plus Kit (Zymo Research) including on-column DNase I treatment. cDNA synthesis was performed as described [104], using 10 μg total RNA, 3 μg anchored oligo-dT (T20VN), and SuperScript III (Thermo Fisher Scientific). One ng cDNA was used as template for qPCR with the following parameters: initial denaturation at 95°C for 3 minutes followed by 40 cycles of 95°C for 15 seconds and 55°C annealing and elongation for 1 minute. Cq was determined using regression analysis, with baseline subtraction via curve fit. The presence of a single amplicon for each reaction was validated by melt curve analysis. The average of two technical replicates were used to determine relative CTT1 mRNA abundance via the ΔΔCq method [105], by normalizing to an internal control gene (ERV25) whose expression is unaffected by ethanol stress and does not vary in expression between S288c and YPS163 [45]. Primers for CTT1 and ERV25 were designed to span ~200 bp in the 3’ region of each ORF (to decrease the likelihood of artifacts due to premature termination during cDNA synthesis), and for gene regions free of polymorphisms between S288c and YPS163 (see S3 Table for primer sequences). Three biological replicates were performed and statistical significance was assessed via a paired t-test using Prism 7 (GraphPad Software).
For peroxidase activity assays, mid-exponential phase cells were collected immediately prior to and 60 minutes post-ethanol treatment, to assess peroxidase activity levels during the induction of cross protection. Cells were collected by centrifugation at 1,500 x g for 3 minutes, washed twice in 50 mM potassium phosphate buffer, pH 7.0 (KPi), flash frozen in liquid nitrogen, and then stored at -80°C until processed. For preparation of whole cell extracts, cells were thawed on ice, resuspended in 1 ml KPi buffer, and then transferred to 2-ml screw-cap tubes for bead beating. An equal volume (1 ml) of acid-washed glass beads (425–600 micron, Sigma-Aldrich) was added to each tube. Cells were lysed by four 30-second cycles of bead beating in a BioSpec Mini-Beadbeater-24 (3,500 oscillations/minute, 2 minutes on ice between cycles). Cellular debris was removed by centrifugation at 21,000 x g for 30 minutes at 4°C. The protein concentration of each lysate was measured by Bradford assay (Bio-Rad) using bovine serum albumin (BSA) as a standard [106]. Peroxidase activity in cellular lysates was monitored as described [107], with slight modifications. Briefly, 50 μg of cell free extract was added to 1 ml of 15 mM H2O2 in KPi buffer. H2O2 decomposition was monitored continuously for 10 minutes in Quartz cuvettes (Starna Cells, Inc.) at 240 nm (ε240 = 43.6 M-1 cm-1) using a SpectraMax Plus Spectrophotometer (Molecular Devices). One unit of catalase activity catalyzed the decomposition of 1 μmol of H2O2 per minute. For each sample, results represent the average of technical duplicates. To assess statistical significance, four biological replicates were performed and significance was assessed via a paired t-test using Prism 7 (GraphPad Software).
|
10.1371/journal.pgen.1001280 | Segregating Variation in the Polycomb Group Gene cramped Alters the Effect of Temperature on Multiple Traits | The phenotype produced by a given genotype can be strongly modulated by environmental conditions. Therefore, natural populations continuously adapt to environment heterogeneity to maintain optimal phenotypes. It generates a high genetic variation in environment-sensitive gene networks, which is thought to facilitate evolution. Here we analyze the chromatin regulator crm, identified as a candidate for adaptation of Drosophila melanogaster to northern latitudes. We show that crm contributes to environmental canalization. In particular, crm modulates the effect of temperature on a genomic region encoding Hedgehog and Wingless signaling effectors. crm affects this region through both constitutive heterochromatin and Polycomb silencing. Furthermore, we show that crm European and African natural variants shift the reaction norms of plastic traits. Interestingly, traits modulated by crm natural variants can differ markedly between Drosophila species, suggesting that temperature adaptation facilitates their evolution.
| The fruitfly Drosophila melanogaster was initially endemic to sub-Saharan Africa and started to colonize Europe and Asia around 10,000 years ago. Northern populations have adapted to these colder environments and differ from Sub-Saharan populations for temperature sensitive traits. Here we analyze cramped (crm), a gene previously identified as a putative target of adaptive selection during the colonization of northern latitudes. crm is involved in the regulation of chromosome structure, a process known to be strongly modulated by temperature. We show that crm is limiting for distinct processes at different temperatures and that crm natural variation modulates temperature sensitive phenotypes. Our results suggest that environmental heterogeneity maintains functional variation in environment sensitive gene networks and might facilitate evolution.
| Environmental conditions can strongly modulate the phenotypes produced by particular genotypes (phenotypic plasticity). Recent studies have stressed how important it is to take into account genetic variation and environmental conditions to analyze the properties of gene regulatory networks [1], [2]. Indeed gene regulatory networks have been selected to cope with variable genetic backgrounds and environmental conditions. This explains, for example, the redundancy of particular regulatory sequences [3], [4]. Selection can limit the influence of the environment (environmental canalization) [5] or compensate it so that different genotypes maintain the same phenotype in different environments (genetic compensation) [6]. Alternatively, the influence of the environment can be selected for so that the different phenotypes produced in distinct environments by a given genotype are optimal in each environment (adaptive phenotypic plasticity) [7]. The spatial and temporal heterogeneity of environmental conditions leads to continuous adaptation of natural populations. It maintains a high genetic variation in environment sensitive gene networks, which can be easily revealed in artificial selection experiments [5], [8], [9]. This high genetic variation is thought to facilitate evolution. These ideas are actively discussed [10]–[12], but experiments analyzing the effect of variation in individual genes in different environmental conditions are required to test them.
We analyze here a candidate gene for adaptation in Drosophila melanogaster. This species was initially endemic to sub-Saharan Africa and started to colonize the rest of the world around 10,000 years ago. Sub-Saharan and European populations now differ dramatically for temperature sensitive traits (such as ovariole number, pigmentation, size, and cold resistance) [13]–[15], indicating that flies have adapted to these new, partially colder, environments. A region of the X chromosome containing the gene cramped (crm) was identified as being putatively involved in adaptation to northern latitudes [16]. crm is involved in Polycomb silencing and position effect variegation (PEV) [17]. These two processes, linked to chromatin regulation, are known to be strongly modulated by temperature [18]. Here, we investigate the involvement of crm in thermal plasticity and whether crm natural variation could have contributed to the adaptation of Drosophila populations to environmental temperature.
In natural D. melanogaster populations, apart from rare amino acid polymorphisms segregating in some African alleles only, amino acid polymorphisms are observed in three regions (A, B, C) in the C-terminal half of CRM [16] (Figure 1A). Analysis of the ratio of non-synonymous to synonymous substitutions (Ka/Ks), when compared to the CRM homologue of the closely related species D. simulans using a sliding window (Figure 1B) suggests that recurrent selection is driving amino acid replacements in these regions. These rapidly evolving regions are located between conserved domains, and are predicted to be intrinsically disordered [19] (Figure 1C). Disordered domains are natively unfolded. They are frequently found in transcription factors and are thought to facilitate protein complex formation. They often carry post-translational modifications modulating protein interactions and stability [20], [21]. The geographic partitioning of variation in D. melanogaster strongly suggests that natural selection has recently shaped the distribution of CRM variants during and after the out of Africa expansion: the derived mutations A639G and T842M are absent in the African sample but found at high frequency in Europe (Text S1; Figure S1, Figure S2).
Altogether the patterns of coding variation and divergence in crm suggest that spatially varying selection favors particular amino acid replacements and leads to the rapid evolution of specific domains. We first sought to identify processes particularly sensitive to crm in controlled temperature conditions in order to test later how crm natural variants affect these processes.
Processes particularly sensitive to crm can be identified by the aberrant phenotypes observed in crm mutants. We analyzed the phenotypes of crm mutants at different temperatures using a null allele, crm32. The phenotypes observed with the crm32 allele at different temperatures are caused by this allele and not by another mutation in the genome because they are not observed when a crm genomic rescue construct is present in the genome (see below). Wild type Drosophila melanogaster males have a single sex comb located on the first tarsal segment of each prothoracic leg. crm null mutant males, however, display ectopic sex combs, both distally on the prothoracic leg and on more posterior legs [17] (Figure 2A). Posterior sex combs represent the canonical Polycomb phenotype, caused by a defect in the repression of the homeotic gene Sex comb reduced (Scr) [22]. We observed that both ectopic sex comb phenotypes are increased at higher temperatures (Figure 2D: N = 10 for each temperature point), as previously described with another crm allele, crm7 [17], [23]. Conversely, other mutant phenotypes (abdominal dorsal fusion defects and wing margin defects) are enhanced at lower temperatures (Figure 2B, 2C). This is most obvious for the dorsal fusion defects, which are observed at 18°C but not at 25°C or 29°C. These experiments indicate that crm is involved in at least two distinct processes, and that its ability to buffer these processes is differentially required at low or high temperatures.
Two additional pieces of evidence demonstrate the pleiotropic nature of crm. First, the different phenotypes found in crm mutants can be enhanced by mutations in different chromatin regulators. For example, the ectopic posterior sex comb phenotype can be enhanced by mutations in Polycomb Group genes (PcG) such as Polycomb-like (Pcl) (Figure S3). The distal sex comb phenotype was shown previously to be dramatically enhanced by heterozygosity for a null allele of the transcription factor bric-à-brac (bab) [23]. In contrast, the ectopic sex comb phenotypes are decreased and the dorsal fusion defects are enhanced by mutations in Su(var)3-9, required for the formation of centromeric heterochromatin [24] (Figure S3). Indeed, dorsal fusion defects are visible at 25°C in crm32; Su(var)3917 males, and are not observed in Su(var)3917 males. Second, deletions of different conserved domains in CRM, affect different phenotypic read-outs. Indeed, it is possible to find CRM homologues in organisms as distantly related as plants [25] and to identify several conserved domains, labeled I to VII (Figure 1C, Figure S4). We generated transgenes allowing the conditional expression of mutant forms with deletions of particular conserved domains (Figure S4). We observed that several of them behave as dominant negatives. Although these types of mutants are very artificial and may induce defects difficult to interpret, they represent a way to disturb the network in vivo from inside as these truncated forms are able to interact with some cofactors, but not with others. In our case, this approach turned out to be very informative as the dominant mutants differently affect the processes sensitive to crm loss of function at low or high temperature: ubiquitous expression of CRM mutant forms with deletion of domain II or VI induces the formation of ectopic sex combs (both distal and posterior) (Figure S4B, S4C). Ubiquitous expression of a mutant form with a deletion of domain V leads to dorsal fusion defects (Figure S4B), and when expressed in the wing imaginal disc using the bi/omb-Gal4 driver, induces a strong wing growth defect (Figure S4E). Furthermore, the expression of this del-V dominant mutant in the dorsal region using the driver Pnr-Gal4 increases thoracic and abdominal pigmentation (Figure S4D). In contrast, it induces much more weakly ectopic sex combs. In conclusion, CRM is a highly pleiotropic protein with a modular structure. Furthermore, at particular temperatures (cold/hot), CRM is limiting for distinct processes (dorsal fusion/posterior sex comb restriction) through interactions involving different domains (V/II, VI) and different factors (Su(var)3.9/Pcl).
We used the strong phenotypes induced by the crm dominant negative crm-del-V to identify genes particularly sensitive to crm disruption. By testing genetically several genes involved in wing development we identified cubitus interruptus (ci) encoding the effector of the Hedgehog signaling pathway as interacting with crm. We tested whether ci is mis-regulated in crm mutants using the LacZ enhancer trap inserted in ci (Figure 3A). In the wing disc, ci is spatially regulated. It is expressed in the anterior compartment and repressed in the posterior compartment. We observed that driving the expression of a CRM-del-V dominant negative in the wing disc leads to an ectopic expression of ci-LacZ in the posterior compartment (Figure 3B). To confirm the relevance of this finding, we also tested crm LOF mutants. Loss of crm function also leads to ectopic (though weaker) ci expression in the posterior compartment (Figure 3B). Interestingly, ci-LacZ expression is also reduced in the anterior compartment in crm null mutants (Figure 3B), suggesting that crm is required for both repression and activation of ci in the wing disc.
The PcG protein Polyhomeotic (PH) has been shown to repress ci directly [26]. Accordingly, in salivary gland nuclei, where ci is repressed, the PH protein is bound to the ci locus [26]. We therefore tested whether crm is required for PH binding to the ci locus, by comparing PH distribution in crm null mutant and wild type salivary glands. Interestingly, while most PH bands on salivary gland polytene chromosomes remain unchanged in crm mutants, the PH band at ci disappears (Figure 3C), suggesting a very specific role for crm in the regulation of ci. A role of crm in regulating ci is further demonstrated by the derepression of a ci-LacZ reporter gene in salivary glands of crm mutants (Figure 3D). Note that the ci-LacZ de-repression is stochastic as the intensity of the staining differs strongly between nuclei (Figure 3E). Altogether, these experiments on salivary glands indicate that crm is specifically required for the direct and stable repression of ci by PH.
Further evidence supporting a role for crm in the silencing of ci comes from examining white marked transposons inserted near the ci locus. Previously, a Polycomb response element (PRE) has been mapped on a 1Kb fragment, 4 kb upstream of ci [26] (see Figure 3A). Two Drosophila deletion lines, Df(4)ED6364 and Df(4)ED6366, differing by a 10 kb region covering the ci PRE, contain a copy of the white gene, in an identical transgene, just upstream of ci [27] (Figure 3A). The silencing effect of the region containing the PRE can be seen in these two lines, which have markedly different eye colors. Df(4)ED6366, which retains the PRE is variegated and much lighter than Df(4)ED6364, in which the PRE is absent (Figure 3F). When the deletion retaining the PRE (Df(4)ED6366), is introduced in ph heterozygote females or crm32 mutant males, the eye color becomes uniform and much darker, showing that both ph and crm are required for the repression of the white transgene (Figure 3G, 3H). As in the line Df(4)ED6366, the white gene is inserted around 10kb upstream of ci (approximately 400 bp downstream of the dTcf promoter), we believe that more than just ci is sensitive to the PRE. In agreement with this hypothesis, previous mapping of Polycomb (Pc) binding sites revealed that the whole ci-dTcf region corresponds to a Pc enriched domain, centered on ci, but reaching the promoter of dTcf [28] (Figure 3A).
Interestingly, ci lies close to the fourth chromosome pericentromeric heterochromatin. This heterochromatic environment is required for ci regulation as translocations separating ci from the pericentromeric heterochromatin of the fourth chromosome do not fully complement the ci1 mutant allele [29]–[31]. Furthermore, previous data suggested that silencing by constitutive heterochromatin is stronger at low temperatures, whereas PcG silencing is stronger at higher temperatures [18]. We therefore investigated whether this antagonistic effect of temperature also applies to the genomic region around ci.
We analyzed the effect of temperature on chromatin regulation in the ci-dTcf region using the line Df(4)ED6366. To compare it to the effect of temperature on neighboring constitutive heterochromatin, we used the line 118E10, carrying a transgene inserted in fourth chromosome pericentromeric heterochromatin [32]. We observed that high temperature reduces silencing in the ci-dTcf region (Figure 4). In contrast, we observed a strong reduction of PEV at low temperature with the line 118E10. The down regulation of the reporter gene near pericentromeric heterochromatin therefore shows temperature sensitivity opposite to that of the neighboring ci-dTcf region in the eye (Figure 4). Furthermore, mutating crm de-represses the monitored reporter genes at all temperatures tested, in both lines. Silencing of the ci-dTcf region is weaker in crm mutants at all temperatures, but this effect is particularly pronounced at low and intermediate temperature (Figure 4). Thus, crm is essential for the stronger down regulation observed at low temperature. Furthermore, LOF mutation in crm, has a strong effect on the 118E10 transgene in particular at intermediate and high temperature (Figure 4). This confirms a role for crm in the formation of constitutive heterochromatin, previously shown with X chromosome pericentromeric heterochromatin [17] and shows that crm is required for the stronger silencing by pericentromeric heterochromatin observed at high temperature.
ci and dTcf are the effectors of the Hedgehog (Hh) and Wingless (Wg) signaling pathways. These two major signaling pathways (and the dpp signaling pathway that is regulated by them) are involved in the development of many morphological traits, including several temperature sensitive traits under the control of crm, such as abdominal dorsal fusion and sex comb development [33]–[35]. In addition, Hedgehog and Wingless regulate together dpp and optomotor-blind/bifid (omb/bi), both involved in the control of abdominal pigmentation, a very plastic trait [23], [36]–[38]. As the regulation of the ci-dTcf region is temperature sensitive, it likely contributes to the thermal plasticity of these traits.
We designed sensitive phenotypic tests based on the results above to analyze functional differences between natural alleles. As we expected these differences to be much weaker than those observed with typical laboratory mutants and to be obscured by variation at other loci, we performed the tests in isogenic backgrounds to remove the effect of variation at other loci (see Methods). Four crm alleles corresponding to the pairwise combinations of the two amino acid polymorphisms A639G and T842M that are presumably affected by adaptation in cosmopolitan D. melanogaster were tested. We scored abdominal pigmentation in females, and distal sex combs or posterior sex combs in males (Figure S5). Because D. melanogaster males do not have distal or posterior sex combs, we analyzed the modification by the crm variants of the ectopic distal or posterior sex comb phenotypes induced by mutations in genes interacting with crm (see Methods). In addition, we tested the effect of crm natural variants on the ci-dTcf region. For this, we analyzed how they modify the wing vein phenotype caused by the dominant allele ciD, which induces an ectopic expression of ci in the posterior compartment [31]. This phenotype was shown previously to be very sensitive to both natural genetic variation and temperature [39]. We observed indeed a strong interaction with temperature as the ciD vein gap phenotype was not visible in flies grown at 18°C. We analyzed it only for flies grown at 25°C (Figure S5). These dominant sex comb and wing vein phenotypes are therefore not used as wild type traits, but as read out to compare the effects of the crm natural variants on crm sensitive processes.
The four crm variants have different reaction norms for all traits: abdominal pigmentation (Two-way ANOVA PC1 p<0.001 and PC2 p<0.005), distal sex combs (Scheirer-Ray-Hare test p<0.05), posterior sex combs (Two-way ANOVA p<0.01), and ciD phenotype (One-way ANOVA p<0.001). However, depending on the phenotypic trait, the crm variants behave differently (Figure 5). The AT combination, which represents an African variant is different from the two non-African variants (GT and GM), for female abdominal pigmentation, distal sex combs and the ciD phenotype. A significant genotype X temperature interaction is observed for female abdominal pigmentation with this allele (principal component 2). The artificial AM form is clearly different from all three other variants for the posterior sex comb and the ciD phenotypes. It groups together with the AT variant for distal sex combs.
To partition the effect of each site on the phenotype, we used three-way ANOVA. In addition to temperature (p<0.001 for all traits), we found significant main effects for each of the polymorphic sites with posterior sex combs and pigmentation. For distal sex combs, the A639G polymorphism, (but not the T842M polymorphism) has a significant main effect (Scheirer-Ray-Hare test p<0.01). In addition, the two polymorphic sites show significant interactions for all traits. In other words, our data show strong intra-molecular epistasis (posterior sex comb p<0.001; pigmentation p<0.001; ciD phenotype <0.001) (Figure 5). For the distal sex comb, the interaction A639G×T842M is significant for the ANOVA on ranks (p = 0.01), but not with the very conservative Scheirer-Ray-Hare test (p = 0.37).
In a genome wide screen for adaptation during the out of African expansion of Drosophila melanogaster, crm was identified as a candidate gene [16]. Consistent with a role for crm in adaptation to temperature, we show here, using mutants, that crm is limiting for different developmental regulatory processes at distinct temperatures. We identified the ci-dTcf region as particularly sensitive to crm. crm affects ci regulation through both constitutive heterochromatin and Polycomb silencing. Our results based on white expression in the eye suggest that pericentromeric heterochromatin is negatively correlated to Polycomb silencing in the ci-dTcf region and that temperature affects inversely Polycomb silencing and pericentromeric heterochromatin. This temperature sensitivity of the ci-dTcf regulation contributes to phenotypic plasticity. By maintaining Polycomb silencing and constitutive heterochromatin, crm appears to contribute to environmental canalization of plastic traits.
Our analysis shows that crm natural variants shift the reaction norms of plastic traits. We observe two effects: a change in slope indicating a modification of the environmental canalization of these traits, or a change in mean, interpretable as genetic compensation (see below). The ancestral combination A639T842 present in Africa reduces the ectopic distal sex comb and the ciD phenotypes. Compared to the European combinations GT and GM, it reduces the effect of high temperature on these traits, allowing different genotypes to maintain similar phenotypes in different environments, a process called genetic compensation [6]. The ancestral African combination AT also decreases female abdominal pigmentation and interacts with temperature for this trait (principal component 2). Abdominal pigmentation shows a complex pattern of geographical variation with both pale and dark phenotypes observed in Sub-Saharan and European regions [40], [41]. In general, flies living in colder regions (high altitude in Africa, Europe versus Indian) are more pigmented [42], [43]. It is thought to be adaptive because darker flies warm up more quickly [15]. The reduced pigmentation of the ancestral African combination A639T842 fits therefore this thermal budget hypothesis.
An effect of the G842M mutation is observed in our tests only in combination with A639. The AM form is observed neither in Africa nor in Europe, because the T842M mutation likely occurred on a G639T842 haplotype. This combination could theoretically be produced by recombination between African AT and European GM forms. Interestingly, admixture between European and African Drosophila melanogaster populations is known to have happened in America [44], [45]. In a limited sequencing survey, such a combination (A639M842) was identified in a population from the Caribbean island Gossier, indicating that it actually exists in the wild (Harr and Schlötterer, unpublished data). As our phenotypic tests show that this recombinant form differs functionally strongly from the others, future studies on Caribbean flies could provide interesting insights on the modulation of the crm network.
The strong epistasis we observe between the two amino acid positions suggest that the elevated divergence observed in these regions between closely related Drosophila species might be caused by a permanent adjustment of one region to the other in particular environmental conditions. Although the molecular interactions modulated by these amino acid polymorphisms are unknown, we note that these regions contain several potential sites of phosphorylation. Our results suggest that amino acid replacements in pleiotropic factors play a significant role in evolution. In contrast, recent studies have stressed the role of regulatory sequences in evolution as their modification has less pleiotropic consequences than amino acid changes [46]. However comparison of closely related species such as D. melanogaster and D. simulans reveals that many amino acid replacements seem to be under positive but weak selection [47]. Our results suggest that many of these changes might correspond to compensatory mutations adjusting to variation in other genomic regions and environmental conditions to stabilize gene regulatory networks.
Other studies analyzing natural variation have suggested also a role of several chromatin regulators in adaptation to different temperature regimes [48]. In addition, African alleles of PH, the Polycomb group factor that requires crm to bind to the ci region, also shows the footprint of positive selection [49]. Furthermore, the bab locus, which interacts genetically with crm, harbors natural variation that strongly modulates abdominal pigmentation [50]. Thus, a whole network of chromatin regulators is apparently involved in local temperature adaptation in Drosophila melanogaster. Interestingly, distal sex comb and abdominal pigmentation, which are modulated by crm polymorphisms and are also regulated by bab and ph in Drosophila melanogaster, can differ markedly between Drosophila species [51]–[53]. It suggests that local adaptation might facilitate the evolution of these traits, although it probably primarily optimizes other more crucial correlated temperature sensitive traits. For example, ovary development or male fertility are also temperature sensitive and show geographical variation [15], [54]. We do not mean that selection does not play any role in the evolution of sex combs or pigmentation. Indeed, the sex comb is under strong selection as its genetic ablation reduces mating performance [55]. Similarly, clinal and altitudinal variation suggest that abdominal pigmentation is under selection [42], [43]. However, selection can act only if genetic variation has phenotypic consequences. We suggest that by maintaining functional genetic variation in temperature sensitive gene networks, environment heterogeneity contribute to the high evolvability of particular correlated traits.
We used the data set previously published [16] and added the crm haplotype from the sequenced strain of Drosophila melanogaster [56]. As outgroup we used a crm haplotype from Drosophila sechellia [16] and a consensus between the sequenced traces from different Drosophila simulans haplotypes (Genome sequencing center, Washington School of Medicine).
We used TREE-PUZZLE 5.2 [57] with the Tamura-Nei (1993) model of sequence evolution to build maximum likelihood phylogenies using the nucleotide sequences of the region A, B and C. Region A (1000 bp) and B (576bp) are two parts of exon 4. Region C (374bp) corresponds to the exon 5.
Ka/Ks ratio was analyzed using all the D. melanogaster and the Drosophila simulans CRM coding sequences with 50 sites windows step size of 10, excluding sites with gaps, using DNASP5 [58].
CRM homolgues were identified by BLAST in other Drosophila (D. simulans, D. sechellia, D. yakuba, D. erecta, D. ananassae, D. pseudoobscura, D. persimilis, D. wilistoni, D. mojavensis, D. virilis and D. grimshawi), the mosquito Aedes aegypti, the beetle Tribolium castaneum, the aphid A. pisum, the wasp Nasonia vitripennis, the chelicerate I. scapularis, several vertebrates including Homo sapiens and the zebrafish B. rerio, two basal metazoans, the cnidarian N. vectensis and the placozoan T. adherens, several plants (including A. thaliana and O. sativa) and an alga M. pusilla.
Sequences were aligned using clustal w [59]. Alignments were improved manually. In order to associate the prediction of disordered domains in Drosophila melanogaster CRM and the level of conservation with its homologues (Figure 1C), positions with a gap in the D. melanogaster CRM were removed from the alignment. Similarity index was calculated with the JProfieGrid Software with a window size of 5 and a threshold of 0.7 [60]. Disorder tendency was calculated for the D. melanogaster CRM sequence using the IUpred server [19].
The crm32 null allele was generated by imperfect excision of the P-element inserted in the line P{EPgy2}crmEY05302. The promoter and the first 286 codons are deleted. All other fly stocks are described in flybase (http://flybase.org). Flies were grown on standard agar-corn medium. We used standard balancer chromosomes.
WT or mutant crm coding sequences with deletion of particular conserved domains were inserted in frame 3′ to the coding sequence of the fluorescent protein Venus [61] in the vector PUAST-attB [62]. The amino acid deleted in the mutant forms were: 68–271 (replaced by a proline, CRM-Del-I), 351–476 (CRM-Del-II), 487–621 (CRM-Del-III), 620–747 (replaced by an alanine, CRM-Del-IV), 736–821 (CRM-Del-V), 848–950 (CRM-Del-VI) and 951–982 (CRM-Del-VII). The crm cDNA used to construct these forms was the clone LD29481 from the BDGP Gold collection. The obtained plasmids were integrated in the plattform 22A on the second chromosome using the PhiC31 transgenesis method [62]. They are thus under the same position effects and can be compared to one another.
We generated in vitro the four pairwise combinations of the amino acid variants A639G (GCC/GGC) and T842M (ACG/ATG). The mutagenesis was performed on a 4912 pb genomic fragment containing the crm allele from the OregonR stock (G639T842 allele). The genomic fragments (including crm regulatory sequences) were cloned in a modified PUAST-attB vector where a PstI-EcoRI fragment containing the HSP70 promoter and the UAS sequences was replaced by a primer containing the restriction sites PstI, NdeI, StuI and EcoRI. We used a natural NotI site present in the middle of the crm genomic fragment to amplify and clone independantly the 5′ half (2621 pb including a BglII restriction site added to the direct primer) and 3′ half (2312 pb including an XbaI site added to the reverse primer) of OregonR crm in the pGEMT easy cloning vector. Sequences were checked for error. The 5′ half was cloned first using the restriction sites BglII and NotI in the modified PUAST-attB vector. The allele G639M842 was amplified from a Drosophila melanogaster dutch line, Tex12, differing only by the substitution responsible for the T842M replacement from OrR. The reverse mutation G639A was induced using a primer by PCR in a strategy similar to the QuickChangeR Multi Site-Directed Mutagenesis Kit (Stratagene) and checked by sequencing. The four different 3′ crm halves were cloned using the restriction sites NotI and XbaI 3′ to the 5′ half of crm. The obtained plasmids were integrated in the plattform 22A on the second chromosome using the PhiC31 transgenesis method [62]. We introduced them in an isogenic y, crm32 mutant background using balancer chromosomes and a fourth chromosome labeled by a Pw+ transgene. The third chromosome used to construct these stocks was extracted from the OregonR stock.
LacZ staining on drosophila tissues [63] and immunostaining on squashed polytene chromosomes were done as previously described [23]. Immuno-staining on whole mount salivary glands and imaginal discs were done according to standard protocols. We used monoclonal mouse anti Beta-galactosidase (Promega, Z378), and rabbit anti-PH [26]. Observation and image capture of immuno-fluorescent staining were made on an Axioplan microscope (Zeiss) with an Optronix camera and Magnafire software.
We scored abdominal pigmentation in females homozygotes for each of the four alleles in a crm32 isogenic mutant background grown at 18, 20 or 25°C. They were identical and thus not informative at 29°C. We scored four pigmentation traits on each side: width of the transversal melanin stripe in the lateral region of the sixth tergite (A6L) and in the middle of the sixth hemitergite (A6M), width of the melanic pigmentation along the anteroposterior axis in the lateral region of the seventh tergite (A7H) and length of melanic pigmentation along the dorso-ventral axis in the seventh tergite (A7L). For the restriction of the sex comb, because wild type flies do not show ectopic sex comb, we analyzed how crm natural alleles modify the ectopic sex comb induced by mutations in particular genes interacting with crm. We used heterozygosity for a loss of function mutation at the bric-à-brac locus (babAR07) to induce a distal sex comb phenotype as it was shown previously to interact genetically particularly strongly with crm [23]. In a distinct phenotypic test, we used a second chromosome carrying mutant alleles of Polycomblike (Pcl11) and Additionnal sex comb (AsxXF23) to induce a strong ectopic sex comb phenotype on the second and third leg. We scored the number of sex comb teeth on the third leg.
For the two sex comb traits, we crossed, at particular temperatures, females homozygote for particular crm alleles (on the second chromosome) in a crm32 mutant background with males from freshly isogenized stocks carrying the tester mutation over a balancer chromosome (babAR07/TM6b or AsxXF23, Pcl11/CyO). We scored sex comb phenotypes in the male progeny (carrying the crm32 allele on the X chromosome, heterozygote for the rescue allele inserted on the second chromosome and heterozygote for the tester mutations). Bristles partially transformed into sex comb teeth (identified by the presence of a large socket) were scored as sex comb teeth in the babAR07 test. Ectopic sex comb teeth were always unambiguously identifiable in the Pcl11AsxXF23 test.
For the wing vein phenotype caused by the ciD allele, we crossed, at 18 and 25°C, females homozygote for particular crm alleles (on the second chromosome) in a crm32 mutant background with males from a freshly isogenized ciD/eyD. We scored the wing vein phenotype in the male progeny (carrying the crm32 allele on the X chromosome, heterozygote for the rescue allele inserted on the second chromosome and heterozygote for ciD). We did not detect any vein gaps in flies grown at 18°C, which shows that the ciD phenotype is extremely temperature sensitive as previously described [39]. We therefore used only the flies grown at 25°C for statistical analyzes. We calculated the ratio between the length of the intact portion of the fourth vein distally to the posterior crossvein and the length of the third vein portion distal to the anterior crossvein (Figure 5D). Vein lengths were measured using the “Measure” function and the “line” tool of the ImageJ software (http://rsb.info.nih.gov/ij).
Statistical analyses were performed with SPSS 17. Phenotypes were scored on both sides and averaged between left and right side for each individual. Means and standard deviation are given in Figure S5. Plotting of the data showed that the four pigmentation traits were strongly correlated (Figure S6) so we extracted principal components. The first component captured more than 82% of the total variance and had an eigen value of 3.311. The second component captured 9.9% of the total variance (eigen value 0.396). We therefore used only these two components. We used ANOVA when the residuals did not deviate significantly from normality (Kolmogorov-Smirnov, p>0.05). The phenotypic data were analyzed using two way ANOVA (genotype, temperature) and three way ANOVA (Temperature, A639G, T842M) except the distal sex comb data which did not respect the assumptions of parametric test. For this data set we used non parametric ANOVA on ranked data (Scheirer-Ray-Hare extension of the Kruskal-Wallis test) and calculated p values using SS/MS total and degrees of freedom [64]. One way ANOVA (genotype) and two-way ANOVA (A639G, T842M) were used to analyzed ciD data. In order to evaluate the effect size we calculated classical eta squared from the type III Sum of Squares provided in the ANOVA output by SPSS for abdominal pigmentation and posterior sex combs as SSfactor/SScorrected total. Tukey post hoc tests were performed to identify the differences between genotypes for pigmentation and posterior sex combs. For the distal sex comb phenotype we use Tamhane post hoc test, which does not assume equality of variance. Statistical analysis can be found in Figure S7.
|
10.1371/journal.pntd.0006635 | Human plague associated with Tibetan sheep originates in marmots | The Qinghai-Tibet plateau is a natural plague focus and is the largest such focus in China. In this area, while Marmota himalayana is the primary host, a total of 18 human plague outbreaks associated with Tibetan sheep (78 cases with 47 deaths) have been reported on the Qinghai-Tibet plateau since 1956. All of the index infectious cases had an exposure history of slaughtering or skinning diseased or dead Tibetan sheep. In this study, we sequenced and compared 38 strains of Yersinia pestis isolated from different hosts, including humans, Tibetan sheep, and M. himalayana. Phylogenetic relationships were reconstructed based on genome-wide single-nucleotide polymorphisms identified from our isolates and reference strains. The phylogenetic relationships illustrated in our study, together with the finding that the Tibetan sheep plague clearly lagged behind the M. himalayana plague, and a previous study that identified the Tibetan sheep as a plague reservoir with high susceptibility and moderate sensitivity, indicated that the human plague was transmitted from Tibetan sheep, while the Tibetan sheep plague originated from marmots. Tibetan sheep may encounter this infection by contact with dead rodents or through being bitten by fleas originating from M. himalayana during local epizootics.
| Plague is mainly a disease of wild rodents, and their parasitic fleas are considered the transmitting vectors. However, human plague originating from Ovis aries (Tibetan sheep) is found in the Qinghai-Tibet plateau in China, where Marmota. himalayana is the primary plague host. Tibetan sheep-related human plague infection is always associated with slaughtering or skinning diseased or dead Tibetan sheep. The plague in Tibetan sheep clearly lags that in M. himalayana. In this study, we performed a genome-wide single nucleotide polymorphism analysis of Tibetan sheep-related plague events, including pathogens isolated from humans, Tibetan sheep, and marmots. Through genomic analysis, together with the epidemiological connections, we confirmed that human plague came from Tibetan sheep, and the Tibetan sheep plague originated from marmots. Tibetan sheep account for about 1/3 of the total number of sheep in China. Tibetan sheep and goats are important domestic livestock on the Qinghai-Tibet plateau. Therefore, the hazards of Tibetan sheep plague should not be underestimated.
| Plague is an acute infectious disease caused by Yersinia pestis that killed millions of people in Europe in the 14th century and tens of thousands in China in the 19th century [1]. Plague is mainly a disease of wild rodents, and their parasitic fleas are considered the transmitting vectors. So far, four subspecies of Y. pestis have been recognized on the basis of their biochemical properties: Y. pestis antiqua, mediaevalis, orientalis, and pestoides (microtus) [2,3]. To date, at least 12 plague foci covering >1.4 million km2 have been identified in China [4]; the largest focus is the Marmota himalayana focus on the Qinghai-Tibet plateau in northwestern China. The overwhelming majority of Y. pestis pathogens on the Qinghai-Tibet plateau are biovar antiqua, with the exception of biovar microtus (qinghaiensis) in the Microtus fuscus focus, which is located in Chengduo county in Qinghai Province and in Shiqu county in Sichuan Province [4].
The Qinghai-Tibet plateau is the highest risk area for human plague in China and M. himalayana is the primary host in this area. The pathogen Y. pestis (biovar antiqua) in the Qinghai-Tibet plateau M. himalayana natural plague focus frequently causes pneumonic and septicemic plague with high mortality. Other rodents (Allactaga sibirica, Mus musculus, Cricetulus migratorius, Microtus oeconomus, and Ochotona daurica), some wild animals (foxes, lynxes, and badgers), and domestic animals (sheep, cats, and dogs) have been found to be infected by Y. pestis [5]. Human plague originating from Ovis aries (Tibetan sheep) was first reported in 1956 in Qinghai Province [5], though no bacterial evidence was obtained at that time. Tibetan sheep account for ~1/3 of the total number of sheep in China [6]. And the distribution areas of Tibetan sheep plague broadly overlap with the habitat of marmots in the Qinghai-Tibet plateau M. himalayana plague focus [6,7]. In August 1975, a patient suffered from plague after butchering a dead Tibetan sheep in Yushu Prefecture, Qinghai Province. The meat of the sheep was eaten by 10 people; two individuals suffered intestinal plague that then developed into pneumonic plague, and one died [5]. Three Y. pestis strains were isolated from the dead individual, Tibetan sheep, and Capra aegagrus hircus (Tibetan goat). This incident was the first time that human plague associated with Tibetan sheep or Tibetan goats was confirmed with bacteriological evidence in China [5]. In this study, we report human plague cases associated with Tibetan sheep on the Qinghai-Tibet plateau since the 1950s. Meanwhile, to further determine the ecological function of Tibetan sheep in Y. pestis endemic epidemics, we performed a genome-wide single nucleotide polymorphism (SNP) analysis of Tibetan sheep-related plague events, including pathogens isolated from humans, Tibetan sheep, and marmots. The genome-wide SNP analysis confirmed that the human plague strains were transmitted from Tibetan sheep, while the Tibetan sheep plague strains originated from marmots.
This study was approved by the Ethics Committee of the Qinghai Institute for Endemic Disease Control and Prevention (FLW2013-001) and the Institute for Communicable Disease Control and Prevention (ACUC2013-002). All animal plague surveillance procedures were performed in accordance with the National Regulations for the Administration of Affairs Concerning Experimental Animals approved by the State Council. All procedures were in accordance with the ethical standards of the National Research Committee.
Y. pestis was isolated and identified by Gram staining, the reverse indirect hemagglutination assay, and the bacteriophage lysis test. All Y. pestis strains isolated from Tibetan sheep (15) or humans (7) associated with Tibetan sheep on the Qinghai-Tibet plateau were included (S1 Fig and S2 Table). The 18 outbreaks of human infection were designated from A to R (Fig 1 and S1 Table). The Tibetan sheep involved in human plague outbreaks based on epidemiological investigations were designated using the same alphabetic code (see S1 Table). In addition, 14 Y. pestis strains isolated from M. himalayana were selected; whenever possible, they were from the same region as the Tibetan sheep and in the same year in order to match the isolates from Tibetan sheep plague and human plague. Furthermore, two Y. pestis strains isolated from patients infected by M. himalayana in Nangqian County (2004) were also included [7]. All the strains were collected from the Qinghai Institute for Endemic Disease Control and Prevention, Xining, China. In addition, we plotted the geographical distribution of human plague, Tibetan sheep plague, and the isolates involved on a satellite map sourced from the Institute of Geographical Sciences and Natural Resources Research, Chinese Academy of Sciences, and we have received permission to publish it under a CC BY license from the institute.
A total of 38 Y. pestis strains isolated from Tibetan sheep or humans or M. himalayana were included in this study. Genomic DNA from each bacterium was extracted using the following method in a Biosafety Level 3 Laboratory of the Qinghai Institute for Endemic Disease Control and Prevention. Y. pestis strains were cultivated in Luria–Bertani broth at 28°C for 48 h, and the collected strains were suspended in 0.5 ml of TE buffer (10.0 mM Tris-HCl [pH 8], 1.0 mM EDTA) and incubated at 28°C for 20 min, Then 80 μl of 10% SDS was added to the preparation (10 μg in 1 ml PBS), and maintained at 65°C for 10 min. Next, 20 μl RNase (10 mg/ml) was added, and the solution incubated at 37°C for 2 h. Following the addition of 10 μl of proteinase K, the preparation was incubated at 37°C for 2 h. The DNA was extracted twice with equal volumes of phenol and once with an equal volume of chloroform. The DNA was precipitated by adding two volumes of absolute ethanol. The precipitated DNA was washed with 70% ethanol and re-suspended in TE buffer (pH 8.0).
The 38 isolates were sequenced using the Illumina HiSeq 2000 platform (Illumina, San Diego, CA). Two paired-end libraries were constructed with average insertion lengths of 500 bp and 3,000 bp. The raw data were filtered by FastQC, and then the clean data were assembled into contigs using SPAdes v3.9.1. Gene prediction was performed using Glimmer with the default parameters. The whole-genome raw SNPs were detected through pairwise comparisons of Y. pestis genomes to the reference genome of the Angola strain (NC_010159) [8] using Bowtie 2 software [9] and MUMmer [10] with the default parameters. Twenty-one completed genomes or draft genomes obtained from the NCBI database were also included in the analysis [1,11–17] (S2 Table). Then the SNPs were combined, and those of low quality (read depth <5) and those located within 5 bp on the same chromosome were removed to avoid the effect of recombination. A phylogenetic tree of Y. pestis was established based on these SNPs with the Bayesian evolutionary method in BEAST software [18] using the 38 Y. pestis genomes from our study and the 21 genome sequences of Y. pestis from GenBank and rooted with Y. pseudotuberculosis (IP32953) [1,13].
The sequencing data of the Y. pestis strains are available in GenBank under accession numbers SRP131404, and the genome sequences of 38 Y. pestis strains sequenced in our study have been deposited in GenBank with accession Nos SRR6512812 to SRR6512849.
According to the epidemic history of plague on the Qinghai-Tibet plateau and the annual national plague surveillance data in China, a total of 18 human outbreaks (events, designated A to R) associated with Tibetan sheep have occurred since 1956 (S1 Table and S1 Fig). Among these events, a total of 78 human cases associated with Tibetan sheep (cases of original infection and successive secondary generation) and 47 deaths were reported, of which 70 cases and 42 deaths occurred in Qinghai. In addition, 8 human cases (5 deaths) associated with Tibetan sheep occurred in Tibet. All index infectious cases had an exposure history of butchering or skinning diseased or dead Tibetan sheep. Massive deaths or larger numbers of infection cases mainly occurred in four events before 1975; for example in 1956, the index case (Tianjun county) suffered pneumonic plague and died after skinning a dead Tibetan sheep, and this individual infected a total of 13 individuals of whom 11 died. Eating meat from infected sheep that is not fully cooked is another cause of human plague infection, such as those in 1961 (Dulan county), 1963 (Yushu county), and 1965 (Zhaduo county), that caused 26 cases of infection due to eating the meat; only the index individual in each outbreak slaughtered or skinned a diseased or dead Tibetan sheep.
Considering the months in which Tibetan sheep plague, M. himalayana plague, and human plague events have occurred on the plateau since 1956, 14 of the 27 Tibetan sheep plague events occurred during October and November. In contrast, the peak occurrence of M. himalayana plague was during June and July and usually ended in October (National Plague Surveillance data and reference [5]). The plague in Tibetan sheep clearly lagged that in M. himalayana (Wilcoxon signed rank test, P <0.05). In addition, 9 of the 18 human plague events in which the index case(s) originated from Tibetan sheep occurred during October and November, while the peak months of human plague originating from M. himalayana were during August and September [5] (Fig 1). From 1997 to 2016, no human plague cases were caused by Tibetan sheep due to active prevention and intervention measures in Qinghai, even though Y. pestis was still isolated from local Tibetan sheep and Tibetan goats on the Qinghai-Tibet plateau.
The genomic sequences of the 38 isolates of Y. pestis were assembled de novo, producing 52 contigs and 70 scaffolds on average. The number of genes per strain ranged from 2,623 to 2,990. The phylogenetic tree of Y. pestis was established using all isolates in our study as well as 21 complete genomes or draft genome sequences from the NCBI GenBank database (S2 Table). We identified 1663 high-quality SNPs compared with the reference genomes and 216 within our isolates, with 39–63 SNPs per genome. Among these SNPs, 149 were located in 143 genes, including 41 synonymous SNPs and 108 nonsynonymous SNPs, with 1–2 in each gene, whereas the remaining 67 SNPs were located in intergenic regions. The 108 nonsynonymous sites were distributed among 106 genes.
The phylogenetic relationships we constructed (Fig 2A) were very similar to the genomic maximum parsimony tree reported previously [1]. The nomenclature of the lineages in the phylogenetic tree are according to the literature [1,19]. The pathogens associated with Tibetan sheep plague were clustered into the 1.IN2 lineage in the phylogenetic tree. These strains were comparatively closer to Y. pestis Z176003, which was isolated from M. himalayana in Naqu County, Tibet, in 1976 [11]. In addition, strains H21 and H22 (human plague isolates originating from M. himalayana in Bangqian Village, Nangqian County in 2004) were clustered in 2.ANT1.
Y. pestis isolated from Tibetan sheep or local M. himalayana all fermented glycerin and reduced nitrate to nitrite, i.e., they belonged to biovar antique, the same as human plague in this focus. Combining the epidemiological information (S1 Table and S2 Table) and the population structure based on the genome-wide SNP analysis, we divided the 36 Y. pestis in the 1.IN2 lineage (including those originating from Tibetan sheep (15) and humans (7) associated with Tibetan sheep, as well as 14 Y. pestis strains isolated from M. himalayana) into four clusters (I–IV), corresponding to eight clades (1–8) (Fig 2B). Generally, the clade-based classification agreed well with the geographical area, i.e., the strains isolated from the same area were found in the same clade (Fig 2B). In fact, where no geographic barrier existed between adjacent areas, the pathogens isolated from adjacent areas also grouped together; for example, Juela Village in Nangqian County and Xialaxiu Village in Yushu are adjacent, and the strains isolated from the two villages grouped into Clade-1; Shanglaxiu, Batang, and Guoqing Villages are neighbors, and the lineages were grouped in Clade-4. This shows that the genomic phylogenetic analysis of the Tibetan sheep-related strains have territory-specific characteristics.
In addition, in Clade-1 and Clade-4, the strains isolated in different years were grouped together. For example, the human plague cases and those corresponding to Tibetan sheep plague occurring in 1979 (in Xialaxiu Village, Yushu County) and in 1997 (in Juela Village, Nangqian County) were grouped together into Clade-1. In 1975, in Yushu County, the first human plague associated with Tibetan sheep was confirmed by bacteriological evidence. However, in 2005 in Yushu County, a larger-scale Tibetan sheep plague occurred, in which a total of 13 Tibetan sheep and 1 Tibetan goat in the same flock died. The isolates from these two events were grouped into Clade-4. This indicated that the same strains of Y. pestis successively caused Tibetan sheep or human plague outbreaks in these areas. Of course, some isolates could not be grouped together by event although the strains were isolated in the same area, such as lineages M8 and S20. In fact, finding any clear epidemiological connection between these two isolates and the rest was difficult. One possible explanation is the genomic diversity of the strains in these foci.
In Clade-1, the Y. pestis isolated from patients (H19) and Tibetan sheep (S30 and S24) in Juela Village, Nangqian County in 1997, as well as isolates from M. himalayana (M39), were grouped together. According to the epidemiological information, the diseased herdsman (H19) suffered pneumonic plague after processing a dead Tibetan sheep (S24), and isolate S30 was obtained from a sheep in the same breeding herd as the dead sheep (S24). In addition, one strain (M39) from a dead M. himalayana in a sheep grazing area had been isolated four months earlier. In fact, a raging animal plague epidemic had occurred one year previously (in 1996) in Juela Village, and a total of three strains (M18, M32, and M38) were collected in the area (National Plague Surveillance data). The above isolates were grouped together in Clade-1. Two strains (S31 and S23) isolated from Tibetan sheep in Xialaxiu Village, Yushu County (adjacent to Nangqian County) also grouped into Clade-1. Furthermore, the strain (H5) from the human plague in 1979, the corresponding Tibetan sheep strains (S6 and S7), and some strains isolated from M. himalayana also clustered into Clade-1.
As noted above, in 2005, three Y. pestis isolates (S11, G12, and S13) were identified in two Tibetan sheep and one Tibetan goat from an outbreak of Tibetan sheep plague in Guoqing Village, Yushu. The Y. pestis isolated from the dead M. himalayana found in the same village and in the same year (2005) were clustered into the same clade (Clade-4). In fact, it was in Shanglaxiu Village, Yushu County, that the first human plague case associated with Tibetan sheep was confirmed in 1975. In addition, three Y. pestis strains isolated from dead patients and Tibetan sheep and Tibetan goats in the same herd were also clustered in Clade-4. Similar clustering of Tibetan sheep and M. himalayana was also found in Zongwulong Village, Delingha County, in 1996 (Clade-5). The above findings, together with the epidemiological connections, support the conclusion that human plague came from Tibetan sheep and Tibetan sheep plague originated from marmots.
The Qinghai M. himalayana natural plague focus was first identified in 1954 as a result of the isolation of Y. pestis from a dead marmot in Qinghai Province [20]. M. himalayana is the primary plague host in this area [5]. According to plague surveillance data in Qinhai Province, a total of 468 human plague cases with 240 deaths were reported, of which 162 cases originated from M. himalayana (34.62%), 39 from Tibetan sheep (8.33%), 16 from carnivorous animals (3.42%), and 216 from successive infection of pneumonic plague by person-to-person transmission (46.15%)[5]. Tibetan sheep plague was sporadic on the Qinghai-Tibet plateau, and was restricted to areas that had M. himalayana plague epidemics. One previous investigation in Yushu Prefecture in 2005 found that the infection rate of Y. pestis in Tibetan sheep was 6.08% (64/1051) with serum titers in the range of 1:20 to 1:1280 [7]. Tibetan sheep-related human plague infection is always associated with slaughtering or skinning diseased or dead Tibetan sheep. Eating incompletely cooked meat from infected sheep or goats is another cause of human infection [5]. In previously research, the incidence of Tibetan sheep-related human plague outbreaks occurring in Qinghai Province between 1975–2009 were counted, and a total of 10 Tibetan sheep-related human plague outbreaks occurred during this period, resulting in 25 cases and 10 deaths, including bubonic plague (9), primary pneumonic plague (6), secondary pneumonic plague (6), septicemic plague (3), and intestinal plague(1) [2,3].
The even-toed ungulates (Artiodactyla), including camels and goats [21–24], donkeys, and cows [25], can be naturally infected by Y. pestis. Previous studies have shown that the sheep is a plague reservoir with high susceptibility and moderate sensitivity [26,27]. And, under natural circumstances, only individual Tibetan sheep in a flock are infected, and they do not become infected directly by sheep-to-sheep contact, even when the same flock contains a mixture of sick and healthy sheep [26]. These findings indicate that the ecological function of the Tibetan sheep in associated human plague should be considered as an intermediate or accidental host.
Another piece of supporting evidence is the fact that the occurrence of Tibetan sheep plague during the year lags behind the occurrence of M. himalayana plague. October and November were the high incidence months for the Tibetan sheep plague and human plague originated from Tibetan sheep. On the Qinghai-Tibet plateau, marmots begin hibernation from October to early November. One possible reason is that the fleas living in the caves escape after the marmots enter hibernation in October and attack other animals, such as Tibetan sheep. A minor peak for the human plague associated with Tibetan sheep occurs in June to July and presumably is caused by the massive death of marmots. Such an ecological change could also result in more fleas escaping from dead hosts and colonizing Tibetan sheep or human beings.
Several possible scenarios may explain how Tibetan sheep become infected by marmots. First, they could be infected by contact with the bodies of dead marmots. Our field observations showed that Tibetan sheep have a habit of licking the bodies of dead rodents such as marmots, which may be a means of ingesting micronutrients in the plateau environment. Previously, a study successfully induced plague infection by feeding or smearing Y. pestis in the mouths of Tibetan sheep [27]. Another possible cause is that Tibetan sheep could be infected by fleas such as Callopsylla dolabris or Oropsylla silantiewi. These are the main parasitic fleas in M. himalayana. Even though they have comparatively specific host selection, they have been found to attack human beings or other animals after the death of their preferred host [6]. Previous research has shown that C. dolabris and O. silantiewi bite and can suck the blood of Tibetan sheep in the laboratory, and the sheep can become infected and die after being challenged for 10 days [27]. The above evidence shows that fleas play an important role in Y. pestis transmission from marmots to Tibetan sheep.
Through genomic analysis, we confirmed that human plague came from Tibetan sheep, and Tibetan sheep plague originated from marmots. To the best of our knowledge, natural infection of sheep with Y. pestis is rare elsewhere in the world. The Tibetan sheep plague epizootic has some novel features, such as a complex transmission route, an extended epizootic period, and the possibility of transmission across long distances. Therefore, the hazards of Tibetan sheep plague should not be underestimated.
|
10.1371/journal.ppat.1006446 | Alpha-defensin-dependent enhancement of enteric viral infection | The small intestinal epithelium produces numerous antimicrobial peptides and proteins, including abundant enteric α-defensins. Although they most commonly function as potent antivirals in cell culture, enteric α-defensins have also been shown to enhance some viral infections in vitro. Efforts to determine the physiologic relevance of enhanced infection have been limited by the absence of a suitable cell culture system. To address this issue, here we use primary stem cell-derived small intestinal enteroids to examine the impact of naturally secreted α-defensins on infection by the enteric mouse pathogen, mouse adenovirus 2 (MAdV-2). MAdV-2 infection was increased when enteroids were inoculated across an α-defensin gradient in a manner that mimics oral infection but not when α-defensin levels were absent or bypassed through other routes of inoculation. This increased infection was a result of receptor-independent binding of virus to the cell surface. The enteroid experiments accurately predicted increased MAdV-2 shedding in the feces of wild type mice compared to mice lacking functional α-defensins. Thus, our studies have shown that viral infection enhanced by enteric α-defensins may reflect the evolution of some viruses to utilize these host proteins to promote their own infection.
| Enteric α-defensins are an ancient form of host defense against pathogens, but until recently there was no robust in vitro system available to study their functions upon secretion from the cells that produce them naturally in vivo. Here, using small intestinal enteroids as a source of naturally secreted α-defensins, we show that enteric viral infection is increased in the presence of these peptides. We then show that α-defensins also enhance infection in vivo, as predicted by the results of the enteroid model. This study points to a new role for enteric α-defensins in promoting viral infection and has implications for the ability of these peptides to dictate viral tropism in the gastrointestinal tract.
| Enteric α-defensins are effector peptides of the innate immune system that are abundantly expressed in both the human and mouse small intestine. Produced exclusively by specialized epithelial cells called Paneth cells, enteric α-defensins are broadly antimicrobial against bacterial and viral infections [1–3]. In humans, the enteric α-defensin repertoire consists of human defensin 5 (HD5) and HD6, while mice have undergone an expansion of their enteric α-defensin locus and encode upwards of 20 enteric α-defensins, termed cryptdins.
We have observed viral species-specific effects of purified α-defensins on human adenovirus (HAdV) infection in cell culture. For most species of HAdV, infection is potently neutralized, while infection by serotypes of HAdV-D and F is either resistant to enteric α-defensins or moderately enhanced [4]. Which of these phenotypes occurs in the presence of naturally secreted α-defensins has not been studied. To address this issue, we took advantage of our previously described small intestinal enteroid infection model [5].
Enteroids are an adult intestinal stem cell-derived primary culture system that models the intestinal epithelium by forming self-organizing “mini-guts” containing all of the major epithelial cell types of the small bowel including enterocytes, goblet cells, Paneth cells, and enteroendocrine cells [6]. We and others have shown that Paneth cell contents, including α-defensins and lysozyme, are constantly released and accumulate in the lumen of small intestinal enteroids to bactericidal levels [5, 7]. An important control is provided by enteroids from Mmp7-/- mice, which lack functional α-defensins due to the absence of pro-defensin processing by matrix metalloproteinase 7 (MMP7) [5, 8]. Thus, comparison of viral infection in wild type and Mmp7-/- enteroids affords an opportunity to examine the specific effects of naturally secreted α-defensins.
Since HAdVs do not replicate in mouse cells, we examined infection by two natural pathogens of mice, mouse adenovirus 1 (MAdV-1) and MAdV-2, in mouse enteroids. MAdV-1 infects macrophages and endothelial cells in vivo and causes acute encephalitis in infected mice, while MAdV-2 is naturally tropic for the small intestine and causes no overt disease [9]. Upon infection of mouse small intestinal enteroids, MAdV-2 but not MAdV-1 was able to replicate, consistent with their in vivo tropisms. Strikingly, more cells were infected and more MAdV-2 was produced from wild type enteroids compared to enteroids from Mmp7-/- mice, demonstrating that naturally secreted α-defensins can enhance viral infection. Furthermore, increased fecal shedding was observed from wild type mice compared to Mmp7-/- mice after oral infection with MAdV-2. Thus, the results from enteroid culture experiments predicted the phenotype of the infected mice. Finally, we show that purified mouse enteric α-defensins enhanced entry and infection by MAdV-2 in a receptor-independent manner. Therefore, our studies demonstrate that enhanced infection due to α-defensins, which was previously only observed in traditional cell culture, also occurs under physiologic conditions, leads to increased infection and viral shedding in vivo, and establishes α-defensins as a previously unknown host factor that contributes to viral tropism and promotes enteric viral infection.
To use enteroids to study the effect of α-defensins on adenoviral infection, we first determined whether or not MAdV-1 and -2 are able to replicate in these cultures. Enteroids from C57BL/6 mice were dissociated with a needle and syringe to expose both apical and basolateral surfaces to infection with MAdV-1 or MAdV-2. The mixture was then re-embedded in Matrigel extracellular matrix and cultured over 8 d. Virus released into the supernatant was titered on CMT-93 cells to quantify viral replication. Small intestinal enteroids supported the growth of MAdV-2, as evidenced by a >3-log increase in levels of progeny MAdV-2, which appeared to plateau around day 3 post-infection (Fig 1A). Although detectable levels of MAdV-1 were observed over 8 d in culture, no increase in infection was observed over this timeframe, leading to the conclusion that MAdV-1 was unable to replicate in small intestinal enteroids (Fig 1A). MAdV-1 was also unable to replicate in Mmp7-/- enteroids. These results were somewhat surprising, since MAdV-1 reached higher titers than MAdV-2 in the mouse rectal cell line CMT-93 (Fig 1B). Due to the absence of MAdV-1 replication in enteroids, we used MAdV-2 for all subsequent experiments.
To visualize and quantify infection without immunostaining, we created a replication competent MAdV-2 expressing enhanced green fluorescent protein fused to the minor capsid protein IX (MAdV-2.IXeGFP). The design of this virus is similar to a previously described MAdV-1.IXeGFP [10]. MAdV-2 and MAdV-2.IXeGFP grow with similar kinetics in CMT-93 cells, although MAdV-2.IXeGFP does not replicate to the same levels as MAdV-2 (Fig 1B).
To determine if naturally secreted α-defensins impact viral infection, we compared MAdV-2.IXeGFP infection of wild type enteroids, which secrete bactericidal concentrations of α-defensins, to that of Mmp7-/- enteroids, which lack functional α-defensins [5]. We utilized microinjection to model natural apical infection via the enteroid lumen. In addition, enteroids were treated for 30 min prior to infection with the cholinergic agent carbamylcholine chloride (CCh), which induces Paneth cell degranulation [5], thereby maximizing α-defensin concentrations in the enteroid lumen. We infected parallel cultures, and a sample from each genotype was collected daily. Virus production was measured by titering the combined supernatant and lysate on CMT-93 cells. We observed equal progeny production on day 1; however, greater than 3-fold more virus was produced in wild type enteroids than in Mmp7-/- enteroids on day 2 (Fig 2A). This increased to >4-fold more virus on day 3. Thus, rather than blocking infection, the presence of α-defensins enhanced the production of MAdV-2 in wild type enteroids.
Greater virus production in wild type enteroids could be explained by a substantial increase in initial infection or by a small cumulative effect of α-defensins on each round of replication and infection of new cells. To measure the number of initially infected cells for each genotype, we counted GFP positive cells in individual enteroids 24 h after microinjection. Given the ~20 h life cycle of adenovirus, GFP positive cells at this time point are exclusively from the initial infection. For this experiment, MAdV-2.IXeGFP was mixed with Texas Red-conjugated dextran in order to mark microinjected enteroids, and our analysis was restricted to enteroids that were still stained with Texas Red after 24 h (Fig 2C). Overall, approximately1.6-fold as many infected cells were observed in wild type enteroids compared to Mmp7-/- enteroids (Fig 2C and 2D). To confirm that this effect was not due to the presence of the protein IX-GFP fusion protein, this experiment was repeated using a replicating MAdV-2 expressing firefly luciferase (MAdV-2.IX2AFFluc). The presence of a 2A sequence from porcine teschovirus-1 between the protein IX and firefly luciferase coding regions prevents the formation of a fusion protein (S1 Fig) [11]. Thus, only wild type protein IX is made in infected cells and incorporated into virus. We obtained a nearly identical 1.6-fold increase in infection of wild type compared to Mmp7-/- enteroids (Fig 2E). Together, this data suggests that the major effect of enhancement by α-defensins in the wild type enteroids is due to an initial increase in infected cells.
In these cultures, deletion of MMP7 is not limited to Paneth cells. To rule out an effect of MMP7 other than α-defensin activation, we used two approaches. First, we compared MAdV-2.IXeGFP infection by two additional routes: basolateral, which bypasses the luminal α-defensins, and enteroid disruption and washing, which disperses the α-defensins. Under both of these conditions, we observed no difference in infection levels between genotypes (Fig 2A), confirming that small intestinal cells of both genotypes are equally able to support MAdV-2 infection and replication. Assuming that MAdV-2 progeny release from polarized cells is not directional (e.g., apical), these results also support our previous conclusion that α-defensin-dependent enhancement primarily acts on initial infection, since α-defensins in the lumen are only initially bypassed after basolateral infection or by mixing. As a second approach, we enumerated infected cells in colonic enteroids. Like the colon, colonic enteroids lack α-defensin expression (Fig 2B) due to the absence of Paneth cells. Unlike our results in small intestinal enteroids, no difference in the number of initially infected cells was seen between wild type and Mmp7-/- colonic enteroids (Fig 2D). Moreover, the number of infected cells in colonic enteroids and Mmp7-/- small intestinal enteroids was equivalent; indicating that the increase in infected cells observed in wild type small intestinal enteroids was due exclusively to α-defensins.
To determine if α-defensin-dependent viral enhancement also occurs in vivo, wild type and Mmp7-/- mice were infected orally with 1x107 iu/mouse of MAdV-2. Since MAdV-2 causes no overt pathology, infection and replication were monitored by quantifying fecal shedding at 6, 20, and 28 h post-infection. We observed a peak of shedding at 6 h, which likely represents passage of the inoculum. Between 20 and 28 h post-infection, wild type mice shed significantly more virus than Mmp7-/- mice (Fig 3A and 3B). We then repeated this experiment over a longer time course. Wild type and Mmp7-/- mice were infected orally with 1x106 iu/mouse of MAdV-2, and fecal shedding was monitored at 2, 4, and 7 d post-infection. As in the previous experiment, wild type mice shed more virus than Mmp7-/- mice (Fig 3C and 3D). Therefore, α-defensin-mediated enhancement of MAdV-2 infection results in increased viral shedding in vivo on a scale comparable to increased virus production observed in enteroids.
To gain some insight into the mechanism of enhancement, we determined if purified α-defensins could enhance MAdV-2 infection in cell culture. We first compared MAdV-2 and MAdV-2.IXeGFP infection in the presence of cryptdin 2, one of the more abundant cryptdins in vivo [12]. Enhancement of infection for both viruses was dose-responsive (Fig 4A and 4B), although the effect of cryptdin 2 was greater for MAdV-2.IXeGFP (~2.4-fold) than wild type MAdV-2 (~1.8-fold). Importantly, the unprocessed pro-form of cryptdin 2 expressed in Mmp7-/- mice had no effect on infection of either virus.
Enhanced infection suggested a direct interaction between the virus and α-defensins. To test this, we measured aggregation as a correlate of binding [13] and found that all cryptdin 2 concentrations tested led to an increase in the average z-diameter of both viruses, consistent with defensin binding to the viruses, while pro-cryptdin 2 did not bind to either virus (Fig 4C). Although the dose-responsiveness of binding to the two viruses was similar, larger aggregates of MAdV-2.IXeGFP than MAdV-2 formed in the presence of cryptdin 2. Therefore, defensin binds to the WT capsid and enhances infection; however, the fusion of GFP to protein IX may present additional surface area for defensin binding that increases both the degree of enhancement and the size of viral aggregates. Nonetheless, fusion of GFP to protein IX is not sufficient for enhanced infection, since MAdV-1.IXeGFP infection is neutralized by cryptdin 2 [14].
Because mice express multiple enteric α-defensin paralogs, we tested the activity of additional purified cryptdins on MAdV-2.IXeGFP infection. Cryptdin 3 increased infection ~1.7-fold, while incubation with cryptdin 4 had no effect on infection (Fig 4D). Thus, not all cryptdin paralogs are equally effective. Interestingly, we previously found that cryptdin 2 but not cryptdin 4 blocked infection by MAdV-1, and we attribute the inactivity of cryptdin 4 for both viruses to a three amino acid deletion that is not found in other cryptdins [14].
We previously found that an interaction with the human enteric α-defensin HD5 increases the amount of HAdV bound to cells, despite the ability of HD5 to block endosome escape of internalized HAdV [4, 13]. Intriguingly, HD5-enhanced binding of HAdV to cells was independent of HAdV binding to its primary receptor, which could be blocked by outcompeting the virus with a molar excess of its receptor-binding domain, the fiber knob (FK) [4]. We speculated that cryptdins might enhance MAdV-2 infection by increasing virus binding to the cells but failing to block a downstream step in entry. To investigate this possibility, we first purified a recombinant construct consisting of the MAdV-2 FK and part of the distal shaft (residues 517–787) and verified its ability to specifically inhibit MAdV-2 but not MAdV-1 infection (Fig 5A). We then identified a minimal concentration of FK (1.0 μM) that could maximally (~84%) block MAdV-2.IXeGFP infection of CMT-93 cells in the absence of α-defensin. When MAdV-2.IXeGFP was pre-incubated with an enhancing α-defensin (cryptdin 2) before being added to cells pretreated with 1.0 μM MAdV-2 FK, we observed the same amount of enhancement as in the absence of MAdV-2 FK (Fig 5B). Taken together, these results indicate that the presence of α-defensins allows MAdV-2 to enter cells in a receptor-independent fashion. Moreover, since the degree of enhancement was equal in the presence and absence of FK, receptor-independent binding to cells is likely the major mechanism of enhanced infection.
We have uncovered a new means to promote enteric viral infection that is mediated by an effector of the host innate immune system, α-defensins, and is distinct from previously described mechanisms involving the intestinal microbiota or bile [15–19]. Since α-defensins are highly expressed at the site of natural MAdV-2 infection, our findings suggest that the virus has evolved to hijack these host-derived peptides to increase its own infection. For both mouse and human AdVs, serotypes with primarily gastrointestinal tropism (e.g., MAdV-2 and HAdV-F) are resistant to or enhanced by enteric α-defensins, while closely related, non-enteric serotypes (e.g., MAdV-1 and HAdV-C) are neutralized by α-defensins [4]. This trend is also consistent with some of the earliest observations regarding the antiviral activity of these proteins: echovirus and reovirus are enteric viruses that were found to be unaffected by α-defensins, establishing a dogma in the field regarding a general inability of α-defensins to neutralize non-enveloped viruses that was only recently dispelled [20, 21]. Thus, α-defensins are primarily antiviral, but we propose that certain non-enveloped viruses have evolved to become selectively resistant to and even enhanced by enteric α-defensins due to pressure imposed during fecal-oral transmission.
The closest equivalents to α-defensin enhancement of viral infection are antibody-dependent enhancement (ADE) of viruses (e.g., dengue) and increased uptake of viruses such as HIV-1 or West Nile Virus (WNV) that are opsonized with complement or lectins [22–25]. No other examples of enhancement of non-enveloped viral infection by α-defensins have been reported; however, HIV attachment and infection is increased by HD5 and HD6 [26, 27]. Although other innate immune effectors such as cytokines can increase viral infection, including adenovirus infection, their effects on infectivity have been indirect by modulating receptor distribution [28]. In contrast, ADE is mediated by an adaptive immune effector binding directly to the viral capsid, allowing viral entry through Fc or complement receptors [22]. It has been proposed that the divergence of dengue serotypes is driven by ADE [29, 30]. Similarly, the selective enhancement of enteric AdVs by enteric α-defensins suggests that this is an evolutionary adaptation of the virus to the intestinal milieu.
Unlike ADE, where the mechanism of enhancement is well understood, the precise mechanism of α-defensin enhancement remains to be determined. Since α-defensins are highly cationic, MAdV-2 enhancement may simply be due to charge neutralization allowing more virus to bind to the cell in the presence of α-defensins, similar to the effects of polybrene or other cationic compounds [2]. Indeed, we and others have documented increased cell binding due to α-defensins even for viruses that are subsequently neutralized [4, 13, 31]. Another possibility is that α-defensins bound to MAdV-2 facilitate an interaction with an alternate, defensin-specific cell surface receptor. α-defensins overcome the effects of a soluble competitor for primary receptor binding, suggesting receptor-independent entry of MAdV-2. This concept is difficult to test directly, as the cellular receptors for MAdV-2 and enteric α-defensins are unknown [32–35]. In ADE, interaction of the virus-antibody complex leads to Fcγ receptor- or complement-mediated entry into cells independent of dengue receptors, thereby expanding the range of cells susceptible to infection [22]. Similarly, a mosquito C-type lectin has been shown to enhance WNV infection of mosquitoes through interactions with a mosquito homolog of CD45 [25]. And, enhancement of HIV-1 infection by human α-defensins HD5 and HD6 is also independent of CD4 and co-receptors [26]. For HAdVs in particular, an analogous mechanism governs liver and spleen cell tropism via blood coagulation factors that bind to hexon and bridge interactions with heparin sulfate proteoglycans on the cell surface [36]. Alternatively, some mouse α-defensins can permeabilize eukaryotic membranes [37], perhaps leading to increased infection by forming pores at the cell surface or facilitating disruption of the endosomal membrane, resulting in more virus reaching the nucleus. Since α-defensins bind to both enhanced and neutralized HAdV and MAdV serotypes and therefore enter the endosomal pathway [4, 13, 14, 38, 39], they may remain bound to and prevent uncoating of neutralized AdVs but dissociate from enhanced AdVs to interact with the endosomal membrane. Although our data provide the strongest support for α-defensins impacting cell binding, enhanced infection could be explained by facilitating internalization. Complement opsonized HIV-1 uptake by dendritic cells is enhanced in this manner [24]. We found no alteration in internalization kinetics in our studies of neutralization of HAdVs by α-defensins [38]; however, this has not yet been investigated for enhanced infection. Defensins could also act post-entry by promoting uncoating of enhanced viruses, perhaps by destabilizing the internalized capsid in a manner analogous to their activity against bacterial toxins [40]. Irrespective of the specific mechanism, our results suggest that the distribution of α-defensins in the gastrointestinal tract may dictate the location of enteric viral infections.
MAdV-2 infection is enhanced by α-defensins in three different systems: upon addition of purified α-defensins to two-dimensional cell culture in vitro, due to naturally secreted α-defensins in small intestinal enteroids, and in vivo. While there has been much interest in intestinal enteroids as a novel model for the human and mouse intestinal tract, there is a need to determine whether enteroids are truly predictive of in vivo effects in a manner distinct from traditional cell culture systems. We demonstrate increased viral replication specific to wild type small intestinal enteroids, which is not observed in colonic enteroids or Mmp7-/- enteroids. These results were recapitulated in vivo, since orally infected wild type mice shed more virus than Mmp7-/- mice in multiple experiments. Therefore, our data as well as other recently published papers support the idea that mouse enteroids are highly predictive of in vivo effects [41–43]. These studies also imply that results from human intestinal enteroids would be directly applicable to the human intestine. As human enteroids are a renewable source of primary cells, they represent an unprecedented opportunity for experimentation in a system that more accurately reflects human biology than cell lines or animal testing.
All mouse experiments were performed in strict accordance with the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health and following the International Guiding Principles for Biomedical Research Involving Animals. Animals were humanely euthanized by CO2 inhalation for infection studies or by bilateral thoracotomy under anesthesia to harvest tissue for enteroid culture. Protocols were approved by the Institutional Animal Care and Use Committee of the University of Washington under Protocol Number 4245–01.
Small intestinal enteroids were derived from crypt enriched ileal fractions from 6–10 week old wild type and Mmp7-/- mice on a C57BL/6NHsd background as previously described [5] and maintained in Complete Crypt Culture Medium (CCCM) [44]. Colonic enteroids were similarly derived and cultured from crypt enriched colonic fractions with the following modifications: 1) To avoid adherence of colonic enteroids during derivation, plastic ware was pretreated with 10% FBS. 2) Pelleted crypts were treated with TrypLE (Thermo Fisher Scientific) for 5 min at 37°C prior to washing and embedding in Matrigel. 3) CCCM was supplemented with murine Wnt-3A (0.1 μg/mL, Sigma-Aldrich), CHIR99021 (2.5 μM final, Stemcell Technologies), and valproic acid (1 mM, Sigma-Aldrich). Once enteroids were established, culture media was supplemented with 200 μl CCCM every 2–3 days. Enteroids were subcultured every 6–7 days [44].
Wild type MAdV-2 was a gift from Susan Compton (Yale University School of Medicine). MAdV-1 and MAdV-1.IXeGFP were gifts from Katherine Spindler (University of Michigan) [10]. Replication-competent MAdV-2.IXeGFP was patterned on MAdV-1.IXeGFP and produced using recombineering with a bacterial artificial chromosome containing the genome of MAdV-2 (pKSB2 MAdV-2, a kind gift from Silvio Hemmi, University of Zurich) [45]. MAdV-2.IX2AFFluc was similarly constructed by inserting a sequence encoding the 2A peptide from porcine teschovirus-1 (AGATNFSLLKQAGDVEENPGPAAA) [11] followed by a sequence encoding firefly luciferase immediately before the stop codon of protein IX by recombineering. All MAdVs were propagated in CMT-93 mouse rectal carcinoma cells (ATCC CCL-223) and purified as described [14]. To quantify virus production in Fig 1B, multiple wells of CMT-93 cells were infected in parallel with MAdV-2, MAdV-2.IXeGFP, or MAdV-1 at an MOI of 3. Beginning on the day of infection, cells and supernatant were collected from one well daily for 5 days. After freeze/thaw and centrifugation, lysates were serially diluted on a fresh monolayer of CMT-93 cells. Infection levels were quantified 48 h post-infection by staining fixed cells with anti-hexon antibody 8C4 and an Alexa Fluor 488-conjugated secondary antibody. Total well fluorescence was quantified with a Typhoon 9400 scanner. For each virus at each time point, the TCID50 was determined and used to calculate FFU/mL based on the Poisson distribution.
For infection by mixing in Figs 1A and 2A, enteroids were subcultured prior to infection and overlaid with 500 μL CCCM in a 24-well tissue culture dish. After 3–4 days, enteroids in Matrigel plugs from 6 wells were pooled and disrupted with a pipette tip, and the suspension was dissociated with a 25-gauge needle. Cells were then pelleted by pulse centrifugation in a mini-centrifuge and resuspended in 250 μl CCCM. MAdV-2.IXeGFP, MAdV-1, or MAdV-2 (1-2x105 infectious units) in 250 μl of CCCM was added to the cells and incubated at 4°C for 30 min to allow for virus binding. Cells were then washed twice in CCCM to remove unbound virus and resuspended in 30 μl CCCM and 120 μl Matrigel. 30 μl was plated into each of 5 wells of a 24 well plate, allowed to polymerize, and then overlaid with 500 μl CCCM.
For basolateral infections, media was removed from 2–3 day old enteroids seeded in 24-well plates and replaced with 500 μl of fresh CCCM containing MAdV-2.IXeGFP (1-2x105 infectious units).
For infection by microinjection, enteroids were cultured in Matrigel on glass coverslips. Prior to injection, cells were treated with 10 μM CCh (Sigma-Aldrich) for 30 min. Ten enteroids per coverslip were injected with 200–250 pL per enteroid of MAdV-2.IXeGFP (4x108 infectious units/mL) alone or mixed 1:1 with 80,000 MW Texas Red-conjugated dextran using the same needle for all wild type and Mmp7-/- enteroids from a single experiment. After injection, wells were washed twice and overlaid with 1 mL CCCM. For Fig 2C and 2D infected (GFP positive) cells per injected enteroid (Texas Red-conjugated dextran positive) were manually counted using an epifluorescence microscope. For Fig 2E, fifteen enteroids of each genotype were injected with 200–250 pL per enteroid of MAdV-2.IX2AFFluc from a virus stock with a concentration of 3.6x1011 genomes/mL. Enteroids were lysed 24 h post-infection, and luciferase activity was measured using the Promega Dual Luciferase Reporter assay.
For Figs 1A and 2A, samples of combined cells and supernatant were collected immediately after washing (day 0) or daily for 8 or 3 days respectively. The combined cells and supernatant underwent three rounds of freeze/thaw to release intracellular virus. After centrifugation, 50 μl of the virus-containing supernatant was applied to a fresh monolayer of CMT-93 cells. Infection levels were quantified by staining for hexon as described above. Data are the integrated density of total well fluorescence (Fig 1A) or the ratio of total well fluorescence of monolayers infected with progeny from wild type compared to Mmp7-/- enteroids (Fig 2A).
Total RNA was extracted from enteroids after 4 days in culture using RNAbee (Tel-Test, Inc.). RNA was primed with Oligo DT and reverse transcribed. qPCR was performed using the primers listed in Table 1. Defcr4 expression was calculated relative to the housekeeping gene Rpl5.
Mice were housed under specific pathogen-free conditions and ABSL-2 containment and were infected between 5 and 7 weeks of age via oral gavage with purified wild type MAdV-2 diluted in sterile PBS. For experiments in Fig 3A and 3B, mice were housed 5 per cage and segregated by genotype. Fresh fecal pellets were collected beginning immediately before infection and at 6, 20, and 28 h post-infection. For the experiments in Fig 3C and 3D, mice were individually housed beginning 3 or 4 d prior to infection and for the duration of the experiment. Fecal samples consisted of ten fecal pellets that accumulated in the cages since the previous collection. Accordingly, on the day of infection and after every fecal collection, mice were transferred to new cages with clean bedding.
Viral DNA was extracted from fecal pellets using the QIAamp DNA Stool Mini Kit (Qiagen) into a total volume of 200 μl. MAdV-2 genomes were quantified by qPCR against a standard curve of pKSB2-MAdV2 using primers M2FF2 and M2FR2 (Table 1). Reaction conditions consisted of 40 cycles of PCR with 55°C annealing temperatures. The limit of detection was defined by the number of viral copies detected in feces from uninfected mice.
Cryptdin 2 (UniProtKB: P28309, LRDLVCYCRTRGCKRRERMNGTCRKGHLMYTLCCR) and pro-cryptdin 2 (DPIQNTDEETKTEEQSGEEDQAVSVSFGDREGASLQEESLRDLVCYCRTRGCKRRERMNGTCRKGHLMYTLCCR) were obtained by oxidative refolding of partially purified linear peptides (synthesized by CPC Scientific and LifeTein, respectively) and purifying the correctly folded species by reverse-phase high-pressure liquid chromatography (RP-HPLC). Purity was determined by analytical RP-HPLC, and the mass of the disulfide-bonded peptides was verified by high mass accuracy liquid chromatography-mass spectrometry. Additional cryptdin 2 was a gift from Wuyuan Lu (University of Maryland). Cryptdin 3 (UniprotKB: P28310, LRDLVCYCRKRGCKRRERMNGTCRKGHLMYTLCCR) and Cryptdin 4 (UniprotKB: P28311, GLLCYCRKGHCKRGERVRGTCGIRFLYCCPRR) were a gift from André Ouellette (University of Southern California). All α-defensin concentrations were quantified by UV absorbance at 280 nm using calculated molar extinction coefficients.
A recombinant protein containing the MAdV-2 fiber knob and ~6.5 distal shaft repeats was generated from a bacterial expression plasmid (pET28(+)) encoding MAdV-2 residues 517–787 with an N-terminal 6× His tag (a gift from Mark J van Raaij, Spanish National Center for Biotechnology). Protein was expressed in BL21-CodonPlus (DE3)-RIPL cells (Agilent Technologies) upon induction with 0.4 mM IPTG for 4 hr at 37°C; purified using a TALON column (Clontech), as previously described [46]; and stored in 50 mM Na2HPO4/NaH2PO4, 130 mM NaCl, 10% glycerol, pH 8.0.
Dilutions of MAdV-2 or MADV-2.IXeGFP producing ~50% maximal signal 48 h post-infection of CMT-93 cells were chosen for inhibition studies. Defensin studies were performed as described with purified virus in serum-free DMEM (SFM) [14]. For Fig 5A, increasing concentrations of MAdV-2 fiber knob were mixed with virus in SFM in a total volume of 50 μl, added to cells, and incubated at 37°C for 2 h. Cells were then washed and cultured with DMEM containing 10% FBS for 48 h prior to quantification. For experiments in Fig 5B, virus was pre-incubated with a final concentration of 5 μM cryptdin 2 in a total volume of 37.5 μl on ice. In parallel, a confluent monolayer of CMT-93 cells was pre-treated at 37°C with SFM containing 1 μM MAdV-2 fiber knob or an equal concentration of BSA in a final volume of 50 μl SFM. After 45 min, 12.5 μl of SFM containing 4 μM MAdV-2 fiber knob or BSA was added to the virus and defensin mixture. The pre-treated CMT-93 monolayer was washed twice with SFM, and 50 μl of the virus, defensin, and fiber knob or BSA mixture was added to cells and incubated for 2 h at 37°C. Cells were then washed and cultured with DMEM containing 10% FBS for 48 h prior to quantification.
α-defensins were serially diluted in 10 mM Tris, 150 mM NaCl, pH 7.5 and mixed with 6.5x108 particles of wild type MAdV-2 or MAdV-2.IXeGFP in a total volume of 50 μl. Control samples of virus or α-defensin only were diluted in the same buffer. Samples were incubated for 45 min on ice and then equilibrated for 3 min at 37°C prior to analysis. The z-average particle size was obtained by cumulant analysis with a Malvern Zetasizer Nano ZS and manufacturer’s software (Malvern Instruments).
Experiments were analyzed using Prism (v. 7.0b, GraphPad). For Fig 2A, 2-way ANOVA with Tukey post-test was used to compare the mean of each infection route with the mean of every other infection route for each time point. In Fig 2D, data was analyzed using one-way ANOVA comparing the means of wild type and Mmp7-/- samples. In Fig 2E, relative light units of triplicate samples were averaged. The log-transformed averages from three independent experiments were then compared by paired, two-tailed t test. In Fig 3B and 3D, genome copy numbers were log transformed prior to determining total viral shedding by calculating the area under the curve (AUC) over the indicated time period. AUCs were compared by unpaired t test. In Fig 4A, 4B or 4C data for each virus in the presence of cryptdin 2 was compared to data in the presence of pro-cryptdin 2 at each concentration by two-way ANOVA with Sidak’s multiple comparison tests. In Figs 4D and 5B, infection for each defensin concentration or condition was compared to infection in the absence of defensin by one-way ANOVA with Dunnett’s multiple comparison test. *P<0.05, **P<0.001, ***P<0.0001
|
10.1371/journal.ppat.1007760 | Viral engagement with host receptors blocked by a novel class of tryptophan dendrimers that targets the 5-fold-axis of the enterovirus-A71 capsid | Enterovirus A71 (EV-A71) is a non-polio neurotropic enterovirus with pandemic potential. There are no antiviral agents approved to prevent or treat EV-A71 infections. We here report on the molecular mechanism by which a novel class of tryptophan dendrimers inhibits (at low nanomolar to high picomolar concentration) EV-A71 replication in vitro. A lead compound in the series (MADAL385) prevents binding and internalization of the virus but does not, unlike classical capsid binders, stabilize the particle. By means of resistance selection, reverse genetics and cryo-EM, we map the binding region of MADAL385 to the 5-fold vertex of the viral capsid and demonstrate that a single molecule binds to each vertex. By interacting with this region, MADAL385 prevents the interaction of the virus with its cellular receptors PSGL1 and heparan sulfate, thereby blocking the attachment of EV-A71 to the host cells.
| Enterovirus A71 (EV-A71) is the virus responsible for most of the severe forms of hand, foot and mouth disease (HFMD) associated with neurological involvement and mortality in young children under the age of 5. Seasonal outbreaks of HFMD -with a 2–3 years epidemic cycle- are recurring around the world, especially in the Asia-Pacific region. To date, no antiviral agent has been approved for the treatment of EV-A71 infections. Here, we report on a recently uncovered class of tryptophan dendrimers with an extraordinary antiviral activity in vitro against circulating EV-A71 clinical isolates. Mode of action studies revealed that this class of compounds targets the 5-fold vertex of EV-A71, in turn blocking receptor binding. Our finding may open an entirely novel line of research and largely aid in anti-enterovirus drug development.
| Since the first large outbreak in 1997, enterovirus A71 (EV-A71) (genus Enterovirus, family Picornaviridae) has been reported to cause 2–3 year cyclic epidemics in the Asia-Pacific region [1,2]. In the last two decades, the increasing number of EV-A71 cases and the spread of the virus across Asia have raised major concerns about its pandemic potential. The virus is primarily transmitted by the oral-fecal route [3,4]. Most EV-A71 infections are characterized by mild symptoms, with the typical signs of the hand, foot and mouth disease (HFMD): slight fever, red rashes on the palms of hand and soles of feet, and ulcers in the mouth. However, EV-A71 infections are also associated to severe neurological complications (such as encephalitis, aseptic meningitis and poliomyelitis-like syndrome) and acute pulmonary edema, which may be highly limiting and fatal particularly in children under the age of 5 years [5,6]. In 2010, a large outbreak of HFMD in China resulted in an estimated 1.7 million cases and 905 deaths [7] and an outbreak in Cambodia in 2012 resulted in the death of 54 children [8,9]. A sub-genogroup C4 EV-A71-inactivated vaccine has recently been approved in China, but worldwide coverage and long-term protection still need to be addressed [10–12]. There are no antiviral agents approved against EV-A71 nor against any other enteroviruses.
EV-A71 has been reported to bind to several cell surface receptors, including scavenger receptor B2 (SCARB2) [13,14], P-selectin glycoprotein ligand-1 (PSGL1) [14,15] and heparan sulfate (HS) glycosaminoglycan [16]. Other host factors such as cyclophilin A, annexin II, sialylated glycans, vimentin, nucleolin, fibronectin and prohibitin have also been reported to promote infection, although their importance in viral entry is still less noted [17–23]. It has been shown that EV-A71 interaction with PSGL1 on leukocytes requires the presence of sulfated tyrosine (Tyr) residues at the N-terminal region of PSGL1 [24] and depends on two highly conserved lysine residues, VP1_244K and VP1_242K, near the 5-fold vertex of the viral capsid [25]. A spatially close residue, VP1_145, is another determinant for PSGL1 binding [26]. Similarly to PSGL1, HS has also been proposed to interact near the 5-fold vertex of the viral capsid [15,16,22].
A well-known class of inhibitors of entero- and rhinovirus entry (such as pirodavir, pleconaril and vapendavir) bind into a hydrophobic pocket under the floor of the viral “canyon” formed by VP1. Drug binding prevents receptor interaction and/or increases particle stability, which in turn blocks the conformational changes required for viral uncoating [27–32].Despite their notable potency in vitro, none of these compounds reached advanced clinical trials.
Attachment of EV-A71 to host cells can also be blocked by suramin and other sulfated and sulfonated analogs, including NF449, which bind the positively charged residues clustered at the five-fold axis of the viral capsid, in turn preventing PSGL1 and HS attachment [33–35]. According to the proposed mechanism of action by Ren et al., the VP1_145 residue was found to be critical for the inhibitory profile of suramin [33]. On the other hand, amino acid changes at position VP1_244 and VP1_98 modulated the antiviral effect of NF449 [34]. These findings reveal a pivotal role of the 5-fold vertex of the viral capsid for binding to host receptors and lodging molecules able to inhibit EV-A71 replication.
Recently, we discovered a class of inhibitors with dual activity against HIV and EV-A71 [36,37]. The lead compound of this family, MADAL385, is a tetrapodal derivative with a pentaerythritol core, 4 trivalent spacer arms and 12 tryptophan (Trp) residues on the periphery [38]. Because the tryptophan dendrimers are linked to the central scaffold through their amino groups, their carboxylates are free and exposed to the solvent. Our earlier biological studies demonstrated that MADAL derivatives inhibit HIV entry into its target cell by interaction with glycoproteins gp120 and gp41 of the viral surface [36]. For EV-A71, we demonstrated that MADAL derivatives exhibit low micromolar activity against the lab-adapted strain BrCr and low-nanomolar/high-picomolar activity against a large panel of EV-A71 clinical isolates from different genogroups and various geographic origins [38]. Structure-activity relationship (SAR) studies on the periphery and central scaffold of MADAL highlighted the importance of free carboxylic groups for optimal antiviral activity, those carried by Trp or Tyr residues.
In the present work, we elaborate on the precise molecular mechanism of action of the lead compound MADAL385 by means of in vitro biological assays, cryo-EM analysis and molecular modeling. Our data support a model of activity by which a single MADAL385 molecule binds on each of the 5-fold vertices of the EV-A71 capsid, thereby blocking the engagement of the virus with host receptors PSGL1 and/or HS.
From the series of Trp dendrimers endowed with high in vitro potency against EV-A71 replication, MADAL385 (Fig 1A) was selected for further mechanistic studies (EC50 against the lab-adapted BrCr strain: 0.29 ± 0.07 μM, CC50: 30.0 ± 2.5 μM) [38]. In a cytopathic effect (CPE)-based reduction assay, MADAL385 is ~1,800 to ~20,000-fold more effective against a representative EV-A71 clinical isolate (EV-A71_11316, geno-group B2) than two earlier reported EV-A71 inhibitors, namely pirodavir [31] and suramin [39] (S1 Fig). MADAL385 inhibits, in a dose-dependent manner, (i) the formation of infectious EV-A71 BrCr particles (Fig 1B) and (ii) the replication of a mCherry-reporter EV-A71 BrCr (Fig 1C). MADAL385 inhibits the early stages of EV-A71 replication [38] and to precisely determine which step of viral entry is targeted, binding and internalization assays were performed with pirodavir and suramin as reference. MADAL385 (10 μM) markedly reduces -akin to suramin- the binding of EV-A71 to the host cells and prevents as well the internalization of the virus (Fig 1D). In contrast, pirodavir does not show any inhibitory activity in either the binding or the internalization assays, which is in line with the reported mechanism of action (i.e. inhibition of viral uncoating). By tightly binding in the hydrophobic pocket, “classical’ capsid binders such as pirodavir and pleconaril increase the rigidity of the viral particle and, as a consequence, its resistance to heat inactivation. MADAL385, in marked contrast to the capsid-binder pirodavir protects the virus only slightly against heat inactivation (Fig 1E). Taken together, these results suggest that MADAL385 blocks viral attachment by binding to the virus particle in a region outside the hydrophobic pocket.
To gain first insights into the putative binding site of MADAL385, two independent, MADAL385-resistant strains were generated via a stepwise clonal resistance selection procedure (Fig 2A). Full genome sequence analysis revealed two amino acid replacements in the capsid protein VP1 of the BrCr strain: S184T and P246S (Table 1). To confirm that S184T and P246S are responsible for resistance to MADAL385, the single or double mutant(s) were engineered by site-directed mutagenesis in the EV-A71 BrCr strain infectious clone. All mutant variants show replication kinetics comparable to that of the wild-type virus (S2A Fig). The susceptibility of the single S184T and P246S mutants to MADAL385 is decreased by 7- and 17-fold, respectively; the double mutant (S184T_P246S) confers the highest (32-fold) resistance to the compound (Table 2). Next, to address the role of viral VP1 in antiviral susceptibility, we engineered a recombinant virus by swapping the VP1 of the BrCr strain with that of the EV-A71 11316 clinical isolate, against which MADAL385 is ≥1000-fold more potent (BrCr strain EC50: 0.28 ± 0.01μM versus 11316 strain EC50: 0.21 ± 0.03nM) (Fig 2B). The susceptibility of the recombinant EV-A71 BrCr_VP1 (11316) strain to MADAL385 increases dramatically to the level observed with the clinical isolate (EC50: 0.12 ± 0.02nM) (Fig 2C). A comparison of sequences of BrCr and selected circulating EV-A71 clinical isolates (including EV-A71_11316) revealed 8 amino acid differences in the viral VP1 (S2B Fig). These residues were individually engineered into the BrCr infectious clone. Only R148P and L241S mutants show an increased sensitivity to the drug, without however fully restoring the susceptibility of the EV-A71_11316 strain (Fig 2D). Altogether, these data further point towards the VP1 as the molecular target of MADAL385. Notably, most of the sensitivity and resistance residues are located in the proximity of the 5-fold vertex of the EV-A71 capsid, in a region known to be involved in HS and PSGL1 receptor binding [25,40] (Fig 2E). As expected, both the MADAL385-resistant strain and the more susceptible recombinant strain EV-A71 BrCr_VP1(11316) were shown to retain wild-type sensitivity to the capsid binders pirodavir and vapendavir. However, to our surprise, no cross-resistance with suramin was observed, despite the fact that this molecule was earlier reported to interact with the positively charged region surrounding the 5-fold axis of the capsid [33]. In addition, when combined, MADAL385 and suramin result in a strong synergistic in vitro antiviral effect with mean volume of 771 μM2% (S2C Fig). According to MacSynergy method [41], values over 100 μM2% indicate strong and biologically relevant synergy. The lack of cross-resistance and the in vitro synergistic effect thus suggest a different mode of antiviral action of MADAL385 and suramin.
To define the binding area of MADAL385 and to understand the possible mechanism underlying resistance and susceptibility to the compound, cryo-EM single particle analysis of MADAL385 bound to the EV-A71_11316 strain was performed. Purified EV-A71_11316 particles were vitrified before and after incubation with MADAL385 (cryo-EM 2D class average images S3 Fig). The corresponding three-dimensional (3D) maps were reconstructed at 3.3 and 3.6 Å resolutions, respectively, by applying icosahedral-symmetry averaging and using the images from a Falcon-3 direct electron detector (S1 Table). The atomic models of free and drug-bound EV-A71_11316 (Fig 3A) superimpose with an overall RMSD value of 0.292 Å (Fig 3B), indicating that the incubation with MADAL did not cause any significant conformational change on the viral capsid proteins. In the two atomic models, the pocket factor lipid (sphingosine) has slightly different conformations, but the corresponding cryo-EM densities have similar intensities (S4 Fig), suggesting that the drug binding likely induces subtle conformational changes around the pocket region but does not initiate pocket factor release. When the two 3D reconstructions were compared, strong extra densities were identified on the 5-fold vertices of the EV-A71_11316-MADAL385 complex (Fig 3C), indicating that MADAL385 binds on the 5-fold vertex. The drug density fills the pore on the 5-fold symmetry axis and is connected to the capsid density. Also, the density intensity is comparable to that of the capsid shell (Fig 3D). Thus, MADAL385 binds on the 5-fold vertex and full saturation of the binding sites is achieved by the incubation. In particular, the intensity of MADAL385 density was highest at the symmetry axis, suggesting that the bound drug molecule occupies the very centric area of the 5-fold vertex. Due to steric hindrance and electrostatic repulsion, these observations suggest that only one molecule of MADAL385 binds on each vertex (Fig 3D).
The location of the two MADAL385-resistance mutations and the eight sensitivity mutations were mapped on the atomic model of EV-A71 (Figs 2E and 3E). Among those, the sensitivity variant VP1_148P, is located on the surface of the channel at the 5-fold axis (Figs 2E and 3F) and is found within the electron density envelope connected to MADAL385 density. The presence of a positively charged arginine residue in the BrCr strain (instead of a proline) may provide an explanation for the reduced drug sensitivity of this lab strain. Two other residues, VP1_244K and 245Y, with their corresponding side-chains extended towards MADAL385, are also found within the density envelope at the interface between capsid and drug (Fig 3G). The two resistance mutations, VP1_184T and 246S, are located closely to the drug density but are not found within the density envelope (Fig 3F and 3G), therefore interactions between MADAL385 and those residues cannot be verified or excluded based on the cryo-EM result. The rest of the other sensitivity mutations are distal from the cryo-EM densities corresponding to MADAL385, suggesting origins for the susceptible phenotypes (such as adaptive mutations) other than the physical binding of the drug. Further interpretation was limited because of the icosahedral symmetry averaging applied during the cryo-EM reconstruction on the asymmetrically bound drug molecules.
To complement the cryo-EM analysis, molecular dynamics (MD) simulations in explicit water were performed on both the free MADAL385 (for conformational sampling) and its complex with the VP1 pentamer (for assessing pose stability and characterization of intermolecular interactions). In the case of the Cα-restrained complex, the pentaerythritol core of the tetrapodal MADAL385 was initially lodged in the external pore region corresponding to the highest electron density while three of its ‘legs’ projected into the inter-subunit crevices and the remaining one occupied the outer part of the pore (S5 Fig). In this location at the 5-fold axis, the Trp carboxylates and indole rings of MADAL385 establish interactions with the positively charged residues and the hydrophobic cavities. This binding mode is reminiscent of the binding to sulfated Tyr residues (S1 Movie) of PSGL1 and HS. To identify the VP1 residues at the 5-fold axis that contribute predominantly to MADAL385 binding, van der Waals and solvent-screened electrostatic interactions, together with the cost of desolvation upon complex formation, were calculated [42]. In line with the cryo-EM analysis, VP1_244K, 245Y, 145Q, and 148P are the residues with which MADAL385 establish the strongest favorable interactions (S6 Fig).
HS and PSGL1 have been proposed to play critical roles in the early steps of EV-A71 infection by interacting with the positive charges at the 5-fold vertex of the EV-A71 capsid. In addition, our structural analysis suggests that the binding site of MADAL385 overlaps with the binding sites of both receptors. We hence hypothesized that MADAL385 exerts its antiviral activity by blocking virus attachment to PSGL1 and/or HS. The experimental setup is depicted in Fig 4A. Binding-inhibition assays were performed with heparin or PSGL1-coupled beads in either the presence of MADAL385 or reference compounds and the extent of inhibition was quantified by RT-qPCR or Western-blot. First, the ability to bind soluble heparin was assessed by neutralization assay for the BrCr, MADAL385-resistant and sensitive EV-A71 strains. Similar sensitivity is noted across the different viruses (Fig 4B). Next, HS-binding inhibition assay revealed that binding affinity between heparin and EV-A71 BrCr is gradually lost with increasing concentration of MADAL385 (Fig 4C). In contrast, MADAL385 does not inhibit heparin binding of the resistant strain (EV-A71 VP1_S184T_P245S), even at the highest concentration tested (Fig 4D). Interestingly, and in line with the high susceptibility to the antiviral action of MADAL385, binding of the recombinant EV-A71 BrCr_VP1(11316) strain to heparin is inhibited by MADAL385 more efficiently as compared to the BrCr strain, with more than 50% binding affinity lost at low micromolar concentrations (Fig 4E). As expected, the pocket binder pirodavir does not significantly affect binding of EV-A71 to heparin, whereas suramin completely blocks heparin binding of all viruses (Fig 4C–4E). Next, the effect of MADAL385 on the interaction of PSGL1 with EV-A71 was studied. For this purpose, PSGL1-Fc was over-expressed in HEK293T cells (S7 Fig) and concentrated on protein G beads. MADAL385, akin to suramin, blocks binding of the EVA71-PB strain to PSGL1 in a concentration-dependent manner. The capsid binder pirodavir has, as expected, no effect on this binding event (Fig 4F). To determine whether MADAL385 affects attachment to PSGL1 and HS during infection, we employed human SCARB2- or PSGL1-overexpressing L929 cells. EV-A71_812 strain, a PSGL1 binding strain sensitive to MADAL385 (EC50: 1.78 ± 0.81nM) was used to study binding to the cell receptors in presence of MADAL385 (0.1 μM). To account for (and exclude) the contribution of HS, L929-SCARB2 and PSGL1 cells were treated in parallel with sodium chlorate (NaClO3, 50mM), an inhibitor of cell-surface sulfation. MADAL385 greatly reduced the binding of EV-A71 to both L929-SCARB2 and PSGL1 cells; however, in NaClO3-treated cells, MADAL385 effectively reduced binding of EV-A71 to L929- PSGL1 but not to L929-SCARB2 cells (Fig 5A). Quantification of the infectious content of EV-A71_812 in L929-SCARB2 and PSGL1 cells treated with MADAL385 in presence of NaClO3 recapitulate the data of the binding assay; i.e. MADAL385 is only effective against EV-A71 in L929- PSGL1 cells (Fig 5B). These data support the hypothesis that MADAL385 blocks EV-A71 infection by preventing viral attachment to either HS, PSGL1 or both.
We report here on the mechanism of action of MADAL385, the lead compound of a novel class of tryptophan dendrimers with exquisitely potent in vitro antiviral activity against EV-A71. Cryo-EM studies revealed that the highly conserved lysine residue at position 244 of VP1 (VP1_244K), near the icosahedral 5-fold vertex, is closely connected to the density of MADAL385. This residue also plays a key role in the interaction of EV-A71 with PSGL1 and HS. Both receptors are sulfated molecules (i.e. endowed with a negative charge at physiological pH) whose interaction with the positively charged VP1_244K capsid residue is thought to involve a strong electrostatic interaction. As a result of this high-affinity interaction, we showed that MADAL385 inhibits EV-A71 binding with PSGL1 and HS. Together with biochemical evidence, we also demonstrate that MADAL385 inhibits the binding of EV-A71 to human SCARB2- or PSGL1-expressing L929 cells. We observed that the activity of MADAL385 in L929-SCARB2 cells was exclusively dependent on the inhibition of HS binding since the activity of MADAL385 was lost in cells treated with sodium chlorate (NaClO3), a molecule that prevents cell-surface sulfation. In addition, this experiment also demonstrates the importance of HS binding for efficient entry and replication of EV-A71 in both L929-SCARB2 and L929-PSGL1 overexpressing cells.
Recent SAR studies performed with MADAL derivatives [37] point to the crucial role of the carboxylic acid groups for the antiviral efficacy. The importance of these carboxylates is corroborated by the lack of activity observed with the corresponding tryptamine (a “decarboxylated” analogue of Trp) and methyl ester derivatives (COOCH3 instead of COOH) [37]. The nature of the amino acid side chains is also very important for activity since the indole ring of Trp is preferred, most likely due to its relative polarity and hydrogen-bonding potential, particularly towards the hydroxyl of Thr141 (according to our MD simulations). These observations suggest that the carboxylic acid (-COOH) or carboxylate (-COO-) groups of MADAL385 can mimic the sulfate groups (-SO3H or -SO3-) of human PSGL1 or HS. By competing with the sulfate groups, MADAL385 may prevent virus attachment to these host receptor(s) and thereby the entry into and infection of host cell.
In agreement with the cryo-EM results showing that MADAL385 is lodged in the external pore region, the MD simulations revealed the preference of the drug for binding inside the cavity lined by the adjoined 141TPTGQVVP148 and 242QSKYP246 loops of the five VP1 subunits. Three of the MADAL385 ‘legs’ projected into the inter-subunit crevices and the remaining one occupied the outer part of the pore. It seems, therefore, that a certain conformation of these two exposed VP1 loops is necessary for MADAL385 binding, most likely for providing relative accessibility of VP1_244K to interact with the negative charges of MADAL385. We propose that the location of the MADAL385-sensitive variants VP1_148P and VP1_245Y is critical for stacking and establishing van der Waals interactions with the indole moieties of MADAL385. Furthermore, our data demonstrate that the MADAL385-resistant variants may only reduce viral sensitivity to MADAL385 in the context of the BrCr strain VP1. Indeed, the clinical isolate EV-A71_11316 carries the VP1_184S residue, a resistant variant for the BrCr strain.
Suramin and its derivative NF449 are known to specifically interact with residues VP1_145Q and VP1_98E_244K at the 5-fold vertex, respectively [33,35]. Of interest, we show that the susceptibility to suramin is not affected in the presence of both S184T and P246S amino acid substitutions. In addition, we observed a prominent in vitro antiviral synergistic effect between MADAL385 and suramin. These results indicate that, despite the common binding site for these two classes of drugs, the subtleties of their binding modes are different, which is not entirely surprising given their very distinct chemical structure. However, based on molecular modelling considerations, the simultaneous binding of MADAL385 and suramin on the same 5-fold vertex seems most unlikely due to the large size and negatively charged character of both entities.
Cyclophilin A is a newly reported uncoating regulator for EV-A71 entry, and its binding site is very close to that of PSGL1 and HS on EV-A71 virion [22]. The MADAL class of tryptophan dendrimers may thus have the potential to block Cyclophilin A binding. However, the Cyclophilin A inhibitor cyclosporin A (CsA) did not affect binding of EV-A71 BrCr to RD cells (S8 Fig) nor could we observe any antiviral activity of CsA or Debio-025 (another specific Cyclophilin A inhibitor) against BrCr and the clinical isolates 11316 and 812 (CsA EC50>30μM and Debio-025 EC50>21μM), suggesting that this host factor does not play a prominent role in the entry of the strains that we used in our study.
Medicinal chemistry efforts are currently ongoing to simplify and reduce the backbone of MADAL385 without affecting the antiviral activity. Half-sized compounds (~1500 Da versus 3575.84 Da) have now been identified that are equipotent to MADAL385 against EV-A71 and that have little or no adverse effect on the host cells (at concentrations up to 100 μM). In vivo studies to assess tolerability and antiviral efficacy will be complementary to medicinal chemistry efforts to pursue the development of this class of compounds as novel EV-A71 antiviral agents.
RD cells (Human rhabdomyosarcoma cells) and Hela cells, obtained from ATCC, were cultured in DMEM (Life Technologies) supplemented with 10% heated-inactivated fetal bovine serum (FBS); Enterovirus A71 BrCr strain (EV-A71 BrCr), a kind gift from Prof. F. van Kuppeveld (University of Utrecht), was propagated on RD cells with 2% FBS-DMEM. L929-human SCARB2 cells and L929-human PSGL1 cells were kindly provided by Prof. Satoshi Koike (Tokyo Metropolitan Institute of Medical Science, Japan) and were grown in DMEM supplemented with 10% FBS, 1% sodium bicarbonate, 1% L-glutamine and 10μg/ml puromycin. To reduce sulfate from the cell surface, 50mM of sodium chlorate (NaClO3) was added to the culture medium one week before the experiment. EV-A71 clinical isolates were obtained from Dr. Shih-Cheng Chang (Chang Guang University) or purchased from The National Collection of Pathogenic Viruses (NCPV).
The Trp-containing dendrimer MADAL385 was synthesized as described previously [38]. Vapendavir and pirodavir were kindly provided by Aviafen Therapeutics and A. Muigg, respectively. Suramin sodium salt and heparin sodium were purchased from Sigma-Aldrich. All compounds were dissolved in DMSO and stored at 4°C. The plasmid pEV-A71 (Nagoya-VP231) was generously provided by Dr. Arita (National Institute of Infectious Disease, Japan) and was modified to carry the viral genome of EV-A71 BrCr (pEV-A71-BrCr) by classic cloning. For the construction of EV-A71BrCr-mCherry, the genome of EV-A71 BrCr was inserted to a pShuttle-BAC vector. In the spacer region between the 5'UTR and VP4, mCherry gene was inserted and flanked by 2Apro cutting sites. The plasmids pLX304-PSGL-1 was purchased from DNASU, and vector pFUSE-hIgG1-Fc1 was from InvivoGen. PSGL-1 extracellular region was cloned into the expression vector pFUSE-hIgG1-Fc1 to create recombinant protein with human IgG Fc-tag at the C-terminal domain (pPSGL1-hFc). The EV-A71 VP1 purified MaxPab rabbit polyclonal antibody was obtained from Abnova. Goat anti-Human IgG Fc cross-absorbed secondary antibody, HRP was purchased from Thermo Scientific. Heparin Sepharose CL-6B and Dynabeads Protein G were supplied by Pharmacia and Thermo Scientific, respectively.
CPE reduction assays were performed as described previously [43]. Briefly, RD cells in 96-well cell culture plates were seeded, after which serials of diluted compounds and EV-A71 inoculum were added. CPE was quantified by MTS assay at 3 days post infection and expressed as percentage of untreated controls. The 50% effective concentration (EC50) was calculated by logarithmic interpolation and is defined as the concentration at which the virus-induced CPE is reduced by 50%.
EV-A71 viral RNA was isolated using NucleoSpin RNA kit (Macherey-Nagel) according to the manufacturer’s protocol. The following primers and probe were used for EV-A71 BrCr qRT-PCR: EV-A71 forward primer 5’ CCGCATTGAACCACTGTAATTT 3’, reverse primer 5’ GGAGCCAACGTGATAGTGATAG 3’ and probe 56-FAM/ACTATTGGT/ZEN/GGTGCCTATCA/3IABKFQ
Recombinant plasmid pPSGL-1-Fc was transfected in HEK-293T cells. After 36-48h, cells were harvested and lysed (1% Triton X-100 in NTE buffer supplemented with protease inhibitor cocktail) for 30 min in ice. Insoluble cell debris was removed by centrifugation 15,000 × g for 15 minutes. To confirm pPSGL-1 expression, the supernatant was subjected to Western-blot with goat anti-Human IgG Fc antibody. After confirmation, the soluble recombinant PSGL1-Fc was incubated with protein G Dynal beads (Sigma) overnight at 4° C on a rotary mixer. Unbound PSGL1-Fc was removed after 3 washes with cold TBS. The EV-A71 viruses were mixed with indicated concentrations of compounds at 37° C for 1 hour. Next, the beads loaded with PSGL1 were incubated with viruses in presence/absence of compounds at 4° C for 2 hours. After 3 washing with TBS to remove the unbound virus, beads were treated with 2x laemmli buffer (Sigma-Aldrich) and subject to 10% SDS-PAGE and further Western-blot with EV-A71_VP1 MaxPab rabbit polyclonal antibody and goat anti-Human IgG Fc antibody. For heparin-Sepharose coated beads, after 3 washing with TBS to remove the unbound virus, lysis buffer was added and viral RNA was extracted and quantified by real time qRT-PCR.
EV-A71_11316 was purified as described previously [44]. Briefly, EV-A71_11316 was propagated in HeLa cells for 24 h. The media and cells were collected and processed by freezing and thawing three times. Cell debris was pelleted by centrifugation and the supernatant was precipitated with sodium chloride and polyethylene glycol (PEG) 8000. After ultracentrifugation through a 30% sucrose buffer cushion, the pellets were resuspended and applied to a 10 to 35% K-tartrate step gradient. The virus was collected and dialyzed against 10 mM Tris, 200 mM NaCl, 50 mM MgCl2, pH 7.5.
EM samples were prepared and data sets were recorded at the Pennsylvania State University—Huck Institutes of the Life Sciences in the following way: prior to incubation and vitrification of the sample, the virus buffer was exchanged to phosphate-buffered saline (PBS). MADAL385 was incubated at 2.8 μM with about 58 nM of the virus at 37°C for 1 hour, which equates to four copies of molecule per each of vertex on the virus capsid. Three microliters of each sample was pipetted onto a Quantifoil R2/1 grid (Quantifoil Micro Tools GmbH, Jena, Germany), blotted to remove excess, and plunge-frozen in liquid ethane using a Vitrobot (Thermo Fisher, USA). Grids were imaged in a Titan Krios G3 under automated control of the FEI EPU software. An atlas image was assembled from micrographs taken at 165x magnification on a FEI BM-Ceta camera, and suitable areas were selected for imaging on the FEI Falcon 3EC direct electron detector. The microscope was operated at 300 kV with a 70 μm condenser aperture and a 100 μm objective aperture. Magnification was set at 59,000x yielding a calibrated pixel size of 1.1 Å. Images were recorded in movie mode saving 44 fraction images, each fraction including 56 frames. The total accumulated exposure was 46 e-/Å2.
The cryo-EM density maps for the EV-A71 virus and the two virus-complexes are available at the Electron Microscopy Data Bank via accession codes EMD-7905 (sharpened and unsharpened virus maps) and EMD-7913 (sharpened and unsharpened virus-MADAL385 complex maps). The atomic coordinates for the viruses built in the two maps are available at the PDB via accession codes 6DIJ (virus) and 6DIZ (virus-drug complex).
PyMOL was used for model building of MADAL385 fragments, their assembly into larger fragments and the full dendrimer, as well as for trajectory visualization and 3D figure generation [45]. Geometry optimization and point charge derivation for suitably capped fragments, namely, the pentaerythritol core, the trivalent spacer, and the peripheral N-acetylated Trp residues, were achieved by means of the PM3 hamiltonian available in the sqm program [46]. The standard ff14SB force field parameter set in AMBER 16 was used for both ligand and protein atoms [47,48]. A three-dimensional cubic grid consisting of 65x65x65 points with a spacing of 0.375 Å centered on the VP1 pore region displaying electron density for MADAL was defined for docking purposes. Electrostatic, desolvation, and affinity maps for the atom types present in MADAL385 were calculated using AutoGrid 4.2.6 and thereafter the Lamarckian genetic algorithm implemented in AutoDock4 was used to generate 100 docked conformations of a large variety of incrementally larger fragments [49]. Intra- and intermolecular energy evaluation of each configuration allowed the selection of the 10 best scoring solutions for each fragment. Significant clustering of solutions with the best scores was apparent for the smaller fragments and visual inspection confirmed the feasibility of the best binding poses, which were used for model building of the VP1 complex with the full dendrimer. For conformational sampling of the full dendrimer, MADAL385 was immersed in a cubic solvent box (TIP3P water molecules) extending 12 Å away from any ligand atom and neutralized by addition of 12 sodium ions to the locations with the most negative molecular electrostatic potential. The same procedure was later used for the VP1:MADAL385 complex; in this case, a weak harmonic restraint of 5 kcal mol-1 Å-2 on the protein Cα atoms was used throughout. Thereafter, all hydrogens and water molecules in both systems were first reoriented in the electric field of the solute and then all atoms were relaxed by performing 25 000 steps of steepest descent followed by 100 000 steps of conjugate gradient energy minimization. The resulting geometry-optimized coordinate sets were used as input for the ensuing molecular dynamics (MD) simulations at 300 K and 1 atm using the pmemd.cuda engine, as implemented in AMBER 16 [50]. The application of SHAKE to all bonds allowed an integration time step of 2 fs to be used. A cutoff distance of 9 Å was selected for the nonbonded interactions and the list of nonbonded pairs was updated every 25 steps. Periodic boundary conditions were applied and electrostatic interactions were represented using the smooth particle mesh Ewald method with a grid spacing of 1 Å [51]. The coupling constants for the temperature and pressure baths were 1.0 and 0.2 ps, respectively. Water molecules and counterions were first equilibrated around the positionally restrained solute for a first run of 0.5 ns. For the remaining 150 ns of simulation the whole systems were allowed to relax and coordinates were saved every 0.1 ns for further analysis by means of the cpptraj module in AMBER [52]. Subsequently, a simulated annealing procedure was followed to cool down snapshots taken every 5 ns from 300 to 273 K over a 1-ns period [53]. The geometries of these “frozen” systems were then optimized by following an energy minimization protocol until the root-mean-square of the Cartesian elements of the gradient was less than 0.01 kcal·mol-1·Å 1. The final ensemble containing 30 energy-minimized frozen molecules, which can be expected to be closer to the global energy minimum, were taken as representative of the dendrimer and the VP1-MADAL385 complex.
|
10.1371/journal.pgen.1003999 | A forkhead Transcription Factor Is Wound-Induced at the Planarian Midline and Required for Anterior Pole Regeneration | Planarian regeneration requires positional information to specify the identity of tissues to be replaced as well as pluripotent neoblasts capable of differentiating into new cell types. We found that wounding elicits rapid expression of a gene encoding a Forkhead-family transcription factor, FoxD. Wound-induced FoxD expression is specific to the ventral midline, is regulated by Hedgehog signaling, and is neoblast-independent. FoxD is subsequently expressed within a medial subpopulation of neoblasts at wounds involving head regeneration. Ultimately, FoxD is co-expressed with multiple anterior markers at the anterior pole. Inhibition of FoxD with RNA interference (RNAi) results in the failure to specify neoblasts expressing anterior markers (notum and prep) and in anterior pole formation defects. FoxD(RNAi) animals fail to regenerate a new midline and to properly pattern the anterior blastema, consistent with a role for the anterior pole in organizing pattern of the regenerating head. Our results suggest that wound signaling activates a forkhead transcription factor at the midline and, if the head is absent, FoxD promotes specification of neoblasts at the prior midline for anterior pole regeneration.
| Regeneration is widespread in the animal kingdom. Planarians are able to regenerate entire bodies from almost any fragment type. This ability requires a cell population called neoblasts, which include pluripotent stem cells, for the production of all missing tissues, as well as the information to form and pattern correct new tissue types. Two discrete regions of the body, called poles, are found at the anterior and posterior ends of the animal. Here we investigate the role of a gene encoding a Forkhead-family transcription factor, FoxD, in formation of the anterior pole. FoxD is expressed at the anterior pole and following injury, FoxD expression is induced in a restricted midline region of the animal. Next, FoxD is expressed in a subset of neoblasts at the midline. Inhibition of FoxD with RNA interference results in defective anterior pole regeneration, and subsequent failure to regenerate an organized head pattern around a new midline. FoxD is specifically required for anterior regeneration. These results suggest that there is a regenerative connection between the midline and the anterior pole.
| Planarians can regenerate from nearly any injury, but how missing tissues are recognized and replaced is poorly understood. The adult population of proliferating cells (neoblasts) in Schmidtea mediterranea includes pluripotent stem cells [1] and is responsible for new tissue production in regeneration. New cell production at wounds produces an outgrowth called a blastema, which will replace some of the missing tissues [2]. At the molecular level, injuries trigger a rapid wound response program that includes conserved immediate early genes and patterning factors required for normal regeneration [3]. Wnt and Hedgehog (Hh) signaling pathways instruct the regeneration of the anterior-posterior (AP) axis, whereas the Bmp signaling pathway controls the regeneration of the dorsal-ventral (DV) axis [4]–[13]. Multiple genes required for positional identity control during embryonic development of other organisms, such as Wnt and Bmp signaling ligands, display constitutive regionalized expression in adult planarians and also guide pattern maintenance during tissue turnover [14].
Two distinct regions composed of a small cluster of cells located at the anterior and posterior animal extremities are referred to here as the anterior and posterior planarian poles. The poles are found at the midline of the animal and are subjects of current intense study. The anterior pole expresses notum, a Wnt inhibitor required for head regeneration [8], whereas the posterior pole expresses wnt1 [7], [14]. A number of genes encoding proteins predicted to be involved in signaling pathways that regulate planarian body plan patterning, and that display regional expression in planarian muscle cells have been described [15]. The genes that display these unique attributes will be referred to as position control genes (PCGs) for simplicity of discussion, but it is important to note that patterning phenotypes have not yet been described for many of these genes. PCGs expressed broadly in the planarian head include the candidate Wnt inhibitor, sFRP-1; candidate FGF inhibitors nou darake (ndk) and ndl-4; and a homeodomain transcription factor, prep [6], [7], [14]–[17]. PCGs expressed broadly in the posterior region of the animal include genes encoding additional Wnt ligands, wnt11-1, wnt11-2, and wntP-2/wnt11-5 and the Wnt receptor frizzled-4 [6], [7], [14], [15], [18]; Hox genes are also expressed in the posterior [14]. Several PCGs are expressed at the planarian poles, but have broader expression domains that extend beyond the cluster of notum+ or wnt1+ cells at the animal head and tail tips.
How the poles are formed and the role they have in organizing regeneration of an animal with a properly patterned body plan is poorly understood. Two genes encoding transcription factors of the TALE-class homeodomain family, prep and pbx, are required for regeneration of the expression domains of anterior PCGs and the anterior pole marker notum [17], [19], [20]. pbx [20] and a LIM-homeobox gene (djislet, [21]) are required for regeneration of expression domains of posterior PCGs and the posterior pole marker wnt1. follistatin, which encodes a secreted regulator of TGF-β proteins, is also expressed at the anterior pole and is required for normal head regeneration [22], but strong inhibition of this gene can also result in the absence of blastema formation indicating a broader role of this gene during regeneration [23].
Forkhead-box (Fox) genes are an evolutionary ancient family of winged-helix transcription factors involved in a wide variety of biological processes such as regulation of cell proliferation, growth, and differentiation [24]. During embryogenesis, Fox genes are involved in organogenesis and patterning of several tissues from all three germ layers [25]. Mutations in Fox genes have a profound impact in human disease causing a variety of phenotypes, from eye abnormalities to speech impediments [25]–[27]. Some members of the Fox gene family are expressed in restricted regions of embryos. In Drosophila, genes encoding Forkhead-family proteins, fkh, sloppy paired 1 and 2, and crocodile, are all expressed in the anterior region of the embryo, and are required for midline establishment as well as head patterning [28]–[31]. In amphioxus, FoxQ is expressed at the anterior pole during embryogenesis [32]. In Xenopus, the forkhead family gene, XDF1 is expressed in Spemann's organizer and at later stages in the anterior neural region [33]. In planarians, few Fox genes have been described. In particular, DjFoxD is expressed in few cells at the anterior pole region of the planarian D. japonica [34] and FoxD influences expression of follistatin in planarian heads [22], [34]. Given the potential importance of the planarian anterior pole in organizing head regeneration, we investigated the role of Schmidtea mediterranea FoxD in regeneration.
A number of genes have been identified that are expressed in different domains of planarian heads. To provide a molecular definition of the anterior-most end of the planarian head, the anterior pole, we investigated the expression of a number of genes expressed near the planarian head tip. FoxD is expressed in a very small number of cells at the head tip (Fig. 1 and [22], [34]), but the pole and its role(s) are poorly defined; we focused our investigation of the anterior pole on the Schmidtea mediterranea ortholog of DjFoxD, Smed-FoxD, or FoxD in short (Fig. S1). FoxD expression in intact animals was dorsal-biased and most FoxD+ cells also expressed the gene notum. notum is required for the head-versus-tail regeneration decision known as AP regeneration polarity, and encodes a secreted inhibitor of Wnt signaling [8]. Like FoxD, notum expression in uninjured animals is largely restricted to a very small number of cells at the head tip (Fig. 1 and [8]). FoxD+ cells at the head tip also expressed multiple anterior markers (PCGs) that are expressed in AP transcriptional domains extending beyond the FoxD+ cells, including sFRP-1, ndk, ndl-4, and prep, but not sFRP-2 (Fig. 1). We also assessed whether FoxD is co-expressed with two planarian genes expressed at the DV boundary (a lateral domain surrounding the animal at the midpoint of dorsal and ventral surfaces) and/or at the midline (the median plane about which bilateral symmetry is organized): admp, encoding a BMP-family signaling ligand expressed in the ventral midline and at the DV boundary [10], [11] and slit, a conserved midline cue with a prominent role in regulation of axon guidance [35] expressed in the ventral and dorsal planarian midline [35]. admp and slit expression did not substantially coincide with pole cells expressing FoxD; midline slit expression did not reach the anterior-most region where FoxD-expressing cells are found (Fig. 1). We propose a definition for the planarian anterior pole in the mature head as the few cells restricted to the head tip and that co-express the highly restricted FoxD and notum genes together with a set of anterior PCGs, but displaying little expression of DV boundary and midline genes. The cellular basis for formation of these cells and the roles of these cells in regeneration are investigated below.
FoxD expression was highly induced by three hours following wounding in subepidermal cells, with expression peaking at approximately six hours after wounding (Fig. 2A). FoxD expression after amputation occurred at both anterior- and posterior-facing wounds, raising the possibility that FoxD is a generically wound-induced gene (Fig. 2A). FoxD expression after amputation was greatly diminished by 18 to 24 hours following injury (Fig. 2B), but increased again between 24 and 48 hours after amputation. At this later time FoxD expression was only observed at anterior-facing wounds that required the formation of a new anterior pole (Fig. 2B and Fig. S2A). Irradiation eliminates neoblasts [36], which comprise the entire population of dividing adult planarian cells; therefore, amputation experiments in irradiated animals can determine whether new gene expression in regeneration occurs in pre-existing cells at wound sites, or requires new cell production. Early wound-induced expression of FoxD was irradiation-insensitive, indicating that it is a transcriptional response in cells present at the time of wounding (Fig. 2B).
The large numbers of planarian wound-induced genes described so far are expressed broadly at the wound site [3], [8], [18], [37]. By contrast, wound-induced FoxD expression following amputation was unique: restricted to cells found in the ventral midline (Fig. 2A,B). To better understand the regulation of FoxD expression by wounding, we examined the impacts of several injury types on FoxD expression (Fig. 2C). First, we observed that expression of FoxD was not exclusive to wounds requiring pole regeneration; both parasagittal and sagittal amputations induced FoxD expression in the midline broadly along the wound site (first two cuts from the left, Fig. 2C). Incision into the planarian side with a scalpel, a wound not requiring blastema formation for repair, was sufficient to induce FoxD expression. Strikingly, FoxD was expressed at anterior and posterior areas of the incised wound site, but only within the midline on the ventral side (third cut from left, Fig. 2C). This wound-induced FoxD expression occurred even in the absence of the anterior pole, indicating that local cues rather than signals from the pole control wound-induced expression of FoxD. An even more minor injury, a dorsal puncture with a needle (third panel from right, Fig. 2C), induced FoxD expression in the vicinity of the dorsal puncture, but only ventrally in the midline. All of these wound types impinged upon the midline of the animal, raising the possibility that any midline injury is sufficient to trigger FoxD expression, regardless of whether blastema formation would be required for repair or not. Supporting this hypothesis, two wound types that do not damage the midline (lateral puncture and lateral edge removal; first and second panels from right, Fig. 2C) did not trigger expression of FoxD. These two wounds did, by contrast, elicit normal expression of other wound-induced genes (Fig. 2D).
The wound-induced expression of FoxD occurred within ventral midline cells expressing the midline markers admp and slit (Fig. 2E). Furthermore, FoxD was also co-expressed at wounds with other defined wound-induced genes, such as noggin-like1 (nlg1) [3], [38], wntless [3], [12], [38], notum [8], and wnt1 [37] (Fig. 2E and Fig. S2B). Wound-induced genes that peak in expression around six hours following injury and that are mostly expressed subepidermally at wound sites are known as W2 genes [3]. Although most W2 genes are predicted to be secreted proteins, such as signaling and matrix remodeling factors, our results indicate that the transcription factor FoxD itself belongs in this category. W2 gene expression is induced in muscle cells expressing collagen [15], and most wound-induced FoxD expressing cells (92.5±4.6%) also co-expressed the muscle gene collagen (Fig. 2E). We conclude that FoxD is a wound-induced gene in planarian muscle, but unique among known planarian wound-induced genes with expression occurring only in ventral midline cells and only following injury that impinges on the midline. FoxD also defines a fourth gene (together with wnt1, notum, and follistatin) that is wound-induced and subsequently expressed in either the anterior or posterior pole.
The second phase of FoxD expression during regeneration, initiating around 24 hours post-amputation, occurred in cells that were coalesced at the anterior pole. This expression phase presented the opportunity to define the cellular steps of anterior pole formation. Irradiated animals did not form an anterior pole and did not express FoxD at 24 to 72 hours following wounding (Fig. 2B), indicating the requirement of neoblasts for this process. FoxD-expressing cells at the regenerating anterior pole at this time also co-expressed multiple anterior PCGs as well as the anterior pole-expressing gene notum (Fig. 3A). follistatin, which is required for anterior regeneration [22], [23], was also co-expressed with FoxD at the regenerating anterior pole at this regeneration stage (Fig. 3A). FoxD expression gradually became restricted with time to fewer cells at the regenerating anterior pole, adopting the same appearance as in intact animals (Figs. 1 and 3A).
The planarian smedwi-1 gene encodes a PIWI family protein and is expressed in all dividing adult planarian cells, marking the neoblast population [1]. We found that some FoxD-expressing cells located at the forming anterior pole co-expressed smedwi-1 in regenerating blastemas 72 hours following amputation (Fig. 3B). In addition to FoxD, expression of the anterior pole gene notum and the anterior-patterning gene prep was also found in smedwi-1-expressing neoblasts at the regenerating anterior pole (Fig. 3B). These data suggest that by three days of regeneration some neoblasts have been specified to produce the new pole cells of the regenerating head, potentially marking the first cellular step in formation of a new anterior pole.
Both Wnt and Hh signaling pathways are required for the head-versus-tail regeneration decision made at planarian amputation planes [4]–[8], [39]. To test whether wound-induced FoxD expression was regulated by these signaling pathways, we inhibited Wnt (β-catenin and APC) and Hh (patched (ptc) and hedgehog (hh)) pathway genes with RNAi and examined FoxD expression six hours following amputation. As controls, we analyzed the numbers of notum- and wnt1-expressing cells in the different RNAi conditions and, as expected from prior reports [5], [8], notum expression was exclusively affected in β-catenin and APC RNAi animals and wnt1 expression was affected following perturbation of the Hh signaling pathway (Figs. 4 and S3A). Perturbation of Wnt signaling did not affect wound-induced FoxD expression (Figs. 4 and S3A). By contrast, ptc(RNAi) animals displayed fewer than normal FoxD-expressing cells at wounds and hh(RNAi) animals had increased numbers of FoxD-expressing cells (Figs. 4 and S3A). patched (ptc) encodes a receptor for Hh that antagonizes pathway output [40]. Hh signaling therefore negatively regulates FoxD induction at the midline following wounding. Because hh impacts wound-induced wnt1 expression oppositely to FoxD [4], [5], this result does not reflect a generic requirement for hh in wound-induced gene activation. FoxD expression at the anterior pole was normal in hh(RNAi) anterior blastemas (Fig. S3B), indicating a specific role for hh in regulating the FoxD wound-induced phase of expression.
ptc(RNAi) animals with a strong phenotype regenerate tails in place of heads, but we found no evidence that FoxD influences the head-versus-tail regeneration choice.
However, ptc(RNAi) animals with a weak phenotype (e.g., cyclopic heads) do resemble FoxD(RNAi) animals [4], [5]. Cyclopic or headless ptc(RNAi) animals showed decreased expression of the anterior PCG sFRP-1 ([5] and Fig. S3D) and decreased expression of the anterior pole marker notum (Fig. S3D), indicating a defect in anterior pole regeneration. Therefore, the defect in FoxD expression might contribute to the ptc(RNAi) phenotype. In vertebrates, Sonic hedgehog (a member of the Hh family) signaling is required for normal forebrain development and midline induction [40], [41]. In planarians, hh is expressed ventrally and medially [4], [5]; the impact on wound-induced expression of FoxD at the midline raises the interesting possibility that hh might also have a role in midline biology in planarians. In ptc(RNAi) animals, midline expression of slit at six hours following amputation was normal (Fig. S3C). 86.3±5.5% of FoxD-expressing cells following wounding co-express the midline gene slit in wild-type animals. Therefore, these results indicate that the reduced wound-induced expression of FoxD in ptc(RNAi) animals is not a consequence of the absence of the midline cells that normally express FoxD.
FoxD(RNAi) animals displayed defective regeneration, with variable blastema size and heads that regenerated either one or no eyes (Fig. 5A). FoxD(RNAi) head blastemas had abnormal anatomy, with medial collapse of cephalic ganglia (labeled with an RNA probe to choline acetyltransferase (chat) [42]), one or no eyes (detected with an anti-ARRESTIN antibody (VC1) [43]), and a slightly abnormal anterior intestine morphology (labeled with an RNA probe to mat [1]) (Fig. 5A). Tail fragments regenerated pharynges (Fig. S4B), demonstrating that some missing tissues can regenerate in FoxD(RNAi) animals. Moreover, intestinal branches were normally regenerated in FoxD(RNAi) tail blastemas (Fig. S4C), indicating that FoxD has a largely specific role in anterior regeneration. Parasagittal thin fragments of FoxD(RNAi) animals also regenerated pharynges (Fig. S6B), further demonstrating that neoblasts can replace missing tissues. However, most of these fragments regenerated only one eye, showed slightly reduced expression of the anterior PCG sFRP1, and regenerated asymmetric cephalic ganglia (Fig. S6B).
Blastema size abnormalities in FoxD(RNAi) animals did not appear to be an overt consequence of a neoblast maintenance defect, because normal mitotic cell numbers were present in FoxD(RNAi) animals at the time of the amputation (0 hours following wounding, Fig. S4A). Following FoxD RNAi, neoblast (smedwi-1-expressing cells) proliferation (6 hours post-amputation, Fig. S4A) and migration (18 hours post-amputation, Fig. 5B, upper panel) in response to wounding were normal. At 48 hours following wounding, neoblasts of regenerating FoxD(RNAi) tail, but not trunk, fragments displayed slightly reduced neoblast proliferation (Fig. 5B and Fig. S4A). Reduced proliferation in FoxD(RNAi) tail fragments persisted five days following wounding (Fig. S4A).
FoxD(RNAi) animals have reduced numbers of follistatin-expressing cells at the anterior pole [22]. At 48 hours following wounding notum+ and notum+ follistatin+ cells were reduced from regenerating FoxD(RNAi) anterior blastemas (Fig. 5B and Fig. S5A). Furthermore, neoblasts expressing notum and prep were also fewer or absent in FoxD(RNAi) animals (Fig. S7), suggesting that FoxD is required for neoblast specification into anterior pole cell progenitors. These observations suggest a role for FoxD in anterior pole regeneration by specifying pole progenitors.
At 72 hours following head amputation, we observed significantly lower or complete absence in expression of the anterior PCGs prep, ndl-4, and sFRP-1, and decreased or no notum-coalesced cells in FoxD(RNAi) animals (Fig. 5C). After seven days of regeneration, FoxD(RNAi) anterior blastemas also showed severe defects in the expression of sFRP-1 and ndl-4, as well as the pole-restricted gene notum (Fig. 5D and Fig. S5B). Together, these results establish a role for FoxD and the anterior pole in head patterning during regeneration.
FoxD(RNAi) animals had normal numbers of notum- and wnt1-expressing cells at six hours following wounding (Fig. 6A), indicating that FoxD does not prevent the wound-induced phase of notum expression during the specification of head regeneration. Furthermore, wound-induced expression of follistatin [22], [23] was normal in FoxD(RNAi) animals (Fig. S5A). We also did not observe ectopic expression of posterior markers (wnt11-2 and wnt1) in regenerating anterior blastemas of FoxD(RNAi) animals (Fig. S4D), demonstrating that the choice to regenerate a head instead of a tail (AP regeneration polarity) was not detectably affected. In addition, we observed normal expression of posterior markers (wnt11-2 and wnt1) in regenerating posterior blastemas, indicating a specific role for FoxD in anterior regeneration (Fig. 6B).
The TALE-homeodomain-encoding pbx and prep genes are required for regeneration and maintenance of anterior PCG expression, including for markers of the anterior pole [17], [19], [20]. Because prep is expressed broadly at the head tip and pbx in most cells of the regenerating head [17], [19], [20], it was not previously possible to determine whether poles promoted anterior PCG expression or vice versa. However, because FoxD expression is restricted to the regenerating pole, the FoxD RNAi phenotype described above suggests the pole is required for anterior PCG expression. This model predicts that pbx and prep might be required for FoxD expression. Wound-induced FoxD expression at six hours following injury was normal in both pbx and prep RNAi animals (Fig. 7A). Similarly, other wound-induced genes (notum and wnt1) are expressed normally in pbx(RNAi) animals [19], [20], indicating that pbx and prep act downstream of wound-induced expression of these genes. By contrast, anterior pole-specific expression of FoxD and notum was completely absent in both pbx and prep RNAi animals 72 hours following head amputation (Fig. 7B). Because both FoxD and notum expression is induced in neoblasts at 72 hours following amputation of wild-type animals, the complete absence of expression of these two genes at this time point suggests a requirement for pbx and prep in the specification of neoblasts into anterior pole progenitors and that this defect might underlie the pbx and prep phenotypes.
The restricted, wound-induced expression of FoxD in the midline raises the possibility of a regenerative connection between the midline and the anterior pole. To explore this possibility further, we analyzed the expression pattern of several midline genes, such as slit, admp, and ephR1, in FoxD(RNAi) animals. Indeed, these midline genes were not properly expressed in regenerating FoxD(RNAi) anterior blastemas (Fig. 5D). In addition, parasagittal thin pieces showed reduced slit expression at the anterior midline (Fig. S6B). By contrast, tail blastemas of transversely amputated FoxD(RNAi) animals displayed normal extension of slit-expressing cells to the posterior pole (Fig. S4C), further demonstrating the specific role of FoxD in the biology of the anterior pole.
To further investigate the possible connection of the regenerating anterior pole and the midline, we examined the midline-expressed slit and ephR1 genes in pbx and prep RNAi animals, which are unable to regenerate an anterior pole (Fig. 7B and [19], [20]). pbx(RNAi) animals have been reported to have defective slit expression in anterior and posterior blastemas [19], [20]. As predicted, expression of both slit and ephR1 was defective in anterior blastemas of pbx and prep RNAi animals seven days following head amputation (Fig. 7B). Moreover, the midline expression of ephR1 was defective in all ptc(RNAi) animals showing intermediate phenotypes (cyclopia and headless; and therefore, reduced or absent anterior pole) at seven days following head amputation (Fig. S3D); because of the broad role of ptc it is unknown whether this defect is solely explained by a pole defect in ptc(RNAi) animals. hh(RNAi) animals, which can regenerate a normal anterior pole (Fig. S3B), showed proper expression of the ephR1 gene in anterior blastemas at seven days following amputation (Fig. S3B). Intact FoxD(RNAi) animals undergoing long-term RNAi did not show an obvious phenotype (Fig. S6A); following transverse amputation of these animals, small blastemas with one or no eyes were regenerated, and the anterior pole was rarely formed or very reduced (Fig. S6C, D). In some of these amputated FoxD(RNAi) animals a small cluster of anterior pole cells were regenerated but offset from the midline (n = 5/19 offset, with 14/19 showing medial but severely reduced poles, Fig. S6D). Altogether, our results suggest a role for the anterior pole in organizing the regeneration of the new midline and patterning of the head.
The ends of embryos that will become the poles of the primary body axis can be involved in organizing embryonic tissue pattern in many animals. For example, bicoid mRNA in Drosophila is locally translated at the prospective anterior end of the oocyte and controls establishment of AP body pattern [44], [45]. PAR-3/PAR-6/aPKC-3 proteins localize to one side of the C. elegans zygote after fertilization, establishing the anterior embryo end [46]. In neurula-stage Xenopus embryos, Wnt genes regulate AP neurectoderm patterning and are active at the posterior, with Wnt inhibitory genes active at the anterior [47]. Early steps in development, such as asymmetric deposition of maternal determinants in oocytes or events triggered at the sperm entry site can be involved in establishment of polarized regions that impact primary axis pattern formation. In animals capable of whole-body regeneration, by contrast, the anterior and posterior poles must be formed de novo without reliance on events specific for embryogenesis. This problem exemplifies a general challenge faced by all regeneration paradigms: the establishment of pattern in an adult tissue context utilizing initial steps that can differ from the cues accessible to embryos. Planarian pole regeneration presents an ideal setting to explore the mechanistic basis for this process.
FoxD encodes a Forkhead-family transcription factor that functions during planarian regeneration to specify cells at a key signaling center occurring at the intersection of the anterior extreme and the midline. Transcription factors expressed regionally in embryos can promote establishment of signaling domains that control tissue pattern. A classic example is goosecoid, which encodes a homeodomain transcription factor required for expression of many patterning factors at Spemann's organizer in frog embryos [48]. In regeneration, how transcription factors might be suitably activated to organize regeneration of tissue pattern is poorly understood. The early and restricted midline expression of FoxD following wounding, the subsequent expression of FoxD at the anterior pole (which is located at the midline), and the anterior pole and midline regeneration phenotype in FoxD, pbx, and prep RNAi animals suggest that the anterior pole organizes many aspects of head regeneration. We propose the following model for head regeneration (Fig. 7C):
In the first phase, the anterior wound response (3–18 hours post-amputation), signaling mechanisms occurring in pre-existing muscle cells at wounds determine whether a new head is to be regenerated. Wound signaling triggers rapid (within ∼6 hours) expression of wnt1 at all wounds [37]. Wnt signaling is selectively inhibited at anterior-facing wounds, involving activation of the notum gene [8]. The mechanism that leads to selectivity in notum activation is unknown. This first phase of head regeneration results in inhibition of Wnt signaling at anterior-facing wounds, whereas Wnt signaling is active at posterior-facing wounds. FoxD is activated at the midline of most wounds during this initial phase of wound-induced wnt1 and notum expression, but our data do not indicate a role for this gene in the decision to regenerate a head-versus-tail.
If an anterior-facing wound is not juxtaposed by anterior tissue, a second phase of head regeneration involving anterior pole progenitor specification ensues. This phase (24–72 hours post-amputation) involves formation of a new anterior pole in an emerging blastema. The initial medial FoxD expression presents a candidate mechanism for establishing the location of pole regeneration–at the prior midline. This location of FoxD induction highlights the connection between the midline and pole regeneration as an important area for further investigation. We found that a small cluster of cells at the midline expressing notum, follistatin, and FoxD emerges from neoblasts in this second phase. Neoblasts can be specified to form eyes [48] or protonephridia [49] in regeneration, and here we demonstrate that some neoblasts near the forming anterior pole express FoxD, notum, and prep. We propose that neoblast induction of these genes near the midline at wounds represents an initial cellular step in formation of the cells of the new anterior pole.
In a third phase of head regeneration (∼48 hours+ post-amputation), as the blastema grows substantially, pattern of the head blastema is established. FoxD RNAi severely reduced anterior pole regeneration. Whereas these FoxD(RNAi) anterior blastemas still displayed anterior character (such as the presence of brain cells), the blastemas lacked the normal tissue organization of wild-type head blastemas. FoxD(RNAi) blastemas also had defects in the expression of a number of anterior PCGs with broad anterior expression domains (e.g., prep, sFRP-1). This raises the possibility that the anterior pole is involved in establishing and maintaining more general anterior patterns of gene expression during this third phase of head regeneration. Consistent with this hypothesis, head tissue can also be regenerated in pbx or prep RNAi animals lacking poles, but establishment of anterior patterning gene expression domains is severely affected [17], [19], [20]. Finally, expression domains of genes expressed at the midline were aberrant in FoxD(RNAi) head blastemas. Together, these results suggest that the regenerating anterior pole promotes midline regeneration for proper bilateral patterning of the head and promotes establishment of gene expression domains for AP patterning of the head.
Asexual Schmidtea mediterranea strain (CIW4) animals starved 7–14 days prior experiments were used. Animals were exposed to a 6,000 rads dose of radiation using a dual Gammacell-40 137 cesium source and amputated three days after irradiation. No vertebrate animals were used in this study, and usage of planarians (invertebrates) is unregulated.
Animals were injected with control (the C. elegans gene unc-22), FoxD, or pbx dsRNA. Animals were transversely amputated and trunk pieces injected within 1 hour post-amputation with dsRNA. A booster dsRNA injection of trunk pieces was performed the following day. This procedure (amputation, injection and booster injection) was performed a total of three times, every two to three days. Following the third cycle of amputation and injections, animals were scored for phenotype at 72 hours and seven days post amputation, fixed and analyzed with in situ hybridizations [50] and for many experiments involved azide quenching as described [50].
For RNAi by feeding experiments, dsRNA-expressing bacteria cultures were mixed with 70% liver solution in a 1∶300 ratio to culture volume. β-catenin, APC, and prep RNAi animals have been fed four times (days 0, 4, 7 and 11), and amputated at d12. 72 hours or seven days following amputation, animals were fixed and in situ hybridizations performed. hh and ptc RNAi animals were fed six times every three to four days. Long-term FoxD(RNAi) homeostasis experiments were performed by feeding the animals every three to four days during a ten week-period.
Animals were wounded and fixed at six hours following injury in 4% formaldehyde [50] and nitroblue tetrazolium/5-bromo-4-chloro-3-indolyl phosphate (NBT/BCIP) colorimetric whole-mount in situ hybridizations or fluorescence in situ hybridizations (FISH) were performed as described [50]. For immunostainings, animals were fixed as for in situ hybridizations and then treated as described [51]. A mouse anti-ARRESTIN antibody was kindly provided by Kiyokazu Agata and used in a 1∶5,000 dilution, and an anti-mouse-Alexa conjugated antibody was used in a 1∶500 dilution. For the neoblast wound response assay, RNAi animals were fed during the course of eight weeks every three to four days, then were transversely amputated and trunk or tail fragments were fixed at 0, 6, 18, 48, and 120 hours following wounding. Animals were immunostained using a rabbit anti-phospho histone 3 antibody and an anti-rabbit HRP in a 1∶100 dilution as previously described [52]. Fluorescent images were taken with a Zeiss LSM700 Confocal Microscope. Light images were taken with a Zeiss Discovery Microscope.
|
10.1371/journal.pgen.1006852 | The population genomics of archaeological transition in west Iberia: Investigation of ancient substructure using imputation and haplotype-based methods | We analyse new genomic data (0.05–2.95x) from 14 ancient individuals from Portugal distributed from the Middle Neolithic (4200–3500 BC) to the Middle Bronze Age (1740–1430 BC) and impute genomewide diploid genotypes in these together with published ancient Eurasians. While discontinuity is evident in the transition to agriculture across the region, sensitive haplotype-based analyses suggest a significant degree of local hunter-gatherer contribution to later Iberian Neolithic populations. A more subtle genetic influx is also apparent in the Bronze Age, detectable from analyses including haplotype sharing with both ancient and modern genomes, D-statistics and Y-chromosome lineages. However, the limited nature of this introgression contrasts with the major Steppe migration turnovers within third Millennium northern Europe and echoes the survival of non-Indo-European language in Iberia. Changes in genomic estimates of individual height across Europe are also associated with these major cultural transitions, and ancestral components continue to correlate with modern differences in stature.
| Recent ancient DNA work has demonstrated the significant genetic impact of mass migrations from the Steppe into Central and Northern Europe during the transition from the Neolithic to the Bronze Age. In Iberia, archaeological change at the level of material culture and funerary rituals has been reported during this period, however, the genetic impact associated with this cultural transformation has not yet been estimated. In order to investigate this, we sequence Neolithic and Bronze Age samples from Portugal, which we compare to other ancient and present-day individuals. Genome-wide imputation of a large dataset of ancient samples enabled sensitive methods for detecting population structure and selection in ancient samples. We revealed subtle genetic differentiation between the Portuguese Neolithic and Bronze Age samples suggesting a markedly reduced influx in Iberia compared to other European regions. Furthermore, we predict individual height in ancients, suggesting that stature was reduced in the Neolithic and affected by subsequent admixtures. Lastly, we examine signatures of strong selection in important traits and the timing of their origins.
| Ancient genomics, through direct sampling of the past, has allowed an unprecedented parsing of the threads of European ancestry. Most strikingly, longitudinal studies of genomewide variation have revealed that two major technological innovations in prehistory, agriculture and metallurgy, were associated with profound population change [1–5]. These findings firmly address the longstanding archaeological controversy over the respective roles of migration, acculturation and independent innovation at such horizons; migration clearly played a key role. However, this may not be universal and genomes from several important European regions and time periods remain unexamined. In particular, at the southwestern edge of Europe several aspects of the archaeology suggest that some querying of the emerging paradigm is necessary.
First, whereas dating and similarity of the Portuguese Neolithic sites to other Mediterranean regions point to a rapid spread of agriculture at around 5500 BC [6], local Mesolithic communities were sedentary, dense and innovative; they appear to have persisted for at least 500 years after the onset of the Neolithic [7] and, along with those Brittany, may have had a role in the subsequent emergence of the earliest Megalithic tradition [8].
Second, in the transition to metallurgy, the Tagus estuary region of Portugal was a source for innovation. The distinctive Maritime Beaker, a key component of the Bell Beaker Package, characterised by grave goods including copper daggers and archery equipment first emerged there during the first half of the 3rd millennium BC. The Beaker package subsequently spread through Western Europe, where it is thought to have met and hybridized with the Steppe-derived Corded Ware or Single-Grave culture [9,10]. It remains an open question whether the influx of Steppe ancestry into North and Central Europe [4,5,11] associated with Corded Ware, also had a third millennium impact in Iberia.
Third, modern Iberia has a unique diversity of language with the persistence of a language of pre-Indo European origin in the Basque region. Interestingly, the population of Euskera speakers shows one of the maximal frequencies (87.1%) for the Y-chromosome variant, R1b-M269 [12], which is carried at high frequency into Northern Europe by the Late Neolithic/Bronze Age steppe migrations [4,5,13], although its arrival time in Iberia remains unknown.
In order to investigate the nature of cultural progression at Europe’s south Atlantic edge we analyse genomes from 14 ancient Portuguese samples from the Middle Neolithic through to the Middle Bronze Age (4200–1430 BC). For broader context we also impute genomewide diploid genotypes in these and other ancient Eurasians and investigate ancient population structure and examine temporal change in individual height.
DNA was extracted from the dense portions of fourteen petrous bones [3] excavated from eight archaeological sites across Portugal (S1 Fig), dated from the Middle Neolithic (MN) and Late Neolithic/Copper Age (LNCA) to the Bronze Age (BA) (S1 Text). Genomic coverage obtained was between 0.05x-2.95x and endogenous DNA estimates ranged from 5.6% to 70.2% (Table 1). Data authenticity was attested to by post-mortem deamination at the end of reads (S2 Text, S3 Fig) and low contamination estimates; X-chromosomes in males gave an average of 1.3% (0–2.3%) (S1 and S2 Tables) and mtDNA 1.07% (0–1.71%) (S4 Table).
In order to harness the power of haplotype-based methods to investigate substructure in our ancient samples, we imputed missing genotypes in 10 out of 14 ancient Portuguese together with 57 published ancient DNA genomes, choosing those with >0.85X coverage and using the 1000 Genomes phase 3 reference haplotypes [2,3,5,11,14–21].
Comparison of imputed variants from down sampled genomes with those called from full coverage has shown that this approach gives genotype accuracy of ~99% in ancient Europeans and we confirmed this using four down-sampled genomes from different time horizons included within our analysis [3,19] (S5 Text, S6 Fig). We observed that lower minor allele frequencies (MAF) imputed less accurately (S7 Fig). We also used D statistics to test whether imputed calls from down-sampled high coverage genomes shared significantly more drift with reference populations, relative to high quality diploid calls from those same genomes. For comparison, we also tested the most commonly used form of ancient variant data; pseudo-haploid calls. Both types of call demonstrated bias towards reference panel populations, with pseudo-haploid data showing the most extreme deviations. The extent of bias was dependent on a number of variables, such as the MAF filters imposed, reference population ancestry and sample ancestry, which are discussed in S5 Text.
We subsequently use both low coverage calling approaches in complementary analyses of our ancient data and filtered for MAF > 0.05 (S5 Text). This gave 1.5 million markers with phase information called across each of the 67 samples. With these we first used CHROMOPAINTER [22] to generate an ancestry matrix which was utilized by fineSTRUCTURE [22] to identify clusters (Fig 1). The 67 Eurasian samples divided into 19 populations on the basis of haplotype sharing which are highlighted in a principal component analysis (PCA) calculated from the coancestry matrix (Fig 1A). Geographical and temporal locations are shown for these also, where Fig 1B shows four populations of hunter-gatherers (HG) individuals, Fig 1C, three populations belonging to Neolithic farmers and Fig 1D other groups containing samples ranging from Copper Age to the Anglo-Saxon period.
Hunter-gatherer samples fall into 4 clusters (Fig 1B); interestingly the Paleolithic Bichon and Mesolithic Loschbour fall together (Western_HG1), despite 6,000 years separation, hinting at some level of continuity in the Rhine basin. Earlier Neolithic individuals are separated into two groupings, one comprising NW Anatolian and Greek samples, as well as two LBK individuals from Hungary and Germany. The second consists of Hungarian individuals from the Middle Neolithic to Copper Age alongside a Spanish Cardial Early Neolithic. A large cluster of individuals from Atlantic Europe, spanning the Middle Neolithic to Copper Age, is also seen, including all Portuguese MN and LNCA samples.
Samples belonging to the Copper Age and subsequent time periods in Russia showed strong stratification, with previous insights into ancient population structure in the steppe [5] corroborated by the formation of the Yamnaya_Afanasievo cluster and the Sintashta_Andronovo. In contrast, Central/Northern European samples stretching from the Copper Age to Anglo Saxon period all clustered together with no detectable substructure (CopperAge_to_AngloSaxon). However, the Portuguese Bronze Age individuals formed a distinct cluster. This was seen to branch at a higher level with the Atlantic_Neolithic rather than CopperAge_to_AngloSaxon, and in the PCA plot placed between the two.
It has been previously shown that an individual (CB13) dating from the very beginning of the Neolithic in Spain showed ancestry closer to a Hungarian hunter-gatherer (KO1, found within a very early European Neolithic context) than to the more western HGs from LaBrana in Spain and Loschbour in Luxembourg [18]. Furthermore, recent studies have highlighted an increase in western hunter-gatherer (WHG) admixture through the course of the Spanish Neolithic [17,23]. To investigate suspected local HG introgression in Iberia we compared relative haplotype donation between different hunter-gatherers within European farmers and other samples from later time-periods (Fig 2). In Iberia, a clear shift in relative HG ancestry between the Early Neolithic (EN) to MN was observed, with greater haplotype donation from the Hungarian HG within the Cardial Neolithic sample CB13 [18], when compared to other HG of more western provenance (Bichon, Loschbour and LaBrana). A reversal of this trend is seen in the later Neolithic and Chalcolithic individuals from Portugal and Spain, but intriguingly not in other Atlantic Neolithic samples from Ireland and Sweden. This is confirmed by a Mann-Whitney test demonstrating that Iberian Neolithic samples receive significantly more (p = 1.02x10-6) haplotypes from west European HG (Bichon, Loschbour and LaBrana) than KO1 relatively to Neolithic samples from elsewhere in Europe suggesting a more prolonged hunter-gatherer interaction at the littoral. In the transition to the Portuguese Bronze Age, a second shift can be seen in relative hunter-gatherer ancestry with some increase in relative haplotype donation from KO1, which is seen more prominently in the majority of post-Neolithic Eurasian samples, hinting at some difference between the Portuguese Neolithic and Bronze Age.
Next, to further investigate this apparent shift between the Neolithic and Bronze Age in Iberia, we explored haplotype sharing patterns of ancient samples in the context of modern populations. We merged our dataset of imputed variants with 287,334 SNPs typed in 738 individuals of European, Middle Eastern, North African, Yoruba and Han Chinese ancestry [24] and ran CHROMOPAINTER/FineSTRUCTURE as above.
When comparing vectors of haplotype donation between Neolithic and Bronze Age individuals of different European regions to modern populations, a geographical pattern emerges (Fig 3) [25]. As expected, Neolithic samples present an excess of genetic contribution to southern Europeans, in particular to modern Sardinians, when compared to Bronze Age samples, which in turn consistently share more haplotypes with northern/eastern groups.
Consistent with this, when comparing Portuguese Neolithic to Bronze Age samples, the former presented an excess of haplotype donation to Sardinian and Spanish (p = 0.017). Northern/eastern ancestry is evident in the Bronze Age, with significantly increased enrichment in Chuvash, Orcadian (p = 0.017), Lezgin and Irish (p = 0.033). However, this shift from southern to northern affinity is markedly weaker than that observed between Neolithic and Bronze Age genomes in Ireland, Scandinavia, Hungary and Central Europe. These findings suggest detectable, but comparatively modest, Steppe-related introgression present at the Portuguese Bronze Age.
Height can be expected to give the most reliable predictions due its strong heritability and massive scale of genome wide association studies; the GIANT consortium has estimated 60% of genetic variation as described by common variants [27]. Using the imputed data of >500 thousand diploid SNP calls [27] we combined genetic effects across the whole genome to estimate this phenotype in individuals. Fig 5 plots genetic height in ancient individuals and reveals clear temporal trends. European hunter-gatherers were genetically tall and a dramatic decrease in genetic height is associated with the transition to agriculture (p<0.001). During the Neolithic period, we see a steady increase, probably influenced by admixture with hunter-gatherers. Within this trend, Iberian individuals are typical of the Middle and Late Neolithic and we see no evidence of an Iberian-specific diminution as has been previously suggested from a 180 SNP panel [23] (Fig 5; S7 Text, S30 Fig). This increase continues through the Bronze Age, influenced in part by admixture with Steppe introgressors who have high predicted values (Neolithic vs Yamnaya_Afanasievo, p<0.018) and into the early centuries AD where ancient Britons and Anglo-Saxons are among the tallest in the sample (ignoring the undoubted influence of differing environments). That Yamnaya and hunter-gatherer introgressions are major determinant of height variation is supported by strong correlations between these ancestral components and genetic height in modern European populations (Fig 4, S7 Text, S32 Fig).
The role of positive selection in shaping diversity at specific loci in European populations has been of enduring interest and thus we tested whether our imputed genomes could directly reveal the imprint of adaptation in the past. For this we used the extended haplotype homozygosity (EHH) method [28] with the six loci related to diet and pigmentation highlighted in the analysis by [23]: LCT (rs4988235), SLC24A5 (rs1426654), SLC45A2 (rs16891982), HERC2 (rs12913832), EDAR (rs3827760) and FADS1 (rs174546) (S8 Text, S37 Fig). Two of these, LCT and FADS1 showed strong signals consistent with selective sweeps; homozygous haplotypes that are longer than those surrounding the derived selected allele and that are also markedly longer than those observed in modern populations (Fig 6). The selective sweep signal for LCT (driven by adaptation to a dietary reliance on raw milk) appears in the Bronze Age and that associated with FADS1 shows first in the Neolithic sample, supporting that this may be a response to changes in the spectrum of fatty acid intake afforded after the transition to an agricultural diet [29]. We caution that the limited success demonstrated in the imputation of rare/low frequency variants in ancient samples (S7 Fig), together with potential phasing inaccuracy may result in overestimation of the length of homozygous genomic segments.
Our genomic data from 14 ancient individuals from 8 Portuguese archaeological contexts ranging from the Middle Neolithic to Middle Bronze Age throws light on how the two fundamental transitions in European prehistory affected populations at the Atlantic edge. Previous data from north Mediterranean regions in Iberia have shown that the first farmers had predominantly Anatolian ancestry [4,18,21], with some increase in hunter-gatherer admixture occurring between the Early and Middle Neolithic. Our analyses, using both observed haploid SNPs and imputed diploid haplotypes show this pattern extends to the Atlantic coast of the peninsula, a region where a dense Mesolithic population persisted in the Neolithic for some 500 years. We support Middle Neolithic HG admixture having occurred locally, as there is greater haplotypic affinity of these Iberians to HG genomes from western Europe than to a hunter-gatherer genome excavated from a much earlier point of contact within the spread of the Neolithic; that within a Hungarian settlement representative of the earliest agricultural cultures of southeast Europe. This affinity is not shared by the earlier genome from the classical Neolithic Cardial phase (7500–7100 BP) which supports the geographical adjacency of this Middle Neolithic HG admixture.
Imputation of ancient European genomes sequenced to 1x coverage has been shown to give diploid genotypes at ~99% accuracy [3]. Our investigation of bias in both imputed and haploid calls suggests value in complementary approaches to genotype determination in the analysis and interpretation of palaeogenomic data. Our imputation of 67 genomes yielded genome-scale diploid calls which we surmised should allow the prediction of polygenic traits at the individual level. We illustrate this for height, in which combined genomewide locus effects are known to explain a high proportion of trait variance and which has been shown to have been under selection in Europeans [23,30,31]. Most strikingly, we find that European hunter-gatherers are significantly taller than their early Neolithic farming counterparts. A pattern of increasing genetic height with time since the Neolithic is clear in these European individuals, which may be influenced by increasing admixture with populations containing higher ancestral components of Eurasian hunter-gatherers. This concords with the increased forager-farmer admixture in the transition from the Early to Middle Neolithic; including within Iberian Neolithic individuals. Interestingly, this is in contrast to previous results which estimated a height decrease within this group. However, that work used more limited data, 169 predictive loci, and predicted at a population rather than individual level using a minimum of only two chromosomes called per SNP [23]. Genetic height increases through the Bronze Age are further influenced by Yamnaya introgression and continue through to a series of early Britons sampled from the early centuries AD. Within this time frame, the genetically tallest individual is an Anglo-Saxon from Yorkshire, followed by a Nordic Iron Age sample.
Our analyses yield both signals of continuity and change between Portuguese Neolithic and Bronze Age samples. ADMIXTURE analysis showing similar ancestral components, and higher order branching in fineSTRUCTURE clustering suggest a level of continuity within the region. Also, both show a degree of local European HG admixture (relative to central European HG influence) that is not observed within other samples in the data set. However, final fineSTRUCTURE clustering and the PCA plot places the Portuguese BA as a separate group which is intermediate between Atlantic Neolithic samples and the Central European Bronze Age individuals. D-statistics support some influx of ancestral elements derived from the east, as is seen in the northern Bronze Age, and a distinct change in Y-chromosome haplotypes is clear—all three Iberian BA males are R1b, the haplogroup that has been strongly associated with Steppe-related migrations. Patterns of haplotype affinity with modern populations illustrate the Portuguese population underwent a shift from southern toward northern affinity to a distinctly reduced degree to that seen with other regional Neolithic-BA transitions.
Taken together this is suggestive of small-scale migration into the Iberian Peninsula which stands in contrast to what has been observed in Northern, Central [4,5] and Northwestern Europe [11] where mass migration of steppe pastoralists during the Copper Age has been implied. The Y-chromosome haplotype turnover, albeit within a small sample, concords with this having been male-mediated introgression, as suggested elsewhere for the BA transition [26].
Several candidate windows for the entry of Steppe ancestry into Portugal exist. The first is the possible emergence of Bell Beaker culture in Southwest Iberia and subsequent establishment of extensive networks with Central and NW European settlements, opening up the possibility of back-migration into Iberia. Indeed, Central European Bell Beaker samples have been observed to possess both steppe-related ancestry and R1b-P312 Y-chromosomes [4,5]. Furthermore, through the analysis of modern samples, it has been proposed that the spread of Western R1b-lineages fits with the temporal range of the Corded Ware and Bell Beaker complexes [32].
An alternative is in the Iberian Middle to Late Bronze Age when individualized burials became widespread and bronze production began [33]. At this time the spread of horse domestication enabled unprecedented mobility and connectedness. This was coupled with the emergence of elites and eventually led to the complete replacement of collective Megalithic burials with single-grave burials and funerary ornamentation reflecting the status of the individual in society. These changes are seen in the Iberian Bronze Age, with the appearance of cist burials and bronze daggers [34]. Indeed, two of the Bronze Age samples analysed in the present work belong to an archaeological site in SW Iberia where the earliest presence of bronze in the region was demonstrated, as well as high status burials with elaborate bronze daggers [34,35].
Two alternate theories for the origin and spread of the Indo-European language family have dominated discourse for over two decades: first that migrating early farmers disseminated a tongue of Neolithic Anatolian origin and second, that the third Millennium migrations from the Steppe imposed a new language throughout Europe [36,37] [4]. Iberia is unusual in harbouring a surviving pre-Indo-European language, Euskera, and inscription evidence at the dawn of history suggests that pre-Indo-European speech prevailed over a majority of its eastern territory with Celtic-related language emerging in the west [38]. Our results showing that predominantly Anatolian-derived ancestry in the Neolithic extended to the Atlantic edge strengthen the suggestion that Euskara is unlikely to be a Mesolithic remnant [17,18]. Also our observed definite, but limited, Bronze Age influx resonates with the incomplete Indo-European linguistic conversion on the peninsula, although there are subsequent genetic changes in Iberia and defining a horizon for language shift is not yet possible. This contrasts with northern Europe which both lacks evidence for earlier language strata and experienced a more profound Bronze Age migration.
All ancient DNA (aDNA) work was done in clean-room facilities exclusively dedicated to this purpose at the Smurfit Institute, Trinity College Dublin, Ireland. We extracted DNA from ~100 mg of temporal bone samples belonging to 14 samples from 8 archaeological sites in Portugal ranging from the Mid Neolithic to the Mid Bronze Age in Portugal (S1 Text) using a silica-column-based method [39] with modifications [40]. We incorporated DNA fragments into NGS libraries using the library preparation method described in [41] and amplified these with 2–4 different indexing primers per samples and purified (Qiagen MinElute PCR Purification Kit, Qiagen, Hilden, Germany) and quantified (Agilent Bioanalyzer 2100). Samples were sequenced to ~1.15X (0.05–2.95X) in an Illumina HiSeq 2000 (100 cycle kit, single-end reads mode; Macrogen) (S2 Text).
We used Cutadapt v. 1.3 [42] to trim NGS read adapters and aligned reads to the human reference genome (UCSC hg19) and mtDNA (rCRS, NC_012920.1) with the Burrows-Wheeler Aligner (BWA) v.0.7.5a-r405 [43], trimming low quality bases (q ≥ 20), removing PCR duplicates and reads with mapping quality inferior to 30 using SAMtools v.0.1.19-44428cd [44]. We estimated genomic coverage with Qualimap v2.2 [45] using default parameters (S2 Text). Raw data and aligned reads have been submitted to http://www.ebi.ac.uk/ena/data/view/PRJEB14737, secondary accession ERP016408.
In order to assess the level of contamination in ancient samples, we considered the number of mismatches in mtDNA haplotype defining mutations and determined the number of X-chromosome polymorphisms in male samples (S2 Text) [46]. We analysed aligned reads using mapDamage v2.0 [47] to inspect patterns of aDNA misincorporations, which confirm the authenticity of our data.
We used the method published in reference [48] to determine the sex of the ancient individuals (S2 Text, S2 Fig). Y-chromosome lineages of ancient male samples were identified using Y-haplo software [49] (https://github.com/23andMe/yhaplo, S4 Table). For mtDNA analysis, reads were separately aligned to the revised Cambridge Reference Sequence (rCRS; NC_012920.1) [50], trimming low quality bases (q ≥ 20) and filtering by mapping quality (q ≥ 30) and duplicate reads as above. mtDNA haplogroups were identified using mtDNA-server (http://mtdna-server.uibk.ac.at/start.html, with default parameters.
Smartpca version 10210 from EIGENSOFT [51,52] was used to perform PCA on a subset of West Eurasian populations (604 individuals) from the Human Origins dataset [2], based on approximately 600,000 SNPs (S3 Text, S4 Fig). The genetic variation of 239 ancient Eurasian genomes [2,4,5,11,14–21,23,53–56] was then projected onto the modern PCA (lsqproject: YES option). A model-based clustering approach implemented by ADMIXTURE v.1.23 [57] was used to estimate ancestry components in 10 of the Portuguese samples, alongside 1941 modern humans from populations worldwide [2] and 166 ancient individuals. Only ancient samples with a minimum of 250,000 pseudo-haploid calls were included. The dataset was also filtered for related individuals, and for SNPs with genotyping rate below 97.5%. A filter for variants in linkage disequilibrium was applied using the—indep-pairwise option in PLINK v1.90 with the parameters 200, 25 and 0.4. This resulted in a final 219,982 SNPs for analysis. ADMIXTURE was run for all ancestral population numbers from K = 2 to K = 15, with cross-validation enabled (—cv flag), and replicated for 40 times. The results for the best of these replicates for each value of K, i.e. those with the highest log likelihood, were extremely close to those presented in [11]. The lowest median CV error was obtained for K = 10.
Formal tests of admixture were implemented using D-statistics [58] and F-statistics [59,60] using the AdmixTools package (version 4.1). These were carried out on WGS ancient data only, using autosomal biallelic transversions from the 1000 Genomes phase 3 release (S4 Text, S5 and S6 Tables).
We selected for genotype imputation relevant published samples that had been sequenced by whole-genome shotgun sequencing and for which coverage is above 0.85X, including 5 ancient individuals downsampled to 2X which were included for estimating accuracy and possible bias in imputation. Within these were called ~77.8 million variants present in the 1000 Genomes dataset using Genome Analysis Toolkit (GATK) [61], removing potential deamination calls. These were used as input by BEAGLE 4.0 [62] which phased and imputed the data (S5 Text, S7 Table). This resulted in a VCF file with approximately 30 M SNPs. Imputed genotypes for 67 ancient samples analysed are available at Dryad (DOI: http://dx.doi.org/10.5061/dryad.g9f5r).
In analysis I (Fig 1), imputed variants in 67 ancient Eurasian samples were filtered for posterior genotype probability greater or equal to 0.99. Variants not genotyped across all individuals were removed with vcftools [63], also excluding SNPs with MAF < 0.05, resulting in approximately 1.5 M SNPs and the resulting VCF was converted to IMPUTE2 format with bcftools version 0.1.19-96b5f2294a (https://samtools.github.io/bcftools/). Hap files were converted to CHROMOPAINTER format with the script “impute2chromopainter.pl”, available at http://www.paintmychromosomes.com/ and created a recombination map with “makeuniformrecfile.pl”. We then split the dataset by chromosome with vcftools and ran CHROMOPAINTER and fineSTRUCTURE v2 [22] with the following parameters: 3,000,000 burn-in iterations, 1,000,000 MCMC iterations, keeping every 100th sample. In S6 Text we describe all 5 analysis CHROMOPAINTER and fineSTRUCTURE analyses in more detail: I—aDNA samples only (S12–S25 Figs, S8 Table); II—aDNA samples and present-day Eurasians and Yoruba (Fig 3; S9 Table); III—Comparison of linked and unlinked analyses (S26 Fig); IV—Analysis with unfiltered genotype probabilities; V—Detection of biases in CHROMOPAINTER analyses derived from genotype imputation in ancient samples (S27 and S28 Figs).
Genetic scores for polygenic traits including height [27], pigmentation [64], Anthropometric BMI [65] and T2D [66] in 67 ancient samples were estimated using PLINK [67] using the—score flag. Odds ratio in the T2D summary statistics [66] were converted to effect size by taking the logarithm of OR/1.81 [68]. In our analyses, we compared p-value filtering (unfiltered p-value threshold against p<0.001) when possible to qualitatively evaluate robustness of signals observed (S7 Text, S29–S31 and S33–S36 Figs).
In order to investigate the correlation between ancient ancestry in present-day populations and height genetic scores, we first calculated polygenic risk in Eurasian populations from the Human Origins dataset. This was followed by the estimation of the percentage ancestry of five distinct ancient populations (EHG, CHG, WHG, Yamnaya, Anatolian Neolithic) in the same dataset, which was done through the implementation of the F4 ratio method described in [60] using the Admixtools package (version 4.1). Two individuals possessing the highest genomic coverage from each population were used in the test, which took the form f4(Mbuti, Ancient_Ind1; Modern_WEurasian, Dai)/f4(Mbuti, Ancient_Ind1; Ancient_Ind2, Dai) (S10 Table; S32 Fig).
We used Selscan [69] to investigate extended haplotype homozygosity (EHH) around SNPs of interest previously described in [23]: LCT (rs4988235), SLC24A5 (rs1426654), SLC45A2 (rs16891982), HERC2 (rs12913832), EDAR (rs3827760) and FADS1 (rs174546). First, SNPs within 5 Mb of each SNP were included for analysis, removing SNPs which are multiallelic and with multiple physical coordinates. EHH requires large populations, and therefore we used selscan in 3 groups: HG, Neolithic farmers and Copper Age to Anglo-Saxon, using the—ehh and—keep-low-freq flag (S8 Text; S37 Fig).
|
10.1371/journal.pbio.0060157 | Immunoglobulin Heavy Chain Exclusion in the Shark | The adaptive immune system depends on specific antigen receptors, immunoglobulins (Ig) in B lymphocytes and T cell receptors (TCR) in T lymphocytes. Adaptive responses to immune challenge are based on the expression of a single species of antigen receptor per cell; and in B cells, this is mediated in part by allelic exclusion at the Ig heavy (H) chain locus. How allelic exclusion is regulated is unclear; we considered that sharks, the oldest vertebrates possessing the Ig/TCR-based immune system, would yield insights not previously approachable and reveal the primordial basis of the regulation of allelic exclusion. Sharks have an IgH locus organization consisting of 15–200 independently rearranging miniloci (VH-D1-D2-JH-Cμ), a gene organization that is considered ancestral to the tetrapod and bony fish IgH locus. We found that rearrangement takes place only within a minilocus, and the recombining gene segments are assembled simultaneously and randomly. Only one or few H chain genes were fully rearranged in each shark B cell, whereas the other loci retained their germline configuration. In contrast, most IgH were partially rearranged in every thymocyte (developing T cell) examined, but no IgH transcripts were detected. The distinction between B and T cells in their IgH configurations and transcription reveals a heretofore unsuspected chromatin state permissive for rearrangement in precursor lymphocytes, and suggests that controlled limitation of B cell lineage-specific factors mediate regulated rearrangement and allelic exclusion. This regulation may be shared by higher vertebrates in which additional mechanistic and regulatory elements have evolved with their structurally complex IgH locus.
| Lymphocytes provide a limitless repertoire of antigen receptors, but each lymphocyte expresses only one kind of receptor per cell in order to provide specific recognition and response to pathogen invasion. The restriction, called allelic exclusion, operates in tetrapod vertebrates from frogs to human beings. In mouse, immunoglobulin (Ig) heavy chain (H) exclusion depends on ordered activation of component parts of the highly complex, three-megabase IgH locus in a process that differentiates between the two alleles. However, the regulation and mechanisms ensuring allelic exclusion remain uncertain. Sharks represent the earliest vertebrates with an adaptive immune system; their IgH organization, consisting of multiple miniloci, is considered primitive and ancestral to the classical IgH locus in other vertebrates. We show that allelic exclusion nonetheless exists in shark B lymphocytes, although attained by alternative means. Thus, major aspects of the complex pathway described for allelic exclusion in mammals evolved with their IgH organization. Elucidating shared and divergent regulatory processes allows us to gain insight into the basis and evolution of allelic exclusion, which provides the foundation for the functioning of the adaptive immune system.
| The adaptive immune system in vertebrates is founded on lymphocytes expressing a vast, anticipatory repertoire of antigen receptors. Only a single species of immunoglobulin (B cells) or T cell receptor (T cells) is allowed per cell. This restriction is termed allelic exclusion, and it describes the requirements for monoallelic receptor gene expression in each cell (for a recent review, see [1]). Allelic exclusion is considered the basis of adaptive immune system function, but how this founding principle was established in evolution has been only a matter of conjecture [2]. In this report, we present data that clarify the basis of allelic exclusion in the shark, representative of the most primitive organism with an adaptive immune system shared by mice and human beings.
The diversity of antigen receptors in lymphocytes is somatically generated by a recombination process in which various gene segments are joined together [3]. The number of gene segments and their organization vary amongst species but are generally comparable among all vertebrate classes, with the exception of the cartilaginous fishes, sharks and skates [4,5]. Figure 1 illustrates the differences between the complex mouse IgH locus and the multiple, minimalist shark clusters [4]. Elucidating their divergent and shared regulatory processes will allow us to understand the basis for allelic exclusion, the phenomenon that ensures specific recognition and response to pathogen invasion.
Each mammalian B lymphocyte must express an immunoglobulin (Ig) antigen receptor with a single specificity, although there are three loci that potentially encode two heavy (H) chains and four light (L) chains. The mouse IgH consists of an array of 200 VH gene segments spaced over 2 Mb and located upstream from 10–13 D, four JH gene segments, and eight constant (C) region genes [6] (Figure 1). Initiated by the RAG recombinase, the joining of VH, D, and JH gene segments generates the ligand-binding V region that encodes the N-terminus of the H chain polypeptide [7]. Accessibility [8] of the gene segments to the recombinase is tissue-, developmental stage-, and gene-specific [9] and is associated with their transcription, although the nature of this connection is not entirely elucidated [10,11]. During B cell differentiation, chromatin domains encompassing the D, J, and Cμ genes become activated, probably through the intronic enhancer [12,13], allowing recombination of one of the D genes to a J gene segment. Only in B cells does VH to DJH rearrangement take place to form VDJ, and this stage not only requires lineage-specific regulation, but sets in motion the process resulting in monoallelic H chain expression at the IgH locus. The chromatin domains encompassing the VH become activated, and locus contraction is required to bring most of them into close proximity with the DJ [14,15]. After the completion of a productively rearranged VDJ, the H chain is expressed at the cell surface, and this initiates a feedback process by signaling the next step in differentiation [16,17]. Since the V to DJ step is asynchronous between the alleles, the first functional VDJ rearrangement will encode the antigen receptor.
Based on these mouse studies, a model for the regulation of allelic expression was developed. Recent work has shown that many factors at various levels—stage-specific expression of RAG [18], differentially activated chromatin domains [19], locus contraction and decontraction [14,20], and subnuclear relocation [15,21]—are involved. Because of the large distances between a rearranged DJ and the available VH gene segments in animals such as mouse and humans, the locus contraction mechanism would appear to be part and parcel of the rearrangement process as well as its regulation. Moreover, the model is based on only one locus, that is, distinguishing one gene from its allele.
In contrast, IgM H chain in cartilaginous fishes (sharks and skates) is encoded by multiple (15–200), independently rearranging IgH loci (Figure 1). What is more, these miniloci may bypass the need for locus contraction, which seems to be a key regulatory step for monoallelic expression at a single IgH locus in the tetrapod model. The question of how allelic exclusion is managed in sharks has thus been a long-standing puzzle.
In the nurse shark, Ginglymostoma cirratum, there are about 15 IgM H chain loci per genome, and every functional gene contains one VH, two D, and one JH gene segments located within 2 kb ([22]; Figure 1, bottom). These miniloci are located at least 120 kb apart, and aside from two IgH genes depicted in Figure 1, their linkage relationships are not known [23]. Among outbred individuals there can be 9–12 active IgH, classified into subfamilies called Groups 1–5. A detailed characterization of two functional loci [22–24] and 78 of their rearrangements show that V(D)J recombination took place within the minilocus ([22,23] and V. Lee and E. Hsu, unpublished data). There do not seem to be long-distance recombination events between the widely separated IgH loci or, presumably, a major role for chromatin contraction in nurse shark IgH rearrangement.
To elucidate the rules for V(D)J recombination in the shark, we first investigated rearrangement patterns at the two defined shark H chain loci, asking whether differential VH, D, and JH activation existed in the short (∼400 bp) intersegmental distances. We have found that all combinations are possible, and a completed VDJ is accomplished during one stage only, as if it were like the initial D to J step in mammals. Our results also confirmed that long-distance recombination between different IgH loci in B cells is rare, if it exists. Thus, two elements thought to be intrinsic to regulating the rearrangement process and resulting in allelic exclusion in mammals—ordered long distance recombination and chromatin contraction—are absent in sharks. Thus these findings tell us that certain mechanics of the rearrangement process can be dissociated from the phenomenon of allelic exclusion and that the two processes developed separately in evolution.
We investigated rearrangement in shark lymphocytes at the population and the single-cell level and established that H chain exclusion does occur in shark B cells, where only one or a few of its many IgH loci rearrange in any one cell. We also looked at IgH loci in shark thymocytes (precursor T cells, see Text S1). Although T cells do not express Ig, the IgH genes were extensively, although partially, rearranged; Ig transcripts were not detected. The differences between B cells and thymocytes demonstrated here suggest there exists in precursors to B and T cells an IgH chromatin state already permitting rearrangement, but in B cells it is further potentiated by lineage-specific factors, leading to efficient recombination at one or a few H chain genes and results in H chain exclusion. We propose that the molecular basis establishing allelic exclusion was achieved in the earliest vertebrates possessing Ig genes, and it is independent of the wide variation in Ig gene number observed in different species.
The experiments are summarized as follows. We first focused on how rearrangement takes place in one IgH subfamily, Group 2, in tissues and isolated cell populations. We demonstrated that partially and fully rearranged VH sequences can be amplified from lymphoid tissue DNA, but not from red blood cell (RBC) control DNA (Figure 2). All anticipated genomic rearrangement configurations were obtained (Table 1). These data demonstrated that rearrangement in sharks is different from the ordered, two-stage process observed in mammals.
Lymphoid tissue can carry B cells, which express IgM, and T cells, which do not. The initial experiments revealed the unexpected finding that somatic IgH recombination is also present in the thymus, an almost exclusively T cell–containing tissue. We further investigated this finding, using genomic Southern analyses. Compared to RBC and heart DNA, new—somatically rearranged—bands were observed in all lymphoid tissue DNA tested. These new bands were demonstrated to be V(D)J rearrangements mapping to predictable locations (Figure 3). The recombined bands were identified by probes detecting all VH or only the Group 2 subfamily. A comparison of surface IgM-positive (sIgM+) B cell DNA with thymus DNA showed that their respective patterns differed in the quantity and quality of the rearranged bands (Figure 4). Thus, we discovered that V(D)J recombination is much more extensive but mostly incomplete in T cells. This result was confirmed through investigating the status of all functional IgH loci in thymocytes and in B cells by single-cell genomic PCR (Figure 5).
As predicted from the genomic Southern analysis results, only a few rearrangements could be obtained from single B cells, and these were fully rearranged VDJ; the other IgH loci were in the nonrearranged, or germline (GL), configuration (Figure 6 and Table 3). Unlike in thymocytes, partial rearrangements (Figure 7) were infrequent in B cells. These results show that in the developing B cell, there was a limited number of genes activated to rearrange, but once initiated, the recombination process went efficiently to completion.
In contrast, multiple and mostly incomplete Ig rearrangements were found in single thymocytes (Table 2), and neither Ig H chain transcripts nor L chain expression and rearrangement could be detected in the thymus (Figure 8). This ability of DNA to act as substrate for RAG in the absence of transcription suggests a previously unknown state of chromatin activation. It was possible to detect this state only in an animal with multiple, independently rearranging sites, but such an observation signals that RAG may act on nontranscribing loci in other organisms as well. We propose that IgH in all shark precursor lymphocytes can be acted upon by RAG recombinase but that B lineage-specific factors are responsible for regulated rearrangement—and H chain exclusion—in the B cells.
PCR primers (Int/JH2; Figure 2) targeting the leader intron of Group 2 VH and the JH gene segment amplified DNA sequences of 1.6 kb from an individual shark (-JS) whole blood DNA (Figure 2B, lane 2); this band contained the two functional Group 2 genes in the nonrearranged, or GL, configuration (see PCR and probes, [22]). The Int/JH2 primers amplified the same 1.6-kb fragment from erythrocyte DNA in other genetically unrelated individuals (Figure 2B, lanes 4 and 6), demonstrating that Group 2 GL gene segments are in the organizational configuration depicted in Figure 1.
The intersegmental distances in Group 2 as well as other IgM genes are all about 400 bp (Figure 1). A single, initial somatic rearrangement event (1R), such as joining of V to D1, would delete this interval and reduce the total V to J genomic span detected by the Int/JH2 primers to about 1,200 bp. Likewise, two rearrangement events (2R) would give rise to a PCR product of about 800 bp, and three rearrangements (3R) 400 bp.
From lymphoid (spleen and peripheral blood leukocytes [PBL]) genomic DNA, a ladder of PCR products hybridizing to the vh probe (Figure 2) can be detected, corresponding to the anticipated sizes of partial and completed genomic rearrangements (Figure 2B, lanes 1, 3, and 5). The arrows at the left of lane 1 point to the fragments that were later cloned from all three sharks and identified as having one, two, or three rearrangements. Splenic lymphocytes and PBL from adult sharks do not express RAG recombinase ([25]; W. Feng and E. Hsu, unpublished data), so that these rearrangement intermediates would be relics from earlier stages of lymphocyte differentiation.
PCR products obtained from shark-JS splenic DNA were cloned, and the insert sequences are classified by size in Table 1. Within each size group, different rearrangement combinations were found, but some are more frequent than others. The junctions (Figure S1) show that each clone is unique, with the typical diversity generated by trimming and N/P region addition.
In order to determine which cells contained these rearrangements, surface IgM-expressing (sIgM+) cells were isolated from PBL (see Figure S2). The PCR reactions performed on this population and the thymus both amplified sequences that showed higher frequency of VH to D1 joining (Table 1), but all combinations exist. The fully rearranged VDJ (i.e., VDDJ) from the sIgM+ cells tended to be in-frame, whereas in those from the spleen and thymus, the nonfunctional ones are in the majority (last two rows, Table 1). Thus, it appeared that IgH recombination had occurred in both precursor T and B lymphocytes. The pattern of rearrangement for Group 2, as demonstrated by the frequencies of the intermediate configurations, was similar in all samples.
Although the 5′ primer is specific for Group 2, the 3′ primer could target any IgH JH. If rearrangement occurred between Group 2 VH and another locus, it would have been possible to detect non-Group 2 intersegmental sequence in the partial configurations shown in Table 1. (All but one of the partially recombined clones had rearranged within Group 2 loci, as ascertained by sequencing or restriction enzyme analyses. In one thymus VDD-J clone, the VH originated from Group 2, but the D2, D2-J intersegmental sequence and JH were from Group 1; only four nucleotides belong to the D1 of either Group. If this sequence were a PCR artifact, the area of homology would have to have been in the N region sequence between VH and D1 or D1 and D2. We screened a total of 58 thymic VDD-J [unpublished data], but this was the only apparent instance of interlocus rearrangement outside these Group 2 IgH.)
Two kinds of probes were used during screening, all of which had been derived from a GL Group 2 bacteriophage clone (Materials and Methods): the vh probe and the intersegmental probes, vd2, dd2, and dj2 (Figure 2, top). The latter were used to detect the infrequent recombination intermediates in these experiments (Table 1, footnote a; e.g., the Group 2 V-D-DJ configuration is vd2+, dd2+, dj2−). On genomic Southern blots, they proved to be specific for Group 2 only, whereas in contrast, the vh probe cross-hybridizes with nurse shark V gene segments from all subfamilies (see Figure S3). In the following genomic Southern blotting experiments, these probes were used to detect rearrangement globally (vh probe) as well as specifically at three Group 2 IgH (vd2, dj2 probe) in lymphoid tissue DNA.
All the genes encoding nurse shark IgM H chain have been cloned and the functional genes can be classified into five subfamilies, Groups 1–5 (see legend, Figure 3); the VH gene segments share >75% nucleotide identity [24]. The various vh-hybridizing bands in RBC DNA can be correlated with anticipated fragment sizes after BamHI/NcoI digestion (Figure 3, RBC lane in vh panel with map). Although the DNA amounts are similar in RBC and PBL lanes (Figure 3, right, ns3v panel), there are novel vh bands in the PBL sample (Figure 3, PBL lane in vh panel). Compared to RBC DNA, the bands at 1,500 bp and 700 bp in PBL are more intense, and a new band appears at 1,100 bp. These three bands correspond to predicted configurations of rearranged DNA from the various IgH, but mostly from Groups 2 and 4 (Figure 3, 1R-3R in blue). At the same time, the 1.9-kb band encompassing the Groups 2+4 GL gene segments in PBL is ca. 23% less than that of the RBC counterpart (see Figure 3, legend), demonstrating loss of the GL band after acquisition of rearranged configurations.
We observed that the relative amount of DNA rearranged was different between thymus (predominantly T cell) and sIgM+ cells (B cells from PBL). Although the images in Figure 4 are from X-ray films, phosphorimager analyses were performed for a quantitative analysis. We centered our analyses on depletion of the 1.9-kb band because it is a single GL configuration of known genes Group 2+4, whereas “gain” measurements cannot be so clearly resolved. For instance, gain of signal in the 1.5-kb region means a combination of Group 2/4 1R plus nonrearranged GL Group 5, but minus an unknown amount of loss by Group 5 rearrangement.
To obtain a rough idea of the proportion of rearranged IgH in B cells only, the DNA from sIgM+ cells from shark-GR PBL was compared to DNA from its RBC (Figure 4A). The “flow-through” sample is from the population mostly depleted of sIgM+ cells and consists of thrombocytes, granulocytes, and lymphocytes (T cells and some B cells that slipped through). An obvious difference between sIgM+ and the flow-through population is the greater intensity of the 700-bp band in the former (Figure 4A). This band mostly contains 3R species, suggesting that most Group 2+4 rearrangements in B cells are VDDJ.
There is a 19% signal reduction of the 1.9-kb Group 2+4 GL band in the sIgM+ lane. The “flow-through” DNA also contained few rearrangements, as assessed by both loss of GL (8%) and gain of rearranged bands. However, unlike the sIgM+ sample, the “flow-through” was a mixture of cell types, and lymphocytes in PBL can range from 5%–30%.
DNA from shark-JS spleen, thymus, whole blood, and heart were compared. There were rearranged Ig bands in spleen and thymus, but these were not detected in blood or heart DNA (Figure 4B). The frequency of lymphocytes in whole blood is very low (0.02%–0.12%, 1 PBL/250 RBC), and in shark-JS, the heart tissue was bled out. The amounts of DNA in the first three lanes in Figure 4B are similar, and a comparison of the intensities of the 1,500-bp, 1,100-bp, and 700-bp bands between the spleen and thymus samples in Figure 4B suggests that more Ig rearrangements were present in the thymus DNA. Indeed, upon calculation, 60% of the thymus vh-hybridizing GL 1.9-kb Group 2+4 band was depleted.
In summary, in one B cell-enriched sample (sIgM+ cells from PBL), 19% of Group 2+4 genes were rearranged and mostly to VDDJ, whereas in thymus, 60% of Group 2+4 genes were rearranged, mostly to intermediate configurations.
In order to analyze these blotted DNA samples in more detail, we performed hybridizations with probes that detect only Group 2 IgH (G2A, G2B, and pseudogene G2C). The resulting bands can be correlated with Group 2 rearrangement intermediates characterized in Table 1 (Figures 4C, vd2, and 4D, dj2). The 1,500-bp (VD-D-J/V-DD-J) and 1,100-bp (VDD-J) bands detected by dj2 probe in thymus appear to be as intense as the 1.9-kb Group 2 GL signal and reflect the high frequency of these events (Table 1). Again, using the GL band as an internal reference, the Group 2 vd2 signal demonstrates that other Group 2 configurations (V-DD-J/V-D-DJ at 1,500 bp and V-DDJ at 1,100 bp) do exist but are less frequent, consistent with results from Table 1.
All the previous experiments were performed on mixed and purified cell populations, and although we can anticipate the general trend in T cells (many and partial rearrangements) and in B cells (few and completed rearrangements), this remains to be shown at the individual cell level. Single-cell analysis was made possible by previous studies in which all the GL IgH sequences in nurse shark have been characterized [22,23] so that degenerate, universal primers could be synthesized, targeting and detecting only the functional genes. Likewise, it was possible to devise primers specific for each Group, just as Int was specific for Group 2 genes. We first focused on thymocytes. We picked single thymocytes, performed single-cell PCR with the universal primers in a two-stage assay (Materials and Methods) and demonstrated the existence of multiple IgH rearrangements. For controls, an erythrocyte was picked after every two thymocytes. Out of 24 RBC, five failed to amplify and the remaining 19 showed only the GL bands. Of the 48 thymocytes, 44 had a variety of 1R, 2R, and 3R bands. Figure 5 (top) shows the results from the first 18 cells after the second round of PCR with nested universal primers. Other nested PCR was also performed with Group-specific primers to Group 2 (Figure 5, middle panel, and Figure S4), Group 1 (Figure S5), Group 4 (Figure S6), and Group 5 (Figure S7). The summary of these results is shown in Table 2. For the most part, little GL sequence can be detected, except in the RBC controls, suggesting either that most of the IgH had rearranged or that the many rearrangements caused the longer GL fragments to be out-competed. Either possibility is the result of widespread IgH rearrangement in the single thymocyte. The various anticipated rearrangements could be cloned from any thymocyte (Table 2, footnote).
The thymocyte result is in contrast to what we obtained in B cells (Figure 5, top and bottom, respectively). Using the identical PCR conditions and reagents, the PCR performed with the universal primers on surface L chain–positive B cells produced predominantly 3R bands. Moreover, GL bands were also present in almost every one of these samples. When several samples of B cell 3R fragments were analyzed on denaturing gels, they appeared to consist of only one or two species per sample (Figure S8).
We went on to identify the rearranged and nonrearranged IgH in single B cells. Using the Group-specific primers, we performed nested PCR on the first-round products of the single B cells (Figure 6, A amplifications) and found that each B cell carried one or only a few rearrangements. Each 3R band was cloned and the number of VDJ species determined per cell (detailed in Figure S9 and legend). We also amplified nonrearranged GL sequence from each cell by using Group-specific primers directed to the intersegmental regions (Figure 6, B amplifications) and identified the genes in each fragment by restriction enzyme sites. These tests were tailored for the donor, shark-GR, all of whose IgH were isolated and sequenced for this experiment (Figure S10). Table 3 summarizes the results from 13 B cells. The CDR3 sequences of these VDJ are shown in Figure S11; the rearrangements in Table 3 are indicated as out-of-frame (VDJø) or in-frame (VDJ+) or nonfunctional (in-frame but containing stops, VDJn). All 13 B cells contained 3R rearrangements, and one cell (KS23) carried a 2R species as well (Figure 7).
We have shown a remarkable disparity between T cells and B cells in IgH gene configuration. In thymocytes, there are multiple and mostly partial IgH rearrangements per cell. Although we cannot claim to detect every VDJ rearrangement present in a B cell, the many IgH genes that remain in GL configuration support the observation that few IgH were rearranged in a single B cell. Many VDJ in Table 3 are out-of-frame or contain stops, consistent with there being only one functional VDJ per cell.
In one cell, KM13, we found two VDJ that were both in-frame (G1; G4CG) and carried no stops in CDR3 (Table 3), whereas the third VDJ is out-of-frame (G2A). One of the former (G4CG) encodes a CDR3 of 24 amino acids, an aberration among nurse shark cDNA CDR3, which range from 4–17 codons (average 11.6 codons, n = 64) in one study [24] and 7–16 codons in another (adult G4 cDNA, average 11.3 codons, n = 41, W. Feng and E. Hsu, unpublished data). However, the G1 VDJ not only contains a CDR3 of average size (11 codons) but is also the only one that has been hypermutated, and its mutations show evidence of positive selection (National Center for Biotechnology Information [NCBI; http://www.ncbi.nlm.nih.gov/] accession number EU719628). There are eight substitutions, with only those in the CDRs resulting in replacement changes. Three point mutation changes in FR2 and FR3 are synonymous, but the CDR1 point mutation (R to W), and the point mutation (Q to K) and 3-bp tandem mutation (S to R) in CDR2 all result in nonconserved changes. Tandem mutations are characteristic of the nurse shark hypermutation process [22], and the frequency of PCR-induced changes after 70 cycles in these studies is 0.14% (13/10,371 bp), or less than one change per 400-bp VDJ fragment. We do not know whether the VDJ with the 24-codon CDR3 encodes an IgM protein, but it is clearly not part of the selection process acting on this hypermutating B cell.
Perhaps, considering their very different CDR3 sizes, there is L chain preference for one polypeptide enabling its expression. We then ask, how often do two rearrangements result in similar CDR3? There are four cells (KM5, KM13, KS3, and KS23) in which more than one VDJ is present although most of these are nonfunctional. The junction sizes range widely. The number of nucleotides between TGT in the VH flank and TGG in the JH flank are 34 bp/45 bp in the KM5 VDJ, 39 bp/44 bp/78 bp in KM13, 24 bp/59 bp/72 bp in KS3, and 33 bp/41 bp in KS23 (Figure S11). With six flanks trimmed and three sites for N region addition per VDJ, it seems unlikely that any two VDJ in a B cell, even if both are potentially functional, would have such similar CDR3 sequence content and loop sizes that they would combine equally well with the available L chain. Thus, constraints operating at two levels—the combination of the random nature of V(D)J rearrangement and L chain compatibility—serve to enforce H chain exclusion.
We propose that rearrangement ceases with the production of a successful H and L chain combination. There are few partially rearranged IgH present in B cells, as the 2R in KS23. Here, the constellation of in-frame (presumed functional) G4 VDJ, the out-of-frame G2A VDJ, and the partially rearranged G2A 2R allele suggest that there was a signal for cessation of rearrangement for the G2A in VDD-J configuration once a viable μ protein was generated.
Ig transcripts from functional and nonfunctional rearrangements can be cloned from B cell-containing shark-JS lymphoid tissue using Int/JH2; we found that the use of a primer in leader intron selects for Ig transcripts unspliced in this region, the majority of which are from aberrant (out-of-frame, partially rearranged) genes. The 3R (VDDJ) sequences were obtained from spleen cDNA, and many were mutated regardless of whether they were productive VDJ or not. Of the 17 2R events we cloned, two were VD-DJ, and one of them carried several mutations in the V region although not in the D-D intergenic sequence. Of 15 independent VDD-J clones, nine were mutants, of which seven contained substitutions throughout V and the D-J intergenic sequence. The mutation patterns are typical of the type previously described in shark Ig, consisting of point and tandem mutations [26]. One such example, A36, is shown in Figure S12.
In contrast, there is very little Ig mRNA in shark-JS thymus, as observed by northern blotting (Figure 8), whereas these and other probes for nurse shark L chain isotypes detect abundant mRNA in spleen and epigonal organ. TCR β chain is abundant in thymus RNA. Reverse transcriptase PCR (RT-PCR) experiments using Int/JH2 to detect Group 2 2R in thymus cDNA were negative (unpublished data). Given the extent of thymic IgH rearrangement described in the preceding section, we conclude that if Ig transcription does occur in precursor T cells, the RNA species are at extremely low levels.
As the IgH rearrangements in thymus were a surprising observation, we investigated whether Ig L chain genes were also active in any way. The nurse shark L chains are encoded by three isotypes, NS3, NS4, and NS5 [27]. NS4 is most abundant (about 60 to 70 IgL), consists of both rearranging and germline-joined loci, and contributes about 90% of the L chain cDNA clones; neither NS4 nor the germline-joined NS3 could be detected in thymus RNA (Figure 8). In NS5, there are four genes, two of which can rearrange; they each consist of one VL and one JL gene segment and one C exon. Whereas somatically rearranged NS5 genomic sequences can be amplified from any source that contains B cells, none was observed in the shark-JS thymus DNA sample (Figure S13). The rearranged NS5 band in the control spleen sample was visually apparent in ethidium gels.
In thymus, few if any NS5 genes somatically rearrange, and certainly not on the scale of the IgH. Thus, like in mouse, Ig rearrangements in thymocytes involve only the H chain loci.
The mechanisms that contribute to generating H chain exclusion—differential chromatin domain activation, locus contraction—have evolved with and are a consequence of the complex mammalian Ig organization. In this study, we have shown that these processes are not necessary to effect H chain exclusion in all vertebrates. Our model, the nurse shark, provides a naturally minimalist IgH locus with four rearranging gene segments. Because rearrangement can be initiated by any gene segment pair, it seems unlikely that the spatially close V, D, and J elements are regulated separately from each other or subject to different chromatin accessibility constraints. Preliminary data from non-Group 2 subfamilies show that rearrangement patterns can vary considerably; for instance, in Group 5, V-DDJ is a prominent configuration that is rarely observed for Group 2 (Tables 1 and 2). Such observations suggest that, once the gene is accessible to recombinase, a preferred order of rearrangement is probably governed by locus-specific factors, for instance, the relative recombination efficiency of particular RSS pairs.
With one possible exception, the 97 1R/2R rearrangements isolated in this study (Table 1) occurred within the minilocus, supporting conclusions drawn from cDNA observations. Long-distance recombination events and sequential chromatin activation do not occur during the shark IgH V(D)J recombination process, demonstrating that in the absence of major aspects of the complex pathways described for mouse allelic exclusion, H chain exclusion will still be managed by limitation of rearrangement.
We established in this report that IgM receptors appear to be clonally expressed in nurse shark and likely all elasmobranch fishes. In one study in the clearnose skate [28] one to three different CDR3 μ junctions were obtained by RT-PCR from single cells. Unfortunately, most of the 100–200 clearnose skate IgH are not characterized, and a number of them are germline-joined VDJ, which make these results difficult to evaluate. We have classified all nurse shark Ig H chain genes in a BAC library and determined those that are functional [23]. Our PCR primers target these genes only. We found ten functional H chain genes in the individual shark-GR and detected in its B lymphocytes one to three VDJ rearrangements per cell. At best, only one VDJ per cell was potentially functional. The other IgH were nonproductive VDJ or in GL configuration.
We believe that most elasmobranch B cells express one dominant H chain mRNA and one IgM receptor. Eason and coworkers [28] hypothesized that one gene is activated at a time, like in the multigene olfactory receptor system. A mechanistic connection seems unlikely, in the absence of an evolutionary relationship between genes encoding Ig superfamily and seven transmembrane domain proteins. From our studies, it appears that either a few IgH loci are rearranging at the same time in the pro-B cell or there are sequential “tries” before a viable H chain protein is generated. The answer is possibly in between.
Partially rearranged 1R and 2R configurations do exist in B cells as best demonstrated by cloning of mutated cDNA (Figure S12) and the 2R species detected in B cell KS23 (Figure 7 and Table 3). The relic incomplete rearrangement configurations in B cells might suggest a feedback mechanism that functions with staggered initiation of rearrangement among loci. Alternatively, a few IgH are fully activated to rearrange, more or less simultaneously; hence the infrequent laggard 2R in the population. In such a scenario, there would be a limited but clear possibility for allelic inclusion. That we do not find many such examples suggests that the probability for two viable rearrangements is low, and as illustrated in the case of KM13, L chain preference DNA might permit only one H chain polypeptide for the receptor. As in mammals, ongoing shark IgH rearrangement probably ceases with the formation of a functional VDJ and expression of the IgM receptor. If L chain rearrangement occurs subsequently, the H chain loci might be transiently inactivated, as occurs with the non-expressed allele in mouse [20]. We have speculated that H and L chain rearrangement occur simultaneously in shark [29], and all rearrangement ceases with the formation of a viable cell surface receptor. However, there currently is no experimental evidence favoring either possibility.
The question remains, how are 15 or 100 IgH loci to be regulated if more than one gene can be activated per cell?
In point of fact, genetically manipulated model systems with more than two H chain genes have been studied. In interspecies hybrid tetraploid and triploid Xenopus [30] and in mice triallelic for IgH [31] allelic exclusion of H chain was observed, despite the increased number of potentially competing genes. There is no reason to believe that in these animals Ig expression is regulated any differently than their diploid version. If that is the case, H chain exclusion is initiated by nonsynchronously occurring rearrangement, and it does not matter how many available genes there are. It is generally accepted that the crucial step differentiating two alleles or multiple genes should be at V to DJ stage [32]. However, in the shark, there is no such second stage; the asynchrony must occur at the initiating step of rearrangement.
Liang and coworkers [33] inserted a GFP reporter into the kappa locus to mark its activation and found that the gene was transcribed at an unexpectedly low frequency in pre-B cells. They suggested that allelic exclusion at the kappa locus is based on probabilistic enhancer activation. Possibly a predetermined allele preference [34,35] contributes to the initial choice, but it was also suggested [33] that a competition for transcription factors would forestall activation of the second gene.
We observed few but mostly fully recombined IgH per shark B cell and propose that there are limiting amounts of trans-factors that target a gene for highly efficient, processive rearrangement, such that however many IgH genes are in the genome, recombination in B cells does not commence at the same time at more than one location. The focused activity at a few IgH also may have the effect of draining other components from general use. Since shark IgH genes lack the usually well-conserved upstream octamer motif [22,36], their trans-factors must differ from and are not competed for by L chain genes if they rearrange at the same time. The first compatible and viable H and L chain combination forming a receptor will generate the feedback signal. If by chance more than one viable H chain is produced at the same time, they may be differentiated by their ability to pair with the available L chain.
The surprising finding in these studies is V(D)J recombination at multiple IgH loci in every thymocyte, and despite the numerous H chain rearrangements present, Ig transcripts are not detected. The majority of thymic IgH are left incomplete as 1R or 2R, further underlining the difference of their estate from that in B cells. Since these IgH genes are not transcribed as in B cells, despite the extensive rearrangement, and are mostly not fully recombined, essential components are obviously lacking in thymocytes. Taken altogether, we propose that in those thymocytes which are in the process of actively recombining their TCR genes also harbor IgH in a rearrangement-permissive state, and this is possibly a prelude to full activation of the chromatin, which can only be achieved in the presence of B lineage-specific components that would include IgH transcription factors. Since cDNAs of rare, aberrant rearrangements of Ig V gene segments to TCRγ have been observed in a shark thymocyte cDNA library (M. Criscitiello and M. Flajnik, unpublished data), we conclude that factors capable of binding the Ig promoter (and perhaps eliciting local chromatin remodeling [37]) could be present in thymocytes.
Most recently, transcription has been shown correlated with rearrangement competence and induction of chromatin changes [11]. One commentary [38] speculated on the connection between transcription, chromatin remodeling, and recruitment of RAG, pointing out that RAG2 contains a methyl lysine-binding region that may act as a reader for the histone code of the chromatin and thus may act differentially depending upon the pattern of the histone modifications. It is currently thought that the formation of a Ig/TCR promoter-enhancer holocomplex, consisting of a complex of nuclear factors-DNA interaction, directs the chromatin remodeling and DNA modifications that promote chromatin interaction with RAG [39,40]. We propose that the limited number of rearranged IgH per shark B cell is a result of infrequent formation of the holocomplex, which contains lineage-specific factors. These ideas are summarized in Figure 9.
In the absence of B cell-specific factors participating in this holocomplex, IgH in shark precursor lymphocytes may still achieve alternative states of accessibility that are not optimal but not prohibiting for unregulated rearrangement. We propose that a quasi-activated level of chromatin accessibility can exist, supports interaction with RAG, and has distinguishable characteristics.
Shark-JS, -GR, -J, -Y, -BL, and -PI (G. cirratum) were captured off the coast of the Florida Keys and maintained in artificial seawater at approximately 28 °C in large indoor tanks at the National Aquarium at Baltimore. Shark-GR was 7 y of age at the time of bleeding. Whole blood was obtained from the caudal sinus and passed through a Ficoll gradient to separate PBL from RBC. Shark-JS was about 5–6 y of age when sacrificed, and its organs were harvested and frozen. Shark-PI was 3–4 y of age. The thymus was dissociated, passed through a cell strainer mesh (Falcon 2235), and subjected to magnetic cell sorting (see below). DNA and RNA were obtained from PBL and frozen tissues using routine procedures. DNA can be extracted from the RBC, which are nucleated.
There are three Group 2 genes, G2A–C, formerly called V2–4 in [18] (NCBI accession numbers DQ192493, DQ192494, and DQ857389). Mismatches in the primers caused the pseudogene (G2C) not to be amplified in this study. Group 2-specific primers used to detect recombined DNA of G2A and G2B include oligonucleotides targeting the leader (V18–1) or the leader intron (Int, 5′-ATTCAGCAATCAGATAAT-3′). These were paired with primers detecting sequence 3′ of D1 (RSS-D1, 5′-GAATGAGGATGTCGGTAT-3′) or in JH (JH2, 5′- TCACGGTCACCATGGT-3′). Primers for the intersegmental sequence between the two D genes (IntDD-F, 5′-GACGATTCAGAACATAGC-3′) and the unique 5′end of the G2-V2 Cμ2 (V18C2–3′, 5′-CGGAGGGTCACCGTTTCC-3′) detected transcripts from partially rearranged Group 2.
The names of the probes are in lower case (e.g., ns3v probe to the NS3 L chain V region gene, ns4c probe to NS4 L chain C exon). The vh probe (vh: V18–1 5′-ACCAGAATGACGACGATG-3′ and V18–2 5′-GTCTTCGATCTTCAGGC-3′, 461 bp), although derived from a Group 2 sequence, cross-hybridizes with all nurse shark VH [18]. Locus-specific probes can be obtained by using the intersegmental sequences, which are relatively nonconserved among the H chain subfamilies. Probes (Figure 2) for the V-D and D-J intervening DNA (vd2: primers IntVD-F 5′-GTACATTGCACCGTAAAC-3′ and IntVD-R 5′-CGCTCATTCTCTGTTC-3′, 352-bp PCR product; dj2: IntDJ-F 5′-ACAGTGCAGTGTTTACT-3′ and IntDJ-R, 5′-TCACGGTAAATCGTCATC-3′, 239 bp) that were generated from the bacteriophage V18 [22] carrying a G2A gene will detect only the three Group 2 loci, G2A, G2B, and G2C (Figure S3). A probe to the conserved cDNA Cμ membrane sequence was also obtained (mem: mem1, 5′-GATTCGATAGATCACACT-3′ and mem2, 5′-AAACAGGACTGATTGTAT- 3′, 216 bp).
Probes to the three nurse shark L chain types NS4 [41], NS3 [26], and NS5 [29] were described previously, and probe names specify whether they detect the V or C sequence (ns3v, ns3c, etc.) ns3v hybridizes to the germline-joined VJ genes of the nonrearranging NS3 L chains [26] and used to standardize DNA on genomic Southern blots because the position of the 3.7-kb band did not overlap with any of the H chain probes. Nurse shark TCRβ C region was cloned from genomic DNA (TCRB-CF, 5′-TCACCAGCAGAGCTGAGA-3′; TCRB-CR, 5′-ATACAGGATGCTCTTGCA-3′). Shark nucleotide diphosphate kinase (ndpk: NDK-F, 5′GGTAACAAGGAACGAACC-3′; NDK-R, 5′-AAAGTTAGTTTATTGTAG-3′) was cloned using PCR primers derived from the available sequence (accession number M63964) [42].
The blots were subjected to autoradiography, and signal intensities of bands were quantified using a Storm 860 phosphorimaging system with ImageQuant software (Molecular Dynamics).
For IgM+ selection, the buffy coat was resuspended in a mixture of shark IgM-specific mAbs (CB5, CB11, and CB16; [43]), and then with goat-anti-mouse IgG Microbeads (Miltenyi Biotec). Approximately 1.5–5 × 107 cells were collected after two rounds of column purification (Miltenyi Biotec). The negative population was collected as the “flow-through” from the first round of magnetic activated cell sorting (MACS). The positive cells were small and round (lymphocyte-like) cells, whereas the negative population contained cells of different shapes and sizes. Thymocytes were mixed in medium containing a mAb specific for nurse shark NS4 L chain C region (LK14; [44]; E. Hsu, unpublished data), and the L chain-negative cells were collected as “flow-through” from the MACS LS column.
Cells were collected after magnetic cell sorting, and RBC were obtained from the same individual for negative controls. Single cells were picked by hand under an inverted microscope with a finely drawn microcapillary pipette (Fisherbrand, #21-164-2G). MAC-sorted lymphocytes were picked, alternating with RBC from another dish; the pipette was rinsed three times in between. The cell was deposited in a 1–1.5-μl volume in shark PBS; 5 μl of lysis solution (1× PCR buffer, 10 mM DTT, 0.5% NP40) was added, topped by mineral oil, and the tubes were heated at 65 °C for one minute to break the nuclear membrane. The tubes were stored at −20 °C until needed. One hundred microliters of 1× PCR solution with dNTP, 0.5 units AmpliTaq (Roche) and primers targeting the VH and JH sequences of Groups 1–5 (two 5′ primers: 20% GR1, 5′-GTTTCTCTACCTCAGCAAT-3′ and 80% GR2–5, 5′-GTTAGTCTMCCTCTGGAAT-3′ with the 3′ primer JH5, 5′-TCACIGTCACCATGGT-3′) were added and the reactions run for 39 cycles at 95 °C 1 min, 58 °C 1 min, 72 °C 1 min, and in the 40th cycle the elongation step was prolonged to 15 min. In the nested reaction, one microliter of the PCR products was added to 50 μl of a second mixture containing two 5′ primers (20% VG1, 5′-AAGGTGTCCAATCGCAA-3′ and 80% VG2–5, 5′-AAGGTGTCCAGTCGGAG 3′) with the 3′ primer JH6 (5′-TCACCATGGTYCCTTGT-3′), and this reaction was run for 20–30 cycles at 95 °C 1 min, 54 °C, 1 min, 72 °C 1 min; again the elongation step was prolonged in the last cycle. The DNA patterns were identical for 20, 25, and 30 cycles. The universal 5′ primers used in the first PCR round are located in the leader intron whereas the nested universal 5′ primers are in FR1 of the VH, about 60 bp downstream.
There are two sets of nested Group-specific reactions used to analyze the B cells, one set to identify the 3R fragments observed above, the other to ascertain which IgH remained in GL configuration. For both, the first-round PCR samples were subjected to ExoSAP-IT (USB) to remove remaining “GR” primers. Six microliters of the PCR sample was incubated with 2 μl of ExoSAP-IT for 15 min at 37 °C, followed by inactivation for 15 min at 80 °C. One microliter of the product was used in a 50-μl PCR reaction. For nested reactions to ascertain VDJ identity, the 5′ primers targeted unique, Group-specific sequences in the leader intron, up to 15 bp downstream of the GR primers (G1: Fam1, 5′-AATGTAAAAGACTCAGCC-3′ used at 58 °C; G2: Int, at 58 °C; G3: GR3N2, 5′-TCATGGATTTTTTCATCT-3′ at 54 °C; G4: Fam4, 5′-AATCATTTCATCAGTAAC-3′ at 54 3 °C; G5: Fam5, 5′-GGCTCAGGATTCATTTCG-3′ at 54 °C). In combination with the JH6 primer, 3R products of 400–440 bp were amplified.
The Group-specific primers for GL configuration targeted intersegmental sequences in V-D and D-J. Both 5′ and 3′ primers are specific for the Group, and PCR products of 1.1–1.2 kb were obtained at 58 °C: G1 (G1DF: 5′-CTGTGCAAAAAGCCACG-3′, G1JR: 5′-TGTCCCCAGTGATCAAG-3′; 1,226 bp); G2 (FD2–1: 5′-CACTTTGTACATTGCACC-3′, RD2: 5′-AATAACTGGCTCTGCACG-3′; 1,154 bp); G3 (G3DF: 5′-AACAATGGCTGGACACG-3′; G3JR: 5′-CCCCAGTTACCGAAGTC-3′; 1,242 bp); G4 (G4DF: 5′-ACCACAGAACGAGGAAG-3′; DR3/4: 5′-GCAAAACAAAATCACGAC-3′; 1,143–1,147 bp); G5 (G5DF: 5′-AACAACGGGTGGACCCG-3′; G5JR: 5′-TTGTCCCCAGTAACCGG-3′; 1,224 bp). The cycling parameters for all nested reactions were the same, except for the annealing temperatures.
|
10.1371/journal.pcbi.1002186 | Local Orientation and the Evolution of Foraging: Changes in Decision Making Can Eliminate Evolutionary Trade-offs | Information processing is a major aspect of the evolution of animal behavior. In foraging, responsiveness to local feeding opportunities can generate patterns of behavior which reflect or “recognize patterns” in the environment beyond the perception of individuals. Theory on the evolution of behavior generally neglects such opportunity-based adaptation. Using a spatial individual-based model we study the role of opportunity-based adaptation in the evolution of foraging, and how it depends on local decision making. We compare two model variants which differ in the individual decision making that can evolve (restricted and extended model), and study the evolution of simple foraging behavior in environments where food is distributed either uniformly or in patches. We find that opportunity-based adaptation and the pattern recognition it generates, plays an important role in foraging success, particularly in patchy environments where one of the main challenges is “staying in patches”. In the restricted model this is achieved by genetic adaptation of move and search behavior, in light of a trade-off on within- and between-patch behavior. In the extended model this trade-off does not arise because decision making capabilities allow for differentiated behavioral patterns. As a consequence, it becomes possible for properties of movement to be specialized for detection of patches with more food, a larger scale information processing not present in the restricted model. Our results show that changes in decision making abilities can alter what kinds of pattern recognition are possible, eliminate an evolutionary trade-off and change the adaptive landscape.
| Animals differ in how they sense and process information obtained from the environment. An important part of this information processing is used to find food. In terms of foraging, local decision making determines how successful individuals are at finding food on longer timescales. Using an artificial-world model, we studied different kinds of decision making to understand how local information processing affects larger scale behavioral patterns and their evolution. We compared a restricted decision making (less memory) to extended decision making (more memory). We then compared the evolution of decision making and behavioral actions (moving and scanning for food) in patchy and uniform environments. Our results show that with restricted decision making individuals face a trade-off in the patchy environment: they try to stay in patches by not moving forward too far, but to do so they sacrifice how fast they travel between patches. With extended decision making this trade-off completely disappears because decision making allows moving forward to be avoided in patches. Instead moving forward can be used exclusively for faster traveling between patches and for selecting bigger patches. Our results show how changes in local decision making can significantly alter what evolutionary forces are faced and can eliminate evolutionary trade-offs.
| The evolution of behavior is to a large extent the evolution of information processing [1]–[4]. On short timescales individuals respond to local information in the environment. For instance in foraging, a basic local information processing is that animals detect food, turn and move to food, and eat. On the long term this generates behavioral patterns. The latter shapes how individual behavior relates to patterns in the environment (e.g. resource distributions) and affects aspects of Darwinian fitness (e.g. foraging success). At present it is poorly known how local information processing mechanisms (e.g. cognition) determine larger scale pattern detection and evolve [3], [5]–[8]. Here we study the evolution of local information processing and orientation to the environment, and its relation to environmental pattern detection.
In evolutionary theory on foraging, the focus is often on how well individuals match (fitness relevant) patterns in the environment. In optimal search theory (OST) the main focus has been on what kinds of random turning strategies optimize search [9]–[11]. A second focus has been on the value of alternating between intensive searching, once a food patch is found, to extensive search when food has not been found for a while, using combinations of correlated random walks differing in turning rates [12]. Simulations show that such switching between search strategies can enhance foraging efficiency because it concentrates search effort in the right places (i.e. it allows patches to be “detected”), so called area-concentrated search. This is true for models in which “continuous” patchy environments are assumed [12], [13], where resource items are only locally detectable, but aggregated on a scale that is beyond the perception of individuals, as apposed to models in which discrete and fully detectable patches are assumed (e.g. the marginal value theorem [14]).
Random-walk models have been used to statistically characterize animal movement trajectories, including bi-modal search patterns similar to area-concentrated search [15], [16]. However, such model fitting does not necessarily reveal underlying movement mechanisms [6], [17]. Interaction with, and orientation to, the external environment can generate similar movement patterns as those generated by internal turning strategies [6], [17], [18]. Moreover, Benhamou showed that local orientation via memory of where an individual last found a food item, can further improve foraging efficiency relative to “random” area-restricted search without such memory [19], indicating the adaptive value of reacting to external cues. However, like the random-walk search models, an important assumption is that food is detected and consumed on the same range. Instead, if food can be detected beyond the range at which it can be eaten (as is often the case), an animal will be able to approach foraging opportunities from some distance via direct visual cues. This is probably one of the most simple ways through which animals can orientate themselves relative to food. Important is that such opportunity-based adaptation (or responsiveness) stands in direct relation to feeding opportunities in the environment. Therefore, on longer timescales, behavioral patterns emerge that are “a reflection of complexity in the environment” [20].
To conceptualize how interaction of individuals with the environment can structure behavior, Hogeweg and Hesper [21] coined the TODO principle. This envisages behavior as multi-scale information processing [22], [23] (see Figure 1): (i) TODO: individuals behaviorally adapt to local opportunities by “doing what there is to do”, and (ii) Pattern formation and detection: behavioral patterns self-organize on larger spatio-temporal scales through the continual feedback between behavior and local environmental contexts (This use of the term “information processing” differs from that in behavioral ecology where it generally refers only to individual-level behavioral flexibility, often specifically in relation to energy-dependent behavioral choices). A simplistic example of TODO is that as food density declines individuals end up moving more and eating less, because there is no opportunity to eat. As such, the environment is like a “behavioral template” to which individuals can respond, allowing individuals to effectively “detect” patterns of opportunities in the environment beyond their own perception.
In order to fit models to movement data and elucidate underlying mechanisms, requires a thorough understanding of how both internal and external structuring of behavior can generate foraging patterns. This can be done using pattern oriented modeling [24] and other multi-level modeling approaches [25], where model fits are evaluated based on patterns on multiple levels: small scale movement decisions, mesoscale patterns such as trajectories and space use and more global patterns such as population distributions. The requirement of fitting models to multiple levels places the focus on the mechanisms that generate the inter-relation between small-scale processes and patterns on larger scales. A thorough understanding of how small scale behavior interactions generate behavioral patterns through TODO could be an important contribution to such modeling approaches.
Essentially, TODO and the longer term behavioral patterns it generates, come to expression (in models) when individuals interact with the environment and need to make behavioral decisions based on local information. In this light, Hogeweg [26] showed that foragers with simple TODO rules could forage much more efficiently than those with much more complicated rules. This was because foragers with simple rules could react to local opportunities and therefore automatically adapt to larger-scale patterns in the environment (i.e. generalize their behavior). More counter-intuitive and complex behavioral patterns emerge in models with more detailed environmental structure and multiple types of behavior. Examples include “self-structuring” explanations for social dynamics in bumblebee colonies [27], grouping patterns in chimpanzees [28], diet learning and cultural inheritance in group foragers [29], [30].
At present, the role of pattern recognition through TODO is most likely underestimated in most approaches to the evolution of foraging behavior. For instance in OST the simple orientation mechanism of turning and moving to food is generally not included. Moreover, behavior is usually assumed to be continuous in that movement, search and food consumption occur in parallel (although a trade-off between movement speed and search accuracy is often assumed [12]). Decision-making is therefore restricted to changes in direction. However, if movement, scanning for food and eating are at least partially mutually exclusive, then individuals must decide about what to do next (e.g. search again at a certain location, or move on). Such foraging behavior can be referred to as pause-travel [31], or intermittent search [7]. Here we focus on local orientation towards food in such a setting where individuals must make decisions, and study the role of TODO in the evolution of simple foraging behavior. We ask: how does local information processing evolve in order to determine how individuals “do what there is to do”? More specifically, how does the responsiveness and orientation of individuals to feeding opportunities in the environment evolve in light of the larger spatio-temporal pattern recognition that this generates?
To address this question, we study the evolution of foraging behavior in a model with individuals that have to choose amongst alternative behavioral actions according to information they obtain through searching. This happens in a spatial environment with patchy and uniform patterns of feeding opportunities. To address how local information processing (sensing and decision making) affects information processing on larger spatio-temporal scales (pattern recognition and genetic adaptation, see Figure 1), we compare the evolution of decision making and properties of behavioral actions in two model variants. In a “restricted” model we limit information individuals can remember and use relative to an “extended” model. The comparison across environments is used to understand evolutionary adaptation to prevailing ecological conditions (patchy or uniform). The comparison across models (restricted versus extended) is used to understand how differences in the evolutionary freedom (or constraints) for evolving decision making affect evolution. This has similarities to artificial neural network approaches to the evolution of behavior, where behavior is not predefined, but emerges from neural architecture and learning processes [32]–[35]. Such models have been used to show, for instance, that risk-averse foraging can emerge as a side-effect of an evolved reinforcement learning process [33]. In our case there is no learning, but the “architecture” of decision making can evolve such that non-predefined behavior can evolve. Therefore we do not prespecify a selection function, but only define that inter-birth intervals decrease with increased food intake, and allow natural selection to arise from competition in a world with finite resources. We then study how Darwinian fitness arises as an emergent property of how micro-scale interactions generate longer-term behavioral patterns. Thus, we study evolution as the interplay of information processing on multiple timescales (Figure 1), based on bioinformatic (processes) theory [22], [23], [36]–[38].
Using this approach, we show that local information processing and opportunity-based adaptation can play a significant role in detecting patterns of resources in the environment, and the evolution of foraging. In particular, we find that the differences in decision making capabilities affect how individuals interact with the environment (TODO), and this can alleviate evolutionary trade-offs and allows for novel pattern recognition specializations.
Our model incorporates (i) individual foragers and (ii) a 2-dimensional environment with resource items in either a patchy or uniform distribution, adapted from van der Post and Hogeweg [29]. Individuals have a decision making algorithm which determines the sequence and context dependency of the following behavioral actions: MOVE, FOODSCAN, MOVETOFOOD and EAT. Each of these behavioral actions has specific properties (such as distances, angles etc). Our model is event-based, which means that actions take time. When individuals complete an action they choose a new one. The individual with the shortest time to complete its action is next to choose a new action.
We study two model variants (“restricted” and “extended”) which differ in the type of decision making algorithm that can evolve. Both the parameters of the decision making algorithm and the details of behavior are “genes” which change through mutation. This generates genetic variation, which may result in differences in foraging efficiency and rates of reproduction. Natural selection then arises from resource competition. For a full list of model parameters please see Table 1 and 2. Next we discuss the model in more detail.
Our environment is 5660 by 5660 lattice, where grid points are scaled to be 1 meter apart, giving 32,035,600 grid points (32.035 km squared). This size was chosen to support a population size (about 100–150 individuals). This was the minimal population where: (i) parameters evolved, (ii) the population is self-sustaining, and (iii) simulations are completed in a reasonable time span. It also ensures that individuals need to move through space to find food, survive and reproduce. Resource items were placed on grid points. Resource items appeared at fixed, but randomly assigned time points within a year, and remained there until eaten. If eaten the resource item was depleted, and appeared again at its fixed time point in the year. Days are 720 minutes (12 hours of “daylight”) and years are 365 days (262800 minutes).
We implement a patchy and a uniform environment, where we keep the total number of food items constant and only vary the resource distribution. In the patchy environment we placed 8000 patches, each with about 2500 items depending on overlap of randomly positioned patches. Each patch is a circle with a radius of 20 meters. Within this circle, 2 resource items are placed at each grid point. All resource items in a patch appear at the same time point, and different patches appear at random fixed times in the year. In the uniform environment resources are placed with probability 0.535 per grid location to match the total number of resources placed in the patchy environment (17150000 items). In the uniform environment, resource items appear at randomly assigned fixed times throughout the year.
The restricted and extended model differ in the decision making that can evolve. Figures 1a and b show the basic decision making algorithms: the behavioral actions that are possible (ovals) and in the case of FOODSCAN, the information this provides (rectangles). Arrows indicate what can be done next, or what information is obtained (after FOODSCAN), and an individuals last action (+ information obtained) represents its “state” (or memory). EAT and MOVETOFOOD can only occur after food is detected. EAT occurs when food is detected in range, otherwise individuals first MOVETOFOOD (MTF) and then EAT. Without any information about food, individuals can either MOVE or do FOODSCAN. As a starting condition, we set these to alternate so that individuals always do FOODSCAN after MOVE and vice versa.
To allow decision making to evolve we define parameters which determine the probability of moving again after MOVE () and scanning again after FOODSCAN () (Figure 2a), searching again after EAT (), or searching again after NO FOOD () (Figure 2b, see also Table 2). This is indicated by decision points (black diamonds) after MOVE, NO FOOD and EAT, where arrows split. For each of these probabilities, the alternative decision has a probability of . For the restricted model we only allow and to evolve, where is a general probability to do FOODSCAN again, irrespective of whether individuals have eaten or did not find food (Figure 2a). Thus in the restricted model, the probability to do FOODSCAN again after EAT or after NO FOOD, is determined by the same parameter (). For the extended model we allow , , and to evolve (Figure 2b), where , and can be seen as context dependent forms of . In the extended model, the probability to do FOODSCAN again after EAT or after NO FOOD, can therefore evolve independently. Thus, in the restricted model individuals cannot remember and make use of the additional information “just ate” or “didn’t find food” to determine the probability to do FOODSCAN again, while in the extended model they can. Moreover, in the restricted model, we assumed individuals always MOVETOFOOD when food is out of reach. In the extended model we allowed this probability () to evolve, and it always evolved to (see section 2 in Text S1 and Figure S1).
The parameters of specific behavioral actions determine how individuals move and sense their environment (see Figure 2c). Unless stated otherwise, we allow all these parameters to evolve:
where is the area scanned (), and where 1 second of scanning for 1 gives . The closest detected item is chosen for consumption. If there are multiple items equally close, a random closest item is chosen. This scanning algorithm therefore represents the case where individuals eat the first item they find. Note also that we assume that MOVE and FOODSCAN cannot occur at the same time, and thus we focus pause-travel foraging [31] or “intermittent search” behavior [7].
Individuals gain energy through food ( energy units per item) which is added to their energy store (with a maximum: ). To survive, individuals must have energy (), which means energy intake must compensate basal metabolism (, which is subtracted from every minute). Because resources become locally depleted individuals must move to eat. We do not add explicit movement costs, but time spent moving cannot be spent eating. Individuals reproduce when . Energy is then halved and the other half goes to a single offspring. The time taken to get back to defines a birth interval. Individuals with shorter birth intervals achieve greater lifetime reproductive success. Individuals can die with a probability of 0.1 per year, and can reach a maximum age of 10 years. This adds some stochasticity in survival and limits lifespans to 10 years. Since resources are limited in the environment, the population grows until the reproduction is at replacement rate (carrying capacity).
Our model requires that the population is viable in relation to resource availability, thus energy and life-history parameters are chosen such that at low population sizes individuals can definitely gain sufficient energy to reproduce. Moreover, to focus on movement and foraging in differently patterned environments, we set the energy required to give birth in relation to energy per food time, and the density of food items in space, such that individuals have to move to and forage from many food patches and experience the full scale of environmental patterns during a reproductive cycle (i.e. they cannot complete reproductive cycles within a single patch). Lifespan is set to allow multiple reproductive events per individual. We expect most parameter combinations that satisfy these qualitative relationships (see section 1 in Text S1 for more detail), to give similar results.
When individuals reproduce, the parameters of decision making and behavioral actions are inherited by offspring, with a probability of mutation of 0.05 per gene (this rate of mutation was chosen after observing that natural selection lead to consistent evolutionary change with increases in foraging efficiency). We allow all action durations, distances and angles to evolve except and . The mutation “step” is defined by drawing the parameter value from a normal distribution with the mother’s parameter value as mean and standard deviation scaled to about 20% of the range of values that is relevant for that parameter (see Table 2). Moreover, in order to keep simulations running fast enough, we limited the minimal action duration to seconds. Most mutations are close the mother’s parameter value, but larger jumps are possible. This was chosen to make evolution of parameters possible without predefining their ranges.
We cannot predict what parameter settings are viable and take a “zero” state (all parameters zero) as initial condition. To make sure the population does not die out initially, we use a birth algorithm in which the non-viable population is maintained at a minimum of 10 individuals, and let it evolve to a viable state. During this time, if the population drops below this minimum then an individual is chosen to reproduce according to a probability () relative to its energy ():(2)
Energy costs of reproduction and energy of offspring as the same as before. Once the population grows above 10 individuals and becomes viable, this algorithm is not used anymore. At this point the population grows to carrying capacity and becomes stable.
For our study we used the following types of simulations:
We find that in both models the population evolves to environment specific attractors. We refer to these evolved states as “specialists”: uniform specialists in the uniform environment, and patch specialists in the patchy environment. These four specialists differ from each other and these differences depend on the following parameters: (i) probabilities to SEARCH again (, , ), (ii) probability to MOVE again (), (iii) MOVE distance (), (iv) turning angle (), and (v) FOODSCAN angle () (see Figure 3). For ease of reference we name the specialists and summarize their distinguishing features as follows (illustrated in Figure 4). Parameter values shown are means of ancestor traces between year 800 and 900 (see also Table S1):
Further analysis revealed that variation of both probability to repeat move () and turning angles () did not impact food intake significantly. For both parameters we found that evolved values result from evolutionary drift because of a very flat adaptive landscape (for more detail see Text S1 section 2 and Figure S1 and Text S1 section 4 and Figure S3 and S4). Moreover, other parameters did not differ between specialists: durations evolved to minimal values (see section 2 in Text S1 and Figure S1) and food scan range () converged to between 2–2.5 meters (see sections 2 and 3 in Text S1 and Figure S2). From here on we focus on those parameters that generated differences in foraging efficiency between the specialists, namely: , , , and . We use the means of evolved parameter values to characterize each specialist (see Table S1 for a complete list of average evolved parameter values).
The values of the evolved decision making parameters mean that in the extended model decision making evolves to: always do FOODSCAN after EAT, always MOVE after NO FOOD ( and , Figure 4c and d). This generates a clear differentiation of behavior in food and non-food contexts (Figure 4c and d, blue and yellow loops respectively). Thus in a food context individuals continue to do FOODSCAN until they no longer find food (blue loop). This generates efficient FOODSCAN - EAT - FOODSCAN - EAT sequences and allows systematic depletion of resources at a given location. During this time any movement is via MOVETOFOOD when food is out of range, always towards food. Only when no more food is found do individuals MOVE. Thus in a “no food” context, individuals switch behavior and no longer repeat FOODSCAN (yellow loop).
In the restricted model only the patch specialist (R-Patchy) has a certain degree of repeated scanning for food (, Figure 4a). However this happens equally after EAT and NO FOOD, because differentiating behavior relative to FOOD and NOFOOD is not possible. This specialist therefore can only to a certain extent avoid MOVE in the presence of food, and is more limited in generating time efficient FOODSCAN-EAT sequences and to only MOVETOFOOD when food is beyond REACH. In contrast the uniform specialist (R-Uni) of the restricted model never repeats FOODSCAN (Figure 4b). It only searches once per location and generates MOVE - FOODSCAN - EAT or MOVE - FOODSCAN - MOVETOFOOD - EAT sequences.
For behavioral actions the most obvious difference between the specialists is that between the patch specialists of the different models (illustrated in Figure 4a and c). R-Patchy’s maximum FOODSCAN angle in combination with its short move distance leads to a behavioral pattern with a large overlap in areas searched after each MOVE. In contrast, Ext-Patchy’s smaller FOODSCAN angle with long move distance generates a pattern with long distances in which it does not scan, followed by food directed movement when food is detected. The difference between the uniform specialists is more subtle (Figure 4b and d). The shorter MOVE of R-Uni leads to considerable overlap in areas scanned after each MOVE. Ext-Uni’s longer MOVE leads to hardly any overlap in areas scanned after each MOVE.
To qualitatively reveal larger-scale behavioral patterns, we visualize the movement trajectories of all evolved specialists in both environments using ecological simulations (Figure 5) . Most striking is that it is difficult to distinguish between the specialists in the same environment, because they all adapt flexibly to both environments, whether they evolved there or not. This is because all specialists are responsive to opportunities in the environment, and have the same basic TODO (“do what there is to do”): move when there is no food, turn and move to food when out or reach, and stop to eat. In the uniform environment this generates random-walk-like patterns reflecting the random encounters with food. In the patchy environment TODO generates a bi-modal pattern of straight movements between patches and frequent turning and remaining localized for some time within patches. Thus irrespective of genetic adaptations, through (automatic) opportunity-based adaptation all specialists are able to generalize their behavior to an environment in which they did not evolve.
The large-scale behavioral patterns of individuals reflect patterns of feeding opportunities in the environment (patchy or uniform). The more accurate this reflection, the better individuals “detect” resource patterns, and this affects their foraging success. An individual’s genotype determines how it responds to opportunities in the environment, and we find that the genetic adaptations of specialists increase their foraging success relative to the environment they evolved in (Figure 6). Overall, differences in food intake rates of evolved specialists, as measured in ecological simulations, are as follows:
(Figure 6a).
(Figure 6b).
where represents a minor difference, and a large difference.
In both environments, specialists from the extended model are the most successful foragers. Interestingly, Ext-Uni is not only the best forager in the uniform environment, but the second best in the patchy environment. In the uniform environment, Ext-Uni has about 9% greater food intake than R-Uni (this difference is significant: Wilcoxon rank sum test, , . For Ext-Uni: ; ; . For R-Uni: ; ; ). In the patchy environment, Ext-Uni has on average about 11% lower food intake than Ext-Patchy (this difference is significant: Wilcoxon rank sum test, , . For Ext-Uni: ; ; . For R-Patchy: ; ; ). However, Ext-Uni has nearly 2 times greater food intake than R-Patchy, even though it did not evolve in the patchy environment (unlike R-Patchy). In contrast, Ext-Patchy is the least successful forager in the uniform environment, although average food intake is only about 3% lower than R-Patchy (but this difference is significant: Wilcoxon rank sum test, , . For Ext-Patchy: ; ; . For R-Patchy: ; ; ). Overall, differences in the patchy environment are greater (2 fold versus a 1.5 fold maximum difference in the uniform environment), indicating more room for specialization. To understand these results we look in detail at how changes in decision making and behavioral actions affect food intake.
The difference in decision making capabilities of the two models has a profound effect on the evolutionary landscape. This is most clear in the patchy environment, where the enhanced information use in the extended model allows a trade-off on within- and between-patch behavior to be eliminated. Therefore, while we find that evolved parameters in both patch specialists reflect a tendency to maximize food intake by (i) trying to stay in patches, and (ii) minimizing inter-patch travel, how this is achieved depends on how the underlying decision making capabilities shape the evolutionary landscape.
This is most clearly illustrated with a local adaptive landscape characterization around the evolutionary attractors relative to the probability to search again ( and ) and move distance (). We consider how parameters affect yearly food intake (“fitness”), and how this depends on inter-patch travel, patch visits time (i.e. how much they manage to eat in a patch) and size of patches visited (Figure 7).
The comparison between the extended model (top) and the restricted model (bottom) reveals a significant shift in the location of the adaptive peak (Figure 7a top and bottom, yellow zone), which coincides with evolved parameter values (indicated by black circles). In the restricted model we can understand the location of the adaptive peak (and evolved parameters) in terms of a trade-off between inter-patch travel rate, and patch visit times. As one increases, the other declines (compare Figure 7b and c bottom row). This is because in order to stay in patches (and find food), individuals need short move distances and repeated food scans, otherwise they prematurely leave the patch. However, this slows down inter-patch travel with redundant search. The evolutionary attractor is therefore located where interpatch-travel time and intrapatch-travel time are such that food intake is maximized (Figure 7a, bottom). As a result R-Patchy has the slowest inter-patch travel of all specialists (see section 5 in Text S1 and Figure S5). Moreover, this is also why R-Patchy has such a large food scan angle, because this allows it to “turn back” when it inadvertently leaves a patch (see section 3 in Text S1 and Figure S2), and why it does not evolve repeated moving (see section 4 in Text S1 and Figure S3).
In the extended model this trade-off does not arise. Here decision making allows differentiation of behavior: food scanning is only repeated after eating and does not occur during inter-patch travel (no food encountered). Repeated food scanning can therefore evolve to maximal values, which allows individuals to move systematically from one food item to the next within patches via MOVETOFOOD. This leads to longer patch visit times (Figure 7c top) and enhanced patch depletion. Unlike in the restricted model, MOVE is now used purely for inter-patch travel. Move distance () is then freed from the trade-off between inter- and intra-patch travel because it no longer affects patch visit times. The enhanced decision making in the extended model therefore eliminates the trade-off, allowing both extended model specialists to be more efficient than R-patchy.
As a consequence of the trade-off disappearing, move distance evolves to much longer distances (Figure 3c) because this allows individuals to bias foraging to larger patches (Figure 7d top). (Note that while we implement patches of a fixed size, partial depletion of patches generates smaller patches.) In fact there are two feedbacks which affect that individuals bias their patch visiting to larger patches: (i) by extending patch visiting times, an individual visits on average larger patches longer, and (ii) by reduced scanning for food while moving during inter-patch travel (i.e. due longer move distances) individuals are less sensitive to each food item on their way. Thus they are more likely to find food and stop moving when local resource densities are higher. Effectively this allows individuals to “select” larger patches. Therefore, for the same time spent traveling, Ext-Patchy manages to find on average larger patches and eat more than Ext-Uni (see section 5 in Text S1 and Figure S5 for more detail). Long move distances also generate more neutrality for repeated move and turning angles, allowing them to evolve (see section 4 in Text S1 and Figure S3 and S4).
For the uniform specialists we also find a difference between the extended and restricted model. Both specialists tend to maximize food intake by (i) not wasting time searching depleted areas, and (ii) not moving too far and skipping food items on the way. However, in the extended model food intake peaks at maximal repeated search after finding food, while in the restricted model food intake peaks at minimal repeated search and slightly shorter move distances (Figure 8a, top and bottom respectively).
In both cases, local depletion of food causes that individuals who move further during MOVE, find a greater average density of food during their next food scan (Figure 8b). However, the further individuals move the longer they travel between food items (Figure 8d). By repeating food scans, travel between food items can be reduced because several food items can be eaten at a given location (Figure 8d, see interaction between and ). However, for the restricted model, redundant food scanning (when no food is found) rises quickly with repeated food scanning (Figure 8c, bottom), because FOODSCAN also happens after not finding food. The best option is therefore not to repeat food scanning (and therefore not systematically deplete a given location), but not move too far, as to not miss undepleted food items on the way. In the extended model, repeated food scanning only occurs after eating, and redundant food scanning is avoided, unless individuals do not move far enough (Figure 8c, top). Here the best option is therefore to always repeat food scans, systematically deplete a given location and move somewhat further than in the restricted model, to avoid a larger depleted area.
Overall Ex-Uni is more efficient than R-Uni (Figure 6a). Both are more efficient than patch specialists in the uniform environment (Figure 6a), because these either have too much redundant overlap in search (R-Patchy, due to repeated search) or skip too many resources on the way (Ext-Patchy, due to long MOVE distance) (see section 5 in Text S1 and Figure S6 for more detail).
To further evaluate our results we studied evolution in an intermediate patchy environment (twice as many patches, but half the density of resources) and a mixed environment (half resources uniform half patchy, only with extended model). In the intermediate patchy environment we find that foraging parameters evolve to be qualitatively the same as our main patchy environment both in the extended (parameter averages are: , , , ) and restricted model (parameter averages are: , , , ). This indicates that the behavioral adaptations in the patchy environment are relatively robust to this change in patchiness although the difference in search angles is less pronounced. It is however likely that much smaller patches would select for smaller move distances, because in the mixed environment we find that the extended model evolves to be most similar to Ext-Uni (parameter averages are: , , , ). This makes sense given that Ext-Patchy does much worse than Ext-Uni in the uniform environment compared to the performance of Ext-Uni relative to Ext-Patchy in the patchy environment (see Figure 6). Selection for generalizability in more heterogeneous environments will therefore probably lead to Ext-Uni type genotypes.
Our results show how responsiveness to opportunities in the environment, and the behavioral pattern formation this generates on longer timescales, can play a significant role in the evolution of foraging behavior. This is because the behavioral pattern formation is also a type of pattern “recognition”, i.e. a larger-scale information processing. To illustrate this more explicitly we consider foraging in the patchy environment. When an individual hits a patch, it only detects a single food item. From that position it can find a neighboring food item and move towards it. Through this feedback between sensing and responding to the environment, the individual effectively uses the spatial auto-correlation of positions of food items as a template to move through the patch (see Figure 5). Effectively, “by doing what there is to do” on a very local scale, the individual generates a behavioral pattern that reflects the position of the patch. Because the behavioral pattern determines food intake, it has value in terms of rates of reproduction. Through natural selection, information about rates of reproduction is processed, effectively selecting behavioral patterns that better match, or “recognize” patterns of feeding opportunities in the environment. This drives changes in the population frequencies of genotypes which define how individuals “do what there is to do”. As such, the evolution of local information processing occurs through information processing on multiple timescales: (i) responses to local opportunities, (ii) formation of behavioral patterns and (iii) natural selection based on behavioral pattern formation (as illustrated in Figure 1).
Our comparison of extended and restricted decision making reveals that decision making capabilities determine the specificity with which individuals can respond to opportunities in the environment and the types and accuracy of pattern recognition. Specificity is greater in the extended model, where individuals could remember and use the information “found food here last scan” or “did not find food last scan”. This allows the context-dependent responses “always scan for food after eat” and “always move after no food found” to evolve, and behavioral differentiation between food and non-food contexts (Figure 4). In the restricted model this was not possible and individuals were less able to characterize local contexts when deciding to scan again: they only had the information that they had scanned, but not what the outcome was. The behavioral differentiation between food and non-food in the extended model allows systematic depletion of resources at a given location, and a more accurate recognition of patterns in the environment (e.g. patches), which is why the extended use of information evolves. Moreover, because larger-scale environmental patterns are spatial arrangements of local opportunities, greater specificity relative to local opportunities via TODO leads to greater behavioral generalization across environments. As a consequence, Ext-Uni performs better than R-Patchy in the patchy environment, even though only the latter evolved there. This reveals that generalization capacity, which leads to behavioral flexibility on the large-scale, can evolve in individuals via “hard wired” TODO tuned to local variation in foraging opportunities.
The differentiation of behavior in food and non-food contexts in the extended model significantly changes the adaptive landscape (selection pressures) and possibilities for larger scale pattern recognition (Figure 7). In the restricted model, in order to repeat FOODSCAN in patches, individuals also had to repeat scanning for food when moving between patches. Moreover, MOVE could not be avoided in patches. This lead to a trade-off on within- and between-patch behavior (Figure 7b and c, bottom). In the extended model, due to behavioral differentiation, MOVE is only used in non-food contexts, and repeated scanning only occurs in a food context. As a consequence there is no trade-off (Figure 7b and c, top), and MOVE is dissociated from selection pressures in the food context. Instead MOVE can become specialized for inter-patch travel, generating a refinement in larger-scale pattern recognition in order to detect a sub-pattern: patches with more food. This is achieved by reducing the responsiveness to opportunities for feeding when in the “no food” behavioral pattern, and to switch to highly responsive behavior once food is detected. In this way extensive and intensive search are generated. Thus we observe that the “modularity” of behavior (the two behavioral modes in food and non-food context generated by TODO), provides evolution with structure in which it can generate new specializations (the adaptation of MOVE) and new forms of larger scale information processing (detection patches with more food).
Much of foraging theory focuses on foraging efficiency, and uses optimality predictions to assess the foraging behavior of animals (e.g. optimal foraging theory [4], [39], and optimal search theory [9]–[11]). Foraging optima are often specified relative to constraints (e.g. body size, morphology, mode of locomotion, information processing abilities) [4]. However, this does not necessarily give insight into why a species faces particular constraints, since “constraints” are also often evolvable. At present little is known about how constraints arise and change in the evolution of behavior, though presumably this has been a driving factor in the evolution of morphology and information processing abilities (e.g. sensing and cognition). If we assume that in our model the change from restricted to extended decision making represents an evolutionary innovation in information processing, our results show how small evolutionary changes in decision making can lead to a “release from constraints” on a larger scale and shift the system to a new local optima (i.e. going from the bottom to top landscape in Figure 7a). This reveals how the inter-relation between local information processing and larger scale behavioral patterns allows a small increment in memory (i.e. remembering the outcome of a previous search event) to generate a cascade of consequences: (i) differentiation of behavior, (ii) altering the adaptive landscape and eliminating trade-off constraints and (iii) allowing novel foraging specializations.
Such insights are relevant for studying the evolution of cognition, which is likely to involve changes in constraints and behavioral opportunities [40], [41]. Moreover, in light of evolving cognitive complexity our model provides a useful reference. For instance, to establish the impact of elementary spatial cognition such as “remembering where one last found food”, it is probably more appropriate to use TODO-based patch detection as a baseline, rather than random-walks (as in [12]), if individuals can orientate towards food on a local scale without memory. This is also true in terms of model fitting to data to establish mechanisms used by animals during movement. An interesting study by Morales et al. [35] used a spatial grid based model to study movement behavior in elk, assuming that individuals perfectly know the vegetation state of 8 neighboring cells around an individual’s location, and know with less accuracy the state of cells 1 and 2 steps further. Their results show interesting similarities to movement patterns in real elk, and like in our study, shows how orientation to cues in the environment structure movement patterns. However, given the relatively coarse grained resolution of their lattice (28.5 by 28.5 meters), their model does not allow for smaller-scale processes via local visual cues, but assumes spatial cognition. In principle it is possible that if food availability patterns traverse the larger scale grid boundaries of Morales et al.’s model, TODO-based processes could allow individuals to move from grid cell to grid cell according to food availability without using spatial memory. The point here is not to claim the elks couldn’t use spatial memory, but that pattern recognition via TODO could be underestimated. To address this requires models and data with a greater spatial resolution.
Our results also have implications for understanding extensive and intensive search behavior. First, we show that a bi-modal search pattern easily self-organizes from TODO in patchy environments in all evolved specialists whether they evolved there or not. This bi-modal pattern is not an evolved strategy, but simply a reflection of the environment. Bi-modal movement patterns are therefore the default expectation in patchy environments. Secondly, in terms of the extended model, we show how a simple mechanism generating extensive and intensive search modes can be created by evolution. Here there is a difference with the model of Benhamou [12], where bi-modal search is assumed as an adaptive strategy, and studied as a combination of random walks. We find that the regulation of switching between extensive and intensive search does not evolve as a specific strategy in the patchy environment, because it also evolves in the uniform environment (Ext-Uni also shows intensive and extensive search). Instead we find that the specific adaptations in Ext-Patchy function to refine the self-organized extensive and intensive search in order to enhance a new kind of pattern detection: implicitly finding larger patches. This latter pattern detection is not usually considered in foraging theory, but may play an important role in foraging success.
Given the focus of optimal search theory on internally-driven turning strategies [6], [7], [9], [11], it is surprising that we do not find any significant evolution of turning angles. This suggests that in some cases externally-driven turning behavior may pre-empt any need for internally-driven turning strategies and that opportunity-based orientation towards food may be an under-represented aspect in this field [6], [17]. Moreover, we show how individuals can generalize their behavior across environments via TODO, while fixed internally-driven turning strategies are less robust because they need to be specified to a given environment. However, our results depend on the fact that individuals can detect food items from beyond their reach. This may often be the case in animals, but not always, especially if food items are very cryptic. Moreover, given our simplistic implementation of turning behavior, and other model assumptions (e.g. random turning at environment boundary, intermittent searching), more work is needed to specifically address the relationship between internally- and externally-driven turning.
In terms of the evolution of behavior, the value of our results lie in revealing how small changes in decision making and memory have profound influences on multiple scales relevant for individuals foragers. Clearly our foragers are simplistic (especially cognitively) and therefore it is unlikely that the local optima we find are directly relevant for a given animal species. However, we show that TODO can be a means through which animals could detect larger-scale environmental patterns, which should be taken into account. Moreover we find that extensive search modes can be used to implicitly detect larger food patches in the environment. These findings can be useful to consider when modeling foraging processes and its fitness consequences. Thus our results provide a useful baseline for understanding the evolution of behavioral flexibility and how evolutionary changes in cognition can alter trade-off constraints and adaptive landscapes.
|
10.1371/journal.pcbi.1003282 | Communication and Common Interest | Explaining the maintenance of communicative behavior in the face of incentives to deceive, conceal information, or exaggerate is an important problem in behavioral biology. When the interests of agents diverge, some form of signal cost is often seen as essential to maintaining honesty. Here, novel computational methods are used to investigate the role of common interest between the sender and receiver of messages in maintaining cost-free informative signaling in a signaling game. Two measures of common interest are defined. These quantify the divergence between sender and receiver in their preference orderings over acts the receiver might perform in each state of the world. Sampling from a large space of signaling games finds that informative signaling is possible at equilibrium with zero common interest in both senses. Games of this kind are rare, however, and the proportion of games that include at least one equilibrium in which informative signals are used increases monotonically with common interest. Common interest as a predictor of informative signaling also interacts with the extent to which agents' preferences vary with the state of the world. Our findings provide a quantitative description of the relation between common interest and informative signaling, employing exact measures of common interest, information use, and contingency of payoff under environmental variation that may be applied to a wide range of models and empirical systems.
| How can honest communication evolve, given the many incentives to deceive, conceal information, or exaggerate? In recent work, it has often been supposed that either common interest between the sender and receiver of messages must be present, or special factors (such as a special cost for dishonest production of signals) must be in place. When talk is cheap, what is the minimum degree of common interest that will suffice to maintain communication? We give new quantitative measures of common interest between communicating agents, and then use a computer search of signaling games to work out the relationship between the degree of common interest and the maintenance of signaling that conveys real information. Surprisingly, we find that informative signaling can in some cases be maintained with zero common interest. These cases are rare, and we also find that the degree of common interest is a good predictor of whether informative signaling is a likely outcome of an interaction. The upshot is that two agents with highly incompatible preferences may still find ways to communicate, but the more they see eye-to-eye, the more likely it is that communication will be viable.
| Many theorists have seen communication as a fundamentally cooperative phenomenon [1]–[4]. In an evolutionary context, however, cooperation cannot be taken for granted, because of problems of subversion and free-riding [5]. In the case of communication, these problems include both refusal to share information, and deception, or lying for one's own advantage. If lying is common, there is no point in listening to what anyone says. If no one is listening, there is no point in talking.
In recent work the situation is often sketched as follows: it is easy to see how communication can be viable if there is complete concordance of interests between senders and receivers of signs. Then communication can result in useful coordination and division of labor. There is no mystery about signaling within multicellular organisms, for example, including hormonal and cell-to-cell signaling (although conflicts of interest may arise even here: [6]). In between-organism contexts, the problem of conflict of interest rapidly becomes acute. Special mechanisms are needed to explain how honesty is maintained. The main approach taken in recent years has been costly signaling theory [7]–[9]. Intrinsic costs of signaling prevent dishonesty, by differential expense to liars or differential benefits to the honest.
“Cheap talk” models, where signaling has no costs, have seen some development [10]–[15] but have been minor players in recent years. Here we use a novel method to examine ways that informative signaling can be sustained without cost in a range of situations of partial and low common interest. We use a version of the Lewis sender-receiver model [1], [16], and employ a method of sampling and analyzing cases drawn from a large space of games with different relationships between sender and receiver payoffs. We then offer generalizations based on analysis of the sample of cases. The analysis uses coarse-grained measures of common interest between sender and receiver, and attends also to a feature that interacts with common interest: the degree to which payoffs for an agent depend on different acts being produced in different states, the contingency of payoff for that agent.
We find that using a simple and intuitive measure of common interest based on comparisons of preference orderings over actions, it is possible, though rare, for informative signaling to be maintained at equilibrium with complete divergence of interests. We then construct a more fine-grained measure of common interest, one that is more demanding in its classification of a case as one of zero common interest, and find that informative signaling with zero common interest is possible in this stronger sense as well. Defining an information-using equilibrium as one where the receiver makes use of informative signals to guide behavior, the proportion of games that include at least one information-using equilibrium increases monotonically and rather smoothly with both measures of common interest. (See below, in the Methods section, for the equilibrium concept we use throughout the paper.) We then look at the equilibria that support the highest amount of information use for a given level of common interest, and again find a monotonic, though less smooth, relationship between degree of common interest and maximum information use. A third analysis, looking at the relationship between common interest and contingency of payoff for sender and receiver (defined below), yields more complicated results.
We conclude that informative signaling can be stable in situations of minimal, even zero, common interest. A combination of mixed strategies of signal use by both senders and receivers, and the selective pooling of states by the sender, makes possible the extreme cases of this phenomenon. Pooling alone can suffice in cases where divergence of interests is not so extreme. As interests converge, stability of informative signaling becomes easier to achieve. Our model complements other recent work on the adaptive importance of mixed strategies and partially informative signaling in evolution.
Our modeling framework draws on Lewis [1] and Skyrms [16]. We assume that the world varies exogenously and has three equally probable states (, , ). The sender perceives (without error) the state of the world and responds by mapping states to signals (, , ). The mapping need not be one-to-one as the sender may “pool” some states, treating them equivalently, and the sender may also probabilistically “mix” signals in response to a given state. The receiver perceives (without error) the signal sent and maps signals to acts (, , ), with pooling and mixes possible again. So a combination of sender and receiver rules can be represented as follows:
Sender:
Receiver:
For example, the sender here sends message 1 whenever they see state 1, message 2 whenever they see state 2, and in state 3 they flip a biased coin to send message 1 two thirds of the time and message 3 one third of the time. Both sides receive payoffs as a consequence of the combination of the receiver's action and the state of the world. Sender and receiver payoffs may differ, and can be represented in the form seen in Table 1.
The payoff matrix defines a preference ordering over acts in each state for both sender and receiver. For example, in Table 1, the preference ordering for the sender in state 1 is [>>], and for the receiver [>>]. A simple measure of the degree of common interest in a game tracks how similar the orderings for sender and receiver are, for each state: there is complete common interest when sender and receiver have the same preference ordering over acts in every state, and complete conflict of interest when these orderings are reversed in every state. Between these extremes are various kinds of partial common interest: sender and receiver might agree on the best act in each state, but disagree otherwise; they might always agree on what is worst, but not otherwise; they might agree entirely in some states but disagree in others.
In cases of complete common interest, some consequences for informative signaling are easily seen. With complete common interest, sender and receiver can both receive their maximum payoffs when the sender maps states to signals one-to-one and the receiver uses these signals to guide appropriate actions. This is a signaling system in the sense of Lewis [1], and neither party has any incentive to change what they are doing. This state might not be attained by the selection process shaping sender and receiver behaviors, but if it is reached it is stable [17]. With complete conflict of interest, it would appear that signaling cannot be maintained, as any information about the state of the world carried by signals can be used by the receiver to produce acts contrary to the sender's interests, and any sensitivity to signals in the receiver can be exploited by the sender. Exploring the generality of this phenomenon is one aim of this paper. Another is quantifying the relationship between common interest and informative signaling.
The varieties of partial common interest described above do not form a complete ordering. However, a coarse-grained measure of the overall degree of common interest can be constructed by modifying the Kendall tau distance. This measure describes the similarity in the ordering of the items in two lists, by counting discordant pairs of items across the lists. The first two items in the two lists form a discordant pair with respect to a preference ordering, for example, if in list 1 the first item is preferred to the second item, whereas in list 2 the second item is preferred to the first. We define a measure C of the common interest in a payoff matrix of the form in Table 1 by counting the discordant pairs in the sender's and receiver's preference orderings over acts in each state of the world, and then averaging across states and rescaling the results to yield a number between 0 and 1, where corresponds to complete common interest and corresponds to complete conflict of interest. In response to results outlined below we also make use of a refinement of ; which compares not only the agents' preference orderings of the actions in each state, but also tracks how the agents' payoffs for each action relate to the mean value of the payoffs the agent might receive in that state. (For details see Text S1.) As discussed below, is one among several ways of refining the simpler measure, , and we do not claim it is best for all purposes.
We also make use of a further description of payoff matrices. For each agent, how much does payoff depend on matching different actions to each state of the world? A simple illustration of the importance of this feature is seen in a case where the receiver has the same best act for every state (has a dominant strategy available). Then the receiver can achieve maximum payoff no matter what the sender does, by mapping all signals to that cover-all act. Even if no one act is best in all states, there may be a cover-all act that works well for an agent nearly all the time. This is a within-agent matter. So we define and , also making use of the Kendall tau distance. For each agent, we compare the preference orderings over acts that apply in different states of the world, comparing each pair of states in turn. K is high for an agent with respect to a pair of states if good acts in one state are bad acts in the other state. K for an agent averages all comparisons of states, rescaled to lie between zero and one, where corresponds to the highest degree of contingency of payoff. (For details see Text S1.)
Our aim is to generalize about games with different levels of common interest and contingency of payoff for the agents. The method used is to generate samples from the space of games with three states where sender and receiver payoffs are integers between 0 and 99. Payoffs for each player for each act in a state are chosen randomly, so 18 random choices specify payoffs for a game. We then use the implementation of Lemke's [18] algorithm provided by the software package Gambit [19] to search for equilibria in that game where informative signals are being sent and used. The equilibrium concept used is the Nash equilibrium: a pair of strategies form a Nash equilibrium if neither player can improve their payoff by unilaterally modifying their strategy.
We measure the degree to which agents engage in informative signaling with mutual information, a symmetrical measure of the degree of association between two variables, measured in bits [20, p. 7]. An equilibrium is an information-using equilibrium if there is non-zero mutual information between states of the world and the receiver's acts. We focus on mutual information between states and acts for the following reasons. If there is mutual information between states and acts, the only way for this to arise is for senders to send informative signals and receivers to use these signals to guide variation in their actions to some extent. It is possible for senders to send signals with information about the state of the world that is not used – informative signals that are ignored by the receiver. It is possible also for receivers to guide actions with different signals sent randomly by the sender. The first of these – informative signals that are ignored – is a situation which may be an equilibrium and in which there is informative signaling, but it is not a situation in which the receiver is making use of that information. Our primary focus is situations in which informative signals are both sent and used. This requires that the signals carry information about states and acts carry information about signals. Given that receivers only have access to the state of the world by attending to signals, by the data processing inequality [20, p. 34] it is not possible for acts to carry more information about states than signals do. (States, signals, and acts form a Markov chain.) Any mutual information between states and acts arises from the use by the receiver of information about states in the signals.
Computational methods are described in Text S1 but one feature should be noted here: Lemke's algorithm is not guaranteed to find every equilibrium in a game [21]. So the reports of information-using equilibria below may be under-counts.
To investigate the role of C we generated a random sample from the space of games with three equiprobable states, three receiver actions, and independently chosen payoffs for sender and receiver associated with each receiver action in each state of the world. (Each value of C is represented by 1500 games.) These sender and receiver payoffs are integers between 0 and 99. For each game we asked whether there is at least one information-using equilibrium in that game – an equilibrium with nonzero mutual information between states and acts – and then asked what proportion of games at each level of C have at least one information-using equilibrium. (All these games also have equilibria that are not information-using equilibria). The results are shown in Figure 1.
Very low degrees of C suffice to enable information-using equilibria, but at low C levels, only a small minority of games do so (unless the algorithm used has significant bias). As C increases, the fraction of games with information-using equilibria increases monotonically.
The curve in Figure 1 does not reach 100% for the case of complete common interest. Some games with are games with zero and . (When , K is the same for sender and receiver.) The same act is best in every state. Around 1/9 games with will also be . In such a game, the receiver can always take the system to an equilibrium by mapping all signals to the same, optimal, act. Then there is no mutual information between states and acts, regardless of what the sender is doing, as there is no variation in acts.
Surprisingly, a small number of games with , where sender and receiver have reversed preference orderings over acts in every state, have information-using equilibria. Table 2 shows a case of this kind – not a case from one of our samples, but a simplified case constructed using the computer-generated cases as a guide.
Despite zero , the game in Table 2 has an information-using equilibrium, whose sender and receiver rules are as follows:
Sender:
Receiver:
The mutual information between states and acts at this equilibrium is 0.67 bits, where the highest possible value for a game with three equiprobable states (a Lewisian signaling system) is 1.58 bits.
A feature of the case in Table 2 is that although sender and receiver have reversed preferences in every state, in they share a second-best outcome () that is almost as good as their best. This is ignored by our measure , and it is one kind of common interest between the two agents. A way to modify C that takes this factor into account is to compare, across sender and receiver, their preference orderings over both the payoffs that arise from different actions and also the average of the payoffs for that agent in that state. This is done by defining a “dummy act” for the receiver in each state, an act that secures for each agent the mean of the other payoffs possible in that state. This dummy act and its payoff are then included in the determination of each agent's preference ordering over acts in that state; the two agents might agree, or disagree, for example, about whether the payoff of Act 1 is higher than the mean of their payoffs possible in that state. , like , counts discordant pairs of preferences and is scaled to lie between 0 and 1. (For further details see Text S1). yields a similar relationship between common interest and the proportion of games with an information-using equilibrium to that seen in Figure 1.
The game in Table 2 has a nonzero , as sender and receiver agree about how one of their second-best outcomes compares to their means for that state, so is a more demanding criterion for complete conflict of interest. Even in this stronger sense, though, it is possible for a game to have an information-using equilibrium with complete conflict of interest. A case of this kind, also one modeled on a less transparent computer-generated case, is shown in Table 3. This game has the following information-using equilibrium:
Sender:
Receiver:
In all the cases with and/or with information-using equilibria we have found, the underlying pattern is as follows. Two signals are used by the sender and three acts are used by the receiver. In one state the receiver produces an act that is intermediate in value for both sides. In the cases in Tables 2 and 3, this is . The receiver is prevented from shifting to their optimal act for this state by the fact that the signal sent in that state is ambiguous, and is sometimes also sent in a state for which the act that might “tempt” the receiver in would be very bad. In another state, the receiver mixes their actions between optimal acts for each side. (This is in both Tables 2 and 3.) Again, the receiver is prevented from settling on their optimal act in by the fact that the message the sender sends in that state is ambiguous; state 2 is used by the sender to deter exploitation in the other two states, and in this state all three acts are produced.
In both cases in Tables 2 and 3 the information-using equilbria are very fragile, as either the sender (in 3) or the receiver (in 2) can shift without penalty to a strategy in which the mutual information between states and acts goes to zero. Not all cases of information-using equilbria and zero common interest have this feature, however; sometimes information-use is less easily lost. The lowest level of common interest at which an information-using equilibrium is found in which neither sender nor receiver plays a mixed strategy, probabilistically varying their response to a state or a signal, is (see Text S1 for examples of both phenomena described in this paragraph).
A valuable feature of C is the weakness of the assumptions required for its measurement; C assumes only ordinal, not cardinal, utilities. assumes cardinal utilities. does not, however, assume that sender and receiver utilities are commensurable. If that further assumption is made, the notion of zero common interest can be analyzed instead by requiring that in every state, sender and receiver payoffs sum to a constant and the choice of action determines only how the division is made (a “constant-sum game”). We do not claim in this paper that information-using equilibria exist in constant-sum games. All constant-sum games have , though the converse does not hold. Some constant-sum games have nonzero , on the other hand, and not all games are constant-sum. Due to its simplicity and weak assumptions, in the remainder of the body of this paper we will use C to measure common interest. and constant-sum games are discussed in Text S1.
Once we know how likely a given level of C is to maintain at least one information-using equilibrium, we can also ask what is the highest level of mutual information between states and acts that can be maintained in a game with a given degree of . Figure 2 shows the maximum amount of mutual information between states and acts generated by an equilibrium pair of strategies from any game examined with a given level of . In constructing the pool of cases for this analysis, we have included not just the sample of games used in Figure 1 but also games found in earlier samples.
Figure 2 shows that the highest value for information use grows monotonically with common interest, as expected, but in a step-like way and with quite high values of mutual information between states and acts seen even at the lowest values of . Conversely, our sample includes cases with high values of C and very minimal information use at equilibrium (, mutual information = 0.03 bits; see Text S1).
A further analysis of these cases takes into account the contingency of payoff for sender and receiver, as well as common interest. The importance of this factor has been evident already in some extreme cases. When there is complete common interest but K is zero for both sides, there is no problem for signaling to solve – a single act always delivers an optimal payoff. When there is less common interest, the contingency of payoff for sender and receiver can diverge, and in most cases will be different. Figure 3 charts the proportion of games with at least one information-using equilibrium as a function of both common interest and contingency of payoff for an agent; separate graphs are given for and (left), and for and (right). The sample used for this chart is not the same one used for Figure 1, as a random sample of all games with a certain under-represents some combinations of and . Figure 3 uses a sample in which every combination of and is represented by 1500 games.
As expected, higher values of generate more information-using equilibria than lower values of . A difference is seen, however, between the consequences of low values of and . When the sender's contingency of payoff is very low, the intermediate values of present a local maximum in the proportion of games with information-using equilibria. When is low and is intermediate, will be appreciable. The receiver seeks to vary their actions with the state of the world, and though the sender would ideally like the same act to always be performed, equilibria exist in which a compromise is reached. When the receiver's is low, on the other hand, they can achieve optimal payoffs by mapping every signal to the same act. The receiver can “go it alone” (though information-using equilibria arise in a few cases with high because of ties for the optimal act in a state).
We have given a treatment of the relation between informative signaling and common interest between sender and receiver, in a framework where signal use is associated with no differential costs and no role is given to iteration of interactions between agents. We find that informative signaling is possible in situations where sender and receiver have reversed preference orderings over receiver actions in every state of the world. This situation, where , is one sense of “complete conflict of interest,” and a sense that has been employed more informally in a range of earlier discussions (eg., [22], [23]. In the light of our results, is shown to be a somewhat undemanding sense of complete conflict. We discussed one refinement of , which requires stronger assumptions about payoffs, and found that information use at equilibrium is possible with complete conflict even in this stronger sense, where . Another way to refine the idea of complete conflict, a way that uses still stronger assumptions, is by appeal to the notion of a constant-sum game. We do not claim that informative signaling is possible at equilibrium in constant-sum games. Another way to interpret our results is to suggest that the degree of conflict of interest in a game cannot be analyzed by noting the relationships holding between preferences in particular states, and then generalizing across states. Moving beyond consideration of these extreme values, we find that is a good predictor of the existence of information-using equilibria in the space of games studied in this paper.
We note several limitations of our model. First, the model assumes a particular relationship between sender and receiver, one where the sender has private knowledge of a state of the world, and payoffs result from the coordination of receiver actions with this state. This “state” of the world might be the condition or quality of the sender. Another kind of model assumes that neither side has privileged information about the state of the world, and the role of signaling is to coordinate acts with acts rather than acts with states (the “battle of the sexes,” for example). In further work we hope to extend our analysis to cover these cases. Another limitation involves our use of the Nash equilibrium concept. A Nash equilibrium need not be an evolutionarily stable strategy (because rivals may increase in frequency due to “drift”). In addition, equilibria of this kind may not be easily found by an evolutionary process [17]. Further work is needed to explore the dynamic properties of the games discussed in this paper. Thirdly, our analysis gives no role to the biological plausibility of games.
We close by comparing our treatment with two other papers, one classic and one recent. First, Crawford and Sobel [10] treated agreement in interests as a matter of degree, and found that when interests diverge, honest signaling is possible, but with lower informational content than there would be with complete agreement: “equilibrium signaling is more informative when agents' preferences are more similar.” In their model, the state of the world (sender quality) and the available actions both vary continuously in one dimension, and the difference between sender and receiver interests corresponds to a constant that is the difference between the actions seen as optimal by sender and by receiver in a given state of the world. In their model the degree of common interest across games can be measured exactly, but the model makes strong assumptions about the pattern of variation in the world. Our model makes weaker assumptions in this area, with the consequence that common interest is only partially ordered, motivating the introduction of coarse-grained measures such as C and . Crawford and Sobel found that as agents' interest converge, a larger number of distinct signals can be sent at equilibrium. We found that informative signaling can exist with zero common interest, through a combination of pooling and mixing, though games of this kind are rare and the proportion of games with an information-using equilibrium increases as interests converge. Crawford and Sobel's model also did not allow for variation in , which we find has significant effects on the viability of information use.
Second, Zollman et al. [24] investigated biologically plausible games with two possible states of the world (again, sender quality) that are usually analyzed with substantial differential costs enforcing honesty. These authors found that very small differences in cost or benefit across different types of senders can maintain honest signaling when both sender and receiver mix strategies in a particular way. Senders in one state mix two signals, and senders in another state send just one of those signals. Receivers mix their responses to the ambiguous signal and do not mix their responses to the other. A conclusion from their model is that variation in signal-using behavior within a given situation, on both sender and receiver sides, need not be a matter of mere “noise” but can be an essential feature of an equilibrium state. Our results, within a framework of zero signal cost, lead to a conclusion of the same kind: probabilistic mixing of strategies, along with partial “pooling” of inputs, by both sign producers and sign interpreters can be important in maintaining signaling in situations of low common interest.
|
10.1371/journal.pgen.1000123 | Identification and Dissection of a Complex DNA Repair Sensitivity Phenotype in Baker's Yeast | Complex traits typically involve the contribution of multiple gene variants. In this study, we took advantage of a high-density genotyping analysis of the BY (S288c) and RM strains of Saccharomyces cerevisiae and of 123 derived spore progeny to identify the genetic loci that underlie a complex DNA repair sensitivity phenotype. This was accomplished by screening hybrid yeast progeny for sensitivity to a variety of DNA damaging agents. Both the BY and RM strains are resistant to the ultraviolet light–mimetic agent 4-nitroquinoline 1-oxide (4-NQO); however, hybrid progeny from a BY×RM cross displayed varying sensitivities to the drug. We mapped a major quantitative trait locus (QTL), RAD5, and identified the exact polymorphism within this locus responsible for 4-NQO sensitivity. By using a backcrossing strategy along with array-assisted bulk segregant analysis, we identified one other locus, MKT1, and a QTL on Chromosome VII that also link to the hybrid 4-NQO–sensitive phenotype but confer more minor effects. This work suggests an additive model for sensitivity to 4-NQO and provides a strategy for mapping both major and minor QTL that confer background-specific phenotypes. It also provides tools for understanding the effect of genetic background on sensitivity to genotoxic agents.
| Complex traits often display a range of phenotypes due to the contribution of multiple gene variants. Advances in statistical models, genetic mapping, and DNA genotyping and sequencing have made baker's yeast an excellent system to identify quantitative trait loci (QTL), regions of the genome linked to a quantitative phenotypic trait. We focused on a complex DNA damage sensitivity phenotype in yeast in which parental strains are unaffected but give rise to progeny with a sensitive phenotype. We used a whole-genome approach to isolate defects in DNA repair caused by gene variants in two strains of baker's yeast that display approximately 0.5% sequence divergence. The parental strains are resistant to the ultraviolet light–mimetic agent 4-nitroquinoline 1-oxide (4-NQO); however, a large number of spore progeny displayed varying sensitivities to the drug. Through linkage and bulk segregant analyses we identified one major QTL, RAD5, and two minor QTL linked to sensitivity to 4-NQO, and we provide evidence that sensitivity is due to additive effects involving several QTL. These observations provide a powerful model in which to understand the basis of disease penetrance and how phenotypic variation can be mapped at the gene level.
| Complex traits often display phenotypic variation due to additive and interactive effects of gene variants located at multiple loci. Because of the above complications, only a relatively small number of quantitative trait loci (QTL) have been identified [1]–[7]. Baker's yeast has become an excellent system to model and dissect complex traits. For example, Steinmetz et al. [8], Brem et al. [9], and Perlstein et al. [10] took advantage of the natural variation between wild and laboratory yeasts to dissect the complex phenotypes that underlie high-temperature growth, transcriptional regulation, and drug sensitivity, respectively. Steinmetz et al. [8] mapped a trait present in one parental strain while Brem et al. [9] mapped transcriptional traits in both parents and hybrid progeny by looking for linkage between phenotypic variation and specific molecular markers. Perlstein et al. [10] performed linkage analysis to identify QTL linked to sensitivity to small molecule drugs. Such analyses involved the use of high-density oligonucleotide arrays to analyze a large collection of F2 hybrid progeny.
Complex traits can be identified in one parental background and mapped by crossing the strain to one lacking the phenotype. However, a complex phenotype or disease can be caused by variants of genes derived from different strain backgrounds, resulting in offspring showing phenotypic variation not present in the parents. Such complex phenotypes have been shown to affect the penetrance of human cancers (e.g. [11]–[13]). For example, in rare cases of hereditary non-polyposis colorectal cancer, the disease is thought to result from the combined effect of non-pathogenic DNA mismatch repair alleles [13]. Such phenotypes could be due to impaired protein/protein interactions or other molecular interactions that occur only in the hybrid progeny; alternatively, they could be due to additive effects involving several QTL.
The above observations encouraged us to identify and dissect a complex trait in baker's yeast associated with defects in DNA repair. We focused on such phenotypes because of our interest [14] in understanding how DNA damage repair pathways contribute to maintaining genomic stability (reviewed in [15],[16]). Also, identifying loci that underlie such traits provides a model to explain differences in the penetrance of phenotypes observed in humans (e.g. [3],[8],[17]). We used a whole-genome approach to identify such loci in the meiotic progeny of a cross between the BY (S288c) and RM strains of S. cerevisiae. BY is a commonly used lab strain and RM is a wild isolate from a grape vineyard in California that displays approximately 0.5–1% sequence divergence relative to S288c [9],[18]. We tested RM and BY strains and the meiotic progeny of a BY/RM diploid for sensitivity to a variety of DNA damaging agents. Both the BY and RM strains are resistant to the ultraviolet light-mimetic agent 4-nitroquinoline 1-oxide (4-NQO); however, a large number of spore progeny from the BY/RM diploid showed varying sensitivity to the drug. Through linkage analysis and a backcrossing strategy involving a bulk segregant analysis and SNP genotyping, we identified one major and two minor QTL linked to the 4-NQO-sensitive phenotype. These observations provide a powerful model in which to understand the basis of disease penetrance and how genetic variation can be mapped at the gene level.
To identify novel complex traits in Saccharomyces cerevisiae associated with DNA repair, we examined 123 haploid meiotic progeny of a BY/RM diploid [9] for sensitivity to DNA damaging agents (Tables 1, 2; Materials and Methods). Using high-density oligonucleotide arrays, 2956 genetic markers were identified between the two parental strains that cover over 99% of the genome [9],[19]. Meiotic spore progeny from the BY/RM hybrid were genotyped using the same high-density oligonucleotide array, creating an inheritance map for each of the 123 hybrid spore progeny [9],[20],[21].
We looked for a reproducible DNA damage sensitivity phenotype in the hybrid progeny that was not seen in either parental strain. This phenotype was assessed by plating serial dilutions of saturated cultures from each hybrid segregant onto plates containing varying concentrations of DNA damaging agents (Table 1 and Figure 1). The DNA damaging agents tested included methyl methane sulfonate (MMS), a DNA alkylating agent, 4-nitroquinoline 1-oxide (4-NQO), an ultraviolet light (UV) mimetic, bleomycin, a radiomimetic antitumor drug, and caffeine, a compound that sensitizes cells to genotoxic agents [22]. We identified hybrid progeny that displayed varying sensitivities to 4-NQO, bleomycin, and caffeine treatments (Table 1). Surprisingly, the RM parent displayed sensitivity to MMS, and this sensitivity was observed in half of the progeny (Table 1); however, as shown below this phenotype does not appear to be entirely monogenic.
We performed linkage analysis for each of the phenotypes tested (Figure 2), determined significance cutoffs for each trait by permutation, and calculated support intervals (see Materials and Methods). MMS-sensitivity, which was seen in the RM parent and half of the hybrid segregants tested, showed strong linkage to a chromosome 12 locus near SMF3 (Figure 2A, LOD score of 16.6 for YLR034C_1989). Linkage analysis of 4-NQO-sensitivity identified the same region (LOD score 10.1 for YLR034C_1989, support intervals completely overlap), but in this case the phenotype was seen in the hybrid progeny but not in the parental strains, indicating that 4-NQO-sensitivity involves additional loci (Figure 1B, Table 1). The bleomycin-sensitive phenotype showed linkage to chromosome 2 (LOD score 5.2 for YBR138C_275; Figure 2C). The caffeine-sensitive phenotype also showed linkage to chromosome 2 (LOD score 5.0 for YBR161W_293; Figure 2D). Both loci overlap a region previously linked to growth-related transcripts and daughter cell separation [19]; however, the linkages described here are not likely to be due to the same locus because their support intervals do not overlap. We further pursued the MMS- and 4-NQO-sensitivity linkages because they showed the highest statistical significance.
Twenty-two genes lie within the support interval associated with the MMS-sensitive and 4-NQO sensitive phenotypes. A prime candidate is RAD5, located ∼2 kb away from the peak marker. RAD5 encodes a DNA-dependent ATPase involved in the error-free branch of post-replicative repair [23]. rad5 null mutants are sensitive to MMS, UV, and ionizing radiation [24],[25]. Only one other gene within the region is required for resistance to MMS, AAT2, which encodes an aspartate aminotransferase involved in nitrogen metabolism [25],[26]. It is located on the edge of the region, approximately 12.5 kb from the peak marker. Although required for MMS-resistance, aat2 deletion strains are not UV-sensitive [25].
Based on the above information, we focused on RAD5 as a candidate gene associated with MMS and 4-NQO sensitivity observed in the hybrid progeny. We tested the role of RAD5 polymorphism by homologous allele replacement and plasmid suppression (Figures S1and S2) approaches. RM strains, which are sensitive to MMS, became resistant when the RAD5 gene in RM was replaced with the BY copy. In addition, the MMS-sensitivity observed in RM strains could be created in the BY strain by replacing RAD5 with the RM copy (Figure S1). Replacement of the BY open reading frame of RAD5 with the RM allele also re-created the 4-NQO sensitive phenotype in the BY strain (BY::RAD5-RM; Figure 3A, Materials and Methods). The sensitivity was not observed when the RAD5 gene in the RM strain was replaced with the BY copy (RM::RAD5-BY), consistent with the RM copy of RAD5 being associated with 4-NQO sensitivity. These data indicate that that the RM copy of RAD5 contains polymorphisms that confer the MMS-sensitive phenotype and contribute to sensitivity to 4-NQO. Interestingly, although the MMS-sensitivity phenotype appeared to be monogenic and showed strong linkage to RAD5, the locus does not completely account for the phenotype. Genotyping analysis of the 121 meiotic progeny indicated that although there is a very strong linkage to RAD5, there is overlap in which progeny displaying moderate (−, −/+) sensitivity to MMS fell into both the RAD5-BY and RAD5-RM genotype classes (data not shown).
The RAD5 open reading frames in BY and RM differ by two non-synonymous substitutions that map to amino acid positions 783 (glutamic acid in BY, aspartic acid in RM) and 791 (isoleucine in BY, serine in RM). These polymorphisms both map to the helicase domain of RAD5 (Figure 3B; [23]). Using site-directed mutagenesis and homologous gene replacement methodologies we found that the serine residue at position 791 in RM RAD5 contributed to sensitivity to 4-NQO. This was accomplished by introducing the RAD5-I791S substitution into the BY strain and showing that the resulting strain was sensitive to 4-NQO (Figure 3C, Materials and Methods). This allele was also responsible for the MMS-sensitivity phenotype observed in the RM parental strain (Figure S3). No other strains of Saccharomyces cerevisiae that have been sequenced (http://www.sanger.ac.uk/Teams/Team71/durbin/sgrp/) contain this serine 791 polymorphism. We also phenotype-tested a number of wild yeast strains in our collection and none of them showed sensitivity to MMS equivalent to that observed in the RM parental strain (data not shown). Finally, this polymorphism is not the same as the loss of function mutation that was previously identified in the W303-derived copy of RAD5, G535R [27].
We identified one locus, RAD5, from the RM strain involved in sensitivity to 4-NQO. The fact that the parental strains were resistant to 4-NQO indicated that at least one locus from BY was also required. We searched for additional loci by partitioning the segregants by their genotype at RAD5 and performed linkage analysis on the subgroups, but there were no regions showing a significant linkage. This negative result could simply reflect the reduced sample size of each subgroup.
A candidate gene approach was also pursued to identify additional factors from the BY parent involved in the sensitivity phenotype. Because we were able to recreate the 4-NQO hybrid sensitivity phenotype by replacing RAD5 in the BY strain with the RM copy, we used the BY::RAD5-RM strain to search for candidate RM genes that could suppress the 4-NQO-sensitive phenotype. We pursued this approach because we were able to suppress the 4-NQO-sensitive phenotype of the BY::RAD5-RM strain by introducing a single-copy ARS CEN plasmid containing RAD5-BY (data not shown). Single-copy plasmids containing the entire open reading frame and ∼250–500 bp of downstream and upstream sequence from the RM parent strain were created and transformed into the BY::RAD5-RM strain (Materials and Methods). Plasmids were made containing several candidate genes that were shown to physically and/or genetically interact with RAD5. These include POL30 and RAD18, whose gene products interact with RAD5 through two-hybrid analysis [28],[29], and CTF18, CSM3, RAD50 and TOF1, mutations in which are synthetically lethal with rad5 mutations [30]. None of these candidate genes derived from the RM parent strain suppressed the 4-NQO-sensitivity of the BY::RAD5-RM strain (Figure S4; data not shown).
A backcrossing strategy was pursued to identify a subset of BY loci linked to 4-NQO-sensitivity. To isolate these loci we backcrossed the BY::RAD5-RM strain to the RM parent. Because both parental strains contained the RAD5-RM allele, all progeny from this cross have the potential to show sensitivity. BY::RAD5-RM was crossed to the RM parental strain and a sensitive spore from this cross was backcrossed to the RM parent (3D-BK1, Figure 4). This backcross was repeated two more times (progeny labeled BK2 and BK3 for second and third backcross, respectively). After each backcross, haploid hybrid progeny were phenotype tested and segregants displaying sensitivity to 4-NQO were chosen and backcrossed to RM. After the first backcross, roughly 35% of the progeny displayed the 4-NQO-sensitive phenotype. Even after the third backcross a 2:2 4-NQO sensitive: resistant segregation pattern for each tetrad was not seen, suggesting that a large number of loci modify sensitivity to 4-NQO. At the end of backcrossing, the parental contribution of the final segregants was approximately 94% RM and 6% BY.
To determine if multiple genetic loci were segregating in the second backcross, two 4-NQO-sensitive segregants from the second backcross (∼12.5% BY and ∼87.5% RM) were mated and sporulated. Incomplete complementation (varying degrees of sensitivity) of the 4-NQO-sensitive phenotype was seen in progeny from this cross and in progeny from other crosses involving second backcross segregants (Figure 4; data not shown). Thus, multiple loci that contribute to the hybrid 4-NQO-sensitive phenotype are likely to be segregating in the second backcross because all the progeny from this cross display some degree of sensitivity (Figure 4).
Bulk segregant analysis followed by microarray SNP genotyping was used to identify loci that conferred 4-NQO sensitivity in the third backcross lines. Genomic DNA was isolated from pooled resistant and sensitive segregants (40 from each set). The two pools were then hybridized to an Affymatrix reference sequence microarray that allows for the determination of inheritance between the two parental strains. Mapping was then performed (Materials and Methods; [31],[32]). If a locus is unlinked to the phenotype, it will not show preferential inheritance in either pooled sample and therefore will show an averaged baseline of inheritance (Figure 5C). For a linked locus, the genomic DNA in one pool will be enriched for one parental inheritance, and therefore will show a peak of inheritance when comparing the array results across the pooled progeny of the phenotype (Figures 5A and B). This method allowed us to screen a large number of segregants and measure the average genotype of progeny within each pool. From this analysis, two regions were identified that segregated preferentially with the 4-NQO phenotype (Figure 5). These regions are on chromosome 7, between 433,546 bp and 551,683 bp (smoothed value of >1), with the peak at 463,019 bp, and on chromosome 14, between 437,935 bp and 473,640 bp (smoothed value of >1.5), with the peak at 467,209 bp (Figure 5A and B). The regions mapped are large, approximately 118 kb and 36 kb in length for chromosome 7 and 14, respectively. This may be due to the backcrossing and the nature of pooled segregant analysis, but also could be due to several genes within each region contributing to the phenotype. We focused on candidate genes located near the peak of each region.
For the chromosome 7 linkage region, the peak includes KAP122, which encodes a protein involved in importing the TOA1-TOA2 complex into the nucleus. This gene has been implicated in regulation of pleiotropic drug resistance [33], making it an excellent candidate to test for suppression. A second candidate gene, PDR1, maps to within 6 kb of the peak and was chosen because it also encodes a protein involved in the pleiotropic drug response [34]. Due to the large size of this region (>100 kb), we focused our analysis on these two genes. For the chromosome 14 linkage region, the peak lies near MKT1, which was named for it role in the maintenance of K2 killer toxin [35], and has been implicated in the posttranscriptional regulation of HO endonuclease [36]. This region has also been implicated in high temperature growth, sporulation, and expression quantitative traits [3],[37],[38],[39]. Sequence analysis of the RM and BY strains indicated that KAP122, PDR1, and MKT1 each contain non-synonymous and synonymous polymorphisms within their coding regions.
KAP122, PDR1, and MKT1 containing ∼500 bp of upstream and downstream sequence were cloned from both the RM and BY strains and inserted into ARS CEN plasmids (Table 3). These plasmids were introduced into 4-NQO-sensitive spores derived from the third backcross of BY::RAD5-RM×RM (3DBK3-12 and 3DBK3-52, included in the bulk segregant analysis). None of the KAP122 and PDR1 plasmids suppressed the 4-NQO phenotype of these segregants (data not shown). However, MKT1 derived from the RM parent suppressed the 4-NQO phenotype of these segregants and other segregants from the third backcross (Figure 6; data not shown). Neither of the parental derived alleles of MKT1 conferred increased sensitivity or resistance when inserted into a backcrossed segregant that did not show the 4-NQO-sensitive phenotype (3D-BK3-16, Figure 6).
As a further test, we deleted the MKT1 gene present in the third backcross of BY::RAD5-RM×RM (3DBK3-12 and 3DBK3-52) and introduced into these strains the ARS CEN plasmids bearing the RM and BY derived MKT1 alleles. As shown in Figure 6, these strains displayed the same phenotype as was observed for the plasmid suppression in the parental third backcross strains, indicating that the MKT1-BY allele contributed to 4-NQO-sensitivity. We attempted to test suppression of the 4-NQO-sensitive phenotype in the backcrossed strains using homologous replacement of MKT1; however, we were unable to do so because the selectable marker used for integration interfered with MKT1 function.
Because we did not see a significant LOD score for the MKT1 locus in the original linkage analysis, we suspected that additional alleles are contributing to sensitivity to 4-NQO in the original BY/RM hybrid progeny. To test this, we looked for allele-specific suppression by MKT1-RM in the original BY/RM meiotic spore progeny that showed sensitivity to 4-NQO (Figure 7). Some of the original hybrid progeny showed allele-specific suppression of 4-NQO-sensitivity (segregants 16 and 27, genotype MKT1-BY and RAD5-BY) while others did not (segregants 20 and 29, genotype MKT1-RM and RAD5-RM). In a resistant segregant, no phenotypic difference was seen between the plasmids (segregant 52, genotype MKT1-RM and RAD5-BY). These observations are consistent with multiple genes contributing to the 4-NQO sensitivity phenotype. Finally, to determine whether the MKT1 and RAD5 alleles show non-additive interaction with respect to sensitivity to 4-NQO, we plotted the semi-quantitative phenotype for 4-NQO shown in Table 1 against the RAD5- and MKT1-region genotypes (Figure 8). This analysis shows that the loci do not strongly interact (see Discussion) and that the most extreme genotypes are recombinant: the RAD5-RM, MKT1-BY genotype is the most sensitive, while the RAD5-BY, MKT1-RM genotype is the most resistant. This is consistent with what we observed in the backcrosses where the MKT1-BY allele was associated with sensitivity to 4-NQO in a RAD5-RM background (Figure 6).
We identified and dissected a complex trait, sensitivity to the DNA damaging agent 4-NQO, in hybrid progeny of baker's yeast. Through linkage mapping of genotyped hybrid progeny, we identified a major effect locus, RAD5, and identified the causative polymorphism. Using a backcrossing strategy along with microarray-assisted bulk segregant analysis we identified other QTL that contribute to the variability seen in the 4-NQO-sensitivity phenotype. The identification of a major effect locus likely contributed to our ability to map minor effect loci for sensitivity to 4-NQO.
Determining the number of loci that underlie a quantitative trait can be difficult due to epistasis, penetrance, and environmental factors that influence the phenotype. The 4-NQO- sensitive phenotype was seen at varying levels in about a quarter of the hybrid RM/BY progeny tested, indicating a minimum of two contributing loci. Backcrossing and bulk segregant analysis indicated that at least two other loci, including MKT1, contribute to a lesser degree to the sensitive phenotype observed in the hybrid progeny. The number of modifier QTL is likely greater than two due to the lack of suppression observed in some of the F1 hybrid progeny by MKT1 (Figure 7).
After backcrossing, we were able to identify a large QTL region (∼118 kb) located on chromosome 7; however, a candidate gene approach was unsuccessful. There are many difficulties in finding functional variants in large QTL. Steinmetz et al. [8] identified two QTL in yeast (chromosome XIV, 52 kb, and chromosome XVI, 8 kb-not pursued further) that mapped to a heat resistance trait. Using reciprocal hemizygosity, they identified three genes in the chromosome XIV QTL linked to this phenotype. Using a backcrossing approach, Deutschbauer and Davis [38] identified four QTL in yeast (30-71 kb) linked to sporulation efficiency. They also performed reciprocal hemizygosity analysis and identified three genes linked to the phenotype. Finally, Perlstein et al. [10] mapped cell growth in the presence of 83 small molecule drugs to 219 QTL (42 kb on average) that localized to eight main regions in yeast. In their analysis they were able to identify two loci, primarily by candidate gene testing, that were linked to a particular drug sensitivity. Each of the above examples illustrates the complexity of mapping QTL to individual genes and the need for candidate gene testing. In our studies the large size (∼118 kb) of the Chromosome VII QTL suggests that candidate gene testing is likely to be very time consuming and more importantly, that the QTL is likely to be complex. The extended nature of the locus, despite extensive backcrossing, points to the possibility of a cluster of closely linked polymorphisms within the region causing the phenotype, making the identification of the individual genes difficult.
We found a major effect QTL in RAD5, and a minor effect modifier locus, MKT1. The RAD5 gene was a logical candidate gene within its QTL because it is required for resistance to a number of DNA damaging agents and is a component of an extensively studied DNA repair pathway [25],[28],[29]. In contrast, MKT1 has not been associated with any DNA damage pathway but has been identified in multiple QTL mapping studies performed in yeast, including sporulation and high temperature growth [3],[8],[37],[38]. In these studies, the same polymorphism in MKT1 (D30G) was associated with the phenotype. Sinha et al. [3] hypothesized that BY derived MKT1 is a loss of function allele and Deutschbauer et al. [38] showed that the MKT1 polymorphism in the BY strain is rare; it was not found in 13 other S. cerevisiae strains. This same allele is likely to be involved in sensitivity to 4-NQO.
It is unclear how the RAD5 and MKT1 alleles interact to create the 4-NQO-sensitive phenotype in the hybrid progeny. The observations that these loci do not strongly interact each other, and that spore progeny displayed a wider range of sensitivity phenotypes, suggest an additive model for sensitivity to 4-NQO (Figure 8). In such a model, each parental strain contains negative and positive alleles, making them appear similar to each other. Segregation of such alleles in progeny would be expected to yield a wide range of phenotypes, as was seen. In this model, the 4-NQO-sensitive effect is transgressive rather than reflecting a defective interaction between the MKT1 and RAD5 gene products. Consistent with the above argument is the fact that genome-wide analyses of the response to DNA damaging agents (deletion, global expression studies) have not shown a direct connection between RAD5 and MKT1 that could explain the 4-NQO-sensitive phenotype with respect to known DNA repair pathways [40]–[42].
It is important to note that many QTL interactions cannot be easily explained in terms of an established genetic pathway. For example, for high temperature growth, the genes MKT1, END3 (involved in endocytosis) and RHO2 (a non essential GTPase) have been identified as QTL; how these three genes interact to confer heat resistance is unclear because no other genetic interactions involving these genes have been identified [3],[8]. MKT1 has been shown to contribute to 4-NQO sensitivity, high temperature growth, and sporulation phenotypes. The fact that a single modifier could be involved in such a variety of phenotypes suggests that a candidate gene approach that tests previously identified modifiers should be considered when searching for loci that underlie a complex trait.
The RAD5-RM polymorphism conferred varying degrees of sensitivity to 4-NQO in RM/BY hybrid progeny. The scientific literature contains numerous examples in which an allele confers a more severe phenotype in one genetic background relative to another (e.g. [9],[14],[43]). In addition, environmental and background effects have been shown to affect the penetrance of many cancers (e.g. [11],[12]). Such effects are thought to be due to DNA sequence differences at multiple genetic loci that lead to molecular incompatibilities between gene products, between gene products and cis-acting sequences that function in specific pathways, negative epistatic interactions uncovered by haploinsufficiency, or as shown here, additive effects involving multiple QTL. Identifying such complex traits in a genetically tractable system is of great interest because they provide testable models to study disease penetrance (e.g. [3],[8],[17]).
S. cerevisiae parental strains, BY and RM, and BY/RM hybrid segregants (Table 1) were tested in this study [9]. Additional strains used in this study are listed in Table 2. Yeast strains were grown in yeast extract/peptone/dextrose (YPD), minimal complete or minimal selective media [44]. When required, nourseothricin (Werner Bioagents) or G418 sulfate (Cellgro) were included in YPD at 200 mg/l [44],[45]. Sporulation plates and procedures were as described previously [43],[46]. EAY253 (rad52Δ::LEU2, ura3-52, leu2Δ1, his3Δ200) was used as a control strain in the 4-NQO, MMS, bleomycin, and caffeine plating assays.
Plasmids used in this study are shown in Table 3. All of the plasmids are derived from the pRS vector series [47]. pRS317 was modified to contain the Gateway cloning cassette (Invitrogen). Each gene was amplified from BY or RM genomic DNA [44] using Pfu turbo polymerase (Stratagene). Primer sequences used to amplify these genes are available upon request. Each PCR product, which contained the entire open reading frame and at least 250 bp of upstream and downstream sequence, was gel purified and cloned into pENTR/D-TOPO entry vector (Invitrogen). The structure and sequence of all entry clones were verified by restriction endonuclease digestion followed by DNA sequencing. Subcloning of each gene from the pENTR/D-TOPO vector into pEAA403 (LYS2, ARSH4, CEN6, aatR1::CmR-ccdB::attR2) was performed via LR recombination (Invitrogen). All of the resulting constructs were expressed via their native promoters. Plasmids were transformed into each strain using standard methods [48] and were selected for on lysine minimal dropout plates.
The RAD5 integration vectors, pEAI209 (RAD5-BY::NatMX) and pEAI210 (RAD5-RM::NatMX) contain RAD5 amplified from BY and RM genomic DNA, respectively [44]. The sequences were amplified using Pfu turbo polymerase (Stratagene) and primers AO990 (5′CAGGACACTGACAACGAATTGC) and AO991 (5′GTTTGCGTTAGAGCAATTCC). The PCR amplified product containing the entire RAD5 open reading frame plus 450 bp upstream and 700 bp downstream sequence was digested with PstI and SalI and inserted into the corresponding sites of pUC19. The entire PCR fragment was DNA sequenced. Using overlapping PCR and subcloning, BamHI and NotI sites were added 30 bp downstream of the RAD5 stop codon using AO946 (5′GAGAAAGAGCTAACTCATACTT), AO1023 (5′CGACTAGTGCGGCCGCTAGTCGGGATCCAAAGTCTTTATATATGAGTATG), AO1024 (5′CGACTAGTGGATCCCGACTAGCGGCCGCATTTATTATTATTTTCAACC) and AO616 (5′CGCCATTCAGGCTGCGCAACT). The NatMX gene from pAG25 [49] was then inserted into these BamH1 and NotI sites. Integration plasmids were also made that contained KanMX downstream of RAD5, pEAI207 (RAD5-BY::KanMX) and pEAI208 (RAD5-RM::KanMX), using the same procedure. All RAD5 point mutations were made in pEAI207 using the QuickChange XL Site-Directed Mutagenesis protocol (Stratagene, USA). A fragment containing the point mutation was then subcloned into unmutagenized pEAI207 and sequenced to determine that only the desired mutation was created.
For homologous replacement of RAD5, integration plasmids pEAI209, pEAI210, pEAI213, and pEAI214 were digested with XbaI and NheI and the fragments were transformed into BY and RM using standard methods [48]. Integrations were selected on YPD media containing G418 sulfate or nourseothricin [45]. Each allele was PCR amplified and sequenced to determine that only the desired mutations were created.
Two strains from the third backcross of RM×BY-RAD5-RM, 3DBK3-12 (EAY2295) and 3DBK3-52 (EAY2298), were transformed with a mkt1Δ::NatMX DNA fragment to create EAY2317 and EAY2323, respectively (Table 2). This fragment contains NatMX flanked by 50 bp of MKT1 sequence upstream of the MKT1 ATG and 50 bp of MKT1 sequence downstream of the MKT1 stop codon. It was created by PCR amplifying pAG25 with primers AO2014 (5′TGAACTATAAAGTACTAAAGGCAGAAAAATTAATAGCAAATTAAGCGATGCGTACGCTGCAGGTCGAC) and AO2015 (5′TGCTTTTTAAATAGTTCCACTATTTCCATCATA CTCATTCTCACGCTTCAA TCGATGAATTCGAGCTCG). Integrations were selected for on YPD media containing nourseothricin and the mkt1Δ::NatMX mutation was confirmed by PCR.
RM, BY, and hybrid segregants were grown to saturation in YPD liquid media. The cultures were then diluted in water and spotted in 10-fold serial dilutions (undiluted to 10−5) onto YPD media containing 0.03% MMS (vol/vol) (Sigma), 0.25 µg/ml or 0.30 µg/ml 4-NQO (Sigma), 10 mM caffeine (Sigma), or 10 µg/ml bleomycin (Sigma) [22]. Plates were photographed after a 2-day incubation at 30 C. Each segregant was scored according to growth on non-treatment and treatment plates and were tested at least three independent times for each phenotype reported in Table 1. Phenotypes were scored for linkage analysis using a semi-quantitative method based on the phenotypes reported in Table 1 in which each phenotype group was assigned a unique category (below).
Linkage analysis on 121 segregants with genotype information was performed in R/qtl [50] using the non-parametric model. Genotypes were previously generated by DNA hybridization to Affymetrix 25-mer arrays [20]. Permutations (4000x per phenotype) were performed to obtain a significance cutoff where p = 0.0125 (0.05/the number of traits: 4). This corresponded to LOD scores of 3.9, 3.8, 3.8, and 3.7 for MMS-, 4-NQO-, bleomycin-, and caffeine-sensitivity, respectively (Figure 2). Support intervals defined by a 1.5 drop in LOD score from the peak were calculated using the lodint function in R/qtl.
DNA samples were prepared by pooling 5 ml YPD overnight cultures from 40 backcrossed individuals that were resistant and 40 that were sensitive to 4-NQO. Genomic DNA was prepared separately from the resistant and sensitive pooled cultures using the QIAGEN Genomic-tip 500/G kit (Qiagen). The pooled gDNA samples were hybridized to Affymetrix tiling arrays containing Watson strand 25-mers tiled every 4 bp (BY reference genome). The log10 ratio of intensity values was calculated for every probe that was considered to contain a sequence variant between BY and RM by SNPscanner [51]. The log10 (intensity difference) was plotted along the chromosome and smoothed using the smooth.spline function in the stats package of R. Regions where the smoothed value exceeded +/−1 were further investigated. |
10.1371/journal.ppat.1005952 | The Kallikrein-Kinin System: A Novel Mediator of IL-17-Driven Anti-Candida Immunity in the Kidney | The incidence of life-threatening disseminated Candida albicans infections is increasing in hospitalized patients, with fatalities as high as 60%. Death from disseminated candidiasis in a significant percentage of cases is due to fungal invasion of the kidney, leading to renal failure. Treatment of candidiasis is hampered by drug toxicity, the emergence of antifungal drug resistance and lack of vaccines against fungal pathogens. IL-17 is a key mediator of defense against candidiasis. The underlying mechanisms of IL-17-mediated renal immunity have so far been assumed to occur solely through the regulation of antimicrobial mechanisms, particularly activation of neutrophils. Here, we identify an unexpected role for IL-17 in inducing the Kallikrein (Klk)-Kinin System (KKS) in C. albicans-infected kidney, and we show that the KKS provides significant renal protection in candidiasis. Microarray data indicated that Klk1 was upregulated in infected kidney in an IL-17-dependent manner. Overexpression of Klk1 or treatment with bradykinin rescued IL-17RA-/- mice from candidiasis. Therapeutic manipulation of IL-17-KKS pathways restored renal function and prolonged survival by preventing apoptosis of renal cells following C. albicans infection. Furthermore, combining a minimally effective dose of fluconazole with bradykinin markedly improved survival compared to either drug alone. These results indicate that IL-17 not only limits fungal growth in the kidney, but also prevents renal tissue damage and preserves kidney function during disseminated candidiasis through the KKS. Since drugs targeting the KKS are approved clinically, these findings offer potential avenues for the treatment of this fatal nosocomial infection.
| Candida albicans is the causative agent of oropharyngeal candidiasis (OPC, thrush), dermal and vaginal candidiasis. However, the most severe C. albicans-induced disease is disseminated candidiasis, a frequent nosocomial infection associated with a high mortality rate. During disseminated candidiasis, C. albicans form invasive hyphae that damage target organs, particularly kidney and liver. Previous studies have identified an essential role of interleukin-17 (IL-17) in controlling systemic infection through regulation of neutrophils. We show here for the first time that IL-17 also regulates the renal protective Kallikrein-kinin system (KKS). Our discovery of a connection between IL-17 and the KKS suggests a new, previously unanticipated avenue for the treatment of renal damage in disseminated candidiasis. These findings have potential translational significance, as agonists of the KKS are in routine clinical use. Therefore, these results not only identify downstream mediators that could serve as novel drug targets, but could possibly be used to guide decisions on whether targeting these mediators could be a useful therapeutic option in conjunction with current antifungal therapies.
| The commensal fungus Candida albicans causes several clinical conditions in immunocompromised individuals, including oropharyngeal candidiasis (OPC, thrush) and vaginal candidiasis [1]. However, the most severe Candida-induced disease is a systemic form of bloodstream candidiasis. Disseminated candidiasis is the fourth most common hospital acquired infection and is associated with a 40–60% mortality rate [2,3]. Intravascular catheters, abdominal surgery, prolonged use of antibiotics and immunosuppressive therapy are risk factors for this disease, and contribute to the concerning rise in the incidence of candidiasis [1]. Available antifungal medications are limited by drug-drug interactions, drug resistance, toxicity and high treatment costs. To date, there are no effective vaccines to fungal pathogens [1]. Thus, there is an unmet clinical need to develop alternative, safe and ideally inexpensive approaches to treat this fatal infection.
Candidiasis is often treated effectively with azoles, amphotericin B and echinocandins [4,5]. The extensive and prolonged use of antifungal medications to treat systemic fungal infections, however, has led to drug resistant fungal strains and host toxicity [5,6]. Thus, novel antifungals or improved therapeutic strategies are still needed. Indeed, in vitro studies combining azoles with other drugs such as tacrolimus, cyclosporine A, amiodarone or retigeric acid B yielded encouraging results [7–11]. These data justify the concept of novel combination therapies to treat candidiasis at lower dosage in preclinical animal models.
Death due to sepsis is a frequent outcome of disseminated candidiasis [1]. However, in 30–40% adults and 50% neonates, Candida hyphae invade and injure the kidney, leading to irreversible damage and fatal renal failure [12]. Once Candida invades the kidney, a robust innate response dominated by neutrophils and monocytes/macrophages contributes to pathogen clearance and sets the stage for the adaptive immune response [13]. During the course of fungal clearance, both innate effector cells and kidney-resident cells release tissue repair enzymes and anti-inflammatory proteins. While necessary to repair injured tissue, these factors also limit bystander damage caused by innate immune cells.
Considerable data implicate IL-17 (IL-17A) in immunity to C. albicans. For example, IL-17RA-/-, IL-17RC-/-, RORγt-/- and IL-17A-/- mice are all susceptible to systemic C. albicans infection [14–17]. At mucosal surfaces, IL-17 mediates antifungal activity by driving the expression of antimicrobial peptides and chemokines that facilitate neutrophil influx [18,19]. Unlike mucocutaneous candidiasis, which affects individuals with compromised IL-17 signaling, systemic C. albicans infection normally impacts individuals with no known underlying genetic defects in IL-17 signaling pathways [1]. One recent study reported that IL-17 also acts on NK cells to drive the production of GM-CSF, with protective activities in disseminated candidiasis [15]. However, the mechanisms of local IL-17-mediated antifungal activities within the kidney still remain unclear.
The Kallikreins (Klk) are a family of fifteen related serine proteases. Klk1 in particular plays a critical role in renal function and pathology [20]. Klk cleave kininogens to generate kinin peptides, known as bradykinin and kallidin. Collectively, this system is termed the Kallikrein-kinin system (KKS) (Fig 1C). Bradykinin signals through two receptors; bradykinin receptor β2 (Bdkrb2) is constitutively expressed, whereas Bdkrb1 is inducible upon inflammatory signals. The primary known function of the KKS is to regulate blood pressure by promoting vasodilation [20]. In addition, studies in animal models implicate the KKS in regulating inflammation, tissue repair and homeostasis during kidney injury [21]. The renal protective function of the KKS is mediated through upregulation of tissue repair proteins, inhibition of profibrotic factors, and control of apoptosis [21]. Consistently, polymorphisms in KKS-related genes (ACE, BDKRB2, NOS3, KLK1) are associated with an increased risk of acute and chronic renal injury in humans [22–25]. While it is clear that the KKS protects the kidney in disorders associated with sterile inflammation, its role in renal immunity in infectious settings is less well defined.
In some bacterial and viral infections, bradykinin enhances vascular permeability to facilitate pathogen spread [26,27]. However, there are very few studies linking the KKS and C. albicans pathogenesis. Kininogens have been shown to bind the C. albicans cell wall, causing the fungal SAP2 protease to induce release of biologically active kinins [28,29]. Additionally, the KKS was implicated in IL-17-mediated skin inflammation and an IL-17-dependent model of autoimmunity (experimental autoimmune encephaloymyelitis, EAE) [30,31].
Here, we identified kallikrein genes as novel IL-17 targets in disseminated candidiasis, revealing an unanticipated link between IL-17 and KKS-mediated renal protection. Mice lacking the IL-17 receptor A subunit (IL-17RA) exhibited diminished Klk expression in the kidney. Moreover, overexpression of Klk1 restored protective renal immunity against systemic C. albicans infection in IL-17RA-/- mice. The IL-17-KKS-axis activated bradykinin receptors, which served to enhance renal anti-C. albicans immunity. Addition of exogenous bradykinin in immunocompetent mice prevented renal damage by inhibiting apoptosis of kidney-resident cells, and prolonged animal survival during candidiasis. Finally, addition of bradykinin to a minimally effective dose of fluconazole significantly improved survival. These data identify a previously unrecognized link between IL-17 and KKS-mediated renal protection against disseminated candidiasis, which may provide the basis for clinical intervention in this disease.
In C. albicans intravenous challenge in mice, the kidney is the most heavily colonized organ [32]. With a higher inoculum (>106 cfu), mice succumb to infection within 48–72 h due to sepsis. However, mice infected with a low dose of C. albicans (105 cfu) exhibit progressive loss of renal function over a period of ~2 weeks, which more accurately reflects disease progression in humans [32]. During candidiasis, IL-17 is rapidly upregulated in the kidney, but its function in that organ is unknown [14]. To understand how IL-17 mediates kidney-specific immunity, we performed Illumina microarray analyses comparing WT and IL-17RA-/- renal gene expression at 48 h p.i. Confirming previous reports, IL-17RA-/- mice demonstrated significantly increased kidney fungal burden in comparison to WT following infection with C. albicans (CAF2-1 or SC5314) (Fig 1A) [14,16,33].
The classic IL-17 gene signature includes neutrophil-related genes and antimicrobial peptides (AMPs), such as CXC chemokines (Cxcl1,2,5), defensins (Defb3), calprotectin (S100a8) and lipocalin 2 (Lcn2) [18]. Although a few genes previously shown to be controlled by IL-17 were differentially expressed in the kidney during candidiasis, overall we saw a surprisingly distinct gene profile compared to analyses of IL-17-dependent genes in other settings (Fig 1B and S1A Fig) [18]. The expression of classical IL-17-responsive genes such as Cxcl1, Cxcl2 and Lcn2 were unaffected in IL-17RA-/- mice (S1A Fig). Using the DAVID Gene Functional Classification algorithm (which uses a gene-to-gene similarity matrix based shared functional annotation), we identified several functional groups with enrichment scores over 1.0. Most striking to us based on their known role in kidney physiology was the enrichment of genes encoding the KKS (Fig 1B and 1C). Multiple Klk genes were suppressed in the kidney of IL-17RA-/- mice compared to WT following C. albicans infection. These results were verified by measuring the renal expression of Klk1 and Klk1b27 by qPCR (Fig 1D).
We next analyzed the impact of differential fungal load on Klk gene expression in the kidney by inoculating WT mice with either a high (106 cfu) or low (105 cfu) dose of C. albicans and measuring expression at 48 h p.i. Either dose caused comparable Klk1 expression (S1B Fig), indicating that the differences in Klk expression are not due to differential fungal loads. We then verified protein expression of Klk1 by immunoblotting. Klk1 was constitutively expressed at comparable, albeit low, levels in the sham-infected WT and IL-17RA-/- mice. However, Klk1 was upregulated in WT kidney following C. albicans infection. This observation was true upon infection with either the CAF2-1 or SC5314 strains of C. albicans. Confirming the gene expression data, we observed lower expression of renal Klk1 in the absence of IL-17RA during systemic infection (Fig 1E and 1F). Collectively, these data indicate that IL-17 signaling in the kidney regulates Klk1 expression during disseminated candidiasis. To our knowledge, this is the first demonstration that renal Klk1 is upregulated in an infectious setting, and certainly first demonstration that the KKS is controlled by IL-17.
To understand the role of Kallikreins in candidiasis, we focused on Klk1 based on its connection to IL-17-driven diseases including EAE and systemic lupus erythematosus [30,31]. Kidney sections from C. albicans-infected WT mice were stained for Klk1 at 72 h p.i. Only kidney-resident cells, particularly renal tubular epithelial cells (RTEC), expressed Klk1 during infection (Fig 2A). These results agree with previous reports indicating that RTEC are the major producers of Klk1 in chronic kidney diseases [34,35]. Although Klk1 controls vital kidney functions, its regulation and function under inflammatory conditions are not well defined. We therefore asked whether changes in Klk1 expression were a direct result of IL-17 signaling in RTEC, or a by-product of generalized renal inflammation due to increased fungal burden. We treated primary RTEC in vitro with IL-17 together with TNFα, a cytokine with which IL-17 exhibits strong signaling cooperativity [36]. Indeed, IL-17 and TNFα triggered strong synergistic upregulation of Klk1 and Klk1b27 mRNA (Fig 2B). Neither IL-17F nor IL-17C induced the expression of Klk genes (S1C Fig). Thus, IL-17 in conjunction with TNFα directly regulates Klk gene expression in RTEC, revealing a previously unrecognized class of IL-17-dependent target genes.
To verify the finding that IL-17 induces Klk1 in the kidney, we overexpressed IL-17 in WT mice using adenovirus (Ad-IL-17) [14]. Mice infected with Ad-IL-17 exhibited 400-fold more serum IL-17 than with a control vector (Ad-ctrl) (S2A Fig). This increased level of IL-17 was not associated with systemic inflammation, as serum TNFα and IL-1β levels were undetectable. By IHC, RTEC within kidney stained positively for Klk1 following overexpression of IL-17. The expression of Klk1 was restricted to RTEC, as no staining could be detected in the glomerular and vascular compartments of the kidney (Fig 2C). Additionally, Ad-IL-17 administration upregulated multiple IL-17 target genes in the kidney (Il6, Cxcl5 and Lcn2) (Suppl. Fig 2B). Nonetheless, kidneys of Ad-IL-17-treated mice exhibited no overt inflammatory changes (S2C Fig). Thus, IL-17 induces the expression of Klk1 in RTEC following disseminated candidiasis.
Klk1 protects the kidney against acute and chronic disorders in sterile inflammation, but has not been linked to candidiasis or IL-17 signaling. To test the hypothesis that Klk1 plays a critical role in IL-17-driven renal protection against disseminated candidiasis, we overexpressed Klk1 with the adenoviral system (Ad-Klk1) and assessed disease susceptibility. Remarkably, overexpression of Klk1 significantly improved the survival of C. albicans-infected IL-17RA-/- mice and in WT mice (Fig 2D). A previous study suggested that Klk1 may induce inflammatory cytokines in human RTECs, at least in vitro [37]. However, we found that overexpression of Klk1 had very little impact on renal inflammatory gene expression or fungal load (Fig 2E and S2D Fig). Overall, these data indicate that Klk1 enhances anti-C. albicans immunity in the kidney in an IL-17-dependent manner, but is not responsible for inducing inflammatory gene expression.
Klk1 mediates cleavage of kininogens to generate bradykinin, which signals through Bdkrb1 and Bdkrb2 (Fig 1C) [21]. Additionally, Klk1 activates protease-activated receptors (PAR) such as PAR4 to trigger the release of inflammatory mediators from RTEC [37]. To define the role of Bdkrb activation in renal immunity, IL-17RA-/- mice were treated with bradykinin (300 nmol/kg) and survival evaluated following infection. As shown, 90% of the untreated IL-17RA-/- mice succumbed to infection by day 5 p.i., while mortality in IL-17RA-/- mice was delayed with bradykinin treatment (Fig 3A). Additionally, untreated WT mice had a modestly increased survival benefit compared to bradykinin treated IL-17RA-/- mice, suggesting that there may also be a Bdkrb-independent pathway occurring in candidiasis.
Disseminated candidiasis typically impacts individuals with no known underlying immune defects [1]. Therefore, we assessed impact of Bdkrb signaling in mice with intact IL-17 signaling capacity and normal levels of Klk1. Accordingly, WT mice were treated with Bdkrb1 and Bdkrb2 antagonists and evaluated for survival following systemic C. albicans infection. Mice given Bdkrb1 and Bdkrb2 antagonists had significantly reduced survival compared to untreated controls (Fig 3B). Collectively, these results indicate participation of Bdkrb signaling in the renal host defense during disseminated candidiasis.
Despite advances in antifungal therapy against disseminated candidiasis, mortality in patients with systemic C. albicans infection remains unacceptably high [1]. Since IL-17 is implicated in controlling candidiasis in experimental mouse models, targeting downstream mediators of IL-17 signaling pathway is an attractive approach to treat disseminated candidiasis [1]. In this regard, we hypothesized that manipulation of the IL-17-KKS pathway with bradykinin would ameliorate candidiasis in a host with intact IL-17 signaling. To provide proof-of-principle, WT mice were treated with bradykinin and survival assessed with two strains of C. albicans (CAF2-1 or SC5314). Indeed, the bradykinin-treated cohort exhibited significantly delayed mortality compared to an untreated control group (Fig 4A and 4B). Although bradykinin has been implicated in the development of angioedema and hypotesion [38], mice treated with bradykinin did not show an increased incidence of angioedema at day 7 p.i. (Fig 4C). Moreover, to investigate the hypotensive effect of bradykinin, we performed a study where blood pressure (BP) alterations were measured in real time in uninfected WT mice following bradykinin treatment. We find that i.p. injection of bradykinin exerted only a very brief hypotensive response that peaked within ~1 minute with full recovery by 6 minutes (maximum reduction of Mean Arterial BP was 36%) (S3A Fig). This result indicates that bradykinin is not likely to have any long term consequences on renal function during candidiasis. Taken together, these results demonstrate a beneficial effect of exploiting the IL-17-KKS axis in treating disseminated candidiasis even in a host capable of mounting normal IL-17 response.
Following hyphal invasion, renal injury is mediated by unchecked fungal replication and bystander tissue damage caused by the local inflammatory response. Therefore, clearance of C. albicans and timely repair of damaged tissues is crucial to preserve renal function. To understand the mechanisms by which bradykinin mediates renal protection during disseminated candidiasis, WT mice were treated with bradykinin starting on day -1 and then daily for 7 days. Mice treated with bradykinin demonstrated significantly diminished serum blood urea nitrogen and creatinine levels compared to untreated animals (Fig 4D), indicating that bradykinin preserves normal renal function in systemic C. albicans infection.
We then asked whether improved renal function in bradykinin treated mice was due to reduced damage of kidney parenchyma following fungal invasion based on histological analyses. While sham-infected mice treated with bradykinin showed normal kidney histology, the renal parenchyma of infected mice showed overt pathological changes characterized by loss of brush border epithelium and tubular atrophy at day 7 p.i. Moreover, the damage was primarily restricted to the renal cortex and outer medullary region. In line with the kidney function results, renal damage was ameliorated upon bradykinin treatment (Fig 4E). Interestingly, the influx of inflammatory cells was comparable between the treated and untreated groups. Furthermore, we observed significantly diminished expression of neutrophil gelatinase-associated lipocalin (NGAL), a prototypical kidney injury marker, in bradykinin treated animals (Fig 4F and 4G). Collectively, these results indicate that bradykinin prevents C. albicans-mediated renal damage and preserves renal function during disseminated candidiasis.
To identify the effector mechanisms by which bradykinin prevents renal insufficiency, we evaluated fungal load and inflammatory cell influx at days 3 and 7 p.i. Although differences in kidney function were already evident early as day 7 p.i. (Fig 4D), fungal loads were comparable between the bradykinin treated and untreated groups at these time points (Fig 5A). Previous studies have shown that neutrophils and monocytes/macrophages mediate fungal clearance in candidiasis [39,40]. Thus, we examined the frequency of infiltrating myeloid cells in kidney upon bradykinin treatment. In agreement with the fungal clearance rates, the percentages of kidney infiltrating inflammatory cells (CD45+), neutrophils (Gr1+) and macrophages (F4/80+) were similar between groups (Fig 5B and S3B Fig). We next assessed whether bradykinin impacted the candidacidal activity of innate cells. Bradykinin treatment of BM-derived neutrophils and macrophages (BMDM) did not alter their ability to kill unopsonized C. albicans yeasts, as determined by an in vitro fungal killing assay (Fig 5C). Additionally, RTEC and BMDMs were stimulated in vitro with bradykinin and culture supernatants evaluated for cytokines and nitrite production. Bradykinin induced IL-6 in RTECs (Fig 5D), but did not induce IL-6, TNFα or nitrite in BMDMs (Fig 5E and 5F). Overall, these results suggest that renal protection observed in bradykinin treated mice is not due to diminished fungal load, inflammatory cell infiltration or proinflammatory function of innate cells in the kidney.
Activation of bradykinin receptors protects the kidney from end-stage renal damage by inducing tissue-protective growth factors and matrix-degrading enzymes [21]. Therefore, we examined the effect of bradykinin treatment on profibrotic changes and tissue-protective growth factors expression in candidiasis. We observed minimal extracellular matrix protein deposition in bradykinin treated mice (S4A Fig). Additionally, there were no significant differences in expression of genes encoding tissue-protective growth factors (Hgf and Ctgf) or matrix-metalloproteinases at day 7 p.i. (Mmp2 and Mmp9) (S4B Fig).
Previous studies have described an anti-apoptotic function of bradykinin in kidney injury [41,42]. Therefore, we assessed apoptosis in kidney-resident cells following C. albicans infection. At day 7 p.i., flow cytometry analysis revealed a significantly reduced frequency of both early (AnnexinV+PI-) and late (AnnexinV+PI+) apoptotic kidney-resident cells (CD45-) in bradykinin treated compared to untreated mice (Fig 6A). Concurrently, there was a significant reduction in the number of TUNEL positive cells in the renal cortex following bradykinin treatment (Fig 6B). C. albicans can regulate survival of kidney-resident macrophages via Caspase-3 [39]. In agreement with the reduced apoptosis (Fig 6A), bradykinin treatment resulted in a significantly diminished number of cleaved Caspase-3+ kidney-resident cells (CD45-) (Fig 6C). Moreover, IHC indicated that cleaved Caspase-3 is localized in the renal cortex and outer medullary region of untreated mice (Fig 6D). The extent of cleaved Caspase-3 was markedly reduced following bradykinin treatment (Fig 6D). Although expression of Bax (a pro-apoptotic gene) in kidney-resident cells was comparable between the groups, there was an increase in expression of Bcl-xL (an anti-apoptotic gene) after bradykinin treatment (Fig 6E). Thus, bradykinin preserves renal function during systemic fungal infection by limiting apoptosis of kidney-resident cells.
There is an unmet clinical need to reduce the dosage of current antifungal drugs to overcome the problem of drug resistance and toxicity. Based on the renal protective function of bradykinin in disseminated candidiasis (Figs 3A, 3B and 4A), we hypothesized that a minimally effective dose of fluconazole (FLC) combined with bradykinin would confer better protection against disseminated candidiasis than either agent alone. To test this hypothesis, we first determined a minimally effective dose of FLC on WT mice with disseminated candidiasis. The lowest dose of FLC (5 mg/kg) was the least effective in clearing the fungus at day 4 p.i. (S5A Fig). Fungal clearance correlated well with survival, as mice treated with FLC (5 mg/kg) showed the same susceptibility as untreated animals (S5B Fig). Therefore, the 5 mg/kg dose of FLC was chosen to evaluate the impact of the combination therapy. Next, C. albicans infected WT mice were treated with FLC and/or bradykinin and evaluated for survival over 14 d (Fig 7A). Strikingly, mice receiving the combination of bradykinin and FLC showed a significant increase in survival compared to untreated mice or mice given FLC or bradykinin alone (Fig 7B). As expected, mice treated with bradykinin alone but not FLC demonstrated increased survival compared to untreated mice (Fig 7B). These data show that a combination of bradykinin and FLC confers better protection against disseminated candidiasis than either drug individually. Consequently, use of bradykinin could potentially permit reducing the dose of antifungal drugs without compromising efficacy against fungal infection.
Kidneys in a healthy state are sterile. However, renal infections occur via hematogenous routes or from ascending spread from the bladder or urethra [43]. In recent years, considerable data have implicated IL-17 in immunity against disseminated candidiasis [14–16]. Nevertheless, it is unclear how IL-17 regulates immunity within the kidney, the most heavily colonized organ during blood-borne C. albicans infection. In the present report, we have identified renal-protective kallikreins as novel IL-17 target genes in systemic candidiasis, thereby revealing a new connection between IL-17 and KKS-mediated renal defense. IL-17 not only limits fungal growth in the kidney, but also prevents renal tissue damage and preserves kidney function during C. albicans invasion (S6 Fig). Consequently, therapeutic manipulation of the IL-17-KKS pathways protected mice from early mortality in disseminated candidiasis. Our data provide important and potentially translatable insights into the renal functions of IL-17 in the context of this fatal hospital-acquired infection.
Kidney-specific immune responses are mediated by both kidney-infiltrating immune effectors and kidney resident cells. Although IL-17RA is ubiquitously expressed, most documented IL-17R signaling occurs in non-hematopoietic cells, particularly cells of epithelial and mesenchymal origin [44,45]. A recent study also suggested a role for IL-17RA signaling in NK cell development in the context of disseminated candidiasis [15], although this report conflicts with a study showing that NK cells are redundant for antifungal defense in immunocompetent hosts [46]. Consistent with the latter finding, we and others have observed upregulation of transcripts encoding IL-17A and IL-17-responsive genes in the kidney [14]. We also showed that kidney resident cells express the IL-17R and are responsive to IL-17 [47]. Taken together, these results argue that there is a bona fide, kidney-specific role of IL-17 in immunity to candidiasis.
Unlike mucocutaneous candidiasis, disseminated infection typically occurs in individuals with no known defects in IL-17 signaling pathways. In line with these observations, we were intrigued by the surprisingly different gene profile seen in C. albicans-infected kidney compared to prior studies involving mucosal Candida infection such as the tongue [18]. These results highlight the fact that different cell types “interpret” IL-17 signals differently, with non-identical patterns of gene expression depending on setting. Therefore, lessons derived from studies of anti-C. albicans immunity at mucosal sites cannot always be applied to local kidney immune responses. Although we show that IL-17 (in conjunction with TNFα) induces Klk1 gene in primary RTEC, there is little known about Klk1 gene regulation at the transcriptional level. We have identified conserved CCAAT Enhancer Binding Protein (C/EBP)-β binding sites within the putative proximal promoters of the Klk1 gene, hinting that IL-17 may regulate Klk1 expression in a C/EBPβ-dependent manner. Indeed, C/EBPβ-/- mice are susceptible to systemic candidiasis [48,49]. Since TNFα is required for protection against disseminated candidiasis [50], further studies should focus on Klk gene expression in the kidney of TNFα-deficient mice.
Klk1 cleaves kininogens to form bradykinin, a process known as the bradykinin-dependent pathway [20]. In addition, Klk1-mediated activation of PAR4 induces cytokine production and prevents apoptosis in RTEC, known as the bradykinin-independent pathway [37]. Mice deficient in specific PARs or treated with PAR antagonists exhibited compromised renal inflammatory changes [51,52]. Nevertheless, the contribution of PARs in IL-17-Klk1-mediated renal protection is unknown. Moreover, the relative contributions of Bdkrb1 and Bdkrb2 in renal defense against candidiasis need to be determined, which is possible with specific knockout mouse strains.
Our data show that bradykinin treatment improved renal function as early as 7 days p.i. Interestingly, serum creatinine and BUN levels did not correlate with the survival benefit at late time points, the basis for which is unclear. Late mortality despite improved renal function likely indicates side effects of bradykinin in mice. Although bradykinin has been implicated in causing angioedema [37], mice treated with bradykinin did not show any signs of angioedema following fungal infection (Fig 4C). Additionally, bradykinin at the dose levels used in this study did not seem to exert any long term consequences on blood pressure. Therefore, these studies argue against the likelihood of bradykinin toxicity and point towards alternative possibilities to explain the lack of correlation between kidney function and survival benefit. First, extremely short half-life of bradykinin (approx. 30 sec) and internalization of bradykinin receptors may lead to bradykinin unresponsiveness at the late time points. Second, daily administration of bradykinin may lead to the activation of a negative feedback regulatory mechanisms leading to increased degradation of bradykinin by Angiotensin converting enzymes. Finally, bradykinin has been shown to modulate glomerular function by regulating podocyte permeability [53], an effect independent of its renal tissue protective function. Given the potential side effects in bradykinin therapy, future studies investigating these questions could help better understand the means to harness the beneficial impact of IL-17-KKS axis without compromising the safety in treatment against disseminated candidiasis.
An Ab targeting IL-17 (secukinumab) was approved in 2016 to treat moderate-severe plaque psoriasis [54]. Abs against IL-17/IL-17RA are in clinical trials for other autoimmune conditions [55]. One obvious concern with anti-IL-17 therapy is compromising IL-17-driven antifungal immunity. Although systemic C. albicans infection has not been reported thus far with secukinumab, patients on this medication may have a higher risk of developing disseminated candidiasis in the face of predisposing factors, such as an indwelling catheter, abdominal surgery or long-term antibiotic use. Our data show that activation of the KKS pathway restored protection in IL-17RA-/- mice. Thus, drugs targeting the IL-17-KKS axis may be considered to treat or prevent disseminated C. albicans infection in patients receiving anti-IL-17 therapy.
Amphotericin B, azoles and echinocandins are used to treat systemic C. albicans infections, but there are concerns due to drug-resistance and toxicity. Here, we show proof-of-concept that combination therapy with bradykinin and low-dose FLC is effective in treating candidiasis in mice. Notably, the survival rate of mice treated with the combination therapy was similar to the survival rate in mice given four times the FLC dose. This approach may be especially valuable for anti-fungal drugs such as amphotericin B due its intrinsic renal toxicity. Future studies will test the efficacy of combination therapy with amphotericin B and bradykinin, once we have access to oral or injectable form of amphotericin B suitable for administration in mice. In addition, the efficacy of ACE inhibitors, known to increase the levels of bradykinin and are routinely used to treat patients with obstructive nephropathy, need further evaluation in pre-clinical animal models of disseminated candidiasis. Overall, our data show that defining the IL-17 anti-fungal pathway has highlighted a potentially translatable and testable approach to treating systemic candidiasis. Additionally, the novel convergence between IL-17 and the KKS pathways in renal defense against fungal infection represents a major advance in our understanding of IL-17 signaling in the kidney inflammation.
Wild type (WT) C57BL/6J mice were purchased from The Jackson Laboratory (Bar Habor, ME). IL-17RA-/- mice were kindly provided by Amgen (San Francisco, CA) and bred in-house. All mice were housed under specific pathogen-free conditions, and age-matched male mice were used for all experiments. Animal protocols were approved by the University of Pittsburgh IACUC (Protocol # 14094427), and adhered to the guidelines in the Guide for the Care and Use of Laboratory Animals of the National Institute of Health.
C. albicans strains CAF2-1 or SC5314 were used as indicated. C. albicans was grown in YPD at 30°C for 18–24 h. Mice were injected via the tail vein with PBS (sham-infected) or 1x105 cfu (unless otherwise indicated) C. albicans yeast cells resuspended in PBS. Mice were weighed and monitored daily. Mice were sacrificed if they showed >20% weight loss or signs of severe pain or distress. Mice were evaluated for survival over a period of 14 days. At sacrifice on days 2, 3 and 7 p.i., as indicated, kidneys were weighed and homogenized in sterile PBS using a GentleMACS (Miltenyi Biotec, Cambridge MA). Serial dilutions of organ homogenates were plated on YPD agar with antibiotics, and fungal burden represented as colony forming units (cfu) per gram of tissue.
Serum was collected by retro-orbital bleeding at day 7 p.i. Creatinine and blood urea nitrogen levels were assessed using the QuantiChrom Creatinine Assay kit (BioAssay Systems, Hayward CA) and MaxDiscovery Blood Urea Nitrogen Enzymatic kit (Bioo Scientific Corp., Austin TX), respectively.
Vascular permeability in mice was assessed as described before [56]. Briefly, 100 μl sterile solution of Evans Blue dye (30mg/kg) (Sigma Aldrich., St Louis, MO) in PBS was injected intravenously. The stain was allowed to circulate for 30 min. After 30 min, mice were sacrificed and the hind feet were removed, blotted dry and weighed. The Evans blue was extracted from the feet with 1 ml of formamide overnight at 55°C and measured spectrophotometrically at 600 nm. Evans Blue stain was quantified according to a standard curve. The results are presented as ng of Evans Blue stain/mg of tissue.
Primary renal tubular epithelial cells (RTEC) from C57BL/6J mice (Cell Biologics, Chicago, IL) were cultured as per manufacturer’s instructions. RTEC (1x106 cells/well) were treated with IL-17A (50 or 200 ng/ml) or TNFα (5 ng/ml) or IL-17C (50 or 200 ng/ml) or IL-17F (50 or 200 ng/ml) and IL-17 and TNFα in combination for 24 h. Recombinant murine IL-17A, IL-17C, IL-17F and TNFα were purchased from Peprotech (Rocky Hill, NJ).
Bone marrow derived macrophages (BMDM) from C57BL/6J mice were cultured for 7 days in the presence of L929 supernatants. BMDM and RTEC (1x106 cells/well) were treated with bradykinin (R&D Biosystems, Minneapolis MN) or left untreated for 24 h. LPS (Sigma Aldrich, St Louis, MO) was used as positive control. Supernatants were subjected to analyses using commercially available IL-6, IL-1β and TNFα ELISA kits (Ebiosciences, Dallas TX. Nitrite concentrations were measured by tri-iodide based reductive chemiluminescence as previously described [57]. Briefly, samples were injected into tri-iodine to reduce nitrite to NO gas that was detected by a Nitric Oxide Analyzer (Sievers, GE).
Neutrophils isolated from bone marrow using Neutrophil Isolation Kit (Miltenyi Biotech, San Diego, CA) were plated at 1×105 cells/well. Non-opsonized C. albicans was added to neutrophils at 0.5 × 105 yeast cells/well (ratio of 2:1). If indicated, bradykinin was added to the wells. Cultures were incubated with unopsonized C. albicans for 3 h and lysed in cold double-distilled H2O. Killing was assessed by cfu counts in triplicate. The results are reported as percentage killing of C. albicans which is calculated as 1-[cfu of treatment group/cfu of control group] X 100. The same protocol was followed for bone marrow derived macrophages which were cultured with L929 supernatants supplemented media for 7 days prior to carrying out the killing assay.
At sacrifice, kidneys were stored at -80°C. Total RNA was extracted from the homogenized kidney tissue with the RNeasy Micro Kit (Qiagen, Valencia CA) and submitted to Genomics Research Core at University of Pittsburgh. Gene expression analysis was performed using Mouse WG6 Gene Expression Bead Chip (Illumina). All test, normalization and transformation analyses were performed using caGEDA, a feely available informatics tool. The data sets were analyzed for differentially expressed genes. Efficiency analysis was performed by Random Resampling Validation using a Naïve Bayes Classifier and PACE analysis. The cluster analysis was performed by Unweighted Pair Group Method with Arithmetic Mean and similarity measure was determined by Euclidean distance. The raw and normalized microarray data have been submitted to the Gene Expression Omnibus (GEO), and study ID is GSE88800 at: https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE88800. For real time PCR analysis, complementary DNA was synthesized with SuperScript III First-Strand (Invitrogen, Carlsbad CA). Gene expression was determined by qPCR with PerfeCTa SYBR Green FastMix ROX (Quanta BioSciences, Gaithersburgh MD) on a 7300 Real-Time PCR System (Applied Biosystems, Carlsbad CA). Primers were obtained from Quantitect (Qiagen, Valencia CA). The expression of each gene was normalized to Gapdh.
At sacrifice on day 7 p.i., kidneys were harvested following perfusion with PBS. Briefly, kidney homogenates were digested in PBS with 1 mg/ml collagenase type I (Worthington, Lakewood NJ) for 30 m at 37°C. Cells were stained with the following antibodies: CD45 (BioLegend; clone 30-F11), Ly6G (BD Biosciences; clone IA8), F/480 (eBiosciences; clone BM8). For detection of apoptotic cells and cleaved Caspase-3 positive cells, cells were stained with Annexin V (BD Pharmingen) and CaspGLOW Fluroscein active Caspase-3 staining kit (eBioSciences), respectively as per manufacturer’s protocol. Samples were acquired and sorted on a Fortessa and FACS ARIA II, respectively (BD Biosciences, San Jose CA) and analyzed with FlowJo (Tree Star, Ashland OR).
Kidneys were fixed in formalin, dehydrated and paraffin embedded. Serial kidney sections were stained with H&E, Periodic-acid Schiff or Masson Trichrome stains for morphological analysis and determination of kidney injury.
Immunohistochemistry staining was done on formalin-fixed, paraffin embedded sections. Sections were rehydrated and antigen retrieval was performed with heated citrate. Primary antibodies against the following proteins were used: Klk1 (LifeSpan Biosciences, Seattle WA), NGAL (Santa Cruz Biotechnology, Dallas TX) and cleaved caspase-3 (Cell Signaling, Danvers MA). Secondary antibodies used were horseradish peroxidase coupled antibodies (Jackson ImmunoResearch, West Grove, PA). To detect apoptotic cells TUNEL staining was done on frozen kidney sections using the TUNEL apoptosis detection kit according to manufacturer’s protocol (Millipore, Temecula CA). The number of TUNEL+ cells was counted in 15 randomly selected high powered fields (400X) per slide. All images were obtained with EVOS FL Auto microscope (Life Technologies CA).
Kidneys were homogenized in RIPA buffer. Concentration of protein was quantified by the BCA quantitation assay (Thermo Scientific, Pittsburgh PA). Equal amounts of sample were subjected to electrophoresis and transferred to PVDF membranes (Millipore, Billerica MA). After blocking with 5% milk in TBS, the blots were incubated with anti-mouse Klk1 (LifeSpan Biosciences, Seattle WA), anti-mouse NGAL (R&D Biosystems, Minneapolis MN), anti-mouse cleaved Caspase-3 (Cell Signaling, Danvers MA) or anti-mouse beta-actin (Abcam, Cambridge, MA) overnight in 4°C. The blots were then washed and incubated for 1 hour at room temperature with individual secondary antibodies. Bands were detected using an enhanced chemiluminescence detection system (ThermoScientific, Pittsburgh PA) and developed with a FluorChem E imager (ProteinSimple, San Jose CA). Band corresponding to proteins of interest were analyzed by ImageJ software.
Adenoviruses expressing IL-17A (Ad-IL-17) and control vector (Ad-ctrl) were kindly provided by Dr. J. Kolls (U. Pittsburgh). Ad-Klk1 and corresponding Ad-ctrl vector were from Applied Biological Materials Inc. (Richmond, British Columbia, Canada). Mice were injected via the tail vein with 1x109 pfu virus 72 h prior to induction of disseminated candidiasis.
Mice were injected with bradykinin (300 nmol/kg/day) (R&D Systems, Minneapolis MN) in a 200 μl volume i.p. Mice received i.p. injection of a combination of Bdkrb1 (R-715: 1 mg/kg/day) and Bdkrb2 (HOE-140: 1 mg/kg/day) antagonists (R&D Systems, Minneapolis MN). Untreated mice received equal volume of PBS.
C. albicans infected mice were treated with Fluconazole (FLC) (Diflucan: obtained from University of Pittsburgh Medical Center, Pittsburgh PA) as described before with minor modifications [58]. Briefly, mice were treated with 5, 10, 20 and 40 mg/kg body weight FLC by oral gavage at 2 h and 26 h post infection. Untreated mice received equal volume of PBS.
Data were analyzed by Kaplan-Meier, ANOVA, Mann-Whitney or unpaired Student's t test using GraphPad Prism (La Jolla, CA). P values <0.05 were considered significant. All experiments were performed a minimum of twice to ensure reproducibility.
|
10.1371/journal.pgen.1004552 | Ku-Mediated Coupling of DNA Cleavage and Repair during Programmed Genome Rearrangements in the Ciliate Paramecium tetraurelia | During somatic differentiation, physiological DNA double-strand breaks (DSB) can drive programmed genome rearrangements (PGR), during which DSB repair pathways are mobilized to safeguard genome integrity. Because of their unique nuclear dimorphism, ciliates are powerful unicellular eukaryotic models to study the mechanisms involved in PGR. At each sexual cycle, the germline nucleus is transmitted to the progeny, but the somatic nucleus, essential for gene expression, is destroyed and a new somatic nucleus differentiates from a copy of the germline nucleus. In Paramecium tetraurelia, the development of the somatic nucleus involves massive PGR, including the precise elimination of at least 45,000 germline sequences (Internal Eliminated Sequences, IES). IES excision proceeds through a cut-and-close mechanism: a domesticated transposase, PiggyMac, is essential for DNA cleavage, and DSB repair at excision sites involves the Ligase IV, a specific component of the non-homologous end-joining (NHEJ) pathway. At the genome-wide level, a huge number of programmed DSBs must be repaired during this process to allow the assembly of functional somatic chromosomes. To understand how DNA cleavage and DSB repair are coordinated during PGR, we have focused on Ku, the earliest actor of NHEJ-mediated repair. Two Ku70 and three Ku80 paralogs are encoded in the genome of P. tetraurelia: Ku70a and Ku80c are produced during sexual processes and localize specifically in the developing new somatic nucleus. Using RNA interference, we show that the development-specific Ku70/Ku80c heterodimer is essential for the recovery of a functional somatic nucleus. Strikingly, at the molecular level, PiggyMac-dependent DNA cleavage is abolished at IES boundaries in cells depleted for Ku80c, resulting in IES retention in the somatic genome. PiggyMac and Ku70a/Ku80c co-purify as a complex when overproduced in a heterologous system. We conclude that Ku has been integrated in the Paramecium DNA cleavage factory, enabling tight coupling between DSB introduction and repair during PGR.
| DNA double-strand breaks (DSBs) are potential threats for chromosome stability, but they are usually repaired by two major pathways, homologous recombination or non-homologous end joining (NHEJ). DSBs can also be essential during physiological processes, such as the programmed removal of germline sequences that takes place in various eukaryotes, including ciliates, during somatic differentiation. We use the ciliate Paramecium tetraurelia as a unicellular model to study how DNA breakage and DSB repair are coordinated during programmed genome rearrangements. In this organism, assembly of the somatic genome involves the elimination of ∼25% of germline DNA, including the precise excision of thousands of short Internal Eliminated Sequences (IES) scattered along germline chromosomes. A domesticated piggyBac transposase, PiggyMac, is required for double-strand DNA cleavage at IES ends and IES excision sites are very precisely repaired by the NHEJ pathway. Here, we report that a specialized Ku heterodimer, specifically expressed during programmed genome rearrangements, is an essential partner of PiggyMac and activates DNA cleavage. We propose that incorporation of DSB repair proteins in a pre-cleavage complex constitutes a safe and efficient way for Paramecium to direct thousands of programmed DSBs to the NHEJ pathway and make sure that somatic chromosomes are assembled correctly.
| DNA double strand breaks (DSBs) are among the most deleterious DNA lesions: if left unrepaired, a single DSB may trigger cell death, while incorrect repair can give rise to chromosome rearrangements [1]. Cells rely on two major pathways to repair DSBs. Homologous recombination (HR) uses a homologous template to restore the sequence of the broken chromosome, while non-homologous end joining (NHEJ) proceeds through the ligation of free DNA ends. Even though they can be very toxic, programmed DSBs are obligatory intermediates in essential biological processes, such as meiosis or acquired immune response. During meiosis, the Spo11 endonuclease cleaves DNA and DSB repair is carried out by HR [2]. In addition to favoring the exchange of parental alleles, HR ensures that homologous chromosomes are correctly paired before they are segregated during the first meiotic division. During lymphocyte differentiation, programmed genome rearrangements (PGR) mediated through V(D)J recombination generate the large diversity of immunoglobulin genes [3]. During V(D)J recombination, the domesticated transposase RAG1, associated with its partner RAG2, cleaves specific recombination sites. The resulting DSBs are repaired through classical NHEJ (C-NHEJ).
A critical step in C-NHEJ is the binding of the Ku70/Ku80 heterodimer to broken DNA ends [4]. Upon binding, Ku protects DNA ends from extensive resection [5] and, together with its facultative partner DNA-PKcs, facilitates the synapsis of two broken ends. Following recruitment of DNA processing enzymes, the Ligase IV-Xrcc4 complex mediates the joining of DNA ends. An alternative end joining pathway, referred to as alt-NHEJ (or MMEJ, for microhomology-dependent end joining), has been reported [6]. This poorly characterized pathway is independent of Ku and, to some extent, of Ligase IV. Because of the absence of Ku, alt-NHEJ involves limited 5′ to 3′ resection of broken DNA ends and generates deletions at DSB repair sites, which often involve microhomologies. When DSBs are repaired through HR, 5′ to 3′ resection also takes place, but two steps can be distinguished: initial short-range end resection relies on the same factors as alt-NHEJ [7], while subsequent long-range resection generates the long 3′ single strand that will invade a homologous DNA duplex [8].
Ciliates provide extraordinary models to study the interplay between DNA cleavage and DSB repair during PGR [9]. In these unicellular eukaryotes, two kinds of nuclei coexist in the same cytoplasm. The highly polyploid somatic macronucleus (MAC) is essential for gene expression but it is destroyed at each sexual cycle, while the diploid micronucleus (MIC) undergoes meiosis and transmits the germline genome to the new MIC and MAC of the next generation. In Paramecium, massive PGR take place in the new developing MAC, while the genome is amplified from 2n to 800n [10]. These PGR consist in the elimination of two types of germline-specific DNA. Regions of up to several kbp in length, often containing repeated sequences, are eliminated in a heterogeneous manner, leading to chromosome fragmentation or intra-chromosomal deletions. In addition, thousands of single-copy, short and non-coding Internal Eliminated Sequences (IES) are excised precisely. Because 47% of genes are interrupted by at least one IES in the germline genome [11], the precise excision of IESs is essential for the assembly of functional genes in the new MAC and the survival of the sexual progeny. Paramecium IESs are invariably flanked by one TA dinucleotide on each side and little additional information can be found in their nucleotide sequence, which raises the question of how these sequences are recognized and targeted for excision. In fact, the excision of an estimated one-third of Paramecium IESs is controlled maternally through a sequence homology-dependent mechanism [12]. For these so-called maternally-controlled IESs, a genome-wide comparison of the germline and rearranged versions of the genome involves non-coding RNAs [13], [14]. This epigenetic control drives the trans-generational inheritance of rearrangement patterns, from the old to the new MAC.
IES excision proceeds through a two-step “cut-and-close” mechanism. A domesticated piggyBac transposase, PiggyMac (Pgm), is essential to introduce the DNA cleavages that initiate the reaction, generating 4-bp staggered DSBs centered on the conserved TA at each IES boundary [15]. Following IES release, precise DSB repair is carried out through the C-NHEJ pathway and leaves a single TA at the IES excision site [16]. Providing their length allows enough DNA flexibility, the excised linear IESs are circularized, also through the C-NHEJ pathway, before they are actively degraded. During DSB repair, the flanking broken ends are thought to anneal through the pairing of the complementary TAs carried by their 4-base 5′ overhangs, and undergo limited 5′ and 3′ processing [17]. The final ligation step is mediated by the NHEJ-specific ligase complex, Ligase IV and its partner Xrcc4, both of which are essential for PGR. When LIG4 or XRCC4 genes are knocked down, Pgm-dependent DSBs are introduced normally, but they accumulate in the developing new MAC, which correlates with severely compromised DNA amplification and impairs the recovery of viable progeny [16].
Next-generation sequencing of the non-rearranged genome of Paramecium tetraurelia led to the identification of at least 45,000 IESs [11]. Therefore, a huge number of programmed DSBs have to be repaired precisely during MAC development. In the present study, we have addressed the question of how the C-NHEJ pathway is recruited to IES excision sites to carry out efficient and precise DSB repair. We have focused our analysis on the Ku heterodimer, which is the most upstream actor of the C-NHEJ pathway. Two KU70 and three KU80 genes were identified in the somatic genome [16]. We report here that development-specific KU genes are essential for PGR. Surprisingly, we demonstrate that Ku is required for the introduction of programmed DSBs at IES boundaries, and provide evidence that Ku interacts functionally and physically with Pgm. We propose that the Ku70a/Ku80c heterodimer forms a complex with Pgm and activates DNA cleavage during PGR in P. tetraurelia.
Two genes encoding Ku70 homologs, KU70a and KU70b, were identified in the macronuclear genome assembly of P. tetraurelia (Figure 1A and [16]). These two closely related copies arose from a recent whole genome duplication (WGD) that took place during the evolution of this species [18], and are referred to as ohnologs. We also identified three homologs of the human KU80 gene (Figure 1A). KU80a and KU80b are ohnologs from the recent WGD, while the more distant KU80c diverged after an earlier, “intermediate”, WGD.
Sexual processes in Paramecium may occur in two different ways: during conjugation, following the mixing of two reactive partners with compatible mating types, or during a self-fertilization process called autogamy, in which MIC meiosis is induced upon starvation in cells of a single mating type [19]. A microarray analysis of the P. tetraurelia transcriptome during autogamy revealed that KU70a and KU80c are specifically induced during the development of the new MAC, when PGR take place [16], [20]. The microarray data were confirmed by high-throughput sequencing of polyadenylated RNAs extracted during an autogamy time-course (Figure 1B, Arnaiz et al. in prep.), and by northern blot hybridization with specific probes (Figures 1C and 3C, left panels; Malinsky et al. in prep.). After Pgm depletion, the transcription of KU70 and KU80c is switched on at the expected time-point during autogamy (Figure 1C, right panels), indicating that KU genes are not induced as a response to Pgm-induced DSBs, but more likely as part of a general transcription program during MAC development. Moreover, in contrast to control cells, the levels of KU70 and KU80c mRNAs do not decrease at later time-points after Pgm depletion, suggesting that the completion of PGR is a signal for transcriptional switch-off.
The expression of KU70a and KU80c is specifically induced during autogamy and reaches a peak when the developing new MACs start to be detected in the culture (T5 and T11 time-points in Figure 1B). Using N-terminal GFP fusions, we followed the cellular localization of Ku70a and Ku80c during autogamy. Transgenes expressing each fusion protein under the control of their respective endogenous promoters were microinjected into the MAC of vegetative cells. The resulting transformants were grown and starved to induce autogamy. Consistent with the transcriptome analysis, GFP fluorescence was stronger in autogamous cells with respect to vegetative cells for each fusion transgene (Figure 2). Both fusion proteins accumulated in the developing new MACs, suggesting a possible involvement of Ku70a and Ku80c in PGR during MAC development.
To test the implication of the different KU genes in MAC development, we knocked them down systematically by feeding wild-type P. tetraurelia cells on dsRNA-producing bacteria to induce RNA interference [21]. The very high percentage of identity between KU70a and KU70b made it impossible to design specific RNAi constructs for each individual gene. Therefore, we silenced them both together. In contrast, we designed gene-specific RNAi constructs for KU80a, KU80b and KU80c. Whenever possible, to make sure that the observed phenotypes would be attributable to the silencing of each targeted gene, we used RNAi constructs homologous to two different regions in each KU80 gene (Figure S2). After three days of starvation in each silencing medium, individual autogamous cells were transferred to standard growth medium and allowed to resume vegetative growth. The survivors were grouped in three categories: (i) those that were able to undergo a new round of autogamy following a few divisions, most likely because they had regenerated their old MAC, were not considered as bona fide post-autogamous progeny [16]; (ii) slow-growing survivors were counted as progeny with a defective new MAC; (iii) those survivors that remained in the vegetative state when starved after a few divisions were classified as fully viable post-autogamous progeny. Only the recovery of survivors from the third category indicated that the silenced cells had been able to form a functional new MAC (Figure 3A).
When KU80a or KU80b were knocked down, either individually or together, the progeny exhibited good survival rates when compared to a control RNAi against the nonessential ICL7 gene, or with respect to cells that underwent autogamy in standard medium (Figure 3A). In contrast, RNAi against KU80c or KU70 yielded only 10 to 30% viable sexual progeny, showing that the developmentally-induced KU80c gene and one or both KU70 genes are required for the completion of autogamy. Interestingly, we observed that new MACs develop normally at the cytological level in cells silenced for KU70 or KU80c (Figure 3B), while cells silenced for LIG4 or XRCC4 were previously shown to harbor small new developing MACs that remain faintly stained with DAPI, correlating with a block in DNA amplification [16]. Bright DAPI staining of the new MACs indicates that DNA amplification takes place normally in Ku-depleted cells.
Residual survival was observed reproducibly following Ku70 or Ku80c depletions, in contrast to the severe lethality phenotype of Xrcc4 depletion (Figure 3A). This phenotypic difference, which was also noted during mouse embryonic development [22], may be explained by the fact that Ku acts upstream of Ligase IV/Xrcc4 during C-NHEJ. Upon Ku depletion, alternative DSB repair pathways could restore chromosome integrity to some extent, while Ligase IV/Xrcc4 depletion would constitute a dead-end for DSBs, committed to C-NHEJ in the presence of Ku. We noted, indeed, that normal DAPI staining of the developing MAC is restored in Ligase IV-depleted cells by simultaneously depleting Ku80c (Figure 3B). At this stage of our study, this observation is consistent with the hypothesis that activation of alternative Ku-independent DSB repair pathways may rescue DNA amplification in the new MAC, therefore allowing the recovery of low amounts of post-autogamous survivors. An alternative explanation for the residual survival observed upon Ku80c depletion could also be that KU80c transcripts are only partially degraded by RNAi, as testified by the restoration of significant amounts of full-length mRNA at T30 and T40 (Figure 3C, top lane).
We showed previously that the NHEJ-specific Ligase IV/Xrcc4 ligation complex joins broken DNA ends at IES excision sites [16]. Strikingly, after Ligase IV depletion, unrepaired broken ends remain very stable throughout autogamy, suggesting that Ku protects them against degradation. To gain further insight into the role of Ku during IES excision, we performed a molecular analysis of PGR in Ku-depleted cells. As predicted from the known role of Ku during C-NHEJ, we expected that DSBs would be introduced normally at IES boundaries, but that the broken DNA ends would not be directed to the C-NHEJ pathway.
We first focused our analysis on cells depleted in Ku80c. Using Southern blot hybridization, we monitored the excision of one particular IES, IES 51G4404 from the surface antigen G51 gene [23], during an autogamy time-course (Figure 4A). In a control RNAi (left panel), a major band corresponded to the IES− molecules from the old MAC and the newly rearranged molecules from the developing new MAC. A higher molecular weight species corresponding to the non-excised IES+ form was transiently and only barely detectable during MAC development (lanes 2′ to 4), and was completely absent at late time-points as a result of IES excision. In contrast, the IES+ form was clearly amplified during autogamy after Ku80c depletion (right panel). Using a related strategy, we tested chromosome fragmentation at one particular fragmentation site located downstream of the G51 gene (Figure 4B). We observed that non-rearranged molecules are amplified during MAC development in Ku80c-depleted cells, which indicates that Ku is also required for the elimination of the germline DNA that is associated with chromosome fragmentation.
The amplification of non-rearranged DNA indicates that PGR do not proceed normally after Ku80c depletion. However, because of the presence of old MAC DNA in our samples, the above experiment could not detect whether residual rearrangements had taken place in the new MAC. We could circumvent this problem, because the strain used in this experiment (51ΔA) harbors a wild-type germline genome, but carries a somatic deletion of the nonessential surface antigen gene A51. During autogamy, all IESs are excised normally from the A51 gene before the whole locus is deleted from the new somatic MAC [24]. Therefore, in the 51ΔA variant, all the IES− molecules originating from this locus may be attributed to de novo IES excision from the new MAC. Using PCR primers hybridizing in the flanking sequences, excised (IES−) and non-excised (IES+) molecules were readily detected in a control RNAi experiment, for IESs belonging to different classes (Figure 4C, left panel): short (51A1835: 28 bp), intermediate (51A4404: 77 bp) or long (51A2591: 370 bp), maternally (51A2591) or non-maternally controlled (51A1835 and 51A4404). This stands in sharp contrast to the complete absence of de novo IES excision junctions after Ku depletion (right panel). Likewise, using IES-specific internal divergent primers, we followed the appearance of excised IES circular molecules during the autogamy of control and KU80c-silenced cells, but no circle junctions were detected following Ku80c depletion (Figure 4D).
To extend our analysis to the other KU genes, we submitted small-scale cultures of 51ΔA cells to RNAi against KU70 or individual KU80 genes, and tested their ability to complete IES excision during autogamy. When KU80a and KU80b were knocked down, together or separately, rearranged IES− molecules appeared at day 2 of starvation, with the same timing as in a control RNAi (Figure S3). In contrast, the appearance of de novo IES excision junctions was strongly impaired upon Ku70 depletion, similar to Ku80c depletion. Taken together, our molecular data indicate that Ku70 and the development-specific Ku80c are essential for the recovery of both precise chromosomal junctions and excised IES circles during PGR. Moreover, after Ku70 or Ku80c depletion, the non-rearranged version of the genome is amplified in the developing MAC, a strikingly different phenotype from that of Ligase IV or Xrcc4 depletions [16], but quite similar to Pgm depletion [15].
The absence of precise de novo IES excision junctions after Ku70 or Ku80c depletions could either reflect a problem in DSB repair, as established for Ligase IV depletion, or an inhibition of DNA cleavage, as demonstrated for Pgm depletion. Because Ku is essential for C-NHEJ in all organisms, a defect in end-joining was expected after Ku depletion in P. tetraurelia. However, the observation that non-excised IESs were amplified in the new MAC of cells depleted for Ku70/Ku80c raised the issue of whether DSBs were actually introduced at IES boundaries. We therefore used a sensitive ligation-mediated PCR assay (LMPCR) to search for DSBs at IES boundaries after Ku depletion. As previously published [17], for those IESs that were tested, free broken DNA ends at IES boundaries were detected at early autogamy time-points in cells subjected to a control RNAi (Figure 5A). DSBs disappeared later on during MAC development, indicative of efficient repair. After Ku80c depletion, no specific broken ends were detected on the MAC side or the IES side of the DSB (Figure 5A).
An important control was to verify whether Pgm is produced and imported normally into the developing MAC upon Ku depletion. Cells expressing a Pgm-GFP fusion under the control of the endogenous PGM transcription signals were subjected to a control RNAi or to RNAi against KU80c (Figure 6). In the control, the fusion protein appeared in the developing new MAC at day 2 of starvation, and formed small foci until day 3 (Figure 6A). Pgm-GFP foci disappeared at day 4, which corresponds to the time when most PGR are completed in control cells (see Figure S3). In a KU80c RNAi, the Pgm-GFP fusion also localized specifically to the developing new MAC (Figure 6B), indicating that Ku is not required for the nuclear import of Pgm. However, Ku80c depletion triggered a dramatic increase in the nuclear amount of the Pgm-GFP fusion (Figure 6B), which parallels the accumulation of the endogenous PGM mRNA observed upon KU80c RNAi in cells harboring no fusion transgene (Figure 3C, right panels). We also observed a striking difference in the subnuclear localization of Pgm, in a KU80c RNAi relative to the control. Indeed, at day 3, Pgm-GFP accumulated in large nuclear bodies, which were clearly detectable under differential interference contrast (DIC) and coincided with DAPI-free regions (Figure 6B, right panels). At day 4, GFP fluorescence was still high, but large nuclear bodies were no longer detectable. The apparently normal subnuclear organization of Pgm-GFP observed at day 4 correlated with the recovery of functional amounts of KU80c mRNA at very late time-points (Figure 3C, top right panel).
In other organisms, Ku is known to protect broken DNA ends against degradation [1], and Ku depletion may reveal alternative DSB repair pathways, such as alt-NHEJ or HR, by allowing 5′ to 3′ DNA end resection. Should resection occur upon Ku depletion in P. tetraurelia, the resected 5′ DNA ends would not be appropriate substrates for the LMPCR assay displayed in Figure 5, because the linkers that were used only allow detection of DSBs with a specific geometry (4- or 3-base 5′ overhangs). We first investigated whether alt-NHEJ might rescue Ku depletions, by repeating the PCR assays shown in Figure 4C with more distant primers hybridizing 1 kb away from the IES excision site. This would have allowed us to detect alternative repair junctions with small deletions attributable to alt-NHEJ. Even under these conditions, however, no heterogeneous excision junctions could be detected (Figure S4). Second, because IES excision is concomitant with genome endoduplication, we reasoned that HR could use yet non-rearranged DNA molecules as templates to repair DSBs at IES excision sites: this would account for the amplification of IES+ molecules that we observed in the new MAC (Figure 4A). During HR-mediated DSB repair, the free 3′OH ends resulting from Pgm-dependent DNA cleavage would not be degraded during 5′ to 3′ resection, and should be detectable through polynucleotidyl terminal transferase (TdT) tailing [17]. In a control RNAi, indeed, free 3′OH ends were detected at IES boundaries (Figure 5B), with the same timing as the DSBs observed using LMPCR (Figure 5A). However, no free 3′OH ends were found at the expected position in a KU80c RNAi (Figure 5B). Taken together, our data do not support the hypothesis that, in cells depleted for Ku, broken IES excision sites are repaired through an alternative pathway involving 5′ to 3′ resection. Our results rather suggest that Ku is required for Pgm-dependent DNA cleavage itself.
A functional interaction between Ku and Pgm to activate DNA cleavage at IES boundaries may rely on the formation of a protein complex containing both proteins. To investigate this hypothesis, HA-tagged versions of Ku70a or Ku80c were produced in a heterologous insect cell system, together with Pgm fused to the maltose-binding protein (MBP) at its N-terminal end. For each condition, the MBP-Pgm protein was precipitated from soluble cell extracts using amylose magnetic beads, and co-precipitation of the Ku subunits was monitored on western blots (Figure 7A). We observed that Ku70a and Ku80c co-precipitated with MBP-Pgm, either individually or when the two subunits were co-expressed in the same cells. Control experiments confirmed that the enrichment in either Ku subunit is specific for the presence of MBP-Pgm in the extracts (Figure 7A). Reciprocally, we could co-immunoprecipitate Pgm with HA-tagged Ku from extracts of insect cells co-expressing both proteins, using magnetic beads coated with anti-HA antibodies (Figure 7B). The association of Pgm and Ku was resistant to DNase I (Figure 7C), suggesting that the formation of a Pgm/Ku complex does not depend upon the presence of DNA. Taken together, our data indicate that Pgm and Ku assemble in a higher-order complex in soluble cell extracts.
Gene duplication has been proposed to be a driver of genome evolution, allowing the sub-functionalization of duplicated genes and sometimes leading to the emergence of novel cellular functions [25]. In P. tetraurelia, the presence of two KU70 and three KU80 genes has been the result of successive WGDs [18]. The KU80 family provides a nice example of sub-functionalization, with KU80a and KU80b being constitutively expressed throughout the life cycle, and KU80c exhibiting a characteristic induction pattern correlating with its specific function during MAC development. Alignment of the three Ku80 proteins reveals that Ku80c differs from Ku80a and Ku80b at several positions (Figure S5), some of which may lie on the exposed surface of the α/β domain [26] and might possibly interact with additional partners. Future studies should address the question of whether the specific function of KU80c has resulted only from its overexpression or also involves a specialization of the protein in assisting PiggyMac-dependent genome rearrangements. The function(s) of the KU70 genes could not be investigated separately through RNAi. The Ku70a and Ku70b proteins are 98% identical (see Figure S5), with only conservative amino acid changes in their α/β and β-barrel domains, which suggests that they perform similar functions. However, using RNA deep sequencing, we confirmed that the two recently duplicated KU70 genes exhibit distinct transcription patterns, with KU70a being overexpressed during MAC development. We propose that, similar to KU80c, KU70a might have been undergoing specialization to carry out an essential function in PGR.
According to previous microarray hybridization analyses, successive transcription induction peaks (early, intermediate and late) were identified at the genome-wide level during autogamy [20]. KU80c and PGM belong to the same “intermediate” cluster of genes that are induced by the time PGR take place during MAC development, and are repressed at later time-points during autogamy. Our study shows that the transcription of PGM is induced normally after Ku80c depletion. However, while PGM expression decreases at late time-points in a control, after PGR are essentially completed, it is continuously turned on in cells depleted in Ku80c, while PGR are strongly inhibited. Reciprocally, after Pgm depletion, which blocks PGR, the transcription of KU80c is induced normally but is not switched off. Quite interestingly, normal progression of autogamy is observed at the cytological level upon Ku80c or Pgm depletion, indicating that cytological progression of MAC development (DNA amplification, segregation of the new developing MACs into daughter cells) in Paramecium is uncoupled from the completion of PGR. Although alternative explanations might be proposed, the observation that similar transcriptional deregulation is observed at late autogamy time-points in KU80c and PGM knockdowns suggests that the cause of aberrant transcript accumulation for genes from the intermediate cluster is a failure to complete PGR. For instance, a transcriptional activator specific for the intermediate gene cluster may be expressed from an IES-containing gene in the developing MAC, and switched off as soon as IES excision has been completed. A regulatory mechanism relying on the retention of an IES overlapping a gene promoter has been shown to control the expression of a mating-type gene in P. tetraurelia [27]. In the ciliate Euplotes crassus, PGR also regulate the expression of a development-specific telomerase gene that is localized in the germline-restricted part of the genome and is, therefore, switched off naturally once it is eliminated [28]. The existence of feedback regulatory loops provides a nice illustration of how ciliates may take advantage of PGR to fine-tune gene expression during their sexual cycle.
Upon Ku70 or Ku80c depletions, we found that germline sequences are retained in the new MAC: both IES excision and chromosome fragmentation are inhibited, which confirms that the developmental-specific Ku70/Ku80c heterodimer plays an essential role in PGR. We confirmed that PGM expression, which is required for the two types of genome rearrangements [15], is still induced in Ku-depleted cells and that Pgm still localizes to the developing new MACs. As discussed above, the persisting overproduction of Pgm at late autogamy time-points appears to be a consequence, and not the cause, of defective genome rearrangements in Ku-depleted cells.
With regard to IES excision, no de novo precise excision junctions were detected upon Ku depletion, consistent with a defective C-NHEJ pathway. Neither did we detect any imprecise junction that may have resulted from alt-NHEJ. More surprisingly, we observed that the non-rearranged version of the genome is amplified in the new MAC, and that no DSBs with the expected geometry can be detected at IES boundaries using a sensitive LMPCR molecular approach. Because IES excision starts after 3 to 4 rounds of genome amplification in the new MAC [29], we considered the possibility that HR substitutes for end-joining during the repair of IES excision sites in Ku-depleted cells, which could restore IES+ chromosomes, supposing that a yet non-rearranged sister chromatid were used as a template. During HR-mediated repair, DSBs would be processed through 5′ to 3′ resection, which would make broken ends inappropriate substrates for LMPCR. However, using a sensitive TdT-tailing assay, we obtained no evidence that HR intermediates are formed, based on the absence of detectable free 3′ ends at IES boundaries. We therefore conclude that IESs are retained in the genome of Ku-depleted cells, as a consequence of defective programmed DNA cleavage at their boundaries. Our findings point to the participation of the development-specific Ku70-Ku80c heterodimer in Pgm-dependent DNA cleavage, upstream of its likely function in C-NHEJ-mediated DSB repair.
In another ciliate, Tetrahymena thermophila, IES excision is mediated by a Pgm homolog, Tpb2p, which is responsible for DNA cleavage at IES boundaries [30], [31]. T. thermophila harbors one KU80 and two KU70 genes: Tpb2p-dependent DNA cleavage does occur in a TKU80Δ strain, but the resulting DSBs are not repaired, leading to an arrest in MAC development and to DNA loss in the new MAC [32]. These phenotypes, which are quite similar to those described for Ligase IV or Xrcc4 depletions in P. tetraurelia [16], are fully consistent with a classical scenario, in which C-NHEJ factors are recruited to broken DNA ends after DNA cleavage at IES boundaries. Interestingly, the presence of Ku is not a prerequisite for DNA cleavage in T. thermophila, and Tku80p localization remains dispersed throughout the developing MAC when Tpb2p concentrates in large heterochromatin bodies, in which DNA elimination is thought to take place [32]. It is unclear, therefore, whether Ku interacts directly with Tpb2p during genome rearrangements in this ciliate. Together with the presence of a single KU80 gene in T. thermophila, the observed differences between the two ciliates suggest that specialization of KU80c in P. tetraurelia may have occurred after the divergence between Paramecium and Tetrahymena. Noteworthy, IES excision is rather imprecise in T. thermophila, and heterogeneity at IES excision junctions could be attributed to variability in the choice of Tpb2p cleavage sites [33] or to the participation of different mechanisms in the formation of IES excision junctions, such as C-NHEJ-mediated DSB repair [32] or trans-esterification if a single boundary is cleaved [34], [35]. Accordingly, T. thermophila appears to have avoided IES insertion into coding regions [36], where imprecise excision could be deleterious.
The situation is quite different in P. tetraurelia, in which 47% of genes are interrupted by at least one IES [11]. The pressure to assemble functional open reading frames in the somatic genome has driven the emergence of a highly efficient and precise IES excision mechanism. Several known features of the excision process contribute to this precision: (i) the establishment of a crosstalk between IES ends before DNA cleavage, within a transpososome-like complex [11], [24], and (ii) the precise positioning of Pgm-dependent cleavages on each flanking TA [17]. Here, we show that the development-specific Ku70/Ku80c heterodimer is required to activate Pgm-dependent DNA cleavage at IES boundaries in vivo. Although we excluded an inhibitory effect of Ku80c depletion on PGM transcription, we cannot formally rule out that a Ku-dependent upstream developmental event activates DNA cleavage. However, our observation that Ku70a and Ku80c interact with Pgm in pull-down experiments suggests that DNA cleavage and C-NHEJ-mediated DSB repair are tightly intertwined through the incorporation of the development-specific Ku70a/Ku80c heterodimer into the DNA cleavage complex itself (Figure 8): programmed DSBs, therefore, would be efficiently directed towards C-NHEJ-mediated precise repair. We propose that the Ku70/Ku80c heterodimer is an essential Pgm partner that activates the assembly of a transpososome-like nucleoprotein complex competent for DNA cleavage. In support to this model, we found that Ku70a and Ku80c interact with Pgm in cell extracts. During IES excision, Ku and Pgm may associate once Pgm is bound to IES boundaries (Figure 8A: activation of Pgm-dependent cleavage), or before Pgm interacts with DNA, as suggested by our observation that the Pgm/Ku complex is resistant to DNase I in vitro (Figure 8B: activation of the DNA binding activity of Pgm). Further biochemical work is needed to analyze the Ku/Pgm interaction and clarify the mechanisms involved in Ku-mediated activation of DNA cleavage. We cannot exclude at this stage that the DNA-PKcs is also a component of the DNA cleavage complex (Figure 8): indeed, a parallel study established that DNA-PKcs depletion impairs the excision of at least a subset of IESs, which exhibit a reduced efficiency of DNA cleavage at their boundaries (Malinsky et al., in prep.). The incorporation of early C-NHEJ proteins in the cleavage complex would allow the cell to face the challenge of repairing efficiently and precisely tens of thousands of DSBs during the short period of MAC development. In contrast, Ligase IV and Xrcc4 are clearly not required for DNA cleavage [16] and are probably not part of the cleavage complex.
Domesticated transposases from cut-and-paste DNA transposons might have been recruited to perform PGR in various systems, not only because of their DNA cleavage activities, but perhaps also because of the particular features of cut-and-paste transposition [37]. When transposons integrate into their target site, they duplicate a short sequence, the TSD (target site duplication), on each side of the integrated element. During the next round of transposition, the transposase cuts the DNA, excises the transposon and leaves a DSB at the donor site, which is repaired through cellular pathways. The study of cut-and-paste transposons has revealed that C-NHEJ, as opposed to alt-NHEJ, accounts for most end-joining events during DSB repair at transposon donor sites (reviewed in [5]). Moreover, several cut-and-paste transposons and/or their transposases interact with Ku. For instance, Ku70 binds the ends of the P element from Drosophila melanogaster and stimulates DSB repair at the donor sites [38]. In vitro, the transposase of Sleeping Beauty, a reconstructed Tc/mariner transposon, forms a complex with Ku70, and efficient transposition of Sleeping Beauty in a cellular system depends on the presence of DSB repair proteins [39]. However, it is not clear in vivo whether Ku/transposase interaction activates DNA cleavage at transposon ends or simply facilitates the recruitment of the C-NHEJ pathway once the donor site has been broken.
With regard to PGR, the formation of a complex involving a nuclease and DSB repair factors has been hypothesized during V(D)J recombination [40]. In vitro, Ku interacts with the domesticated transposase RAG1 [41], but no evidence has been provided that Ku is required in vivo for the introduction of RAG1-dependent programmed DSBs. As in Tetrahymena, the formation of nuclear recombination centers, or recombination factories, may facilitate V(D)J recombination by bringing together the recombination sites, the RAG1/RAG2 endonuclease and DSB repair factors. However, in this system, the presence of C-NHEJ proteins appears dispensable for DNA cleavage itself. The interplay between DNA cleavage and DSB repair has been pushed one step further in Paramecium. Indeed, the present study of IES excision in P. tetraurelia provides the first evidence that Ku is absolutely required in vivo to introduce programmed DSBs during PGR. The demonstration that DNA cleavage mediated by a transposase-related protein and C-NHEJ mediated repair are coupled during PGR in P. tetraurelia is reminiscent of the observation that, during meiosis in S. cerevisiae, the Mre11p HR protein is required for DNA cleavage by the topoisomerase-like Spo11 endonuclease [42]. These two systems support the notion that the presence of DSB repair factors in recombination factories may be a prerequisite for DNA cleavage during programmed genome rearrangements.
For autogamy time-course experiments, we used P. tetraurelia strain 51 new (hereafter called 51) and its 51ΔA variant carrying a heritable deletion of the A gene in its MAC but harboring a wild-type MIC [24]. To facilitate the screening of transformants in microinjection experiments, we introduced the nd7-1 mutation [43] into strain 51 by conjugation, or used somatic variants carrying a MAC deletion of the ND7 gene [44]. Cells were grown at 27°C in a wheat grass infusion (WGP; Pines International Inc.) inoculated with Klebsiella pneumoniae. Autogamy was carried out through starvation as described [29]. Total RNA and genomic DNA were extracted from ∼400,000 cells for each time-point and quantified as described [15].
Oligonucleotides were purchased from Sigma-Aldrich or Eurofins MWG Operon (Table S1).
PCR amplifications were performed in a final volume of 25 µL, with 10 pmol of each primer, 5 nmol of each dNTP and 1 U of DyNAzyme II DNA polymerase (Finnzymes) or DreamTaq (Thermo Scientific) according to the enzyme suppliers' recommendations. PCR products were analyzed on 3% NuSieve GTG agarose gels (BioWhittaker Molecular Applications). LMPCR detection of double-strand breaks was performed as described [24]. Poly(C) tailing of free 3′ ends using terminal transferase was performed as described [17], using 500 ng of input total genomic DNA. Sanger DNA sequencing was performed at GATC Biotech, or using the fmol DNA Cycle Sequencing System (Promega). Northern and Southern blot hybridization with 32P-labeled probes was carried out as described [15]. The KU70 and KU80c probes are described in Figure S2 and Table S1. The sequence of the 17S rRNA oligonucleotide probe is shown in Table S1.
Strand-specific RNA-seq libraries were prepared from 50 ng polyA+ RNA following the directional mRNA-seq library preparation protocol provided by Illumina: RNAs were fragmented using fragmentation buffer, purified and treated with phosphatase and kinase prior to sequential ligation with different RNA adapters to the 3′ and 5′ ends. The ligated RNA fragments were reverse-transcribed, followed by PCR amplification. Each library was sequenced using an Illumina Genome Analyzer IIx to generate 75-nt paired-end reads. Reads were mapped with TopHat2 [48] (read-mismatches 1; min-intron-length 15; max-intron-length 100) on the P. tetraurelia strain 51 reference MAC genome [11]. Alignments were indexed using Samtools [49] and a custom perl script was used to count uniquely mapped fragments for each gene model. The counts were normalized to account for gene size and the total number of mapped fragments.
For the construction of in-frame GFP-KU fusions, a GFP-coding fragment adapted to Paramecium codon usage [50] was added by PCR fusion to the 5′ end of the KU70a or KU80c genes. Each construct was inserted in a pUC18 plasmid between the SphI and SacI sites. As a result, the GFP is fused to the N-terminus of Ku70a and Ku80c and the fusion proteins are expressed under the control of the KU70a and KU80c transcription signals (promoters and 3′UTR), respectively. The PGM-GFP fusion will be described in detail elsewhere (Dubois et al., in prep.). Briefly, the GFP-coding fragment was added by PCR fusion to the 3′ end of the PGM gene carried by plasmid pPBL49g [15]: in the resulting construct, the PGM-GFP coding sequence is flanked by 96 bp upstream of the ATG (i.e. the putative endogenous PGM promoter) and 54 bp downstream of the TGA stop codon (including the endogenous 3′ UTR and the polyadenylation site).
Plasmids encoding GFP fusion proteins were linearized by appropriate restriction enzymes and microinjected into the MAC of vegetative 51 nd7-1 or 51ΔND7 cells, as described [15]. A complementing plasmid carrying a functional ND7 gene [43] was coinjected with the fusion transgenes to facilitate the selection of transformants. During autogamy, cells were permeabilized for 4 min in PHEM (60 mM Pipes, 25 mM Hepes, 10 mM EGTA, 2 mM MgCl2 pH 6.9) +1% Triton, then fixed for 10 min in PHEM +2% paraformaldehyde. All observations were performed using a Zeiss Axioplan 2 Imaging epifluorescence microscope. Developing MACs were identified using Nomarski differential interference contrast (DIC) combined with 4′,6-diamidino-2-phenylindole (DAPI) staining. No lethality was observed in the post-autogamous progeny of transformed cells.
|
10.1371/journal.ppat.1000205 | Alteration of Blood–Brain Barrier Integrity by Retroviral Infection | The blood–brain barrier (BBB), which forms the interface between the blood and the cerebral parenchyma, has been shown to be disrupted during retroviral-associated neuromyelopathies. Human T Lymphotropic Virus (HTLV-1) Associated Myelopathy/Tropical Spastic Paraparesis (HAM/TSP) is a slowly progressive neurodegenerative disease associated with BBB breakdown. The BBB is composed of three cell types: endothelial cells, pericytes and astrocytes. Although astrocytes have been shown to be infected by HTLV-1, until now, little was known about the susceptibility of BBB endothelial cells to HTLV-1 infection and the impact of such an infection on BBB function. We first demonstrated that human cerebral endothelial cells express the receptors for HTLV-1 (GLUT-1, Neuropilin-1 and heparan sulfate proteoglycans), both in vitro, in a human cerebral endothelial cell line, and ex vivo, on spinal cord autopsy sections from HAM/TSP and non-infected control cases. In situ hybridization revealed HTLV-1 transcripts associated with the vasculature in HAM/TSP. We were able to confirm that the endothelial cells could be productively infected in vitro by HTLV-1 and that blocking of either HSPGs, Neuropilin 1 or Glut1 inhibits this process. The expression of the tight-junction proteins within the HTLV-1 infected endothelial cells was altered. These cells were no longer able to form a functional barrier, since BBB permeability and lymphocyte passage through the monolayer of endothelial cells were increased. This work constitutes the first report of susceptibility of human cerebral endothelial cells to HTLV-1 infection, with implications for HTLV-1 passage through the BBB and subsequent deregulation of the central nervous system homeostasis. We propose that the susceptibility of cerebral endothelial cells to retroviral infection and subsequent BBB dysfunction is an important aspect of HAM/TSP pathogenesis and should be considered in the design of future therapeutics strategies.
| The blood–brain barrier (BBB) forms the interface between the blood and the central nervous system (CNS). BBB disruption is considered to be a key event in the pathogenesis of retroviral-associated neurological diseases. The present paper deals with the susceptibility of the endothelial cells (i.e., one of the main cellular components of BBB) to retroviral infection, and with the impact of infection in BBB function. This study focuses on the Human T-Lymphotropic Virus (HTLV-1), which infects 20 million people worldwide, and is the etiological agent of a neurodegenerative disease called HTLV-1 Associated Myelopathy/Tropical Spastic Paraparesis (HAM/TSP). We first demonstrated that the cerebral endothelial cells express the receptors for the retrovirus in vitro, and on spinal cord autopsy sections from non-infected and HAM/TSP patients. We found on these latter that vascular-like structures were infected and confirmed in vitro that the endothelial cells could be productively infected by HTLV-1. We demonstrated that such an infection impairs BBB properties in vitro, as well as tight junctions, that are cell adhesion structures. This study is the first to demonstrate the impact of HTLV-1 infection on human BBB integrity; such a susceptibility has to be considered in the design of future therapeutics strategies.
| The Blood-Brain barrier (BBB) constitutes the interface between the blood and the central nervous system (CNS). It is composed of astrocytes, pericytes and brain microvascular endothelial cells. This latter cell type forms the major structural and functional element of the BBB, with endothelial cells sealed together with Tight Junctions (TJs). Under physiological conditions, the BBB maintains CNS homeostasis and selectively regulates intracellular and paracellular passage of ions, molecules and cells [1].
BBB integrity is compromised during retroviral infection; for example, BBB breakdown has been reported during Human Immunodeficiency Virus Type 1 (HIV) infection, especially during HIV-related encephalitis and HIV-associated dementia [2]. One to three percent of the 20 million people infected worldwide by the retrovirus HTLV-1 (for human T-lymphotropic virus type 1) develop HTLV-Associated Myelopathy/Tropical Spastic Paraparesis (HAM/TSP) [3]. This is a slowly progressive paraplegia of the lower extremities, involving demyelination and neuronal degeneration mainly in the thoracic spinal cord. BBB disruption has been attested in HAM/TSP patients by several lines of evidence, such as fibrinogen leakage and IgG deposits in CNS parenchyma [4] as well as lymphocyte passage through brain endothelium [4]–[7]. As previously shown, BBB disruption is associated with alterations in tight junctions between endothelial cells in the vasculature of a HAM/TSP patient [8].
The mechanisms of BBB disruption during retroviral-associated pathologies are not yet fully understood. Most studies focus on the effect of soluble molecules secreted by infected lymphocytes on BBB functions and intercellular TJ organization. In the case of HIV infection, the viral protein Tat has been shown to induce an inflammatory process in brain endothelial cells, or endothelial cell apoptosis [9], and to be able to disrupt the intercellular TJs [10]. In the context of HTLV-1 infection, we recently demonstrated that proinflammatory cytokines, such as IL-1α and TNFα, secreted by infected lymphocytes, are sufficient to disrupt TJs between human brain endothelial cells and induce permeability changes [8].
Alternative mechanisms could contribute to BBB dysfunction associated with HTLV-1 infection. Although neurological disease in mice infected with the PVC-211 Murine Leukemia Virus has been associated with infection of brain endothelial cells [11], the infection of brain endothelial cells by human retroviral agents and its role in BBB breakdown is still a matter for debate. In the case of HIV infection, a number of earlier studies reported infection of endothelial cells in adult brain tissue [12]–[14], based upon morphological appearance and vascular localization of cells found positive by immunocytochemistry, in situ hybridization or PCR-in situ hybridization for viral transcripts. Conflicting results were obtained in vitro from brain-derived endothelial cells (for review, see [15]).
In the case of HTLV-1, no evidence for infection of human brain endothelial cells has been reported so far, most likely due to the rarity of material from patients with HAM/TSP, and the low level of HTLV-1 expression in tissues. Although an increased adherence of T lymphocytes from HAM/TSP patients to human brain endothelial cells has been observed [16], the main data concern extra-neural endothelial cells: it has been demonstrated in vitro that human venous endothelial cells derived from umbilical cords are susceptible to HTLV-1 infection [17],[18], and that HTLV-1 proviral DNA could be detected in dermal endothelial cells ex vivo [19].
In this study, we investigated the susceptibility of human brain endothelial cells to HTLV-1 infection, and its possible consequences on BBB integrity, both in vitro, in a human brain endothelial cell line, and ex vivo on spinal cord autopsy sections from HAM/TSP patients. We found that human brain endothelial cells can be productively infected in vitro by HTLV-1, with consequent alterations in the BBB, evidenced by increased lymphocyte migration and passage of small molecules through endothelium. These data provide a basis for and transient BBB alterations that may be observed during BBB pathogenesis.
Three cellular components have been identified as forming part of the HTLV-1-entry complex: heparan sulfate proteoglycans (HSPGs) [20],[21], Neuropilin-1 [22], a co-receptor for VEGF165 and semaphorin 3a, and the glucose transporter Glut-1 [23].
The expression of HSPGs in BBB endothelial cells has previously been reported; in vivo, HSPGs are ubiquitously expressed at the cell surface or throughout the extracellular matrix of all mammalian tissues [24]; in particular, HSPGs have been previously detected in cerebral blood vessels [25].
In this study, we examined the expression of Glut-1 and Neuropilin-1 in the endothelial cells that form the BBB in situ. As HAM/TSP is characterized by lymphocyte infiltration and inflammation mainly within the thoracic spinal cord, we focused our investigation on this region.
In tissue sections derived from the thoracic spinal cord of uninfected control cases, Glut-1 expression was diffuse and patchy on fibres throughout the grey matter, intense on fibers in the dorsal root and also detected in the meninges surrounding the cord (Fig. 1A). Glut-1 was expressed prominently on blood-vessels within both the white and the grey matter (Fig. 1A and B). The Neuropilin-1 (NP-1) was expressed diffusely in the posterior, lateral and anterior columns, and in the dorsolateral fasciculus/dorsal root (Fig. 1D). Cellular expression was particularly noted at the apex of the posterior column, and on motor neurons in the anterior column (data not shown). Significantly, NP-1 was highly expressed on blood vessels, and within the meninges in all segments of the thoracic spinal cord (Fig. 1D and E). Vascular endothelial cell expression of Glut-1 and NP-1 was confirmed by double immunolabeling with Factor VIII, a specific marker for endothelial cells (Fig. 1C and F).
We then determined whether expression of these receptors by endothelial cells was conserved in HAM/TSP. In tissue sections of the thoracic spinal cord derived from HAM/TSP patients, Glut-1 immunoreactivity was detected in blood vessels, in the absence (Fig. 2A) or presence of cell infiltrates (Fig. 2B and C). Neuropilin-1 could also be detected in blood vessels in sections from HAM/TSP patients (data not shown).
Since microvascular endothelial cells that constitute the BBB express HTLV-1 receptors, we examined whether the infection of these cells by HTLV-1 could be detected in situ, by performing in situ hybridization for a viral mRNA (the messenger that encodes the viral transactivator Tax) on the spinal cord sections. Cellular infiltrates were positive for viral Tax mRNA (data not shown). However, we focused our analyses on spinal cord regions where the infiltrates were absent, to prevent the signal within the infected lymphocytes from masking the signal from resident cells within the CNS parenchyma. Since astrocytes are known to be targets of HTLV-1 infection [26],[27], the detection of a positive signal in several GFAP immunoreactive cells constituted a suitable positive control (Fig. 2D), as shown in previous studies [4]. We also encountered rare positive signals associated with vascular structures (Fig. 2E). This observation suggested the possibility of infection of cerebral endothelial cells by HTLV-1.
Although viral transcripts were found to be associated with blood vessels, this observation could not be taken as definitive evidence for the infection of endothelial cells forming the BBB specifically as other cell types such as pericytes are closely associated with endothelial cells. We therefore took advantage of a human-derived brain endothelial cell line, hCMEC/D3, that had been previously reported to retain many BBB characteristics [28]. We first investigated the expression of the 3 (co-)receptors for HTLV-1 entry by flow cytometry analysis.
Glut-1 expression was highly detected in permeabilized hCMEC/D3 cells. In order to detect expression of this protein at the cell surface, immunostaining was performed on fixed but non-permeabilized cells. The expression of Glut-1 at the cell surface was detected in 31% of the hCMEC/D3 cells (Fig. 3A). Similarly, cell surface expression of NP-1 and HSPGs were detected in 82% and 59% of the hCMEC/D3 cells respectively (Fig. 3B and C).
In order to determine if the expression of surface receptors on endothelial cells allows HTLV-1 entry, we investigated whether hCMEC/D3 cells could be infected by HTLV Env-pseudotyped LacZ vectors. β-Gal production was detected upon infection of hCMEC/D3 cells by A-MLV as well as by H-MLV pseudotypes. Addition of serum from an uninfected donor did not reduce infection by either pseudotype. In contrast, addition of serum from an HTLV-1-infected HAM/TSP patient abolished the infection by H-MLV, while high level of A-MLV pseudotype infection was still observed (Fig. 3D). These results indicate that hCMEC/D3 cells allow HTLV Env-mediated entry.
The ability of a retrovirus such as HTLV-1 to enter a particular cell type is usually correlated with the ability of target cells to fuse and form syncytia with infected T lymphocytes. We therefore quantified the number of syncytia in hCMEC/D3 and lymphocyte cocultures. Whereas hCMEC/D3 did not fuse with C81-66 lymphocytes (HTLV-1-infected cells that do not express Env glycoproteins), numerous syncytia were observed in co-cultures with infected MT2 lymphocytes at 24 hours. The number of nuclei per syncytium ranged from 3 to 11 (Fig. 4A). Syncytia were found immunoreactive for viral p24 (Fig. 4C). In order to ascertain the endothelial origin of the syncytia, we prestained the hCMEC/D3 cells with a vital fluorescent molecule (CellTracker Red CMTPX). Syncytia were shown to be fluorescently labeled (Fig. 4D).
Addition of serum from an uninfected patient did not prevent the formation of syncytia. In contrast, a dramatic reduction in the number and the size of syncytia was observed when serum from an HAM/TSP patient was added to the medium (Fig. 4A), showing that the fusion is HTLV-mediated. Similarly, syncytia formation could also be inhibited by addition of VEGF165, a physiological ligand of NRP-1 that has been shown to inhibit the binding of the HTLV-1 Env proteins to target cells [22], of dextran sulfate, an inhibitor of HSPG-mediated HTLV-1 entry [20], or of an antibody directed against the glucose transporter Glut-1 (Fig. 4B). It is worth noting that no significant alterations in syncytia formation were observed following addition of an irrelevant isotype-matched antibody to the culture medium.
These data suggest that HTLV-1 entry into hCMEC/D3 cells is dependent on the interactions between viral envelope proteins and the three putative cellular receptors for HTLV-1 infection (heparan sulfate proteoglycans, neuropilin-1 and Glut-1).
As HTLV-1 can enter hCMEC/D3 cells, we then determined whether this event allows a productive infection. Endothelial cells were co-cultivated with irradiated lymphocytes. The irradiation dose was lethal for these cells, as confirmed by Trypan blue staining which indicated that 100% of the lymphocytes were dead by day 8 post-irradiation (data not shown).
HTLV p19 was detected in the supernatants of co-cultures but decreased during the 10 first days: this correlated with lymphocyte and syncytia cell death (data not shown). At 10 days post coculture, no syncytium could be observed. From day 13 post-contact, p19 production to reach about 510 pg/day at day 22, indicating that human brain endothelial cells produced viral proteins (Fig. 5A). HTLV-1 productive infection could be prevented by addition of AZT, an inhibitor for the reverse transcriptase, to the culture medium. The percentage of positive endothelial cells for viral p24 increased in parallel to p19 levels detected in the supernatant reach maximal infectivity rates at day 22, with 18% of the endothelial cells positive for p24 (Fig. 5B). No cytopathic effect of HTLV-1 infection endothelial cells was observed.
At day 22 post co-culture (2 days after renewal of the medium), the supernatant was collected and ultracentrifuged. The pellet was resuspended and added to culture medium of reporter 293T-LTR-GFP cells. After 6 days of culture, fluorescent cells were visualized, demonstrating the presence of Tax protein within these reporter cells. Furthermore the fluorescent cells could form syncytia, and expressed the viral envelope protein (Fig. 5C). In addition, the number of the fluorescent cells was dramatically reduced (up to 37%) in the presence of AZT indicating that the detected fluorescence signal in 293T-LTR-GFP cells was specific to de novo infection of reporter cells. These results indicate that cerebral endothelial cells produced infectious viral particles.
Lastly, we assessed the impact of HTLV-1 infection of endothelial cells on BBB function, by evaluating the paracellular permeability of hCMEC/D3 cell monolayers and the transmigration of lymphocytes through the barrier in an infected and non-infected context.
hCMEC/D3 cells, incubated with irradiated C81-66 (HTLV-1-infected lymphocytes that are not productively infected) or with MT2 cells 15 days prior to the experiment were seeded on filters and allowed to reach confluence. The paracellular permeability of monolayers of hCMEC/D3 cells infected with HTLV-1 was much higher than the permeability of hCMEC/D3 cells previously cocultured with control C81-66 T-lymphocytes (Fig. 6A). Similarly, transmigration of uninfected T-lymphocytes (CEM and Jurkat) across monolayers of hCMEC/D3 cells infected with HTLV-1 was increased compared to that across a monolayer of uninfected hCMEC/D3 cells (Fig. 6B). In addition, the migration of HTLV-1-infected lymphocytic cell lines through hCMEC/D3 cells was increased compared to control lymphocytes as previously reported [8]. However, no differences in HTLV-1-infected lymphocyte transmigration across HTLV-1-infected or non-infected hCMEC/D3 cells were observed. These data indicate that HTLV-1-infection of endothelial cells alters classical BBB functions.
We then analyzed the expression levels of proteins that constitute the tight junctions by Western blot. As indicated in Fig. 6C, ZO-1 levels were dramatically reduced in HTLV-1-infected hCMEC/D3 cells whereas, in the case of occludin, the expression levels of the 55 kDa isoform, but not those of the 60 kDa isoform were decreased.
Since we have previously shown that the Myosin Light Chain Kinase (MLCK) is important in tight junction regulation of endothelial cells incubated with HTLV-1 infected lymphocytes, we determined whether the inhibition of MLCK activity could prevent long-term barrier impairment of HTLV-1 infected endothelial cells (Fig. 6D). Treatment of the endothelial cells for 24 h with the MLCK inhibitor ML7 failed to restore low paracellular permeability in HTLV-1 infected hCMEC/D3 cells. The molecular mechanisms of TJ disruption in endothelial cells induced by direct HTLV-1 infection appear to be different to those induced by HTLV-1-infected lymphocytes, as previously demonstrated [8].
The BBB constitutes an interface between the bloodstream and CNS parenchyma, and regulates the intracellular and paracellular passage of molecules and cells between the two compartments [1].
Disruption of the BBB in HAM/TSP is strongly suggested by observations of perivascular cuffing with lymphocyte and macrophage infiltrates [4],[5],[7], and confirmed by reports of fibrinogen and IgG deposits in the CNS parenchyma of HAM/TSP patients [4]. BBB breakdown is an important step in HAM/TSP pathogenesis, especially by facilitating migration of lymphocytes into the CNS. Infiltrated lymphocytes are believed to cause demyelination and axonal degeneration that are hallmarks of HTLV-1-associated neuropathology [29]. The mechanisms underlying BBB alteration during HAM/TSP are not yet well determined. In an in vitro model of the BBB, composed of human cerebral microvascular endothelial cells [28], we previously demonstrated the importance of proinflammatory cytokines secreted by infected lymphocytes in the early stages of BBB disruption [8],[30]. In the present study we investigated whether BBB endothelial cells were susceptible to HTLV-1 infection and the impact of such an infection on BBB integrity.
BBB dysfunction associated with retroviral infections has been previously described. In a murine model, the infection of brain endothelial cells has been reported both in vitro and in vivo in the case of PVC-211 murine leukemia virus (a neuropathogenic variant of the Friend MuLV), with a direct correlation between the replication efficiency of a virus in brain endothelial cells in vitro and its ability to cause neurological disease in vivo [11]. Moreover, in the case of infection by the Feline Immunodeficiency Virus, infection of brain endothelial cells has been proposed to represent one of the ways of viral entry into the CNS [31]. Meanwhile, infection of brain endothelial cells by human retroviral agents is still a matter of debate. In the case of HIV infection, previous studies have reported infection of endothelial cells in adult brain tissue [12]–[14], but these have been mainly based upon interpretation of the morphological appearance and vascular localization of cells found positive by immunocytochemistry, in situ hybridization or PCR-in situ hybridization, whereas conflicting results have been obtained in vitro from brain-derived endothelial cells (for review, see [15]). Up until now, no study has focused on the subject of human endothelial cell infection by HTLV-1.
Recently three membrane proteins have been described as components of the receptor for HTLV-1 entry: the glucose transporter Glut-1 [23], and a receptor for VEGF, Neuropilin-1 [22] and heparan sulfate-proteoglycans [21],[32]. Vascular expression of Glut-1 [33],[34] and Neuropilin-1 [35],[36] has been previously reported within the CNS, under normal and pathological conditions. We have shown that the HTLV-1 receptors are expressed on blood vessels of the adult human thoracic spinal cord, a region which is characterized by BBB impairment, lymphocyte infiltration and inflammation during HAM/TSP. Expression of these proteins was found in endothelial cells within the spinal cord of HAM/TSP patients, irrespective of the extent of lymphocyte infiltration. We could also detect expression of these receptors at the cell surface of a human cerebral endothelial cell line, hCMEC/D3.
We then looked for HTLV-1 infected endothelial cells by in situ hybridization using DNA probe directed against transactivator Tax transcripts on spinal cord sections of a HAM/TSP patient. Astrocytes are reportedly susceptible to HTLV-1 infection in vitro [26],[27]. Astrocytic infection with HTLV-1, also reported using this technique in CNS tissues (Ozden et al. 2002), served as a positive control in our samples. HTLV-1 transcripts were found only very rarely associated with the vasculature. The difficulty in detecting the virus within endothelial cells in situ may be in part due to the specific immune response developed against HTLV-1 in HAM/TSP patients [37]. HTLV-1 infected endothelial cells may be detected and lysed by cytotoxic lymphocytes, thereby constituting a new mechanism for transient BBB disruption, which remains to be explored further. Moreover, the tissue sections studied are from a patient with a rapid progressing HAM/TSP [4]. Additional studies in other HAM/TSP patients, with slower disease progression, might facilitate the detection of HTLV-1-infected endothelial cells.
As the ex vivo detection was difficult to ascertain, we studied the susceptibility of an in vitro model, hCMEC/D3, to HTLV-1 infection. We confirmed that the expression of the receptors at the cell surface of hCMC/D3 cells allowed viral entry. In fact, it has been demonstrated that cholesterol, a central component of lipid rafts, was necessary for HTLV-1 infection, especially on post-binding entry steps [38]. Thus we examined whether the spatial organization of viral receptors on the membrane of cerebral endothelial cells would enable infection with HTLV. Using LacZ-reporting viral particles pseudotyped with the HTLV envelope [23], we demonstrated that the envelope recognized and fused correctly with the membrane of endothelial cells. Expression of functional receptors at the cell surface are usually also demonstrated via the ability of target cells to form syncytia with infected lymphocytes [39]. Syncytia were indeed formed during co-cultures of infected lymphocytes and hCMEC/D3, and this could be prevented by addition of serum from an HAM/TSP patient, confirming the role of viral envelope in the process of cell-cell fusion. Similarly, the formation of syncytia could be prevented by addition of binding inhibitors directed to the previously described HTLV-1 receptors, demonstrating that it is a proper receptor-mediated entry. The death of cells forming syncytia was observed after one week in culture, as reported for other cell types [40], but p19 detection in the supernatant at later stages confirmed the persistent infection of endothelial cells. The production of viral proteins corresponds to that of the infectious virus, as shown using 293T-LTR-GFP reporter cells assays.
Of note, direct infection of cerebral endothelial cells may constitute a new mechanism of entry for the retrovirus into the CNS. In fact, endothelial cells and astrocytes are tightly associated at the BBB; infected endothelial cells could then transmit infectious particles to astrocytes, which have been shown to be susceptible to HTLV-1 infection both in vitro and in vivo [4],[26],[27],[41].
Finally, we analyzed the impact that infection of endothelial cells had on BBB integrity. At day 20 following co-culture with irradiated lymphocytes, the endothelial cells showed no cytopathic effect and no syncytia was observed. We demonstrated that HTLV-infected hCMEC/D3 cells could no longer form confluent monolayers resistant to molecular diffusion and cellular migration. Indeed, BBB dysfunction results in enhanced transendothelial migration of both HTLV-1-infected and uninfected lymphocytes. Both lymphocytes carrying HTLV-1 and tight junction opening could facilitate the spread of the virus and cytotoxic lymphocytes into the CNS. The mechanisms for such an alteration are not completely understood and may be multifactorial. We previously demonstrated that inflammatory cytokines produces by HTLV-1 infected lymphocytes induced BBB disruption, by increasing the expression and activity of the Myosin Light Chain Kinase (MLCK), which is transcriptionally regulated by the NF-κB pathway [8]. Since Tax expression induces cytokine secretion [42],[43],[44], disruption of the BBB could be a consequence of the autocrine or paracrine effects of proinflammatory cytokines secreted by HTLV-1 infected endothelial cells. This could explain why a low infection rate (<20%) of the cells is sufficient to significantly alter BBB-related functions. However, ML-7 treatment (an inhibitor of MLCK activity) of virally infected endothelial cells could not prevent such a disruption. This is consistent with the observations of McKenzie and Ridley who recently showed that MLCK activity is not required for long term modification of TJ protein expression, mediated by TNFα [45].
Further investigations could focus on possible interactions between viral Tax protein and proteins that constitute TJs. For example, the protein ZO-1 bears a PDZ domain [46] and Tax a PDZ binding domain [47] and suggest that these proteins interact with each other. This interaction could induce relocalization of ZO-1 and disorganize the TJ, or the protein could directly target the proteasome, as Tax does for the Retinoblastoma protein [48].
In conclusion, we have shown that endothelial cells, which constitute the BBB, are susceptible to infection by HTLV-1. This represents a new mechanism for BBB disruption in HAM/TSP, either directly as the expression of Tax induces a loss of BBB functions, or indirectly as the infected endothelial cells will be targeted by the immune system. Moreover, as the infection is productive, endothelial cells could allow entry of the virus into the CNS and facilitate the infection of astrocytes within the CNS parenchyma. Some may argue that the disruption of the BBB due to the infection of the endothelial cells seem to be a minor event in the natural course of HAM/TSP pathogenesis when compared to the proinflammatory cytokines [8]. However, considering this possibility should not be neglected for the design of potential new treatments. For example, the inhibition of the VEGF has been proposed previously as a way to prevent lymphocyte migration throughout the vasculature [49],[50]. Our results suggest that such a treatment should be envisioned with caution in the context of BBB, as it could increase the availability of the Neuropilin-1 to the virus and thereby, could facilitate infection of the BBB by HTLV-1.
The human Cerebral Microvascular Endothelial Cell line, hCMEC/D3, was immortalized after transduction with lentiviral vectors encoding the catalytic subunit of human telomerase hTERT and SV40 T antigen, as described previously [28]. hCMEC/D3 cells were grown in Endothelial Growth Medium-2 (EGM-2MV, Clonetics, Cambrex Biosciences, Workingham, UK) without hydrocortisone, on Biocoat tissue culture flasks (BD Biosciences, Bedford, MA).
MT-2 and C81-66 were used as HTLV-1-infected T cell lines. These are cell-lines derived from human umbilical cord blood T cells, following culture with irradiated cells from at ATLL patient. Both cell lines express the viral transactivator Tax-1, although C81-66 does not produce any viral particles. CEM and Jurkat were used as uninfected control T-cell lines. Non-adherent cell lines were grown in RPMI 1640 medium (Gibco BRL, Gaithersburg, MD) supplemented with 1 mM glutamine and 10% FCS.
The HEK epithelial cells containing an integrated HTLV-1 long terminal repeat (LTR) coupled to a green fluorescent protein (GFP) reporter gene (called herein 293T-LTR-GFP) [51] were grown in Dulbecco's modified Eagle's medium (Gibco BRL) supplemented with glutamine (1 mM), 100 U/ml penicillin, and 10% FCS.
Frozen tissue autopsy sections from a HAM/TSP patient were obtained as previously described [4]. For one experiment, we used paraffin embedded material from the thoracic spinal cord of a HAM/TSP patient from Chile, whose case has been previously described [52]. Frozen spinal cord tissues from uninfected control patients were obtained from the UK Multiple Sclerosis Tissue Bank and as previously described [53]. All tissue specimens were obtained in accordance with the respective hospital and national regulations and ethical rules.
Fixed frozen (4% paraformaldehyde, PFA) and snap frozen blocks of tissues were sectioned serially at 10 µm using a cryostat; sections were then air-dried, and fixed in methanol before processing for immunohistochemistry. Sections were either labeled by immunoperoxidase technique (3-step procedure, DAKO, Glostrup, Denmark) or alcaline phosphatase technique (Vectastain, Vector Laboratories, CA, USA). Endogenous peroxidase was blocked with 2,5% hydrogen peroxidase in methanol, and non-specific labeling blocked with normal sera corresponding to the secondary antibody species. Sections were dehydrated in graded alcohols, cleared in xylene and coverslipped using permount. The primary antibodies used were a mouse antibody to Glut-1 (MAB 1418, R&D systems, Minneapolis, MN, USA), to NP-1 (clone A-12, Santa Cruz Biotechnology, Santa Cruz, CA, USA), or to Factor VIII (DAKO).
Cultures were fixed with 4% PFA. Staining with primary antibody was performed after incubation for 30 minutes with 10% normal goat serum and 0.05% saponin diluted in PBS. The following primary antibodies were used: mouse antibodies to HTLV-1-p24 (ab9081, Abcam, Cambridge, UK), to Glut-1 (R&D systems), to NP-1 (Santa Cruz Biotechnology). Specific secondary antibodies were coupled with Fluorescein (Vector Laboratories). After washes, preparations were mounted in DAPI-containing Vectashield medium (Vector laboratories).
ISH was performed on serial sections of frozen tissues by means of 32P antisense and sense riboprobes corresponding to the complete tax mRNA, as described elsewhere [54]. Sections were stained prior to ISH by immunoperoxidase using a rabbit polyclonal antibody against Glial Fribrillary Acidic Protein (anti-GFAP, DAKO).
Cells were analyzed for surface or intracellular receptor expression, with or without permeabilization with triton. Cells were resuspended in PBS-EDTA and incubated at 4°C for 30 min with a primary antibody to NP-1, Glut-1, HSPGs (clone F69-3G10, Seikagaku Corp., Tokyo, Japan) or HTLV-1 p24. Secondary antibody staining was performed by incubating the cells with fluorescein-isothiocyanate (FITC)-labeled antibody (Vector Laboratories) at 4°C for 30 min. Cells were washed twice with phosphate-buffered saline (PBS) and analyzed by using a FACScan flow cytometer and Cellquest software (Becton Dickinson, San Jose, CA, USA).
Replication-defective LacZ retroviral vectors pseudotyped with either the HTLV (H-MLV), or amphotropic murine leukemia virus (A-MLV) envelope proteins were produced by transfection of 293T cells with Gag/Pol, Env, and LacZ plasmids as described elsewhere [23]. Target cells were plated on 24-well plates (5.104) for 24 h and supernatants from H-MLV (1/5), or A-MLV (1/100) pseudotype-producing cells were added. Infectivity was assessed 48 h later by measuring the level of lacZ activity with the β-Gal reporter gene assay kit (Roche, France).
Syncytia were obtained after 24 h coculture of hCMEC/D3 cells and MT-2 infected lymphocytes (1∶1 ratio). The p24 staining was performed by an indirect immunofluorescence assay using the mouse anti-HTLV-1 p24 antibody on PFA fixed cells. To ascertain the endothelial origin of the syncytia, the hCMEC/D3 cells were prelabeled with CellTracker™ Red CMTPX (Molecular Probes, Invitrogen, Carlsbad, CA, USA). The preparations were visualized with a Zeiss Axiovert apparatus (Iena, Germany) or Leica DMRB (Wetzlar, Germany). The syncytia formation was inhibited by addition of dextran sulfate (100 µg/mL, Sigma Aldrich, St Louis, MO, USA), VEGF165 (50 ng/mL, R&D systems) or rabbit polyclonal antibody directed to Glut-1 (ab15309, Abcam; anti Cbl-b antibody H454 was used as an irrelevant antibody).
The extent of syncytia formation was assessed after 24 h in coculture by counting their numbers, as well as nuclei per syncytia, using Giemsa staining. All microscopic fields from 16 mm diameter coverslips were evaluated, from three different cultures.
For inhibition experiments, serum from a HAM/TSP patient or control serum (lacking HTLV-1 antibodies detected by Western blot assay) was added at the beginning of the coculture.
Chronically infected cells (MT-2 lymphocytes or C81-66 cells as control as they can not transmit infection) were irradiated at 10 Gy and washed twice with PBS to eliminate free radicals. hCMEC/D3 monolayers were cocultivated overnight at 37°C with the irradiated lymphocytes at a 1∶1 ratio and extensively washed three times with medium lacking serum. Human endothelial cell cultures were then maintained in normal culture medium. Aliquots of culture medium were collected at different time to detect viral proteins. HTLV-1 p19 was detected in culture media using the HTLV p19 antigen ELISA assay (Zeptometrix, Buffalo, NY, USA). In order to prevent the viral infection, 25 µM of AZT, an inhibitor of the reverse transcriptase, were added to the culture medium.
Secondary infection was performed as described previously [55]. Briefly, HCMEC/D3 cells were cocultured with irradiated MT-2 lymphocytes (or control C81-66 cells) for 15 days. The cells were then extensively washed and the medium replaced. Forty-eight hours later, the growth medium was collected and clarified by low-speed centrifugation (2,500 rpm for 5 min) then filtered through a 45-µm filter. The resulting media was then layered on a 20% glycerol gradient and the virus in it was pelleted by centrifugation in a SW28 rotor at 22,000 rpm for 2 h. The pellet was then resuspended in 200 µl of serum-free DMEM. 293T-LTR GFP indicator cells were incubated with 200 µl of the virus suspension in a total volume of 2 ml of serum-free DMEM for 2 h. Complete medium was then added and changed twice a week. One week later, the cells were fixed in 4% PFA, and visualized with a Zeiss Axiovert apparatus. The specificity of the signal was confirmed by addition of AZT into the culture medium.
TJ proteins expression in HCMEC/D3 cells (cocultured for 15 days with MT-2 or C81-66 irradiated lymphocytes) was investigated by immunoblot analysis. Cells were washed twice with PBS, lysed in appropriate buffer (50 mM Tris-HCl,pH 7.4, 120 mM NaCl, 5 mM EDTA, 0.5% Nonidet P-40, 0.2 mM Na3VO4, 1 mM dithiothreitol, 1 mM phenylmethylsulfonyl fluoride) in the presence of a cocktail of protease inhibitors (Roche Applied Science, Indianapolis, IN, USA), and incubated on ice. Protein concentration was determined by the Bradford method (Bio-Rad, Hercules, CA, USA). Samples were loaded into 4–20% Tris/Gly gels (NOVEX, Invitrogen), subjected to SDS-PAGE, and transferred onto a nitrocellulose membrane (Immobilon-P, Millipore, Billerica, MA, USA). Following incubation with specific antibodies (rabbit anti ZO-1 and Occludin from Zymed or mouse anti-HTLV-p24 from AbCam) and extensive washing in PBS-Tween 0.05%, membranes were incubated with horseradish peroxidase-conjugated secondary antibodies (Vector Laboratories) and developed using either the SuperSignal WestPico or SuperSignal West Femto Chemiluminescent substrate kit (Pierce, Rockford, IL, USA). To ensure equal amount of protein loaded per well, membranes were stripped with the Re-blot Plus Kit (Chemicon International, Temecula, CA, USA) and reprobed with a specific anti β-tubulin antibody (Santa Cruz Biotechnology).
Permeability of hCMEC/D3 cell monolayers was measured using a method adapted from Dehouck et al. [56] and Romero et al. [30] on Transwell-ClearTM filters (polyester, 12 mm diameter, pore size 3 µm, Costar, Brumath, France). Briefly, 105 cells/well were seeded on filters previously coated with rat-tail collagen I (BD Biosciences) and bovine plasma fibronectin (Sigma Aldrich). At confluence, hydrocortisone was added to EBM-2 medium as recommended by the manufacturer. After 24 hours, cells were used for experiments. Coculture experiments were set up by adding of 105 lymphocytes to the endothelial monolayer in the presence or absence of inhibitors.
For the permeability test, the culture medium was replaced by DMEM without phenol red. FITC-labelled dextran (molecular weight 70 kDa, Sigma Aldrich) was added to the upper compartment and inserts were transferred sequentially at 5 minutes intervals from well to well for 30 min. The quantity of FITC-dextran that had diffused through the monolayer into the lower compartment at each time point was determined using a fluorescence multiwell plate reader (Wallac VictorTM 1420, PerkinElmer, Wellesley, MA, USA). The permeability coefficients of the endothelial monolayers were then calculated as previously described [30].
Lymphocytes were labeled with CellTracker™ Green BODIPY® (Molecular Probes) according to manufacturer's instructions. Labeled lymphocytes were added to the upper chamber of Transwell-ClearTM insert filters (polyester, 12 mm diameter, pore size 3 µm, Costar) containing confluent hCMEC/D3 monolayers. After 24 hours at 37°C, the monolayer was extensively washed with PBS/EDTA, in order to collect lymphocytes adherent to each side of the membrane. Cells were lyzed using water. Fluorescence intensity was determined using a fluorescence multiwell plate reader.
|
10.1371/journal.pbio.1000238 | Concordant Regulation of Translation and mRNA Abundance for Hundreds of Targets of a Human microRNA | MicroRNAs (miRNAs) regulate gene expression posttranscriptionally by interfering with a target mRNA's translation, stability, or both. We sought to dissect the respective contributions of translational inhibition and mRNA decay to microRNA regulation. We identified direct targets of a specific miRNA, miR-124, by virtue of their association with Argonaute proteins, core components of miRNA effector complexes, in response to miR-124 transfection in human tissue culture cells. In parallel, we assessed mRNA levels and obtained translation profiles using a novel global approach to analyze polysomes separated on sucrose gradients. Analysis of translation profiles for ∼8,000 genes in these proliferative human cells revealed that basic features of translation are similar to those previously observed in rapidly growing Saccharomyces cerevisiae. For ∼600 mRNAs specifically recruited to Argonaute proteins by miR-124, we found reductions in both the mRNA abundance and inferred translation rate spanning a large dynamic range. The changes in mRNA levels of these miR-124 targets were larger than the changes in translation, with average decreases of 35% and 12%, respectively. Further, there was no identifiable subgroup of mRNA targets for which the translational response was dominant. Both ribosome occupancy (the fraction of a given gene's transcripts associated with ribosomes) and ribosome density (the average number of ribosomes bound per unit length of coding sequence) were selectively reduced for hundreds of miR-124 targets by the presence of miR-124. Changes in protein abundance inferred from the observed changes in mRNA abundance and translation profiles closely matched changes directly determined by Western analysis for 11 of 12 proteins, suggesting that our assays captured most of miR-124–mediated regulation. These results suggest that miRNAs inhibit translation initiation or stimulate ribosome drop-off preferentially near the start site and are not consistent with inhibition of polypeptide elongation, or nascent polypeptide degradation contributing significantly to miRNA-mediated regulation in proliferating HEK293T cells. The observation of concordant changes in mRNA abundance and translational rate for hundreds of miR-124 targets is consistent with a functional link between these two regulatory outcomes of miRNA targeting, and the well-documented interrelationship between translation and mRNA decay.
| The human genome contains directions to regulate the timing and magnitude of expression of its thousands of genes. MicroRNAs are important regulatory RNAs that tune the expression levels of tens to hundreds of specific genes by pairing to complimentary stretches in the messenger RNAs from these genes, thereby reducing their stability and their translation into protein. Although the importance of microRNAs is appreciated, little is known about the relative contributions of degradation or repression of translation of the cognate mRNAs to the overall effects on protein synthesis, or the links between these two regulatory mechanisms. We devised a simple, economical method to systematically measure mRNA translation profiles, then applied this method, in combination with gene expression analysis, to measure the effects of the human microRNA miR-124 on the abundance and apparent translation rate of its mRNA targets. We found that for the ∼600 mRNA targets of miR-124 that were identified by their association with microRNA effector complexes, around three quarters of the reduction in estimated protein synthesis was explained by changes in mRNA abundance. Although the apparent changes in translation efficiencies of the targeted mRNAs were smaller in magnitude, they were highly correlated with changes in the abundance of those RNAs, suggesting a functional link between microRNA-mediated repression of translation and mRNA decay.
| MicroRNAs (miRNAs) are small noncoding RNAs whose complementary pairing to target mRNAs potentially regulates expression of more than 60% of genes in many and perhaps all metazoans [1]–[6]. Destabilization of mRNA and translational repression have been suggested as the mechanisms of action for miRNAs [1],[3],[7]–[15], and recent work directly measuring endogenous protein levels in response to altered miRNA expression levels found that specific miRNAs modestly inhibit the production of hundreds of proteins [16],[17].
The importance and functional range of miRNAs are evidenced by the diverse and often dramatic phenotypic consequences when miRNAs are mutated or misexpressed, leading to aberrant development or disease [7],[18]–[24]. Although regulation by miRNAs is an integral component of the global gene expression program, there is currently no consensus on either the mechanism by which they decrease the levels of the targeted proteins or even the steps in gene expression regulated by miRNAs [3],[25]–[29].
The proposal that miRNAs decrease protein levels without affecting mRNA stability arose from the observation that the miRNA lin-4 down-regulates lin-14 expression in the absence of noticeable changes in lin-14 mRNA abundance in Caenorhabditis elegans [7],[30]–[33]. Subsequent studies in mammalian cell culture provided further support for this model [34]–[37]. Several studies have found that repressed mRNAs as well as protein components of the miRNA regulatory system accumulate in P-bodies, suggesting that repressed mRNAs may be sequestered away from the translation pool [38]–[44]. Other evidence points to deadenylation of miRNA-targeted mRNAs, an effect that can inhibit translation [14],[45]–[53]. Some studies have argued that initiation of translation is blocked at either an early, cap-dependent stage or later during AUG recognition or 60S joining [10],[44],[52],[54]–[58]. Others have argued that a postinitiation step is targeted, resulting in either slowed elongation, ribosome drop-off, or nascent polypeptide degradation [7],[59]–[62].
One factor contributing to the lack of a consensus model for miRNA function is the evidence that miRNA targeting of an mRNA significantly reduces message levels (despite previous reports to the contrary) [9],[11],[12],[14],[52],[63],[64]. Indeed, very recent studies from Baek et al. and Selbach et al. found that the changes in mRNA abundance are not only correlated with the repression of many targets, but also can account for most of the observed reduction in protein expression [16],[17]. mRNA targets of the same miRNA can either be translationally repressed with little change in mRNA abundance, translationally repressed and have concordant changes in mRNA abundance, or have little translation repression with large changes in mRNA abundance [52],[65],[66]. That miRNAs can affect both protein production and abundance of their mRNA targets raises the question of to what extent these outcomes of miRNA regulation are mediated by a common mechanism or by competing or complementary processes. The regulatory consequence of a particular miRNA–mRNA interaction might be influenced by miRNA-independent factors such as cellular context or by additional information encoded by the target mRNA, e.g., presence of binding sites for other RNA-binding proteins and miRNAs, secondary structure around miRNA binding sites, or the intrinsic decay rate of the mRNA [25],[51],[67],[68].
Experiment-specific effects of in vitro translation assays, reporter constructs, or procedural differences that alter properties of gene expression could account for some of the wide variation in the apparent mechanisms by which miRNAs alter expression [25],[27]. To date, most studies on translational regulation by miRNAs have used reporter assays. Although assays that rely on engineered reporter transcripts are powerful, assay-specific anomalies are a concern; artificial mRNAs may lack key pieces of regulatory information, overexpression of reporter mRNAs could mask subtle regulatory functions, and DNA transfection can lead to indirect effects on cell physiology [26]. Indeed, recent reports have found that differences in experimental setup, such as the method of transfection, type of 5′-cap, or the promoter sequence of the DNA reporter construct can drastically alter the degree or even the apparent mode of regulation by miRNAs [59],[69]. In addition, some models have been based on studies in which only one or a few targets were studied, which introduces the possibility of generalizing the behavior of a single miRNA–mRNA interaction that may not represent the dominant biological mechanism.
Two recent studies avoided many of these caveats by overexpressing, inhibiting, or deleting specific miRNAs and systematically measuring changes in endogenous mRNA and protein levels using DNA microarrays and stable isotope labeling with amino acids in cell culture (SILAC), respectively [16],[17]. Both studies found mostly concordant changes in mRNA levels and protein levels, with changes in mRNA levels accounting for much, but not all, of the changes in protein abundance. With data for hundreds of endogenous targets, these studies were the first to provide genome-wide evidence that mRNA degradation accounts for much of the reduction in protein levels. And whereas these results suggest that translation inhibition accounts for some of the observed changes in protein abundance of miRNA targets, they do not provide direct evidence of this, nor do they provide insight into which steps in translation are regulated, the extent this regulation contributes to reduced gene expression of specific mRNAs, or its possible links to mRNA decay.
To investigate how miRNAs regulate gene expression, we systematically identified direct targets of the miRNA miR-124 by measuring the recruitment of target mRNAs to Argonaute (Ago) proteins, the core components of the miRNA effector complex, as previously described [70]–[72]. We then measured, in parallel, mRNA abundance and two indicators of translation rate, ribosome occupancy and ribosome density, for more than 8,000 genes, using DNA microarrays and a novel polysome encoding scheme. This strategy allowed us to directly investigate the behavior of miRNA–mRNA target pairs with respect to both mRNA fate and translation, on a genomic scale.
To study the effects of miR-124 on expression of mRNA targets, we first had to identify those targets. Recruitment to Ago complexes in response to the expression of a particular miRNA appears to be the most reliable criterion for target identification [70]. To this end, we lysed human embryonic kidney (HEK) 293T cells transfected with miR-124 and isolated Ago-associated RNA by immunopurification (IP) using a monoclonal antibody that recognizes all four human Ago paralogs [73]. We measured mRNA enrichment in Ago IPs by comparative DNA microarray hybridization of samples prepared from immunupurified RNA and total RNA from cell extracts. Three replicates of Ago and control IPs were performed from both miR-124 and mock-transfected cells (Datasets S1 and S5).
To examine the enrichment profiles of the IPs, we first clustered the microarray results by their similarity and visualized the results as a heatmap, with the degree of enrichment of each RNA shown on a green (least enriched) to red (most enriched) scale (Figure S1). The Ago IP enrichment profiles were reproducible as evidenced by an average Pearson correlation coefficient between mRNA enrichment profiles of Ago IPs in mock-transfected cells and miR-124–transfected cells of 0.90 and 0.94, respectively.
Thousands of mRNAs were reproducibly enriched in the Ago IPs from mock-transfected cells (Figures S1 and S2, and Text S1). We found that the presence of sequence matches to two highly expressed microRNA families, miR-17-5p/20/92/106/591.d and miR-19a/b, in the 3′-untranslated regions (UTRs) of mRNAs significantly correlated with Ago IP enrichment (Text S2), suggesting that association with Ago is in large part a reflection of the relative occupancy of each mRNA with the suite of miRNAs endogenously expressed in HEK293T cells. High-confidence Ago-associated mRNAs (at least 4-fold enriched over the mean, 1,363 mRNAs) disproportionately encode regulatory proteins (409, p = 0.001), with roles including “transcription factor activity” (95, p = 0.01), “signal transduction” (230, p = 0.02), and “gene silencing by RNA” (7, p = 0.02).
To identify RNAs specifically recruited to Agos by miR-124, we compared the mRNA enrichment profiles of Ago IPs from miR-124–transfected cells to Ago IPs from mock-transfected cells using the significance analysis of microarrays (SAM) modified two-sample unpaired t-test (Datasets S1 and S5). At a stringent 1% local false-discovery rate (FDR) threshold, we identified 623 distinct mRNAs significantly enriched in Ago IPs from lysates of miR-124–transfected cells compared to Ago IPs from mock-transfected cells (Figure 1A).
Previous work established that the 5′-end of the miRNA, the “seed region,” is particularly important for interactions with mRNA targets [4],[11],[37],[74]–[77]. In most cases, there is a 6–8 bp stretch of perfect complementarity between the seed region of the miRNA and a “seed match” sequence in the 3′-UTR of the mRNA [4],[11],[37],[74]–[77]. We reasoned that if the mRNAs specifically recruited to Agos by miR-124 transfection were physically associated with miR-124, seed match sequences would be significantly enriched in miR-124–specific IP targets compared to nontargets. Indeed, we found strong enrichment of 6–8 base seed matches to miR-124 in the 3′-UTRs of miR-124 Ago IP targets (Figure 1B). We also found enrichment within the coding sequences of miR-124 Ago IP targets, as previously reported (Figure 1B) [11],[16],[17],[70],[71],[78],[79]. For instance, 60% of miR-124 Ago IP targets contain a perfect match to positions 2–8 of miR-124 (called 7mer-m8) in their 3′-UTRs, compared to 10% of nontargets (p<10−185, hypergeometric distribution), and 23% of miR-124 Ago IP targets contain a perfect match to positions 2–8 of miR-124 in their coding sequence, compared to 10% of nontargets (p<10−21). After removing mRNAs with 7mer seed matches in their 3′-UTRs, the remaining miR-124 IP targets were still significantly, albeit weakly, enriched for 3′-UTR 6mer matches to miR-124 (6mer 2–7, p = 0.008, 6mer 3–8, p<10−5). These data argue that most miR-124 Ago IP targets were recruited to Agos by direct association with miR-124, via seed matches in their 3′-UTRs or coding sequences.
The standard approach to assess translation in vivo has been the analysis of “polysome profiles.” After treatment with cycloheximide to trap elongating ribosomes, mRNAs with no associated ribosomes and those with varying numbers of ribosomes bound can be separated by velocity sedimentation through a sucrose gradient. The polysome profile of a gene's mRNA provides information on two key parameters in translation: (1) the fraction of the mRNA species bound by at least one ribosome, and presumably undergoing translation, referred to as “ribosome occupancy,” and (2) the average number of ribosomes bound per 100 bases of coding sequence to mRNAs that have at least one bound ribosome, referred to as the “ribosome density.”
We previously developed a method to systematically measure ribosome occupancy and ribosome density by measuring the relative amount of each gene's mRNA in each fraction of a polysome profile using DNA microarray hybridization [80].
We have since developed and implemented a more streamlined approach that uses one DNA microarray hybridization to measure ribosome occupancy and only a single additional microarray hybridization to measure ribosome density (Figure 2 and Figure S3). We measured ribosome occupancy by first pooling ribosome bound fractions and unbound fractions and adding exogenous doping control RNAs to each (Figure 2A). Poly(A) RNA from bound and unbound pools was isolated, amplified, coupled to Cy5 and Cy3 dyes, respectively, and comparatively hybridized to DNA microarrays. The ribosome occupancy for each gene's mRNA was obtained after scaling the microarray data using the doping controls (see Materials and Methods for details).
We determined the ribosome density for each gene's mRNA by a “gradient encoding” strategy in which a graded ratio of each fraction from the ribosome bound fractions was split into a “heavy” and a “light” pool, respectively. For instance, 99% of the first fraction (∼one ribosome bound) was added to the light pool and 1% was added to the heavy pool. Then, 98% of the second fraction (1.5–2 ribosomes bound) was added to the light pool and 2% was added to the heavy pool, and so on, such that the light pool was enriched for mRNAs associated with fewer ribosomes, and the heavy pool was enriched for mRNAs associated with a greater number of ribosomes (Figure 2B). The RNA in each pool was amplified, labeled with Cy5 or Cy3, mixed, and comparatively hybridized to DNA microarrays. Thus, the Cy5/Cy3 ratio measured at each element on the array is a monotonic function of the mass-weighted average sedimentation coefficient of the corresponding mRNA, which is primarily determined by the number of ribosomes bound to it. The validity of this approach is supported by the very strong concordance between ribosome density measurements in yeast obtained with the gradient-encoding method and our previously published ribosome density measurements obtained using the traditional approach of analyzing each fraction on separate DNA microarrays (Pearson r = 0.95) (unpublished data) [80]. Further details of the methodology, as well as control experiments and additional analyses, will be described elsewhere.
To measure the effects of miR-124 on translation, we performed translation profiling on cell extracts generated from the same miR-124–transfected, or mock-transfected cell cultures that were used for Ago IPs and mRNA expression profiling (see below). We obtained high-quality ribosome occupancy and ribosome density measurements on 16,140 sequences (representing 10,455 genes) from three independent mock-transfected cultures and two miR-124–transfected cultures (Datasets S2, S3, and S5). There was a strong concordance between replicate experiments for both the ribosome occupancy and ribosome number/density measurements, both in terms of the correlation of the gene-specific measurements (Pearson correlation for ribosome occupancy = 0.85–0.89, ribosome number = 0.91–0.97) and the means (mean ribosome occupancy = 0.83–0.87, mean ribosome number = 5.6–6.1 per mRNA), which were derived independently for each experiment based on the exogenous doping controls.
The measurements from mock-transfected cells provide some general insights into the translation regulatory program in proliferating human cells. Here, we focus on 8,385 genes that correspond to a RefSeq mRNA for which we obtained high-quality measurements in both Ago IP and mRNA expression DNA microarray experiments. The average ribosome occupancy for the mRNAs from these 8,385 genes was 85% (25th and 75th quartiles = 0.81 and 0.94, respectively) (Figure 3A) suggesting, that for most genes, most polyadenylated mRNAs are associated with ribosomes under these growth conditions and that there are not abundant pools of polyadenylated mRNAs in an untranslated “compartment.” For more than 97% of the genes analyzed, a majority of the transcripts were associated with ribosomes; mRNA transcripts of 3% of these genes (224) were predominantly unassociated with ribosomes (ribosome occupancy <50%). The reason for the relative exclusion of this small set of mRNAs from the highly translated pool remains to be determined: possibilities include sequestration from the translation machinery or a relatively short half-life that results in these mRNAs spending a correspondingly small fraction of their lives in the translated pool. We searched for common biological themes among these non–ribosome-associated mRNAs using gene ontology (GO) term analysis, and found that an unexpectedly large fraction of these mRNAs encode proteins involved in “regulation of transcription” (64, p<10−7). On the flip side, there were 342 genes whose mRNAs were almost completely (98% or greater) associated with ribosomes. Many of these mRNAs encoded proteins involved in metabolism and gene expression, including “oxidative phosphorylation” (21, p<10−10), “nuclear mRNA splicing” (23, p<10−5), “proteasome complex” (11, p = 0.0003), and “glycolysis” (10, p = 0.0002). mRNAs with low ribosome occupancy (less than 50%) were significantly less abundant than mRNAs with high ribosome occupancy (greater than 98%) (Kolmogorov-Smirnov test, p<10−15), consistent with the hypothesis that a lower rate of decay, and hence a greater fraction of the lifespan spent in the translated pool, contributes to ribosome occupancy.
The average ribosome density for the 8,385 genes with technically high-quality data across this set of experiments was 0.53 ribosomes per 100 nucleotides (nts) (25th and 75th quartiles = 0.27 and 0.67, respectively), which corresponds to one ribosome per 189 nts (Figure 3B). Given that ribosomes are believed to span ∼30 nts of the mRNA, the average ribosome density would be approximately one sixth of the maximal packing density [81]. This spacing suggests that translation initiation is rate limiting for most mRNAs.
We previously observed a strong negative correlation between an mRNA's ribosome density and its coding sequence length in yeast cells rapidly growing in rich medium [80]. Subsequent experiments suggested that this relationship is due to either a strong inverse correlation between initiation rate and coding sequence length [82], or a decrease in ribosome density as a function of position along the mRNA [83]. We found the same inverse relationship between the size of a coding sequence and ribosome density in proliferating mammalian cells (Spearman r = −0.90) (Figure 3C). Sucrose gradient sedimentation did not clearly resolve polysomes containing more than seven ribosomes, so it is possible that our method underestimates the number of ribosome bound to mRNAs with long coding sequences, which could, in principle, lead to a spurious negative correlation between coding sequence length and ribosome density. However, the inverse relationship between coding sequence length and ribosome density is still readily evident when only mRNAs with coding sequences less than 1,000 nts are considered (r = −0.73), strongly supporting the validity of this relationship.
These broad similarities between translational programs in proliferating HEK293T cells and proliferating S. cerevisiae grown in rich medium, suggest that the overall organization of the program, and perhaps some of the fundamental mechanisms underlying the regulation of translation, may be similar in rapidly growing yeast and human cells [80].
To measure the effects of miR-124 on mRNA expression levels, we profiled mRNA expression in the same cell cultures that we used for the Ago IPs and translation profiling. We obtained high-quality measurements for 15,301 genes from three independent mock-transfected cultures and three independent miR-124–transfected cultures (Datasets S4 and S5). There was strong concordance between replicate experiments (Pearson r = 0.95–0.97).
To study the effects of miR-124 on the expression of its mRNA targets, we first compared the changes in mRNA abundance of Ago IP targets of miR-124 (560 mRNAs; 1% local FDR) and nontargets (7,825 mRNAs) between cells transfected with miR-124 and cells that were mock transfected. Samples were taken 12 h after the respective treatments. We plotted the cumulative distributions of miR-124–dependent Ago IP targets (Figure 4A, green curve) and nontarget mRNAs (Figure 4A, black curve) as a function of the differences in their mRNA abundance between miR-124 and mock-transfected cells. miR-124 target mRNAs were much more likely to decrease in abundance after miR-124 transfection than nontargets (p<10−173, one-sided Kolmogorov-Smirnov test). For example, 74% of miR-124 IP targets decreased at least 15% at the mRNA level, compared to 13% of nontargets. The average abundance of miR-124 Ago IP targets decreased by 35% compared to nontargets (Figure 4C, green bar on the left). The results are consistent with miRNAs having significant, but modest effects on the mRNA levels of most of their endogenous mRNA targets.
Previous work has established that perfect seed matches to the miRNA in 3′-UTRs are important to elicit effects on mRNA abundance [4],[11],[37],[74]–[77]. To test the importance of 3′-UTR seed matches on the expression of miR-124 targets, we plotted the cumulative distributions of miR-124 IP targets with at least one 7mer 3′-UTR seed match (379, Figure 4A, red curve) and miR-124 IP targets that lacked a 7mer 3′-UTR seed match (181, Figure 4A, blue curve). We found that mRNA targets with 7mer 3′-UTR seed matches were more likely than targets that lacked a 7mer 3′-UTR seed match to decrease in abundance in the presence of miR-124 (90% of miR-124 IP targets with a 3′-UTR seed match decreased at least 15%, compared to 49% of targets that lacked a 7mer 3′-UTR seed match). On average, IP target mRNAs with 7mer 3′-UTR seed matches decreased 40%, whereas IP targets that did not have a 7mer seed match in their 3′-UTR decreased 17%, compared to nontargets (Figure 4C, left). These results underscore the importance of 3′-UTR seed matches for regulation at the mRNA level, but also demonstrate that a large fraction of miR-124 IP targets that lack 7mer seed matches to miR-124 in their 3′-UTR are nevertheless regulated at the mRNA level by miR-124.
To study the effects of miR-124 on translation of targeted mRNAs, we estimated the change in the translation rates between miR-124-transfected and mock-transfected cells (Tr) for each mRNA as:(1)where multiplying O, the fraction of the mRNA that is ribosome-bound (ribosome occupancy), by D, the average number of ribosomes per 100 nts for bound mRNAs (ribosome density) provides the weighted ribosome density for each mRNA; Er is an unmeasured value for the elongation rate of any given mRNA and was assumed not to change (discussed further below). Values Tr obtained from miR-124 transfected cells were divided by those from mock-transfected cells to estimate the change. We plotted the cumulative distribution of Tr for miR-124 Ago IP targets and nontargets (Figure 4B). miR-124 targets (Figure 4B, green curve) were much more likely to decrease in translation rate than nontarget mRNAs (Figure 4B, black curve) (p<10−62, one-sided Kolmogorov-Smirnov test). The apparent translation rate of 47% of miR-124 Ago IP targets, but only 10% of nontargets, decreased by at least 10%. In line with what we observed for changes in mRNA abundance, miR-124 IP targets with at least one 7mer seed match in their 3′-UTR were more likely to decrease in translation rate than miR-124 IP targets that lacked a 7mer 3′-UTR seed match (56% percent of miR-124 IP targets with a 7mer 3′-UTR seed match decreased at least 10% in translation versus 27% of IP targets that lacked a 7mer 3′-UTR seed match). The overall effects on translation, while highly significant, were very modest; on average, the estimated translation rates of miR-124 Ago IP targets decreased by 12% relative to nontargets (15% for miR-124 IP targets with a 7mer 3′-UTR seed match and 5% for miR-124 IP targets without a 7mer 3′-UTR seed match) (Figure 4C, right). These results show that miR-124 has modest effects on the abundance, translation rate, or both for most its targets.
In some cases, mRNAs that are translationally repressed are deadenylated and stored, rather than degraded [84]–[86]. All of our measurements were of mRNAs amplified based on their poly(A) tails. Therefore, it was possible that the effects on translation were underestimated and the effects on abundance were overestimated because a large percentage of targets mRNAs were translationally repressed, and stored without a poly(A) tail. To test this possibility, we measured the differences in total RNA levels irrespective of poly(A) tail for each gene between miR-124–transfected and mock-transfected cells. We found that the differences in RNA abundance between miR-124–transfected and mock-transfected cells as measured with unamplified total RNA were similar to those measured for amplified poly(A)-selected mRNA for miR-124 targets (Pearson r = 0.82, slope of least-squares regression fit in linear space = 0.82) (Figure S4). These data suggest that the apparent decrease in abundance of miR-124 target mRNAs results primarily from degradation rather than deadenylation alone.
Many steps in protein synthesis have been proposed to be regulated by miRNAs. The proposed mechanisms include: (i) blocking initiation, e.g., by preventing eiF4F binding to mRNA caps or joining of the 40S and 60S ribosomal subunits; (ii) promoting poly(A) tail deadenylation, which can slow initiation by preventing interactions between the poly(A) tail and 5′-cap, and by increasing the rate of mRNA decay, which reduces the fraction of the mRNA's lifespan spent in the translated pool; (iii) promoting premature ribosome release during elongation; (iv) slowing translation elongation; (v) promoting cotranslational proteolysis; and (vi) concerted slowing of initiation and elongation [7],[10],[30],[34]–[37],[45],[51],[52],[54]–[62]. The first four proposed mechanisms make specific predictions about the effects of miRNAs on the ribosome occupancy and ribosome density of targets. Proposed mechanisms (i), (ii), and (iii) predict that both occupancy and density will decrease; mechanism (iv) predicts that ribosome density will increase as a function of the extent to which elongation is slowed. In contrast, proposed mechanism (v) does not predict that ribosome occupancy or ribosome density will change, and the effects on ribosome occupancy and ribosome density in mechanism (vi) depend on the relative effects of the miRNA on the two steps.
We tested these predictions by comparing ribosome occupancy and density profiles of mRNAs from miR-124 and from mock-transfected cells. We found that miR-124 Ago IP targets were much more likely than nontarget mRNAs to exhibit both reduced ribosome occupancy (Figure 5A) (p<10−31, one-sided Kolmogorov-Smirnov test) and reduced ribosome density (Figure 5B) (p<10−51, one-sided Kolmogorov-Smirnov test) following miR-124 transfection. Thirty-nine percent of miR-124 Ago IP targets decreased at least 5% in ribosome occupancy, compared to 13% of nontargets; 55% of miR-124 Ago IP targets decreased at least 5% in ribosome density, compared to 18% of nontargets. On average, the ribosome occupancy of miR-124 Ago IP targets decreased by 4%, and their ribosome density decreased by 8% (Figure 5C, green bars). We hypothesized that mRNAs with fewer associated ribosomes might exhibit larger changes in ribosome occupancy as a result of the increased likelihood of losing all ribosomes. In support of this hypothesis, on average, all ten miR-124 target mRNAs with ribosome occupancy changes greater than 20% had significantly shorter coding sequences and fewer bound ribosomes than mRNAs that changed less than 20% (p = 0.0003, one-sided Mann-Whitney test) (Figure S5A).
The effects on ribosome occupancy and ribosome density were significantly larger for miR-124 Ago IP targets that contain at least one 3′-UTR 7mer seed match (45% and 65% decreased at least 5% in ribosome occupancy and ribosome density, respectively), compared to miR-124 Ago IP targets that lack a 3′-UTR 7mer seed match (26% and 34% decreased at least 5% in ribosome occupancy and ribosome density, respectively), providing direct evidence for the general importance of 3′-UTR seed matches for miRNA-mediated translational repression of endogenous mRNAs [16],[17].
The observed effects on ribosome occupancy and density could, in principle, be the result of multiple independent regulatory mechanisms. For instance, the decrease in ribosome occupancy and density could be a result of mechanisms (i), (ii), and (iii). If however, the effects on ribosome occupancy and ribosome density were due to the same regulatory mechanism, we would expect a large overlap between mRNAs that show appreciable decreases in ribosome occupancy and ribosome density in the presence of miR-124. Indeed, 77% of miR-124 IP targets that decreased at least 5% in ribosome occupancy also decreased at least 5% in ribosome density (30% of all miR-124 IP targets decreased at least 5% in both ribosome occupancy and ribosome density compared to 2% of nontargets), which is significantly more than expected by chance (p<10−18, hypergeometric distribution). There was also a modest, but highly significant, correlation between changes in ribosome occupancy and ribosome density of miR-124 Ago IP targets (Spearman r = 0.45, p<10−25) (Figure S6A), although many mRNAs appeared to differentially change in either ribosome occupancy or ribosome density (some miR-124 mRNA targets even appeared to increase appreciably in ribosome occupancy; Figure S6 and Text S3). These results are consistent with the effects on ribosome occupancy and ribosome density arising from the same regulatory mechanism.
If miR-124 induced ribosome drop-off (mechanism (iii)) stochastically along the coding sequence, the change in ribosome density would be exponentially related to the length of the coding sequence. To test this, we plotted the change in ribosome density as a function of mRNA length for miR-124 IP targets and found that although they are correlated (Spearman r = 0.30), it is highly unlikely there is a first-order exponential relationship between the change in density and the length of the mRNA's coding sequence (p<10−211, F-test with the null hypothesis that the observed change in density fits the predicted change in density from an exponential least-squares fit) (Figure S5B). Thus, if ribosome drop-off is the predominant mode of miR-124 regulation, it occurs preferentially near the translation start site.
The observation that many miR-124 targets decreased in both ribosome occupancy and ribosome density after transfection with miR-124 is consistent with regulation of translation initiation (mechanisms (i) or (ii)) or ribosome drop-off preferentially near the translation start site (mechanism (iii)) by miR-124 and suggests that slowed elongation (model (iv)) is not the predominant mode of regulation of translation by miR-124 under these conditions. Without measurements of the actual effects on protein synthesis, these results, however, do not rule out the possibility that miR-124 also induces cotranslational proteolysis (v) or coordinately represses translation initiation and translation elongation (vi), resulting in modest decreases in ribosome occupancy and ribosome density, but large effects on protein synthesis.
To analyze the overall effect of the observed decreases in mRNA abundance and translation on protein production, we calculated the estimated change in protein synthesis as:(2)where the estimated change in protein synthesis (Pc) can then be derived by multiplying the change in mRNA abundance by the estimated change in translation rate (Tr). The change in relative mRNA abundance is calculated by dividing relative mRNA abundance values from miR-124 transfection experiments by values from the mock condition . Although the overall effect on predicted protein production was on average quite modest (∼2-fold decrease compared to nontargets), for a small fraction of miR-124 targets, the predicted changes in protein production were fairly large; 45 of the 560 identified miR-124 targets were predicted to have a decrease of at least 4-fold in protein production 12 h after miR-124 transfection. A disproportionate fraction of the most significantly affected mRNAs encoded proteins associated with membrane compartments (28, p = 0.001), including endoplasmic reticulum (seven) and plasma membrane (nine); these mRNAs are likely to be translated on the rough endoplasmic reticulum. A similar observation was reported with different miRNAs in a recent study [17]. These results suggest that mRNAs that are translated on the rough endoplasmic reticulum might be particularly susceptible to miRNA-mediated regulation, possibly while stalled prior to engagement with the endoplasmic reticulum [87].
To test whether our estimated changes in protein synthesis predict actual changes in protein abundance, we measured changes in protein abundance of a diverse set of proteins encoded by mRNAs that are highly enriched in miR-124 Ago IPs by Western blot analysis based on the availability of reliable antibodies. We chose 14 proteins encoded by mRNAs that are highly enriched in miR-124 Ago IPs, with predicted decreases in protein synthesis ranging from no change to 3-fold (Table S1). We collected cell lysates 60 h (four to five cell divisions) after miR-124 or mock-transfection to reduce the likelihood of underestimating the change in protein synthesis for long-lived proteins. Twelve of the 14 antibodies detected bands at the predicted molecular weight (Figure 6A). We observed a significant correlation between the estimated changes in protein synthesis (Figure 6B, x-axis) and the measured changes in protein levels (Figure 6B, y-axis) in response to miR-124 transfection (Spearman r = 0.54, p = 0.07, slope of least-squares regression fit = 0.54, grey line in Figure 6B), with one exception. Only RNF128, with a predicted 3.7-fold reduction in protein synthesis, drastically disagreed with our measured decrease of 1.2-fold reduction. It is possible that the discordance in RNF128 protein levels is due to posttranslational autoregulation, which is common among ring finger proteins [88]–[90]. After excluding RNF128 from analyses, there is a strong concordance between the two measurements (Spearman r = 0.90, p = 0.0001, slope = 0.95; red line, Figure 6B) for the remaining 11 proteins. The high correlation and the fact that the slope of the best-fit line excluding RNF128 is close to one, suggests that miR-124–induced changes in transcript abundance and translation rate can almost completely account for the changes in abundance of the targeted proteins. Thus, cotranslation proteolysis (proposal (v)) and coordinate repression of initiation and elongation (proposal (vi)) are unlikely to play more than a minor role in miR-124 regulation under these conditions.
Multiple distinct miRNA regulatory pathways have been proposed, such that translational repression and mRNA degradation can be regulated independently, and these two regulatory consequences are differentially affected by specific features of miRNA–mRNA interaction [34],[37],[67],[71],[91]. The relative magnitude of effects on translation and decay of targeted mRNAs might be influenced by the sequence context of the miRNA–mRNA interaction and the particular suite of RNA-binding proteins associated with the mRNA [38],[43],[65],[92]. If the balance between effects on translation and effects on decay were influenced in a gene-specific way by features of the mRNA, we would expect that some targets of miR-124 would have relatively large changes in translation with little change in mRNA abundance or vice versa. If, however, miRNA–mRNA interactions act through a single dominant regulatory pathway that affects both translation and decay, we would expect a strong correlation between the changes in abundance and translation of mRNA targets of miR-124.
We compared the changes in mRNA abundance (Figure 7, x-axis) to apparent changes in translation rate (Figure 7, y-axis) for miR-124 Ago IP targets following miR-124 transfection. There was strong correlation between these two regulatory effects (Pearson r = 0.60, see Text S4 and Figure S7 for estimates of significance of the correlation), and we found no subpopulation of mRNAs whose translation was appreciably diminished without corresponding changes in mRNA abundance, and few mRNAs whose abundance changed significantly without a corresponding change in translation. To test whether the apparent correlation might be driven solely by mRNAs with the largest measured changes in abundance and translation, we calculated the average changes in mRNA abundance and translation in moving windows of ten mRNAs ranked by their change in mRNA abundance. As shown in Figure 7 (red curve), we found a persistent, nearly monotonic, relationship between changes in mRNA abundance and translation that closely matches the least-squares fit of all the data (Pearson r = 0.91). We obtained similar results when we analyzed miR-124 Ago IP targets with 7mer 3′-UTR seed matches and those that lacked a 7mer 3′-UTR seed match (Figure S8), although the correlation was stronger for targets with 7mer 3′-UTR seed matches (r = 0.60 versus 0.42).
The correlation between changes in mRNA abundance and estimated translation rate, and the absence of a subgroup of mRNAs regulated at the translational level without corresponding effects on abundance, is consistent with a model in which these two regulatory programs are functionally linked. Although there was a measurable decrease in mRNA abundance for almost all miR-124 targets that significantly decreased in translation, only about half of the targets that decreased in mRNA abundance registered a measurable reduction in translation. It is possible that some mRNA targets are degraded without any appreciable effect on translation (e.g., the mRNAs are degraded while still associated with ribosomes) or that translation of these mRNAs is indirectly stimulated in response to miR-124, resulting in no apparent effect on translation at the time we performed translation assays. Alternatively, as the changes in translation tended to be smaller than the changes in mRNA abundance, we may have been unable to accurately measure the small effects on translation of many targets.
Although most functional microRNA seed matches are located in 3′-UTRs as judged by mRNA expression data, phylogenetic conservation analysis, Ago IPs, and reporter studies, some sites in coding sequences and 5′-UTRs can also confer regulation by miRNAs [4],[16],[17],[59],[70],[71],[78],[79],[93]–[95]. The 560 high-confidence miR-124 Ago IP targets for which we obtained high-quality measurements in expression and translation analyses were strongly enriched for mRNAs that contained miR-124 seed matches in 3′-UTRs and coding sequences (Figure 1B), but they were also significantly, albeit weakly, enriched, for seed matches in 5′-UTRs (16, p = 0.009).
We compared the effectiveness of 7mer seed matches in the 3′-UTR, coding sequence, and 5′-UTR, and 6mer seed matches in the 3′-UTR in effecting changes in mRNA abundance and estimated translation rate. We found that both the abundance and translation rate of IP targets, regardless of the location of seed matches, decreased relative to nontarget mRNAs in miR-124 transfected cells compared to mock-transfected cells (Figure S9). The estimated effects on protein production were greatest for mRNAs with 7mer seed matches in the 3′-UTR, consistent with previous studies reporting that 3′-UTR seed matches confer the highest degree of regulation [11],[70],[71],[93]. Changes in mRNA abundance were significantly greater than changes in translation for miR-124 Ago IP targets with 3′-UTR and coding sequence seed matches (Figure S9). IP targets that did not contain any 6mer seed matches were also significantly more likely to decrease in mRNA abundance than nontargets (Figure S9), which suggests that many of these mRNAs are specifically recruited to Agos by miR-124 and regulated by miR-124, even though they do not appear to have canonical recognition elements.
The extent to which each of the thousands of genes expressed in a given mammalian cell is regulated by the suite of (often hundreds of) miRNAs expressed in that cell is not known. We reasoned that our Ago immunopurification strategy, by quantitatively measuring association of mRNAs with microRNA effector complexes, could serve as a direct readout of the potency of the regulatory effects of miRNAs on each mRNA. We compared the change in Ago IP enrichment following transfection with miR-124 to the estimated changes in protein production (Equation 2) for mRNAs with seed matches to miR-124 in their 3′-UTR or coding sequence (Figure S10). For mRNAs with 7mer or 8mer seed matches to miR-124 in their 3′-UTR, there was a strong negative correlation between the magnitude of their enrichment by the Ago IP and the estimated changes in production of the protein they encode (3′-UTR 7mer: Pearson r = −0.72, p<10−192; 3′-UTR 8mer: r = −0.72, p<10−26) (Figure S10A). For mRNAs with 7mer or 8mer seed matches to miR-124 in their coding sequences, but no 7mer seed matches in their 3′-UTRs, there was also a significant, albeit weaker, correlation (coding sequence 7mer: r = −0.39, p<10−33; coding sequence 8mer: r = −0.38, p<10−4) (Figure S10B). There was also a weak, but still significant, correlation between IP enrichment and the estimated change in protein production for mRNAs that lacked 7mer seed matches in their 3′-UTR or coding sequence or that lacked even 6mer seed matches in their 3′-UTR or coding sequence, respectively (3′-UTR 6mer: r = −0.40, p<10−75; no 3′-UTR or CDS 6mer: r = −0.23, p<10−24) (Figure S10C). Most of the mRNAs with 7mer or 8mer seed matches in their 3′-UTR or coding sequence that decreased significantly in protein production were enriched in the Ago IPs. Thus, Ago IP enrichment following transfection with a specific miRNA appears to be a good predictor of the corresponding effects on protein production. Because changes in mRNA abundance and translation following transfection of a specific miRNA are quantitatively smaller and less specific than their change in association with Agos, the IP method appears to be a more sensitive assay to identify the direct regulatory targets of specific miRNAs.
miRNAs regulate the posttranscriptional fates of most mammalian mRNAs, yet for endogenous mRNAs, the effects of miRNAs on translation, the steps in translation that are regulated by miRNAs, and the relationship between regulation of translation and mRNA decay by miRNAs have not been systematically explored. To address these effects and relationships, we determined the effect of a human miRNA, miR-124, on translation and abundance of hundreds of endogenous mRNAs that were recruited to Argonaute proteins in response to ectopic expression of miR-124 in HEK293T cells.
We developed a simple and economical method to quantitatively measure two key parameters of translation, ribosome occupancy and average ribosome density, on a genome-wide scale with single DNA microarray hybridizations for each (Figure 2). This method allowed us to address the effects of miR-124 on translation of endogenous mRNAs; it is also more broadly applicable to the study of translational regulation. In this initial application, we found many parallels between the translation programs in proliferating human embryonic kidney cells and S. cerevisiae (Figure 3), suggesting common features of translational programs in eukaryotes [80].
Direct identification of the mRNAs specifically recruited by miR-124 to Ago proteins, core components of miRNA-effector complexes, defined functional targets of this miRNA in this model system, providing a starting point for dissecting miRNA regulation [70]–[72],[96],[97]. mRNA expression profiling then allowed us to recognize the specific effects of miR-124 on the abundance of these targets.
Three major conclusions emerged from our studies: (i) miR-124 reduces translation and abundance of its mRNA targets over a broad range; changes in mRNA abundance accounted for ∼75% of the estimated effect on protein production; (ii) miR-124 predominantly targets translation at the initiation stage or stimulates ribosome drop-off preferentially near the translation start site; and (iii) miR-124–mediated regulation of translation and mRNA decay are correlated, indicating that most mRNAs are not differentially targeted for translational repression versus mRNA decay.
Transfection of miR-124 consistently reduced the translation and abundance of most of its several hundred high-confidence targets; the resulting decrease in translation averaged 12% and the decrease in target mRNA abundance averaged 35% (Figure 4). The observation that there were several mRNAs (CD164, VAMP3, and DNAJC1) that had about 10-fold reductions in mRNA levels (Figure S7), and the fact that 90% of control-transfected cells expressed the transfected GFP marker, suggests that more than 90% cells were transfected with functionally significant quantities of miR-124; thus the small magnitude of the effects on translation and abundance of most of the mRNA targets of miR-124 identified by Ago IP was not likely a result of poor transfection efficiency. The correlation between predicted changes in protein synthesis and observed changes in protein levels for 11 of 12 proteins following miR-124 transfection (Figure 6), suggests that our assays capture most (or all) of the effects of miR-124 on protein synthesis.
Although we need to be cautious in generalizing from these model systems, in these cells under the condition examined, miRNAs appears to modulate production for hundreds of proteins through joint regulation of target mRNA translation and stability over a strikingly large dynamic range. While the repressive effects on most targets were modest (1–3-fold), there were eight targets (DNAJC1, VAMP3, CD164, SYPL1, MAGT1, HADHB, ATP6V0E1, and SGMS2) that were substantially down-regulated with decreases in protein synthesis of 10-fold or greater. In addition, 45 targets were estimated to have greater than 4-fold changes in protein synthesis. Regardless of the magnitude of regulation, mRNA destabilization accounted for ∼75% of the change in estimated protein synthesis. This range of regulation is in good accord with previous studies with genetically characterized endogenous miRNAs as well as with studies introducing exogenous miRNAs introduced into human tissue culture [7],[9],[16],[17],[33]. However, our observation that miR-124 had only modest effects on the translation of hundreds of targets contrasts dramatically with several previous studies in which miRNAs reduced protein expression by 5–25-fold while only modestly decreasing mRNA levels (1.1–2-fold), suggesting substantial inhibitory effects on translation [37],[44],[61],[69],[91]. The previous studies, however, measured the effect of a specific miRNA on reporter constructs in which the 3′-UTRs of the encoded mRNAs were not derived from mammalian mRNAs, but were either short (∼250 nts) modified viral sequences or artificial. In contrast, mammalian mRNA 3′-UTRs tend to be much longer (on average ∼1,000 nts) and include regulatory sites for RNA-binding proteins and regulatory RNAs that influence mRNA localization, translation, and decay. The basis for the discrepancy in the results from these two experimental designs remains to be determined, and the answer is likely to provide useful mechanistic insights. One possibility is that mRNAs containing exogenous 3′-UTRs might have anomalously long mRNA half-lives that obscure the normal contribution of mRNA degradation to the miRNA-directed inhibition of protein expression. The large magnitude of effects observed in reporter-based assays, compared to what we and others have observed with endogenous mRNAs, is likely to be partially due to the multiple (four to eight) engineered miRNA binding sites in the reporter constructs used in those studies [37],[44],[61],[69],[91]. Further, these sites were in close proximity, and adjacent miRNA binding sites have been reported to act cooperatively [36],[93],[98]. Indeed, two studies that measured the effects of specific miRNAs on protein and mRNA levels of reporters with endogenous mammalian 3′-UTRs found more modest effects on translation, less than 2-fold on average [11],[72]. Moreover, the magnitude of the effects we observed on translation of the mRNAs targeted by miR-124 were in agreement with two recent studies that inferred the repressive effect of miRNAs on translation by measuring miRNA-mediated effects on mRNA and protein abundance [16],[17]. Those reports, based on directly measured changes in protein levels by quantitative mass spectrometry, concluded that the effects of miRNAs on translation were small (less than 2-fold for hundreds of target mRNAs).
Although we believe that our experimental design provided a good model of miRNA regulation as it normally operates in vivo, our results do represent the full range of possible regulatory consequences of miRNA–mRNA interactions. Our results suggest that miRNAs have a large dynamic range of effects on endogenous protein expression, achieved via regulation of both translation and mRNA abundance; this pattern is generally quite consistent with previous results from cells grown in culture and limited in vivo observations. However, in specific developmental or physiological programs, or for specific mRNAs, the effects on abundance and translation, as well as the apparent mode of translation regulation may differ from what we observed in this study [7],[33],[60],[62],[99]. Thus, the effects we observed for miR-124 targets after ectopically expressing the microRNA in Hek293T cells may not capture the full scope of regulation by miRNAs in their endogenous context; miR-124 is endogenously expressed in neuronal cells, and the regulatory effects of miR-124 interactions may be modulated by the physiological demands of the cell and the specific suite of specific RNA-binding proteins and regulatory RNAs that also associate with miR-124 target mRNAs.
miR-124 negatively affected both the ribosome occupancy and ribosome density of hundreds of its targets (Figure 5). These parallel effects, combined with the close match between changes in protein synthesis predicted from miRNA-induced effects on mRNA abundance and translation and changes in protein levels for 11 of 12 proteins, suggest that the step in translation principally targeted by miR-124 and presumably other miRNAs is initiation or elongation processivity near the translation start site. We favor the initiation model because it is in accord with several previous studies that focused on one or a few mRNAs [10],[44],[52],[54]–[58], and there is a paucity of empirical evidence supporting ribosome drop-off, which predicts that ribosome density of miRNA-regulated mRNAs declines between the 5′- and 3′-ends of the coding sequence and that there should be an overrepresentation of incompletely synthesized N-terminal nascent polypeptides [61].
The small apparent magnitude of the effects on translation initiation, combined with the strong correlation between changes in translation and mRNA abundance, can be explained by a model in which repression of translational by miR-124 rapidly leads to mRNA decay. Such a model would explain why observable effects on translation appear to be smaller than the changes in mRNA abundance: if mRNAs whose translation is inhibited are quickly destroyed, their diminished translation would not be detected in our translation assay. There is already compelling evidence that translational repression and mRNA decay are linked [39],[100]–[112]. Our observation that an overwhelming majority of polyadenylated mRNAs are associated with ribosomes in HEK293T cells may be a manifestation of this relationship (Figure 3A). Thus, miRNA-mediated inhibition of translation may be linked to a general system for removal of the mRNA from the translational pool, involving recruitment to P-bodies and subsequent destruction [39]–[44]. Regulated decoupling of miRNA-mediated translation repression and mRNA decay would then allow organisms to tilt the balance of effects in favor of translational repression during physiological and developmental conditions where mRNA destruction is a disadvantage [99]. Our results are also consistent with a model in which miRNA-mediated regulation of translation and mRNA decay are functionally independent, but are similarly controlled by the same cis-elements. Determining whether the concordant regulation of translation and mRNA abundance represents a mechanistic coupling of miRNA-mediated regulation of translation and mRNA decay, and understanding the molecular links between these two regulatory consequences of miRNA–mRNA interactions are important goals for future investigation.
miR-124 siRNA:
sense: 5′-UAA GGC ACG CGG UGA AUG CCA-3′
antisense: 5′-GCA UUC ACC GCG UGC CUU AAU-3′
HEK293T cells were obtained from ATCC (Cat# CRL-11268) and grown in Dulbecco's modified Eagle's medium (DMEM) (Invitrogen) with 10% fetal bovine serum (Invitrogen) and supplemented with 100 U/ml penicillin, 100 mg/ml streptomycin, and 4 mM glutamine at 37°C and 5% CO2. Transfections of HEK293T cells were carried out with calcium phosphate. Cells were plated in 15-cm dishes 12 h prior to transfection at 2×105 cells per ml (25 ml total). We made mock-transfection mixture (1/10 volume of growth medium) by diluting 152 µl of 2 M CaCl2 into 1.25 ml of nuclease-free H2O and then slowly adding this solution to 1.25 ml of 2× HBS (50 mM Hepes [pH 7.1], 280 mM NaCl, 1.5 mM Na2HPO4). After 1 min, the mixture was added to a 15-cm plate at a medium pace. Transfections with miR-124 oligonucleotides were performed analogously with 30 nM of oligonucleotides in 2.5 ml of transfection mixture.
Ago-specific 4f9 hybridoma was grown in suspension and adapted to 10% FBS-enriched DMEM [73]. We purified the antibody by passing supernatant from 1 l of culture over a 5-ml protein L-agarose column (Pierce Cat# 89929) as per the vendor's instructions. Eluent fractions were pooled and dialyzed into PBS with Slide-A-Lyzer Dialysis Cassettes (Pierce Cat# 66382). We then biotinylated the purified 4f9 antibody with No-Weigh NHS-PEG4-Biotin Microtubes (Pierce Cat# 21329). We quantified biotinylation with EZ Biotin Quantitation Kit (Pierce Cat# 28005). Biotinylated 4f9 antibody was aliquoted and stored at −80°C until use. For Ago immunopurifications, we coupled biotinylated 4f9 antibody to DYNAL Dynabeads M-280 Streptavidin magnetic beads (Invitrogen Cat# 112-06D) (50 µg of antibody per ml of beads) as per vendor's instructions and stored the coupled beads at 4°C for up to 1 wk before use.
Twelve hours after transfection, we washed each 15-cm plate once with phosphate-buffered saline (usually two plates were used per IP), then added 1 ml of 4°C lysis buffer (150 mM KCl, 25 mM Tris-HCl [pH 7.4], 5 mM Na-EDTA [pH 8.0], 0.5% Nonidet P-40, 0.5 mM DTT, 10 µl protease inhibitor cocktail [Pierce Cat# 78437], 100 U/ml SUPERase•In [Ambion Cat# AM2694]). Following a 30-min incubation at 4°C, we scraped the plates, combined the lysates, and then spun them at 4°C for 30 min at 14,000 RPM in a microcentrifuge. We collected the supernatant and filtered it through a 0.45-µm syringe filter. We froze an aliquot of lysate in liquid nitrogen for reference RNA isolation. We then added 0.22 mg/ml heparin to the lysate. We mixed the lysate with 2.5 mg of Dynal M-280 Streptavidin beads (250 µl from original storage solution) coupled to biotinylated 4F9 ago antibody (∼12.5 µg), which we equilibrated immediately prior to use by washing twice with 1 ml of lysis buffer. We incubated the beads with the lysate for 2 h at 4°C and then washed the beads twice with 1.25 ml of ice-cold lysis buffer for 5 min each. Five percent of the beads were frozen for SDS PAGE analysis after the second wash. RNA was extracted directly from the remaining beads using lysis buffer from Invitrogen's Micro-to-Midi kit (Invitrogen Cat# 12183-018). We purified RNA from the lysate and RNA extracted from the beads with the Micro-to-Midi kit as per vender's instructions, except that the percentage isopropanol used for binding to the column was 70%, instead of 33%, to promote the binding of small RNAs.
Sixty hours after transfection, HEK293T cell lysate was prepared using the same protocol for immunoaffinity purifications. The concentration of protein in each sample was quantified using the BCA assay (Pierce Cat#23255). For SDS-PAGE separation, 25 µg of protein from each sample was used. Protein was then transferred on to a polyvinylidene fluoride (PVDF) membrane for detection with the following specific antibodies: DUSP9 (Abcam Cat# ab54941-100); PTPN11 (Bethyl Laboraties Cat# a301–544a); ITGB1 (BD Transduction Laboratories Cat# 610467); AURKA, DHCR24, MAPK14, and PLK1 (Cell Signaling Cat# 4718, 2033s, 9212, and 4513, respectively); AHR, ACTN4, CDK4, RNF128, NRAS, and PTBP2 (Santa Cruz Biotechnology Cat# sc-5579, sc-17829, sc-260, sc-101967, sc-519, and sc-101183, respectively); and TUBA1A (Sigma Cat# 096K4777). GAPDH and TUBB1 (Abcam Cat# ab9484, ab6046) were used as loading controls to check for lane-specific differences from loading, transfer, and detection errors. Protein bands were quantified using the BioRad Quantity One software package.
For translation experiments, two 15-cm dishes of cells (per condition) were seeded, grown, and transfected as described above. Twelve hours after transfection, high-purity cyclohexamide (Calbiochem Cat# 239764) was added at a final concentration of 0.1 mg/ml directly into growth media, and the plate was agitated for 1 min at room temperature. Plates were then placed on ice and washed twice with 10 ml of ice-cold buffer A (20 mM Tris [pH 8.0], 140 mM KCl, 5 mM MgCl2, 0.1 mg/ml cycloheximide). After the second wash was aspirated, the plates were tilted and left for 1 min on ice to facilitate removal of excess liquid. Each plate was then washed 1× with 2 ml of ice-cold buffer A that contained 0.22 mg/ml of heparin. After removal of excess liquid, cells were scraped from each dish and collected in a 1.5-ml microcentrifuge tube on ice. Each plate typically yielded about 300 µl (for 600 µl total) of cells and residual buffer. This mixture was then brought to 1× protease inhibitor cocktail (Pierce Cat# 78437), 100 U/ml SUPERASin, and 0.5 mM DTT. To lyse the cells, the cell-buffer mixture was brought to 0.1% Brij 58 (Sigma Aldrich Cat# P5884-100G) and 0.1% sodium deoxycholate (Sigma Aldrich Cat# D6750-100G) and vortexed for 1 min. The lysate was subsequently spun at 3,500 rpm in a microcentrifuge for 5 min at 4°C. Supernatant was collected in a fresh tube and spun at 9,500 rpm in a microcentrifuge for 5 min at 4°C. Supernatant was collected, flash frozen in liquid nitrogen, and then stored at −80°C until use.
Sucrose gradients were prepared using the Gradient Master (Biocomp) according to the manufacturer's suggestions. Five percent and 60% (w/v) sucrose solutions were prepared by dissolving sucrose in Gradient Buffer (20 mM Tris-HCl [pH 8.0], 140 mM KCl, 5 mM MgCl2, 0.5 mM DTT, 0.1 mg/ml cycloheximide) at room temperature. The 60% solution was dispensed into an SW41 ultracentrifuge tube through a cannula underneath the 5% solution. Using an 11-step program (Biocomp, SW41 SHORT SUCR 5–50 11), the two solutions were mixed on the Gradient Master to form a linear gradient. After preparation, gradients were placed in chilled SW41 ultracentrifuge buckets and equilibrated for several hours at 4°C.
Immediately before centrifugation, 300 µl of lysate (∼300 µg of total RNA) was transferred to the surface of the gradient. Gradients were centrifuged at 41,000 rpm (RCFave = 207,000) for 70 min at 4°C using a SW41 rotor and then stored at 4°C until fractionation. The Gradient Station (Biocomp) trumpet tip was pushed into the ultracentrifuge tube at a rate of 0.17 mm per second. Fractions (550 µl) were collected into a 96-well plate containing 600 µl of lysis solution (Invitrogen) using a fraction collector (Teledyne-Isco). The absorbance of the gradient at 260 nm was measured during fractionation using a UV6 system (Teledyne-Isco).
Immediately after fractionation, a unique set of four to five polyadenylate-tailed control RNAs, corresponding to Methanococcus jannaschii mRNAs that do not share significant identify to sequences in the human genome, were added at 100 pg each to fractions that contained the 80S ribosome and polysomes (Table S2). The solution was mixed well by inverting the plate several times, and liquid was collected in the well bottom by a brief centrifugation. A Precision XS liquid handler (BioTek Intruments) was used to transfer a defined volume of each of the fractions to one of four tubes (Fisher Cat# 14-959-11B) (Table S2); the solutions in each tube are referred to as pool “A,” “B,” “C,” and “D,” respectively. Upon completion of liquid handling, eight additional control RNAs (Ambion Cat# 1780) (Table S1) were added to each pool, and the pools were stored at −20°C.
Pools A–D were thawed at room temperature for 30 min. Two volumes of isopropanol was added to each pool, and the RNA in each pool was isolated from the mixture using the Micro-to-Midi RNA isolation kit (Invitrogen Cat# 12183-018).
HEEBO oligonucleotide microarrays were printed on epoxysilane-coated glass (Schott Nexterion E) by the Stanford Functional Genomic Facility. The HEEBO microarrays contain ∼45,000 70-mer oligonucleotide probes, representing ∼30,000 unique genes. A detailed description of this probe set can be found at (http://microarray.org/sfgf/heebo.do) [113].
Prior to hybridization, slides were first incubated in a humidity chamber (Sigma Cat# H6644) containing 0.5× SSC (1× SSC = 150 mM NaCl, 15 mM sodium citrate [pH 7.0]) for 30 min at room temperature. Slides were snap-dried at 70–80°C on an inverted heat block. The free epoxysilane groups were blocked by incubation with 1M Tris-HCl (pH 9.0), 100 mM ethanolamine, and 0.1% SDS for 20 min at 50°C. Slides were washed twice for 1 min each with 400 ml of water, and then dried by centrifugation. Slides were used the same day.
Amplified RNA was used for most DNA microarray experiments. Poly-adenylated RNAs were amplified in the presence of aminoallyl-UTP with Amino Allyl MessageAmp II aRNA kit (Ambion Cat# 1753). For mRNA expression experiments, universal reference RNA was used as an internal standard to enable reliable comparison of relative transcript levels in multiple samples (Stratagene Cat# 740000). Amplified RNA (3–10 µg) was fluorescently labeled with NHS-monoester Cy5 or Cy3 (GE HealthSciences Cat# RPN5661). Dye-labeled RNA was fragmented (Ambion Cat# 8740), then diluted into in a 50-µl solution containing 3× SSC, 25 mM Hepes-NaOH (pH 7.0), 20 µg of human Cot-1 DNA (Invitrogen Cat# 15279011), 20 µg of poly(A) RNA (Sigma Cat# P9403), 25 µg of yeast tRNA (Invitrogen Cat# 15401029), and 0.3% SDS. The sample was incubated at 70°C for 5 min, spun at 14,000 rpm for 10 min in a microcentrifuge, then hybridized at 65°C using the MAUI hybridization system (BioMicro) for 12–16 h. For translation experiments, amplified RNA from pools A and C was fluorescently labeled with NHS-monoester Cy5, and RNA from pools B and D was fluorescently labeled with NHS-monoester Cy3. Amplified RNA from pools A and B were comparatively hybridized to a DNA microarray to obtain the average ribosome density, and amplified RNA from pools C and D were comparatively hybridized to a DNA microarray to measure ribosome occupancy.
To compare total RNA levels in miR-124 and mock-transfected cells (Figure S3), 5–10 µg of total RNA from miR-124–transfected cells or mock-transfected cells or universal reference RNA (Stratagene Cat# 740000) was reverse transcribed with Superscript III (Invitrogen Cat# 18080085) in the presence of aminoallul-dUTP 5-(3-aminoallyl)-dUTP (Ambion Cat# AM8439) and natural dNTPs (GE Healthsciences Cat# US77212) with 10 µg of N9 primer (Invitrogen). Subsequently, amino-allyl–containing cDNAs from miR-124 and mock-transfected cells were covalently linked to Cy5 NHS-monoesters, and universal reference cDNA was covalently linked to Cy3 NHS-monoesters (GE HealthSciences Cat# RPN5661). Cy5- and Cy3-labeled cDNAs were mixed and diluted into 50 µl of solution containing 3× SSC, 25 mM Hepes-NaOH (pH 7.0), 20 µg of human Cot-1 DNA (Invitrogen Cat# 15279011), 20 µg of poly(A) RNA (Sigma Cat# P9403), 25 µg of yeast tRNA (Invitrogen Cat# 15401029), and 0.3% SDS. The sample was incubated at 95°C for 2 min, spun at 14,000 rpm for 10 min in a microcentrifuge, then hybridized at 65°C for 12–16 h.
Following hybridization, microarrays were washed in a series of four solutions containing 400 ml of 2× SSC with 0.05% SDS, 20058 SSC, 1× SSC, and 0.2× SSC, respectively. The first wash was performed for 5 min at 65°C. The subsequent washes were performed at room temperature for 2 min each. Following the last wash, the microarrays were dried by centrifugation in a low-ozone environment (<5 ppb) to prevent destruction of Cy dyes [114]. Once dry, the microarrays were kept in a low-ozone environment during storage and scanning (see http://cmgm.stanford.edu/pbrown/protocols/index.html).
Microarrays were scanned using AxonScanner 4000B (Molecular Devices). PMT levels were autoadjusted to achieve 0.1–0.25% pixel saturation. Each element was located and analyzed using SpotReader (Niles Scientific) and GenePix Pro 6.0 (Molecular Devices). For IP and mRNA expression experiments, the microarrays were submitted to the Stanford Microarray Database for further analysis [115]. Data were filtered to exclude elements that did not have one of the following: a regression correlation of ≥0.7 between Cy5 and Cy3 signal over the pixels compromising the array element, or an intensity/background ratio of ≥3 in at least one channel.
Ribosome density (pool A versus B) and ribosome occupancy (pool C versus D) measurements were normalized using exogenous doping control RNAs to correct for experimental variation between the two pools from RNA isolation, labeling efficiency, and scanning levels. In most cases, oligonucleotides that were designed to measure the exogenous doping control RNAs were represented multiple times on the DNA microarray (up to eight) and printed from different plates with different print tips. For ribosome occupancy experiments, the measured Cy5/Cy3 ratios of features on the microarray that correspond to the eight RNA controls added to pools C and D were fit to their expected Cy5/Cy3 ratios using least-squares linear regression in the statistical computing program R. The slope and y-intercept were used to rescale the measured Cy5 value of all other features on the DNA microarray. The ribosome occupancy for each RNA was then calculated as the corrected Cy5 intensity/(corrected Cy5 intensity + Cy3 intensity) (Figure S3C).
To calculate the average number of ribosomes bound to each mRNA, the measured Cy5/Cy3 ratios of features on the microarray that correspond to the 85 M. jannaschii doping control RNAs that were added to fractions that contained ribosomes pools was fit to their expected Cy5/Cy3 ratios using least-squares linear regression. The slope and y-intercept were used to rescale the measured Cy5 value of all other features on the DNA microarray (Figure S3B).
The average ribosome density was calculated by dividing the average ribosome number by coding sequence length and then multiplying the result by 100 to give density per 100 nts. The average ribosome number was calculated using two relationships. For each ribosome peak in the profile, the distance traveled from the start point was determined. In all gradients, we could clearly resolve peaks for up to seven bound ribosomes, and we used least-squares regression to relate the ribosome peaks 1–7 to their distance traveled in the gradient according to the following equation:(3)where R represents the number of ribosomes bound, DT represents the distance traveled, and a and b are the slope and y-intercept, respectively. We then recorded the distance between the midpoint of each fraction to the start of the profile for each of the 15 ribosome-bound fractions and used the slope and y-intercept from Equation 3 to calculate the number of ribosomes at each fraction midpoint. The gradient encoding ratio at each fraction midpoint is the result of differential partitioning of each fraction in a predetermined manner into the heavy and light pools, and the ratio can be related to the ribosome number at each fraction midpoint using least-squares linear regression as described by Equation 4:(4)where R represents the average ribosome number, and ER represents the encoding ratio. Finally, the average number of ribosomes bound for each gene's mRNAs was calculated using the slope and y-intercept from Equation 4.
Prior to normalization, spots with intensity/background of less than 1.5 for either Cy3 or Cy5 channel were filtered.
The microarray data are available from Gene Expression Omnibus (GEO) (http://www.ncbi.nlm.nih.gov/geo/) and Stanford Microarray Database.
Hierarchical clustering was performed with Cluster 3.0 [116] and visualized with Java TreeView 1.0.12 [117].
For SAM, unpaired two-class t-tests were performed with default settings (R-package samr; http://cran.r-project.org/web/packages/samr/index.html). Microarray features that passed quality filtering in all experiments were used as input. Ago IP experiments (Dataset S1) and mRNA expression experiments (Dataset S4) were mean centered at log2 0 prior to running SAM. The ribosome occupancy (Dataset S2) and ribosome number/density measurements (Dataset S3) from miR-124 and mock-transfected cells were highly correlated, but had slightly different means (see main text). Because of the small changes in ribosome occupancy and ribosome density between miR-124–transfected and mock-transfected samples, we conservatively adjusted the means of each experiment to be the same by subtracting the difference between the mean of that experiment and the mean of all the experiments to ensure that differences observed between miR-124–transfected and mock-transfected cells were not due to the doping control normalization.
Enrichment of GO terms was performed with Genetrail [118]. p-Values were corrected for multiple-hypothesis testing by the Bonferroni method [119].
The significance of correlations was estimated in R by recalculating the correlations with 10,000 permuted sets of data, then estimating the p-value with the normal distribution function using the mean and standard deviation from the permuted data.
We used a bootstrap method to estimate 95% confidence intervals for the average changes in mRNA abundance, estimated translation rate, ribosome occupancy, and ribosome density (Figures 4 and 5, and Figure S8) of IP targets compared to nontargets. To do this, we sampled with replacement measurements for each gene from the mock and miR-124 replicates, respectively, 10,000 times, then calculated the respective changes between miR-124 IP targets and nontargets for the 10,000 bootstrapped samples.
For molecular features that mapped to genomic loci with an Entrez ID, the RefSeq sequence with the longest 3′-UTR was used. In cases with multiple RefSeqs with the same 3′-UTR length, the one that was alphanumerically first was used. RefSeq 3′-UTR, coding, and 5′-UTR sequences were retrieved from UCSC genome browser (hg18) http://genome.ucsc.edu/. Seed match sites in these sequences were identified with Perl scripts. miR-124 seed matches: 6mer_n2-7 “UGCCUU,” 6mer_n3-8 “GUGCCU,” 7mer-m8 “GUGCCUU,” 7mer-A1 “UGCCUUA,” 8mer “GUGCCUUA.” In many instances, there were multiple probes on the DNA microarrays that mapped to the same RefSeq. In these cases, we used the probe that was most enriched in Ago IPs from miR-124–transfected cells compared to mock-transfected cells.
|
10.1371/journal.pntd.0000685 | The Impact of Schistosoma japonicum Infection and Treatment on Ultrasound-Detectable Morbidity: A Five-Year Cohort Study in Southwest China | Ultrasonography allows for non-invasive examination of the liver and spleen and can further our understanding of schistosomiasis morbidity.
We followed 578 people in Southwest China for up to five years. Participants were tested for Schistosoma japonicum infection in stool and seven standard measures of the liver and spleen were obtained using ultrasound to evaluate the relationship between schistosomiasis infection and ultrasound-detectable pathology, and the impact of targeted treatment on morbidity. Parenchymal fibrosis, a network pattern of the liver unique to S. japonicum, was associated with infection at the time of ultrasound (OR 1.40, 95% CI: 1.03–1.90) and infection intensity (test for trend, p = 0.002), adjusting for age, sex and year, and more strongly associated with prior infection status and intensity (adjusted OR 1.84, 95% CI: 1.30–2.60; test for trend: p<0.001 respectively), despite prompt treatment of infections. While declines in parenchymal fibrosis over time were statistically significant, only 28% of individuals with severe parenchymal fibrosis (grades 2 or 3) at enrollment reversed to normal or grade 1 within five years. Other liver abnormalities were less consistently associated with S. japonicum infection.
Parenchymal fibrosis is an appropriate measure of S. japonicum morbidity and can document reductions in disease following control efforts. Other ultrasound measures may have limited epidemiological value in regions with similar infection levels. Because severe fibrosis may not reverse quickly following treatment, efforts to reduce exposure to S. japonicum should be considered in combination with treatment to prevent schistosomiasis morbidity.
| Schistosomiasis is a water-borne parasite that infects approximately 200 million people worldwide. Schistosoma japonicum, found in Asia, causes disease by releasing eggs in the liver, leading to fibrosis, anemia, and, in children, impaired growth. Ultrasound can assess liver pathology from schistosomiasis; however more information is needed to evaluate the relevance of standard ultrasound measures. We followed 578 people for up to five years, testing for schistosomiasis infection and conducting ultrasound examinations to assess the relationship between infection and seven ultrasound measures and to evaluate the impact of treatment with anti-schistosomiasis chemotherapy (praziquantel) on morbidity. All infections were promptly treated. Fibrosis of the liver parenchyma, pathology unique to S. japonicum, was associated with schistosomiasis infection, and was most advanced in people with high worm burdens. Liver fibrosis declined significantly following treatment, but reversal of severe liver fibrosis was rare. Other ultrasound measures were not consistently related to schistosomiasis infection or treatment. These findings suggest parenchymal fibrosis can be used to measure morbidity attributable to S. japonicum and evaluate the impact of disease control efforts. Because reversal of severe fibrosis was limited, disease control efforts will be most effective if they can not only treat existing infections but also prevent new infections.
| Schistosomiasis causes morbidity in the human host through the schistosome egg, which triggers inflammation and fibrosis that can lead to anemia, impaired growth and in severe cases, gastrointestinal bleeding and death [1]–[4]. The major intestinal schistosomes, Schistosoma japonicum, found in Asia, and S. mansoni, found in Africa, the Americas and the Middle East, mature, mate and lay eggs in the portal and mesenteric blood vessels. Eggs are transported to the liver where they are encapsulated and the granulomas that form induce an inflammatory cascade that includes the deposition of collagen and extracellular matrix proteins, a normal liver repair process that can lead to fibrosis when fibrogenesis exceeds the replacement of scar tissue with healthy cells [5], [6]. The immune regulation of this process is currently being explored [7]. Approximately 700,000 people are infected with S. japonicum in China [8]. As in other parts of the world, schistosomiasis control efforts have focused primarily on the distribution of the antischistosomal drug, praziquantel [9], [10]. However the success of such efforts hinges on the ability to reduce not only schistosomiasis infections, but also morbidity. A means of documenting S. japonicum morbidity is essential to the evaluation of disease control efforts [11].
Ultrasound is a non-invasive method that can be used to evaluate fibrosis resulting from schistosomiasis infection. Fibrosis along the portal vein and its branches produces a clay pipestem pattern, as the portal tracks are lined with fibrous tissue, and is observed following both S. mansoni and S. japonicum infection. This periportal fibrosis can be assessed qualitatively through image classification [12] or quantitatively by measuring the diameter of three secondary portal branches [13]. Unique to S. japonicum infection is parenchymal fibrosis, a network pattern that is often described as fish scale or tortoise shell-like. Parenchymal fibrosis is likely due to the smaller S. japonicum egg size which allows the parasite eggs to enter smaller portal veins and reach a greater portion of the liver [7]. S. japonicum adult females produce ten times more eggs per day than S. mansoni, and the eggs often are deposited in clusters, two additional factors that exacerbate the severity of S. japonicum morbidity relative to other schistosome species and may contribute to the unique fibrotic pattern [3]. Ultrasound can also be used to assess hepatomegaly, splenomegaly and dilation of the portal vein, all of which result from portal hypertension. Recognizing the potential of ultrasound to be used to evaluate the impact of disease control efforts, draft protocols for the use of ultrasound in assessing schistosomiasis morbidity were established for each of the three major schistosome species in 1990 [13], [14]. Follow-up meetings to refine the protocols for S. haematobium and S. mansoni were held in 1996 and 1997 [15]. Ultrasound is considered the gold standard for schistosomiasis morbidity assessment for these species [11]. However, due to insufficient evidence the protocol for S. japonicum has not been revised.
To date, a comprehensive evaluation of the S. japonicum ultrasound measures proposed in Cairo, including their relationship to S. japonicum infection and the way in which they change following treatment, is lacking. Li et al. [16], [17] offer the most complete examination of the proposed ultrasound measures to date, tracking a highly exposed cohort over five years, but as standard organ sizes have only recently been published [18], assessments of liver and spleen enlargement did not account for participant height as currently recommended [15]. Failure to account for height-specific variation in organ sizes can lead to underestimates of morbidity, particularly in pediatric populations [19]. Ideally, ultrasound measures are associated with S. japonicum infection and decline following treatment [13]. The available data is conflicting on both criteria. Treatment led to declines in parenchymal and periportal fibrosis in a highly exposed Chinese cohort; however 68% of participants experienced no change in parenchymal fibrosis over a five-year period [16], [17] and other studies have not found significant declines in periportal fibrosis following S. japonicum treatment [20], [21]. Further, a direct relationship between infection and fibrosis or measures of portal hypertension has not been demonstrated consistently [16], [17], [22].
We followed 578 individuals over five years in order to examine the relationship between S. japonicum infection and five liver ultrasound measurements recommended in the draft protocol as well as two spleen ultrasound measurements included in the standard Chinese examination. Specifically, we hypothesized that S. japonicum infection is associated with ultrasound-detectable measures of hepatic fibrosis and that treatment of infected individuals leads to declines in ultrasound-detected morbidity.
Participants were drawn from a cross-sectional survey in 2000 that revealed high infection prevalence (ranging from 3% to 73%) in 20 villages distributed in the hilly terrain of Sichuan Province in Southwest China where irrigated agriculture for the cultivation of rice, wheat, corn and tobacco is the primary source of S. japonicum infection [23]. As part of the survey, all residents aged 4 to 60 years in 20 villages Xichang County, Sichuan Province, China were invited to be tested for S. japonicum infection and to answer a brief questionnaire. We used a random number generator to select approximately 30% of the population, stratified by village and occupation, for ultrasound examination. Ten villages with high infection prevalence in 2000 were selected to be followed longitudinally using infection surveys and ultrasound examinations over the next five years. Participants in our cohort included individuals who had an ultrasound examination in 2000 and lived in one of the ten villages followed through 2005.
All participants provided oral informed consent and were provided treatment following S. japonicum positive stool examinations. As the survey procedures used in this study are the same as those used by the Institute of Parasitic Diseases, Sichuan Center for Disease Control and Prevention (IPD) for schistosomiasis surveillance and given the high rates of illiteracy in the population, oral informed consent was obtained and documented by IPD staff. The research protocol and consent procedures were approved by the Sichuan Institutional Review Board and the University of California, Berkeley Committee for the Protection of Human Subjects.
Ultrasound examinations were conducted in fall 2000, 2002 and 2005. All examinations were conducted by one trained ultrasonographer (YZ) using a single portable ultrasound machine (Hitachi EUB 405, Hitachi Corporation, Tokyo, Japan) and 3.5 MHz probe (Hitachi EUP-C314T, Hitachi Corporation, Tokyo, Japan) with participants in the supine position at a central location in each village. The ultrasonographer was blind to infection status.
Liver ultrasonography was conducted according to the 1990 draft guidelines [13], [14]. Liver parenchymal fibrosis was graded 1 through 3 based on observed lesions, or 0 if none were present. Periportal fibrosis was assessed by grading the average diameter, from outer wall to outer wall, of three peripheral branches of the portal vein between the first and third branching point (grade 0: <3mm; grade 1: 3 to 5 mm; grade 2: >5 to 7 mm; grade 3: >7 mm). As done previously, grades 0 and 1 were combined for analysis [16]. The internal diameter of the portal vein was measured at the entry point of the portal vein into the liver. The length of the left liver lobe was measured in a longitudinal section along the left parasternal line, and the length of the right liver lobe was measured as the maximum oblique diameter using a right anterior axillary view according to the revised guidelines established for S. mansoni [15].
Two measurements of the spleen that are part of the standard Chinese examination were also included: spleen thickness, measured from the hilum to the opposite section, and the internal diameter of the spleen vein, measured at the entry point to the spleen [24].
Because organ and vein sizes vary with height, left and right liver lobe length, portal vein diameter and spleen thickness were evaluated using height-specific standard sizes drawn from a Chinese population where schistosomiasis is not endemic [18]. Measurements greater than two standard deviations above the mean size for height were classified as abnormal. Height was measured in 2002 in 440 of the 578 cohort members. The 343 adults (≥18 years old in 2000) with height measurements were assigned their 2002 height throughout the study. Children (<18 years old in 2000) were measured again in 2007 as part of a study described elsewhere [19]. The height measurements from 2002 and 2007 from 119 children (60 with 1 height measurement, 59 with two height measurements) were used to generate an age- and sex-dependent random intercept model in order to impute heights during the years children weren't measured. Height for child i at time j was calculated using the following equation:where Aij represents a child's age at time j, Si represents his or her sex (S = 1 for males) and represents his or her random intercept. The model was fit using xtmixed in Stata 10.1 (StataCorp, College Station, TX, USA) and was predicted using empirical Bayes [25]. As random intercept models assume parametric distribution of residuals, first and second order residuals were examined and were normally distributed. The selected model was superior to a model that did not include a random intercept (likelihood ratio test, χ2 = 38.20, 1 d.f., p<0.001). Organ and vessel measures could not be height adjusted for the 86 adults and 26 children without height measurements. Participants with any versus no height data did not differ significantly by any of the predictors examined in the analyses including sex, mean age or infection status. No height standardized values were available for spleen vein diameter, so it was evaluated by the standard threshold used in China: >8 mm [24].
Participants were tested for infection with S. japonicum in the fall of 2000 and 2002 using the miracidial hatch test [26] and the Kato-Katz thick smear procedure [27]. For each hatch test, a stool sample weighing at least 30 g was suspended in aqueous solution, filtered using copper mesh to remove large particles (40–60 mesh) followed by nylon gauze (260 mesh) to concentrate schistosome eggs. This sediment was re-suspended with distilled water in a 250 ml Erlenmeyer flask. Flasks were examined for miracidia 30–60 minutes, 4 hours and 8 hours after suspension if temperatures were above 30 degrees Celsius, or at 6, 12 and 18 hours at lower ambient temperatures. In 2000, three miracidial hatch tests were conducted per person using stool samples from three distinct days. In 2002, due to logistical constraints, one miracidial hatch test was conducted per person; however the Kato-Katz protocol was identical both years. The Kato-Katz procedure involved the preparation of three 41.5 mg slides from one homogenized stool sample in 2000 and 2002. Infection intensity, in eggs per gram of stool (EPG), was calculated as the total number of S. japonicum eggs present on the slides divided by the total sample weight. Participants were classified as infected if at least one test was positive. Everyone testing positive for S. japonicum was provided praziquantel treatment by the county Anti-Schistosomiasis Control Station. In addition, praziquantel was administered to all residents in the study villages in 2003, as part of a nation-wide effort to control infectious diseases following the outbreak of severe acute respiratory syndrome.
Participant age, sex, occupation and highest level of schooling were obtained by interview in fall 2000.
In order to assess whether the participants in the ultrasound cohort were representative of the village populations from which they were selected, cohort participants were compared to individuals who participated only in the cross-sectional demographic and infection surveys in terms of age, sex, occupation, educational attainment and 2000 S. japonicum infection status. Similarly, cohort members with complete vs. incomplete follow-up were compared in terms of age, sex, baseline morbidity and infection in order to assess non-differential loss to follow-up. In both cases, comparisons were conducted using the χ2 and t-test.
Analyses that included multiple observations from individuals accounted for within subject correlation of outcomes. Liver parenchymal fibrosis grade, an ordinal measure, was examined using ordinal logistic regression, a population averaged model using a sandwich type estimator for inference accounting for within-subject residual correlation [28]. Because ordinal logistic regression assumes the effect of a predictor on an outcome is constant for each stepwise increase in the outcome, the Brant test [29] was used to check that this parallel regression assumption was not violated. For all other liver and spleen measures and for predictors of S. japonicum infection status, generalized estimating equation (GEE) logistic regression with exchangeable correlation was conducted [30]. The Huber/White/sandwich estimator of variance was used which is robust to misspecification of the outcome distribution [31], [32].
The relationship between S. japonicum infection and ultrasound-detected abnormalities was first assessed by examining the impact of current infection status and, separately, infection intensity, on current ultrasound measures. Because infection intensity was highly right skewed, it was categorized into approximate quartiles among those who were infected. In order to examine the role of past infection on current morbidity, we also examined the relationship between infection status and intensity two to three years prior to the ultrasound examination and ultrasound-detected abnormalities (for example: 2000 infection status as a predictor of ultrasound-detected morbidity in 2002, and 2002 infection status as a predictor of ultrasound-detected morbidity in 2005). We hypothesized that age, sex and year of examination could modify the effect of infection on morbidity, so each model was run including all possible first-order interaction terms. The Wald test was used to test for significant interactions and if present, terms were removed from the model step-wise until only interaction terms significant at p-values <0.05 remained. In the absence of effect modification, these same variables could confound the relationship between infection and ultrasound-detected morbidity. While occupation, village and educational status were considered potential predictors of S. japonicum infection, they are unlikely to effect ultrasound morbidity independent of infection status. As they do not fit the definition of confounders [33], they were not controlled for in models examining the relationship between infection and ultrasound-detected abnormalities. The change in ultrasound-detected abnormalities over time was examined adjusting for age and sex.
We modeled age as a categorical variable when examining the relationship between age and ultrasound-detected morbidity to allow a non-linear relationship. Because liver abnormalities increase with age, age was treated as a continuous variable when included as a confounder in models testing the relationship between infection and morbidity, or changes in morbidity over time. Periportal fibrosis grade was modeled as a binary variable, as grades 0 and 1 were combined and no grade 3 fibrosis was detected. Tests for trend were calculated by treating categorical variables as ordinal. All results were assessed for statistical significance setting α = 0.05. Statistical analyses were conducted using Stata 10.1 (StataCorp, College Station, TX, USA).
In 2000, 578 people from ten villages were examined using ultrasound. The mean age of participants was 29.8 years (range 4–61 years). Half (51%) were female (Table 1). Most adults (≥18 years) were farmers (91%) and had no formal schooling beyond elementary school (59%). Participants were similar in terms of occupation, educational attainment and 2000 infection status to the 1,333 residents in these villages who participated in infection and demographic surveys only, although cohort members were slightly older than the rest of the population (mean age 29.8 vs. 28.2, p = 0.037).
We conducted ultrasound examinations with 444 people in 2002 and 321 people in 2005. The 320 people with complete ultrasound follow-up were no more likely to be infected with S. japonicum at enrollment than those who missed at least one follow-up examination, but they were older (mean age 33.0 vs. 25.6, p<0.001) and more likely to have at least one liver abnormality in 2000 (57% vs. 37%, p<0.001).
Nearly half (47%) of participants tested positive for S. japonicum in 2000. Mean infection intensity was 53.4 EPG. In 2002, infection prevalence declined to 32% and intensity to 9.4 EPG. As shown in Table 2, adults aged 50 and older were less likely to be infected than younger participants. Neither sex nor occupation was associated with infection prevalence, but among adults, higher educational attainment was protective. Infection prevalence in 2002 was significantly higher among those who were infected vs. uninfected in 2000 despite the distribution of treatment to everyone who tested positive (41% vs. 23%, p<0.001). Infection prevalence varied significantly by village.
Table 3 describes the prevalence of liver parenchymal fibrosis, periportal fibrosis and abnormal liver and spleen measurements from 2000 to 2005. Following the initiation of schistosomiasis testing and treatment of all infections in 2000, parenchymal fibrosis, periportal fibrosis and right liver lobe enlargement decreased significantly through 2005, controlling for age and sex (Table 4). Decreases in spleen enlargement were also observed, although the trend was of marginal significance (p = 0.091).
Liver abnormalities increased significantly with age, most notably for parenchymal fibrosis (Table 4). Spleen enlargement was not associated with age. Individuals aged 18 to 29 years had the highest odds of spleen vein dilation. The relationship between liver abnormalities and sex varied: men were more likely to have periportal fibrosis and enlarged right liver lobes; women were more likely to have enlarged left liver lobes.
Schistosomiasis infection at the time of ultrasound was associated with an increase in liver parenchymal fibrosis grade (OR 1.40, 95% CI: 1.03–1.90) adjusting for age, sex and year of ultrasound (Table 5). Infection intensity at the time of ultrasound was also associated with an increase in liver parenchymal fibrosis grade (test for trend: p = 0.002). Individuals with greater than 50 EPG had 2.10 times greater odds of more advanced fibrosis than those not excreting eggs (95% CI: 1.33–3.32).
Schistosomiasis infection two to three years prior to ultrasound was associated with an increase in liver parenchymal fibrosis grade, and the association was stronger than that of current infection (OR 1.84, 95% CI: 1.30–2.60). Prior infection intensity was also associated with liver parenchymal fibrosis (test for trend, p<0.001). Individuals with greater than 50 EPG two to three years prior to the ultrasound examination had 2.84 times greater odds of advanced fibrosis than those not excreting eggs two to three years prior to ultrasound (95% CI: 1.71–4.73).
The other hepatosplenic ultrasound measures were not associated with current infection status or intensity. Several measures were associated with prior infection, although the relationships were not as consistent as those observed for periportal fibrosis. Prior infection appeared to elevate the probability of left liver lobe enlargement (OR 1.45, 95% CI: 0.97–2.17). Prior infection intensity but not prior infection status, was associated with increased odds of portal vein dilation (test for trend, p = 0.051). The impact of prior infection status on right liver lobe enlargement varied by the year of examination and the sex of the participant. Prior infection was associated with increased odds of right liver lobe enlargement among males examined in 2005 (OR 3.95, 95% CI: 1.82–8.57) and decreased odds of right liver lobe enlargement among females examined in 2002 (OR 0.33, 95% CI: 0.13–0.86). Prior infection intensity was not associated with right or left liver lobe enlargement. Due to the limited number of individuals with periportal fibrosis, models were unable to yield stable estimates of the effect of prior infection on this measure Spleen enlargement and spleen vein dilation were not associated with current or prior infection.
Most people with liver and spleen enlargement, portal vein dilation, periportal fibrosis or spleen vein dilation in 2000 had normal pathology by 2002 (Table 6). However, this was not the case for parenchymal fibrosis: 67% of people with grade 2 fibrosis and 100% with grade 3 fibrosis in 2000 remained at or above grade 2 throughout the five-year follow-up.
We found evidence of a direct, exposure-response relationship between S. japonicum infection and parenchymal fibrosis. While there has been suggestive evidence of an association between infection and parenchymal fibrosis, including an association between progression of parenchymal fibrosis and current infection [16], this is the first study to show the risk of parenchymal fibrosis is higher in people who are infected vs. uninfected, and highest in individuals with the greatest infection intensities. Parenchymal fibrosis declined significantly following treatment however, improvements were limited among individuals with advanced fibrosis: 72% of people with severe fibrosis at enrollment (grades 2 or 3) had not resolved below grade 2 by the end of the five-year study. This suggests parenchymal fibrosis is an appropriate measure of S. japonicum morbidity and can document improvements in morbidity following treatment, although little improvement may be observed among those with advanced fibrosis.
In contrast, the remaining measures were less consistently associated with S. japonicum infection and are of questionable epidemiological use in regions with similar infection levels. Periportal fibrosis was rare in this population and could not be associated with S. japonicum infection, although it did decline significantly following the initiation of targeted treatment. Others have used image-based classification to assess periportal fibrosis [34], [35], which has been shown to have better reproducibility for S. mansoni-related fibrosis, perhaps because it does not require the ultrasonographer to measure narrow vessel widths [36]. Our measures of periportal fibrosis were not height adjusted, as Chinese standards for the diameter of portal vein branches as measured here have not been published. The lack of height adjustment may have led to underestimates of the prevalence of periportal fibrosis and may explain the higher prevalence of periportal fibrosis in men, however higher prevalence of fibrosis has also been observed in males using image-based classification [35].
The remaining three hepatic measures, left and right liver lobe enlargement and portal vein dilation, were associated with S. japonicum infection or declined following chemotherapy. However, the relationships between S. japonicum infection and these morbidity measures were less consistent than those observed with parenchymal fibrosis and several associations were of marginal statistical significance. The spleen measures were neither associated with infection nor did they decline with treatment. All five measures require the ultrasonographer to measure organ or vessel sizes and participant's height. Like the measures of periportal fibrosis used here, these methods present opportunities for measurement error which may bias estimates of exposure-disease relationships toward the null, producing attenuation and exacerbating non-linearity [37]. The reproducibility of some of these measures has been shown to be poor in a pediatric population, and further efforts to evaluate the accuracy of organ and vein size measurements are needed [19].
Overall, the prevalence of morbidity in this cohort was lower than has been observed elsewhere in China as was infection intensity [16], [17], [34], [38]. Measures of liver and spleen enlargement, portal vein dilation and spleen vein dilation were less informative than measures of parenchymal fibrosis but may be appropriate in areas where infection intensity and associated morbidities are higher. Declines in hepatomegaly and splenomegaly have been demonstrated following treatment in regions with higher average worm burdens [16], [17]. The strong associations between age and all liver measures highlight the importance of considering age as a potential confounder when examining the relationship between S. japonicum infection and morbidity and declines in morbidity over time. For example, the unadjusted prevalence of parenchymal fibrosis did not change over time, however as loss to follow-up was highest in younger populations which were least likely to have fibrosis, the age and sex adjusted prevalence did decline.
Our findings shed light on the development of hepatic fibrosis following S. japonicum infection. The lack of reversal of grade 3 parenchymal fibrosis to grade 1 or normal and the association between past infection and current fibrosis suggest severe fibrosis may persist or even progress following treatment. Prior studies have also found minimal reversal of severe fibrosis following treatment [16], [39] and hepatic fibrosis has been documented in people living in areas where the parasite, and therefore exposure, was eliminated 20 years previously [40]. While treatment with praziquantel can reduce infection prevalence, intensity and fibrosis, this analysis provides further evidence that severe hepatic fibrosis may be unlikely to reverse quickly. Minimal declines in severe hepatic fibrosis associated with S. mansoni infection have also been detected [41] but other studies have noted complete reversal of severe morbidity [42].
Further, parenchymal fibrosis was associated with current infection status and intensity but more strongly associated with infection status and intensity two to three years prior to ultrasound, despite timely treatment of infections. The surprising association between past infection and fibrosis raises questions about the progression of fibrosis following infection and treatment. Praziquantel kills the adult worm but schistosome eggs can remain trapped in host tissues decades after exposure [43]. While schistosome eggs are thought to disintegrate within weeks of granuloma formation [7], it is possible, due to an inflammatory cascade, fibrogenesis continues over a longer time period. Collagen deposition has been shown to occur following treatment with praziquantel in S. japonicum infected mice, suggesting fibrosis may continue despite removal of adult worms [44]. In humans, progression of fibrosis has been observed following treatment and was not explained by reinfection [21], [45]. Alternatively, it is possible that the relationship between past infection and fibrosis is due to reinfection. Past infection predicted subsequent infection. However, if reinfection rather than past infection determined fibrosis, one would expect stronger relationships between current infection and fibrosis than between past infection and fibrosis, which was not observed. It is also possible that some individuals who were treated were not cured. A single dose of praziquantel cures 90% of S. japonicum infections [46], suggesting 10% of those treated may continue to harbor adult worms and face continued egg production. In our work in this region, we have also found individuals who decline to take praziquantel even after a positive infection test. Non-compliance and treatment failure are realities of any chemotherapy-based control program, however the marked declines in infection prevalence, intensity and morbidity suggest the number of individuals who were not cured was limited.
This study examined one of the largest populations to be followed over five years in order to assess ultrasound-detectable morbidity and S. japonicum infection. Participants were randomly sampled from a comprehensive cross-sectional survey in order to minimize selection bias. Loss to follow-up was greater among those without ultrasound-detectable morbidity, which suggests the prevalence of morbidity in 2002 and 2005 may be overestimated, and therefore the true declines in morbidity over time may be greater than observed. Retention was independent of infection status, minimizing bias in estimates of the impact of infection on pathology. No information was available on alcohol consumption or infection with hepatitis B virus (HBV), two factors that can induce liver pathology and may exacerbate schistosomiasis morbidity. The parenchymal network pattern due to S. japonicum is distinct from the lesions produced by HBV, as HBV produces a finer, meshwork texture [47], suggesting the observed prevalence of parenchymal fibrosis is specific to schistosomiasis. While HBV has been shown to hinder regression of periportal fibrosis following treatment of S. mansoni infections [41], the extent to which HBV exacerbates morbidity due to S. japonicum remains unclear and warrants further study. Reductions in parenchymal fibrosis may be impaired by alcohol consumption, which in our study areas is confined almost exclusively to males [17]. Unless alcohol consumption or HBV impacts a person's probability of infection, they are unlikely to confound the relationships between infection and ultrasound-detected morbidity.
Community-wide testing and treatment of all infections with praziquantel yielded marked declines in infection prevalence and intensity. However, reinfection was high: 32% of people were infected two years after the first round of targeted treatment. High rates of infection following treatment are not uncommon and underscore the challenges of sustainably reducing human schistosomiasis [48]. Adults aged 50 and over were less likely to be infected than younger populations, possibly corresponding to a decline in water contact later in life or acquired immunity [49].
In summary, we present evidence that ultrasound-detectable liver fibrosis is associated with S. japonicum infection status and intensity, controlling for age, sex and year of ultrasound examination, and this measure can be used to monitor S. japonicum induced morbidity. Other ultrasound measures including hepatomegaly and splenomegaly were less clearly associated with infection. Our findings also suggest that some morbidity may not reverse within five years of treatment and may even progress despite treatment. Praziquantel has yielded remarkable declines in schistosomiasis morbidity in China and throughout the globe but reinfection following treatment is common as observed in this study population. S. japonicum infection, particularly high egg loads, may lead to fibrosis that is not rapidly reversed by treatment, underscoring the importance of measures to prevent new infections as well as treating current disease.
|
10.1371/journal.pgen.1002099 | A Novel Approach to Selectively Target Neuronal Subpopulations Reveals Genetic Pathways That Regulate Tangential Migration in the Vertebrate Hindbrain | Vertebrate genes often play functionally distinct roles in different subsets of cells; however, tools to study the cell-specific function of gene products are poorly developed. Therefore, we have established a novel mouse model that enables the visualization and manipulation of defined subpopulations of neurons. To demonstrate the power of our system, we dissected genetic cascades in which Pax6 is central to control tangentially migrating neurons of the mouse brainstem. Several Pax6 downstream genes were identified and their function was analyzed by over-expression and knock-down experiments. One of these, Pou4f2, induces a prolonged midline arrest of growth cones to influence the proportion of ipsilaterally versus contralaterally settling neurons. These results demonstrate that our approach serves as a versatile tool to study the function of genes involved in cell migration, axonal pathfinding, and patterning processes. Our model will also serve as a general tool to specifically over-express any gene in a defined subpopulation of neurons and should easily be adapted to a wide range of applications.
| In mammals, many genes execute a unique set of distinctive and common functions in different cell types. Strategies to address these individual roles often involve the generation of series of transgenic animals. Here, we present a novel approach that combines a single transgenic mouse line with tissue-specific transfection protocols and organotypic cultures to enable the quick analysis of numerous genes in a cell-specific manner. As a proof of principle, we analyzed the function of transcription factors in tangentially migrating neurons of the developing vertebrate hindbrain. We identified a temporary halt in migration as a novel mechanism for neurons to decide whether to cross or not cross the midline. Our model may serve as a general tool to quickly study axonal pathfinding, neuronal cell migration, or patterning processes in a well-defined population of neurons.
| Understanding cell-specific regulatory mechanisms is a major challenge in the post-genome era. Particularly in mammals, the reiterated usage of the same transcription factor in distinct subsets of cells or during distinct developmental time points provides the basis to generate thousands of individual cell types with a relatively small number of genes. A single transcription factor may therefore elicit variable downstream effects depending on the context of its expression. Tissue-specific knockout strategies based e.g. on the Cre-lox-system, or promoter-driven transgenic models allow a cell-specific manipulation of genes. However, as these techniques rely on the generation of new transgenic animals for each gene-combination analyzed they are laborious and time-consuming. Here, we combined a transgenic model with tissue-specific transfection protocols and organotypic cultures to enable the quick analysis of numerous genes in a cell-specific manner. As a proof of principle we applied our system to decode molecular pathways initiated by the transcription factor Pax6 which is involved in neuronal cell migration and axonal pathfinding processes.
Pax6, a homeodomain and paired domain containing transcription factor, is a major determinant of visual and olfactory sensory structures and is essential for a variety of patterning and pathfinding processes throughout the nervous system [1]–[3]. Depending on the context and area of expression Pax6 initiates varying downstream effects. Homozygous small eye (Pax6Sey/Sey) mouse and rat embryos, which lack functional Pax6, do neither generate eye nor nasal structures and are deficient in ventral diencephalic structures [4]–[9]. In the ventral hindbrain and spinal cord, Pax6 controls the dorso-ventral patterning of motorneurons and of interneurons [5], [10]. In the cerebral cortex Pax6 determines the neurogenic potential of radial glial cells [11], [12]. Throughout the developing nervous system, with the exception of the midbrain, Pax6 is expressed in a ventral and a dorsal pool of progenitor cells. Although the dorsal Pax6 expression domain has achieved much less attention than the ventral domain there is evidence that Pax6 plays a pivotal role in the specification and migration of neurons derived from this domain [13]–[16].
The dorsal domain of Pax6 positive neuronal precursors of the hindbrain includes the rhombic lip (RL) [14], [16] which comprises the interface between the dorsal neuroepithelium and the roof plate. The RL is the source of several tangentially migrating neurons (see also Figure 1A) [14], [17]–[24]. The most notable are the neurons of the marginal migratory stream (mms; also pes) which migrate from the rhombic lip circumferentially around the medulla towards their contralateral destinations to settle in the ECN (external cuneate nuclei) and the LRN (lateral reticular nuclei) [17], [25]. Owing to the superficial nature of the mms migration these neurons serve as paradigm to study neuronal migration and axonal pathfinding processes.
The highly complex neuronal circuits of the vertebrate nervous system are established during development when growing axons travel considerable distances towards their targets to generate the appropriate connections. This wiring process depends on attractive and repulsive factors which emanate from final or intermediate cellular targets and which are interpreted by cell surface receptors located on axonal growth cones [26], [27]. Although the general principles were uncovered during the past years our understanding of axonal pathfinding processes is far from being complete. Current methods to analyze candidates regulating neuronal migration and axonal navigation processes are laborious and often involve the generation of transgenic animals for each gene analyzed. Non-transgenic methods, as DiI labeling of neurons or vector-driven mis-expression of gene constructs, are suitable for use with certain applications, however, they are neither cell specific nor can they be targeted to distinct neuronal subpopulations.
Here we describe a novel transgenic mouse model, which allows the specific and exclusive visualization and manipulation of subsets of neurons in the developing brain. To demonstrate the power of this system we have analyzed the role of Pax6 in migrating neurons of the brainstem. In Pax6 mutant mice migration of these neurons is distorted and some neurons differentiate at ectopic positions. Using transplantation, knock-down and over-expression experiments we show that distinct migratory features are controlled by discrete sets of Pax6 downstream genes. These results demonstrate the potential of our transgenic mouse model as a tool to study the role of Pax6 in individual neurons. Moreover, our system should be widely applicable to study virtually any gene that acts during cell determination, axonal pathfinding and/ or cell migration processes.
The functional analysis of genes in restricted tissues often involves the generation of inducible knockout mice or mice over expressing transgenic constructs. To simplify this time-consuming process we developed an in vitro model that enables the visualization and manipulation of defined populations of neurons. To label neurons in a largely unlabelled background we searched for genes that were expressed in only a subset of neuronal precursors and in migrating neurons. Pax6 meets these criteria ideally. Pax6 is expressed in several groups of tangentially migrating neurons and their precursors as well as in a small population of radially migrating neurons and their precursors (Figure 1A–1C) [5], [10], [12], [14], [16], [28].
We adopted the Tet binary system [29] and generated YAC (yeast artificial chromosome) transgenic mice which expressed the tetracycline dependent transactivator (tTA) in all Pax6 positive cells. A 420 kb YAC spanning the human PAX6 locus (Y593) [30] was modified such that the PAX6 coding region was replaced with a cassette containing an IRES (internal ribosomal entry site) and the tTA (Figure 1D). Previously, we and others had shown that the unmodified YAC Y593 contains all elements driving full functional PAX6 expression [30]–[32] and, in agreement with this, Tg(PAX6-tTA) mice showed a wide overlap of tTA and endogenous murine Pax6 expression (Figure S1). Tg(PAX6-tTA) mice were entirely normal and control experiments insured that neuronal patterning and migration was unaltered.
tTA is a transcriptional activator that at moderate levels of expression is completely inert in vertebrates, yet, enables the activation of artificial constructs containing a tTA-DNA-binding element (TRE = tetracyline responsive element). To examine whether our transgenic model specifically allows the labeling of only Pax6 positive cells we introduced by electroporation reporter constructs driving the green fluorescent protein into transgenic embryos. In all instances only Pax6 positive cells, e.g. retinal precursor cells, cortical precursors, or cerebellar granule cells, expressed the reporter genes (Figure S1). Non-transgenic embryos or Pax6 negative tissues did not induce reporter gene expression (Figure 1H, Figure S1). Together these results demonstrate that Tg(PAX6-tTA) mice enable the targeting of reporter gene constructs specifically to Pax6 positive cells and tissues during development.
As Tg(PAX6-tTA) mice allow any gene to be targeted to Pax6 expressing cells, they are of potential value to study neuronal migration and axonal pathfinding processes and for the analysis of Pax6 downstream effects. As a proof of principle, we chose to focus on the marginal migration stream (mms). Like other tangentially migrating neurons, mms neurons use the same or similar navigational cues as do growing axons, and migration of mms neurons is severely disturbed in Pax6 mutant Pax6Sey/Sey mice (see below). Neurons of the mms are generated at the rhombic lip and migrate circumferentially around the embryonic brainstem to generate the contralateral lateral reticular (LRN) and the external cuneate (ECN) nuclei (Figure 1A) [16], [21]–[23], [25]. Migration starts at E13.0 and is completed by E16.5. Pax6 is expressed in precursors at the rhombic lip, in all migrating neurons of the mms and during the initial period of settling in the target nuclei (Figure 1B, 1C and data not shown). Antibody staining and in situ hybridization (not shown) of Tg(PAX6-tTA) mice confirmed a complete overlap of Pax6 and tTA expression in these neurons (Figure 1I).
To visualize migrating mms neurons in Tg(PAX6-tTA) mice, reporter constructs were introduced into neuronal precursor cells in the left rhombic lip by electroporation at E12.5 before migration had begun (Figure 1E). Whole brainstems including the cerebellar primordium were then sustained in organotypic filter cultures for up to 14 days as an open book preparation which allowed the observation of migrating neurons with a fluorescence microscope from above (Figure 1F, 1G). Our approach to use a binary system ensured that only Pax6 positive neurons containing tTA and a TRE reporter construct expressed the desired reporter genes. This procedure resulted in the specific labeling of mms neurons originating only from one rhombic lip. Pax6 positive neurons originating from the opposite rhombic lip remained unlabelled as were Pax6 negative (and therefore tTA negative) neurons originating from regions close to the rhombic lip. Unlabelled neurons included neurons of the submarginal migration stream (sms) which generate the inferior olive (IO) thus demonstrating the specificity of our model. To allow the simultaneous visualization and manipulation of neurons we designed reporter constructs containing two TRE elements (Figure 1J). Control constructs co-expressed a cytoplasmic green fluorescent protein (EGFPm) and a nuclear red fluorescent protein (DsRed2nls) in 99% (±1; n = 10) of labeled neurons demonstrating that our reporter constructs enable the co-expression of two genes in the same neurons (Figure 1J). To enable statistical analysis of the cultures, the territories of the LRN and the ECN were delineated using visible landmarks (Figure 1K, 1K′, 1L; see also the Materials and Methods section). Immunolabeling of cultures expressing a HA-tagged Pax6 construct demonstrate that over-expression of TRE constructs in Tg(PAX6-tTA) transgenic cultures result in moderate levels of protein expression that are in the range of physiological Pax6 concentrations (Figure 1M, 1M′, 1M'').
Pax6 mutant Pax6Sey/Sey mice display multiple neuronal patterning and migration defects. We therefore wished to determine whether Pax6 also regulates the mms. At the anatomical level, several features of the mms are severely disturbed in Pax6 mutant Pax6Sey/Sey embryos. Most noticeable, the initiation of migration and the midline crossing was delayed by 0.5 days (asterisks in Figure 2A, 2A′ and data not shown; see also Figure S2 and Figure 4A, 4A′ The expression patterns of Pax2, Dcx, NK1R, and DopH was unaltered indicating that there is no general developmental delay in the mutant brainstem (data not shown). In Pax6Sey/Sey embryos some migrating mms neurons used a sub-marginal instead of a marginal migration path (black arrowhead in Figure 2A′; see also Figure S2 and Figure 4A′) and at E14.5 a large number of mutant neurons accumulated around the midline suggesting a reduced pace in midline crossing (black arrow in Figure 2B, 2B′). Furthermore, a subset of Pax6 positive neurons migrated along the midline into the parenchyma of the hindbrain (white arrowheads in Figure 2C, 2C′). We used a variety of markers, e.g. antibodies against the potassium channel Kcnj6, and DiI tracing of mossy fiber projections to discriminate between mms neurons (generating the ECN and LRN) and neurons of the sms (generating the IO). These experiments all indicated a complete loss of neurons in the ECN and a disorganized settling of neurons in the LRN (Figure 2D, 2D′, 2E, 2E′; Figure S2, and data not shown). Many Kcnj6 positive neurons were even observed within the inferior olivary territory (black arrows in Figure 2E′) and dorsally to the IO at the midline (open arrow in Figure 2E′). In agreement with previous reports [16], we found a slight enlargement of the IO at E14.5 when we used the Ets transcription factor Etv1 as a IO specific marker [33] (Figure S3 and data not shown). However, by labeling for the axon-guidance-molecule B (RgmB) no alterations in the general architecture of the IO were seen [34] (Figure S3). This is consistent with our observation that misguided Pax6Sey/Sey mms neurons make only a negligible contribution to the IO or settle in the periphery of the IO. In summary, these data demonstrate that migration of Pax6Sey/Sey mms neurons is severely disrupted. Mutant neurons of the mms are initially delayed. Later, a number of neurons use a sub-marginal migration path, migration is disturbed at the midline and several neurons migrate to ectopic positions along the midline. Lastly, the normal structure of the LRN is lost, the ECN is completely missing, and the IO is enlarged.
In order to dissect the complex neuronal cell migration defects observed in Pax6Sey/Sey mice, Pax6Sey/Sey mice were crossed to the Tg(PAX6-tTA) transgenic line. Comparison of cultures obtained from wt and from Pax6Sey/Sey embryos confirmed the anatomical observations described above. Pax6Sey/Sey mms cells showed an initial delay in the onset of the migration (Figure 3A, 3B) and a disturbance at midline crossing later on. After 5 DIV (5 days in vitro) Pax6Sey/Sey mms neurons settled randomly in the LRN (Figure 3B′) but failed to form any ECN structures (Figure 3B''). In contrast, wt cultures formed a well organized LRN in which cells settled in a dorsal and ventral sub-nucleus of the LRN (Figure 3A′) and in a distinguished ECN (Figure 3A''). To quantify the effect we counted labeled cells in cultures from wt (LRNi = ipsilateral LRN = 112±7; LRNc = contralateral LRN = 266±16; ECN = 86±14; total number of cells = 610±34; n = 50) and Pax6Sey/Sey embryos (LRNi = 313±23; LRNc = 426±46; ECN = 8±2; total number of cells = 670±71; n = 8) (Figure 3K). These data suggested that there were no alterations in the gross number of migrating neurons between wt and Pax6Sey/Sey embryos and confirmed the complete absence of an ECN in Pax6Sey/Sey embryos. In both, mutant and wt tissues, a proportion of LRN neurons settled ipsilaterally (Figure 3K). All LRN neurons, however, projected to the contralateral cerebellum in respect to their origin from one rhombic lip, explaining the observations by Bourrat and Sotelo of an ipsilateral and contralateral contribution of mossy fibers [17].
The Pax6Sey/Sey migration defect could be caused either by a direct cell autonomous action of Pax6 in migrating RL precursors or via an indirect non-autonomous effect, for example in the ventral domain of Pax6expression (e.g. by altering migration cues at the midline). We performed three types of experiments to discriminate between these alternatives. First, we transplanted transfected Pax6Sey/Sey rhombic lips onto wt brainstems and vice versa (Figure 3C, 3D). Unexpectedly, migrating Pax6Sey/Sey neurons (in a wt host) formed a well organized LRN (Figure 3C′), but no ECN (Figure 3C''). In contrast, wt neurons (in a mutant host) failed to form a correctly organized LRN (Figure 3D′), but were able to generate a normal ECN (Figure 3D''). These data suggested, that Pax6 may act cell autonomously in generating ECN neurons, but non-autonomously in specifying the correct sub-organization of LRN neurons. To further validate this assumption we rescued the Pax6Sey/Sey migration defect by re-expression of Pax6. We tested the two major splice variants and of these, the expression of the Pax6(-5a) isoform in PaxSey/Sey6 rhombic lips resulted in a full recovery of the ECN (Figure 3E''), but a disorganized LRN (Figure 3E′), whereas, the Pax6(+5a) variant was ineffective (not shown). Thus, in the RL Pax6 splice variants differ in their biological activity, similar to the embryonic cortex [35]. Lastly, we diminished the endogenous Pax6 mRNA by using siRNAs. Transfection of siRNAs or over expression of shRNA constructs directed against Pax6 (Figure S4) resulted in a massive reduction of ECN cells in wt explants (Figure 3F''), whereas, control constructs had no effect (not shown). The effect was quantified by counting labeled cells that had settled in the ECN (Figure 3L). Taken together, these experiments demonstrate that our model system enables the simultaneous visualization and manipulation of tangentially migrating cells in the mouse brainstem. In addition, we have shown that Pax6 plays numerous distinct roles in the formation and migration of mossy fiber producing neurons. Moreover, the combination of a binary model and organotypic culture assays facilitates a quick discrimination between cell-autonomous and non-autonomous effects.
We identified several genes whose expression was altered in Pax6Sey/Sey mms neurons (Figure S5.; see also Materials and Methods). To gain more insights into the function of these putative Pax6 downstream targets all genes were over-expressed or their expression level was diminished with shRNAs. Those genes which showed the most noticeable effects are summarized in Table 1.
The altered migration and settling behavior of Pax6Sey/Sey ECN/LRN neurons suggested that migration cues were changed in Pax6Sey/Sey embryos. The most prominent candidates are ligand/ receptor couples of the Slit/ Robo- and Netrin/ Dcc- pathways [36], [37]. Expression of Netrin1, Dcc, and Robo1,2 and 3 was unaltered in migrating Pax6Sey/Sey mms neurons (Figure 4A, 4A′, Figure S2, and data not shown). However, Slit1 and Slit2 which were expressed in the hypoglossal nuclei were both lost in Pax6Sey/Sey embryos (Figure 4B, 4B′ and Figure S2) [5], [10]. Motorneurons of the hypoglossal nuclei are in close proximity to the LRN settling territories suggesting that Slit1 and Slit2 expression provided from these neurons may determine the place of LRN settlement. To test this hypothesis we performed transplantation experiments and shRNA driven knock-down of the Slit-receptor Robo3 in migrating mms neurons. Both types of experiments resulted in a disorganized LRN similar to the phenotype observed in Pax6Sey/Sey mice (Figure 4C-4H). The above results indicate that factors provided from the hypoglossal nucleus, (most likely Slit1 and Slit2) determine the place of LRN settlement. These data also explain the cell non-autonomous role of Pax6 during this process. Hypoglossal neurons are Pax6 negative, but are completely lost in Pax6Sey/Sey embryos (Figure S2) [5], [10]; hence, Slit1 and Slit2 are most likely not direct targets of Pax6. Additional experiments suggest that Slit1 and Slit2 may also act as repellent to push mms neurons to the marginal migration route during the initial phase of migration (data not shown).
Two POU transcription factors were among the genes whose expression pattern was altered in the mms of Pax6Sey/Sey embryos. Pou4f2 (also: Brn3b) was strongly expressed in about 18.6% (±4.6%, n = 3) of E14.5 and 23.3% (±6.2%, n = 3) of E15.5 wt mms neurons but was completely lost in the Pax6Sey/Sey mms (Figure 5A, 5A′, 5B, 5B′). Pou4f1 (also: Brn3a) was expressed between E13.5 and E15.5 in a subset of mms neurons, but was up-regulated in the E14.5 and E15.5 Pax6Sey/Sey mms (Figure 5C, 5C′). Expression of Pou4f1 and Pou4f2 in Pax6Sey/Sey IO neurons was unaltered (Figure 5A, 5A′, 5C, 5C′). Pou4f2 plays several roles in specifying and guiding retinal ganglion cells and their axons. We therefore asked whether Pou4f2 may accomplish similar tasks in rhombic lip derived neurons. Pou4f2 was only expressed in a subset of wt mms neurons. We therefore over-expressed Pou4f2 in all migrating mms neurons. Remarkably, growth cones of all Pou4f2 over-expressing neurons were arrested at the midline for about 1.5 days (±0.5 days, n = 17), whereas the majority axons in control cultures crossed the midline instantly (Figure 5D, 5F). Interestingly, in control cultures the growth cones of some neurons also appeared to be arrested at the midline: 5% (±3%) at 1DIV, 15% (±6%) at 2DIV, 25% (±5%) at 3DIV, and 6% (±3%) at 4DIV (n = 11). This correlates well to the peak of Pou4f2 expression at E14.4 and E15.5 (in cultures: 2DIV and 3DIV). Over-expression of Pou4f2 had also a noticeable effect on the settling behavior of LRN neurons. Quantification of LRN neurons at 5DIV revealed that Pou4f2 expressing LRN neurons preferably settled at the ispilateral side (LRNc/LRNi = 0.8 ± 0.1, n = 17; Figure 5G, 5M) compared to control cultures in which the majority of LRN neurons settled at the contralateral side (LRNc/LRNi = 2.5 ±0.1, n = 50; Figure 5E, 5M). Similar relations were obtained at 6DIV and 8DIV suggesting that Pou4f2 over-expression altered the migration behavior of mms neurons and did not cause a delayed settlement of these neurons. The effect was specific to Pou4f2 and could not be mimicked by over-expression of Pou4f1, Pou4f3 or Pou6f1 (Figure 5M and data not shown). Together these data suggest that Pou4f2 acts through a novel mechanism which induces an arrest of growth cones at the midline to regulate the ratio of ipsilaterally versus contralaterally settling neurons.
We altered expression levels of about 25 potential Pou4f2 retinal target genes [38]–[40] and of these two showed an effect on the migration behavior of mms neurons. Over-expression of Gfi1, a zinc finger transcription factor, reduced the contra-/ipsi-lateral ratio of LRN neurons (Figure 5M). In contrast, the down-regulation of Gap43 by shRNA constructs caused a higher contra-/ipsi-lateral ratio of LRN neurons (Figure 5M and Figure S4). Gap43 is slightly reduced in the Pax6Sey/Sey mms (Figure S5). In addition, mis-expression of Pou4f2 resulted in a massive down-regulation of Pou4f1 in transfected, but not in control, rhombic lips (Figure 5N), suggesting that the loss of Pou4f2 in Pax6Sey/Sey mms neurons leads to an up-regulation of Pou4f1 (Figure 5C, 5C′). Pou4f1 over-expression or down-regulation, however, did not alter migration behavior of mms neurons (Figure 5M).
Pou4f2 is expressed only in a subset of Pax6 positive mms neurons suggesting that other factors together with Pax6 may co-regulate Pou4f2. In the developing retina Pou4f2 expression depends on two transcription factors: the bHLH protein Math5 and the zinc finger gene Wt1 [41]–[44]. Wt1 was found to be expressed in the rhombic lip, though, in a region just dorsally to the Pax6 positive domain (Figure 5J). Math5 was neither expressed in the rhombic lip nor in migrating mms neurons, however, a close homologue, Math1, was expressed in neuronal precursors at the rhombic lip and in a subset of early migrating mms neurons [22], [23] (Figure 5K). Thus, Math1, but neither Math5 nor Wt1, was the most likely candidate to regulate Pou4f2 or Pou4f1 expression in mms neurons. Consistent with this, mis-expression of Math1, but not of Wt1 (+ and – KTS splice variants) or Math5, led to a midline arrest of migrating mms neurons and a reversed settling behavior of LRN neurons (Figure 5H, 5I, 5M). Mis-expression of Math1 resulted in an up-regulation of Pou4f2 and a down-regulation of Pou4f1 (Figure 5O, 5P). Together, these data suggest, that Pou4f2 expression in rhombic lip derived mms neurons depends on Pax6 and Math1 and that Pou4f2 may regulate Pou4f1 and Gap43 in mms neurons.
In summary, our work led to the identification of a gene cascade acting in tangentially migrating neurons of the brainstem, in which Pou4f2 plays a central role to induce a previously unknown mechanism that controls midline crossing behavior. Furthermore, our results imply that our model system is applicable to quickly analyze genetic hierarchies in Pax6 positive cells and may therefore serve as a general tool.
The extraordinary complexity of cell determination, migration and wiring processes in the developing mammalian brain creates a major challenge for developmental neurobiologists. Here, we introduced a simple yet powerful technology to quickly analyze any gene potentially involved in these processes. Our model is of threefold use: first to study the function of Pax6 and of Pax6 downstream genes in their genuine environment, second to investigate genes involved in general patterning, axonal pathfinding and cell migration processes, and third to enable the analysis of tissue-specific gene functions. The Tg(PAX6-tTA) model complements and improves existing approaches and has certain benefits: it combines cell specific transfection protocols and organotypic culture assays, thus, facilitating the quick analysis of genes in a natural tissue environment. The experimental design and the binary nature of the Tg(PAX6-tTA) model is fundamentally simple and has several advantages over systems that are based purely on transgenic animals. First, the electroporation and subsequent culture of embryonic tissues allows the screening of large number of genes without the need of generating new transgenic animals for each construct. In fact, less than 10% of the constructs we have tested revealed phenotypes. Thus, only those genes showing positive results in culture assays may be used subsequently to generate transgenic lines. Of note however, some of the phenotypes reported here, for example, the midline arrest or the altered ipsi- to contra-lateral ratio, would have been missed in purely transgenic systems. Second, variation of the electroporation protocol allows transfections ranging from just a few cells to a complete Pax6 expression domain with thousands of cells. Hence, our approach allows adjustment according to the needs: either to monitor single migrating cells or to determine global patterning effects. In addition, neighboring cells and non-electroporated contra-lateral sides serve as internal controls. The usefulness of our system critically depends on the tightness of the TRE based promoter and on the ability of the constructs to express two genes simultaneously. To ascertain the tightness of our system we used repeated electroporations and high DNA concentrations (up to 5 µg/µl). Even under these extreme conditions we were never able to detect any reporter gene expression in Pax6 negative cells at any developmental stage. Thus, under the conditions used in this report the combination of Tg(PAX6-tTA) mice and TRE based promoters allow expression of reporter gene constructs only in Pax6 positive cells. It is also important to note, that our strategy to use a YAC based technology combined with an internal ribosomal entry site (IRES) resulted in moderate levels of reporter gene expression which were in the range of physiological concentrations. To ensure the simultaneous expression of two reporter genes we tested several types of TRE constructs. Only our approach, to use two consecutive TRE based promoters led to the activation of nearly equal amounts of two genes at the same time in the same cell. A bidirectional TRE element that previously had been shown to work in transgenic animals failed in our system [45]. One obvious difference is that in transgenic animals typically multiple copies of constructs are stably integrated into the genome, whereas, in our assay transfections were transient.
Pax6 loss of function phenotypes are often highly complex involving massive malformations in the affected organs. Pax6 is expressed in neuronal precursors of the telencephalon, commissural neurons in the dorsal spinal cord, in adult neuronal stem cells, the early eye cup, in the pancreas, in precursors and in migrating cells of several tangential and radial migration streams of the rhombencephalon and of the forebrain [5], [6], [10], [11], [14], [28], [46], [47]. In addition to its technical advances, the Tg(PAX6-tTA) model represents a novel, highly versatile technology to study the function of Pax6 or any other gene in these tissues. As a paradigm, we have dissected the role of Pax6 in tangentially migrating cells of the brainstem. In principle, however, this system shall be applicable to any Pax6 positive tissue and we have initial evidence that our model allows to specifically target Pax6 positive telencephalic precursor cells, cerebellar granule cells, the developing retina, the rostral migratory stream, the pontine migration and ventral precursor cells of the brainstem and spinal cord (Figure S1 and data not shown). With the help of this model it should therefore be possible to systematically analyze cell fate decisions and the migratory behavior of Pax6 expressing cells at any developmental stage.
Several studies have revealed that Pax6 is required for hindbrain and spinal cord development [5], [7], [10], [14], [15]. Our work adds that Pax6 also controls the determination and migration of rhombic lip derived neurons (for a summary see Table 1 and Figure 6). Pax6 functions twofold: first, Pax6 controls guidance cues which push migrating mms neurons to the marginal path and which control the settling pattern of LRN neurons. The most likely sources of these cues are the hypoglossal nuclei which are located close to the midline and in proximity to the LRN. Slit1 and Slit2 are expressed in the hypoglossal nuclei and the Slit receptor Robo3 is expressed in migrating mms neurons [48]. Slit expression provided by the hypoglossal nuclei may therefore act as repellent to push mms neurons to a marginal migration route and may also specify the settlement of neurons in the LRN. The loss of Slit-expressing hypoglossal nuclei in Pax6Sey/Sey embryos [5], [10] causes a major reduction of the repellent (a minor source of Slit is still present in midline cells). Consequently, migrating mms neurons would use a more sub-marginal migration route and settle less organized in Pax6Sey/Sey embryos. Furthermore, Slit expression at the RL may be involved during the initial phase of mms migration. Secondly, Pax6 functions cell-autonomously in migrating mms neurons to control the determination, the timing of migration, and midline crossing. Several genes show altered expression in Pax6Sey/Sey mms neurons (Table 1: Pou4f1, Pou4f2, Unc5h1, Mafb, Chordin) and may convey individual aspects of migration.
We and others find that several transcription factors relay Pax6 downstream effects in dorsal brainstem neurons: Ngn1 in precursor cells ventral to the RL [16], and Pou4f1, and Pou4f2 in migrating neurons (this report). Mis-expression of Ngn1 or Ngn2 in Pax6Sey/Sey embryos failed to rescue the migration defects observed in the Pax6 mutant (Table 1). Neither did the mis-expression or down-regulation of these genes generate small eye - like migration defects in wt embryos (Table 1). On the other hand, Pou4f2, which is lost in the mms of Pax6Sey/Sey embryos (Figure 5B and also in the pontine migration and in the cerebellum, data not shown), alters migration behavior of mms neurons. Together these data suggest that Pou4f2 may regulate genes involved in pathfinding processes, whereas, Ngn1 acts earlier in the cell determination process.
There are striking similarities in gene expression pattern between sensory neurons and RL derived neurons. We found that at least two thirds of the genes which are co-expressed with Pax6 and Pou4f2 in retinal ganglion cells are also co-expressed with these genes in mms neurons. Furthermore, genetic hierarchies seem to be analogous: in the retina Math5 controls Pou4f2, which then acts upstream of Pou4f1 [38], [41]–[43], whereas, in RL derived neurons Math1, a close homologue of Math5, initiates related pathways. General genetic pathways are conserved between retinal and RL derived neurons and our model may therefore help to elucidate some of the phenotypes observed in Pou4f1-/- and Pou4f2-/- mice. Both mouse models have revealed distinct axonal pathfinding errors [39], [49]–[52]. Mis-expression of Pou4f2 (or Math1) in RL derived neurons stalls growth cones at the midline for several hours. To our knowledge, this is the first report of such a midline arrest and it may thereby be a paradigm for a novel mechanism controlling midline crossing. The arrest does neither induce a growth cone collapse nor does it inhibit midline crossing per se as all neurons generate axons that cross the midline after a “waiting period”. These axons all migrated into the cerebellum like those of control cultures. Gfi1 mis-expression and Gap43 knockdown were able to partially mimic the Pou4f2 induced phenotype, however, additional unknown targets or a combinatory code may be needed to elicit the full phenotype.
As Pou4f2 was only expressed in about 1/4 of wt mms neurons, the loss of Pou4f2 in Pax6Sey/Sey embryos mimics only minor aspects of the Pax6Sey/Sey phenotype. The down regulation of Pou4f2 by shRNA constructs resulted in a severe midline disturbance of neuronal processes at similar to the phenotype observed in Pax6Sey/Sey cultures, whereas, in control cultures neuronal processes crossed the midline instantly (data not shown). Comparable phenotypes were also observed in Pax6Sey/Sey cultures, in cultures transfected with Pax6 shRNA constructs, and in Pax6Sey/Sey brainstem sections.
Tangentially migrating neurons follow similar navigational cues as developing axons [14], [19], [48], [53]–[58]. Hence, tangentially migrating neurons of the mms provide an excellent system to study axonal pathfinding and neuronal cell migration processes. Migrating mms neurons are easily accessible as they navigate along the pial surface. Our model takes advantage of the superficial migration of these neurons and provides a straightforward assay to specifically label and manipulate these cells without affecting their surroundings. Members of most families of guidance receptors (Netrin receptors, Slit receptors, Semaphorins, Eph receptors, and Ephrins) are expressed in migrating mms neurons [48], [53], [54] (Engelkamp, unpublished) and at least two of these pathways are essential for the correct guidance of mms neurons: the Slit/ Robo - [48] and the Netrin/ Dcc- pathways [53], [56], [57] (see also Table 1). Our system should therefore also have important implications for the study of the signal cascades entailed in these pathways.
In summary, we have established a novel model system which allows the simultaneous visualization and manipulation of neuronal subpopulations. As a prototypical model we have focused on the role of Pax6 in migrating brainstem neurons. Yet, our results imply that our model system is applicable to a range of other cells in the developing brain and may therefore serve as a general tool to quickly study axonal pathfinding, neuronal cell migration or patterning processes.
The Small Eye allele [4] was maintained on a CD1 background. Embryos were obtained from matings of heterozygote (Pax6Sey/+) mice. 0.5 denotes the morning when the vaginal plug was found. Experiments were always performed on matching pairs of control (wt) and Pax6Sey/Sey embryos that were carefully staged. All phenotypes described were confirmed on at least six individual Pax6Sey/Sey embryos obtained from different crossings. There was no noticeable phenotypic difference between Pax6Sey/+ and wt embryos and therefore, in our experiments wt designates wt and Pax6Sey/+ embryos. For brainstem cultures, matings between heterozygote Tg(PAX6-tTA) and wt CD1 mice or between heterozygote Pax6Sey/+/ Tg(PAX6-tTA) and heterozygote Pax6Sey/+ mice (to generate Pax6Sey/Sey cultures) were used. Genotyping was performed by PCR with primers directed against the Tet repressor (upper: GCGCTGTGGGGCATTTTACTTTAG; lower: CCGCCAGCCCCGCCTCTTC). All animal procedures were carried out in accordance to the guideline approved by institutional protocols.
YAC Y593 [30] was modified such that exons 8 to 11 of the Pax6 gene were replaced by homologous recombination with a construct containing the following elements in 5′ to 3′ order: Pax6k30 – IRES – tTA – loxP – LYS2 – loxP – Pax6k32. Pax6k30 and Pax6k32 corresponded to the sequences 29.792 to 30.296 and 31.587 to 32.095 of the Pax6 cosmid cFAT5 (NCBI accession no. Z95332), respectively, and were generated via PCR. The IRES (internal ribosomal entry site) was derived from pIRES-EGFP (Invitrogen), however, the original ATG-11 start codon was reconstituted to enhance translational initiation. tTA (Tet-On-system), a fusion of the tetracycline repressor and the activation domain of VP16, was derived from pUHD15-1neo (Clontech). The LYS2 gene from S. cerevisiae was derived from pAF107, which was obtained from B. Dujon, Institute Pasteur, Paris, France [59]. LoxP sequences were generated via PCR. All constructs were sequence verified. Homologous recombination in yeast was performed using standard techniques. The integrity of the recombined YAC was then verified by PCR and southern blotting. Preparation of the YAC DNA and the generation of transgenic mice were as described [30].
In situ hybridization was performed on free floating vibratome sections as previously described [14]. Probes for Math1 [60], Neurod1 and Neurod2 [61], Pax6 [47], Unc5h3 [62] and Slit1, Slit2 [63] were obtained from H. Zoghbi, A. Bartholomä, R. Hill, S. Ackerman, and M. Little, respectively. Probes for Dcc and RgmB were as published [34], [64]; other probes were obtained by RT-PCR. The PCR products were subcloned and their identities were confirmed by sequencing. The general staining patterns of all probes matched published expression patterns. Probes were as follows: Etv1 (bp 853–1820 of NM_007960); Fgfr2 (bp 343–1192 of NM_201601); Pou4f1 (bp 1321–2199 of NM_011143), Pou4f2 (bp 216 – 1762 of S68377); Robo3 (bp 3648–4673 of NM_011248). Several genes which are down- or up-regulated in Pax6Sey/Sey embryos were identified with the help of a large scale in situ screen using >300 putative candidates. Individual probes are available on request.
The vector for the co-expression of two constructs in Tg(PAX6-tTA) mice contained the following elements in 5′ to 3′ order: MCSI – TRE – PminCMV – IntronA - BGHPolyA – MCSII – TRE - SV40PolyA; MCS = multiple cloning sites; TRE = 7 repeats of the tetracycline responsive element, PminCMV = minimal CMV promoter, and IntronA were from ptetOi-MCS (obtained from Martin Spiegel, Tübingen); SV40polyA and BGHPolyA = polyadenylation signals (derived from pTetOi-MCS and pRc/CMV, Invitrogen, respectively). Fluorescent markers to label migrating cells were a modified EGFP or DsRed2 (Clontech). Full length clones for Gfi1, Math1, Math5, Ngn1, and Ngn2 were obtained from the German Resource Center for Genome Research (RZPD) and sequence verified; clones for all other genes were obtained by RT-PCR and confirmed by sequencing. Fusions with a triple HA-tag or the engrailed repressor domain (EnR) were generated by PCR. shRNA constructs were generated in the psiSTRIKE vector (Promega) using the Promega Web tool for designing the hairpin oligonucleotides. In the psiSTRIKE vector shRNAs are expressed under control of the U6 RNA polymerase promoter. Efficiency of shRNA knockdown was demonstrated in HEK293 cells using the psiCHECK/ Dual Luciferase system according to the manufacturers protocol (Promega). All constructs were sequence verified.
Responder constructs (2-4 µl at 0.5 µg/µl in GBSS/ 0.01% Methyl Fast, Sigma) were injected into the fourth ventricle of E12.5 wt and Tg(PAX6-tTA) mouse embryos by using glass needles. Electroporation was then performed with forceps-like electrodes with platinum ending (Ø = 0.5 mm) (one Electrode above the right RL and the other under the left jaw). Conditions were 8 pulses at 50V, 50msec with a pulse interval of 1sec. We used the square pulse generator EPI2500 (L. Fischer, Heidelberg). After electroporation, the hindbrain (rhombomeres 1–8 including the cerebellar anlage) was dissected, opened at the roof plate and cultivated with the ventricular site onto MillicellCM filters (Millipore) in culture medium (DMEM/F12 (1∶1); 0.6% Glucose; 0.02 mM Glutamine; 5 mM HEPES; 5% Fetal Calf Serum; 5% Horse Serum) at 37°C and 5%CO2.
Depending on the antibodies used, brainstem preparations were fixed with 4% or 0.2% PFA in PBS for 12 hours at 4°C. Cryosections were cut at 14 µm. Primary and secondary antibodies used for staining were as follows: mouse monoclonal (mAb) α-Pax6 ([10], 1∶1000, DSHB); rabbit pAb α-Pou4f2 (also Brn3b, 1∶300, Covance); rabbit pAb α-Kcnj6 (also Girk2, 1∶300, Chemicon); rabbit α-Wt1 (Santa Cruz); rabbit α-VP16 (Clontech); α-HA-tag (1∶100, Roche) and α-mouse and α-rabbit secondary antibodies conjugated with Alexa488 or Alexa596 (Molecular Probes). Quantification of Pou4f2 positive mms neurons was done on every 3rd of serial sections double stained for Pax6 and Pou4f2.
Images were taken at a Zeiss Axiophot microscope equipped with a Spot camera, at a confocal Zeiss LSM microscope, or at a Leica MZ12 equipped with a camera device. Images were processed using the MetaView software (Universal Imaging Corporation) and Adobe Photoshop. To perform statistical analysis the position of the ECN and the LRN were determined in wt un-manipulated cultures by in situ RNA staining of Pax6 and Kcnj6. The resulting territories were then overlaid onto the electroporated cultures with the help of three landmarks: a) the position of the rhombic lip; b) the position of the floor plate; and c) the position of the superior and inferior olivary complexes, which both are visible in phase contrast images of the cultures. This procedure allowed classifying 97% of labeled neurons on the contralateral side and 90% on the ipsilateral side as either ECN or LRN neurons. The remaining 3% (or 10% for the ipsilateral side) of labeled cells were scattered neurons mainly in between the ECN and the LRN. Quantification of growth cones arrested at the midline in wt cultures was done by counting all growth cones in a 25 µm wide territory at the midline. Continuous observations of cultures implied that mms growth cones traveled at an average speed of at least 500 µm/day, suggesting that within any 25 µm interval only 5% of growth cones should be detected if migration would not pause. Quantification of the volume of the inferior olive was done with AxioVision (Zeiss).
|
10.1371/journal.ppat.1002936 | DNA Mutations Mediate Microevolution between Host-Adapted Forms of the Pathogenic Fungus Cryptococcus neoformans | The disease cryptococcosis, caused by the fungus Cryptococcus neoformans, is acquired directly from environmental exposure rather than transmitted person-to-person. One explanation for the pathogenicity of this species is that interactions with environmental predators select for virulence. However, co-incubation of C. neoformans with amoeba can cause a “switch” from the normal yeast morphology to a pseudohyphal form, enabling fungi to survive exposure to amoeba, yet conversely reducing virulence in mammalian models of cryptococcosis. Like other human pathogenic fungi, C. neoformans is capable of microevolutionary changes that influence the biology of the organism and outcome of the host-pathogen interaction. A yeast-pseudohyphal phenotypic switch also happens under in vitro conditions. Here, we demonstrate that this morphological switch, rather than being under epigenetic control, is controlled by DNA mutation since all pseudohyphal strains bear mutations within genes encoding components of the RAM pathway. High rates of isolation of pseudohyphal strains can be explained by the physical size of RAM pathway genes and a hypermutator phenotype of the strain used in phenotypic switching studies. Reversion to wild type yeast morphology in vitro or within a mammalian host can occur through different mechanisms, with one being counter-acting mutations. Infection of mice with RAM mutants reveals several outcomes: clearance of the infection, asymptomatic maintenance of the strains, or reversion to wild type forms and progression of disease. These findings demonstrate a key role of mutation events in microevolution to modulate the ability of a fungal pathogen to cause disease.
| Many diseases are contracted from the environment, rather than from sick people. It is unclear why those species are able to cause disease, since the selective pressures in the environment are presumed to be very different from those found within the host. Cryptococcus neoformans is a fungus that causes life-threatening lung and central nervous system disease in approximately one million people each year. The fungus is inhaled from environmental sources. One hypothesis to account for C. neoformans virulence is that amoeba are predators for this fungus, and surviving strains are pre-selected to be virulent in the human host. On the other hand, experiments have found that amoeba eat C. neoformans. A pseudohyphal cell type can survive, and while protecting against amoeba these cells are unable to cause disease in mouse models. We predicted that the pseudohyphal morphology reflected a change in function of a pathway of genes, and found that all pseudohyphal isolates contain mutations within genes for this pathway. The pseudohyphal trait is unstable, with reversion to normal yeast growth by counter-acting mutations. These mutations can occur during the course of mammalian infection. Our results show that mutation events account for a microevolution system currently described as phenotypic switching, and that mutations, at least under experimental conditions, can regulate pathogen adaptation and influence its host range.
| Pathogens across all major microbial groups – viruses, bacteria, fungi and protists – have representative species that owe their success to rapid change during infection or within a population. Microevolution is thus essential to pathogenesis, yet due to its stochastic nature it can be difficult to study and the underlying mechanisms challenging to elucidate.
Cryptococcus neoformans is a fungal pathogen that is acquired directly from the environment through inhalation of desiccated yeast cells or the sexual basidiospores. The fungus is found world wide and it causes disease predominantly in immunocompromised individuals, especially AIDS patients [1], [2]. The global mortality rate is estimated at 624,000 per annum [3]. The closely-related species C. gattii causes disease mostly in healthy individuals, and is responsible for an ongoing and expanding outbreak of cryptococcosis in the Pacific Northwest of Canada and the United States [4], [5], [6]. Both Cryptococcus species are extensively studied, have a suite of experimental resources, and serve as general models for understanding pathogenesis and its evolution in pathogenic eukaryotes [7], [8]. Cryptococcus species undergo microevolution in vitro, within animal models, and during the course of disease in humans [9], [10], [11], [12].
A current hypothesis is that the Cryptococcus species are pre-selected for virulence within mammalian animals because of interactions with predatory microbes like amoeba or nematodes [13], [14], [15], [16]. Evidence for this comes from studies on interactions with non-mammalian hosts. C. neoformans has been co-isolated with three different Acanthamoeba species [15], [17], [18] and these amoebae can take up the fungus by phagocytosis [19], [20]. Genes that are essential for mammalian virulence are also required for virulence in non-mammalian models [21]. Furthermore, screens of insertional mutants of the fungus with the nematode Caenorhabditis elegans identified fungal genes required for nematode viability: deletion of these genes also reduces virulence in mouse models of cryptococcosis [22], [23]. Additional support for the hypothesis comes from passage of C. neoformans through a slime mold host, Dictyostelium discoideum, since this produces strains with increased virulence in mice [24]. However, one caveat is that Acanthamoeba species ingest and kill C. neoformans. Surviving subpopulations can be isolated, including one common class that has pseudohyphal cells rather than the normal yeast shape [17]. The pseudohyphal strains were avirulent in animal models [17], [25], [26].
Strikingly, the pseudohyphal phenotype is not always stable. One of the eight pseudohyphal strains originally isolated, after wild type strains were exposed to amoeba, was inoculated into mice and it demonstrated wild type virulence [17]. Closer examination revealed, to quote directly, that “a high percentage of the cells in the inoculum of this isolate had reverted to the encapsulated yeast form” [17], a description of an unstable trait governed by epigenetics or microevolution. In repeat experiments, including use of different wild type strains of C. neoformans and a different species of Acanthamoeba, pseudohyphal isolates were again obtained: the phenotype exhibited instability in some, but not all, strain backgrounds [15], [25]. More recently, a similar pseudohyphal morphology was reported from in vitro experiments, and that it too could revert back to wild type at a high frequency (e.g. an average of 1 revertant per 1600 colonies) [27]. This high frequency of phenotypic change is referred to as phenotypic switching. The basis for this rapid evolution in C. neoformans remained unknown.
The pseudohyphal morphology of strains from amoeba or phenotypic switching appear similar to those of C. neoformans strains with loss-of-function mutations in the RAM/MOR pathway of genes [28]. This pathway is conserved in eukaryotes and characterized primarily in Saccharomyces cerevisiae where the abbreviation is from Regulation of Ace2p activity and cellular Morphogenesis [29], [30]. In Schizosaccharomyces pombe the pathway is known as MOR, for the Morphogenesis-related NDR kinase network. The pathway comprises six components, centered around the action of two protein kinases Cbk1 and Kic1 and the accessory proteins Sog2, Hym1 and Mob2. The large size and physical interaction of Tao3 with Cbk1 and Kic1 suggests that Tao3 may act as a scaffold protein [29], [31]. Within the fungi, the pathway can have dramatically different effects on cell type, for instance promoting cell polarity in the ascomycete yeast S. cerevisiae whereas inhibiting it in the basidiomycete yeast C. neoformans [28], [29]. It is required for virulence in plant and human pathogenic fungi [32], [33], [34], [35], and polymorphisms in KIC1 recently emerged from genome wide association studies between clinical and non-clinical isolates of S. cerevisiae [36]. The role of the RAM pathway in pathogenesis has been most thoroughly analyzed in Candida albicans, in which components mediate cell separation and polarity, and thus mutations block filamentation, impair biofilm formation and surface adhesion, and reduce virulence [34], [37], [38], [39], [40], [41], [42]. Mutation of the RAM pathway in C. neoformans causes a pseudohyphal morphology and other phenotypic changes [28], although effects on virulence had not been tested.
We hypothesized that the RAM pathway is affected in strains that “switch” morphology either in vitro or upon exposure to amoeba, and that the analysis of this process would provide insight into the mechanism of microevolution. In this study, we identify mutations within a RAM pathway gene in original pseudohyphal mutants derived from amoeba and phenotypic switching. We recapitulate the isolation of pseudohyphal strains by both means, and find that these strains all have mutations in the RAM pathway. A strain derived from amoeba with a point mutation in the MOB2 gene was unable to cause disease in a murine model, unless during the course of infection the mutation reverted to wild type. Switching back to yeast morphology relies on a multifactor system for microevolution that is also driven by DNA mutations. These findings demonstrate that DNA mutation contributes to fungal pathogenesis.
The C. neoformans species consists of two varieties, var. grubii (serotype A) and var. neoformans (serotype D). Pseudohyphal strains have been isolated from both varieties (table S1). Three isolates of C. neoformans var. grubii from the 1970s that originated after exposure to amoeba (strains C, D and E; ATCC 42343–42345 [17]) and one C. neoformans var. neoformans isolate from the 1990s that originated from a phenotypic switching study (strain F7 [27]) were compared to defined deletion strains of the RAM pathway genes. The strains had similar pseudohyphal cell morphologies distinguishing them from wild type yeast cells (Fig. 1A). Most strains showed decreased growth at mammalian body temperature (Fig. 1B). Consistently, all were highly sensitive to the immunosuppressive chemical FK506 (Fig. 1B). This drug inhibits the calcineurin pathway, the impairment of which is synergistically lethal with RAM pathway mutation in C. neoformans [28]. The similarity in phenotypes between the strains isolated from amoeba and phenotypic switching with defined deletion strains suggests the same genes or pathways are affected, and that this could be the RAM pathway.
We hypothesized that the historical pseudohyphal strains represent loss of the RAM pathway. The nature of the deficiency was sought in the strains that had been isolated from amoeba and the phenotypic switched isolate. Epigenetic or microevolutionary changes can arise from a suite of different causes. As a first approach towards gene identification, constructs containing wild type copies of four of the genes in the pathway (MOB2, CBK1, KIC1 and SOG2) were introduced into strains C, D, E and F7 with the endeavor to identify the affected gene by expressing additional copies. None of the four genes restored growth to the wild type yeast form, whereas these constructs complemented deletion strains. A fifth RAM pathway gene, HYM1, has not emerged as a RAM component in C. neoformans from random mutant screens, and has thus far eluded gene replacement experiments. The sixth component of the pathway, TAO3, is large and it was a challenge to generate a vector for this gene for transformation experiments. However, the lack of evidence for a direct role by the other members of the RAM pathway provoked closer examination of the TAO3 gene in the pseudohyphal strains.
The TAO3 gene was sequenced from the four historical pseudohyphal strains. The three strains isolated after exposure to amoeba (strains C, D and E) all contained an identical predicted g-t base pair substitution, that results in a codon for glutamic acid being substituted for a premature stop codon (Fig. 2A; dataset S1). The wild type strain for the three amoeba derived isolates is unknown, but is likely to be strain G (ATCC 42437) based on co-deposition of this isolate with C, D and E to the American Type Culture Collection. The TAO3 gene was sequenced from strain G to confirm that the stop codon is not present. The TAO3 gene from the strains C, D, E and G strains, with the exception of the stop codon in three isolates, was identical in sequence to the gene from C.n. var. grubii strain H99 sequenced by the Broad Institute (CNAG_03622). Sequence analysis of strain F7, the phenotypic switched strain, has an allele of TAO3 bearing a predicted a-t bp substitution that also causes a premature stop codon (Lys-Stop, Fig. 2B and dataset S1). The wild type parent for F7 is strain ATCC 24067A, and TAO3 was amplified and sequenced from this strain. The wild type strain does not contain the stop codon. Thus, the historical pseudohyphal strains derived from exposure to amoeba or phenotypic switching are tao3 mutants.
While the stop codons within the TAO3 gene are consistent with impaired RAM function, we aimed to demonstrate that these mutations cause the pseudohyphal phenotype. A Mendelian genetic segregation approach was first considered. However, strain F7 and the three amoeba-derived strains are infertile and thus genetic linkage tests were not feasible. Next, a construct was generated by overlap PCR that introduces a silent point mutation into the coding region adjacent to the Glu-stop codon, and at the same time introduces a new BglII restriction enzyme site (Fig. 3A, B). The rationale behind this approach was that the chance of two identical nucleotide changes occurring in independent strains is highly unlikely, and this construct could be used to distinguish reconstituted strains from reverted strains. The TAO3BglII construct was transformed into strain D using a biolistic apparatus that facilitates homologous gene replacement events. After transformation the cells were plated on FK506-containing media. Strains were isolated with the wild type yeast morphology, and when the TAO3 fragment was amplified and the PCR product digested with BglII, a subset of strains now contained a novel BglII restriction enzyme site (Fig. 3C, D). This indicates successful gene targeting and reconstitution to a wild type copy of the gene, and that the pseudohyphal phenotype is due to mutation of TAO3.
Additional pseudohyphal strains were sought in order to explore the basis of this trait and whether or not it was specific to the TAO3 gene. Two C.n. var. grubii candidate TAO3 mutant strains isolated in a previous study [28], but not further characterized, were examined through PCR and DNA sequencing (Fig. 2; dataset S1). One has a T-DNA insertion in the promoter of the TAO3 gene, and the other has a deletion of 7 bp (gcgtagc). Two pseudohyphal T-DNA mutants were isolated during other experiments. One contains an insertion in the KIC1 gene and the other in TAO3. In addition, strains with complete deletion of TAO3 were generated in three backgrounds (C.n. var. grubii KN99α and strain G, and C.n. var. neoformans ATCC 24067A). Overlap PCR products were created to replace the TAO3 ORF with the nourseothricin acetyltransferase cassette via biolistic transformation and homologous recombination. The tao3 point mutant strains have the same phenotype as complete deletion or T-DNA insertion alleles, consistent with complete loss-of-function.
We screened and isolated 13 spontaneous pseudohyphal strains in the C.n. var. neoformans ATCC 24067A strain, six using a UV-induction method [27] and seven by plating colonies for random mutation events (table S1). The strains were transformed with the wild type copies of CBK1, KIC1, MOB2 or SOG2 to test for complementation, and/or the TAO3 gene from these strains was sequenced. Eight contained changes in the TAO3 sequence, as illustrated in Fig. 2. Complementation experiments implicated mutations in the KIC1, MOB2, SOG2 and CBK1 genes in the remaining five, and these mutations were identified by sequencing those genes (Fig. 2 and dataset S1). The SOG2 gene in one had a mutation in which a stretch of 16 bp (tgcacaacgcaactct) in the fifth exon was duplicated and inserted adjacent to the original sequence. This insertion would result in a frame shift mutation. The cbk1 mutant bears an a-g mutation in the 3′ g splicing site of an intron. Likewise, the two kic1 mutants are both bp substitutions within splice sites. MOB2 was sequenced from the mob2 mutant, and has a bp deletion that will cause a translational frameshift. In summarizing, TAO3 is most often mutated in spontaneous pseudohyphal strains but other genes in the RAM pathway can also be affected.
The provenance of the C, D and E strains with pseudohyphal morphology that were isolated in the 1970s by exposing C. neoformans to amoeba is not well documented. Isolating RAM mutants using amoeba was tested. The original Acanthamoeba polyphaga strain from mouse feces and the A. palestinensis strain from pigeon guano, used previously to isolate pseudohyphal colonies, were not saved in a culture collection. The ATCC 30234 strain of A. castellanii was co-isolated with C. neoformans by Aldo Castellani [18], and is commonly used to study the interactions of amoeba with other pathogenic microbes. First, C. neoformans wild type and RAM mutants were exposed to this amoeba in the expectation that a similar interaction would occur as had been described in the 1970s. The outcome of the fungus-amoeba interaction depended on strain background and medium type. In some combinations, such as that illustrated in Fig. 4, the amoeba consumed the wild type strain whereas the RAM pathway mutants were resistant. Second, the wild type strains G (C.n. var. grubii) and ATCC 24067A (C.n. var. neoformans) were exposed to the amoeba on proteose peptone agar, and examined 2–3 weeks later for the presence of surviving colonies. A subset was of colonies comprised of pseudohyphal cells. The nature of the mutation within the RAM pathway for those strains was sought through complementation experiments and gene sequencing. Mutations were identified again in the TAO3 gene, as well as MOB2 (Fig. 2, dataset S1). These findings thus extend the diversity of the C. neoformans/amoeba interaction to include both C. neoformans varieties and a third Acanthamoeba species. In doing so, the results provide further evidence that the RAM pathway is integral to pseudohyphal morphology, and the isolation of these mutants reflects a common underlying ability across divergent C. neoformans strains.
RAM pathway mutants are sensitive to FK506, providing a simple means to select for strains that revert to wild type. Revertants were sought from tao3 mutant strains D and F7. A high rate of spontaneous resistance to FK506 was observed for strain F7, but not all of these FK506 resistant strains had reverted to the wild type yeast cell morphology. Rather, the strains had acquired FK506 resistance through some other means, possibly through mutation of the gene encoding the FKBP12 protein to which FK506 physically binds [43]. The frequency of reversion and FK506 resistance was noticeably lower in strain D than F7. Part of the TAO3 gene was sequenced from >30 revertants derived from F7 with the wild type yeast morphology: four types of changes were identified (Fig. 5A). In the first, an a-t bp mutation had restored the original lysine codon. In the second, an a-c bp change replaced the stop codon with a glutamine codon. In the third, the adjacent nucleotide had mutated (a-t) to change the stop codon to a leucine codon. In the fourth, the original stop codon was still present. In addition, another reversion event was also common, leading to a suppression of the RAM phenotype and partial return to the wild type phenotype. These strains were distinctive because of the yellow-colored colonies, with cells exhibiting a lemon-shape and inefficient cell separation. Sequence analysis revealed that the stop codon mutation was still present in TAO3 in these types of strains. In contrast to the revertants isolated from strain F7, when the region containing the stop codon in TAO3 was sequenced from 24 revertants of strain D, all 24 still contained the stop codon. Thus, mutation within TAO3 allows reversion back to wild type, although another mechanism(s) can account for some reversion events.
The pseudohyphal strain DM03 contains a 16 bp duplicated region within the SOG2 gene, and as such is a different type of mutation compared to the bp substitutions in the TAO3 gene described above. After selection on FK506 for wild type colonies from strain DM03, three reversion types were observed (Fig. 5B). In one, the duplicated piece of DNA was excised. In a second case, 4 bp were deleted downstream of the insertion event, returning the gene to the correct reading frame and producing an allele encoding 45 different amino acid residues. In the third, the original mutation was still present. Thus, excision of a duplication region or insertions or deletions correcting the open reading frame provide yet additional mechanisms to revert RAM mutants.
Northern blot analysis was use to examine changes in transcript levels or sizes in RAM pathway genes in response to mutation or reversion in a selection of var. neoformans and var. grubii strains (Fig. S1). The KIC1 and TAO3 transcripts were below detectable levels. CBK1 and MOB2 have overall constitutive transcript levels. HYM1 and SOG2 showed variation in transcript levels, although there was no perfect correlation between loss of a RAM gene and upregulation. Of note, the mob2 mutant DM09 used in the virulence analysis described below exhibited altered transcript sizes, consistent with a mutation in a predicted intron splice site (Fig. S1).
If mutation is the primary source of pseudohyphal strains, it was surprising that they arise at such a high frequency. This together with the observation of high rates of spontaneous resistance to FK506 in the strains in the ATCC 24067A strain background used in phenotypic switching experiments [27], [44], led us to test the mutation rate in this strain. ATCC 24067A is derived by laboratory passage from strain ATCC 24067 [45]. 20 separate cultures of strains ATCC 24067 and ATCC 24067A were plated onto medium to select for spontaneous uracil auxotrophy. The mutation rate for ATCC 24067 was 2.66 per 1×108 (95% confidence interval 1.78–4.75). In contrast, in ATCC 24067A the rate was 67.88 per 1×108 (95% CI 57.06–79.38). Thus, ATCC 24067A has greater than 25 fold higher mutation rate that its progenitor parent ATCC 24067.
To ensure that the uracil auxotrophs were due to mutations in the same gene, the URA5 gene enoding orotate phosphoribosyltransferase was amplified from 15 5-FOA resistant strains derived from separate starting colonies, and sequenced (Fig. 6A; dataset S2). 5-FOA resistance in fungi can result from mutation of either URA3 or URA5 homologs; prior studies suggest that URA5 is the main target in C. neoformans [46]. All 30 strains had mutations in URA5. Comparing the mutation profiles for ATCC 24067 and ATCC 24067A revealed similarities, e.g. three in each strain background had the same t-c mutation. A main difference was in mutations involving more than one base pair. Two indels for ATCC 24067A were single bp. In contrast, for ATCC 24067 three alleles have large insertions or rearrangements identified by altered or absent PCR products (data not shown). Another three uracil auxotrophs have insertions or deletions between 2 and 18 bp (dataset S2). We interpret this to imply a higher level of bp substitutions in the ATCC 24067A strain. Comparison of the ATCC 24067 and ATCC 24067A strains under stress conditions also revealed altered response to stress agents, especially oxidative stress agents and ethidium bromide (Fig. 6B). We hypothesize that during laboratory passage ATCC 24067A acquired a mutation in a DNA repair pathway gene.
The RAM pathway itself could potentially influence mutation rates, thereby enhancing the rate of reversion. Mutation rates were compared between 15 cultures each of ATCC 24067A and the tao3 point mutant strain F7. Uracil auxotrophs were isolated at a rate of 42.00 per 1×108 (CI 34.30–53.11) for the ATCC 24067A strain, and 15.74 per 1×108 (CI 13.37–18.93) for the F7 strain. These results suggest that there is no increase in mutation rate in the RAM mutants, since the wild type strain had 2.6 times higher frequency for isolation of uracil mutants compared to the F7 strain. A caveat in this comparison is the challenge of accurate quantification of viable cells for RAM pathway mutants.
Additional evidence was sought that pseudohyphal strains and reversions are due to mutation events. A “yellow” suppressor or partially-reverted strain was examined by Mendelian genetic analysis. The strain was isolated in the tao3::NAT deletion strain background (strain AI235) by selection on FK506. The AI235ya strain was crossed to a wild type of opposite mating type. 20 progeny were obtained: three pseudohyphal NATR FK506S, ten wild type yeast NATS FK506R, and seven “yellow” NATR FK506R (Fig. 7). These results show that the suppression phenotype is meiotically stable, and the progeny ratio is consistent with its segregation as a single genetic locus. This finding further illustrates that mutation leads to some types of reversion events that are yet to be defined.
An alternative hypothesis for the high frequency of isolation of RAM mutants would be if the pathway or parts of it were hot spots for mutations. To test this, aliquots from the identical 20 cultures of strain ATCC 24067A used to isolate uracil auxotrophs were plated onto YPD. Six RAM mutants were identified from ∼143,000 colonies. These six came from four starting cultures. In two examples, pairs on strains were isolated from the same plate. Characterization of the pairs (AI273–AI274 and AI275–AI276) revealed that they shared the identical mutation within TAO3, reflecting an attached pair of pseudohyphal cells that were physically separated when spread on the plate. There are 225 codons in URA5 vs. 5750 codons combined for the six RAM pathway genes. Based on size alone we expected a 25.6 fold higher frequency of isolation of pseudohyphal strains compared to 5-FOA resistant strains, or 1 in every 57,546 colonies. The isolation of six mutants from 143,000 screened would be unlikely (P<0.03; Poisson distribution). However, if the pairs of identical tao3 mutants are considered as one event, then four from 143,000 is not statistically significant (P<0.24).
To circumvent bias due to cell separation in the RAM mutants, an alternative measure of mutation was taken. Uracil auxotrophs were isolated in the AI228 mutant, and the URA5 gene sequenced. AI228 has an a-g transition in an intron splice site in CBK1. Strain AI228 ura#3 was isolated with a g-a transition in a highly-conserved glycine codon. We reasoned that the only way to revert the strain to wild type would be the perfect reversal of the mutation. 15 separate colonies were inoculated into liquid medium, cultured overnight, and from each 2.5×108 cells plated onto media supplemented with FK506, to select for a mutation in CBK1, and onto media without uracil, to select for a mutation in URA5. No wild type yeasts were obtained. In contrast, nine uracil prototrophs were isolated, reflecting reversion in URA5. These results indicate that at least one gene in the RAM pathway, CBK1, is not a general hot-spot for mutation.
If amoeba select for pseudohyphal strains of C. neoformans in the wild, then why are pseudohyphal strains not isolated on a regular basis? We explored this question by testing isolation methods for C. neoformans and the fitness of the RAM pathway mutants under different growth conditions.
First, we explored the ability of RAM pathway mutants to grow on medium that mimics an environmental substrate, pigeon guano, with which C. neoformans is associated in nature [47]. Both wild type and RAM mutants grew equally well on pigeon guano medium, suggesting that RAM mutants have equivalent growth as wild type on this substrate (Fig. 8A).
Second, RAM mutants were grown on bird seed agar. This medium is made from Guizotia abyssinica seed and is a standard medium for isolation of C. neoformans from environmental sources, aided by melanization of C. neoformans colonies [48], [49]. In two serotype A genetic backgrounds, the RAM mutant strains were delayed in pigmentation and produced smaller colonies compared to wild type (Fig. 8B). Melanin is a well-established virulence factor for C. neoformans. Another virulence trait is the biosynthesis of a polysaccharide capsule, which was found previously to be produced like wild type in RAM pathway mutants [28].
Third, phenotypes were explored under various stress conditions. No visual differences were observed between wild type and pseudohyphal strains growing on YPD at pH 4.5 or pH 8, or on YPD supplemented with high levels of salt NaCl, detergent sodium dodecyl sulfate, antifungal flucoazole, or oxidative stress agents (H2O2 and t-butyl hydroperoxide). One phenotype identified that did differ is altered colony integrity. The RAM pathway mutant cells were easily dispersed by washing (Fig. 8C), possibly a detrimental trait leading to reduced protection of the cells within a structured colony in the wild. Mutation of the CBK1 homolog in the basidiomycete fungus Ustilago maydis causes sterility [33]. C. neoformans RAM mutants also have reduced fertility in crosses in which both parents bear mutations, since no filaments, basidia or basidiospores are produced in crosses on V8 juice or Murashige-Skoog medium (Fig. 8D).
Thus, explanations for the lack of pseudohyphal C. neoformans isolated from the wild could include their discard due to reduced melanization, reduced growth on bird seed agar or at elevated temperatures, and unconventional cell morphology. Alternatively, while it is possible that RAM mutation confers benefits under some environmental conditions, under others the pseudohyphal strains are less fit thereby countering the advantages gained in avoiding predation by amoeba.
The mechanism by which the RAM mutants of C. neoformans evade amoeba was explored using light microscopy, with confocal microscopy of GFP-expressing fungal strains and amoeba stained with FM4–64 used to confirm internalization of the fungal cells (data not shown). These results indicated that one possible mechanism of action is less efficient phagocytosis of RAM pathway mutants by amoeba. The pseudohyphal strains, especially when in clusters of cells, are larger than the amoeba thereby forming a physical impediment to phagocytosis (Fig. 4B). Consistent with this hypothesis, pseudohyphal cells are less frequently found inside amoeba. For instance, a mixture of wild type and tao3 mutants was exposed to amoeba. Only 3% of amoeba that harbored Cryptococcus had cells that were pseudohyphal (n = 178). In the cases in which pseudohyphal cells were present in amoeba, these were as single or two attached cells (Fig. 4B). While a physical block may account in part for evasion of amoeba, other reasons for resistance, such as increased intracellular survival of pseudohyphal cells, cannot be excluded.
Based on previous observations of the virulence of pseudohyphal C. neoformans and the role of the RAM pathway in virulence in other fungi, we hypothesized that our mutants would be attenuated or avirulent. Wild type strain G, the mob2 mutant strain DM09 that was isolated by exposing strain G to amoeba, and a complemented strain AI255 (DM09+MOB2-NEO) were used to test the role of the RAM pathway in virulence in wax moth larvae and mouse models.
The mob2 mutant strain DM09 exhibits reduced growth at 37°C, which is predicted to influence virulence in mammalian hosts. To address virulence at a more permissive temperature, the three strains were inoculated into wax moth (Galleria mellonella) larvae and the larvae maintained at 30°C. 1×105 cells of wild type and complemented strains were used as inocula. Two inocula were used for DM09: 1×105 cell-clusters and one at 1/10 that concentration. Microscopic analysis of the cultures of the mob2 mutant and accompanying plating assays indicated that approximately 10 pseudohyphal cells formed the equivalent of one colony forming unit, due to the cell separation defect of the strain.
The larvae inoculated with wild type or complemented strains started dying five days after inoculation, and 22 of 23 were dead by day nine (Fig. 9). In contrast, the larvae inoculated with the mob2 mutant strain survived longer, e.g. on day nine only two of the 21 larvae had died. The experiment was terminated when the surviving larvae, including the control group inoculated with PBS, formed cocoons. Log-rank statistical comparisons indicated that the differences in survival between wild type or complemented strains with the mob2 mutant were significant (P<0.0001). Thus, the RAM pathway controls fungal virulence in an insect model and reduced virulence is independent of temperature.
Next, the three strains were tested in a mouse inhalation model of cryptococcosis. A subset of mice were sacrificed 24 and 96 h post-inoculation, and colony counts measured and lung tissue prepared for histology. The wild type proliferated during the three day interval, while the mob2 mutant strain maintained a level of ∼1×105 cfu per gram (Fig. 10A). The colonies isolated from mice inoculated with the wild type were smooth and comprised of yeast cells. Colonies from mice inoculated with the mob2 mutant were all wrinkled and comprised of pseudohyphal cells. Yeast or pseudohyphal cells were evident in histological samples of the lungs at both 24 and 96 h from mice infected with the wild type and mob2 mutant, respectively (Fig. 10B, C). These results show that C. neoformans pseudohyphal cells can penetrate the lung and survive at least four days.
The remaining sets of inoculated mice were monitored daily for signs of cryptococcal disease. The mice infected with the wild type or complemented strains succumbed to disease and were sacrificed by day 26, a time at which all of the mice inoculated with the mob2 mutant strain were alive and healthy. Interestingly, four of these mob2-inoculated mice developed symptoms of cryptococcosis and were sacrificed between days 38 and 49 post-inoculation (Fig. 11A). Cells isolated from either the lungs or brains of these mice were the round yeast morphology like wild type strains (Fig. 11C). The MOB2 gene was amplified from two colonies isolated from each organ. All sequences showed the original g nucleotide found in the wild type gene sequence (Fig. 11C, dataset S3). Histological examination of the lungs and brain of the diseased animals also revealed yeast cells, as well as extensive host tissue damage (Fig. S2). The interpretation of these data is that the mob2 mutation had reverted to wild type in the four animals during the course of infection.
At day 70, the remaining six mice inoculated with the mob2 mutant had no symptoms of disease, so were sacrificed and organ homogenates plated. Three mice had cleared the infection since they had no fungal cells present in either lung or brain tissue. One mouse (number 5) had wild type morphology cells in the brain and lung (Fig. 11B,C). Sequence of MOB2 from these cells showed the wild type gene sequence. Two mice (numbers 7 and 10) had pseudohyphal strains only in the lung, and the brains were free of fungal cells (Fig. 11B,C). When the MOB2 alleles in the pseudohyphal strains were examined by DNA amplification and sequencing, those strains contained the original splicing mutation in MOB2 (Fig. 11C, dataset S3).
In summarizing the results from the mouse model, the mob2 mutant was attenuated for virulence (Log-rank test P<0.0001). However, the pseudohyphal cells can persist during infection, and reversion mutations occur stochastically over time to restore cell shape to the wild type and fully pathogenic form.
In this study we describe a new mechanism for microevolution in the human pathogenic fungus C. neoformans that controls the host range of the organism. Cryptococcus species have plastic genomes, with experiments showing changes in chromosome length over time, microevolution during human infection and in culture, and changes in chromosome numbers conferring azole drug resistance [9], [10], [45], [50], [51]. Microevolution modulates the polysaccharide capsule composition that surrounds the cell: this is the best-studied aspect about microevolution in the fungus [44], [52], [53].
The Cryptococcus genus is found in association with trees, soil, bird excreta and additional environments that are also homes to other microbial species [1], [14]. One hypothesis is that selection for traits that defend against small predators, such as amoeba or nematodes, has led to species capable of causing disease in humans [13], [14], [15], [16]. The evidence for amoeba-C. neoformans interactions date to over half a century ago with the work of Castellani, who showed that the species later named Acanthamoeba castellanii could kill C. neoformans cells [20]. In the 1970s, amoeba were again co-isolated with C. neoformans [15], [17]. Subsequent studies found that a subset of C. neoformans colonies changed cellular morphology after exposure to Acanthamoeba species, from yeast to pseudohyphal cells, and that some isolates reverted rapidly to wild type yeast. Based on the more recent observation of phenotypic switching between pseudohyphal and yeast forms [27] and the discovery of a set of genes that produces a pseudohyphal phenotype when mutated [28], we hypothesized that phenotypic switching involved the RAM pathway.
Here we show that the RAM pathway is the integral component of “switching” in C. neoformans because it is mutated in pseudohyphal strains isolated from amoeba and spontaneously in culture. The largest gene in the pathway, TAO3, most commonly bears point mutations leading to the introduction of premature stop codons. Some strains like F7 revert to wild type at a high frequency in vitro. Analysis of the mutated region in those strains reveals that there are multiple ways in which the strain can revert (Fig. 5). This may be through a bp substitution leading to a coding triplet being reformed. Alternatively, the stop codon may still be present. The basis for the latter situation is unknown. It could be mediated by stop codon read through, tRNA suppressor mutations, changes in downstream gene expression, or epigenetic phenomena. Mutation in another RAM pathway component may suppress the phenotype, as occurs with a specific residue in the Cbk1 kinase of S. cerevisiae to rescue mutations in other pathway components [54], or be modified by interacting pathways such as seen for the Neurospora crassa cot-1 suppressors [55], [56]. An alternative mechanism of reversion is illustrated by strain DM03 bearing a mutation in SOG2. Excision of the duplicated region or deletion of another region downstream reverts Sog2 sequence back to wild type or in frame, respectively. Taken together, the conversion between yeast and pseudohyphal cells reported previously as a form of phenotypic switching is based on DNA mutations, rather than epigenetic changes.
The effect of a defective RAM pathway on mammalian virulence was tested. The mob2 mutant strain used carries a bp substitution mutation within an intron splice site, and is phenotypically stable. Three strains were inoculated into mice. The wild type and complemented strains caused cryptococcal disease. In contrast, different outcomes were observed for mice infected with the mob2 mutant (Fig. 11). Four mice succumbed to disease, albeit weeks after those infected with the wild type and mutant had been sacrificed, and when their organs were harvested only yeast cells were recovered rather than the expected pseudohyphal cells that were used to inoculate the animals. The MOB2 gene was sequenced, and now had the wild type sequence (Fig. 11C). Sacrifice of the remaining and asymptomatic animals at day 70 and characterization of fungal material in lung and brain tissue shows that three mice had cleared the infection, one carried wild type cells, and the other two still maintained the original mob2 mutant phenotype and genotype. These results are consistent with previous virulence studies using pseudohyphal strains selected by amoeba in which reversion back to wild type for some occurred at a high frequency within mouse models [17], [25], [26]. A third animal experiment has been performed with pseudohyphal strains, presumably also RAM mutants, in a rat tracheal model [27]. In this experiment, two of the four animals cleared the infection while the other two did not, potentially representing another case of reversion within the host.
Morphological differentiation is important for Cryptococcus pathogenesis. Recently, a role has been assigned for a giant cell form during disease development [57], [58], while constitutive filamentation by altered regulation of the ZNF2 gene impairs virulence [59]. There are reports of pseudohyphal cells in histopathological samples of patients infected with C. neoformans [60], [61]. One speculation is that pseudohyphal forms could allow escape of the fungus from mammalian or amoeba cells, in addition to escape of yeast cells from macrophages or A. castellanii by exocytosis [62], [63], [64].
How can “switching” occur at high frequency? The formation of pseudohyphal strains relates to mutation rates in cells; switching also increases upon exposure to UV light [27]. Two factors influence frequency. The first is that the strain used in phenotypic switching studies has a 25 times higher mutation rate compared to a standard wild type strain. Second, because there are six genes in the RAM pathway there is a large amount of target DNA available for spontaneous mutation. Protein-coding sequence alone, the six genes comprise more than 17 kb of DNA, or ∼0.1% of the genome (Fig. 2). TAO3, as the largest member (42% of total cumulative size), is therefore the most likely gene to be hit. The original historical isolates bear mutations in this gene. Of 14 unique mutations that we defined in the ATCC 24067A background, eight are in TAO3, supporting this hypothesis. One challenge for quantitative analysis of the pseudohyphal strains is their defect in cell separation. Reversion to wild type has been estimated as high as 1.6×10−3 [27], but this may be an over-estimate by up to an order of magnitude if the colony forming units were derived from attached cells.
It is unknown why an organism like Cryptococcus, which is normally found in the environment, can cause disease in humans or other mammals. Further, the fungal behavior upon entering the human host that ends in life-threatening disease remains unclear. Three points are worth raising. First, people are exposed to C. neoformans during their lifetimes yet most do not develop disease. For instance, children in city environments where there is a high prevalence of pigeons as sources of C. neoformans become antigen positive at an early age [65]. Second, cases of re-activation from quiescent infections brought about by immunosuppression supports a hypothesis that the fungus enters a latent state [66]. Third, in comparing virulence of strains derived directly from the environment vs. a human host, many environmental strains do not cause disease in animal models although they persist in the lungs [67], [68], [69], [70]. Collectively, a high rate of exposure to C. neoformans, to strains that may not necessarily be able to cause disease immediately, and the potential for latency provide the scenario in which microevolution of C. neoformans by DNA mutations could influence clinical outcomes.
Multiple mechanisms can facilitate microevolution in pathogens. Among the fungi, mutations during infection lead to the emergence of antifungal drug resistance. However, these arise under conditions with a high fungal burden in the host, promoting the generation of strains from rare events. More broadly, there is evidence from diverse microbes that DNA is mutated to generate phenotypic variation. In the protist Trypanosoma brucei, mutation is required for evasion of the host immune response whereby double stranded breaks are generated and then repaired to generate antigenic diversity via the VSG genes [71]. Adaptation via mutation is implicated in bacterial disease progression. For example, 20% of Pseudomonas aeruginosa isolates from cystic fibrosis patients are “hypermutators” compared with 0% of environmental isolates [72]. Nevertheless, this trend in bacteria is not universal, as correlations between clinical isolates of E. coli and higher mutation rates have been found in some, but not all, studies [73], [74]. A very different system to develop variation is the low fidelity of the human immunodeficiency virus' reverse transcriptase: along with other factors this results in rapid genetic changes within the human host and reduces immunological recognition of the virus. In contrast, evidence for mutations affecting the pathogencity of C. neoformans are rare at present. Curiously, a C. neoformans mutant in the MSH201 gene predicted to function in mismatch repair has a competitive advantage in mouse lungs compared to control strains [75].
This research points towards future directions into investigating the contribution mutation and mutation rates play in the ability of C. neoformans and other pathogenic fungi to adapt and cause disease. Specific directions are to assess mutation rates within the host and test if mutations that arise in the host result in more virulent strains. Second is to test for correlations between clinical and environmental isolates and rates of mutation. A third direction is to explore what role, if any, transcriptional, translational or epigenetic regulation of the RAM pathway plays in the interaction of C. neoformans with amoeba and mammalian cells.
Cryptococcus neoformans wild type strains used were KN99α (var. grubii), G (var. grubii, ATCC 42347), ATCC 24067, ATCC 24067A (var. neoformans), and JEC21 (var. neoformans). ATCC 24067A is derived from laboratory passage of ATCC 24067. Historical pseudohyphal strains were F7 (var. neoformans), and C, D, and E (var. grubii; ATCC 42343-5). Strains were kindly provided by Dr. Joseph Heitman and Dr. Bettina Fries. C. neoformans strains were cultured on yeast extract-peptone dextrose (YPD) ±2% agar medium, and stored as glycerol stocks at −80°C. The Acanthamoeba castellanii strain was obtained from the American Type Culture Collection (ATCC 30234) and maintained according to ATCC instructions and stored at 4°C [76]. To isolate new pseudohyphal strains, ATCC 24067A was grown in overnight YPD cultures, then spread on YPD plates. A subset of plates were subject to a low dose of UV light in a UV transilluminator. Colonies were screened by eye for those with a dry appearance, which were streaked to isolate single colonies. To isolate RAM mutants using the A. castellanii amoeba, C. neoformans strains ATCC 24067A and G were inoculated in a cross pattern on a selection of different agar medium types (potato dextrose, proteose peptone, YPD, V8 juice, Murashige-Skoog and trypan blue), and a drop of amoeba placed at the intersection, following the original protocol [17]. T-DNA insertional mutants were generated as described previously and RAM pathway mutants isolated based on colony morphology [28]. Wild type revertants were selected from RAM pathway mutants by plating on YPD agar supplemented with FK506 (1 µg/ml). Pigeon guano medium was 10% w/v of unfiltered pigeon guano (collected under the I-35 overpass of Southwest Blvd, Kansas City, MO) that had been homogenized in a coffee grinder and autoclaved with 4% agar. Bird seed agar was prepared as described [48]. Crosses were set up on 5% V8 juice or Murashige-Skoog agar [77]. Strains used and generated during this study are listed in Table S1.
Genomic DNA was extracted using a CTAB buffer [78] from 50 ml overnight cultures of strains. The TAO3 gene was amplified by PCR with two primer sets for each variety: ALID0013–ALID0061 and ALID0014–ALID0060 for var. grubii strains, and ALID0127–ALID0128 and ALID0129–ALID0138 for var. neoformans strains. SOG2 was amplified with primers ALID0123–ALID0162. MOB2 was amplified with primers DM062–DM063. CBK1 was amplified with primers ALID0977–ALID0978. Part of the KIC1 gene was amplified with primers ALID1681–ALID1682 The PCR products were sequenced with the primers used for amplification and additional internal primers. Primer sequences used for amplification are listed in Table S2.
Tests were performed on pseudohyphal strains using vectors that complement the deletion mutants of mob2, cbk1, kic1, and sog2. The first three plasmids were generated in a previous study on the RAM pathway [28]. The SOG2-NEO1 construct was generated by amplification of SOG2 from strain JEC21 with primers ALID0123–ALID0162, cloning into TOPO pCR2.1 (Invitrogen, Life Technologies, Grand Island, NY), and a SpeI-XbaI fragment subcloning into the XbaI site of pPZP-NEO11. The four genes were in plasmids enabling their introduction into C. neoformans cells via Agrobacterium-mediated transformation [79]. Transformants were selected on YPD medium containing cefotaxime (200 µg/ml) and either nourseothricin (100 µg/ml) or neomycin (200 µg/ml). The TAO3 gene in strain D was reconstituted by homologous recombination. A construct with an engineered BglII restriction enzyme site was generated by overlap PCR using primers ALID0061–ALID0227 and ALID0111–ALID0228, and introduced into strain D cells plated on YPD+1 M sorbitol by biolistic transformation with a PDS-1000/He Particle Delivery System (Bio-Rad, Hercules, CA), using standard methods [80]. Cells were allowed to recover for 3 h and transferred to YPD+FK506 (1 µg/ml).
The TAO3 gene was deleted in the KN99α, ATCC 24067A and G strains. For var. grubii strains, the 5′ and 3′ flanks were amplified from genomic DNA of strain KN99α using primers damp5-ALID0013 and damp6–damp7, respectively. For var. neoformans, the 5′ and 3′ flanks were amplified from strain ATCC 24067A using primers damp1–damp2 and damp3–damp4, respectively. Nourseothricin acetyltransferase (NAT) was amplified from plasmid pAI3 using primers ai006–ai290 [79]. The primers ALID0013-damp7 or damp1–damp4 were used for overlap PCR. The SOG2 gene was deleted in strain KN99α. Primers ALID0483–ALID0484 and DM036–DM043 were used to amplify the 5′ and 3′ flanks, and ALID0483–DM036 used for overlap PCR with these and the NAT cassette. The DNA molecules were transformed into C. neoformans cells using the biolistic apparatus, cells allowed to recover for 3 h, and transferred to YPD medium containing nourseothricin (100 µg/ml). Correct gene replacement was confirmed by PCR analysis and Southern blotting with [32P]-dCTP-labelled fragments of the genes.
Isolation of spontaneous uracil auxotrophs was used to measure mutation frequency. For comparison between strains ATCC 24067, which was acquired from the ATCC, and ATCC 24067A, strains were grown on yeast nitrogen base (YNB) medium, then 20 separate cultures of each strain established at 1×105 cells/ml in YPD medium. After overnight culture in a roller drum incubated at room temperature, 5×107 or 1×108 cells were plated onto YNB supplemented with uracil (20 mg/L) and 5-fluoroorotic acid (5-FOA; 1 g/L) medium. The resulting colony numbers were analyzed to determine the mutation rate (Lea-Coulson method of the median) with FALCOR software [81]. To ensure mutations targeted the same gene in both strains, the URA5 gene was amplified with primers ALID0375–ALID0376 and sequenced.
To compare mutation rates in URA5 to the RAM pathway genes, aliquots from the same 20 ATCC 24067A cultures used to isolate URA5 mutants were diluted and plated onto ten YPD plates. Dry colonies were screened visually and pseudohyphal morphology confirmed by microscopy. The nature of the mutation in these strains was identified by complementation tests and DNA sequence analysis. Statistical analysis used the Poisson distribution, testing the probability of isolating n or more pseudohyphal strains. Strain AI228 ura#3 (cbk1 ura5) was inoculated into 15 YPD cultures, grown overnight, and 2.5×108 cells plated onto YPD+2 µg/ml FK506 and YNB.
50 ml cultures in liquid YPD medium were incubated overnight at 150 rpm at room temperature. The cells were frozen and lyophilized. Total RNA was extracted with TRIzol (Invitrogen) or TRI reagent (Sigma-Aldrich, St. Louis, MO). For northern blots, 10 µg of RNA purified from each strain were resolved on 1.4% agarose/formaldehyde gels. RNA was blotted to Zeta-Probe membrane (Bio-Rad). [32P]-labeled probes of the six RAM genes (the primers used for amplification are in Table S2) were hybridized to blots. Blots were stripped and reprobed with actin (ACT1) as a loading and RNA transfer control. RNA purified from wild type strain KN99α was also used to confirm the intron-exon boundaries of TAO3, by sequencing cDNAs reverse transcribed with Superscript III (Invitrogen).
Wax moth larvae (G. mellonella) were purchased from Vanderhorst Wholesale (Saint Mary's, OH). Overnight cultures of C. neoformans grown in liquid YPD were washed in phosphate buffered saline (PBS), the concentration determined by counting cells with a hemocytometer, and diluted such that 1×105 cells of wild type and the complementation mob2+MOB2 strain were injected into the larvae as described previously [82]. For mob2 mutant strain DM09, 1×105 and 1×106 cells were inoculated. Concentrations were confirmed by plating serial dilutions onto YPD agar plates.
Groups of female A/JCr mice (NCI-Frederick, MD) were infected intranasally with 105 cfus of each strain, as previously described [83]. Inocula were confirmed by plating onto YPD agar. Animals that appeared moribund or in pain were sacrificed by CO2 inhalation. For cfu assays, lungs and brain were dissected from animals, homogenized in PBS, and plated onto YPD medium containing ampicillin and chloramphenicol. Colonies were determined after incubation for 3 d at 30°C. For histology, lung and brain samples were fixed and hematoxylin and eosin (H&E) stained. Survival data from the murine experiments were statistically analyzed between paired groups using the log-rank test in the PRISM program 4.0 (GraphPad Software). P values of <0.01 were considered significant.
The mouse experiments were performed in full compliance with a protocol approved by the University of Medicine and Dentistry of New Jersey Institutional Animal Care and Use Committee, and in compliance with the United States Animal Welfare Act (Public Law 98–198). The experiments were carried out in facilities accredited by the Association for Assessment and Accreditation of Laboratory Animal Care.
Sequences of CBK1, MOB2 and TAO3 from strain ATCC 24067A have been deposited to GenBank, under accessions HM770879, JX297541 and GU903010, respectively.
|
10.1371/journal.pcbi.1004152 | Predicting Epidemic Risk from Past Temporal Contact Data | Understanding how epidemics spread in a system is a crucial step to prevent and control outbreaks, with broad implications on the system’s functioning, health, and associated costs. This can be achieved by identifying the elements at higher risk of infection and implementing targeted surveillance and control measures. One important ingredient to consider is the pattern of disease-transmission contacts among the elements, however lack of data or delays in providing updated records may hinder its use, especially for time-varying patterns. Here we explore to what extent it is possible to use past temporal data of a system’s pattern of contacts to predict the risk of infection of its elements during an emerging outbreak, in absence of updated data. We focus on two real-world temporal systems; a livestock displacements trade network among animal holdings, and a network of sexual encounters in high-end prostitution. We define the node’s loyalty as a local measure of its tendency to maintain contacts with the same elements over time, and uncover important non-trivial correlations with the node’s epidemic risk. We show that a risk assessment analysis incorporating this knowledge and based on past structural and temporal pattern properties provides accurate predictions for both systems. Its generalizability is tested by introducing a theoretical model for generating synthetic temporal networks. High accuracy of our predictions is recovered across different settings, while the amount of possible predictions is system-specific. The proposed method can provide crucial information for the setup of targeted intervention strategies.
| Following the emergence of a transmissible disease epidemic, interventions and resources need to be prioritized to efficiently control its spread. While the knowledge of the pattern of disease-transmission contacts among hosts would be ideal for this task, the continuously changing nature of such pattern makes its use less practical in real public health emergencies (or otherwise highly resource-demanding when possible). We show that in such situations critical knowledge to assess the real-time risk of infection can be extracted from past temporal contact data. An index expressing the conservation of contacts over time is proposed as an effective tool to prioritize interventions, and its efficiency is tested considering real data on livestock movements and on human sexual encounters.
| Being able to promptly identify who, in a system, is at risk of infection during an outbreak is key to the efficient control of the epidemic. The explicit pattern of potential disease-transmission contacts has been extensively used to this purpose in the framework of theoretical studies of epidemic processes, uncovering the role of the pattern’s properties in the disease propagation and epidemic outcomes [1, 2, 3, 4, 5, 6, 7, 8]. These studies are generally based on the assumption that the entire pattern of contacts can be mapped out or that its main properties are known. Although such knowledge would be a critical requirement to conduct risk assessment analyses in real-time, which need to be based on the updated and accurate description of the contacts relevant to the outbreak under study [9], it can hardly be obtained in reality. Given the lack of such data, analyses generally refer to the most recent available knowledge of contact data, implicitly assuming a non-evolving pattern.
The recent availability of time-resolved data characterizing connectivity patterns in various contexts [10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22] has inevitably weakened the non-evolving assumption, bringing new challenges to the assessment of nodes’ epidemic risk. Traditional centrality measures used to identify vulnerable elements or influential spreaders for epidemics circulating on static networks [1, 2, 4, 23, 24, 25, 26, 27, 28, 29, 30] are unable to provide meaningful information for their control, as these quantities strongly fluctuate in time once computed on the evolving networks [19, 31]. An element of the system may thus act as superspreader in a past configuration of the contact network, having the ability to potentially infect a disproportionally larger amount of secondary contacts than other elements [32], and then assume a more peripheral role in the current pattern of contact or even become isolated from the rest of the system [19]. If the rules driving the change of these patterns over time are not known, what information can be extracted from past contact data to infer the risk of infection for an epidemic unfolding on the current (unknown) pattern?
Few studies have so far tried to answer this question by exploiting temporal information to control an epidemic through targeted immunization. They are based on the extension to temporal networks [33, 34] of the so-called acquaintance immunization protocol [4] introduced in the framework of static networks that prescribes to vaccinate a random contact of a randomly chosen element of the system. In the case of contacts relevant for the spread of sexually transmitted infections, Lee et al. showed that the most efficient protocol consists in sampling elements at random and vaccinating their latest contacts [33]. The strategy is based on local information gathered from the observation and analysis of past temporal data, and it outperforms static-network protocols. Similar results are obtained for the study of face-to-face contact networks relevant for the transmission of acute respiratory infections in a confined setting, showing in addition that a finite amount of past network data is in fact needed to devise efficient immunization protocols [34].
The aim of these studies is to provide general protocols of immunization over all possible epidemiological conditions of the disease (or class of diseases) under study. For this reason, protocols are tested through numerical simulations and results are averaged over starting seeds and times to compare their performance. Previous work has however shown that epidemic outcomes may strongly depend on the temporal and geographical initial seed of the epidemic [35], under conditions of large dynamical variability of the network and absence of stable structural backbones [19]. Our aim is therefore to focus on a specific epidemiological condition relative to a given emerging outbreak in the population, resembling a realistic situation of public health emergency. We focus on the outbreak initial phase prior to interventions when facing the difficulty that some infected elements in the population are not yet observed. The objective is to assess the risk of infection of nodes to inform targeted surveillance, quarantine and immunization programs, assuming the lack of knowledge of the explicit contact pattern on which the outbreak is unfolding. Knowledge is instead gathered from the analysis of the full topological and temporal pattern of past data (similarly to previous works [33, 34]), coupled, in addition, with epidemic spreading simulations performed on such data under the same epidemiological conditions of the outbreak under study. More specifically, we propose an egocentric view of the system and assess whether and to what extent the node’s tendency of repeating already established contacts is correlated with its probability of being reached by the infection. Findings obtained on past available contact data are then used to predict the infection risk in the current unknown epidemic situation. We apply this risk assessment analysis to two large-scale empirical datasets of temporal contact networks—cattle displacements between premises in Italy [19, 36], and sexual contacts in high-end prostitution [16]—and evaluate its performance through epidemic spreading simulations. We also introduce a model to generate synthetic time-varying networks retaining the basic mechanisms observed in the empirical networks considered, in order to explain the results obtained by the proposed risk assessment strategy within a general theoretical framework.
The cattle trade network is extracted from the complete dataset reporting on time-resolved bovine displacements among animal holdings in Italy [19, 36] for the period 2006–2010, and it represents the time-varying contact pattern among the 215,264 premises composing the system. The sexual contact network represents the connectivity pattern of sexual encounters extracted from a Web-based Brazilian community where sex buyers provide time-stamped rating and comments on their experiences with escorts [16].
The five-years data of the livestock trade network show that stationary properties at the global level co-exist with an active non-trivial local dynamics. The probability distributions of several quantities measured on the different yearly networks are considerably stable over time, as e.g. shown by the in-degree distribution reported in Fig. 1A, where the in-degree of a farm measures the number of premises selling cattle to that farm. These features, however, result from highly fluctuating underlying patterns of contacts, never preserving more than 50% of the links from one yearly configuration to another (Fig. 1C), notwithstanding the seasonal annual pattern due to repeating cycles of livestock activities [37, 38] (see S1 Text). Similar findings are also obtained for the sexual contact network (Fig. 1B-D), where the lack of an intrinsic cycle of activity characterizing the system leads to smaller values of the overlap between different configurations (< 10%). In this case we consider semi-annual configurations, an arbitrary choice that allows us to extract six network configurations in a timeframe exhibiting an approximately stationary average temporal profile of the system, after discarding an initial transient time period from the data [16]. Different time-aggregating windows are also considered (see the Materials and Methods section and S1 Text for additional details).
The observed values of the overlap of the time-resolved contact networks in terms of the number of links preserved are a measure of the degree of memory contained in the system. This is the outcome of the temporal activity of the elements of the system that reshape up to 50% or 90% of the contacts of the network (in the cattle trade case and in the sexual contact case, respectively), through nodes’ appearance and disappearance, and neighborhood restructuring. By framing the problem in an egocentric perspective, we can explore the behavior of each single node of the system in terms of its tendency to remain active in the system and re-establish connections with the same partners vs. the possibility to change partners or make no contacts. We quantitatively characterize this tendency by introducing the loyalty θ, a quantity that measures the fraction of preserved neighbors of a node for a pair of two consecutive network configurations in time, c−1 and c. If we define 𝒱 i c − 1 as the set of neighbors of node i in configuration c−1, then θ i c − 1 , c is given by the Jaccard index between 𝒱 i c − 1 and 𝒱 i c:
θ i c − 1 , c = 𝒱 i c − 1 ∩ 𝒱 i c 𝒱 i c − 1 ∪ 𝒱 i c . (1)
Loyalty takes values in the interval [0, 1], with θ = 0 indicating that no neighbors are retained, and θ = 1 that exactly the same set of neighbors is preserved (𝒱 i c − 1 = 𝒱 i c). It is defined for discrete time windows (c, c+1) and in general it depends on the aggregation interval chosen to build network configurations.
In case the network is directed, as for example the cattle trade network, θ can be equivalently computed on the set 𝒱 i n , i c of incoming contacts or on the set of neighbors of outgoing connections, 𝒱 o u t , i c, depending on the system-specific interpretation of the direction and on the interest in one phenomenon or the opposite. This measure originally finds its inspiration in the study of livestock trade networks, where a directed connection from holding A to holding B indicates that B purchased a livestock batch from A, which was then displaced along the link direction A → B. If we compute θ on the incoming contacts of the cattle trade network, we thus quantify the propensity of each farmer to repeat business deals with the same partners when they purchase their cattle. This concept is at the basis of many loyalty or fidelity programs that propose explicit marketing efforts to incentivize the reinforcement of loyal buying behavior between a purchasing client and a selling company [39], and corresponds to a principle of exclusivity in selecting economic and social exchange partners [40, 41]. Analogously, in the case of the sexual contact network we consider the point of view of sex buyers. Formally, our methodology can be carried out with the opposite point of view, by considering out-degrees with loyalties being computed on out-neighbors. Our choice is arbitrary and inspired by the trade mechanism underlying the network evolution.
Other definitions of similarity to measure the loyal behavior of a node are also possible. In S1 Text we compare and discuss alternative choices. For the sake of clarity all symbols and variables used in the article are reported in Table 1. Finally, other mechanisms different from fidelity strategies may be at play that result in the observed behavior of a given node. In absence of additional knowledge on the behavior underlying the network evolution, we focus on the loyalty θ to explore whether it can be used as a possible indicator for infection risk, as illustrated in the following subsection.
The distributions of loyalty values, though of different shapes across the two datasets, display no considerable variation moving along consecutive pairs of configurations of each dataset (Fig. 2A-B and S1 Text), once again indicating the overall global stability of system’s properties in time and confirming the results observed for the degree. A diverse range of behaviors in establishing new connections vs. repeating existing ones is observed, similarly to the stable or exploratory strategies found in human communication [42]. Two pronounced peaks are observed for θ = 0 and θ = 1, both dominated by low degree nodes for which few loyalty values are allowed, given the definition of Eq. (1) (see S1 Text for the dependence of θ on nodes’ degree and its analytical understanding). The exact preservation of the neighborhood structure (θ = 1) is more probable in the cattle trade network than in the sexual contact network (P(θ = 1) being one order of magnitude larger), in agreement with the findings of a higher system-wide memory reported in Fig. 1. Moreover, the cattle trade network exhibits the presence of high loyalty values (in the range θ ∈ [0.7, 0.9]), differently from the sexual contact network where P(θ) is always equal to zero in that range except for one pair of consecutive configurations giving a positive probability for θ = 0.8. Farmers in the cattle trade network thus display a more loyal behavior in purchasing cattle batches from other farmers with respect to how sex buyers establish their sexual encounters in the analyzed sexual contact dataset.
For the sake of simplification, we divide the set of nodes composing each system into the subset of loyal nodes having θ greater than a given threshold ϵ, and the subset of disloyal nodes if instead θ < ϵ. We call hereafter these classes as loyalty statuses L and D, respectively, and we will later discuss the role of the chosen value for ϵ.
Both networks under study represent substrates offering potential opportunities for a pathogen to diffuse in the corresponding populations. Sexually transmitted infections spread among the population of individuals through sexual contacts [43, 44], whereas livestock infectious diseases (e.g. Foot-and-mouth disease [45], Bluetongue virus [46], or BVD [47]) can be transmitted from farm to farm mediated by the movements of infected animals (and vectors, where relevant), potentially leading to a rapid propagation of the disease on large geographical scales.
As a model for disease-transmission on the network of contacts we consider a discrete-time Susceptible-Infectious compartmental approach [48]. No additional details characterizing the course of infection are considered here (e.g. recovery dynamics), as we focus on a simplified theoretical picture of the main mechanisms of pathogen diffusion and their interplay with the network topology and time-variation, for the prediction of the risk of infection. The aim is to provide a general and conceptually simple framework, leaving to future studies the investigation of more detailed and realistic disease natural histories.
At each time step, an infectious node can transmit the disease along its outgoing links to its neighboring susceptible nodes that become infected and can then propagate the disease further in the network. Here, we consider a deterministic process for which the contagion occurs with probability equal to 1, as long as there exist a link connecting the infectious node to a susceptible one. Although a crude assumption, this allows us to simplify the computational aspects while focusing on the risk prediction. The corresponding stochastic cases exploring lower probabilities of transmission per link are reported in S1 Text.
We focus on the early phase of the spreading simulations, defined as the set of nodes infected up to simulation time step τ = 6. This choice allows us to study invasion stage only, while the epidemic is no more trivially confined to the microscopic level. Additional choices for τ have been investigated showing that they do not alter our findings (see S1 Text). Network configurations are kept constant during outbreaks, assuming diseases spread faster than network evolution, at least during their invasion stage. Examples of incidence curves obtained by the simulations are reported in S1 Text.
Livestock disease spread is often modeled by assuming that premises are the single discrete units of the spreading processes and neglecting the possible impact of within-farm dynamics [49]. This is generally considered in the study of highly contagious and rapid infections, and corresponds to regarding a farm as being infected as soon as it receives the infection from neighboring farms following the transport of contagious animals. Under this assumption, both case studies can be analyzed in terms of networks of contacts for disease transmission. In addition, for sake of simplicity, we do not take into account the natural definition of link weights on cattle network, representing the size of the moved batches. In S1 Text we generalize our methodology to the weighted case, including a weighted definition of loyalty, reaching results similar to the unweighted case.
We consider an emerging epidemic unfolding on a network configuration c and starting from a single node (seed s), where the rest of the population of nodes is assumed to be initially susceptible. The details on the simulations are reported in the Material and Methods section. We define 𝓘 s c the set of nodes infected during the early stage invasion. In order to explore how the network topology evolution alters the spread of the disease, we consider an outbreak unfolding on the previous configuration of the system, c−1, and characterized by the same epidemiological conditions (same epidemic parameters and same initial seed s). By comparing the set of infected nodes 𝓘 s c − 1 obtained in configuration c−1 to 𝓘 s c, we can assess changes in the two sets and how these depend on the nodes’ loyalty. We define a node’s infection potential π L c − 1 , c ( s ) (π D c − 1 , c ( s )) measuring the probability that a node will be infected in configuration c by an epidemic starting from seed s, given that it was infected in configuration c−1 under the same epidemiological conditions and provided that its loyalty status is L (D):
π L c − 1 , c ( s ) = d e f Prob i ∈ ℐ s c | i ∈ ℐ s c − 1 and i ∈ { L } , π D c − 1 , c ( s ) = d e f Prob i ∈ ℐ s c | i ∈ ℐ s c − 1 and i ∈ { D } ,
where i is a node of the system. πL and πD thus quantify the effect of the temporal stability of the network at the local level (loyalty of a node) on the stability of a macroscopic process unfolding on the network (infection). They depend on the seed chosen for the start of the epidemic, on the pair (c−1, c) of network configurations considered along its evolution, and also on the threshold value ϵ assumed for the definition of the loyalty status of the nodes.
By exploring all seeds and computing the infection potentials for different couples of years, we obtain sharply peaked probability distributions of πL and πD around values that are well separated along the π axis. Results are qualitatively similar in both cases under study, with peaks reached for πL/πD ≃ 2.5 in the cattle trade network and πL/πD ≃ 3 in the sexual trade network (Fig. 3A-B). An observed infection in c−1, based on the knowledge of the epidemiological conditions and no information on the network evolution, is an indicator of an infection risk for the same epidemic in c more than twice larger for loyal farms with respect to disloyal farms. Analogously, loyal sex buyers have a threefold increase in their infection potential with respect to individuals having a larger turnover of partners. Remarkably, small values of loyalty threshold ϵ are able to correctly characterize the loyal behavior of nodes with status L. Results shown in Fig. 3A-B are obtained for ϵ = 0.1. Findings are however robust against changes in the choice of the threshold value, as this is induced by the peculiar bimodal shape of the probability distribution curves for the loyalty (see S1 Text). This means that intermediate values of the local stability of the nodes (i.e. θ > ϵ) imply that a possible risk of being infected is strongly stable, regardless of the dynamics of the network evolution. Valid for all possible seeds and epidemiological conditions, this result indicates that the loyalty of a node can be used as an indicator for the node’s risk of infection, which has important implication for the spreading predictability in case an outbreak emerges.
These results are obtained for temporally evolving networks where no further change induced by the epidemic is assumed to occur. Focusing on the initial stage of the outbreak, we disregard the effect of interventions (e.g. social distancing, quarantine of infectious nodes, movements bans) or of adaptive behavior following awareness [37, 50, 51, 52, 53, 54]. Such assumption relies on the study’s focus on the initial stage of the epidemic that may be characterized by a silent spreading phase with propagation occurring before the alert or outbreak detection takes place; or, following an alert, by a contingent delay in the implementation of intervention measures.
The observed relationship between loyalty and infection potential can be used to define a strategy for the risk assessment analysis of an epidemic unfolding on an unknown networked system at present time, for which we have however information on its past configurations. This may become very useful in practice even in the case of complete datasets, as for example with emerging outbreaks of livestock infectious diseases. Data on livestock movements are routinely collected following European regulations [55], however they may not be readily available in a real-time fashion upon an emergency, and a certain delay may thus be expected. Following an alert for an emerging livestock disease epidemic, knowledge of past network configurations may instead be promptly used in order to characterize the loyalty of farmers, simulate the spread of the disease on past configurations and thus provide the expected risk of infection for the farms under the ongoing outbreak. The general scheme of the strategy for the risk assessment analysis is composed of the following steps, assuming that the past network configurations {c−n, …, c−1, c} are known and that the epidemic unfolds on the unknown configuration c+1:
identify the seed s of the ongoing epidemic;
characterize the loyalty of the nodes from past configurations by computing θ i c − 1 , c from Eq. (1);
predict the loyalty of the nodes for the following unknown configuration c+1: θ i c , c + 1;
simulate the spread of the epidemic on the past configuration c under the same epidemiological conditions of the ongoing outbreak and identify the infected nodes 𝓘 s c;
compute the node epidemic risk for nodes in statuses L and D.
This strategy enables the assessment of the present infection risk (i.e. on configuration c+1) for all nodes hit by the simulated epidemic spreading on past configuration c (𝓘 s c), not knowing their present pattern of contacts. It is based on configurations from c−n to c as they are all used to build the probability distributions needed to train our approach. In the cases under study such distributions are quite stable over time so that a small set of configurations ({c−2, c−1, c}) was shown to be enough.
To make the above strategy operational, we still need to determine how we can exploit past data to predict the evolution of the loyalty of a node in future configurations (step 3) and use this information to compute nodes epidemic risk (point 5). As with all other variables characterizing the system, indeed, also θ may fluctuate from a pair of configurations (c−1, c) to another, as nodes may alter their loyal behavior over time, increasing or decreasing the memory of the system across time. Without any additional knowledge or prior assumption on the dynamics driving the system, we measure from available past data the probabilities of (dis)loyal nodes staying (dis)loyal across consecutive configurations, or conversely, of changing their loyalty status. This property can be quantified in terms of probabilities of transition across loyalty statuses. We thus define T L L c ( k ) as the probability that a node with degree k being loyal between configurations c−1 and c will stay loyal one step after (c, c+1). It is important to note the explicit dependence on the degree k of the node (here defined at time c), which may increase or decrease following neighborhood reshaping (it may also assume the value k = 0 if the node becomes inactive in configuration c). Analogously, T D D c ( k ) is the probability of remaining disloyal. The other two possible transition probabilities are easily obtained as TLD = 1−TLL and TDL = 1−TDD.
Fig. 3C-D show the transition probabilities of maintaining the same loyalty status calculated on the two empirical networks for ϵ = 0.1. Stability in time and non-trivial dependences on the degree of the node are found for both networks. In the cattle trade network, loyal farmers tend to remain loyal with a rather high probability (TLL > 0.6 for all kin values). In addition, this probability markedly increases with the degree, reaching TLL ≃ 1 for the largest values of kin. Interestingly, the probability that a disloyal farmer stays disloyal the following year dramatically decreases with the degree, reaching 0 in the limit of large degree. Among the farmers who purchase cattle batches from a large number of different premises, loyal ones have an increased chance to establish business deals with the same partners the following year, whereas previously disloyal ones will more likely turn to being loyal.
A similar qualitative dependence on the degree is also found in the sexual contact network, however in this case the probability of remaining disloyal is always very high (TDD > 0.7) even for high degrees. TLL shows a relatively more pronounced dependence on k, ranging from 0.3 (low degree nodes) to 0.6 (high degree nodes). Differently from the farmers behavior, sex buyers display a large tendency to keep a high rate of partners turnover across time. Moreover, the largest probability of preserving sexual partners is obtained when the number of partners is rather large.
Remarkably, in both networks, transition probabilities are found to be stable across time and are well described by logarithmic functions (with parameters depending on the system and on ϵ) that can be used to predict the loyalty of nodes in configuration c+1 from past data (Fig. 3C-D). With this information, it is then possible to compute the epidemic risk of a node i in configuration c+1, having degree k = k i c in configuration c and known loyalty status {L, D} between configurations c−1 and c as follows:
if loyalty class = D : ρ i c + 1 = π D c , c + 1 ( s ) T D D ( k ) + π L c , c + 1 ( s ) T D L ( k ) ; if loyalty class = L : ρ i c + 1 = π D c , c + 1 ( s ) T L D ( k ) + π L c , c + 1 ( s ) T L L ( k ) . (2)
It is important to note that in our framework the epidemic risk is a node property, and not a global characteristic of a specific disease.
To validate our strategy of risk assessment, we test our predictions based on past data for the risk of being infected in configuration c+1 on the results of an epidemic simulation explicitly performed on the supposedly unknown configuration c+1. We consider the set of nodes 𝓘 s c for which we are able to provide risk predictions and divide it into two subsets, according to their predicted risk of infection ρ i c + 1. We indicate with 𝓘 s , h c the top 25% highest ranking nodes, and with 𝓘 s , l c all the remaining others. We then compute the fraction Ph of nodes in the subset 𝓘 s , h c, i.e. predicted at high risk, that belong to the set of infected nodes 𝓘 s c + 1 in the simulated epidemic aimed at validation. Analogously, Pl measures the fraction of nodes in 𝓘 s , l c that are reached by the infection in the simulation on c+1. In other words, Ph (Pl) represents the probability for a node having a high (low) risk of infection to indeed get infected. The accuracy of the risk assessment analysis can thus be measured in terms of the relative risk ratio ν = Ph/Pl, where values ν ≤ 1 indicate negative or no correlation between our risk predictions and the observed infections, whereas values ν > 1 indicate that the prediction is informative. For both networks we find a significant correlation, signaled by the distributions of the relative risk ratio ν peaking around values ν > 1 (Fig. 4A-B). The peak positions (ν ≃ 1.4 and ν ≃ 1.7 for cattle and sex, respectively) are remarkably close to the benchmark values represented by the distributions computed on the training sets (red lines in Fig. 4A-B). In addition, the comparison with the distributions from a null model obtained by reshuffling the infection statuses of nodes (dotted curves peaking around ν = 1 in Fig. 4A-B) further confirms the accuracy of the approach. Findings are robust against changes of the value used to define 𝓘 s , h c or against alternative definitions of this quantity (see S1 Text).
One other important aspect to characterize is the predictive power of our risk assessment analysis. Our predictions indeed are limited to the set 𝓘 s c of nodes that are reached in the simulation performed on past data, proxy for the future outbreak. If a node is not infected by the simulation unfolding on configuration c or it is not active at that given time, our strategy is unable to provide a risk assessment for that node in the future. We can then quantify the predictive power ω as the fraction of infected nodes for which we could provide the epidemic risk, i.e. ω s c , c + 1 = ∣ 𝓘 s c + 1 ∩ 𝓘 s c ∣ / ∣ 𝓘 s c + 1 ∣. High values of ω indicate that few infections are missed by the risk assessment analysis. Fig. 4C-D display the distributions P(ω) obtained for the two case studies, showing that a higher predictive power is obtained in the cattle trade network (peak at ω ≃ 60%) with respect to the sexual contact network (peak at ω ≃ 40%). Our methodology can potentially be applied to a wide range of networks, other than the ones presented here, as shown with the example of human face-to-face proximity networks relevant for the spread of respiratory diseases reported in S1 Text.
We also tested whether our risk measure represents a significant improvement in prediction accuracy with respect to simpler and more immediate centrality measures (namely, the degree). Through a multivariate logistic regression, in S1 Text we show that our definition of node risk is predictor of infection even after adjusting for node degree.
The results of the risk assessment analysis obtained from the application of our strategy to the two empirical networks show qualitatively similar results, indicating that the approach is general enough to provide valuable information on the risk of infection in different settings. The observed differences in the predictive power of the approach are expected to be induced by the different temporal behavior of the two systems, resulting in a different amount of memory in preserving links (Fig. 1) and different loyalty of nodes and their time-variations (Fig. 2 and 3C-D).
In order to systematically explore the role of these temporal features on the accuracy and predictive power of our approach, we introduce a generic model for the generation of synthetic temporal networks. The model is based on a set of parameters that can be tuned to reproduce the empirically observed features of the two networks, i.e.: (i) the topological heterogeneity of each configuration of the network described by a stable probability distribution (Fig. 1A-B); (ii) a vital dynamics to allow for the appearance and disappearance of nodes; (iii) a tunable amount of memory characterizing the time evolution of the network contacts (Fig. 1C-D). These specific properties differentiate our approach from the previously introduced models that display instantaneous homogeneous properties for network configurations [56, 57, 58, 59], reproduce bursty inter-event time distributions but without the explicit introduction of memory [33, 60, 61] or of its control [58].
Based on an iterative network generation approach (see Materials and Methods), we can build an arbitrarily large number of configurations of networks with 104 nodes. They are characterized by stable in-degree and out-degree heterogeneous distribution across time (Fig. 5A where high memory and low memory regimes are displayed) and by profiles for the probability distribution of the loyalty as in the empirical networks (Fig. 5B). The number of nodes with zero loyalty can be computed analytically (see Materials and Methods) and it is confirmed by numerical findings (see S1 Text). A high memory regime corresponds to having nodes in the system that display a highly loyal behavior (e.g., θ > 0.7), whereas values in the range θ ∈ [0.7, 1) are almost absent in a low memory regime, in agreement with the findings of Fig. 2.
Applying the introduced risk assessment analysis to the synthetically generated temporal network, we recover a significant accuracy for both memory regimes (Fig. 5C). Different degrees of memory are however responsible for the fraction of the system for which a risk assessment can be made. In networks characterized by higher memory, the distribution of the predictive power ω has a well defined peak, whereas for lower memory it is roughly uniform in the range ω ∈ [0, 0.4] (Fig. 5D). Such a regime implies that not enough structure is maintained in the system to control more than 40% of the future infections. Our risk assessment analysis allows therefore accurate predictions across varying memory regimes characterizing the temporal networks, but the degree of memory impacts the amount of predictions that can be made. The model also shows that the analysis is not affected by the choice of the aggregating time window used to define the network configurations [61, 62, 63], as long as the heterogeneous topological features at the system level and the heterogeneous memory at the node level are kept across aggregation, as observed for the empirical networks under study (see [19] and S1 Text).
We introduce a simple measure to characterize the amount of memory in the time evolution of a networked system. The measure is local and it is empirically motivated from two case studies relevant for disease transmission. By focusing on the degree of loyalty that each node has in establishing connections with the same partners as time evolves, we are able to connect an egocentric view of the system (the node’s strategy in establishing its neighborhood over time) to the system’s larger-scale properties characterizing the early propagation of an emerging epidemic.
We uncover a non-trivial correlation between the loyalty of a node and its risk of being infected if an epidemic occurs, given fixed epidemiological conditions, and use this to inform a risk assessment analysis applicable to different settings with no information on the network evolution dynamics. A theoretical model generating synthetic time-varying networks allows us to frame the analysis in a more general perspective and disentangle the role of different features. The accuracy of the proposed risk assessment analysis is stable across variations of the temporal correlations of the system, whereas its predictive power depends on the degree of memory kept in the time evolution. The introduced strategy can be used to inform preventive actions in preparation to an epidemic and for targeted control responses during an outbreak emergency, only relying on past network data.
The cattle trade network is obtained from the database of the Italian national bovine registry recording all cattle displacements due to trade transactions. We consider animal movements during a 5 years time period, from 2006 to 2010, involving 215,264 premises and 2,973,710 directed links. Nodes may be active or inactive depending whether farms sell/buy cattle in a given timeframe. The cattle network is available as S1 Dataset. From the dataset we have removed slaughterhouses (∼ 1% of the nodes) as they are not relevant for transmission.
The sexual contact network is extracted from an online Brazilian forum where male sex buyers rate and comment on their sexual encounters with female sex sellers [16]. Time-stamped posts are used as proxies for sexual intercourse and multiple entries are considered separately, following previous works [16, 31]. A total of 13,855 individuals establishing 34,509 distinct sexual contacts are considered in the study, after discarding the initial transient of the community growth [16]. Nodes may be active or inactive depending whether individuals use or not the service, and join or quit the community. Six-months aggregating snapshots are chosen. A different aggregating time window of three months has been tested, obtaining similar results (see S1 Text).
The distributions of the risk potentials πL and πD reported in Fig. 3 are modeled with a sum of Landau distribution and an exponential suppression. This family of functions depending on four parameters (see S1 Text for the specific functional form) was chosen as it well reproduces the distribution profiles of the risk potentials, and it was used to compute the nodes’ epidemic risk. A goodness of fit was not performed, as this choice was automatically validated in the validation analysis performed on the whole prediction approach.
The basic iterative network generation approach allows to build configuration c+1 from configuration c through the following steps:
vital dynamics: nodes that are inactive in configuration c become active in c+1 with probability b, while active nodes become inactive with probability d;
memory: active nodes maintain same in-neighbors each with probability pα; then they form βin new in-stubs, where βin is extracted from a power-law distribution: P ( β i n ) ∼ β i n − γ;
out-degree heterogeneity: each node is assigned βout out-stubs, where βout is drawn from another power-law distribution: P ( β o u t ) ∼ β o u t − δ. Then each of the in-stubs is randomly matched to an out-stub.
The total set of nodes is considered to be fixed in time, and nodes may be active (i.e. establishing connections) or inactive (i.e. isolated) in a given configuration. All five parameters b, d, γ, pα, δ are assumed constant in time and throughout the network. The amount of memory in the system is tuned by the interplay of the two parameters pα and d. Starting from an arbitrarily chosen initial configuration c = 0, simulations show that the system rapidly evolves towards a dynamical equilibrium, and successive configurations can be obtained after discarding an initial transient of time. The parameters values used in the paper are: N = 104; b = 0.7; d = 0.2; γ = 2.25; δ = 2.75; pα = 0.3, 0.7. The influence of such parameters on the network properties is examined in S1 Text.
If we denote with α the number of neighbors that a given node keeps across two consecutive configurations (c−1, c), we can express the loyalty simply as:
θ i c − 1 , c = α c − 1 , c k i c + β i n c (3)
where the superscript c for α, βin indicate the values used to build configuration c. The number of nodes with θ = 0 as a function of the degree can be computed analytically: P ( θ c , c + 1 = 0 ) = d + ( 1 − d ) ( 1 − p α ) k c. Similarly, it is possible to compute the probability fc, c+1 that a link present in configuration c is also present in configuration c+1. In the S1 Text we show that fc, c+1 ≃ (1−d)pα and confirm this result by numerical simulations.
|
10.1371/journal.pgen.1003817 | Fragile Site Instability in Saccharomyces cerevisiae Causes Loss of Heterozygosity by Mitotic Crossovers and Break-Induced Replication | Loss of heterozygosity (LOH) at tumor suppressor loci is a major contributor to cancer initiation and progression. Both deletions and mitotic recombination can lead to LOH. Certain chromosomal loci known as common fragile sites are susceptible to DNA lesions under replication stress, and replication stress is prevalent in early stage tumor cells. There is extensive evidence for deletions stimulated by common fragile sites in tumors, but the role of fragile sites in stimulating mitotic recombination that causes LOH is unknown. Here, we have used the yeast model system to study the relationship between fragile site instability and mitotic recombination that results in LOH. A naturally occurring fragile site, FS2, exists on the right arm of yeast chromosome III, and we have analyzed LOH on this chromosome. We report that the frequency of spontaneous mitotic BIR events resulting in LOH on the right arm of yeast chromosome III is higher than expected, and that replication stress by low levels of polymerase alpha increases mitotic recombination 12-fold. Using single-nucleotide polymorphisms between the two chromosome III homologs, we mapped the locations of recombination events and determined that FS2 is a strong hotspot for both mitotic reciprocal crossovers and break-induced replication events under conditions of replication stress.
| Loss of heterozygosity (LOH) at tumor-suppressor genes contributes to cancer, and deletions resulting in LOH are frequently observed in tumor cells at certain chromosomal regions known as common fragile sites. LOH can also result from repair of DNA damage by mitotic recombination, if the homologous chromosome rather than the sister chromatid is used as a repair template. The extent to which fragile site instability causes LOH by mitotic recombination with the homologous chromosome is unknown. We evaluated mitotic recombination on the yeast Saccharomyces cerevisiae chromosome III, which contains a naturally-occurring fragile site known as FS2. We report that yeast chromosome III has a high frequency of spontaneous mitotic recombination that involves the homologous chromosome. Under conditions that stimulate instability at the fragile site, LOH resulting from mitotic recombination on yeast chromosome III is increased 12-fold, and FS2 is a hotspot for initiating these events. These results suggest that instability at human common fragile sites may drive mitotic recombination repair pathways that cause LOH and promote tumorogenesis.
| Cancer cells contain a variety of genomic changes that result in altered gene expression affecting cell growth. Amplification or over-expression of oncogenes and loss of heterozygosity (LOH) at tumor-suppressor genes are both significant contributors to tumorogenesis. Human common fragile sites have been extensively investigated for their contribution to genomic changes that cause tumor initiation and progression. Common fragile sites are large genomic regions of 250 kb–1 Mb that are unstable under conditions that partially inhibit DNA replication (reviewed in [1]). Treatment with aphidicolin, which inhibits DNA polymerases [2], [3], or hydroxyurea, which inhibits ribonucleotide reductase and results in unbalanced nucleotide pools [4], both cause replication stress that induces instability at fragile sites. Several mechanisms have been proposed to explain why breaks form in human common fragile sites, including secondary structure formation within single-stranded DNA (ssDNA) at stalled replication forks [5], [6], paucity of replication origins [7], [8], replication fork pausing between early- and late-replicating regions [9], [10], and collision between RNA and DNA polymerases [11]. Multiple mechanisms may contribute to breaks, and each mechanism may be responsible for breaks at a particular site or group of sites. The mutations at common fragile sites appear to often be early drivers of tumorogenesis rather than later “passenger” events [12]–[14]. This may be because replication stress resulting from nucleotide deficiency and oncogene-induced hyper replication occurs early in the progression of cancer [15]–[17].
Research to date has focused on the ability of common fragile sites to cause deletions at tumor-suppressor genes, initiate oncogene amplification by breakage-fusion-bridge cycles, generate non-reciprocal translocations, and promote integrations of human papilloma virus (reviewed in [18]). However, common fragile sites are also hotspots for sister chromatid exchange [19], and down-regulation of Rad51 in human cells, a key protein in homologous recombination, results in increased gaps and breaks at common fragile sites [20], which suggests the potential for fragile site lesions to also cause LOH through homologous recombination. Double-strand breaks are the canonical inducer of homologous recombination, but this repair pathway can also be stimulated by single-strand gaps and stalled replication forks, lesions that are likely to occur at fragile sites [21]–[23]. Homologous recombination in mitosis favors use of the sister chromatid as a repair template and use of non-crossover resolution pathways [24]–[27], but inter-homolog events can occur and result in LOH from crossovers, break-induced replication (BIR), and local gene conversion events [28], [29]. Mitotic recombination events that cause LOH have been understudied, and it is unknown to what extent the replication stress present early in cancer development causes LOH by mitotic recombination, and whether fragile sites contribute to these events.
In Saccharomyces cerevisiae, a fragile site named FS2 was identified on chromosome III [30]. FS2 is composed of two, 6 kb Ty1 elements in inverted orientation separated by ∼280 bp. Like human fragile sites, FS2 is a hotspot for double-strand breaks under conditions of DNA replication stress when DNA polymerases are partially impeded [30], [31]. In cells with normal levels of polymerase, FS2 is more stable but it is a hotspot for BIR events leading to non-reciprocal translocations between Ty1 elements, indicating that the fragile site is active even in the absence of replication stress [30], [32], [33]. In cells with low levels of DNA polymerase, it is likely that long stretches of single-stranded DNA form at the replication fork, which we hypothesize allows the inverted Ty1 elements of FS2 to self-pair into a hairpin structure, and cleavage of this hairpin results in a DSB [30], [34]. Here, we have used this yeast model to examine the role of fragile site instability in stimulating LOH during mitosis. In diploid cells, we determined the frequency of mitotic recombination events on chromosome III occurring spontaneously and under conditions of replication stress by low levels of polymerase alpha. Frequent single-nucleotide polymorphisms (SNPs) between the two chromosome III homologs were used to map the location of crossovers and BIR events. We find that chromosome III has a higher than expected level of spontaneous mitotic BIR, compared to reports for chromosomes IV, XII, and XV [35]–[37], and that replication stress elevates mitotic recombination by 12-fold. Reciprocal crossovers and BIR events occur at approximately equal frequencies under replication stress, and fragile site FS2 is a strong hotspot for causing LOH by both of these types of events. Our analysis of gene conversions associated with crossovers indicates that lesions at FS2 during replication, and not during G1, are the primary stimulation for these mitotic recombination events.
The naturally-occurring fragile site FS2 is located on S. cerevisiae chromosome III [30]. To evaluate mitotic recombination stimulated by this fragile site, we constructed diploids based on the detection system developed to study mitotic crossovers on yeast chromosome V [38], [39] (also see Text S1). An event that causes loss of heterozygosity at the SUP4-o locus in a mitotic division at the time of plating results in a red/white or red/light pink sectored colony. Therefore by their nature, each sectored colony represents an independent event.
The relevant features of the five diploid strains we created are shown in Figure 1. These strains are homozygous for ade2-1, which is an ochre stop codon null mutant allele. Cells with mutant ade2 are adenine auxotrophs and appear red due to a build-up of a red precursor in the metabolic pathway for adenine synthesis. We inserted a single copy of the tRNA ochre suppressor SUP4-o on the right arm of one homolog of chromosome III approximately 159 kb distal to the centromere. SUP4-o suppresses the ochre stop mutation; therefore the diploids are adenine prototrophs and light pink in color. The diploids are also homozygous for the GAL-POL1 construct, except for strains AMC324 and AMC331 [30], [34]. This construct links the GAL1/10 promoter to the POL1 gene, so that the level of Pol1p in the cell is regulated by galactose in the growth medium, which allows us to induce replication stress and instability at FS2. Under high galactose conditions (0.05%), the level of Pol1p is approximately 300% of wild-type levels, and under low galactose conditions (0.005%), it is limited to approximately 10% of wild-type levels, thereby putting the cell under replication stress [30].
Sectored colonies can result from several types of events that cause loss of heterozygosity at the SUP4-o locus in our diploids: crossover, BIR, local gene conversion, and chromosome loss. A crossover between the chromosome III centromere and SUP4-o is diagrammed in Figure 2A. As shown, one daughter cell is homozygous SUP4-o/SUP4-o, and the other daughter cell lacks this gene. Only half of crossover events are detected, due to chromosome segregation patterns. If the two recombined chromosomes segregate together in cell division, no red/white sectoring will occur. The two possible segregation patterns are equally likely in yeast [40] therefore the frequency of crossovers observed in our experiments is multiplied by two to obtain the total frequency of crossovers. Sectoring can also result from a BIR event initiated between the centromere and SUP4-o that proceeds centromere-distal from invasion, local gene conversion at SUP4-o, or loss of the chromosome containing SUP4-o (Figures 2B, 2C, and 2D), although in these cases the sectoring is red/light pink. BIR initiated by a lesion in the homolog that does not contain SUP4-o results in white/light pink sectoring. This color difference is difficult to consistently detect, and therefore white/light pink sectors were not examined. A BIR event initiated on the right arm that proceeds centromere-proximal would not be detected; however, this type of event is unlikely because BIR is impeded by the centromere [41]. Loss of HygR in the red side of a sectored colony suggests chromosome loss (Figure 2), although BIR or crossover on the left arm of chromosome III can affect this phenotype. A point mutation in SUP4-o also results in red/light pink sectoring (not shown).
Our diploids have ∼0.5% sequence divergence between homologous chromosomes, as a result of mating a haploid derived from YJM789 with an S228c-related haploid [42]. This divergence in sequence does not cause a significant change in the rate of mitotic crossovers [39]. In our diploids, the S228c-related haploid is MS71, and it contains fragile site FS2 on chromosome III [31]. The YJM789-derived chromosome III does not contain FS2; therefore, to provide homology for recombination, we inserted one Crick-orientation Ty1 element in the corresponding location on this chromosome. We used single nucleotide polymorphisms (SNPs) between homologs that change a restriction enzyme site to map and analyze recombination events.
Because BIR events and chromosome loss are readily detectable in our system as red/light pink sectors only when the initiating lesion occurs on the homolog containing SUP4-o, we created two different experimental diploid strains (Figure 1). In Experimental Diploid #1, both SUP4-o and FS2 are on the MS71-derived homolog of chromosome III, which allows us to evaluate the frequency of BIR and chromosome loss from initiating lesions on this homolog. To evaluate BIR and chromosome loss that result from initiating lesions on the YJM789-derived chromosome III, which does not contain FS2, we created Experimental Diploid #2 (strain AMC 310) by moving SUP4-o to the YJM789-derived homolog.
Spontaneous mitotic events on chromosome III were initially evaluated in Experimental Diploid #1 (Y332) grown in medium with high galactose. This diploid is homozygous for the GAL-POL1 construct and the single copy of SUP4-o is located on the same homolog of chromosome III as fragile site FS2 (Figure 1). We identified 31 sectored colonies among 30,543 total colonies. The event responsible for each sectored colony was determined through a combination of phenotype analysis and SNP genotyping, and frequencies for all event classes are reported in Table 1.
In Experimental Diploid #1 on high galactose, the total frequency of spontaneous mitotic events resulting in LOH on the right arm of chromosome III is 115×10−5. We observed three categories of events: reciprocal crossovers, BIR, and chromosome loss. The spontaneous frequency of crossovers is 26×10−5. Since the interval between CEN3 and SUP4-o is 159 kb, this is 1.65×10−6 crossovers per kb. The frequency of spontaneous BIR events initiated between CEN3 and SUP4-o is 46×10−5, or 2.89×10−6 BIR events per kb. Because BIR is unidirectional in its transfer of genetic information, only BIR initiated by a lesion in the homolog containing SUP4-o results in red/light pink sectoring (Figure 2). Initiating lesions on the other homolog that are repaired by BIR result in white/light pink sectoring that is not easily detected. Therefore, BIR in Experimental Diploid #1 reported in Table 1 reflects only events initiated by a break on the MS71-derived homolog of chromosome III, which contains both SUP4-o and fragile site FS2. Similarly, loss of the SUP4-o containing homolog of chromosome III is detectable by red/light pink sectoring. The frequency of spontaneous loss of the MS71-derived chromosome III is 43×10−5 in Experimental Diploid #1.
Cells with GAL-POL1 grown on high galactose contain an excess of Pol1p and have a modest increase in instability at fragile site FS2 relative to strains with POL1 under its native promoter [30]. To evaluate the effect of excess Pol1p on mitotic recombination, we created Control Diploid #1, which is isogenic to Experimental Diploid #1 but homozygous for POL1 under its native promoter. After growth in medium with high galactose, the total frequency of spontaneous mitotic events resulting in LOH is reduced by half in this control diploid compared to Experimental Diploid #1 on high galactose (p = 0.0187) (Table 1). However, the relative proportions of each type of mitotic event (crossover, BIR, and chromosome loss) are not significantly different between these two diploids (p = 0.089).
FS2 is a hotspot for Ty1-mediated translocations under normal polymerase conditions [32], [33]. To evaluate the effect of FS2 instability on mitotic recombination in cells with normal levels of POL1, we modified Control Diploid #1 by replacing the entire FS2 region on the MS71-derived homolog, including both Ty1 elements and the nucleotides between them, with the NAT gene [43]. The same region on the YJM789-derived homolog was also replaced with the NAT gene. This diploid is referred to as Control Diploid #2. We found that there is no difference in the total frequency of spontaneous mitotic LOH events on the right arm of chromosome III between Control Diploids #1 and #2 after growth in medium with high galactose (p = 1.0) (Table 1).
Partial inhibition of replication by lowering the level of DNA polymerase alpha causes breaks on yeast chromosome III at fragile site FS2. In haploid cells these breaks are frequently repaired by BIR or result in loss of chromosome III and can be detected by increased illegitimate mating [30], [34]. In diploid cells, low polymerase alpha increases mitotic reciprocal crossovers within the yeast rDNA array by 7-fold [44]. Here, we have further evaluated the role of replication stress in stimulating events that cause LOH in diploid cells, and the role of fragile site instability in initiating these events. To study stress-induced mitotic events on yeast chromosome III, we grew Experimental Diploid #1 in medium with no galactose for six hours to lower the level of polymerase alpha, followed by plating on high galactose. We identified 140 sectored colonies among 22,640 total colonies. Replication stress in this diploid increases the total frequency of mitotic LOH events by 6.5-fold relative to high galactose conditions (p<0.001) (Table 1). However, the relative proportions of the categories of crossover, BIR, and chromosome loss in this diploid are the same in both high galactose and no galactose (p = 0.463).
As explained above, BIR events and chromosome loss are readily detectable in our system only when the initiating lesion occurs on the homolog containing SUP4-o. In Experimental Diploid #1, both FS2 and SUP4-o are on the MS71-derived homolog. We used Experimental Diploid #2, which has SUP4-o on the non-FS2 containing YJM789-derived homolog, to evaluate the frequency of BIR and chromosome loss from initiating lesions on this homolog of chromosome III. We grew this diploid in medium lacking galactose for six hours to induce replication stress then plated cells on high galactose. We identified 47 sectored colonies among 14,876 total colonies. Under replication stress, the total frequency of mitotic LOH on chromosome III in Experimental Diploid #2 is half that of Experimental Diploid #1 (p<0.001) (Table 1). The frequency of crossovers is similar in Experimental Diploids #1 and #2 under replication stress (p = 0.543) (Table 1). This result is consistent with our expectation, given that crossovers are detected in our system irrelevant of which chromosome III homolog contains the initiating lesion. The frequency of replication stress-induced BIR is one-third lower in Experimental Diploid #2 than in Experimental Diploid #1 (p = 0.0629). This difference results from the absence of FS2 in the SUP4-o marker homolog. In Experimental Diploid #2, only BIR events that are initiated by lesions on the YJM789-derived chromosome can be detected as red/light pink sectors. Events caused by a lesion at FS2 on the MS71-derived homolog of chromosome III will result in white/light pink sectoring, which is not easily detected and thus not scored in this diploid. The lower frequency of BIR in Experimental Diploid #2 suggests that FS2 instability drives 1/3 of the stress-induced BIR observed in Experimental Diploid #1. There is an even stronger reduction in the frequency of chromosome III loss in Experimental Diploid #2 under replication stress, such that the loss frequency is below that observed in Experimental Diploid #1 with high levels of polymerase. Therefore, the primary cause of chromosome III loss in Experimental Diploid #1 is FS2 instability.
To further evaluate the role of FS2 instability in driving mitotic LOH under replication stress, we created Control Diploid #3 (Y382), in which we stabilized fragile site FS2. Normally, the inverted Ty1 elements at FS2 are separated by ∼280 bp. We inserted the NAT gene [43] between these two Ty1 elements, separating them by ∼1.8 kb. The increased distance effectively stabilizes the fragile site [30]. We grew Control Diploid #3 in medium with no galactose for six hours to induce replication stress conditions, and then plated cells on high galactose medium. We identified 7 sectored colonies among 21,666 total colonies, and analyzed these sectored colonies as before. Under replication stress, the total frequency of mitotic LOH on chromosome III in Control Diploid #3 is less than half that of Experimental Diploid #1 (p<0.0001) (Table 1). However, the relative proportions of the categories of crossover, BIR, and chromosome loss events in these diploids is similar (p = 0.1106).
We used the ∼0.5% sequence divergence between the two homologs of chromosome III in our diploids to map the locations of the events causing sectoring. This divergence results in many single nucleotide polymorphisms (SNPs) between the homologs, some of which alter restriction enzyme sites. We purified a single cell from each half of the sectored colony and evaluated a set of 27 SNPs on the right arm of chromosome III by PCR and restriction enzyme digest (Table S3). On average, these SNPs are spaced 6.9 kb apart. The closest SNP centromere-distal to FS2 that changes a restriction enzyme site is at chromosome III base 175324; this is 5.4 kb from the end of the fragile site. The closest SNP centromere-proximal to FS2 is at base 167720; this is 0.8 kb from the end of the fragile site. As shown in Figure 3A, in the case of BIR, all SNPs in the light pink cell remain heterozygous while in the red cell, SNPs proximal to the event remain heterozygous and SNPs distal to the event are homozygous for the homolog lacking SUP4-o. In the case of BIR events initiated by a lesion in unique sequence, we assume that invasion of the broken end into the corresponding region on the homologous chromosome results in homozygosity for SNPs distal to the BIR. However, if the broken end occurs near a Ty1 element, it may invade a Ty1 or Ty2 element on a non-homologous chromosome to initiate replication. In such cases of non-allelic repair, SNPs distal to the BIR would be hemizygous.
In Experimental Diploid #1 we detect only BIR initiated by lesions on the MS71-derived homolog of chromosome III, which contains both SUP4-o and FS2. As anticipated, only YJM789-derived SNPs are present in the red cell distal to each BIR event. Our SNP mapping indicates that fragile site FS2 is a hotspot for initiation of BIR in this diploid (Figure 4A). Of 66 total BIR events under replication stress, 18 were initiated between the SNPs flanking FS2. BIR can be initiated centromere-proximal to a break location due to exonuclease processing at the break that usually exposes 3–6 kb of ssDNA [45]; therefore, the nine BIR events initiated between the pair of SNPs immediately centromere-proximal to FS2 likely also result from lesions at FS2, for a total of 27/66 events (41%) stimulated by FS2. We evaluated the significance of this distribution by dividing the CEN3 – SUP4-o interval into four equal-sized bins of 39.7 kb, then counting the number of BIR events initiated within each bin. There are 16 events in bin #1 (CEN3 to SNP152), 43 in bin #2 (SNPs 164 to193), 5 in bin #3 (SNPs 195 to 233), and 2 in bin #4 (SNPs 246 to SUP4-o). This distribution is significantly different from random (p<0.0001 by chi-square goodness-of-fit). To determine whether BIR events were allelic or non-allelic, we determined the sizes of chromosome III in a subset of 35 BIR events from Experimental Diploid #1 under replication stress. Allelic events will produce chromosome III repair products of normal size, and non-allelic events will produce a chromosome that may be smaller or larger than the normal chromosome III size. Intact yeast chromosomes from each event were separated by pulsed-field gel electrophoresis, and chromosome size was evaluated by Southern blotting with a CHA1 probe to the right arm of chromosome III. Of events tested that were initiated between the SNPs flanking FS2 or within 6 kb proximal of FS2, 7/18 (39%) had non-allelic BIR products (Figure S1). All of the BIR events tested that were initiated more than 6 kb proximal of a Ty1 element were allelic.
In high galactose conditions that permit high levels of POL1 transcription, the number of BIR events initiated in Experimental Diploid #1 at or within 6 kb proximal to FS2 is reduced to 4/14 (29%) (Figure 4B). This fact that this reduction is relatively modest is likely attributable to excessive polymerase alpha causing FS2 instability in this diploid, because Control Diploid #1, which has POL1 under its native promoter, and Control Diploid #3, which has a stabilized version of FS2, do not have any BIR events initiated at or within 6 kb proximal of FS2. We note that the pair of tandem-oriented Ty1 elements centromere-proximal to FS2 on the MS71-derived homolog is not a BIR hotspot in Experimental Diploid #1 (Figure 4A), although this was a frequent site of recombination in the illegitimate mating assays previously used to study fragile sites on yeast chromosome III [30], [34]. This difference will be further discussed below.
In Experimental Diploid #2, 10/28 BIR events (36%) were initiated at or within 6 kb proximal of the location allelic to FS2 (Figure 4C). In this diploid, only events initiated by a lesion on the YJM789-derived homolog are detected. As before, we evaluated the significance of this distribution by dividing the CEN3 – SUP4-o interval into four equal-sized bins of 39.7 kb, then counting the number of BIR events initiated within each bin. There are 7 events in bin #1, 15 in bin #2, 4 in bin #3, and 2 in bin #4. This distribution is significantly different from random (p = 0.0029 by chi-square goodness-of-fit). Therefore, despite the fact that FS2 is not present on the YJM789-derived homolog, the site allelic to this fragile site is a hotspot for initiation of BIR events. The YJM789 homolog of chromosome III contains a pair of inverted delta elements (the ∼300 bp long terminal repeat portion of Ty1 elements) at the location allelic to FS2. The spacing between these inverted deltas is the same as between the Ty1 elements of FS2. As explained above, we modified the YJM789 homolog to expand the Crick-orientation delta element to a full Ty1, to provide homology for recombination without creating a fragile site. However, the inverted delta elements also have the potential for intra-strand base pairing to form a hairpin under conditions of replication stress. Since the overall frequency of stress-induced BIR is lower in Experimental Diploid #2 than in Experimental Diploid #1, the frequency of BIR stimulated by the “full” version of FS2 is higher than that stimulated by the “delta only” version of FS2 (frequencies of 123×10−5 and 67×10−5 FS2-stimulated BIR, respectively). There were three BIR events in Experimental Diploid #2 that had adjacent gene conversion tracts; two with a 4∶0 tract (SC100 and SC104) and one with a 3∶1 tract (SC121) (Figure S2). Gene conversion associated with BIR has previously been reported, and appeared to result from repair of two double-strand breaks in the same location on both sister chromatids, [35], [36]. The 3∶1 tract observed here does not fit that mechanism, and may instead represent repair of heteroduplex mis-matches in the region of invasion for BIR initiation. The 4∶0 tracts are unusual and may represent an internal deletion prior to BIR initiation.
Figure 3B shows an example of the SNP pattern in a sectored colony from a reciprocal crossover on the right arm of chromosome III. For crossovers un-associated with gene conversion, SNPs proximal to the crossover remain heterozygous, and distal to the crossover, are homozygous for the homolog lacking SUP4-o in the red cell, and homozygous for the homolog containing SUP4-o in the white cell. Gene conversion that is associated with a crossover can be of two types, either a typical 3∶1 segregation in which SNPs are heterozygous in one cell and homozygous in the other (as shown in Figure 3B), or a 4∶0 pattern in which SNPs are homozygous for the same version in both the red and white cells [39]. The 3∶1 conversions appear to result from repair of damage that occurs during S-phase and 4∶0 conversions result from DNA double-strand breaks that occur during G1 that are replicated, followed by repair of both broken sister chromatids in G2 using the unbroken homolog as a template [46].
As shown in Figure 5A, our SNP mapping results indicate that fragile site FS2 is a hotspot for crossover events under replication stress caused by low Pol1p. We identified 41 crossover events in Experimental Diploid #1 under stress (Figure 5B). These crossover events were collected in two ways; 29 crossover events were collected among the 22,640 colonies in Table 1 that were fully analyzed for crossover, BIR, and chromosome loss events, and 12 crossover events were collected among another set of 14,792 colonies that was not fully analyzed for BIR and chromosome loss events. Of the 41 crossovers in Experimental Diploid #1, 21 have no associated gene conversion tract, 19 have a gene conversion adjacent to the crossover, and one has a conversion tract that is not contiguous with the crossover. Of the 21 crossovers without gene conversion, 8 occur between the SNPs flanking FS2. Of 19 crossovers with adjacent gene conversion tracts, 12 tracts cross a SNP flanking FS2. Therefore, 20/41 crossover events (49%) in Experimental Diploid #1 under replication stress are associated with FS2. The crossover data from Experimental Diploid #2 under replication stress is similar to Experimental Diploid #1, which is consistent with the expectation that our system detects all crossovers between CEN3 and SUP4-o irrespective of which homolog contains SUP4-o or which homolog has the initiating lesion. In Experimental Diploid #2, 8/15 events (53%) are associated with FS2 (Figure 5B). Of 15 total crossover events, 11 are unassociated with gene conversion, and 6 of these occur between the SNPs flanking FS2. Of the 4 crossovers with adjacent gene conversion, two have tracts that cross a SNP flanking FS2; these two tracts have information transferred from the non-FS2 containing homolog indicating the initiating event was at or near the fragile site (Figure 5B).
In Experimental Diploid #1 in high galactose conditions, five crossovers were collected. These crossover events were collected in two ways; 4 crossover events were collected among the 30,543 colonies in Table 1 that were fully analyzed for crossover, BIR, and chromosome loss events, and 1 crossover event was collected among another set of 4,792 colonies that was not fully analyzed for BIR and chromosome loss events. Of these five crossovers, one is located at FS2 (Figure 5D). This crossover is associated with a gene conversion tract in which the transfer of genetic information indicates that the initiating event is on homolog with the “delta-only” FS2. In Control Diploid #1, which has POL1 under its native promoter, two of the five crossovers are located between the nearest SNPs flanking FS2 (Figure 6). Although the number of events detected under high galactose and in Control #1 is low, it is intriguing that 20–40% of crossovers in these diploids were at FS2. We address this result in the discussion below. In Control Diploid #2, which has POL1 under its native promoter and no FS2, the single crossover detected was not near the deleted fragile site, and in Control Diploid #3, which is GAL-POL1 and has a stabilized version of FS2, one of the three crossovers was at FS2 (Figure 6).
Several characteristics of the crossover-associated gene conversion tracts in Experimental Diploids #1 and #2 under replication stress are of interest. First, 12 of the 14 tracts crossing a SNP at FS2 have three copies of the information from the chromosome lacking FS2. This result is consistent with damage at FS2 responsible for crossover stimulation, since in both mitotic and meiotic recombination events, the damaged chromosome typically receives genetic information from the unbroken homolog [47]–[49]. However, the two tracts that were stimulated by an initiating lesion on the YJM789-derived homolog are consistent with our BIR results above, in which a “delta-only” version of FS2 is capable of stimulating a lower level of recombination than the “full” version of FS2. Second, of 24 total tracts, only two are 4∶0 type tracts and the others are 3∶1. The 3∶1 conversions have been reported to result from repair of S-phase damage and the 4∶0 conversions result from DNA double-strand breaks in G1 that are replicated, followed by repair of both broken sister chromatids during G2 [46]. Therefore, our results indicate that the crossover-associated gene conversion tracts under replication stress are consistent with damage occurring primarily during S phase. Third, six of the gene conversion tracts associated with FS2 cross both SNPs flanking FS2, four cross only SNPs centromere-proximal, and four cross only SNPs centromere-distal. Therefore, repair of a lesion at FS2 that occurs during S-phase can result in gene conversion that extends either bi- or uni-directionally. Fourth, our mitotic gene conversion tracts are relatively long, with a median length of 14.7 kb (95% confidence interval of 7.0 kb to 34.5 kb) for the 23 tracts contiguous with a crossover. Both 4∶0 and 3∶1 tracts were included in our analysis of median tract length.
We have determined the frequencies of spontaneous and replication stress-induced mitotic events resulting in loss of heterozygosity (LOH) on the right arm of yeast chromosome III. Yeast fragile site FS2 is present on this chromosome, and we report that it is a hotspot for mitotic reciprocal crossovers and BIR events.
St Charles and Petes [28] defined the microStern (µS) as a unit to measure mitotic crossovers, with 10−6 crossovers/division equal to one microStern, and they estimated the entire yeast genome has a mitotic genetic map length of 620 µS. The portion of chromosome III we evaluated accounts for 1.3% of the physical yeast genome, therefore we expect a genetic map length of 8 µS. We detect a map length of 6 µS for the right arm of chromosome III in Control Diploid #2, which has normal Pol1p levels and does not contain FS2. In Control Diploid #1, which has normal Pol1p but contains FS2, we detect a map length of 310 µS, and two of the five crossover events are at FS2. These data are in accordance with reports that FS2 can be unstable under normal polymerase conditions [32], [33]. However, there is no difference in the total frequency of spontaneous mitotic LOH events between these two diploids (p = 1.0)(Table 1).
Previous studies of mitotic LOH in yeast have reported that BIR is less frequent than crossovers. On yeast chromosomes IV and XII, spontaneous BIR is three to four-fold less frequent than crossovers [35], [37], and on chromosome XV, repair by BIR of a mitotic double-strand break from an I-SceI cut site is extremely rare compared to repair that results in crossover or non-crossover outcomes [36]. The exception to this pattern is in old yeast mother cells, in which nearly 90% of spontaneous mitotic LOH results from BIR [37]. We observed that BIR is nearly 5-fold more frequent than crossovers in Control Diploid #2, and that BIR is only 20% less frequent than crossovers in Control Diploid #1. None of the BIR events in Control Diploid #1 were initiated at or near FS2, which indicates that a mechanism other than fragile site instability drives spontaneous BIR on yeast chromosome III in this strain. The BIR pathway is primarily used to repair one-ended double-strand breaks, such as those that exist at collapsed replication forks [50]. Therefore, our results may suggest a higher frequency of spontaneous replication fork stalling and collapse on the right arm of chromosome III than on other chromosomes similarly examined to date.
Here, we report that the total frequency of mitotic LOH is elevated 12-fold in Experimental Diploid #1 with low levels of polymerase, relative to Control Diploid #1 with wild-type levels of polymerase (Table 1). In our analysis, replication stress induces reciprocal crossovers, BIR, and chromosome loss with approximately equal frequency. In haploids with low levels of polymerase alpha, physical analysis of chromosome III indicates that a double-strand break at FS2 occurs in approximately 7% of cells [34]. If a similar percentage of diploid cells with low polymerase alpha have breaks at FS2, then our results indicate that LOH is a rare outcome in responding to these breaks. This is not unexpected, because LOH as a result of mitotic recombination requires crossover and BIR events involving the homologous chromosome. However, during mitosis the sister chromatid is favored as a repair template during S-phase [27], [51], [52], and crossover resolution of Holliday junctions is normally suppressed [53]. A related issue is the detection of gene conversion events at FS2 that are un-associated with crossover. Our system does not permit analysis of such events unless those gene conversions are large enough to also encompass SUP4-o. Recent studies have demonstrated that approximately 35% of conversions are crossover-associated [35], [36], [54]. Therefore, we would not expect that undetected local gene conversion events at FS2 would change the relative rarity of LOH at this site compared to the frequency of breaks.
Here, we report that FS2 is a hotspot for driving mitotic events that result in LOH on the right arm of yeast chromosome III. Unexpectedly, a smaller inverted repeat consisting of two long terminal repeat delta elements separated by the same ∼280 bp distance as between the two full Ty1 elements of FS2, is similarly a hotspot for mitotic recombination under replication stress. However, the delta-only FS2 stimulates only half as many BIR events as the full FS2. There are no other inverted delta-delta pairs on the right arm of chromosome III to investigate for fragile site activity. However, inverted pairs of delta elements have been reported to fuse and generating acentric and dicentric chromosomes in yeast when replication is impeded; faulty template switching at stalled replication forks was proposed as a mechanism to generate these rearrangements [55]. Human Alu sequences, which are similar in length to the yeast delta element, stimulate breakage and recombination when inserted in inverted orientation on yeast chromosome II, although this is strongly influenced by the distance between the repeats [56].
It is somewhat surprising that crossovers and BIR events are stimulated approximately equally at FS2. Under replication stress, it is hypothesized that extended single-stranded DNA at the replication fork allows a hairpin to form between the pair of inverted Ty1 elements of FS2 [30], [34], and cleavage at this secondary structure would result in replication fork collapse to a one-ended double-strand break, which should primarily drive BIR (Figure 7) [50]. Crossover formation requires a double Holliday junction intermediate. The stimulation of crossovers at FS2 is primarily replication-dependent, because the gene conversion tracts adjacent to crossovers are nearly all of the 3∶1 type. Two possible ways that a double Holliday junction intermediate could form at FS2 are (1) template switching at a stalled replication fork or single-strand gap left at FS2 during replication, or (2) convergence of a collapsed fork with replication from a nearby origin, producing a canonical double-strand break (Figure 7) [22], [23]. Although physical analysis of chromosome III demonstrates that double-strand breaks do form at FS2 under replication stress [34], it is unclear whether breaks are the initiating lesion for crossovers at this fragile site, since template switching during replication can generate a double Holliday junction in the absence of a break.
Mitotic crossovers are rare in cells with wild-type levels of polymerase, but of the five events we collected in Control Diploid #1, two were at FS2 and did not have an adjacent 3∶1 gene conversion tract. The two inverted Ty1, Ty2 pairs on chromosome IV have been reported as hotspots for spontaneous crossovers [28]. Crossover events at these are usually associated with 4∶0 gene conversion, indicating an initiating lesion in G1. Although the number of spontaneous crossover events we collected is too low for a conclusive comparison with this data on inverted Ty1, Ty2 pairs from chromosome IV, FS2 likely behaves as a similar hotspot for G1-lesion stimulated crossover events in un-stressed cells.
In cells with low levels of polymerase, over 90% of gene conversion tracts associated with a crossover on the right arm of chromosome III are of the 3∶1 pattern, indicating an initiating lesion during S-phase. Analysis of crossovers induced by low alpha DNA polymerase on yeast chromosomes IV and V indicates that these also are associated primarily with 3∶1 rather than 4∶0 conversion tracts (W. Song and T. D. Petes, personal communication). These results are consistent with stalled or collapsed forks under replication stress stimulating crossover formation. In studies of un-stressed cells, crossover-associated gene conversion tracts on yeast chromosomes IV and V are either a mixture of 3∶1 and 4∶0, or are primarily 4∶0 [35], [39].
Our median gene conversion tract length in cells under replication stress is 14.7 kb, which is much longer than the 1–4 kb tracts observed during meiosis [57]–[59]. Other studies of mitotic crossovers in yeast have highlighted similarly extensive gene conversion with median tract lengths of 4.7 to 20.3 kb [28], [29], [35], [39], [46]. The density of SNP markers evaluated affects our ability to evaluate how often crossovers have adjacent gene conversion. In a previous report on chromosome IV where a high density of SNPs was used, 87% of crossovers had an adjacent gene conversion [28]. In our cells under replication stress, which were evaluated using fewer SNPs, only 41% of crossovers had an adjacent gene conversion.
As discussed, we find that spontaneous mitotic BIR events on the right arm of chromosome III are more frequent than expected, compared to other yeast chromosomes. Under replication stress by low levels of polymerase alpha, the frequency of mitotic LOH is elevated approximately 12-fold, resulting from crossover, BIR, and chromosome loss. Fragile site FS2 is a hotspot for initiating LOH events under replication stress, and S-phase lesions at this site stimulate crossovers and BIR events approximately equally. More than one-third of the BIR events initiated at or near FS2 are non-allelic, resulting in gross chromosomal alteration of chromosome III. These results have important implications for adding to the mechanisms in which human common fragile sites promote tumorogenesis. Human common fragile sites, like yeast FS2, are unstable under conditions of replication stress [60] and replication fork stalling at sequences with secondary-structure forming potential has been observed in some fragile sites [6], [7]. The contribution of deletions, amplifications, and translocations at common fragile sites to tumor development and progression has been extensively documented [18], [61], [62]. However, LOH at tumor suppressor genes has long been known as a driver of tumorogenesis [63], and this mechanism has not been well studied at fragile sites. Based on our results, further research is warranted to determine the role of common fragile sites in stimulating LOH in tumors through BIR and reciprocal mitotic crossovers.
The five diploid strains used for analysis of mitotic recombination were Y332, Y382, AMC310, AMC324, and AMC331 (Figure 1). Each of these diploids was created by mating an MS71-derived haploid cell [64] with a YJM789-derived haploid cell [42], resulting in ∼0.5% sequence divergence between homologous chromosomes [42]. Each diploid is homozygous for the ade2-1 mutation and contains one copy of SUP4-o. Strains Y332, Y382, AMC310, and AMC324 contain one copy of fragile site FS2; strain AMC331 does not contain any Ty1 elements at the location of FS2 on either chromosome III homolog. Strains Y332, Y382, and AMC310 are homozygous for the GAL-POL1 construct [30]; strains AMC324 and AMC331 are homozygous for POL1 driven under its native promoter. The configuration of genes and markers in each diploid strain is diagrammed in Figure 1. The steps in construction of these strains are described in the Text S1, and construction details and genotypes for all strains are in Tables S1 and S2. All transformations and matings were done using standard protocols.
All five diploid strains, whether they contained GAL-POL1 or not, were maintained at 30°C in standard rich media [65], with the exception that the medium contained 3% raffinose instead of dextrose. Raffinose was used as a carbon source because it does not suppress the GAL1/10 promoter, allowing us to control expression of the GAL-POL1 construct by varying the amount of galactose in the medium.
All diploid strains were purified to individual colonies, and were inoculated for growth overnight in standard rich media containing high galactose (0.05%). Three or four cultures were inoculated for each diploid in each condition. Cells were then washed and diluted 1∶5 in fresh rich media liquid culture with no galactose for 6 hours (to induce replication stress by lowering the level of polymerase alpha), or were diluted 1∶5 in rich media liquid culture with high galactose for 6 hours. The galactose treatment for each strain is indicated in Table 1. The density of each culture was determined, and cells were spread at low density to form colonies (∼350 colonies per plate) on plates containing high galactose and 10 µg/ml adenine (two-fold less than standard omission medium). Twenty to forty plates were spread from each culture, to obtain 7,000 to 14,000 colonies per culture. Cells were allowed to grow at 30°C for 3 days and then plates were incubated overnight at 4°C to intensify red color development in the colonies. The total number of colonies was counted for each culture, and culture counts from each diploid were totaled. If a crossover, BIR event, local gene conversion at SUP4-o, or chromosome loss event occurs in the first or second division at the time the diploid is plated, a sectored colony is produced. Therefore, each sectored colony is an independent event. All red/white and red/light pink sectored colonies in which the red portion was at least one-fourth of the colony were identified, and a single cell from each half of the sector was purified for subsequent analysis of the mitotic event that resulted in sectoring. The frequency of BIR events and of chromosome loss events reported in Table 1 was calculated as [number of sectored colonies of the event type/total colonies]. The frequency of crossovers reported in Table 1 was calculated for each strain as [2*number of crossover sectored colonies]/[total number of colonies].
95% confidence intervals for the proportion [66] of each mitotic event were calculated using VassarStats (http://vassarstats.net/). Chi-square contingency tables were used to compare the frequencies of mitotic events between strains.
Phenotype analysis was initially used to classify sectored colonies. Sectors from Y332, Y382, AMC324, or AMC331 with phenotype His+ HygS (red cell) and His+ HygR (light pink cell) usually result from chromosome loss. Sectors from AMC310 of phenotype His− HygR (red cell) and His+ HygR (light pink cell) are usually chromosome loss. Sectors from all strains in which cells from both sides of the sectored colony remain His+ HygR are crossover, BIR, local gene conversion at the SUP4-o locus, or mutation at the SUP4-o locus.
There is ∼0.5% sequence divergence between the two homologs of chromosome III in the experimental diploids (Wei et al. 2007), and several of the single nucleotide polymorphisms (SNPs) alter restriction enzyme sites. For example, a SNP on chromosome III at base 266045 results in an HpyCH4III site on the YJM789-derived chromosome but not the MS71-derived chromosome. This region was amplified by PCR, generating a 374 bp product (Table S3). If the site is heterozygous in the cell being examined, digestion of the amplified product with HpyCH4III followed by gel electrophoresis reveals three band sizes: the uncut 374 bp product and the cut 259 bp and 115 bp products. Genotype analysis of the SNP at chromosome III base 266045, which is 7 kb centromere-proximal of SUP4-o, was used for initial evaluation of event type in all sectored colonies. In chromosome loss and BIR events, the red cell of a sector has only the form of the SNP from the copy of chromosome III lacking SUP4-o, and the light pink cell is heterozygous at this site. In crossover events, the red cell of a sector has only the form of the SNP from the copy of chromosome III lacking SUP4-o, and the white cell has only the form of the SNP from the copy of chromosome III containing SUP4-o. Sectors with red cells that remained homozygous at SNP 266045 may result from a point mutation in SUP4-o or a small gene conversion tract surrounding SUP4-o; these were not further analyzed.
All sectored colonies with a change of zygosity at SNP 266405 were further analyzed. Genomic DNA was harvested from purified cells from each side of sectored colonies and subjected to polymorphism analysis. For all sectored colonies, an initial set of 8 SNPs were tested for zygosity to reconfirm the event type. BIR and crossover events were then tested with additional SNPs to further refine the location of the event. In total, we used 25 SNPs in the 159 kb interval between CEN3 and SUP4-o on chromosome III, plus 2 additional SNPs centromere-distal from SUP4-o. Polymorphic sites, primers, and diagnostic restriction enzymes are listed in Table S3. Gene conversion tract lengths were calculated as described in Lee et al. (2009) with the modification that the size of Ty1 elements present in the MS71-derived homolog that are not present in the sequenced genome are accounted for in our distance calculations when this homolog is used as the template for repair.
Genomic DNA from 1×108 cells was harvested in agarose blocks to prevent shearing as described in [67]. Chromosomes were separated by PFGE in a 1.2% gel in 0.5× TBE at 14°C using a Gene Navigator system (Pharmacia Biotech). Switch times at 6 V/cm were as follows: 50 sec switch for 4.5 hr, 90 sec switch for 5.5 hr, 105 sec switch for 7.5 hr, 124 sec switch for 7.5 hr, 170 sec switch for 7.5 hr. DNA was transferred to Hybond N+ membrane (GE Healthcare Life Sciences) by a neutral transfer according to standard protocol, then probed with CHA1, a gene located on the left arm of chromosome III. The CHA1 probe was made by PCR, using primers 5′ CTGGAAATATGAAATTGTCAGCGAC and 5′ TGAATGCCTTCAACCAAGTGGCCCTTTC. Probes were radioactively labeled by random-prime labeling using Ready To Go beads (-dCTP) (GE Healthcare Life Sciences). Southern blot hybridization and washes were standard. Membranes were exposed to a phosphor screen and images were captured with an FLA-3000 scanner (Fujifilm). There are two normal sizes for chromosome III in our diploids; the YJM789-derived homolog of this chromosome is ∼18 kb smaller than the MS71-derived homolog, because the YJM789 homolog has only one Ty1 element and the MS71 homolog has four Ty1 elements on the right arm of chromosome III. Diploids with two normal-size copies of chromosome III were considered allelic BIR events. Diploids with one normal-size III and one chromosome III of abnormal size were considered non-allelic BIR events.
|
10.1371/journal.pbio.1001803 | Structural Insights Into DNA Repair by RNase T—An Exonuclease Processing 3′ End of Structured DNA in Repair Pathways | DNA repair mechanisms are essential for preservation of genome integrity. However, it is not clear how DNA are selected and processed at broken ends by exonucleases during repair pathways. Here we show that the DnaQ-like exonuclease RNase T is critical for Escherichia coli resistance to various DNA-damaging agents and UV radiation. RNase T specifically trims the 3′ end of structured DNA, including bulge, bubble, and Y-structured DNA, and it can work with Endonuclease V to restore the deaminated base in an inosine-containing heteroduplex DNA. Crystal structure analyses further reveal how RNase T recognizes the bulge DNA by inserting a phenylalanine into the bulge, and as a result the 3′ end of blunt-end bulge DNA can be digested by RNase T. In contrast, the homodimeric RNase T interacts with the Y-structured DNA by a different binding mode via a single protomer so that the 3′ overhang of the Y-structured DNA can be trimmed closely to the duplex region. Our data suggest that RNase T likely processes bulge and bubble DNA in the Endonuclease V–dependent DNA repair, whereas it processes Y-structured DNA in UV-induced and various other DNA repair pathways. This study thus provides mechanistic insights for RNase T and thousands of DnaQ-like exonucleases in DNA 3′-end processing.
| DNA repair relies on various enzymes, including exonucleases that bind and trim DNA at broken ends. However, we know little about how an exonuclease precisely selects and trims a DNA broken end in specific repair pathways. In this study, the enzyme RNase T, previously known for its involvement in processing RNA substrates, is shown to also possess DNase activity. RNase T is a DnaQ-like exonuclease and is characterized in this work as the exoDNase responsible for trimming the 3′ ends of structured DNA in various DNA repair pathways. Based on the high-resolution crystal structures of RNase T-DNA complexes, an insightful working model is provided showing how RNase T processes bulge, bubble, and Y- structured DNA in various DNA repair pathways. RNase T thus represents a unique structure-specific exonuclease with multiple functions not only in processing 3′ overhangs of duplex RNA during RNA maturation, but also in processing 3′ ends of bubble, bulge, and Y-structured DNA during DNA repair. These findings advance our understanding of the precise function of an exonuclease in DNA repair and suggest possible roles for thousands of members of DnaQ superfamily exonucleases in DNA repair and replication.
| It is well known that DNA repair mechanisms maintain genomic integrity and are essential for cell survival. Damaged DNA can be restored by a variety of DNA repair processes, such as direct reversal, base excision, nucleotide excision, mismatch, and recombination repair pathways [1]. Although diverse proteins play different roles in these pathways, DNA repair is generally accomplished by a coordinated effort via several types of DNA enzymes, including endonucleases that nick DNA near the damaged site, exonucleases that trim DNA from the broken end, helicases that unwind duplex DNA, polymerases that make new strand DNA with correct sequences, and ligases that seal the restored DNA strands. Among all these DNA enzymes, the molecular functions of exonucleases, which bind at the 3′ or 5′ end of DNA and cleave one nucleotide at a time, are least understood. How they select, rather than randomly bind to, a broken end of DNA and process it up to the site for the next-step processing remains to be investigated.
Here we use the bacterial exonuclease RNase T as a model system to study the processing of DNA in various DNA repair pathways. RNase T is a member of the DnaQ-like 3′–5′ exonucleases with a DEDDh domain that contains four acidic DEDD residues (D23, E25, D125, and D186) for binding of two magnesium ions, and one histidine residue (H181) for functioning as the general base in the active site for the hydrolysis of the 3′-terminal phosphodiester bond of a nucleic acid chain [2]. The family of DnaQ-like exonucleases constitutes thousands of members, all with exonuclease activity either processing RNA during RNA maturation, interference, and turnover, or processing DNA during DNA replication, degradation, repair, and recombination. A number of the DnaQ-like exonucleases have been shown to play a role in DNA repair. Usually the DEDDh domain can be linked to a DNA polymerase domain for proofreading during DNA replication, such as the DnaQ domain of the ε subunit of E. coli pol III holoenzyme and the exonuclease domain of human pol δ, ε, and γ [3]. Mutations in or deletion of the proofreading 3′ exonuclease domain for these polymerases are either lethal or induce high mutation rates and high incidence of cancers [4]. The DEDDh domain can also be linked to a helicase domain and functions in processing of broken DNA strands during DNA repair and recombination, such as that of human WRN [5]. Mutations in the DEDDh exonuclease domain of WRN are associated with Werner syndrome that results in premature aging and increased risk of cancer [6].
However, most of the DEDDh domain functions as an autonomous protein and is not linked to a polymerase or a helicase domain. Some of these exonucleases participate in DNA 3′-end processing in DNA repair, such as ExoI and ExoX from E. coli and TREX1 and TREX2 from human [7]–[9]. ExoI and ExoX are monomeric enzymes that digest single-stranded DNA in mismatch and DNA recombination repair pathways [10]–[12], whereas the human TREX1 and TREX2 are dimeric enzymes, likely processing single-stranded DNA in mammalian cells [13],[14]. Mutations in TREX1 are linked to the autoimmune diseases Aicardi-Goutieres syndrome and systemic lupus erythematosus, probably due to the accumulation of nonprocessed intermediate DNA during replication and repair pathways [15]–[17]. The crystal structures of ExoI [18], TREX1 [19], and TREX2 [20] reveal that they all bear a classical α/β fold of the DEDDh domain; nevertheless, their precise functions in DNA processing remain uncertain.
RNase T has also been implicated in the UV-repair pathways based on the observations that the cells lacking RNase T are less resistant to UV radiation and overexpression of RNase T can rescue the UV sensitivity of the ExoI knockout E. coli strain [21],[22]. Yet RNase T was originally recognized as an RNase based on its indispensable role in tRNA 3′-end processing during tRNA maturation [23]. RNase T also performs the final trimming for various stable RNA, including 5S and 23S rRNA [24],[25]. RNase T can digest both DNA and RNA and it has a unique specificity that its exonuclease activity is reduced by a single 3′-terminal C or completely abolished by a dinucleotide 3′-terminal CC in digesting either DNA or RNA, referred to as the C effect [26]. Moreover, its exonuclease activity is inhibited by duplex structures, referred to as the double-strand effect; therefore, a 3′ overhang of a duplex DNA or RNA is only digested near the duplex region by RNase T [26],[27]. Previous crystal structures of RNase T in complex with various single-stranded DNA (3′-terminal G versus C) and double-stranded DNA (1 versus 2 nucleotide 3′ overhang) reveal the structural basis for C effect and double-strand effect [27],[28]. The binding of an uncleavable substrate, such as a single-stranded DNA with a 3′-terminal C or a duplex DNA with a short 2-nucleotide 3′ overhang, induces an inactive conformational change in the active site and thus inactivates the exonuclease activity. Therefore, in digesting a duplex DNA with a 3′ overhang, RNase T can accurately differentiate its cleavable or noncleavable substrates based on the C effect and double-strand effect, and it produces a precise final product of a duplex with a 1-nucleotide (if the last base pair in duplex region is AT) or 2-nucleotide (if the last base pair in duplex region is GC) 3′ overhang if the CC dinucleotide is not present within the 3′ tail, or else it stops at the 3′-CC end. RNase T hence is capable to trim various precursor RNA to produce mature RNA with a precise 3′ overhang depending on the structure and sequence of these precursors: 1 nt for 5S rRNA, 2 nt for 23S rRNA, 4 nt for 4.5S RNA, and 4 nt for tRNA [27],[28].
In fact, RNase T is a more efficient DNase than RNase in that it digests DNA with a 10-fold efficiency as compared to RNA (see figure S2 in [28]), supporting its possible cellular role in DNA processing. However, it is not certain if RNase T indeed processes DNA in DNA repair, and if it does, how it selects and processes its DNA substrates. To determine the molecular function of RNase T, we show here by biochemical and structural approaches that it is a structure-specific DNase capable of digesting intermediate structured DNA during DNA repair. We found that RNase T not only digests bubble and bulge DNA in Endonuclease V (Endo V)–dependent DNA repair but also digests Y-structured DNA in UV-induced DNA repair pathways. The crystal structures of RNase T in complex with a bulge and a Y-structured DNA further demonstrate how this dimeric enzyme elegantly binds and processes these structured DNA molecules in different ways. Our results reveal, for the first time, the precise molecular role of an exonuclease in the 3′ end DNA processing and may hint at the molecular function for other members of DnaQ-like exonucleases.
To confirm the possible roles of RNase T in DNA repair, we measured the chronic and acute sensitivity of the RNase T knockout E. coli strain (Δrnt)?against various DNA-damaging agents, including hydrogen peroxide (H2O2), methyl methanesulfonate (MMS), 4-nitroquinoline-1-oxide (4NQO), mitomycin C (MMC), and UV light. A number of exonucleases that have been shown to play a role in DNA repair, including ExoI (ΔexoI) [7],[10], ExoX (ΔexoX) [7], PNPase (Δpnp) [29], and RecJ (Δrecj) [30], were tested in parallel for a comparison (Table 1 and Figures S1 and S2). The wild-type K-12 strain resisted all DNA-damaging agents when present at a chronic dose, whereas RNase T-deficient strain (Δrnt) had a slow growth phenotype and was sensitive to the chronic dose of H2O2, MMS, 4NQO, and UV-C (Figure 1A). The RNase T-deficient strain (Δrnt) was also sensitive to the acute dose of H2O2 in various concentrations from 20 to 80 mM (Figure 1B). The sensitivity of Δrnt strain to UV-C was different from those observed in the previous report [21]; therefore, we further confirmed the UV and H2O2 sensitivity by rnt-rescued experiments, which restored the resistance of Δrnt cells against UV-C and H2O2 (see Figure S1). This result shows that the sensitivity of the RNase T knockout cells to UV-C and H2O2 is indeed due to the deficiency of RNase T.
H2O2 produces a wide variety of DNA lesions, including single-strand/double-strand breaks (DSB), oxidation and deamination of bases, and sugar modifications [31],[32], that are usually restored by direct repair (DR), base excision repair (BER), and alternative repair (AR) [1],[29],[30],[33]. DNA alkylating agent MMS produces methylated DNA bases that can be restored by DR and BER [34],[35]. MMS also leads to the accumulation of single-strand gaps (SSGs) and DSB-related DNA damage [29],[35]. MMC is a DNA cross-linking agent that can trigger the SOS response and creates damage repaired by NER [29],[36],[37]. UV light-mimetic agent 4NQO can produce replication-blocked DNA base adducts, SSGs, and DSB-related DNA damages [29],[38] that are mainly repaired by NER [38]. UV-C irradiation (100–290 nm) leads to three major base modifications and DSB-related DNA damage that are usually repaired by BER and DNA recombination [31],[39],[40]. The sensitivity of the Δrnt strain to H2O2, MMS, 4NQO, and UV-C suggests that RNase T may play a role in BER, AR, and DSB-related DNA repair pathways.
In comparison to the known DNA-repair exonucleases, RNase T had a wider sensitivity to various DNA-damaging agents. The ExoI-deficient cells were only sensitive to H2O2; the ExoX-deficient cells were not sensitive to any DNA-damaging agents; the PNPase-deficient cells were sensitive to H2O2, MMS, and UV-C; and the RecJ-deficient cells were sensitive to 4NQO and UV-C (Table 1 and Figure S1). RNase T had a more apparent and wider sensitivity as compared to those of ExoI, ExoX, PNPase, and RecJ, suggesting that RNase T plays more extensive and crucial roles in various DNA repair pathways.
To further characterize the role of RNase T in DNA repair pathways, the single-stranded DNA containing a methylated, deaminated, or oxidized base, or an abasic site at the 3′-terminal end (5′-GAGTCCTATAX-3′) were incubated with RNase T in the DNA digestion experiment. We found that RNase T digested the DNA with a methylated base—O4-methylthymine (O4-mT) and O6-methylguanine (O6-mG)—and a deaminated base—uracil and hypoxanthine. However, the DNA with a 3′-terminal oxidized base, 8-oxoguanine (8-oxo), and an abasic site were more resistant to RNase T digestion (see Figure 1C). This result suggests that RNase T can function as an exonuclease in the excision step for methylated and deaminated bases in BER and AR.
The next question we tackled was what types of DNA that can be processed by RNase T in DNA repair, besides the single-stranded DNA with a lesion. RNase T is not an appropriate exonuclease for digesting single-stranded DNA since its exonuclease activity is easily blocked by any C within a DNA strand. A variety of intermediate structured DNAs are generated during DNA repair, such as bulge, bubble, and Y-structured DNA. Bulge DNAs are produced in frameshift DNA mutations during DNA replication of repetitive sequences [41], whereas bubble DNA are generated in mismatch replication or deamination of DNA bases [42]. Y-structured DNAs are generated in various DNA repair pathways, such as mismatched DNA repair and DNA recombination (see Discussion). To test if RNase T processes these intermediate structured DNA, we incubated RNase T with different DNA and found that RNase T can digest Y-structured DNA and blunt-end bubble DNA with an I-T or I-G bubble (Figure 2A). In digesting the Y-structure DNA, the exonuclease activity of RNase T was blocked by the duplex structure—that is, double-strand effect—and RNase T generated a final product of 1-nucleotide 3′ overhang duplex (Figure 2A). In digesting bulge and bubble DNA, the double-strand effect did not occur, and the blunt-end bulge and bubble DNA was cleaved by RNase T (Figure 2A).
We further tested the sequence preference of RNase T in digesting the structured DNA. In digesting a classical duplex DNA with a short 3′ overhang, the exonuclease activity of RNase T was blocked by a dinucleotide 3′-end CC sequence (Figure 2B). In contrast to the duplex DNA, the bulge and Y-structured DNA with terminal 3′-CC were processed by RNase T into a final product with a 1-nucleotide 3′ overhang (Figure 2B). Therefore, in digesting bubble and bulge DNA, RNase T has no sequence preference, and it removes the last paired nucleotide of any sequence to generate a 1-nucleotide 3′ overhang. In digesting Y-structured DNA, RNase T also has no sequence preference and processes the 3′ tail of any sequence close to the duplex region to generate a 1-nucleotide 3′ tail as the final product.
We were intrigued by how RNase T could bind and process a bubble or bulge in DNA with a blunt end. Previous studies showed that the double-strand effect of RNase T requires a 3′ overhang of a duplex with a length of more than 2 nucleotides for inserting into the active cleft for digestion (see Movie S1). To reveal how RNase T binds and processes a DNA bulge with a blunt end, we co-crystallized RNase T with two bulge DNA molecules, one with a 3′-end TC and one with a 3′-end CC sequence in acidic conditions, pH 5.5 and pH 6.0, respectively (see Table S2). RNase T only digests nucleic acids in basic conditions because the general base H181 has to be deprotonated to activate a nucleophilic water for hydrolysis. Therefore, due to the low pH, the bulge DNA in the crystal were not cleaved by RNase T. The crystal structure of the two complexes was solved by X-ray diffraction methods at a resolution of 1.8 and 2.0 Å, respectively (see Figure 3). In the RNase T–bulge DNA complex structures, the dimeric RNase T bound to two bulge DNAs, with the 3′ end of DNA binding at the active site of each protomer (Figure 4). The bulge DNA was bound between the two RNase T protomers, in a way similar to that of the classical duplex DNA with a 3′ overhang [28].
However, in contrast to the previous duplex DNA complex, the aromatic side chain of Phe29 was inserted into the bulge and stacked with the two neighboring GC base pairs in both of the bulge DNA complexes (see Figure 4A and 4B). In the previous duplex DNA complex, Phe29 was stacked with the 5′-end base of the opposite nonscissile strand, and the stacking stopped the further cleavage of the scissile strand at the 3′ end, resulting in the double-strand effect (see the schematic comparison in Figure 4C). The crystal structure of the bulge DNA complex revealed how RNase T can overcome the double-strand effect by inserting Phe29 into the bulge so that the 3′-end scissile phosphate was moved accordingly into the active site (see Movie S2). We found that the active site of the bulge DNA complex indeed had an active conformation with two bound Mg2+ ions, and the general base His181 was located close to the scissile phosphate (Figure 4B).
The crystal structure of the bulge DNA complex also revealed how RNase T could overcome the C effect. The 3′-end cytosine was paired with the 5′-end guanine, and this base pairing prevented Glu73 from interacting with the 3′-end C to induce the C effect (Figure 2B, Figure S2B). Therefore, the bulge DNA could be processed by RNase T without any sequence preference. Moreover, the 5′ end of bulge DNA was not hindered by any residue and could further extend (Figure S2A), suggesting that RNase T can cleave bulge DNA with a long single-stranded region at the opposite strand, similar to those DNA in the frameshift DNA mutations (see Discussion) [41]. The crystal structure thus reveals at the atomic level how RNase T binds and processes a bulge DNA with a blunt end without a sequence preference.
The bubble and bulge DNA can be produced by Endo V, which makes a nick at the 3′ side one base pair away from a damage site with a deaminated base in the alternative DNA repair [42]. Endo V also processes mismatched DNA, hairpin-containing DNA, bulge DNA, and flap DNA [43]–[45], however the downstream process following Endo V nicking has not been characterized. The bulge DNA in our crystal structures had a conformation similar to the bubble DNA produced by Endo V nicking, suggesting that RNase T might be the downstream exonuclease of Endo V, responsible for removing the last base-paired nucleotide at the 3′ end to release the single-stranded DNA or the damaged DNA bases, such as hypoxanthine, xanthine, and uracil (Figure 2D).
To test this possibility, we prepared the hypoxanthine-containing—that is, inosine-containing—heteroduplex DNA for examination of Endo V–dependent inosine excision repair in vitro [46]. The heteroduplex DNA plasmid contained the I-G base pair with an AlwNl cutting site and a potential XhoI cutting site. Once the inosine was restored to cytosine, the plasmid could be cleaved by AlwNl and XhoI into two linear double-stranded DNA molecules of 4.1 and 3.1 kilobases (Figure 2D). The I-G–containing plasmid was then incubated with Endo V, RNase T, ligase, and the Klenow fragment exo− (Polymerase I Klenow fragment with a defected 3′–5′ exonuclease activity). The inosine in the plasmid was restored to cytosine with a higher repair efficiency (86.7%) as compared with those incubated with the wild-type DNA Polymerase I with a proofreading exonuclease domain (61.4%) (Figure 2C). The repair efficiency was positively correlated with the RNase T concentration and the time of incubation (Figure S3). Interestingly, ExoI and ExoX could not work with Endo V to restore the inosine to cytosine (unpublished data). These results show that RNase T can work with Endo V in the Endo V–dependent DNA repair.
Beside bubble/bulge DNA, RNase T also processed Y-structured DNA, which can be generated during various DNA repair pathways, such as mismatch repair and DNA recombination. However, it remained unknown how an exonuclease can specifically process the 3′-end tail of the intermediate Y-structured DNA. To reveal how RNase T binds and processes a Y-structured DNA, we co-crystallized RNase T with a Y-structured DNA and solved the complex crystal structure at a resolution of 1.9 Å (Figure 3 and Figure 5). In the crystal structure, the Y-structured DNA was bound to RNase T in a unique way, different from those of the bulge DNA and the duplex DNA that were bound between the two protomers with one strand of DNA bound to one protomer (see Figure 4C). In contrast, both strands of the Y-structured DNA were bound to a single protomer, one Y-structured DNA bound to protomer A and the other DNA molecule bound to protomer B (Figure 5). This unique binding mode can avoid the hindrance produced by Phe29, which might stack with the 5′-end base of the opposite nonscissile strand if the Y-structured DNA was bound in a way similar to that of a duplex DNA. Therefore, in this complex, the opposite nonscissile strand of the Y-structured DNA rotated about 180° to interact with the same protomer of RNase T (see Figure 5B). Several residues, including Gln169, Asp174, Phe175, and Ser177, interacted with the nonscissile strand forming hydrogen bonds with the first and second phosphates in the 5′-overhang region, making it fit snugly onto the molecular surface of RNase T (Figure S5).
The 3′ tail of the Y-structured DNA in the crystal structure had a dinucleotide 3′-CC sequence. However, the 3′-CC did not induce the C effect and inhibit the exonuclease activity of RNase T. A close look at the crystal structure of the Y-structured DNA complex showed that the 3′-end C did not interact with Glu73 as it did in the duplex complexes (left panel in Figure S4B). Moreover, the scissile phosphate of the 3′-end C did not shift away from the active site, and as a result, two Mg2+ ions were bound in the active site in an active conformation (right panel in Figure S4B). Therefore, due to the unique binding mode, the C effect did not occur when RNase T was bound to a Y-structured DNA with a 3′-end CC. In summary, this crystal structure reveals how RNase T binds a Y-structured DNA in a unique way and how it processes the 3′ tail of any sequence close to the duplex region (see Movie S3).
Besides RNase T, two monomeric DnaQ-like exonucleases ExoI and ExoX also process DNA during DNA repair in E. coli. ExoI is suggested to play a role in BER [47], mismatch repair [7],[10],[48],[49], UV-related repair [41],[50], and DNA replication [51]. ExoX is involved in mismatch repair [11],[12] and UV-related repair [8]. ExoI binds and cleaves long single-stranded DNA [52], whereas ExoX digests both single-stranded and double-stranded DNA [8]. The exonuclease activity of RNase T, ExoI, and ExoX probably overlap and are redundant in these pathways or they may target different substrates. To compare the substrate preference of RNase T to those of ExoI and ExoX, we further expressed and purified ExoI and ExoX for DNA digestion assays. The dynamic light scattering confirmed that ExoI and ExoX were monomeric proteins in contrast to RNase T, which existed as dimeric proteins (Figure S5).
We found that RNase T, ExoI, and ExoX digested single-stranded 11-nucleotide DNA with similar efficiencies (Figure 6A). However, in digesting Y-structured DNA with a short 3′ overhang, only RNase T and ExoX could process the 3′ overhang close to the duplex region, whereas ExoI did not digest Y-structured DNA at low concentrations but did digest Y-structured DNA randomly into small nucleotides at high concentrations (Figure 6B). In digesting duplex DNA with a short 3′ overhang, RNase T processed DNA into a specific length close to the duplex region, generating a final duplex product with a 1-nt 3′ overhang at low concentrations (Figure 6C). ExoX also digested duplex substrates but was less specific, generating various end products with 3′ overhangs of different lengths. On the contrary, ExoI could not digest the duplex substrates at the low concentration (0.02 µM) (Figure 6C). At the high exonuclease concentrations (0.1 and 1 µM), both ExoI and ExoX digested the duplex DNA substrates in the single-stranded and double-stranded regions into small nucleotides. However, RNase T still retained its specificity, only cleaving in the 3′ overhang but not in the duplex region (Figure 6C). These results suggest that RNase T is a highly specific exonuclease that targets the 3′ overhang of structured DNA and produces a precise final product. On the other hand, ExoX is less specific and generates 3′ overhangs of different lengths in digesting duplex substrates with 3′ overhangs, whereas ExoI is specific for single-stranded DNA.
Besides DNA digestion assays, the gel shift assays further showed that RNase T bound with similar affinities to single-stranded DNA, duplex DNA with 4-, 6-, and 10-nucleotide 3′ overhangs (Figure S6). In contrast, ExoI had lower binding affinity for duplex DNA with short 3′ overhangs, such as 4 and 6 nucleotides, in agreement with its low activity for these substrates. ExoX also preferred to bind to single-stranded DNA, but not duplex DNA with short 3′ overhangs at similar concentrations (Figure S6). Combining these results of the exonuclease activity and DNA-binding assays, we conclude that RNase T is the ideal exonuclease for trimming the 3′ overhang of structured DNA closely to the duplex region, including Y-structured DNA and duplex DNA, whereas ExoI and ExoX mainly process single-stranded DNA in DNA repair.
Our results suggest that RNase T is likely involved in the Endo V–dependent DNA repair pathway. Endo V is a conserved endonuclease playing critical roles in maintaining genome stability in prokaryotes and eukaryotes [53]. Endo V recognizes bubble DNA with mismatched base pairs and deaminated DNA lesions and initiates the Endo V–dependent DNA repair pathway that is independent of BER and MMR [42],[44],[45],[53],[54]. Moreover, Endo V nicks frameshift and structured DNA, such as insertion/deletion loops, hairpins, and flap DNA [43],[55]. Frameshift DNA mutations are mistakenly generated during replication of repetitive sequences [41], and as a result, the bulge DNA are produced by slipped misalignment of tandem repeats [56],[57]. Rearrangements between tandem repeated DNA are important factors for genome instability and have been implicated in Friedreich ataxia in humans [58],[59]. Slipped misalignment of tandem repeat DNA may cause palindrome-stimulated deletion or expansion by two RecA-independent recombination mechanisms—that is, single-strand annealing and replication slipped mispairing [60],[61]. Single-strand-specific exonucleases, such as ExoI, ExoX, and RecJ, were reported to stabilize tandem repeats and limit RecA-independent recombination [56],[62]. However, the downstream structure-specific exonuclease of Endo V for the further trimming of the DNA from the broken end has not yet been identified.
Our structural and biochemical data of RNase T show that it can bind and digest bulge/bubble and Y-structured DNA. Moreover, RNase T can work with Endo V, DNA Polymerase I (Klenow fragment exo−), and ligase to restore an inosine to cytosine in a heteroduplex DNA molecule in vitro. The crystal structures of RNase T bound with a blunt-end bulge DNA further show how RNase T removes the last base pair at the 3′ end by a special Phe-inserting binding mode. All these results suggest that RNase T may function as the downstream exonuclease of Endo V in alternative DNA repair. Taking together these lines of evidence, we suggest that RNase T likely recognizes these bulge and bubble DNA structures generated by Endo V and trims at the 3′ end of the nicked site to remove the last base pair next to the lesion. The single-stranded DNA or damaged DNA is then released for the next step of processing (see Figure 7A).
After removing the 3′-end base-paired nucleotide by RNase T, insertion DNA, hairpin DNA, and deaminated DNA lesions are released as single-stranded DNA. These single-stranded insertion DNA and hairpin DNA are probably further trimmed by the single-strand-specific exonucleases, such as ExoI and/or ExoX, with the help of single strand binding protein (SSB) and helicases. RNase T can further digest the 3′-end short overhang close to the duplex region in a way that we observed in the crystal structure of the Y-structured DNA complex. Deaminated DNA lesions are likely also removed by RNase T since we show that RNase T can digest single-stranded DNA containing oxidized bases and deaminated bases (Figure 1C). It has been shown that the dimeric Exo I from Thermus thermophilus shares a sequence homology to RNase T and plays a similar role in digesting damaged DNA with methylated and deaminated bases [30]. It is very likely that Exo I from Thermus thermophilus is a functional homologue of RNase T and both of them play key roles in DNA repair. Therefore, after nicking by Endo V, the single-strand-specific exonucleases and structure-specific RNase T likely work together to further trim DNA from the broken end. After this trimming, polymerases and ligases can complete the DNA repair pathway.
RNase T plays crucial roles in various DNA repair pathways, as shown by the sensitivity of the rnt knockout strain to a wide range of DNA-damaging agents. The indispensable role of RNase T might be due to its unique specificity for structured DNA that are generated during various DNA repair pathways. For instance, UV radiation can lead to single/double-strand breaks and base modifications, such as cross-linked pyrimidine dimers, photoproducts, and thymine glycols, and as a result, three different UV-induced DNA repair pathways are initiated [39],[63]. In the first pathway, the base modification induced by UV may stall replication forks. In such a case, RecFOR and RecA bind to the lagging strand template and the invasion-containing leading strand to promote double-strand formation and repair by NER [64]. During this process, the Y-structured DNA formed on the leading strand requires a structure-specific exonuclease, very likely RNase T, to trim its 3′ overhang (see Figure 7C).
In the second pathway, UV radiation can induce single-strand breaks that can be repaired by homologous recombination [65]. During this process, Y-structured DNA is formed as an intermediate during gap-filling recombination (see Figure 7D). ExoI was reported to promote this RecA-dependent 5′-end strand exchange by digesting the 3′ competitor strand [66],[67]. However, ExoI cannot digest the 3′ overhang close to the duplex region, and thus most likely RNase T is responsible for processing the Y-structured DNA intermediates in the gap-filling recombination pathway.
In the third pathway, the double-strand breaks induced by UV radiation are generally repaired by the RecA-dependent homologous recombination in bacteria [68]. This DNA repair pathway is initiated by RecBCD or RecJ to generate 3′ overhangs and is followed by RecA and RecFOR to promote strand invasion. DNA repair synthesis is then primed by PolI and PolIII from the invaded strand of the D-loop structure. RuvC resolvase cleaves the Holliday junctions that are synthesized after branch migration and LigA seals the nick to complete the homologous recombination [1]. In this process, ExoI was reported to affect RecBCD-mediated recombination [69] since the 3′–5′ exonucleases are required to degrade the 3′ tail of the intermediate Y-structured DNA after RecA dissociation [48],[51],[70],[71]. Yet ExoI is not an appropriate exonuclease for digesting the 3′ tail near the duplex region. Based on our results, we suggest that most likely RNase T is involved in digesting the 3′ tail close to the duplex region in the UV-induced DNA homologous recombination (Figure 7E). Moreover, in comparison with FEN1, which is a flap endonuclease that binds DNA with one 3′-flap nucleotide and cleaves one nucleotide into the double-stranded DNA at the 5′ flap end to produce a ligatable product during DNA replication and repair [72], RNase T is likely required to produce a DNA with a short 3′ overhang with one or two nucleotides that can be further processed in DNA homologous recombination.
Besides UV-induced DNA repair, RNase T may also participate in other DNA repair processes that require a structure-specific 3′–5′ exonuclease, such as MMR. It has been shown that ExoI and ExoX are essential for methyl-directed mismatch repair in E. coli [7],[10]–[12],[49],[50]. These two monomeric exonucleases are responsible for removing the 3′ single-stranded tail in Y-structured DNA during MMR (see Figure 7B). However, they cannot process the 3′ single-stranded tail close to the double-stranded region [1],[49]. ExoI only processes DNA with a long single-stranded region (over 13 nucleotides) in a processive manner, while a SSB stimulates its exonuclease activity [52],[73],[74]. ExoX, however, interacts with MutL during MMR and is not specific for processing Y-structured DNA [8],[12]. On the other hand, the RNase T homolog Thermus thermophilus ExoI is suggested to excise the 3′ overhang of a Y-structured DNA and plays a role in MMR [30]. Therefore, it is very likely that RNase T processes the 3′ tail of the Y-structured DNA in MMR in E. coli. Our structure and biochemical assays show that the C effect does not occur when RNase T digests short 3′ overhang of a Y-structured DNA, and hence RNase T is capable of processing any sequence of the 3′ overhang of a Y-structured DNA during MMR. Therefore, we propose here that the monomeric ExoI and ExoX work with a helicase or SSB to process long 3′ tails, while the dimeric RNase T further trims the short 3′ overhang of Y-structured DNA during MMR.
In conclusion, RNase T is a unique structure-specific exonuclease responsible for processing the 3′ ends of structured DNA in various DNA repair pathways. RNase T has an ideal dimeric architecture for binding and processing the 3′ end of various structured DNA in diverse ways, including duplex, bulge/bubble, and Y-structured DNA. Therefore, this intriguing exonuclease has multiple functions not only for processing duplex RNA during RNA maturation, but also processing bubble/bulge and Y-structured DNA during DNA repair. The diverse functions and different specificities of RNase T are closely correlated to its dimerization architecture and various binding modes against different substrates. We provide solid data here showing how the dimeric RNase T processes structured DNA in DNA repair that will serve as a model for understanding the molecular functions of thousands of members of DnaQ-like exonucleases.
Wild-type E. coli K-12, single gene knockout (Δrnt, Δsbcb, Δexox, Δpnp, and ΔrecJ) strains used in the survival studies were from the Keio collection [75]. All E. coli cells were grown to an OD600 of 0.5–0.6 in LB medium at 37°C. To measure the acute sensitivity to hydrogen peroxide (H2O2), cells were exposed to 0, 20, 40, and 80 mM H2O2 for 20 min. After removing H2O2, cells were diluted 100-fold into 10 ml LB medium and further grown on a rotary shaker (200 r.p.m.) at 37°C for the measurement of A600 (OD) at 60 min intervals. To measure the chronic sensitivity to H2O2, MMS, mitomycin (MMC), and 4NQO, serial dilutions of cells were spotted on plates containing indicated concentrations of the DNA-damaging agents and incubated overnight at 37°C. To measure the sensitivity against UV-C, serial dilutions of cells were spotted on plates and exposed to UV-C (254 nm) in 20 J/m2 for 10 s by Hoefer UVC 500-Ultraviolet Crosslinker (Hoefer Inc.). After UV-C irradiation, cells were incubated overnight at 37°C.
The full-length rnt, sbcb, and exox genes were amplified by PCR using E. coli genomic DNA from JM109 or K-12 strains and cloned into NdeI/XhoI sites of expression vectors pET-28a (Novagen) to generate the N-terminal His-tagged fused recombinant proteins. The expression plasmid was transformed into the E. coli BL21-CodonPlus(DE3)-RIPL strain (Stratagene) cultured in LB medium supplemented with 35 µg/ml kanamycin. Cells were grown to an OD600 of 0.5–0.6 at 37°C and induced by 0.8 mM IPTG at 18°C for 18 h. The harvested cells were dissolved in 50 mM Tris-HCl (pH 7.5) buffer containing 300 mM NaCl and disrupted by a microfluidizer. Each exonuclease was purified by chromatographic methods using a HiTrap TALON column (GE Healthcare), a HiTrap Heparin column (GE Healthcare), and a gel filtration column (Superdex 75, GE Healthcare). Purified RNase T, ExoI, and ExoX samples were concentrated to 15–35 mg/ml in 300 mM NaCl and 50 mM Tris-HCl (pH 7.0).
DNA oligonucleotides used for nuclease activity assays were synthesized (BEX Co., Tokyo, Japan or MDBio, Inc., Taiwan) and labeled at the 5′ end with [γ-32P]ATP by T4 polynucleotide kinase and purified on a Microspin G-25 column (GE Healthcare) to remove the nonincorporated nucleotides. Purified substrates (20 nM; see Table S1 for sequences) were incubated with RNase T, ExoI, or ExoX at various concentrations in a buffered solution of 120 mM NaCl, 2 mM MgCl2, and 50 mM Tris-HCl (pH 7.0) at room temperature for 20–60 min. The reaction was quenched by addition of the stop solution (2× TBE) and heating at 95°C for 5 min. Reaction samples were then resolved on 20% denaturing polyacrylamide gels and visualized by autoradiography (Fujifilm, FLA-5000).
DNA binding affinities of RNase T, ExoI, and ExoX were measured by gel shift assays. The 5′-end 32P-labeled DNA substrates (20 nM) were incubated with RNase T, ExoI, or ExoX in a solution of 100 mM NaCl, 30 mM EDTA, 10 mM EGTA, and 50 mM Tris-HCl (pH 7.0) for 20 min at room temperature. The concentrations of each protein used in the assays were 0, 5, and 50 µM. Reaction samples were then resolved on 20% TBE gels (Invitrogen) and visualized by autoradiography (Fujifilm, FLA-5000).
The E. coli strain NM522.RS5033 was used in the assay as described in Fang et al. [76]. DNA polymerase I (E. coli), the Klenow fragment exo− (DNA polymerase I Klenow fragment without the 3′–5′ exonuclease activity), E. coli DNA ligase, T4 polynucleotide kinase, recombinant Endo V, and restriction endonucleases were obtained from New England Biolabs. RecBCD nuclease was purchased from EPICENTRE Biotechnologies.
Construction of dI-G heteroduplex DNA substrates was prepared as described in Lee et al. [46]. M13mp18 replicative form DNA was hydrolyzed with HindIII and mixed with a 4-fold molar excess of M13LR1 viral DNA, followed by alkaline denaturation and re-annealing. The excess ssDNA was removed by hydroxyapatite (Biorad) chromatography and benzoylated naphthylated DEAE cellulose (Sigma) chromatography, and the linear dsDNA was removed by RecBCD nuclease (EPICENTRE) treatment. The resulting circular duplex DNA containing 22-nt gaps was further purified by Vivaspin 20 ultrafiltration (GE Healthcare). A 5′-phosphorylated deoxyinosine-containing 22-bp synthetic oligonucleotides, 5′-AGCTCTIGAGGCTGCTGCTGCT-3′ (Blossom Biotech), was then annealed to the gap and sealed by T4 DNA ligase in the presence of ethidium bromide. The covalently closed dG:I heteroduplex DNA was isolated by CsCl-EtBr density gradient centrifugation.
The repair conditions were modified from Lee et al. [46]. The dI-G heteroduplex substrates (0.1 µg) were incubated with repair enzymes (1.1 nM Endo V, 0.13 µM DNA polymerase I/0.13 µM Klenow fragment exo-, and 5 µM RNaseT) for 30 min at 37°C in 15-µl reactions containing 50 mM NaCl, 10 mM Tris-HCl (pH 7.9), 10 mM MgCl2, 1 mM dithiothreitol, 50 µg/ml bovine serum albumin, 0.3 mM NAD+, and 125 µM of each dNTP. The reactions were terminated by heat inactivation at 75°C for 20 min. The DNA was then analyzed by restriction endonuclease hydrolysis and agarose gel electrophoresis. The gel images were captured by a gel documentation CCD camera (UVP Ltd.) using Viewfinder 3.0, and band intensities were then measured by NIH Image J 1.45s software.
Wild-type RNase T (25–35 mg/ml) in 300 mM NaCl and 50 mM Tris-HCl (pH 7.0) were mixed with different stem-loop DNA substrates in the molar ratio of 1∶1.2. Detailed information for DNA sequences and crystallization conditions of the three structures is given in the Table S2. All crystals were cryo-protected by Paraton-N (Hampton Research, USA) for the data collection at 100 K. X-ray diffraction data were collected using synchrotron radiations at SPXF beamline BL13B1 at NSRRC, Taiwan, or at the BL44XU beamline at SPring-8, Japan. All diffraction data were processed by HKL2000, and the diffraction statistics are listed in Table 1. Structures were solved by the molecular replacement method using the crystal structure of E. coli RNase T (PDB ID code 3NGY) as the search model by program MOLREP of CCP4. The models were built by Coot and refined by Phenix.
Structural coordinates and diffraction structure factors have been deposited in the RCSB Protein Data Bank with the PDB ID codes of 4KB0 and 4KB1 for RNase T-bulge DNA complexes and 4KAZ for the RNase T-Y structured-DNA complex.
|
10.1371/journal.pcbi.1004468 | Comprehensive Meta-analysis of Ontology Annotated 16S rRNA Profiles Identifies Beta Diversity Clusters of Environmental Bacterial Communities | Comprehensive mapping of environmental microbiomes in terms of their compositional features remains a great challenge in understanding the microbial biosphere of the Earth. It bears promise to identify the driving forces behind the observed community patterns and whether community assembly happens deterministically. Advances in Next Generation Sequencing allow large community profiling studies, exceeding sequencing data output of conventional methods in scale by orders of magnitude. However, appropriate collection systems are still in a nascent state. We here present a database of 20,427 diverse environmental 16S rRNA profiles from 2,426 independent studies, which forms the foundation of our meta-analysis. We conducted a sample size adaptive all-against-all beta diversity comparison while also respecting phylogenetic relationships of Operational Taxonomic Units(OTUs). After conventional hierarchical clustering we systematically test for enrichment of Environmental Ontology terms and their abstractions in all possible clusters. This post-hoc algorithm provides a novel formalism that quantifies to what extend compositional and semantic similarity of microbial community samples coincide. We automatically visualize significantly enriched subclusters on a comprehensive dendrogram of microbial communities. As a result we obtain the hitherto most differentiated and comprehensive view on global patterns of microbial community diversity. We observe strong clusterability of microbial communities in ecosystems such as human/mammal-associated, geothermal, fresh water, plant-associated, soils and rhizosphere microbiomes, whereas hypersaline and anthropogenic samples are less homogeneous. Moreover, saline samples appear less cohesive in terms of compositional properties than previously reported.
| We here set out to map the entirety of available environmental microbiomes in order to discover the underlying compositional characteristics. For us it is intriguing to see which environmental factors influence the assembly of microbiomes. We collected many diverse environmental samples and annotated them with a restricted, yet structured set of ecosystem terms. We then cluster all samples and automatically detect which ecosystems cluster together.
The resulting map provides an overview of relatedness between microbial communities from all types of ecosystems, their global patterns of diversity, i.e. their compositional similarities and differences. The comprehensive structure of relations provides insights to habitat adaptation and assembly rules.
Finally, utilizing the background database of samples we can now put new samples into an environmental context.
| The often quoted tenet:“Everything is everywhere, but the environment selects” by Lourens Baas Becking has been subject to intense debate [1]. It gave rise to a series of hypotheses, how exactly the environment selects, i.e., which ecological rules are driving selection in which environment and whether they do so deterministically. The two competing theories for addressing this question are the ecological inference theory with a niche-based perspective [2, 3] and the neutralist random process theory [4]. Compact clusters of low beta diversity in microbial communities from the same environment indicate assembly determinism (i.e.), environmental factors predictably govern community composition. Conversely, if random processes and founder effects were the main drivers during community assembly, we would expect that this is reflected in high beta diversity and consequently low cluster homogeneity for samples from the same environment type. In order to elucidate these mechanisms as well as the environmental factors that drive bacterial community composition, it is necessary to develop a framework for comprehensive meta-analyses of microbial communities. To this end it is desirable to collect large, representative sets of samples from independent studies and diverse environments. In this light it is encouraging that the ever decreasing cost of DNA sequencing has led to a recent deluge of Metagenomics projects and Microbial Community profiling experiments in many, diverse ecosystems on the planet, e.g. the Human Microbiome Project and the Earth Microbiome Project [5, 6]. The primary data type for studying the community composition is the 16S rRNA gene, where hypervariable regions serve as phylogenetic markers [7]. Among the advantages of the 16S rRNA gene are the chronometric properties suitable for phylogeny construction and its widespread use to profile communities in all types of environments [8], i.e. to determine the relative abundance of the community members. Thanks to multiplexing and high read counts on Next Generation Sequencing (NGS) platforms, it is possible to generate a large number of samples with a single run on various recent platforms [7]. The 16S rRNA genes for many different bacteria have been sequenced and deposited in primary (GenBank) and secondary (GreenGenes, SILVA, RDP) databases [9–11]. However, community composition information, revealing co-occurrence of organisms, is lacking from these databases. On the other hand, microbial community collections such as those stored in the Sequence Read Archive [12] mainly focus on raw data deposition. MG-RAST [13] and CAMERA [14] (now defunct) are predominantly a repository for full shotgun Metagenomics. They do not maintain unified standards for Operational Taxonomic Unit (OTU) calling, i.e., the grouping of sequences into taxonomic levels of minimal sequence identity (commonly 97%). QIIME-DB (microbio.me/qiime) and the associated Global Environmental Sample Database (www.earthmicrobiome.org) are current efforts to overcome the above shortcomings but data deposition and retrieval methods are currently in a nascent stage.
Moreover, it is clear that the current community collections require formalisms to integrate metadata. Standards for data deposition (like “Minimum information about a marker gene sequence”, MIMARKS) have been introduced to address this problem [15], with ontology based knowledge management systems being an integral part of this. The introduction of an environmental ontology for ecosystems and -subsystems enables the semantic grouping and comparison of environments in an entirely new way: for example corals, dugong feces, ocean water, brine pools can all be associated to marine ecosystems; a relationship not automatically recognizable from pure text annotation. As a result, we can compare environments on various levels of abstraction and determine how widespread ecosystem-specific OTU compositions are. Nevertheless, quality control of these submissions remains difficult, as submitters might not be aware of the entire ontology structure and thus make non-optimal choices.
Clusterability of microbial communities has been investigated in the Human Microbiome Project, and various techniques and results, e.g. enterotypes have been presented [16] and debated [17, 18]. The effects of clustering methodology, distance metrics and taxonomic level of OTU picking are of great importance for the process of detecting clear-cut clusters of microbial communities. Importantly, traditional clustering algorithms for microbial communities are “uninformed” with respect to meta-data: the decision of partitioning is solely based on clustering structure and coefficients derived from beta diversity distances, disregarding useful semantic clues that ontologies can provide. Our philosophy is to postpone the decision to find meaningful clusters after a traditional hierarchical clustering structure is produced and ontology information for samples is taken into account. This is achieved by correlating the clustering structure with environmental categories—a novel approach in the realm of microbial community analysis. It is conceptually similar to the CLustering Enrichment ANalysis (CLEAN) described in [19], which integrates clustering of genes and their membership in functional categories such as Gene Ontology in the context of gene expression. We systematically generate and analyze a series of hypotheses to identify the extent that environments deterministically govern microbial community assembly. Interesting cases are those where the clustering structures (reflecting OTU compositions) and the post-hoc added environmental annotations coincide. This indicates which environmental factors were responsible for community composition. On the other hand, discrepancies can be further reconciled by considering additional meta-data (pH value, temperature) reflecting environmental differences or stochastic processes.
The feasibility of microbial community profile comparisons in meta-analyses depends on a number of aspects, such as standardization steps and sequencing platforms. In [20], Caporaso et. al show that biological conclusions were highly reproducible across lanes, read directions and Illumina HiSeq and MiSeq platforms. Other meta-analyses have demonstrated that microbial community samples are comparable across studies and platforms [7, 21, 22].
Previously Lozupone and Knight studied global patterns of bacterial diversity on the basis of a data set that comprised 202 samples from 111 studies [23]. The authors postulated that salinity and human-association are major environmental factors that drive community composition. They manually assigned 15 distinct environmental categories to all samples. While our approach draws a great deal of inspiration from this work, we extend and automatize it in various ways so to minimize sampling bias and to cope with the current and expected volumes of input data. First, the mentioned previous studies were performed on a small set of independent studies with comparatively low sequence counts. In fact, modern NGS platforms allow sample sizes far beyond 50 sequences (as used in [23]). This is more suitable for diversity studies, as more low abundance OTUs are accounted for. Our data collection is orders of magnitude larger, thus adds rigor to clustering observations but also demands appropriate storage and retrieval systems. The acquisition does not rely on single sequence retrieval from GenBank but builds on emerging repositories dedicated for 16S rRNA profiles that also provide community information and metadata. Together they provide a more representative snapshot of the Earth Microbiome and hence allow a more differentiated review of the early hypotheses on community assembly. Second, instead of just 15 manually defined environmental categories, we here describe how to to annotate samples using a suitable ontology, namely the Environmental Ontology [24]. Finally, we propose an algorithm that conducts cluster analysis with > 10,000 samples including exhaustive testing for enrichment of environmental attributes, borrowing techniques from Information Retrieval (for a thorough introduction, see [25]).
The overview of all steps involved in our data acquisition and analysis is provided in Fig 1 displaying three major components: (i) an integrated, comprehensive database of OTU-clustered 16S rRNA profiles annotated with ontological metadata descriptors employing EnvO, the Environment Ontology [24], (ii) a module to compare microbial communities utilizing a phylogeny based distance measure ([26]) and Hierarchical Clustering (iii) an Information Retrieval based post-hoc cluster analysis to test for enrichment of EnvO terms in the clusters of the dendrogram.
Large scale community comparisons must overcome innate sample differences such as uneven sampling size, different OTU calling methods and inconsistent or lacking environmental annotation. We composed a large meta-dataset from heterogeneous sources such as the QIIME-DB microbio.me/qiime/, Sequence Read Archive (SRA, [27]), a data collection provided by [28] (henceforth referred to as Chaffron dataset) and some locally sampled data. In total, we collected 20,472 distinct 16S rRNA from 2,461 different studies and stored them together with additional sample descriptions and meta data in a relational database (MySQL) for fast retrieval. Although the Chaffron data collection is composed of predominantly small samples (mainly non-NGS, lacking MIMARKS annotation), it proves to be valuable as samples are from 2297 independent studies, which contributes to the comprehensive nature of meta-analyses composed of global microbiomes and increases confidence beta diversity patterns. he sequence preprocessing includes: quality filtering, demultiplexing samples using QIIME’s split_libraries_fastq.py, and consistent closed-reference OTU calling against GreenGenes (version 13.5, 97% sequence similarity, using QIIME’s pick_closed_reference_otus.py). Closed reference OTU picking has a number of advantages (see also Discussion for caveats), namely it allows comparison of samples with different 16S rRNA regions, it comes with a high-quality phylogeny based on full-length sequences (thus facilitating phylogeny-based beta diversity calculation) and it is likely to filter chimera sequences. Also, for the sake of consistent data processing, note that UCLUST-ref [29] and GreenGenes were also used in the data sets acquired from QIIME-DB. According to [30], UCLUST-ref performs well in comparison to other reference and non-reference based methods. The projection of datasets on a limited set of reference OTUs incurs a loss in diversity. We therefore validate that the impact on beta diversity is within acceptable boundaries, i.e., beta diversity distances between open and closed reference OTU picking are correlated: for six environmental samples [31] where original sequences for the same 16S rRNA region are available, we compare beta diversity. FastTree [32] was used to construct the phylogeny for de novo OTU representatives, using QIIME’s make_phylogeny.py and pick_representatives.py with default parameters.
See Fig 1 for the different, data source specific steps and the overall workflow. To enable the integration with QIIME tools, we developed a web interface for custom BIOM table creation, a widely used format for ecological sample survey data [33]. The selection of samples is possible on the basis of EnvO annotation (including recursive sub-category) or presence/absence of lineages, metadata, and a combination of these criteria, thanks to a function for recursive EnvO traversal and the expressive power of SQL algebra.
The aim of Ontology annotation is to use controlled vocabulary for different types of environments to hierarchically catecgorize samples so to be able to relate clusters of communities to environmental determinants. We used text-mining (weighted Jaccard Index for phrase similarity [34]) to automatically annotate samples lacking MIMARKS annotation (SRA, Chaffron’s dataset and our own) with EnvO-terms based on sample description texts like isolation-source: we regard both sample description and EnvO-terms as bags of stemmed words to accomodate word order permutations and inflections. The weighted Jaccard expressing the similarity of two phrases A and B is then given by:
J ( A , B ) = ∑ w ∈ ( A ∩ B ) i d f ( w ) ∑ v ∈ ( A ∪ B ) i d f ( v ) (1)
where idf is the inverse document frequency of a word with respect to the Brown corpus [35].
EnvO is used by various projects [6, 13] to facilitate a principled approach towards environment classification by formalizing adequate naming conventions. It contains a rich, structured vocabulary (including synonyms), and it is arranged as a Directed Acyclic Graph, maintaining a general-to-specific order. We obtained the obo version of EnvO from obofoundries.org. We extend EnvO by creating new subclasses with terms that best describe environments of our database, including for example the human body site descriptions of the Human Microbiome Project. We also introduce missing semantic relationships, for example we describe “feces” (ENVO:00002003) as part of the gut (ID:0000002) to connect samples annotated with these respective terms that would otherwise appear unrelated. Multiple EnvO-terms can be associated to a sample, describing the biome, environmental feature and environmental material. We generated a subgraph for those EnvO-terms for which we found associations to microbial community samples. The graph coloring recursively assigns shades of the overarching ecosystem to its child nodes, while also reflecting multiple inheritence (Fig 2).
In order to programmatically make use of EnvO as a knowledge management system, we created a simple text parser that transforms obo-format into a richly annotated graph structure using networkx [36]. We also develop an API that lets us navigate along the graph structure. For validation of our annotation algorithm, we predict the best EnvO-term based on Eq 1 for 712 non-redundant sample descriptions and their annotations from all EnvO annotated QIIME-DB studies. We then calculate the minimal distance of the predicted and the manually annotated EnvO-terms (given in QIIME-DB as Environmental matter, Environmental feature and Environmental Biome). The distance is the shortest path in the undirected EnvO graph, such that exact matches have graph distance 0, EnvO-terms in direct subclass-superclass relation have distance 1, direct sibling nodes 2, cousins 4, etc. The results are shown in S5 Fig. It can be seen that from the 712 samples, most automatic annotations are in exact agreement (294) or in direct sub/superclass relation (117) with the manual annotation. By statistically summarizing all possible 7,122 graph distances in the EnvO graph, it turns out that the probabilities for two random nodes to be 0, 1, 2 and 3 steps apart are 0.06%, 0.20%, 1.18% and 3.42% respectively. Thus 73.3% of our annotations would have been predicted by chance with less than 3.5% probability. Note that manual EnvO annotations are a source of error as well and contribute to disagreement. We corrected for only very few blatant misannotations and consider therefore our accuracy estimates to be a conservative lower bound. Note that highlighting discrepancies can be used instructively during manual curation step in sample submission tools like QIIME-DB or MG-Rast.
Due to the multiple inheritence DAG (Multi-tree) structure of EnvO, it is possible to easily extend EnvO with further terms that capture high level concepts such as abstract ecosystems or environmental properties (saline, hypersaline environments, contamination and various subtypes of it, chemical enrichment such as nitrate, hydrocarbons). These overarching groupings also deal with EnvO’s attempt to include various, classification systems (including WWF, Udvardy and Baileys biomes). E.g., “Tundra”/“Tundra mire” or “Forest”/”Forest biome” are in entirely different branches of the ontology, despite the obvious semantic relation. We here propose high level ecosystems composed of distinct, yet related EnvO categories, see Table 1. Note that due to multiple inheritence and multiple annotations, a sample can fall into several ecosystems.
In order to compare samples it is highly recommendable to have comparable sample sizes. It is therefore common practice to apply rarefaction to samples that are to be compared, usually by randomly down-sampling. If we down-sample to the smallest sample size in the data set, this method unfortunately leads to a big loss of information when comparing samples of strongly differing sizes. E.g., representing communities of 10,000 different OTUs with only 10 or 50 OTUs neglects the majority of community members in the larger communities. On the other hand, if a larger target subsampling size is chosen for rarefaction, many small size communities have to be excluded.
We here device a method that for (almost) each pairwise sample comparison subsamples only to a size necessary for that individual pair, rather than subsampling all samples to the size of the smallest sample (as is commonly done with tools that perform rarefaction on BIOM tables, such as QIIME/UniFrac). By keeping the subsample size as large as possible, we increase the pairwise beta diversity distance precision. E.g., given 1001 samples A1 … A1000 of size 10,000 and B of size 1,000. We calculate beta diversity on subsamples of size 1,000 for Ai-B and 10,000 for all pairs Ai-Aj. The latter, comprising 99.8% of all comparisons, would have been substantially less accurate, had we downsampled all samples to 1,000. We validate the claimed precision improvement by performing jack-knifing (multiple subsampling) on traditional one-size-fits-all and adaptive rarefaction: we repeat the random subsampling and beta diversity calculation ten times for 60 samples of five different size categories. For each repetition, distance matrices are calculated using Weighted UniFrac. Principal Coordinate Analysis (PCoA) and three-dimensional PCoA plots were produced with QIIME scripts principal_coordinates.py and make_3d_plots.py, respectively. Finally we calculate standard deviations for corresponding positions in the upper half of the distance matrix, convert it to a one-dimensional vector and apply the t-test for related samples (using ttest_rel from scipy.stats).
We calculate a complete distance matrix for all samples, effectively requiring n(n-1)/2 comparisons, with n = 10,313. In order to account for phylogenetic similarities of involved OTUs, we use Weighted UniFrac as beta diversity. We motivate this choice as follows: although the usability as a distance metric has been questioned in [37], the criticism was addressed in [22]. Moreover, it was shown to be an instance of the more general Earth mover’s (aka Kantorovich-Rubinstein) distance metric [38]. UniFrac has been applied in over 1500 research publications in a wide range of microbial community comparison tasks.
Note that our approach of retaining only closed-reference OTUs avoids the construction of phylogenetic trees, a very time consuming and error-prone task, by trimming a comprehensive, high-quality tree (provided by GreenGenes) to the relevant OTUs. It further enables phylogenetic beta diversity calculation of samples with different 16S rRNA regions. Subsequently we apply Unweighted Pair Group Method with Arithmetic Mean (UPGMA), a form of agglomerative Hierarchical Clustering, using SciPy.
We calculate alpha diversity using QIIME’s alpha_diversity.py for each ecosystem independently, considering both phylogenetic (Phylogenetic Distance) and non-phylogenetic (Chao1, observed species) methods. Every sample is downsampled to a range of suitable sizes between 60 and 60,000 counts. The results are stored in the provided MySQL database.
After applying hierarchical clustering to all samples, we systematically analyze the resulting dendrogram structure and subcluster constituents. Hierarchical clustering yields a dendrogram encoded as a (n − 1) × 4 linkage matrix. For each possible cluster, we systematically test, whether it is enriched in any EnvO-category. We quantify this intuition by calculating precision, recall and F-measure: these tools, borrowed from information retrieval, express how well samples from a certain category cluster. A cluster containing predominantly members from one EnvO-category receives a high precision value. On the other hand, high recall is achieved if most category members are also subsumed under a cluster. The pseudo code is provided in Algorithm 1 (available at https://goo.gl/70LsQi). Moreover we determine the cluster coefficiencts describing the compactness of a cluster. High homogeneity (high intra-cluster similarity) and separation (low inter-cluster similarity) are indicators for a distinct, compact set of samples [39]. Homogeneity is determined by average distance between all members of the cluster, separation is the average distance to all members outside the cluster.
Our method can be best compared to the CLustering Enrichment ANalysis (CLEAN) score [19], where environmental samples are the equivalent to genes (clustered by gene expression levels) and Environment Ontology annotation corresponds to membership of genes in functional categories such as those from Gene Ontology. We therefore compare the F-measure to Fisher’s exact test, which is central to the CLEAN score: we test to what extend the Fisher’s exact test and F-measure coincide wrt. significance for a given contingency table. We perform an overarching grid search for the two significance thresholds for both tests. For each significance threshold setting we consider a set of representative category and cluster sizes, calculate all possible contingency tables and measure when each test would call this significantly enriched. Note that we define the two groups of the contingency table to be all samples belonging to a cluster and all those that do not, respectively. We then count the percentage of cases of (dis-)agreement between the two tests.
We imposed the property “saline” on high level EnvO-terms such as “marine feature” (ENVO:01000031), “marine water body” (ENVO:00001999) and “saline hydrographic feature” (ENVO:00000017). Consequently all subsumed EnvO-terms (incl. saline lake, ocean, lagoon etc.) inherit this property. The database lookup for EnvO-sample associations then facilitated fast and convenient identification of salinity related samples. Those samples are then marked in Suppl. S3 Fig.
Algorithm 1 Bottom-up algorithm to determine dendrogram-clusters enriched in EnvO-terms. The dendrogram is the result from the hierarchical clustering (UPGMA) and encoded as linkage matrix.
LinkageMatrix = UPGMA(beta diversity Distance Matrix)
for all rows row in LinkageMatrix do
form new cluster c bottum-up from two subclusters as specified in row
homogeneity = average distance (sample1, sample2) ∀sample1, sample2 ∈ c
separation = average distance (sample1, sample2) ∀sample1 ∈ c, sample2 ∉ c
if homogeneity/separation < thresholddensity then
Document dense cluster (c, homogeneity, separation)
end if
for all EnvO-terms e present in c do
r
e
c
a
l
l
=
samples with e in c samples with e in total
p
r
e
c
i
s
i
o
n
=
samples
with
e
in
c
samples
with
in
c
F
1
=
2
⋅
p
r
e
c
i
s
i
o
n
⋅
r
e
c
a
l
l
p
r
e
c
i
s
i
o
n
+
r
e
c
a
l
l
if F1 > thresholdenrichment and studies(c) > 1 and Ontology-Depth(e) > 1 then
Document enriched cluster (c, e, homogeneity, separation)
Color linkages of c in dendrogram according to e
end if
end for
end for
We here describe a microbial community analysis framework extending conventional pipelines by including three novel components, which all help to enable all-encompassing meta analyses of 16S rRNA samples: (i) the creation of a 16S profile database from heterogeneous sources, (ii) adaptive rarefaction to compare large sets of samples of strongly differing sizes with minimal loss of information, and (iii) a post-hoc clustering algorithm that tests for enrichment of environmental categories (and their respective abstractions) after conventional hierarchical clustering of phylogeny based beta diversities. As a result we obtain the hitherto most differentiated and comprehensive view on global patterns of microbial community diversity with automatically detected enriched subclusters. It provides indicators for environmental factors that drive community assembly. Fig 1 summarizes all steps in our framework.
The creation of a comprehensive relational database of 16S rRNA community profiles constitutes an early result of our work. The database integrates various heterogeneous data sources. In total, we collected 20,472 samples comprising 6,331,600 sequence-sample associations from 2,462 independent studies. 10,313 of these samples are of suitable size for diversity studies (i.e. > 2000 sequences, according to [7]). Performing closed-reference OTU picking against a consistent reference (GreenGenes 13.5) yielded 40,164 OTUs, corresponding to 40.44% of GreenGenes’ 97% sequence identity clusters. Our pipeline makes use of OTUs being from a closed reference as it allows fast, yet phylogeny-sensitive beta diversity calculations without reconstruction of phylogenies.
To estimate the loss of diversity incurred by closed reference OTU picking, we list the number of dropped sequences for inhouse samples in Supplementary S2 Table. It shows that also for the case of environmental samples, a substantial amount of sequences is retained. We also measure the impact on beta diversity and compare to the corresponding results from de novo OTU picking. The beta diversity distances of the two methods are strongly correlated (Pearson correlation 0.82), thus justifying closed reference OTU picking as a proxy (see further contemplations in Discussion).
An overview of the data integration steps (including results) is shown in Fig 1(a). The central tables of the database are for sample description (ID, isolation source, associated publication, further meta-data), OTUs (GreenGenes identifier, RDP lineage, sequence) and sample-OTU association. Further, the database contains various sample annotations (EnvO, meta-data, alpha diversity, ecosystem coloring, sample size). The database scheme is provided in Supplementary Material, S2 Fig. All tables are appropriately indexed. The database is freely accessible through http://ecophyl.info/phpmyadmin/ (User login is provided upon request from corresponding author) or downloadable as an SQL dump. We argue that this form of storage is a viable concept to integrate the current and anticipated amounts of microbial community data such that fast and powerful queries are possible. To elaborate meta-study compositions, the relational algebra of SQL allows to combine conditions on sample annotations and properties as well as OTU annotations. For example, it is straightforward to retrieve all samples of a given alpha diversity from all subtypes of marine environments containing OTUs belonging to the Vibrio genus. For identifying environment subtypes, we employ EnvO’s General-to-Specific ordering of environments arranged along a Directed Acyclic Graph. EnvO-terms that were used to annotate samples are shown in a color coded Ontology subgraph in Fig 2. Finally, our framework includes a tool for BIOM table creation upon sample selection (see Fig 1(b)) for customized meta-analyses using adaptive rarefaction or further integration with tools like QIIME [40].
The acquired profiles stem from different sequencing platforms and hence differ strongly in sampling effort/sequencing depth. In order to calculate alpha and beta diversity for a collection of heterogeneous samples, it is common practice to subsample all samples using rarefaction to a size smaller than the smallest sample to be included. This one-size-fits-all rarefaction seems unsuitable for a large set of samples with strongly varying depth as it incurs either strong downsampling on large samples (which might yield non-representative subsets) or exclusion of many smaller samples below a certain threshold. To address this problem, we provide an algorithm that for each pairwise beta diversity calculation rarefies samples only to a size necessary for the individual pair at hand, rather than subsampling all samples to a size below the smallest sample to be included. We are able to show that adaptive rarefaction produces more accurate beta diversity distances: multiple subsampling repetitions (jack-knifing) with static and adaptive rarefaction lead to significantly smaller distance variances for adaptive rarefaction (as shown by Student’s t-test, p = 3.3 × 10−258, S4(c) Fig. We also observed that for nearly all cases the distances from adaptive rarefaction were strictly contained by the range of distances from static rarefaction. We also visualize this process using three-dimensional PCoA plots, where larger uncertainty ellipsoids for traditional rarefaction indicate larger variance, see S4(a) and S4(b) Fig. The runtime of the Adaptive rarefaction algorithm is O(n2∣P∣), where n is the number of samples to be compared and ∣P∣ is the size of the reference phylogeny (the GreenGenes phylogeny, in our case).
The adaptive rarefaction component is shown in Fig 1(c). The main result of adaptive rarefaction is a high accuracy distance matrix for 10,313 samples.
Fig 3 shows that soil-, plant- and marine ecosystems are most diverse in terms of Alpha diversity. We can further break these findings down to subsumed environments. This reveals that among soil environments, farm soils are most diverse (Suppl. S1 Fig).
In order to test, whether a cluster is significantly enriched in environmental categories, we determine to what extend EnvO annotations —at all abstraction levels— are predominant in any part of the dendrogram, see Fig 4. We developed an algorithm that performs this task as a post-hoc clustering analysis, Fig 1(e), i.e, after Hierarchical Clustering Fig 1(d). The pseudo code is given in Algorithm 1 and a more detailed explanation is in Fig 5. It constitutes a rigorous formalization and implementation of the manual process outlined in [23], performed on tens of thousands of microbial community samples. It automatically identifies clusters enriched in environmental categories based on precision, recall, F1-score, number of studies and cluster coefficients homogeneity and separation (see Methods). The algorithm output of enriched clusters together with their respective EnvO-terms (F1-score > 0.5) is provided in Table 2. Ocean floor, bed (the portion of the ground surface which lies below water), grassland soil, small lake bioms, gut and animal associated habitat all are enriched in identified clusters and therefore seem to bear significant compositions that are driven by their respective environmental conditions. The color coding of these findings into the Hierarchical Clustering dendrogram shows that environmental samples cluster non-randomly (Fig 4). Although few inconsistencies persist, major ecosystems are recognizable, as soil samples, freshwater, rhizosphere, geothermal and the majority of human/animal-associated all fall together in respective clades of the UPGMA dendrogram. The clusterability of these environments is corroborated by visually inspecting Principal Coordinate Analysis plots, see Fig 6. There, the first Principal Component largely separates human and and environmental samples, while the second component helps to identify clusters for soil, marine, freshwater and plant-associated samples.
We regard clusters with F1-score > 0.5 as enriched in an EnvO-term and list them ordered by descending F1-score. We further report the count of samples associated to the dominating Envo-Term in a cluster (count), the cluster size (cluster), the total count of samples associated to the dominating Envo-term (Total), the number of independent studies (Studies), generality of an EnvO-term (Generality), measured as the hierarchical level in the Directed Acyclic Graph of EnvO, precision, recall, F1-score, homogeneity and separation. The list is filtered by EnvO-terms that appear at least in three independent studies.
Marine samples fall into two separate clusters, as shown by both UPGMA and PCoA. Cluster “Marine 1” is mainly composed of samples from QIIME-DB studies 1222 (Bergen Ocean Acidification Cosms), 1235 (Fjord mesocosms) and 1240 (Western English Channel time series). Interestingly, given the overall picture of beta diversity distribution, these marine samples cluster well despite their geographically different sampling locations. Further, small scale studies also fall into this cluster, see ecophyl.info/html/SI/PCOA/PCoA_Marine. Cluster “Marine 2” is composed of marine sediments/contaminated marine environments (QIIME-DB studies 1046, 1039 and 1198). It is remarkable that “Marine 2” appears closer to soils, again confirmed by both UPGMA and PCoA.
Plant samples also fall into two main clusters. Samples from QIIME-DB study 1792 (maize rhizosphere) appear close to marine samples in the PCoA plot (principal components 1 and 2), however, UPGMA separates them more clearly from marine samples. QIIME-DB studies 2019, 1690 and 1689 constitute the second plant cluster.
The human/animal-associated cluster is clearly separate from environmental samples: it has a separation score of 0.8512, the highest of all clusters. On the other hand it is not very homogeneous, as evidenced by large branches within the cluster and a homogeneity of 0.5849, i.e., it is among the least homogeneous of the detected clusters. Moreover, a smaller cluster of human/animal associated samples groups better with soil samples, possibly due to compositional similarities to soil samples with fecal contamination.
We also scrutinized the compactness of ecosystems by investigating cluster coefficients of all samples related to an EnvO-term, regardless of the clustering structure. We observe that hypersaline samples show the least homogeneity as compared to all other ecosystems, see Table 3. suggesting that these extremophiles get recruited predominantly through non-deterministic processes or unaccounted environmental factors.
Our approach differs from traditional “uninformed” clustering methods, as it identifies EnvO-category enriched clusters in a posthoc analysis. Any (hierarchical) clustering method is suitable in combination. Note that samples from the same environmental category can naturally occur in remote parts of the dendrogram as a result of stochastic assembly processes in some environments or insufficient subcategorization of environmental categories. I.e., this phenomenon is not a short-coming of the clustering process nor the enrichment test. We show that the systematic enrichment test discovers meaningful clusters that would have been otherwise overlooked by classic “uninformed” clustering methods. Compact clusters (low homogeneity/separation ratio) are not necessarily enriched in any EnvO-category (in terms of F-measure). As shown in Supplementary S1 Table, amongst the 245 most compact clusters, only eight have an F-measure above 0.5.
We further validate the F-measure in our post-hoc analysis by comparing to the Fisher’s exact test (as used in [19], see Methods). The results are shown in Fig 7. It can be seen that for the commonly used thresholds for Fisher’s exact test (−log(p) ∈ {2, 3, 5}), the disagreement comes exclusively from cases that are considered significant by Fisher’s exact test but not by F-measure threshold (green bars). In other words, the F-measure is for most thresholds a stricter test, as it only reports a small subset of Fisher’s exact test as significant.
We revisited the hypothesis postulated in [23], stipulating that salinity largely explains observed patterns of community assembly. In contrast to this, our (much larger) collection of saline samples falls mainly into two clearly separate clusters (see Supplementary S3 Fig), where marine samples and marine sediment samples are the largest contributors, respectively (also cf. Figs 4 and 6b). The observed seafloor/seawater split is in accordance with recent studies on these two realms (e.g. [41]), but our multi-environment contextualization additionally puts these relative differences into a global perspective. Moreover, we observe various outliers. Visual inspection showed that closely related (incl. non-saline) samples in and around the marine sediment cluster are rich in hydrocarbons or molecular nitrogen, thus promoting high relative abundance of Proteobacteria (which are known for their functions in nitrogen fixation and oil-spill related hydrocarbon degradation). We thus hypothesize that these factors are more influential for the assembly of certain communities than salinity.
We here presented a comprehensive effort towards revealing global patterns of beta diversity. To this end we collected a broad range of 16S rRNA profiles of environmental and host associated microbiomes from diverse sources and independent studies. After data integration in a relational database, we showed how to efficiently calculate mutual phylogenetic beta diversity distances (weighted UniFrac) without the information loss in comparison to normal rarefaction. Moreover, our meta-analysis was driven by ontological environmental meta-data information: EnvO-term enriched clusters were automatically detected and used to visualize the emerging global patterns of the Earth meta-community.
Systematically correlating beta diversity dendrograms with environmental annotations is motivated by the idea that oftentimes similar environments select their constituents in similar ways. We also observe samples from the same environment with low homogeneity (i.e. large distances amongst category members). This phenomenon can be either explained by incorrect annotation, by random processes governing assembly or true differences in the same environmental category, which is then instructive for meaningful EnvO subcategorization. On the other hand, it is intriguing to study dense beta diversity clusters with seemingly inconsistent, i.e. diverse environment annotations. Exhaustive scrutiny of environmental parameters might then reveal commonalities amongst those samples and thus explain the low beta diversity. This approach depends on a sufficient, consistent metadata collection for a large set of samples in the future. The systematic integration of metadata into future visualization techniques, as shown here with EnvO-terms, will serve as a new form of hypothesis generation. For example, dense clusters of mixed categories can be explained by latent variables such as high levels of hydrocarbons. If eventually all these potential assembly drivers are consistently captured in the metadata, our method can then be adapted to data mine beyond EnvO annotation. Conversely, exploiting the semantic hierarchy of ontologies, it might also be convenient to extend EnvO to capture over-arching concepts such as “contaminated” or “hydrocarbon-rich” sites. We demonstrated the feasibility of this approach for the analysis of saline vs. non-saline samples.
Occasionally, EnvO mis-annotation or inadequate choice of EnvO-terms incorrectly accounts for mixed clusters. Our method can then point to the source of error and function as an annotation curator, in particular now that our reference dataset is large and robust enough to redline suspecious outliers as demonstrated for the alleged soil microbiome (Fig 6).
With the advent of large comprehensive microbial community databases, we anticipate that it will be possible to provide a context of similarly composed environments for new 16S rRNA profiles, just as BLAST and other sequence comparison tools in conjuction with large sequence databases help to elucidate single sequences. To classify these in a hierarchy (dendrogram) rather than just ranking gives a contextual impression similar to phylogenies for sequences over sequence similarity rankings.
A number of challenges and current limitations prevail, though. Arguably, while the discussed advantages of the chosen components in the proposed pipeline in our opinion justify their application, we also note their caveats here and mention potential alternatives. Cross-plattform meta analyses have to deal with inherent protocol and technology biases: DNA extraction kits and sequencing platforms are afflicted with specific errors, different variable regions differ in informative power to distinguish phylogenetic clades, so the choice of primers impacts relative OTU abundances. Additionally, alignment quality for reference phylogenies and differences in sequence filtering can skew beta diversity calculations [42]. Environment Ontology annotations are far from being perfect, as neither automatic nor manual methods guarantee 100% accuracy. OTU clustering methods often produce surprisingly different results, as argued in [43]. Closed-reference OTU picking discards all sequences that do not match a provided reference (GreenGenes in our case) and thus represents the original sample inaccurately, especially in understudied environments with high diversity such as soils. Alternatively, open-reference or de novo OTU clustering could be employed, yet this is a daunting task with its own caveats: the considerably higher computational effort is less parallelizable, the number of comparable samples is restricted by the choice of 16S rRNA region and the lack of a high-quality phylogeny based on full sequences severely impedes downstream phylogenetic beta diversity calculation. The latter can be addressed by resorting to non-phylogenetic beta diversity but that would disregard evolutionary relatedness of OTUs. Moreover, as pointed out in [7], short reads stretching only over one or few variable regions might be unsuitable for de novo OTU picking.
Rarefaction introduces a loss of information, and albeit substantially reduced, it still persists with our proposed method of adaptive rarefaction. One concern regarding the generality of our results is due to the sampling bias induced by very large scale studies appearing as sole contributors to a certain environment. As can be seen in the interactive coloring of PCoA plots (ecophyl.info/html/SI), the ecosystem distribution and particularly EnvO-distribution is similar to study distribution. Beta diversity patterns could thus be explained to some extend by study methods (and possibly systematic artefacts). However, many independent studies do confirm characteristic beta diversity distributions, especially for soils and human-associated samples, when including small scale studies. Our visualization techniques reflect this important measure of confidence.
Likewise, some large studies like high resolution time series can give a false impression of community similarity (i.e, tight clustering in PCoA and UPGMA plots) in a certain environment, if a single study dominates the contribution. As a remedy, we would be tempted to downsample on study level, i.e. select only few representative samples for each study. However, genuine beta diversity has been captured in large scale studies such as [44], so that simple exclusion of samples from large studies will lead to loss of important information. In addition to uneven sample size, the aspect of uneven study size should be addressed by future meta analyses. The environmental equivalents of enterotypes (clusters of gut microbiomes and conceptually extended in [17] to all human-associated environments), are barely observable throughout the entirety of samples used in our meta-analysis. In accordance with Koren et al [17], we detect gradients of microbiomes for nearly all ecosystems. Compact clusters with high homogeneity/separation ratio often only exist when single studies constitute the bulk of an ecosystem (or -subsystem) and thus require further evidence to be considered an “environmental enterotype”.
Our generated beta diversity maps also suggest that many studies still tap into unchartered territory, indicating that the beta diversity space for the entirety of Earth microbiomes is far from being fully explored. Despite our best attempts to collect a comprehensive dataset, various samples still appear rather unique. Note that the large collection of millions of pairwise beta diversities was a prerequisite to develop a sense of uniqueness. However, increased coverage of similar environments in the future will inevitably either improve confidence in uniqueness or find closer matches. The effect of microbiome uniqueness is exacerbated, when we start including OTUs beyond closed references.
Systematic errors specific to studies and/or sequencing technologies are still a concern. We therefore report in our final result table the number of studies that support a cluster enrichment and developed an interactive visualization that allows to inspect PCoA clusters color-coded by EnvO-annotation, by ecosystem or by study. Despite the discussed limitations of cross-study meta analyses, we believe that by integrating as many samples as possible, a global picture of diversity is emerging due the law of large numbers. We are encouraged by a number of similar (albeit substantially smaller) works corroborating the observation that samples do cluster by environment rather than just by study [7, 21, 22].
|
10.1371/journal.ppat.1003180 | Sustained Activation of Akt Elicits Mitochondrial Dysfunction to Block Plasmodium falciparum Infection in the Mosquito Host | The overexpression of activated, myristoylated Akt in the midgut of female transgenic Anopheles stephensi results in resistance to infection with the human malaria parasite Plasmodium falciparum but also decreased lifespan. In the present study, the understanding of mitochondria-dependent midgut homeostasis has been expanded to explain this apparent paradox in an insect of major medical importance. Given that Akt signaling is essential for cell growth and survival, we hypothesized that sustained Akt activation in the mosquito midgut would alter the balance of critical pathways that control mitochondrial dynamics to enhance parasite killing at some cost to survivorship. Toxic reactive oxygen and nitrogen species (RNOS) rise to high levels in the midgut after blood feeding, due to a combination of high NO production and a decline in FOXO-dependent antioxidants. Despite an apparent increase in mitochondrial biogenesis in young females (3 d), energy deficiencies were apparent as decreased oxidative phosphorylation and increased [AMP]/[ATP] ratios. In addition, mitochondrial mass was lower and accompanied by the presence of stalled autophagosomes in the posterior midgut, a critical site for blood digestion and stem cell-mediated epithelial maintenance and repair, and by functional degradation of the epithelial barrier. By 18 d, the age at which An. stephensi would transmit P. falciparum to human hosts, mitochondrial dysfunction coupled to Akt-mediated repression of autophagy/mitophagy was more evident and midgut epithelial structure was markedly compromised. Inhibition of RNOS by co-feeding of the nitric-oxide synthase inhibitor L-NAME at infection abrogated Akt-dependent killing of P. falciparum that begins within 18 h of infection in 3–5 d old mosquitoes. Hence, Akt-induced changes in mitochondrial dynamics perturb midgut homeostasis to enhance parasite resistance and decrease mosquito infective lifespan. Further, quality control of mitochondrial function in the midgut is necessary for the maintenance of midgut health as reflected in energy homeostasis and tissue repair and renewal.
| Malaria is a major public health problem in the world and various strategies are under development for control, including vaccines and transgenic mosquitoes that block parasite transmission. We previously reported that overexpression of the major signaling protein Akt in the midgut of female Anopheles stephensi mosquitoes could impart resistance to infection with the most important human malaria parasite and also reduce the duration of mosquito infectivity to human hosts. However, to use this strategy for malaria transmission control in endemic areas, we must understand the mechanism by which parasites are killed to ensure that transmission of other human pathogens (e.g., viruses, nematodes) is not unexpectedly enhanced and to allow the design of rational, preventive interventions. Here, we report that overexpression of a constitutively active Akt in the mosquito midgut alters important cellular, and in particular, mitochondrial processes – in a manner similar to Akt control of these processes in mammalian cells – to generate high levels of toxic compounds that kill parasites within hours after infection. However, the same alterations in mitochondrial processes that result in parasite killing ultimately reduce mosquito infective lifespan for transmission, indicating that mitochondrial dynamics in the mosquito midgut could be targeted for multi-faceted genetic control of mosquito biology to reduce malaria transmission.
| Malaria is one of the greatest public health threats worldwide and is caused by infection with protozoan parasites of the genus Plasmodium that are transmitted by Anopheles mosquitoes. Shortly after an infective bloodmeal is consumed by the female mosquito, which can occur as early as 3 d of age, zygotes form and develop into motile ookinetes in the midgut lumen. Ookinetes must successfully traverse the midgut epithelium to form non-motile oocysts that grow and develop on the outside of the midgut for a minimum of 12 d. Within 14–16 d of ingesting a parasite-containing blood meal (or at 17–19 d post-emergence of the mosquito), oocyst-derived sporozoites invade the salivary glands to yield a mosquito that is infective to humans for the duration of her life. Despite this need for lengthy development, only a small percentage of mosquitoes under natural conditions live long enough to become fully infective [1]–[3].
Akt is a key signaling molecule in nearly all eukaryotes and regulates a variety of physiological processes in a tissue dependent manner. In mosquitoes, Akt regulates immunity, lifespan, reproduction, metabolism and diapause [4]. We previously demonstrated that increased Akt signaling in the midgut of the female malaria mosquito Anopheles stephensi disrupted development of the human malaria parasite Plasmodium falciparum and concurrently reduced the duration that mosquitoes are infective to humans [5]. Specifically, overexpression of constitutively active Akt (myristoylated Akt or myrAkt) in heterozygous (HT) transgenic An. stephensi reduced parasite infection by 60–99% relative to non-transgenic (NTG) controls. Of those mosquitoes that were infected, we observed a 75–99% reduction in parasite load. Homozygous (HM) transgenic mosquitoes were resistant to parasite infection. The increase in midgut-specific Akt signaling also reduced the average mosquito lifespan by 18–20% and the window of opportunity to transmit malaria parasites by 50% relative to controls. Thus, activation of Akt signaling reduced the number of infected mosquitoes, the number of malaria parasites per infected mosquito, and the duration of mosquito infectivity relative to NTG controls. While these findings are significant, the safe application of this or any similar strategy for malaria transmission control requires identification of the mechanism(s) whereby parasites are killed to ensure that transmission of other human pathogens (e.g., viruses, nematodes) is not unexpectedly enhanced and to allow the design of rational, preventive interventions.
The signaling protein Akt is not only a key mediator in insulin and insulin-like growth factor signaling (IIS), but it also integrates signals from other growth factor-activated tyrosine kinase receptors as well as activated G-protein-coupled receptors and integrins to regulate a wide array of downstream proteins involved in cell growth, cell survival, and metabolism [6]. Although activation of Akt is required for fundamental processes, sustained activation of the Akt pathway – as observed in PTEN haploinsufficient or null mice – can lead to abnormal behavior [7], [8], disrupted autophagy [9], and mitochondrial dysfunction with accumulation of mtDNA deletions [8]. Furthermore, enhanced or defective clearance of damaged mitochondria by autophagy or mitophagy [10], [11] has been reported in several disorders with dysregulation of Akt [12], [13], [14]. Increases in PTEN, in contrast, lead to increased oxidative phosphorylation and increased tumor resistance [15].
Removal of damaged mitochondria occurs via general autophagy or by a selective form of autophagy termed mitophagy [16]. In general, signals that promote mitophagy and autophagy, in addition to oxidative stress-mediated damage, include activation of the stress-associated kinases (p38 MAPK and JNK), ERK-dependent signaling, energy deprivation, some pro-inflammatory cytokines as well as Toll-like receptor (TLR) and peptidoglycan recognition protein (PGRP) signaling [17], [18]. Signals that inhibit mitophagy and autophagy include the signaling axis surrounding Akt (phosphatidylinositol-3 kinase [PI-3K] upstream of Akt and target of rapamycin [TOR] downstream of Akt), allergic inflammation-associated signals including interleukin (IL)-4, IL-13, and some microbial virulence factors [17], [18]. Autophagy is critical for the regulation and coordination of cellular homeostasis, epithelial barrier integrity, stem cell maintenance and differentiation, lifespan, and immunity [17]–[21]. For example, autophagy in C. elegans daf-2 mutants – which are both long-lived and infection-resistant – is necessary and sufficient for pathogen resistance and lifespan extension under reduced IIS activation [22], [23]. Thus, balance between mitochondrial biogenesis and these positive and negative signals for the clearance of damaged mitochondria is required for adequate levels of steady-state functional mitochondria.
In light of observations from nematodes to mammals and the phenotypes of enhanced anti-parasite resistance and reduced lifespan in our transgenic mosquitoes, we hypothesized that myrAkt overexpression in the midgut of An. stephensi would lead to mitochondrial dysfunction that could impact resistance to P. falciparum infection and infective lifespan. Specifically, we predicted that Akt overexpression would increase steady-state oxidative stress through a reduction in antioxidants, including FOXO-dependent mitochondrial MnSOD and perhaps glutathione-S-transferases [24], [25] and through enhanced NO production, based on previous observations of ROS-induced NOS expression in An. stephensi cells [26]. Given the reciprocal interaction between PTEN levels and mitochondrial dysfunction [8], [15], we hypothesized that reduced oxidative phosphorylation, resulting from oxidative damage by increased toxic reactive nitrogen and oxygen species (RNOS) contributed in part perhaps from enhanced mitochondrial ROS production, would feedback to increase damage if not accompanied by concomitant increases in antioxidant and repair enzymes. While a significant upregulation of toxic RNOS could be sufficient to kill parasites, we reasoned that RNOS-damaged midgut architecture could also contribute to anti-parasite resistance and, at the same time, alter infective lifespan. For example, low MnSOD activity, which in the presence of normal or high NO production would result in peroxynitrite formation within mitochondria [27], [28], could disrupt the activity of several targets, in particular, Complexes I [9], [29] and V [30], [31] and, possibly, remaining MnSOD [32], [33]. Although the resulting high oxidative stress should be a potent signal for activating autophagy and/or mitophagy, Akt overexpression would antagonize RNOS-mediated activation, resulting in reduced clearance of damaged mitochondria, altered biogenesis, and concomitantly higher RNOS. Hence, despite the reduction in parasite infection, Akt-induced metabolic changes would lead to significantly altered mitochondrial dynamics and damage to the midgut epithelium, the primary site for nutrient acquisition from ingested blood.
Previous studies from Oliveira et al. [34] and Kumar et al. [35] using a strain of Anopheles gambiae (L3–5) genetically selected for refractoriness (R) to infection with the simian parasite Plasmodium cynomolgi, reported that resistance to infection was related to a heightened state of oxidative stress resulting from lower antioxidant defenses based on morphological and microarray mRNA expression analysis. In particular, they observed decreased State 3-dependent oxygen uptake in midgut and thorax, increased midgut ROS production (from Nox, Duox or mitochondria), and decreased fat stores with increased transcript expression of several glycolytic enzymes. When Oliveira et al. [34] silenced the mitochondrial adenine nucleotide transporter (ANT) in sensitive mosquitoes, they observed increased ROS production and a recapitulation of “resistance to infection” observed in R mosquitoes. Although the only glycolytic transcript increased was lactate dehydrogenase (LDH), the authors concluded that a “metabolic shift” sustained oxidative stress to favor killing of parasites, driving resistance to infection. A decrease in State 3-mediated oxygen uptake (especially in thorax and less in midgut) and increased LDH transcript, however, would not favor beta-oxidation of fatty acids, a mitochondrial process, to explain the decrease in fat deposits. Rather, these changes would favor glycolysis as a main source of ATP in an attempt to cope with the lower output of mitochondrial ATP. Midgut epithelium is a highly aerobic tissue, for which the electrolyte balance would not be supported by glycolysis alone. Thus, these changes do not appear to be “adaptive” with a temporary implication, rather they are terminal, dramatic changes that would result in increased resistance (less infection) but with a higher cost (shorter lifespan). Although this model of malaria resistance is appealing, it is not possible to definitively identify which (or any) of the detected metabolic changes define the mechanism of resistance given that resistance could result from altered expression of more than one gene in more than one tissue. Indeed, Oliveira et al. [34] suggested that the changes associated with increased resistance might be affected by a regulatory protein such as “a constitutively active transcription factor or a non-functional suppressor of a signal transduction pathway”. In this study, we have used a well-defined model of malaria resistance based on expression of constitutively active myrAkt in the midgut of female mosquitoes [5] to elucidate the mechanism underlying resistance to infection. Here, sustained activation of Akt functions broadly to control mitochondrial dynamics in malaria resistance. This control is initiated as an overproduction of NO and resistance is sustained as an imbalance of mitochondrial biogenesis and autophagy. This fundamental imbalance perturbs midgut homeostasis or “midgut health” to mediate the mosquito response to infection and the infective lifespan.
To identify the molecular targets affected by Akt overexpression in An. stephensi, we used differential LC-MS/MS to identify proteins that were over- or underrepresented in 3–5 d old, female myrAkt HM and HT An. stephensi relative to age-matched, NTG female An. stephensi. Analysis of the An. gambiae and An. stephensi genome data sets yielded largely concordant results; additional related matches from other mosquito species were detected through limited analyses of available data.
Among 477 identified proteins, a total of 38 proteins (excluding the transgenesis eye marker DsRed and multiple hits to the same protein in different mosquito genomes) were shared between HM and HT An. stephensi that were not evident or were reduced in NTG mosquitoes (Table 1). Notably, of these 38 proteins, a large proportion (14/38 or 37%) were associated with mitochondrial processes, including Krebs' cycle, electron transport chain subunit assembly, protein folding, and mitochondria-specific oxidative stress responses. The latter included an antioxidant enzyme, a peroxiredoxin V (PrxV) ortholog, and three additional proteins generally associated with oxidative stress response, including chaperones (heat shock protein), protein refolding (protein disulfide isomerase) and reduction of oxidative modifications (aldo-keto reductase). In Drosophila melanogaster, prxV −/− mutants exhibited reduced survivorship under oxidative stress, while overexpression of PrxV enhanced oxidative stress resistance and lifespan [36]. In other studies, PrxV was associated with redox regulation during bacterial infection of the midgut in D. melanogaster, but overexpression of PrxV was unexpectedly associated with reduced fly survivorship relative to controls following infection [37]. While the association of PrxV overexpression with reduced lifespan under some circumstances is intriguing, it is more likely that overexpressed PrxV here is a limited response to oxidative stress, perhaps protecting only a small subset of mitochondrial proteins [38].
The overrepresentation of proteins from Krebs' cycle and the electron transport chain (ETC) was interpreted as increased mitochondrial biogenesis to enhance oxidative phosphorylation (OXPHOS) output and perhaps to replace oxidatively-modified mitochondrial proteins. An increase in OXPHOS output would be consistent with elevated levels of the alpha subunit of Na+/K+-ATPase (Table 1). This enzyme, known to be located in the basolateral plasma membrane close to mitochondria-enriched fractions [39], catalyzes the hydrolysis of ATP coupled with the exchange of Na+ and K+ across the plasma membrane creating an electrochemical gradient, which sustains the resting membrane potential as well as provides the energy for active transport of various nutrients [40], and is one of the major ATP-consuming units of the cell [30].
Among the 14 proteins that were underrepresented in myrAkt An. stephensi, 7 proteins (50%; Table 1) were associated with epithelial and chromatin integrity. In particular, perlecan, collagen, and laminin – key components of the extracellular matrix (ECM) – were underrepresented in HM and HT An. stephensi. ECM proteins can be degraded and fragmented by oxidative stress [41], [42], suggesting that ECM/epithelial integrity was compromised in myrAkt An. stephensi. Histones H2B, H3, and H4 were also underrepresented in myrAkt An. stephensi relative to NTG mosquitoes. Histones are essential for chromatin packaging and DNA damage has been identified as a major mediator of chromatin reorganization and histone loss [43], suggesting that oxidative damage to DNA in myrAkt An. stephensi at 3–5 d post-emergence was likely to be high.
Four cytoskeletal proteins were overrepresented in myrAkt An. stephensi, including orthologs of microtubule-associated protein 1A and myosin light chain, proteins that are critically associated with mitophagy and autophagy, respectively [44], [45]. A third cytoskeletal protein, troponin T, a member of the troponin complex, regulates the interaction of myosin light chain with actin via tropomyosin, and, therefore, is associated with actin-myosin contractility. Mutations in troponin T in mice have been associated with mitochondrial degradation and the formation of increased numbers of small, round mitochondria with loss of well-defined membranes and cristae [46], indicating the importance of this protein in mitophagy and mitochondrial structure and function. Spectrin was also overrepresented in HM and HT An. stephensi relative to NTG females. Changes in levels of spectrin, a protein that links the actin cytoskeleton to the plasma membrane, precede autophagic cell death in salivary glands of D. melanogaster [47].
The upregulation of cytoskeletal proteins associated with the progression of autophagy and mitophagy suggested some involvement of these processes in the phenotype of myrAkt An. stephensi. To address this hypothesis, we examined activation levels (phosphorylation) of autophagy-promoting ERK, JNK, and p38 MAPK in the midguts of 3–5 d old, age-matched NTG and HM An. stephensi. Activation levels of all three MAPKs were reduced in the midgut of HM females relative to NTG females (by 40–60%; Fig. 1), which in the context of Akt overexpression, suggested a state of disruption of normal, autophagic repair processes in the midgut epithelium. Together with our LC-MS/MS data, these data suggested that proteins associated with mitochondria and maintenance of the structure of the midgut were altered by tissue-specific Akt overexpression in An. stephensi.
To investigate the impact of myrAkt overexpression on general midgut morphology and on midgut mitochondrial size, number, and shape, we completed morphometric analyses on TEM micrographs from posterior midguts of age-matched, 3 d and 18 d NTG, HT, and HM female An. stephensi. While the majority of blood digestion occurs in the posterior part of the midgut [48], this analysis provided the additional benefit of exploring a midgut region that that is enriched for endocrine cell-associated intestinal stem cells that are required for epithelial repair and maintenance [49]–[51]. The analyzed micrographs from the posterior midgut covered an average area of 462.9 µm2 or the equivalent of 5–8 cells per posterior midgut region. Each image section included an average of 72.9 µm of the midgut epithelium brush border and the average depth of the midgut sample from the edge of the brush border to the edge of the image was 85.9 µm.
The overexpression of myrAkt resulted in significant morphological changes to the An. stephensi posterior midgut (Table 2). Posterior midgut tissue of 3 d and 18 d NTG mosquitoes had a well-defined brush border, intact basal lamina, well-defined nuclei, normal lysosomes, and numerous mitochondria that tended to localize near the brush border (Figs. 2A, B) as observed in other insect species [52]–[54]. The midgut morphology of 3 d HT and HM females were similar to those of NTGs with most structures comparable to NTG mosquitoes. However, a significantly higher prevalence of midguts from 3 d transgenic mosquitoes had cytoplasmic inclusions (40% HT and 60% HM) that were less common or absent in midguts from NTG females (20%; Table 2). Based on morphology (outer double membrane around electron dense material, membrane-like structures, membrane-bound structures with cristae-like organization; [55], [56]), these inclusions were likely stalled autophagosomes of various sizes (Figs. 2E). Mitochondrial autophagy in particular was evident as either mitochondrial membranes engulfed by a developing phagolysosome or as autophagosome-associated mitochondrial degradation (Fig. 2E, F insets).
Midgut epithelium is a highly aerobic tissue and deficits in energy from mitochondria, perhaps associated with altered mitophagy, would likely result in electrolyte imbalance, defective secondary active transport of cations and solutes, and increased permeability [57]. To determine whether histological changes observed in HM midguts were accompanied by altered epithelial integrity, 3–5 d old NTG and HM mosquitoes were fed with fluorescent beads in reconstituted human blood meals. After 48 h – to allow complete digestion of the blood – bead numbers were quantified by flow cytometry in washed, lysed midguts and in whole mosquitoes to estimate transport of beads across the epithelium. Body bead counts (minus midgut beads) from NTG females were 2,805±593 (mean ± SEM), whereas body bead counts from HM females were nearly 2-fold higher (5,073±534; P = 0.011; Fig. 3). Therefore, the permeability of the midgut epithelium was significantly increased in 3–5 d old HM females compared to age-matched NTG females, confirming that observed histological changes in the midgut epithelium of 3–5 d old HM myrAkt An. stephensi were functionally significant.
In 18 d NTG mosquitoes, midgut morphology was comparable to 3 d NTG mosquitoes although an increased number of lysosomes were present (Fig. 2B). However, profound changes were observed in midguts from 18 d HT and HM transgenic mosquitoes (Figs. 2D, F). In both HT and HM females we observed multiple giant stalled autophagosomes containing stacks of membrane-like material (Figs. 2D, F; Fig. S1A). Some giant autophagosomes contained brush border microvilli (Figs. 2D, F; Fig. S1B), which likely formed as a result of invagination of brush border membrane (Fig. S1B). These giant stalled autophagosomes with remnants of brush border were present only in TG mosquitoes at 18 d (50% HT and 100% HM versus 0% NTG; Table 2). In addition, in midguts of 18 d HM females mitochondria were no longer localized to the brush border, but were instead evenly distributed throughout the cell cytoplasm (Fig. 2F; Fig. S1B).
Although autophagy is required for normal mitochondrial turnover, the accumulation of inclusions suggested (i) stalled autophagy in HT and HM relative to NTG mosquitoes, (ii) normal autophagy overwhelmed by the high increase in damaged mitochondria, and (iii) over-reactive autophagy that could perhaps eliminate normal as well as dysfunctional mitochondria. In support of the first hypothesis, altered Atg6 and Atg8 mRNA expression levels were observed in the midgut of 18 d HM An. stephensi (Fig. 4). Atg6, also known as Beclin-1, is required for the generation of pre-autophagosome structures [58], while Atg8-phosphoethanolamine conjugates and the Atg5–Atg12 complex are essential components of the autophagosomal membrane [59]. In 18 d HM females, expression of Atg8 was significantly reduced relative to NTG females (Fig. 4), while expression of Atg6 showed a similar, non-significant trend, suggesting that autophagosome maturation is decreased in 18 d HM females.
Quantitative morphometric analysis of mitochondria in the posterior midgut revealed a significant decrease in the average size of mitochondria in midguts from 18 d HM females compared to similarly aged NTG females (by 47%, P<0.05; Fig. 5A). Midguts from 18 d HT also showed a decrease in average mitochondrial size (33% relative to NTGs) although this was not significantly different from 18 d NTG controls. In contrast to mitochondrial size, the number of mitochondria per µm2 of midgut and the total number of mitochondria per midgut decreased by 37% and 36% respectively in 3 d HM compared to NTG females; however, these decreases were not significant (Figs. 5B, C). The total of number of mitochondria per µm2 and the total number of mitochondria per midgut in 3 d HM were significantly lower than 3 d HT TG (45% decrease; P<0.05).
Total mitochondrial content – represented as a percentage of posterior midgut area occupied by mitochondria and based on mitochondrial number and size – was significantly lower in 3 d HM females compared to 3 d NTG females (by 33%, P<0.05; Fig. 5D), which was consistent with the reduction in mitochondria numbers in 3 d HM females (Fig. 5C). Interestingly, there was no corresponding decrease in the total mitochondrial content of 3 d HT females relative to 3 d NTG females. However, at 18 d after adult emergence, midguts from both HT and HM females had a significant reduction in total mitochondrial content relative to NTG females (by 29% in HT, P<0.05; by 40%, P<0.05 in HM; Fig. 5D). These differences were reflective of reductions in size of mitochondria in midguts of 18 d HT and 18 d HM females (Fig. 5A).
The overrepresentation of mitochondrial proteins (3–5 d; Table 1), the accumulation of autophagosomes (3 d HM and HT; Table 2), apparent altered autophagy (18 d HM; Fig. 4), and changes in size, number, and distribution of mitochondria in the midguts of 3 d HM (Fig. 5D, Table 2) and 18 d HT (Fig. 5D) An. stephensi were indicative of defective organelle maintenance in TG mosquitoes. Accordingly, we quantified the percentages of round versus non-round mitochondria to assess the balance of fission-fusion with the assumption that functional midgut mitochondria exhibit a tubular, elongated shape while round mitochondria form in cells undergoing a response to cellular oxidative damage. We examined the shape of 2524 mitochondria in 3 d NTG, 2979 in 3 d HT, 1638 in 3 d HM, 2080 in 18 d NTG, 2154 in 18 d HT, and 2147 mitochondria in 18 d HM An. stephensi. The number of midgut epithelial cells from which we analyzed mitochondrial shape could not be determined for 18 d HT and 18 d HM mosquitoes due to extensive tissue damage induced by transgene expression in older mosquitoes. Thus, mitochondria were counted over an identical midgut area for all treatments. As expected, a high percentage of elongated, tubular mitochondria were observed in NTG midguts (Fig. 6). Contrary to our expectations, however, no significant differences in the percentages of round mitochondria among midguts of NTG, HM, and HT females at 3 d or 18 d or between matched genotypes at 3 d or 18 d were observed (P = 0.088). However, analysis of the distributions of small (<50,000 nm2), medium (50,000–100,000 nm2), and large (>100,000 nm2) round mitochondria showed significant differences between and among NTG, HT, and HM females. The distributions of small, medium, and large round mitochondria were comparable in midguts from 3 d and 18 d NTGs (Fig. 6). However, the distributions of round mitochondria in NTGs were significantly different from those in HT and HM midguts at both 3 d and 18 d (P<0.0001). In addition, within each transgenic genotype (HT or HM), there were significant differences in mitochondrial size distributions between samples analyzed from 3 d and 18 d An. stephensi (P<0.0001). At 18 d, the occurrence of small, round mitochondria showed a gene-dose dependence from NTG to HT to HM females (30% to 48% to 65%; Fig. 6). The increased percentages of small mitochondria were accompanied by losses in both medium and large sized mitochondria, with percentages of both sizes trending downward from NTG to HT to HM females at 18 d. These changes in morphology appeared to be consistent with persistence of oxidative stress-induced fission, resulting in the formation of small, round mitochondria that can persist when fusion is inhibited [60]–[63].
Despite obvious abnormal mitochondrial morphology, our LC-MS/MS data indicated that mitochondrial proteins were overrepresented in midguts of myrAkt transgenic An. stephensi. Successful mitochondrial biogenesis requires a correct sequence of events consistent with increased and coordinated synthesis of mitochondrial precursors from the nuclear and mitochondrial genomes, followed by import and assembly of nuclear-encoded subunits. Importantly, if any processes downstream from overexpression of mitochondrial proteins (including post-translational processes such as increased RNOS-mediated stress) are altered, then mitochondrial OXPHOS would be compromised despite an available excess of individual subunits. To evaluate this possibility, individual Complex activities were evaluated in whole midguts from 3 d and 18 d NTG and HM An. stephensi females along with citrate synthase, a marker for mitochondrial mass [8] (Table 3). Citrate synthase activity in midguts from HM females was approximately 10% higher than NTG females at 3 d. This value contrasted with the 33% reduction in posterior midgut mitochondrial content of 3 d HM females by morphometric analyses (Fig. 5D). By 18 d, citrate synthase was decreased by 40% in the whole midgut, consistent with the 40% reduction in total mitochondrial content revealed by morphometric analyses of posterior midgut (Fig. 5D). The discrepancy at 3 d could be due to spatial changes in mitochondrial mass over time in the midgut. In particular, mitochondrial loss may be evident at 3 d only in the posterior region of the HM midgut, which was examined morphometrically, but not detectable biochemically at the level of the whole midgut at this time. By 18 d, changes in total mitochondrial content/mass may have spread from the posterior region throughout the midgut, as revealed by concordance between poster midgut morphometrics and whole midgut biochemical analyses.
Examination of OXPHOS capacity revealed that in HM mosquitoes, activities of Complex I, Complex II–III and Complex V were 75% (for Complex I average of both NQR and NFR activities), 30% and 70% of NTG controls at 3 d, respectively, whereas at 18 d, these values were 50%, 40% and 50% of NTG controls, respectively (Table 3). These activities were still lower than controls when normalized to citrate synthase, making them independent of the number of mitochondria present at any given time point in whole midguts (Table 3). Relatively lower ETC activities can result in lower OXPHOS and energy deficits. To test this hypothesis, the energy charge potential (ECP) – defined as ([ATP]+0.5*[ADP])/([ATP]+[ADP]+[AMP]; [64]) – was evaluated in NTG and HM female midguts at 3 d and 18 d post-emergence and in whole bodies of 3 d NTG and HM females. In midguts from HM females at 3 d, the ECP showed some decline (94% of controls), which was clearly lower at 18 d (89% of controls; Table 3, Fig. 7). In addition to local tissue effects, overexpression of myrAkt in the An. stephensi midgut was associated with significantly reduced whole body total adenosine metabolites. In particular, [ADP] (in percentage of total nucleotides) at 3 d was 2-fold higher, [AMP] was 7.7-fold higher, and ECP was significantly reduced relative to age-matched NTG controls (Table 4) despite the fact that Akt overexpression was targeted to the midgut of An. stephensi. These results (decreased ECP, higher [ADP]/[ATP] and [AMP]/[ATP] ratios) indicated clear energy deficiencies locally in the midgut and systemically in the body of HM mosquitoes, suggesting that (i) mitochondrial biogenesis could not be correctly completed (defects at import/assembly) or (ii) oxidative/nitrative stress-mediated damage overpowered this putative compensatory response.
To test these possibilities, oxidative/nitrative damage was assessed by evaluating Tyr nitration of the beta subunit of ATPase (ATPB; Fig. S2), a sensitive marker for mitochondrial protein nitration [30], [31], given that Complex V activity was significantly lower than controls (50 to 70% of NTG). NOS activity was evaluated by detecting NO using electron paramagnetic resonance in conjunction with spin trapping technique. Nitration of midgut proteins in HM females was significantly increased at 3 d (8-fold of controls; P = 0.02) and at 18 d (2-fold of controls; P = 0.05) while nitration of mitochondrial ATPB was 3-fold (P = 0.003) and 2-fold (P = 0.048) of controls at each time point (Table 3, Fig. 8). The production of NO was 2-fold of controls at 3 d (Table 3, Fig. 9). Thus, nitrative/oxidative stress was enhanced in midguts from HM An. stephensi – even at 3 d post-emergence – consistent with the increased ATPB nitration and activity loss [65]. Further, NO is an inhibitor of Complex IV [66], [67] through competitive and noncompetitive pathways [66], suggesting increased NO production could block electron transport at the terminal oxidase, even when no changes in activity are detected, enhancing the RNOS-mediated damage of individual Complexes and/or mitochondrial targets and negating compensatory biogenesis.
While these data provided clear indications of mitochondrial dysfunction in myrAkt An. stephensi even in the absence of infection, they did not provide a mechanistic explanation for inhibition of P. falciparum development in myrAkt An. stephensi observed previously [5]. Collectively, our data suggested that inhibition of parasite development could result from (i) direct, toxic effects of RNOS on developing P. falciparum [68]–[70], (ii) indirect effects of RNOS-mediated mitochondrial dysfunction in the host (e.g., reduction in host energy required for parasite development, damage to midgut epithelial receptors/proteins required for parasite development), or (iii) some combination of these direct and indirect effects of Akt overexpression on developing parasites.
To first assess whether overproduction of NO was responsible for resistance to P. falciparum in myrAkt An. stephensi, four separate cohorts of HM An. stephensi females were provided with water, Nω-Nitro-L-arginine methyl ester (3.7 mM, L-NAME; [69], [71]) or the biologically inactive isomer D-NAME from 72 h before blood feeding though P. falciparum infection and thereafter until dissection. Age- and cohort-matched NTG females were infected side-by-side as controls. After 10 days, females were dissected to visualize and count P. falciparum oocysts. Among those NTG An. stephensi that were fully gravid (an indicator of complete engorgement), 49% had at least one midgut oocyst (Fig. 10A). Infected NTG mosquitoes averaged 2.4 P. falciparum oocysts per midgut (Fig. 10B). As expected, HM females provided only with water or with water with D-NAME were resistant to infection (Fig. 10B). However, provision of L-NAME to HM An. stephensi reversed the phenotype of resistance to infection, resulting in a prevalence and intensity of infection that were not significantly different from control, NTG females fed on the same P. falciparum-infected blood (Figs. 10A, B). Neither L-NAME nor D-NAME had significant effects on P. falciparum growth in the absence of the mosquito (Fig. S3). Although this growth assay cannot be performed efficiently on mosquito-stage parasites, we assert that our results suggest that observed infection patterns were due to L-NAME effects on the mosquito host.
To determine whether the effects of NO were attributable specifically to direct, toxic effects on developing parasites, we examined more closely the timing of parasite death in HM females relative to parasite development in NTG An. stephensi. For these studies, we used a mouse malaria parasite infection model (GFP-expressing Plasmodium yoelii yoelii 17XNL; kindly provided by A. Rodriguez [72]) in addition to mosquito infection with P. falciparum. This design allowed us to examine mosquito infection using independent, quantitative measures and to determine whether NO-dependent parasite killing in myrAkt An. stephensi was unique to P. falciparum or more broadly effective against unrelated parasite species. Infection with P. y. yoelii was monitored using fluorescence quantitation, while P. falciparum infection levels were assessed with quantitative, reverse-transcriptase PCR for two markers of sexual stage parasite development, Pfs16 and Pfs25 [73]–[75]. In infections with both species of Plasmodium, significant parasite death was first observed by 18–20 h after infection (Figs. 11A, B). At 20 h post-infection, P. y. yoelii 17XNL parasites are present as mature ookinetes, with some in the midgut lumen and a large percentage in transit across the midgut epithelium [76]. At 18 h post-infection, all P. falciparum parasites are present only as ookinetes in the midgut lumen of An. stephensi [77]. A secondary significant drop in infection levels of HM An. stephensi relative to NTG females was observed by 48 h after infection for both parasite species – a time that coincides with ookinete to oocyst transition on the outside of the midgut epithelium for P. falciparum [77] and early oocyst development for P. y. yoelii [76] – suggesting that RNOS-mediated anti-parasite killing occurs over a broad period of parasite development. These data confirmed that anti-P. falciparum resistance in myrAkt An. stephensi is initiated as direct, early toxic effects of mosquito NO/RNOS on parasites prior to invasion of the midgut epithelium.
Overexpression of a constitutively active Akt targeted to the midgut of An. stephensi inhibited P. falciparum infection and reduced the duration of mosquito infectivity [5]. Here, we have elucidated the mechanism for this process, demonstrating that Akt-dependent anti-parasite resistance is due to early, toxic effects of NO/RNOS followed by sustained mitochondrial dysfunction that cannot be mitigated via biogenesis and/or mitophagy/autophagy. Collectively, these phenomena lead to midgut epithelial damage (Figs. 2, S1; Table 1) and systemic energy deficiencies (as judged by ECP) that would be consistent with a reduction in lifespan and, as a consequence, a reduced infective lifespan. Activation of nuclear factor (NF)-kB-dependent immunity does not contribute to parasite resistance in myrAkt An. stephensi: Pakpour et al. [78] showed that activation of PI-3K-dependent signaling represses NF-KB activation in response to immune signals in An. stephensi cells in vitro and in vivo. Rather, overwhelming RNOS production with overexpression of myrAkt not only confers resistance to parasite infection, but also adversely impacts host infective lifespan. We would assert that balance and successful resolution of oxidative stress-induced mitophagy and mitochondrial biogenesis are the driving forces behind these phenotypes (Fig. 12) and that genetic manipulation of mitochondrial processes can provide a basis to alter multiple mosquito phenotypes to inhibit malaria parasite transmission.
In this study, we identified the molecular processes downstream of Akt overexpression in myrAkt An. stephensi that are involved in parasite infection resistance and reduction in lifespan. First, myrAkt protein levels in the midgut of transgenic An. stephensi increase significantly by 2 h after bloodfeeding and remain elevated above control levels through 12 h; these levels decline for the latter half of blood digestion and the reproductive cycle (24–48 h post-feeding; [5]). Hence, inducible overexpression of active Akt soon after blood ingestion would be available to repress FOXO-dependent antioxidants, including mitochondrial MnSOD, allowing a rise in damaging mitochondrial RNOS [27], [28]. We have previously demonstrated that toxic RNOS, which likely include peroxynitrite, can rise to high levels in the parasite-infected mosquito midgut after infection and that NOS-dependent killing is central to anti-parasite resistance in wild type mosquitoes [79]. However, the critical differences in the current studies are that Akt overexpression (i) leads to the overproduction of NO even in the absence of infection and (ii) represses cellular mitophagy, which would normally occur in response to infection-associated oxidative/nitrative stress, resulting in much more efficient parasite killing at a cost of both local and systemic energy synthesis and renewal/repair of the midgut epithelium that are essential for host survival following infection.
In addition to Akt-dependent signaling inhibition of mitophagy/autophagy, activated Akt controls mitochondrial biogenesis through phosphorylation of FOXO and exclusion of this transcriptional activator from the nucleus. In the absence of FOXO, there is no induction of PPAR-gamma coactivator-1 alpha (PGC-1a), a key mediator of mitochondrial biogenesis in mammalian cells [80]. Rera et al. [81] demonstrated that overexpression of FOXO-dependent Drosophila ortholog of PGC-1a led to an increase in abundance of respiratory complexes I, III, IV, and V and an increase in respiratory chain activity, indicating that control of biogenesis is conserved. Hence, in transgenic An. stephensi, activated Akt likely represses mitochondrial biogenesis through FOXO phosphorylation, which results in lower total mitochondrial content at 3 d and 18 d by morphometric analyses of the posterior midgut (Fig. 5C) and at 18 d as indicated by citrate synthase activity in the whole midgut (Table 3).
Despite the decrease in mitochondrial mass, we observed an upregulation of mitochondrial protein levels. A second strong signal in our system – NO synthesis (Table 3; Figs. 8, 9) – provides a possible explanation for our observations. In brief, Moncada and others demonstrated in a variety of mammalian cell types that NO is a potent inducer of mitochondrial biogenesis via NO-dependent activation of guanylate cyclase, which induces PGC-1a expression [82]–[84]. Hence, strong activation of NO synthesis, which also occurs in response to Akt overexpression, could explain the early upregulation of mitochondrial proteins in myrAkt An. stephensi. Alternatively, direct inhibition of OXPHOS by NO results in increased AMP [67], which can activate AMP-activated protein kinase (AMPK) to increase PGC-1a expression [80]. In agreement with these results, we observed increased levels of mitochondrial proteins without increases in OXPHOS, an effect likely due to a combination of direct NO inhibition of OXPHOS and indirect RNOS-mediated damage to individual Complexes (Table 3; Figs. 8, 9).
Conflicts in signaling – RNOS induction of mitophagy, Akt-dependent inhibition of mitophagy, Akt-dependent inhibition of mitochondrial biogenesis, NO/AMP-dependent induction of mitochondrial protein synthesis – are apparent in the midgut epithelium of myrAkt An. stephensi. We suggest that conflict of these signals over time results in a lack of resolution of both mitophagy and mitochondrial biogenesis, which ultimately results in energy deficits as judged by ECP. Deficits in energy would result in loss of midgut tissue architecture and electrolyte balance, both of which are critical for proper absorption of nutrients and barrier function. In support of these inferences, incomplete resolution of mitophagy is evident as shifting distributions toward increased small, round mitochondria and decreased large, round mitochondria (Fig. 6) from 3 d to 18 d in midguts of HM females and also in the progression of gene dosage from NTG to HT to HM females at 18 d. We suggest that these changes have resulted from fission of damaged mitochondria into small fragments that are not eliminated from the cell. Excessive fission has also been linked to S-nitrosylation of Drp-1 (SNO-Drp-1) [85], which could occur in the context of high level NO synthesis in myrAkt An. stephensi. Because proteolysis and, ultimately, mitophagy are key to resolution of oxidative damage in the cell, excessive fission and accumulated damage ultimately trigger bulk autophagy, evident as giant stalled autophagosomes in midguts from 18 d HM females (Figs. 2, S1). An imbalance of fission and fusion, whereby undamaged mitochondrial fragments are reassembled in recovered cells, has been reported in a number of human disease states. In particular, Parkinson's disease-specific proteins associated with fission/fusion include PTEN-inducible kinase 1/parkin, alpha-synuclein, and HTRA2/OMI, while mutant huntingtin appears to be associated with alterations in mitochondrial fission and fusion in Huntington's disease [86]. Based on these observations and our data, an appropriate balance of mitochondrial dynamics in the mosquito midgut epithelium – in a manner analogous to that in the gut of C. elegans and the midgut of D. melanogaster [87] – is likely to be key to overall mosquito vigor and vector capacity in malaria transmission.
Given the complexity of mitochondrial dynamics – and the conflicting signals that balance response to and recovery from damage – how can this knowledge be harnessed for novel strategies for malaria control? That is, is it possible to genetically engineer optimal mitochondrial dynamics to promote parasite killing while maintaining competitive fitness of mosquitoes under natural conditions? In light of successful manipulation of autophagy genes and mitochondrial proteins in invertebrates and mammals to alter immunity, lifespan/cell senescence, and stem cell differentiation, we suggest that this is entirely possible in vector mosquitoes. In particular, overexpression of autophagy-related genes enhances anti-pathogen immunity, including clearance of Mycobacterium tuberculosis-containing phagosomes in mouse macrophages and human myeloid cells in vitro [88], [89], protection against fatal Sindbis virus infection in vivo in mice ([90], and proper localization of anti-pathogen hypersensitive responses in Arabidopsis thaliana [91]. Hence, independent manipulation of mosquito autophagy genes is likely to impact pathogen resistance. In addition, moderate repression of neuronal ETC genes in D. melanogaster [92] and in neuronal and intestinal cells in C. elegans [93] can promote longevity, which in C. elegans is dependent on the upregulation of the mitochondrial unfolded protein response (mtUPR) in the nematode intestine [93]. The mtUPR is activated in response to mitochondrial stress that is communicated to the nucleus to increase the expression of the mitochondrial protein chaperones HSP-6 and HSP-60 [94]. In addition to the relationship with longevity, mtUPR-associated HSP-60 has profound effects on immunity. In particular, HSP-60 is released from damaged or stressed cells and can act a as a potent inducer of innate immune responses, including release of pro-inflammatory cytokines and NO, and is believed to be a major extracellular mediator in linking infectious agents with immune cells in response to stress [95], [96]. Further, HSP-60 peptides and protein, which bind a variety of receptors including CD36 and other class B scavenger receptors [97], have been delivered in vivo to enhance the immunogenicity of anti-microbial vaccines, resulting in the description of HSP-60 as a “natural adjuvant” for innate and adaptive responses to infection [98]. Intriguingly, the mosquito ortholog of HSP-60 (AGAP004002; Table 1), was upregulated in myrAkt An. stephensi, suggesting that anti-parasite resistance may be regulated in part by actions of mosquito HSP-60 via scavenger receptor binding [99], which could be enhanced directly or via ETC manipulation without epithelial damage as induced by Akt overexpression.
In this study, we used a well-defined model based on midgut-specific expression of a constitutively active myrAkt [5] to elucidate the mechanism underlying mosquito resistance to infection. Activated Akt leads to increased steady-state reactive nitrogen and oxygen species, which leads to mitochondrial dysfunction. In a compensating response, the expression of mitochondrial proteins in the midgut is upregulated, likely by PGC-1a, to promote mitochondrial biogenesis. However, this response is countered by (i) Akt-mediated repression of autophagy and (ii) the loss of ATP from mitochondria which would ensue in dysfunctional regulation of electrolyte balance and nutrient transport [57]. These two effects, which resemble “accelerated aging” [100], result in accumulation of damaged mitochondria and general loss of tissue structure and barrier function. Additionally, Akt activation inhibits apoptosis, suggesting that sustained activation of Akt would undermine elimination of damaged cells through a controlled death process [101]. Thus, a critical point for malaria resistance is quality control of mitochondrial function in the mosquito midgut, which would support – as confirmed in mammalian and invertebrate models – the maintenance of epithelial integrity through energy homeostasis, stem cell viability, and tissue repair and renewal [57], [81], [102]. Accordingly, we assert that genetic manipulation of mitochondrial processes in the midgut as a “signaling center” can be used as the basis for an efficient and novel strategy to block malaria parasite transmission.
All protocols involving animals for mosquito rearing and feeding were approved and in accordance with regulatory guidelines and standards set by the Institutional Animal Care and Use Committee of the University of California, Davis (protocol #15990 to SL, approved on June 28, 2012) under institutional approvals by the Association for Assessment and Accreditation of Laboratory Care International (AAALAC International) accreditation program (approval #00029), the Public Health Service Office of Laboratory Animal Welfare (PHS OLAW assurance #A3433-01), and the United States Department of Agriculture Animal and Plant Health Inspection Service (USDA APHIS registration #93-R-0433). Human blood for mosquito feeding for TEM protocols was acquired as anonymously donated, expired human blood from the American Red Cross under Institutional Review Board (IRB) distribution protocol #2010-014 to MR. Anonymously donated human blood products for infection studies and midgut permeability assays were purchased from Interstate Blood Bank (Memphis, TN); these materials are deemed exempt from Human Subjects Use by the University of California Davis IRB.
Sodium succinate, sucrose, HEPES, calcium chloride, NADH, pyruvate kinase, lactic dehydrogenase, ATP, oligomycin, rotenone, acetyl-CoA, bovine serum albumin (fatty-acid free), antimycin A, 2,3-dimethoxy-5-methyl-1,4-benzoquinone, bovine heart cytochrome c, magnesium chloride, L-arginine, 5,5′-dithiobis(2-nitrobenzoic acid), and propidium iodide were obtained from Sigma-Aldrich (St. Louis, MO). Potassium cyanide, NADPH, and oxaloacetic acid were purchased from Calbiochem/EMD (Rockland, MA). Phosphoenol pyruvate was purchased from MP Biomedicals (Solon, OH). Ferrous sulfate was purchased from Mallinckrodt (St. Louis, MO). Sodium N-methyl-D-glucamine dithiocarbamate (MGD) was purchased from the OMRF Spin Trap Source (Oklahoma City, OK). Monoclonal anti-phospho-ERK1/2 (pT183/pY185) was obtained from Sigma-Aldrich. Anti-phospho-p38 MAPK (T180/Y182) was purchased from Cayman Chemical (Ann Arbor, MI) and anti-phospho-JNK 1/2 (pT183/pY185) was purchased from Invitrogen/Life Technologies (Grand Island, NY). Anti-GAPDH antibody was purchased from Abcam (Cambridge, MA). Horseradish peroxidase-conjugated polyclonal rabbit anti-mouse IgG was purchased from Sigma-Aldrich. Horseradish peroxidase-conjugated goat anti-rabbit F(ab′)2 fragment was purchased from Invitrogen/Life Technologies (Grand Island, NY). The SuperSignal West Pico chemiluminescent detection kit was purchased from Pierce Biotechnology (Rockford, IL). All biochemical reagents were of analytical grade.
Non-transgenic (NTG) as well as homozygous (HM) and heterozygous (HT) myrAkt An. stephensi Liston (Indian wild-type strain; myrAkt line characterized in [5]) were reared and maintained at 27°C and 75% humidity with a 16 h light and 8 h dark photoperiod. All mosquito rearing and feeding protocols were approved and in accordance with regulatory guidelines and standards set by the Institutional Animal Care and Use Committee of the University of California, Davis. For mosquito feedings, 3–5 d female mosquitoes were maintained on water or experimental treatment via soaked sterile cotton balls, changed twice daily, for 24–72 h prior to any experiment. For TEM studies mosquitoes were provided a single meal of whole human blood containing 1% sodium citrate (acquired as anonymously donated, expired blood from the American Red Cross under Institutional Review Board distribution protocol #2010-014 to MR) at 3 d after adult emergence and oviposition substrates were offered 48 h later. To obtain HT TG and NTG mosquitoes for experiments, HT male and wild-type colony female mosquitoes were mated together. The resulting progeny, consisting of approximately 50% TG and 50% NTG individuals, were reared together to eliminate differences in crowding and resources. TG and NTG siblings were separated at the pupal stage by eye fluorescence using a fluorescent dissecting microscope with DsRed filters. HM myrAkt mosquitoes were maintained as a separate line.
For these assays, 3–5 d old NTG and HM myrAkt An. stephensi females were maintained on water for 24 h and then allowed to feed for 30 min on reconstituted blood provided through a Hemotek Insect Feeding System (IFS; Discovery Workshops, Accrington, UK). This blood meal contained washed human RBCs and saline (10 mM NaHCO3, 15 mM NaCl, pH 7.0). Midguts were dissected after blood feeding from 30 mosquitoes in each treatment group. Detection of phosphorylated MAPKs followed the protocols of Surachetpong et al. [103]. In brief, midgut protein lysates were separated by gel electrophoresis on 10% sodium dodecyl sulfate-polyacrylamide gels (SDS-PAGE), transferred to nitrocellulose membranes (BioRad, Hercules, CA), and probed for proteins of interest with target-specific antibodies. Membranes were blocked in 5% dry milk/Tris-buffered saline with 0.1% Tween-20 for 1 h at room temperature, then incubated overnight in each antibody solution. Primary and secondary antibodies, respectively, were used at the following dilutions: 1∶10,000 phospho-ERK/1∶20,000 rabbit anti-mouse IgG; 1∶1250 phospho-p38/1∶20,000 goat anti-rabbit IgG; 1∶1250 phospho-JNK/1∶20,000 goat anti-rabbit IgG; 1∶10,000 GAPDH/1∶20,000 goat anti-rabbit IgG.
For these assays, 20–30 midguts were dissected each from non-bloodfed 18 d NTG and HM An. stephensi, homogenized by pulse sonication in TriZOL reagent (Invitrogen) and RNA was extracted according to the manufacturer's protocol. cDNA was synthesized from RNA samples using the SuperScript III First-Strand Synthesis System (Invitrogen) according to the manufacturer's protocol. Prior to quantitative analysis, cDNA samples were pre-amplified by PCR using gene specific primers. Cycling conditions for pre-amplification were as follows: 1× (95°C for 5 min), 20× (95°C for 30 sec, 55°C for 30 sec, 70°C for 30 sec), 1× (70°C for 5 min), 1× (4°C for 5 min). Pre-amplified cDNA was used for quantitative analysis with Maxima SYBR green/ROX qPCR Master Mix (Fermentas ThermoScientific, Waltham, MA) on an ABI 7300 real-time PCR machine. Cycling conditions for real-time PCR were as follows: 1× (95°C for 5 min), 1× (50°C for 2 min), 35× (95°C for 15 sec, 60°C for 1 min). Expression levels were calculated using the 2−ΔΔCt method relative to the ribosomal protein s7 gene. Primers were designed based on published sequences from An. gambiae using Primer3 software: ATG6F 5′ GCGCGAGTATACGAAGCAT 3′, ATG6R 5′ GCTTCTCTAGCTGGCTCTGG 3′, ATG8F 5′ GCCATCATTCTTTGGAGAGC 3′, ATG8R 5′ TGCTATTAAAATGCGTAGAATGG 3′, RPS7F 5′ GAAGGCCTTCCAGAAGGTACAGA 3′, RPS7R 5′ CATCGGTTTGGGCAGAATG 3′.
A total of 50 NTG, 75 HM transgenic and 55 HT transgenic 3–5 d old adult female An. stephensi were each completely homogenized for total protein isolation in 1.3 ml lysis buffer (50 mM Tris HCl pH 7.5, 100 mM NaCl, 5% glycerol, 1 mM DTT, 1× SigmaFAST protease inhibitor cocktail, 1% n-octyl glucoside). Initial homogenization was achieved by 5× grinding with a PCT Shredder (Pressure Biosciences, South Easton, MA). Partially homogenized samples were placed in a Barocycler NEP2320 (Pressure Biosciences) and subjected to pressures rotating between 31,000 PSI and atmospheric for 35 cycles (20 min total). Following quantification by Bradford assay, 30 µg of total proteins from each sample were electrophoretically separated by SDS-PAGE. Three lanes (biological replicates) each were analyzed for NTG, HM and HT samples as described the Supporting Materials and Methods.
Mitochondrial enzymatic activities were performed on 150 hand-homogenized whole midguts from NTG and HM transgenic An. stephensi in a cold hypotonic solution (300 µl of 20 mM HEPES, pH 7.4, with protease inhibitors and phosphatase inhibitors). Spectrophotometry with a microplate reader (Tecan infinite M200; Tecan Systems, Inc., San Jose, CA) was used to evaluate the samples and data were recorded and analyzed with the Magellan software V6.6 (Tecan Systems, Inc.). All samples were run in triplicate on a 96-well microplate, all reagents were scaled down from 1 ml to 0.2 ml, using water or buffers as blanks, along with the modifications indicated below. Rates were expressed as nmol×(min×mosquito midgut)−1. Values with CV >10% were excluded from calculations and repeated with available material. Complex activities were analyzed as described in the Supporting Materials and Methods.
Extraction of ATP, ADP, AMP, NAD, and NADH were carried with two 1 ml whole mosquito midgut suspensions. One vial was spiked with 7.5 nmol of each standard during the extraction to determine their recovery. The two vials (spiked and unspiked) were extracted in parallel. The suspensions were resuspended and centrifuged at 4°C and 190×g for 3 min. Supernatant was discarded and the pellet was resuspended in 1 ml of ice-cold PBS buffer, pH 7.4, followed by a 3 min spin at 4°C and 190×g. Supernatant was discarded and the cell pellet was treated with 75 µl of ice-cold 0.5 M HClO4 (Sigma-Aldrich). Both vials were incubated on ice for 2 min. The suspension was centrifuged for 3 min at 4°C and 2000×g. The supernatant was kept cold and the pellet was extracted a second time by resuspending the cell pellet in another 75 µl of ice-cold 0.5 M HClO4, keeping on ice for 2 min, and centrifuging for 3 min at 4°C and 2000×g. The supernatants were collected and neutralized to pH 6.5 by adding ice-cold 2.5 M KOH (JT Baker/Avantor, Center Valley, PA) in 1.5 M K2HPO4 (Fisher Scientific, Fairlawn, NJ), and stored on ice for 15 min. The KClO4 precipitate was removed by centrifuging at 2000×g for 1 min at 4°C. The clear supernatant was filtered through a 0.45 µm nylon microspin filter (Grace, Deerfield, IL) and centrifuged for 10 min at 10,000×g at 4°C. Filtered supernatants were then spiked with 4 nmol of hypoxanthine (Sigma-Aldrich) to serve as a loading control to normalize the areas of the standards. Samples were analyzed immediately. Preparation of standards, peak identification, quantification, and HPLC conditions for these assays is described in the Supporting Materials and Methods.
These protocols were published previously [30], [31]. In brief, proteins were denatured in SDS-PAGE sample buffer (BioRad) plus 0.5% β -mercaptoethanol at 100°C for 3 min. Thirty µg of whole midgut protein were loaded onto 12% SDS-PAGE gels (BioRad) and electrophoretically separated at 200 V for approximately 50 min. Proteins were then transferred via semi-dry transfer (20% methanol, 0.0375% SDS) to a 0.45 µm PVDF membrane for 30 min at 15 V, 300 mA. Membranes were washed once for 5 min in Tris-buffered saline plus tween-20 (TBST; 150 mM NaCl, 25 mM Tris, pH 7.4, 0.1% Tween-20), blocked in 5% nonfat dry milk TBST for 1 h and then incubated with anti-nitrotyrosine (1∶1,000 dilution; EMD Millipore, Billerica, MA) or anti-beta subunit ATPase (1∶5,000 dilution; BD Biosciences) antibodies overnight at 4°C. Membranes were washed 3× for 5 min with TBST and then incubated with goat anti-mouse HRP antibody (1∶10,000; Zymed/Invitrogen Grand Island, NY) for 1 h at room temperature. After washing for 3× for 10 min with TBST proteins were then visualized with chemiluminescent reagents (ECL) on a Kodak 2000 MM Imager. The loaded protein amounts were plotted against the densitometry readings to ensure that the ECL response was within a linear range of the protein range. Images were analyzed with the Kodak Imager 2000 MM software provided by the manufacturer. All values were normalized to actin or VDAC1 as loading controls. Data were obtained from triplicates performed on different days. Statistical analyses were performed by using the Student's t-test, alpha = 0.05.
Whole midguts from 150 NTG and 150 HM An. stephensi were each hand-homogenized in 300 µl of 20 mM HEPES, pH 7.4 with protease inhibitors and phosphatase inhibitors), then incubated for 3 h at 20–22°C following addition of 100 µl reaction buffer (3 mM sodium N-methyl-D-glucamine dithiocarbamate [MGD] complexed with ferrous sulfate prepared fresh, 0.1 mM NADPH, 1 mM calcium chloride and 1 mM L-arginine in degassed 20 mM HEPES, pH 7.4; [104]). After the incubation, 50 to 100 µl of sample was loaded into an EPR tube and measured using a Bruker EPR and XEpr software. The conditions were as follows: average of 2 scans, sampling time of 0.163 s, sample temperature was 105 K, field modulation amplitude at 0.0008 T, field modulation of 100,000 Hz, microwave frequency of 9.4E9, microwave power of 0.0126 W, receiver gain 60, receiver time constant was 1.31 s, receiver phase of 0 deg, receiver harmonic 1, and receiver offset of 0% FS.
Midguts were dissected from female mosquitoes 3 d and 18 d post adult emergence from all three treatment groups (non-transgenic, HT myrAkt, and HM myrAkt) into 1× TBS buffer (0.025 M Tris, 0.15 M sodium chloride, pH 8.0). Midguts were immediately fixed in 2.5% glutaraldehyde in 0.1 M PIPES buffer pH 7.4 for 1 h at room temperature and then transferred to 0.1 M PIPES buffer, pH 7.4 on ice for same day submission to the AHSCI Imaging Core Facility at University of Arizona. Five midguts per genotype for 3 d mosquitoes and four midguts per genotype for 18 d mosquitoes were processed as follows. Samples were incubated in 1% osmium tetroxide in 0.1 M PIPES for 1 h and then washed 3 times for 5 min in deionized water. Washed samples were then transferred to 2% aqueous uranyl acetate for 20 min, rinsed in deionized water for 5 min and incubated in increasing concentrations of ethanol: 50% ethanol for 5 min, 70% ethanol for 5 min, 90% ethanol for 5 min, 100% ethanol for 5 min; 100% ethanol for 20 min, and 100% ethanol for 5 min. This was followed by three incubations (5 min each) in propylene oxide, followed by an overnight incubation in EmBed 812/propylene oxide (1∶1). The next day samples were incubated 3 times for 60 min in EmBed 812 resin, and, finally, flat embedded for 24 h at 60°C. Semi-thin sections (0.5 µm) were stained with 1% toluidine blue and examined with light microscopy. Thin silver sections were cut onto uncoated 150 mesh copper grids, stained with 2% aqueous lead citrate for 2 min and examined with a CM12S electron microscope operated at 80 kv. TIFF images were collected with an AMT 4Mpix CCD camera.
Twenty adjacent images from individual posterior midguts were acquired at 15,000× magnification. For each individual midgut, 12 to 17 adjacent images (equivalent to ∼5–8 cells) were analyzed for morphometric analysis, with care taken to ensure that areas of the analyzed midguts were the same size for each mosquito. For each micrograph all mitochondria were outlined by hand using ImageJ software [105]. The area of each individual mitochondrion, the total mitochondrial area and the total number of mitochondria were determined using ImageJ. The numbers of round (presumably damaged) and elongated (healthy) mitochondria were counted for each image, and round mitochondria were further classified by size (<50 K, 50–100 K, and >100 K nm2). Vacuoles with electron dense content, small stalled autophagosomes with membrane material, large stalled autophagosomes with membrane material and giant stalled autophagosomes with brush border were counted for all midguts. Two 95% confidence intervals (CI) were constructed using NTG values at 3 and 18 d ([4.4, 17.9] and [7.7, 33.8]) to identify posterior midguts that contained a number of autophagosomes above the highest 95%CI limit. Chi-square test (alpha = 0.5) was used to compare NTG versus HT, NTG versus HM, and HT versus HM. In addition, the area of each midgut epithelial cell on each micrograph, the length of the brush border, and depth from the brush border were measured. Total mitochondrial content was determined by dividing the total mitochondrial area in one midgut by the total area examined in that midgut. The data for mitochondria size, total mitochondrial area and total number of mitochondria were analyzed using two-way ANOVA followed by Tukey-Kramer HSD test. Proportions of round mitochondria were analyzed with ANOVA following arcsin transformation. Size distributions of round mitochondria among treatments were analyzed by contingency analysis followed by specific pairwise comparisons.
Cultured parasites were grown in 10% heat-inactivated human serum and 6% washed human red blood cells (RBCs; Interstate Blood Bank, Memphis, TN) in RPMI 1640 with HEPES (Gibco) and hypoxanthine for 15 days, or until stage V gametocytes were evident. Exflagellation rates of mature gametocytes were evaluated on the day prior to and the day of mosquito infection. Mosquitoes were fed on mature gametocyte cultures diluted with human RBCs and heat-inactivated human serum (Interstate Blood Bank) for 30 min. Experimental treatments were added to the diluted culture just before blood feeding. After feeding, blood fed mosquitoes were maintained on the same treatments, provided as water-soaked sterile cotton balls and changed twice daily, until dissection. Protocols involving the culture and handling of P. falciparum for mosquito feeding were approved and in accordance with regulatory guidelines and standards set by the Biological Safety Administrative Advisory Committee of the University of California, Davis.
For these studies, 125 NTG and 125 HM myrAkt An. stephensi were provided with water, Nω-Nitro-L-arginine methyl ester (1 mg/ml, L-NAME; Sigma-Aldrich) or the biologically inactive D-NAME (1 mg/ml; Sigma-Aldrich) from 72 h before blood feeding through P. falciparum infection and thereafter until dissection. This experiment was repeated four times with four separate cohorts of mosquitoes. After 10 d, fully gravid females were dissected and midguts were stained with 0.1% mercurochrome to visualize P. falciparum oocysts. Oocysts were counted for each midgut and mean oocysts per midgut (infection intensity) and percentages of infected mosquitoes (infection prevalence; infection defined as at least one oocyst) were calculated for all dissected mosquitoes. Infection prevalence data were analyzed by Fisher's exact test to determine whether infection status differed between treatment conditions. Infection intensity data were analyzed by ANOVA to determine whether the oocysts per midgut in the controls differed among replicates. Since no differences were evident, the data were pooled across replicates. The numbers of infected mosquitoes in the water and D-NAME groups were outside the 95% confidence intervals for the NTG (1.79,3.05) and L-NAME (1.41,4.01) groups and, therefore, were excluded from analyses of the latter data. Oocyst counts between L-NAME and NTG groups were compared using the Mann-Whitney U test for non-parametric data.
To determine whether L-NAME or D-NAME experimental treatments had a direct effect on parasite growth that could contribute to the infection phenotypes in An. stephensi, aliquots of P. falciparum NF54 culture were synchronized and subjected to a standard growth assay [103]. After synchronization, parasites were plated in 96-well flat-bottom plates in complete RPMI 1640 with HEPES, hypoxanthine, and 10% heat inactivated human serum. Parasites were treated for 48 h at 37°C with equivalent volumes of PBS and L-NAME or inactive D-NAME at 0.74 or 3.7 mM (latter concentration provided in the infectious blood meal to An. stephensi). Assays were terminated by replacing culture media with RPMI 1640/1% formalin. Infected RBCs were stained with 10 µg/ml of propidium iodide in phosphate-buffered saline (PBS; Cellgro, Manassas, VA) for 1 h at room temperature, then counted with FACS Calibur flow cytometer, Becton Dickinson (BD Biosciences, San Jose, CA). Relative levels of parasite growth in response to treatment were normalized to PBS-treated controls, which were set to 100%.
To determine the timing of parasite killing in myrAkt HM An. stephensi relative to NTG controls, we examined infection of HM and NTG mosquitoes with GFP-expressing Plasmodium yoelii yoelii 17XNL and with P. falciparum NF54. For P. y. yoelii studies, CD1 mice were infected with the 17XNL strain stably transfected with green fluorescent protein [72] and parasitemia and gametocytemia were monitored daily via Giemsa staining of thin blood films. At approximately 9 d post infection (parasitemia of 15–18%), mice were anesthetized and used to feed laboratory-reared 3–5 d, female HM and NTG An. stephensi (n = 125 per group). Mosquitoes were maintained on water for 24 h prior to blood feeding, and then allowed to feed for 30 min (mice were rotated among cartons to ensure uniform infections). At 6, 20 and 48 h post-feeding, midguts were dissected into PBS (Cellgro). As an experimental control, HM and NTG mosquitoes were also fed on uninfected CD1 mice. Three pools of 10 midguts were collected at each time point, homogenized using QIAshredder homogenizer columns (Qiagen, Valencia, CA), serially diluted and then scanned at 485 nm excitation/535 nm emission wavelengths. Additionally, infected and uninfected blood was collected directly from mice via cardiac puncture and fluorescence readings were obtained for serial dilutions of this blood. Parasitemia and hematocrit counts were determined by Giemsa stained thin blood smears and by hemocytometer, and the fluorescent values used as a standard for P. yoelii fluorescence in midgut samples. This experiment was repeated three times with three separate cohorts of mosquitoes and infected mice.
For assessment of timing for P. falciparum killing, laboratory-reared 3–5 d female myrAkt HM and NTG An. stephensi (n = 150 per group) were maintained on water for 24 h prior to blood feeding. TG and NTG mosquitoes were provided blood meals containing P. falciparum NF54-infected RBCs and allowed to feed for 30 min. Midguts from blood-fed females were dissected into TriZOL reagent (Invitrogen) at 6, 18 and 48 h post infection and homogenized using pulse sonication. RNA was extracted from homogenates following the manufacturer's protocol. Contaminating genomic DNA was removed from RNA samples using Turbo DNAse Free Kit (Applied Biosystems/Life Technologies, Foster City, CA). Reverse transcription was carried out with the Superscript III cDNA Synthesis Kit (Invitrogen). Real time quantitative PCR reactions were performed using Fermentas Maxima SYBR Green Master Mix (Fermentas ThermoScientific, Waltham, MA). Reactions were performed in 25 µl volumes containing 200 µg cDNA and 0.5 µM gene specific primers. Each cDNA sample was analyzed in triplicate (all Cts within 0.5 units to confirm amplification consistency) on an Applied Biosystems 7300 Real-Time PCR System. Primers were based on Berry et al. [106] to P. falciparum genes Pfs16, Pfs25, and A18S rRNA. PCR cycling conditions were as follows: 50°C/2 min; 95°C/10 min; 50 cycles with 95°C/15 sec denaturing and 60°C/1 min annealing-elongation. Since expression of A18S rRNA is high in all parasite stages and varies with the parasitemia [106], this gene and the ribosomal protein S7 gene (An. stephensi) were used to normalize parasite gene target data. Each plate also included no template negative controls. These assays were completed with four separate cohorts of P. falciparum-infected An. stephensi.
Laboratory reared 3–5 d old female NTG or HM mosquitoes were kept on water for 48 h and then allowed to feed for 30 min on reconstituted human blood meals [1∶1 washed human RBCs (Interstate Blood Bank) in PBS (Cellgro)] with 1×107 fluorescent beads/ml (3.0–3.4 µm, Sphero Rainbow Calibration particles RCP-30-5A-2; Spherotech, Lake Forest, IL) provided through a Hemotek Insect Feeding System (Discovery Workshops). Non-blood fed mosquitoes were removed and, at 48 h post blood feeding, samples of three whole mosquitoes or three dissected midguts were placed in cell lysis buffer (Invitrogen), pulse sonicated, and filtered through a 35 µm nylon mesh to remove tissue debris. Pelleted samples were rinsed once with phosphate-buffered saline and then analyzed by flow cytometry. Data acquisition was performed with a FACScan flow cytometer (BD Biosciences), and analysis was conducted using FlowJo software (version 6.4.1; Tree Star, Ashland, OR). The number of beads per three midguts was quantified and subtracted from each analyzed sample of three whole mosquitoes to remove the contribution of beads remaining in the midgut to whole body bead counts. Statistical significance was determined by Student's t-test.
|
10.1371/journal.pgen.1007184 | Kinesin Khc-73/KIF13B modulates retrograde BMP signaling by influencing endosomal dynamics at the Drosophila neuromuscular junction | Retrograde signaling is essential for neuronal growth, function and survival; however, we know little about how signaling endosomes might be directed from synaptic terminals onto retrograde axonal pathways. We have identified Khc-73, a plus-end directed microtubule motor protein, as a regulator of sorting of endosomes in Drosophila larval motor neurons. The number of synaptic boutons and the amount of neurotransmitter release at the Khc-73 mutant larval neuromuscular junction (NMJ) are normal, but we find a significant decrease in the number of presynaptic release sites. This defect in Khc-73 mutant larvae can be genetically enhanced by a partial genetic loss of Bone Morphogenic Protein (BMP) signaling or suppressed by activation of BMP signaling in motoneurons. Consistently, activation of BMP signaling that normally enhances the accumulation of phosphorylated form of BMP transcription factor Mad in the nuclei, can be suppressed by genetic removal of Khc-73. Using a number of assays including live imaging in larval motor neurons, we show that loss of Khc-73 curbs the ability of retrograde-bound endosomes to leave the synaptic area and join the retrograde axonal pathway. Our findings identify Khc-73 as a regulator of endosomal traffic at the synapse and modulator of retrograde BMP signaling in motoneurons.
| Retrograde axonal transport is essential for normal synaptic function and neuronal survival. How endosomes are specifically sorted from the synaptic terminal for retrograde axonal transport is currently not known. At the Drosophila neuromuscular junction, receptors for the Bone Morphogenic Protein signaling pathway are transported from the synapse to the neuron cell body for the proper establishment of synaptic growth and function of motoneurons. Using this system we demonstrate that a kinesin motor protein, Khc-73, is necessary for the efficient sorting of retrograde bound vesicles to the retrograde transport route.
| Bidirectional communication between the neuronal cell body and distant synaptic terminals is essential for synapse formation, plasticity and neuronal survival [1, 2]. This is achieved primarily through highly regulated axonal transport. Anterograde transport is mediated by plus-end directed kinesin motor proteins that deliver synaptic vesicles and newly synthesized proteins to the synapse, while retrograde transport of cargo destined for the cell body, such as activated receptor complexes, is accomplished by dynein protein complexes [1, 3–5]. Kinesin and dynein motors are also required for endosomal traffic within the cell. The coordinated action of anterograde and retrograde motors ensures the proper sorting and delivery of signaling complexes, proteins and organelles [6]. Although defects in endosomal traffic and axonal transport have been associated with a number of nervous system diseases including Charcot-Marie-Tooth disease, Amyotrophic Lateral Sclerosis, Huntington’s disease and Parkinson’s disease, we know little about how signaling endosomes are routed from the synapse to the retrograde pathway [7–13]. Retrograde signaling has been extensively studied at the Drosophila larval neuromuscular junction (NMJ). In particular the Bone Morphogenic Protein signaling pathway (BMP) has been identified as a major regulator of synaptic growth and function. As such, many regulators of synaptic endosomal sorting have been identified in the regulation of BMP signals at synaptic terminals. Nevertheless, how activated receptors are preferentially sorted to travel to the nucleus is currently unknown.
The movement of endosomes within the cytoplasm is directed through the actions of microtubule binding proteins such as minus end dynein motors, and plus end directed kinesins. Co-ordination and competition between these opposing motors for endosome cargoes regulates the transport of proteins to their correct targets [1, 4, 14–18]. In this study, we have discovered a surprising role for the plus-end directed microtubule motor protein Khc-73 in retrograde sorting of signaling vesicles at the Drosophila larval NMJ.
Khc-73 and its vertebrate homolog KIF13B/GAKIN are kinesin 3 motor protein family members with multiple protein domains and diverse roles in both vertebrates and invertebrates [15, 19–32]. At its N-terminal, Khc-73 contains a kinesin motor necessary for its association with microtubules and plus-end directed transport to synaptic terminals, and at its C-terminal, a Cytoskeletal Associated Protein GLYcine rich (CAP-GLY) domain that provides microtubule association properties [30, 32]. In the nervous system, through microtubule cytoskeleton interactions, both KIF13B and Khc-73 have been shown to participate in mechanisms that control neuronal polarity: Khc-73 has a role in spindle orientation in neuroblasts [30], and KIF13B is involved in the establishment of axonal structures in post-mitotic neurons [20]. Interestingly, KIF13B/Khc-73 has been implicated in the regulation of endosomal dynamics [33, 34] and axonal transport [15] through interaction with Rab5-GTPases. Our previous findings suggested that Khc-73, under strong inhibitory control of the microRNA miR-310-313 cluster in motoneurons at the Drosophila NMJ, plays an important role in the regulation of synaptic function by influencing presynaptic neurotransmitter release [35].
In order to investigate the mechanism of action of Khc-73, we have generated loss of function deletions in Khc-73 gene in D. melanogaster and examined the motoneurons of third instar larvae. While the number of synaptic boutons at the NMJ and the amount of neurotransmitter release per action potential are unaffected in Khc-73 mutant larvae, we find a small but significant decrease in the number of presynaptic release sites. Our experiments indicate the presence of Khc-73 function in BMP signaling by demonstrating a strong genetic interaction between Khc-73 and members of the BMP signaling pathway. We further show that activation of retrograde BMP signaling that normally leads to accumulation of pMad in the nuclei of motoneurons is significantly suppressed when Khc-73 is genetically removed. Our findings suggest that Khc-73 exerts its function by influencing the sorting of endosomes at the NMJ and promoting retrograde routing of endosomes. Our findings identify, for the first time, a plus-end directed microtubule motor protein as a regulator of retrograde signaling in motoneurons.
Khc-73 is a member of the KIF superfamily of kinesin motor proteins and the homologue of the vertebrate KIF13B/GAKIN (S1A Fig) [30, 35]. We have previously shown that Khc-73 is a target of the micro RNA miR-310-313 cluster in motoneurons. We found that loss of the miR-310-313 cluster led to abnormally enhanced neurotransmitter release at the NMJ; this enhancement was fully reversed to wild type levels as a result of neuronal knock down of Khc-73 [35]. In order to investigate the role of Khc-73 in more detail, we generated deletions in the Khc-73 gene by imprecise excision of a P-element transposon insert in the vicinity of the 5’ UTR of Khc-73 (S1B Fig). We isolated two deletion flies Khc-73149 and Khc-73193 missing portions of the Khc-73 5’UTR and the ATG start (S1B Fig); we also isolated a fly where the P-element was excised precisely leaving the entire genetic region of Khc-73 intact (Khc-73100) (S1B Fig). Our western blot analysis with an antibody against the C-terminal end of Khc-73 indicates that both Khc-73149 and Khc-73193 are protein null alleles (S1C Fig). Khc-73 is maternally expressed and is expressed in the embryo [32]. We examined the expression pattern of Khc-73 protein in motoneurons with transgenic overexpression, since we were not able to detect a specific signal using our antibody against Khc-73 in larval preparations. We overexpressed HA-tagged Khc-73 transgene in motor neurons and detected punctate accumulation of HA-Khc-73 both in axons and in synaptic boutons at the NMJ (S1D and S1E Fig). In addition, we tested transcriptional activity of Khc-73 by generating a Khc-73-Gal4 fly (containing 4kb of Khc-73 genomic sequence driving Gal4 expression, see methods). Crossing this fly to UAS-mCD8-GFP transgene led to the expression of GFP in nearly all neurons including motor neurons, suggesting that Khc-73 transcription is active in all motor neurons in third instar larvae (S1F and S1G Fig). We also found Khc-73 transcription widely expressed in the brain of adult flies (S1H Fig).
Based on the previously published roles for Khc-73 in neuroblasts, endosomal sorting, axon morphology and synaptic function [21, 29, 30, 34–36], we expected loss of Khc-73 to cause significant defects in the normal synaptic function and/or structure. To our surprise, we found only mild defects (S2E and S2F Fig) in Khc-73 mutant larvae in our assessment of gross synaptic structure at the larval NMJ. The number of synaptic boutons and the muscle surface area (MSA) at NMJs were not significantly different comparing Khc-73 mutant and control larvae; this was true for muscle 4 NMJs (Fig 1A–1C) as well as muscle 6/7 NMJs (S1I and S1J Fig).
To test whether loss of Khc-73 might affect synaptic function, we examined the baseline electrophysiological properties including miniature excitatory postsynaptic currents (mEPSCs), evoked excitatory postsynaptic currents (EPSCs) and quantal content (QC) and found no differences between Khc-73 mutants and wild type larvae (Fig 1D and 1E). Similarly, we tested synaptic vesicle recycling dynamics in Khc-73 mutant NMJs with high frequency stimulation and found no significant difference in the decay of the synaptic response compared to controls (S1K and S1L Fig). Consistent with the lack of defects in baseline synaptic function, we found no significant changes in the fluorescent intensity of the synaptic vesicle calcium sensor synaptotagmin (SYT), synaptic vesicle marker cysteine string protein (CSP) or synaptic vesicle recycling protein Epidermal growth factor receptor pathway substrate clone 15 (EPS15) in Khc-73 mutant larvae (S2A–S2D Fig). We found a mild reduction in the staining intensity for the postsynaptic marker Discs large (Dlg) (S2E and S2F Fig) but no differences in the expression level of postsynaptic glutamate receptor subunit A (GluRIIA) (S2G Fig).
Altogether these findings indicate that synaptic growth and function are largely normal in Khc-73 mutant.
We have previously reported an increase in the accumulation of the active zone protein Bruchpilot (Brp) in miR-310-313 cluster mutant larvae that could be reduced by transgenic knockdown of Khc-73 [35]. Therefore, we set out to conduct a deeper examination of Khc-73 mutants to understand the mechanism of action of Khc-73 in motoneurons. As our previous data would predict, we found a significant decrease in the number of presynaptic release sites per NMJ in Khc-73 mutant larvae, as indicated by a reduction of the number of Brp Puncta (Fig 2A and 2B). Inclusion of a genomic fragment containing the entire genetic region of Khc-73 gene restored synaptic defects in Khc-73 mutant larvae, indicating that this defect is related to loss of Khc-73 (Fig 2B). Previously we showed that Khc-73 is under control of the microRNA cluster miR-310-313 [35]. To maintain this relationship in our tissue specific rescue, we used a Khc-73 transgene Khc-73-3’UTR(K014) [35] that retains this negative regulatory control. We found that transgenic expression of Khc-73 in presynaptic motoneurons, but not in postsynaptic muscles was sufficient to establish a normal number of presynaptic release sites (Fig 2C and 2D). This result indicates that Khc-73 function in motoneurons is required for normal maturation of synaptic release sites.
During larval development, both the coordinated growth of synaptic boutons and the establishment of synaptic strength at the NMJ are largely dependent on a retrograde signaling cascade that is initiated by the release of the Bone morphogenic protein Glass bottom boat (Gbb) in postsynaptic muscles. Gbb signals through type I and type II BMP receptors, leading to phosphorylation of and subsequent accumulation of the BMP transcription factor Mad (Mothers against decapentaplegic) in the nuclei of motor neurons [37–41]. Through this signaling cascade, genes that control synaptic growth and function are transcriptionally regulated [42–44]. The decrease in the number of presynaptic release sites in Khc-73 mutant larvae, therefore, prompted us to examine the state of BMP signaling in these mutants.
The first indication of Khc-73 involvement with BMP signaling came from genetic interaction experiments between Khc-73 and the Drosophila homolog of vertebrate SMAD4, Medea. Medea is a transcriptional co-factor that is required for normal BMP signaling in motor neurons [45]. We found that a combination of previously published alleles MedeaG112 and MedeaC246 resulted in a very small reduction in the number of boutons at the NMJ compared to MedeaC246 homozygous loss of function mutant [45], suggesting that G112 is a hypomorphic allele (Fig 3A and 3E). Interestingly, in transheterozygous combinations of Khc-73 and Medea, we found a significant reduction in the number of presynaptic release sites per NMJ (Fig 3A and 3B), no change in Brp puncta per bouton (Fig 3C), a significant reduction in synaptic area (Fig 3D) and a significant reduction in bouton number with MedC246 allele but not MedG112 (Fig 3E), as compared to heterozygous MedeaG112 controls. This transheterozygous genetic interaction suggested that Khc-73, while having a mild influence on baseline BMP signaling, becomes critical when BMP signaling is compromised. In support of these results, we also found a strong genetic interaction between Khc-73 mutants and a mutation in the BMP type II receptor wishful thinking (wit): transheterozygous combination between Khc-73 and wit A12 mutants led to a significant reduction in the number of presynaptic release sites (Fig 3F and 3G), synaptic area (Fig 3I) and bouton number (Fig 3J) but no change in Brp puncta per bouton (Fig 3H).
To further explore the functional link between Khc-73 and BMP signaling, we generated double mutant combinations of Khc-73 and Medea. We assessed these double mutant combinations for defects in Brp puncta number at the NMJ and for accumulation of Brp in axons (as previously reported [46]). We found that defects in active zone number and abnormal Brp accumulation in axons in the transallelic combination of MedeaC246/MedeaG112 were not further enhanced upon removal of Khc-73 (S3A–S3F Fig), indicating that Khc-73 and Med likely function in the same and not parallel pathways with respect to these phenotypes.
We then tested whether defects in Brp puncta number in Khc-73 mutants can be restored by overexpressing BMP signaling in motoneurons. Indeed, overexpression of a constitutively active form of BMP type I receptor Thick veins (TKVACT) in motoneurons was capable of reversing the reduction in Brp puncta defect in Khc-73 mutant larvae (Fig 4A–4C).
These results prompted us to compare the degree of axonal accumulation of Brp and another synaptic marker, synaptotagmin (SYT) between Khc-73 and Mad mutants. For this we used a Mad mutant allele (MadK00237) that is known to show a strong reduction of synaptic growth and function at the NMJ and exhibit defects in axonal transport of synaptic markers [41]. Both endogenous Brp and SYT accumulated in large aggregates in axons of Mad mutant larvae compared to wild type or Khc-73 mutant larvae, highlighting the fact that Khc-73 related axonal defects would be comparable to a hypomorphic loss of function of BMP signaling (Fig 5A–5D). In support of this interpretation, the increase in Brp accumulation in axons of Khc-73 mutant larvae was fully reversed as a result of overexpression of TKVACT (Fig 5E and 5F).
We also tested whether abnormal axonal accumulation of Brp in Khc-73 mutant larvae could be due to changes in microtubule structures; however, we found no significant changes in the expression of acetylated tubulin in axons or terminals in Khc-73 mutant larvae when compared to wild type counterparts (S3G and S3H Fig). These results further support that the defects associated with active zone numbers in Khc-73 mutant larvae are most likely related to defects in BMP signaling.
While we did not detect measurable changes in the accumulation of pMad in response to loss of Khc-73 (S4A and S4B Fig), or overexpression of Khc-73 transgene (S4C and S4D Fig), the genetic interactions described above provide strong evidence for a functional link between Khc-73 and BMP signaling. In order to strengthen this link and extend it to the regulation of synaptic function, we conducted a number of electrophysiological examinations. Mild to moderate overexpression of TKVACT in motor neurons can lead to an enhancement in synaptic release without significantly affecting the number of synaptic boutons [44, 47]. We found that loss of Khc-73 could significantly block the ability of TKVACT to enhance synaptic strength (Fig 6A and 6B).
Similarly, we found that overexpression of the BMP ligand Gbb in postsynaptic muscles led to a significant enhancement in quantal content (Fig 7A and 7B). As in the case of TKVACT induced enhancement in neurotransmitter release, loss of Khc-73 led to a significant suppression of Gbb-induced enhancement in neurotransmitter release (Fig 7A and 7B).
In addition, we quantified the accumulation of pMad in motoneuron nuclei in the ventral nerve cord (VNC) as a result of postsynaptic overexpression of Gbb. It is well accepted that the accumulation of pMad in the nuclei of motor neurons is a reliable readout of the strength and efficiency of retrograde BMP signaling in motor neurons and is essential for BMP-dependent transcriptional regulation as well as regulation of synaptic function [38, 40, 44, 48, 49]. Muscle overexpression of Gbb led to a statistically significant increase in pMad in the nuclei of motoneurons, which was fully reversed as a result of loss of Khc-73 (Fig 7C and 7D).
Finally, we tested whether Khc-73 gain-of-function would be dependent on normal BMP signaling in motoneurons. We have previously shown that Khc-73 overexpression in motoneurons leads to an enhancement of neurotransmitter release [35]. We found that heterozygosity for the BMP type II receptor wishful thinking (wit) was sufficient to suppress this enhancement to a large extent (Fig 8A and 8B), further supporting the presence of a functional link between Khc-73 and BMP signaling.
From these results a picture emerges, indicating a strong functional link between Khc-73 and BMP signaling in motor neurons. But how does Khc-73 interact with BMP signaling? BMP signaling in motoneurons depends on tightly regulated endosomal traffic. For example, pMad accumulation in motoneuron nuclei in response to activation of BMP signaling at the synapse is dependent on retrograde routing of signaling endosomes containing BMP receptor complexes from the nerve terminal along axons to the cell body [5]. Conversely, routing of BMP receptor complexes to lysosomal pathways appears as one of the mechanisms that attenuates BMP signaling in motor neurons [50–52]. Therefore, we considered a role for Khc-73 in both retrograde routing as well as lysosomal sorting of BMP receptor complexes.
To test these possibilities, we assessed the level of BMP receptors Wit and TKV using a combination of Western blot analysis and immunohistochemistry. Western blot analysis of CNS and body wall muscle tissue (containing NMJ terminals) revealed no change in the level of endogenous Wit protein as a result of genetic removal of Khc-73 (Fig 9A–9D). The available antibody to Wit does not detect endogenous Wit in immunohistochemistry. Thus we turned to transgenic Wit and Tkv to visualize their localization at the synapse. Static images of the boutons in live preps of WIT-GFP revealed punctate accumulations at the NMJ and an increase of Wit receptor intensity in Khc-73 mutants at muscle 4 and muscles 6/7 (Fig 9E–9H). Similarly, TKV:YFP transgene expression appeared more punctate at muscle 4 (S5A Fig) and muscles 6/7 (S5C Fig), trending towards increased intensity at muscle 4 (S5B Fig), while significantly increasing in intensity at muscles 6/7 (S5D Fig) in Khc-73 mutants. We ruled out changes in TKV:YFP transgene transcription by quantitative PCR (S5E Fig) and did not observe obvious changes in axonal traffic of TKV:YFP in motoneurons (S5F Fig).
We next tested our model that Khc-73 loss can suppress BMP signaling by examining pMAD levels in larvae overexpressing the Wit receptor in presynaptic neurons and in larvae overexpressing the Gbb ligand from postsynaptic muscle. Overexpression of Wit enhanced presynaptic pMad levels (Fig 10A and 10B). In Khc-73 mutants, this enhancement was significantly suppressed (Fig 10A and 10B). Similarly, muscle overexpression of Gbb enhanced pMAD levels in presynaptic boutons. Khc-73 loss also suppressed this increase (Fig 10C and 10D).
It has been demonstrated that BMP receptor activity can be dampened when trapped inside the lumen of multivesicular bodies (MVBs) at the NMJ [53]. Generally, MVBs are intracellular vesicles that contain one or more smaller vesicles within their lumen and play an important role in signal transduction and endosomal sorting [54, 55]. Current evidence suggests that MVBs may be at the crossroads for endosomal cargo joining the lysosomal pathway, the retrograde pathway or the exosomal secretory pathway [55, 56]. We find that fluorescence intensity of the MVB localized protein Hrs (hepatocyte growth factor related tyrosine kinase substrate) is increased by 20% at the NMJ in Khc-73 mutant larvae overexpressing the BMP receptor Wit (S6A and S6B Fig). Suggesting that there are more MVBs in Khc-73 mutants in this Wit overexpressing background. Therefore, a scenario can be considered in which retrograde bound BMP receptors are encapsulated in multivesicular bodies and may be stalled at the NMJ in Khc-73 mutants.
Together, these results suggest that degradation of BMP receptors is not a likely explanation for the inhibition of BMP signaling in Khc-73 mutant larvae. Secondly, our findings suggest that while BMP receptors appear to accumulate at the NMJs in Khc-73 mutants, they are in an endosomal state that prevents these receptors from signaling.
Previous studies on Khc-73/KIF13B have identified endosomal sorting roles for this protein [15, 21, 22, 27, 28, 34, 57]. In order to gain additional insight into the role of Khc-73 in the regulation of endosomal traffic, we conducted an ultrastructural analysis of NMJ synapses in Khc-73 mutant larva. Our analysis revealed no gross abnormalities in presynaptic boutons (Fig 11A–11F): different morphometric measures of active zones and synaptic vesicles appeared normal in Khc-73 mutant larvae (Fig 11B–11E); however, we did detect a small increase in the depth of the subsynaptic reticulum (SSR) (Fig 11F). Interestingly, although we find no statistical difference in the mean MVBs per bouton (1.28±0.29 control and 1.68±0.41 Khc-73), we found a proportion of boutons with an abnormally higher number of MVBs (7–9 MVBs per bouton) in Khc-73 mutant larvae (Fig 11G and 11H). The trend towards more MVBs in Khc-73 mutant boutons suggested a role for Khc-73 in endosomal sorting. Therefore, we turned to exploring a possible role for Khc-73 in the regulation of endosomal dynamics by examining the expression of transgenic Rab-GTPases at the synapse. Rab-GTPases are small GTPases that associate with endocytic vesicles and are known to mediate many aspects of endosomal traffic in all eukaryotes [58]. Based on previous reports on interaction between Khc-73 with the early endosome associated Rab5 in vitro [34], we tested the expression pattern of Rab5 at the NMJ in Khc-73 mutant larvae with a UAS-Rab5:YFP transgene. However, we found that in Khc-73 mutants the punctate appearance of Rab5:YFP was unaffected in terms of fluorescence intensity or localization (Fig 12A and 12B). Similarly, we did not detect any effect on the expression level of the recycling endosomal marker Rab11 (Fig 12C and 12D). In most eukaryotic cells Rab5 positive internalized vesicles become associated with Rab7 along their path of maturation [59–62]; Rab7 containing late endosomes are then either routed to the lysosomal pathway or the recycling pathway [58, 63]. In neurons, the transition from Rab5 to Rab7 is also necessary for routing late endosomes onto the retrograde pathway [64]. The retrograde pathway is necessary for transporting signaling complexes, neurotrophic factors and other cellular proteins from nerve endings to the cell body [2]. Interestingly unlike the case of Rab5, we found an abnormal increase in Rab7 accumulation at synaptic boutons in Khc-73 mutants (Fig 12E and 12F). These results suggested to us that the normal dynamics of Rab7 positive vesicles and by extension those of BMP receptors are disrupted in Khc-73 mutant larvae.
In order to examine the dynamics of late endosomal traffic in more detail, we set out to conduct live imaging in dissected larvae expressing Rab7:GFP. To see if our observations of Rab7:GFP would be relevant to the dynamics of Wit/Tkv complexes, we confirmed in fixed samples that Wit and Rab7:GFP colocalized when expressed simultaneously (S7A Fig, Pearson’s r coefficient 0.68). We also confirmed that Tkv and Wit colocalized at the NMJ (S7B Fig, Pearson’s r coefficient 0.60). In live dissected larval preparations, Rab7:GFP showed dynamic movement within synaptic boutons in both wild type and Khc-73 mutants (Fig 13A–13C and S1 Movie and S2 Movie). We noticed that occasionally a Rab7 marked vesicle left the synaptic area and moved retrograde towards the shaft of the axon. Vesicles entering the axon moved, paused and continued moving out of the NMJ. We measured the velocity of these vesicles when in motion and calculated the mean velocity in the anterograde and retrograde directions (Fig 13B–13E, S8A–S8D Fig and S3 Movie and S4 Movie) and found no statistical difference in their velocities. We also recorded the time spent paused in a single spot (Fig 13F), the number of pauses for each spot (Fig 13G) and summed the total time paused in the axon (Fig 13H). Here, our assessment of Rab7 dynamics revealed a significant difference between control and Khc-73 mutant larvae. We recorded long periods of pausing or stalling of Rab7 positive vesicles in Khc-73 mutants, which showed statistical difference compared to our recordings in control larvae (Fig 13F and 13H, S1 Movie and S2 Movie). This pausing phenotype provides one explanation for the increase in Rab7:GFP in Khc-73 NMJs, however alternative explanations related to Rab7:GFP protein turnover are also possible.
We next performed time lapse imaging on TKV-YFP expressing Khc-73 mutant larvae focusing on the axon shaft near the synapse. Here we observed a similar stalling phenotype of TKV-YFP puncta in Khc-73 mutants whereas in control larvae the axonal shaft was devoid of stalled puncta (Fig 13I and 13J, S5 Movie and S6 Movie)).
As an additional test for axonal retrograde transport, we used a peripheral axon injury model developed by Collins and colleagues for activating Jun-N-terminal kinase (JNK) signaling in motor neurons [65]. In this model, crushing peripheral axons in larvae leads to a strong transcriptional upregulation of the JNK phosphatase puckered (puc) in the injured motoneurons [65]. The puc transcriptional response to axon injury is dependent on axonal retrograde transport [65]. Using a puc-LacZ transcriptional reporter line, we assessed JNK activation in motoneurons in response to nerve crush. In Khc-73 larvae, we found that puc transcriptional upregulation as a result of axonal injury was indistinguishable from that of control larvae (S8E–S8H Fig). Thus we can rule out a defect in retrograde axonal transport in Khc-73 mutants. Similarly, we did not find any significant changes in axonal transport of mitochondria in Khc-73 mutant larvae (S7 Movie and S8 Movie). These results provided strong evidence for a model in which Khc-73 is required primarily in synaptic terminals for efficient routing of retrograde vesicles onto the retrograde path with little influence on bidirectional axonal transport.
Khc-73 function plays a supporting role in retrograde BMP signaling under basal conditions. However under conditions of enhanced BMP signaling, this endosomal coordination by Khc-73 becomes critical to transmit the retrograde signal from the synapse to the neuronal cell body.
Efficient retrograde signaling from synaptic terminals back to the neuronal soma is critical for appropriate neuronal function and survival [2, 7–11]. Nevertheless, we know very little about the molecular steps that facilitate the routing of synaptic endosomes destined for retrograde axonal pathways. Here we describe several lines of evidence for a potential role for Khc-73 in this process. Khc-73 mutant larvae develop grossly normal synaptic structure and function at the Drosophila larval neuromuscular junction (NMJ), but we find a reduction in the number of presynaptic release sites. Through genetic interaction experiments, we show that this defect is most likely the result of abnormal BMP signaling in motoneurons: transheterozygous combinations of Khc-73 and Medea or wit mutants show a significant loss of presynaptic release sites compared to control. Khc-73 becomes even more critical, when higher demand is put on the motoneuron by activating BMP signaling: loss of Khc-73 largely blocks the retrograde enhancement in synaptic release in response to activation of BMP pathway in motor neurons. Consistently we have previously shown that transgenic knock down of Khc-73 in motoneurons blocks the ability of the NMJ to undergo retrograde synaptic homeostatic compensation [35]. Our findings show that when BMP signaling is activated, loss of Khc-73 reduces the accumulation of pMad in motoneuron nuclei, suggesting a role for Khc-73 in the regulation of retrograde signaling. Our immunohistochemical assessment and live imaging analysis of Khc-73 mutant larvae provide evidence for involvement of Khc-73 in at least two steps in endosomal dynamics in motoneurons. On the one hand, Khc-73 is required for normal dynamics of internalized endosomes through late endosomal and multivesicular stages, and on the other Khc-73 plays a role in facilitating the routing of endosomes onto the retrograde pathway (see Fig 14A for model). These defects have two main consequences: first, we find an accumulation of BMP receptors at the NMJ (possibly in multivesicular bodies) without increased local signaling, suggesting that these receptor containing endosomes might be trapped in a state between late endosomal and lysosomal stage (see Fig 14B for model). Second, we see a dampening of the ability of retrograde bound Rab7:GFP tagged endosomes to join the retrograde pathway, illustrating a defect in retrograde movement of vesicles and possibly providing an underlying explanation for the reduction in pMAD when retrograde BMP signaling is activated in Khc-73 mutants. These results together present Khc-73, a plus-end microtubule motor, in the unexpected role of regulation of endosomal traffic from synapse to the soma in motoneurons with a role for ensuring the efficiency of retrograde BMP signaling.
While our findings provide compelling evidence for the proposed model above, we cannot, at this time, rule out the possibility that the abnormal accumulation of BMP receptors at the NMJ and the slowing of retrograde movement of Rab7 positive endosome in Khc-73 mutant larvae could be due to a defect in an intermediate molecule, whose anterograde transport is dependent on Khc-73. In support of such model, we do report an abnormal accumulation of Brp and SYT (two synaptic proteins) in axons. While our data suggests that this abnormal accumulation can be remedied by transgenic activation of BMP signaling in Khc-73 mutants, we cannot rule out the possibility that an anterograde transport defect might exist for other proteins independent of the interaction between Khc-73 and BMP signaling.
Our findings point to a model in which Khc-73 facilitates the routing of retrograde bound vesicles onto the retrograde axonal pathway. This model predicts coordination between endosomes, dynein motors and kinesin Khc-73. The coordinated involvement of dynein and kinesin motor proteins in the transport and sorting of endosomes has been previously proposed and examples supporting this model are mounting [14, 66, 67]. Previously published data for Khc-73 and KIF13B have provided evidence that interaction between early endosomes, dynein motors and microtubules are possible. Khc-73/KIF13B is capable of binding to the GTPase Rab5 (found on early endosomes), thus allowing Khc-73 to localize directly to Rab5 endosomes [15, 34, 68]. As a kinesin motor protein, Khc-73 could then transport these endosomes to the retrograde pathway by moving along the microtubule network in the synapse.
Compelling evidence for a dynein interaction with Khc-73 has been previously demonstrated during mitotic spindle formation [24]. The Khc-73/KIF13B stalk domain is phosphorylated by Par1b and this creates a 14-3-3 adapter protein binding motif [29]. It has been proposed that physical interaction between Khc-73 stalk domain and the dynein interacting protein NudE via 14-3-3 ε/ζ might underlie the interaction between Khc-73 and dynein that is necessary for appropriate mitotic spindle formation [24]. Interestingly, transgenic knock down of NudE in Drosophila larval motoneurons leads to a reduction in the number of presynaptic release sites, a phenotype reminiscent of Khc-73 loss of function [69]. Thus, Khc-73 contains domains and protein-protein interactions that are capable of coordinating endosomes, microtubules and dynein. We propose that Khc-73 is necessary for the normal endosomal sorting and exit of endosomes from the NMJ to support efficient retrograde BMP signaling.
Flies were cultured at 25°C on standard medium except for Gene Switch experiments where RU486 was added to the media (50μM). The following stocks were used: MedC246 (Y324term mutation)[45] and MedG112 (mutation in splice donor site of exon 4)[45] from Herman Aberle [45]. witA12 [37, 39, 70]. witHA4 [37, 45, 49]. UAS-TKVACT and UAS-Gbb99 [40] provided by M.B. O’Connor (University of Minnesota, Minneapolis, MN), UAS-Wit [39], UAS-HA-Khc-73 and UAS-HA-Khc-73-3’UTR(K014) [35], UAS-Wit-GFP, UAS-TKV-YFP [5], BG380-Gal4 [71], Elav-Gal4 [72], OK371-Gal4[73], MHC-Gal4 [74]. Bloomington stocks used were P{y[+m8] = Mae-UAS.6.11}Khc-73[DP00530] (RRID:BDSC_22058), UAS-Rab5:YFP (RRID:BDSC_9775), UAS-Rab7:GFP(RRID:BDSC_42706), UAS-Rab11:GFP (RRID:BDSC_50782), VGlut-Gal4 (RRID:BDSC_24635), MadK00237 (RRID:BDSC_10474), UAS-Mito-HA-GFP (RRID:BDSC_8442), nSyb-Gal4 (RRID:BDSC_51635). UAS-luciferase (RRID:BDSC_35788). puckered LacZ insertion pucE69 [75]. Wild type stock used was w1118.
Khc-73 deletions were created by mobilizing the P-element from y[1] w[67c23]; P{y[+m8] = Mae-UAS.6.11}Khc-73[DP00530]. Virgin y[1] w[67c23]; P{y[+m8] = Mae-UAS.6.11}Khc-73[DP00530] female flies were mated to Cyo/+; Δ2–3, Sb/TM6b males. Male progenies of y[1] w[67c23]; P{y[+m8] = Mae-UAS.6.11}Khc-73[DP00530]/Cyo; Δ2–3, Sb/+ were mated to virgin y, w; CyoGFP/Adv females. Yellow, non Sb, yellow eyed progeny were singly mated to y, w; Adv/CyoGFP virgins and individual stocks were established. P-element excisions were screened with the following primers: OED91: CTGACGGCGCTGTTGCTTG and OED96: GATCTAGAGATGATTCTGCATCACTAG TAAAAATT.
Khc-73 promoter GAL4 construct was generated by cloning a 4kb fragment upstream of the translational start site with primers OED453: CAG GTA CCG CCG AGG AAC CGC TAA CG and OED452:CAG GTA CCC GCG GAT GTG GAT GCA GC. Vector pW+SN attB was modified with a GAL4 sequence cloned as a KpnI/NotI fragment.Khc-73 promoter was subsequently inserted into the unique KpnI site. Genomic Khc-73 is from BACPAC clone CH321-36I16 (BACPAC Resources Center).
Transgenic fly CH321-36I16 was made by standard embryo injection of BACPAC clone CH321-36I16 (BACPAC Resources Center) with ΦC31 –mediated integration into attP site at position 86F of chromosome III.
Wandering third instar larvae were dissected, prepared and embedded as described in [76]. Ultra-thin serial sections of 50 nm thickness were cut from muscle 6, 7 and 12 of hemisegment A3. Electron micrographs were taken at a magnification of 25,000x for measurements, 25,000x and 40,000x for figures. Serial Reconstruction and analysis was conducted on FIJI (Fiji is Just ImageJ) (NIH) [77] and Reconstruct v.1.1.0.0 Software [78].
Wandering third instar larvae were dissected as previously described [74]. Third Instar larvae were dissected in cold HL3 and fixed with 4% Paraformaldehyde for 10 min or 5min ice cold Methanol for GluRIIA staining. Larvae were washed with PBS (Phosphate buffered saline), permeabilized with PBT (PBS with 0.1% Triton X-100), blocked with 5% Normal Goat Serum (NGS) in PBT and placed in primary antibody overnight at 4°C. The larvae were then washed three times for 15min in PBT, placed in secondary antibody for 2 hrs, washed three times for 15min with PBT and mounted in Vectashield (Vector labs). Antibodies used are as follows: anti-GluRIII (1:500) (gift from A. DiAntonio, Washington Univ. St. Louis, MO), anti-Hrs (1:200), anti-SYT (1:1000) (gift from H. Bellen, Baylor College of Medicine, Houston, TX), anti-pMAD (PS1)(1:200) (gift from M.B. O’Connor, University of Minnesota, Minneapolis, MN). anti-Dlg (1:500), anti-nc82 (1:500), anti-GluRIIA(1:500), anti-CSP(1:500), anti-EPS15 (1:50), anti-LacZ (1:100) and anti-Wit (1:10) (Developmental Studies Hybridoma Bank(DSHB)), anti-HA (1:500) (HA.11 clone 16B12) (Covance Research Products), anti-GFP (1:500) (A6455) (Molecular Probes), anti-GFP (1:500) (Rat IgG2a, GF090R)(Nacalai Tesque Inc.), anti-HRP conjugated Alexa 647(1:250) (Jackson ImmunoResearch), anti-acetylated tubulin (1:500) (T7451, clone 6-11B-1 Sigma-Aldrich) and anti-pSmad3 (EP823Y)(Epitomics).
Western blots were performed as previously described [41]. Muscle tissue (without the nervous system and motor axons or imaginal discs) or Brain tissue (VNC and axons) were isolated from wandering third instar larvae dissected in cold HL3. Western blot analysis was performed according to manufacturer’s protocols. Antibodies used: anti-Khc-73 (1:2000)[35], anti-Wit (1:10) (DSHB), anti-actin (Millipore, MAB1501). Gel images were scanned and band intensities were quantified using FIJI (Fiji is just ImageJ software) (NIH) [77].
Synapses were imaged using a ConfoCor LSM710 and Zeiss LSM 780 on an Axiovert 200M inverted microscope (Carl Zeiss, Inc.) with 63x/1.4 oil objective. Image analysis was performed with ImageJ 1.46j (NIH) [79], Imaris (Bitplane Scientific Software), Image Analyst MKII (Image Analyst Software, Novato, CA) and Metamorph (Molecular Devices).
Wandering third instar larvae were dissected in room temperature HL3 to remove the guts and fat bodies. The larval filet was then inverted and stretched onto a coverslip using magnetic dissection pins inside a chamber consisting of a coverslip surrounded by magnet strips. Larval prep was maintained at room temperature in an HL3 bath during imaging. NMJs at hemisegment A3 and A4, muscles 6/7 and 4 were imaged. Axons were imaged at hemisegment A3 to A4. Larvae were imaged for a maximum of 30 minutes after dissection. Axons and NMJs were imaged with 63x 1.4NA oil objective on Axiovert 200 inverted microscope with Zeiss LSM780 confocal (Carl Zeiss, Inc.).
The nerve crush assay was performed as previously described [65]. Briefly, third instar larvae were anaesthetized with carbon dioxide. The segmental nerves at the midbody were then pinched with Dumostar number 5 forceps for five seconds. The larvae were then recovered on standard media for 25 hours at 25°C after which time they were dissected and stained for LacZ.
Wandering third instar larvae were dissected in cold HL3 solution following standard protocol [80]. The spontaneous (mEJC) and evoked (EJC) membrane currents were recorded from muscle 6 in abdominal segment A3 with standard two-electrode voltage-clamp technique [41]. All the recordings were performed at room temperature in HL3 solution containing 0.5mM Ca2+ unless otherwise indicated. The current recordings were collected with AxoClamp2B amplifier (Molecular Devices Inc.) using Clampex 9.2 software (Molecular Devices Inc.). The nerve stimulation was delivered through a suction electrode which held the cut nerve terminal cord. In all voltage clamp recordings, muscles were held at -80 mV. The holding current was less than 5 nA for 90% of the recordings and we rejected any recording that required more than 10 nA current to maintain the holding potential.
The amplitudes of mEJC and EJC were measured using Mini Analysis 6.0.3 software (Synaptosoft) and verified by eye. QC was calculated by dividing the mean EJC amplitude by mean mEJC amplitude. The recording traces were generated with Origin 7.5 software (Origin Lab).Spontaneous and evoked potentials were measured as previously described [49]. Standard two-electrode voltage-clamp technique was used as described in [44].
|
10.1371/journal.pmed.1002586 | Anticipated burden and mitigation of carbon-dioxide-induced nutritional deficiencies and related diseases: A simulation modeling study | Rising atmospheric carbon dioxide concentrations are anticipated to decrease the zinc and iron concentrations of crops. The associated disease burden and optimal mitigation strategies remain unknown. We sought to understand where and to what extent increasing carbon dioxide concentrations may increase the global burden of nutritional deficiencies through changes in crop nutrient concentrations, and the effects of potential mitigation strategies.
For each of 137 countries, we incorporated estimates of climate change, crop nutrient concentrations, dietary patterns, and disease risk into a microsimulation model of zinc and iron deficiency. These estimates were obtained from the Intergovernmental Panel on Climate Change, US Department of Agriculture, Statistics Division of the Food and Agriculture Organization of the United Nations, and Global Burden of Disease Project, respectively. In the absence of increasing carbon dioxide concentrations, we estimated that zinc and iron deficiencies would induce 1,072.9 million disability-adjusted life years (DALYs) globally over the period 2015 to 2050 (95% credible interval [CrI]: 971.1–1,167.7). In the presence of increasing carbon dioxide concentrations, we estimated that decreasing zinc and iron concentrations of crops would induce an additional 125.8 million DALYs globally over the same period (95% CrI: 113.6–138.9). This carbon-dioxide-induced disease burden is projected to disproportionately affect nations in the World Health Organization’s South-East Asia and African Regions (44.0 and 28.5 million DALYs, respectively), which already have high existing disease burdens from zinc and iron deficiencies (364.3 and 299.5 million DALYs, respectively), increasing global nutritional inequalities. A climate mitigation strategy such as the Paris Agreement (an international agreement to keep global temperatures within 2°C of pre-industrial levels) would be expected to avert 48.2% of this burden (95% CrI: 47.8%–48.5%), while traditional public health interventions including nutrient supplementation and disease control programs would be expected to avert 26.6% of the burden (95% CrI: 23.8%–29.6%). Of the traditional public health interventions, zinc supplementation would be expected to avert 5.5%, iron supplementation 15.7%, malaria mitigation 3.2%, pneumonia mitigation 1.6%, and diarrhea mitigation 0.5%. The primary limitations of the analysis include uncertainty regarding how food consumption patterns may change with climate, how disease mortality rates will change over time, and how crop zinc and iron concentrations will decline from those at present to those in 2050.
Effects of increased carbon dioxide on crop nutrient concentrations are anticipated to exacerbate inequalities in zinc and iron deficiencies by 2050. Proposed Paris Agreement strategies are expected to be more effective than traditional public health measures to avert the increased inequality.
| Increasing atmospheric carbon dioxide concentrations are anticipated to decrease the zinc and iron concentrations of some crops.
The disease burden induced by changes in zinc and iron concentrations in crops, and optimal mitigation strategies, are unknown.
We developed a country-level microsimulation model of zinc and iron deficiency to project the health effects of zinc- and iron-deficiency-related diseases in terms of disability-adjusted life years accumulated over the period 2015 to 2050, worldwide.
We estimated that carbon-dioxide-induced reductions in the zinc and iron concentrations of crops would be expected to induce approximately 125.8 million disability-adjusted life years globally over the period from 2015 to 2050, disproportionately affecting South-East Asian and sub-Saharan African countries.
We also estimated that a climate mitigation strategy concordant with the Paris Agreement would avert a greater portion of this burden (approximately 48.2%) than traditional public health interventions (approximately 26.6%).
The effects of increased carbon dioxide concentrations on crop nutrient concentrations may increase existing inequalities in nutritional deficiencies by 2050.
Climate mitigation strategies, such as the Paris Agreement, may be more effective than traditional public health interventions in averting this increased inequality.
| Increasing atmospheric carbon dioxide concentrations are anticipated to affect public health through increased unsafe weather events, extreme heat, air pollution, and allergen and disease vector exposure [1]. Increasing carbon dioxide concentrations are also anticipated to reduce the concentrations of zinc and iron in many agricultural crops, particularly C3 plants, which rely solely on C3 carbon fixation and include common rice and wheat varieties; C3 plants constitute 95% of terrestrial plants, account for over half of global caloric consumption, and serve as the primary source of dietary zinc and iron for populations at highest risk of nutritional deficiencies [2–5]. The empirical observation of lowered zinc and iron concentrations under elevated carbon dioxide is well established, although the mechanisms are not yet well understood and likely relate in part to changes in crop transpiration [2,5,6]. Zinc and iron deficiencies, in turn, increase the risk of infections, diarrhea, and anemia [7,8]. It remains unclear, however, where and to what degree increasing carbon dioxide concentrations may increase the global burden of nutritional deficiencies, and which strategies may best mitigate the increase. An active area of public health policy debate is whether direct climate change mitigation strategies, such as the Paris Agreement, will be sufficient or comparable to traditional public health measures to combat the complications of carbon-dioxide-induced zinc and iron deficiencies—particularly supplementation and disease control programs to combat the heightened risk of infections, diarrhea, and anemia [9].
The impact of increased carbon dioxide on nutritional deficiencies and their associated diseases depends on 5 related factors: the magnitude of increase in atmospheric carbon dioxide concentrations; the resultant changes in crop nutrient concentrations; current and future trends in consumption of crops and other dietary nutrient sources; the risks of deficiencies, given dietary consumption; and the risks of disease, given nutritional deficiencies.
We developed a model (Fig 1) to estimate where and to what degree increasing carbon dioxide concentrations would be expected to increase the global burden of zinc and iron deficiencies and their associated diseases over the period 2015 to 2050. We used this model to evaluate which strategies may best mitigate the burden of disease attributable to increased carbon dioxide in the atmosphere. Specifically, we created a microsimulation in which demographically representative populations in each country were or were not exposed to carbon-dioxide-induced changes in the zinc and iron concentrations of crops, and experienced related disease burdens quantified in disability-adjusted life years (DALYs).
To estimate the impact of increasing carbon dioxide concentrations on nutritional deficiencies and associated disease rates, we used 5 main data sources. We focused our analysis on country-level estimates of the following: carbon dioxide concentration changes from the Intergovernmental Panel on Climate Change [10], crop zinc and iron concentrations from the US Department of Agriculture (USDA) [11], changes in crop zinc and iron concentrations due to increasing carbon dioxide concentrations from Myers et al. [5], dietary supply from the Statistics Division of the Food and Agriculture Organization of the United Nations (FAOSTAT) [12], and disease prevalence among different populations globally from the Global Burden of Disease Project (GBD) [13]. Example values for Nigeria are shown in Table 1, with links to primary data for other countries in S1 and S2 Texts.
We incorporated the data into a microsimulation model (Fig 1) that simulates individuals within each country, reflecting the range of consumption of zinc and iron given the variations in dietary patterns in each country, and associated zinc and iron deficiency status and attendant risks for infections, diarrhea, and anemia. For each of 137 countries with data, we simulated 2 identical samples of 10,000 individuals demographically distributed to match country-specific age, sex, and secular birth/death trends. Each individual was probabilistically given values of zinc and iron consumption, morbidity risk, and mortality risk (Table 1; Fig 1). One population sample was subjected to carbon-dioxide-induced declines in the zinc and iron concentrations of crops while the other maintained current levels of zinc and iron concentrations. The primary model outcome was the difference between the 2 populations in DALYs per 1,000 persons. DALYs are the sum of years of life lost from disease and years of life lived with disability, weighted by the disability associated with each disease endpoint (Table 1).
In the simulation, we assumed that current dietary patterns (number of kilocalories/capita/day consumed of each food category) would remain the same despite increasing carbon dioxide concentrations, meaning that individuals would not adjust their total intake of food, or types of food eaten, based on reduction in nutrients in each crop. Average food consumption per capita in each country was determined from FAOSTAT data on approximately 100 mutually exclusive and collectively exhaustive food categories, such as potatoes, bovine meat, and beans [12]. The baseline zinc and iron contents per 100 grams and energy density of each food type were obtained from USDA data [11]. Country-specific zinc consumption distributions were constructed assuming normality and coefficients of variation (CVs) of 25% [16–18]. This method of constructing zinc consumption distributions from FAOSTAT and USDA data is similar to that used by Kumssa et al. [14]. Country-specific iron consumption distributions were also constructed assuming normality and CVs of 25% following prior literature [16]. An alternative assumption for iron consumption distributions, lognormality with CVs of 40% [19], was applied in sensitivity analysis (S4 Text). The annual crop-specific declines in zinc and iron concentrations due to increasing carbon dioxide were estimated from nutrient concentrations of crops grown under carbon dioxide concentrations expected by the year 2050 [5] with assumed linear declines in crop concentrations over the period from 2015 to 2050. The expected carbon dioxide concentrations by the year 2050 used in [5] are based on the Intergovernmental Panel on Climate Change’s assessment of carbon dioxide concentration projections [10], which are the same in all countries. Specifically, projected declines in the micronutrient concentrations of particular crops were extrapolated to broader crop categories, in keeping with previous studies (S1 Text) [9,20].
Given the annual level of zinc and iron consumed by each simulated individual, we determined whether an individual fell below a critical threshold level of zinc or iron intake that would be labeled “deficiency” and would predispose the individual to disease, using the estimated average requirement cut-point method [21–23], which compares a person’s average annual intake to the corresponding weighted estimated average requirement (WtdEAR). A WtdEAR is an average micronutrient requirement for a population and can be calculated by multiplying the requirement for each demographic group by the fraction of people in that demographic group and summing; the requirement for each demographic group is based on population-specific absorption factors [24].
The microsimulation model estimated the total morbidity and mortality burden associated with zinc and iron deficiencies per 1,000 people and for the overall population in each country. Zinc deficiency is associated with a heightened relative risk of morbidity and mortality from malaria, pneumonia, and diarrhea among children under 5 years, with no heightened disease risks among older persons [7]; the disability weights attributable to each of these conditions were obtained from prior assessments (Table 1) [8,13,15]. Prevalence rates, mortality rates, and trends over time for malaria, pneumonia, and diarrhea in each country were obtained from the GBD [13]. Mortality trends assumed continuation of the annual percent change in the per capita mortality for each disease as calculated by the GBD. The GBD calculated these values based on historical data from 1990 to 2013. Similarly, disease prevalence trends assumed continuation of the annual percent change in the prevalence rate for each disease from 1990 to 2013 as obtained from GBD data; in sensitivity analysis, disease prevalence rates were held constant (S8 Table). Iron deficiency was simulated as producing associated anemia, with disability weights, mortality rates, and mortality trends over time also from GBD data [8,13]. We note that iron deficiency and iron deficiency anemia (IDA) are not equivalent, but because IDA occurs when iron deficiency is sufficiently severe to reduce red blood cell production [25,26], the estimated average requirement cut-point method is a viable strategy for estimating the relative increase in IDA [27].
Validity of model estimates was assessed by comparing model estimates to values reported in the literature [8,13,14] (S1–S3 Tables). Model estimates subjected to validation included current zinc and iron dietary supplies and percent deficiencies, current yearly disease burdens, and projected disease burdens from 2015 to 2050 (calculated from current yearly burdens and assuming continuation of secular trends in the scenario without climate change).
We considered alternative strategies to mitigate the disease burden from increased zinc and iron deficiencies attributable to increasing carbon dioxide concentrations. Specifically, we compared 5 traditional public health strategies to strategies consistent with the Paris Agreement to maintain global temperatures within 2°C of pre-industrial levels, adjusting each strategy for current anticipated feasibility and population reach. The 5 public health strategies were (i) zinc supplementation [28], in which zinc is administered daily to cover 80% of children under 5 years of age with random selection; (ii) iron supplementation [29], in which iron is administered weekly to cover 80% of females over 5 years of age; (iii) malaria mitigation [30], in which a previously described portfolio of interventions (e.g., long-lasting insecticide-treated nets, indoor residual spraying, and mass screening and treatment) is adopted at an 80% coverage level; (iv) pneumonia mitigation [31], in which a previously described portfolio of interventions (e.g., Haemophilus influenzae type b vaccine, pneumococcal vaccine, antibiotics for pneumonia, promotion of breastfeeding, vitamin A supplementation, and environmental improvements) is adopted at an 80% coverage level (except for the H. influenzae and pneumococcal vaccines, which are adopted at 90% coverage); and (v) diarrhea mitigation [31], in which a previously described portfolio of interventions (e.g., oral rehydration solution, rotavirus vaccine, antibiotics for dysentery, promotion of breastfeeding, vitamin A supplementation, and environmental improvements) is adopted at an 80% coverage level (except for the rotavirus vaccine, which is adopted at 90% coverage). Each strategy is detailed in S5 Text.
We performed 1-way sensitivity analysis, in which parameters were varied one at a time ± 10%, across all model inputs (S4 Text; S2–S8 Figs). In particular, we varied initial zinc and iron intake distributions across ranges to account for potential changes in dietary patterns and uncertainties in the quality of current measurements, initial disease prevalence rates to account for potential underreporting, and disease mortality rates conditional on deficiency to account for potential changes in healthcare availability. We also performed probabilistic sensitivity analysis by running the model 10,000 times per country while Monte Carlo sampling from probability distributions around each input parameter to compute uncertainty intervals for each outcome (S4 Text; S5–S8 Tables). Probability distributions for input parameters were determined based on previously reported probability distributions, uncertainty estimates, and natural bounds. The model was programmed in R version 3.2.2 [32], with input data and statistical code for replication and extension of our analysis published at https://purl.stanford.edu/tx325yy8269 concurrent with publication.
Our model-based estimates of nutrient supply, nutrient deficiency, and associated disease had a high degree of correspondence to current estimates from other sources (S1–S3 Tables). Specifically, our model-based estimates of global population-weighted zinc supply and deficiency were within 0.8% and 3.3% relative error from values in the literature, respectively [14]. Our estimate of global population-weighted iron deficiency was within 0.6% of values in the literature [8]. Our estimates of the 2015 DALY burdens due to malaria, pneumonia, diarrhea, and IDA were within 2.4%, 0.3%, 1.4%, and 0.2%, respectively, of values in the literature [13]. Estimated 2015 to 2050 DALY burdens due to malaria, pneumonia, diarrhea, and IDA globally were within 0.7%, 1.3%, 2.0%, and 2.7%, respectively, of values in the literature [13]. All other model-based results showed high concordance with prior estimates of current nutrient deficiency and disease (S1–S3 Tables).
In the absence of increasing carbon dioxide concentrations, zinc and iron deficiencies would be expected to induce 1,072.9 million DALYs globally over the period 2015 to 2050 according to our model. This global burden would represent approximately 2.0% of the expected DALYs due to all diseases of any kind over this period, which would make zinc and iron deficiencies the 13th leading risk factor for DALYs if the relative burdens attributable to other risk factors did not change [33].
In the presence of increasing carbon dioxide concentrations, decreased zinc and iron concentrations of crops would be expected to lead to an additional 125.8 million DALYs globally over the period 2015 to 2050, with a disproportionate burden in the World Health Organization’s South-East Asia and African Regions (Fig 2). Countries in the study sample would be expected to experience increases in carbon dioxide concentration from approximately 400 ppm to 550 ppm (a 37.5% increase) from 2015 to 2050 and resultant declines in zinc and iron concentrations of C3 crops of approximately 5%–10% [5,9,10,34]. Given current consumption patterns, per our model, the nutrient concentration changes would induce an additional 8.2 percentage point increase in zinc deficiency (from the current baseline rate of 32.0% of the population deficient) and 6.1 percentage point increase in iron deficiency (from the baseline of 21.8% deficient) in the South-East Asia Region—the most highly affected region for zinc/iron deficiency—by 2050. The least-affected region, by contrast, would be the European Region, which would be anticipated to experience an additional 1.9 percentage point increase in zinc deficiency (from a baseline of 3.7% deficient) and 3.4 percentage point increase in iron deficiency (from a baseline of 11.1% deficient). The difference in the fraction of people placed at risk of zinc deficiency was not primarily due to a greater dependence on C3 crops in the South-East Asia Region for dietary zinc, but because a higher portion of the current population is only marginally above the threshold for zinc deficiency.
Model results suggested that carbon-dioxide-induced reductions in zinc and iron concentrations among crops would increase between-country and between-region inequalities in DALY burden per person (Fig 3). The regions with highest initial per capita burdens due to carbon-dioxide-induced zinc and iron reductions would also be expected to be the most affected over the 35-year period, with increased disparities over time. For example, the WHO African Region would be expected to experience approximately twice the per capita burden of other regions, such as the European Region, over the study period. Moreover, the regions with the highest per capita burdens due to carbon-dioxide-induced reductions (WHO African and South-East Asia Regions) are those with the highest current estimated burdens due to zinc and iron deficiencies (e.g., the current per capita burden of zinc and iron deficiencies in the African Region is approximately 4 times that of the European Region).
Compared with the 5 modeled traditional public health interventions, a climate mitigation strategy such as the Paris Agreement would be expected to avert the greatest portion of the cumulative projected burden of carbon-dioxide-attributable zinc and iron deficiencies between 2015 and 2050 (Fig 4). Even with 80% coverage with zinc and iron supplementation and with intensive disease control programs for malaria, pneumonia, and diarrhea, the sum total of the public health measures would be anticipated to reduce only 26.6% of the global carbon-dioxide-attributable zinc and iron deficiency burden of disease (95% credible interval [CrI]: 23.8%–29.6%). The countries that benefited most from the public health programs were those with high current zinc and iron deficiency DALY burdens (e.g., countries in the African Region), as these countries are expected to experience the greatest increases in deficiencies and highest disease rates over the study period.
By contrast, interventions consistent with the Paris Agreement to keep global temperatures within 2°C of pre-industrial levels, which was previously anticipated to feasibly prevent approximately 47% of the increase in carbon dioxide concentrations by 2050 [35,36], would be expected to avert 48.2% of global carbon-dioxide-attributable zinc and iron deficiency burden of disease (95% CrI: 47.8%–48.5%). Since the 2015 carbon dioxide concentration was approximately 400 ppm and the 2050 carbon dioxide concentration is projected to be approximately 550 ppm [5,9,10,34], adhering to the Paris Agreement would be expected to result in a 2050 carbon dioxide concentration of approximately 480 ppm.
The greater effectiveness of the climate mitigation strategy relative to the public health measures was driven by the fact that it benefited the entire population, whereas the public health measures were targeted to subsets of the population with the highest per capita DALY burden, missing substantial populations with small per capita, but large aggregate, DALY burdens. Furthermore, the Paris Agreement reversed carbon-dioxide-induced deficiencies and related diseases, unlike the public health measures, which each addressed only a single form of deficiency or disease manifestation. The credible intervals for the percent reductions in disease burdens with the mitigation strategies were relatively small as all the mitigation strategies effectively reduce deficiencies or related diseases on a percent basis.
The relative benefits of the climate mitigation strategy versus the traditional public health strategies were unaltered by using a wide range of alternative literature-based input parameters (S2–S8 and S10 Figs; S7 and S8 Tables). In particular, the relative superiority of the climate mitigation strategy did not qualitatively change with alterations to baseline zinc and iron intakes, disease prevalence, or mortality rates. Model projections were generally most sensitive to changes in inputs related to iron (e.g., iron consumption, WtdEARs for iron, and IDA disability weights; S2–S8 Figs), but no scenario altered the comparative effectiveness of climate mitigation relative to traditional public health measures.
Carbon-dioxide-induced reductions in zinc and iron concentrations among crops are anticipated, by our model, to lead to an additional 125.8 million DALYs globally over the period 2015 to 2050. A disproportionate burden of these DALYs is anticipated to affect countries with high existing burdens from zinc and iron deficiencies, thereby increasing existing global inequalities in nutritional deficiency by disproportionately affecting South-East Asia and Africa.
Our study suggests that climate mitigation strategies such as the Paris Agreement would be expected to avert a greater portion (approximately 48%) of the projected increase in zinc and iron deficiencies, about 1.8 times greater than the mitigation from traditional public health interventions including nutrient supplementation and disease control programs to counteract the heightened risk of diseases associated with zinc and iron deficiencies. A 48.2% reduction in DALY burden due to reduced zinc and iron concentrations of crops with the climate mitigation strategy is reasonable, as the climate mitigation strategy reduces the increase in 2050 carbon dioxide concentrations by approximately 47%, and a linear relationship between carbon dioxide concentrations and plant micronutrient concentrations was assumed. However, this linear relationship need not be the case and is not an obvious result due to potential nonlinearities created by relationships between micronutrient supplies, deficiencies, diseases, and resulting DALY burdens.
Our study extends prior work projecting changes in the zinc and iron concentrations of crops due to increasing carbon dioxide concentrations, changes in the prevalence of zinc deficiencies, and countries at risk from the changing iron concentrations of crops (by computing changes in iron supplies and cross-referencing them to estimates of the current prevalence of anemia) [2–5,9,20]. We extended this body of work by modeling diseases stemming from zinc and iron deficiencies using a microsimulation model to project disease burdens attributable to the nutrient concentration effects of increased atmospheric carbon dioxide and to evaluate potential mitigation strategies. Critically, this approach enabled us to understand this important health impact of climate change in the context of existing disease burdens and health inequalities between countries and regions.
Among the primary limitations of our analysis is our inability to anticipate what changes in dietary patterns might occur simultaneously with increasing carbon dioxide concentrations. We varied consumption parameters over wide ranges in sensitivity analysis and found that our findings of an overall increased inequality in disease burden between countries and the impact of climate mitigation relative to public health measures were consistent across a broad range of scenarios. This is an important area for further research. Another limitation is that we assumed that secular exponential trends in disease mortality rates would continue, reflecting economic development and health services improvement. Should recent trends not continue in positive directions, our projections may be viewed as optimistic, rendering our results conservative forecasts of the complications from zinc and iron deficiencies. A third limitation is that we assumed that crop zinc and iron concentrations decline linearly with increasing carbon dioxide concentrations. If the declines are convex, our disease burden projections are likely underestimates, while if they are concave, our disease burden projections are likely overestimates. Finally, because we used WtdEARs to project deficiencies, we felt that disaggregating results by gender and age group would imply a level of precision we cannot reliably provide at this point.
Future studies may directly experiment with interventions to induce higher zinc and iron concentrations in plants that can be grown in the most heavily affected countries. Further studies should also evaluate whether dietary changes may be directed to reduce the burden of disease associated with nutritional deficiency attributable to increased carbon dioxide in the atmosphere. Finally, subsequent studies might seek to evaluate the cost and cost-effectiveness of different mitigation strategies.
While such studies are underway, our findings indicate that rising atmospheric carbon dioxide concentrations and their impacts on nutrient concentrations in crops are likely to increase health inequalities from nutritional deficiencies by disproportionately impacting countries with the highest existing health burdens attributable to nutritional deficiencies. This projected increase in health inequalities is driven by differences in existing vulnerability of populations marginally above the threshold for deficiency and associated disease risk. Mitigation strategies concordant with the Paris Agreement, however, would be anticipated to substantially reduce the carbon-dioxide-induced rise in zinc and iron deficiency burdens, as compared to traditional public health measures, which may have a lesser effect even with optimistic degrees of population coverage and efficacy.
|
10.1371/journal.ppat.1000932 | Suppressing Glucose Transporter Gene Expression in Schistosomes Impairs Parasite Feeding and Decreases Survival in the Mammalian Host | Adult schistosomes live in the host's bloodstream where they import nutrients such as glucose across their body surface (the tegument). The parasite tegument is an unusual structure since it is enclosed not by the typical one but by two closely apposed lipid bilayers. Within the tegument two glucose importing proteins have been identified; these are schistosome glucose transporter (SGTP) 1 and 4. SGTP4 is present in the host interactive, apical tegumental membranes, while SGTP1 is found in the tegumental basal membrane (as well as in internal tissues). The SGTPs act by facilitated diffusion. To examine the importance of these proteins for the parasites, RNAi was employed to knock down expression of both SGTP genes in the schistosomula and adult worm life stages. Both qRT-PCR and western blotting analysis confirmed successful gene suppression. It was found that SGTP1 or SGTP4-suppressed parasites exhibit an impaired ability to import glucose compared to control worms. In addition, parasites with both SGTP1 and SGTP4 simultaneously suppressed showed a further reduction in capacity to import glucose compared to parasites with a single suppressed SGTP gene. Despite this debility, all suppressed parasites exhibited no phenotypic distinction compared to controls when cultured in rich medium. Following prolonged incubation in glucose-depleted medium however, significantly fewer SGTP-suppressed parasites survived. Finally, SGTP-suppressed parasites showed decreased viability in vivo following infection of experimental animals. These findings provide direct evidence for the importance of SGTP1 and SGTP4 for schistosomes in importing exogenous glucose and show that these proteins are important for normal parasite development in the mammalian host.
| Schistosomes are parasitic worms that live in the blood streams of ∼200 million people globally. They import glucose from host blood directly across their skin (the tegument). In the tegument the parasites possess glucose transporter proteins designated SGTP1 and SGTP4. SGTP4 sits on the outermost tegumental membranes while SGTP1 sits in the tegumental basal membrane (and on internal tissues). We have long hypothesized that SGTPs are involved in taking in glucose from the host but until the advent of advanced molecular technologies for use with schistosomes (notably RNA interference), we could not test this fundamental notion. In this work we employed RNAi to suppress expression of both SGTP genes in schistosomes. In support of our hypothesis, we find that SGTP1 or SGTP4-suppressed schistosomes do exhibit an impaired ability to import glucose compared to control worms and that this effect is compounded by suppression of both genes simultaneously. When suppressed parasites are cultured in glucose-depleted medium fewer of them survive. In addition, suppressed parasites showed decreased viability in experimental animals. These findings provide direct evidence of the importance of these tegumental transporters for schistosome feeding and show that these SGTPs are important for normal parasite development in the mammalian host.
| Schistosoma mansoni is a parasitic platyhelminth that causes the chronic, often debilitating disease, schistosomiasis affecting several hundred million people globally. Infection is initiated following skin penetration by larval parasites called cercariae which rapidly adapt to the intra-mammalian environment in a process called cercarial transformation. These transformed juvenile parasites are now called schistosomula and they move from the epidermal tissues into the blood stream where they mature. Adult worms reside in the mesenteric veins of their mammalian hosts, where they are generally found as male-female pairs.
The entire worm is surrounded by a continuous cytoplasmic unit, or syncytium, called the tegument. The host interactive surface of the tegument is unusual in that it consists of two tightly apposed, lipid bilayer membranes that are highly invaginated. The internal, basal membrane of the tegument consists of a normal (trilaminate) lipid bilayer containing many invaginations. This bilayer extends periodically beneath the underlying muscle to enclose areas called “cell bodies” (or cytons) which contain nuclei and protein synthetic machinery [1].
Adult worms use large quantities of host glucose; they are reported to consume the equivalent of their dry weight in glucose every 5 hours [2]. While the adults possess a functional gut, they have been shown to take up glucose directly across their external body surface by facilitated diffusion [3], [4]. Three glucose transporter mRNAs were originally identified from Schistosoma mansoni and these were designated schistosome glucose transporter protein (SGTP) 1, 2 and 4 [5]. Only SGTP1 and SGTP4 displayed glucose transport activity when expressed in Xenopus laevis oocytes. In the Xenopus uptake assay, both proteins functioned as typical facilitated diffusion glucose transporters, exhibiting glucose stereospecificity, relaxed specificity for other hexoses, sodium independence and marked inhibition by cytochalasin B [5].
Immunolocalization of SGTP1 and SGTP4 revealed that both of these proteins are localized in the tegument of schistosomula and adult worms [6]. SGTP4 appears to be localized uniquely to the tegument, while SGTP1 can also be detected within the body of the worm, particularly in muscle [6]. The presence of facilitated diffusion transporters in the tegument implies that schistosomes have the capacity to take up glucose by passive diffusion. Localization of the SGTPs by immuno-electron microscopy reveals that SGTP4 is present predominantly or exclusively within the apical membranes, while tegumental SGTP1 is found only within the basal membrane [7], [8]. This asymmetrical localization of the two SGTPs in the tegument suggests that the host interactive protein (SGTP4), acts to import sugar from the bloodstream into the tegument and that SGTP1 acts to transport some portion of this sugar to underlying tissues. The Km for glucose transport by the apical tegumental membrane transporter SGTP4 is greater than that of the basal transporter SGTP1 (in Xenopus oocytes) [5]. This should give an advantage to the basal membrane transporter to associate with any free glucose that is not utilized in the tegumental syncytium so that it can be moved more deeply into the body of the worm.
We have long hypothesized that the SGTPs function to transport exogenous glucose across the tegumental membranes and into body of the worms [9]. However, until the advent of RNAi methods for use with schistosomes, we could not effectively test this fundamental notion. In this work, we show that suppressing SGTP gene expression using RNAi does impair schistosome glucose uptake capabilities and can debilitate the parasites in vitro and in vivo.
The availability of a nearly complete draft of the S. mansoni genome [10] permits a careful bioinformatic analysis for facilitated glucose transport protein genes and this identifies a total of four SGTP genes. In addition to the three genes previously identified, another facilitated glucose transporter homolog can now be identified. The gene, which we designate SGTP3, is currently identified as hypothetical protein Smp_127200. Searches of dbEST reveal that ESTs exist for all four SGTP genes demonstrating that these genes are expressed in mammalian stage schistosomes. Because SGTP1 and SGTP4 are clearly demonstrated to be expressed in the adult tegument and appear to be the predominant facilitated glucose transporters in adult S. mansoni, we focused our RNAi studies on these two genes [6].
To determine whether SGTP1 and SGTP4 are amenable to gene silencing in schistosomula via the RNAi pathway, parasites were treated with two siRNAs spanning distinct positions for each target. All siRNAs were effective and showed comparable knockdown for each target (not shown). One of each target-specific siRNA was then selected for all subsequent experiments: SGTP1siRNA1 for SGTP1 and SGTP4siRNA1 for SGTP4. Parasites were electroporated with SGTP1siRNA1, SGTP4siRNA1, or a mix or both siRNAs. Control parasites were treated with an irrelevant siRNA or were not exposed to siRNA at all. Parasites were then cultured for 14 days in Basch medium before being harvested for gene expression analysis. Figure 1 shows that the transcript levels of both targets were substantially reduced when parasites were treated with each siRNA separately or in combination, compared to controls (Figure 1A). Gene knockdown is specific; siRNAs targeting SGTP1 have no effect on SGTP4 expression levels and vice versa. The reduction in transcript levels was more striking for SGTP4 (∼85%) than for SGTP1 (∼55%). Schistosomula treated with an siRNA targeting SGTP1 alone or with a mix of siRNAs targeting both SGTP1 and SGTP4 exhibited a similar decrease in SGTP1 gene expression. Likewise, parasites treated with an siRNA targeting SGTP4 alone or with a mix of siRNAs targeting both SGTPs exhibited a similar decrease in SGTP4 gene expression, as depicted in Figure 1.
To monitor SGTP gene suppression in adults, six week old worms were recovered from infected mice and electroporated with SGTP1 plus SGTP4 siRNAs. Transcript levels were measured 14 days after siRNA treatment by qRT-PCR and the results are shown in Figure 1B. About 70% suppression of SGTP1 and 90% of SGTP4 was observed in adult parasites following treatment, compared to control parasites electroporated with an irrelevant siRNA or parasites electroporated in the absence of siRNA. Western blotting analysis was undertaken in order to assess the impact of gene suppression on target protein levels. Figure 1C shows that the suppression of both targets resulted in a substantial diminution of SGTP1 (top panel) and SGTP4 (middle panel) protein levels in these parasites. In contrast, both proteins were easily detected in extracts of control parasites. In the bottom panel, the amino acid permease control protein SPRM1hc was detected in all extracts, demonstrating that comparable levels of protein were present in each lane.
The glucose uptake capacity of SGTP-suppressed schistosomula versus controls was compared. As shown in Figure 2, parasites treated with SGTP1 or SGTP4 siRNAs had a significant (P = 0.0005) and similar (∼50%) reduction in glucose uptake capacity compared to the control group. Parasites treated with a mix of both SGTP siRNAs showed an even more pronounced reduction in glucose uptake (to ∼70%) and this decrease was significantly different from the values obtained using single SGTP-suppressed parasites (P = 0.003). In the presence of cytochalasin B, a glucose transporter inhibitor, the intake of glucose by control parasites was decreased further (to ∼80%, P = 0.008) (Figure 2).
To determine whether SGTP suppression and the resulting decrease in glucose uptake capacity affected the phenotype of cultured schistosomula, parasite viability in culture was measured by Hoechst staining 14 days after siRNA treatment. Schistosomula were maintained either in complete RPMI (containing 10 mM glucose) or in glucose-depleted RPMI (containing 0.05 mM glucose). Figure 3 shows that suppressing the SGTP1 and SGTP4 genes did not significantly affect the viability of parasites kept in medium containing relatively high levels of glucose. However SGTP-suppressed parasites cultured in RPMI containing low glucose were significantly less viable (by >40%) than their control counterparts (P = 0.02). Parasites cultured in RPMI with no glucose do not survive beyond 48 hours. It is noteworthy that control parasites experience stress under low glucose conditions such that 62.4% ±5.6 of them remain viable after 14 days in culture (compared with 93.3% ±14.2 of control parasites cultured under high glucose conditions).
To investigate whether RNAi-mediated gene silencing of SGTP1 and SGTP4 affects parasite viability in vivo, we infected groups of 7–8 mice with 1 day old control or SGTP1+ 4-suppressed schistosomula. Figure 4 shows the number of worms recovered from these mice 28 days after infection. There was a significant reduction in worm burden in the SGTP-suppressed group compared to either control group.
SGTP gene expression analysis was undertaken on the worms recovered from the infected mice and the data were compared to the gene expression pattern of suppressed and control schistosomula that had been maintained in culture. SGTP-suppressed parasites cultured for 7 days (white bars, Figure 5) exhibited ∼65% suppression of SGTP1 (Figure 5A) and close to 100% suppression of SGTP4 (figure 5B). After 28 days cultured ex vivo (grey bars, Figure 5), mRNA levels in the SGTP dsRNA-treated worms were rising but remained substantially lower than control levels (∼50% for SGTP1 (Figure 5A) and ∼70% for SGTP4 (Figure 5B)). In contrast, after 28 days in vivo (black bars, Figure 5), SGTP-suppressed parasites recovered from infected mice were no longer suppressed; SGTP transcript levels had returned to normal, or above normal, levels.
Essentially the same results were observed when schistosomula were treated with long dsRNAs specific for SGTP1 and SGTP4 by soaking. The level of gene suppression using this metholology was comparable to that reported above for parasites exposed to SGTP-specific siRNA by electroporation. Parasites treated with SGTP-specific, and control, long dsRNA were used to infect mice and were recovered by perfusion 28 days later. As for the siRNA work, significantly fewer worms were recovered from the SGTP-suppressed group versus controls in this experiment using long dsRNA (not shown). Finally, as seen with siRNA-treated parasites, recovered parasites were no longer suppressed (data not shown).
In this work we show that two Schistosoma mansoni glucose transporter (SGTP) genes, SGTP1 and SGTP4, are susceptible to suppression via RNAi. Of the two SGTPs targeted we find that SGTP4 is consistently better suppressed than SGTP1 using different siRNAs, long dsRNA and at two different life stages tested. This is consistent with the notion that genes expressed in schistosome tissues that are in direct contact with the environment (e.g. the tegument or the gut) are more efficiently suppressed by RNA interference compared to genes expressed in other tissues. SGTP4 is predominantly and perhaps exclusively expressed in the tegument [6], [7] whereas SGTP1 is additionally expressed in the internal tissues of the parasite, notably in the muscle [6], [8]. In the past we have noted that genes expressed predominantly or exclusively in the tegument (e.g. SmAQP) can be potently suppressed while those expressed both in the tegument and in internal tissues (e.g. SPRM1hc) are more poorly suppressed using the same protocols [11], [12], [13]. This may reflect differences in the ability of dsRNAs to enter internal tissues or to the differential expression of RNAi pathway components in different organs.
The level of SGTP4 gene suppression in schistosomula is ∼80%. This is the case when parasites are treated with dsRNA targeting SGTP4 alone or when treated with two siRNAs targeting SGTP4 and SGTP1. In a similar manner, the level of suppression of SGTP1 remains essentially the same when SGTP1 alone is targeted for suppression or when SGTP1 and SGTP4 are both targeted. These results support previous work [14] showing that more than one gene can be suppressed at one time in schistosomes. Our quantitative data show that the RNAi machinery is not saturated by multiple siRNAs targeting different mRNAs.
The level of inhibition of glucose uptake into SGTP1-suppressed parasites is comparable to that seen for SGTP4-suppressed parasites. When SGTP1 alone is suppressed, glucose should still be able to enter the parasite tegument freely via the outer tegumental membrane transporter SGTP4. However, the movement of imported glucose further into the body of the SGTP1-suppressed parasites would then be impaired since this transporter is present on the tegumental basal membranes and on the membranes of other internal tissues. The inability of imported glucose to be efficiently transported out of the tegument and into the deeper tissues using SGTP1 would increase tegumental glucose concentrations and likely impede the further import of glucose by facilitated diffusion from the external environment. This is reflected in lower radiolabeled glucose being taken in to the SGTP1-suppressed parasites compared to controls.
When SGTP4 alone is suppressed, less glucose should enter the worms across the tegument compared to controls but any glucose that does enter and that is not utilized within the tegument should be efficiently transported inward via SGTP1. This would promote further glucose diffusion into the parasites via residual tegumental SGTP4 transporters. Parasites with both SGTP1 and SGTP4 genes suppressed exhibit a significantly greater impairment of radiolabeled glucose uptake compared with parasites that have had just one of the transporter genes suppressed. This likely reflects both a lower level of glucose uptake into the tegument via SGTP4 and an impaired ability to move that glucose into the internal tissues via SGTP1. Note that the level of glucose uptake in the doubly suppressed parasites is higher than that seen in parasites treated with a chemical inhibiter of facilitated glucose transporter protein function - cytochalasin B. This compound has been shown to block SGTP1 and SGTP4 function since it inhibits radiolabeled glucose uptake into Xenopus oocytes that are expressing SGTP1 or SGTP4 [5]. The double SGTP knockdown parasites exhibited a higher glucose uptake (of ∼30% versus untreated controls) compared to parasites treated with cytochalasin B (whose uptake was ∼20% of untreated control parasites). Likely this reflects the high potency of cytochalasin B in almost completely shutting down all SGTP function. In contrast, RNAi leads to SGTP gene knockdown (but not gene knockout) and the presence of residual functional SGTP protein in the siRNA-treated groups does permit some label uptake. Residual protein includes any protein generated before siRNA administration as well as new protein derived from transcripts that survive the RNAi treatment. The diminished ability of SGTP-suppressed schistosomes to import glucose unequivocally demonstrates that these parasites do use both SGTP1 and SGTP4 to efficiently take in sugar. In a similar vein, earlier work reported that glucose uptake is impaired in schistosomes following exposure to SGTP antisense oligonucleotides [15]. However, this work noted non-specific effects with some oligonucleotides and considerable variability between treatments, making the data equivocal [15]. Previous work has demonstrated that SGTP1 is important for glucose uptake from the environment in the sporocyst life stage [16].
In order to establish whether the inability to import glucose by the SGTP-suppressed parasites had a detrimental impact on the worms, their viability was compared with that of control parasites in vitro and in vivo. Parasites in culture whose glucose transporter genes are suppressed show no significant phenotypic differences compared with controls, when they are maintained in medium with a high glucose concentration (10 mM) for up to 14 days. However, when these parasites are instead cultured in low glucose medium (0.05 mM) for 14 days, significantly fewer suppressed parasites survive compared with controls. This suggests that, in the sugar-poor environment, an impaired ability to import glucose upsets parasite metabolism and decreases viability. When SGTP-suppressed parasites infect mice, fewer of them survive to adulthood relative to controls. This is the case despite the fact that glucose concentrations in blood are high (∼5 mM). These data suggest that the parasites' glucose demands in vivo are higher than in culture and this likely reflects the need for parasites in vivo to generate more energy (through glucose catabolism) to allow them migrate through tissues, invade the vasculature and combat host immune effectors.
The level of RNAi-mediated target gene suppression diminishes with time in culture. After 4 weeks in vitro the level of suppression of SGTP1 is ∼50% compared with ∼65% at day 7 post treatment. For SGTP4 the suppression level at week 4 in culture is 70% compared with >95% at day 7. These data demonstrate that the RNAi effect remains substantial even after a month in culture. In contrast, equivalent parasites recovered from infected mice 4 weeks after RNAi treatment exhibit no remaining SGTP gene suppression. Those parasites that have survived in vivo have SGTP mRNA levels at or even above control levels. Similar variable outcomes of RNAi in schistosomes ex vivo compared to in vivo have been reported in other studies [17], [18]. One hypothesis is that RNAi is variably effective in different parasites and/or that different individuals in the treated parasite population received different amounts of siRNA. Those in which SGTP knockdown is least effective, or that received less dsRNA, survive because the expression of their SGTP genes is minimally impaired. Another hypothesis is that worms in vivo are more metabolically robust and this leads to a shorter half life of the dsRNA and/or its downstream effectors. In mammalian cells the longevity of the RNAi effect can depend on cell type: in non-dividing cells suppression can persist for several weeks whereas in rapidly dividing cells the effect may last only from 3 to 7 days. [19]. Schistosomes in culture appear quiescent; they do not develop as quickly and fully as do parasites in infected animals and this may contribute to the persistence of gene suppression observed in the cultured worms.
In summary, this work shows that by demonstrably suppressing glucose transporter gene expression in schistosomes using RNAi, parasite feeding is hindered and this can significantly lower parasite viability. These findings provide direct evidence for the importance of SGTP1 and SGTP4 for schistosomes in importing exogenous glucose and show that the proteins are important for normal parasite development within the mammalian host.
Infection of mice with schistosome parasites was carried out following review and approval by the Institutional Animal Care and Use Committee of Tufts University or Instituto René Rachou - FIOCRUZ. The Tufts animal management program is accredited by the American Association for the Accreditation of Laboratory Animal Care, meets the National Institutes of Health standards as set forth in the “Guide for the Care and Use of Laboratory Animals” (National Academy Press, Washington DC, 1996), and accepts as mandatory the PHS “Policy on Humane Care and Use of Laboratory Animals by Awardee Institutions” and NIH “Principals for the Utilization and Care of Laboratory Animals Used in Testing, Research and Training”.
Biomphalaria glabrata snails infected with S. mansoni were obtained from Dr. Fred Lewis (Biomedical Research Institute, Rockville, MD). In some experiments parasites were obtained from snails infected at Instituto René Rachou - FIOCRUZ, Belo Horizonte, MG, Brazil. Schistosomula were prepared from cercariae released from infected snails and were cultured in Basch medium at 37°C, in an atmosphere of 5% CO2 as described [20]. Parasite viability was ≥90% at the beginning of each experiment as assessed by Hoechst staining [11]. In some experiments schistosomula were kept in complete RPMI medium which is RPMI supplemented with 10 mM Hepes, 2 mM glutamine, 5% fetal calf serum and antibiotics (100 U/ml penicillin and 100 µg/ml streptomycin). Adult worms were recovered by vascular perfusion from Balb/c mice that were infected with 125 cercariae, 6 weeks previously. Adult parasites were maintained in Basch medium for RNAi experiments.
Schistosomula and adult worms were treated either with synthetic siRNAs (IDT, Coralville, IA) or with long dsRNAs specific for SGTP1 or SGTP4 (GenBank accession numbers L25065 and L25067, respectively). The siRNAs were designed with the help of the RNAi Design Tool at http://www.idtdna.com/Scitools/Applications/RNAi/RNAi.aspx. The siRNAs targeting SGTP1 are SGTP1siRNA1: 5′-GGAGCATTCAGTTGTGGTTGGGTTG-3′spanning the coding sequence at positions 229–254 and SGTP1siRNA2: 5′-ACATAAAGAAGCTGAGGCACGTAAA-3′ spanning the coding sequence at positions 647–672. The siRNAs targeting SGTP4 are SGTP4siRNA1: 5′-GAAATAGCTCCCTTATCTCTTCGTG -3′, which cover positions 447-472 of the coding sequence and SGTP4siRNA2: 5′-GTGACACCAAGTTTCTTATATGCTC-3′ which cover positions 186-211 of the coding sequence. The negative control siRNA (5′-CTTCCTCTCTTTCTCTCCCTTGTGA-3′) is the “DS Scrambled Neg” obtained from IDT, Inc. This sequence does not match any in the S. mansoni genome. Target-specific siRNA delivery to the parasites was performed by electroporation as described previously, using 2.5 µg/50 µl (2.8 µM) of each siRNA for schistosomula and 5 µg/50 µl (5.6 µM) for adults [21], [22].
Long dsRNA was prepared as described previously [21]. The primer sequences for preparing long dsRNA targeting SGTP1 are SGTP1-T7, 5′-ggtaatacgactcactatagggCTAATCGGATACAATCT-3′ and SGTP1-T3, 5′-ggaattaaccctcactaaagggAATGAAATACGAGAAA-3′ which spans the coding sequence at positions 79–514. The lower case sequences represent T7 or T3 RNA polymerase promoter sequences. SGTP4 long dsRNA was prepared as described [17]. A non-schistosome derived long dsRNA used as an irrelevant control was generated from the yeast expression plasmid pPIC9K, as described earlier [17]. Long dsRNA was delivered to the parasites by soaking cultured schistosomula overnight with 50 µg/ml of irrelevant or SGTP-specific long dsRNA [21], [22]. Gene suppression was assessed post-treatment by comparing mRNA and protein levels in target versus control groups.
The levels of expression of SGTP1 and SGTP4 genes in schistosomula and adult worm pairs treated with gene-specific dsRNA were measured by quantitative real time PCR (qRT-PCR), using custom TaqMan gene expression systems from Applied Biosystems (Foster City, CA). The procedure, involving total RNA extraction and quantitative real time PCR, has been described [21]. The following primers and probe were selected to detect SGTP1: SGTP1 forward, 5′-CTGCAGCTTATTCACTGAGTCAATC- 3′; SGTP1 reverse, 5′-CCACCGATGTTTTTCTGTATAACAGGAT-3′ and SGTP1 probe, 5′-FAM- TCAATGGTTATCCAATCTAATTGT- 3′. To detect SGTP4 expression, the following primers and probe were used: SGTP4 forward 5′-AGCCAAGGAGTTAACTTATTATGCAATTTATTG 3′-; SGTP4 reverse, 5′- TCCAACAGATAATAACGATAACTAAAAATGGTAAGAA-3′ and SGTP4 probe, 5′-FAM- CAATGGCATCATTAATGC- 3′. Alpha tubulin was used as the endogenous control gene for relative quantification employing the ΔΔCt method [23]. Results were graphed as gene expression level relative to the group treated with control irrelevant dsRNA.
Parasite lysates were prepared by adding 50 µl of ice cold cell disruption buffer (PARIS Kit, Ambion, TX) followed by incubation for 30 minutes on ice. The protein content in each extract was estimated using the BCA Protein Assay Kit (Pierce, IL) according to the manufacturer's instructions. Soluble protein (5 µg in 20 µl SDS-PAGE sample buffer) was subjected to SDS-PAGE under reducing conditions, blotted onto PVDF membrane and blocked using detector block solution (KPL, Inc.) for 1 h at room temperature. The membrane was then probed overnight at 4°C with affinity purified rabbit anti-SGTP1 or anti-SGTP4 serum at 1∶500 [5] or antibody directed against a control schistosome protein (SPRM1hc) [13]. Bound primary antibody was detected using goat anti-rabbit IgG conjugated to horseradish peroxidase (Invitrogen, Inc.), diluted 1∶5000, followed by incubation with the chemiluminescent substrate LumiGLO (KPL, Inc.) and the membrane was exposed to X-ray film. The same membrane was probed three times to detect SGTP1, SGTP4 and the loading control protein, SPRM1hc. For each re-use, the membrane was first incubated for 30 min at room temperature with 2% SDS and 0.7% β-mercaptoethanol to strip bound antibody and was then washed in phosphate buffered saline twice for 30 min each.
To evaluate if SGTP1 and SGTP4 gene knockdown affected parasite survival in culture, viability was assessed by Hoechst staining [11]. Three day old schistosomula electroporated in the presence of a mix of SGTP1 and SGTP4 siRNAs at 2.5 µg each, were cultured in RPMI medium containing either 10 mM or 0.05 mM D-glucose for 14 days. Parasite mortality was determined in samples containing ∼100 parasites each, by adding 1 µg/ml Hoechst 33258 to the cultures at room temperature. After 10 min dead parasites were counted using a 460 nm reading filter. Viability in each group was calculated as the average value (+/− standard deviation) from triplicate experiments relative to controls treated with irrelevant siRNA.
Schistosomula treated with siRNA specific for SGTP1, SGTP4, or a mix of equal amounts of both siRNAs, were compared for their ability to take up glucose relative to control parasites treated with an irrelevant siRNA. Schistosomula, 14 days after siRNA treatment, were washed four times in wash medium (RPMI without glucose and supplemented with 10 mM Hepes, 2 mM glutamine, and antibiotics (100 U/ml penicillin and 100 µg/ml streptomycin)) and resuspended in 30 µl of wash medium supplemented with 0.1 M D-glucose. Each sample received 1 µl of [1,2-3H]2-deoxyglucose at 1 µCi/ml (Amersham, Piscataway, NJ) followed by a 30 min incubation at room temperature. Parasites were subsequently washed four times in wash medium before being disrupted in 30 µl of 2% SDS. The parasite lysate was added to 1 ml Scintiverse scintillation fluid (Fischer) and subjected to liquid scintillation counting. Radiolabel uptake was calculated per 1000 schistosomula. Assays were performed in triplicate for each group and averaged for analysis. Glucose uptake in control parasites was also measured in the presence of cytochalasin B (40 µM) which was added to the parasite culture for 30 min prior to the start of the uptake experiment.
One day old cultured schistosomula were electroporated with SGTP, or control, or no, siRNA and the groups were divided into three samples each. The first sample was immediately used to infect female BALB/c mice (∼1,000 parasites/mouse) and the other two samples were kept in culture for 7 days or for 28 days to determine the efficiency of gene knockdown (at day 7) and to monitor long-term suppression in vitro (at day 28). Mice were infected by injecting schistosomula in 100 µl of RPMI without phenol red into the thigh muscle of the animals using a 1 ml tuberculin syringe and a 25G-1 needle. Twenty eight days later, the mice were euthanized and adult worms recovered by portal vein perfusion. Recovered worms were counted, examined under a light microscope and subsequently their SGTP gene expression levels were determined, as described above. The same procedure was followed in experiments in which schistosomula were exposed to long dsRNA by soaking, except that mice were infected 24 h after parasite exposure to long dsRNA.
All data were analyzed using GraphPad Prism 4 software. One Way ANOVA was used to compare median values among three or more groups. Student's t-tests were used to compare the means between a target group and a control group and p values close to or less than 0.05 were considered significant.
The following GenBank accession numbers apply to the DNAs targeted in this work: SGTP1: L25065 and SGTP4: L25067.
|
10.1371/journal.pntd.0006738 | Phylogenomic analysis unravels evolution of yellow fever virus within hosts | The yellow fever virus (YFV) recently reemerged in the large outbreaks in Africa and Brazil, and the first imported patients into Asia have recalled the concerns of YFV evolution. Here we show phylogenomics of YFV with serial clinical samples of the 2016 YFV infections. Phylogenetics exhibited that the 2016 strains were close to Angola 1971 strains and only three amino acid changes presented new to other lineages. Deep sequencing of viral genomes discovered 101 intrahost single nucleotide variations (iSNVs) and 234 single nucleotide polymorphisms (SNPs). Analysis of iSNV distribution and mutated allele frequency revealed that the coding regions were under purifying selection. Comparison of the evolutionary rates estimated by iSNV and SNP showed that the intrahost rate was ~2.25 times higher than the epidemic rate, and both rates were higher than the long-term YFV substitution rate, as expected. In addition, the result also hinted that short viremia duration of YFV might further hinder the evolution of YFV.
| The first importation of infections into China in 2016 and the following outbreaks in Africa and Brazil of yellow fever virus (YFV) have raised again the concerns of the potential viral spread into new territories. In this study, we aimed to know the evolution dynamics of YFV by using intrahost phylogenomics and to assess the risk of virus epidemics. Through deep sequencing of consecutive samples from 12 patients, we identified hundreds of genomic variations (iSNVs and SNPs), and noticed the nearly linear accumulation of variations within individuals. The estimated evolutionary rate within host is much higher than the epidemic evolutionary rate. In comparison with Dengue virus (DENV) and Zika virus (ZIKV), which share similar host vectors (Aedes spp.), life cycles, mutation rates and replication strategies to YFV, the lower epidemic evolutionary rate of YFV might have been hindered by the shorter viremia duration, which decreased the accumulated variations to get into the transmission cycle.
| Yellow fever is a notorious mosquito-borne viral disease emerged during the 15th-19th centuries in the Americas, Africa and Europe, causing severe hemorrhagic fever and liver injury with high mortality rates. Although control of mosquitoes and the use of the live-attenuated yellow fever virus (YFV)-17D vaccine strain have effectively prevented and controlled the epidemics, YFV is estimated to cause approximately 30,000 deaths out of 200,000 infections annually worldwide, mostly in Africa (https://www.cdc.gov, last accessed 10th April, 2018). The etiological agent of the disease, YFV, is a single-stranded, positive-sense RNA virus with a ~11 kb genome in length. The virus is frequently transmitted between nonhuman primates and mosquitoes in African and American jungles, known as the sylvatic cycle [1]. Occasionally, YFV can escape from the sylvatic cycle to infect humans, with subsequent transmission between mosquitoes and humans, forming the urban cycle. Historically, YFV was believed to have originated from Central or East Africa, and transmitted to America during the slave trade [2, 3]. Currently, YFV in Africa and America are classified into seven genotypes, with two in South America, two in West Africa, and three in East and Central Africa [2, 4, 5].
In January 2016, the Ministry of Health of Angola notified the World Health Organization (WHO) of a yellow fever disease outbreak: as of October 28th, 4,347 suspected cases, including 377 deaths, were reported from all 18 provinces of Angola. Outbreaks were also reported simultaneously from the Democratic Republic of Congo and Uganda [6]. In February 2018, 464 confirmed human cases of yellow fever have also been reported in Brazil, with 154 deaths (http://www.who.int, last accessed 10th April, 2018). Meanwhile, epizootics have expanded to areas previously not considered at risk for yellow fever (http://www.who.int/csr/don/27-february-2018-yellow-fever-brazil/en/, last accessed on 10th April, 2018). All these recalled the concerns of the evolution of YFV. During the Angolan outbreak, YFV infections were detected in Chinese workers returning to China from Angola [7–9], marking the first time YFV infections were documented in Asia. By using the consecutive samples from those imported YFV patients, we unraveled the intrahost and epidemic evolutionary dynamics of YFV.
We collected samples from twelve out of thirteen YFV patients in China, sequenced the viral genomes, and performed analyses on phylogenetics, sequence comparison, and intra-host dynamics. Of the 12 patients, two had severe disease, in which one survived (YF-BJ3) and one died (YF-BJ1), and the remaining patients displayed mild symptoms (Table 1). We tested blood samples of these 12 patients from the first day of admittance to the hospital by real-time reverse transcription PCR (RT-PCR), and four patients (YF-BJ1, YF-BJ2, YF-BJ3, and YF-BJ5) were positive. YFV RNA fragments could be detected in serum until 9, 12, 10 and 6 days, respectively, after the onset of symptoms (Table 1, and Fig 1). Urine samples from all patients were also tested by real-time RT-PCR and all were YFV-positive. Higher viral RNA loads were observed in the urine compared to the blood samples, except for those from the non-survivor (Table 1). The virus persisted in urine samples for at least 15 days after the onset of symptoms in survivors. Specifically, the urine sample from patient YF-BJ3 was still PCR-positive when tested 32 days after the onset of symptoms (Table 1 and Fig 1). We then sequenced all available YFV-positive samples by using total RNA sequencing, amplicon sequencing and/or Sanger sequencing (Table 1 and S1 Table). In total, nearly complete virus genomes (>10,222 bp) from 9 patients, and deep-sequenced genome datasets (average sequence depth 34,978x) from 3 patients covering 12 time points were obtained (Table 1).
We performed phylogenetic analysis of the coding region of YFVs (S2 Table), and found that YFV sequences from the returning workers in 2016 outbreak closely clustered with the 1971 Angola strains (Fig 2A). Of note, the vaccine strain YF-17D and its derivatives are located far from the Angola strains on the phylogenetic tree. A closer inspection of the Angola strains shows that the viruses in the 2016 outbreak are likely from a single origin and genome sequences from both severely ill patients are closely clustered (Fig 2B). There are 188 nucleotide substitutions between the 1971 and 2016 consensus sequences and 6 of them are nonsynonymous substitutions. Comparison of the YFV polypeptides shows that only three amino acid changes that appear to be specific to the Angola 2016 strains (Fig 2C). Two are in the capsid protein and one in NS5, the RNA-dependent RNA polymerase. The above data are consistent with previous studies using partial or whole genomes showing that YFV exhibits a slow evolutionary rate [2, 4, 10].
For the deep-sequenced samples, we identified the iSNVs by using methods in a previous study for Ebola virus [11]. The mean sequencing depth of genomic regions is between 16,558x and 37,822x (Fig 3A), and a total of 101 iSNV sites were discovered, including 69 in the coding region and 32 in non-coding regions. In each iSNV sites, we only found two types of nucleotides. Taken that sequencing errors are generated randomly and may result in multi-nucleotide heterogeneity in a single site, it is unlikely those iSNVs were the results of sequencing bias. Fewer iSNVs appeared in the 1st and 2nd codon positions than the 3rd codon position (Fig 3B) and fewer non-synonymous than synonymous iSNVs (Fig 3C), implying that the coding region is generally under purifying selection. We plotted the distributions of all mutated allele frequencies of the iSNVs (Fig 3D and 3E). The mean mutated allele frequency of non-coding iSNVs was at 0.18, while that of synonymous and non-synonymous iSNVs was 0.12 and 0.09, respectively (Fig 3D). Among them, non-synonymous iSNV is significantly lower than that in non-coding regions (P = 0.02). The distribution of non-coding iSNVs is close to the expected neutrality, whereas curves of synonymous and nonsynonymous iSNVs have higher portions of iSNVs in the area of low mutated allele frequencies (Fig 3E). This further supports the notion that the coding region was under purifying selection. Additionally, we discovered two variant types at the 3’ untranslated region (UTR), one with 5 concurrent iSNVs (G10360A, U10365G, C10367U, G10373A, and U10398C) and the other having an additional iSNV (A10425G) (Fig 3F). Phasing analysis reveals that these iSNVs tend to be concurrent in the same reads (S3 Table and S1 Fig). Predicted RNA structure shows that these substitutions are likely to affect the structure of the 3’ UTR (Fig 3F and S2 Fig), and probably influence viral replication in hosts [12–14].
Subsequently, we placed all of the polymorphic sites (SNPs and iSNVs) along viral genome (Fig 4A). SNPs and iSNVs described the viral difference in two levels of between and within the host, respectively. Within 65 days (from the onset of the first patient to the sample of the last patient in this study) we observed a total of 234 SNPs in 148 sites in 18 samples, including 102 synonymous and 117 nonsynonymous SNPs. A total of 138 SNPs that appear only once are scattered along the viral genome. In particular, the SNPs of 5 sites (T900A, A2352C, C3918A, G6463A, and A7320G) could be used to characterize the 2016 YFVs. All five SNPs were detected in seven patients, four SNPs (without A7320G) were detected in one patient (YF-BJ5), and only one patient (YF-BJ3) does not possess any of the above SNPs (Fig 4A). Of note, YF-BJ1 is likely a co-infection with more than one YFV variant, as the virus genome on day 7 does not contain any of the five SNPs, whereas all five SNPs became dominant on day 8 (Fig 4A). Of the five SNPs, only G6463A is nonsynonymous, resulting in an amino acid substitution from Val to Ile in NS4A.
We then compared the accumulation of polymorphic sites within patients over time (Fig 4B). Each sample was set to days after onset of the symptoms of the patients it belongs to. Linear regressions revealed that iSNVs and SNPs accumulated from the day of onset of symptoms at a rate of 0.72 and 0.32 variations/day, respectively. The accumulation rate of iSNV may reflect the intrahost evolution. Intriguingly, the YFV evolutionary rate within humans is roughly 2.48 x10-2 variations/site/year (0.72 variations/day = 0.72 x 365 / 10591 variations/site/year = 2.48 x10-2 variations/site/year) in our study. Of those synonymous variations, the intrahost evolutionary rate estimated by iSNV was 0.7 x10-2 variations/site/year; whereas the rate of nonsynonymous variations was 1.2 x10-2 variations/site/year. Compared to the rates estimated by SNP within these samples, the intrahost evolutionary rate of non-synonymous sites (by iSNV) is higher, while the rate of synonymous sites is slightly higher than the SNP rate.
The evolution of YFV is a basic question to unfold to predict the risk of the pathogens. Phylogenetic analysis suggests that the 2016 outbreak is likely the spillover of YFV viruses in Angola. The high identity between the 2016 strain and the vaccine strain YF-17D would explain the effectiveness of the current vaccines [15]. Although there is a diversity of yellow fever virus genotypes, all genotypes are indistinguishable in serological assays. There is not any evidence that the Angola outbreak was due to immune escape and it is accepted among public health professionals that low vaccine coverage in the area prior to the outbreak was the driving factor and the resulting aggressive mass vaccination campaign was vital for containing and eliminating the outbreak. However, it should be noted that the Angola lineage is clearly diversified from the 17D-lineage (Fig 2A). Moreover, we did notice 6 patients who had been vaccinated before [16], but the limited clinical and epidemiological data cannot tell the immune escape. Future immune escape from current YFV vaccines cannot be fully ruled out, although YFVs evolve very slowly [17]. A novel vaccine based on a Central/East African YFV isolate would be necessary as a complement vaccine to prepare against future outbreaks. Owing to the slow evolutionary rate and the past experience of using YF-17D, the novel vaccine is likely to work for another decades.
Virus evolution experienced mutation, selection within and between hosts, genetic drift, and transmission, and finally formed the genetic variants (lineages) over long-term evolutionary timescales [18]. Deep sequencing of viral genomes can provide novel insights into viral evolutionary dynamics [19, 20]. In this intra-host phylogenomic analysis, we depicted the evolutionary process of YFV from intra-host to epidemic substitutions, which were reflected by iSNV to SNP, respectively. Other than using the metric for long-term evolutionary rate, we applied the substitution/site/day to measure the intrahost evolutionary rate, which appeared more appropriate in outbreak or epidemic scenarios. Short-term rates may provide better insight into the intrahost YFV evolutionary trends (e.g., iSNV dynamics over time) relative to clinical symptomology, severity, and/or co-morbidities. However, further in vitro or in vivo studies would be needed. The estimated evolutionary rates clearly showed that the intrahost evolutionary rate is ~2.25 times higher than that of the epidemic evolutionary rate, and both are much higher (59 times and 26 times) than the substitution rate of YFV along longer evolutionary timescale [10]. The higher rate estimated by iSNV reflected purifying selection occurred, when intrahost variations were spreading out. The rates estimated by synonymous iSNV and SNP, respectively, were similar, implying that the synonymous variations were under similar evolutionary constraints during transmission. Meanwhile, only a few of the nonsynonymous variations were spread out in different individuals, and only G6463A (Val to Ile in NS4A) were detected in multiple samples.
Compard to Dengue virus (DENV) and Zika virus (ZIKV), YFV shares similar host vectors (Aedes spp.) and life cycles [21, 22], and has similar mutation rates and replication strategies [10, 23, 24]. However, the long-term substitution rates of DENV (0.77–0.99 x10-3 substitutions/site/year) and ZIKA (0.98–1.06 x10-3 substitutions/site/year) are higher than that of YFV [23, 25, 26]. Previous studies have posed that additional constrained selection forces are probably existed [10, 27]. Based on our results, we noticed the nearly linear accumulation of variations within individuals. Since the transmission of YFV variants needs mosquitoes as vector, the window (viremia duration) that mosquitoes could use is essential, as only the variations within this period could be brought into the transmission cycle. A wider window size would possibly allow more non-synonymous variations into the transmission cycle (Cristina Domingo, EID, 2018). Given that YFV, DENV and ZIKV have ~3, 3–7, and 4–7 days of viremia, respectively, after symptom onset (S4 Table), it hinted that the shorter viremia duration in YFV infections might result in the lower evolutionary rate than that of DENV or ZIKV. Most Flaviviruses are vector-borne viruses and the evolutionary rates in mammals and arthropods are different. The intrahost analysis of virus could show the short-term evolution dynamics, especially from serial samples, and this would be a helpful complement to the long-term evolution of Flaviviruses.
In addition, deep sequencing and viral genomics have provided new insights into viral evolutionary dynamics, especially in a continuous mode. Our findings strongly suggest that the urine samples should be considered for the clinical diagnosis, genomic surveillance, and large-scale screening of YFV infections. Further investigations are also encouraged to characterize the factors that maintain viremia in vector-borne diseases.
The study was conducted in accordance with the Declaration of Helsinki and approved by the ethics committee of Beijing Ditan Hospital (DSRB 2008/00293 and DSRB 2013/00209). Clinical samples and information were obtained after written informed consents from all participants. All human subjects were adult, and if not, whether a parent or guardian of any child participant provided informed consent on their behalf.
Twelve YFV patients returning from Angola were recruited (Table 1). We sampled urine and blood sample of these patients from routine examination and collected their clinical data.
YFV infection was diagnosed according to the WHO guidelines (http://www.who.int/csr/disease/yellowfev/case-definition/en/, last accessed 10th April, 2018), including clinical symptoms (fever, headache, jaundice, etc.) and laboratory detection of virus by using real-time RT PCR. The serum and urine samples were collected and stored at -80 oC for further analyses, including real-time RT-PCR, PCR sequencing and RNA sequencing.
RNA from both serum and urine samples were deployed to two approaches of deep sequencing. Approach I: Total RNA sequencing. RNA was purified by Trizol (Thermo Scientific) for each sample, and then total RNA libraries were constructed with 500bps insertion size using the NEBNext UltraTM RNA Library Prep Kit (NEB, MA, USA), and then sequenced by Illumina Hi-seq platform, and generated 2 × 125 bp paired-end reads. Approach II: RNA amplicon sequencing. Two sets of YFV-specific primer pairs were designed (Set A: 5 pairs, amplicon length = 2,248 ± 168 nt; Set B: 12 pairs, amplicon length = 1,018 ± 48 nt; S5 Table). PCR products were sequenced by Illumina platform with 2 × 250 bp paired-end reads. Quality control and error correction were implemented according to previous report [28, 29]. The probably low quality regions in sequencing, including 1) Q<30, and 2) the first 10 bps after primers were removed from the high quality sequences, according to a previous study [28]. To minimize sequencing error that affects the accuracy of iSNV calling, we only kept the read-pair that has >100 high quality bases in both ends as clean reads. All reads were deposited in the NCBI SRA database under the accession no. SRP096859.
Consensus sequences were aligned by MUSCLE [30]. Phylogenetic analyses were performed by using RAxML v8.1.6, with the GTR model of nucleotide substitution and γ-distributed rates among sites. Phylogenetic tree was constructed by maximum-likelihood using YFV genome sequences [31]. A NJ tree based on YFV genome sequences was also constructed to show the results were robust (S3 Fig). A total of 1,000 bootstrap replicates were performed. All SNPs were listed in S6 Table. The genome sequences of YF-BJ1, YF-BJ2, YF-BJ3, YF-BJ4, YF-BJ5, YF-FZ2, YF-FZ4, YF-FZ6, and YF-FZ7 were deposited in the NCBI GenBank database under the accession no. MH633684-MH633692.
To minimize the potential sequencing errors generated by system, we performed the following two steps before we called iSNV, including 1) mapping the clean reads to the reference genome, 2) qualification of the samples. For mapping the clean reads to genome, we use the traditional protocol used widely. First, we mapped the clean reads with pair-ended aligned to the assembled genome from YF-BJ1/7D by using Bowtie2 v2.2.5 [32] with default parameters. SAMtools v1.2 [33] was used to generate ‘mpileup’ files with no limit for the maximum site depth. To find the potential mutation in each site, we do not use the following pipeline in SAMtools. Instead, we developed homemade PERL scripts (available at http://github.com/generality/iSNV-calling/), which were used for iSNV calling to identify all potential mutations by using the mpileup files as input. The depth of bases in each sites were used to measure the mutation rate of iSNV. Then, we selected the samples with more than 3,000 sites with a sequencing depth ≥300× as candidate samples for iSNV calling. Using this protocol, we called iSNV in 15 samples in this study (S1 Table), and obtained the frequencies of each allele in each genome site. Interestingly, we only found the heterogeneous nucleotides with two types in each sites. Thus, we defined the nucleotide with higher frequencies as the major allele, and fewer one as the minor allele in the following iSNV calling.
The iSNV calling was according to the site depth and strand bias with following: 1) remove the ambiguous iSNV and keep the following iSNVs: (1.1) minor allele frequency of ≥5%, a conservative cutoff based on an error rate estimation described before [11, 28]; (1.2) depth of the minor allele of ≥15; and (1.3) strand bias of the minor allele less than tenfold. 2) In our analysis, these two types of nucleotides in each site have contained the nucleotide in the reference genome. Mutated allele frequencies were calculated by the rate of the reads numbers for mutated alleles to the references and total reads. Minor allele and Major allele frequencies were calculated by the majority nucleotides and minor nucleotides in each site respectively. 3) The effect of iSNV to the gene was calculated by the mutation whether caused the amino acid changes.
The iSNVs that co-occurred in same viral haplotype(s) within quasispecies was defined as phased iSNVs. 1)We screened the iSNV sites along genome with a window of 250 bp, and then identified all the windows which probably containing a phasing iSNVs. 2) All the sequencing reads aligned to YFV genome region harboring these iSNVs were extracted. 3) The stretches of nearest iSNV were determined by the mutations in the reads aligned to the genome region with the same phasing (supporting reads > = 2). 4) The fractions of reads supporting phased and non-phased neighbor iSNVs were respectively calculated.
RNA secondary structures were predicted by using RNAfold program of ViennaRNA Package 2 with default settings [34]. Nucleotide sequences started at the beginning of 3' UTRs of Angola 2016 strain. The length of window for secondary structure prediction was equal to 70 nucleotides. We also extended the additional 200 nucleotides at the end of the sequence, with a step length equal to 50.
All reads from next generation sequencing were deposited in the NCBI SRA database under the accession no. SRP096859. The genome sequences of YF-BJ1, YF-BJ2, YF-BJ3, YF-BJ4, YF-BJ5, YF-FZ2, YF-FZ4, YF-FZ6, and YF-FZ7 were deposited in the NCBI GenBank database under the accession no. MH633684-MH633692.
|
10.1371/journal.pgen.1003104 | Deciphering the Transcriptional-Regulatory Network of Flocculation in Schizosaccharomyces pombe | In the fission yeast Schizosaccharomyces pombe, the transcriptional-regulatory network that governs flocculation remains poorly understood. Here, we systematically screened an array of transcription factor deletion and overexpression strains for flocculation and performed microarray expression profiling and ChIP–chip analysis to identify the flocculin target genes. We identified five transcription factors that displayed novel roles in the activation or inhibition of flocculation (Rfl1, Adn2, Adn3, Sre2, and Yox1), in addition to the previously-known Mbx2, Cbf11, and Cbf12 regulators. Overexpression of mbx2+ and deletion of rfl1+ resulted in strong flocculation and transcriptional upregulation of gsf2+/pfl1+ and several other putative flocculin genes (pfl2+–pfl9+). Overexpression of the pfl+ genes singly was sufficient to trigger flocculation, and enhanced flocculation was observed in several combinations of double pfl+ overexpression. Among the pfl1+ genes, only loss of gsf2+ abrogated the flocculent phenotype of all the transcription factor mutants and prevented flocculation when cells were grown in inducing medium containing glycerol and ethanol as the carbon source, thereby indicating that Gsf2 is the dominant flocculin. In contrast, the mild flocculation of adn2+ or adn3+ overexpression was likely mediated by the transcriptional activation of cell wall–remodeling genes including gas2+, psu1+, and SPAC4H3.03c. We also discovered that Mbx2 and Cbf12 displayed transcriptional autoregulation, and Rfl1 repressed gsf2+ expression in an inhibitory feed-forward loop involving mbx2+. These results reveal that flocculation in S. pombe is regulated by a complex network of multiple transcription factors and target genes encoding flocculins and cell wall–remodeling enzymes. Moreover, comparisons between the flocculation transcriptional-regulatory networks of Saccharomyces cerevisiae and S. pombe indicate substantial rewiring of transcription factors and cis-regulatory sequences.
| Flocculation is a process that involves yeast cells adhering to one another to form clumps called flocs. This trait is important for industrial yeast applications as it provides a cost-effective and efficient method to remove yeast cells. The adherence between cells occurs by the binding of glycoproteins known as flocculins and carbohydrate molecules located on the cell surface. To better understand how flocculation works, the genes that encode for flocculins and the transcription factors that regulate their expression need to be identified. In the fission yeast S. pombe, many of the flocculins and transcription factors that function in flocculation are not known. To address this gap in knowledge, we have employed molecular genetics and functional genomic approaches to uncover transcription factors and their target genes that play a role in flocculation. We discover that flocculation in S. pombe is regulated by a complex network of transcription factors that activate and repress themselves, as well as multiple target genes that encode for flocculins and cell wall–remodeling enzymes. The comparison of the flocculation regulatory networks between fission and budding yeasts indicates that they mainly differ in the types of transcription factors and their binding sequences.
| Flocculation is an inherent characteristic of yeasts involving asexual aggregation of cells into flocs that separate rapidly from the medium (reviewed recently in [1], [2]). Individual yeast cells transition into this morphological state as an adaptation to various environmental stresses by shielding the inner cells of the flocs [3]. The flocculent trait has also proven highly beneficial in industrial yeast applications by allowing efficient and cost-effective removal of cells [4]. The ability of yeast strains to flocculate is dependent on the expression of specific cell surface glycoproteins known as flocculins. Cell-to-cell adhesion occurs via binding between the flocculin and surface carbohydrates in a calcium-dependent manner [5]. The bound carbohydrates consist of various sugars including mannose, glucose, and galactose that are specific to the type of flocculin and yeast species [6]–[8]. There has been considerable interest in elucidating the genetic control of flocculation to better understand this phenomenon and generate biotechnological advances in yeast-based industries.
In Saccharomyces cerevisiae, a transcriptional-regulatory network composed of interactions between transcription factors and their flocculin gene targets is central in controlling flocculation. The primary flocculins that function in flocculation are encoded by the FLO1, FLO5, FLO9, and FLO10 genes [9]–[11]. Overexpression of the individual FLO genes is sufficient to trigger flocculation [8], [12]. However, the degree of flocculation by FLO overexpression varies from FLO1 to FLO10 exhibiting the strongest to weakest flocculation, respectively. The flocculin FLO11 also exhibits weak flocculation when overexpressed [8], but its function is mainly in cell-to-surface adhesion [13], diploid pseudohyphal growth [14], and haploid invasive growth [15]. The transcription factors required for flocculation include Flo8p and Mss11p, which primarily activate FLO1 transcription [16]. The Sacc. cerevisiae laboratory strain S288C containing a nonfunctional FLO8 gene is not able to flocculate, but flocculation is restored in this strain by the overexpression of FLO8 or MSS11 [16], [17]. In addition, Sfl1p has been shown to inhibit transcription of FLO1 in the W303-1A strain and not in S288C, likely through interactions with the Ssn6p-Tup1p global repressor and components of Mediator [18], [19].
The control of flocculation is much less known in Schizosaccharomyces pombe. The ability of the heterothallic wild-type strains 972 h− and 975 h+ to flocculate has not been observed presumably because the inducing environmental conditions have not been identified. Phenotypic analysis of constitutive flocculent mutant strains show that flocculation is dependent on the presence of calcium, but unlike Sacc. cerevisiae, the flocculin-carbohydrate interactions involve galactose rather than mannose and glucose residues [7]. Moreover, the transcriptional-regulatory network governing flocculation in S. pombe remains poorly characterized. Only a single interaction between the Mbx2 MADS box transcription factor and the gsf2+ flocculin gene is currently known [20], [21]. The gsf2+ gene was initially identified as highly upregulated in response to heterologous expression of FLO8 [20]. Overexpression of gsf2+ is sufficient to trigger flocculation while its deletion abrogates the flocculent phenotype of tup12Δ, lkh1Δ, and gsf1 mutants. In addition, gsf2+ displays additional roles in cell-to-surface adhesion and invasive growth [20]. The induction of gsf2+ during flocculation and invasive growth is mediated by Mbx2 [21]. Two other transcription factors implicated in flocculation have been reported. The CSL transcription factors Cbf11 and Cbf12 play opposing roles in flocculation where mutant strains lacking cbf11+ or overexpressing cbf12+ flocculate [22]. The direct targets of these transcription factors functioning in flocculation have not been identified, but could be several putative flocculin genes that show protein sequence homology to other yeast-related proteins [23]. Indeed, these putative flocculin genes, as well as gsf2+ are transcriptionally upregulated in certain Mediator mutants that flocculate indicating that these genes are likely repressed by Mediator [24]. Similar to Sacc. cerevisiae, the global transcriptional regulators Tup11 and Tup12 function in flocculation but their influence on the expression of these flocculin genes has not been addressed [25]. Importantly, it has not been directly demonstrated that these putative flocculin genes in S. pombe actually play a role in flocculation and the identity of the transcription factors that regulate them remains unknown.
In this study, we have initiated an extensive characterization of the transcriptional-regulatory network of S. pombe flocculation by identifying the relevant transcription factors and their flocculin gene targets. Importantly, we have also determined that heterothallic wild-type S. pombe is able to flocculate when grown in rich medium containing ethanol and glycerol as a carbon source. A screen of transcription factor deletion and overexpression strains for flocculent phenotypes revealed five novel transcriptional regulators of flocculation (Rfl1, Adn2, Adn3, Sre2, Yox1) in addition to our independent finding of Mbx2, Cbf11, and Cbf12. The strongest flocculation was observed upon overexpression of mbx2+ and deletion of rfl1+ (SPBC15D4.02) which encodes an uncharacterized fungal Zn(2)-Cys(6) transcription factor. Microarray expression profiling of the mbx2OE and rfl1Δ strains revealed good overlap in the upregulation of several flocculin genes, while ChIP-chip analysis of HA-tagged Mbx2 and Rfl1 under control of the nmt41 promoter indicated that these transcription factors bound to some of the flocculin gene promoters. Nine flocculin gene targets (pfl1+–pfl9+) including gsf2+/pfl1+ were identified. The single overexpression of these genes triggered flocculation to varying degrees and cumulative effects on flocculation were observed in double overexpression experiments. Only loss of gsf2+ could abrogate the flocculent phenotype of all the transcription factor mutants indicating that gsf2+ encodes the dominant flocculin in S. pombe. Interestingly, we discovered that certain cell wall-remodeling enzymes can also function in flocculation, and some of these genes are likely regulated by the LisH transcription factors Adn2 and Adn3. In addition to the identification of target genes within the transcriptional-regulatory network, autoregulatory and inhibitory feed-forward loops involving several transcription factors were also detected. These results provide a significant insight into the transcriptional control of flocculation in S. pombe.
Our understanding of the transcriptional-regulatory network that governs flocculation in S. pombe remains limited. To further decipher this network, we sought to systematically identify transcription factors that play a role in flocculation. A list of 101 genes encoding sequence–specific transcription factors containing a bona-fide DNA-binding domain was assembled from [26] and GeneDB [27]. From this gene list, we constructed 101 nmt1-driven overexpression strains and 92 nonessential deletions in which the entire ORF was replaced with the KanMX6/NatMX6 cassette. A detailed description of the construction and phenotypic characterization of this transcription factor mutant collection will be described elsewhere (unpublished data). The transcription factor array of overexpression and deletion strains were screened for flocculation in EMM lacking thiamine and YES media, respectively. We recovered a total of eight transcription factors in which four overexpression strains (mbx2OE, adn2OE, adn3OE and cbf12OE) and four deletions (rfl1Δ, sre2Δ, yox1Δ and cbf11Δ) exhibited flocculation. These transcription factors represent positive and negative regulators of flocculation, respectively. Among these transcription factors, only the overexpression of cbf12+ and mbx2+ and deletion of cbf11+ have been reported to cause flocculation [20], [22].
The strongest flocculation was observed in the mbx2OE and rfl1Δ strains. The flocs of the rfl1Δ strain in YES medium were larger and sedimented faster than the flocs produced in the mbx2OE strain after 48 hour induction (Figure 1A). The mbx2+ gene encodes a MADS-box transcription factor which was originally isolated in a screen for genes functioning in the biosynthesis of cell surface pyruvated galactose residues [28]. Recently, Mbx2 has been shown to function in flocculation and invasive growth by regulating the flocculin gene gsf2+ [20], [21]. The rfl1+ (repressor of flocculation) gene encodes an uncharacterized fungal Zn(2)-Cys(6) transcription factor.
The flocculation exhibited by these overexpression and deletion transcription factor mutants recovered from our screens could be abolished with the addition of galactose, but not mannose or glucose (data not shown). The amount of galactose required to completely deflocculate cells depended on the degree of flocculation. For example, mbx2OE strain could be deflocculated with 2% galactose while rfl1Δ strain required 5–10 times more galactose to completely deflocculate. Reflocculation of these strains was achieved in CaCl2 or in YES medium (data not shown).
The growth conditions that trigger flocculation in heterothallic wild-type S. pombe are not well known. To identify the inducing conditions, 972 h− and 975 h+ cells were tested on different carbon sources at different cell densities for flocculation. We determined that heterothallic wild-type cells were able to flocculate when cultured for five days at an initial concentration of 1×106 cells/ml in medium containing 1% yeast extract, 3% glycerol and, 4% ethanol (referred to as flocculation-inducing medium, Figure 1B). The degree of flocculation was slightly enhanced in strains auxotrophic for leucine, uracil, and/or adenine indicating that nutrient status may also play a role in triggering flocculation (data not shown). However, these wild-type strains flocculated significantly less in flocculation-inducing medium than the mbx2OE and rfl1Δ mutants in EMM and YES media, respectively. The weaker flocculation in these strains was more easily observed in petri-dishes incubated on an orbital rotator than in test tubes. In contrast to wild type, deletion of mbx2+ did not produce any visible flocs in the flocculation-inducing medium (Figure 1B).
Fungal genes that function in flocculation are usually associated with filamentous invasive growth [17], [20]. We hypothesized that the rfl1Δ strain would exhibit hyperfilamentous invasive growth because of its strong flocculent phenotype. Indeed, the amount of cells resistant to removal from the agar by washing in the invasive assay on LNB medium with an underlayer of YE+ALU was much greater in the rfl1Δ strain than in wild type (Figure 1C). Under the microscope, the filamentous growth like those detected by Dodgson et al. [29] was observed below the agar surface for both wild type and rfl1Δ strain with the latter showing much larger and more frequent formation of filamentous growth (data not shown). Similarly, adn2+ and adn3+ which were previously observed to have defects in invasive growth when deleted were recovered in our screens as flocculent when overexpressed [29].
The strongest flocculation observed in the mbx2OE and rfl1Δ strains indicated that these two genes encode the major regulators of flocculation. Therefore, we initially focused on the characterization of these two transcription factors and proceeded to identify their target genes involved in flocculation. The nmt41-driven mbx2-HA strain was subjected to microarray expression profiling with a custom-designed S. pombe 8×15 K Agilent expression microarray (Table S2). The intermediate strength nmt41 promoter was sufficient for mbx2OE flocculation and was utilized in the microarray experiments in order to reduce possible secondary transcriptional effects compared to the strong nmt1 promoter. To better distinguish the direct target genes, ChIP-chip was also carried out concurrently on the same strain using the S. pombe 4×44 K Agilent Genome ChIP-on-chip microarray (Table S3). For the rfl1+ expression profiling and ChIP-chip experiments, the flocculent deletion mutant and nmt41-driven rfl1-HA strain were used, respectively (Tables S4 and S5). The highly-induced putative target genes identified by microarray expression profiling of these transcription factor mutant strains were validated by qPCR (Table S13).
The list of genes that were induced at least two fold in the mbx2OE or rfl1Δ strain was subjected to gene ontology analysis using the Princeton GO Term Finder (http://go.princeton.edu/cgi-bin/GOTermFinder). These induced genes were highly enriched in cell wall components with p-values of 9.0e-9 and 6.3e-6 for the mbx2OE and rfl1Δ strains, respectively. Strikingly, the most-induced genes in the mbx2OE strain encoded cell surface glycoproteins. The cell surface glycoprotein genes up-regulated above two-fold were SPAC186.01, gsf2+, SPAC977.07c/SPBC1348.08c, SPCC188.09c, fta5+, SPBC947.04, SPBC359.04c, SPBC1289.15, SPAPB2C8.01, SPAC1F8.02c, SPAPB18E9.04c, SPCC553.10, and SPBPJ4664.02, which all but gsf2+ and the last 4 genes were predicted to be pombe adhesins based on BLAST sequence analysis (Figure 2A; [23]). SPAC977.07c and SPBC1348.08c are gene duplications with 100% sequence identity. To our knowledge, these genes with the exception of gsf2+ have not been characterized further. The induction of these genes in the mbx2OE strain ranged from 2 to 112-fold relative to the empty vector control (Figure 2A, Table S13). In addition, several genes (agn2+, psu1+, SPAC4H3.03c and gas2+) encoding cell wall-remodeling enzymes such as glucan glucosidases and a betaglucanosyltransferase were induced up to 91-fold compared to the empty vector control when mbx2+ was overexpressed (Figure 2A). In the rfl1Δ expression data, a similar set of cell surface glycoprotein genes were upregulated at a comparable level as the mbx2OE expression data except for SPAC1F8.02, SPBC359.04c, SPAPB18E9.04c and SPBPJ4664.02 (Figure 2A, Table S13). In contrast to the mbx2OE strain, the same genes encoding the cell wall-remodeling enzymes were not highly upregulated in the rfl1Δ strain (Figure 2A).
Of the thirteen highly-induced cell surface glycoprotein genes in the mbx2OE expression data, nine of them were detected with ChIP-chip indicating that these genes are very likely the direct transcriptional targets of Mbx2 (Figure 2A). Four of the nine highly-induced cell surface glycoprotein genes in the rfl1Δ strain were detected with ChIP-chip confirming that these genes are probably direct transcriptional targets of Rfl1 (Figure 2A). For both Mbx2 and Rfl1, gsf2+, fta5+ and SPAPB2C8.01 were detected in the expression microarray and ChIP-chip experiments (Figure 2A).
Next, we sought further evidence that these cell surface glycoprotein genes were targets of Mbx2 and Rfl1 by epistasis studies. We decided to study a subset of these genes, which included the majority of the gene sequences analyzed by Linder and Gustafsson [23], [24]. The mbx2+ gene was overexpressed in single deletions of these putative target genes and their degree of flocculation was determined visually in petri-dishes, as well as quantitatively (Table S14). The putative glycoprotein gene SPAPB15E9.01c was included in these studies, because even though the transcript was downregulated in both mbx2OE and rfl1Δ strains, ChIP-chip analysis detected Mbx2 and Rfl1 association with its promoter (Figure 2A). Deletion of gsf2+ decreased mbx2OE flocculation to the greatest extent while the reduction of flocculation was less extensive in the other single deletion mutants (Figure 2B, Table S14). The degree of reduction in mbx2OE flocculation roughly corresponded to the pfl numbers, which were assigned based on the degree of flocculation when overexpressed (see below). Moreover, mbx2OE flocculation was completely abrogated in the gsf2Δ pfl9Δ double mutant indicating that the reduction of mbx2OE flocculation in these mutants were additive in some cases (Figure 2B). Similar experiments were performed for rfl1+ in which flocculation was assayed in the same putative target deletions in the rfl1Δ background. The flocculation exhibited in the rfl1Δ strain was completely abolished by the deletion of gsf2+, but not by the deletion of pfl9+ (Figure 2C).
To further analyze the expression microarray datasets of Mbx2 and Rfl1, the promoter regions of the differentially-expressed genes were subjected to the motif-finding algorithms RankMotif++ and MEME to identify their binding specificities [30], [31]. Mbx2 is a member of the MEF2-MADS box transcription factor family which has been shown to bind to the consensus sequence 5′-(C/T)TA(T/A)4TA(G/A)-3′ [28], [32], [33]. The Mbx2 binding specificity obtained by RankMotif++ closely resembled this known consensus sequence (Figure 2D). Similarly, RankMotif++ generated an Rfl1 binding specificity that resembled known consensus sequences of several members of the fungal Zn(2)-Cys(6) transcription factor family (Figure 2E). The binding specificity of Zn(2)-Cys(6) DNA-binding domains is composed of conserved GC-rich trinucleotides spaced by a variable sequence region differing in length among members of the transcription factor family [34]. Analyses of the Mbx2 and Rfl1 expression microarray and ChIP-chip datasets by MEME did not generate any candidate DNA motifs.
Altogether, these results demonstrate that Mbx2 and Rfl1 are transcription factors responsible for regulation of flocculation in fission yeast by activating or repressing the transcription of candidate S. pombe flocculin genes, respectively.
Besides gsf2+, the other putative target genes of Mbx2 and Rfl1 that encode for cell surface glycoproteins share some amino acid sequence homology with domains found in other fungal adhesins [23]. However, the role of these glycoprotein genes in flocculation has not been demonstrated. Overexpression studies were employed to the aforementioned set of putative flocculin target genes of Mbx2 and Rfl1 to determine whether they function directly in flocculation. Each single overexpression of these flocculin genes was able to induce flocculation to varying degrees with the strongest flocculation observed in the gsf2OE strain which produced visible flocs within one day (Figure 3A; Table S14). Weaker flocculation was observed from the overexpression of the other flocculin genes after total incubation of 2–7 days in EMM minus thiamine medium with sub-culturing into fresh medium in Day 3. The flocculation images of these overexpression strains shown in Figure 3A were captured after total of 7 days of induction. As a result of these observations, we named these genes pfl+ for Pombe Flocculins and numbered them according to their degree of flocculation when overexpressed: pfl1+/gsf2+ (referred as gsf2+ hereafter), pfl2+/SPAPB15E9.01c, pfl3+/SPBC947.04, pfl4+/SPCC188.09c, pfl5+/SPBC1289.15, pfl6+/SPAC977.07c, pfl7+/SPBC359.04c, pfl8+/fta5+ (referred as fta5+ hereafter) and pfl9+/SPAC186.01. Furthermore, we overexpressed some double combinations of the weaker flocculin genes to determine whether flocculation could be additive. Indeed, the pfl4+ pfl9+, pfl6+ pfl9+, and fta5+ pfl9+ double overexpression strains flocculated earlier and formed larger flocs than their corresponding single overexpressors, thus, demonstrating the additive effect of these flocculins (Figure 3B, Table S14). We next tested the single deletions of the pfl+ genes for their ability to flocculate in flocculation-inducing medium. No visible flocculation was observed in the gsf2Δ strain while wild type was flocculent (Figure 1B). In contrast, flocculation still occurred in the pfl2Δ–pfl9Δ strains in the inducing medium indicating that gsf2+ encodes the dominant flocculin and the other flocculin genes are dispensable for flocculation (data not shown).
These observations revealed that the contribution in flocculation by these pfl+ genes varied and certain combinations of pfl+ were additive. The strength of flocculation by the single overexpression of pfl+ genes was directly correlated with the reduction of mbx2OE flocculation in the corresponding deletion strains (Figure 2B and Figure 3A, Table S14). For example, the pfl2OE strain which produced larger flocs than the pfl3OE–pfl9OE strains exhibited a greater inhibition of mbx2OE flocculation when deleted. Similarly, the flocculation of the rfl1Δ strain was completely abrogated by the deletion of gsf2+, but not at all by the deletion of pfl9+ (Figure 2C). Consistent with the above results, the deletion of both gsf2+ and pfl9+ led to a greater abrogation of mbx2OE flocculation compared to each deletion alone (Figure 2B). In summary, we have demonstrated that these pfl+ genes encode for S. pombe flocculins and Gsf2 is the dominant flocculin.
Interestingly, ChIP-chip analysis also detected binding of Mbx2 and Rfl1 to their own promoters, as well as Rfl1 binding to the mbx2+ promoter (Figure 2A), indicating autoregulation and mbx2+ regulation by Rfl1 within the transcriptional-regulatory network of S. pombe flocculation. Mbx2 also appeared to be associated with the rfl1+ promoter, but this interaction was marginal as it was found just above the detection threshold for ChIP-chip (Figure 2A). To investigate the autoregulation of mbx2+, the gene was C-terminal tagged with GFP at its native locus (mbx2-GFP). However, the GFP-tagged strain resulted in a hypermorphic allele that displayed constitutive flocculation and nuclear localization of Mbx2-GFP (see below). We speculated that the removal of the 3′-untranslated region of mbx2+ during the C-terminal tagging may be the cause of the hypermorphic allele. To bypass this potential problem, we created an N-terminal GFP-tagged allele (GFP-mbx2) with an intact 5′-untranslated region and approximately 1 kb of native promoter sequence. In contrast to the C-terminal tagged hypermorphic allele, the N-terminal tagged GFP-Mbx2 expression was comparable to background levels and the strain did not exhibit constitutive flocculation (Figure 4A). Moreover, the GFP-mbx2 strain flocculated when grown in glycerol-inducing medium indicating that the tagged protein is functional (Table S14). When nmt1-driven mbx2+ expression was induced for 9 hours in the GFP-mbx2 strain, nuclear GFP-Mbx2 expression was detected, indicating that Mbx2 can activate its own expression (Figure 4A). As expected, this strain was now flocculent. Longer induction of nmt1-driven mbx2+ expression resulted in greater GFP-Mbx2 expression with multi-nucleated GFP foci (data not shown). The positive autoregulation of mbx2+ is likely to be direct as several putative MEF2-binding sequences (e.g. 5′-TTAAAAATAG-3′) are located within 1000 bp upstream from the mbx2+ start codon (data not shown).
To determine whether negative autoregulation occurs with rfl1+, a C-terminal GFP-tagged strain under native control was generated (rfl1-GFP). The localization of Rfl1-GFP was nuclear in the rfl1-GFP strain (Figure 4B). The induction of nmt1-driven rfl1+ expression for 18 hours in the rfl1-GFP strain led to a reduced nuclear Rfl1-GFP signal and a slightly increased cytoplasmic Rfl1-GFP signal (Figure 4B). However, overall Rfl1-GFP expression in the cell was reduced when Rfl1 was overexpressed compared to the empty vector control (Figure 4B; two-tailed t-test; p value<0.01). In contrast to our observations with the Rfl1-GFP protein expression, we found that there was no decrease of the Rfl1-GFP transcript when rfl1+ was overexpressed (Table S13). These results indicate that although Rfl1 can bind to its own promoter, negative autoregulation appears marginal or may not be occurring.
The observation that Rfl1 is associated with the mbx2+ promoter by ChIP-chip suggests that Rfl1 may oppose Mbx2 function in flocculation by repressing its expression. To test this hypothesis, we first examined the genetic interactions between mbx2+ and rfl1+. The mbx2Δ rfl1Δ double mutant did not display flocculation indicating that mbx2+ is epistatic to rfl1+ (Figure 5A). In addition, the flocculation associated with mbx2OE was abrogated by co-overexpression of rfl1+ (Figure 5A). These results are consistent with mbx2+ being downstream of rfl1+ and that rfl1+ opposes mbx2+ function in flocculation.
We next utilized the C-terminal and N-terminal GFP-tagged mbx2+ strains to further determine if Rfl1 represses mbx2+ expression. First, Rfl1 was overexpressed in the hypermorphic C-terminal tagged mbx2-GFP allele which shows constitutive nuclear Mbx2-GFP expression and flocculation. This resulted in the near-abolishment of both the GFP signal (Figure 5B) and flocculation (data not shown) in the hypermorphic mbx2 allele. Second, when the N-terminal tagged GFP-mbx2 strain was crossed into the rfl1Δ background, the resulting strain displayed dramatic increase in nuclear GFP-Mbx2 expression (Figure 5C) and flocculation strength equivalent to the rfl1Δ strain (data not shown). These results support the hypothesis that mbx2+ expression is repressed by Rfl1 in non-flocculent cells.
Cbf12, a member of the CSL transcription factor family has previously been reported to trigger flocculation when overexpressed [22]. However, the target genes of Cbf12 that function in flocculation have not been identified. To further elucidate the role of cbf12+ in flocculation, we took a similar approach to identify its direct target genes by concurrent expression microarray profiling and ChIP-chip analysis of the nmt41-driven cbf12-HA strain (Tables S6 and S7, respectively).
When cbf12+ was deleted and cultured in flocculation-inducing medium, flocculation was abolished (Figure 6A). In contrast, overexpression of cbf12+ by the nmt1 promoter triggered flocculation (Figure 6C) and produced a bowling pin–shaped phenotype after 24 hours in medium lacking thiamine (data not shown). Further induction of the nmt1-driven cbf12+ caused the strain to become sick and granulated, eventually leading to growth arrest (data not shown). To reduce the toxic effects of cbf12+ overexpression, an nmt41-driven cbf12-HA strain was used for concurrent expression profiling and ChIP-chip analysis.
Gene ontology analysis was carried out separately on the top 50 most highly-induced genes and all 160 promoter-occupied genes by Cbf12 with the Princeton GO Term Finder. Functional enrichment of genes in cell surface (p = 1.8e-7) and plasma membrane (p = 5.7e-4) was detected for the highly-induced and promoter-occupied genes, respectively. These genes included several flocculin genes, (Figure 6B). Both gsf2+ and pfl7+ were among the five highest induced genes (18.1 and 27.6-fold, respectively) in the cbf12OE strain and were also detected by ChIP-chip (Figure 6B) suggesting that Cbf12 directly activates the transcription of gsf2+ and pfl7+ for flocculation. The flocculation triggered by cbf12+ overexpression was completely abrogated in the gsf2Δ background, whereas deletion of pfl7+ had little effect (Figure 6C, Table S14). This was consistent with the hypothesis that gsf2+ encodes the dominant flocculin. In addition, loss of gsf2+ or pfl7+ did not alter the bowling-pin cell shape or the reduced fitness phenotypes of the cbf12OE strain indicating that these two phenotypes were not due to the upregulation of the flocculin genes (data not shown). The much weaker flocculation observed in the cbf12OE strain in comparison to the mbx2OE and gsf2OE strains may be attributed to additional defects in cell and nuclear division, which would cause early growth arrest before the full flocculation potential could be reached [22].
Consistent with previous findings, C-terminal GFP-tagged Cbf12 under native control was expressed predominantly in the nucleus in stationary phase cells while expression in logarithmic cells was comparable to background (Figure 6D; [22]). Compared to logarithmic growth in rich medium, Cbf12-GFP nuclear expression increased in cells grown in flocculation-inducing medium, thus supporting its role in flocculation (Figure 6D). Interestingly, Cbf12 was also detected by ChIP-chip to bind to its own promoter (Figure 6B). Indeed, positive autoregulation appears to occur as native Cbf12-GFP expression increased greater than three-fold when nmt1-driven cbf12+ was ectopically expressed in logarithmically growing cells (Figure 6E).
Recently, it was demonstrated that an N-terminal-truncated Cbf12 bound to probes containing a canonical CSL binding motif (5′-GTGGGAA-3′) by gel mobility shift assay [35]. We next searched for a similar DNA binding sequence for Cbf12 from the expression microarray and ChIP-chip cbf12OE datasets by RankMotif++ and MEME. RankMotif++ and MEME analyses of the expression microarray and ChIP-chip data, respectively, did not identify a binding specificity for Cbf12. However, when the promoters of up-regulated genes in the cbf12OE strain belonging to the cell surface GO category were subjected to MEME analysis, a motif closely matching the canonical CSL binding motif (6/7 nucleotide match) was recovered (Figure 6F).
These results demonstrate Cbf12 as part of the transcriptional-regulatory network of fission yeast flocculation by controlling the transcription of several flocculin genes including gsf2+.
From our transcription factor screens, the deletion of yox1+, sre2+, or cbf11+ also resulted in flocculation, although the size of the flocs were smaller than observed in mbx2OE, cbf12OE and rfl1Δ strains (Figure 7A, Table S14). Yox1 has been implicated in a negative autoregulatory loop to prevent inappropriate transcriptional expression of MBF gene targets, while the function of Sre2, which shows homology to the human sterol regulatory element binding protein SREBP-1A remains largely unknown [36], [37]. A role of Yox1 and Sre2 in flocculation has not been reported. In contrast, cbf11+ encodes a CSL transcription factor that plays a role in flocculation, but its target genes are not known [22].
To elucidate the transcriptional flocculation program of yox1+, sre2+ and cbf11+, expression microarray profiling was conducted on the corresponding flocculent deletion strains in rich medium (Tables S8, S9, S10). The expression microarray profiles of yox1Δ and sre2Δ most resembled each other compared to the other strains described in this study (Figure 7B). Genes upregulated by at least two-fold in the yox1Δ and sre2Δ strains showed enrichment for ribosomal subunits (p = 2.8e-31 and 7.4e-25 for yox1Δ and sre2Δ, respectively) and mitochondrial membrane transporters (p = 7.5e-5 and 1.2e-3 for yox1Δ and sre2Δ, respectively). These findings did not intuitively answer our questions as to how these two transcription factors might be related or associated with the flocculation pathway. We next examined whether any of the flocculin genes and their putative regulators were induced in the yox1Δ and sre2Δ strains. In the sre2Δ strain, gsf2+, pfl3+ and fta5+ transcripts were upregulated 3.7, 2.5 and 3.1-fold, respectively, indicating that the expression of these genes could be contributing to the flocculent phenotype (Figure 7C). In contrast, mbx2+ and cbf12+ transcripts were downregulated approximately 2-fold suggesting that the elevated levels of gsf2+, pfl3+ and fta5+ transcripts in the sre2Δ strain were not mediated by Mbx2 and Cbf12 (Figure 7C). Similarly in the yox1Δ strain, we observed that gsf2+ and pfl3+ transcripts were upregulated although less than in the sre2Δ strain, and mbx2+ and cbf12+ were also downregulated (Figure 7C). Therefore, this suggests that sre2+ and yox1+ may be involved in the repression of flocculation through a pathway independent from mbx2+ and cbf12+.
The microarray expression profile of the cbf11Δ strain revealed greater than 2-fold increase of gsf2+ and pfl3+ transcripts and a 60-fold increase of the SPAC1F8.02c transcript suggesting that these two flocculin genes and this uncharacterized glycoprotein gene may be responsible for the flocculent phenotype in this mutant. In contrast to the yox1Δ and sre2Δ mutants, mbx2+ did not show differential expression in the cbf11Δ strain compared to wild type. However, the cbf12+ transcript was upregulated 1.8-fold in the cbf11Δ strain. This suggests that cbf11+ may regulate flocculation through cbf12+, in agreement with previous reports of the antagonistic functions of cbf11+ and cbf12+ in this process [22].
We next determined whether the flocculation caused by the deletion of yox1+, sre2+ or cbf11+ was also dependent on gsf2+. The absence of gsf2+ was sufficient to abolish the flocculation in yox1Δ, sre2Δ, and cbf11Δ strains, even though gsf2+ was not always the most highly-expressed flocculin gene (Figure 7A and 7C, Table S14). Taken together, these results suggest that the expression of the dominant flocculin Gsf2 is responsible for the bulk of flocculation observed in yox1Δ, sre2Δ and cbf11Δ strains.
The transcription factor genes adn2+ and adn3+ are orthologous to Sacc. cerevisiae FLO8 (http://www.pombase.org/) and exhibit defects in invasive growth and cell-to-surface adhesion when deleted during nitrogen starvation [29]. From our screens, we discovered that the overexpression of adn2+ and adn3+ triggered minor flocculation while loss of adn2+ and adn3+ prevented flocculation in flocculation-inducing medium (Figure 8A and 8B, respectively). The flocculent phenotype of adn2OE and adn3OE strains was disrupted by the addition of galactose (data not shown). To identify the target genes of Adn2 and Adn3 that are involved in flocculation, expression microarray profiling was performed on nmt1-driven adn2OE and adn3OE strains (Tables S11 and S12). Surprisingly, gsf2+ transcript levels were relatively unchanged and the majority of pfl+ genes were downregulated in both overexpression strains (Figure 8C). Consistent with these results were the observations that mbx2+ and cbf12+ transcripts were downregulated greater than 2-fold in both adn2OE and adn3OE strains, whereas rfl1+ transcript levels were not differentially regulated (Figure 8C). Therefore, it appeared that the flocculent phenotype of adn2+ and adn3+ overexpression could not be attributed to the pfl+ genes identified in this study. These results led us to consider that perhaps the expression of other genes besides these encoding for flocculins could be responsible for triggering flocculation in adn2OE and adn3OE strains.
Interestingly, some of the aforementioned cell wall-remodeling enzymes (gas2+, psu1+ and SPAC4H3.03c) were also highly upregulated in both adn2OE and adn3OE strains (Figure 8C, Table S13). For example, gas2+ and SPAC4H3.03c were the highest induced genes in the adn2OE strain (17.9 and 36.8-fold, respectively) and also appeared within the top 20 most induced genes in the adn3OE strain. These genes were also induced in the mbx2OE strain except for psu1+ (Figure 2A). Overexpression analysis was subsequently carried out to determine if these genes possessed some role in flocculation. Although agn2+ was not upregulated in the adn2OE and adn3OE strains, it was included in the overexpression analysis because it was the second most induced gene (91-fold), as well as detected by ChIP-chip in the mbx2OE strain. Indeed, the single overexpression of these four genes resulted in flocculation after 5-days (including 3rd day sub-culturing into fresh medium) in medium lacking thiamine, implicating the involvement of these cell wall-remodeling enzymes in flocculation (Figure 8D, Table S14). Since deletion of adn2+ and adn3+ results in defects of invasive growth and cell-to-surface adhesion in response to nitrogen starvation, we wanted to determine if the single overexpression of gas2+, agn2+, psu1+ and SPAC4H3.03c could cause enhancement of these processes. We discovered that the single overexpression of these four cell wall-remodeling genes increased cell-to-surface adhesion, but not invasive growth relative to wild type under the nitrogen-deprivation condition (Figure S1). Because gsf2+ encodes the dominant flocculin, we also investigated whether the flocculation caused by adn2+ and adn3+ overexpression was dependent on gsf2+. Deletion of gsf2+ completely abrogated the flocculation in adn2OE and adn3OE strains (Figure 8A, Table S14).
In addition, the adn2OE and adn3OE strains exhibited cell separation defects such as the formation of multisepta and forkhead phenotypes (Figure 8E). The cell separation defect was more severe when adn3+ was overexpressed. We next determined whether the putative target genes involved in the flocculation of adn2OE and adn3OE strains also played a role in the multisepta phenotype. Overexpression of adn2+ and adn3+ in the gsf2Δ background did not alter the multisepta phenotype (Figure 8E), while the overexpression of gas2+, SPAC4H3.03c, psu1+ and agn2+ did not lead to formation of multisepta (data not shown). These results suggest that Adn2 and Adn3 may regulate cell separation and flocculation independently through different sets of target genes. Our microarray expression data suggests that Adn2 and Adn3 may control cell separation through ace2+, which encodes a major transcriptional activator of this process (Alonso-Nuñez et al., 2005). Overexpression of adn2+ and adn3+ resulted in the down-regulation of ace2+ and many of its known target genes such as adg1+, adg2+, adg3+, cfh4+, agn1+, eng1+, and mid2+ by 1.5 to 3.4-fold (Figure 8C).
In summary, the regulation of flocculation by adn2+ and adn3+ is likely mediated by the induction of genes encoding the cell wall-remodeling enzymes Gas2, SPAC3H3.03c, and Psu1. The regulation of these genes is independent from Mbx2 because mbx2+ was downregulated in the adn2OE and adn3OE strains. Although gsf2+ transcript level was not significantly upregulated by adn2+ and adn3+ overexpression, it was sufficient to abrogate the flocculation when deleted. However, it is possible that other cell surface glycoprotein genes not investigated in this study but were upregulated may also play a significant role in the flocculation function of adn2+ and adn3+.
In this study, we have deciphered a significant portion of the transcriptional-regulatory network governing flocculation in S. pombe. To date, few transcription factors and their target genes that function in flocculation have been identified. The MADS box transcription factor Mbx2 positively regulates flocculation by induction of the flocculin gene gsf2+, while the CSL transcription factors Cbf11 and Cbf12 repress and activate flocculation, respectively, but their target genes are not known [21], [22]. We have substantially expanded our limited knowledge of the flocculation transcriptional-regulatory network by the identification of several novel transcriptional activators (Adn2 and Adn3) and repressors (Rfl1, Yox1 and Sre1), and their putative target genes that function in flocculation. In addition, novel target genes of Mbx2, Cbf11 and Cbf12 were identified. The putative target genes of the transcription factors implicated in flocculation encode for several cell surface glycoproteins (gsf2+ and pfl2+–pfl9+) and cell wall-remodeling enzymes (agn2+, psu1+, SPAC4H3.03c and gas2+). These target genes were sufficient to trigger flocculation when overexpressed. Moreover, instances of regulation between transcription factors (Rfl1 repression of mbx2+), as well as positive (mbx2+ and cbf12+) autoregulation were detected within the flocculation network.
Mbx2 and Rfl1 appeared to be the major positive and negative regulators of flocculation, respectively, based on the largest flocs observed in the mbx2OE and rfl1Δ strains compared to the other flocculent mutants in this study. Our initial efforts to identify the target genes of Mbx2 and Rfl1 revealed several putative flocculin genes that were strikingly upregulated in the mbx2OE and rfl1Δ flocculent mutants. Previously, Gsf2 was the only S. pombe flocculin demonstrated to be directly involved in flocculation, and its transcription was influenced by the activity of Mbx2 [20], [21]. Similar to these studies, we also found that overexpression of gsf2+ triggers flocculation while loss of gsf2+ abrogates the flocculent phenotype of several mutants including mbx2OE. Here, we identified an additional eight flocculin genes (pfl2+–pfl9+) as putative target genes of Mbx2. Seven of these target genes (pfl3+–pfl9+) were reported to contain tandem repeats found in fungal adhesins, while pfl2+ is a sequence orphan predicted to encode a GPI-anchored protein [23], [27]. Seven pfl+ genes (gsf2+/pfl1+, pfl3+, pfl4+ and pfl6+–pfl9+) have been reported to be upregulated in loss-of-function flocculent mutants of Cdk8 module genes (cdk8+/srb10+, med12+/srb8+) suggesting that the transcriptional repression of these putative flocculin genes may be controlled by Mediator [24]. The transcriptional repression of flocculin genes by Mediator may not be direct, but could be through mbx2+ since its expression is highly upregulated in the cdk8 kinase-mutant and med12Δ strain (9 and 13-fold increase, respectively, within top 11 up-regulated genes, found in supplementary data [24]). This proposed role of Mediator appears conserved in Sacc. cerevisiae as FLO genes are similarly upregulated in cdk8 mutants [38]. Despite these observations, no direct evidence has been shown aside from the gsf2+ study by Matsuzawa et al. [20] that the pfl+ gene products are actually flocculins. We have shown that this is indeed the case as single and double overexpression of the pfl+ genes is sufficient to trigger flocculation and that this flocculation is galactose-specific.
The degree of flocculation triggered by single overexpression of the pfl+ genes varied, with gsf2+ and pfl9+ producing the largest and smallest flocs, respectively (the pfl numbers correspond roughly to the degree of flocculation upon overexpression). This result indicates that gsf2+ encodes the most dominant flocculin compared to the other pfl+ genes. In agreement are the observations that only deletion of gsf2+ and not the other pfl+ genes prevented flocculation in flocculation-inducing medium, and reduced the constitutive flocculent phenotype to the greatest extent of all the transcription factor mutants examined in this study. Moreover, the strength of the flocculins was directly correlated with the amount of reduction in mbx2OE flocculation observed in the various pfl deletion backgrounds (Figure 2B). These observations are similar in Sacc. cerevisiae, where overexpression of FLO1 produces the strongest flocculation compared to FLO5, FLO9, FLO10 and FLO11 [8],[12]. Furthermore, the flocculation mediated by pfl+ genes was additive as observed in our double deletion and co-overexpression experiments (Figure 2B and Figure 3B). These results suggest that the varying strengths of flocculation exhibited by S. pombe strains could be attributed to the upregulation of different combinations of pfl+ genes.
We identified Rfl1, an uncharacterized Zn(2)-Cys(6) transcription factor as a novel repressor of flocculation in fission yeast. The repression of flocculation by Rfl1 appears to be primarily mediated by the inhibition of gsf2+ expression since loss of gsf2+ can abrogate the constitutive flocculent phenotype of the rfl1Δ mutant. Rfl1 represses gsf2+ either directly by association with its promoter or indirectly by inhibition of mbx2+ transcription, thereby forming an inhibitory feed-forward loop (coherent type 2) within the transcriptional-regulatory network (Figure 9). These results indicate that Mbx2 and Rfl1 are opposing transcription factors, and the latter inhibits mbx2+ and gsf2+ expression under non-inducing conditions of flocculation.
Aside from its role in flocculation, Rfl1 may have a role in regulating genes involved in carbohydrate metabolism such as glycolysis and gluconeogenesis. Rfl1 appeared to be associated with promoters of genes enriched in glucose catabolic and metabolic processes (p-values = 0.00092 and 0.00269, respectively) including adh1+, hxk2+, pfk1+, tpi1+, adh4+, pgi1+, gpd3+, tdh1+, pgk1+, fba1+, eno101+, pyr1+, SPCC794.01c (predicted glucose-6-phosphate 1-dehydrogenase), SPBC2G5.05 (predicted transketolase) and SPBC660.16 (phosphogluconate dehydrogenase). Most of these genes with the exception of fba1+, eno101+, SPCC794.01, and SPBC660.16 were upregulated 1.2 to 26-fold in the rfl1Δ strain (Table S4). From these data, we speculate that Rfl1 could serve as a negative transcriptional regulator of several enzymes involved in the glycolysis and gluconeogenesis. Because flocculation and invasive growth are associated with nutritional limitation, Rfl1 may coordinate the expression of genes involved in flocculation and carbohydrate metabolism in fission yeast.
Previously, the CSL proteins Cbf11 and Cbf12 were shown to exhibit antagonistic roles in flocculation [39]. Overexpression of cbf12+ or loss of cbf11+ triggers flocculation. However, none of their target genes have been identified. We present supportive evidence that Cbf12 induces flocculation by directly activating the transcription of gsf2+. In addition, gsf2+ expression is up-regulated approximately 2.4-fold in the cbf11Δ strain suggesting that the repressive flocculation function of Cbf11 may also be directly mediated through gsf2+. The activation and repression of gsf2+ transcription by Cbf12 and Cbf11, respectively, may occur by competitive binding to promoter sites since both transcription factors have been shown to interact with a canonical CSL consensus sequence in vitro [39]. Several putative sites with six out of seven nucleotide match to the canonical CSL consensus sequence are located within 900 base pairs of the gsf2+ promoter (data not shown). Further experimentation would be required to verify this proposed mechanism of gsf2+ transcriptional regulation by Cbf11 and Cbf12. It is likely that cbf12+ plays a lesser role in activating flocculation compared to mbx2+ since the floc size resulting from cbf12+ overexpression is considerably smaller than the mbx2OE strain. Also unlike mbx2+, deletion of cbf12+ is not sufficient to abrogate the flocculation of the rfl1Δ strain (data not shown). These data suggest that the flocculent phenotype of the cbf12Δ rfl1Δ double mutant is probably caused by the presence of mbx2+ activity.
CSL transcription factors are components of the conserved Notch signaling pathways in metazoans which primarily function in cell-to-cell communication during development [40]. Although multiple fungal CSL proteins have been discovered, their exact roles remain unclear in unicellular organisms [39]. Flocculation has been described as a manifestation of social behaviour in yeast with a purpose of enhancing survival under stressful conditions [3]. Therefore, it is conceivable that CSL transcription factors originated as regulators of this primitive form of cell-to-cell communication, and later evolved into the metazoan Notch signaling pathway.
We also discovered novel functions of the Yox1 and Sre2 transcription factors in the repression of flocculation. Loss of yox1+ or sre2+ results in a mild flocculent phenotype. The Yox1 homeodomain transcription factor functions as a repressor of MBF (Mlu1 binding factor) target genes to prevent their inappropriate expression at the end of S-phase [36]. Transcriptional repression of MBF target genes is mediated by the direct interaction of Yox1 and Nrm1 to the MBF complex [41]. Deletion of yox1+ causes a cell cycle delay and results in elevated constitutive expression of MBF gene targets [36]. Similarly, these genes (e.g. cdc18+, cdc22+, cdc10+, cdt1+, cdt2+, cig2+ and nrm1+) were also found to be upregulated 2.2 to 6.4-fold in our yox1Δ microarray expression data (Table S8). We found that the flocculent phenotype of the yox1Δ strain is also dependent upon gsf2+. However, the pfl+ genes including gsf2+ were not highly expressed in the yox1Δ strain. One possible explanation why pfl1+ genes were not highly expressed in the yox1Δ strain is that our experiments were performed under asynchronous culturing conditions, and therefore, the upregulation of pfl+ genes including gsf2+ could have been obscured if their expressions were periodically controlled in the vegetative cell cycle. However, it is unlikely that Yox1 regulates the pfl+ genes directly because previous chIP-chip analysis did not detect binding of Yox1 to the promoters of pfl+ genes [36]. Although there has been solid evidence linking yeast morphogenesis events such as pseudohyphal and hyphal growth to cell cycle regulators [42]–[47], the relationship between cell cycle control and flocculation remains unclear. A flocculation function for yox1+ has not been reported in other yeasts. However, disruption of YOX1 in the Sacc. cerevisiae ∑1278b strain inhibited filamentous invasive growth, a process usually associated with flocculation during nutritional limitation, while deletion of C. albicans NRM1 reduced flocculation [48], [49].
Sre2 is an uncharacterized membrane-tethered helix-loop-helix transcription factor predicted to be an ortholog of mammalian SREBP-1a, which is responsible for the transcriptional activation of genes needed for uptake and synthesis of cholesterol, fatty acids, triglycerides, and phospholipids [50]. While sre1+, a paralog of sre2+ has been shown to function in the transcriptional activation of sterol-biosynthetic and hypoxic-adaptation genes, there has been no direct evidence that sre2+ plays similar biological roles [37]. Loss of sre2+ results in the upregulation of gsf2+, pfl3+ and fta5+ transcripts (3.76, 2.51 and 3.08-fold, respectively) (Figure 7C, Table S9) which may contribute to its flocculent phenotype. The sre2Δ flocculent phenotype requires gsf2+ activity and is independent of mbx2+ and cbf12+ since these transcripts are downregulated in the deletion mutant.
In addition, the microarray expression profiles of yox1Δ and sre2Δ strains displayed similar differential gene expression despite the supposedly different functions of these transcription factors (Figure 7B). Mitochondrial genes were found to be highly upregulated in both deletion mutants (Tables S8 and S9). This occurrence may not be unexpected for Sre2 if it has a similar role in hypoxia as Sre1 where mitochondrial function is probably impaired [37]. It is currently not clear whether the mitochondrial genes are direct targets of Yox1 and Sre2 or induced in response to an altered physiological state in the deletion mutants. Interestingly, mitochondrial activity has been reported to be important for flocculation and invasive growth in Sacc. cerevisiae [48], [51]. Disruption of mitochondrial activity has been shown to alter the synthesis and structure of the cell wall, possibly by interfering with the interactions of flocculins and their substrates [52]. Based on these observations, the flocculent phenotype of yox1Δ and sre2Δ strains could be partially the result of enhanced mitochondrial activity from the upregulation of mitochondrial genes.
A genome-wide systematic deletion screen previously uncovered a cell-to-surface adhesion function that is sensitive to the presence of galactose for the Adn2 and Adn3 transcription factors [29]. Here, we discovered that adn2+ and adn3+ have additional functions in flocculation. Overexpression of adn2+ and adn3+ induced minor flocculation while loss of these genes prevented flocculation in inducing glycerol medium (Figure 8A and 8B). However, the flocculent phenotype of the adn2OE and adn3OE strains appeared to be primarily caused by the differential regulation of genes encoding cell wall-remodeling enzymes rather than flocculins. Several genes encoding cell wall-remodeling enzymes (gas2+, agn2+, psu1+ and SPAC4H3.03c) were highly induced when mbx2+, adn2+ or adn3+ was overexpressed. In the adn2OE strain, gas2+ and SPAC4H3.03c were the most highly induced genes (17.9 fold and 36.8-fold, respectively) (Figure 8C, Table S11) while in the adn3OE strain, these two genes and psu1+ appeared within the top 20 up-regulated genes (Table S12). Similarly, in the mbx2OE strain, gas2+, agn2+, and SPAC4H3.03c appeared within the top 100 up-regulated genes (greater than 3.7-fold increase, Figure 2A). We found that the single overexpression of these four genes could trigger flocculation (Figure 8D). Cell wall remodeling is an essential process for proper growth and adaptation to environmental stresses in yeast cells. Part of the cell wall-remodeling process involves the dissolution of sugar moieties in the glucan layer and elongation of glucan chains by glycoside hydrolases and glycosyltransferases, respectively. Among these four genes, three (agn2+, psu1+ and SPAC4H3.03c) encode for glycoside hydrolases while the fourth (gas2+) encodes for a glycosyltransferase. Agn2 is an endo-(1,3)-α-glucanase that hydrolyzes (1,3)-α-glucans of the ascus wall for ascospore release [53], [54]. Although agn2+ function appears only specific for sporulation, its ectopic expression could alter the cell wall structure during vegetative growth by inappropriate hydrolysis of (1,3)-α-glucan. Similarly, inappropriate glucan hydrolysis of the cell wall could be occurring as a result of ectopic expression of SPAC4H3.03c which encodes a putative (1,4)-α-glucanase (Hertz-Fowler et al., 2004). Psu1, which exhibits close homology to the members of the SUN family in Sacc. cerevisiae and C. albicans, as well as the BglA beta glucosidase of C. wickerhamii, has an essential function in cell wall synthesis [55]. Loss of psu1+ activity conferred resistance to (1,3)-β-glucanase suggesting that Psu1 may influence the amount or structure of (1,3)-β-glucan in the cell wall [55]. In addition, the (1,3)-β-glucanosyltransferase Gas2 has been shown to lengthen glucan chains during cell wall assembly and its overproduction is able to suppress the cell wall defect and lethality of gas1Δ cells [56]. Then how does the overexpression of these cell wall-remodeling genes trigger flocculation in S. pombe cells? The expression of flocculin genes during vegetative growth is not well characterized in yeasts, but studies in Sacc. cerevisiae indicate that flocculin synthesis and insertion into the cell wall initiate in early exponential phase prior to the onset of flocculation during stationary phase [57]. This suggests that the flocculins are already present in the cell wall, but cannot induce flocculation because of inaccessibility to cell surface oligosaccharides. We speculate that the restructuring of the β-glucan layer during cell wall remodeling may result in the rearrangement of flocculins that enhances galactose oligosaccharide binding, thereby promoting flocculation. Several lines of evidence in Sacc. cerevisiae support this hypothesis. First, alteration of cell wall structure by disruption of PKC1 activity results in flocculation [58]. Second, heat shock induces flocculation and regulation of cell wall-remodeling genes via the Hsf1 transcription factor [57], [59]. Currently, we cannot rule out that agn2+, psu1+, SPAC4H3.03c and gas2+ are the only cell wall remodeling enzymes that can trigger flocculation when overexpressed. Other genes with potential functions in cell wall modification and integrity such as gas4+, gma12+, meu7+, agl1+, meu10+ and mde5+ were also detected as putative target genes of Mbx2 (Table S2). In contrast, there was little change in gsf2+ transcript levels in the adn2OE or adn3OE strains compared to the empty vector control. However, the flocculation triggered by adn2+ and adn3+ overexpression was abrogated in a gsf2Δ background indicating that gsf2+ was indispensible for this process (Figure 8A). Altogether, these results suggest that Gsf2 is likely expressed in the cell wall as an inactive flocculin, and the cell wall remodeling resulting from adn2+ and adn3+ overexpression alters the arrangement of Gsf2 and possibly other flocculins that now becomes favorable for flocculation.
The single overexpression of the cell wall-remodeling genes triggered flocculation to a greater extent than the adn2OE and adn3OE strains. A possible explanation for the different degrees of flocculation between the transcription factor and its target genes could be that overexpression of adn2+ and adn3+ causes reduced fitness due to toxicity effects associated with a greater misregulation of genes compared to the aberrant production of a single enzyme. Consistent with this theory is that adn2OE and adn3OE strains exhibited additional phenotypes including septation defects (Figure 8E) which were not observed when gas2+, agn2+, psu1+ and SPAC4H3.03c were overexpressed (data not shown). Furthermore, a systematic overexpression analysis of 5280 genes in Sacc. cerevisiae revealed that genes encoding for transcription factors, signalling molecules and cell cycle regulators were more likely to cause reduced fitness [60].
In S. pombe, cell separation involves the transcriptional activation of adg1+, adg2+, adg3+, agn1+, eng1+, cfh4+ and mid2+ by the Ace2 transcription factor, which is in turn regulated by the Sep1 forkhead transcription factor [61]–[64]. We discovered that the adn2OE or adn3OE strains displayed multisepta and forkhead phenotypes similar to loss-of-function mutations of these cell separation genes. The cell separation defect in adn2OE and adn3OE strains is likely due to the downregulation of ace2+ transcription since ace2+ and its target genes were substantially downregulated in these strains (Figure 8C). However, sep1+ transcript levels remained unchanged in the adn2OE and adn3OE strains indicating that their involvement in cell separation phenotype could be either downstream of sep1+ or parallel to the sep1+ pathway. In additional to its flocculation role, Adn2 and Adn3 appear to have a separate function in cell separation perhaps by directly or indirectly repressing ace2+ transcription. Experiments are planned in the future to address these possibilities.
Interestingly, we also found some evidence that supports a role of Mbx2 and Cbf12 in cell separation perhaps through repression of ace2+ activity. Overexpression of mbx2+ and cbf12+ results in significant down-regulation of all seven Ace2 target genes approximately 1.5 to 3.4-fold relative to the empty vector control. (Tables S2 and S6). The mbx2OE strain indeed showed septation defects but were slightly different in nature than the adn2OE and adn3OE strains with less multi-septation and more mislocalization of septum material (data not shown). Moreover, overexpression of cbf12+ has been reported to produce multisepta phenotypes albeit at a low frequency [39]. These observations indicate the possible existence of crosstalk between flocculation and cell separation pathways mediated by the Mbx2, Cbf12, Adn2 and Adn3 transcription factors (Figure 9).
A comparison between the flocculation network of budding and fission yeast revealed both conserved and divergent features within the transcriptional circuitry. In Sacc. cerevisiae, the positive and negative transcriptional controls of the dominant flocculin gene FLO1 by Flo8p or Mss11p, and Sfl1p, respectively, draw parallel to gsf2+ regulation by the Mbx2-Rfl1 and Cbf12-Cbf11 opposing transcription factors in S. pombe [16], [19]. The conservation of the flocculin genes between these two yeasts is apparent among these transcription factors. Similar to Mbx2, Mss11p and Flo8p appear to activate multiple flocculin genes (FLO1, FLO9 and FLO11), while the latter may also regulate genes encoding cell wall enzymes (STA1 and SGA1 which both encode glycoside hydrolases) [9], [16]. The putative target genes of the Sfl1 repressor have been reported to include FLO1 and FLO11 [19], [65]. In contrast to the conservation of the flocculin genes, the types of transcription factors involved in flocculation are quite different between the two yeasts. Mbx2 belongs to the MADS box family, while the DNA-binding domains of Flo8p and Mss11p have not been defined. In addition, the Sfl1p and Rfl1 repressors contain the heat shock factor and Zn(2)-Cys(6) DNA-binding domains, respectively. Moreover, CSL transcription factors (Cbf11/Cbf12) are not found in Sacc. cerevisiae. These observations would imply that the cis-regulatory elements controlling transcription of the flocculin genes have likely undergone considerable rewiring within the transcriptional-regulatory network between the two yeasts. However, it was recently demonstrated that heterologous expression of Flo8p and Mbx2 could induce gsf2+ and FLO1 transcription in fission and budding yeast, respectively [20], [21]. Therefore, despite the divergent types of transcription factors controlling flocculin gene expression in the two yeasts, there may be some degree of conservation among the cis-regulatory sequences.
Although the transcription factors regulating flocculation appear to be quite different between the two yeasts, the downstream transcriptional events involved in the repression of flocculin genes are likely to be conserved. Disruption of genes encoding the Ssn6p-Tup1p general corepressor or the Cdk8 module of Srb/Mediator complex have been shown to cause upregulation of flocculin genes and constitutive flocculation in S. pombe and Sacc. cerevisiae [65]. In the latter yeast, Sfl1p represses FLO1 and FLO11 transcription through physical interactions with Ssn6p and Srb proteins (Srb8p, Srb9p and Srb11p) [19], [65], [66]. Moreover, Sfl1p has been reported to repress FLO8 in Saccharomyces diastaticus. The observation that Sfl1p can repress FLO1 transcription directly and indirectly through FLO8 seems very similar to the inhibitory feed-forward regulation of gsf2+ by Rfl1 in S. pombe. If these connections are truly analogous, then there is a possibility that Rfl1 repression could also be mediated through physical interactions with the Cdk8 module proteins. In the srb10− mutant, gsf2+ and mbx2+ expression are upregulated suggesting that its flocculent phenotype could be caused by a failure to repress mbx2+ transcription [24]. In addition, the flocculent phenotype of tup11+/tup12+ mutants [25] and the abrogation of lkh1Δ flocculation in the absence of mbx2+ [21] supports the role of Tup11/12 corepressor in Mbx2-Rfl1-mediated flocculation. Taken together, we speculate that Srb10 and Tup11/12 activity and binding may be required for Rfl1-mediated repression of gsf2+ and mbx2+. Future experiments focusing on the interactions between Rfl1, Tup11/12 and Srb8-10 in relation to flocculation would provide clarification to our speculation.
Our analyses of the transcription factors implicated in flocculation of S. pombe revealed the possible existence of several network motifs including positive autoregulation of mbx2+ and cbf12+ and regulation of gsf2+ by a inhibitory feed-forward loop (coherent type 2). The latter involves the Rfl1 transcriptional repression of gsf2+ directly and indirectly by inhibition of mbx2+ expression. Autoregulatory motifs have not been detected so far for FLO8, MSS11 and SFL1. The discovery of these network motifs in S. pombe suggests that the transcriptional inhibition of gsf2+ could occur more rapidly than its transcriptional activation. Experimental and modeling studies have proposed that positive and negative autoregulation of transcription factors generate slow and fast response times, respectively, within a transcriptional-regulatory network [67]. Under positive autoregulation, the synthesis rate of the transcription factor is initially slow at low concentrations, but increases as the concentration of the transcription factor reaches the activation threshold of the promoter, while negative autoregulation accelerates the attainment of steady state levels of the transcription factor [67]. Moreover, the inhibitory feed-forward motif of Rfl1 seems to indicate that repression of gsf2+ expression likely happens in a shorter period compared to its activation. Altogether, these data suggest that the onset of flocculation may occur gradually while repression of the flocculation pathway is a much faster process. Consistent with this speculation is the observation that it requires several days for wild-type S. pombe cells to undergo flocculation when grown in inducing medium.
In summary, we have provided an initial and substantial view of the transcriptional-regulatory network governing flocculation in S. pombe. Found within this network are the master regulators Mbx2, Cbf12, Adn2 and Adn3, which are able to trigger flocculation when overexpressed by the activation of their target genes encoding for flocculins and cell wall-remodeling enzymes. In addition, several repressors including Rfl1 were uncovered that play a major role in the regulation of these target genes. However, significant gaps of knowledge surrounding the transcriptional-regulatory network still remain. The environmental cues that impinge upon the activity of the positive and negative regulators, as well as the dynamics of transcription factor binding and regulation of target genes during the onset of flocculation remain to be elucidated. Also, although gsf2+ encodes the dominant flocculin, it is currently unclear whether the other flocculins have nonessential or more specialized roles during flocculation. Detailed analyses of the temporal and spatial expression of the pfl+ genes would be required to address these questions. Moreover, the exact mechanism of how other biological processes such as cell wall restructuring and mitochondrial function influence flocculation is unknown. Further studies to expand our knowledge of this transcriptional-regulatory network would provide a more comprehensive understanding of flocculation control and contribute to a valuable resource for the improvement of industrial yeast applications.
All strains used in this study are listed in Table S1 and were maintained on YES or EMM medium. Geneticin, nourseothricin, and thiamine hydrochloride were added to media at a concentration of 150 mg/L, 100 mg/L, and 15 µM, respectively. EMM medium was supplemented with amino acids when necessary at 225 mg/L each for adenine, leucine, and uracil. Matings were performed on SPAS medium. Wild type and deletion strains were assayed for flocculation in YEGlyEtOH (flocculation-inducing) medium containing 1% (w/v) yeast extract, 3% (v/v) glycerol, and 4% (v/v) ethanol. Overexpression strains containing ORFs under control of the nmt1 or nmt41 promoter were grown in EMM minus thiamine medium. Standard genetics and molecular biology techniques were performed as described in [68].
A PCR-based stitching method was utilized to construct the deletion and epitope-tagged strains. For construction of deletion strains, ∼500 bp fragments upstream and downstream of the ORF and the KanMX6 or NatMX6 cassette were PCR-amplified and gel-purified. The 3′ end of the upstream fragment and 5′ end of the downstream fragment contained ∼25 bp homology to the selectable marker cassette sequence. Approximately equimolar amounts (∼40 ng) of each PCR fragment were combined and stitched together in a 20 µl PCR reaction (0.2 mM dNTPs and 0.4 units of Phusion HF DNA polymerase (New England Biolabs), and subjected to one cycle of 98°C (30 sec), 5 cycles of 98°C (15 sec), 60°C (1 min), and 72°C (1–2 min) and a final extension at 72°C (5 min). The stitched product was then amplified in a 50 µl PCR reaction by combining the entire stitched reaction with 6 nmol dNTPs, 0.6 units of Phusion HF DNA polymerase and 20 pmol each of the outer pair of primers and then subjected to one cycle of 98°C (30 sec), 30 cycles of 98°C (10 sec), 60°C (30 sec) and 72°C (2 min), and a final extension at 72°C (5 min). The amplified product was gel-purified and transformed into the appropriate strain by lithium acetate transformation. A similar strategy was used to construct GFP-tagged transcription factors under the control of the native promoter. To tag the transcription factor with GFP at the C-terminus, ∼500 bp upstream and downstream fragments flanking the stop codon and the GFP-KanMX6 cassette (amplified from pYM27 plasmid, [69]) were PCR-amplified for the stitching reaction as described above. To conserve the native promoter in the N-terminal GFP fusion of Mbx2, 1 kb upstream of the mbx2+ start codon was amplified along with four other fragments for PCR stitching: (1) ∼500 bp upstream of the aforementioned 1 kb fragment; (2) ∼500 bp downstream of the mbx2+ start codon; (3) KanMX6 cassette and; (4) the GFP ORF with its stop codon removed and a GDGAGL linker added (adapted from [70]). All five fragments contained ∼25 bp overlapping homology to their respective flanking fragments and were PCR-stitched as described above. Proper gene deletion and GFP tagging were confirmed by colony PCR screen and the resulting amplicons sequenced.
Genes were overexpressed with the nmt1 promoter by cloning the entire ORFs of interest into the pREP1 or pREP2 vector. For ChIP-chip experiments, C-terminal triple HA-tagged Mbx2, Rfl1, and Cbf12 were expressed with the nmt41 promoter by cloning the corresponding ORFs into pSLF272 [71]. All the clones were PCR-confirmed, sequenced, and transformed into appropriate strains by the lithium acetate method. Expression of the HA-tagged proteins was verified by western blotting with anti-HA F-7 antibody (Santa Cruz Biotechnology, Santa Cruz, CA).
Strains overexpressing the triple HA-tagged Mbx2, Rfl1, and Cbf12 were grown in 200 ml of EMM medium containing appropriate supplements without thiamine for 18-20 hr to induce the nmt41 promoter. The empty vector control strain was cultured concurrently to a matching cell density of ∼8×106 cells/ml prior to harvesting. The experimental culture was divided into two, each for ChIP-chip and microarray expression profiling while the control culture was only utilized in the latter. The expression profiling cultures were harvested by centrifugation (1800× g, 3 min, 20°C), followed by immediate freezing of the cell pellets in liquid nitrogen. Culturing of adn2OE and adn3OE strains were performed similarly except that these genes were driven by the nmt1 promoter and were not epitope-tagged. For transcription factor deletion strains (rfl1Δ, cbf11Δ, sre2Δ, and yox1Δ), the mutant and an isogenic wild-type strain were concurrently grown in YES medium and harvested at a similar cell density as described above. Total RNA extraction, mRNA isolation, reverse transcription with aminoallyl-dUTP (Sigma-Aldrich, Oakville, ON), and Cy™3/Cy™5 (GE Healthcare, Buckinghamshire, UK) dye coupling of cDNA were performed with dye reversal as previously described [72]. Purified Cy™3- and Cy™5-labelled cDNA (1 µg in total) was hybridized onto custom-designed 8×15 K Agilent expression microarrays containing 60mer probes to all S. pombe ORFs in 2–3 times coverage per gene. The hybridization procedure was carried out according to the manufacturer's instructions (Agilent Technology, Santa Clara, CA) with the exception for the use of Human Cot-1 DNA. The microarrays were washed in 6× SSPE/0.005% sodium N-lauroylsarcosine at room temperature for 5 min followed by a second wash in pre-heated 42°C 0.6× SSPE for 2 min.
The microarrays were scanned with a GenePix4200A scanner (Molecular Devices, Sunnyvale, CA). The raw microarray data was lowess normalized [73] and the average log2 ratios with the corresponding t-test p values [74] from the dye-swap experiments were obtained using the R Bioconductor Limma package. Heat map images of the microarray expression and ChIP-chip data were constructed with Cluster 3.0 [75] and Java Treeview 1.1.6r2 [76]. The microarray expression data has been submitted to the NCBI Gene Expression Omnibus Database (GSE41730).
Culturing of the HA-tagged transcription factor strains are described above. The culture was fixed by the addition of a final concentration of 1% formaldehyde and agitation for 30 min at room temperature. The formaldehyde was quenched by the addition of 2.5 M glycine to a final concentration of 125 mM and agitation for 5 min at room temperature. The cells were then centrifuged (800× g, 5 min, 4°C), washed twice in 25 ml 1× ice-cold PBS (137 mM NaCl, 2.7 mM KCl, 10 mM Na2HPO4, 1.8 mM KH2PO4 pH 7.4) and washed once with 2 ml ice-cold lysis buffer (50 mM NaCl, 50 mM HEPES-KOH pH 7.5, 0.1% SDS, 1% Triton X-100, 1 mM EDTA, 0.1% sodium deoxycholate and 1 tablet/50 ml Protease Inhibitor Cocktail (Roche Applied Science, Indianapolis, IN)). The cell pellet was resuspended in 1.6 ml lysis buffer and stored at −80°C.
The cell suspension was transferred to two 2 ml bead beating vials containing 800 µl of 0.5 mm Zirconia/Silica beads (BioSpec Products, Bartlesville, OK) and subjected to 3 cycles of alternating 2 min beating and 2 min incubation on ice with a Mini Beadbeater 16 (BioSpec Products, Bartlesville, OK). The lysed cells were collected by puncturing the bottom of the bead-beating vial with a flame-heated inoculating needle and placing the vial on a sonication tube nested in 10 ml disposable culture tubes prior to centrifugation (800× g, 3 min, 4°C). The cell pellet was resuspended, transferred to chilled microcentrifuge tubes, centrifuged (16,000× g, 15 min, 4°C) to remove unbound soluble proteins, and the resulting pellet resuspended in 800 µl of fresh lysis buffer in a sonication tube. Total cell lysate volume was adjusted to 2.2 ml with lysis buffer and subjected to 4 cycles of sonication and 1 min on ice incubation at 30% amplitude, 30 sec setting using a Sonic Dismembrator with a 1/8 tapered microtip probe (Thermo Scientific, Waltham, MA). The sonicated cell lysate was centrifuged (4600× g, 2 min, 4°C) and the supernatant stored at −80°C. The supernatant was tested to ensure that greater than 90% of the sonicated DNA was in the size range of 100 bp–1 kb by subjecting a sample (∼50 µl) of the supernatant to overnight reverse-crosslinking at 65°C and phenol-chloroform extraction, followed by gel electrophoresis of 3–5 µg of DNA. To immunoprecipitate the chromatin-bound transcription factor, 100–200 µl of Dynabeads conjugated with sheep anti-mouse IgG (Invitrogen Life Technologies, Carlsbad, CA) were washed twice in 800 µl ice cold 1× PBS-BSA (5 mg/ml BSA, 1× PBS), resuspended in 800 µl cold 1× PBS-BSA with 5 µg of anti-HA F-7 antibody (Santa Cruz Biotechnology, Santa Cruz, CA) and shaken gently for 2 hr at 4°C on a Labquake Tube Shaker (Thermo Scientific, Waltham, MA). The beads were washed twice in 1 ml cold deoxycholate buffer (100 mM Tris-HCl pH 8, 1 mM EDTA, 0.5% (w/v) sodium deoxycholate, 0.5% (v/v) NP-40, 250 mM LiCl) and twice in 1 ml cold lysis buffer. The beads were resuspended in 200 µl 1× PBS-BSA, combined with 400 µl of sonicated cell lysate, and shaken gently for 2 hr at 4°C. Four washes of 5 minutes each were next carried out: (1) 1.4 ml cold lysis buffer at 4°C; (2) 1.4 ml cold lysis buffer with 400 mM NaCl at 4°C; (3) 1.4 ml deoxycholate buffer at room temperature and; (4) 1.4 ml TE (pH 8) at room temperature. The transcription factor and bound DNA were eluted twice from the Dynabeads by incubating with 250 µl TES each (TE pH 8, 1% (w/v) SDS) at 65°C for 6 min. Dynabead washing and the supernatant collection were performed using DynaMag™−2 (Invitrogen, Carlsbad, CA). For the input DNA, 200 µl of the cell lysate was added to 300 µl TES. Both the immunoprecipitated and input cell lysates were incubated at 65°C overnight to reverse the DNA-protein cross-linking. Western blotting with anti-HA antibody was performed to confirm proper pull-down of the transcription factor.
For protein removal, both immunoprecipitated and input samples were incubated with 200 µg Proteinase K (Promega, Madison, WI) and 20 µg glycogen (Roche Applied Science, Indianapolis, IN) at 56°C for 2 hr. The DNA was then extracted by phenol-chloroform extraction, ethanol-precipitated overnight, washed once with 70% EtOH, resuspended in 42 µl TE containing 0.1 µg DNAse-free RNaseA (Roche Applied Science, Indianapolis, IN), and incubated for 30 min at 37°C.
Blunt ends were generated in the entire immunoprecipitate and input DNA samples with 1 unit of T4 DNA Polymerase (Invitrogen Life Technologies, Carlsbad, CA), 1× NEB Buffer #2 (New England Biolabs, Ipswich, MA), 5 µg NEB BSA, and 10 nmol dNTPs in a 110 µl reaction by incubation at 12°C for 20 min, followed by phenol-chloroform extraction and ethanol precipitation with 10 µg glycogen and 1/10 volume 3 M NaOAc. The DNA pellets were washed in 70% EtOH and resuspended in 25 µl water. Approximately 1/5 of precipitated input DNA was used in the subsequent ligation reaction as input DNA concentration was >100 times greater than that of immunoprecipitated DNA. For ligation of linkers to blunt ends, the resuspended DNA was incubated with 1000 units of concentrated T4 DNA Ligase (New England Biolabs, Ipswich, MA), 1× T4 DNA Ligase Buffer (Invitrogen Life Technologies, Carlsbad, CA), and 200 pmol annealed linker (15 µM Oligo #1 5′-GCGGTGACCCGGGAGATCTGAATTC-3′ and 15 µM Oligo #2 5′-GAATTCAGATC-3′ in 250 mM Tris) at 16°C overnight. The annealed linker and the ligation mix were kept on ice at all times prior to overnight incubation. The DNA was ethanol-precipitated, washed, and resuspended in 25 µl water as described above.
The ligated DNA was PCR-amplified by adding 15 µl of labeling mix (2 µl aa-dUTP dNTP mix containing 5 mM each dATP, dCTP and dGTP, 3 mM dTTP, and 2 mM aminoallyl-dUTP (Sigma-Aldrich, St. Louis, MO)), 1.25 µl 40 µM Oligo #1 (5′-GCGGTGACCCGGGAGATCTGAATTC-3′), 4 µl 10× ThermoPol Buffer (New England Biolabs, Ipswich, MA) and 7.75 µl water) in a PCR cycler paused at 55°C. A 10 µl enzyme mix containing 5 units of GoTaq DNA polymerase (Promega, Madison, WI), 0.001 units of Pfu Turbo DNA polymerase (Stratagene, La Jolla, CA), and 1× ThermoPol Buffer (New England Biolabs, Ipswich, MA) was added and the PCR proceeded with one cycle of 55°C (4 min); 72°C (5 min); 95°C (2 min) and 30 cycles of 95°C (30 sec); 55°C (30 sec); 72°C (1 min), followed by a final extension at 72°C (4 min). The PCR products were purified (QIAGEN, Valencia, CA) with a few modifications: (1) buffer PE was replaced with phosphate wash buffer (5 mM KPO4 pH 8.5, 80% ethanol) and (2) buffer EB was replaced with phosphate elution buffer (4 mM KPO4 pH 8.5). A sample of the purified PCR product was run on an agarose gel to check for fragment sizes ranging between 100 bp and 1 kb. The purified PCR products were quantified, and equal amounts of immunoprecipitated samples and corresponding input samples were coupled to Cy™3 and Cy™5 dyes as described above.
The labelled samples (total amount of 3–5 µg) were hybridized onto an Agilent 4×44 K S. pombe Genome ChIP-on-chip microarray according to the manufacturer′s instructions (Agilent Technology, Santa Clara, CA) except for the use of Human Cot-1 DNA. The washing and scanning of the microarrays were performed as described above. The ChIP-chip data was normalized by scaling in Limma [73] and analyzed by ChIPOTle Peak Finder Excel Macro [77] with the default setting of log2 ratio cut-off of 1. Peaks located within 3 kb upstream of a start codon and 2 kb downstream of a start codon within a coding region or 3′-UTR, in the case of short ORFs, were assigned to the gene. ChIP-chip data sets are found in Tables S3, S5, and S7. Genes with multiple peaks are noted in the data set with the peak values. The ChIP-chip data has been submitted to the NCBI Gene Expression Omnibus Database (GSE41730).
The transcription factor binding specificities were determined by RankMotif++ [31] and MEME [30]. S. pombe promoter sequences 1000 bp upstream of the translational start site were used for these motif-finding algorithms. For MEME, promoter sequences of genes with various log ratio thresholds from expression microarray and ChIP-chip experiments were input into the MEME online server. RankMotif++ was applied to the entire expression microarray data since its motif-searching algorithm is threshold independent. The consensus sequences of the transcription factor binding sites were displayed by submitting the position weight matrices obtained from RankMotif++ analysis into the enoLOGOS online server [78].
Strains were grown in flasks at 30°C for the appropriate time, and 10 ml of culture was transferred to culture tubes for strains with larger floc sizes. Images were acquired immediately after vigorous shaking in glass culture tubes with a Canon G10 digital camera. For strains with mild flocculation, flocs were harder to visualize in culture tubes, and therefore, were observed in 90 mm plastic petri dishes. 10–15 ml of culture was transferred to petri dishes, followed by gentle shaking [8] on an orbital low-speed shaker (Labnet International, Woodridge, NJ) at maximum speed for one hour in room temperature. Floc images in petri dishes were captured using a SPImager (S&P Robotics Inc., Toronto, ON). Deflocculation of flocculent strains was performed by the addition of 2–20% D-(+)-galactose or 10 mM EDTA. The reflocculation of the deflocculated cells was performed by washing with water, resuspending the cells in YES or EMM medium or 100 mM CaCl2 and allowing the culture to sit for 30 min at room temperature.
For the overexpression of pfl+ genes, the strains were inoculated at a concentration of 107 cells in 100 ml of EMM without thiamine and cultured for 3 days at 30°C. For the weaker flocculent strains (pfl2+–pfl9+), 5 ml of the 3-day culture was then inoculated into 100 ml of fresh EMM without thiamine and incubated for another 3–4 days at 30°C followed by the petri dish flocculation assay as described above. Fresh EMM medium was added on the third day to prevent cells from remaining in stationary phase. Flocculation assays for the more flocculent overexpression strains were similarly carried out except the induction times were less than three days and did not require refeeding with fresh EMM medium. It should be noted that the empty vector control cells also eventually flocculate after refeeding with fresh EMM medium, but the onset of flocculation and flocs were delayed for several days and less pronounced, respectively, compared to the weakest flocculent overexpression strains. Wild-type strain and deletion mutants (mbx2Δ, gsf2Δ, cbf12Δ, adn2Δ and adn3Δ) were induced to flocculate by inoculating cells at a concentration of 108 cells in 100 ml of YEGlycEtOH medium and culturing for 5 days at 30°C followed by the petri dish flocculation assay as described above.
A patch of cells approximately 1/6 of a 90 mm petri dish was grown on YES medium for two days at 30°C and transferred as described in [29] onto a LNB plate (0.067 g/L yeast nitrogen base without amino acids (Bacto), 20 g/L glucose, 20 g/L agar, salts and vitamins as for EMM) with an underlying layer of YE + ALU (0.5% YE, 225 mg/L adenine, leucine, and uracil each) [79]. The plates were incubated at 30°C for 2 weeks before testing for cell-to-surface adhesion by washing cells off under a gentle stream of water and for invasive growth by rubbing the remaining cells off the agar with a finger under a stream of water. For strains showing resistance to rigorous washing by finger, a small section of the agar was cut out and observed under a Zeiss AxioScope A1 tetrad microscope (Zeiss, Thornwood, NY). Invasive growth was observed by the presence of elongated and branched cells remaining underneath the agar [29], [79].
Images of GFP-tagged cells were acquired with a Zeiss AxioScope 2 microscope (Zeiss, Thornwood, NY) and Scion CFW Monochrome CCD Firewire Camera (Scion Corporation, Frederick, MD). Fluorescence intensity was quantitated using the open source software ImageJ (version 1.44) (National Institutes of Health). First, the background signal for each image was subtracted using the “Subtract Background” function (50 pixel rolling ball radius). Individual cells were then selected as regions of interest using the freehand or polygon selection tools. Using the “Set Measurements” function both the area and integrated density were determined for each selected cell (n ranged between 27 and 50). Corrected GFP intensity was determined for each cell and was defined as the quotient of integrated density/area in background subtracted images. The averaged integrated density/area measurements for a given number cells is presented as the mean corrected GFP intensity with standard deviation. Significant differences between means were calculated by the Student t-test. To view nuclei and cell wall material, cells were methanol-fixed and stained with DAPI (1 µg/ml) and calcofluor white (50 µg/ml), respectively.
|
10.1371/journal.ppat.1002952 | Comparative Pathogenomics Reveals Horizontally Acquired Novel Virulence Genes in Fungi Infecting Cereal Hosts | Comparative analyses of pathogen genomes provide new insights into how pathogens have evolved common and divergent virulence strategies to invade related plant species. Fusarium crown and root rots are important diseases of wheat and barley world-wide. In Australia, these diseases are primarily caused by the fungal pathogen Fusarium pseudograminearum. Comparative genomic analyses showed that the F. pseudograminearum genome encodes proteins that are present in other fungal pathogens of cereals but absent in non-cereal pathogens. In some cases, these cereal pathogen specific genes were also found in bacteria associated with plants. Phylogenetic analysis of selected F. pseudograminearum genes supported the hypothesis of horizontal gene transfer into diverse cereal pathogens. Two horizontally acquired genes with no previously known role in fungal pathogenesis were studied functionally via gene knockout methods and shown to significantly affect virulence of F. pseudograminearum on the cereal hosts wheat and barley. Our results indicate using comparative genomics to identify genes specific to pathogens of related hosts reveals novel virulence genes and illustrates the importance of horizontal gene transfer in the evolution of plant infecting fungal pathogens.
| Cereals are our most important staple crops and are subject to attack from a diverse range of fungal pathogens. A major goal of molecular plant pathology research is to understand how pathogens infect plants to allow the development of durable plant protection measures. Comparing the genomes of different pathogens of cereals and contrasting them to non-cereal pathogen genomes allows for the identification of genes important for pathogenicity toward these important crops. In this study, we sequenced the genome of the wheat and barley pathogen F. pseudograminearum responsible for crown and root-rot diseases, and compared it to those from a broad range of previously sequenced fungal genomes from cereal and non-cereal pathogens. These analyses revealed that the F. pseudograminearum genome contains a number of genes only found in fungi pathogenic on cereals. Some of these genes appear to have been horizontally acquired from other fungi and, in some cases, from plant associated bacteria. The functions of two of these genes were tested by creating strains that lacked the genes. Both genes had important roles in causing disease on cereals. This work has important implications for our understanding of pathogen specialization during the evolution of fungal pathogens infecting cereal crops.
| Crop losses due to fungal pathogens represent one of the most serious threats to global food production. Staple cereal crops such as wheat, barley, rice and maize are subject to attack from a diverse array of fungal pathogens including biotrophs such as rust fungi that feed on living cells and necrotrophs such as Fusarium pathogens that kill host cells to obtain nutrients. Many Fusarium pathogens not only reduce crop yields but also produce mycotoxins that are harmful to humans and livestock when consumed in food and feed. A better understanding of the infection strategies used by these pathogens would help develop novel plant protection strategies. Comparative analysis of pathogen genomes offers a new and powerful approach to identify common and divergent virulence strategies as well as evolutionary history of pathogen lineages.
Shared virulence strategies may be used by different fungi to invade specific plant hosts. Presumably in many cases, the existence of common virulence strategies in different pathogen species may be explained by conservation of virulence gene function through vertical inheritance and/or exposure to common host defensive selection forces during pathogenesis on the same or related hosts. However, in some instances, horizontal gene transfer events have been identified in fungal pathogens and subsequently shown to have roles in pathogenicity [1]–[3]. A striking example of a locus-specific horizontal gene transfer event emerged from the sequencing of the wheat pathogen Phaeosphaeria nodorum (anamorph Stagonospora nodorum) genome, where a gene encoding a host-specific protein toxin (ToxA) was identified by homology to a known toxin from another wheat pathogen Pyrenophora tritici-repentis. Functional analyses demonstrated that ToxA was necessary for virulence in both pathogens [1]. It was proposed that transfer of ToxA from P. nodorum to P. tritici-repentis resulted in the emergence of the tan spot disease of wheat caused by P. tritici-repentis in the 1930s [1], [4]. In another example, genome analysis of the tomato vascular wilt pathogen F. oxysporum f. sp. lycopersici revealed the presence of several supernumerary chromosomes. Non-sexual transfer of one of these chromosomes to a non-virulent and genetically diverged recipient strain was shown to be sufficient to confer virulence on tomato [2]. Recently, association genomics has been used to identify the fungal effector Ave1 (for Avirulence on Ve1 tomato) in Verticillium dahliae. Ave1 homologs were shown to be present in diverse plant pathogenic fungi and important for virulence in at least one fungal species and one plant pathogenic bacterium [5]. In addition, Ave1 had strong homologies to plant proteins, suggesting that a cross-kingdom gene transfer event from plant to fungi may have occurred [5].
Ancient horizontally acquired virulence genes that have been retained because of their selective advantage may have more subtle sequence homologies and therefore are harder to detect [6], [7]. Such genes are best identified by gene phylogenies where a gene associates with a clade of sequences from unrelated species and relationships are incongruent with established ancestries [7]. For example, Klosterman et al [3] found the presence of a glucosyltransferase encoding gene in the genomes of three different wilt causing fungi, F. oxysporum f. sp. lycopersici, V. dahliae, and V. albo-atrum as well as an insect pathogen Metarhizium anisopliae that was otherwise found only in bacteria. These authors proposed that the acquisition of this gene predates the Verticillium spp. split and probably occurred independently in F. oxysporum f. sp. lycopersici [3]. Using targeted gene disruption approaches, the glucosyltransferase was shown to be important for virulence in V. dahliae toward tobacco (Nicotiana benthamiana) but not lettuce (Lactuca sativa). Similarly, it was recently suggested that the ability to synthesize auxin in members of both the Fusarium and Colletotrichum genera has probably resulted from horizontal transfer of auxin biosynthetic genes from bacteria [8].
Recent advances in genomics are revolutionizing the analysis of fungal species, which are particularly suited to analysis by the current generation of DNA sequencing technologies due to their relatively small genomes and in many cases minimal repetitive DNA contents. The de novo detection of shared virulence strategies without a priori information on the roles of shared genes in pathogen virulence offers an exciting possibility of uncovering new insights into the pathogenesis-related processes. The work of Klosterman et al [3] is one of the few examples where nothing was known about the role of the identified genes in pathogen virulence prior to being identified through comparative genomics. Therefore, it is reasonable to expect that further unbiased comparative genomic analyses will uncover examples of shared virulence genes or niche specialization genes in these pathogens, and ultimately provide insights into the co-evolution of virulence/niche specialization functions and the mechanisms of plant defense.
In this paper, we report the sequencing, assembly and annotation of the genome of Fusarium pseudograminearum (Aoki and O'Donnell) using Illumina sequencing and a comparative genomics analysis of its gene content. In many parts of the world, Fusarium pseudograminearum is the principal cause of Fusarium crown rot (FCR) of wheat and barley. FCR is significant in arid cereal-growing regions worldwide including South Africa [9], Northern Africa [10], the Middle East [11], Europe [12] and Australia [13] as well as the northwest of the United States of America [14]. In Australia, FCR is a chronic problem and among the most economically significant diseases of both wheat and barley [15], [16]. The recent increase in incidence of FCR is ascribed to the increased use of conservation farming practices such as zero tillage and stubble retention. These practices permit the survival of fungal inoculum on residual plant matter across planting seasons. Fusarium root-rot is a related cereal disease caused by F. pseudograminearum as well as other fusaria [17], [18]. F. pseudograminearum was initially distinguished from F. graminearum by host tissue preferences [19] and on the basis of molecular data was formally recognized as a separate species [20], [21]. Multi-locus sequence analysis of diverse isolates has shown that F. pseudograminearum is a single phylogenetic species globally [22] in contrast to the F. graminearum species complex, which shows geographical structure [23], [24]. In addition F. pseudograminearum is heterothallic whilst F. graminearum has a homothallic mating system. Whilst F. pseudograminearum, F. graminearum and F. culmorum are present throughout the Australian wheat growing regions and can all cause FCR, F. pseudograminearum is the species most commonly recovered from plants showing FCR symptoms [25], [26]. Field surveys in Australia have revealed that F. graminearum is the most frequently isolated species from wheat plants showing Fusarium head blight (FHB) symptoms [26], although F. pseudograminearum can also cause FHB disease. These observations suggest that F. pseudograminearum, while a broadly adapted cereal pathogen, may have evolved adaptations for niche specialization or infection processes that favor stem and/or crown infection.
We hypothesized that at least some genes in the F. pseudograminearum genome that were either uniquely or predominantly present in other cereal fungal pathogens may have specialized functions related to cereal pathogenesis and niche specialization. Throughout the manuscript the term ‘niche specialization’ is used to encompass other potential aspects of the biology of these fungi for which are poorly understood such saprophytic colonization of dead plant material, non-pathogenic interactions with other plants or potential endophytism of other hosts. A broad comparative genomic analysis indicated that the F. pseudograminearum genome contains genes that have strong homology to genes that are unevenly distributed across cereal pathogens while being apparently absent in other fungal genomes. Some of these genes shared by cereal pathogens also encode proteins with significant similarity to those from plant-associated bacteria. This finding is consistent with multiple horizontal acquisition events and indeed phylogenetic analysis of selected genes supported the hypothesis of horizontal gene transfer into diverse cereal pathogens. Functional analysis of two potentially horizontally acquired genes revealed important roles in the virulence of F. pseudograminearum on cereal hosts. Our results illustrate an important role for horizontal gene transfer in the evolution of cereal associated fungi.
The genome of F. pseudograminearum isolate CS3096 was sequenced using a combination of paired and single-end Illumina reads. De novo assembly of these reads resulted in a nuclear genome size of 36.8 Mbp assembled in 656 contigs with 50% of all nucleotides in contigs of 189 Kbp in length or greater (AFNW00000000). The average sequence coverage across these contigs was 179-fold. Compared to many other fungal genome assemblies using next generation sequencing technologies, the F. pseudograminearum genome sequence assembly has a relatively high N50 (Table 1). Gene model predictions from three programs were combined to identify 12,448 protein coding gene (see materials and methods).
The repeated sequence content of the F. pseudograminearum genome, assessed using the RepeatMasker program, is only 1.6%, which is slightly higher than that of the F. graminearum genome (0.7%) assessed using identical parameters, albeit that the sequencing and assembly methodology differed. RepeatMasker recognizes both simple sequence repeats and transposable elements present in RepBase [27]. Although approximately four times as many base pairs were flagged as being derived from Gypsy type Long Terminal Repeat (LTR) elements in the F. pseudograminearum genome (26 Kbp) compared to F. graminearum (6 Kbp), the difference in repetitive DNA content could mostly be attributed to a higher level of simple repeats and low complexity DNA (1.5% versus 0.41% of the genome, respectively). One high coverage contig in F. pseudograminearum encodes a LTR-type retrotransposon with best match to an F. oxysporum transposon, probably present in 9–10 copies based on an average coverage of 1689 fold. This transposon matched to sequences in a number of different fungi and also to transposons in both monocot and dicot plants. The contig showed no polymorphism in the sequence read assembly across 5.5 Kbp, suggesting all copies are identical and thus could not be placed in other assembled contigs. Also not included in the overall repeat counts are the contigs that represent rRNA encoding genes. The Illumina sequencing approach used here was unable to resolve these repeats in F. pseudograminearum and these currently appear in the assembly as high-coverage contigs.
The F. pseudograminearum genome sequence assembly was also compared globally to that of F. graminearum [28] by aligning the two genomes after masking simple repeat sequences and known fungal repetitive elements. In total, 89.8% of the F. pseudograminearum genomic sequence could be aligned to the F. graminearum genome at >70% nucleotide identity (Figure 1). An alignment with increased sensitivity was performed using a six frame translations of both genomes enabling an alignment of 94% of the F. pseudograminearum genome to that of F. graminearum. Thus, at least 6% of the low copy region of the genome (approximately 2.2 Mbp) appears to be completely unique to F. pseudograminearum. Very few rearrangements between the F. pseudograminearum and F. graminearum genomes in the aligned regions were found (Figure 1A). The amount of aligned sequence between the two species decreases towards the ends of the F. graminearum chromosomes (Figure 1B) and also in regions previously reported to be undergoing higher rates of genome innovation [28].
Genes showing a distribution of homologues limited to cereal pathogens would be both candidates for strong selection, because of a possible involvement in pathogenicity, and for horizontal acquisition. To determine whether the F. pseudograminearum predicted gene set contained such candidates, a BLASTmatrix analysis pipeline based on the identification of gene homologues by reciprocal best BLASTp hits was developed using predicted proteomes from the genome sequence of 16 different cereal pathogens and 11 non-pathogens of cereals. The entire BLASTmatrix analysis is presented in Dataset S1 along with the filtered protein sets described below. A numerical summary of the matches is shown in Table S1. In these analyses, 156 predicted F. pseudograminearum proteins did not have reciprocal best BLAST hits of any strength in any of the 27 organisms, while 239 predicted proteins had reciprocal best BLASTp matches only in cereal pathogens, with a score of at least 200 bits. In contrast, applying the same filtering criteria to identify proteins that are found in F. pseudograminearum and only non-cereal pathogens in this BLASTmatrix analysis identified only 32 proteins, 24 of which were present in the Fusarium lineage.
Amongst the 239 cereal pathogen proteins, 214 had matches within the two other Fusarium cereal pathogens (F. graminearum and F. verticillioides) and not in other species. Amongst the remaining 25 F. pseudograminearum proteins, only five had no match in other fusaria while matching other cereal pathogen proteins. Since the genome sampling described herein was limited to 27 fungal genomes, the sets of 239 and 156 proteins were combined and additionally curated using BLASTp against the NCBI nr (non-redundant) database. This process identified a total of 17 proteins with strong BLASTp matches (>200 bits) to bacterial proteins, and six of these 17 proteins were present only in a small number of other fungi (Table 2). The corresponding genes for these proteins are therefore good candidates for having been ancestrally acquired from bacteria. The GC content for each of these six genes was not obviously different to that of surrounding genes, possibly with the exception of FPSE_07765 (see below). Other candidates for cereal pathogen specificity, which to varying degrees showed distributions limited to cereal pathogens, were also identified through the BLASTmatrix filtering. These are therefore also of interest with respect to putative functions in niche specialisation and virulence towards plants. Examples of predicted genes from this latter category suggesting conserved roles in virulence towards cereals, and/or horizontal gene transfer are shown in Table 3. These candidate genes were subjected to more detailed analysis of sequence relationships as described below.
Amongst the F. pseudograminearum genes of potential bacterial origin (Table 2), the intronless gene FPSE_07765 and its orthologues in F. graminearum and F. verticillioides encode proteins highly similar to a bacterial protein with both aminotransferase and homoserine kinase domains. FPSE_07765 is 75% identical at the amino acid level across the entire length to a protein from Microbacterium testaceum, which is a bacterial endophyte of a variety of plants including the cereals sorghum and maize [29], [30]. Other fungal hits to this sequence in GenBank only align across less than half of the protein and at much lower identities, with the best fungal match in Trichophyton verrucosum at only 31% amino acid identity and with only partial query coverage of 44%. Phylogenetic analysis of FPSE_07765 identifies a Fusarium sequence clade embedded amongst a range of related bacterial sequences and strongly supports an origin of this gene via horizontal acquisition from bacteria, with retention in the cereal infecting fusaria (Figure S1). The GC content of FPSE_07765 was 58% compared to the genome average of 51.8±3.9% (±SD) for coding sequences.
FPSE_11233, another F. pseudograminearum gene with possible horizontal acquisition, encodes a putative secreted hydrolase that has reciprocal best BLAST hits only in cereal pathogens among the 27 selected genomes as well as hits in GenBank to putative pectin- and xylan-hydrolases in bacteria (best match is 38% identical with an e-value of 1e−125 to a Streptomyces hygroscopicus protein). Further manual database queries showed FPSE_11233-like sequences in the cereal endophyte Chaetomium globosum and the Brassica pathogen Leptosphaeria maculans. This gene sequence was analyzed further because, although L. maculans is not a cereal pathogen, it causes blackleg disease of canola which is a common rotation crop used in cereal farming systems [31], and L. maculans is related to other cereal pathogens in the Pleosporales family [32], [33]. Again, phylogenetic analysis supported the relatedness of these fungal genes to bacterial sequences (Figure S2). However, any potential acquisition from bacteria is clearly extremely ancient as the sequence seems to have undergone considerable vertical diversification within the fungi following acquisition from bacteria. This hypothesis is not only supported by the phylogenetic analysis but also computational predictions, suggesting that the position and number of introns in FPSE_11233-like sequences are not conserved between fungal species (data not shown). The placement of the L. maculans sequence outside of the Dothideomycete clade in the phylogenetic analysis (Figure S2) may also indicate a more complex mode of inheritance within the fungi.
A number of other enzyme-encoding genes of putative bacterial origin were also identified in the F. pseudograminearum genome, including genes encoding a hydrolase of unknown specificity (FPSE_07775) and a NAD+ dependent dehydratase/epimerase (FPSE_11221). Two putative amidohydrolase encoding genes (FPSE_00725 and FPSE_05717) that have clear bacterial homologues were also identified in this analysis. One of the amidohydrolase genes, FPSE_05717 hereafter termed FpAH1, also appeared to have a clear homologue in P. nodorum but not in any other fungus, suggesting a potential role for this gene in cereal pathogenesis and an unusual evolutionary history. The genomic organization, sequence diversity and virulence function of FpAH1 will be described in more detail later in the manuscript.
Many F. pseudograminearum genes also had homologues in other fungal cereal pathogens but no clear bacterial matches, suggesting that these genes are either laterally inherited or rapidly diversifying and therefore have been selectively retained only in a limited number of pathogenic fungi. Examples of these (Table 3) include three genes (FPSE_06956, FPSE_10646 and FPSE_02381) encoding small secreted proteins (candidate effector molecules) that were present in F. pseudograminearum and F. graminearum as well as other cereal pathogens. FPSE_06956 had orthologous matches in the Magnaporthe and Fusarium lineages but no other matches in the BLASTmatrix or GenBank. FPSE_10646 is a member of the killer protein 4 family (PFAM09044) of toxins that were shown to be extensively laterally shared between multiple fungal lineages including non-pathogens [34]. FPSE_02381 is a member of a two-gene family encoding small secreted cysteine rich proteins in F. pseudograminearum and F. graminearum and has strong, albeit non-reciprocal BLASTp matches in M. oryzae. The other member of this family in F. pseudograminearum, FPSE_05488, had reciprocal best BLASTp hits only in cereal pathogens, with the exception of a single match in the canola pathogen Colletotrichum higginsianum in the BLASTmatrix analysis (Dataset S1). Three other gene products shown in Table 3 (FPSE_05718, FPSE_05719 and FPSE_05720) were encoded in a gene cluster in the F. pseudograminearum genome. Interestingly, FpAH1, which is predicted to be of bacterial origin (see above and Table 2), also seems to be part of this cluster. The function of FpAH1 will be described in more detail later in the manuscript.
Also shown in Table 3 is a gene encoding a putative dienelactone hydrolase (FPSE_08136), hereafter termed FpDLH1, with very strong orthologues in F. verticillioides and C. graminicola with 98% and 88% identities, respectively at the amino acid level. The next best BLASTp match (69% identical) of FpDLH1 in GenBank, which is not a reciprocal best hit, is to FGSG_00089 from F. graminearum. The absence of strong reciprocal matches in all other fungi suggests that this gene may have been either shared between, or independently acquired from another donor species, by these Fusarium and Colletotrichum species through horizontal transfer mechanisms. The genomic organization and function of FpDLH1 will be described in more detail later in the manuscript.
FpAH1 encodes a putative amidohydrolase (Pfam domain PF07969). The best BLASTp match (e-value 2e−141, 88% identical, 94% similar at the amino acid level) of FpAH1 was to a predicted protein from the wheat glume blotch pathogen P. nodorum (encoded by SNOG_04819 hereafter termed PnAH1) [35]. The lack of FpAH1 homologues in other fusaria and presence in P. nodorum was confirmed via hybridization of a FpAH1 probe to genomic DNA of various Fusarium and a P. nodorum isolates (Figure S3). The overwhelming majority of subsequent BLASTp matches of FpAH1 in GenBank were to predicted proteins from bacteria (Table 4). The next fungal match after P. nodorum was to the F. graminearum protein FGSG_10599 (25% identical 2e−25) but this was much weaker than the bacterial matches (Table 4). FGSG_10599 also had another much stronger, indeed nearly identical (97%) protein (FPSE_00725) encoded in the F. pseudograminearum genome. The FPSE_00725/FGSG_10599 orthologs also had next best matches in bacteria. Phylogenetic analysis of these two F. pseudograminearum amidohydrolases strongly supports the hypothesis that they are of bacterial origin (Figure 2).
The F. graminearum genome contains six entries in this class of amidohydrolases (PF07969), all of which contain predicted orthologous matches with other genes in the F. pseudograminearum genome (Table S2). In other fusaria, clear orthologous relationships exist between the remaining five amidohydrolases with Pfam domain PF07969 found in F. pseudograminearum, although both F. oxysporum and F. verticillioides contain additional members in this class of amidohydrolases (Table S2). In P. nodorum, only four proteins fall into this class of amidohydrolases. Thus FPSE_00725 and FpAH1 encode two amidohydrolases with extremely restricted distribution in fungi and are most likely of bacterial origin. BLASTp analysis of the other F. pseudograminearum amidohydrolases identified a much wider distribution in fungi with FPSE_00474, FPSE_02365, FPSE_03227 and FPSE_11444 having more than fungal 30 hits of greater strength than the best bacterial hit (Table S2). FPSE_05738 was less widely distributed with seven hits of greater score than the best bacterial hit (Table S2) and may also be a candidate for acquisition from bacteria.
The predicted FpAH1 protein of 570 amino acid residues was encoded by an uninterrupted open reading frame of 1710 bp that was confirmed by RNAseq analysis of cDNA (data not shown). The P. nodorum genome annotation for PnAH1 contained a single intron, but it is likely that this prediction was not correct as the PnAH1 genomic region has a single uninterrupted open reading frame. In the coding region, FpAH1 was conserved between F. pseudograminearum and P. nodorum with 89% identity at the nucleotide level and 174 bp upstream of the predicted start codon and 85 bp downstream of the predicted stop codon could also be readily aligned.
The chromosomal complement of F. pseudograminearum has not been characterized and therefore the location of FpAH1 in the F. pseudograminearum genome is unknown. FpAH1 is present near the end of a ∼90 kb sequence contig, the first 70 kb of which aligns with the end of F. graminearum chromosome 1, as shown in Figure 3. Interestingly, the genes surrounding FpAH1 did not have clear orthologues in the F. graminearum genome. On the contig containing FpAH1, 28 genes (FPSE_05686 through to FPSE_05714, excluding FPSE_05712) had clear orthologues in F. graminearum. However, FPSE_ 5715 to FPSE_05720 did not. Furthermore, three of the gene products (FPSE_05718, FPSE_05719 and FPSE_05720) were also identified in the BLASTmatrix analysis to be cereal pathogen-specific, suggesting that parallel selection or coordinated acquisition may have affected this region (Table 3). The putative function and position of the genes in this region in F. pseudograminearum are shown in Figure 3. Of the remaining 40 genes on the end of chromosome 1 in F. graminearum, only eight encoded proteins with reciprocal best BLAST hits in F. pseudograminearum, and these were distributed across six different contigs in the F. pseudograminearum assembly (data not shown). In contrast, PnAH1 is ∼100 kb from the end of supercontig seven of the P. nodorum genome and appears to be surrounded by genes encoding proteins conserved in other fungi. In both P. nodorum and F. pseudograminearum, the GC content in the region of the PnAH1 and FpAH1 was similar to that of other gene rich regions of the respective genomes and some of the surrounding genes in the regions had introns. In summary, the genomic region containing FpAH1 has no equivalent sequence at the syntenic location in the F. graminearum genome nor the region containing PnAH1 in the P. nodorum genome.
The lack of close orthologues of FpAH1 and PnAH1 in other fungi and the presence of similar genes in bacteria (Table 4) based on BLASTp searches suggested an origin for these fungal genes via horizontal acquisition from bacterial species. fusaria belong to the order/class Hypocreales/Sordariomycetes while Phaeosphaeria is in the distantly related Pleosporeales/Dothidiomycetes. Acquisition of the gene may have occurred independently in both species or alternatively was horizontally transferred between the fungal species. To differentiate these possibilities, the sequence diversity of each gene was assessed in several globally sourced Fusarium and Phaeosphaeria isolates (Table S3). Isolates identified as F. pseudograminearum by multi-locus DNA sequencing [22] or by sequencing the elongation factor 1 alpha (EF1α) gene, were the only Fusarium sp. out of six tested that produced positive amplicons. PCR analyses of the distribution of FpAH1 in fusaria were confirmed for a limited number of isolates by hybridization analysis (Figure S3). PnAH1 sequences were PCR amplifiable from all P. nodorum isolates tested and also from two of four sister species, P. avenaria f. sp. tritici (Pat) 1 and Pat3, suggesting PnAH1 was present in a common ancestor in this lineage.
In order to test the hypothesis of horizontal transfer between F. pseudograminearum and P. nodorum, a ∼500 bp region of the AH1 gene was sequenced in isolate collections. A haplotype network showing the sequence relationships across both genera is shown in Figure 4. In F. pseudograminearum, seven haplotypes observed did not correspond to the geographic origin of isolates, consistent with global gene flow within the species as previously described [22]. Diversity in the Phaeosphaeria spp. sequences was more limited with only one haplotype observed in a global sample of P. nodorum and two haplotypes detected within each of the two Phaeosphaeria sister species containing PnAH1 orthologues. There were no shared sequence haplotypes between Phaeosphaeria spp. and F. pseudograminearum. The presence of AH1 orthologues in up to four related Phaeosphaeria spp. suggests that the acquisition of this gene occurred before the divergence of these species. Divergence of PnAH1 between the Phaeosphaeria spp. was comparable to divergence observed at neutral sequence loci (M.C. McDonald, unpublished data). Furthermore, the haplotypes observed in the Phaeosphaeria species complex and F. pseudograminearum were distinct, suggesting independent acquisitions of AH1-like sequences by each lineage.
The likely independent acquisition and retention of AH1 orthologues by F. pseudograminearum and P. nodorum suggests that this gene may encode a protein that is necessary for virulence, at least in some hosts. To test this hypothesis, a functional analysis of the FpAH1 and PnAH1 genes was undertaken. FpAH1 was expressed in both infected barley and wheat leaves and roots (Figure S4). The role of the FpAH1 amidohydrolase in FCR of barley was also assessed by generating gene deletion mutants in F. pseudograminearum by replacing 635 nucleotides of the FpAH1 locus with a geneticin resistance gene cassette (Figure S5A). There were no obvious defects in appearance or sporulation of the mutants. A culture time-course was used to compare the growth rates of one mutant to that of the parental strain and these were indistinguishable (Figure S5C). As shown in Figure 5, two independently-derived knock-out mutants consistently showed reduced virulence towards barley (cv. Golden Promise) seedlings across multiple independent experiments using a previously established FCR inoculation assay [36]. A second barley cultivar (cv Gairdner) also showed similar reduced virulence (data not shown). FCR disease severity caused by the mutant strains was significantly (P-value<0.01) reduced compared to those caused by the wild type strain. Genetic complementation of an FpAH1 mutant with a cassette containing FpAH1 under the control of the Aspergillus nidulans TrpC promoter restored virulence towards barley (Figure 5C and 5D), providing further evidence that FpAH1 is required for virulence against barley.
F. pseudograminearum has a wide host range within cereals with no evidence for race specialization [37], [38] for FCR disease on wheat. Unlike the experiments conducted on barley, highly replicated FCR assays on wheat using the FpAH1 knock-out mutants failed to reveal reduction of virulence on two unrelated varieties of wheat (Figure S6). However, reduced virulence was detectable when inoculated directly on roots toward both wheat and barley and again complementation restored virulence (Figure 6). P. nodorum is not pathogenic on barley and therefore was not tested on this host but mutant strains of this pathogen with deletions of the PnAH1 gene were also generated (Figure S5D) and tested on wheat plants in replicated leaf infection assays. No significant differences in virulence were observed between mutant and wild type strains on wheat (Figure S7). These experiments indicate that FpAH1 is required for full virulence on both wheat and barley but the importance of PnAH1 in pathogenesis remains unknown.
The F. pseudograminearum locus FpDLH1 encoding a dienelactone hydrolase was identified in the BLASTmatrix analysis as a cereal-pathogen associated gene with homologues detected only in F. verticillioides and C. graminicola. Strikingly, the gene (FPSE_08135 hereafter termed FpAMD1) immediately adjacent to FpDLH1 in the F. pseudograminearum genome encodes an amidase that is also present in the same divergently transcribed arrangement in the genomes of F. verticillioides and C. graminicola (Figure 7), indicating this is likely to be a two-gene cluster. The amidase family is much larger in ascomycetes than the dienelactone hydrolase family, making one-to-one orthologous relationships between species difficult to detect. However, for both FpAMD1 and FpDLH1, strong homology is detected between all three species at the nucleotide level (>80%) across the coding sequences of these genes. Furthermore phylogenetic analysis also supports the grouping of both of these genes in the three species in clades incongruent with expected evolutionary relationships (Figure S8). Thus, their genomic arrangement, discontinuous distribution in the fungi and strong homology to each other all suggest these genes represent a two-gene cluster that may have a common origin.
The FpDLH1 gene has the same intron-exon structure as another dienelactone hydrolase that is present in both F. pseudograminearum and F. graminearum (FPSE_08131 and FGSG_00089, Figure 7). This intron arrangement is not a generic feature of dienelactone hydrolase encoding genes in F. pseudograminearum, indicating a possible gene duplication event in the Fusarium lineage. Furthermore, FPSE_08131 had reciprocal best BLAST hits in all fusaria included in the analysis, with the exception of F. solani, as well as in two Magnaporthe spp., and the intron-exon structure was maintained in all species. The synteny in this region is also well conserved in the F. pseudograminearum-F. graminearum comparison (albeit split into two regions that align near the ends of different chromosomes in F. graminearum) but outside of this comparison the synteny weakens with the orthologues of the genes flanking this two-gene cluster in F. pseudograminearum appearing on multiple different contigs in F. verticillioides (Figure 7). In the F. verticillioides-F. pseudograminearum comparison, the conservation of FpDLH1/FvDLH1 (98% identical at the amino acid level) is much greater than the more widespread dienelactone hydrolase (FPSE_08131 and its F. verticillioides orthologue FVEG_12625) at 75% identity, suggestive of either an intra-fusaria transfer or strong conservation and selection. In C. graminicola the DLH1-AMD1 gene cluster is located on chromosome 2 in a region where only two of 11 genes flanking the cluster have clear orthologous relationships between C. graminicola and the closely related C. higginsianum (data not shown). However, three of the nine genes on one flank of the C. graminicola DLH1-AMD1 gene cluster have orthologues present in the flank adjacent to FpDLH1-FpAMD1 (Figure 7). These genomic associations suggest an ancient relationship in these regions.
Both FpDLH1 and FpAMD1 are expressed during infection of root by F. pseudograminearum and leaf tissue with higher expression in wheat than in barley (Figure S4). The role of FpDLH1 in fungal pathogenesis was assessed by creating knockout strains of two different F. pseudograminearum strains (CS3096 and CS3427), where the FpDLH1 gene was replaced by the geneticin resistance cassette (Figure S9). No obvious differences in sporulation were observed in the mutants nor were there differences in growth rate compared to their respective parents (Figure S9C). However, FpDLH1 mutants in both strain backgrounds showed significantly reduced virulence towards both wheat and barley in both root-rot and FCR assays (Figure 8 and Figure S10), suggesting that FpDLH1 contributes to fungal virulence against cereal plants.
The new DNA sequencing technologies are well suited to characterizing low copy, gene-rich regions of fungal genomes that are relatively small in size. Using Illumina technology it was possible to assemble de novo an almost complete sequence of the F. pseudograminearum genome. A large proportion (∼94%) of the F. pseudograminearum genome showed high similarity to F. graminearum and a great deal of co-linearity was observed. Alignment of the F. pseudograminearum genome sequence to the chromosomes of F. graminearum revealed that regions of poor sequence match were concentrated in specific regions, such as the ends of chromosomes and what are thought to be regions of ancient chromosome fusion in this Fusarium lineage and probable regions of genome innovation [2], [28]. Although more strains will need to be sequenced to confirm the species-specificity of these regions of the F. pseudograminearum genome, it is possible that the genes contained in these regions may be responsible for various phenotypes that distinguish F. pseudograminearum from F. graminearum, such as its propensity to cause FCR rather than FHB, its broad geographical adaptation in arid cereal production areas [39].
We hypothesized that genes in the F. pseudograminearum genome that were either uniquely or predominantly present in other cereal fungal pathogens may have a specialized function related to cereal pathogenesis and niche specialization. To identify these genes, we undertook a BLASTmatrix analysis and found that many genes present in the F. pseudograminearum genome were also present exclusively in cereal pathogens. 214 of these genes appeared to be conserved in the three cereal-infecting fusaria but had no equivalent matches in the genomes of three fusaria that infect dicotyledonous plants. These genes may have undergone specialized selection in these cereal pathogen Fusarium lineages but been diversified or lost in other fusaria. Several F. pseudograminearum genes also appeared to have equivalents in cereal pathogens outside the fusaria, and these were present mostly in other necrotrophic or hemibiotrophic Ascomycete fungi. Two of these genes, encoding an amidohydrolase (FpAH1) and a dienelactone hydrolase (FpDLH1) were selected for functional analysis in F. pseudograminearum. In both cases, we demonstrated roles in virulence on cereal hosts. This suggests that the comparative genomic analysis that we undertook is a powerful approach for the identification of genes with specialized pathogenesis on related hosts. Our analyses also identified many other candidate F. pseudograminearum genes showing similar distributions and a systematic analysis of their functions in fungal virulence is now warranted.
Our analysis identified genes in the genome of F. pseudograminearum with matches in bacterial genomes, and a number of genes were restricted to other fungal pathogens of cereals. These findings are consistent with possible acquisition of these genes by horizontal transfer. These observations also suggest considerable genome plasticity in F. pseudograminearum and provide a number of candidates for further study of potential horizontal acquisition. Compelling evidence for acquisition of a gene of bacterial origin and retention in cereal-infecting fusaria is illustrated by FPSE_07765 encoding a putative aminotransferase. Most matches to this aminotransferase were from bacteria, with the closest match showing a remarkable 75% amino acid identity to a predicted protein in the genome of Microbacterium testaceum, a common bacterial endophyte of cereals. Horizontal transfer from co-habiting endophytic bacteria into the Fusarium lineage with selective retention in cereal pathogens is a simple explanation for these strong, but restricted gene homologies and organismal relationships. Another significant bacterial match was the putative cell wall hydrolase encoded by FPSE_11233, with equivalent proteins present in many cereal pathogens. In this case, phylogenetic analysis indicated that all identified proteins from diverse fungal species clustered into a single clade. The biased occurrence of this gene in plant pathogens is likely due to selective retention after an ancient acquisition event during fungal evolution.
FpAH1 represents a striking example of likely horizontal movement of a gene from a bacterium into the F. pseudograminearum genome. The only closely related gene to FpAH1 in the fungi examined was PnAH1 found in the genome sequence of P. nodorum, a pathogen of wheat. Sequencing of several globally distributed isolates of F. pseudograminearum revealed several distinct haplotypes for this gene. This suggests that acquisition of FpAH1 was not recent or that there has been significant selection driving the creation of new alleles. There was no evidence of horizontal transfer directly between F. pseudograminearum and a Phaeosphaeria spp. The limited divergence of PnAH1 between the Phaeosphaeria spp. orthologues, suggests that a common ancestor of these Phaeosphaeria species acquired the gene. The presence of the gene in at least two lineages of cereal pathogens suggests that the gene may play an important role in wheat pathogenesis.
Both FpAH1 and PnAH1 were clearly dispensable for growth and their retention in two otherwise unrelated fungal pathogens suggests a specialized function for these genes in fungal virulence. Indeed, a virulence function for FpAH1 against two different cereal hosts was confirmed. However, PnAH1 knockout strains did not show altered virulence against wheat. Pathogenesis in any one species is the sum of many different components and the relative contribution of these genes to pathogenesis in the different species may be substantially different. The genomic context of the genes in these two fungal pathogens was also very different. PnAH1 is located in a region of the P. nodorum genome adjacent to several conserved genes. In contrast, FpAH1 occurs at the end of a long contig in a cluster of genes found in other cereal pathogens, but not in F. graminearum, and it could be that FpAH1 is functioning in F. pseudograminearum as part of this group of genes. It appears that most of the genes on the end of chromosome 1 in F. graminearum are absent from F. pseudograminearum. This observation provides further support to the notion that chromosome ends or ancient chromosome fusion sites are regions of genome innovation in the Fusarium lineage [28] and regions like this may have played a role in niche separation between F. pseudograminearum and F. graminearum. The role and reason for retention of PnAH1 in P. nodorum however, remains unknown.
The two-gene cluster represented by FpDLH1-FpAMD1 is likely to be of fungal origin in the genomes of F. pseudograminearum, F. verticillioides and C. graminicola. A comparison of the genomic context of the DLH1-AMD1 genes in F. pseudograminearum and C. graminicola also supports a common origin with some related genes being present in flanking regions of this cluster in both pathogens. The conservation (∼80%) of the nucleotide sequence across the coding regions of both genes between C. graminicola and the two Fusarium spp. suggests that any possible genetic exchange event between these lineages is ancient and has thereby allowed accumulation of considerable sequence divergence. The close physical proximity in the F. pseudograminearum genome of FpDLH1 to another dienelactone hydrolase encoding gene, FPSE_08131, with identical intron-exon structure and with orthologues in other fusaria, suggests FpDLH1 may have arisen vertically by gene duplication and subsequent diversification within the Fusarium lineage. The FpDLH1/FvDLH1 and FpAMD1/FvAMD1 orthologues of F. pseudograminearum and F. verticillioides show remarkable identity (98%) at the amino acid level but very different genomic context. However, currently, it is not possible to resolve whether an inter-species transfer had occurred between the Fusarium lineages, or alternatively whether strong DLH1 and AMD1 gene conservation with regional genomic rearrangements had occurred in these two cereal infecting fusaria. Further diversity surveys will be required to resolve the origin of FpDLH1- and FpAMD1- related genes in these cereal pathogens.
Both FpAH1 and FpDLH1 are thought to be catabolic enzymes based on conserved domain matches. Amidohydrolases are a diverse superfamily of enzymes that catalyze the hydrolysis of C-N bonds in small molecules. This family includes enzymes functioning in central metabolism (eg urease), enzymes that degrade xenobiotics (eg atrazine) as well as those that are known to catalyze reactions other than C-N cleavage, including P-O cleavage and also isomerisation [40]. Likewise, the dienelactone hydrolase family of enzymes appears to be large, with members involved in the degradation of chloroaromatic compounds [41], [42] by bacteria and activation of prodrugs containing lactone-like side chains in humans [43]. The specific biochemical roles of FpAH1 and FpDLH1 during this particular host-microbe interaction remain elusive. However, the predicted catabolic activity of these enzymes suggests they could be either targeting specific plant defense compound(s) or involved in production of fungal toxin(s). Although the defensive compounds important in the response to F. pseudograminearum are currently unknown in barley and wheat, candidates may include hordatine, hordenine and gramine in barley and the benzoxazolinones in wheat [44]–[47]. All these compounds contain C-N bonds and are known to have antifungal properties [44]–[47]. However, fungal growth inhibition assays conducted in liquid culture medium with synthetic hordatine and gramine showed that while F. pseudograminearum is moderately sensitive to both compounds, FpAH1 knockout strains were equally as sensitive as the wild type (data not shown). Likewise, the FpDLH1 mutants were as sensitive as the wild-type to the benzoxazolinones, BOA (2-3H-Benzoxazolinone) and MBOA (6-methoxy-2(3)-benzoxazolinone). However, in F. verticillioides BOA detoxification is a two-step process with only one of the two responsible genes having been cloned [48]. A more detailed understanding of the barley and wheat metabolites involved in defense against F. pseudograminearum would be an important starting point to identify molecular mechanism(s) of FpAH1- and FpDLH1- mediated virulence.
Genomics is allowing increased discovery of examples of likely horizontal gene transfer events [6], [7]. Our work provides additional evidence for this hypothesis although detection of horizontal gene transfers, particularly of ancient events, requires robust phylogenetic analyses [3], [7], [49]. The extent of the horizontal gene transfer phenomenon has not yet been fully ascertained amongst the sequenced fungal genomes but there will presumably be many more examples. A robust methodology needs to be developed to enable identification of horizontally acquired genes on a scale of kingdoms and beyond. The BLASTmatrix analysis presented here is one possible method for identifying these genes, although it is limited in its ability to cope with post-acquisition family expansion. It is also important to note that the analysis reported here was centered on F. pseudograminearum and limited by the species included in the comparative analyses. A broader systematic analysis for each fungal pathogen genome is warranted to obtain a more complete perspective of potential gene sharing and its relation to host range and virulence.
In summary, in this paper we report the first genome sequence for F. pseudograminearum and demonstrate that a broad comparative genomic analysis can identify genes that show a biased distribution in fungal pathogens of plants with putative roles in virulence processes or niche specialisation. Functional analysis in F. pseudograminearum of two such genes demonstrated novel virulence functions on cereals.
The F. pseudograminearum isolate (CS3096) chosen for genome sequencing was originally isolated from a wheat crown collected in 2001 near Moree in Northern New South Wales, Australia [26]. Isolates for phylogenetic analysis of the FpAH1 gene were selected from a collection housed at CSIRO Plant Industry, Brisbane, Australia, consisting of isolates from Australia, New Zealand, Canada, United States of America and Turkey (Table S3). Phaeosphaeria isolates were selected from a collection housed at ETH Zurich and the Australian isolate SN15.
Two lanes of single-end 75 bp and one lane of 100 bp paired-end sequencing was performed on an Illumina GAIIx genome sequencer by the Australian Genome Research Facility, Melbourne Australia. Reads were imported into CLCBio Genomics Workbench 3 and quality trimmed (default parameters) prior to assembly. A total of 60 million paired reads and 33 million single end reads were used in de novo assembly plug in (version 3.03), again using default parameters with a minimum contig size of 500 bp. Low coverage (<40×) contigs (compared to the genome average of 179×) were excluded from further analyses. BLASTn comparison to the F. graminearum mitochondrial sequence and coverage information (3,500–6,000×) was used to separate nuclear and mitochondrial sequence. 29 contigs with total length 107 Kbp were identified in this process and excluded from the genome annotation but are included in the F. pseudograminearum genome submission to GenBank. This Whole Genome Shotgun project has been deposited at DDBJ/EMBL/GenBank under the accession AFNW00000000. The version described in this paper is the first version, AFNW01000000.
Repeat masking was performed using RepeatMasker version open-3.2.8 [50], run in sensitive mode with cross_match version 0.990329 as the search engine and RepBase version 20090604 [27]. The species descriptor “fungi” was used.
Whole genome alignments were performed with the NUCmer and PROmer algorithm in the MUMmer package [51]. The minimum cluster length was set to 100 bp. The total length of the genome alignment between F. graminearum and F. pseudograminearum was calculated by summing the length of all non-redundant alignments extracted by using the show-coords program (part of the MUMmer package).
Protein coding genes were ab initio predicted in the F. pseudograminearum genome using FGENESH [52] based on the F. graminearum gene models as part of the MolQuest2 package from Softberry, AUGUSTUS [53] and GeneMark-ES version 2 [54]. AUGUSTUS was run with a training set of F. graminearum genes that had 100% nucleotide alignment and identity to the F. pseudograminearum genome. When the three programs disagreed in the splicing for the same gene, the following strategy was used to prioritize the predictions. When two of the three programs agreed, this prediction was taken. When all three programs differed for what appeared to be the same gene, priority was given in the order FGENESH, GeneMark and AUGUSTUS. Predicted transcripts encoding <20 amino acids were removed manually from the main prediction set. In total 17,503 unique transcripts were predicted, which after removing 4833 alternative predictions for the same gene left a putative set of 12448 unigenes.
The proteins coded by 27 different fungal genomes were downloaded from the Broad Institute (www.broadinstitute.org), Joint Genome Institute [55] or NCBI for the F. oxysporum strain 5176 genome [56]. Reciprocal best BLASTp hits were used to determine the orthologous protein relationships between F. pseudograminearum and each of the 27 fungal protein sets. This was performed using a pair-wise all versus all BLASTp analysis was conducted on the National Computational Infrastructure Specialized Facility in Bioinformatics Cluster located at the University of Queensland, Australia. Perl scripts obtained from http://sysbio.harvard.edu/csb/resources/computational/scriptome/unix/Protocols/Sequences.html were used to extract best reciprocal hits and the associated blast score from each species comparison BLASTp outputs. The F. pseudograminearum–F. graminearum relationship was treated as a special case and used to curate the dataset. In total, 11481 genes had clear one-to-one relationships between the two species, but differences in gene prediction (eg splice sites or predictions split/fused with respect to each other in each of the genomes) resulted in false identification of genes without orthologous matches in this analysis. These were accounted for by performing BLASTn analysis using F. pseudograminearum gene sequences containing introns as the query against the F. graminearum genomic sequence. Transcripts showing strong matches to the F. graminearum genome (match length greater than 50%) were classified as having different annotation between F. graminearum and F. pseudograminearum. 456 genes were removed from the analysis using this process.
Logical formulae and filtering functions in Microsoft Excel were used to identify potential cereal pathogen specific proteins via the presence of reciprocal best BLAST hits in one or more of the 16 cereal pathogens at a BLAST bit score of greater than 200 and in the complete absence of a reciprocal best BLAST hit in any of the 11 non-cereal pathogens. Manual curation of putative cereal pathogen specific genes was conducted to account for the possibility that genes were predicted in F. pseudograminearum that were missed in the annotation of other genomes. This was performed by querying the intergenic sequences of the nearest non-cereal pathogen, F. oxysporum f. sp. lycopersici in a tBLASTn search using the putative cereal pathogen specific proteins. Through this process, 112 proteins were removed due to potential missed predictions in the F. oxysporum f. sp. lycopersici genome.
Phylogenetic analyses were implemented using phylogeny.fr [57]. Briefly, multiple alignments were generated using MUSCLE [58] with default parameters, and curated using Gblocks [59]. Phylogeny was performed using PhyML [60] with the WAG amino acid substitution matrix [61] using an approximate likelihood-ratio test for branch support [62]. Trees were drawn using TreeDyn [63]. Trees were exported to Adobe Illustrator to allow shading of fungal branches.
A slot blot membrane was prepared using a Hoefer PR 648 apparatus using Amersham Hybond-XL membrane (GE Healthcare) prewetted with 0.2 M NaOH. Approximately 500 ng of DNA was prepared in 0.2 M NaOH to a final volume of 60 µL and incubated at 37°C for 15 min prior to application to the membrane. Following application of the DNA the membrane was cross-linked using a GS GeneLinker UV cross-linker (BioRad) with 150 mJ of energy. Two identical membranes were prepared and probed with either a FpAH1 or rRNA probe. Probes were generated using the PCR DIG Probe synthesis kit (Roche) according to the manufacturer's instructions using primers FpAHDiversityF and FpAHDiversityR for the FpAH1 probe and ITS1-F and ITS4 primers for the rRNA probe (Table S4). Hybridization and washing was performed with the DIG Easy Hyb solution and DIG wash and block buffer set (Roche). Detection was performed with the DIG Luminescent detection kit (Roche).
For Phaeosphaeria spp. PCR amplification was performed in 20 µL reactions containing 0.05 µM of PnAHDiversityF and FpAHDiversityR (supplied by Microsynth), 1× Dream Taq Buffer (Fermentas), 0.4 µM dNTPs (Fermentas) and 0.5 units of Dream Taq DNA polymerase (Fermentas). For F. pseudograminearum isolates, the forward primer was replaced with FpAHDiversityF and PCR was conducted using Invitrogen Taq DNA polymerase as per the manufacturer's instructions. Primer sequences are listed in Table S4. The PCR cycle parameters were: 2 min initial denaturation at 96°C followed by 35 cycles of 96°C for 30 s, 58°C for 45 s and 72°C for 1 min and a final 5 min extension at 72°C. The same cycling conditions were used for F. pseudograminearum isolates except the denaturation was performed at 94°C. Fusarium-derived PCR products were purified using the QIAgen PCR purification kit and sequencing was carried out by the Australian Genome Research Facility using both the forward and reverse primers. Sequencing reactions for Phaeosphaeria-derived products were conducted in 10 µL volume using the BigDye Terminator v3.1 Sequencing Standard Kit (Applied Biosystems, Foster City, CA) with both the forward and the reverse primer. The cycling parameters were 96°C for 2 min followed by 55 or 99 cycles of 96°C for 10 s, 50°C for 5 s and 60°C for 4 min. The products were cleaned with the illustra Sephadex G-50 fine DNA Grade column (GE Healthcare) according to the manufacturer's recommendations and then sequenced with a 3730×/Genetic Analyzer (Applied Biosystems). Alignment of forward and reverse sequences for each isolate was performed in SeqScape software V2.5 (Applied Biosystems). Translation and identification of protein haplotypes was also performed using this software. The program TCS v1.2 was used to visualize the most-parsimonious haplotype network [64].
The vectors for deletion of the FpAH1 and FpDLH1 genes were constructed using lambda phage mediated recombination as described previously [65] using primers listed in Table S4. For FpAH1 the targeting construct consisted of 1858 bp of sequence immediately upstream of the start codon and 1132 bp covering the 3′ end of the gene and downstream sequence. 635 bp were deleted. For FpDLH1 the targeting construct consisted of 1523 and 1898 bp flanks, deleting 891 bp of the gene, leaving 125 bp of the 5′ coding sequence and 22 bp of the 3′ coding sequence. F. pseudograminearum protoplasting and transformation was performed as previously described for F. graminearum [66] except the STC (sorbitol, Tris and calcium chloride solution) was made with 0.8 M sorbitol instead of sucrose and the 40% PEG8000 solution also contained 0.8 M sorbitol.
Transformants were selected on 50 mg L−1 geneticin (Sigma, St. Louis, MO, USA), subcultured onto geneticin and DNA was prepared using the REDextract and Amp kit as per the manufacturer's instructions (Sigma). Transformants were screened for gene deletion using AHKOscr1, AHKOscr2 and gpdAr primers. In total, 42 transformants were screened and six were identified as carrying a deletion of FpAH1. Mutant strains were single-spored and then stored as water cultures from ½ PDA plates.
The PnAH1 deletion construct was made using overlap PCR as previously described [67]. PnAH1KO5'F and PnAH1KO5'R were used to amplify a 1456 bp region upstream of PnAH1 whilst PnAH1KO3'F and PnAHKO3'R amplified a 1519 bp region downstream. These flanks were fused using overlap PCR with a hygromycin resistance cassette amplified from pAN7.1, resulting in a deletion construct of 5.9 kb. The deletion cassette was transformed into P. nodorum isolate SN15 protoplasts as previously described [68] and screened by using primers designed outside the flanking DNA (PnAH1KOscr-F and PnAH1KOscr-R). Copy number of the transformed construct was determined as previously described [69].
The ΔFpAH1 mutant was complemented by co-transformation of ΔFpAH1 mutant 96T492 with pAN7.1 [70] and a construct designed to expresses FpAH1 under the control of the Aspergillus nidulans TrpC promoter. This construct was created by PCR amplification of the coding sequence and 341 bp of terminator of FpAH1 from genomic DNA of isolate CS3096 with forward and reverse primers (AH-OXf and AH-OXr) containing ClaI and EcoRI sites respectively and replacement of the hygromycin cassette in pUChph [71] using the same restriction sites. The TrpC-FpAH1 section of the construct was fully sequenced prior to transformation into ΔFpAH1 mutant 492. Selection was with hygromycin at 200 mg L−1. To verify that transformants were over-expressing the FpAH1 gene after successive rounds of sub-culture and single-sporing, they were grown in 6-well tissue culture plates in liquid basal media with 10 mM (NH4)2HPO4 as the nitrogen source [72]. Mycelia were harvested, freeze-dried and RNA extracted using TRIzol (Invitrogen) according to the manufacturer's instructions after four days of growth at 28°C in the dark. Relative FpAH1 gene expression was measured using quantitative reverse transcriptase PCR as previously described [73]. β-tubulin and FpAH1 (FpAH1f and FpAH1r) primers are shown in Table S4. Expression of FpAH1 in the two transformants used was ∼3–5-fold higher than the wild type strain during in vitro culture and undetectable in the mutant as expected.
A microtitre plate assay to monitor growth was performed based on the method described by Schmelz [74]. Briefly, each well contained 200 µL of basal media [72] with 5 mM glutamine as the nitrogen source with a final spore concentration of 1×104 spores mL−1. Absorbance at 405 nm was measured using an iEMS microplate reader. For the FpAH1 mutant test the plate remained in the instrument for the duration of the assay at ambient temperature. In the case of the FpDLH1 mutant assay after an initial reading the plate was incubated at room temperature and readings commenced 22 hours post inoculation.
Assessment of fungal virulence during FCR was carried out using the method described previously [36]. For virulence testing of FpAH1 mutant and complemented strains, isolates were inoculated onto 14 cm Synthetischer Nährstoffarmer Agar plates from potato dextrose agar plugs stored as water cultures and allowed to grow for 14 days under 12/12 hour day/night cycle under white and black fluorescent light at room temperature (∼22°C). Spores were harvested by flooding the plates with water and adjusted to the same concentration. To test the role of FpDLH1 in virulence, spores were produced in shaking 1 L flasks containing 100 mL of 25% Campbell's V8 juice inoculated with a single plug taken from a 20% V8 juice plate. Flasks were incubated at room temperature for 8 days and spores harvested by filtration through miracloth followed by centrifugation (5,500×g) to remove media. Spores were resuspended in sterile water, counted and adjusted to the same concentration. Seeds were plated out two days prior to inoculation on wet paper towel in a 14 cm Petri dish and allowed to germinate on the laboratory bench. Germinated seeds were transferred to a 50 mL falcon tube containing 2 mL of the spore suspension and rolled gently to coat the seeds. Six to ten seeds were placed in a single paper towel, rolled up and tapped closed and placed in a jar containing water. Paper towel rolls were kept moist throughout the experiment. For the analysis of FpAH1 mutants a concentration of 1×105 spores mL−1 was used and plants were maintained on a laboratory bench without supplementary lighting. To test complemented FpAH1 mutants and FpDLH1 mutants, 4×105 spores mL−1 were used and plants were maintained in a closed bench top biohazard cabinet with a cool white fluorescent light suspended from the roof of the cabinet (approximately 15 cm from the plants) providing 16 hours of light per day. Assays were scored by recording the number of plants that were alive as a proportion of the total plants.
Fungal strains were plated from PDA plug water storage onto 20% Campbell's V8 juice 1.2% agar plate and incubated for seven days at room temperature under white and black fluorescent light on a 12/12 hour day/night cycle. Seed for the assay were surface sterilized by soaking for 5 minutes in a 0.64% sodium hypochlorite-10% ethanol solution, followed by several rinses in sterile distilled water. Seed were plated onto three sheets of pre-wetted Whatmann no 3 12.5 cm filter disks in a 14 cm Petri dish. Seeds were incubated in the dark for 5 days at 4°C prior to germination at 20°C also in the dark. Germinated seedlings were distributed to Petri plates assembled with three filter papers wetted with 20 mL of sterile water prior to inoculation. Each plate contained 15–16 seedlings and was used for inoculation with one isolate. Inoculum consisted of agar plugs taken using either an inverted 1-mL pipette tip or a number 4 (6 mm) cork borer from the edge of the growing V8 agar plate. Plugs were placed on a single root per seedling about 1 cm below the seed with the mycelia in direct contact with the root. Plates were sealed with sealing film (PhytoTechnology Laboratories), incubated in a Thermoline illuminated incubator for 6 days at 20°C with 12 hours of light provided by fluorescent lamps. Assays were scored by measuring the length of the shoot.
Parental and mutant strains of P. nodorum were tested on the susceptible wheat cultivar Amery as previously described [75]. Latent period sporulation was assessed using detached leaves as described previously [76].
Gene expression during infection of wheat and barley by F. pseudograminearum was performed using both a detached leaf infection assay and a root infection time course. For the latter, plants were inoculated as described for the root-rot virulence assay and root segments were harvested into liquid nitrogen at 24, 48 and 96 hours post inoculation. For expression in detached leaves, fourth leaf segments (7–8 cm long) of both hosts were taken from glasshouse-grown plants and each end of the cut leaf was sandwiched between water agar in a 14-cm Petri plate. Leaf segments were pierced at two points in a central region using a 200 µL pipette tip. A spore suspension of the wild type isolate (1×106 sp mL−1) was placed on the wound sites and the plate was sealed with sealing film. Plates were maintained in a fluorescent bulb lit growth chamber (Thermoline) at 20°C with 12 hours of light. The whole leaf segment was harvested for RNA extraction. RNA extraction was performed using TRIzol reagent (Invitrogen) according to the manufacturer's instructions. Relative expression was compared to β-tubulin. Primers used for gene expression analysis are shown in Table S4. qRT-PCR was performed as previously described [73].
|
10.1371/journal.pcbi.1000788 | Deciphering Diseases and Biological Targets for Environmental Chemicals using Toxicogenomics Networks | Exposure to environmental chemicals and drugs may have a negative effect on human health. A better understanding of the molecular mechanism of such compounds is needed to determine the risk. We present a high confidence human protein-protein association network built upon the integration of chemical toxicology and systems biology. This computational systems chemical biology model reveals uncharacterized connections between compounds and diseases, thus predicting which compounds may be risk factors for human health. Additionally, the network can be used to identify unexpected potential associations between chemicals and proteins. Examples are shown for chemicals associated with breast cancer, lung cancer and necrosis, and potential protein targets for di-ethylhexyl-phthalate, 2,3,7,8-tetrachlorodibenzo-p-dioxin, pirinixic acid and permethrine. The chemical-protein associations are supported through recent published studies, which illustrate the power of our approach that integrates toxicogenomics data with other data types.
| Exposure to environmental chemicals and drugs may have a negative effect on human health. An essential step towards understanding the effect of chemicals on human health is to identify all possible molecular targets of a given chemical. Recently, various network-oriented chemical pharmacology approaches have been published. However, these methods limit the protein prediction to already known molecular drug targets. New findings can for example be made by using high-confidence protein-protein association databases. Here, we describe a generic, computational systems biology model with the aim of understanding the underlying molecular mechanisms of chemicals and the biological pathways they perturb. We present a novel and complementary approach to existing models by integrating toxicogenomics data, chemical structures, protein-protein interaction data, disease information and functional annotation of proteins. The high confidence protein-protein association network proposed reveals unexpected connections between chemicals and diseases or human proteins. We provide literature support to demonstrate the validity of some predictions, and thereby illustrate the power of an approach that integrates toxicogenomics data with other data types.
| Humans are daily exposed to diverse hazardous chemicals via skincare products, plastic cups, computers and pesticides to mention but a few sources. The potential effect of these environmental compounds on human health is a major concern [1]–[2]. For example chemicals such as phthalate plasticizers have been widely linked to allergies, reproductive disorders and neurological defects. Humans are intentionally exposed to drugs used for treatment and cure of diseases. Many drugs affect multiple targets and may interact or affect the same proteins as environmental chemicals [3]–[5]. The mechanism of action of these small molecules is often not completely understood and can be associated to adverse and toxic effects through for example drug-drug interactions [6]. There is thus a need to improve our understanding of the underlying mechanism of action of chemicals and the biological pathways they perturb to fully evaluate the impact of small molecules on human health.
An essential step towards deciphering the effect of chemicals on human health is to identify all possible molecular targets of a given chemical. Various network-oriented chemical pharmacology approaches have been published recently to identify novel protein candidates for drugs, using structural chemical similarity [7]–[10]. For example Keiser et al. [8] applied network analysis to drugs and their targets. The authors identified unexpected molecular targets such as muscarinic acetylcholine receptor M3, alpha-2 adrenergic receptor and neurokinin NK2 receptor for methadone, emetine and loperamide, respectively. Additionally, recent studies have demonstrated that chemicals could be classified based upon their effect on mRNA expression detected by microarrays [11]–[12]. Lamb et al. showed that genomic signatures could be used to recognize drugs with common mechanism of action allowing discovery of unknown modes of action. Despite the explosion of chemical-biological networks, the chemical toxicity remains a major issue in human health. Analysis of environmental chemicals with similar gene expression profiles is still lacking. With the recent advances in toxicogenomics, information on gene/protein activity in response to small molecule exposures becomes more available. This provide necessary data to develop computational systems biology models to predict both high level associations (linking chemical exposures to diseases) and more detailed associations (linking chemicals to proteins)
In this paper we present a method that can associate chemicals to disease and identify potential molecular targets based on the integration of toxicogenomics data, chemical structures, protein-protein interaction data, disease information and functional annotation. The core of our procedure is derived from the “target hopping” concept defined previously [3]. But instead of considering only binding activity, we extended the concept to gene expression. If two proteins are affected with two chemicals, then both proteins are deemed associating in chemical space. Our approach is not only a statistical model but mimics the true biological system by constructing a network of associations between human proteins defined as Protein-Protein Association Network (P-PAN). We have validated our network by comparison with two high confidence protein-protein interaction (PPI) networks, and by assessing the functional enrichment of clusters in the network generated. The P-PAN revealed both known as well as many novel surprising connections between chemicals and diseases or proteins. We provide literature support for some of the unexpected associations, such as the connection between diethylhexylphthalate (DEHP) and gamma-aminobutyric acid A receptor beta target [13], as well as between apocarotenal, a chemical found in spinach, and necrosis. This illustrates the usefulness of an approach that integrates toxicogenomics data with other diverse data types.
Based on the Comparative Toxicogenomics Database (CTD) [14], we constructed a human P-PAN. A workflow of the strategy is shown on Figure 1. We extracted 42,194 associations between 2,490 chemicals and 6,060 human proteins from the CTD. We mapped compounds to chemical structures from PubChem and extracted their indication of use from Medical Subject Headings (MeSH, http://www.nlm.nih.gov/mesh/MBrowser.html) to classify them as either drugs (MeSH: “Pharmaceutical Actions”) or environmental chemicals (MeSH: “Toxic Actions” and “Specialty Uses of Chemicals”).
In the CTD, drugs and environmental compounds are claimed to be associated with toxicologically important proteins. To estimate how much the information from the CTD differs from available data on pharmacological action of drugs, we compared the data shared between CTD and DrugBank, as of May 2009 [15]. DrugBank is a repository of pharmacological action for ‘Food and Drug Administration’ approved drugs. From the 1358 drugs gathered in DrugBank, 420 drugs matched in CTD. Interestingly, whereas 1403 proteins are associated to these drugs in DrugBank, only 194 proteins are found in both databases. For example, according to Drug Bank celecoxib, a known non-steroidal anti-inflammatory drug, is associated to two metabolizing enzymes: the Cytochrome P450 2C9 (CYP2C9) and the Cytochrome P450 2D6 (CYP2D6) and to two drug targets: the Prostaglandin G/H synthase 2 (COX-2) and the 3-phosphoinositide-dependent protein kinase 1 (PDPK1). In the CTD, celecoxib is linked to 33 human proteins including CYP2C9 and COX-2. The toxicity information extracted from CTD is relatively different to the known pharmacological action of drugs and should be considered as a complementary source of information.
To investigate the assumption that two compounds sharing similar structure can potentially affect the same molecular targets, we compared chemical properties of the compounds collected from the CTD. The chemicals were characterized by 50 properties calculated from the structure, including the molecular mass and affinity for a lipid environment. The distribution of properties, as it appears in a multi-dimensional properties space, was projected and visualized in two dimensions using principal component analysis (PCA) (shown in Figure 1). There is substantial overlap in the PCA projections between environmental chemicals and drugs indicating that they can potentially affect the same protein targets. We also compared the oral bioavailability profiles of compounds based on standard Lipinski [16] and Veber [17] rules. Again, overlaps were observed, indicating that environmental chemicals mimic drug properties (see Figure S1). These results confirm that it is reasonable to generate a network by integrating toxicogenomics knowledge from both drugs and environmental compounds, as they share many properties.
The human P-PAN was generated based on the assumption that if two proteins are biologically affected with the same chemicals (defined as shared chemicals), they are likely to be involved in a common mechanism of action of the chemicals. Then, two proteins are connected to each other if they are linked to the same chemical in the CTD. The resulting P-PAN consists of 2.44 million associations. To reduce noise and select the most significant associations, we assigned two reliability scores to each protein-protein association: a score based on hypergeometric calculation and a weighted score. The weighted score was calculated as the sum of weights for shared chemicals, where weights were inversely proportional to the number of associated proteins for a given compound.
We went one-step further and compared the P-PAN with two human PPI databases: (1) a high confidence set of experimental PPIs extracted from a compilation of diverse data sources [18] and (2) PPIs based on an internal consistent single data source [19]. Our P-PAN performed well compared to both PPIs. Based on the calibration curves (Figure S2), we considered a threshold that capture good overlaps between our P-PAN and the PPI networks for different reliability scores thus reducing our P-PAN to ∼200,000 reliable associations. Using this approach, the molecular target predictions are limited to the 3,528 proteins present in the P-PAN. To confirm that biological information is not lost when selecting only 8% of the entire P-PAN, we compared functional enrichment for the complete network (6,060 proteins) and for the high confidence sub-network (3,528 proteins) using Gene Ontology (GO) [20]. For example cell proliferation (p-values of 3.22e-36 and 1.46e-27 for the large network and the sub-network, respectively) and protein binding (p-values of 1.2e-72 and 4.13e-47 for the large network and the sub-network, respectively) were the most overrepresented terms.
Since proteins tend to function in groups, or complexes, an important step has been to verify that our high confidence network mimics true biological organization. This task is commonly executed using graph clustering procedures, which aim at detecting densely connected regions within the interaction graph. Two clustering methods have been applied to our network. The molecular complex detection (MCODE) approach [21] that allows multiple clusters assignation for a protein, mimicking the reality as a protein can participate in several complexes simultaneously. On the other hand, the markov cluster algorithm (MCL) [22] which assign one protein to a unique cluster has been shown to be superior to other graph clustering methods in recent studies [23]–[24]. Applied on our network, MCODE extracted few large core clusters and several tiny clusters (possibly singleton clusters). The MCODE approach results in a clustering arrangement with a weak cluster-wise separation. Compared to MCL, MCODE yielded a lower number of clusters, with a higher number of proteins per cluster. Only 35 clusters varying in size from five to 845 proteins were extracted. Using the MCL algorithm we obtained a more heterogeneous separation with 58 clusters varying in size from five to 462 proteins. Therefore, to identify the biologically meaningfulness of our network, we used complexes extracted using the MCL method. Each cluster was then investigated for functional enrichment based on GO terms. To ensure the high quality of functional annotations we used only annotations experimentally supported or with traceable references. Hypergeometric testing was used to determine GO functional annotation overrepresented amongst each cluster. The two top scoring molecular functions found were heme binding (p-value of 6.60e-25, cluster 4) and glucuronosyl transferase activity (p-value of 2.34e-21, cluster 12). Regulation of apoptosis (p-value of 1.67e.17, cluster 2) and oxidation reduction (p-value of 6.67e-14, cluster 4) were the most highly enriched categories in the biological process branch of the GO. This analysis thus confirms that clusters in the network, and therefore the proteins associated with each other, are functionally coherent. This was further evidence that the organization of the network is meaningful.
In the clusters of the P-PAN, proteins are more connected with other proteins within the cluster than with the other targets in the network. As proteins are associated based on their shared relationship with chemicals, proteins within a given cluster tend to be more linked to specific compounds. It is thus possible to find associations between diseases and the chemicals that underlie the protein-protein associations within the cluster using protein-specific disease annotations. For each cluster, we investigated if specific disease annotation was found more frequently than expected by using protein-disease information [25]. We identified several diseases associated with specific clusters. These included the two most common types of cancer, breast cancer (cluster 1, p-value of 9.67e-18) and lung cancer (cluster 12, p-value of 4.84e-12), as well as necrosis (cluster 2, p-value of 2.26e-12), ichthyosis (a skin disorder associated to cluster 4, p-value of 1.41e-5), retinoblastoma (cluster 7, p-value of 9.46e-8) and inflammation (cluster 8, p-value of 1.55e-5).
To predict which chemicals may affect human health, we then analyzed selected clusters to identify new chemical-disease associations (see Table 1). When linking diseases to compounds, it is important to keep in mind that there is no direction in the association, i.e. it is not possible from the network to separate positive from negative associations between a chemical and a disease. Discriminating between whether a compound prevents or causes disease requires manual interpretation of the association.
One of the clusters showed high enrichment for breast cancer. The most significantly associated chemicals are already known from the literature to be related to cancer, thus supporting the clustering quality of the P-PAN. Among the most significantly associated chemicals are the well-known polychlorinated biphenyls (PCBs). PCBs are used for a variety of applications i.e. flame retardants, paints and plasticizers. After being banned due to their toxicity, they still persist in the environment. Previous results suggest that specific PCBs may indeed be associated with breast cancer [26]. Several organizations (EPA, IARC) have classified PCBs as probable human carcinogens. When we inspected another cluster highly connected to lung cancer using our P-PAN method, thimerosal, dinitrochlorobenzene (DNCB) and styrene were significantly associated with this cluster. Thimerosal and DNCB are not known lung cancer-causing chemicals, while the last compound, styrene has been classified as a possible carcinogen. Thimerosal is an organomercury chemical widely used as preservative in health care products and in vaccines. It may have possible adverse health effects such as a role in autism and in nervous system disorders [27] as well as possible gene-toxic effects to human lymphocytes [28]. No study has previously related it to lung cancer. The second chemical DNCB is known to be a skin allergen that may cause dermatitis. Genes associated with allergies were shown to be up regulated in rat lung tissue after DNCB exposure [29], but no direct link to lung cancer has been demonstrated so far. Another interesting finding is the association between apocarotenal and necrosis. Apocarotenal, a natural carotenoid found in spinach and citrus, is used as a red-orange coloring agent (E160E) in foods, pharmaceuticals and cosmetics products. No direct evidence has been found that links apocarotenal to necrosis. However, in vitro and in vivo studies [30] have suggested that spinach may be a good anti-cancer agent. This is in line with epidemiologic studies that have shown that those who consume higher dietary levels of fruits and vegetables have a lower risk of certain types of cancer [31] due to the presence of carotenoids. Furthermore, carotenoids have been defined as chemopreventive agents [32]. Studies have established associations between carotene and beta-carotene with reduced risk of prostate cancer [33] or breast cancer [34]. The prediction that apocarotenal is positively associated to necrosis and could prevent certain types of cancer is thus indirectly supported by other studies. The other chemicals significantly associated to disease (Table 1) are discussed in the supplementary text (see Text S1).
Besides revealing disease-chemical associations, the network can be used to predict novel targets for chemicals. It has been shown that many small molecules affect multiple proteins rather than a single target, and that proteins sharing an interaction with a chemical are targeted by the same chemicals [8]. Based on the CTD data available, strong promiscuities between some proteins exist. For example, more of 25% of chemicals annotated to estrogen receptor 1 (ESR1) affects also progesterone receptor (PGR). In the same order, cytochrome p450 2D6 (CYP2D6) and cytochrome p450 2C9 (CYP2C9) shared one-third of their respective associated compounds. By the term “affected”, we consider effects such as up regulated, down regulated, agonist, antagonist and inhibitor. Then, our network can not be used to identify chemical synergies or opposite effect on proteins. Thus, if two proteins are affected by two chemicals and one of the proteins is further deregulated by an additional chemical, then it might be that both proteins are in fact deregulated with the same three chemicals. Based on this assumption and in order to suggest novel associations between chemicals and proteins, a neighbor protein procedure was used which scored the association between each protein and each chemical (see Materials and Methods). Molecular targets known to be associated with a chemical were extracted from the CTD, and the P-PAN was scanned for proteins associated with a high score. The significance of enrichment was calculated by random testing (for the confidence scores see Text S2), and sub-networks were subsequently ordered according to their significance. Four examples of various chemicals are presented in Table 2 (other case stories are shown in Table S1). To estimate the performance of our approach for approved drugs, we analyzed the level of recall and precision obtained for the 420 common drugs between DrugBank and CTD. We obtained a recall and a precision of 5.91% and 3.77% respectively, corresponding to the percentage of interactions in DrugBank retrieved and percentage of interactions in DrugBank from all interactions predicted obtained from CTD data and from the neighbor protein procedure. These values illustrate that information between the two data sources are relatively different.
Phthalates, mainly used as plasticizers, have received a lot of attention as environmental compounds because they are potential human carcinogens. As there are many phthalates, we focused on Di-EthylHexyl Phthalate (DEHP) that has been associated with more proteins compared to other phthalates such as additional information on kinases (e.g. mitogen-activated protein kinase 1, and mitogen-activated protein kinase 3) [35]. DEHP is widely used due to its suitable properties and low cost, and is present in the general environment at high levels. Exposure to DEHP is of particular concern with regard to developing fetuses where it is believed to cause malformation of reproductive organs and neurological defects [36]. Using our approach, several proteins were identified as being associated with DEHP (Table 2). Cysteine dioxygenase type I (CD01) and peroxisome proliferator-activated receptor alpha (PPARA), the two top scoring proteins, are already known in the CTD and from the literature [37]–[38] as molecular targets for DEHP. Six other high ranking proteins are new potential DEHP molecular targets which are not recorded in the CTD (thus not input data). Among them, four gamma-aminobutyric acid A (GABA) receptors were predicted as potential DEHP molecular targets. These associations are supported by a recent study showing that DEHP can modulate the function of ion channels as GABA receptors in a manner similar to volatile anesthetics in experiments on expressed receptors [13]. This makes sense because the GABA neurotransmitter system has been implicated in the pathogenesis of bipolar disorders (neurological disorders) via gamma-aminobutyric acid receptor subunit alpha-1 (GABAα1) [39], and DEHP is also associated with neurological defects [36]. In addition to GABA receptors, we identified several other candidates including proopiomelanocortin (POMC) and a cytochrome P450 (CYP3A11) (discussed in the Text S2). We looked at another environmental chemical, the 2,3,7,8-TetraChloroDibenzo-p-Dioxin (TCDD), which originates from burning or incineration of chlorinated industrial compounds. TCDD is believed to cause a wide variety of pathological alterations, with the most severe being progressive anorexia and body weight loss [40]. TCDD is also known to be a neurotoxin leading to neurodevelopmental and neurobehavioral deficits [41]–[42], and accumulating in the brain as well as other organs [43]. We identified six proteins associated with TCDD that are not recorded in the CTD for human (Table 2). Among them five are supported by literature (see Text S2). This included protein kinase C elipson (PRKCE), known to be involved in brain tumors [44], carnitine palmitoyltransferase I (CPT1A), 11β-hydroxysteroid dehydrogenase type 1 (HSD11B1) and apolipoprotein B (APOB) which are all linked to obesity [45]–[47]. Furthermore, we investigated in detail the drug pirinixic acid (PA) (also named WY14,643), which is a peroxisome proliferator-activated receptor (PPAR) agonist with strong hypolipidemic effects. PA was never approved for clinical use due to hepatocarcinogenesis adverse effect shown in animal studies [48]. To date there is no evidence that PA promotes carcinogenesis in humans [49], and this has spurred new studies for identifying cellular processes that are capable of responding to PA. Among 11 molecular targets identified and not recorded in the CTD (Table 2), only five are supported by the literature (see Text S2). For example the expression of the C3 protein, an acylation stimulating protein involved in necrosis and afibrinogenemia (blood disorders), has been shown to be affected by PA in rats [50]. Finally we studied proteins associated with permethrin in more detail. Permethrin is a widely used insecticide, acaricide and insect repellent, classified by the US EPA as a likely human carcinogen, but still used in healthcare for the treatment of lice infestations and scabies. Four proteins not recorded in the CTD were identified as associated with permethrin. Three of them are supported by literature (see Text S2 for details) including a cytochrome P450 (CYP2B1) [51]–[52] and sex hormone-binding globulin (SHBG) [53], which are proteins linked to the endocrine system. These findings suggest a mechanism by which chronic exposure of humans to pesticides containing this compound may result in disturbances in endocrine effects related to androgen action.
The examples we provide include both known and new protein associations with a given chemical, and many of the novel associations are supported by the literature. We compared our approach with STRING (version STRING 1) [54] a high-confidence protein-protein association network, to see if the findings generated by the current approach are also found by other existing methods. The STRING network includes direct (physical) and indirect (functional) associations derived from diverse sources as genomic context, high throughput experiments, co-expression and literature. As a test example, we used the 15 proteins associated with DEHP in the CTD to query the P-PAN by a neighbor protein procedure. The same 15 proteins were also used to query the STRING network. Subsequently we compared the predicted molecular targets between the two networks (P-PAN and STRING). In the resulting STRING network none of the GABA receptors were found (see Figure S3). The STRING network showed a clear tendency to associate phthalates with kinases and nuclear receptors. This example demonstrated that our approach was complementary to other association approaches. This highlights the value of integrating various sources of data to understand potential toxic effects on human health caused by chemical exposure.
We propose an approach different from existing computational chemical biology networks, which primarily integrate drugs information, to identify new molecular targets for chemicals and to link them to diseases. In our approach we have integrated toxicogenomics data for drugs and environmental compounds. The ability to make new findings using a different network is illustrated by a comparison with a similar method, showing the capacity of our P-PAN to identify novel chemical-protein associations. Using phthalate as an example, our model suggests potential associations between DEHP and GABA receptors, which have not been predicted previously.
An extension of this network by integrating more data, for example other chemical-protein associations or dose levels for which a compound may affect human health, would be beneficial to the proposed approach. Paracelsus (1493–1541) is often cited for his quote, “all things are poisons and nothing is without poison, only the dose permits something not to be poisonous”. This emphasize that the dose of a chemical is an issue to consider in the deregulation of systems biology. Nevertheless, a global mapping could allow a better understanding of adverse effects of drugs and toxic effects of environmental compounds. This could be used as a new approach for risk assessment and regulatory decision-making for human health.
Among the examples presented, some predictions are supported by literature for other organisms. Regarding toxicogenomics, the available human data are generally sparse compared to rodents. Data on toxicity - adverse effects of chemicals on humans – can be acquired through epidemiologic studies and from occupational, accident-related exposures as intentional human testing of environmental compounds remains limited. However, differences exist between model animal and human responses to chemicals, including differences in the type of adverse effects experienced and the dosages at which they occur. The differences may reflect variations in the underlying biochemical mechanisms, in metabolism, or in the distribution of the chemicals. As an example, bisphenol A (BPA) does not affect proteins in a similar way across species (Figure 2). In the human systems studied to date, BPA does not affect the proto-oncogene c-FOS (FOS) and the mitogen-activated protein kinase 8 (MAKP8) but seems to modify their expression in rodent species. BPA binds and modifies the activity of the estrogen receptor alpha (ESR1) in a very conservative way across organisms [14]. BPA has an ability to function as an estrogen like receptor (ER) agonist, and thus has the potential to disrupt normal endocrine signaling through regulation of ER target genes e.g. androgen receptors, estrogen receptor, progesterone receptors. There is a need to integrate data with cross-species extrapolation in order to have a more accurate understanding of the human risk from chemical exposure.
The major limitation of our integrative systems biology approach is that the molecular target predictions are limited to the 3,528 proteins present in our P-PAN, which represent only 15% of the estimated human proteome [55]. Hence, the current lack of high quality data is the limiting factor in approaches such as the one described here. Today high throughput methodologies result in available large scale data in both chemical biology and systems biology, but these data are discipline specific [56]. There is an evident need for the development of databases [57] to integrate disparate datasets such as toxicogenomics data in order progress in systems biology research. In addition, the results of the disease-compound association analysis will improve in the future as newer, more complete and curated data will become available.
We downloaded the publicly available Comparative Toxicogenomics Database (CTD) as of June 26, 2008 [14]. The CTD contains curated information combining drug and environmental chemical data associated with proteins. We selected 42,194 associations between 2,490 unique compounds and 6,060 molecular targets known to be involved in human disease. Different associations are presented in the CTD such as “chemical x results in increased expression of protein z” or “compound x binds to protein z”. Gene expression data are essentially present in the CTD such as a chemical can increase, decrease or affect a gene expression. However, only few binding data are present in CTD and therefore integrated in our network: 3189 in total among the 42,194 associations. Scripts were used to remove associations with negation such as “chemical x does not affect protein z”.
To verify the uniqueness of chemicals, chemical names extracted from the CTD were checked using PubChem (http://pubchem.ncbi.nlm.nih.gov/) as of June 26, 2008 to avoid synonymous names for the same compound. The few chemical names not retrieved via the database were manually verified. To determine overlaps with protein-protein interaction databases and facilitate further data integration, the CTD protein names were mapped to the corresponding Ensembl IDs [58] as of June 26, 2008. Only 1.5% of the 42,194 chemical-protein associations could not be clearly identified.
To investigate chemical space of drugs and environmental compounds, 50 two-dimensional properties were calculated for each structure extracted from PubChem. To visualize them, principal component analysis (PCA) was performed. All necessary data were calculated using the MOE software (Chemical Computing Group version 2007.09)
Relevant human chemical-protein associations collected from the CTD were used to create a P-PAN. The maximum number of molecular targets assigned to one compound ‘tert-Butylhydroperoxide’ was 1,189 and the maximum of chemicals assigned to one protein, the cytochrome P450 3A4 (CYP3A4), was 276. The P-PAN was generated by instantiating a node for each protein, and linking by an edge any protein-protein pair where at least one overlapping chemical was identified. Scripts were used to convert the protein-protein associations into a non-redundant list of associations. If proteins A and B are associated, the network may have two associations, A–B and B-A. Only one of these associations was retained in the P-PAN. We assigned two reliability scores to each protein-protein association: a score based on hypergeometric calculation and a weighted score. The weighted score was calculated as the sum of weights for overlapping compounds, where weights were inversely proportional to the number of assigned proteins. The resulting P-PAN is a complex structure containing a total of 2.44 million unique associations between 6,060 human proteins.
The reliability of the weighted score was confirmed by fitting a calibration curve of different scores against Lage's PPIs18 (version 2.9) and Vidal's PPIs19. Only 35,000 high confidence experimental interactions were extracted from Lage's PPI, which contains interactions present in the largest databases (Reactome, KEGG…) and data inferred from model organisms. Vidal's PPIs are based on an internal consistent single data source defined using yeast two-hybrid system and contains 3111 interactions.The overlaps of our P-PAN scores and Lage/Vidal PPIs are shown in Figure S2. The benchmark revealed that the weighted score is superior to a score calculated as the negative logarithm of p-values from a test in hypergeometric distribution and a simple overlap count. To estimate the robustness of the model, four thresholds selected from the ‘weighted score’ curves (5%, 8%, 12.5% and 17%) of the complete P-PAN were used to perform prediction for DEHP. At 5%, 73,000 associations between 2105 proteins were extracted. The number of proteins is relatively stable at 8% and 12.5%. However, the number of associations increased significantly from 200,080 to 306,000 including lower score associations in the output file of prediction. The threshold of 17% corresponds to 415,000 associations between 3894 proteins. All thresholds showed a good prediction with the GABA receptors for DEHP. As the 12% threshold already added some more noise in the prediction, we decided to not include more proteins, in order to keep the most significant associations. We then considered a threshold of 8%, represented by the vertical line in Figure S2, which captured a good overlap between our P-PAN and the PPI networks. This selection represents 200,080 associations of the complete P-PAN.
Among the ∼200,000 high confidence associations selected, 3,528 proteins were identified, and these were significantly enriched among the high scoring protein-protein associations as shown in Figure S2 (861 Lage's PPI interactions corresponding to 24.4% were found among the top 5% of the high scoring protein-protein associations). By comparison, only 1,852 of the high confident interactions from Lage were identified in a random P-PAN created by node permutation, and no enrichment was seen for the random network. As example, the selection of high confidence associations allowed to conserve only 803 proteins from the 1189 proteins assigned to the ‘tert-Butylhydroperoxide’.
A high confidence sub-network of ∼200,000 protein-protein associations was selected which contained 3,528 proteins. This sub-network was highly interconnected, with the majority of proteins belonging to a single large cluster. In order to increase the resolution and facilitate biological interpretation, two clustering methods were applied to the sub-network, MCODE [21] and MCL [22]. We used the default settings for MCODE (fluff option set to 0.1, mode score cutoff set to 0.2, degree cutoff set to 2), and obtained 35 clusters. One major drawback of this algorithm is that not all the proteins in the network were clustered. We used the MCL algorithm with scheme and granularity parameters set to 7 for highest performance and granularity. With the MCL approach we identified a total of 58 clusters as strongly interconnected, with a minimum size of 5 proteins. These clusters were linked together into a new network consisting of a scored cluster-cluster association network. The association score between each cluster pair was calculated from the mean of the P-PAN between each pair of clusters. Each cluster was investigated for functional analysis based on the three Gene Ontology categories (a) molecular function, (b) biological processes, and (c) cellular components as of January 2009. To reduce the noise and improve the quality of the functional annotation, we only used the functional annotation if it was experimentally supported or had traceable references. The following GO evidence codes were allowed: IMP (Inferred from Mutant Phenotype), IGI (Interfered from Genetic Interaction), IPI (Inferred from Physical Interactions) and IDA (Inferred from Direct Assay) and TAS (Traceable Author Statement). At time of use the molecular function category contained 5,981 proteins, the biological processes category 5,196 proteins, and the cellular components 5,151 proteins. We compared human proteins present in GO categories with proteins extracted from the CTD; 14.3% of the CTD proteins could not be annotated for the molecular function, 16.6% for biological processes and 14.9% for cellular components.
To identify chemicals associated with disease, protein-specific information such as involvement in disease was integrated in each cluster. The Online Mendelian Inheritance in Man database (OMIM) [59] (July, 2009) and the GeneCards database [25] (February, 2008) were considered as sources of protein-disease connections. Various clusters were investigated. For example, cluster 1 contained 462 proteins. Using GeneCards, 269 proteins were retrieved with disease annotations. Amongst these 269 proteins, 128 were associated to breast cancer (with give a p-value of 9.67e-18 for breast cancer to cluster 1). Using OMIM, only 90 proteins among the 462 were retrieved with disease annotations. Looking at the cluster enrichment with OMIM, we obtained at the top a non significant p-value of 0.0048 (corresponding to two proteins for paget disease of bone). As another example, we analyzed the second cluster. Cluster 2 contained 433 proteins. 281 proteins were annotated to diseases in Genecards, for only 78 proteins in OMIM. Additionally, cluster 2 has a significant p-value of 2.26e-12 using GeneCards information for necrosis. According to these results we decided to use GeneCards as a source of protein-disease relationships. To avoid too many false positive from Genecards, we set a significance cut-off value of the GeneCards-AKS2 score based on a comparison with OMIM. This was done by overlapping common protein-disease associations from Genecards against OMIM (see Figure S4). The protein-disease connections were kept with a minimum AKS2 score of 60 and p-values were calculated for each disease present in clusters. Then, chemical information from the CTD was integrated with each cluster and p-values were assigned to each chemical. All p-values obtained were calculated using hypergeometric testing, and were corrected for multiple testing with Bonferroni correction [60]. The significance cutoff for the corrected p-values was set to 0.05.
To predict molecular targets for a chemical, a network-neighbor's pull down was done in a three steps procedure: (1) Selection of the input protein(s): Extraction of the protein(s) known to be associated with the selected chemical from the CTD. (2) Identification of network(s) surrounding the input proteins by a neighbor proteins procedure. In this procedure, our P-PAN was queried for the input proteins, and associations between these were added. Next, the first order interactors of all the input proteins were queried and added. For each neighbor, a score was calculated taking into account the topology of the surrounding network, based on the ratio between total associations and associations with input proteins. Molecular targets with a score higher than the threshold (0.1) were kept in the final sub-network(s). This node inclusion parameter is in the conservative end of the optimal range for protein-protein interaction networks18. As a final step all proteins in the complex were checked for associations among them and the missing one were added. (3) Establishment of a confidence score for the surrounding network (cscore) and of a score for each protein (cpscore): Each of the pulled down complexes was tested for enrichment on our input set by comparing them against 1.0e4 random complexes for the protein-protein association set to establish a cscore for each sub-network and a cpscore for each connected proteins. The cpscore was used to rank proteins to select potential molecular targets for chemicals. An illustration of cpscore is available on Table S2 for approved drugs.
All the CTD human protein-chemical associations were extracted from the CTD on June 26, 2008. Subsequent updates of CTD, as of June 25, 2009, did not change the overall trends or conclusions of the present study.
|
10.1371/journal.ppat.1007561 | The switch between acute and persistent paramyxovirus infection caused by single amino acid substitutions in the RNA polymerase P subunit | Paramyxoviruses can establish persistent infections both in vitro and in vivo, some of which lead to chronic disease. However, little is known about the molecular events that contribute to the establishment of persistent infections by RNA viruses. Using parainfluenza virus type 5 (PIV5) as a model we show that phosphorylation of the P protein, which is a key component of the viral RNA polymerase complex, determines whether or not viral transcription and replication becomes repressed at late times after infection. If the virus becomes repressed, persistence is established, but if not, the infected cells die. We found that single amino acid changes at various positions within the P protein switched the infection phenotype from lytic to persistent. Lytic variants replicated to higher titres in mice than persistent variants and caused greater infiltration of immune cells into infected lungs but were cleared more rapidly. We propose that during the acute phases of viral infection in vivo, lytic variants of PIV5 will be selected but, as the adaptive immune response develops, variants in which viral replication can be repressed will be selected, leading to the establishment of prolonged, persistent infections. We suggest that similar selection processes may operate for other RNA viruses.
| As well as causing acute infections that result in mild to serious disease, many RNA viruses can establish prolonged or persistent infections in some infected individuals, that occasionally lead to chronic or reactive disease. Little is known about the molecular mechanisms involved in the establishment of such infections. Using parainfluenza virus type 5 (PIV5) as a model, we show how lytic and persistent variants of the virus can be selected on the basis of single amino acid substitutions and propose that the selection of persistent variants as the adaptive immune response develops following an acute infection might be a mechanism these viruses have evolved to enhance their transmission rates. As well as being of fundamental interest, understanding the molecular basis by which RNA viruses establish persistent infections may improve our understanding of virus epidemiology (and hence improve the control of virus infections) and of virus:host interactions that influence the relationship between virus persistence and chronic/relapsing disease. Furthermore, the knowledge of how RNA viruses, such as PIV5, establish persistent infections may lead to improve vaccine design since vectors which can establish persistent infections may induce longer-lasting more robust immunity.
| Paramyxoviruses are primarily known for the acute infections and associated diseases they cause, such as mumps, measles and respiratory illnesses. However, under certain conditions they can establish persistent (or prolonged) infections [1], which can be considered as infections that continue for longer than would be expected from a prototypical acute infection (2–3 weeks). For example, both immunocompromised and immunocompetent patients have been shown to shed parainfluenza virus (PIV) 2, 3 and 4 for weeks and even years after infection [1], whilst dogs infected with canine distemper virus can shed virus for months after initial infection [2]. Prolonged/persistent infections can lead to chronic diseases, for instance subacute sclerosing panencephalitis (SSPE) is associated with measles [3], postviral olfactory dysfunction is associated with PIV3 [4] and chronic kidney disease is associated with feline paramyxovirus [5, 6].
Little is understood of the molecular mechanisms by which paramyxoviruses establish and maintain persistent infections. Like other viruses, they must avoid elimination by the host immune response, while maintaining their genomes in at least some infected cells. Since it is highly probable that continual high-level viral replication in a cell will either directly kill the cell or lead to viral recognition and killing by innate and adaptive immune responses, it is likely that viral replication must be at least partially repressed in some cells in order for a virus to establish a persistent infection [7].
The interferon (IFN) response, and the ability of viruses to circumvent it, is also likely to play an important role in the establishment and maintenance of persistent infections. Thus, patients who have a defective ability to respond to type I IFN can become persistently infected with measles, mumps and rubella viruses following MMR vaccination with serious sequelae [8, 9]. On the other hand, we have suggested that the IFN response may also dampen down viral replication in some cells, thereby facilitating persistence [10, 11]. Using parainfluenza virus type 5 (PIV5) as a model, we show here that viral replication at late times after infection can also be repressed in an IFN-independent mechanism, thereby leading to the establishment of persistent infections.
PIV5 (previously known as simian virus 5; species Mammalian rubulavirus 5, genus Rubulavirus, family Paramyxoviridae) has been isolated from numerous mammals including, humans, primates, pigs, ungulates (cattle), dogs [12, 13] and lesser pandas [14]. There is also some evidence that cats, hamsters, rabbits and guinea pigs can be infected [15, 16]. The association of PIV5 with acute disease is often not obvious, although the virus causes kennel cough in canines [17] and may, or may not, cause acute respiratory symptoms in pigs [18, 19] and calves [20]. PIV5 can also cause unsuspected persistent infections of tissue culture cells, including the AGS line [21, 22], guinea pig kidney and mouse fibroblast cells [23], and is likely to establish persistent infections in vivo [24–26].
PIV5 has a non-segmented, negative-sense RNA genome of 15,246 nucleotides (nt) containing seven tandemly arranged genes, that encode eight proteins, flanked by 3’-leader and 5’-trailer sequences at the genome ends. From the 3’-leader sequence, the genome encodes the nucleocapsid protein (NP), V protein (V), phosphoprotein (P), matrix protein (M), fusion protein (F), small hydrophobic protein (SH), haemagglutinin-neuraminidase (HN), and the large protein (L). The genomic RNA is encapsidated by NP, forming a flexible helical nucleocapsid complex that is associated with the viral RNA-dependent RNA polymerase complex (vRdRP) consisting of L and P (extensively reviewed in ref [27]). Previous work has highlighted the importance of the phosphorylation of the P protein in regulating the activity of the vRdRP and influencing viral growth [28]. Furthermore, P plays a role in limiting the induction of host cell responses by influencing the fidelity of viral RNA synthesis [29]. The use of mass spectrometry has identified multiple sites on the P protein that can be phosphorylated, including serine residues at positions 36, 126, and 157 and a threonine residue at position 286 [30]. In addition, it has been shown that host cell Polo-like kinase 1 (PLK1) can phosphorylate a serine residue at position 308 [31]. Mutation of the serine residues to alanine residues at either position 157 or 308, thereby preventing phosphorylation at these residues, significantly enhanced the activity of the vRdRP in mini-genome assays and the replication of recombinant viruses that bear these mutations [31]. In contrast, phosphorylation of the threonine residue at position 286 may enhance viral replication, since mutating it to an alanine residue reduced vRdRP activity and viral growth [30]. Here we show that amino acid substitutions, found in natural isolates of PIV5, at residues 157 and 308, as well as at other sites, including some which cannot be phosphorylated, also influence the activity of the vRdRP in mini-genome assays and determine whether replication is or is not repressed at later times post infection leading to lytic or persistent phenotypes respectively. As only single nucleotide changes in all the wild-type isolates of PIV5 sequenced are predicted to convert them from lytic to persistent phenotypes, or visa versa, we propose that the selection of lytic or persistent variants, from a quasispecies population (defined as the mutant distributions that are generated upon replication of viruses in infected cells and organisms [32]), at early and late times after infection, respectively, may be a mechanism that PIV5, and some other RNA viruses, have evolved to increase their success of transmission.
In contrast to infection with most strains of PIV5 held within our laboratory (see below), a high multiplicity of infection (moi) of A549, MRC5 or Vero cells with the PIV5-W3 isolate led to >90% of the cells surviving to become persistently infected; these infected cell-lines can be readily passaged (S1 Fig). We therefore decided to investigate the molecular events that lead to the establishment and maintenance of PIV5-W3 persistence in the expectation that this may lead to a better understanding of paramyxovirus persistence in vivo. Initially, we monitored the synthesis of viral proteins and the levels of viral mRNA and genomic RNAs following a high moi of A549 cells (Fig 1). Ongoing viral protein synthesis at various times post-infection (p.i.) was visualised by metabolically labelling A549 cells that had been infected with PIV5-W3 at a high moi, with [35S]-L-methionine for 1h (Fig 1, panel a). At 24h post infection (p.i.) NP and M were synthesised at sufficiently high levels to be clearly detectable above the background of cellular protein synthesis, which is not significantly repressed in PIV5-W3-infected cells. However, by 48 and 96h p.i., synthesis of NP and M had fallen below the levels that could be detected above the background of cellular protein synthesis. Immune precipitation analysis revealed that even by 36h p.i. there was an obvious reduction in the synthesis of all viral proteins (S2 Fig). In contrast, immunoblot analysis of the same samples showed that the relative levels of (accumulated) NP were slightly higher at 96h p.i. than at 24h (Fig 1, panel b), even though at 96h p.i. there was little, if any, de novo virus protein synthesis (Fig 1, panel a).
It was possible that the decrease in viral protein synthesis observed between 24 and 48h p.i. was due to an IFN-induced antiviral state within the infected cells. Although, this appeared unlikely because MxA, an IFN-induced protein, was not induced in the infected A549 cells (Fig 1, panel b), we monitored the kinetics of viral protein synthesis in cells that were infected in the presence of ruxolitinib (an inhibitor of JAK1 that blocks IFN signalling [33]). There were no changes in the switch-off kinetics in the presence of ruxolitinib (Fig 1, panel a). Furthermore, viral protein synthesis was also switched off with the same kinetics in A549/Npro cells which cannot produce IFN because BVDV-Npro targets IRF3 for proteasome-mediated degradation ([34]; S3 Fig), confirming that the decrease in virus protein synthesis observed was independent of the IFN response.
To investigate whether the observed switch-off of viral protein synthesis was due to inhibition of viral transcription, we used high-throughput sequencing (HTS) to quantitate the levels of viral mRNA and viral genomic RNA during infection. Maximal levels of viral transcription were observed between 12 and 18h p.i., at which times the amount of viral mRNA comprised almost 5% of total cellular mRNA (Fig 1 panel c). Thereafter, the amount of viral mRNA slowly declined such that by 96h p.i. it amounted to less than 0.2% of total cellular mRNA, thus indicating that it was the reduction in viral mRNA that was responsible for the observed switch-off of viral protein synthesis (note: although viral mRNA and antigenome sequences cannot be distinguished by directional sequencing, antigenome sequences contributed <2% of the total viral mRNA and antigenome reads, see Fig 1C legend). Levels of viral genomic RNA continued to increase until 48h p.i. (Fig 1, panel c). Strikingly, high levels of genomic RNA were also present at 96h p.i., by which time there was very little viral transcription occurring. Defective virus genomes (DVGs) were not detected by HTS at 96h p.i. (or in persistently infected cultures) suggesting that they do not play a role in the switch-off of viral transcription and protein synthesis (discussed in greater detail below).
The switch-off of viral protein synthesis could also be inferred from immunofluorescence studies (Fig 2, panel a) aimed at visualising the presence within infected cells of HN (which has a half-life of 2.5 hours [35]) and NP (which has a half-life of days). At 24h p.i. all cells were strongly positive for both NP and HN. However, by 96h p.i., while all the cells remained positive for NP, less than 50% of the cells were also positive for HN, and many of those were only weakly positive (Fig 2, panel a). Since HN possesses neuraminidase activity, the levels of HN expression at later times p.i. could also be inferred by staining cells for the presence of sialic acid. At 24h p.i. none of the infected cells expressed detectable amounts of sialic acid on their surface (Fig 2, panel b). However, by 72h p.i. some cells were positive for sialic acid. The fact that a high proportion of cells were negative for HN, and positive for sialic acid, strongly suggests that there was little or no ongoing viral protein synthesis in these cells at late times p.i.. These results demonstrate a degree of cellular asynchrony in the relative levels of viral gene expression at late times p.i..
Cells persistently infected with PIV5-W3 grew slightly slower than uninfected cells for the first couple of passages but by passage 3 (p3) replicated as fast as uninfected cells and showed few visual signs of being infected. Immunostaining of persistently infected cells at p3 showed that whilst all the cells were infected there was heterogeneity in expression of the HN protein (Fig 2, panel b). Thus, some cells were positive for HN and negative for sialic acid, and others were negative for HN and positive for sialic acid. All cells were positive for NP and P expression, although the amount of NP and P present in the cells varied considerably (Fig 2, panels b and c). In general, cells that were strongly positive for NP (and P) were negative for sialic acid, and those that were weakly positive for NP (and P) were positive for sialic acid. HTS of the cells at 96 h p.i., and of persistently infected cells, showed that viral mRNA constituted less than 0.2% of the total RNA (Fig 1, panel c). This level of viral mRNA characterised the population of cells as a whole but, given the heterogeneity of virus expression, the levels of viral mRNA must have varied considerably among cells within the persistently infected population. These results therefore strongly suggest that within the persistently infected population as a whole, active viral transcription was occurring in some cells (HN-positive cells) but was largely, if not completely, repressed in others (HN-negative cells).
We next investigated whether infectious virus could be rescued from cells in which viral transcription and replication were repressed. Persistently infected cells at passage 3 were stained for surface expression of the HN protein, and FACS was used to sort HN-positive and HN-negative cells into individual wells of 96 well plates (Fig 2, panel d). Colonies from each population were grown out in the presence, or absence, of a pool of PIV5-neutralizing antibodies to prevent viral spread between cells. Immunofluorescence showed that all the colonies remained infected (Fig 2, panel e), regardless of whether the sorted cells were derived from HN-positive or HN-negative cells, and whether or not they had been cultured in the presence or absence of neutralizing antibody. Furthermore, upon removal of the medium containing neutralizing antibody, infectious virus was recovered from all colonies tested. The fact that all the cells remained positive for NP in the presence of high titres of neutralizing antibody demonstrates that the cells remained infected as they divided, and that the production of infectious virus was not required for the maintenance of persistence within the colonies.
Previous work has shown that phosphorylation of PIV5 P regulates the activity of the vRdRP and that phosphorylation of the serine (S) residue at position 157 (S157) of the P protein plays a role in down-regulating viral gene expression [31, 36]. To determine whether S157 plays a critical role in the switch-off of viral transcription and replication, and in the establishment of persistence, we generated a recombinant virus, rPIV5-W3:P(F157), in which the serine residue at position 157 was replaced by a phenylalanine (F) residue in the PIV5-W3 backbone. The F substitution was chosen because several strains of PIV5 have this residue at position 157. Sequence analysis of rPIV5-W3:P(F157) confirmed that this was the only amino acid substitution in the recombinant virus. Following infection with rPIV5-W3:P(S157) at high moi, switch-off of viral protein synthesis occurred 24 and 72h p.i., and >95% of the cells survived the infection (Fig 3, panels a and b). In striking contrast, no detectable switch-off of viral transcription or replication occurred in cells infected with rPIV5-W3:P(F157), and by 72 h p.i. ~90% of infected cells had died (Fig 3, panels a, b and c). Furthermore, rPIV5-W3:P(S157) generated poorly defined plaques, which needed to be immunostained for clear visualisation, whereas plaques produced by rPIV5-W3:P(F157) were easily visualised by crystal violet staining because the cells within the plaques died, leaving obvious holes in the monolayer (Fig 3, panel d). Significantly, in single-step growth curves, rPIV5-W3:P(F157) grew more rapidly and to higher titres than rPIV5-W3:P(S157) (Fig 3, panel e).
It has been shown previously that a cellular kinase, Polo-like kinase 1 (PLK1), interacts with the P protein through the phosphorylated S157 residue and phosphorylates other sites on the P protein, including S308; phosphorylation at either of these sites reduces virus transcription [31]. We used mass spectrometry to compare the phosphorylation of the P protein in cells infected with rPIV5-W3:P(S157) and rPIV5-W3:P(F157) (S4 Fig). These results confirmed that S157 and S308 were phosphorylated in cells infected with rPIV5-W3:P(S157). Despite identifying multiple other phosphorylation sites on P in this way, we did not identify any sites, other than S157, that were phosphorylated on rPIV5-W3:P(S157) but not on rPIV5-W3:P(F157), and vice versa, including S308 (S4 Fig). However, we could not rule out the possibility that the relative levels of phosphorylation at the different sites did not vary significantly between rPIV5-W3:P(S157) and rPIV5-W3:P(F157). In a previous study, Sun et al [31], showed that a PLK1 kinase inhibitor (BI2536) increased PIV5-W3 gene expression in infected cells at 18–20h p.i.. Therefore, we tested whether BI2536 treatment prevented or delayed the switch-off of rPIV5-W3:P(S157) protein synthesis (S5 Fig). BI2536 had no discernible effect on the switch-off of viral protein synthesis, or the ability of rPIV5-W3:P(S157) to establish a persistent infection, suggesting that PLK1 is not the only cellular kinase that can phosphorylate serine-157 and inhibit viral gene expression.
Having established that both transcription and replication of the W3 strain of PIV5 are significantly down-regulated within 48h, we next investigated whether this was the case for other PIV5 strains. A549 cells were infected at high multiplicity with the W3, CPI+, MEL, LN, SER and H221 strains of PIV5 and were metabolically labelled at various times p.i. with [35S]-L-methionine. Expression of NP and M proteins was repressed with time in cells infected with PIV5-W3, but no obvious switch-off of viral protein synthesis was observed for the other strains of PIV5 (Fig 4, panel a). HTS confirmed that high levels of virus protein synthesis at late times p.i. in CPI+-infected cells were because virus transcription was not repressed (S6 Fig, panel a). These studies also showed that the maximal levels of viral mRNA were significantly higher in CPI+ and rPIV5-W3:P(F157), approximately 17% and 13% respectively (S6 Fig, panel a and b), than in rPIV5-W3:P(S157)-infected cells (approximately 5%; Fig 1, panel b).
A comparison of PIV5 P protein sequences published in GenBank strains revealed that CPI+, MEL and LN have F157 (Fig 5), consistent with their failure to shut-down expression as observed with rPIV5-W3(F157). However, we had expected that viral protein synthesis would be repressed at late times p.i. with H221 and SER because they have a serine at residue 157, but it was not. There are three amino acid sequence differences in P between PIV5-W3 and PIV5-SER (S69L, T155P and T293K) [13]; strikingly, T155P is in close proximity to S157. There are four amino acid differences in P between PIV5-W3 and PIV5-H221 (V226M, T293K, N306K and I381D); N306K is in close proximity to S308. We next checked whether P155, K306, and other amino acid changes around S157, have a direct effect on vRdRP activity by using a minigenome system. We initially compared the ability of P from PIV5-W3 (S157) and PIV5-CPI+ to stimulate viral RNA synthesis (Fig 4, panel b). In agreement with previously published data [36], P from PIV5-CPI+ was considerably more active than that derived from PIV5-W3. As predicted, substituting S for F at position 157 stimulated vRdRP activity (Fig 4, panel b). We next examined the phenotypes of other changes in the W3 P protein. The T155P substitution (as observed in PIV5-SER) significantly enhanced the activity of PIV5-W3 P in the minigenome assay (Fig 4, panel b). We also noted stimulatory effects of amino acid substitutions at positions 156 and 159. The N306K substitution (observed in PIV5-H221) and a substitution at amino acid 308 (S308A) also significantly enhanced P activity. In contrast, the T293K substitution (observed in both PIV5-SER and -H221) had only a small effect. These data show directly that the potential for phosphorylation in two motifs TSSPI (residues 155–159) and NDS (residues 306–308) represent targets for the negative regulation of P activity.
To further test the effect of changes in P on the gene expression profile of W3, we generated recombinant viruses with a T to P substitution at position 155, rPIV5-W3:P(P155), or an N to K substitution at position 306, rPIV5-W3:P(K306). Strikingly, both rPIV5-W3:P(P155) and rPIV5-W3:P(K306) behaved similarly to rPIV5-W3:P(F157), in that viral protein synthesis was not inhibited at late times p.i. (Fig 4, panel c), and viral infection resulted in increased cell death. In contrast, substituting a lysine for an arginine residue at position 254 (rPIV5-W3:P(R254)), which is part of a putative sumoylation site [37], had no effect on the switch-off of PIV5-W3 transcription, neither did deletion of the SH gene (Fig 4, panel c). These results demonstrate that single amino acid substitutions at multiple sites within P (observed in natural isolates of PIV5) can switch PIV5 from a virus with a persistent phenotype to one with a lytic phenotype.
Having shown that the repression of viral transcription and replication, and the establishment of persistence, depends on the integrity of the TSSPI motif at residues 155–159 in PIV5-W3, we compared all the PIV5 P protein and gene sequences available in the GenBank database (Fig 5). It was striking that in all strains residue 155 was either threonine or proline, residue 156 was either serine or asparagine, residue 157 was either serine or phenylalanine, and residue 159 was either threonine or isoleucine. Residue 158 (proline) was invariant. No strain had more than one difference from W3 in this region (e.g. none had both a proline at residue 155 and a phenylalanine at residue 157), and codon redundancy at these residues was such that a single nucleotide substitution was sufficient to change the virus from one with a predicted lytic to a W3-like persistent phenotype. Similarly to PIV5-W3, some strains had threonine at residue 155 and serine at residue 157 but differed from PIV-W3 sequence at neighbouring residues that increased vRdRP activity in the minigenome assays (e.g. residue 156 could be serine or asparagine, and residue 159 isoleucine or threonine; Fig 4, panel b). However, again only one nucleotide substitution was required to convert the sequence of P in these strains to that of PIV5-W3. It is also notable that PIV5-H221 was the only strain with lysine instead of asparagine at residue 306, and again codon redundancy ensured that a single nucleotide substitution was sufficient to convert it to a PIV-W3 phenotype.
Defective virus genomes (DVGs) have been shown to play a role in the establishment and maintenance of persistent infections of tissue culture cells by many positive and negative sense RNA viruses [38]. We therefore used HTS to determine whether DVGs may play a role in the establishment of persistence with PIV5-W3. Firstly, we showed that HTS both of purified nucleocapsids and of total cell RNA (following the physical removal of ribosomal and mitochondrial RNA) could be used to successfully detect the presence of DVGs (S7 Fig). From this analysis we determined that we would detect any DVGs if their breakpoint sequences contributed more than 0.02% of genomic sequences, or if the DVG contributed more than 0.2% (or even less, see below) of the total DVG and non-defective genomes. Using this approach, no DVGs were detected, either in purified nucleocapsids or in total cell RNA, isolated from passage 3 persistently infected PIV5-W3 cells. HTS both of nucleocapsid RNA and total cell RNA extracted from p3 persistently infected cells also revealed that there were no changes in the consensus sequence of PIV5-W3 in the persistently infected cells.
Although, approximately 80% of cells die by 3 days p.i. following infection with CPI+ (as determined by measuring cell viability using PrestoBlue as described in Fig 3), some cells survive and, with difficulty, it is possible to establish a persistently infected cell-line from these surviving cells. This necessitated that the surviving cells be cultured for many weeks without sub-culturing, replacing the culture medium regularly. Eventually the surviving cells began to grow and could be passaged. However, even then they continued to grow poorly and showed obvious signs of virus cytopathic effects within the monolayers. High levels of DVGs (the ratio of DVGs to non-defective genomes was 1.7:1, S7 Fig) were detected in cells persistently infected with CPI+, suggesting that they may play a role in the establishment of persistent infections under these conditions. Also, in marked contrast to cells persistently infected with PIV5-W3 in which the amount of viral mRNA was less than 0.2% of total cellular mRNA, the levels of CPI+ mRNA in persistently infected cells was significantly higher, approximately 6% of total cellular mRNA. Furthermore, several polymorphic mutations were identified in the CPI+ persistently infected cell-lines but all of these, except for one, were synonymous mutations. The exception was located at position 13093 of the genome, an A to T change, resulting in a phenylalanine to leucine substitution in L, that was present in 17% of the reads, but the biological significance of this is unclear.
We wished to investigate whether the phenotypic differences observed between rPIV5-W3:P(S157) and rPIV5-W3:P(F157) were reflected in differences in their biological properties in a mouse model system. However, first, and in agreement with our previously published data [39], we demonstrated that, as observed in A549 cells, rPIV5-W3:P(S157) protein synthesis was switched off in BALB/c fibroblasts and that the cells survived the infection. In contrast, rPIV5-W3:P(F157) protein synthesis was not switched off in murine fibroblasts by 72 h p.i. and most of the cells died following infection (Fig 6, panels a and b).
To determine whether rPIV5-W3:P(S157) and rPIV5-W3:P(F157) had different biological properties in vivo, BALB/c mice were infected intranasally with rPIV5-W3:P(S157) or rPIV5-W3:P(F157) and sacrificed at 1, 2, and 7 days p.i., and the amount of virus present in the lungs was estimated by quantitative PCR (Fig 6, panel c). In addition, the amount of inflammation at the time of sacrifice was assessed by measuring the number of cells in the lungs (Fig 6, panel c) [40]. These results showed that, although rPIV5-W3:P(F157) had replicated to higher titres than rPIV5-W3:P(S157) by 2 days p.i., it was cleared more rapidly. Thus, by 7 days p.i. there was significantly less rPIV5-W3:P(F157) present in the lungs than rPIV5-W3:P(S157) (p<0.001). Interestingly, the amount of rPIV5-W3:P(S157) present in the lungs at 7 days p.i. was similar to that observed at 1 and 2 days p.i.. There were significantly more cells in the lungs after rPIV5-W3:P(F157) infection at days 2 (p<0.05) and 7 (p<0.01) p.i. than rPIV5-W3:P(S157), indicating greater inflammation [40]. However, neither virus caused overt disease as measured by weight loss (Fig 6, panel c).
Within-host RNA viral persistence has many potential consequences for both virus and the host [7]. For example, persistently infected individuals may act as viral reservoirs within host communities, and persistent infections may be important in the development of long-lasting protective immunity. However, little is known about the molecular mechanisms by which RNA viruses establish such infections. In part this may be because, unlike the better understood situations with DNA viruses or retroviruses, and despite the examples of hepatitis C virus and bornaviruses, it is often difficult to determine whether certain RNA viruses have evolved specific molecular mechanisms to establish and maintain persistent infections. To establish such infections within a host following lytic infection, it is likely that viral replication must be repressed in at least some cells in order either to prevent viral replication from killing the cell or to avoid the infected cell being eliminated by innate and adaptive immune responses [7]. In the case of paramyxoviruses and other members of the order Mononegavirales, and given their general mode of replication, it is not obvious how viral replication could be specifically repressed in order to facilitate virus persistence.
Using PIV5 as a model, we report that viral transcription and replication can be repressed by phosphorylation of P, resulting in the establishment of persistently infected cell cultures (without the need for the presence of DVGs) in which the virus can flux between active and repressed states within individual cells. Since the consensus genome sequence of PIV5-W3 does not change in persistently infected cells we suggest that the amount of P (Fig 2, panel c), as well as its level of phosphorylation, varies heterogeneously over time within persistently infected cells and it is this which determines whether the virus is active or repressed within individually infected cells. We also speculate that in vivo, depending on the status of the adaptive immune response, variants with lytic or persistent phenotypes will be selected for or against. Since the quasispecies nature of RNA viruses will generate a cloud of virus mutants in vivo [32, 41, 42] as the virus spreads from cell to cell, during acute phases of infection rapidly replicating PIV5 variants will be selected for. However, eventually cells that continuously synthesize high levels of viral proteins will be efficiently killed by cytotoxic T cells and possibly ADCC (antibody dependent cell cytotoxicity). Consequently, as the adaptive immune response develops, variants whose replication may be repressed, thus avoiding cell killing, will be selected, leading to the establishment of persistence. Since virus can be reactivated from cells in which it has been repressed, it is likely that small amounts of infectious virus will continuously be produced in persistently infected individuals perhaps as the immune response waxes and wanes. If such variants are transmitted to a new host again initially rapidly replicating variants will have a selective advantage, setting up a cycle of alternative selection of acute vs persistent variants in at least some infected individuals. Given that single amino acid (nucleotide) changes can determine whether a particular PIV5 variant has a lytic or persistent phenotype this mechanism may have evolved to allow PIV5 to establish both productive acute infections as well as persistent infections, thereby potentially increasing its chances of transmission [43]. For similar reasons, it is possible that other RNA viruses may have evolved analogous mechanisms in which single amino acid (nucleotide) changes can determine whether a particular variant has a lytic or persistent phenotype.
Our results support and extend those published by the He group showing that phosphorylation of S157 and S308 on P results in repression of viral RNA synthesis [31]. However, our mass spectroscopy analysis showed phosphorylation of S308 in peptides from cells infected with rPIV5-W3:P(S157) and from cells infected with rPIV5-W3:P(F157) (although we could not quantitate the degree of phosphorylation at this, or other, residues which might be quite heterogeneous), and yet rPIV5-W3:P(F157) replication was not inhibited at late times p.i.. Furthermore, BI2536, an inhibitor of PLK1, did not influence the observed switch-off of PIV5-W3 protein synthesis nor did it prevent the establishment of persistence (S5 Fig). This strongly suggests that cellular kinases other than PLK1 are (also) responsible for phosphorylating P. Although, we do not know the other kinase(s) responsible, the fact that PIV5-W3 transcribes and replicates its genome efficiently until 12-18h p.i. suggests that they are unlikely to be constitutively expressed cellular kinases but may be induced kinases, for example the ER stress response kinases activated by PIV5 infection [44]. If the latter, then perhaps this may also be a mechanism that the virus has evolved to help establish persistence. We note that P is also highly phosphorylated in other paramyxoviruses (as are the phosphoproteins of other members of the order Mononegavirales, including Ebola virus) and that the level of phosphorylation influences viral transcription [28, 45, 46]. Furthermore, between different strains/isolates there can be substitutions of amino acids, including serine, that may be phosphorylated, opening up the possibility that lytic and persistent variants of these viruses may be selected in vivo.
Work presented here and elsewhere [31] clearly shows that phosphorylation of residues within the TSPPI motif (amino acids 155–159) and the NDS motif (amino acids 306–308) strongly influence the activity of the vRdRP. Residues 155–159 and 306–308 are outside the known N-terminal and C-terminal binding sites on P for NP [47, 48] and are outside its predicted oligomerisation domain [49]; no binding site for L has yet been mapped. It is interesting to note that a structural prediction for the P protein of the closely-related rubulavirus, PIV2 [50] places the 155–159 and 306–308 motifs in long non-structured regions that flank the predicted oligomerisation domain; it is tempting to speculate that phosphorylation alters the conformation of the non-structured regions thereby influencing the properties of the P protein. In this regard, it is of note that the threonine to proline substitution at residue 155 enhanced the vRdRP activity more than the serine to phenylalanine substitution at residue 157. Indeed, P155 led to the most active vRdRP in minigenome assays, even more active than A308 (and K306). The reason for this is unclear but suggests a model in which active and repressed states of vRdRP are in equilibrium, and that phosphorylation of P at residues 157 and 308 moves this equilibrium towards the repressed state. In this case, substituting threonine with proline at position 155 may cause a major structural change that locks vRdRP in the active state.
DVGs may also play a role in the establishment of persistent infections with PIV5, at least in vitro, as observed with other RNA viruses [38]. We show here that, with difficulty, persistently infected cell-lines can be established following a high moi of A549 cells with CPI+ lytic isolate of PIV5. In contrast to cells persistently infected with PIV5-W3, the CPI+ persistently infected cell-line had high levels of DVGs (S7 Fig), suggesting under these circumstances DVGs may play a role in CPI+ virus persistence. The characteristics of this persistently CPI+-infected cell-line was very different from that established by PIV5-W3. Ongoing virus transcription was much higher, the cells grew much more slowly and there were clear signs of a virus cytopathic effect. There was no evidence from HTS for selection of variants of CPI+ (e.g. S157) that would be predicted to have a persistence phenotype. However, this would be expected as the cells were initially infected at a high moi making it unlikely that such variants could be selected. Rather, as discussed above, we speculate that the selection of virus variants with a persistence phenotype, such as PIV5-W3, would likely occur following low moi infections in vivo in the presence of an ongoing adaptive immune response. Indeed, a critical point we are making here is that PIV5, and thus potentially other paramyxoviruses, may have evolved specific molecular mechanisms for the establishment of persistent infections which do not rely on the production of DVGs.
Although we have highlighted the importance of the phosphorylation status of P in determining whether or not a particular variant can establish persistence, it is possible that other single amino acid (nucleotide) changes in other genes, including L, may also play a role. We have also previously suggested that the interferon response may play an important role in repressing viral replication in some cells, thereby facilitating the establishment of persistence, and that there may be alternating selection of IFN-resistant and IFN-sensitive viruses during the acute and persistent phases of infection [11]. Interestingly, single amino acid (nucleotide) substitutions in the V protein, which is the viral IFN antagonist, can also determine whether a variant is IFN-sensitive or IFN-resistant [51]. Given that P and V are encoded by the same gene and share their N-terminal sequences, this gene may have evolved in such a manner as to facilitate the establishment of persistent infections. In this regard, it is of interest that PIV5 and other paramyxoviruses block the IFN response in such a way as not to cause cell death, which is a pre-requisite for establishing persistence.
PIV5 has been isolated on numerous occasions from a variety of host species but its association with disease is often tentative and unclear. Of possible relevance is that the disease potential of lytic and persistent variants of PIV5 is likely to be different. Thus, lytic variants may cause more cell death and spread more rapidly in vivo than persistent variants. Although PIV5 does not replicate to high titres in, or naturally infect, mice, this idea is supported by the observation that, rPIV5-W3:P(F157) replicated better than rPIV5-W3:P(S157) in mice and induced more cellular infiltration into the lungs of infected mice, which is a clear sign of greater pathology. Also if there are mixed populations of persistent and lytic variants in vivo, then the balance of the two may also influence disease outcomes. Interestingly, both lytic and persistent variants were detected by HTS in our stocks of PIV5-H221, which was isolated from a dog with kennel cough, but which had only been passaged a limited number of times in tissue culture cells following its initial isolation [13]. Thus, although the consensus sequence at position 157 of P was serine, phenylalanine was predicted in 4% of the viral population, and although the consensus at position 306 was a lysine, 5% of the sequenced population encoded asparagine [13].
It is of note that all the PIV5 strains sequenced, apart from the W3 strain, are predicted to have a lytic phenotype, which argues against the suggestion that viral transcription and replication are reduced at late times p.i. in order to limit the production of viral PAMPs and hence the induction of antiviral cytokines [31]. However, as lytic strains are more likely to induce a cytopathic effect than persistent variants, their selection may be favoured during clinical isolation. Furthermore, lytic variants, which give an obvious cytopathic effect in tissue culture cells, may evolve during the isolation of PIV5 from clinical material. This may have occurred during the isolation of PIV5 from human bone marrow cells, which were co-cultured with either MRC5 or Vero cells, as immunofluorescence was initially used to detect PIV5 during virus isolation as there was often an absence of a clear virus induced cytopathic effect [24, 52]. Similarly, during the isolation of cryptovirus (a strain of PIV5), human lymphocytes from a patient with SSPE (there is no suggestion that PIV5 can cause SSPE) were cultured with AV3 (continuous human amnion) cells, but the first clear signs of cytopathic effect only became visible after 20 passages [26]. On the other hand, tissue culture cell-lines can be persistently infected with PIV5 with no overt signs of infection, for example AGS cells which are commercially available from ATCC and ECACC [21].
Understanding the mechanisms by which paramyxoviruses, and other RNA viruses, can establish persistence in vivo is important for both fundamental and practical reasons. It may lead to a more complete view of viral epidemiology, and thus to better control measures. In addition, if the induction of long-lasting immunity is enhanced by viral persistence, then understanding the mechanisms by which viruses can establish such infections may lead to improved vaccine design.
Vero, 293 and A549 cells (all from the European Collection of Authenticated Cell Cultures; ECACC) and derivatives were grown at 37°C as monolayers in 25 cm2 or 75 cm2 cell culture flasks, in Dulbecco's modified Eagle's medium (DMEM) supplemented with 10% (v/v) foetal bovine serum at 37°C. Stocks of PIV5 strains W3, LN, MEL, H221, SER and CPI+ (described in [13]) were grown and titrated in Vero cells. Commercial cell-viability assays (PrestoBlue (ThermoFisher Scientific) were performed according to manufacturer instructions.
Infected or uninfected cells were metabolically labelled for 1h with [35S]-L-methionine (500Ci/mmol, MP Biomedical, USA) at various times p.i. as indicated in the text. After labelling, cells were lysed in disruption buffer, sonicated and heated for 5 min at 100°C and proteins were analysed by sodium dodecyl sulphate-polyacrylamide gel electrophoresis (SDS-PAGE). The gels were fixed, stained, dried, and resolved radiolabelled bands visualized by phosphorimager analysis.
Procedures for immunoprecipitation, immunoblotting and immunofluorescence have been described previously [53, 54]. The antibodies used included monoclonal antibodies (mAbs) against PIV5 HN, P and NP [55] and against cellular MxA and β-actin (Sigma, A5441). Sialic acid on the surface of cells was visualised by staining with a recombinant protein in which green fluorescent protein (GFP) had been fused to two carbohydrate-binding modules derived from Vibrio cholerae [56]. Viral plaques were immunostained with a pool of mAbs against PIV5 followed by alkaline phosphatase-conjugated goat anti-mouse immunoglobulin G (Abcam, ab97020), and plaques were visualised with SigmaFast BCIP/NBT. For FAC-sorting, cells were prepared as a single-cell suspension by trypsinisation and immunostained with a pool of mAbs against HN. Single cells were sorted into individual wells of 96-well microtiter plates on the basis of whether or not they were positive for HN using a Becton Dickinson FACSJazz instrument.
The changes in the P gene of the PIV5 W3 genome [T155P, S157F, K254R, N306K and S308A] were generated by primer-mediated mutagenesis using oligonucleotides purchased from Sigma and the modified fragments inserted into the rPIV5-W3 backbone plasmid, pBH276 [57] using standard molecular biology approaches. Base changes were confirmed by DNA sequencing. The pBH276-derived template plasmids (1μg) were transfected together with pCAGGS-based helper plasmids directing the synthesis of PIV5-NP (100ng), PIV5-P (100ng) and PIV5-L (500ng) into 6-well dishes containing ~106 BSRT7 cells per well using linear polyethyleneimine (PEI) of molecular weight 25,000 (Polysciences Inc., Warrington PA, USA), or Fugene, under standard conditions. Successful recovery was confirmed by immunofluorescence screening using a monoclonal antibody (PIV5-Pk) conjugated to a FITC fluorophore, which recognises PIV5 V and P [55]. Working stocks of virus were produced from positive wells by two successive passages at low multiplicity of infection (moi) in Vero cells, and stocks were harvested, clarified by centrifugation, and flash frozen in liquid nitrogen.
Confluent monolayers of A549 cells, grown in 25 cm2 flasks, were infected at a high moi with either rPIV5-W3:P(S157) or rPIV5-W3:P(F157) and at 24h p.i. were lysed in disruption buffer and submitted to the FingerPrints Proteomics Facility (University of Dundee, UK) for SDS-PAGE gel analysis. The samples were run on a 4–12 Bis-Tris gel with MOPS running buffer (Thermo Fisher Scientific) and the gel stained with SimplyBlue SafeStain (Thermo Fisher Scientific). PIV5 P bands (~45kDa) were excised for in-gel processing and trypsin digestion prior to analysis by mass spectrometry using a RSLCnano UHPLC system coupled to a LTQ Orbitrap Velos Pro mass spectrometer (Thermo Scientific). The resultant data were analysed using the Mascot Search engine (Version 2.4.1) using the Sprot Human database and the sites of phosphorylation annotated using the Mascot delta score.
Infected cells in 25cm2 flasks were lysed in 1 ml of Trizol and RNA was extracted using a Direct-zol RNA miniprep kit (Zymo). A directional sequencing library was prepared from rRNA-depleted RNA using a TruSeq stranded total RNA library prep kit (Illumina, U.K.). Quality control and quantification of the cDNA library were monitored using DNA-specific 1000 or 5000 chips on a Bioanalyzer 2100 (Agilent Technologies) and a Qubit fluorometer (Invitrogen). Individual libraries were pooled at 10 nM each and sequenced on the MiSeq platform (Illumina). Abundances of genome and antigenome/mRNA reads were calculated relative to total read numbers (including cellular reads) from which residual rRNA and mitochondrial RNA reads had been removed. These reads were identified by aligning the trimmed, filtered data to reference genomes for human 18S, 28S, 5S and 5.8S rRNA and mitochondrial DNA (accession numbers NR_003286.2, NR_003287, X51545, J01866, NC_012920), and then removed.
The presence of defective virus genomes (DVGs) was assessed using ViReMa (Routh and Johnson 2013). ViReMa detects potential recombination by identifying reads that contain sequences mapping to different regions of the genome, and thus facilitates the identification and quantification of DVG populations.
Using standard techniques the Renilla luciferase gene in pSMG-RL [36] was replaced by a gene encoding firefly luciferase. An additional modification to reduce transcriptional readthrough from cryptic promoters within the vector backbone was made by incorporating two copies of a 237 bp fragment from SV40 (coordinates 2533–2770) that includes the bidirectional polyadenylation site and transcriptional terminator site downstream from the T7 RNA polymerase terminator sequence. To determine the activity of the minigenome, 25ng of the resulting plasmid (pPIV5MG-Fluc.ter), was transfected into 293 cells together with pCAGGS-based helper plasmids directing the synthesis of PIV5-NP (100ng), PIV5-P (100ng) and PIV5-L (500ng), 500ng of a pCAGGS-based plasmid directing the synthesis of T7 RNA polymerase (codon-optimised for expression in human cells), and 50ng of a β-galactosidase-expressing transfection control plasmid, pCATlac. Transient transfections used PEI and were left for 40 h before harvesting. Luciferase and β-galactosidase activity assays were carried out and normalised as previously described [58]. Variants of the P gene were generated by primer-mediated mutagenesis as described above.
Female BALB/c mice were obtained from Charles River (Bath) at 7–9 weeks of age and housed in accordance with the United Kingdom Home Office guidelines. All work was conducted with approval from the Animal Welfare and Ethical Review Board of Imperial College London. Mice were infected intranasally with 2 x 106 plaque-forming units (pfu) of virus in 100 μl. Mice were provided with food and water ad libitum and monitored daily for signs of illness. Statistical comparisons of mouse data were as described in Figure legends were performed using Prism 6 (GraphPad Software Inc., La Jolla, CA, USA).
Viral load in lung tissue was assessed by quantitative PCR (qPCR) of bulk PIV5 M gene RNA. RNA was extracted from frozen lung tissue using Trizol extraction after homogenisation in a TissueLyzer (Qiagen) and converted into cDNA using random primers (GoScript, Promega). qPCR of PIV5 M gene was carried out using SYBRselect master mix and 250 nM forward (5’-TCATGAGCCACTGGTGACAT-3’) and reverse (5’-TGGAATTCCCTCAGTTGTCC-3’) primers on a Stratagene Mx3005p instrument (Agilent Technologies). In order to normalise M gene levels, levels of cellular Gapdh mRNA were measured using forward (5’-AGGTCGGTGTGAACGGATTTG-3’) and reverse (5’-TGTAGACCATGTAGTTGAGGTCA-3’) primers.
Lung tissue was homogenised through a 100μm cell strainer and centrifuged at 500 x g for 5 minutes, as described previously [59]. Supernatants were removed, and red blood cell lysis buffer (ACK lysing buffer, ThermoFisher) was added to the cell pellet and mixed for 5 min before a further centrifugation at 500 x g for 5 minutes. Remaining cells were resuspended in DMEM and viable numbers were quantified by trypan blue exclusion.
All animal experiments were performed in accordance with the United Kingdom’s Home Office guidelines under PPL P4EE85DED and all work was approved by the Animal Welfare and Ethical Review board (AWERB) at Imperial College London. Studies followed the ARRIVE guidelines and all animal infections and infectious work was carried out in biosafety level two facilities.
|
10.1371/journal.pntd.0007268 | A 5-Year intervention study on elimination of urogenital schistosomiasis in Zanzibar: Parasitological results of annual cross-sectional surveys | The Zanzibar Elimination of Schistosomiasis Transmission (ZEST) project aimed to eliminate urogenital schistosomiasis as a public health problem from Pemba and to interrupt Schistosoma haematobium transmission from Unguja in 5 years.
A repeated cross-sectional cluster-randomized trial was implemented from 2011/12 till 2017. On each island, 45 shehias were randomly assigned to receive one of three interventions: biannual mass drug administration (MDA) with praziquantel alone, or in combination with snail control or behavior change measures. In cross-sectional surveys, a single urine sample was collected from ~9,000 students aged 9- to 12-years and from ~4,500 adults aged 20- to 55-years annually, and from ~9,000 1st year students at baseline and the final survey. Each sample was examined for S. haematobium eggs by a single urine filtration. Prevalence and infection intensity were determined. Odds of infection were compared between the intervention arms.
Prevalence was reduced from 6.1% (95% confidence interval (CI): 4.5%-7.6%) to 1.7% (95% CI: 1.2%-2.2%) in 9- to 12-year old students, from 3.9% (95% CI: 2.8%-5.0%) to 1.5% (95% CI: 1.0%-2.0%) in adults, and from 8.8% (95% CI: 6.5%-11.2%) to 2.6% (95% CI: 1.7%-3.5%) in 1st year students from 2011/12 to 2017. In 2017, heavy infection intensities occurred in 0.4% of 9- to 12-year old students, 0.1% of adults, and 0.8% of 1st year students. Considering 1st year students in 2017, 13/45 schools in Pemba and 4/45 schools in Unguja had heavy infection intensities >1%. There was no significant difference in prevalence between the intervention arms in any study group and year.
Urogenital schistosomiasis was eliminated as public health problem from most sites in Pemba and Unguja. Prevalence was significantly reduced, but transmission was not interrupted. Continued interventions that are adaptive and tailored to the micro-epidemiology of S. haematobium in Zanzibar are needed to sustain and advance the gains made by ZEST.
| Schistosomiasis is a disease caused by parasitic blood flukes of the genus Schistosoma. The highest burden is concentrated in sub-Saharan Africa. On the Zanzibar islands, urogenital schistosomiasis has been successfully controlled over the past decades. The Zanzibar Elimination of Schistosomiasis Transmission (ZEST) project implemented from 2011/12 through 2017 aimed to eliminate urogenital schistosomiasis as a public health problem from Pemba and to interrupt S. haematobium transmission from Unguja in 5 years. In a cluster-randomized trial, we investigated the impact of biannual treatment of the population with praziquantel alone or combined with snail control or behavior change interventions. After five years of interventions, the overall S. haematobium prevalence was reduced to <3% in schoolchildren and adults. Heavy infection intensities were observed in <1% of the surveyed population groups. Urogenital schistosomiasis was eliminated as a public health problem from most of the study sites on Pemba and Unguja. The prevalence was significantly lower in 2017 compared with 2011/12, but transmission was not interrupted. To sustain and advance the gains made by ZEST, continued interventions that are adaptive and tailored to the micro-epidemiology of S. haematobium in Zanzibar are needed.
| The transmission of parasitic blood flukes of the genus Schistosoma is reported from 78 countries. The global burden caused by the disease schistosomiasis was conservatively estimated at 1.43 million disability-adjusted life years in 2017 [1]. The highest burden of disease is concentrated in sub-Saharan Africa [2]. In 2012, the World Health Assembly urged countries to intensify schistosomiasis control and initiate elimination campaigns where appropriate [3].
The Zanzibar archipelago of the United Republic of Tanzania, with the main islands Pemba and Unguja, has a history of schistosomiasis control that dates back to the 1980s [4–7]. At that time, urogenital schistosomiasis caused by S. haematobium, was highly prevalent across both islands [7–9]. Thirty years later, considerable progress had been made and the overall prevalence in children and adults had decreased to 8% in Unguja and 15% in Pemba in 2011 [10]. Subsequently, the Zanzibar islands were one of the first areas in sub-Saharan Africa where elimination of urogenital schistosomiasis was considered as a feasible goal [10–12].
The Zanzibar Elimination of Schistosomiasis Transmission (ZEST) alliance, consisting of various stakeholders and institutions, including the Neglected Diseases Program of the Zanzibar Ministry of Health, the Public Health Laboratory-Ivo de Carneri, the Schistosomiasis Consortium for Operational Research and Evaluation (SCORE), the Schistosomiasis Control Initiative (SCI), the World Health Organization (WHO), the Natural History Museum London, and the Swiss Tropical and Public Health Institute, aimed to eliminate schistosomiasis as a public health problem from Pemba (<1% heavy infection intensities in all sentinel sites) and to interrupt transmission on Unguja (zero incidence in all sentinel sites) in 5 years [10, 13]. Moreover, it aimed to learn about the effectiveness of snail control and behavioral change interventions. In that regard, the Zanzibar Neglected Diseases Program supported by SCI and WHO, implemented biannual mass drug administration (MDA) with praziquantel to the whole eligible population from 2012 onward. In addition, SCORE funded a 5-year repeated cross-sectional cluster-randomized trial (CRT), which was implemented in Zanzibar from November 2011 till May 2017. A total of 90 shehias (small administrative areas) were randomly assigned to receive one of three interventions: biannual mass drug administration (MDA) with praziquantel alone (arm 1) or in combination with snail control (arm 2) or behavior change measures (arm 3). The primary outcome of the CRT was defined as S. haematobium infection prevalence and intensity in 9- to 12-year old children after five years of follow-up in 2017; these results are published elsewhere [14]. Here we present the impact of the interventions on the S. haematobium prevalence and intensity in 1st year students and adults, in addition to 9- to 12-year old children, across all 5 years of the project in Pemba and Unguja, and discuss the implications and challenges for interruption of urogenital schistosomiasis transmission in Zanzibar.
The Zanzibar islands (Unguja and Pemba) are part of the United Republic of Tanzania. The islands are divided into districts and the smallest administrative areas are called shehias. The population of Zanzibar was estimated at 1.4 million in 2016 [15]. The average population of a shehia is ~4,700 inhabitants but they vary considerably in size (range: 482–26,275 inhabitants) [16]. Most shehias have at least one public primary school. Children usually enter the first grade at the age of 6 or 7 years. Primary schools include grades 1 to 6. The net enrolment ratio in primary schools was estimated at 84.2% in Zanzibar in 2014/15 [15]. The predominant religion in Zanzibar is Islam. In addition to primary school, many children visit religious schools, called madrassas [17]. Among all households in Zanzibar, 90.5% used a protected water source for drinking in the dry season and 83.7% had a toilet facility in 2014/15 [15].
On both islands, urogenital schistosomiasis was considered as a major public health problem in the past century [5, 8, 9]. Repeated MDAs with praziquantel and likely also improvements in the socio-economic standard, including access to clean water and sanitation, contributed to the reduction of the S. haematobium prevalence. The baseline survey of the CRT in 2011/12 revealed that 8.9% and 4.3% of schoolchildren and 5.5% and 2.7% of adults in Pemba and Unguja, respectively, were infected with S. haematobium according to a single urine filtration result [18].
Ethical approvals were obtained from the Zanzibar Medical Research Ethics Committee in Zanzibar, United Republic of Tanzania (ZAMREC, reference no. ZAMREC 0003/Sept/011), the “Ethikkommission beider Basel” (EKBB) in Basel, Switzerland (reference no. 236/11) and the Institutional Review Board of the University of Georgia in Athens, Georgia, United States of America (project no. 2012-10138-0). All individuals participating in the parasitological surveys were informed in lay terms about the aims and procedures of the study. Written informed consent was obtained from the parents or guardians of participating children and directly from participating adults. The study is registered with the International Standard Randomized Controlled Trial Number register (ISRCTN48837681).
The study was designed as a 5-year repeated cross-sectional CRT. A shehia was defined as cluster and intervention unit (Fig 1). A total of 45 shehias on Pemba and 45 shehias on Unguja, respectively, were assigned randomly to receive one of three interventions (ratio 1:1:1): biannual MDA with praziquantel alone (arm 1), or in combination with snail control (arm 2), or in combination with behavior change measures (arm 3). In the annual cross-sectional surveys conducted from November 2011 through May 2017, we annually enrolled schoolchildren aged 9- to 12-years from grade 3 and 4, and at baseline and the final survey also from grade 1 in the public primary schools of the study shehias, and annually adults aged 20- to 55-years in the communities of the study shehias. From each participant a single urine sample was collected between 10:00 and 14:00 hours and examined for S. haematobium infection as described below. In the surveys conducted in 2014 till 2016, 9- to 12-year old students were asked about their participation in the last school-based treatment (SBT) round and adult participants were interviewed about their participation at the last community-wide treatment (CWT) round preceding the survey and about their compliance with praziquantel intake. In all years, the surveys were followed by MDA, where praziquantel (40 mg/kg) was offered to the whole eligible population as described below.
The sample size calculation, eligibility criteria, and randomization procedures of clusters and study participants were described in detail in our published study protocol [10]. Due to the nature of the interventions, neither participants nor field or laboratory personnel were blinded to the intervention arms.
Praziquantel (40 mg/kg) was administered biannually in intervals of approximately 6 months to the whole eligible population by the Neglected Diseases Program of the Zanzibar Ministry of Health [10]. Between April 2012 and November 2016, a total of 10 CWT rounds were conducted. In each round, over a period of 3 days, praziquantel was provided by trained community drug distributors to adults with exception of severely ill people and pregnant women, and to children (≥3 years) who had not received praziquantel during SBT. From 2013, to increase coverage, schools were used as additional venues to distribute praziquantel to children. Teachers, supervised by staff of the Neglected Diseases Program, administered the tablets to students using a dose pole [19]. Treatment in schools was directly observed. Between November 2013 and November 2016, 6 rounds of SBT were conducted in Pemba and 5 rounds of SBT in Unguja.
Snail control was conducted in 15 shehias in Pemba and Unguja, respectively, in addition to biannual MDA. For this purpose, human water contact sites at streams or ponds inside the study shehia boundaries were identified with the support of the local communities. Each site was surveyed up to five times per year for intermediate host snails of the genus Bulinus in the dry season, when water conditions were suitable for snail collection [10]. At each visit, the number of Bulinus collected by a pre-specified number of fieldworkers in a certain time and length of shoreline was recorded. When Bulinus were present, the trained fieldworkers sprayed the molluscicide Bayluscide (niclosamide) to the shorelines near the human water contact site, using Hudson sprayers or a petrol power sprayer.
Behavior change measures in addition to biannual MDA were implemented in 15 shehias in Unguja and Pemba, respectively. The interventions were developed with formative research and using a human centered design approach in 2011 [20]. They were implemented in a phase-in approach and improved over the study period [14]. From 2012, educational play and interactive learning methods for urogenital schistosomiasis prevention were introduced and applied by trained teachers in the public primary schools of the study shehias. Moreover, in each behavioral shehia, one male and one female urinal were installed close to a known transmission site at a river or pond. From 2013, regular behavior change education meetings were conducted in the community. From 2014, to increase the reach to children not attending public schools and to the community as a whole, trained teachers of religious schools (madrassas) were involved as behavioral change agents [17]. In 2014 and 2015, 21 and 25 washing platforms were installed near safe water sources such as taps or wells in Pemba and Unguja, respectively [14]. At least one washing platform was installed in each shehia of the behavioral arm.
The parasitological surveys were implemented annually, starting with the baseline surveys of adults in November and December 2011 and of schoolchildren in January till April 2012. Annual follow-up surveys were conducted between January and May from 2013 through 2017. Each year before the parasitological surveys started, a meeting was conducted with teachers and shehas of all 90 study schools and shehias. The study aims and procedures were explained and the results from the previous years presented. The date when the survey teams would visit the schools and shehias was communicated in a letter to the headmasters and shehas, respectively. Each school was visited on two subsequent working days. On the first day, children were randomized for participation [10]. From enrolled children, name, sex, age, place of birth and their participation at the previous SBT round were recorded. The study was explained to the children in lay terms and they received an information- and a consent-sheet for their parents to sign. On the second day, each child that submitted written informed consent from a parent or legal guardian received a container for urine collection between 10:00 and 12:00 hours. In each of the 90 study schools, urine samples were collected annually from ~100 children aged 9-to-12 years attending grades 3 and 4. At baseline and the final survey, urine samples were additionally collected from ~100 children from grade 1.
Each of the 90 study shehias was visited on one day, for data and urine collection from ~50 adults living in the shehia. The randomization procedures for selection of households and adult participants are described elsewhere in detail [10]. Before enrolment, the study aims were explained to each potential participant by a trained fieldworker and written informed consent for participation was collected. Each enrolled adult was asked to submit a urine sample between 10:00 and 14:00 hours and interviewed with a pretested questionnaire concerning his/her demographic data and participation during the last round of MDA.
After collection, the urine samples were placed in carrier boxes, transferred to the Public Health Laboratory-Ivo de Carneri in Pemba or the laboratory of the Neglected Diseases Program in Unguja and examined during the afternoon of the same day and within seven hours after production. Each sample was examined by eye for macrohematuria and with reagent strips (Haemastix; Siemens Healthcare Diagnostics GmbH) for microhematuria. From samples of sufficient volume, a single 10 ml urine filtration slide was prepared, using a polycarbonate filter with a pore size of 20 μm. The filters were covered with hydrophilic cellophane and a drop of Lugol’s iodine was added to stain S. haematobium eggs. The filters were examined under the microscope by trained technicians, who recorded the number of eggs detected on each slide. Ten percent of the slides of each technician were kept for quality control by an external microscopist [21]. Once the surveys were concluded, the next round of MDA commenced, where praziquantel (40 mg/kg) was offered to the whole eligible population using a praziquantel dose pole [19].
The primary outcomes were S. haematobium infection status and infection intensity among 9- to 12-year old children in 2017 at individual and cluster level. These results are presented elsewhere [14]. The secondary analyses presented here included S. haematobium infection status and intensity among 9- to 12-year old children and 20- to 55-year old adults in all study years (2011/12 through 2017), and among 1st year students at baseline and the final survey (2012 and 2017). They also included coverage and compliance of the study participants with praziquantel treatment provided in the MDA round preceding the respective survey.
Data were entered in spreadsheets (Excel 2010, Microsoft), cleaned, and analyzed with StataIC 14 (StataCorp.; College Station, Texas, United States of America) and R version 3.4.3 (www.r-project.org). Maps were created with QGIS version 2.14.21 using coordinates of schools collected with a handheld Garmin GPSMAP 62sc device (Garmin, Kansas City, USA) and shape files of shehias provided by the Zanzibar Health Management Information System to the Neglected Diseases Program of the Zanzibar Ministry of Health.
All participants with a urine filtration and/or reagent strip result were included in the analyses. Participants were considered S. haematobium-positive if at least one S. haematobium egg per 10 ml urine was detected by the urine filtration method. In the absence of a urine filtration result, a participant was considered S. haematobium-positive, if microhaematuria was detected with reagent strips. Egg counts were truncated at 1000 eggs per 10 ml urine, and stratified into light (1–49 eggs per 10 ml urine) and heavy (≥50 eggs per 10 ml urine) infections according to WHO guidelines [22]. Based on the manufacturers color code, microhaematuria intensity was graded into negative, trace, +, ++, and +++.
Generalized estimating equation (GEE) models with binary and negative binomial distributed outcomes and independent correlation structure were applied to compare trial arms. Arm 1 was the designated reference group. For unadjusted estimates, trial arm as predictor and school or shehia as cluster, were included in the model. Adjusted odds ratios were calculated using the cluster prevalence at baseline as a continuous predictor in the model. For the inversed probability weighted odds ratios inverse probability weights were calculated for baseline prevalence and number of observations per cluster. The product of both weights was included in the model. All models used GEE with robust standard errors to account for clustering. For better comparability only clusters with baseline information were included.
Coverage in the MDA round preceding the survey and compliance with praziquantel treatment was assessed in 2014, 2015 and 2016. Coverage of schoolchildren was calculated as the proportion of children, who participated in the directly observed SBT, among all 9- to 12-year old schoolchildren surveyed. Compliance of adults was calculated as the proportion of adults, who reported in the questionnaire interview to have had received and swallowed praziquantel tablets all together (instead of not taking the tablets at all, or splitting intake to morning and evening, or over multiple days), among all surveyed adults [23].
Pemba consisted of 88 shehias and Unguja of 203 shehias in 2011. Among them, 201 shehias were excluded in a stepwise procedure and finally 45 shehias on Unguja and Pemba, respectively, were randomized to one of the three intervention arms as shown in Fig 1 and described in detail in the published study protocol [10]. The number of children and adults in each study arm included and excluded from the analysis of annual cross-sectional surveys is indicated in S1 Table. The baseline prevalence of S. haematobium in arm 1 was 4.2% in 9- to 12-year old children, 7.0% in 1st year students, and 2.8% in 20- to 55-year old adults. In arm 2, the prevalence was 7.8%, 9.9%, and 4.5%, respectively. In arm 3, the prevalence was 6.4%, 9.6%, and 4.5%, respectively. Additional baseline results of S. haematobium prevalence, infection intensity and microhaematuria, stratified by island and study group, are presented in Table 1 and S2 Table.
Fig 2 indicates that in none of the study years and in none of the study groups, there was a significant difference in the odds of S. haematobium infection, respectively, between the intervention arms. At the final survey of 9- to 12-year old children in 2017, the GEEs revealed no significant differences between the prevalence of arm 2 (OR 1.3, 95% CI 0.6–2.8) or arm 3 (OR 1.3, 95% CI 0.6–2.9) compared with arm 1 [14]. Also for 1st year students no significant differences between the prevalence of arm 2 (OR 1.8, 95% CI 0.8–3.8) or arm 3 (OR 1.8, 95% CI 0.7–4.4) compared with arm 1 was detected in 2017. Similarly, for adults, there was no significant difference between the prevalence of arm 2 (OR 1.0, 95% CI 0.5–2.1) or arm 3 (OR 0.9, 95% CI 0.4–1.8) compared with arm 1. Adjusting for baseline prevalence and inverse probability weight, respectively, in 9- to 12-year old children and adults in 2017 and preceding years, did not reveal significant differences between the arms either. However, a trend of reduced odds of infection was observed for 9- to 12-year children attending schools in shehias with additional snail control and for adults living in shehias with additional behavioral interventions from 2015 till 2017 (S3 Table).
Table 2 and the trend lines in Fig 3A–3F show that the overall S. haematobium prevalence decreased on Pemba and Unguja islands from 2011/12 to 2017 in all study populations. The relative difference from baseline to the final survey in Pemba was -79.5% (from 8.2% to 1.7%) in 9- to 12-year old children, -76.8% (from 12.2% to 2.8%) in 1st year students, and -68.1% (from 5.5% to 1.7%) in adults. In Unguja the relative difference was -58.2% (from 4.1% to 1.7%) in 9- to 12-year old children, -55.5% (from 5.2% to 2.3%) in 1st year students, and -48.7% (from 2.5% to 1.2%) in adults. The egg reduction rate from baseline to the final survey was 0.8%, 0.8%, and 0.5% for 9- to 12-year old children, 1st year students, and adults in Pemba, and 0.4%, 07% and 0.4%, for respective groups, in Unguja. In 2017, heavy intensity infections occurred in 0.4% of 9- to 12-year old students, 1.2% of 1st year students, and 0.2% of adults in Pemba, and in 0.3% of 9- to 12-year old children, 0.5% of 1st year students, and 0.1% of adults in Unguja. Considering 9- to12-year old children, there were 5/45 schools in Pemba and 3/44 schools in Unguja where heavy infection intensities occurred in ≥1% of the students in 2017. Considering 1st year students and adults, there were 13/45 schools and 4/45 shehias in Pemba and 7/44 schools and 2/45 shehias in Unguja, respectively, with heavy infection intensities ≥1%.
The violin plots in Fig 3A–3F indicate that the number of schools/shehias with high prevalence decreased over the years. Similarly, the number of clusters with a high percentage of heavy infection intensities was reduced. However, temporal heterogeneity was observed.
The maps in Fig 4A–4C show that there was also spatial heterogeneity. The maps indicate the change in the S. haematobium prevalence in all 90 study schools/shehias stratified by study population over time. Most schools/shehias had a baseline prevalence of <10% and decreased to <5% in 2017; also some schools with a baseline prevalence >10% decreased to <5% in 2017. In 2017, considering 9- to 12-year old children, 37/44 schools in Unguja and 42/45 schools in Pemba had a prevalence of <5%. Considering 1st year students in 2017, 37/44 schools in Unguja and 37/45 schools in Pemba had a prevalence of <5%. Considering adults in 2017, 42/45 shehias in Unguja and 42/45 shehias in Pemba had a prevalence of <5%. However, some schools/shehias remained at or bounced back to a prevalence ≥5% in certain years.
Fig 5A, 5B and 5D–5F show that the praziquantel coverage of 9- to 12-year old children receiving directly observed SBT in the 45 study schools in Pemba and Unguja, respectively, improved over the years and was mostly >75% in 2015. The average coverage in Pemba was 85.8% in 2013, 93.9% in 2014, and 96.8% in 2015. In Unguja, some schools did not receive treatment in 2013; the average coverage was 61.9%. In 2014, only CWT but no SBT was conducted; the average coverage was 68.7%. In 2015, all schools were targeted and the average coverage was 92.7%.
Praziquantel compliance of 20- to 55-year old adults receiving CWT in 2013 and 2015 (Fig 5A, 5B, 5C, 5E and 5F) or praziquantel via health posts in Pemba in 2014 (Fig 5D) was <75% in most shehias in all years. The average compliance in Pemba was 48.8% in 2013, 58.8% in 2014, and 42.3% in 2015. In Unguja, the average compliance was 61.5%, 69.2%, and 62.3%, respectively.
Participants who did not receive treatment were at a considerably higher risk of harboring a S. haematobium infection compared with treated participants. The 9- to 12-year old children who participated at the SBT rounds in 2013 till 2015 preceding our annual survey had lower odds of infection (OR: 0.77; 95% 0.58–1.0). Adults who complied with praziquantel treatment had significantly lower odds of infection (OR: 0.68; 95% CI: 0.53–0.87).
The Zanzibar islands are one of the first settings in sub-Saharan Africa aiming for elimination of urogenital schistosomiasis as public health problem and interruption of transmission. To help achieve this goal the ZEST alliance was formed. From 2012 onward praziquantel was administered in two annual treatment rounds to the whole eligible population by the Neglected Diseases Program of the Zanzibar Ministry of Health. In addition, within a cluster randomized intervention trial conducted from 2011/12 through 2017, randomized shehias received snail control or behavior change interventions [10]. We assessed the impact of the interventions on the S. haematobium prevalence and intensity in schoolchildren and adults from 90 schools and shehias, respectively, from 2011/12 through 2017. Annually, more than 12,000 individuals were included in cross-sectional parasitological surveys.
Over this five year period, urogenital schistosomiasis was eliminated as public health problem from most of the study sites in Pemba and Unguja. The S. haematobium prevalence was significantly reduced, but transmission was not interrupted and infections persisted despite intense control interventions.
A trend of lower odds of S. haematobium infection in children visiting schools in shehias that received snail control in addition to biannual MDA in comparison with participants from biannual MDA-only shehias was observed from 2015 onward, when the GEES were adjusted for baseline prevalence. However, confidence intervals were wide and the difference between the arms was not statistically significant. Similarly, adults living in areas with behavioral interventions showed lower odds of infection from 2015 onward than their counterparts who only received biannual MDA, but the difference was not significant. Noteworthy, working in an elimination setting with a very low overall prevalence, it was clear from initial design that our study was not powered to detect the statistical significance of small differences between the study arms [10], and of course only small differences could be observed at these levels.
Our results show, that while transmission was not interrupted in the study period and at the scale the interventions were applied, island-wide biannual CWT and SBT, plus additional snail control or behavior change measures in randomized shehias, were able to reduce significantly the overall S. haematobium prevalence and infection intensity in all participant groups from baseline in 2011/12 till the final survey in 2017. On both, Pemba and Unguja, the overall prevalence was <3% in all participant groups in 2017.
Moreover, in most schools and shehias, urogenital schistosomiasis had been eliminated as a public health problem in 2017. The 1st year students were identified as the group carrying most of the heavy infection intensities. Considering this participant group, 13/45 schools in Pemba and 7/44 schools in Unguja still had students with heavy infection intensities ≥1% in 2017. It seems that praziquantel distribution in CWT does not effectively reach children who are not (yet) attending primary school and receiving SBT. On Pemba and Unguja, an effort to increase coverage of preschool-aged children was made by including nursery and madrassa schools in the biannual treatment rounds since 2015. In future, the reach needs to be further increased so that young children benefit more from treatment. In addition, future island-wide or hotspot-focused behavioral and snail control interventions can help to educate children, parents, teachers and the general community about urogenital schistosomiasis transmission and prevention and consolidate treatment by lowering the risk of reinfection, respectively. While a study published in 2008 reported urogenital schistosomiasis to be rare in preschool-aged children [24] and we detected a low overall prevalence in all surveyed population groups in 2017, a new and larger study assessing the extent of S. haematobium infections throughout all age-groups is needed to provide greater insight into which sectors of the community infections are occurring.
A clear limitation of our study that might have resulted in some bias of our sample is that we recruited children in schools and adults in their homes. Non-school attenders and adults being out of their homes were not included, but might harbor heavy infections and contribute significantly to transmission [6]. Moreover, we used basic parasitological methods for diagnosis, which are not very sensitive, particularly for detecting ultra-light intensity infections [21]. Hence, the S. haematobium prevalence in our study is likely underestimated. Whether or not such undiagnosed ultra-light intensity infections significantly contribute to transmission is, however, unclear [25].
Importantly, our study showed that there was considerable temporal and also spatial heterogeneity. In some schools and shehias, the prevalence decreased in some years but bounced back to considerably high levels in other years. These shehias and schools were mostly not individually isolated, but clustered in areas covering neighbouring shehias in the north, center and east of Pemba and in the north and center of Unguja. Treatment coverage at cluster level did not directly explain an increase in prevalence. More likely to be responsible for the existence and persistence of hotspots are factors such as frequent human contact with natural open freshwater bodies, where intermediate host snails live and transmission occurs. The frequency of water contact is likely determined by the proximity of households and schools to water bodies, attractiveness of the water bodies for farming, household and leisure activities, and inconvenient access to safe water and alternative leisure options [26–28]. While these factors were not explored in our study, the presence of intermediate host snails at human water contact sites and their patent infections with S. haematobium were assessed in the snail control shehias [10]. We found that very few snails (<1%) were shedding cercariae and observed that snails repopulated treated sites soon after mollusciciding [14]. Hence, the risk of reinfection of humans and snails remains, and focal and repeated snail control should contribute to further reducing transmission in remaining hotspot areas.
In future, to achieve elimination as a public health problem across Pemba and Unguja, interventions will need to be improved and sustained. Biannual MDA alone, despite excellent coverage in schools and moderate coverage in adults over 5 years, was not always sufficient to achieve this goal. While in our elimination setting and CRT, the effect of snail control and behavioral interventions on prevalence and intensity was smaller than expected and statistically not significant, the interventions are likely to have a greater impact if applied at different scale and for a longer period. Recent reviews and a meta-analysis have shown that chemical mollusciciding is an effective way to control schistosomiasis and reduce prevalence [29, 30]. Behavior change and improving knowledge, attitudes and practices take time, but are essential since community engagement and compliance with interventions are key to sustained success. Combining all available interventions, including biannual MDA, snail control, behavioral change, and if possible, also improved access to safe water and sanitation, with best possible coverage and across neighboring hotspot shehias, might contribute to a further and sustainable reduction of prevalence and infection intensity.
In parallel to focusing improved intervention packages to hotspot areas, it will be important to start establishing surveillance-response as intervention in areas with very low prevalence. Moving towards interruption of transmission as the ultimate goal, a system will need to be in place that is able to diagnose and treat infected individuals rapidly, in order to avoid reintroduction and recrudescence in areas that become free of transmission [30–32]. To sustain and advance the gains made by ZEST, interventions must be continued long-term, improved, and adapted to the micro-epidemiology of S. haematobium in Zanzibar. Expansion of the MDA program to reach pre-school age children effectively is an essential requirement in Zanzibar and other endemic countries.
|
10.1371/journal.pgen.1000786 | Comparative ICE Genomics: Insights into the Evolution of the SXT/R391 Family of ICEs | Integrating and conjugative elements (ICEs) are one of the three principal types of self-transmissible mobile genetic elements in bacteria. ICEs, like plasmids, transfer via conjugation; but unlike plasmids and similar to many phages, these elements integrate into and replicate along with the host chromosome. Members of the SXT/R391 family of ICEs have been isolated from several species of gram-negative bacteria, including Vibrio cholerae, the cause of cholera, where they have been important vectors for disseminating genes conferring resistance to antibiotics. Here we developed a plasmid-based system to capture and isolate SXT/R391 ICEs for sequencing. Comparative analyses of the genomes of 13 SXT/R391 ICEs derived from diverse hosts and locations revealed that they contain 52 perfectly syntenic and nearly identical core genes that serve as a scaffold capable of mobilizing an array of variable DNA. Furthermore, selection pressure to maintain ICE mobility appears to have restricted insertions of variable DNA into intergenic sites that do not interrupt core functions. The variable genes confer diverse element-specific phenotypes, such as resistance to antibiotics. Functional analysis of a set of deletion mutants revealed that less than half of the conserved core genes are required for ICE mobility; the functions of most of the dispensable core genes are unknown. Several lines of evidence suggest that there has been extensive recombination between SXT/R391 ICEs, resulting in re-assortment of their respective variable gene content. Furthermore, our analyses suggest that there may be a network of phylogenetic relationships among sequences found in all types of mobile genetic elements.
| Integrative and conjugative elements (ICEs) are a class of mobile genetic elements that are key mediators of horizontal gene flow in bacteria. These elements integrate into the host chromosome, yet are able to excise and transfer via conjugation. Our understanding of ICE evolution is rudimentary. Here, we developed a method to capture ICEs on plasmids, thus facilitating their sequencing. Comparative analyses of the DNA sequences of ICEs from the same family revealed that they have an identical genetic structure consisting of syntenous, highly conserved core genes that are interrupted by clusters of diverse variable genes. Unexpectedly, many genes in the core backbone proved non-essential for ICE transfer. Comparisons of the variable gene content in the ICEs analyzed revealed that these elements are mosaics whose genomes have been shaped by inter–ICE recombination. Finally, our work suggests that ICEs contribute to a larger gene pool that connects all types of mobile elements.
| There are three types of self-transmissible mobile genetic elements: plasmids, bacteriophages and integrative conjugative elements (ICEs). All three classes of elements enable horizontal transmission of genetic information and all have had major impacts on bacterial evolution [1]–[4]. ICEs, (aka conjugation transposons), like plasmids, are transmitted via conjugation; however, unlike plasmids, ICEs integrate into and replicate along with the chromosome. Following integration, ICEs can excise from the chromosome and form circular molecules that are intermediates in ICE transfer. Plasmids and phages have been the subject of more extensive study than ICEs and while there is growing understanding of the molecular aspects of several ICEs [5]–[10], to date there have been few reports of comparative ICE genomics [11],[12] and consequently understanding of ICE evolution is only beginning to be unraveled.
Diverse ICEs have been identified in a variety of gram-positive and gram–negative organisms [13]. These elements utilize a variety of genes to mediate the core ICE functions of chromosome integration, excision and conjugation. In addition to a core gene set, ICEs routinely contain genes that confer specific phenotypes upon their hosts, such as resistance to antibiotics and heavy metals [14]–[18], aromatic compound degradation [19] or nitrogen fixation [20].
SXT is an ∼100 Kb ICE that was originally discovered in Vibrio cholerae O139 [16], the first non-O1 serogroup to cause epidemic cholera [21]. SXT encodes resistances to several antibiotics, including sulfamethoxazole and trimethoprim (which together are often abbreviated as SXT) that had previously been useful in the treatment of cholera. Since the emergence of V. cholerae O139 on the Indian subcontinent in 1992, SXT or a similar ICE has been found in most clinical isolates of V. cholerae, including V. cholerae serogroup O1, from both Asia and Africa. Other vibrio species besides V. cholerae have also been found to harbor SXT-related ICEs [22]. Furthermore, SXT-like ICEs are not restricted to vibrio species, as such ICEs have been detected in Photobacterium damselae, Shewanella putrefaciens and Providencia alcalifaciens [23]–[25]. Moreover, Hochhut et al [26] found that SXT is genetically and functionally related to the so-called ‘Inc J’ element R391, which was derived from a South African Providencia rettgeri strain isolated in 1967 [27]. It is now clear that Inc J elements are SXT-related ICEs that were originally misclassified as plasmids. In the laboratory, SXT has a fairly broad host range and can be transmitted between a variety of gram-negative organisms [16].
The SXT/R391 family of ICEs is now known to include more than 30 elements that have been detected in clinical and environmental isolates of several species of γ- proteobacteria from disparate locations around the globe [28]. SXT/R391 ICEs are grouped together as an ICE family because they all encode a nearly identical integrase, Int. Int, a tyrosine recombinase, is considered a defining feature of these elements because it enables their site-specific integration into the 5′ end of prfC, a conserved chromosomal gene that encodes peptide chain release factor 3 [29]. Int mediates recombination between nearly identical element and chromosome sequences, attP and attB respectively [29]. When an SXT/R391 ICE excises from the chromosome, Int, aided by Xis, a recombination directionality factor, mediates the reverse reaction - recombination between the extreme right and left ends (attR and attL) of the integrated element - thereby reconstituting attP and attB [6],[29]. The excised circular SXT form is thought to be the principal substrate for its conjugative transfer. The genes that encode activities required for SXT transfer (tra genes) were originally found to be distantly related to certain plasmid tra genes [30]–[32]. The tra genes encode proteins important for processing DNA for transfer, mating pair formation and generating the conjugation machinery. Regulation of SXT excision and transfer is at least in part governed by a pathway that resembles the pathway governing the lytic development of the phage lambda. Agents that damage DNA and induce the bacterial SOS response are thought to stimulate the cleavage and inactivation of SetR, an SXT encoded λ cI-related repressor, which represses expression of setD and setC, transcription activators that promote expression of int and tra genes [5].
The complete nucleotide sequences of SXT (99.5kb) and R391 (89kb) were the first SXT/R391 ICE family genomes to be reported [14],[32]. Comparative [33] and functional genomic analyses [5],[32] revealed that these 2 ICEs share a set of conserved core genes that mediate their integration/excision (int and xis), conjugative transfer (various tra genes), and regulation (setR, setCD). In addition to the conserved genes, these 2 ICEs contain element specific genes that confer element specific properties such as resistance to antibiotics or heavy metals. Interestingly, many of these genes were found in identical locations in SXT and R391, leading Beaber et al [33] to propose that there are ‘hotspots’ where SXT/R391 ICEs can acquire new DNA. The genomes of two additional SXT/R391 ICEs, ICEPdaSpa1, isolated from Photobacterium damselae [23], and ICESpuPO1, derived from an environmental isolate of Shewanella putrefaciens [24] are now also known. These two genomes also share most of the conserved set of core genes present in SXT and R391 and contain element specific DNA.
Determination of the sequences of SXT/R391 family ICE genomes was a fairly arduous task due to their size and predominantly chromosomal localization. Here, we developed a method to capture and then sequence complete SXT/R391 ICE genomes. In addition, we identified 3 as yet unannotated SXT/R391 ICE genomes in the database of completed bacterial genomes. Comparative analyses of the 13 SXT/R391 genomes now available allowed us to greatly refine our understanding of the organization and conservation of the core genes that are present in all members of this ICE family. Comparative and functional analyses also facilitated our proposal of the minimal functional SXT/R391 ICE genome. Furthermore, this work provides new knowledge of the considerable diversity of genes and potential accessory functions encoded by the variable DNA found in these mobile elements. Finally, this comparative genomics approach has allowed us to garner clues regarding the evolution of this class of mobile elements.
To date, ICE sequencing has been cumbersome because it has typically required construction of chromosome-derived cosmid libraries and screening for sequences that hybridize to ICE probes [23],[32]. We constructed a vector (pIceCap) that enables capture of complete SXT/R391 ICE genomes on a low-copy plasmid to simplify the protocol for ICE sequencing. This plasmid is a derivative of the single-copy modified F plasmid pXX704 [34],[35], which contains a minimal set of genes for F replication and segregation but lacks genes enabling conjugation. We modified pXX704 to include an ∼400bp fragment that encompasses the SXT/R391 attachment site (attB) and thereby enabled Int-catalyzed site-specific recombination between attB on pIceCap and attP on an excised and transferred ICE to drive ICE capture (Figure 1). Conjugations between an SXT/R391 ICE-bearing donor strain and an E. coli recipient deleted for prfC (and thus chromosomal attB) and harboring pIceCap yielded exconjugants containing the transferred ICE integrated into pIceCap (Figure 1). We used the ΔprfC recipient to bias integration of the transferred ICE into pIceCap rather than the chromosome. In these experiments, we selected for exconjugants containing the transferred ICE integrated into pIceCap, using an antibiotic marker present on the ICE as well as a marker present in pIceCap. The low copy IceCap::ICE plasmid was then isolated and used as a substrate for shotgun sequencing. We also found that the IceCap::ICE plasmids were transmissible. Thus, in principle this technique should facilitate capture of ICEs that do not harbor genes conferring resistance to antibiotics, by mating out the IceCap::ICE plasmid into a new recipient and selecting for the marker on pIceCap.
A list of the 13 SXT/R391 ICEs whose genomes were analyzed and compared in this study is shown in Table 1. All of the ICEs included in our analyses contain an int gene that was amplifiable using PCR primers for intsxt [29]. They were isolated on 4 continents and from the Pacific Ocean during a span of more than 4 decades. They are derived from 7 different genera of γ-proteobacteria and the ICEs derived from V. cholerae strains are from both clinical and environmental isolates of 3 different V. cholerae serogroups.
Five of these ICE genome sequences were determined at the J. Craig Venter Institute (JCVI) using the ICE capture system described above (Table 1, rows 1–5). In addition, we sequenced ICEVflInd1, also at the JCVI, by isolating cosmids that encompassed this V. fluvialis derived ICE prior to developing the ICE capture technique (Table 1, row 6). Table 1 (rows 7–10) also includes 4 previously unannotated ICE genomes that we found in BLAST searches of the NCBI database of completed but as yet unannotated genomes; 3 of these ICEs are clearly members of SXT/R391 ICE family since they are integrated into their respective host's prfC locus and contain int genes that are predicted to encode Int proteins that are 99% identical to Intsxt. The fourth element, ICEVchBan8 does not encode an Intsxt orthologue; however, this element contains nearly identical homologues of most of the known conserved core SXT/R391 ICE family genes. ICEVchBan8 will be discussed in more detail below but since it does not contain an Intsxt orthologue it is not considered a member of the SXT/R391 family of ICEs and thus not included in our comparative study. Finally, Table 1 also includes the 4 SXT/R391 ICEs that were previously sequenced (Table 1, rows 11–14).
Despite the diversity of our sources for SXT/R391 ICEs, the genomes of two pairs of ICEs that we analyzed proved to be very similar. SXTMO10 and ICEVchInd4 only differed by 13 SNPs in 7 genes and by the absence from ICEVchInd4 of dfr18, a gene conferring trimethoprim resistance. These ICEs were derived from V. cholerae O139 strains isolated in India from different cities at different times: SXTMO10 from Chennai in 1992 and ICEVchInd4 from Kolkata in 1997. The high degree of similarity of these two ICE genomes suggests that ICEs can be fairly stable over time. ICEVchBan9 and ICEVchMoz10 were also extremely similar although ICEVchMoz10 lacks dfrA1, another allele for trimethoprim resistance. These two ICEs were derived from V. cholerae O1 strains from Bangladesh (1994) and Mozambique (2004) respectively. The great similarity of these ICEs suggests that there has been spread of SXT-related ICEs between Asia and Africa in recent times. Studies of CTX prophage genomes have also suggested the spread of V. cholerae strains between these continents [36].
The ICEs listed in Table 1 were initially compared using MAUVE [37] and LAGAN [38], programs that enable visualization of conserved and variable regions on a global scale. All of the SXT/R391 ICEs we analyzed share a common structure and have sizes ranging from 79,733 bp to 108,623 bp (Table 1 and Figure 2). They contain syntenous sets of 52 conserved core genes (Figure 2A) that total approximately 47kb and encode proteins with an average of 97% identity to those encoded by SXT. All of the individual ICEs also contain DNA that is relatively specific for individual elements (Figure 2B); the differences in the sizes of the variable regions accounts for the range in ICE sizes.
Five sites within the conserved SXT/R391 ICE structure have variable DNA present in all of the ICEs in Figure 2. Four of these sites were previously termed ‘hotspots’ for ICE acquisition of new DNA [33]. Due to similarities between SXT and R391, the fifth hotspot only became apparent through our comparison of the 13 ICEs examined here. Each of these hotspots (HS1 to HS5 in Figure 2B) is found in an intergenic region (see below), suggesting that the acquisition of these variable DNA regions has not interrupted core ICE gene functions. In addition, some of the ICEs have variable DNA inserted in additional intergenic locations or in rumB (labeled I–IV in Figure 2B). Previous analyses [32] indicated that the insertion in rumB, did not impair SXT transmissibility. Overall, comparison of these 13 SXT/R391 ICE genomes suggests that: 1) these elements consist of the same perfectly syntenous and nearly identical 52 core genes that serve as a scaffold (see below) capable of mobilizing a large range of variable DNA; and 2) selection pressure to maintain ICE mobility has restricted insertions of variable DNA into sites that do not interrupt core functions.
The 52 core genes present in all the SXT/R391 ICEs analyzed include sets of genes that are known to be required for the key ICE functions of integration/excision, conjugative transfer and regulation [32] as well as many genes of unknown function. Most genes of known or putative (based on homology) function (coded by gray shading or hatch marks in Figure 2A) are clustered with genes that have related functions. For example, int and xis, genes required for integration and excision, are adjacent and setR, and setC/D, the key SXT regulators are near each other at the extreme 3′ end of the elements, although separated by 4 conserved genes of unknown function. Each ICE also has four gene clusters implicated in conjugative DNA processing and transfer (shown in light gray in Figure 2A). Finally, each of the ICEs has a nearly identical origin of transfer (oriT), a cis-acting DNA site that is thought to be nicked to initiate DNA processing events during conjugative transfer [39], in the same relative location.
The conserved core genes include approximately as many genes of unknown function as genes of known function. Some of the genes of unknown function are found either interspersed amongst gene clusters that likely comprise functional modules (e.g s091 between traD and s043) while others are grouped together (e.g. most genes between traN and traF). In several cases, the interspersed genes appear to be part of operons with genes of known function (e.g. s086-s082 maybe in an operon with setDC).
In addition to sharing 52 core genes, all of the ICE genomes analyzed contain variable DNA regions, ranging in size from 676 to 29,210 bp. Most of the variable DNA sequences are found in 5 intergenic hotspots (Figure 2B). However, some ICEs contain additional variable DNA inserts outside the 5 hotspots. For example, SXT and five other ICEs in Figure 2 have variable DNA segments, corresponding to related ISCR2 elements, disrupting rumB (Figure 2B, site III). ISCR2 elements are IS91-like transposable elements that tend to accumulate antibiotic resistance genes [40]. Interestingly, it is unusual for the contents of the hotspots and other variable regions to be found in only one ICE. Instead, the variable gene content of most of the ICEs shown in Figure 2B is found in more than one ICE. For example, ICESpuPO1, ICEPalBan1, and ICEVflInd1, all have identical contents in hotspot 5 (lavender genes in hotspot 5 in Figure 2B); however, the contents of the other hotspots in these 3 elements are almost entirely different. Thus, the variable gene content of the SXT/R391 ICEs reveals that these elements are mosaics. The overlapping distribution of variable DNA segments seen in the ICEs in Figure 2B suggests that recombination among this family of mobile elements may be extensive. In addition, in some instances, the variable regions appear subject to additional genetic modifications. For example, ICEPdaSpa1 and ICEVchBan9 contain ICE-specific DNA nested within the shared sequences inserted at hotspot 5 DNA (the green and pink genes in hotspot 5 in these elements, Figure 2B).
The variable genes encode a large array of functions and only a few will be discussed here. A complete list of the diverse genes found in the hotspots is found in Table S1. Although we cannot predict functions for many genes found in the hotspots, since they lack homology to genes of known function, at least a subset of the known genes seem likely to confer an adaptive advantage upon their hosts. Most of the ICE antibiotic resistance genes are found within transposon-like structures (e.g., the ISCR2 elements noted above) but four ICEs contain a dfrA1 cassette, which confers resistance to trimethoprim [25], in a class IV integron located in hotspot 3. A disproportionate number of variable genes are likely involved in DNA modification, recombination or repair, as they are predicted to encode diverse putative restriction-modification systems, helicases and endonucleases. Such genes may provide the host with barriers to invasion by foreign DNA including phage infection and/or promote the integrity of the ICE genome during its transfer between hosts. Three ICEs contain genes that encode diguanylate cyclases [41] in hotspot 3. These enzymes catalyze the formation of cyclic-diguanosine monophosphate (c-di-GMP), a second messenger molecule that regulates biofilm formation, motility and virulence in several organisms including V. cholerae [42],[43]. Most SXT/R391 ICEs contain mosA and mosT in hotspot 2. These two genes encode a novel toxin-antitoxin pair that promotes SXT maintenance by killing or severely inhibiting the growth of cells that have lost this element [44]. Not all ICEs in the SXT/R391 family contain mosAT; however, those lacking these genes may encode similar systems to prevent ICE loss. For instance, R391 and ICEVchMex1 contain two genes (orf2 and orf3) encoding a predicted HipA-like toxin and a predicted transcriptional repressor distantly related to the antitoxin HipB.
The variable regions found in the 5 hotspots are found exclusively in intergenic regions, punctuating the conserved ICE backbone (Figure 2). The boundaries between the conserved and variable sequences were mapped on the nucleotide level and compared (Figure 3A–3E). Each hotspot had a distinct boundary. Remarkably, even though the contents of the variable regions markedly differ, with few exceptions the left and right boundaries between conserved and variable DNA for each hotspot was identical among all the ICEs (Figure 3). For example, the left junctions of the inserts in hotspot 2 immediately follow the stop codon of traA and the right junctions are exactly 79 bp upstream of the start of s054 (Figure 3B), despite the fact that the DNA contents within these borders greatly differ. In hotspot 2, the right junction appears to begin with a 15 bp sequence that has two variants (Figure 3B, brown & light brown sequence). These sequences may reflect the presence of earlier insertions that have since been partially replaced. A similar pattern was found adjacent to the left boundary of hotspot 4 in several ICEs (Figure 3D, lines 3–6). Once an insertion is acquired, the number of permissive sites for the addition of new variable DNA likely increases.
There are two exceptions to the precise boundaries between variable and conserved DNA. Hotspot 1 and hotspot 3 in ICEVchMex1 and ICEPdaSpa1, respectively, contain variable DNA that extends beyond the boundary exhibited by all the other ICEs in these locations (Figure 3A, line 3, and Figure 3C, line 7). The only boundary that could not be identified was the left border of hotspot 5, the region containing genes between s026 and traI. As discussed below, s026 is the least conserved core gene and its variability obscured any consensus sequence abutting the variable DNA. Perhaps this border has eroded because s026 is not required for ICE mobility [32].
The relative precision of most boundaries between conserved and variable DNA sequences in all the ICEs analyzed suggests that a particular recombination mechanism, such as bet/exo-mediated recombination, may explain the acquisition of the variable regions. However, at this point, we cannot exclude the possibility that the precise locations for variable DNA insertions simply reflects selection for optimal ICE fitness; i.e., ICEs can optimally accommodate variable DNA in these locations while preserving their essential functions.
Unexpectedly, BLAST analyses revealed that most of the conserved core SXT/R391 genes are also present in IncA/C conjugative plasmids. These multidrug resistance plasmids are widely distributed among Salmonella and other enterobacterial isolates from agricultural sources [45],[46]. Recently, members of this family of plasmids have also been identified in Yersinia pestis, including from a patient with bubonic plague [47], and in aquatic γ-proteobacteria [48], including Vibrio cholerae [49],[50]. To date, the closest known relatives of the SXT/R391 transfer proteins are found in the IncA/C plasmids. Every predicted SXT transfer protein is encoded by the IncA/C plasmid pIP1202 isolated from Y. pestis [50] and the identities of these predicted protein sequences vary from 34 to 78% (Figure 4A). Furthermore, there is perfect synteny between the four gene clusters encoding the respective conjugative machineries of these two mobile elements (yellow and orange genes in Figure 4A). Despite the extensive similarity of the SXT and IncA/C conjugative transfer systems, these plasmids lack homologues of setR and setD/C as well as int/xis, suggesting that regulation of conjugative transfer differs between these elements.
The similarity of IncA/C plasmids and SXT/R391 ICEs is not limited to genes important for conjugal DNA transfer. Ten genes of unknown function (shown in black in Figure 4A), some of which are interspersed within likely tra gene operons and some of which are clustered together between traN and traF, are similar in the two elements. Furthermore, most of these ten genes are in identical locations in the two elements. Both elements also contain homologs of bet and exo (shown in green in Figure 4A); these are the only known homologs of the λ Red recombination genes found outside of bacteriophages. Together, the similarity of DNA sequences and organization of SXT/R391 ICEs and IncA/C plasmids suggests that these elements have a common ancestor. The fact that the contents of the hotspots in the two classes of elements are entirely distinct suggests that their evolutionary paths diverged prior to acquisition of these variable DNA segments.
The conservation of the 52 core genes in all 13 SXT/R391 ICEs analyzed suggested that many or even all of these genes would be required for key ICE functions of excision/integration, conjugative transfer and regulation. The presence of ten ICE core genes of unknown function in IncA/C plasmids (black genes in Figure 4A) is also consistent with the hypothesis that these genes might be required for ICE transfer. However, our previous work demonstrated that not all genes recognized here as part of the conserved core gene set are required for SXT transfer. Beaber et al showed that deletion of rumB – s026 (which includes 5 cores genes) from SXT had no detectable influence on SXT excision or transfer [32]. Therefore, we systematically deleted all of the core ICE genes whose contributions had not previously been assessed, in order to explore the hypothesis that these genes (especially those also present in IncA/C plasmids) would be essential for ICE transfer and to define the minimum functional SXT/R391 gene set.
Surprisingly, deletion of most of the ICE core genes of unknown function, including genes with homologues in IncA/C plasmids, did not alter SXT transfer efficiency. Deletion of s002 or s003, which are located downstream of int in all SXT/R391 ICEs, did not alter the frequency of SXT transfer; similarly, deletion of s082, s083, and s084, core genes of unknown function that are found near the opposite end of SXT/R391 ICEs but not in IncA/C plasmids, also did not influence SXT transfer frequency (Figure 4B). Furthermore, deletion of s091, which is found between traD and s043 in ICEs and IncA/C plasmids, did not reduce SXT transfer (Figure 4B). In contrast, deletion of s043, which has weak homology to traJ in the F plasmid (a gene important in DNA processing) and is located in a transfer cluster containing traI and traD, abolished transfer (Figure 4B, Δd), suggesting that s043, here re-named traJ is required for SXT transfer. It is unlikely that the transfer defect of SXTΔtraJ can be explained by polar effects of the deletion on downstream genes, since traJ appears to be the last gene of an operon found immediately upstream of hotspot 1. Similarly, deletion of s054, which is found immediately 5′ of traC and is homologous to a disulfide-bond isomerase dsbC, also abolished transfer (Figure 4B, Δe). Interestingly, disulfide bond-isomerases are present in several other conjugative systems [51]. However, it is not clear at this point if the deletion of s054 from SXT accounts for the transfer defect of SXTΔs054, since we could not restore transfer by complementation.
Additionally, Beaber et al found that deletion of s060 through s073 in SXT, which includes 7 genes that are also found in IncA/C plasmids reduced SXT transfer more than 100-fold [32]. We constructed several smaller deletions in this region and found that deletion of s063, which is also found in pIP1202, reduced the transfer frequency of SXT by ∼100-fold, nearly the same amount as deleting the entire region (Figure 4B). Complementation analyses revealed that the absence of s063 accounted for the transfer defect of SXTΔs063 (data not shown). Even though SXTΔs063 was still capable of transfer, in our view, the drastic reduction in the transfer frequency of this mutant warrants inclusion of s063 into the minimum functional SXT ICE genome (shown in Figure 4C). Other deletions in this region, including deletions of bet, exo, s067, s068 and s070, which have orthologues in IncA/C plasmids, resulted in ≤10-fold reductions in transfer frequency. We therefore did not include these genes in the minimal functional core SXT/R391 genome (Figure 4C).
The findings from our experiments testing the transfer frequencies of SXT derivatives harboring core gene deletions (shown in Figure 4B), coupled with our previous work demonstrating the requirements for the predicted SXT tra genes in the element's transfer [32], suggest a minimal functional SXT/R391 ICE structure as shown in Figure 4C. This minimum element is ∼29.7 kb and consists of 25 genes. Genes with related functions, which in some cases encode proteins that likely form large functional complexes (such as the conjugation apparatus), are grouped together in the minimal genome. At the left end of the minimum ICE genomes are xis and int, the integration/excision module of SXT/R391 ICEs. In the minimal ICE genome, the ICE oriT and mobI, which encodes a protein required for SXT transfer [39], are no longer separated from the other genes (traIDJ) that are also thought to play roles in the DNA processing events required for conjugative DNA transfer. The genes required for formation of the conjugation machinery, including the pilus, and mating pair formation and stabilization [32],[39] are divided between three clusters (denoted mpf1-3 in Figure 4C). Finally, at the right end of the minimal functional genome are the genes that regulate ICE transfer (setC/D and setR). Thus, the minimal functional SXT/R391 ICE is relatively small and organized into 3 discrete functional modules that mediate excision/integration, conjugation, and regulation.
Even though deletion of 27 out of 52 SXT/R391 ICE core genes proved to have little or no effect on SXT transfer frequency, and hence these genes were not included in Figure 4C, it is reasonable to presume that these genes encode functions that enhance ICE fitness given their conservation. For example, the presence of highly conserved bet and exo genes in all SXT/R391 ICEs suggests that there has been selection pressure to maintain this ICE-encoded recombination system that promotes ICE diversity by facilitating inter ICE recombination (G Garriss, MK Waldor, V Burrus, in press). A key challenge for future studies will be to determine how core genes of unknown function promote ICE fitness.
To identify genes in the SXT/R391 core genome that may be subject to different selection pressures, we compared the percent identity of each ICE's core genes to the corresponding SXT gene (Figure 5). Most of the ICEs' core genes exhibited 94% to 98% identity on the nucleotide level to SXT's core genes. There was no discernable difference in the degree of conservation of most core genes that were or were not part of the minimal ICE, suggesting that there are equal selective pressures on essential and non-essential genes. However, we identified 8 genes (s026, traI, orfZ, s073, traF, eex, s086, and setR) that exhibit significantly different degrees of conservation (Figure 5 and Figure S1). Three of these showed unusually high conservation, while the other 5 had below average conservation. Two of the highly conserved genes, setR and s086, are found at the extreme 3′ end of the elements. The conservation of setR may reflect the key role of this gene in controlling SXT gene expression. S086 may also play a role in regulating SXT transfer [52]. The other highly conserved gene, orfZ, is found between bet and exo and has no known function.
s026 and s073 are the most divergent of all the genes in the backbone. s026 encodes a hypothetical protein with homologues in many gram negative organisms. Although S026 is predicted to contain a conserved domain, COG2378, which has a putative role in transcription regulation, this protein is not required for SXT transfer [32]. The significant divergence of s026 along with its lack of essentiality suggests that this gene could become a pseudogene. A similar argument could be made for s073, which encodes a hypothetical protein that is also not required for ICE transfer. However, this argument does not hold for traI or traF, two genes which are essential for ICE transfer. Although the reasons which account for the different degrees of conservation of these 8 core genes are hard to ascertain at this point, the data in Figure 5 suggests that individual core genes are subject to different evolutionary pressures.
We created phylogenetic trees for each core gene based on their respective nucleotide sequences to further explore the evolution of the conserved backbone of SXT/R391 ICEs. Since we found such a high degree of conservation for most of the core genes, the bootstrap values for most of these trees were relatively low. Thus, we concentrated on the most polymorphic genes found in Figure 5, s026, s073, traI, and eex, for phylogenetic analyses. As shown in Figure 6A, the trees for s026, traI and s073 exhibit 3 distinct branching patterns. The lack of similarity in these phylogenetic trees suggests that either individual core genes have evolved independently or that high degrees of recombination mask their common evolutionary history. The latter hypothesis seems more likely since experimental findings have revealed that SXT/R391 ICEs can co-exist in a host chromosome in tandem [26] and recombination between tandem elements can yield novel hybrid ICEs with considerable frequency [53] (G Garriss, MK Waldor, V Burrus, in press). Also, as noted above, the distributions of variable genes among the ICEs shown in Figure 2 also supports the idea that inter-ICE recombination is commonplace.
Unlike most core genes, the trees for traG and eex were similar. In these two trees, the ICEs segregate into two evolutionarily distinct groups (Figure 6B), confirming and extending previous observations that revealed that there are two groups of eex and traG sequences in SXT/R391 ICEs [54]. These two groups correspond to the two functional SXT/R391 ICE exclusion groups. Interactions between traG and eex of the same group mediate ICE exclusion [55]. Thus, the identical 2 clusters of traG and eex sequences observed in their respective trees reveals the co-evolution of the traG/eex functional unit. The two groups of eex sequences can also be observed in Figure 5 where the bifurcating pattern reveals the 2 exclusion groups. This pattern is difficult to discern for traG, perhaps because of the large size of this multi-functional gene.
The sequence of ICEVchBan8, which was derived from a non-O1, non-O139 V. cholerae strain, is incomplete but it appears to contain 49 out of 52 SXT/R391 core genes. However, since this strain lacks Intsxt it was not included in our comparative analyses above. It is not known if ICEVchBan8 is capable of excision or transmission; however, it contains a P4-like integrase and a putative xis. It is tempting to speculate that the genome of ICEVchBan8 provides an illustration of how acquisition (presumably via recombination) of a new integration/excision module may generate a novel ICE family.
Comparative analysis of the genomes of the 13 SXT/R391 ICEs studied here has greatly refined our understanding of this group of mobile genetic elements. These elements, which have been isolated from 4 continents and the depths of the Pacific Ocean, all have an identical genetic structure, consisting of the same syntenous set of 52 conserved core genes that are interrupted by clusters of diverse variable genes. All the elements have insertions of variable DNA segments in the same five intergenic hotspots that interrupt the conserved backbone. Furthermore, some of the elements have additional insertions outside the hotspots; however, in all cases the acquisition of variable DNA has not compromised the integrity of the core genes required for ICE mobility. Functional analyses revealed that less than half of the conserved genes are necessary for ICE transmissibility and the contributions of the 27 core genes of unknown function to ICE fitness remains an open question. Finally, several observations presented here suggest that recombination between SXT/R391 ICEs has been a major force in shaping the genomes of this widespread family of mobile elements.
Although comparisons of the 13 ICE genomes analyzed here strongly suggest that these mobile elements have undergone extensive recombination during their evolutionary histories, there is a remarkable degree of similarity among the SXT/R391 ICEs. All of these ICEs consist of the same syntenous and nearly identical 52 genes. In contrast, other families of closely related mobile elements, such as lambdoid or T4-like phages for example, exhibit greater diversity [56],[57]. Since the elements that we sequenced were isolated from several different host species and from diverse locations, the great degree of similarity of the SXT/R391 ICE family does not likely reflect bias in the elements that we sequenced. It is possible that this family of mobile elements is a relatively recent creation of evolution and has yet to undergo significant diversification.
To date, relatively few formal comparative genomic analyses of other ICE families have been reported. Mohd-Zain et al [11] identified several diverse ICEs and genomic islands that shared a largely syntenous set of core genes with ICEHin1056, an ICE originally identified in Haemophilus influenzae. However, even though these elements share a similar genomic organization, they exhibit far greater variability in the sites of insertion of variable DNA and in the degree of conservation in their core genes compared to SXT/R391 ICEs. Thus, although this group of elements appears to share a common ancestor, they seem to have diverged earlier in evolutionary history than the SXT/R391 ICEs. However, when comparative genomic analyses were restricted to ICEHin1056-related ICEs found in only two Haemophilus sp., Juhas et al found that, like the SXT/R391 family of ICEs, these 7 ICEHin1056-related ICEs share greater than 90% similarity at the DNA level in their nearly syntenous set of core genes [12]. It will be interesting to learn the extent of conservation of genetic structure and DNA sequence in additional ICE families to obtain a wider perspective on ICE evolution.
Comparative genomic studies of bacteriophages have led to the idea that the full range of phage sequences are part of common but extremely diverse gene pool [58],[59]. The SXT/R391 ICE genomes suggest that there may be an even larger network of phylogenetic relationships linking sequences found in all types of mobile genetic elements including phages, plasmids, ICEs and transposons. The genomes of SXT/R391 ICEs appear to be amalgams of genes commonly associated with other types of mobile elements. Many of the ICE core genes are usually associated with phages, such as int, bet, exo and setR, or with plasmids, such as the tra genes. Additionally, the SXT/R391 ICEs and IncA/C plasmids clearly have a common ancestor, as we found that the entire set of SXT/R391 tra genes are also present in IncA/C plasmids. Thus, the genes present in all types of mobile genetic elements appear to contribute to a common gene pool from which novel variants of particular elements (such as ICEVchBan8) or perhaps even novel types of mobile genetic elements can arise.
ICEPalBan1, ICEVchMex1, ICEVchInd4, ICEVchInd5 and ICEVchBan5 were isolated using the plasmid capture system described in Figure 1. The SXT chromosomal attachment sequence, attB, was introduced into the modified F plasmid pXX704 [34] to create pIceCap. This plasmid was then introduced into a ΔprfC derivative of the TcR E. coli strain CAG18439. Exconjugants derived from matings between this strain and those harboring the 5 ICEs listed above resulted in strains carrying a pIceCap::ICE plasmid. Once captured, the plasmids were isolated using the Qiagen plasmid midi kit for low-copy plasmids (Qiagen). Isolated pIceCap::ICE plasmids were then sequenced.
ICEVflInd genome was determined by sequencing several overlapping cosmids that encompassed this ICE's genome. Briefly, genomic DNA from a Vibrio fluvialis strain carrying ICEVflInd was prepared using the GNome DNA kit (QBIOgene). Sau3A1 restricted genomic DNA was used to create a SuperCos1 (Stratagene)-based cosmid library according the manufacture's instructions. The library was subsequently screened for cosmids containing ICE-specific sequences using PCR with primers to conserved core ICE sequences. Four cosmids containing overlapping ICEVflInd sequences were identified and sequenced.
The genomes of 6 ICEs were sequenced by the Sanger random shotgun method [60]. Briefly, small insert plasmid libraries (2–3 kb) were constructed by random nebulization and cloning of pIceCap::ICE DNA or of cosmid DNA for ICEVflInd. In the initial random sequencing phase, 8–12 fold sequence coverage was achieved. The sequences of either pIceCap or pSuperCos were subtracted and the remaining sequences were assembled using the Celera Assembler [61]. An initial set of open reading frames (ORFs) that likely encode proteins was identified using GLIMMER [62], and those shorter than 90 base pairs (bp) as well as some of those with overlaps eliminated.
Nucleotide and amino acid conservation were assessed with the appropriate BLAST algorithms. ICEs were aligned using clustalW with default settings [63]. MAUVE [37] and LAGAN [38] were used to identify core genes in Figure 2. To map the boundaries of the hotspots, sequence comparisons were made using MAUVE and then manually compared to find boundaries between conserved and variable DNA as shown in Figure 3.
Phylogenetic trees were generated from alignments of nucleotide sequences using the neighbor-joining method as implemented by ClustalX software, version 2.011 [64]. The reliability of each tree was subjected to a bootstrap test with 1000 replications. Trees were edited using FigTree 1.22 (http://tree.bio.ed.ac.uk/software/figtree/).
CAG81439 harboring SXT was used as the host strain to create the SXT deletion mutants shown in Figure 3; the deletions were constructed using one-step gene inactivation as previously described [44],[65]. The primers used to create the deletion mutants are available upon request. Matings were conducted as previously described [16],[44] using deletion mutants and a KnR E. coli recipient, CAG18420. Exconjugants were selected on LB agar plates containing chloramphenicol, 20µg/ml (for SXT selection) and kanamycin, 50 µg/ml. The frequency of exconjugant formation was calculated by dividing the number of exconjugants by the number of donors.
|
10.1371/journal.pcbi.1003800 | Classical Mathematical Models for Description and Prediction of Experimental Tumor Growth | Despite internal complexity, tumor growth kinetics follow relatively simple laws that can be expressed as mathematical models. To explore this further, quantitative analysis of the most classical of these were performed. The models were assessed against data from two in vivo experimental systems: an ectopic syngeneic tumor (Lewis lung carcinoma) and an orthotopically xenografted human breast carcinoma. The goals were threefold: 1) to determine a statistical model for description of the measurement error, 2) to establish the descriptive power of each model, using several goodness-of-fit metrics and a study of parametric identifiability, and 3) to assess the models' ability to forecast future tumor growth. The models included in the study comprised the exponential, exponential-linear, power law, Gompertz, logistic, generalized logistic, von Bertalanffy and a model with dynamic carrying capacity. For the breast data, the dynamics were best captured by the Gompertz and exponential-linear models. The latter also exhibited the highest predictive power, with excellent prediction scores (≥80%) extending out as far as 12 days in the future. For the lung data, the Gompertz and power law models provided the most parsimonious and parametrically identifiable description. However, not one of the models was able to achieve a substantial prediction rate (≥70%) beyond the next day data point. In this context, adjunction of a priori information on the parameter distribution led to considerable improvement. For instance, forecast success rates went from 14.9% to 62.7% when using the power law model to predict the full future tumor growth curves, using just three data points. These results not only have important implications for biological theories of tumor growth and the use of mathematical modeling in preclinical anti-cancer drug investigations, but also may assist in defining how mathematical models could serve as potential prognostic tools in the clinic.
| Tumor growth curves display relatively simple time curves that can be quantified using mathematical models. Herein we exploited two experimental animal systems to assess the descriptive and predictive power of nine classical tumor growth models. Several goodness-of-fit metrics and a dedicated error model were employed to rank the models for their relative descriptive power. We found that the model with the highest descriptive power was not necessarily the most predictive one. The breast growth curves had a linear profile that allowed good predictability. Conversely, not one of the models was able to accurately predict the lung growth curves when using only a few data points. To overcome this issue, we considered a method that uses the parameter population distribution, informed from a priori knowledge, to estimate the individual parameter vector of an independent growth curve. This method was found to considerably improve the prediction success rates. These findings may benefit preclinical cancer research by identifying models most descriptive of fundamental growth characteristics. Clinical perspective is also offered on what can be expected from mathematical modeling in terms of future growth prediction.
| Neoplastic growth involves a large number of complex biological processes, including regulation of proliferation and control of the cell cycle, stromal recruitment, angiogenesis and escape from immune surveillance. In combination, these cooperate to produce a macroscopic expansion of the tumor volume, raising the prospect of a possible general law for the global dynamics of neoplasia.
Quantitative and qualitative aspects of the temporal development of tumor growth can be studied in a variety of experimental settings, including in vitro proliferation assays, three-dimensional in vitro spheroids, in vivo syngeneic or xenograft implants (injected ectopically or orthotopically), transgenic mouse models or longitudinal studies of clinical images. Each scale has its own advantages and drawbacks, with increasing relevance tending to coincide with decreasing measurement precision. The data used in the current study are from two different in vivo systems. The first is a syngeneic Lewis lung carcinoma (LLC) mouse model, exploiting a well-established tumor model adopted by the National Cancer Institute in 1972 [1]. The second is an orthotopic human breast cancer xenografted in severe combined immunodeficient (SCID) mice [2].
Tumor growth kinetics has been an object of biological study for more than 60 years (see e.g. [3] as one of the premiere studies) and has been experimentally investigated extensively (see [4] for a thorough review and [5]–[8] for more recent work). One of the most common findings for animal [9] and human [10]–[12] tumors alike is that their relative growth rates decrease with time [13]; or equivalently, that their doubling times increase.
These observations suggest that principles of tumor growth might result from general growth laws, often amenable to expression as ordinary differential equations [14]. The utility of these models can be twofold: 1) testing growth hypotheses or theories by assessing their descriptive power against experimental data and 2) estimating the prior or future course of tumor progression [9], [15] either as a personalized prognostic tool in a clinical context [16]–[20], or in order to determine the efficacy of a therapy in preclinical drug development [21], [22].
Cancer modeling offers a wide range of mathematical formalisms that can be classified according to their scale, approach (bottom-up versus top-down) or integration of spatial structure. At the cellular scale, agent-based models [23], [24] are well-suited for studies of interacting cells and implications on population-scale development, but computational capabilities often limit such studies to small maximal volumes (on the order of the mm3). The tissue scale is better described by continuous partial differential equations like reaction-diffusion models [19], [25] or continuum-mechanics based models [26], [27], when spatial characteristics of the tumor are of interest. When focusing on scalar data of longitudinal tumor volume (which is the case here), models based on ordinary differential equations are more adapted. A plethora of such models exist, starting from proliferation of a constant fraction of the tumor volume, an assumption that leads to exponential growth. This model is challenged by the aforementioned observations of non-constant tumor doubling time. Consequently, investigators considered more elaborate models; the most widely accepted of which is the Gompertz model. It has been used in numerous studies involving animal [9], [28]–[31] or human [12], [15], [30], [32] data. Other models include logistic [30], [33] or generalized logistic [11], [31] formalisms. Inspired by quantitative theories of metabolism and its impact on biological growth, von Bertalanffy [34] derived a growth model based on balance equations of metabolic processes. These considerations were recently developed into a general law of biological growth [35] and brought to the field of tumor growth [36], [37]. When the loss term is neglected, the von Bertalanffy model reduces to a power law (see [5], [38] for applications to tumor growth). An alternative, purely phenomenological approach led others [39] to simply consider tumor growth as divided into two phases: an initial exponential phase then followed by a linear regimen. Recently, influences of the microenvironment have been incorporated into the modeling, an example being the inclusion of tumor neo-angiogenesis by way of a dynamic carrying capacity [40], [41].
Although several studies have been conducted using specific mathematical models for describing tumor growth kinetics, comprehensive work comparing broad ranges of mathematical models for their descriptive power against in vivo experimental data is lacking (with the notable exception of [30] and a few studies for in vitro tumor spheroids [33], [42]–[44]). Moreover, predictive power is very rarely considered (see [42] for an exception, examining growth of tumor spheroids), despite its clear relevance to clinical utility. The aim of the present study is to provide a rational, quantitative and extensive study of the descriptive and predictive power of a broad class of mathematical models, based on an adapted quantification of the measurement error (uncertainty) in our data. As observed by others [45], specific data sets should be used rather than average curves, and this is the approach we adopted here.
In the following sections, we first describe the experimental procedures that generated the data and define the mathematical models. Then we introduce our methodology to fit the models to the data and assess their descriptive and predictive powers. We conclude by presenting the results of our analysis, consisting of: 1) analysis of the measurement error and derivation of an appropriate error model, subsequently used in the parameters estimation procedure, 2) comparison of the descriptive power of the mathematical models against our two datasets, and 3) determination of the predictive abilities of the most descriptive models, with or without adjunction of a priori information in the estimation procedure.
Animal tumor model studies were performed in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. Protocols used were approved by the Institutional Animal Care and Use Committee (IACUC) at Tufts University School of Medicine for studies using murine Lewis lung carcinoma (LLC) cells (Protocol: #P11-324) and at Roswell Park Cancer Institute (RPCI) for studies using human LM2-4LUC+ breast carcinoma cells (Protocol: 1227M). Institutions are AAALAC accredited and every effort was made to minimize animal distress.
For all the models, the descriptive variable is the total tumor volume, denoted by V, as a function of time t. It is assumed to be proportional to the total number of cells in the tumor. To reduce the number of degrees of freedom, all the models (except the exponential V0) had a fixed initial volume condition. Although the number of cells that actually remain in the established tumor is probably lower than the number of injected cells (∼60–80%), we considered 1 mm3 ( cells [49], i.e. the number of injected cells) as a reasonable approximation for V(t = 0).
For a given animal j and model M, the general setting considered for prediction was to estimate the model's parameter set using only the first n data points and to use these to predict at a depth d, i.e. to predict the value at time , provided that a measurement exists at this day (in which case it will be denoted by ). The resulting best-fit parameter set will be denoted .
The following method was used for analysis of the error made when measuring tumor volume with calipers. One volume per time point per cage was measured twice within a few minutes interval. This gave a total of 133 measurements over a wide range of volumes (20.7–1429 mm3). These were subsequently analyzed by considering the following statistical representationwhere Y is a random variable whose realizations are the measured volumes, is the true volume, ε is a reduced centered Gaussian random variable, and σE is the error standard deviation. The two measures, termed y1 and y2, were, as expected, strongly correlated (Figure 1A, ). Statistical analysis rejected variance independent of volume, i.e. constant E (, χ2 test) and a proportional error model (E = Y) was found only weakly significant (, χ2 test, see Figure 1B). We therefore introduced a dedicated error model, defined by(22)Two main rationales guided this formulation. First, we argued that error should be larger when volume is larger, a fact that is corroborated by larger error bars for larger volumes on growth data reported in the literature (see Figure 4 in [2] for an example among many others). This was also supported by several publications using a proportional error model when fitting growth data (such as [42], [60]). Since here such a description of the error was only weakly significant, we added a power to account for lower-than-proportional uncertainty in large measurements. Second, based on our own practical experience of measuring tumor volumes with calipers, for very small tumors, the measurement error should stop being a decreasing function of the volume because of detectability limits. This motivated the introduction of the threshold Vm. After exploration of several values of Vm and α, we found to be able to accurately describe dispersion of the error in our data (, χ2 test, see Figure 1C). This yielded an empirical value of
We did not dispose of double measurements for the breast tumor data and the error analysis was performed using the lung tumor data set only. However, the same error model was applied to the breast tumor data, as both relied upon the same measurement technique.
This result allowed quantification of the measurement error inherent to our data and was an important step in the assessment of each model's descriptive power.
We tested all the models for their descriptive power and quantified their respective goodness of fit, according to various criteria. Two distinct estimation procedures were employed. The first fitted each animal's growth curve individually (minimization of weighted least squares, with weights defined from the error model of the previous section, see Material and Methods). The second method used a population approach and fitted all the growth curves together. Results are reported in Figure 2 and Tables 1 and 2. Parameter values resulting from the fits are reported in Tables 3 and 4.
Figure 2A depicts the representative fit of a given animal's growth curve for each data set using the individual approach. From visual examination, the exponential 1 (1), logistic (2) and exponential-linear (1) models did not well explain lung tumor growth and the exponential 1 (1) and logistic (2) models did not satisfactorily fit the breast tumor growth data. The other models seemed able to describe tumor growth in a reasonably accurate fashion.
These results were further confirmed by global quantifications over the total population, such as by residuals analysis (Figure 2C) and global metrics reported in Tables 1 and 2. When considering goodness-of-fit only, i.e. looking at the minimal least squared errors possibly reached by a model to fit the data (metric in Tables 1 and 2), the generalized logistic model (3) exhibited the best results for both data sets (first column in Tables 1 and 2). This indicated a high structural flexibility that allowed this model to adapt to each growth curve and provided accurate fits. On the other hand, the exponential 1 (1) and logistic (2) models clearly exhibited poor fits to the data, a result confirmed by almost all the metrics (with the exception of the AICc).
The two models that were shown unable to describe our data in the previous section, namely the exponential 1 (1) and logistic (3) models, were excluded from further analysis. The remaining ones were assessed for their predictive power. The challenge considered was to predict future growth based on parameter estimation performed on a subset of the data containing only n data points (with n<Ij for a given j). We refer to the Materials and Methods section for the definitions of prediction metrics and success scores.
When relatively fewer data points were used, for example with only three, individual predictions based on individual fits were shown to be globally limited for the lung tumor data, especially over a large time frame (Figure 4.A, Table 5). However, this situation is likely to be the clinically relevant since few clinical examinations are performed before the beginning of therapy. On the other hand, large databases might be available from previous examinations of other patients and this information could be useful to predict future tumor growth in a particular patient. In a preclinical setting of drug investigation, tumor growth curves of animals from a control group could be available and usable when inferring information on the individual time course of one particular treated animal.
An interesting statistical method that could potentiate this a priori information consists in learning the population distribution of the model parameters from a given database and to combine it with the individual parameter estimation from the available restricted data points on a given animal. We investigated this method in order to determine if it could improve the predictive performances of the models. Each dataset was randomly divided into two groups. One was used to learn the parameter distribution (based on the full time curves), while the other was dedicated to predictions (limited number of data points). For a given animal of this last group, no information from his growth curve was used to estimate the a priori distributions. The full procedure was replicated 100 times to ensure statistical significance, resulting in respectively 2000 and 3400 fits performed for each model. We refer to the Materials and Methods for more technical details. Results are reported in Figure 5.
Predictions obtained using this technique were significantly improved for the lung tumors, going from an average success score of 14.9%±8.35% to 62.7%±11.9% (means ± standard deviations) for prediction of the total future curve with the power law model (6) (see Figure 5.A). Prediction success rates were improved even at large future depths. For instance, predictions 7 days in the future reached an average success rate of 50.6%, power law model (6), see Figure 5.C, while their success rate was very low with direct individual prediction (6.07%). Prediction successes reached 90% (power law model (6)) at the closer horizon of the next day data point (), while success rate was only 57.1% using an individual approach (Figure 5.B). Other small horizon depths also reached excellent prediction scores (Figure 5.C). The largest improvement of success rates for the power law model was observed for that went from an average score of 6.86% (with standard deviation 7.47) to an average score of 75.2% (with standard deviation 12.9), representing more than an 11-fold increase. We report in Figure S5 the details of predictions with and without a priori information for all the animals within a given forecast group from the lung tumor data set (power law model (6)). It can be appreciated how additional information on the parameter distribution in the estimation procedure significantly improved global prediction of the tumor growth curves. The impact of the addition of the a priori information was however less important when using more data points for the estimation (results not shown).
For the breast data, due to its already high prediction score without adjunction of a priori information, the exponential-linear model did not benefit from the method. For the next day data point of the breast tumor growth curves, predictability was already almost maximal without adjunction of a priori information and thus no important impact was observed.
For both data sets, not all the models equally benefited from the addition of a priori information (Figure 5). Models having the lowest parameter inter-animal variability, such as the power law (6), Gompertz (4), exponential-linear (1), and exponential V0 (1) models (Table 3), which also had better practical identifiability (Tables 2 and S3), exhibited great benefit. In contrast, the models with three parameters showed only modest benefit or even decrease of their success rates (see and for the von Bertalanffy model (6) on the breast tumor data in Figure 5.B), with the exception of the generalized logistic model (3) on the breast tumor data. In these cases, adjunction of a priori information translated into poor enhancement of predictive power because the mean population parameters did not properly capture the average behavior within the population and were therefore not very informative. On the other hand, models such as the power law model (6) on the lung tumor data set, whose coefficient γ characterized particularly well the growth pattern (Table 3), had a more informative a priori distribution that translated into the highest improvement of predictive power. For the generalized logistic model (3) on the breast data, the mean parameters were able to inform the linear regimen of the growth phase and thus protected the model from too early saturation.
These results demonstrated that addition of a priori information in the fit procedure considerably improved the forecast performances of the models, in particular when using a small number of data points and low-parameterized models for data with low predictability, such as the power law model for the lung tumor data set.
In our analysis, constant variance of the error was clearly rejected and although a proportional error (used by others [33]) was not strictly rejected by statistical analysis (p = 0.08), a more adequate error model to our data was developed. However, using a proportional or even constant error model did not significantly affect conclusions as to the descriptive power of the models, identifying the same models (Tables 1, 2) as most adequate for description of tumor growth (results not shown). Nevertheless, the use of an appropriate error model could have important implications in the quantitative assessment of a model's descriptive performance and rejection of inaccurate tumor growth theories. For instance, using the same human tumor growth data, Bajzer et al. [60] found the assumption of proportional error variance to favor the Gompertz model for descriptive ability, whereas Vaidya and Alexandro [30] had observed the logistic model to be favored, under a constant-variance assumption. The error model used might additionally have important implications on predictions. Although detailed analysis of the impact of the error model on prediction power is beyond the scope of the present study, we performed a prospective study of predictive properties when using a constant error model on the lung tumor data and found changes in the ranking of the models (results not shown).
As expected, our results confirmed previous observations [9]–[11], [13], [29], [32] that tumor growth is not continuously exponential (constant doubling time) in the range of the tumor volumes studied, ruling out the prospect of a constant proliferating fraction. A less expected finding was that the logistic model (linear decay in volume of the relative growth rate) was also unable to describe our data, although similar results have been observed in other experimental systems [31], [33]. On the other hand, the Gompertz and power law models could give an accurate and identifiable description of the growth slowdown, for both data sets. More elaborate models such as the generalized logistic, von Bertalanffy, and dynamic CC models could describe them as well. However, their parameters were found not to be identifiable from only tumor growth curves, in the ranges of the observed volumes. Additional data could improve identifiability, such as related to later growth and saturation details. It should be noted in this case that the dynamic CC model was not designed with the intent to quantify tumor growth, but rather to describe the effects of anti-angiogenic agents on global tumor dynamics. Because the model carries angiogenic parameters that are not directly measureable, or even inferable, from the experimental systems we used, it stands to reason that they would not be easily identifiable from the data. Kinetics under the influence of antiangiogenic therapy might thus provide useful additional information that could render this model identifiable. For the breast tumor experimental system, the slowdown was characterized by linear dynamics and was most accurately fitted by the exponential-linear model. Observed was exponential growth from the number of injected cells (during the unobserved phase) that switched smoothly to a linear phase (exponential-linear model). It should be noted that in the breast tumor data set, no data were available during the initiation phase (below 200 mm3) and only the linear part of a putative exponential-linear growth was observed. Explorations of the kinetics of growth during the initial phase (at volumes below the mm3) are needed for further clarification.
Despite structural similarities, important differences were noted in the parameter estimates between the two experimental models, in agreement with other studies emphasizing differences between ectopic and orthotopic growth [61], [62]. Our results and methodology may help to identify the impact on kinetics of the site of implantation, although explicit comparisons could not be made here due to the differences in the cell lines used.
The Gompertz model (exponential decay in time of the relative growth rate) was able to fit both data sets accurately, consistently with the literature [13], [15], [29], [31], [33]. One of the main criticisms of the Gompertz model is that the relative tumor growth rate becomes arbitrarily large (or equivalently, the tumor doubling time gets arbitrarily small) for small tumor volumes. Without invoking a threshold this becomes biologically unrealistic. This consideration led investigators [13], [63] to introduce the Gomp-exp model that consists in an initial exponential phase followed by Gompertzian growth when the associated doubling time becomes realistic. This approach could also be applied to any decreasing relative growth rate model. We did not consider it in our analysis due to the already large initial volume and the lack of data on the initiation phase where the issue is most relevant.
The power law model was also able to describe the experimental data and appeared as a simple, robust, descriptive and predictive mathematical model for murine tumor growth kinetics. It suggests a general law of macroscopic in vivo tumor growth (in the range of the volumes observed): only a subset of the tumor cells proliferate and this subset is characterized by a constant, possibly fractional, Hausdorff dimension. In our results, this dimension (equal to 3γ) was found to be significantly different from two or three (p<0.05 by Student's t-test) in 14/20 mice for the lung tumor data set and 13/34 mice for the breast tumor data, effectively suggesting a fractional dimension. A possible explanation of this feature could come from the fractal nature of the tumor vasculature [64], [65], an argument supported by others who have investigated the link between tumor dynamics and vascular architecture [37]. More precisely, the branching nature of the vascularization generates a fractal organization [37], [64], [65] that could in turn produce a contact surface of fractional Hausdorff dimension. Considering further that the fraction of proliferative cells is proportional to this contact surface (for instance because proliferative cells are limited to an area at fixed distance from a blood vessel or capillary, due to diffusion limitations), this could make the connection between fractality of the vasculature and proliferative tissue. These considerations could therefore provide a mechanistic explanation for the growth rate decay that naturally happens when the dimension of the proliferative tissue is lower than three. Our results were obtained using two particular experimental systems: an ectopic mouse syngeneic lung tumor and an orthotopic human xenograft breast tumor model. Although consistent with other studies that found the power law model adequate for growth of a murine mammary cell line [38] or for description of human mammography density distribution data [5], these remain to be confirmed by human data. This model should also be taken with caution when dealing with very small volumes (at the scale of several cells for instance) for which the relative growth rate becomes very large. Indeed, the interpretation of a fractional dimension then fails, since the tumor tissue can no longer be considered a continuous medium. In this instance, it may be more appropriate to consider exponential growth in this phase [37].
Our results showed that a highly descriptive model (associated to large flexibility) such as the generalized logistic model, might not be useful for predictions, while well-adapted rigidity – as provided by the exponential-linear model on the breast tumor data – could lead to very good predictive power. Interestingly, our study revealed that models having low identifiability (von Bertalanffy and dynamic CC) could nevertheless exhibit good predictive power. Indeed, over a limited time span, different parameter sets for a given model could generate the same growth curves, which would be equally predictive.
For the Gompertz model, predictive power might be improved by using possible correlations between the two parameters of this model, as reported by others [15], [63], [66]–[68] and suggested by our own parameter estimates (R = 0.99 for both data sets, results not shown).
If a backward prediction is desired (for instance for the identification of the inception time of the tumor), the use of exponential growth might be more adapted for the initial, latency phase, e.g. by employment of the Gomp-exp model [13], [63].
Translating our results to the clinical setting raises the possibility of forecasting solid tumor growth using simple macroscopic models. Use of a priori information could then be a powerful method and one might think of the population distribution of parameters being learned from existing databases of previous patient examinations. However, the very strong improvement of prediction success rates that we obtained partly comes from the important homogeneity of our growth data (in particular the LLC data) that generated a narrow and very informative distribution of some parameters (for instance parameter γ of the power law model), which in turn powerfully assisted the fitting procedure. In more practical situations such as with patient data, more heterogeneity of the growth data should be expected that could alter the benefit of the method. For instance, in some situations, growth could stop for arbitrarily long periods of time. These dormancy phases challenge the universal applicability of a generic growth law such as the Gompertz or power law [69]. Description of such dormancy phenomena could be integrated using stochastic models that would elaborate on the deterministic models reviewed here, as was done by others [70] to describe breast cancer growth data using the Gompertz model. Moreover, further information than just tumor volume could be extracted from (functional) imaging devices, feeding more complex mathematical models that could help design more accurate in silico prediction tools [18], [71].
Our analysis also has implications for the use of mathematical models as valuable tools for helping preclinical anti-cancer research. Such models might be used, for instance, to specifically ascertain drug efficacy in a given animal, by estimating how importantly the treated tumor deviates from its natural course, based on a priori information learned from a control group. Another application can be for rational design of dose and scheduling of anti-cancerous drugs [22], [72], [73]. Although integration of therapy remains to be added (and validated) to models such as the power law, more classical models (exponential-linear [39] or dynamic CC [41]) have begun to predict cytotoxic or anti-angiogenic effects of drugs on tumor growth. Our methods have allowed precise quantification of their respective descriptive and predictive powers, which, in combination with the models' intrinsic biological foundations, could be of value when deciding among such models which best captures the observed growth behaviors in relevant preclinical settings.
|
10.1371/journal.pgen.1001338 | Genome-Wide Interaction-Based Association Analysis Identified Multiple New Susceptibility Loci for Common Diseases | Genome-wide interaction-based association (GWIBA) analysis has the potential to identify novel susceptibility loci. These interaction effects could be missed with the prevailing approaches in genome-wide association studies (GWAS). However, no convincing loci have been discovered exclusively from GWIBA methods, and the intensive computation involved is a major barrier for application. Here, we developed a fast, multi-thread/parallel program named “pair-wise interaction-based association mapping” (PIAM) for exhaustive two-locus searches. With this program, we performed a complete GWIBA analysis on seven diseases with stringent control for false positives, and we validated the results for three of these diseases. We identified one pair-wise interaction between a previously identified locus, C1orf106, and one new locus, TEC, that was specific for Crohn's disease, with a Bonferroni corrected P<0.05 (P = 0.039). This interaction was replicated with a pair of proxy linked loci (P = 0.013) on an independent dataset. Five other interactions had corrected P<0.5. We identified the allelic effect of a locus close to SLC7A13 for coronary artery disease. This was replicated with a linked locus on an independent dataset (P = 1.09×10−7). Through a local validation analysis that evaluated association signals, rather than locus-based associations, we found that several other regions showed association/interaction signals with nominal P<0.05. In conclusion, this study demonstrated that the GWIBA approach was successful for identifying novel loci, and the results provide new insights into the genetic architecture of common diseases. In addition, our PIAM program was capable of handling very large GWAS datasets that are likely to be produced in the future.
| Recent studies on the genetic basis of common diseases have identified many loci that confer disease susceptibility. However, much of the heritability of these diseases remains unexplained. Loci involved in gene–gene interactions are considered cryptic, because they confer susceptibility, but may not generate a detectable signal on their own. These interactions may account for the “missing heritability” of common diseases. Theoretically, these interactions can be identified with the genome-wide interaction-based association analysis. But, in reality, very few gene–gene interactions have been identified with that method, and most were based on prior biological knowledge. Here, we applied a parallel computing technique that facilitated the identification of multiple new cryptic susceptibility loci involved in common diseases. We applied stringent control for false positives, and we validated our findings with independent datasets. This study demonstrated that interactions between gene loci could be successfully identified with the genome-wide interaction-based approach. With this approach, we also identified cryptic loci with moderate single-locus effects. The identified loci and interactions merit further investigations for fine mapping and functional analyses. Our results extend the current knowledge of common diseases for future studies in genetic mapping. This approach is applicable to current and future genome-wide association datasets.
| Recent genome-wide association studies (GWAS) have identified many common genetic variants associated with common diseases. This has rapidly expanded our knowledge of the genetic architecture of these diseases. For example, the Wellcome Trust Case Control Consortium (WTCCC) study [1] and other large-scale GWASs (including meta-analyses) have discovered many susceptibility loci for common diseases, including coronary artery disease (CAD) [2], Crohn's disease (CD) [3], [4], type 1 diabetes (T1D) [5], and type 2 diabetes (T2D) [6]. However, compared with the successes of single-locus approaches, the achievements of interaction-based approaches, which seek susceptibilities that derive from gene-gene interactions, have lagged behind [7], [8]. Thus, gene-gene interactions that are largely undetected may explain some of the heritability of common diseases [9]. Most reported interactions are currently found through candidate approaches, which incorporate prior biological knowledge. Moreover, very few interactions have been confirmed in an independent population.
Genome-wide interaction-based association (GWIBA) analysis uses markers to conduct genome-wide screens without prior candidate selection. In addition, GWIBA incorporates interaction effects among genetic variants. Many interaction-based methods for GWIBA are currently available, including a logistic regression-based method [10]; in addition, several methods have been recently developed [11]–[15]. However, no studies on real data have successfully identified novel disease-associated loci. Two studies reported non-significant results on small datasets [11], [14]; several studies with the WTCCC dataset reported problematic interactions [12], [13], [15], and those were found to be probable false positives in this study. Thus, the GWIBA methods have identified very few new loci convincingly, and none of the detected interactions have been replicated to date. In addition, the computational time was a major barrier for GWIBA analyses on large-scale GWAS datasets. Most previous studies resorted to stochastic searches, or partial search strategies based on biological knowledge [12]–[18]. Until recently, genome-wide association studies have followed the traditional single-locus approach and have investigated gene-gene interactions only through candidate approaches.
In this study, our main aim was to discover novel susceptibility loci by identifying interaction effects in a GWIBA analysis with the large-scale WTCCC dataset [1]. We also aimed to confirm these novel loci in independent datasets. To that end, we identified several novel susceptibility loci with replication/validation evidence, and the results provide new insights into the genetic architecture of common diseases.
We performed a complete GWIBA analysis with validation analyses. We started with the WTCCC dataset [1], which contained ∼2,000 cases for seven diseases and ∼3,000 shared controls (Materials and Methods). The quality-controlled WTCCC data were used as input for the “pair-wise interaction-based association mapping” (PIAM) program, and we performed an exhaustive two-locus search for each disease (Materials and Methods). We used the single-locus likelihood ratio test (LRT) p-value (5×10−7) as a cutoff value for incorporating the single-locus effects in the PIAM searches. The cutoff value was based on the significance threshold set by WTCCC for single-locus analyses. This prevented the marginal effects of a few loci from dominating the interactions. The computation was performed with the PIAM program running in parallel on computer clusters.
In the initial search, we used the cases and the shared controls of the WTCCC data to screen single-nucleotide polymorphism (SNP) pairs that passed a p-value threshold of P<50/L, where L was the total number of two-locus combinations for each disease. The threshold allowed SNP pairs with p-values that were 1,000 times larger than the significance level of 0.05/L. During the calculation, the distributions of two-locus statistics were evaluated with the approximate statistical distribution method in PIAM (Materials and Methods), which generated genome-wide two-locus quantile-quantile plots (Figure S1). We obtained 2,570 SNP pairs for seven diseases at these screening thresholds (Table S1A), after excluding 20,968 SNP pairs within the major histocompatability complex (MHC) region for rheumatoid arthritis (RA) and T1D (Table S1B).
Although many SNP pairs had rather significant p-values, there were an overwhelming number of false positive results observed. We found that the initial SNP quality control performed by the WTCCC was not sufficiently stringent for the interaction searches, due to sparse data and poor genotyping quality. The sparseness of the data was due to the constraint that we used two-locus genotype interaction analyses, instead of the single-locus analysis applied by the WTCCC; this relative sparseness of data conferred a higher sensitivity to genotyping errors. Therefore, a stringent additional SNP quality control was applied (Materials and Methods). A total of 1,392 SNP pairs passed this additional quality control (Table S1C).
After the initial search, these 1,392 SNP pairs were tested with the expanded controls to gain greater statistical power (Materials and Methods). We retained 634 SNP pairs that gave Bonferroni corrected P<0.5 (Table S1D), according to the numbers of available two-locus tests (Table S2).
Among the results from the 634 SNP pairs, we observed two major types of problematic results, irrespective of the SNP quality control. The first problem was that we found many “interactions” between known susceptibility loci with large marginal effects. These “interactions” might have resulted from marginal effects, according to the two-locus LRT tests that incorporated both marginal and pure interaction effects. To control for this problem, we used a strategy similar to BEAM [19], where we compared the two-locus p-values with the single-locus p-values. The second problem was that we found 88 SNP pairs with linked SNPs (Table S1E); most of these gave quite significant p-values, but were identified as artificial associations that could be separated into two types, one was a batch effect and the other was a genotype clustering problem. These artificial associations were due to sparse data and genotyping artifacts. Later, we found that some previously reported interactions were probably these kinds of artificial associations [12], [13], [15] (details in Discussion). Therefore, a stringent result filter was applied to filter out these false positive interactions (Materials and Methods). Thus, we removed 536 SNP pairs with excessive marginal effects, and 85 SNP pairs with the two kinds of artificial associations. Within the 88 SNPs pairs with linked SNPs, 3 pairs were not affected by artificial associations; therefore, these interactions were considered true haplotypic associations. These 3 SNP pairs were located in regions known to be associated with CD, thus, we did not present these results in detail here, except in the corresponding regional signal plots (Figure S2) and odds ratio (OR) tables (Table S3). Finally, 10 SNP pairs with unlinked SNPs remained qualified (Table S1F).
After the result filtering, the simultaneous searches identified an interaction between rs7522462 (on C1orf106) and rs11945978 (on TEC) for CD with a Bonferroni corrected P<0.05, and another five pairs of regions associated with CAD, CD, T1D, and T2D with Bonferroni corrected P<0.5 (Table 1; Figure 1). Among the above six pairs of regions, the interaction between rs7522462 and rs11945978 for CD, and the allelic effect of rs6470733 (close to SLC7A13) for CAD were replicated by proxy linked SNPs. In addition, we validated one pair of interacting regions around rs153423 (near SPRY4) and rs748855 (on NOD2) for CD, one single region around rs1501540, and one pair of interacting regions around rs11731175 and rs11236365 (on SLCO2B1) for T2D, all with nominal P<0.05, through local validation analyses (Materials and Methods; Table 1; Figure 2). We then performed the three-locus conditional searches based on the six pairs of SNPs listed in Table 1; this did not produce any significant results.
We did not identify any interactions for bipolar disorder (BD), hypertension (HT), or RA, according to the significance thresholds and result filtering applied (except the interactions within the MHC region for RA). In fact, a single-locus analysis did not identify significant results for HT, and only one significant locus was associated with BD, but this has not been replicated to date [1], [20]. This may indicate that the quality control and result filtering we performed was effective for removing random false positives and artificial associations.
Within each SNP pair in Table 1, the SNPs were independent in the controls and dependent in the cases (Table 2). The SNPs in Table 1 showed good genotype clustering (Figure S3), and did not present any significant deviations from Hardy-Weinberg equilibrium (HWE, P>0.05). Note that the corrected two-locus p-values in Table 1 were only corrected within each disease.
Only one pair of interacting loci was associated with CAD. The SNPs were rs9397512 and rs6470733, located at the intron of SYNE1 and 7 kb downstream from SLC7A13, respectively. Note that this interaction only gave a moderate corrected p-value of 0.380. However, this pair of regions generated wide, block-like interaction signals (Figure 1), with strong linkage disequilibrium (LD) (Figure S4, plotted with Haploview [21]). For this interaction, when the rs6470733 genotypes were paired with the TT genotype stratum of rs9397512, the effects were in the opposite direction compared to those observed when the rs6470733 genotypes were paired with the CC and CT strata (Table 3). The highest OR relative to the most common homozygote combination (2.95) was higher than the OR under the assumption of an additive effect of the two loci (1.95). The two loci also showed moderate single-locus allelic effects, especially rs6470733.
In order to validate the association of the two loci identified for CAD, we used the online results of the German MI Family Study (GerMIFS) [2], which included 875 cases and 1644 controls (Materials and Methods). Surprisingly, we found that rs13262822, which was 1.5 kb downstream from rs6470733 and had an r2 = 0.90 (based on the WTCCC shared controls), showed a rather significant allelic effect with a trend test P = 1.09×10−7 (Table 4). In addition, rs13262822 showed an allelic effect in the WTCCC data with a trend test P = 1.81×10−3 and an association in the same direction. Therefore, the allelic effect of the original SNP, rs6470733, was replicated by its proxy SNP rs13262822, in strong LD. Interestingly, in the GerMIFS data, the minor allele frequency (MAF) of rs13262822 was a bit larger than that in the WTCCC data, and the OR of 1.39 was much higher compared to the OR of 1.16 reported in the WTCCC. The previous paper did not identify the rs13262822 locus because the trend p-value of rs13262822 in the WTCCC data marginally failed the 0.001 threshold before the combined analysis [2]. At this time, the online result from the GerMIFS data is not sufficient to confirm the marginal effect of rs9397512 or the interaction effect. SYNE1 was previously suggested as a potential mediator of cardiomyopathy, because it showed muscle-specific inner nuclear envelope expression and a physical interaction with lamin A/C [22]. Furthermore, a recent study suggested that SYNE1 was involved in the pathogenesis of Emery Dreifuss muscular dystrophy through skeletal muscle cell destruction [23], which emphasized the functional role of SYNE1 in muscles. SLC7A13 is a cationic amino acid transporter, and two early studies showed that cationic amino acid transporters may be related to atherosclerotic lesion formation by regulating L-ornithine transport and polyamine synthesis in vascular smooth muscle [24], [25]. Thus, these two genes may be involved in different, but related aspects of CAD pathogenesis. This could explain the statistical interaction between the two regions.
Two pairs of interacting loci were associated with CD. The first interaction was between rs7522462, which is in the region of C1orf106 gene, and rs11945978, which is in a newly identified region of the TEC gene. The C1orf106 region was previously identified in a meta-analysis after the WTCCC study, which included data from both the WTCCC and from the national institute of diabetes, digestive, and kidney diseases (NIDDK) inflammatory bowel disease genetics consortium (IBDGC) [4]. This interaction gave a Bonferroni corrected P = 0.039. The interaction signal showed a clear block that extended over several tens of kb (Figure 1). In the IBDGC data, a weak interaction signal also appeared at the corresponding regions (Figure 2). For this interaction, the single-locus effect of rs7522462 varied significantly among the genotype strata of rs11945978, which indicated the interaction (Table 3), and the effect of rs7522462 was strongest in the rs11945978 CC stratum.
Validation analysis with the IBDGC data supported the interaction between rs7522462 and rs11945978 (Materials and Methods). We selected proxy SNPs in the IBDGC data instead of the original SNPs, according to HapMap [26] CEU r2 values. For rs7522462, two proxy SNPs were in moderate LD: rs296533, which is 16 kb upstream, with an r2 = 0.44, and rs296547, which is 10 kb downstream, with an r2 = 0.79. For rs11945978, the proxy SNP rs2089509 showed perfect linkage disequilibrium (LD) in the HapMap CEU population. The allelic effects, interaction effect, and combined effect of the proxy SNP combination of rs296533 and rs2089509 were replicated in the IBDGC non-Jewish population data (rs296533 trend P = 0.020, rs2089509 trend P = 0.047, pure interaction P = 0.013, two-locus P = 0.001). The ORs showed trends similar to those in the WTCCC data, particularly in the CC genotype stratum of rs11945978 (corresponding to the GG genotype stratum of rs2089509) (Table 5). The trend P of rs7522462 stratified by rs11945978 CC, and the trend P of rs296533 stratified by rs2089509 GG were 2.05×10−8 and 1.35×10−3, respectively. The risk alleles of rs7522462 and its proxy, rs296533, and the risk alleles of rs11945978 and its proxy, rs2089509, comprised the major haplotypes according to the HapMap data. This indicated the same association direction in the WTCCC data and the IBDGC non-Jewish population data. Although the interaction between rs296533 and rs2089509 was not significant in the IBDGC Jewish population data (with quite a small sample size), the interaction showed a similar pattern (Table S4). Nevertheless, the downstream proxy SNP, rs296547, had a larger r2 value of 0.79 and the interaction was not significant in the IBDGC data. This may be explained by the small sample size and the LD difference between the HapMap data and the IBDGC data for the marker loci and the causing loci. For SNPs that were either ungenotyped in the WTCCC or in the IBDGC non-Jewish population data (rs7522462, rs11945978, rs296533, rs2089509), the corresponding genotypes were imputed (Materials and Methods). We found a consistent interaction between rs296533 and rs2089509, which was significant in both the IBDGC non-Jewish population data (P = 0.013) and the imputed WTCCC data (P = 0.015), and they showed a similar interaction pattern (Table S4). A previous study found that the expression of TEC was up-regulated upon T-cell activation, and Tec overexpression in lymphocyte cell lines was sufficient to induce phosphorylation of phospholipase C gamma and activation of nuclear factor of activated T cells [27]; moreover, over-activation of T cells is a typical feature of CD.
The second interaction for CD was between rs153423 and rs748855, which gave a corrected P of 0.146. The latter SNP lies in the early identified NOD2 gene [1]; the former SNP is located about 100 kb upstream from the SPRY4 gene, and the association signal extended fairly close to the gene (Figure 1). The two-locus pattern showed that rs153423 was epistatic to rs748855, because the most common rs153423 genotype (AA) masked a considerable single-locus effect of rs748855 (Table 3). Locus-based replication for this interaction failed, and local validation of the interaction with the IBDGC non-Jewish population data indicated a nominally significant interaction (P = 0.034; Figure 2). A previous study showed that SPRY4 suppressed vascular epithelial growth factor-induced, Ras-independent activation of Raf1 [28]; moreover, another study suggested that vascular epithelial growth factor-A signaling was related to CD through angiogenesis [29].
Only one pair of interacting loci was associated with T1D. The SNPs, rs7310460 and rs2302270, interacted with a moderate corrected P of 0.252. The 12p13.31 region around rs7310460 was previously found in a meta-analysis study conducted after the WTCCC study [5]. This region harbors many immunoregulatory genes, including CLEC2D. In contrast, rs2302270 is mapped to an intergenic region. The association signal for the interaction effect extended about 100 kb for both regions with clear borders, and it included the previously suggested CD69 gene [5] (Figure 1). The association pattern showed that rs2302270 was epistatic to rs7310460 (Table 3). We currently have no available data to validate the association of the rs2302270 region or the interaction.
Two pairs of interacting loci were associated with T2D. The first interaction was between rs1501540 and rs7359782, which gave a corrected P of 0.082. The rs1501540 SNP is mapped to a region with no annotated genes, and rs7359782 is located 238 kb upstream of C18orf58. The interaction signal was very narrow; however it was not restricted to a single SNP (Figure 1). The interaction and the region around rs7359782 failed in the validation analysis. However, the region around rs1501540 showed large allelic effects and was validated in the GENEVA Diabetes Study data with the local validation strategy (P = 0.022; Figure 2). In contrast, we did not detect any SNPs that were both significant and in the same direction in the two populations. Interestingly, we found that the significant SNPs in each dataset showed different frequencies between the two populations (MAF = 0.27 in the WTCCC data, MAF = 0.14 in the GENEVA data, with respect to the most significant SNPs in the associated region of each dataset, rs1501540 and rs302001); but, within each population, the significant SNPs showed similar frequencies.
The second interaction for T2D was between rs11731175 and rs11236365, which gave a corrected two-locus P of 0.298. Neither of the SNPs showed obvious marginal effect (trend P and genotypic LRT test P>0.05). The rs11731175 SNP lies within a region where the nearest annotated gene is more than 500 kb away, and rs11236365 is mapped to the SLCO2B1 gene (Figure 1). The association pattern clearly showed that rs11731175 was epistatic to rs11236365. The GG and GT genotypes of rs11731175 masked the effect of rs11236365 (Table 3). However, when the genotype of rs11731175 was TT, rs11236365 showed a very strong effect. Moreover, the ORs of the TT and CT genotypes of rs11236365 relative to the most common homozygote combination were 1.83 and 0.33, respectively. It appeared that the C allele of rs11236365 provided a strong protective effect against T2D. The exact replication of this interaction with the GENEVA Diabetes Study data was not significant. Local validation of the interaction was nominally significant (P = 0.029), and the interaction signal was very close to the original signal (Figure 2). SLCO2B1 is an organic anion transporting polypeptide, and one of its substrates, dehydroepiandrosterone-sulfate (DHEA-S, a direct metabolite of DHEA) [30], was found in several early studies to increase insulin sensitivity in a T2D mouse model [31], [32], in rats [33]–[35], and in humans [36].
Recent studies on the genetics of common diseases have revealed a lot of susceptibility loci and produced many tools for data analyses. However, the GWIBA approaches, which are prospective methods for discovering novel interacting loci, had not succeeded in identifying convincing interactions. In the present study, we developed an effective GWIBA approach that facilitated the discovery of novel loci. First, we used the parallel search program, PIAM, and implemented a simple statistical method and an optimized algorithm for detecting interactions. This could complete two-locus exhaustive searches on large-scale GWAS data in a short time. Second, in addition to the initial search, we used expanded controls with large sample sizes to gain statistical power for detecting interactions. Third, the results were carefully examined, and we found the artificial associations as well as the “interactions” with excessive single-locus effects. Finally, we employed independent datasets to validate the detected interactions; moreover, we introduced the “local validation” method for the validation of interactions between populations, where confounding factors may affect the consistency of the observed interactions.
In Table 1, two regions were previously identified through meta-analysis studies. One region associated with CD, the C1orf106, which did not achieve significance in the WTCCC study, was subsequently identified in a meta-analysis study that included the WTCCC data and the IBDGC data [4]. We identified this region by including only the WTCCC data that showed corrected p-values<0.05. Also, the region on 12p13.31 that was associated with T1D was previously identified in a meta-analysis study of T1D [5]. These results demonstrated that this GWIBA approach enhanced the power of detecting loci with moderate single-locus effects; it also implied that some known susceptibility loci with moderate single-locus effects might be interacting with other loci. Moreover, we reasoned that interaction effects could increase the overall effects of loci that only showed marginal effects, and there were very few examples of large-effect common variants for common diseases [9]; therefore, we speculated that interactions of common variants may prefer to reside on loci with moderate to small single-locus effects. This hypothesis could explain a common phenomenon that there was seldom any significant interaction detected by the means of investigating interactions among loci with certain marginal effects after the single-locus analyses [1], [4], [5], [8].
Our exhaustive searches revealed several two-locus associations, where both the individual loci exhibited relatively small single-locus effects. The most extreme case was the interaction between rs11731175 and rs11236365 for T2D. Neither of the SNPs showed obvious marginal effect, but they exhibited an excessive interaction effect. This suggested that some interactions might be missed by using methods based on single-locus analysis or interaction-based approaches with non-exhaustive search strategies. On the other hand, we searched for genetic interactions associated with seven diseases and observed only one pair of loci that fell within that extreme situation. Theoretically, because the allele frequency of genetic loci often varies among different populations, it is relatively unlikely that marginal effects of the interactions will be obscured in all populations.
We have noticed that although some loci (or their good proxies) could not be replicated, the corresponding regions showed apparent association/interaction signals in the validation data. The signals were unlikely to be observed by chance given the local validation p-values. The replication failure for these loci was unlikely to simply result from insufficient statistical power; because unlike the replicated interaction for CD, we did not identify any consistency of the OR values between the datasets for these loci (data not shown), while the signals were observed. In addition, the signals were unlikely to be affected by genotyping artifacts, because multiple loci were considered and the data were initially quality-controlled. To our knowledge, tens of loci have been identified for some common diseases, but no interaction between exact loci has been confirmed in independent populations to date, despite of the fact that many of the loci are in the same pathway. Based on these observations, one plausible hypothesis is that the genetic heterogeneity may affect the consistency of the interactions. We speculated that many disease-causing interacting loci for common diseases might reside among rare variants that have large effects, and these rare variants could vary in frequency between populations, or they could be on adjacent, but distinct loci between populations. This could appropriately explain the lack of consistently observed interactions for common diseases in current GWASs that used common-variant markers.
In this study, we found an overwhelming number of false positives, including artificial associations, in the raw results. The problem of false positives was more severe in our two-locus analyses than in the single-locus analyses, because our two-locus genotype combinations had insufficient sample sizes, which made them very sensitive to the artificial genotyping errors that were widely present in GWAS data. In addition, sparse data caused inaccuracy on asymptotic tests. Therefore, the results of two-locus analyses require careful examination, and particular attention must be paid to incredibly small p-values.
In the raw results with linked SNPs, we identified two kinds of artificial interactions; the batch effect (Figure 3) and the genotype clustering problem (Figure 4). Note that, although these kinds of observations were exaggerated by LD, and therefore, were previously considered as LD effects [8] (rs2532292), they were, in fact, caused by genotyping artifacts (Figure S5). Thus, interactions with unlinked SNPs, particularly SNPs with low MAFs, also require careful examination. In some previous studies, we found probable false positive results of the same kinds. For example, in two previous works [12], [13], we conducted experimental searches on the WTCCC RA data without any quality control procedures; all the interactions that were outside the MHC region contained unqualified SNPs, according to the WTCCC study. Careful examination showed that many of these results were SNP pairs with linked SNPs, which were probably artificial associations of the two kinds mentioned in this study. Only one result was not affected by unqualified SNPs; but, when this was tested with quality-controlled samples, we observed a sharp decrease in significance. Moreover, a recent study [15] tested a new program on part of the WTCCC data and reported many interactions; however, almost all those results were interactions with linked SNPs that showed extremely significant p-values. We observed a large overlap between those reported SNPs and the SNPs that were filtered out in this study. In particular, two of the SNPs that were reported in that study (rs1065705 and rs1420247) were confirmed in this study to be affected by artificial genotyping errors (Table S1E). We also found that three regions, PLXNA2, PTPRT, and PPM1A that were reported to be “associated” with multiple diseases in the WTCCC data were extremely unlikely to be true interactions; in particular, we found that the PPM1A region, with the most significant p-value, was “associated” with all diseases except BD, and the association was probably a false positive. Therefore, we suggest that careful false positive control procedures should be adopted in future GWIBA studies to avoid misleading results and unnecessary endeavors in subsequent replication analyses.
There are a few limitations of this study. First, although GWIBA permits agnostic searching without the need for prior biological knowledge, it loses substantial power due to the penalty introduced by multiple testing corrections for the huge number of potential pair-wise interactions. Therefore, candidate-gene methods should not be discarded, because they offer promising, well-powered detection of interactions based on biological knowledge. For example, a previous study performed a partial search on genes within certain biological networks and obtained some significant interactions [18]. Second, the contingency tables used for fast computing could not incorporate continuous covariates. However, these might be very important in some genetic analyses. This problem might be partially addressed by incorporating the covariates after an initial screen for interactions with a loose threshold. Third, we had to compromise for the huge computational issue by using general tests that assumed no specific genetic models; this resulted in decreased power compared to a test that conforms to a certain specific model. Furthermore, detection of high-order interactions was restricted to the conditional search, in order to conserve the computational time. Fourth, two-locus associations should be interpreted with caution when the single-locus effect of one SNP is very large; validation analyses should be performed to further confirm pure interaction effects. Fifth, the non-pseudoautosomal region of the X chromosome was not included in this analysis due to the imbalanced proportions of males and females between the case and control groups; however, many susceptibility loci of common diseases may reside on the X chromosome. This problem might be addressed by stratifying the contingency tables with a sex covariate, and then removing the corresponding female individuals with heterozygote genotypes for the tested SNPs on the X chromosome. Finally, this method provided inflated test statistics to detect SNPs with low MAFs, which were removed from the analysis. The removal may have caused us to miss low-frequency variants with relatively large effects, and these loci may be more valuable than common variants with smaller effects [37]. These issues require further studies to be fully addressed. Thus, we do not unreservedly recommend the approach used in this analysis for detecting genetic interactions. Rather, we recommend further improvements to this method, and the use of other methods when appropriate. Nevertheless, we would like to emphasize that the procedures described here are important for ensuring the reliability of interactions.
We implemented PIAM with a multi-thread/parallel program, rapid tests for two-locus interactions, optional two-stage strategies for interaction searching, fast algorithms for collecting contingency tables with a binary genotype coding method [38], and an intrinsic CPU instruction for new types of CPUs. These components made PIAM capable of handling very large GWAS datasets that are anticipated to be commonly available in the future. For example, the WTCCC2 study will include much larger numbers of SNP markers and sample sizes for the identification of susceptibility loci with moderate single-locus effects and interactions. We estimated that, for a dataset with up to 1,000,000 SNPs and 10,000 samples, PIAM could complete an exhaustive, two-locus search within 6 days with one computer equipped with a modern quad-core, 3.0 GHz, desktop CPU and 4 G of memory; this speed could be multiplied with parallel computing on multiple computers.
This GWIBA approach can be used routinely, in addition to single-locus analyses in future genome-wide association studies. It is a promising approach for the discovery of novel loci with interaction effects, which may provide important insights into common diseases. By combining various approaches, we could greatly accelerate the discovery of the genetic architecture of common diseases.
The initial data were obtained from the WTCCC (http://www.wtccc.org.uk/). This dataset comprised ∼2,000 samples each for seven diseases (BD, CAD, CD, HT, RA, T1D, and T2D). For half of these samples, there were ∼3,000 shared control samples from the 1958 birth cohort (58C); for the other half, there were control samples from the National Blood Service (NBS). Genotyping was performed with the Affymetrix GeneChip Mapping 500K Array Set. Genotypes were called with the CHIAMO algorithm with the parameter “posterior probability less than 0.9” set to “missing”. The non-pseudoautosomal SNPs on the X chromosome were not included, because the general genotypic test was used in PIAM, and the male/female proportion was imbalanced between the cases and controls. Quality control for the samples and the SNPs was performed as described in the WTCCC. In addition, we excluded SNPs that were significant in the single-locus association analysis, but showed poor clustering according to the WTCCC. After trimming, 459,075 SNPs remained for each disease, and the corresponding data were used as input for the two-locus exhaustive searches.
We developed a fast, multi-thread/parallel program named “pair-wise interaction-based association mapping” (PIAM, available at http://www.ihs.ac.cn/xykong/PIAM.zip) to search for susceptibility SNPs with interaction effects in a set of genome-wide SNPs. PIAM is based on a two-locus logistic regression model and the likelihood ratio test (LRT). For the logistic regression model, the additive effect of a SNP was represented with a variable that was coded 0, 1, and 2 for homozygote, heterozygote, and the other homozygote (e.g., AA, AB, BB), respectively. We added another variable for the heterozygote effect that was coded 1 for heterozygote and 0 otherwise. Therefore, two variables were used for the general effects of one SNP. The interaction was modeled by the multiplication of variables between SNPs; thus, four terms were used for each pair-wise interaction. The interactions can be interpreted by their deviation from the restricted model without the interaction terms. The restricted model only considers the additive effect between the two loci on the log odds ratios, that is, the multiplicative effect between the two loci on the odds ratios. The full two-locus logistic model considers all possible effects of the two loci. Accordingly, the deviation between the two indicates the significance (or relevance) of the interaction.
A previous study proposed the use of a full, two-locus, logistic regression model and evaluated its statistical power [10]. However, when all the SNP pairs were tested with the LRT, and the null model (with only one intercept term) was compared to the full model of two SNPs (with an intercept term and eight terms for all the effects of two SNPs and their interactions), as previously proposed, there were excessive results associated with the single-locus effect of a single SNP. Therefore, for practical use, we modified the previous approach with the following strategy. First, a single-locus LRT for the general effect of each SNP was performed. Then, a family-wise-significant, single-locus, p-value threshold was used to divide the whole set of SNPs into two subgroups. One subgroup was small and significant (subset A with n SNPs) and the other subgroup comprised the remainder of SNPs (subset B with m SNPs). Then, we performed three types of searches:
For the conditional and simultaneous searches, the LRT statistics were calculated by the G2 test with contingency tables for fast computing. This method was equivalent to the LRT for the logistic regression models, but it did not estimate the parameters. In addition, the genotypes were transformed to a set of binary values to accelerate the collection of contingency tables, as proposed by a recent study [38]. For the simultaneous search, we also implemented in PIAM the previously proposed two-stage strategy [10], [39]. The Bonferroni correction was used for N multiple tests, where N was the total number of tests for all search situations. Missing genotypes were addressed by removing the corresponding individual. After the exhaustive two-locus search, the conditional search was extended to high-order interactions. For example, conditional on an existing two-locus interaction, the full two-locus model was compared to a full three-locus model by adding another locus; this resulted in an 18 d.f. LRT test.
The huge number of statistics (up to 1×1011) generated in this study would be extremely computationally demanding to handle directly. Therefore, to check the overall distributions of the observed two-locus test statistics, we implemented the approximate statistical distribution method in PIAM. First, very small, continuous intervals (e.g., 0.001 in length), were predefined for the LRT statistics; a maximum value of the statistic was set to be that with a corresponding p-value equal to 0.01/L, where L was the total number of comparisons; thus, the last interval was the maximum value to infinity. During the computation, PIAM recorded the number of statistics within each small interval, rather than the exact value of the statistic. When applying the statistics to the quantile-quantile plot, the statistics were treated as equal to the lower bound of the corresponding interval; therefore, the error of the statistic was controlled below 0.001. This method can also be used to handle p-values transformed by a negative logarithm. The approximate statistic distributions can further be used in various multiple test correction approaches. This method can, and should, be adopted in other GWIBA studies in the future to obtain the overall statistical distributions, similar to the single-locus analyses.
We found that the initial SNP quality control performed by the WTCCC was not sufficiently stringent for our interaction analysis. That control yielded an abundance of extremely significant interaction results, but these were subsequently identified as false positives, due to sparse data and/or poor genotyping quality. The sparse data were introduced by comparing the two-locus genotypes for the interaction analysis to the single-locus analysis. These relatively sparse data were more sensitive to genotyping errors. To avoid that problem, an additional, more stringent SNP quality control was applied. We removed any SNPs with a missing data rate that was >2% of the cases or controls, with a MAF<0.1 in the controls, or with a HWE p-value<0.001 in the controls. The removed SNPs with high missing data rates typically showed poor genotype clustering; the low frequency SNPs often yielded an inflated two-locus LRT statistic, due to sparse data (at least one expected cell count <5 in the two-locus contingency table for current sample size, assuming HWE for both SNPs with either or both MAFs<0.1); and a deviation from HWE in the control population was probably due to genotyping errors. After applying this additional SNP quality control, the test numbers changed for the Bonferroni correction, due to the removal of SNPs.
To improve the detection of true interactions that may not initially achieve significance within the cases and shared controls, we applied the expanded control analysis, as performed in the WTCCC study. The enlarged, “expanded control” was used for each disease to test pairs of loci with interactions that passed moderate screening p-value thresholds (50/L in this study, where L was the total number of two-locus combinations) in the initial analysis with the cases and shared controls, but did not necessarily achieve significance (i.e., a Bonferroni corrected P<0.05). In the expanded control analysis, the final statistical significance was evaluated. The expanded control for a certain disease was the combination of the shared control and some other disease cohorts. For BD, the expanded control group included the shared control plus the CAD, CD, HT, T1D, and T2D groups. For the three autoimmune diseases (CD, RA, and T1D), the expanded control included the shared control plus all other disease cohorts, except the autoimmune disease cohorts. The same was true for the three metabolism-related diseases CAD, HT, and T2D. These expanded controls were the same as those used in the WTCCC study. Note that, associations caused by diseases other than the disease of interest could be avoided in this expanded control analysis, because the first stage screening with the shared control required a low p-value.
The expanded control analysis was not an independent replication of the initial analysis; therefore, a genome-wide multiple test correction should also be used when testing the interactions retained in the initial analysis. For convenience, we used the same test numbers for correction in the expanded control analysis as those used in the initial analysis. That is, the two SNP subsets (subsets A and B) were the same as those in the initial analysis; therefore, the subset division was not based on the single-locus p-value of the expanded control. This is similar to the “joint analysis” strategy [40] of single-locus analyses in GWAS. However, an additional problem we encountered was the possibility that some SNPs in subset B might not pass the 5×10−7 p-value threshold in the initial analysis, but could pass it in the expanded control analysis. These SNPs had to be removed to avoid associations that were caused by a single-locus effect only. An alternative strategy could be to determine different subsets A and B for the expanded control analysis. These would be chosen according to the single-locus p-value threshold of the expanded control. Then, the corresponding search situation and the appropriate numbers of multiple tests would be used in the expanded control analysis.
First, some interactions detected by tests that incorporate marginal effects may result from marginal effects alone, without any pure interaction effects, and we used a strategy similar to BEAM (the hierarchical significance declaration procedure) [19] to address this problem. We compared two-locus p-values with single-locus p-values, as follows. For SNP pairs obtained in the simultaneous search, we compared the corrected two-locus p-value to the more significant corrected single-locus p-value of the two SNPs; for SNP pairs obtained the conditional search, we compared the corrected two-locus association p-value to the corrected single-locus p-value of the SNP in subset B; we removed SNP pairs that had two-locus p-values that were less significant than the single-locus p-values. The p-values in the expanded control analysis were used for these comparisons. In addition, we also removed SNP pairs with any SNPs that did not pass the 5×10−7 p-value threshold in the initial analysis, but passed the threshold in the expanded control analysis.
Second, we examined all SNP pairs that were located within 1 Mb of each other. Two kinds of artificial associations were found; one was a batch effect and the other was a genotype clustering problem. The batch effect was severe aggregation of samples of some individuals with particularly high risk, two-locus genotypes, in the 96-well plates. The genotype clustering problem was observed on genotype clustering plots; this manifested as an ambiguous extra cluster (beyond the normal three clusters) that the genotype calling algorithm classified differently between the case and control groups. SNP pairs with either of these problems were removed from the analyses.
Third, we further checked the regional interaction signals to avoid artificial associations due to errors in genotyping a given SNP. Only results with consecutive interaction signals were retained; i.e., an elevated interaction signal could be observed on at least two nearby SNPs from both regions. No results were excluded based on this check in this study.
We used the online analysis results of the German MI Family Study [2] (http://www.cardiogenics.imbs-luebeck.de/) to test for allelic effects in order to validate the pair of regions that we had associated with CAD. We did not have access to the individual-level genotype data from that study to validate the interaction. The genotyping platform was the Affymetrix GeneChip Mapping 500K Array Set. The SNPs were quality-controlled; only SNPs with a trend test P<0.001 were shown on the website. We searched the website for any SNPs that showed significant single-locus effects within 50 kb of the loci. Because only SNPs with trend p-values<0.001 were shown, we could not check the regional signals or validate the interaction, due to the lack of individual-level genotypes.
The NIDDK IBDGC data (phs000130.v1.p1) [3] was accessed from the National Center for Biotechnology Information (NCBI) database of genotypes and phenotypes [41] (dbGaP, http://www.ncbi.nlm.nih.gov/dbgap/) to validate the interactions for CD. The dataset was stratified into two populations, the non-Jewish population stratum, which comprised 513 cases and 515 controls, and the Jewish population stratum, which comprised 300 cases and 432 controls. The genotyping platform was the Illumina HumanHap300 Genotyping BeadChip. The SNPs in the association result file (pha002847.1.IBD.analysis.tar.gz) were selected by removing SNPs with call rates <0.9 in cases or controls and SNPs with HWE p-values<0.001 in the controls. Thus, a total of 305,345 SNPs was used as the validation SNP set.
Two subsequent strategies were used for this validation analysis: the proxy replication strategy and the local validation strategy. First, proxy replication was implemented; because a different genotyping platform was used for the IBDGC data compared to the WTCCC data. SNPs in LD with the original SNPs were selected for proxy replication. The measurement of LD was based on the r2 values from the CEU population data (Phase III release #2) of the International HapMap Project [26] (HapMap, http://www.hapmap.org/). The MaCH imputation method [42] was then used to impute ungenotyped SNPs for validations between the WTCCC and the IBDGC datasets. Interactions were considered valid when they could be replicated by proxy SNPs. The reference haplotypes for MaCH were obtained from the HapMap CEU population (Phase III release #2).
Second, “local validation” was used when the locus-based replication (e.g., proxy replication) failed. In this local validation method, all SNPs in the validation dataset within 50 kb of the original loci were tested for allelic effects (by the trend test) or pure interaction effects (by the 4 d.f. LRT test). This method was based on the notion that confounding factors might affect the consistency of the interactions between the original data and the validation data. The significance was evaluated under the null hypothesis that none of the SNPs (interactions) in the 100 kb region (pair of 100 kb regions) was associated with the disease. In addition, the p-values of the SNPs (interactions) should form a uniform distribution for SNPs that were independent; moreover, the number of p-values lower than a certain threshold (we used 0.05) from the total number of p-values should form a binomial distribution. Therefore, a single-tailed binomial test could be performed to determine the significance in the numbers of SNPs with p-values lower than the threshold. However, SNPs were not independent, due to the LD. Thus, to obtain the empirical significance for the tests, we randomly sampled 1,000 pairs of 100 kb regions from this validation dataset and calculated the empirical distribution of the p-values for correction (Table S5). A significant local validation was interpreted to reject the null hypothesis, that none of the SNPs or interactions in a certain region was associated with the disease. However, strictly speaking, this cannot be interpreted as a successful replication of the original association or interaction; thus, we used the term “local validation” instead of “local replication” to avoid confusion.
The GENEVA Diabetes Study data (phs000091.v1.p1) was accessed from the dbGaP to validate the T2D association results. The participants in that study were all female. The genotyping platform was the Affymetrix Genome-Wide Human SNP Array 6.0. Caucasian individuals without missing data on the disease status were included, SNPs were quality-controlled, and 496,606 genotypes were set to “missing”, as initially recommended. After selection, a total of 1,543 cases and 1,770 controls, with 707,301 SNPs were analyzed. According to the genotyping platforms, the SNPs in the GENEVA Diabetes Study dataset contained most of the SNPs in the WTCCC data. Therefore, we could select the exact SNP combinations in the validation dataset for exact replication, without the need for the proxy replication described above. Upon failure of the exact replication, the local validation method was used with this dataset.
|
10.1371/journal.pcbi.1000257 | Predicting Cellular Growth from Gene Expression Signatures | Maintaining balanced growth in a changing environment is a fundamental
systems-level challenge for cellular physiology, particularly in microorganisms.
While the complete set of regulatory and functional pathways supporting growth
and cellular proliferation are not yet known, portions of them are well
understood. In particular, cellular proliferation is governed by mechanisms that
are highly conserved from unicellular to multicellular organisms, and the
disruption of these processes in metazoans is a major factor in the development
of cancer. In this paper, we develop statistical methodology to identify
quantitative aspects of the regulatory mechanisms underlying cellular
proliferation in Saccharomyces cerevisiae. We find that the
expression levels of a small set of genes can be exploited to predict the
instantaneous growth rate of any cellular culture with high accuracy. The
predictions obtained in this fashion are robust to changing biological
conditions, experimental methods, and technological platforms. The proposed
model is also effective in predicting growth rates for the related yeast
Saccharomyces bayanus and the highly diverged yeast
Schizosaccharomyces pombe, suggesting that the underlying
regulatory signature is conserved across a wide range of unicellular evolution.
We investigate the biological significance of the gene expression signature that
the predictions are based upon from multiple perspectives: by perturbing the
regulatory network through the Ras/PKA pathway, observing strong upregulation of
growth rate even in the absence of appropriate nutrients, and discovering
putative transcription factor binding sites, observing enrichment in
growth-correlated genes. More broadly, the proposed methodology enables
biological insights about growth at an instantaneous time scale, inaccessible by
direct experimental methods. Data and tools enabling others to apply our methods
are available at http://function.princeton.edu/growthrate.
| A major challenge for living organisms is the regulation of cellular growth in a
fluctuating environment. Sudden changes in nutrient availability or the presence
of stress factors typically require rapid adjustments of cellular growth. The
misregulation of growth control in higher organisms is a major factor in the
development of cancer. A statistical characterization of cellular growth based
on gene expression levels provides a quantitative perspective to understand the
regulatory network that controls growth. We develop a model of cellular growth
in the yeast Saccharomyces cerevisiae, grounded in the
expression levels of a small set of genes. The model is able to predict the
growth rate of new cellular cultures from expression data and is robust to
changing biological conditions, experimental methods, and technological
platforms. The predictions are informative about changes in growth at very short
time scales, which direct experimental methods cannot generally access. The
model also predicts growth rates in Saccharomyces bayanus and
in Schizosaccharomyces pombe, a yeast diverged by approximately
a billion years of evolution. Our findings suggest that the model describes
fundamental characteristics of the unicellular eukaryotic growth regulatory
program. A case study explores the role of nutrient sensing in the yeast growth
regulatory system.
| Proper regulation of growth rate is a key systems-level challenge for all cells,
particularly microorganisms facing a fast-changing and often hostile environment.
Cell growth, defined as an increase in cellular biomass due to biosynthetic
processes, is one of the primary functions that must be coordinated with the
environment in order for cells to maintain viability and reproduce. The
determination of how cells integrate information from the external environment with
information from their internal state to mount an appropriate
response—growing in the presence of nutrients, arresting growth when
stressed, and resuming afterwards— is of central importance to our
understanding of basic biology. From a genomic perspective, growth also raises the
issue of disentangling correlated systems-level behaviors. When the expression
levels of thousands of genes change due to a growth-related stimulus, which
underlying regulatory parameters are responsible?
In this paper, we identify quantitative aspects of the transcriptional regulatory
mechanisms underlying cellular growth in Saccharomyces cerevisiae,
and we develop a model to predict instantaneous growth rates of cellular cultures
based on gene expression data. The model enables the estimation of growth rates
under any conditions for which expression data is available, even on a very short
time scale, where standard experimental techniques cannot measure cellular growth
directly [1]. For example, a culture undergoing continuous growth
in a chemostat [2] can be perturbed from steady state by means of a brief
heat pulse, but the departure from and the return to steady state growth is too
brief to capture with optical density measurements. Our model allows such a decrease
(and subsequent resumption) of growth rate to be quantified under a variety of
conditions: batch or chemostat cultures, different microarray platforms, and under
any environmental stimulus for which gene expression can be assayed. Surprisingly,
this model also successfully predicts growth rates from Saccharomyces
bayanus and Schizosaccharomyces pombe expression data,
the latter of which is evolutionarily diverged from S. cerevisiae
by an estimated billion years [3].
Our findings suggest that the proposed statistical model of cellular growth provides
a broadly applicable biological characterization of the transcriptional regulatory
network underlying growth rate control. We have previously observed that the
expression of ∼25% of the yeast genome responds to changes in
growth rate [4]. The response is functionally cohesive, with genes
up-regulated with increasing growth enriched for translational and ribosomal
functions, and with down-regulated genes enriched for oxidative metabolism and the
peroxisome. This functional portrait provides a rich environment in which to study
transcriptional regulation of growth; for example, production of new proteins at the
ribosome is vital to cellular proliferation, and yeast devotes some
∼60% of its transcriptional throughput to ribosomal RNA [5].
Similarly, growth rate regulation is highly interconnected with a variety of other
cellular processes (e.g. the environmental stress response [6], metabolic cycling [7], and
the cell cycle [8]), and we discuss potential causative regulatory
signals from the Ras/PKA pathway [9] and growth-related transcription factors.
Our recent analysis of gene expression measurements from a collection of S.
cerevisiae chemostat cultures across several nutrient limitations and
growth regimes [4] offered intriguing evidence for a notion of
instantaneous growth rate. In this paper, we develop a model to characterize such a
notion quantitatively, in a statistically principled fashion. We further assess the
robustness of the proposed characterization by presenting new computational evidence
on six additional published data sets and on four newly collected data sets. More in
detail, we demonstrate that the model can accurately predict relative growth rates
under a variety of conditions and is robust to the conditions of the originating
culture, the technological platform used to assay gene expression, and evolutionary
conservation to other organisms (S. bayanus and S.
pombe). The model allows us to predict growth rates for published
genome-wide collections of expression data (e.g. the stress response [6] or gene
deletions [10]) and for four new data collections we have generated
for this work (Tables S1, S2, S3, S4), providing biological insight into the growth
rate response at very short time scales—minutes, rather than the hours
necessary to experimentally assay doubling times. This biological validation of the
predictions is accompanied by an out-of-sample validation and outlier analysis to
assess the statistical accuracy of the model. We have made an implementation of this
model available to the public at http://function.princeton.edu/growthrate.
Additional analyses offer biological insights that support and further substantiate
the empirically observed robustness of the predictions based on the newly
characterized growth-rate genes. Our insights rely, in part, on the quantitative
identification of binding motifs of known (and uncharacterized) transcription
factors associated with the genes responding to growth. Moreover, our model enables
a quantitative characterization of growth profiles underlying puzzling experimental
evidence that provides a first convincing explanation of observed cell death in
response to a perturbation in the Ras/cAMP/PKA pathway. More in detail, we apply our
model to study two important aspects of cell growth regulation: nutrient sensing and
the cell cycle. Artificial activation of the Ras/cAMP/PKA pathway has been
previously observed to recapitulate approximately 85% of the expression
response associated with increased growth in the presence of glucose [11]; here,
we show that the cell's regulatory state during this activation is
indicative of an up-regulated growth response, even in the absence of appropriate
nutrient availability. This conflict between internal regulatory state and the
external environment leads to rapid cell death. In contrast, analysis of growth rate
regulation during metabolic cycling [12] and synchronous cell cycles [8],[13] indicates that growth
rate regulation is not specific to cell cycle phases, but it is strongly limited to
the oxidative phase of the metabolic cycle. These observations, coupled with an
analysis of putative transcription factors mediating the growth response, establish
a substantial foundation on which to base further experimental work on the
systems-level control of cellular growth rate.
Cellular growth is typically quantified in one of two experimental environments:
batch culture or the chemostat. In a batch culture, cells are provided with a
saturating amount of nutrient [1]. Growth is quantified by measuring the optical
density (OD) of the culture over time, X. Figure 1 illustrates three typical phases of
an OD growth curve: a slow initial phase (lag), a fast exponential growth phase
(exp), and a slow saturation phase (stationary). Solving the appropriate
differential equation leads to an exponential model of cellular growth,
X = e μ·t. In
practice, the OD of a culture is sampled at discrete points over time, and the
growth rate parameter μ (in units of inverse hours
h−1) is estimated from an exponential fit to the OD
measurements.
In the chemostat, a specific growth rate is maintained by limiting the
concentration of a controlling nutrient provided to the cells [14].
Figure 2 illustrates the
principle behind the chemostat. A limited concentration (S0 in the
tank) of the controlling growth factor is provided in media flowing continuously
into a growth tube of limited capacity. Changes in density of the culture, X,
and in concentration of the controlling nutrient (S), in the growth tube, are
driven by Michaelis-Menten dynamics [15]. In this regime, the
growth rate is a function of the concentration of the controlling nutrient,
μ = μ(S). In particular,
dX/dt = [μ(S)−D]
X; at steady state, the density of the culture no longer changes,
dX/dt = 0, and the concentration of the
controlling growth factor also stabilizes,
dS/dt = 0. The growth rate then equals the flow
rate set by the experimenter,
μ(S*) = D.
In a batch culture, the growth rate is generally not controlled; it is determined
by a complex interaction of environmental and genotypic states, and it is
maximal during the exponential phase of growth (μmax). Under
these conditions, the growth rate of the culture (the first derivative of the
curve in Figure 1) changes
with time. In a chemostat, the growth rate is controlled by setting the nutrient
flow rate D below an organism's maximum possible growth rate
μmax as estimated from batch culture. In either
experimental environment, the growth rate μ is directly related to the
doubling time, Td = ln(2)/μ.
Our model is built on a collection of gene expression data from chemostats at
known growth rates, and it allows us to quantify a notion of instantaneous
growth rate in chemostat and batch cultures, even in cultures in which the
growth rate is changing rapidly over time.
We fit a linear model to a collection of expression data drawn from S.
cerevisiae chemostat cultures over several growth rates and nutrient
limitations. Estimates of the parameters characterize each gene's response
to changes in growth rate, and provide insight into the transcription factors and
regulatory network responsible for yeast growth homeostasis. By applying this model
to new expression data sets, we are able to predict instantaneous growth rates for
any yeast culture. The model is robust to the biological and technical conditions of
the originating gene expression data and enables the prediction of growth rates at
instantaneous time scales inaccessible to standard experimental methods (e.g.
optical density). We have also successfully applied the model to the related
organisms S. bayanus and S. pombe. Data and tools
relating to this model are made available at http://function.princeton.edu/growthrate.
Our model is based on a collection of gene expression measurements from steady
state (chemostat) cultures limited across several nutrients and growth regimes.
Briefly, 36 CEN.PK derived S. cerevisiae chemostat cultures
were grown under six nutrient limitations: Glucose (G), Nitrogen (N), Phosphate
(P), Sulfur (S), Leucine (L), and Uracil (U). Six growth rates were used for
each nutrient, ranging by steps of 0.05 h−1 from 0.05
h−1 to 0.3 h−1. Agilent Yeast V2
microarrays were used to measure gene expression in the resulting 36 chemostats;
for details, see [4]. This experimental design provides the
opportunity to discover gene expression patterns correlated with growth rate,
independently of nutrient-specific responses.
Figure 3 highlights the
sources of variability in the gene expression profiles that the experimental
design aims at capturing. The resulting data contain a number of characteristic
gene expression patterns, including genes with strong growth-specific
transcriptional regulation and negligible nutrient-specific response (Figure 3A). Other genes
include a growth-specific expression component but are also strongly up- or
down-regulated under specific nutrient limitations (Figure 3B). Finally, Figure 3C displays expression profiles that
show unsystematic or negligible responses under these conditions. The linear
model described below summarizes the variability in the expression profiles of
individual genes specifically due to changes in growth rate, which leads to a
characterization of growth-specific calibration genes such as
those shown in Figure 3A.
This growth-specific signature enables predictions of the instantaneous growth
rate of any cellular culture based on the relative expression values these
growth-specific genes.
Table 1 summarizes the
collections of expression data analyzed in this study. Six collections were
previously published by others, one was published in our previous work [4], and
four are new to this study: 1. chemostats limited for different nitrogen
sources, 2. heat pulses inducing a temporary departure from steady state growth,
3. artificial activation of the Ras/PKA pathway, and 4. S.
bayanus diauxic shift and heat shock time courses. All gene expression
collections were pre-processed as in [16]. The gene
expression values for all growth-specific genes are provided in Tables S1,
S2,
S3,
S4,
respectively.
We sought to identify a small set of genes providing a quantitative summary of
cellular growth rate regulation. Genome-wide expression measurements underlying
the 36 chemostat cultures provided us with the opportunity to determine which
genes were responding linearly to changes in growth rate, and not to differences
in nutrient limitation. To identify such gens in a statistically principled
fashion, we performed four steps, beginning by using maximum likelihood to fit a
linear model of each gene g's expression under all
training conditions (Y
g
) based on the conditions' known growth rates (X
c
):(1)
This step yields two learned parameters per gene, a baseline expression level
αg and a growth rate response
βg. The model is fit to minimize the
residual error ε
g
, which can capture either non-growth-related biological variability or
technical noise. We fit this model for the yeast genome using the expression
levels from our 36 chemostat conditions, recording each gene's
αg and
βg parameters and its goodness of fit (total
explained variability) R2
g
.
We next used the bootstrap (i.e. a randomized re-sampling technique) to assess
the expected background distributions of these parameters in the absence of a
growth-related biological signal (i.e. the null distributions). We constructed
100,000 randomized expression vectors of length 36 by sampling each component
(with replacement) from the collection of gene expression values in the
corresponding condition, i.e., the same combination of growth rate and nutrient
limitation. For example, the first value randomly chosen for such a vector could
be drawn from any gene or nutrient limitation in our chemostat data at a flow
rate of 0.05 h−1, the second from any flow rate of 0.1
h−1, and so forth. Note that by re-sampling the
expression values of putative genes column-by-column, we do not wash away the
average transcriptional response that is expected to be associated with
nutrient-growth rate pairs. In this sense, the null distribution we derive
carries information about how genes respond to growth across nutrient
limitations, on average. As a consequence, the statistical significance of the
differential response we compute is biologically justified. In other words, this
sampling scheme maintains average nutrient specific and growth rate specific
information, and leads to an estimate of the null distribution in the absence of
gene-specific growth related and nutrient related gene expression. This process
yields null distributions for parameters αg,
βg, and the goodness of fit R2
g
.
Third, from these null distributions, we assign false discovery rate corrected
p-values [17] to each gene's
αg,
βg, and R2
g
values. Finally, a gene was deemed to have a significant expression
response to changes in growth rate if it fit this model well (R2
g
p<0.05) and was up- or down-regulated significantly with growth
(βg p<0.05); this information
is available in [4]. We further characterized a specific set of
growth-specific calibration genes responding only and
significantly to changes in growth rate (βg
p<10−5 and R2
g
p<10−5) that we used to infer
instantaneous growth rates in new expression data (Table S5
and Dataset
S1).
The set of growth-specific genes identified with the four-step procedure above
represents a quantitative signature of a cellular culture's
transcriptional regulation of growth rate, i.e. the speed at which cells are
proliferating. By examining these genes' expression levels in a new
collection, we can predict the instantaneous growth rate of the cellular culture
the expression measurements correspond to. This notion of instantaneous growth
rate is comparable to the derivative of an optical density growth curve, but it
can be inferred robustly by our model on any time scale, e.g. minutes, from
expression data, without the need to measure one or more full doubling times of
a culture.
Given expression data for a new experimental condition, we use an iterative
maximum likelihood approach to infer its growth rate using the parameters
captured by our linear model. Formally, consider a vector of expression
measurements for n growth-specifc genes,
Z1:n. As described above, the
expression of these growth-specific genes varies primarily in response to
changes in a condition's growth rate, which we model as the mean
μ of a Gaussian with variance
σ2. Using our previously calculated
maximum likelihood estimates of the calibration gene parameters
α1:n and
β1:n, the expected value of
a gene's expression is thus:(2)
Here, δ is a condition-specific parameter that captures
the condition's baseline gene expression, i.e. an average offset
between a new experimental condition and our training expression data. In
dual-channel data, this parameter may capture differences between a new
condition's reference channel and our training data's
reference channel; for a single-channel array, δ may
capture the absolute difference between the platform baseline and our training
data. In any event, the expected variability of a new measurement is:(3)
The likelihood of the expression measurements
Z1:n is thus a product of Gaussians:(4)
From this, we derive the maximum likelihood estimate of the condition's
growth rate μML:(5)
Similarly, the maximum likelihood estimate of the condition's baseline
δML is given by:(6)
Note that the estimate of δML depends on the
estimate of μML, and vice versa. To
calculate these estimates, we initialize
μML(0) assuming
δML(0) = 0
and iterate subsequent computations of
μML(t+1)
and
δML(t+1)
to convergence. In practice, individual growth-specific genes with residuals
outside the inner fences of all growth-specific gene residuals (more than 1.5
inter-quartile ranges from the lower or upper quartiles, [18]) are noted as
outliers and do not participate in that condition's growth rate
inference procedure. This allows outlier genes responding to non-growth related
stimuli (which are, in general, infrequent, e.g. six in one of our most variable
conditions as discussed below) to be noted for further investigation, while also
decreasing the cross-validated error of predicted growth rates.
In principle, this model of growth rate can be extended to study and predict
instantaneous growth in any organism for which appropriate homology to
growth-specific genes exists. To analyze growth rates in expression data from
S. bayanus and S. pombe, the S.
cerevisiae calibration genes were mapped to known orthologs. This
mapping was performed using the unambiguous pairings from [19] for S.
bayanus and the curated orthologous groups from [20]
for S. pombe. This resulted in 51 growth-specific genes for
S. bayanus and 74 for S. pombe, the
increase being due to one-to-many mappings; see Table
S5.
The parameter estimates driving our predictions and tools allowing users to
predict growth rates in new data sets are available at http://function.princeton.edu/growthrate. Specifically, users
can upload S. cerevisiae expression data (single- or
dual-channel in standard PCL format) to receive estimates of relative growth
rate for each condition. If a reference with known growth rate is provided,
absolute rate estimates will be generated. This growth rate prediction tool has
been implemented in R and is also available for offline use, allowing further
customization (such as application to additional organisms).
We apply our linear model of growth rate regulation to predict instantaneous growth
rates for a variety of expression data. This includes new chemostat cultures used to
assess prediction quality, publicly available stress response and gene deletion
microarrays from batch cultures, growth differences between metabolic cycling and
the cell cycle, several different microarray platforms, and an out-of-sample
validation to quantify model accuracy. We also observe good predictive performance
for growth rates in S. bayanus and S. pombe data
sets, the latter despite up to a billion years of evolutionary divergence from our
S. cerevisiae training data. This suggests that the
growth-related transcriptional regulation captured by our model is a key feature of
unicellular homeostasis, a feature we explore by examining nutrient sensing inputs
through the Ras/PKA pathway and potential growth rate transcription factors and
binding sites.
Our model of the growth rate transcriptional response can be used to predict
relative instantaneous growth rates from any S. cerevisiae gene
expression data. For example, Figure 4A shows our predicted growth rates for a gene expression
time course sampled from a steady state culture exposed to a brief (<30
s) heat pulse (Table S1). The predictions clearly show a departure from steady
state within five minutes of the heat pulse, followed by recovery within 15
minutes. Similar predictions over a range of chemostat flow rates (Figure S1)
reveal that this cellular behavior is consistent, although there is some
variation in the degree of growth cessation during stress, in agreement with
tolerance and sensitization models of the yeast stress response [21].
Notably, standard experimental assays for growth rate (e.g. optical density)
would be incapable of monitoring such a response, while our model is able to
observe these growth changes on an instantaneous time scale.
A similar application of our model to predict relative growth rates for the
stress response conditions of [6] is presented in Figure 4B (see Figure S2
for complete results). These data represent yeast batch cultures assayed using a
variety of different reference mRNA samples on a custom spotted microarray
platform, none of which differences from our training data impair the growth
rate estimation process. While there are no direct measurements of growth rate
in these non-steady-state conditions, our predictions are consistent with known
yeast biology and agree with expected growth behavior. Most shock time courses,
including all heat shocks, peroxide, diamide, and hyper-osmotic stress, provoke
an initial sharp decrease in growth rate followed by a return to initial or
near-initial rate; shorter shocks, such as DTT, menadione, and peroxide
responses, capture only the rate decrease. Batch growth proceeds at a fairly
constant rate until nutrients become depleted, at which point the rate decreases
sharply; this pattern is also seen in intentional nitrogen depletion. Growth
rates across varying temperatures peak as expected at 25 C [1], falling off at
lower and higher temperatures. Finally, response to varying carbon sources is
also as expected [22], with ethanol inducing the slowest growth and
fructose, sucrose, and glucose allowing the most rapid. Our model's
inference of growth rate from gene expression data alone allows both post hoc
growth analysis (e.g. years after the original experiment) and an estimation of
growth rates for cultures where direct growth measurements would be unfeasible,
difficult, or time consuming.
When applied to expression data from yeast mutant strains, in which one or more
genes have been deleted, predicted growth rates can be used to quantify single
mutant fitness. We used our model to analyze the knockout collection assayed in
[10]; predictions on the complete data set are
available in Table S6. Direct fitness measurements for 199 of the ∼300
mutants assayed via microarrays is available as supporting information [10].
Our predictions for these 199 growth rates correlate very strongly with the
direct fitness measurements (ρ = 0.473,
p<10−11) and are derived solely from expression
data. In contrast, methods for experimentally estimating single mutant fitness
from high-throughput growth curves showed substantially less agreement
(ρ = 0.321,
p<10−6
[23];
ρ = 0.108, p>0.2 [24])
with the original publication's direct fitness measurements. These
results represent a compelling argument as to the relevance of our growth rate
model for fitness estimation.
With a small amount of additional information (i.e., a scalar) the relative
growth rates inferred by our model can be made absolute, in units of chemostat
flow rate (hr−1). Our model's predicted rates for
a collection of arrays are relative estimates, to one another. This is due to
the unknown quantitative effects of the reference mRNA in our dual-channel
training data; it is impossible to know a priori the relationship between this
reference channel and the relative (for dual-channel) or absolute (for
single-channel) expression levels in new microarray data. However, if the
absolute growth rate is known for some array in a given collection, our model
can make absolute rate predictions for other two-color arrays in the collection,
given that they all share the same reference channel.
Figure 4C shows actual growth
rates (dotted, black lines) for a collection of chemostats at various flow rates
limited on one of several different nitrogen sources (Table S2)
along with estimates of the relative instantaneous growth rates (red, dashed
lines) and of the absolute instantaneous growth rates (solid, blue lines).
Absolute growth rates are estimated by recording the growth rate in the Proline
limited chemostat at μ = 0.35
hr−1, and shifting all the estimates accordingly, since
the dual-channel microarrays in this study all share the same transcriptional
readout in the reference channel. We sought to evaluate the goodness of the
predictions in Figure 4C by
computing the statistical significance of their correlation with the actual
growth rates. To this end, we computed the correlation between the true growth
rates and the predicted instantaneous growth rates. The correlation is the same
for both absolute and relative predicted rates, as they differ by a constant,
and equals ρ = 0.956
(p-value≈0). This computation provides statistical support to the
goodness of the predictions produced with the proposed model. More in general,
on normalized dual-channel microarrays, the doubling of any gene's mRNA
level in these conditions results in the same increase in its expression
readout. Thus one unit of predicted relative rates to corresponds to one unit of
absolute chemostat flow rate. However, since the reference channel differs from
that of the arrays used to train the model, all rate predictions are typically
off by a corresponding constant factor. By normalizing to any one of the
N arrays' known growth rates, this shift can be
automatically corrected for the N-1 other arrays, employing the
same reference channel.
We assessed the quality of our growth rate predictions using 1,000 out-of-sample
experiments, according to a hybrid bootstrap/cross-validation setup, using the
data from [4]. Results are shown in Figure 5A. In each experiment, we randomly
withheld 12 of the 36 conditions for testing, fit our linear model on the
remaining 24, derived bootstrapped null distributions using only these data, and
determined growth-specific gene sets to use for growth rate inference on the
held-out conditions. This experimental setup leads to absolute growth rate
predictions directly, as all the dual-channel microarrays share the same
transcriptional readout in the reference channel. This out-of-sample validation
allowed us to assess the accuracy and variability of our predictions on
conditions with known growth rates not included in the model building procedure.
In addition to the performance indicated by Figure 5A, the out-of-sample experiments
demonstrated robustness of p-value cutoffs and number of growth-specific genes;
these ranged in number from ∼50 to ∼110 across the randomized
validations (of a total ∼5,500 possible genes), and changes of this
magnitude in the final calibration gene set had little impact on predicted
growth rates. We further quantified a notion of reliability for each of the 72
growth-specific genes. Specifically, we computed the percentage,
P, of bootstrap experiments in which each individual gene was
selected as a member of the growth-specific gene set. The percentages provide an
expectation about whether each individual gene should be considered reliable in
a new study. We found that 69 genes were selected in more than half of the
experiments, P>0.5. Full results are
reported in Table 2.
In the process of estimating growth rates and determining this confidence score,
growth-specific genes with outlying expression values are also detected. While
most conditions induce few outlying growth-specific genes, when they occur, they
are not indicative of the quality of growth rate predictions.
We have found that neither the number of outliers nor their variability
correlates with prediction error (data not shown), but they call out genes that
may be responding to non-growth stimuli under specific biological conditions.
Excluding outliers from the growth rate estimation process improves the accuracy
of the predictions, and these outliers can in turn be biologically informative:
an outlying growth-specific gene is likely responding specifically to a stimulus
other than change in growth rate. For example, some of the only outliers in the
mild heat shock time course from [6] occur towards the end of a shift from 29 C to
33 C (Figure 5B). These
include HSP26 and HSP78, both known heat shock
chaperones [25],[26]. Three genes of
unknown function (YLR327C, MOH1 and the
neighboring dubious ORF YBL048W, and TMA10)
are also outliers in this condition, which is evidence that these genes may have
specific expression responses (and thus biological functions) during heat shock.
HSP26 and YLR327C are frequent outliers in
stress-related conditions, perhaps suggesting a more general stress response
function.
While our growth rate model is based on a transcriptional growth signature in
S. cerevisiae, the model can be applied to any organism
with sufficiently orthologous transcriptional activity. This is likely to be the
case within the sensu stricto yeasts, separated by ∼25
million years of evolution [27]. By finding the ∼50 S.
bayanus genes orthologous to our ∼70 S.
cerevisiae growth-specific calibration genes [19], we can apply our
model directly to S. bayanus expression data (Table S4).
Figure 6 demonstrates
such a result for two S. bayanus time courses assaying the
diauxic shift and a response to heat shock. These results have comparable
profile to those from S. cerevisiae and are similarly
biologically compelling. For example, the diauxic shift in S.
bayanus results in a very similar growth pattern to the known response
in S. cerevisiae, with a near-cessation of growth during the
shift and subsequent rebound. Conversely, S. bayanus is less
resistant to high temperatures than S. cerevisiae
[28], and our growth rate inferences show a
corresponding failure in its ability to grow following severe heat shock.
We have also extended our model to a significantly further diverged yeast,
specifically the yeast Schizosaccharomyces pombe, separated
from S. cerevisiae by an estimated one billion years of
evolution [3]. A mapping of our growth-specific calibration
genes to S. pombe using information from [20] results in
∼75 genes due to one-to-many correspondences, but these provide
sufficient calibration information to make high quality predictions (Figure 6C). Calibration gene
outliers and expression cohesiveness are not substantially changed relative to
S. cerevisiae and S. bayanus, and the
inferred relative rates reflect various biological expectations. All cultures
(data from [29]) show an initial increase from low growth rates
due to stalled growth during synchronization. An expected decrease in growth
rate is predicted during increased exposure to hydroxyurea (HU), and a
rad3Δ deletion (S. cerevisiae
ortholog MEC1) incurs a mild overall growth impairment as well
as exacerbating HU sensitivity. While MEC1 is essential in
S. cerevisiae, this sensitivity has previously been noted
for deletions sod1Δ and lys7Δ,
both members of the MEC1 pathway [30], which is necessary
for the cell cycle checkpoint function.
The extent to which transcriptional regulation is conserved between S.
cerevisiae and S. pombe, which allows us to
successfully apply the model despite the evolutionary distance that separates
these species, is reflective of cellular growth's central role,
particularly in unicellular organisms. While this model would become less
meaningful in metazoans, where the growth of individual cells is subjugated to
the growth and differentiation of the organism as a whole, certain
transcriptional growth behavior is of necessity conserved in single celled
organisms [31]. This is particularly true of the ribosome, one
of the main contributors to our model's predictive power; rRNA
regulation is purely transcriptional, and ribosomal proteins must be expressed
stoichiometrically. Since any cellular growth requires translation, observation
of ribosomal transcription is a strong indicator of unicellular growth [5].
This is one aspect of the transcriptional growth response made quantitative by
our model.
To further investigate the biological basis of growth rate correlated gene
expression, we used our model to predict relative growth rates for two
interesting cases: the yeast metabolic cycle [12] and the mitotic cell
division cycle [8],[13]. The expression
data published by Tu et al. was obtained for cells grown at high density in a
glucose-limited chemostat. Under this regime, cells within the culture become
metabolically synchronized and undergo periodic consumption of oxygen (defined
as the oxidative phase of the metabolic cycle) followed by periods of
undetectable oxygen consumption (termed the reductive building and reductive
charging phases). The cell cycle data sets by Spellman et al and Pramila et al
were obtained from experiments in which cells were uniformly arrested in the
cell division cycle using a variety of methods and then released to undergo
synchronous cell division cycles.
Growth rate prediction applied to the yeast metabolic cycle data revealed a
striking periodicity (Figure
7A). The cyclical pattern of growth rate variation occurs completely in
concert with the metabolic cycle as defined by Tu et al. Specifically, the
culture's growth rate is predicted to be at minima during the reductive
phase of the metabolic cycle, when oxygen consumption is at a minimum, and reach
maxima during the peak of the oxidative phases when oxygen consumption is
maximal. In contrast, growth rate prediction for the cell cycle (Figure 7B and 7C) show
virtually no variation in predicted growth rate during the different stages of
cell division.
These data support and extend our previous assertions [4] that the there is a
close connection between the metabolic cycle identified in [7] and [12] and the
association we identify between growth rate and gene expression levels. This
result is consistent with two possible explanations. The first is that there is
variation in the growth rate of cells throughout the metabolic cycle. [12] and
[32]
have shown that under their specific experimental conditions, DNA replication
and cell division is restricted to the reductive phases of the metabolic cycle.
It is conceivable that growth per se (i.e. the accumulation of biomass) is
paused during the reductive phases of the metabolic cycle so that the cell can
replicate and segregate DNA and complete the complex processes of cell division;
growth may then be restricted to the oxidative phase of the metabolic cycle.
Alternatively, it is possible that as any heterogeneous culture grows faster, a
greater fraction of cells are in the oxidative phase at any point in time. Thus,
the growth rate gene expression signature we detect might reflect the fraction
of cells in the oxidative and reductive phases of the metabolic cycle in a
metabolically unsynchronized population.
The absence of growth rate differences during the cell division cycle (Figure 7B and 7C) supports our
previous claim [4] that the growth rate expression signature is
unrelated to the cell cycle. Moreover, since the published cell cycle
experiments were performed in rich media using a fermentable carbon source, the
results suggest that rapidly growing cells (which are almost exclusively
fermenting) do not partition metabolic activity into discrete phases, as their
energetic requirements are met in a continuously reductive metabolic state. It
is only when slowed growth is imposed upon the cell, due to stress, nutrient
limitation, or other suboptimal environments, that the metabolic cycle is
required.
We sought to distinguish whether nutrient availability directly determines the
transcriptional state related to growth rate or whether nutrient availability is
integrated through an internal signaling pathway that controls the appropriate
transcriptional state. To address this issue, we examined the regulatory circuit
responsible for transcriptional changes in response to glucose availability in
yeast. Glucose addition to cells growing on glycerol elicits a rapid and massive
change in the pattern of gene expression, with more than half of all genes
changing at least twofold in expression. Previous work has shown that the
Ras/cAMP/PKA pathway is the major source for eliciting this transcriptional
change in response to glucose addition [9],[11]. Activation of the
Ras/PKA pathway in the absence of environmental signals, through induction of an
activated allele of RAS2
(RAS2G19V), recapitulates in magnitude and direction
more than 85% of the changes observed by glucose addition, and
inhibition of PKA (concurrent with addition of glucose) blocks most of the
glucose induced transcriptional changes ([11], Table S3).
This mutation thereby represents a useful model connecting S.
cerevisiae's glucose sensory signaling to its transcriptional
regulation of growth rate.
We used a gal1Δ strain carrying the activated allele
RAS2G19V under control of the galactose
inducible GAL10 promoter. Addition of galactose activates the
Ras/PKA pathway, but since galactose cannot be metabolized by this strain, the
metabolic state of the cell remains unaltered [9]. When grown on
glycerol our model predicts a relative growth rate of ∼0.2 for this
strain (Figure 8A), which
changes to ∼0.6 within twenty minutes following glucose addition,
consistent with the change in doubling time from 5.8 hr to 2.6 hr. When we
performed the same experiment on glycerol media and induced the
RAS2G19V by means of galactose addition, we
detected a transcriptional response within sixty minutes. The predicted growth
rate of the RAS2G19V mutant strain was comparable to
the addition of glucose despite the fact that galactose addition does not yield
an increase in growth, as measured by optical density, since the cells are
unable to metabolize galactose. In fact, while the model's
summarization of gene expression state indicates that the culture is attempting
to increase growth, induction of the RAS2G19V allele
results in an immediate decrease in growth rate and complete cessation of growth
within four hours [33]. These results are consistent with the cell
setting its growth-specific transcription program on the basis of its
perception of nutrients present in the environment, rather
than on the direct availability of energy or metabolites produced from such
nutrients. The mechanism by which the cell integrates this external state in
order to set the appropriate growth rate expression state must be mediated, at
least in part, through the Ras/cAMP/PKA pathway.
To investigate the regulatory basis of growth-associated gene expression, we
identified motifs enriched in the 3′ and 5′ regions of genes
with strong growth rate responses (Figure 8B). We assigned genes to clusters based on their growth rate
response parameter (βg) using k-means
clustering with k = 10. Using the FIRE motif
identification program [34], we identified enriched motifs in seven of
the resulting ten clusters. Consistent with the functional enrichments of
negatively growth rate correlated genes [4], we identified known
binding sites associated with the stress responsive transcription factors Msn2p
and Msn4p in genes negatively correlated with growth rate. Conversely, genes
that increase in expression with increased growth rate are enriched for the
Rap1p consensus motif, which is commonly found upstream of genes encoding
protein components of the ribosome.
We also found enrichment of the Ino4p binding site in genes upregulated with
increasing growth rate. Ino4p forms a heterodimer with Ino2p to activate genes
involved in phospholipid, fatty acid, and sterol biogenesis, all of which are
required in greater abundance with increased growth rates. Furthermore, Ino4p
has been proposed to have an inhibitory effect on a number of genes, including
those that encode the heat shock proteins (Hsp12p, Hsp26p) and catalase (Ctt1p)
[35]. We also identified two additional enriched
motifs in the 5′ UTR for which the binding factor is not known,
suggesting that additional activators of growth-related transcriptional programs
remain to be determined.
In addition to 5′ upstream motifs, we identified five enriched
3′UTR motifs, which are potential binding site for proteins that
promote mRNA degradation. Only a small number of mRNA binding consensus
sequences are known in yeast, all of which belong to the Puf family of mRNA
binding proteins [36]. Our analysis identified five enriched motifs
in 3′UTRs. Two of these motifs, found in genes positively correlated
with growth rate, were identified by the FIRE program as being targets of Puf4p.
As an independent test, we compared the distribution of growth rate responses in
the known gene targets of the five Puf proteins with the overall distribution of
growth rate slopes. Targets of both Puf3p (220 genes) and Puf4p (205 genes) are
enriched for genes that are upregulated with increasing growth
(Wilcoxon-Mann-Whitney two sample p-values
9×10−23 and
7.23×10−16, respectively; Figure S3).
The consensus motifs of Puf3p and Puf4p are very similar; investigation of the
PUF4 motif identified by FIRE suggests that the enrichment signal for at least
one of the motifs denoted PUF4 is likely to result from a composite of Puf3p and
Puf4p target genes (Figure
8B).
Overall, this analysis is consistent with tight transcriptional regulation
underlying the cellular growth program; it is likely that mRNAs involved in this
process are also subject to extensive post-transcriptional control.
Interestingly, since our growth-rate prediction method is sensitive to changes
in gene expression levels that occur within minutes of a perturbation, we expect
that post-transcriptional regulation (both mediated decay of and stabilization
of transcripts) is involved in this response. Experimental analyses of the
effects of perturbations within this regulatory network promise to shed further
light on its organization.
We present a statistical model of the gene expression response to changes in growth
rate in S. cerevisiae. Developed on expression levels from a
variety of steady state growth rates and nutrient limitations, the model captures
information regarding each gene's linear response to growth rate. As
detailed in [4], approximately half of the genome shows a significant
transcriptional response to growth rate with strong functional cohesiveness; here,
we extend this model to show its robustness, applicability to new data, and ability
to provide insight into the biological systems driving cellular regulation of growth
rate. New experiments with more complex models (quadratic and hierarchical)
demonstrated that additional model parameters did not provide substantial
performance gains, in terms of growth rate prediction accuracy, particularly
relative to their added complexity (data not shown). Similarly, variations in the
definitions of responding genes or of growth-specific genes did not substantially
alter results. This stability is reflected in the out-of-sample validation results,
which quantify the model's accuracy in predicting relative growth rates
from gene expression data, and in Table 2, which suggest that growth-specific signal is localized to a
small number of genes consistently across experiments.
The model can be applied to new gene expression data to estimate the instantaneous
growth rate of the originating cellular culture. The estimated instantaneous rate
represents a measurement of the transcriptional state of cellular growth rate
control, and it provides insight into the cell's growth rate at arbitrarily
short time scales inaccessible by experimental measurements (e.g. optical density).
Moreover, genes with unexpectedly high or low expression values can be detected
during growth rate inference, and may indicate biological responses to non-growth
stimuli. The predictions based on the proposed model are robust to changing
biological conditions, experimental methods, and technological platforms; they also
extend to the related yeast S. bayanus and the highly diverged
yeast S. pombe, suggesting that the transcriptional control of
growth rate captured by the model are a fundamental aspect of unicellular biology.
Through further analysis, we discovered several putative transcription factor binding
sites enriched in growth-correlated genes, most notably the stress-responsive Msn2p
and Msn4p, the Rap1p ribosomal factor, and Ino4p. Importantly, we have identified a
likely role for post-transcriptional regulation in modulating transcriptional states
related to growth rates. This finding is consistent with our ability to measure
changes in growth rate over very short time scales using gene expression signatures.
The abundance of any messenger RNA is a function of both its rate of production and
of its rate of degradation; however, since transcription is relatively slow, changes
in mRNA abundance can be most rapidly instantiated by altering the stability of the
existent mRNA population. The Puf proteins have known roles in mediating mRNA
degradation [37] and in mediating the association of functionally
related transcripts [36]. It has recently been proposed that modulation of
mRNA stability is an important factor in metabolic regulation [38]. The association of
Puf protein binding domains in the 3′ UTRs of genes with increased
expression at higher growth rates suggests that modulating mRNA stability is also
important in the regulation of the growth response at short time scales.
From a statistical perspective, it is notable that a simple linear model accurately
and robustly captures a specific biological phenomenon. The model represents a
concise, functionally cohesive set of expression profiles regarding the
genome's transcriptional response to growth. This functional interpretation
of the model agrees with known aspects of the growth response, such as the
transcription of ribosomal components, and provides insight as to the mechanistic
roles of internal feedback, environmental sensing, and the stress response as growth
rate varies. By monitoring a small ensemble of genes—with few parameters
per individual gene—the model is easily applicable to new conditions and
organisms and is robust to technical and biological sources of variation. These
features enable our model to serve both as a practical tool for growth rate
estimation (available at http://function.princeton.edu/growthrate) and as a mechanistic
building block in the pursuit of a systems-level understanding of cellular growth
processes.
|
10.1371/journal.pgen.1004491 | Sox11 Is Required to Maintain Proper Levels of Hedgehog Signaling during Vertebrate Ocular Morphogenesis | Ocular coloboma is a sight-threatening malformation caused by failure of the choroid fissure to close during morphogenesis of the eye, and is frequently associated with additional anomalies, including microphthalmia and cataracts. Although Hedgehog signaling is known to play a critical role in choroid fissure closure, genetic regulation of this pathway remains poorly understood. Here, we show that the transcription factor Sox11 is required to maintain specific levels of Hedgehog signaling during ocular development. Sox11-deficient zebrafish embryos displayed delayed and abnormal lens formation, coloboma, and a specific reduction in rod photoreceptors, all of which could be rescued by treatment with the Hedgehog pathway inhibitor cyclopamine. We further demonstrate that the elevated Hedgehog signaling in Sox11-deficient zebrafish was caused by a large increase in shha transcription; indeed, suppressing Shha expression rescued the ocular phenotypes of sox11 morphants. Conversely, over-expression of sox11 induced cyclopia, a phenotype consistent with reduced levels of Sonic hedgehog. We screened DNA samples from 79 patients with microphthalmia, anophthalmia, or coloboma (MAC) and identified two novel heterozygous SOX11 variants in individuals with coloboma. In contrast to wild type human SOX11 mRNA, mRNA containing either variant failed to rescue the lens and coloboma phenotypes of Sox11-deficient zebrafish, and both exhibited significantly reduced transactivation ability in a luciferase reporter assay. Moreover, decreased gene dosage from a segmental deletion encompassing the SOX11 locus resulted in microphthalmia and related ocular phenotypes. Therefore, our study reveals a novel role for Sox11 in controlling Hedgehog signaling, and suggests that SOX11 variants contribute to pediatric eye disorders.
| Ocular coloboma is a condition in which tissue is missing from a portion of the eye due to its abnormal development. Coloboma is also frequently associated with additional anomalies, including microphthalmia (abnormally small eye) and cataracts. Although some of the genes that cause coloboma have been identified, in the majority of cases the underlying genetic cause has not been determined. One pathway that has been implicated in coloboma is the Hedgehog (Hh) signaling pathway. In this study, we have taken advantage of the ability to titrate levels of gene expression in zebrafish to demonstrate for the first time that the transcription factor Sox11 is required to limit levels of Hedgehog (Hh) signaling during ocular development. We show that in the absence of Sox11, levels of the Sonic Hedgehog (Shh) ligand are greatly elevated, which disrupts the proper patterning of the optic stalk and optic vesicle, resulting in coloboma. We also provide evidence that SOX11 dosage changes or mutations contribute to human coloboma, microphthalmia, and rod photoreceptor dysfunction. Thus, our work establishes a novel link between Sox11 and Hh signaling, and suggests that mutations in SOX11 contribute to pediatric eye disorders such as coloboma.
| Ocular coloboma arises when the embryonic choroid fissure in the ventral optic cup fails to close. It can cause significant pediatric visual impairment [1], and is often associated with other ocular abnormalities such as microphthalmia or anophthalmia (collectively referred to as MAC). Coloboma may also be observed in conjunction with dysgenesis of the anterior segment (front portion of the eye) or optic nerve, lenticular defects (such as cataract), or systemic congenital malformation syndromes [2]. In addition to phenotypic heterogeneity, coloboma is genetically heterogeneous, exhibiting differing patterns of inheritance, variable expressivity, and reduced penetrance [2].
Among the signaling pathways that converge to regulate ocular morphogenesis, Hedgehog (Hh) signaling has a critical role and acts reiteratively during eye development [3]. Hh signaling from the midline promotes the segregation of the single eye field into two optic primordia, and is required for the correct proximodistal and dorsoventral patterning of the optic vesicle [3]–[5]. Once the optic cup has formed, intraretinal Hh signaling regulates the differentiation of retinal progenitor cells [3]. Given its central role in eye development, it is unsurprising that mutations in genes encoding Hh pathway ligands (SHH) or targets (PAX2, VAX1) are associated with congenital ocular malformations in humans [6]–[9]. However, these mutations account for only a minority of patients; for the majority of MAC cases, the molecular defect has yet to be identified. Because of their potency, the spatiotemporal levels of Hh ligands must be tightly regulated throughout eye development; yet, very little is known about the factors that restrict their expression during oculogenesis. Such factors would represent excellent candidate genes for human coloboma and associated ocular defects, and potentially could be used to influence Hh signaling.
Here, we focus on the role of the SRY-box transcription factor Sox11 during eye development. Sox11 is a member of the group C family of SOX proteins, which also includes Sox4 and Sox12 [10]. Sox11 is required for a variety of processes, including organogenesis and neurogenesis, craniofacial and skeletal development [10], as well as being implicated in carcinogenesis (including mantle cell lymphoma, medulloblastoma, and glioblastoma) [10], [11]. Expression and functional studies support a role for Sox11 during several stages of eye development. In the mouse, Sox11 is expressed in the optic cup and periocular mesenchyme during early eye development, and in the developing lens and retina at later stages [12], [13]. In the zebrafish retina, we previously found that Sox11 is upregulated in rod progenitor cells during rod photoreceptor regeneration [14]. Sox11−/− mice exhibit ocular abnormalities such as anterior segment dysgenesis, microphthalmia, a persistent lens stalk, delayed lens formation, and coloboma [13]. Finally, some human chromosomal rearrangements resulting in ocular abnormalities have been mapped to the vicinity of the SOX11 locus at chromosome 2p25.2 [15]–[18]. These data together suggested intriguing roles for Sox11 in ocular morphogenesis and rod photoreceptor differentiation, however the underlying mechanisms were undefined.
In this study, we inhibited Sox11 activity in zebrafish embryos, and based on the resultant phenotypes demonstrate that the function of Sox11 in regulating lens development and choroid fissure closure is evolutionarily conserved, and that Sox11 is required for rod photoreceptor differentiation. We demonstrate that elevated Hh signaling causes the ocular phenotypes in Sox11-deficient zebrafish, and that Sox11 is required to repress expression of the Sonic hedgehog gene (shha). Finally, we identify SOX11 variants with reduced transactivation ability in MAC patients, and in parallel demonstrate that decreased SOX11 gene dosage results in congenital ocular abnormalities. In revealing a previously uncharacterized role for Sox11 upstream of Hh signaling, these studies may substantially extend our understanding of additional Sox11-dependent developmental and pathologic processes.
Zebrafish possess two orthologs of mammalian Sox11, which are expressed in overlapping and distinct domains ([19]–[21], this study). Previous studies have shown that both sox11a and sox11b are maternally expressed prior to the midblastula transition, and are expressed in the region of the anterior neural plate that gives rise to the diencephalon at the onset of the segmentation period [20]. Using in situ hybridization with paralog-specific probes, we investigated the expression of sox11a and sox11b both within the forebrain during optic cup formation, and in the eye at later stages of retinal development. At 18 hours post fertilization (hpf), we detected expression of sox11a and sox11b in the telencephalon, and in the dorsal “corner” formed by the diencephalon and the evaginated optic stalk/optic vesicle (top panel arrows and second row closed asterisks, Figure 1A). We also detected faint expression of sox11a and sox11b at the ventral hinge of the optic stalk/optic vesicle axis (open asterisks, Figure 1A). However, we did not detect expression of sox11a or sox11b within the optic vesicle itself (Figure 1A). At 24 hpf, sox11a/b expression persisted in the diencephalon adjacent to the retina and in the telencephalon, and both paralogs were also expressed in the hypothalamus (Figures 1A, C). Within the developing retina at 24 hpf, sox11b was expressed diffusely across the lens and retinal neuroepithelium, and was distinctly visible in a small cluster of cells in the ventro-nasal retina (arrow, bottom right panel, Figure 1A), corresponding to the location at which retinal neurogenesis initiates [22]. As retinal development proceeded, sox11a expression was observed in the ganglion cell layer (GCL) at 48 hpf, whereas the expression of sox11b was detected in a few scattered cells across the central retina but was mostly restricted to the undifferentiated peripheral retina (Figure 1B; [14]). By 72 hpf, when retinal neurogenesis was mostly complete, both sox11a and sox11b were predominantly expressed in the persistently neurogenic ciliary marginal zone (Figure 1B; [14]); expression of sox11a also persisted in the GCL and in some cells in the inner half of the inner nuclear later (INL). Interestingly, the expression domains of both sox11 paralogs were adjacent to regions of shha expression in the ventral diencephalon at 18 and 24 hpf (Figure 1C), whereas at 48 hpf sox11a expression overlapped with the previously described location of shha in the GCL [23], [24].
To investigate the function of Sox11 paralogs during eye development, translation of sox11a and sox11b was blocked with morpholino oligonucleotides (MOs), whose efficiency and specificity were confirmed using a second sox11 MO and a GFP reporter assay, respectively (Figures S1A, C). Zebrafish embryos were injected with sox11a and sox11b MOs simultaneously (hereafter referred to as sox11 morphants), as co-inhibition of both paralogs induced the highest prevalence of ocular phenotypes (Figure S1B). At 24 hpf, 72.9±7.4% of sox11 morphants displayed a misshapen, rudimentary, or absent lens (Figures 2A, S1B). Sox11 morphant lenses mostly recovered to a spherical shape by 2 days post fertilization (dpf), however at this stage a similar proportion of morphants (70.0±7.7%) displayed coloboma (Figures 2A, S1B). Sox11 morphant eyes were also hypopigmented ventrally, and microphthalmic compared to controls (Figure 2A, B). Histological sections revealed that the colobomatous retinas in sox11 morphants frequently extruded through the open choroid fissure into the brain (Figure 2C). Approximately 54% (15 of 28 individuals examined) of sox11 morphant retinas with coloboma also exhibited poor or reduced retinal lamination, suggesting a delay in retinal differentiation. In contrast, of the sox11 morphant retinas that did not display coloboma, only 14% were poorly laminated (4 of 29). This suggests that similar mechanisms may underlie the ocular morphogenesis and retinal developmental defects observed in sox11 morphants with coloboma. The presence of the coloboma prevented the retinal pigmented epithelium (RPE) from completely enclosing the posterior eye (Figure 2C), which likely accounts for the hypopigmented appearance of the ventral portion of the eye when viewed laterally (Figure 2A). This coloboma phenotype was reminiscent of the zebrafish blowout mutant, which has a mutation in patched2 (formerly named patched1), a negative regulator of Hh signaling [25], [26]. In addition to the ocular phenotypes, sox11 morphants also frequently displayed a downward kink of the tail, as well as brain abnormalities such as widened ventricles, likely reflecting Sox11's expression and function in the posterior somites and developing brain, respectively [10], [20]. All of the morphant phenotypes were rescued by injection of wild type sox11a and sox11b mRNA, consistent with the morpholinos being specific for Sox11 (Figure 2A, D). Importantly, these phenotypes were also rescued by injection of human SOX11 mRNA, indicating that the function of Sox11 in regulating early eye development is evolutionarily conserved (Figure 2D).
One mechanism that has been suggested to contribute to optic fissure closure defects is overproliferation of progenitor cells within the presumptive neural retina [27]. To determine whether changes in mitotic activity underlie the lens and coloboma phenotypes in sox11 morphants, we immunolabeled retinal sections from control and sox11 morphants with an antibody to phosphohistone H3 (PH3). We observed a modest but significant increase in the number of PH3-positive cells in sox11 morphant optic vesicle and retinas at 18 and 24 hpf, and a larger increase in proliferation relative to controls at 48 and 72 hpf (Figure S2C, D). However, the excess PH3-positive cells were not clustered in the ventral retina, optic stalk, or lens at 24 hpf (Figure S2D), by which time the abnormal ocular phenotypes are already apparent. Therefore, we conclude that overproliferation likely does not underlie the early ocular phenotypes of sox11 morphants. We also performed TUNEL staining on sections from control and sox11 morphant retinas (Figure S2A, B). This analysis revealed a variable but significant increase in TUNEL-positive cells in the optic vesicle of sox11 morphants at 18 hpf. At 24 hpf, we did not detect elevated apoptosis in the retina or optic stalk of Sox11-deficient embryos. However, we did observe a significant increase in TUNEL-positive cells in the anterior lens of sox11 morphants, which persisted through 72 hpf (Figure S2A, B). This increase in apoptotic cells in the lens may be related to the abnormal lens morphology we observed by light microscopy (Figure 2A). Finally, we observed an increase in TUNEL-positive cells in the colobomatous tissue of sox11 morphant retinas at 48 hpf (Figure S2B), indicating that this abnormal ocular structure negatively impacted the survival of the cells within it.
Given that expression of sox11a/b is upregulated in adult zebrafish rod progenitor cells during rod photoreceptor regeneration [14], we investigated whether Sox11-deficient embryos displayed altered rod development. Since we found that a significant proportion of sox11 morphant retinas with coloboma also displayed poor lamination, indicating a potential delay in retinal development, for analysis we divided the sox11 morphants into those with and without coloboma. This approach minimized the potential secondary effects on retinal development from the ocular morphogenetic defect masking any additional role for Sox11 in retinal neurogenesis. Using immunohistochemistry with cell-type specific antibodies, we found that sox11 morphants without coloboma (approximately 30% of morphant embryos) possessed well-laminated retinas with normal numbers of ganglion, amacrine, horizontal, and bipolar cells, Müller glia, and cone photoreceptors at 72 hpf (Figure S3A, B). In contrast, when control and sox11 MOs were injected into a rod photoreceptor-GFP transgenic reporter line [28], we observed a significant reduction in mature rod photoreceptors in sox11 morphant retinas without coloboma at 3 dpf (control embryos, 34.9±7.4rods/section; sox11 morphants, 8.7±8.9 rods/section; p<0.00001; Figure 3A, B). Furthermore, several retinal sections from sox11 morphants contained no detectable GFP-positive rods at 3 dpf. The reduction in mature rod photoreceptors in sox11 morphant retinas was confirmed by immunolabeling with the rod-specific antibody 4C12 (not shown), by fluorescent in situ hybridization (FISH) of retinal sections with a probe for rhodopsin (rho), and by quantitative RT-PCR (qPCR) for the rho transcript at 3 dpf (Figure 3C, D). Rod photoreceptor number could be rescued by injection of wild type sox11 mRNA (Figure 3B), demonstrating that the reduction in rods was due to Sox11 deficiency. To determine whether depletion of Sox11 blocks specification of the rod photoreceptor fate, we conducted FISH on 3 dpf retinal sections from control and sox11 morphants using probes for three genes associated with the rod photoreceptor lineage: neuroD, crx, and nr2e3 [29]–[31]. Interestingly, we found that expression of all three rod lineage genes was qualitatively normal in sox11 morphant retinas, even those with coloboma and poor lamination (Figure 3C). We also verified by qPCR that nr2e3 transcript levels were not significantly different in sox11 morphants and controls (Figure 3D) Therefore, these data suggest that Sox11 is required for the terminal differentiation, but not the specification, of rod photoreceptor cells. Because the window of rod photoreceptor differentiation is longer than that of cones or other retinal neurons [32], [33] we investigated whether rod photoreceptor number remained reduced in sox11 morphants later in development. The number of rods in sox11 morphant retinas was higher at 4 dpf than at 3 dpf, but remained significantly reduced relative to controls (sox11 morphants, 15.9±2.9 rods/section; controls, 57.9±5.4 rods/section; p<0.001; Figure S3C). Taken together, these data suggest that terminal differentiation of rods requires Sox11.
As mentioned above, the coloboma phenotype of sox11 morphants is similar to the zebrafish blowout mutant, in which increased Hedgehog signaling results in altered proximodistal patterning of the optic vesicle [25]. To determine whether a similar defect was present in sox11 morphants, we performed FISH on retinal sections to examine the expression of pax2a and pax6a, which mark optic stalk and retinal territories, respectively. This revealed expansion of the pax2a domain in approximately 50% of sox11 morphant embryos at 18 and 36 hpf, while at later stages (48 hpf), expression persisted around the open choroid fissure, whereas it was barely detectable in controls (Figure 4A). These expression changes were verified by qPCR at 18 and 24 hpf. Although we did not observe a concomitant decrease in pax6a expression in the optic vesicle at 18 hpf, there was a significant reduction in transcript levels detected by qPCR at 24 hpf (Figure 4B). To further test whether Hh signaling was elevated in sox11 morphants, we made use of a recently described Hh signaling reporter line of zebrafish, which expresses GFP under the control of the patched2 (ptc2) promoter [34]. Sections through the head of 24 hpf control and sox11 morphants on the ptc2:GFP background revealed both an increase in GFP expression and an expansion of the GFP-positive domains in the brain, retina, and RPE of sox11 morphants (Figure 5A). Taken together, these data strongly suggest that Hh signaling is indeed elevated in sox11 morphants.
To directly test this hypothesis, control and sox11 morphant embryos were treated from 5.5–13 hpf with the Hh inhibitor cyclopamine. This treatment window was chosen because it resulted in maximal rescue of colobomas in the blowout mutant [25]. The proportion of embryos displaying a malformed lens at 24 hpf (21.3±8.8%) or coloboma at 2 dpf (10.1±3.8%) was significantly reduced after cyclopamine treatment, compared to vehicle-treated sox11 morphants (>70% for both phenotypes; p<0.0001; Figures 5B and S4A). Moreover, cyclopamine treatment significantly increased the number of rods at 72 hpf (sox11 MO: 20.3±11.1 rods/section; sox11 MO + cyclopamine: 43.8±10.5 rods/section; p = 0.02; Figures 5C and S4B), and corrected the lamination and differentiation defects that were associated with colobomatous retinas (Figure S4C). In a reciprocal experiment, embryos were injected with half the full dose of each sox11 MO, and treated with either a Hh agonist (purmorphamine) or vehicle control (DMSO) from 5.5–24 hpf. We used a sub-threshold dose of purmorphamine (75 µM), which did not cause coloboma when given alone (Figure 5B). In contrast, when the half dose of sox11 MO was combined with purmorphamine, the prevalence of lens malformations at 24 hpf and coloboma at 2 dpf significantly increased (sox11 MO half dose + purmorphamine: 57.9±10.2% malformed lens, 57.2±5.3% coloboma; sox11 MO half dose + DMSO: 24.2±3.5% malformed lens, 28.9±5.9% coloboma; p<0.0001; Figures 5B and S4A). Together, these data demonstrate that deficiency of Sox11 increases Hh signaling, resulting in defects in ocular morphogenesis and reduced rod photoreceptor number.
Finally, we injected sox11a and sox11b mRNA into wildtype zebrafish embryos and evaluated the prevalence of a cyclopic phenotype, which is classically associated with reduced Hh pathway activity [3], at 24 hpf. Injection of sox11a and sox11b mRNA caused a cyclopic phenotype in 33.8±2.9% of the embryos, whereas only 4.2±0.7% of embryos had cyclopia when injected with a control td-tomato mRNA (p<0.001; Figure 5D). Taken together, these results demonstrate that Sox11 is required to limit Hh signaling during zebrafish ocular development.
Zebrafish possess five Hedgehog ligands (Sonic hedgehog a and b, Indian hedgehog a and b, and Desert hedgehog), two Patched and one Smoothened receptor, and four Gli effectors. To determine whether expression of any of these pathway members was altered in sox11 morphants, we performed qPCR on mRNA prepared from 18 and 24 hpf control and sox11 morphant heads. At 18 hpf, no significant gene expression changes were observed, except for gli2a and gli3, which were both slightly elevated in sox11 morphants (Figure S5A). In contrast, at 24 hpf we observed a very strong increase (189-fold) in the expression of shha in sox11 morphants relative to controls, as well as a modest decrease in ihhb, ptc1, and gli2b expression, and a 2-fold increase in expression of ptc2 (Figure 6A). The increase in shha expression in sox11 morphants appeared to be dose-dependent, as injection of one-half the full dose of sox11 MOs resulted in only a 65-fold elevation in shha (Figure S5B). In situ hybridization revealed greatly increased shha signal intensity in regions of the sox11 morphant embryo that normally express shha, such as the ventral forebrain and the notochord (Figure 6B), with ectopic expression observed in the dorsal midbrain and telencephalon (Figure 6B). These results suggest that the ocular phenotypes in sox11 morphants are caused by elevated levels of Shha. However, we were puzzled that shha transcript levels were not significantly increased at 18 hpf (Figure S5A), and yet cyclopamine treatment from 5.5–13 hpf rescued the ocular defects of sox11 morphants. Therefore, we asked whether shha levels were elevated in sox11 morphants at earlier time points. We performed qPCR analysis on mRNA from control and sox11 morphants at 8, 10, and 12 hpf and found that shha levels are elevated approximately 2-fold in sox11 morphants at these time points (Figure 6C). Moreover, using the ptc2:GFP line, we detected increased GFP levels in the ventral midline of sox11 morphants at 12 hpf, confirming that Hh signaling was elevated at this stage (Figure 6D). Taken together, these results suggest that knockdown of sox11 results in elevated expression of shha, and an increase in Hh signaling, as early as 8–12 hpf when the optic vesicle is evaginating from the midline.
To test the hypothesis that elevated Shha levels cause the ocular phenotypes of sox11 morphants, we knocked down both shha and sox11 simultaneously (using our sox11 MOs and a previously described shha MO [35]) and scored embryos at 24 hpf and 2 dpf for malformed lens and coloboma phenotypes, respectively. We used a low dose of the shha MO (3.14 ng/embryo), which by itself did not produce lens defects, coloboma, or rod photoreceptor defects (Figures 6E and S5C). The prevalence of ocular phenotypes was significantly reduced in the double morphants (sox11 MO: 70.3%±6.7% malformed lens, 75.4%±8.3% coloboma; sox11 + shha MOs: 28.9%±9.2% malformed lens; 34.7%±2.7% coloboma; p<0.0001; Figures 6E and S5C). Rod photoreceptor number was also significantly increased at 3 dpf in the double shha/sox11 morphants, however it did not reach the levels observed in controls (sox11 MO: 5.8±7.1 rods/section; sox11 + shha MOs: 14.6±2.3 rods/section; p<0.05; Figure S5D). We performed qPCR analysis on sox11 morphants treated with cyclopamine and purmorphamine and confirmed that these treatments caused a decrease and an increase in shha transcript levels, respectively (Figure S5E, F). Cyclopamine treatment of sox11 morphants also restored expression of the Hh target gene ptc2 to control levels (data not shown). Moreover, qPCR analysis of embryos injected with control or sox11 mRNA confirmed that overexpression of sox11 resulted in a concomitant decrease in shha expression (Figure S5G). Finally, we determined that there was not a reciprocal regulation of sox11 by shha, because injection of the shha morpholino alone did not result in a change in expression of sox11a or sox11b (Figure S5H). Taken together, these results demonstrate that Sox11 controls levels of Hh signaling primarily through negative regulation of shha expression, and that limiting shha expression is essential for proper ocular morphogenesis.
Thus far, our data strongly suggest that Sox11 is required to limit levels of shha expression during ocular development. However, Sox11 and other members of the SoxC family have previously been shown to function as transcriptional activators rather than repressors [36]–[38]. Furthermore, a scan of the shha promoter revealed no perfect consensus binding sequences for Sox factors (not shown; [37]), and the expression domains of sox11 and shha only partially overlap in the ventral midline during ocular morphogenesis. Therefore, we hypothesized that Sox11 negatively regulates Shha indirectly and perhaps non-cell autonomously, by activating the expression of an upstream inhibitor of Shha. We searched the literature to identify candidate Shha repressors that are expressed in the forebrain during development, and then asked whether expression of any of these factors was reduced in sox11 morphant heads at 24 hpf (Figure 7A). We analyzed five candidate genes: bmp7b, fgfr2, tbx2a, tbx2b, and kras, which had been shown previously to negatively regulate shha expression during development [39]–[44]. Of these five, bmp7b showed significantly decreased expression in sox11 morphants compared to controls (Figure 7A). Bmp7b represents a good candidate intermediary between Sox11 and shha for several reasons. First, Bmp7 null mice display microphthalmia and optic fissure defects, similar to sox11 null mice [45]. Second, bmp7 is expressed in the ventral midline and proximal optic vesicle in the mouse [45], and bmp7b is expressed in the forebrain adjacent to the optic vesicle in zebrafish at 18 hpf in a similar pattern to sox11 [41]. Third, bmp7 expression was reported to be reduced in Sox11−/− mice [13]. And finally, a scan of the bmp7b promoter revealed two perfect Sox consensus binding sites [37] located approximately 950 bp upstream of the transcription start site (not shown).
Because we had detected elevated shha levels as early as 8 hpf in sox11 morphants, we asked whether bmp7b expression is also downregulated at that time. qPCR analysis revealed that bmp7b transcript levels were significantly reduced at 8, 10, and 12 hpf in sox11 morphants when compared to controls (Figure S6). Interestingly, bmp7b expression increased to just above control levels at 18 hpf, before declining significantly again at 24 hpf. This rebound in bmp7b expression at 18 hpf precisely mirrors the normal levels of shha expression in sox11 morphants at this time (Figure S5A). Taken together, these data suggest that the initial decrease in bmp7b expression (and corresponding elevation of shha) caused by knockdown of sox11 induces a compensatory pathway that works to bring transcriptional levels back to normal, but that the continued knockdown of sox11 results in renewed dysregulation of bmp7b and shha.
We reasoned that if Bmp7b functions downstream of Sox11 and upstream of Shha, then expression of bmp7b in sox11 morphants should rescue the ocular phenotypes caused by elevated Hh signaling. To test this hypothesis, we injected bmp7b mRNA into control and sox11 morphant embryos, and determined the proportion of embryos that displayed lens defects and coloboma at 24 hpf and 2 dpf, respectively. We found that co-injection of bmp7b mRNA into sox11 morphants significantly reduced the number of embryos displaying ocular phenotypes (sox11 MO: 72.6±2.22% malformed lens, 74.5±1.8% coloboma; sox11 MO + bmp7b mRNA: 35.6±6.9% malformed lens; 43.8±14.4% coloboma; p<0.001; Figure 7B, C), although the rescue was not as large as that observed with cyclopamine treatment. These data suggest that Sox11 negatively regulates shha at least in part through Bmp7b.
As functional redundancy between SoxC family members has been observed in mouse models [10],[36],[46], we investigated whether another SoxC factor could compensate for the loss of Sox11 during zebrafish ocular morphogenesis. By in situ hybridization and qPCR, we observed elevated expression of the SoxC factor sox4a in sox11 morphants at 24 and 36 hpf, suggesting that sox11 deficiency induces a compensatory increase in sox4 expression (Figure S7A, B). We then injected sox4 mRNA into sox11 morphants and found that this significantly reduced the proportion of embryos with lens and coloboma phenotypes (Figure S7C). This result suggests that increased Sox4 expression may buffer the effects of Sox11 deficiency. Consistent with this hypothesis, we observed a significantly greater proportion of embryos with coloboma in sox4/sox11 double morphants than when either gene was knocked down alone (data not shown).
To investigate whether SOX11 mutations contribute to patient phenotypes, the coding region was sequenced in DNA samples from 79 MAC patients [47]. These DNA samples had been previously screened for mutations in two other coloboma-related genes, GDF3 and GDF6 [47]–[49]. We identified heterozygous sequence changes in two probands (Figure 8A), both of whom are Canadians of white European ancestry. The first, a c.488G→T missense mutation in a coloboma patient, is predicted to result in a G145C amino acid alteration, considered damaging by SIFT analysis (http://sift.jcvi.org/). The second variant, a 12-nucleotide duplication (c.1106–1117) in a patient with bilateral iris and retino-choroidal coloboma (Figure 8B), is predicted to result in an in-frame, four amino acid duplication (S351–354dup). The affected amino acid residues are located outside previously defined functional domains and are conserved in chimp and macaque SOX11 (Figure 8A). These variants were absent from dbSNP and the 1000 Genomes databases, and from the NHLBI database comprising more than ten thousand exomes (Figure S8A). Sequencing of SOX11 from the probands' family members revealed that the S351–354dup alteration was present in the proband's mother, who did not exhibit a phenotype clinically (Figure 8C). In light of the rod photoreceptor phenotype in zebrafish sox11 morphants, an electroretinogram (ERG) was performed on the mother carrying the S351–354dup alteration. This analysis demonstrated a reduction in scotopic b-wave amplitude, indicating reduced rod photoreceptor function (Figure S8B). In addition, her 10Hz dim white flicker response was appreciably reduced, and was associated with a change in latency. The mother was asymptomatic at the time the ERG was performed, which may reflect her young age (37 years). Her cone flicker response was normal.
Intrigued by the presence of phenotypic effects in a heterozygote only on targeted testing, 384 DNA samples derived from patients undergoing screening for hemochromatosis were sequenced, which detected the S351–354dup variant in three individuals, whilst the G145C variant was absent. Unfortunately, these three carriers could not be recalled for clinical examination.
To determine whether the two SOX11 sequence variants had functional consequences, their ability to rescue the lens and coloboma phenotypes of zebrafish sox11 morphants was compared to wild type human SOX11 mRNA. Whereas wild type SOX11 mRNA significantly reduced the proportion of sox11 morphants displaying lens defects and coloboma, no significant rescue was observed with mRNA containing either SOX11 variant (Figures 8D and S8D), suggesting that both sequence changes compromise SOX11 function. Next, we utilized a luciferase reporter containing the promoter region of the SOX11 target gene GDF5 [50] to further analyze the functional consequences of the two mutations. Expression of increasing amounts of wild type SOX11 in COS-7 cells produced a dose-dependent increase in luciferase activity from the GDF5 reporter (Figure 8E). In contrast, transfection of equivalent amounts of either SOX11 variant did not enhance luciferase activity over the empty vector control (Figure 8E), although the variants showed comparable levels of protein expression by Western blot (Figure S8C). Equivalent results were obtained with the luciferase assay in two additional cell lines (HEK293 and HeLa; data not shown). To further confirm that the two SOX11 sequence variants are functionally compromised, we overexpressed them in zebrafish and quantified the proportion of embryos that exhibited a cyclopic phenotype at 24 hpf. Whereas injection of WT human SOX11 mRNA caused a significant increase in the proportion of cyclopic embryos compared to injection of control td-Tomato mRNA (61.16±10.7% in WT SOX11 injected vs. 35.3±11% in control injected; p<0.05), neither of the SOX11 sequence variants produced elevated levels of cyclopia (G145C, 10.0±4.8%, S351–354dup, 13.1±4.11%; Figures 8F and S8E). Taken together, these data suggest that the two variants compromise SOX11's transactivation ability.
Finally, array comparative genomic hybridization (array CGH) was performed on DNA from a patient with microphthalmia, unilateral optic nerve agenesis, and a de novo chromosome 2p25 deletion [18]. This defined a 1.14 Mb segmental deletion (5,206,155–6,343,906; chromosome build GRCh37), encompassing an interval within which SOX11 is the only protein-coding gene (Figures 8G and S8F). Taken together, these data demonstrate that perturbed SOX11 function, either through mutation or decreased gene dosage, contributes to structural (microphthalmia/coloboma) or functional (rod photoreceptor) phenotypes.
This study reveals a novel role for Sox11 in maintaining the correct level of Hedgehog (Hh) signaling during ocular morphogenesis. We demonstrate that knockdown of Sox11 in zebrafish perturbs lens formation, induces coloboma, and reduces the number of differentiated rod photoreceptors – phenotypes that can be rescued by pharmacological inhibition of the Hh pathway (cyclopamine) or morpholino inhibition of shha. Comparable lenticular and coloboma phenotypes have also been observed in murine mutants [13], demonstrating that Sox11's function in vertebrate ocular development is evolutionarily conserved. However, the perinatal lethality of Sox11 null mice has precluded a thorough in vivo assessment of rod photoreceptor differentiation, which mostly occurs postnatally. Expression of the rod photoreceptor genes Nrl, Nr2e3, and Sag (Rod arrestin) is significantly reduced in E16 retinas from Sox11−/− mice [51], suggesting that Sox11 does regulate aspects of rod photoreceptor differentiation in mammals. However, in retinal explants derived from Sox11 null mice and cultured for several days, reduced rod photoreceptor number was not observed [51]. Our data suggesting that early, midline-derived Shh influences rod photoreceptor differentiation (see below), indicates that retinal explants, being removed from the source of extra-retinal Shh, may not accurately reflect the in vivo response of retinal progenitor cells to their environment. In this context, the external embryogenesis, rapid pace of retinal development, and continual rod photoreceptor genesis in the zebrafish have benefitted our studies and permitted us to uncover for the first time both the mechanism of Sox11's action during early ocular development, as well as a role for Sox11 in regulating rod photoreceptor differentiation.
A second key finding of our study is that Sox11 acts upstream of Hh signaling specifically by negatively regulating transcription of the ligand shha. In Sox11-deficient embryos, we observed a strong increase in shha expression in the ventral forebrain, as well as an expansion of the shha territory into the dorsal diencephalon and the telencephalon. Therefore, in addition to regulating expression of shha expression in the ventral midline, our data suggest that Sox11 is also required to prevent activation of shha in the more dorsal regions of the brain. Within the retina, the expression of sox11a in the GCL at 48 hpf suggests that Sox11 continues to regulate Hh signaling during retinal neurogenesis.
The magnitude of the increase in shha expression in the absence of sox11 (over 180-fold) at 24 hpf suggests that loss of shha transcriptional repression is accompanied by a significant positive transcriptional feedback loop. However, the Hh target gene ptc2 demonstrated a much smaller increase in expression (2-fold) at this time, raising the question of why the dramatic upregulation in shha did not produce a correspondingly large transcriptional response. One possible explanation is that post-transcriptional mechanisms narrow the range of Shha protein expression in sox11 morphants. Moreover, additional feedback mechanisms may work to attenuate the transcriptional response of Hh target genes such as patched. In any case, the elevated and expanded GFP expression in the Hh reporter line ptc2:GFP, as well as the rescue by cyclopamine and shha co-knockdown, strongly argue that the rise in shha transcription induced by sox11 deficiency has functional consequences.
In the absence of Sox11, we observed an early expansion of the optic stalk marker pax2.1 in the optic vesicle, and a later reduction in the retinal marker pax6a. Such altered proximodistal patterning of the optic vesicle has been observed in several models of elevated Hh signaling [4], [5], [25], [52], [53]. The increased apoptosis in the lens and its abnormal development may be attributable to reduced pax6a expression in sox11 morphants, since similar phenotypes were observed in a lens-specific Pax6 conditional mutant mouse model [54]. In parallel, we suggest the expansion of pax2.1 expression due to elevated levels of Shh enlarged the area of the optic vesicle that was specified as optic stalk, hindering closure of the choroid fissure and thus causing coloboma. Elevated Hh signaling could also account for the increase in mitotic cells in the retina, as this pathway is known to be mitogenic [55].
Previous studies in zebrafish have shown that blocking early Hh signaling, either with shha and shhb morpholinos or by cyclopamine treatment, caused a reduction in rhodopsin expression in the retina, suggesting that Hh signaling promotes rod photoreceptor differentiation [56]. However, murine studies have found that activation of the Hh pathway results in a non-cell autonomous inhibition of rhodopsin expression [57], which is consistent with our results. Moreover, loss of Shh was shown to cause accelerated differentiation of rods and cones in a conditional mouse model [58]. The seemingly paradoxical response to increased and decreased Shh levels is potentially explained by the requirement for precise Shh dosage, with either alteration resulting in reduced photoreceptor number. This accords with a comparable model for Shh's effect on reactive astrocytes [59], and is a well-recognized feature of transcription factors, as exemplified by the effects of altered Pax6 dosage in inducing microphthalmia [60].
Interestingly, we observed a significant increase in shha expression at 8–12 hpf, when the optic vesicle is evaginating from the midline, and we confirmed that Hh signaling was increased at this time using a ptc2:GFP reporter line (Figure 6). Furthermore, treatment of sox11 morphants with cyclopamine during this developmental window was sufficient to restore rod photoreceptor number at 72 hpf. Thus, taken together, these data indicate that early, midline-derived Shh influences rod photoreceptor differentiation. This is not the first demonstration that early midline Hh signals influence later neurogenesis in the retina. It has been shown previously that the timely progression of ath5 expression in the retina, which coincides with the activation of neurogenesis, depends on axial Shh [61]. As ath5-positive cells contribute significantly to the rod photoreceptor lineage [62], it is plausible that elevated Shh coming from the midline in sox11 morphants delays rod photoreceptor differentiation by influencing the cell-intrinsic neurogenic program of retinal progenitor cells.
Although the phenotypes of zebrafish blowout (blw) mutants and sox11 morphants are similar with respect to coloboma, blw mutants do not appear to have a defect in the differentiation of rod photoreceptors or any other retinal cell types. This is surprising, given the well-described influence of Hh signaling on retinal neurogenesis [3], [4], [33], [56], [61], [63]–[66]. Moreover, patients with elevated Hh signaling due to heterozygous loss of function mutations in PTCH exhibit retinal abnormalities, and PtchlacZ+/− mice display a delay in photoreceptor and horizontal cell maturation at P5, all of which is consistent with our data [67]. One possible explanation as to why sox11 morphants and blw mutants differ in this aspect of their phenotype is that the mutation in ptc2 may be a partial loss of function allele, which is supported by the observation that ptc2 morphants display more severe phenotypes than blw mutants [25].
Since SoxC factors are generally considered to function as transcriptional activators rather than repressors [36], [38], we hypothesized that Sox11 regulates Shha indirectly, through the induction of a repressor. Indeed, we found that bmp7b expression was significantly reduced in sox11 morphants, and that injection of bmp7b mRNA into sox11 morphants could rescue the lens and coloboma phenotypes (Figure 7). As Bmp7 has previously been shown to antagonize Shh signaling [42], [68], our results are compatible with a model whereby Bmp7 functions downstream of Sox11 to limit Shh expression during ocular morphogenesis. However, since the magnitude of the bmp7b rescue was not as large as that observed with cyclopamine treatment, additional mechanisms linking Sox11 with the regulation of Hh signaling are likely.
So far, more than 27 genes are associated with coloboma in humans [2], however mutations in these account for less than 20% of cases. Consequently, it is important to define additional causative genes, both to extend understanding of pathogenesis and define pathways that may be amenable to therapeutic modulation. Our work, in combination with previous studies [13], strongly supports a contribution from SOX11 to coloboma phenotypes, however our data indicate that the relationship is complex. With a 50% reduction in gene dosage (2p25 segmental deletion; Figure 8G), a profound phenotype was observed. In contrast, milder coding changes (S351–354dup) with a low prevalence in the general population, resulted in incompletely penetrant phenotypes, with the unaffected carrier exhibiting a sub-clinical phenotype, only detectable on ERG testing. Since the variants had significantly reduced function on in vitro and in vivo assays, this suggests that such mild alleles contribute to MAC but may be insufficient to induce phenotypes alone in all cases. Coloboma, like many developmental defects, exhibits extensive phenotypic variability, suggesting complex relationships between disease genes and modifying alleles that complicate simple genotype-phenotype correlations. It is also possible that oligogenic inheritance is a factor in coloboma, in which individuals in non-penetrant families carry a combination of pathogenic alleles at two or more disease loci, as has been described for other genetically heterogeneous developmental disorders such as the ciliopathies [69]. Furthermore, functional redundancy between Sox subgroup family members is also commonly observed [36], [46], [70], suggesting that one SoxC family member may buffer the effects of mutation in a second. Consistent with this model, we observed elevated expression of the SoxC factor sox4 in sox11 morphants at 24 and 36 hpf, and found that the lens and coloboma phenotypes of sox11 morphants could be rescued by injection of sox4 mRNA (Figure S7). Finally, in light of the incompletely penetrant phenotypes evident with multiple other MAC-causing genes [47], [49], [71]–[73], a similar additive contribution from other SOX gene variants is highly plausible.
In summary, we describe here a novel role for Sox11 in regulating levels of Shh during ocular morphogenesis. It will be interesting to determine whether dysregulated Hh signaling underlies any of the additional developmental defects observed in Sox11−/− mice, such as congenital cardiac malformations and craniofacial anomalies. Future studies will continue to explore the mechanisms of how Sox11 regulates Hh signaling and Shh transcription, as well as the identification of direct molecular targets of Sox11 transcriptional control.
The Tg (XlRho:EGFP)fl1 transgenic line has been previously described [28], and was generously provided by J.M. Fadool (Florida State University, Tallahassee, FL). The Tg (gfap:GFP)mi2001 line has been previously described [74] and was obtained from the Zebrafish International Resource Center (Eugene, OR). The Tg (3.2TαC-EGFP) line has been previously described [75], and was generously provided by S.E. Brockerhoff (University of Washington, Seattle, WA). Tg(GBS-ptch2:nlsEGFP) has been previously described [34] and was kindly provided by R. Karlstrom (University of Massachusetts, Amherst, MA). Zebrafish (Danio rerio) were reared, bred, and staged according to standard protocols [76], [77]. All animal procedures were carried out in accordance with the policies established by the University of Kentucky Institutional Animal Care and Use Committee (IUCAC).
Morpholinos (MOs) were obtained from Gene Tools, LLC (Philomath, OR) and were prepared and injected as previously described [78]. The following MOs were used in this study: standard control MO, 5′-CCTCTTACCTCAGTTACAATTTATA-3′; sox11a MO1, 5′ –GTGCGTTGTCAGTCCAAAATATCAA-3′; sox11b MO1, 5′ –CATGTTCAAACACACTTTTCCCTCT; shha-MO: 5′CAGCACTCTCGTCAAAAGCCGCATT [35]. The specificity of the sox11 morphant phenotype was confirmed using a second sox11 morpholino placed further downstream of the first set (completely non-overlapping with sox11a MO1, and overlapping by only 4 nucleotides with sox11b MO1). Because the target site for this morpholino extended into the coding region (which is highly similar in sequence for both genes) it simultaneously targets both sox11a and sox11b (sox11 MO2, 5′ –TCCGTTTGCPGCACCATG-3′; the “P” indicates a photo-cleavable moiety that was not used in this study). The sox11 MO2 produced the same coloboma phenotype as the first set of MOs (Figure S1C). All data presented in this study are from embryos injected with sox11a MO1 and sox11b MO1. Unless stated otherwise, embryos were injected with 4.18 ng each of sox11a MO1 and sox11b MO1, 4.18 ng of the standard control MO, or with 3.14 ng of shha MO. We also confirmed that no abnormal phenotypes were observed when embryos were injected with 8 ng of standard control MO. To determine the efficiency of the sox11 MOs, PCR fragments corresponding to the 5′UTRs of sox11a and sox11b encompassing the morpholino target sequences were amplified (using primers listed in Table S1) and cloned upstream and in frame with the EGFP gene in the pEF1α:GFP plasmid (Addgene plasmid 11154). One-cell stage zebrafish embryos were injected with 100 pg/embryo of pEF1α:GFP plasmid containing the MO binding site in the presence or absence of the sox11 MOs. GFP expression in injected embryos was analyzed by fluorescence microscopy at 24 hpf.
Zebrafish sox11a and sox11b or human wild type and variant SOX11 coding sequences were PCR amplified (using primers listed in Table S1) and cloned into the pGEMT-easy vector (Promega). The pCRII-bmp7b plasmid has been previously described [41] and was a kind gift from Dr. S. Fabrizio (The Novartis Institutes for Biomedical Research, Cambridge, MA). The constructs were linearized and mRNA was prepared using the mMESSAGE mMACHINE kit (Ambion) according to manufacturer's instructions. Zebrafish sox11a and sox11b mRNAs (1.0 ng each), human SOX11 mRNA (0.3 ng), zebrafish bmp7b mRNA (1.0 ng) or zebrafish sox4a and sox4b (0.5 ng each) were injected into zebrafish embryos at the one-cell stage. For mRNA rescue experiments, the mRNAs were either co-injected with sox11 MOs, or were injected sequentially after injection of the MOs. As both methods produced similar results, the data presented here are for co-injection of mRNA and morpholino. Injections were always performed in triplicate, and a minimum of 55 injected embryos were analyzed in each experiment. For mRNA overexpression experiments, embryos were injected with either a control (tdTomato) mRNA, zebrafish sox11a/b mRNA, or human WT, G145C (MI), or S351–354dup (MII) SOX11 mRNA, all at equimolar concentrations. The control mRNA was synthesized from pRSET-B-td-Tomato (kindly provided by Dr. D.A. Harrison, University of Kentucky, Lexington, KY). To compare control versus sox11a/b mRNA, 0.003 pmol of each mRNA was injected. To compare control versus human WT and variant SOX11 mRNA, 0.0133 pmol of each mRNA was injected. Zebrafish embryos were injected at the one-cell stage, and embryos were scored for cyclopic phenotypes (one single eye in the center of the head, two eyes that were almost fused at the midline, or one normal eye and one vestigial eye) at 24 hpf.
To screen for mutations in human SOX11, PCR was performed using three sets of overlapping primers that spanned the entire coding region of the single-exon SOX11 gene. The amplicons were sequenced on an ABI Prism 3100 capillary sequencer (Applied Biosystems), analyzed using DNABaser v.3.1.5 and sequence alignments were performed using ClustalW. Mutations were confirmed by bi-directional Sanger sequencing and RFLP analysis of the SOX11 amplicons. Half of the 384 control DNA samples were screened by RFLP analysis, using TseI (NEB) for the G145C variant and SfcI (NEB) for the S351–S354dup variant, and the other half were screened by direct Sanger sequencing of the SOX11 coding region. Array CGH analysis was performed using a custom designed Nimblegen 4×72 whole human genome array. Oligonucleotide probes were spaced approximately every 75 bp across a 2.65 Mb region at 2p25.2, and backbone probes covered the rest of the genome. Four technical replicates were performed on the proband's DNA, and two replicate hybridizations were performed for each parental DNA sample. Array hybridization and scanning were performed by the Roy Carver Center for Genomics at the University of Iowa (Iowa City, IA). Array data were analyzed using the segMNT analysis program (Nimblegen). Informed consent was obtained from all participants. Study approval was provided by the University of Alberta Hospital Health Research Ethics Board and the Ethics Committee of the IRCCS Oasi Maria SS Onlus, Troina, Italy.
Cyclopamine (Sigma) was resuspended at 1 mM concentration in 100% ethanol and diluted in fish water for exposure. A dose response curve was generated by exposing wild type embryos to 0.5, 1.0, and 2.0 µM of cyclopamine from 5.5–13 hpf, and the dose (2.0 µM) at which no abnormal phenotype and negligible toxicity was observed was used for control and sox11 morphants. Purmorphamine (Calbiochem) was resuspended at 50 mM concentration in DMSO and diluted in fish water for exposures. Wild type embryos were exposed to 10–100 µM of purmorphamine from 5.5–24 hpf, and the dose (75 µM) at which no ocular phenotypes were observed was used to treat control and sox11 morphants.
Whole mount in situ hybridization (WISH) and immunohistochemistry were performed essentially as previously described [78]. For FISH embryos were manually dechorionated and fixed in 4% paraformaldehyde (PFA) made with diethyl pyrocarbonate (DEPC)-treated PBS at 4°C overnight. The fixed embryos were sequentially cryoprotected in 10% sucrose-DEPC and 30% sucrose-DEPC at 4°C overnight. Embryos were then embedded in OCT (Ted Pella, Redding, CA) and frozen at −80°C. Ten-micron sections were collected using a cryostat (Leica CM1900, Leica Biosystems, Buffalo Grove, IL), placed on Superfrost plus glass slides (Fisher Scientific, Waltham, MA) and air dried at room temperature overnight. The sections were post-fixed in 1% PFA-DEPC and rehydrated in PBST-DEPC. The sections were permeabilized for 10 minutes with 1 µg/ml proteinase K. Sections were acetylated in triethanolamine buffer plus 0.25% acetic anhydride (Sigma-Aldrich, Saint Louis, MO), and then rinsed in DEPC treated water. Sections were hybridized with digoxigenin (DIG) and fluorescein (FITC) labeled probes (2.5 ng/µl) in hybridization buffer (0.25% SDS, 10% dextran sulfate, 1× Denhardt's solution , 200 µg/ml torula yeast tRNA, 50% de-ionized formamide, 1 mM EDTA, 600 mM NaCl, and 10 mM Tris pH 7.5 in DEPC-treated water) at 65°C in a sealed humidified chamber for a minimum of 16 hours. Following hybridization, the slides were rinsed in 5× SSC and then with pre-warmed 1× SSC/50% formamide. Endogenous peroxidase activity was quenched with 1% H2O2 for 30 minutes. Sections were blocked using 0.5% PE blocking solution (Perkin Elmer Inc, Waltham, MA) for at least 1 hour. For two-color FISH, sections were incubated first with anti-DIG-POD Fab fragment (Roche, Indianapolis, IN) at 4°C overnight. Subsequently, probe signal was detected using the TSA plus Cy3 kit (Perkin Elmer Inc, Waltham, MA) following the manufacturer's instructions. For the second color detection, the sections were treated with 1% H2O2 for 30 minutes and then incubated with anti-FITC-POD Fab fragment (Roche, Indianapolis, IN) at 4°C overnight. Subsequently, the FITC-labeled probe signal was revealed using TSA plus Fluorescein (Perkin Elmer Inc, Waltham, MA). Finally, sections were counterstained with 4′, 6-diamidino-2-phenylindole (DAPI; Sigma-Aldrich, Saint Louis, MO), mounted in 40% glycerol, and imaged on an inverted fluorescent microscope (Nikon Eclipse Ti-U; Nikon Instruments, Melville, NY) using a 40× objective.
The sox11a, sox11b and shha cDNAs were amplified (using primers listed in Table S1) and cloned from 48 hpf whole embryo cDNA. The sox11b and NeuroD antisense probes have been previously described [14]. The pax6a , crx, and nr2e3 probes have been previously described, and were kindly provided by Y.F. Leung (Purdue University, Indiana). The pax2.1 probe has been described previously [25] and was a gift from J.M. Gross (University of Texas, Austin, TX). The following primary antibodies and dilutions were used: Zpr-1 (1∶20; ZIRC), which labels red-green cones; Zn-8 (1∶10; ZIRC), which labels ganglion cells; anti-Prox-1 (1∶2000; Millipore), which recognizes horizontal cells; anti-PH3 (1∶500; Millipore), which marks cells in G2/M phase; 5E11 (1∶10; J.M. Fadool, Florida State University), which labels amacrine cells; and anti-PKCα (1∶300; Santa Cruz Biotechnology), which labels bipolar cells. Alexa Fluor secondary antibodies (Molecular Probes, Invitrogen) and Cy-conjugated secondary antibodies (Jackson ImmunoResearch) were all used at 1∶200 dilution. Sections from the same region of the eye were analyzed for quantification purposes. One section was quantified per individual embryo (for both control and sox11 morphants).
Terminal deoxynucleotide transferase (TdT)-mediated dUTP nick end labeling (TUNEL) was performed on retinal cryosections using the ApopTag Fluorescein Direct In Situ Apoptosis Detection Kit (Millipore) according to the manufacturer's instructions. Sections from the same region of the eye were analyzed for quantification purposes. One section was quantified per individual embryo (for both control and sox11 morphants).
RNA extracted from the heads of control, sox11, and shha morphant embryos at various time points was used to perform first-strand cDNA synthesis (GoScript Reverse Transcriptase System; Promega). Real time PCR was performed using either Maxima SYBR Green qPCR master mix (Thermo Scientific) or FastStart SYBR Green Master (Roche) on an iCycler iQ Real Time PCR Detection system (Bio-Rad) or LightCycler 96 (Roche) with primers listed in Table S1. Three biological replicates were performed for each experiment. The gene expression change was determined using a relative standard curve quantification method with gapdh, atp5h, or 18s rRNA [79] expression as the normalization control.
Statistical analysis was performed on all data using the GraphPad Prism 6.02 software. Continuous data were analyzed using Student's-t-test and Fisher's exact test. For all graphs, data are represented as the mean ± the standard deviation (s.d.).
COS-7 cells were transfected with the pcDNA3 expression vector (Invitrogen) containing the coding region of wild type, G145C, or S351–354dup SOX11; the pGL3 Firefly Luciferase reporter vector (Promega) containing the GDF5 core promoter was a kind gift from Akinori Kan (Harvard Medical School, Boston, MA) [80]; and the pRL-TK vector (Promega) containing Renilla luciferase driven by a ubiquitous tyrosine kinase promoter to control for transfection efficiency. Transfections were performed using Fugene 6 (Promega), following manufacturer's instructions. The total mass of DNA and molar ratios of pGL3 and pRL-TK were held constant across transfections, which were repeated a minimum of 6 times. Dose response curves were generated using wild type SOX11 at 0∶100, 1∶20, 1∶10, and 1∶5 molar ratios to the GDF5 reporter. The mutant SOX11 variants were transfected at a 1∶5 molar ratio to the GDF5 reporter. Firefly and Renilla luciferase activity were measured 24–36 hours post transfection using the DualGlo Luciferase Assay System (Promega). Data was analyzed as follows: Firefly luciferase (FFLuc) was baselined against untransfected control (UTC) samples ( = FFLuc – UTC) and normalized using the Renilla luciferase (RLuc). The Relative Luciferase Activity (RLA) was calculated as (FFLuc-UTC)/RLuc and compared between experimental and control transfections.
|
10.1371/journal.ppat.1000620 | Equine Rhinitis A Virus and Its Low pH Empty Particle: Clues Towards an Aphthovirus Entry Mechanism? | Equine rhinitis A virus (ERAV) is closely related to foot-and-mouth disease virus (FMDV), belonging to the genus Aphthovirus of the Picornaviridae. How picornaviruses introduce their RNA genome into the cytoplasm of the host cell to initiate replication is unclear since they have no lipid envelope to facilitate fusion with cellular membranes. It has been thought that the dissociation of the FMDV particle into pentameric subunits at acidic pH is the mechanism for genome release during cell entry, but this raises the problem of how transfer across the endosome membrane of the genome might be facilitated. In contrast, most other picornaviruses form ‘altered’ particle intermediates (not reported for aphthoviruses) thought to induce membrane pores through which the genome can be transferred. Here we show that ERAV, like FMDV, dissociates into pentamers at mildly acidic pH but demonstrate that dissociation is preceded by the transient formation of empty 80S particles which have released their genome and may represent novel biologically relevant intermediates in the aphthovirus cell entry process. The crystal structures of the native ERAV virus and a low pH form have been determined via highly efficient crystallization and data collection strategies, required due to low virus yields. ERAV is closely similar to FMDV for VP2, VP3 and part of VP4 but VP1 diverges, to give a particle with a pitted surface, as seen in cardioviruses. The low pH particle has internal structure consistent with it representing a pre-dissociation cell entry intermediate. These results suggest a unified mechanism of picornavirus cell entry.
| Picornaviruses are small animal viruses comprising an RNA genome protected by a roughly spherical protein shell with icosahedral symmetry. How the RNA is introduced into the cytoplasm of the host cell to initiate replication is unclear since they have no lipid envelope to facilitate fusion with cellular membranes. Instead, they become internalized into endocytic vesicles whence the viral genome must be delivered through the vesicle membrane, into the cytoplasm. In some picornaviruses (enteroviruses), genome delivery is proposed to be coordinated by an intact particle inducing pore formation in the membrane through which the genome can be transferred directly without exposure to the hostile vesicle environment. In contrast, other picornaviruses (aphthoviruses e.g. ERAV, FMDV) present a dilemma by appearing to simply fall apart in acidified vesicles. Here we show that acid treatment results in the formation of an intact but transient aphthovirus empty particle from which the genome has been released. We have determined the crystal structures of the ERAV particle at native and acidic pH. The acid induced structure is consistent with a destabilized particle en-route to disassembly. We propose that the entry process for this group of viruses involves externalisation of the RNA from a novel capsid intermediate and unifies in principle the entry process for all picornaviruses.
| The Picornaviridae is a family of small non-enveloped RNA viruses, classified into several genera including Enterovirus (e.g. poliovirus, PV; human rhinovirus, HRV), Aphthovirus (e.g. foot-and-mouth disease virus) and Cardiovirus (e.g. Mengovirus). Equine rhinitis A virus (ERAV) shares physicochemical properties such as buoyant density, base composition and acid lability with foot-and-mouth-disease virus (FMDV) [1],[2]. The nucleotide sequence of ERAV also links it most closely to FMDV [3]–[5] and ERAV is now included alongside FMDVs in the aphthovirus genus of the Picornaviridae [6].
The clinical disease caused by ERAV most closely resembles the febrile respiratory tract infections attributable to rhinoviruses rather than the systemic disease observed in entero- and cardiovirus infections [4], [7]–[9]. However, broad host cell range, viraemia and persistent infection are associated with ERAV; features not seen with rhinovirus infections but reminiscent of the consequences of FMDV infection [1]. ERAV also shares unusual features of genome structure and organization with FMDV [4]. For example, it encodes two species of L protein and possesses a 16 amino acid 2A protein. However, a further unusual feature of the FMDV genome, the three copies of 3B (VPg), is not present in ERAV. The similarities between the two viruses suggest that ERAV may be a useful model system for analyzing the biology of FMDV.
The ∼300 Å diameter non-enveloped picornavirus capsid encloses a single-stranded RNA genome coding for a poly-protein which is post-translationally cleaved by viral proteases to yield the structural and non-structural viral proteins. The mature capsid comprises 60 copies of each of four proteins: VP1-4 (with molecular weights of 25, 25, 22 and 11 kDa respectively for ERAV) [2]. VP1-3 are composed of eight-stranded β-sandwiches, with strands denoted CHEF and BIDG respectively on the two sheets, which form a pseudo T = 3 icosahedral lattice (Figure 1A) [10] with the CHEF sheet exposed on the capsid surface and BIDG to the interior. VP4, and the N-termini of VP1 and VP3, are located internally.
Crystallographic structures are available for representative viruses from several genera of the Picornaviridae [11]. These share, within a genus, on average 86% sequence identity, with VP1 being the most variable protein. Although there are underlying structural similarities between all picornaviruses, they cluster into two groups which correlate with their mechanisms of uncoating. The enterovirus capsid proteins possess several extended surface loops giving rise to a circular cleft (canyon) around the icosahedral five-fold axes, which often functions as a site for receptor attachment [12],[13]. Receptor binding frequently destabilizes the virion and triggers the uncoating process, which proceeds via altered (A) 135S particles in which VP4 is partially or completely absent and the VP1 N-terminus is externalized, ultimately producing icosahedral 80S empty particles which have expelled the genomic RNA [14]. The A particle exposes hydrophobic sequences which are thought to facilitate membrane attachment and transport of the viral RNA through the membrane into the cytoplasm. In contrast cardio- and aphthoviruses lack the circular cleft, have weaker inter-pentamer interactions and dissociate directly to pentamers at low pH with no evidence for receptor-binding induced conformational changes [15]. These distinct pathways for genome-release would appear to have profound consequences for the mechanisms by which the viral RNA is delivered to the cytoplasm to initiate infection. The enteroviruses might introduce their RNA into the cytoplasm by forming a pore in the membrane through which the genome is transferred directly from the capsid, or by disrupting the vesicle membrane prior to release of the genome within the cytoplasm. Both of these scenarios could protect the RNA from exposure to the potentially damaging environment of the lumen of the entry vesicle. In contrast it is difficult to envisage how RNA transfer can safely occur for viruses, such as the FMDVs, that appear to simply dissociate into RNA and protein subunits under the influence of reduced pH in the endosome.
To investigate how similar ERAV and FMDV are in terms of structure and to probe further the cell entry mechanism for aphthoviruses we have determined crystal structures of ERAV at two different pHs, using highly optimised methods since the yield of purified virus was very low. These structures confirm that ERAV is most closely similar to FMDV. In addition the structure of a low-pH form shows internal changes consistent with a pre-dissociation state and biochemical analyses show that ERAV dissociates in acid conditions into pentameric subunits via a transient 80S intermediate particle which has lost the genomic RNA. These results suggest that there may be more similarities in the entry mechanisms for different picornaviruses than has been thought.
FMDV dissociates in mildly acidic conditions into pentameric capsid subunits presumed critical for the release of genomic RNA from the particle during the process of cell entry. For ERAV it is known that virus infectivity is sensitive to low pH [2], but the effect of acidification on the integrity of the particle has not been reported. We therefore investigated the effects of exposing radiolabelled ERAV particles to low pH by analyzing their sedimentation in sucrose gradients (Figure 1B). At pH 7.3 or 6.5, particles sedimented at approximately 150S, the expected position for native virus. However, after exposure to pH 5.5, only a minor peak of native virus was seen and the majority of the radioactivity sedimented more slowly, consistent with the dissociation of particles into 12S pentameric capsid subunits. After exposure to pH 4.5, a similar profile was seen, but with no signal for native virus.
Radiolabelled ERAV was exposed to low pH conditions as above but with the addition of 0.1 mg/ml bovine serum albumin (BSA) and 2 mM CaCl2, conditions which stabilise uncoating intermediates of poliovirus [16]. At pH 5.5, a minor peak was detected in the expected position for a 75–80S particle (Figure 1C), potentially equivalent to the 80S empty particles produced during uncoating of poliovirus (PV) and human rhinovirus (HRV). To investigate the kinetics of formation of this particle (hereafter termed 80S), radiolabelled virus was diluted 20-fold into pH 5.5 buffer and re-adjusted to neutral pH (by addition of Tris pH 7.5 to 500 mM) after various exposure times (Figure 1D). Brief exposure (6 s) of ERAV to pH 5.5 had no effect on virus sedimentation. However, after exposure for 150 s, a mixture of 80S and dissociated pentamers was observed while after 750 s only pentamers were detectable. To confirm that the 80S particle was empty of RNA, virus preparations with 35S labelled capsids or 3H labelled RNA were mixed and exposed to pH 5.5. Sedimentation of these samples (Figure 1E) showed a minor protein peak at 80S from the capsid, while no equivalent signal was detected for the RNA, consistent with the formation of an empty particle. Together these data demonstrate that viral RNA is lost from an intact but transient ERAV empty particle, before the dissociation to pentameric subunits.
Micro-crystallization facilities and the microdiffractometer-equipped station BM14 at the ESRF allowed data collection from small crystals grown from limited supplies of virus (Table 1 and data not shown). A particle data set, and a receptor soaked data set (Fry et al, in preparation) were collected from crystals grown at pH 4.6, from approximately 11 µl of 2.7 mg/ml virus solution. With thirty-fold non-crystallographic redundancy, the electron density map of the low pH particle at 3.0 Å resolution clearly differentiated residues that differ between ERAV and FMDV A1061 [17] despite incomplete (41%) sampling of the reciprocal lattice. The model built into this map (Figure 2A) comprised residues 1–246 of VP1, 31–230 of VP2, 1–226 of VP3 and 16–36 of VP4 (residues 1–30 of VP2, 1–15 and 37–80 of VP4 were too flexible to be reliably modelled – a difference map confirmed that these segments were not truncated by the averaging envelope). N-terminal sequencing of VP2 verified that the VP0 cleavage occurs at the homologous position to FMDV with no evidence for an upstream cleavage [18]. The final R factor for the model was 28.3% with 98.1% of the residues having allowed Ramachandran angles [19].
Since the pH of crystallization liquor for this structure was 4.6 (tests confirmed that the virus solution did not significantly perturb this) we might have expected to see dissociated pentamers rather then intact particles. However, the structure was clearly of an intact particle, although the disorder of certain features on the inside of the capsid prompted us to try to solve the structure of the virus at physiological pH. Crystals grown at neutral pH only yielded poor, incomplete data to 3.5 Å (see Materials and methods) but possessed 120-fold non-crystallographic symmetry. Averaging rendered the electron density clear, delineated differences on the inside of the capsid compared to the low pH structure, and confirmed that there were no major changes elsewhere in the particle (Figure 2B).
A structure–based phylogenetic tree constructed from similarities between the biological protomers of picornaviruses (Figure 3A, [20],[21]) indicates that ERAV is most similar to FMDV.
FMDV has short surface loops, so that the virus presents a rather smooth surface (Figure 3B) with an outer radius of 152 Å. In contrast the surface profile for ERAV extends to an outer radius of 159 Å at a prominent crown around the five-fold axis built from the extended VP1 loops (Mengo virus has an outer radius of 164 Å). ERAV also possesses marked surface depressions or pits around the five-fold axis, not dissimilar to those which harbour the site of receptor attachment in Mengovirus [22] (Figure 3B). The receptor-binding and antigenic properties of ERAV will be described elsewhere (Fry et al., in preparation).
FMDV capsids, unlike those of enteroviruses, contain pores penetrable by large molecules and ions such as proflavine and Cs+ [23]. The ERAV structure reveals the presence of capsid pores similar to those seen in FMDV, at the 5-fold and 3-fold axes of symmetry (Figure 3C). To confirm the porosity of ERAV, ERAV or PV was mixed at 25°C with a dye (ribogreen) which fluoresces upon binding to RNA (Figure 3D). With ERAV, a strong increase in fluorescence indicated penetration of the capsid by the dye; however with PV no increase in signal was detected, consistent with the PV capsid being impenetrable by the dye. After thermal uncoating at 60°C, the signal from PV became equivalent to that of ERAV. Properties of the dye such as molecular weight and hydrodynamic radius were withheld by the manufacturer.
The overall structure of the low pH particle is very similar to the native described above (for the protomeric unit 674 residues can be superimposed with rmsd 0.9 Å) i.e. there is no expansion of the low pH particle. There are however significant rearrangements, and an overall loss of order, in several internal loops. Thus the entire first 31 residues of VP2 cannot be clearly defined, including the hairpin structure which is generally well conserved and is a crucial element in stabilizing the pentamer interface. The VP1 N-terminus rearranges to form a loop close to the pentamer interface underlying VP3 and adjacent to the site of residues 78–88 of VP4 in FMDV (Figures 2 and 5). It thus occupies the position that would otherwise be occupied by the VP2 hairpin from an adjacent pentamer and replaces some of the stabilizing interactions at the pentamer interface, however overall the inter-pentamer interface between VP2 and VP3 is weakened. In contrast there do not appear to be any significant rearrangements on the capsid exterior surface, only slight deviations in loop conformations and side-chain orientations. The similarity between the two structures extends to the calcium ions bound on the icosahedral 3-fold axes (liganded by Asp 195 and Thr 194 of VP3) in both the low pH and native particles.
Allowing the electron density to float to different levels on the inside and outside of the capsid during cyclic averaging of the electron density map yielded essentially no difference between the inside and outside (this result was consistent when the analysis was done over the full resolution range and when restricted to data of 6 Å or lower resolution), whereas similar calculations for native FMDV particles showed a clear distinction with higher levels in the RNA rich interior than the solvent [29], suggesting that the low pH particle is empty. Unfortunately, since we were unable to measure data below 20 Å resolution, the reliability of this calculation is limited. However, empty particles which sedimented at 80S were detected following sucrose gradient centrifugation analysis of radiolabelled virus exposed to the low pH crystal buffer conditions (Figure S1).
Structure based comparisons of the ERAV capsid with other picornaviruses (Figures 3 and 4) [20] reveal, as expected, a strong similarity to members of the aphtho and cardioviruses. For VP2 and VP3 it appears closest to the aphthoviruses but VP1 is intermediate between the two genera. Overall the structure supports the classification (based on sequence homology) of ERAV as an aphthovirus and the notion that it may prove a useful surrogate for studying FMDV biology without the need for high security bio-containment facilities.
Picornavirus capsids must provide a mechanism for the viral RNA genome to be safely transported from cell to cell. As stated in the introduction there are suggestions for how this may occur for enteroviruses, however, there is no such proposal for the cardio and aphthoviruses, which are thought to uncoat directly to pentamers.
The seemingly distinct pathways for uncoating probably reflect the different architecture in the extended β-sheet which spans the pentamer interface in all picornaviruses. In aphtho and cardioviruses this comprises the VP3 CHEF strands from one pentamer and the VP2 N-terminal hairpin of the adjacent pentamer. In enteroviruses the extended VP1 N-termini contribute a further strand (residues 36–38 in poliovirus) so that strands from one pentamer sandwich the VP2 from the adjacent pentamer to stabilize the capsid. In FMDVs the protonation of a cluster of histidine residues (pI = 7.8) at the interface acts as a switch to dissociate the pentamers below pH 7, whilst ERAV, which in the native structure follows the canonical FMDV interface structure, has fewer histidines, consistent with the somewhat lower pH at which it dissociates into pentamers (between pH 6.5 and 5.5). In contrast the low pH ERAV structure shows weakened interactions at the pentamer interface, with no ordered electron density for the VP2 N-terminal hairpin, which is partly replaced by the N-terminus of VP1 from an adjacent pentamer which contributes a few stabilizing interactions (Figure 5B), consistent with a particle en-route to dissociation. This structure likely corresponds to the transient empty particle seen in the biochemical analyses but both experiments are suggestive of disassembly intermediates not previously reported. It is possible that the high particle concentration necessary for crystallization may prevent dissociation and/or drive the equilibrium towards the empty capsid intermediate.
Enterovirus uncoating intermediates include the ‘altered’ 135S state [30],[31] and the 80S form devoid of RNA. Cryo-electron microscopy [31]–[33] of these particles revealed significant global alterations before and after genome release: externalization of myristoyl-VP4 and N-terminus of VP1, expansion of the particle, iris-like movement, opening of pores at 5-fold axes and movement of the VP2 ‘plug’. These structural transitions have led to models for mechanisms of membrane interaction, genome release and membrane penetration. In contrast the low pH ERAV structure remains relatively unaltered from the native capsid; the changes seen being restricted to internal features and reducing particle stability. No expansion of the particle is observed, there is no change in the diameter of the pores at the icosahedral 3- and 5-fold axes or in Ca2+ coordination on the 3-fold axes. The N-terminus of VP1 does not appear to be externalized. Interestingly the VP2 N-terminal hairpin is rich in hydrophobic residues and becomes disordered and may potentially be externalized and involved in membrane interactions.
In HRV2 and HRV14, pores at the 5-fold axes were seen to expand in the empty structures [32],[33]. However, for PV, pores large enough to allow passage of the RNA are not present in any of the structures and additional transient capsid forms have been proposed to explain genome release [34]. Similarly, there are no alterations in the ERAV pores between the low-pH and native structures to support their involvement in the exit of the genome. Additional RNA permeable intermediate structures must therefore also exist for ERAV. The formation of such structures may be facilitated by the reduced stability of the low pH particle.
The structure of a low pH form of the acid-dissociable Mengovirus [35] shows very few conformational changes compared to the native structure, with alterations confined to the receptor binding ‘pit’ consistent with a (pH dependent) loss of receptor binding prior to direct involvement with the membrane. This has been reported for HRV2 [36] and PV [16] and may be a generic feature of picornavirus cell entry.
In enteroviruses VP4 appears to play an important role in breaching the membrane. However, the low pH ERAV particle we have captured still contains ordered portions of VP4, although our results would be consistent with a number of copies of VP4 being lost from the particle. In this context we note that not all copies of VP4 are necessarily ejected from the 135S intermediate or empty particles of enteroviruses [33], [37]–[40].
The final stage of picornavirus assembly is the cleavage of the precursor protein VP0 into VP2 and VP4. This is thought to establish a metastable state, priming the particle to initiate the entry process when receptor interactions and/or reduced pH trigger the conformational transition to a lower energy state. In poliovirus, interactions of the 44–56 loop of VP1 with the inner surface of VP2 and VP3 contribute significantly to the stability of the mature capsid and seem likely to have an important role in regulating structural transitions and cell entry. In a poliovirus immature empty capsid structure [41], where the cleavage of VP0 has not occurred, VP0 residues near the cleavage site prevent the N-terminus of VP1 from accessing its position in the mature particle. These final structural rearrangements to form the mature capsid involve similar structures to those externalized reversibly when the virus ‘breathes’ [42] and irreversibly in receptor-mediated conformational rearrangements early in the entry process [43]. The changes we see in ERAV correlate strongly with this (the residues rearranged in VP1 correspond to residues 44–56 in poliovirus), suggesting that cleavage and reorganization to prime the virus for the conformational changes required for cell entry [33],[34] is a general principle in all picornaviruses.
Our results suggest, surprisingly, that there are several common features between the uncoating process in aphthoviruses and the enteroviruses, including the existence of a quasi-stable 80S empty ERAV particle, produced upon acidification, which precedes disassociation into pentameric fragments. Furthermore, conditions have been previously reported which induce the formation of intact, empty FMDV particles missing both RNA and VP4 [44]. We infer from this that the ability to eject the genome while maintaining icosahedral integrity is a feature common to both aphtho and enteroviruses and suggest that there may be a general mechanism by which all picornaviruses protect their genome within intact capsids until the moment the genome is safely transported into the cytoplasm. Future research will more fully characterize these particles and the methods by which RNA is ejected.
ERAV was provided by Dr Janet Daley (Animal Health Trust, Newmarket, UK). Sequencing confirmed the virus strain as NM11/67 (Genbank accession number FJ607143). Ohio HeLa cells were infected at low MOI and maintained at 37°C in DMEM (BioWhittaker) containing 5% FCS (GIBCO) and standard concentrations of glutamine and antibiotics. At complete CPE, cultures were freeze-thawed and pelleted by low speed centrifugation. Supernatants contained peak virus titres of ∼5×107 pfu/ml (similar titres were obtained in Vero or RK-13 cells). The supernatant was precipitated by the addition of an equal volume of cold, saturated ammonium sulphate solution with mixing for 1 hour at 4°C. Precipitated material was pelleted at 10,000 g and 4°C for 30 minutes and resuspended in PBS, pH 7.4. Virus was purified by sedimentation through 15–45% sucrose gradients (w/v in PBS) by centrifugation at 111,000 g (average RCF, Sorvall AH629 at 29,000 rpm) and 4°C for 2.5 hours. The virus was located by measuring the 260 nm absorbance of gradient fractions. Peak fractions were pooled and precipitated with ammonium sulphate as before. Virus was re-purified by sedimentation through 15–30% sucrose gradients (w/v in PBS) at 237,000 g (average RCF, Sorvall AH650 at 50,000 rpm) and 4°C for 40 minutes. Peak fractions were pooled and samples concentrated and buffer exchanged for 50 mM Hepes pH 7.3, 50 mM NaCl, using a centrifugal concentrator (Vivascience). Poliovirus Mahoney strain (PV1) was propagated in HeLa S suspension culture and purified as previously described [16].
Viruses with [35S]methionine and [35S]cysteine labelled capsid or [5,6-3H]uridine labelled RNA were generated by metabolic labelling of infected cells and purified as described above. Radiolabelled viruses were exposed at ∼20°C to differing pH values by diluting at least 20-fold into solutions containing 50 mM NaCl and 50 mM of either HEPES pH 7.3, MES pH 6.5, sodium citrate pH 5.5, or sodium citrate pH 4.5 (other conditions as mentioned in the text). Samples were layered onto 5 ml 10–30% sucrose gradients and centrifuged as described. Radioactive material was located by liquid scintillation counting (Packard Tri-Carb) and 3H and 35S signals distinguished by their energy spectra.
Purified virus and ribogreen reagent (Invitrogen) were combined at room temperature at final concentrations of 10 µg/ml and 1 in 2000 dilution respectively and fluorescence (485/520 nm) measured at ∼25°C, using a BMG Optima plate reader. Thermal uncoating was performed by incubating samples at 60°C for ten minutes and cooling to 25°C before measuring the fluorescence.
|
10.1371/journal.pcbi.1000451 | A Three Species Model to Simulate Application of Hyperbaric Oxygen Therapy to Chronic Wounds | Chronic wounds are a significant socioeconomic problem for governments worldwide. Approximately 15% of people who suffer from diabetes will experience a lower-limb ulcer at some stage of their lives, and 24% of these wounds will ultimately result in amputation of the lower limb. Hyperbaric Oxygen Therapy (HBOT) has been shown to aid the healing of chronic wounds; however, the causal reasons for the improved healing remain unclear and hence current HBOT protocols remain empirical. Here we develop a three-species mathematical model of wound healing that is used to simulate the application of hyperbaric oxygen therapy in the treatment of wounds. Based on our modelling, we predict that intermittent HBOT will assist chronic wound healing while normobaric oxygen is ineffective in treating such wounds. Furthermore, treatment should continue until healing is complete, and HBOT will not stimulate healing under all circumstances, leading us to conclude that finding the right protocol for an individual patient is crucial if HBOT is to be effective. We provide constraints that depend on the model parameters for the range of HBOT protocols that will stimulate healing. More specifically, we predict that patients with a poor arterial supply of oxygen, high consumption of oxygen by the wound tissue, chronically hypoxic wounds, and/or a dysfunctional endothelial cell response to oxygen are at risk of nonresponsiveness to HBOT. The work of this paper can, in some way, highlight which patients are most likely to respond well to HBOT (for example, those with a good arterial supply), and thus has the potential to assist in improving both the success rate and hence the cost-effectiveness of this therapy.
| In the time it takes you to read this paragraph, one person will have undergone a lower limb amputation due to diabetic foot disease. With the global diabetes population on the rise and set to reach 330 million by 2025, the need for research into therapies and technologies that have the potential to prevent amputation is dire. There is much debate about the best way to treat these wounds, and one treatment that is shrouded with controversy is Hyperbaric Oxygen Therapy (HBOT). There are currently no conclusive data showing that HBOT can assist chronic wound healing, but there has been some clinical success. In light of how expensive properly designed clinical trials can be, we must turn to alternative methods of assessment, such as the theoretical model presented here. The mathematical model reproduces a number of clinical observations. A key result is that while HBOT can assist chronic diabetic wounds, it holds little benefit for wounds that would heal of their own accord. This model represents a useful tool to analyse the optimal protocol, and the results and insights gained from the model may be used to improve both the success rate and thus the cost-effectiveness of this therapy.
| Chronic leg ulceration is a significant socioeconomic problem [1]. Those who suffer from leg ulcers experience considerable pain, immobility and decreased quality of life [2]. Approximately 3% of the over 60 age group suffer from lower limb ulceration [3].
A successfully healing wound (or an “acute” wound) is typically thought to progress through four stages; haemostasis, inflammation, proliferation and remodelling [4],[5], although these processes are interconnected and overlapping. Haemostasis should last a matter of hours during which time the blood flow is stopped. Inflammation sees the production of chemoattractants that stimulate fibroblasts, the dominant cell in the proliferative stage of healing, to migrate into the wound site and to produce collagen, the main component of the extracellular matrix (ECM). The cocktail of chemoattractants also stimulate the systematic rearrangement of endothelial cells (ECs) from neighbouring blood vessels [6]. Capillary sprout extension is facilitated by EC proliferation and further migration toward the chemical attractant. The joining of two capillary sprouts within a healing wound forms a loop through which blood can flow and new sprouts develop from this vessel thus propagating angiogenesis [7].
A chronic wound is one in which healing fails to proceed through an orderly and timely process to produce anatomic and functional integrity, or proceeds through the repair process without establishing a sustained anatomic and functional result [8]. The factors responsible for the development of a chronic wound remain unclear, however the most common cause, according to Mathieu [9], is thought to be related to the detrimental effects of prolonged wound hypoxia (oxygen deficiency). Enoch et al. reports that chronic wounds can be arrested in any one of the stages of wound healing, but disruption commonly occurs in the inflammatory or proliferative phases [4].
HBOT involves the intermittent exposure of the body to 100% oxygen at a pressure greater than 1 atmosphere (atm) and its use is supported in the treatment of problem wounds [10]. However, there is much debate about the optimal HBOT protocol in treating such wounds [11]–[14]. Although HBOT is typically used as an adjunctive therapy for treating chronic wounds, many clinicians lack a full knowledge of the evidence-based data that support its use [15].
The primary rationale behind the use of HBOT in the treatment of chronic wounds is to elevate the amount of oxygen delivered to the wound site [16]. For a more detailed review of the wound healing process, the different etiologies of chronic wounds and the use of HBOT to treat nonhealing wounds see Thackham et al. [5]. It should be noted that HBOT is not the only wound healing therapy currently being studied. Gordillo et al. review the use of topical oxygen therapy to assist the closure of chronic wounds [17].
While the wound healing process is undeniably complex, there are useful mathematical models that address various aspects of the phenomenon. The models can be categorised into four broad groups: continuum reaction-diffusion models, mechanochemical models, discrete/stochastic models and multiscale models.
Discrete models have the ability to contain a level of detail that is not possible from a continuum model and, in general, allow for quicker numerical simulation however, continuum models allow mathematical analysis that discrete models do not.
Continuum reaction-diffusion models are arguably the most commonly used theoretical approach for studying the angiogenesis process. One of the first models of this kind is due to Balding and McElwain, who developed a model to investigate tumour-induced angiogenesis [18]. Their model consisted of distinct equations for the blood vessel and capillary tip density. Edelstein had previously used a similar approach to model fungal growth [19]. The concept behind including both blood vessel and tip species is that the ECs in the tip of a vessel guide the ECs in the sprout. This aspect of the model, the so-called “snail-trail” production of blood vessels, means that if the capillary tip density, , moves with a velocity , then the rate of increase (that is, production/extension) of blood vessels is given by , where is a unit vector in the direction of . The model by Balding and McElwain accounts for branching and anastomosis but does not account for the extension of vascular loops.
In 1996, Pettet and coworkers developed two models of wound healing angiogenesis [20],[21]. Sherratt, in 2002, described the work in [21] as “the most important theoretical work on wound angiogenesis to date”. In [20] Pettet et al. proposed a three species simplification of the more detailed six species model of wound healing angiogenesis presented in [21] and used an analytic approach to obtain an approximate solution. The first and foremost feature of these models is that they incorporate the dependence of chemoattractant production on the local wound oxygen concentration, with chemoattractant production occurring in a specific oxygen concentration range. Both of these publications modelled the extension of blood vessels using the Balding and McElwain “snail-trail” concept. Other authors have chosen this approach to model wound and tumour angiogenesis (see for example [7],[22],[23]).
In 2002, Gaffney et al. used a two-species model to investigate cutaneous wound healing [24]. Here a travelling wave analysis was used to identify a lower bound on the wave speed of the wound healing unit in terms of two key model parameters, namely, the random motility of capillary tips and the rate of budding of tips. Importantly, Gaffney et al. chose not to use a snail-trail approach, opting rather to consider the EC density explicitly in addition to the capillary tip density [24]. The flux of the capillary tips is determined by random motion and directed motion. The flux of the EC is then assumed to be proportional to the tip flux, where the rate constant is the number of EC that makes up an average capillary tip. More recently, Schugart et al. developed a seven species model of acute wound healing angiogenesis [25], using a similar approach to Gaffney et al. [24] in their treatment of the EC flux, while Addison-Smith et al. used a simple mechanistic model for the sprouting of vessels during tumour-induced angiogenesis [26]. Mantzaris et al. provides an excellent review of continuum models of angiogenesis and concludes that continuum models are important for providing significant insight into the relative importance of different processes [27].
The interactions between cells and the substratum in wound healing are not just chemical; there is also a mechanical influence [27]. For instance, in order to migrate, ECs extend lamellipodia in the direction of migration, and exert tractional forces on the ECM. Mechanochemical models of wound healing are essentially continuum models that account for the forces that cells exert on the ECM. Mechanochemical models of wound healing include the early work by Tranquillo and Murray [28]–[30] and the extensions by Olsen et al. and Cook [31],[32]. These models consider the connection of cells to the ECM and are thus relevant for deeper, dermal wounds. They are typically applied to acute (normal) healing wounds that heal primarily by contraction.
Mechanochemical modelling is needed when considering the remodelling phase of wound healing since it is during this stage that the wound contracts. However, since we are interested in chronic wound healing, and chronic wounds typically arise due to complications in the inflammatory or proliferative phase, the use of mechanochemical models is not addressed here. Furthermore, healing in human wounds is predominantly due to proliferation and migration of cells from outside the wound, whereas in animal models, the contraction of the wound by mechanical forces is thought to be more substantial. It is interesting to note that while the critical role of oxygen in wound healing is well known there are no mechanochemical models, to date, that incorporate angiogenesis.
Discrete models, using cellular automata for example, have been used to capture key features of angiogenesis including the outgrowth, branching and anastomosis of vessels. Stokes and Lauffenburger pioneered the discrete modelling of blood vessel formation in their series of publications [33]–[35]. Their work is discrete in that they present a stochastic model for the random motility and chemotaxis of individual cells. Other aspects of their model use continuum modelling, for example, in the conservation equation for the chemoattractant. Levine and coworkers extended this approach with a series of publications [36]–[40] in which they investigated the mathematical modelling of tumour-induced angiogenesis. In these models, continuous limits of reinforced random walk equations govern the angiogenesis process while ordinary differential equations (ODEs) model the biochemical kinetic equations. In their review, Mantzaris et al. states that one main advantage of using discrete models rather than continuum ones is that individual cells and sprouting of vessels can be tracked [27]. Although discrete models can be computationally fast and efficient, and provide quantitative numerical data, these models are not as readily amenable to mathematical analysis as continuum models are.
Multiscale techniques have been used to simulate the wound healing process (see for example Dallon et al. and Cai et al. [41],[42]). Sun et al. have developed several models of angiogenesis including a multiscale model where the concentration of chemoattractant is modelled at the tissue scale, while the capillary network is modelled at the cellular scale [43],[44]. More recently, Alarcon and coworkers and McDougall and coworkers have used multiscale techniques to investigate angiogenesis associated with tumour growth [45]–[48].
The overall aim of this paper is to use a theoretical model to evaluate the use of hyperbaric oxygen therapy as an adjunct therapy (a therapy used to assist a primary treatment) for treating chronic wounds. Through numerical simulations we conclude that intermittent hyperbaric oxygen therapy has the potential to aid in the healing of chronic wounds (a chronic wound is one which does not heal in an orderly set of stages and in a reasonable amount of time in the way that most wounds do).
We use the model presented in the Materials and Methods section to simulate a number of different wound healing scenarios, each of which is discussed below.
We simulate an acute healing wound with the choice of parameters outlined in Table 1, noting that this choice of values yields a steady state oxygen concentration behind the wave front (behind the healing front the oxygen concentration, , tends to as ) above the lower threshold for capillary tip production, , so that healing will be initiated. Fig 1 shows such a normal situation in which a wound of length 2 cm (that is, ) is almost completely reoxygenated within 2 weeks. It would take longer than this, roughly 2.5 weeks, for the simulated wound to completely revascularise. We note that the vessel density can rise above the carrying capacity, , due to rapid chemotaxis and may remain elevated until the remodelling process drives the density to return to normal levels.
A chronic wound is simulated by selecting parameter values such that the oxygen concentration behind the injured tissue (near the oxygen concentration tends to ) does not rise above the lower threshold for capillary tip production, . As mentioned above, our assumption is that chronicity is associated with a reduced or impeded supply of oxygen from the vasculature. We therefore reduce the value of used for the simulation in Fig 1 by a factor of 10 to . The resulting simulation produces an oxygen profile that is always within the range (that is, below the lower threshold for capillary tip production) inside the wound. Fig 2 shows the chronic wound simulation. We note no significant change over time, indicating that no healing is taking place.10.1371/journal.pcbi.1000451.g002Figure 2
Simulation of a chronic wound in which no healing occurs.
Multiple day intervals are shown (dark blue = 2, red = 4, green = 6, black = 8, yellow = 10, light blue = 12, pink = 14). Parameter values: as per Fig 1, except .
We now investigate the impact of HBOT on the healing of a chronic wound. The strength and duration of HBOT are given by the parameters and in the model, respectively. The parameter is a measure of the relative increase in supply of oxygen during HBOT compared to times of no treatment. Fig 3 shows such a chronic wound situation under HBOT with for hours per day (that is, of a day). A value of is associated with 100% oxygen at a pressure of just under 3atm (see Materials and Methods Section), which is a reasonable treatment protocol. We note from the simulation that the capillary tip density in the chronic wound reaches highly elevated levels under treament and that healing is quickly initiated in the chronic wound.
Table 2 shows that an value of 5.73 equates to 100% oxygen at 1atm (that is, normobaric oxygen therapy). The analysis presented later (see Expression (22) in “Analysis of Feasible HBOT Protocol”) predicts that a chronic wound simulated with will not heal and this is confirmed by numerical simulations. Thus we have shown, under the assumptions on the model presented here, that normobaric oxygen will not stimulate healing of a chronic wound and we have answered the somewhat controversial question of whether or not normobaric oxygen can be used to substitute for HBOT in the treatment of chronic wounds. Normobaric oxygen fails to stimulate healing in the chronic wound since the oxygen levels under the treatment are still insufficent to initiate capillary tip production. Similarly, our numerical simulations reveal that values of in excess of about 2000 are too high to enable healing to occur. This is because the oxygen levels are raised so much under the treatment that capillary tip production is switched off when the upper oxygen threshold, , is reached and surpassed. Such high values of are not considered physically feasible (see Table 2).
The results presented in Fig 4 reveal what happens when we simulate a situation in which HBOT is halted prematurely (after 5 days). Interestingly, the effects of HBOT seemed to persist for some time after treatment is halted, but the healing progress slows considerably (compare Figs 3 and 4). Thus, if we want the wound to close as quickly as possible, then HBOT should not be terminated until complete healing of the wound is observed. Note that this is in disagreement with typical clinical protocols, which is to apply the therapy daily for about 6 weeks [16]. This restriction is likely based on cost considerations rather than clinical or experimental evidence which indicates that this is more effective in stimulating healing than continuing until the wound is completely healed.
Many hyperbaric centers around the world advocate the use of HBOT to treat ‘normal’ wounds on the basis that HBOT may accelerate healing in sports injuries [49]. This use of HBOT is highly controversial [50]. Typically a sports injury is internal (muscular) rather than dermal, but here we consider the effect of applying HBOT to a normal healing wound. Comparing Figs 1 and 5, we see that there is little benefit (at most, a 10% increase in the rate at which blood vessels are progressing through the wound space) in applying HBOT to a wound that progresses through the healing process of its own accord. Furthermore, the relatively high-cost of HBOT further detracts from the appeal of its use to treat such wounds. This simulation also shows that the capillary tip density falls significantly towards the end of the healing process. Numerical experimentation reveals that healing will occur, even with very small levels of capillary tips, suggesting that it is the presence of capillary tips, rather than their quantity, that is important for initiating healing. Clinically, this means that stimulating capillary tip production is the crucial factor that enables a chronic wound to heal when HBOT is applied.
We note from Fig 1 and 5 that the density of capillary tips at comparable times is lower in the treated wound than the untreated one. This result can be explained as follows. The normal wound develops a vasculature which supplies sufficient oxygen to the wound to initiate angiogenesis. The wound does not require HBOT to heal. The application of HBOT to this wound increases oxygen levels and reduces the net tip production during times of treatment, resulting in a decreased capillary tip density. In the preceding paragraph, we deemed the therapy to have a positive effect on healing however, this was based on the faster progression of blood vessels through the wound site under the treatment.
We now discuss the potential clinical implications of the restrictions derived for the HBOT protocol, the mathematical detail of which is shown in “Analysis of Feasible HBOT Protocol” in “Materials and Methods”. By considering the feasible HBOT protocol (that is, the range of values, where is the relative increase in supply of oxygen during HBOT) to be those that
where and are the lower and upper oxygen concentration thresholds, respectively, for capillary tip production to take place, we are able to derive constraints that depend on key parameter values from the model for the range of feasible HBOT regimes. We note that if a lower bound is too high then a patient would need to be exposed to levels of pressure that are not safe in order for healing to be observed. For instance, if a particular set of wound parameter values lead to a lower bound that exceeds twenty-one, then we must conclude that HBOT will not assist this patient, since only is clinically reasonable (see Table 2).
We consider two approaches to estimating the feasible protocol region. The first approach is to assume that the kinetics dominate the evolution of the oxygen concentration within the wound space and the second is to assume that the blood vessels do not migrate into the wound significantly over the first 24 hours of healing and to solve the resulting partial differential equation (PDE) for the oxygen concentration using Green's functions. We will not consider the upper bound from either approach since both failed to yield clinically relevant restrictions. Instead we focus on the two lower bounds, namely:(1a)and(1b)where and and are the lower bounds for the first and second approach, respectively, described above.
We use Eqs (1a) and (1b) to identify patients who are unlikely to benefit from HBOT: these individuals will have higher values of and . Examining Eqs (1a) and (1b) reveals that patients with the following characteristics are unlikely to benefit from HBOT:
In this paper we have developed a simple mathematical model that simulates the healing of both acute and chronic wounds. The modelling framework is based on the premise that chronic wound healing is associated with poor and/or impeded oxygen delivery from the vasculature. We are now in a position to extend the model to investigate other hypotheses. For example, chronic wound healing may be associated with impaired cellular function such as poor cell chemotactic responsiveness [55]. Alternatively, certain chronic wounds may be extremely hypoxic because there is a high bacterial load in the wound bed [56]. Our model can easily be adapted to study this situation by increasing , the rate of removal/consumption of oxygen in the equation governing the oxygen distribution, Eq (2a).
The development of new blood vessels occurs by two processes, namely, angiogenesis and vasculogenesis. Here we have only modeled the effect of oxygen on angiogenesis. The extension of the model to include the vasulogenesis and its potential as a mechanism for the improved healing associated with HBOT is a further extension of the model. There has already been some work on the role of vasculogenesis in tumour growth (see for example Stamper et al. [57]).
We used our model to evaluate the effect of treating chronic wounds with HBOT. In summary, our simulations have allowed us to make several, clinically-relevant conclusions including the following:
By considering the feasible HBOT protocol (that is, the range of values, where is the relative increase in supply of oxygen during HBOT) to be those that
where and are the lower and upper oxygen concentration thresholds, respectively, for capillary tip production to take place, we were able to derive constraints that depend on key parameter values from the model for the range of feasible HBOT regimes. By considering patients that will need excessive exposure to pressure in order to stimulate healing in conjunction with the lower bounds on the parameter, we predict that patients with any of the following conditions are unlikely to respond well to HBOT:
In conclusion, we have used a simple three species model of wound healing to evaluate the effect of treating chronic wounds with HBOT. While the causal reasons for the improved healing remain unclear, protocols will remain empirical and an unreliable screening process for appropriate patients will remain in place. The work of this paper is a first step towards identifying in a systematic manner patients who are likely to respond well to HBOT and thus has the potential to assist in improving both the success rate and the cost-effectiveness of this therapy.
We adopt a continuum approach to modelling angiogenesis in wound healing, similar to that of Balding and McElwain [18]. We focus on three key species: oxygen: , capillary tips: and blood vessels: . For simplicity, we consider the wound to be one-dimensional, of total length and symmetric about its centre. Here represents the wound centre while denotes the edge of the wound. Equations governing the evolution of , and are presented in turn below.
Oxygen concentration, :(2a)That diffusion is the primary mechanism by which oxygen is transported through an unvascularised wound site is well established [59] and here we assume a linear diffusion equation for oxygen, with assumed constant diffusivity . Clearly the formation of new blood vessels to replace the damaged system will increase the supply of oxygen (and other nutrients) to the wound site [51]. We assume that oxygen is supplied by new vessels at rate . We assume that oxygen is removed via the vasculature at rate . Molecular oxygen is also assumed to decay naturally. HBOT is known to substantially increase oxygen levels within the wound [5],[60] and we thus model the application of HBOT to a wound as an increase in supply during the time of treatment. That is, we take during the session and 0 at all other times.
Capillary Tip density, :(2b)where is the Heaviside function. We assume that the dominant mechanism for capillary tip movement is migration down the oxygen gradient (with coefficient ). That is, the tips are attracted to regions of relative hypoxia, which is consistent with the experimental literature [61]. We note that, typically chemotaxis is used to model attraction of a species towards a chemical gradient, whereas we are using it here to model the migration of capillary tips down the oxygen gradient and thus the term appears slightly differently than a typical chemotaxis term. Tip sprouting occurs from existing vessels at rate within a range of values of oxygen concentration, . This reflects the well-established concept that proliferation of the ECs that constitute the tips occurs in a region just behind the healing front [62]. Capillary tips are known to be lost due to anastomosis during the healing process [63] and here we model this as a linear process with rate constant .
The idea that the oxygen gradient within the wound tissue is responsible for driving the process of angiogenesis is well-supported in the literature [2], [11], [61], [64]–[66]. For instance, Tompach et al. states that “angiogenesis is driven by a gradient of oxygen whereby high arterial oxygen tension drives angiogenesis into hypoxic spaces” [11] while Oberringer et al. make note that “the oxygen gradient serves as a key stimulus for the direction of cell migration” [61]. While we acknowledge that a gradient of chemoattactant (such as vascular endothelial growth factor (VEGF)) may also be present in the wound, we justify modelling the oxygen concentration as a chemical stimulus without including VEGF by referring to the statement by Bao et al. “a gradient of VEGF expression is established that parallels the hypoxic gradient, and ECs subsequently migrate toward the most hypoxic areas” [67].
There is evidence that oxygen regulates the production of chemoattractant by macrophages and that it is the stimulus provided by the chemoattractants that stimulate angiogenesis [68]–[71]. Pettet et al. avoid the inclusion of a macrophage species in their model by assuming that within a window of oxygen , chemoattractants are produced, which in turn stimulate new blood vessel production. Here we avoid the need for explicitly including the chemoattractant by assuming that capillary tips are produced only within the oxygen interval . This is consistent with current literature suggesting that in both chronically hypoxic wounds [72] and those that have high levels of oxygen [68], angiogenesis is halted.
Blood vessel density, :(2c)
We apply the widely used Balding-McElwain “snail trail” approach to modelling the laying down of blood vessels by the moving capillary tips [18],[27]. After the rapid formation of blood vessels during the inflammatory and proliferative stages of healing, the vasculature system is remodelled [73] and we assume that this behaviour can be described by a logistic term with growth rate and the normal blood vessel density, , as the “carrying capacity”.
In order to close Eqs (2a)–(2c), we impose boundary and initial conditions as follows. At the wound centre, , we impose zero flux of oxygen due to the assumed symmetry of the wound about , so that:(3a)
We assume that the wound edge is devoid of capillary tips and that the flux of oxygen there is zero so that the wound and tissue oxygen concentrations rapidly equilibrate under intermittent HBOT and we have(3b)and(3c)
We assume that initially the wound is devoid of tips, the blood vessel density is that of normal tissue within a certain area of the wound edge () and the wound is partially oxygenated throughout this vascularised region so that(3d)(3e)(3f)
For numerical simulations, the discontinuity in the initial condition of the blood vessels is “smoothed” by replacing the two sharp corners with two small quarter-circles.
We now determine parameter constraints on the HBOT protocol by considering the change in oxygen concentration over the first 24 hours of treatment. Consideration of Eqs (2a)–(2c) reveals that unless the oxygen concentration somewhere within the wound space falls in the range , then production of capillary tips is not possible and healing will not occur. We thus define “feasible” values to be those that
Current clinical protocol is to administer 1.5 hours of treatment once per day. Hence the above two conditions can be expressed mathematically as:
We consider two approaches to deriving the aforementioned constraints. The first involves assuming that the kinetics of Eq (2a) dominate the oxygen concentration within the wound space. This allows us to consider a time-dependent ODE for the oxygen concentration which is solved using standard techniques. The second approach is to assume that the blood vessels do not migrate into the wound significantly over the first 24 hours of healing. The resulting partial differential equation (PDE) for the oxygen concentration decouples from the remaining equations and can be solved using Green's functions [90].
By assuming that the kinetics dominate the evolution of oxygen within the wound space we arrive at the following ODE that governs the transition from the steady state oxygen concentration without HBOT to the steady state concentration under treatment(6)where , represents the proportion of each day that a patient is administered HBOT (that is, 90 minutes per day). From Eq (2a), the blood vessel density at steady state is . Substituting this into Eq (6) we obtain(7)with .
Eq (7) has solution(8)
Using Eq (8) it is straight forward to show that our conditions for the feasible HBOT protocol as outlined previously can be written:
These inequalities identify a region of values (that is, HBOT protocols) for treating a chronic wound such that pro-healing effects are predicted (under the given assumptions) to be observed:(9)where capillary tip production can only take place in the oxygen range of , is the rate at which oxygen is removed from the wound via the vasculature, is the characteristic blood vessel density, is the rate at which oxygen is supplied by the blood vessels and is the initial oxygen concentration at the wound edge.
It should be noted that this analysis is based on the assumption that the wound is chronic in that, if it is left untreated, then the steady state oxygen concentration will not rise above the lower threshold for capillary tip production (that is, ).
By substituting the estimated parameter values shown in Table 1 into the inequality in (9), we predict that for this particular set of parameter values, HBOT will assist healing if:Recall that the parameter is dimensionless. It represents the increase in supply of oxygen during HBOT relative to periods without the treatment. It should also be noted that the upper limit of this region of feasible values is outside what would be considered “clinically reasonable”.
Numerical experimentation reveals that a chronic wound exposed to HBOT with will not heal, which is in violation of the inequalities predicted by the above analysis. There are a number of potential reasons why our lower bound does not provide an accurate restriction on the HBOT protocol including:
To emphasize the fact that the steady state approach is inappropriate for deriving the constraints on the feasible HBOT regime, let us consider a chronic wound exposed to HBOT with . Fig 6 compares the oxygen concentration at the wound margin (that is, at ) during the first day of healing with the value predicted from the above steady state analysis. Note from Fig 6 that the concentration of oxygen at the wound edge differs substantially from the value associated with the steady state analysis.
On closer inspection Fig 3 reveals that during the first day, the blood vessels do not migrate deep into the wound. By assuming that the blood vessel distribution throughout the wound domain does not change from the initial conditions during the first day of treatment then the oxygen PDE decouples and we have:(10)where and subject to the boundary conditions . We find the solution to Eq (10) over the first day of treatment using Green's functions:where is the initial distribution of oxygen (see Eq (3d)), is given above, andEvaluating the integrals gives:(11)where , is the duration of the HBOT session on the first day of treatment and is the Heaviside function.
By imposing we find:(12)and similarly, we can ensure that if does not exceed:(13)where and . In terms of the estimated parameter values, shown in Table 1, the region of feasible predicted by using Green's function is given by:(14)
This lower bound is consistent with numerical simulations, which reveal that healing does not occur with . However, we note that the upper limit is significantly higher than what would be considered clinically reasonable.
In Fig 7 we compare the oxygen concentration at the wound margin during the first day of healing with that predicted using the Green's function analysis. We note that there is excellent agreement between the values obtained by solving the full model numerically and by using the Green's function approach.
It should be noted that in practice only the lower bounds presented here are useful since exposing a patient to high levels of oxygen for even short periods of time causes oxygen toxicity [91]. In fact, 100% oxygen for 3 hours at 3 atm can cause central nervous system breakdown [50]. Hence a value of greater than 20.93 (see Table 2) should be considered clinically irrelevant.
|
10.1371/journal.pgen.1007229 | Drosophila Gr64e mediates fatty acid sensing via the phospholipase C pathway | Animals use taste to sample and ingest essential nutrients for survival. Free fatty acids (FAs) are energy-rich nutrients that contribute to various cellular functions. Recent evidence suggests FAs are detected through the gustatory system to promote feeding. In Drosophila, phospholipase C (PLC) signaling in sweet-sensing cells is required for FA detection but other signaling molecules are unknown. Here, we show Gr64e is required for the behavioral and electrophysiological responses to FAs. GR64e and TRPA1 are interchangeable when they act downstream of PLC: TRPA1 can substitute for GR64e in FA but not glycerol sensing, and GR64e can substitute for TRPA1 in aristolochic acid but not N-methylmaleimide sensing. In contrast to its role in FA sensing, GR64e functions as a ligand-gated ion channel for glycerol detection. Our results identify a novel FA transduction molecule and reveal that Drosophila Grs can act via distinct molecular mechanisms depending on context.
| Fatty acids (FAs) are energy-rich nutrients that are detected through the gustatory system to promote feeding. Here, we show FA detection requires a Drosophila gustatory receptor, Gr64e. Although GR64e functions as a ligand-gated ion channel for glycerol detection, in FA sensing, it acts downstream of phospholipase C signaling. We identified a novel signaling molecule for FA sensing in Drosophila. Furthermore, our findings suggest Drosophila GRs have multiple modes of action depending on their cellular and molecular context.
| Animals use gustatory systems to evaluate the quality of food. Gustation is essential not only to prevent ingestion of toxic chemicals but also to ensure ingestion of essential nutrients such as sugars, amino acids, and lipids. The detection and consumption of energy-dense foods can confer a survival advantage, especially when food is scarce. Lipids are more calorie-rich than proteins or sugars, so it is unsurprising that lipid sensing has emerged as a new candidate taste modality in addition to the five basic taste modalities in mammals: sweet, umami, bitter, sour, and salt. Dietary lipid sensing was thought to be mediated by texture and olfaction [1–3], but the recently discovered taste receptors for fatty acids (FAs) in mammals indicate gustatory systems can also detect lipids [4, 5]. Two G-protein coupled receptors (GPCRs), GPR40 and GPR120, are present in the taste receptor cells of mammals [5, 6] and are partly requried for FA preference [5]. FA-induced responses depend on phospholipase C (PLC) and its downstream signaling molecules like transient receptor potential channel type M5 (TRPM5) [7], suggesting that FA taste is mediated by a phosphoinositide-based signaling pathway.
Drosophila melanogaster can detect several taste modalities including sweet, bitter, salt, and amino acids [8, 9]. Most taste modalities are detected by the direct activation of ion channels expressed in gustatory receptor neurons (GRNs). The 68 members of the gustatory receptor (Gr) gene family in the Drosophila genome include the main taste receptors for the sweet and bitter modalities [10, 11]. Although GRs have seven transmembrane domains, these proteins are not GPCRs. They have an opposite membrane topology [12, 13] and function as ligand-gated ion channels [14, 15]. Ionotropic receptors (Irs), which are distantly related to ionotropic glutamate receptors [16], are involved in the detection of low salt, pheromones, polyamines, and amino acids [17–20].
In contrast to other taste modalities, Drosophila FA taste signaling is mediated by the PLC pathway [21]. Mutation of norpA, a Drosophila orthologue of PLC, results in reduced attraction to FAs. The introduction of a norpA cDNA into sweet GRNs of norpAP24 flies rescues their deficit in FA sensing, suggesting PLC in sweet GRNs is essential for FA sensing. FA detection requires PLC signaling in sweet GRNs, but no other signaling molecules have yet been implicated. Here, we show that Gr64e, which is known as a glycerol receptor [22], is required downstream of PLC for the detection of FAs. The precise deletion of the Gr64 cluster via CRISPR/Cas9 reduces FA palatability. By screening individual Gr64 cluster gene mutant flies, we identified a requirement for Gr64e in FA sensing. We also found the re-introduction of Gr64e into Gr64 cluster deletion mutants rescues their behavioral attraction to FAs and FA-evoked action potentials. Gr64e seems to function as a ligand-gated ion channel for glycerol sensing because the co-expression of Gr64e and Gr64b confers glycerol responses independent of PLC on sweet GRNs, the low-salt sensing GRNs, and bitter GRNs of Gr64 cluster mutant flies. In contrast, the introduction of TrpA1, which can couple to PLC signaling [23, 24], in sweet GRNs of flies lacking Gr64e rescues their deficit in FA sensing but not glycerol sensing. In addition, Gr64e expression in TrpA1 mutants can only rescue their deficit in aristolochic acid (ARI) sensing [23], which is PLC-dependent. Gr64e expression does not rescue the TrpA1 mutant defect in N-methylmaleimide (NMM) sensing, which proceeds via direct TRPA1 activation [25]. Together, our results reveal a novel component in Drosophila for signal transduction in FA detection and suggest Drosophila Grs can function via multiple molecular mechanisms depending on their cellular and molecular context.
We were prompted to test whether the Gr64 cluster is involved in FA sensing because the Gr64 cluster is required for the detection of most phagostimulatory substances [26–31]. The Gr64 cluster comprises six tandem Gr genes (Gr64a-Gr64f) transcribed as a polycistronic mRNA (Fig 1A) [26, 29, 31]. Because deletion of the whole Gr64 cluster (ΔGr64) is lethal due to the additional deletion of neighboring genes [31], we used CRISPR/Cas9 to generate a new Gr64 cluster deletion (Gr64af) covering only the Gr64 cluster coding region (Fig 1A). We confirmed the deletion of the Gr64 loci by genomic PCR and DNA sequencing (Fig 1A). In contrast to ΔGr64, Gr64af is viable and fertile. As expected, we found Gr64af flies show a reduced proboscis extension reflex (PER) to sucrose, glucose, fructose, trehalose, and glycerol (Fig 1B). PER responses to low salt are slightly increased compared to wild-type (Fig 1C), suggesting Gr64af does not have a general defect in gustatory function. Furthermore, optogenetic activation of sweet GRNs expressing red activatable channelrhodopsin (ReaChR) [32] induces PER in wild-type and Gr64af flies (Fig 1D), confirming that sweet GRNs of Gr64af are functional. We, next asked whether the Gr64 cluster is required for FA sensing. Although wild-type flies show a robust PER response to hexanoic acid (HxA), octanoic acid (OcA), oleic acid (OA), and linoleic acid (LA), Gr64af flies show severely reduced PER responses to all the FAs we tested (Fig 1E). We were also able to confirm that the other sweet Grs (Gr5a, Gr43a, and Gr61a) are not required for FA sensing (Fig 1F).
To determine which of the six Grs in the Gr64 cluster are required for FA sensing, we examined PER responses to HxA in flies carrying mutations in the individual genes of the Gr64 cluster (S1 Fig). norpAP24 flies, which carry a mutation in the Drosophila orthologue of PLC [33], show reduced PER responses to HxA like Gr64af flies (Fig 2A) [21]. Of the various Gr64 cluster mutants, we found Gr64cLEXA and Gr64eLEXA flies show reduced PER responses to HxA like the norpAP24 and Gr64af mutants (Fig 2A).
To confirm the requirement of Gr64c and Gr64e for HxA sensing, we further characterized the Gr64c and Gr64e mutants. Although Gr64cLEXA flies show reduced PER responses to HxA, glycerol, and sucrose (Fig 2B), the expression of a Gr64c cDNA in Gr64cLEXA flies using Gr5a-GAL4, which labels sweet GRNs [34], does not rescue this defect. This suggests the Gr64cLEXA phenotype cannot be attributed to the loss of Gr64c in labellar sweet GRNs. This result is also consistent with the strong FA preference of ΔGr64a2 flies, which harbor a deletion of the protein-coding sequence of Gr64a and Gr64b as well as a third of the protein-coding sequence of Gr64c at its N-terminus (S1 Fig, Fig 2A). Gr64e is known as a glycerol receptor [22]. Gr64eLEXA flies show reduced PER responses to glycerol and to several FAs (i.e., HxA, OcA, OA, and LA) (Fig 2C and 2D). Expression of a Gr64e cDNA in the Gr64e mutant background using Gr5a-GAL4 rescues glycerol and FA responses to wild-type levels, indicating Gr64e is required for both glycerol and FA detection (Fig 2C and 2D). In addition, the expression of Gr64e using Gr5a-GAL4 rescues the HxA responses of Gr64af flies, suggesting Gr64e is the only Gr in the Gr64 cluster required for FA sensing (Fig 2E).
Silencing the labellar Gr64e-expressing GRNs by expression of the potassium channel Kir2.1 [35] abolishes PER to HxA, suggesting that preference to HxA is mediated by Gr64e-expressing GRNs (S2 Fig). To better understand FA sensing in the labellum, we examined electrophysiological responses to HxA. HxA elicits action potentials mainly in S-type sensilla of wild-type flies (Fig 3A). In a few cases, we also observed HxA-evoked firing in I-type sensilla, but such responses were rare. Consistent with our PER results, we did not observe any responses to HxA in Gr64af or Gr64eLEXA flies (Fig 3B and 3C). Gr64cLEXA flies show robust, wild-type-like HxA responses, indicating that the reduced attraction of Gr64cLEXA flies to HxA cannot be attributed to a peripheral defect in FA detection (Fig 3B and 3C). In addition, Gr5a-GAL4-driven expression of Gr64e in Gr64eLEXA and Gr64af flies restores HxA-evoked action potentials, which suggests Gr64e is the only Gr in the Gr64 cluster required for FA sensing (Fig 3D and 3E).
Gr64e is required in GRNs for electrophysiological and behavioral responses to glycerol [22]. To determine whether the molecular function of Gr64e is the same in the detection of glycerol and FAs, we next asked whether PLC is required for glycerol sensing. We found no difference between wild-type and norpAP24 flies in glycerol-evoked action potentials or PER responses (Fig 4A–4C). This indicates Gr64e plays distinct molecular roles in the detection of glycerol and FAs.
It remains unclear whether Gr64e alone is sufficient for glycerol detection. Ectopic expression of Gr64e in olfactory receptor neurons confers glycerol responses [27], but Gr64e requires Gr64b as a co-receptor to confer glycerol responses on sweet GRNs [36]. To address this ambiguity, we used Gr5a-GAL4 or Ir76b-GAL4, which labels low-salt sensing GRNs [20], to misexpress Gr64b alone, Gr64e alone, or Gr64b and Gr64e together in sweet GRNs or low-salt sensing GRNs of Gr64af flies, respectively. The misexpression of Gr64b and Gr64e together confers glycerol sensitivity in both sweet GRNs and low-salt sensing GRNs of Gr64af flies (Fig 4D–4G). Co-expression of Gr64b and Gr64e together in sweet GRNs of Gr64af flies restores their PER responses to glycerol (Fig 4H). In addition, introduction of Gr64b and Gr64e in bitter GRNs of Gr64af flies under the control of Gr66a-GAL4, which labels bitter GRNs [34], confers glycerol response (S3 Fig). These data suggest glycerol detection occurs through the direct activation of heteromeric ion channels formed by Gr64b and Gr64e.
Although both Gr64e and PLC are required for FA detection in sweet GRNs, it is unclear how they function together. It is possible that Gr64e acts as a GPCR that detects HxA and functions upstream of PLC. This is unlikely, however, because sweet GRNs of L-type sensilla expressing Gr64e do not respond to HxA. To exclude the possibility that sweet GRNs of L-type sensilla lack other factors required for PLC signaling, we used Gr5a-GAL4 to express either Gαq/norpA or Gr64e/Gαq/norpA in sweet GRNs. Neither of these combinations confers HxA responsiveness on the sweet GRNs of L-type sensilla (S4 Fig). A second hypothesis relating the function of Gr64e to PLC is that Gr64e functions downstream of PLC. Drosophila trpA1 is expressed in a subset of bitter GRNs and required for avoidance to NMM [25], a tissue damaging reactive electrophile and ARI [23], a plant drived antifeedant. TRPA1 can be activated directly by NMM[25] and has also been associated with PLC signaling in ARI avoidance [23]. We hypothesize that if both TRPA1 and GR64e function downstream of PLC, TRPA1 and GR64e should be able to substitute for one another with regard to PLC signaling. We misexpressed either the thermosensory isoform TrpA1(B) or the chemosensory isoform TrpA1(A) in sweet GRNs of Gr64af flies to explore whether TRPA1 can replace the function of GR64e in FA sensing but not glycerol detection. We found TrpA1 expression in sweet GRNs of Gr64af flies rescues HxA-evoked electrophysiological responses in their S-type sensilla and their HxA-evoked PER responses (Fig 5A–5C, S5 Fig). It does not, however, rescue glycerol detection. Furthermore, we also confirmed that functional replacement of GR64e with TRPA1 was dependent on PLC. Expression of TrpA1 or Gr64e in sweet GRNs of norpAP24,Gr64af double mutant flies does not restore the response to HxA (S6 Fig).
We next asked whether GR64e can replace the function of TRPA1 in sensing noxious chemicals. We found that ARI elicits similar electrophysiological responses in wild-type and TrpA11 flies expressing Gr64e in their bitter GRNs (Fig 5D and 5E). TrpA11 flies expressing Gr64e in bitter GRNs do not, however, respond to NMM, a direct TRPA1 activator. These data further support Gr64e acts downstream of PLC for FA detection.
Here, we show that Gr64e—a sweet clade Gr required for glycerol detection [22]—is also essential for the gustatory detection of FAs. Although Gr64e is required in sweet GRNs for the detection of both glycerol and FAs, the molecular mechanisms by which it does so are different.
Glycerol evokes action potentials in sweet GRNs in L-, I-, and S-type sensilla in a PLC-independent manner (Fig 4A and 4B) [22]. Freeman et al. reported that single sweet GRs alone confer the responses to various sugars including glycerol when they mis-express them in olfactory neurons [27]. Only the combination of Gr64b and Gr64e, however, confers glycerol responsiveness on the sweet GRNs [36], low-salt sensing GRNs, and bitter GRNs of Gr64af flies. This suggests Drosophila GRs form heteromeric complexes for sensing sugars. Since Gr64b/Gr64e-misexpressing low-salt sensing GRNs or bitter GRNs produce fewer glycerol-evoked action potentials than sweet GRNs, we speculate that there are unknown additional Grs in sweet GRNs that facilitate the formation of high affinity glycerol receptors. This would be similar to our findings with the L-canavanine receptor [15]. Based on the characterization of GRs for bitter sensing [15, 37], the detection of glycerol occurs through the direct activation of ion channels formed by Gr64b and Gr64e (Fig 6A), but it remains unclear whether unknown intracellular signaling components also contribute to the function of sweet GRs.
FAs selectively activate sweet GRNs in S-type sensilla in a PLC-dependent manner. Of the nine sweet clade Grs (i.e., Gr5a, Gr43a, Gr61a, and Gr64a-f), only Gr64e is required for FA detection. Gr64e seems unlikely to be a FA receptor for several reasons. First, the sweet GRNs in L- and I- type sensilla, where endogenous Gr64e is expressed [28], respond only to glycerol, not FAs (Fig 3). Second, overexpression of G-protein signaling components (Gαq and norpA) alone or together with Gr64e (Gr64e, Gαq, and norpA) in sweet GRNs of L-type sensilla does not endow FA sensitivity (S4 Fig). Finally, although there are reports that the distantly related olfactory receptors function as both GPCRs and ionotropic receptors [38, 39], the inverse topology of GRs relative to GPCRs is further evidence that Gr64e is unlikely a direct FA receptor. We were unable to exclude the possibility that Gr64e acts as an accessory protein for an unknown FA-responsive GPCR or the possibility that the absence of other accessory proteins (i.e., CD36 [40]) in sweet GRNs of L-type sensilla explains their inability to respond to HxA. Furthermore, the functional redundancy we identified between GR64e and TRPA1 in PLC-specific functions (e.g., FA but not glycerol detection by GR64e and ARI but not NMM detection by TRPA1) suggests Gr64e functions downstream of PLC (Fig 6B). Although GR64e and TRPA1 are functionally interchangeable downstream of PLC, it remains unclear whether they share the same molecular mechanism of activation. GR64e can be activated by hydrolysis of phosphoinositide by PLC, elevation of intracellular calcium, or diacylglycerol. Alternatively, Gr64e may be a voltage-gated channel that is not directly coupled to the PLC pathway.
Two Drosophila species, D. psedoobscura and D. persimilis carry pseudogenized versions of Gr64e and do not respond to glycerol [22]. If these two species have also lost gustatory sensitivity to FAs, it will confirm the evolutionary conservation of this dual function for Grs.
Because this is the first time a Drosophila GR has been found to function downstream of PLC, our results extend the molecular repertoire of the GR family of proteins. This is particularly intriguing because there are Grs expressed in the antenna [28, 41] and in the enteroendocrine cells of the gut [42]. Rather than acting in the direct detection of ligands in these non-gustatory cells, these GRs may mediate novel sensory modalities via distinct molecular mechanisms.
FAs act as sources of energy, but also as structural components of membranes. In addition, they have multiple biological roles in metabolism, cell division, and inflammation [43]. In flies, changes in the FA composition of membranes via FA deprivation influences cold tolerance and synaptic function [44, 45]. Dietary FAs also modulate mitochondrial function and longevity [46]. Thus, animals must ingest dietary FAs for survival. Indeed, regular laboratory Drosophila foods also contain FAs [45]. It is unsurprising that FA taste is well-conserved between mammals and flies, which are required for PLC pathway in contrast to other taste modalities in flies. Since GPR40 and GPR120 are strong FA receptor candidates in mammals [5], an FA-sensitive GPCR may also be selectively expressed in the sweet GRNs of S-type sensilla in flies. It will be interesting to determine whether the Drosophila orthologue of the mammalian FA receptor or any other GPCRs are involved in FA detection.
Flies were maintained on cornmeal-molasses-yeast medium at 25°C and 60% humidity with a 12h/12h light/dark cycle. The fly medium recipe is based on the Bloomington recipe (https://bdsc.indiana.edu/information/recipes/molassesfood.html) and composed of 3% yeast (SAF Instant Yeast), 6% cornmeal (DFC-30102, Hansol Tech, Korea), 8% molasses (extra fancy Barbados molasses, food grade, Crosby Molasses Co., Ltd. of Canada), and 1% agar (DFA-30301, Hansol Tech) for the nutrients and the hardener. It also includes 0.8% Methyl 4-hydroxybenzoate (H5501, Sigma-Aldrich, Saint Louis, MO), 0.24% propionic acid (P1386, Sigma-Aldrich), and 0.0028% phosphoric acid (695017, Sigma-Aldrich) as preservatives. For optogenetic experiments, instant fly food was purchased from Carolina (Burlington, NC, #173200). Gr64d1 was described previously [47]. Gr5a-GAL4, Gr66a-GAL4, Gr43aGAL4, Gr5aLEXA, Gr64aGAL4, Gr64bLEXA, Gr64cLEXA, Gr64eLEXA, and Gr64fLEXA were provided by H. Amrein. ΔGr64a1, ΔGr64a2, and ΔGr61a1 were provided by J. Carlson. UAS-Gr64b, UAS-Gr64c, and UAS-Gr64e were provided by A. Dahanukar. Gr64ab, Ir76b-GAL4, and TrpA11 were provided by C. Montell, UAS-TrpA1(A)10a, UAS-TrpA1(A)10b, and UAS-TrpA1(B)10a were provided by P. Garrity, and LexAop-Kir2.1 was provided from B. Dickson, respectively. UAS-ReaChR (BL53741), norpAP24 (BL9048), UAS-norpA (BL26273), UAS-Gαq (BL30734), Gr64e-GAL4 (BL57667), and UAS-Kir2.1 (BL6595) were obtained from the Bloomington Stock Center. nos-Cas9 (#CAS-0001) was obtained from NIG-FLY. All the mutant lines and transgenic lines were backcrossed for five generations to the w1118 control genotype. For clarity, the w1118 line is referred to as wild-type throughout the manuscript.
We used CRISPR/Cas9 system to generate Gr64af flies [48]. We selected two target sites for deletion of the whole Gr64 cluster using DRSC Find CRISPRs (http://www.flyrnai.org/crispr) and CRISPR optimal target finder (http://tools.flycrispr.molbio.wisc.edu/targetFinder): one near the 5’ end of Gr64a (GAATCCTCAACAAACTTCGGTGG, the Protospacer Adjacent Motif is underlined) and one near the 3’ end of Gr64f (GGTCGTTGTCCTCATGAAATTGG). We synthesized oligomers and cloned them into the BbsI site on pU6-BbsI-ChiRNA (Addgene #45946). After injecting two pU6-ChiRNA targeting constructs into nos-Cas9 embryos at 500 ng/μl each, we screened the resulting flies for deletions via PCR of genomic DNA isolated from the G0 generation. The primers we used for deletion confirmation were as follows: TCTCGGCAGCTAATCGAAAT and GCGACCATTCTTTGTGGAAT.
We collected 3–5-day-old flies in fresh food for 24 hours. Then, we starved them for 18 hours in vials containing 1% agarose. After anaesthetizing the flies on ice, we mounted them on slide glasses with melted 1-tetradecanol (185388, Sigma-Aldrich). We then allowed the flies to recover for 1–2 hours and ensured they were satiated with water before the assay. For each test solution, we used a 1 ml syringe with a 32-gauge needle to apply a single droplet directly to the labellum. We dissolved FAs in 4% ethanol. Each experimental group contained 24 flies, half were mated males and half were mated females, attached to a slide glass. All PER experiments were performed at the same time to eliminate any circadian effects. We report PER responses as the number of responding flies/total flies.
We performed tip recordings as previously described [49, 50]. Briefly, we immobilized 5–7-day-old flies by inserting a reference electrode—a glass capillary filled with Ringer’s solution—through the thorax and into the labellum. Then, we stimulated the indicated labellar sensilla with a recording electrode (10–20 μm tip diameter) containing test chemicals in 30 mM tricholine citrate (TCC) as the electrolyte. After connecting the recording electrode to a 10X preamplifier (TastePROBE; Syntech, Hilversum, The Netherlands), we recorded action potentials at 12 kHz with a 100–3,000 Hz band-pass filter using a data acquisition controller (Syntech), sorted the spikes based on amplitude, and analyzed them with the Autospike 3.1 software package (Syntech).
We purchased hexanoic acids (153745), octanoic acids (2875), oleic acids (01008), linoleic acids (L1376), sucrose (S9378), α-D-glucose (158968), D-(-)-fructose (F3510), D-(+)-trehalose dihydrate (90210), glycerol (G9012), N-methylmaleimide (389412), aristolochic acid I (A5512), and tricholine citrate (T0252) from Sigma-Aldrich. Sodium chloride (S0520) was purchased from Duchefa Biochemie (Haarlem, Netherland).
3–4-day-old flies were transferred to vials containing instant Drosophila medium with or without 400 μM all trans-retinal (R2500, Sigma-Aldrich), respectively. After feeding the flies retinal for a week, they were mounted into 200 μl pipette tips. Then, they were exposed to LED light (wavelength of 627 nm). PER responses were monitored by video camera and counted manually.
We performed all statistical analyses using SPSS Statistics 23 (IBM Corporation, Armonk, NY). We tested normality and homoscedasticity using the Kolmogorov-Smirnov and Levene tests. PER responses are displayed as means ± SEM. We used unpaired Student’s t-tests or one-way ANOVAs with Tukey post-hoc tests to analyze the PER data. All electrophysiological data are presented as medians with quartiles. We used the Mann-Whitney U-test or Kruskal-Wallis test with Mann-Whitney U post-hoc tests to determine whether the medians for each genotype were significantly different.
|
10.1371/journal.pbio.1001615 | The Rho Exchange Factors Vav2 and Vav3 Favor Skin Tumor Initiation and Promotion by Engaging Extracellular Signaling Loops | The catalytic activity of GDP/GTP exchange factors (GEFs) is considered critical to maintain the typically high activity of Rho GTPases found in cancer cells. However, the large number of them has made it difficult to pinpoint those playing proactive, nonredundant roles in tumors. In this work, we have investigated whether GEFs of the Vav subfamily exert such specific roles in skin cancer. Using genetically engineered mice, we show here that Vav2 and Vav3 favor cooperatively the initiation and promotion phases of skin tumors. Transcriptomal profiling and signaling experiments indicate such function is linked to the engagement of, and subsequent participation in, keratinocyte-based autocrine/paracrine programs that promote epidermal proliferation and recruitment of pro-inflammatory cells. This is a pathology-restricted mechanism because the loss of Vav proteins does not cause alterations in epidermal homeostasis. These results reveal a previously unknown Rho GEF-dependent pro-tumorigenic mechanism that influences the biology of cancer cells and their microenvironment. They also suggest that anti-Vav therapies may be of potential interest in skin tumor prevention and/or treatment.
| GDP/GTP exchange factors (GEFs) involved in Rho GTPase activation have been traditionally considered as potential anticancer drug targets. However, little is known about the best GEFs to inhibit in different tumor types, the pro-tumorigenic programs that they regulate, and the collateral effects that inactivation may induce in healthy tissues. Here, we investigate this issue in HRas-dependent skin tumors using genetic techniques. Despite the large number of Rho GEFs present in both normal and tumoral epidermis, we demonstrate that the co-expression of the exchange factors Vav2 and Vav3 is critical for the development of this tumor type. We also identify a previously unknown Vav-dependent autocrine/paracrine program that favors keratinocyte survival/proliferation and the formation of an inflammatory state during the initiation and promotion phases of this tumor. Finally, our results indicate that inactivation of Vav proteins is innocuous for the homeostasis of normal epidermis. Taken together, these results imply that Vav protein-based therapies may be of interest for skin tumor prevention and/or treatment.
| Rho guanosine nucleotide exchange factors (Rho GEFs) promote the transition of Rho family GTP hydrolases (GTPases) from the inactive (GDP bound) to the active (GTP bound) state during signal transduction [1],[2]. These enzymes can be subdivided into the Dbl-homology (DH) and the Dedicator of Cytokinesis (Dock) families based on the catalytic domain utilized for the GDP/GTP exchange reaction. A common feature of these two families is their extreme diversity because, in mammals, they are composed of 54 and 11 members, respectively. These family members vary widely in terms of catalytic specificity, presence of regulatory and effector domains, mechanism of activation, and expression patterns. As a consequence, they are key elements to adapt the activation of Rho GTPases to specific cell types, membrane receptors, or subcellular localizations [1],[2]. Rho GEFs have been traditionally regarded as important for tumorigenesis and, thereby, as potential drug targets [3]. However, the large number of Rho GEFs and their regulatory complexity have made it difficult to identify which ones were the most important for the development and/or progression of specific tumors. Inferences from sequencing data have not been useful in this case, because their genes seldom undergo mutations in cancer cells [3]. The use of animal models in this functional context has been also rather limited. However, the few studies available do support the idea that these enzymes have pro-tumorigenic functions. Thus, the T-cell lymphoma invasion and metastasis-inducing protein 1 (Tiam1, ID number: 21844) has been shown to be important for both cutaneous squamous and colorectal [4],[5] tumors. The adenomatous polyposis coli-stimulated exchange factor 1 (Asef1, ID number: 226970) and Asef2 (ID number: 219140) proteins have been linked to colorectal cancer [6]. Finally, Vav3 (ID number: 57257) and phosphatidylinositol 3,4,5-triphosphate-dependent Rac exchanger 1 (P-Rex1, ID number: 277360) are involved in the development of p190Brc/Abl-driven acute lymphoblastic leukemia and melanoma, respectively [7],[8].
Vav proteins exemplify well the complexity existing in the large Rho GEF family. Thus, this subfamily has three members in vertebrates (Vav1 [ID number: 22324], Vav2 [ID number: 22325], Vav3) that display overlapping but not identical expression patterns. They all share similar structures that encompass a complex array of regulatory, catalytic, and protein–protein interaction domains [9]. These domains enable them to interact with and become activated by receptors with either intrinsic or associated tyrosine kinase activity, activate GDP/GTP exchange on Rho GTPases, and in addition, engage parallel routes in GTPase-independent manners [9]–[17]. The physiological role of Vav proteins in the immune, nervous, and cardiovascular systems are now well established thanks to the use of genetically engineered mice [9],[18]–[25]. By contrast, the genetic analysis of the role of these proteins in cancer has been restricted so far to acute lymphoblastic leukemia and polyomavirus middle T-antigen-induced breast cancer. These studies have revealed that Vav3 and Vav2 plus Vav3 were required for the development of each of those tumors, respectively [7],[17].
In the present work, we aimed at expanding the spectrum of Vav family-dependent tumors by focusing our attention on cutaneous squamous tumors (CSTs), the second most frequent type of skin cancer worldwide [26]. To this end, we decided to use Vav family knockout mice to evaluate the role of these proteins in the development of 7,12-dimethylben[a]antracene (DMBA)/12-O-tetradecanoylphorbol-13-acetate (TPA)- and DMBA/DMBA-triggered skin tumors. In the former model, a single topic administration of DMBA induces oncogenic mutations (Q61L) in the HRas locus (ID number: 15461) in a small pool of keratinocytes (initiation phase). Subsequent serial topic applications of TPA are then applied to expand this pool of transformed keratinocytes to generate papillomas (promotion phase) and, depending on the genetic background of mice, cutaneous squamous cell carcinomas (cSCCs) (progression phase). Such pro-tumorigenic effect is mediated by the stimulation of intracellular signaling cascades in the initiated keratinocytes and, in addition, through autocrine/paracrine-based crosstalk between cancer and tumor-associated stromal cells that ultimately favor the expansion of the initial pool of transformed keratinocytes [27]–[30]. The latter model uses serial topic applications of DMBA that increase the frequency of cSCC development at the end of the carcinogenic protocol. We selected these models for our experiments because: (i) they are known to be Rac1- (ID number: 19353) and Tiam1-dependent [4],[31]; (ii) high levels of two Vav family proteins and a large number of additional Rho GEFs are present in normal and tumoral skin (see Figure S1), thus making a perfect working model to address intra-GEF family redundancies in a tumorigenic context; and (iii) they are compatible with the analysis of the role of the proteins under study in the tumor initiation, promotion, and progression phases [28]. Using this strategy, we have unveiled a Vav2/Vav3-dependent and cancer-specific autocrine/paracrine program that contributes to the initiation and promotion phases of skin tumors.
Expression analyses indicated that mouse papillomas and cSCCs expressed large numbers of Rho GEFs, including Vav2 and Vav3 (Figure S1). To assess if these two Vav family proteins played nonredundant roles with other Rho GEFs in the skin, we used compound Vav2−/−;Vav3−/− mice to evaluate the impact of the systemic inactivation of these two proteins in both epidermal maintenance and tumorigenesis. Unlike the case of Rac1−/− mice [32]–[34], we could not find any defect in skin development, histological structure, or self-renewal in those mice regardless of the genetic background used (Figure S2 and unpublished data). Wild-type-like parameters were also found in triple Vav1−/−;Vav2−/−;Vav3−/− C57BL/10 mice, indicating that the lack of an epidermal phenotype was not due to functional compensation events by Vav1 (unpublished data). Hence, other Rho/Rac GEFs must be in charge of stimulating Rac1 in skin stem cells and keratinocytes under physiological conditions. By contrast, we observed that Vav2−/−;Vav3−/− FVB mice displayed lower kinetics of tumor development (Figure 1A), a ≈5-fold reduction in the total number of tumors developed per mouse (Figure 1B,C), and 10-fold lower levels of carcinoma in situ (Table S1) when subjected to the two-step DMBA/TPA carcinogenic method. This effect was independent of the mouse genetic background, because C57Bl/10 Vav2−/−;Vav3−/− mice also showed statistically significant reductions in tumor burden when compared to control animals (Figure 1D–F). Vav2−/−;Vav3−/− FVB mice also displayed lower kinetics of tumor development (Figure 1G), a 2-fold reduction in tumor burden per mouse (Figure 1H,I), and a decrease in the percentage of cSCC development (Table S2) when subjected to the complete DMBA/DMBA tumorigenic method. Taken together, these results indicate that Vav2 and Vav3 play important roles in CST development but not in normal epithelial development and homeostasis.
The lower tumor burden observed in DMBA/DMBA-treated Vav2−/−;Vav3−/− mice indicated that Vav proteins may have direct roles during the initiation phase of CSTs. To investigate this possibility, we analyzed the short-term response of the epidermis of wild type and Vav2−/−;Vav3−/− mice to DMBA (Figure 2A). Using immunostaining with antibodies to the cleaved fragment of caspase 3, we observed that DMBA induced ≈2-fold higher apoptotic cell numbers in the epidermis of Vav2−/−;Vav3−/− mice than in controls (Figure 2B,C). This was not due to enhanced absorption and/or metabolization of the carcinogen, because immunohistochemistry experiments with antibodies to phospho-histone H2AX (ID number: 15270) indicated that the DMBA treatment triggered similar levels of DNA double-strand breaks in mice of both genotypes (Figure 2D,E). We also found that primary Vav2−/−;Vav3−/− keratinocytes were more susceptible to programmed cell death upon DMBA treatment or serum starvation in tissue culture, suggesting that the defects detected in vivo were keratinocyte autonomous (Figure 2F). This was a stimulus-dependent defect, because wild-type and mutant keratinocytes displayed similar cell death rates when challenged with other pro-apoptotic agents such as radiomimetic (bleomycin) and endoplasmic reticulum stress-inducing (dithiothreitol) drugs (Figure 2F). The ectopic expression of either HA-tagged Vav2 or Myc-tagged Vav3, but not of the control green fluorescent protein (GFP, ID number: P42212), restored wild-type-like apoptotic rates in both DMBA-treated and serum-starved Vav2−/−;Vav3−/− keratinocytes (Figure 2G,H), further indicating that this survival defect was a direct effect of the Vav2;Vav3 gene deficiency in keratinocytes.
We hypothesized that Vav proteins could be also involved in the TPA-dependent promotion phase of skin tumors, an idea consistent with our prior observations indicating that the effect of the double Vav2;Vav3 gene deficiency in the reduction of tumor burden was significantly more conspicuous in DMBA/TPA- than in DMBA/DMBA-treated mice (Figure 1; compare panels B and H). To investigate this possibility, we evaluated the short-term proliferative and inflammatory reaction induced by TPA in the skin of control and Vav2−/−;Vav3−/− mice (Figure 3A). The TPA-induced proliferative response of the epidermis was severely attenuated in the absence of these two proteins, as demonstrated by the limited hyperplasia (Figure 3B,C) and the low levels of BrdU incorporation into keratinocytes (Figure 3D) detected in the epidermal layers of TPA-treated Vav2−/−;Vav3−/− mice. In vivo BrdU pulse-chase experiments indicated that those proliferative defects were associated with delayed kinetics and a reduced efficiency in the G1/S phase transition induced by TPA in the mutant keratinocytes (Figure 3E). Consistent with this, immunoblot analyses using total cellular extracts obtained from the epidermis of TPA-treated mice showed that the activation (extracellular regulated kinase, [Erk, ID numbers: 26417, 26413], signal transduction and activator of transcription 3 [Stat3, ID number: 20848]) or abundance (cyclin E, ID number: 12447) of proteins involved in such cell cycle transition did not take place efficiently in Vav2−/−;Vav3−/− mice (Figure 3F). Furthermore, we observed that the dermis of these animals did not show any sign of neutrophil infiltration (Figure 3G,H) or edema-associated thickening (Figure 3G,I), indicating that the inflammatory response that takes places during the tumor promotion phase is totally abated in the absence of Vav proteins.
The lack of an inflammatory response led us to explore the potential contribution of Vav2−/−;Vav3−/− inflammatory cells to the proliferative defects found in the epidermis of Vav2−/−;Vav3−/− mice. We observed that such defects still persisted in Vav2−/−;Vav3−/− C57BL/10 mice carrying a wild-type C57BL/6-Ly5.1 hematopoietic system (Figure 4A–C), ruling out the possibility that the defective proliferation of the epidermis of the knockout animals could be an indirect consequence of dysfunctional hematopoietic cells. In agreement with those results, we also found that the grafting of Vav2−/−;Vav3−/− C57BL/10 bone marrow cells into lethally irradiated wild-type C57BL/6-Ly5.1 mice did not have any detectable effect on the responsiveness of the epidermis of host animals to TPA (Figure 4D–F).
To further assess the keratinocyte autonomous nature of the proliferative defects found in the epidermis of Vav2−/−;Vav3−/− mice, we evaluated the proliferative response of wild-type and Vav2;Vav3-deficient primary keratinocytes to TPA in cell culture. In agreement with the in vivo data, we observed that quiescent Vav2−/−;Vav3−/− keratinocytes incorporated less efficiently the S phase marker 5-ethynyl-2′-deoxyuridine (EdU) than their wild-type counterparts upon TPA stimulation (Figure 5A). This was a TPA-specific defect, because Vav2−/−;Vav3−/− keratinocytes showed normal cell cycle progression when stimulated with complete growth media (Figure 5A). Western blot and GTPase-linked immunosorbent (G-LISA) assays revealed that the Vav2;Vav3 gene deficiency was associated to reduced amounts of activation of Erk (Figure 5B, upper panel), Stat3 (Figure 5B, third panel from top), and Rac1 (Figure 5C, upper panel) in TPA-stimulated cells. It also reduced the basal levels of RhoA (ID number: 11848) activation in nonstimutated cells and, in addition, eliminated the inactivation of RhoA that was typically observed in TPA-stimulated wild-type keratinocytes (Figure 5C, lower panel). Normal levels of Erk and Rac1 activation were observed upon overexpression of HA-tagged Vav2 (Figure S3A,C) or Myc-tagged Vav3 in Vav2;Vav3-deficient keratinocytes (Figure S3B,C), thus confirming that those defects were directly due to the Vav2;Vav3 gene deficiency. Furthermore, and consistent with the TPA-specific deficiency of the cell cycle transitions, we observed that those signaling responses were normal when Vav2−/−;Vav3−/− keratinocytes were stimulated with either serum or synthetic CnT07 media (Figure 5D,E).
We surmised that a tyrosine kinase had to be involved in this process, because Vav proteins cannot be activated by direct TPA/diacylglycerol binding or protein kinase C (PKC)–mediated serine/threonine phosphorylation 9,35. In agreement with this idea, we observed that TPA triggered the tyrosine phosphorylation of both endogenous and ectopically expressed Vav2 in keratinocytes (Figure 5F). This phosphorylation was blocked when cells were pre-incubated with either general PKC (GF109203X) or Src family (PP2) inhibitors (Figure 5F) prior to the TPA stimulation step. An inactive PP2 analog (PP3) did not have such inhibitory effect on Vav2 tyrosine phosphorylation (Figure 5F). Similar results were obtained with the TPA-mediated stimulation of Erk route (Figure 5G). Although many PKC family members are present in keratinocytes (Figure S4A), we believe that a classical PKC (cPKC) must be involved, because we could reproduce the lack of Rac1 activation typically seen in Vav2−/−;Vav3−/− keratinocytes when we incubated the wild-type counterparts with Gö6976 (Figure 5H), a cPKC-specific inhibitor that does not inactivate PKCs belonging to the novel or atypical subclasses [36]. A similar effect was observed when Fyn (ID number: 14360) was knocked down in wild-type cells using short hairpin RNA (shRNA) techniques (Figures 5H and S4B), indicating that this kinase was the Src family member preferentially involved in this signaling response. This is consistent with previous results reporting the TPA-mediated stimulation of this kinase in mouse keratinocytes [37]. These results indicate that Vav proteins act downstream of a cPKC/Fyn signaling route that mediates the pro-mitogenic effects of TPA in keratinocytes.
We next considered the possibility that Vav proteins could control, in addition to the intrinsic signaling programs of keratinocytes described above, the stimulation of long-range autocrine/paracrine programs in the skin. This idea was consistent with the long-term defects seen in the activation/expression of G1/S phase-related signaling proteins in the epidermis of TPA-stimulated Vav2;Vav3-deficient mice (Figure 3F) and, in addition, by the total lack of inflammatory response found in the skin of those mice (Figure 3G–I). To explore this idea, we carried out microarray experiments to identify the fraction of the TPA-induced transcriptome of the skin (epidermis plus dermis) that was Vav-dependent. We found a significant subset of TPA-regulated transcripts whose upregulation (Figure S5A; right panel, A1 cluster; for functional annotation, see Table S3A,B) or repression (Figure S5A; right panel, A2 cluster; for functional annotation, see Table S3C,D) was Vav2/Vav3-dependent. Interestingly, bioinformatics analyses using a skin tumor microarray dataset [38] revealed that most A1 cluster transcripts displayed a similar up-regulation in DMBA/TPA-induced papillomas and/or cSCCs when compared to normal skin (Figure S5B, see clusters 2 and 3). Conversely, most A2 cluster mRNAs were expressed in normal skin and down-regulated in papillomas and/or cSCCs (Figure S5B, see clusters 5 and 6). These results indicated that the short-term Vav2/Vav3-dependent gene signature identified in the above microarray experiments is mostly conserved in fully developed tumors.
The functional annotation of the A1 gene cluster revealed a statistically significant enrichment of genes encoding extracellular ligands, including EGF family members (i.e., amphiregulin [Areg, ID number: 11839], tumor growth factor α [TGFα, ID number: 21802], heparin-binding EGF-like growth factor [HbEGF, ID number: 15200]), hepatocyte growth factor (HGF, ID number: 15234), fibroblast growth factor 7 (FGF7, ID number: 14178), vascular endothelial growth factor β (VEGFβ, ID number: 22340), and a large cohort of cytokines and chemokines (i.e., IL1β [ID number: 16176], interleukin 6 [IL6, ID number: 16193]) (Table S4). The detection of cytokine-encoding transcripts was not due to the infiltration of hematopoietic cells in the samples analyzed, because the A1 cluster did not include myeloid- or lymphocyte-specific genes (Table S3A). In silico analyses indicated that many of the genes for those ligands could be regulated by transcriptional factors belonging to the Stat (p = 6.4×10−6), nuclear factor of activated T-cells (NFAT; p = 0.002), nuclear factor kappa-light-chain-enhancer of activated B cells (NFκB; p = 0.004), AP1 (p≤0.05), and E2F (p = 0.03) families.
We confirmed by quantitative RT-PCR (qRT-PCR) that Vav proteins were important for the TPA-mediated induction of mRNAs for EGF family ligands, HGF, FGF7, IL6, and IL1β both in vivo (Figure S5C) and in vitro (Figure S5D). The only exception found for the correlation between whole skin and cultured keratinocytes was the Tgfa mRNA, which showed a Vav-dependent expression pattern in TPA-stimulated skin (Figure S5C) but not in isolated primary keratinocytes (Figure S5D). These results suggest that some of the transcripts detected in the skin microarray experiments probably represent secondary waves of transcriptional activation set in place upon the engagement of other Vav2/Vav3-dependent extracellular ligands. Defects in the production of HGF and IL6 by Vav2;Vav3-deficient mice were confirmed at the protein level using ELISA determinations in skin and serum samples obtained from TPA-treated animals (Figure S6A) and, in the case of IL6, by carrying out immunohistochemical analyses in skin sections (Figure S6B). The TPA-mediated up-regulation of these Vav-dependent transcripts was abolished in wild-type keratinocytes upon the Fyn mRNA knockdown (Figure S7), indicating that this autocrine/paracrine program is one of the downstream responses triggered by the TPA/cPKC/Fyn/Vav pro-mitogenic route previously characterized in keratinocytes (see above, Figure 5).
This Vav-dependent autocrine/paracrine program was keratinocyte-specific, since it was mostly absent in the recently described Vav-dependent transcriptome of mouse breast cancer cells [17]. Indeed, these two transcriptomes only shared the Areg, Hbegf, Tnfa, Il24 (ID number: 93672), Il23a (ID number: 83430), Osm (ID number: 18413), and Cxcl14 (ID number: 57266) transcripts (Table S4). By contrast, we observed that the majority of the Vav2/Vav3-dependent transcripts previously found to be involved in the lung-specific metastasis of breast cancer cells was also present in the Vav-dependent transcriptome of TPA-stimulated skin (Inhba [ID number: 16323], Ptgs2 [ID number: 19225], Tacstd2 [ID number: 56753]; Figure S8) [17]. The only exception was Ilk (integrin-linked kinase, ID number: 16202), which was repressed rather than induced by TPA independently of the expression status of Vav proteins (Figure S8A). This indicates that the Vav-dependent transcriptome contains both cell-type-dependent (i.e., the autocrine/paracrine program) and -independent (i.e., lung metastasis-related genes) subsets.
Given the critical roles that extracellular factors play in both the initiation and promotion phase of skin tumors [29],[30], we investigated whether they could be involved in the tumorigenic defects observed in Vav2−/−;Vav3−/− keratinocytes. If so, we speculated that the experimental manipulation of the extracellular environment had to rescue wild-type-like responses in them. To this end, we first checked whether the survival defects exhibited by DMBA-treated keratinocytes could be overcome when co-culturing them in the presence of equal numbers of wild-type cells. To distinguish each cell subpopulation in the mixed cultures, one of the subpopulation was labeled with a cell permeable chromophore prior to the co-culturing step. Using annexin V flow cytometry, we found that wild-type cells restored normal survival rates to both DMBA and serum deprivation in the co-cultured Vav2−/−;Vav3−/− keratinocytes (Figure 6A). A similar protective effect was found when we included, without wild-type cells, ligands for transmembrane tyrosine kinase receptors (epidermal growth factor [EGF, ID number: 13645], TGFα, HGF, FGF7) in the culture media of Vav2−/−;Vav3−/− keratinocytes (Figure 6B). By contrast, the addition of IL6 protected wild type but not Vav2−/−;Vav3−/− keratinocytes in the same type of experiments (Figure 6B).
The cell cycle defects shown by those cells under TPA-stimulation conditions were also eliminated when the serum-free media was supplemented with either transmembrane tyrosine kinase receptor ligands or IL1β (Figure 6C). However, as in the apoptotic assays, we observed that IL6 could induce cell cycle entry in wild-type but not in Vav-deficient keratinocytes (Figure 6C). This lack of responsiveness was not due to abnormal expression of any of the two IL6 receptor (IL6-R) subunits (Figure S9), indicating that Vav2;Vav3-deficient keratinocytes have, in addition to the general defect in the generation of the autocrine/paracrine program, a specific signaling defect downstream of the IL6-R. Consistent with this idea, immunoblot analyses indicated that EGF (Figure 6D), but not IL6 (Figure 6E), could trigger proper phosphorylation levels of Erk and Stat3 in Vav2−/−;Vav3−/− keratinocytes. This was a direct consequence of the Vav2;Vav3 gene deficiency, because we could restore normal phosphorylation levels of Erk and Stat3 downstream of the IL6-R upon the re-expression of either HA-tagged Vav2 (Figure 6F) or Myc-tagged Vav3 (Figure 6G) in Vav2−/−;Vav3−/− cells. The implication of Vav proteins in the signaling of the IL6-R was further demonstrated by the observation that IL6 triggered tyrosine phosphorylation of endogenous Vav2 in wild-type keratinocytes (Figure 6H).
To further evaluate the relevance of the Vav-dependent autocrine/paracrine program, we investigated whether the intradermal injection of mitogens could eliminate the proliferative and inflammatory defects seen in the skin of TPA-treated Vav2−/−;Vav3−/− mice. We observed that the injection of EGF, TGFα, or FGF7 induced similar levels of hyperplasia (Figure 7A) and BrdU immunoreactivity (Figure 7B) in the epidermis of wild-type and Vav2;Vav3-deficient mice. The simultaneous application of TPA resulted in a synergistic proliferative response in the epidermis of animals of both genotypes (Figure 7C,D). The intradermal injection of IL6 could trigger a robust, TPA-like mitogenic response in the epidermis of control but not Vav2−/−;Vav3−/− animals (Figure 7A–D), thus recapitulating the signaling defects observed in primary keratinocyte cultures (see above; Figure 6). IL6, but not the other ligands tested, did induce a potent infiltration of neutrophils in the skin of both wild-type and knockout mice (Figure 7E,F). This indicates that, unlike the case of keratinocytes, the IL6-R does not require the presence of Vav2 and Vav3 for proper signaling in myeloid cells. This is not due to functional compensation events from Vav1, because IL6-treated Vav1−/−;Vav2−/−;Vav3−/− mice showed neutrophil infiltration rates similar to both wild-type and Vav2−/−;Vav3−/− animals (Figure 7F, right panel).
Finally, we investigated whether the tumors that developed in Vav2;Vav3-deficient mice could be the result of the outgrowth of cancer cells that, due to selection events, could have compensated the lack of Vav proteins by the exacerbation of other signaling routes. We suspected that such compensation could occur in this case because the few tumors that developed in mutant mice displayed a high similarity to those found in control animals in terms of size distribution, proliferation rates, and differentiation stage (unpublished data). Based on the above, we decided to compare the abundance of transcripts encoding a variety of mitogenic ligands in papilloma and cSCC samples obtained from FVB mice of both genotypes. In addition, we monitored the expression pattern in those samples of Vav2/Vav3-dependent genes that were common between breast cancer cells and TPA-stimulated keratinocytes [17]. We observed that the Vav2/Vav3-dependent Tgfa, Hgf, Fgf7, and Il6 transcripts showed reduced abundance in papillomas derived from Vav2/Vav3-deficient mice when compared to the levels present in control mouse tumors (Figure 8A). This reduction, however, was milder than the defect originally seen in the TPA-stimulated skin of Vav2−/−;Vav3−/− mice (see above, Figure S5C). Other Vav2/Vav3-dependent (Areg, Hbegf) and -independent (Egf, Btc [ID number: 12223]) transcripts displayed similar levels in papillomas regardless of the Vav2/Vav3 expression status, further suggesting that the autocrine/paracrine defect was ameliorated in these tumors (Figure 8A). Such compensation event was exacerbated in carcinomas, since we observed a striking up-regulation of the abundance of EGF-R family ligand transcripts in all cSCC samples derived from mutant mice (Figure 8B). Such up-regulation took place irrespectively of whether the ligands were of the Vav-dependent (Tgfa, Areg, Hbegf) or Vav-independent (Egf, Btc) subclasses (Figure 8B). Carcinomas from Vav2;Vav3-deficient mice also displayed up-regulation of the Hgf mRNA (Figure 8B). This was not a sign of a total elimination of the Vav2/Vav3 signaling deficiency, because lower levels of Fgf7 and Il6 transcripts were still detected in cSCCs collected from Vav2−/−;Vav3−/− mice (Figure 8B). A similar behavior was observed in the case of the Inhba and Ptgs2 when their abundance was compared between papilloma and cSCCs (Figure 8C,D). However, the Tacstd2 and Ilk transcripts showed no variations in these experiments (Figure 8C,D). These results suggest that the HRasQ61L-transformed keratinocytes have to progressively bypass part of the Vav2/Vav3 signaling deficiency to generate fully developed tumors. This compensation event is not due to Vav1 overexpression, because this transcript shows a similar 3-fold increase in abundance in cSCCs of both wild-type and Vav2−/−;Vav3−/− mice (n = 4).
Our work indicates that Vav2 and Vav3 play first- and second-level signaling roles in keratinocytes that, although mechanistically different, act in a concerted manner to favor the initiation and promotion phases of CSTs (Figure 8E). From a signaling hierarchical point of view, the earliest role of Vav proteins is to work in a cPKC- and Fyn-dependent signaling cascade that leads to the downstream stimulation of Rac1 and optimal phosphorylation kinetics of both Erk and Stat3 (Figure 8E, step 1). This route has a direct impact on the G1/S transition of primary keratinocytes. A more distal endpoint of this Vav-dependent route is the engagement of a wide autocrine/paracrine program that favors keratinocyte survival to DNA damage, epithelial hyperplasia, and the formation of an inflammatory microenvironment (Figure 8E, step 2). This program is composed of extracellular factors (i.e., HGF, FGF7, IL1β) involved in the monovalent regulation of keratinocyte survival and proliferation during the initiation and promotion phases, respectively. In addition, it contains bivalent extracellular factors (i.e., IL6) that regulate keratinocyte mitogenesis and the engagement of inflammatory responses during the tumor promotion phase. Finally, Vav proteins play a second-level, keratinocyte-specific role in the signaling route of one of the Vav2/Vav3-dependent extracellular factors, the cytokine IL6 (Figure 8E, step 3). This biological program may play roles in both inchoate and fully established tumors, as indicated by the similar regulation of the TPA-induced Vav2/Vav3-dependent gene signature in papillomas and cSCCs obtained from DMBA/TPA-treated mice. By contrast, we have observed that the compound Vav2;Vav3 gene deficiency does not induce any overt dysfunction in normal skin development and homeostasis, indicating that the Vav2/Vav3-dependent route only becomes functionally relevant in the epidermis under conditions that require increased signaling thresholds for the assembly of new, pathophysiological-specific programs. To our knowledge, this is the first demonstration of the implication of Vav proteins, Rho GEFs, or any Rho GTPase in the assembly of such large autocrine/paracrine program in either a physiological or pro-tumorigenic scenario.
Several indications support the idea that the upstream and downstream roles of Vav proteins in this autocrine/paracrine program are critical for the initiation and promotion phase of inchoate skin tumors and, probably, for long-term tumor sustenance. Thus, previous reports have shown that many of the Vav2/Vav3-dependent extracellular factors regulated by Vav2 and Vav3 contribute to keratinocyte survival, proliferation, and tumorigenesis [29],[30]. Furthermore, we have shown that all the defects detected in cultured Vav2−/−;Vav3−/− keratinocytes can be effectively bypassed upon their co-culture with wild-type keratinocytes or, alternatively, upon the addition of specific Vav2/Vav3-dependent ligands (EGF family ligands, HGF, FGF7, IL1β) to their cultures. Similar results were observed when these rescue experiments were carried out in vivo using intradermal injections of those extracellular factors in Vav2−/−;Vav3−/− mice. However, and in good agreement with the critical role of Vav proteins downstream of the IL6-R, IL6 could not restore the survival and mitogenic defects of Vav2/Vav3-deficient keratinocytes when used in vitro or in vivo. It is likely that these second-level signaling defects also contribute to the lower tumor burden observed in Vav2−/−;Vav3−/− animals, since previous reports have shown that IL6-mediated signaling is essential for skin tumorigenesis [39]. Interestingly, we observed that IL6 did restore the defective inflammatory response observed in the skin of TPA-stimulated Vav2−/−;Vav3−/− animals. This result indicates that IL6 is indeed part of the Vav2/Vav3-dependent pro-mitogenic and pro-inflammatory program of keratinocytes and, in addition, that the critical role of Vav proteins downstream of the IL6-R is keratinocyte-specific. The use of Vav1−/−;Vav2−/−;Vav3−/− mice has ruled out the possibility that the normal chemoattraction of Vav2−/−;Vav3−/− neutrophils induced by IL6 could be due to functional compensation events mediated by the hematopoietic-specific Vav1 protein [9]. Thus, these cell-type-specific differences could be due to the presence of other IL6-regulated Rho GEFs in neutrophils (e.g., P-Rex1) [40] or, alternatively, reflect the implication of Vav proteins in a proliferative/survival signaling branch of the IL6-R that is not important for migration and the induction of the pro-inflammatory program. Additional experiments using wild-type and Vav family-deficient neutrophils will be needed to discriminate those possibilities. Finally, the pathophysiological importance of this program during skin tumorigenesis is further highlighted by our in silico analyses indicating that the Vav2/Vav3-dependent transcriptomal signature is conserved in fully developed tumors and, in addition, by the exacerbated up-regulation of most transcripts for those autocrine/paracrine factors consistently seen in cSCCs from Vav2−/−;Vav3−/− mice (Figure 8E).
We surmise that these experiments have only revealed the tip of the iceberg of this biological program, because the Vav2/Vav3-dependent transcriptomal signature encodes other pro-mitogenic and pro-inflammatory factors that have not been yet analyzed. It also contains factors with assigned roles in angiogenesis (i.e., Cyr61 [ID number: 16007], IL6, VEGFβ), the polarization of the TH cell response towards the TH1 (i.e., Ccl3 [ID number: 20302], Ccl4 [ID number: 20303], Cxcl5 [ID number: 20311], IL1β, Spp1 [ID number: 20750], TNFα), and TH17 (i.e., Csf3 [ID number: 12985], Cxcl2 [ID number: 20310], Cxcl5, IL6, IL10 [ID number: 16153]) subtypes or the tumor-induced “education” of dermal fibroblasts (i.e., IL1β) [29],[30],[41]–[43], thus suggesting that the engagement of the Vav2/Vav3 route in keratinocytes could result in a widespread reprogramming of the tissue microenvironment (Figure 8E). This biological program is also endowed with multiple positive feedback mechanisms that can lead to self-amplification and long-term signal sustenance upon its initial engagement in keratinocytes. An obvious signaling relay is the Vav→IL6→IL6-R→Vav→Stat3 route, since it is known that the stimulation of phospho-Stat3 re-feeds this autocrine loop by activating transcriptionally the Il6 gene [44],[45]. The targeted inflammatory and stromal cells provide additional options for amplification and diversification of signals, since those cells can turn on additional paracrine mechanisms that target keratinocytes and stromal cells [41]–[43]. Thus, it is likely that the initial activation of the Vav route in keratinocytes will kindle a “butterfly effect” that will induce widespread and reciprocal signaling interactions among keratinocytes, resident stromal cells, and newcomer inflammatory and immune cells.
Our work also sheds light on issues related to functional redundancies with the Rho GEF family and Vav subfamily during the initiation and promotion phases of skin tumors. On the one hand, the cell reconstitution experiments indicate that Vav2 and Vav3 seem to act redundantly in all the biological processes analyzed in this work. Consistent with this idea, we have observed that single Vav2−/− and Vav3−/− knockout mice do not show the initiation and promotion defects found in the compound Vav2;Vav3-deficient mice (M.M.-M. and X.R.B., unpublished data). On the other hand, this program seems to be quite idiosyncratic for Vav proteins as inferred by the detection of a phenotype despite the large number of Rho GEFs that, according to our bioinformatic array analyses, are present in normal skin, papilloma, and cSCCs. The cancer-linked phenotype of Vav2/Vav3-deficient mice is also quite different from that previously reported for Tiam1-deficient mice during both tumor initiation [4],[5] and papilloma/cSCC malignant progression [4]. In this context, the observation that subsets of Rho GEFs are differentially regulated in normal skin, papilloma, and cSCCs suggests the specific engagement of physiological- and tumor-stage-specific Rac1-, RhoA, and Cdc42 (ID number: 12540)-dependent programs that may contribute to both skin homeostasis and pathophysiology. These results emphasize the importance of extending animal-model-based genetic analysis to all Rho GEF family members.
Our results suggest that the pharmacological targeting of Vav proteins could be a potentially useful strategy in skin cancer. Since our data have been generated using mice lacking Vav proteins from the initiation stage, they could only be formally used to establish the value of such therapies at the prevention rather than the remediation level. However, preventive therapies are interesting in this case because CSTs are known to develop at high frequency and multiplicity in individuals with actinic keratosis or in patients treated with some immunosuppressants, antifungal antibiotics, or antitumoral therapies (i.e., B-Raf (ID number: 109880) inhibitors) [46]. Assessing the value of such potential therapies in the case of fully developed or metastasized CSTs will require the generation of chemically inducible knock-in systems. In any case, we can anticipate from the present data that anti-Vav therapies will elicit much fewer intrinsic side effects in the skin than those based on the inactivation of either Tiam1 or Rac1.
All animal work has been done in accordance with protocols approved by the Bioethics committees of both the University of Salamanca and CSIC.
To analyze the expression of Rho GEFs mRNAs in skin and tumors, raw data containing samples from normal tail skin (n = 83), papillomas (n = 60), and carcinomas (n = 68) was downloaded from the Gene Expression Omnibus website (Accession number: GSE21264). These data were obtained in Balmain's laboratory using Affymetrix Mouse Genome 430 2.0 arrays and animals belonging to a mixed Mus musculus FVB/Mus spretus genetic background [38]. Signal intensity values were obtained from CEL files after RMA. Probesets corresponding to Rho/Rac GEFs were extracted from the dataset and ANOVA analysis used to identify Rho/Rac GEF-encoding transcripts that were differentially expressed between normal tail, papillomas, and carcinomas (p value <0.01). For those genes with more than one probeset significantly deregulated, we selected the one with the lowest p value for graphic representation.
Total RNA was extracted from the indicated cells using Trizol (Sigma) and quantitative RT-PCR performed using the QuantiTect SYBR Green RT-PCR kit (Qiagen) and the iCycler machine (Bio-Rad) or, alternatively, the Script One-Step RT-PCR kit (BioRad) and the StepOnePlus Real-Time PCR System (Applied BioSystems). Raw data were then analyzed using either the iCycler iQ Optical System software (Bio-Rad) or the StepOne software v2.1 (Applied Biosystems). We used the abundance of the endogenous Gapdh mRNA as internal normalization control. Primers used for transcript quantitation included 5′-TTG CCC AGA ACA AAG GAA TC-3′ (forward for mouse Vav1), 5′-AAG CGC ATT AGG TCC TCG TA-3′ (reverse for mouse Vav1), 5′-AAG CCT GTG TTG ACC TTC CAG-3′ (forward for mouse Vav2), 5′-GTG TAA TCG ATC TCC CGG GAT-3′ (reverse for mouse Vav2), 5′-GGG TAA TAG AAC AGG CAC AGC-3′ (forward for mouse Vav3), 5′-GCC ATT TAC TTC ACC TCT CCA C-3′ (reverse for mouse Vav3), 5′-GGC AAA AAG TCA GTC CGA CC-3′ (forward for mouse Fyn, transcript variants 1 and 2), 5′-AAA GCG CCA CAA ACA GTG TC-3′ (reverse for mouse Fyn, transcript variants 1 and 2), 5′-TCG TGG CAA AAG AGC TTG GA-3′ (forward for mouse Fyn, transcript variant 3), 5′-TAG GGT CCC AGT GTG AGA GG-3′ (reverse for mouse Fyn, transcript variant 3), 5′-CGT CCG CCA TCT TGG TAG AGA GAG CAT-3′ (forward for mouse Cd3e), 5′-CTA CTG CTG TCA GGT CCA CCT CCA C-3′ (reverse for mouse Cd3e), 5′-ATG CTA GCG ATG CAT GAG TG-3′ (forward for mouse Tgfa), 5′-CAG GGA CTT TCT TGC CTG AG-3′ (reverse for mouse Tgfa), 5′-CGG TGG AAC CAA TGA GAA CT-3′ (forward for mouse Areg), 5′-TTT CGC TTA TGG TGG AAA CC-3′ (reverse for mouse Areg), 5′-GCT GCC GTC GGT GAT GCT GAA GC-3′ (forward for mouse Hbegf), 5′-GAT GAC AAG AAG ACA GAC G-3′ (reverse for mouse Hbegf), 5′-GCA GAC ACC ACA CCG GCA CAA-3′ (forward for mouse Hgf), 5′-GCA CCA TGG CCT CGG CTT GC-3′ (reverse for mouse Hgf), 5′-TTT GGA AAG AGC GAC GAC TT-3′ (forward for mouse Fgf7), 5′-GGC AGG ATC CGT GTC AGT AT-3′ (reverse for mouse Fgf7), 5′-CTT CCT ACC CCA ATT TCC AAT G-3′ (forward for mouse Il6), 5′-ATT GGA TGG TCT TGG TCC TTA GC-3′ (reverse for mouse Il6), 5′-ACG GAC CCC AAA AGA TGA AGG GCT-3′ (forward for mouse Il1b), 5′-GGG AAC GTC ACA CAC CAG CAG G-3′ (reverse for mouse Il1b), 5′-CGC TGC TTT GTC TAG GTT CC-3′ (forward for mouse Ereg), 5′-GGG ATC GTC TTC CAT CTG AA-3′ (reverse for mouse Ereg), 5′-CCC AGG CAA CGT ATC AAA GT-3′ (forward for mouse Egf), 5′-CCC AGG AAA GCA ATC ACA TT-3′ (reverse for mouse Egf), 5′-GGA ACC TGA GGA CTC ATC CA-3′ (forward for mouse Btc), 5′-TCT AGG GGT GGT ACC TGT G-3′ (reverse for mouse Btc), 5′-TGC ACC ACC AAC TGC TTA GC-3′ (forward for mouse Gapdh), and 5′-TCT TCT GGG TGG CAG TGA TG-3′ (reverse for mouse Gapdh). Primers used for quantitating Inhba, Ptgs2, Tacstd2, and Ilk transcripts were described before [17]. To calculate the number of copies of mouse Vav family mRNAs in cell/tissue samples, the Ct values obtained for the amplified Vav1, Vav2, and Vav3 cDNA fragments by qRT-PCR in each sample were compared with those obtained using serial dilutions of plasmids of known concentration containing the Vav1 (pJLZ52) [13], Vav2 (pCCM33) [17], and Vav3 (pCCM31) [17] cDNAs. The number of copies obtained for each transcript was finally calculated using this titration curve and the size of each plasmid.
Single and compound Vav family knockout mice were described elsewhere [19],[23],[47]–[49]. For long-term carcinogenesis experiments, the backs of animals of the indicated genotypes were shaved and, 2 d later, the two-step carcinogenic DMBA/TPA protocol initiated using a single topic application of DMBA (25 µg diluted in 200 µl of acetone, both from Sigma). The promotion phase consisted of biweekly applications of TPA (200 µl of a 1×10−4 M solution in acetone, Sigma) during a 20-wk-long period. For complete carcinogenesis, mice were treated biweekly with 5 µg of DMBA alone in 200 µl of acetone during 20 wk. The number, size (measured with a digital caliper), and incidence of papilloma was determined weekly. At the end of the experiment, animals were injected intraperitoneally with BrdU (100 µg/g body weight, Sigma) and euthanized 1 h later.
For short-term studies of in vivo epidermal apoptosis, the dorsal skin of mice was treated with a single application of either DMBA (25 nmol in 200 µl of acetone) or acetone (200 µl) 2 d after shaving. For short-term in vivo proliferation assays, the dorsal skin of mice was treated with either one or four applications of either TPA (6.8 nmol in 200 µl acetone) or carrier solution 2 d after shaving. Animals were injected with BrdU and euthanized, as indicated above. For the determination of in vivo cell cycle transitions, animals were treated with a single topic application of TPA and, 1 h before euthanasia, injected intraperitoneally with BrdU. Alternatively, TPA-treated animals were euthanized and the epithelial layer obtained to generate tissue cell extracts.
For intradermal injection of mitogens and cytokines, mice were anesthetized and injected under the epidermal layer of their skin with 10 µl of either the appropriate protein-containing saline solution or the control buffer injected using a 30-gauge needle and a Hamilton microsyringe. The concentration of ligands (Peprotech) in the injection solution was 100 ng/ml. Mice were then BrdU labeled and killed as above.
For bone marrow reconstitution experiments, 3–5×106 cells of donor bone marrow cells were injected intravenously into recipient mice that were previously subjected to sublethal doses of irradiation (600 rad; 1 Gy = 100 rads). In these experiments, we used “wild-type” C57BL/6 Ly5.1 mice in order to distinguish the hematopoietic populations derived from them (which express the CD45.1 surface marker) and from the knockout C57BL/10 mice (expressing the CD45.2 surface marker). Reconstitution experiments were done in both orientations. Eight weeks after injection, peripheral blood samples were collected from the cheek vein and the proper reconstitution of the immune system by the injected donor cells evaluated using flow cytometry. Surface markers used in the cytometry experiments included an allophycocyanin-labeled mouse monoclonal antibody to CD45.2, a biotin-labeled mouse monoclonal antibody to mouse CD45.1, a biotin-labeled rat monoclonal antibody to mouse CD45R/B220, a biotin-labeled rat monoclonal antibody to mouse CD19, a phycoerytrin-labeled rat monoclonal antibody to mouse CD11b, a phycoerytrin-labeled hamster monoclonal antibody to mouse CD3ε (all from BD Pharmingen), a biotin-labeled hamster monoclonal antibody to mouse TCRβ, and a phycoerytrin-labeled rat monoclonal antibody to mouse CD19 (both from eBioscience). Bone-marrow-reconstituted animals were then subjected to topic TPA treatments in the skin as indicated above.
Whereas the long-term carcinogenesis assays were done in animals of both the FVB and C57BL/10 genetic backgrounds, we selected animals of the C57BL/10 background for the short-term studies. This facilitated comparative studies with other Vav family knockout animals (i.e., single Vav1−/− and triple Vav1−/−;Vav2−/−;Vav3−/− mice, which were all homogenized in that background) as well as the bone marrow reconstitution experiments (which required transplantation between genetically compatible donor and recipient mice). At the beginning of each experimental procedure, cohorts of 6–8-wk-old animals with an even gender distribution of each genotype subset were used.
Samples from tumors and short-term skin experiments were processed following three independent protocols: (i) Fixed in 4% paraformaldehyde (Sigma), (ii) cryoprotected and stored at −70°C, and (iii) snap-frozen and stored at −70°C for molecular analyses (i.e., RNA and protein extraction).
Tissues were extracted, fixed in 4% paraformaldehyde (Sigma), cut in 2–3 µm thick sections, and stained with hematoxylin/eosin. Tumor sections were analyzed by pathologists to classify them according to malignancy grade (benign, benign plus carcinoma in situ, malignant) and level of differentiation (high, mild poor). In short-term experiments, the thickness of the epidermal layer and total skin thickness was measured in vertical cross-sections in at least 10 different locations in each mouse to determine epidermal hyperplasia and inflammatory response (edema), respectively.
For standard immunohistochemical staining, paraffin-embedded sections were dewaxed, microwaved in citrate buffer, blocked with 10% nonimmune horse serum (Gibco), and incubated overnight with the appropriate primary antibody at 4°C. After exhaustive washing in 137 mM NaCl, 2.7 mM KCl, 10 mM Na2HPO4, 2 mM KH2PO4, and 0.1% Tween-20 (PBST), sections were incubated with appropriate biotin-coupled secondary antibodies (all used at a 1∶1,000 dilution in PBST) followed by avidin-peroxidase (ABC Elite Kit Vector, Vector labs). Positive staining was determined using diaminobenzidine as a substrate (DAB Kit Vector, Vector labs) following the manufacturer's recommendations. Sections were then counterstained with hematoxylin and mounted. Images were captured using an Olympus BX51 microscope coupled to an Olympus DP70 digital camera. Staining quantification was blindly assessed by two independent investigators and classified according to intensity (0–10) and estimation of ratio of cells stained (0–1). The relative value for each section was scored as the product of intensity×ratio given a general score ranging from 0 to 10. For immunofluorescent staining, sections treated as before were incubated with either a Alexa Fluor 488–labeled goat anti-rabbit IgG antibody (Molecular Probes, 1∶200 dilution) or a Cy3-conjugated goat anti-mouse IgG antibody (Jackson Immunoresearch laboratories, Inc, 1∶200 dilution), washed, and countersatined with 4′,6-diamidino-2-phenylindole (DAPI; 2 ng/ml, Sigma) with or without rhodamine-labeled phalloidin (Molecular Probes, 1∶200 dilution). Primary antibodies used in these experiments included rabbit polyclonal antibodies to keratin 1, keratin 5, keratin 14, and filaggrin (Covance, 1∶500 dilution in each case), rabbit polyclonal antibodies to IL6 (Genzyme, 1∶100 dilution), rat monoclonal antibodies to keratin 8 (Troma1, not commercial; 1/5 dilution of the hybridoma supernatant) and CD45 (BD Biosciences, 1∶50 dilution), mouse monoclonal antibodies to keratin 10 (Santa Cruz Biotechnology, 1∶50 dilution), keratin 13 (Sigma, 1∶50 dilution) and histone H2A.X (Cell Signaling, 1∶100 dilution), and a rabbit polyclonal antibody to myeloperoxidase (Abcam, 1∶250 dilution). To detect proliferating cells, dewaxed sections were denaturalized in 2N HCl at 37°C for 1 h, washed extensively in 0.1 M borate buffer, blocked with 2% nonfat dry milk (Nestlé) in 0.1% Triton-PBS, and incubated overnight with a mouse monoclonal antibody to BrdU (BD Biosciences, 1∶400 dilution). To detect apoptotic cells, we used two different approaches. In some cases, tissue sections were deparafinized, hydrated, digested with proteinase K (Dako) for 30 min at 37°C, and subjected to the Tunel reaction using the Tunel-based In Situ Cell Detection kit (Roche) as indicated by the manufacturer's instructions. Alternatively, tissues sections were cryoprotected using stepwise immersions in 15%–30% sucrose-PBS solutions, embedded in Tissue-Tek OCT (Sakura Finetek Europe), and stored frozen at −70°C. Upon thawing, slides were blocked as described before and incubated with a rabbit polyclonal antibody to cleaved caspase 3 (Cell Signaling, 1∶50 dilution). Images from immunofluoresce experiments were captured using a Leica CTR600 microscope. Quantifications of both standard and immunoflurescence signals were done with the Metamorph-Metaview software (Universal Imaging).
Epidermis from neonates of the indicated genotypes were treated with 250 U/ml of dispase (Roche) overnight at 4°C and keratinocytes prepared using CnT07 media (CELLnTEC) according to the manufacturer's instructions. Keratinocytes were maintained in CnT07 media on type I collagen-precoated plates (BD Biosciences). When genotypically mixed cultures were utilized, the keratinocytes of one the genotypes used were labeled in culture with a cell permeable chromophore (CellTracker green CMFDA, Invitrogen) for 30 min according to the manufacturer's instructions. Labeled and nonlabeled cultures were trypsinized and plated either as genotypically pure or mixed populations in tissue culture plates.
Apoptotic rates in vitro were determined by flow cytometry using the Annexin V kit (Immunostep). In vitro cell cycle transitions were determined using the Click-iT EdU Alexa Fluor 647 Flow Cytometry Assay Kit (Invitrogen). For in vitro apoptotic assays, exponentially growing keratinocytes of the indicated genotypes were either serum starved in Eagle's minimal essential medium (EMEM, Lonza) supplemented with 0.05 mM CaCl2 (Sigma) or, alternatively, treated with 100 µM DMBA, 0.15 µM bleomycin (Sigma), or 2 mM dithiothreitol (Sigma). Cells were then harvested 8–24 h later and the number of apoptotic cells determined by flow cytometry using the Annexin V kit (Immunostep). In apoptotic experiments using keratinocytes ectopically expressing proteins, cells that had integrated the lentiviral particle (GFP+) were gated away from noninfected (GFP−) cells. A similar gating strategy (presence/absence of the CellTracker green CMFDA chromophore) was used to characterize apoptotic rates in genotypically mixed cell cultures. In rescue experiments with extracellular ligands, these factors were added at the beginning of the starvation period at a concentration of either 10 ng/ml (EGF, TGFα, HGF, FGF7; all obtained from Peprotech) or 100 ng/ml (IL6, Peprotech) and apoptosis quantified as above after an overnight incubation.
For keratinocyte G1/S phase transition assays, cells were starved for 3 h as above and then stimulated by the addition EMEM supplemented with 0.2 µM TPA, 10 ng/ml of receptors for transmembrane tyrosine kinases (EGF, TGFα HGF, FGF7), 10 ng/ml IL1β (Peprotech), 100 ng/ml IL6, or 8% calcium-chelated fetal calf serum. After 4 h, cells were incubated with the EdU reactive (Invitrogen) for 30 min, and the percentage of cells in S phase (EdU+) determined using the Click-iT EdU Alexa Fluor 647 Flow Cytometry Assay Kit (Invitrogen).
For intracellular signaling experiments, exponentially growing primary cells of the indicated genotypes were washed and starved for 3 h in EMEM supplemented with 0.05 mM CaCl2. For stimulation, we used EMEM supplemented with 0.2 µM TPA, 10 ng/ml EGF, or 100 ng/ml IL6 for the indicated periods of time. When indicated, the pan-PKC GF109203X (5 µM, Sigma), the classical PKD-specific Gö6976 (3 µM, Calbiochem), and the Src family PP2 (2 µM, Calbiochem) inhibitors were added to the starved keratinocytes 30 min before the stimulation with TPA. The PP3 molecule was used as negative control for the PP2 experiments (2 µM, Calbiochem).
When indicated, HA-Vav2, Myc-Vav3 with or without GFP were expressed in wild-type (GFP) or Vav2−/−;Vav3−/− (each of the above proteins) keratinocytes using lentiviral delivery methods. To this end, the pCCM33 [17], pCCM31 [17], or pCQS1 [21] vectors were transfected into Lenti-X 293T cells (Clontech) using the Lenti-X HT packaging mix (Clontech). Viral particles were collected 48 h after transfection, concentrated using the Lenti-X concentrator kit (Clontech), and then used to infect keratinocyctes of the indicated genotypes during 3 consecutive days using centrifugation at 1,500× g in the presence of 8 µg/ml polybrene (Sigma). As controls, we carried out infections of wild-type and mutant keratinocytes using either the empty pLVX-IRES-Hyg or the GFP-encoding pcDH1-MCS1-EF1-coGFP (System Biosciences) lentivirus. The shRNA-mediated knockdown of Fyn transcripts was carried out using lentiviral particles (TRC lentiviral Mouse Fyn shRNA; Thermo Scientific) as previously described [17]. The TRC number and the shRNA sequence yielding the greatest knockdown was clone number TRCN0000023382 (5′-AAA CCC AGG GCT GCC TTG GAA AAG-3′). This clone was used in the experiments presented in this work.
In the case of tissue extracts, skin samples were excised from the euthanized animals, placed on ice-cold glass plates, the epidermis removed with a razor blade, transferred into a lysis buffer (10 mM Tris-HCl [pH 8.0], 150 mM NaCl, 1% Triton X-100, 1 mM Na3VO4, 10 mM β-glycerophosphate, and a mixture of protease inhibitors [Cømplete, Roche]), and mechanically homogenized using the GentleMACS dissociator (Miltenyi Biotec). In the case of primary keratinocytes maintained in culture, cells were washed with chilled PBS, scrapped in lysis buffer, and disrupted by extensive vortexing (Ika). Extracts were precleared by centrifugation at 14,000 rpm for 10 min at 4°C, denatured by boiling in SDS-PAGE sample buffer, separated electrophoretically, and transferred onto nitrocellulose filters using the iBlot Dry Blotting System (Invitrogen). Membranes were blocked in 2% BSA (Sigma) in TBS-T (25 mM Tris-HCl (pH 8.0), 150 mM NaCl, 0.1% Tween-20) for at least 1 h and then incubated overnight with the appropriate antibodies. Those included rabbit polyclonal antibodies to phospho-Erk1/2 (residues Thr202/Tyr204; Cell Signaling, 1∶1,000 dilution), total Erk1/2 (Cell Signaling, 1∶1,000 dilution), phospho-Stat3 (residue Tyr705; Cell Signaling, 1∶1,000 dilution), total Stat3 (Cell Signaling, 1∶1,000 dilution), cyclin E (Abcam, 1∶1,000 dilution), the Myc epitope (Upstate/Millipore, 1∶1,000 dilution), Gp130 (Cell Signaling, 1∶1,000 dilution), PKC and PKD (all from Cell Signaling, 1∶1,000 dilution), as well as mouse monoclonal antibodies to α-tubulin (Calbiochem, 1∶1,000 dilution), phosphotyrosine (Santa Cruz Biotechnology, dilution 1∶1,000), the HA epitope (Covance, 1∶1,000 dilution), and IL6-Ra subunit (Santa Cruz Biotechnology, 1∶500 dilution). Homemade rabbit polyclonal antibodies to Vav2 and Vav3 have been previously described [16],[20],[24]. After three washes with TBS-T to eliminate the primary antibody, the membrane was incubated with the appropriate secondary antibody (GE Healthcare, 1∶5,000) for 30 min at room temperature. Immunoreacting bands were developed using a standard chemoluminescent method (ECL, GE Healthcare). In the case of immunoprecipitations, keratinocyte lysates obtained as above were incubated overnight at 4°C using either a rabbit polyclonal antibody to Vav2 or a monoclonal antibody to HA. Immunocomplexes were collected with Gammabind G-Sepharose beads (GE Healthcare Life Biosciences), washed, and analyzed by immunobloting as above. ELISAs were used to measure the amount of IL6 (IL6 Mouse ELISA Kit, Invitrogen) and HGF (HGF Mouse ELISA kit, Abnova) according to the manufacturer's instructions. Absorbance at 450 nm was measured immediately at the end of the protocol using a plate reader (Ultraevolution, Tecan).
Total cellular lysates were obtained as above, snap frozen, thawed, quantified, and analyzed using Rac1 and RhoA G-LISA assay kits according to the manufacturer's instructions (Cytoskeleton). Absorbance at 490 nm was measured using a Ultraevolution plate reader.
All microarray experiments were performed by the personnel of the Genomics and Proteomics Unit of our Institution. Total cellular RNA was extracted from the back skin of wild-type or knockout mice treated as described before (n = 3 for each treatment) using the RNAeasy kit (Qiagen), quantified using 6000 Nano Chips (Agilent), and used (2.3 µg/sample) to generate labeled cRNA probes according to the manufacturer's instructions (Affymetrix). Purified RNA was processed and hybridized to Affymetrix GeneChip Mouse Gene 1.0 ST as indicated elsewhere [17],[50]. Signal intensity values were obtained from CEL files after robust multichip average [51],[52]. We performed Pavlidis template matching analysis, in which the Pearson's correlation coefficient is computed between the intensities measured for each gene and the values of an independent variable [53]. The p values to test for the null hypothesis that the correlation is zero are provided. Corrected p values were also calculated (Q value). For functional annotation purposes, genes were considered differentially regulated if their Q values were lower than 0.05. For metagenomic analysis, we used as threshold significance values a Q≤0.025 and fold change variations relative to control skin ≥1.5. Graphical representation of microarray data was generated using hierarchical clustering analysis and the BioConductor HCLUST tool.
Further bioinformatics were used to check the expression of Vav2/Vav3-dependent mRNAs in normal skin and skin tumors. To that end, probesets within GSE21264 dataset corresponding to A1 and A2 gene clusters were analyzed from differential expression between normal tail, papillomas, and carcinomas using the microarray database indicated above and ANOVA analysis (p≤0.01). For those genes with more than one probeset significantly deregulated, we selected the one with the lowest p value for representation.
The raw microarray data generated in this work have been uploaded to the GEO database (www.ncbi.nlm.nih.gov/geo/query/acc.cgi?token=pjcxliqugwowqfq&acc=GSE40849).
Differences in tumor multiplicity and incidence were analyzed by the Mann-Whitney U test and the χ2 test, respectively. Other wet lab data were processed using the Student's t or the one tail Mann-Whitney tests. In all cases, p values lower than 0.05 were considered statistically significant. p values were represented in all figures as * (when ≤0.05), ** (when ≤0.01), and *** (when ≤0.001). Data obtained are given as the mean ± the s.e.m.
|
10.1371/journal.ppat.1002415 | Wolbachia Symbiont Infections Induce Strong Cytoplasmic Incompatibility in the Tsetse Fly Glossina morsitans | Tsetse flies are vectors of the protozoan parasite African trypanosomes, which cause sleeping sickness disease in humans and nagana in livestock. Although there are no effective vaccines and efficacious drugs against this parasite, vector reduction methods have been successful in curbing the disease, especially for nagana. Potential vector control methods that do not involve use of chemicals is a genetic modification approach where flies engineered to be parasite resistant are allowed to replace their susceptible natural counterparts, and Sterile Insect technique (SIT) where males sterilized by chemical means are released to suppress female fecundity. The success of genetic modification approaches requires identification of strong drive systems to spread the desirable traits and the efficacy of SIT can be enhanced by identification of natural mating incompatibility. One such drive mechanism results from the cytoplasmic incompatibility (CI) phenomenon induced by the symbiont Wolbachia. CI can also be used to induce natural mating incompatibility between release males and natural populations. Although Wolbachia infections have been reported in tsetse, it has been a challenge to understand their functional biology as attempts to cure tsetse of Wolbachia infections by antibiotic treatment damages the obligate mutualistic symbiont (Wigglesworthia), without which the flies are sterile. Here, we developed aposymbiotic (symbiont-free) and fertile tsetse lines by dietary provisioning of tetracycline supplemented blood meals with yeast extract, which rescues Wigglesworthia-induced sterility. Our results reveal that Wolbachia infections confer strong CI during embryogenesis in Wolbachia-free (GmmApo) females when mated with Wolbachia-infected (GmmWt) males. These results are the first demonstration of the biological significance of Wolbachia infections in tsetse. Furthermore, when incorporated into a mathematical model, our results confirm that Wolbachia can be used successfully as a gene driver. This lays the foundation for new disease control methods including a population replacement approach with parasite resistant flies. Alternatively, the availability of males that are reproductively incompatible with natural populations can enhance the efficacy of the ongoing sterile insect technique (SIT) applications by eliminating the need for chemical irradiation.
| Infections with the parasitic bacterium Wolbachia are widespread in insects and cause a number of reproductive modifications, including cytoplasmic incompatibility (CI). There is growing interest in Wolbachia, as CI may be able to drive desired phenotypes such as disease resistance traits, into natural populations. Although Wolbachia infections had been reported in the medically and agriculturally important tsetse, their functional role was unknown. This is because attempts to cure tsetse of Wolbachia by antibiotic treatment damages the obligate mutualist Wigglesworthia, without which the flies are sterile. Here we have succeeded in the development of Wolbachia free and still fertile tsetse lines. Mating experiments for the first time provides evidence of strong CI in tsetse. We have incorporated our empirical data in a mathematical model and show that Wolbachia infections can be harnessed in tsetse to drive desirable phenotypes into natural populations in few generations. This finding provides additional support for the application of genetic approaches, which aim to spread parasite resistance traits in natural populations as a novel disease control method. Alternatively, releasing Wolbachia infected males can enhance Sterile Insect applications, as this will reduce the fecundity of natural females either uninfected or carrying a different strain of Wolbachia.
| Tsetse flies are the sole vector of Human African Trypanosomiasis (HAT), also known as sleeping sickness, caused by the protozoan Trypanosoma brucei spp. in sub-Saharan Africa. Recent figures released by the World Health Organization (WHO) indicate that the devastating HAT epidemics, which started in the early 1990s, are coming under control and may no longer represent a major public health crisis [1]–[3]. While this news is welcoming, about 60 million people continue to live in tsetse infested areas at risk for HAT in 37 countries, and those at high risk are in remote areas where disease control is difficult to implement [2]. Diseases caused by trypanosomes in animals continue to be rampant in Africa and result in severe economic and nutritional losses. The ability to curb infections in animals stands to increase both economic and nutritional status of the continent.
Unfortunately, the disease toolbox remains very limited. To date, no vaccines have been developed for HAT, therapeutic treatments are expensive and have serious side effects, and diagnostic tools are inadequate [1]. Reduction of tsetse populations, however has proven as an effective method for disease control [1]. Although effective, implementation of vector control methods in remote regions of Africa where the disease is rampant is difficult, expensive and relies on extensive community participation and thus has not been widely exercised for human disease control [4]. During an endemic period however, vector control can be particularly advantageous in the absence of continued active case surveillance [5]. Mathematical models indicate that parasite infection prevalence in the tsetse host is an influential parameter for HAT epidemiology and disease dynamics [5]. Thus, reducing vector populations or reducing the parasite transmission ability of flies can be most effective in preventing disease emergence.
Advances in tsetse biology offer novel strategies, one being a population replacement approach to modify tsetse’s parasite transmission ability (vector competence) by expressing trypanocidal molecules in the gut bacterial symbiont fauna, termed paratransgenic transformation strategy [6]–[10]. For the paratransgenic approach to be successful, gene drive mechanisms need to be discovered to spread parasite resistant phenotypes into natural populations. An alternative vector control approach currently being entertained on the continent involves a population eradication method, through sterile male releases (SIT) [11]. Genetic methods that induce reproductive male sterility are superior to the currently available SIT strategy that relies on chemical irradiation to induce male sterility.
Tsetse flies are infected with multiple bacterial symbionts. Two of the symbionts are enteric: the obligate Wigglesworthia glossinidia reside within bacteriocytes in the midgut bacteriome organ as well as in milk gland accessory tissue [12], while commensal Sodalis glossinidius reside both inter- and extra-cellularly in various tissues [13]. A large portion of Wigglesworthia’s proteome encodes vitamin products that may be necessary to supplement the strictly vertebrate blood meal diet of tsetse [14]. Without the bacteriome population of Wigglesworthia, tsetse flies have reduced egg development and are infecund [15]–[18]. The third symbiont, Wolbachia resides mainly in the reproductive tissues [13], [19], [20].
Tsetse females have an unusual viviparous reproductive biology. Females develop a single oocyte per gonotrophic cycle. The oocyte is ovulated, fertilized and undergoes embryonic development in-utero. The resulting larva hatches and is carried in the intrauterine environment through three larval instars before being deposited. During its intrauterine life, the larva receives all of its nutrients in the form of milk secreted by the female accessory glands, milk glands. While Wolbachia is transovarially transmitted, the enteric symbionts are maternally transmitted into tsetse’s intrauterine larva through mother’s milk secretions [14]. By providing ampicillin in the blood meal diet, it has been possible to clear the extracellular Wigglesworthia in the milk without damaging the intracellular Wigglesworthia in the bacteriome [21]. Thus, such females remain fecund but give rise to sterile progeny that lack Wigglesworthia (both bacteriome and milk gland populations) but retain Wolbachia and Sodalis. As a result of the obligate role of Wigglesworthia, it has not been possible to use tetracycline treatment to cure Wolbachia infections, and the biological significance of Wolbachia infections in tsetse has thus remained elusive.
Wolbachia infections associated with various insects have been shown to cause a number of reproductive modifications in their hosts, the most common being CI [22]–[24]. CI occurs when a Wolbachia infected male mates with an uninfected female, causing developmental arrest of the embryo. In contrast, Wolbachia infected females can mate with either an uninfected male or a male infected with the same Wolbachia strain and produce viable Wolbachia infected offspring. This reproductive advantage of infected females results in the spread of Wolbachia infections along with other traits that infected insects may exhibit [25], [26]. Empirical studies and previously developed models have shown that the reproductive advantage provided by Wolbachia may be able to drive desired phenotypes along with other maternally inherited genes, organelles and/or symbionts into natural populations [27]–[30]. The Wolbachia type found in the tsetse species Glossina morsitans morsitans belongs to the Wolbachia A super group [20]. In a number of insect systems, Wolbachia strains belonging to the A super group have been associated with the CI phenotype in the different hosts they infect [31].
Here we investigated the possible role of Wolbachia symbionts that can be used to drive desirable tsetse phenotypes into natural populations, or to induce natural reproductive male sterility for field applications. We developed a dietary supplementation method that can restore fecundity of tsetse in the absence of their natural symbiotic fauna, including obligate Wigglesworthia and Wolbachia. We report on the fitness parameters of the engineered symbiont-free lines and on the level of CI expression after wild type and aposymbiotic flies are crossed. A mathematical model was also developed to ascertain whether Wolbachia infections in tsetse could be used to drive a disease refractory phenotype into a natural population.
In many insect systems, tetracycline supplemented diet is used to generate Wolbachia free lines to demonstrate the functional role of Wolbachia through mating experiments. Inseminated tsetse females maintained on tetracycline-supplemented blood meals however do not generate any viable progeny. This is because tetracycline treatment damages the obligate intracellular Wigglesworthia present in the midgut bacteriome structure (Figure S1) [21]. These results are similar to prior reports where damage to Wigglesworthia had been found to reduce host fecundity [17], [21], [32].
The fecundity of fertile females maintained on various diets was evaluated (Figure 1A). Specifically, the diet combinations were as follows: (a) blood only, (b) blood and ampicillin, (c) blood and tetracycline, (d) blood and yeast, (e) blood, ampicillin and yeast, and (f) blood, tetracycline and yeast. We monitored the number of larva deposited in each group over a 40-day period when females undergo two gonotrophic cycles (defined as time required for the development of a single progeny in-utero). Under optimum conditions the first gonotrophic cycle takes about 20–22 days for development from egg to parturition. In subsequent gonotrophic cycles females produce a larva every 9 to 11 days. As we had previously shown, ampicillin treatment does not reduce fecundity since it does not damage Wigglesworthia resident within bacteriocytes in the midgut, unlike tetracycline, which clears all bacteria including Wigglesworthia and Wolbachia and induces sterility. Accordingly, ampicillin-receiving flies remained fecund while tetracycline receiving flies were rendered sterile.
Yeast extract (10% w/v) provisioning of the blood meal rescued fecundity of the females receiving tetracycline to similar levels as that of wild type and ampicillin receiving flies (65%, 55% and 64% over the first gonotrophic cycle and 53%, 58% and 49% over the second gonotrophic cycles, respectively). However, yeast provisioning at 10% w/v had a cost on fecundity when compared to flies maintained on normal blood meals, (92% versus 55% over the first gonotrophic cycle and 92% and 58% over the second gonotrophic cycle, respectively). Nevertheless, yeast supplementation was able to rescue the tetracycline-induced sterility to levels comparable to those observed for GmmWt receiving yeast or ampicillin supplemented blood meals, respectively (Figure 1A). Thus yeast supplemented dietary regiment allowed us to develop two lines to analyze the functional role of Wolbachia symbionts in tsetse biology; one lacking all symbionts (GmmApo) and another lacking Wigglesworthia but still retaining Wolbachia and Sodalis (GmmWig−).
The GmmApo progeny resulting from the first and second depositions of tetracycline treated mothers were tested for the presence of Sodalis, Wigglesworthia and Wolbachia by a bacterium-specific PCR-assay. The PCR-assay demonstrated the absence of all three symbionts as early as the first deposition in both the male and female GmmApo adults (Figure 1B). The absence of Wolbachia from the reproductive tissues of GmmApo females was also verified by Fluorescent In Situ Hybridization (FISH) analysis (Figure 1E). In contrast, Wolbachia was present in egg chambers during both early and late developmental stages in GmmWt females (Figure 1C & D). For analysis of longevity, survivorship curves were compared using the Kaplan-Meier and log rank tests. Longevity of F1 GmmApo females was compared to that of GmmWt adults maintained on the same yeast-supplemented blood meal over 40 days (two-gonotrophic cycles). No difference (X2 = 0.71, df = 1, P = 0.4) was observed in survivorship comparisons between the two groups (Figure 1F).
The second line (GmmWig−) generated from ampicillin treated females still retain their Wolbachia and Sodalis symbionts, while lacking both Wigglesworthia populations as evidenced by FISH and PCR amplification analysis [21]. When maintained on yeast-supplemented blood, this line (similar to GmmApo) also did not display any longevity differences from the GmmWt adults sustained on the same diet.
Tetracycline treatment has been shown to have a negative impact on the fertility of Drosophila simulans males [33]. To determine if the fertility of GmmApo males is negatively affected, we mated GmmWt females with either GmmWt or GmmApo males and maintained all flies on yeast-supplemented blood meals. Larval deposition and eclosion rates from both crosses were compared using arcsin(sqrrt(x)) transformed data to ensure normality. No significant difference was observed between the crosses for two gonotrophic cycles (P>0.05) (Table 1). The mean larval deposition rate for GmmWt females crossed with GmmWt males was 0.68 and 0.65 for the first and second gonotrophic cycles respectively, while the mean larval deposition rate for GmmWt females crossed with GmmApo males was 0.87 and 0.89 for the first and second gonotrophic cycles, respectively (Table 1). Similarly, no difference in eclosion rates was observed between the two groups (P>0.05) (Table 2). Of the pupae obtained in the first gonotrophic cycle from the GmmWt cross, 82% underwent eclosion compared to 83% for the cross between GmmWt females and GmmApo males. For the second gonotrophic cycle, we observed 89% average eclosion for pupae from GmmWt crosses and 93% for pupae from GmmWt females crossed with GmmApo males (Table 2). Taken together, these results demonstrate the preservation of reproductive fitness in GmmApo males and rule out possible paternal effects of Wolbachia in tsetse.
To determine the expression of Wolbachia-induced CI, cage population crosses were setup between GmmWt and GmmApo individuals. Cages were the experimental units and the data were arcsin(sqrrt(x)) transformed to ensure normality. To estimate the possible cost of reproductive fitness due to loss of Wigglesworthia, we made use of GmmWig− flies. Since GmmWig− flies still retained Wolbachia infections but lacked Wigglesworthia (as described earlier and in Figure 1A), this line served as the control for the CI cross in order to measure potential fecundity effects due to loss of Wigglesworthia in the GmmApo line and possible yeast-supplementation effects.
Although CI typically manifests itself as embryonic lethality, given the viviparous nature of reproduction in tsetse, we measured larval deposition rates, which are reflective of both successful embryogenesis and larvagenesis (Table 1). Differences in larval deposition rates (number of larva deposited per female) over the two gonotrophic cycles for all crosses were significant by ANOVA on arcsin(sqrrt(x)) transformed data (ANOVA; first deposition, F4, 9 = 20.6, P<0.0001, second deposition, F4, 10 = 21.9, P≤0.0001). No differences in larval deposition were observed between the crosses GmmWt × GmmWt, GmmWig− × GmmWig− and GmmApo × GmmApo (Table 1). However differences were observed in comparisons of the GmmApo × GmmWt cross with all other crosses for the first and second gonotrophic cycles (Table 1). Given that the GmmWig− females that lack Wigglesworthia are equally fecund as GmmWt, the strong incompatibility we observed in GmmApo females when crossed with GmmWt males is likely due to Wolbachia mediated reproductive affects, and not due to nutritional effects resulting from loss of the obligate symbiont Wigglesworthia.
We found that GmmWt females were compatible with all male infection types, while GmmApo females were only compatible with GmmApo males. Crosses of GmmApo females and GmmWt males demonstrated a pattern of unidirectional CI (Table 1). Spermathecae dissections of females in incompatible crosses that did not deposit a larva revealed the presence of sperm, suggesting females were inseminated and that lack of deposition was the result of CI. We also found that larval deposition rates and pupal eclosion rates showed similar patterns to large cage experiments when measured in single-pair crosses (Table S2). Differences were observed in larval deposition rates (number of larva deposited per female) over the two gonotrophic cycles for all single-pair crosses (Kruskal-Wallis; first deposition, χ2 = 9.3, df = 3, P = 0.03, second deposition, χ2 = 9.5, df = 3, P = 0.02). No differences in larval deposition were observed in pair-wise comparisons of the crosses GmmWt × GmmWt, GmmWt × GmmApo and GmmApo × GmmApo (Table S2). However differences were observed in comparisons of the incompatible GmmApo × GmmWt cross with all other crosses for the first and second gonotrophic cycles (Table S2). These results support strong CI expression driven by the Wolbachia infection status in female flies.
Other than reproductive modifications, Wolbachia infections have been shown to affect the fitness of their insect hosts [34], [35]. In this study, differences in eclosion rates (Table 2) were observed in the first gonotrophic cycle of crosses of GmmApo, GmmWt, and GmmWig− individuals on arcsin(sqrrt(x)) data (ANOVA, first gonotrophic cycle, F4, 11 = 7.5, P = 0.0036, second gonotrophic cycle, F3, 8 = 2.5, P = 0.13) (Table 2). No differences in eclosion rates were observed in single pair crosses for both gonotrophic cycles (Kruskal-Wallis; first gonotrophic cycle, χ2 = 0.74, df = 3, P = 0.86, second gonotrophic cycle, χ2 = 0.31, df = 2, P = 0.85) (Table S2). To determine if observed differences in eclosion rates were due to Wolbachia infection we compared the GmmWig− × GmmWig− and the GmmApo × GmmApo cross, since both strains lack Wigglesworthia infection, but one (GmmWig−) harbors Wolbachia infection. There were no significant differences however between these crosses (P>0.05) (Table 2), suggesting no extensive effect of Wolbachia infection on host eclosion rates.
The CI phenotype was further examined by analyzing the reproductive tract physiology of tsetse females between incompatible and compatible crosses during embryonic development. Under normal conditions a single oocyte undergoes and completes oogenesis during larvagenesis. In compatible crosses (♀ GmmWt × ♂ GmmWt) we observed that the reproductive tract contains a developing larva in the uterus and a developing or completed oocyte in one of the two ovaries (Figure 2A). In an incompatible cross (♀ GmmApo × ♂ GmmWt) a developing oocyte is observed in one of the ovaries in the absence of a developing larva in the uterus, suggesting a disruption of embryogenesis or early larval development (Figure 2C). The observation of an incomplete oocyte in the absence of a developing larva in the uterus suggests the failure and abortion of either an embryo or very young larva. These observations differ from older GmmWt virgin females. Typically, GmmWt virgin females undergo oogenesis but do not undergo ovulation, which results in the development and eventual accumulation of two oocytes in each of the ovaries. Larvae are never observed in the uterus as developed oocytes are never ovulated, or fertilized in adult virgin females (Figure 2B).
From the experimental data, we estimated the impact of CI on tsetse population biology using a Bayesian Markov chain Monte Carlo method. The transmission failure of Wolbachia from mothers to developing oocytes was moderate: 10.7% [0.07%, 22.7%] of progeny produced by GmmWt mothers were Wolbachia uninfected (Table 3). In addition, the incompatibility between GmmWt males and GmmApo females was strong: 79.8% [63.0%, 90.3%] of matings between GmmWt males and GmmApo females did not result in viable larvae as measured by pupal deposition. There was a significant fecundity (number of larval progeny deposited) benefit for Wigglesworhia infection: GmmWt females had 28.4% [8.5%, 54.2%] higher fecundity than GmmWig− females. Furthermore, Wolbachia infection alone was estimated to give a fecundity benefit of 19.3% [−9.2%, 57.9%]. This is an estimate of the fecundity difference between hypothetical females carrying Wigglesworthia and Sodalis but not Wolbachia and the experimental GmmWt females.
Most importantly, our model demonstrates that, given a large enough initial release, Wolbachia infected individuals will successfully invade a tsetse population (Table 4). The fixation prevalence of Wolbachia is estimated to be 96.9% [85.6%, 99.8%]. There may exist a release threshold, which an initial release must be above in order for Wolbachia to invade: the median was no release threshold (i.e. 0%), but the upper end of the 95% credible interval was a release of the size of 39.6% of the native population. The median threshold value is zero because, despite imperfect maternal transmission, the fecundity benefit of Wolbachia is strong enough to allow Wolbachia to invade a naïve tsetse population from any size initial release, no matter how small. In addition, the time to reach fixation from a release of the size of 10% of the native population can be relatively short: the median value was 529 days, however the upper end of the 95% credible interval was undefined because in more than 2.5% of samples, 10% initial release was below the release threshold.
Sensitivity analysis showed that the model results are sensitive to both assumed and estimated parameters (supplementary material Text S1). In particular, time to fixation had the largest sensitivity to the time to first deposition and large elasticities to Wolbachia- and Wigglesworthia-related parameters, suggesting that improving the estimates of these parameters would most effectively improve the fidelity of the estimate of time to fixation.
Here, we report for the first time on the functional role of Wolbachia infections in tsetse, which support the expression of CI. Microscopic analyses of the CI expressing females show that loss of fecundity results from early embryogenic failure. Essential for our studies we have discovered that we can maintain Wolbachia cured tsetse lines fertile by dietary provisioning of tetracycline supplemented blood meals with yeast extract, despite the fact that such flies lack the obligate mutualist Wigglesworthia, which is essential for tsetse’s fecundity. When incorporated into a mathematical model, our results suggest that Wolbachia can be used successfully as a gene driver and, the time to reach fixation is relatively short given a large enough initial release: on the order of 1 to 2 years. These results provide a first insight into the role of Wolbachia infections in a viviparous insect and indicate that Wolbachia mediated CI can potentially be used to drive desirable tsetse phenotypes into natural populations.
Our data presented here as well as previous results from other studies indicate that in the absence of Wigglesworthia, tsetse females are rendered sterile. Our prior studies where we maintained inseminated flies on ampicillin supplemented blood diets resulted in progeny deposition. This is because ampicillin treatment did not affect the intracellular Wigglesworthia resident in the bacteriome organ in the midgut, which provides essential nutrients to maintain tsetse host fecundity [21]. Antibiotic ampicillin treatment however eliminated the extracellular Wigglesworthia population present in the milk gland essential for symbiont transmission, and thus the resulting progeny from such females lacked Wigglesworthia (GmmWig−). Such progeny were reproductively sterile although they retained the symbiont Wolbachia. The tetracycline diet eliminated both intracellular and extracellular forms of Wigglesworthia and thus we did not obtain any viable progeny from inseminated females that were maintained on the tetracycline only diet. Prior studies showed that tetracycline blood meals supplemented with vitamin B1 could partially rescue fertility [15], but in our experiments vitamin supplementation could give rise to at most one progeny deposition, which either did not hatch or did not survive as an adult (data not shown). In sharp contrast, supplementation of the blood meal diet with 10%(w/v) yeast-extract reverted sterility in tetracycline treated flies to levels comparable to GmmWt and GmmWig− females receiving the same diet (Figure 1A). Although we have compared the fecundity of all three lines for two gonotrophic cycles here, yeast supplemented flies continue to deposit four to five progeny (data not shown). Given the complex nature of the yeast extract (peptides, amino acids, vitamins and other yeast cell components), it is difficult to know the exact nature of the essential nutrients it provides, but we believe that it could be working via supplementation of lipids and/or essential vitamins that are lacking in the strict blood diet of tsetse. However, we did observe some negative effect attributable to the yeast diet when the fecundity of GmmWt flies receiving yeast supplemented blood meals is compared to those receiving normal blood diets. As such, we are further investigating the use of different yeast supplementations and/or concentrations in an effort to improve the diet efficiency. Nevertheless the availability of Wolbachia-cured flies (GmmApo) allowed us to begin to understand the functional role of this symbiosis.
In addition to Wolbachia symbiont specific PCR amplification, we confirmed the absence of Wolbachia from the reproductive tissues of GmmApo females by FISH analysis. We show the presence of Wolbachia in GmmWt females, isolates to a pole late in development (Figure 1C). There are a number of studies in other model systems that have investigated the link between Wolbachia localization during spermatogenesis and density effects on CI [36], [37]. However, other studies have found no correlation between Wolbachia density and CI during spermatogenesis [38], [39]. There have also been a number of studies investigating Wolbachia localization during oogenesis [40]–[42]. Different Wolbachia strains in Drosophila embryos display posterior, anterior, or cortical localization congruent with the classification based on the wsp gene sequence [39]. A positive correlation between levels of Wolbachia at the posterior pole and CI has been suggested, but this has yet to be examined in detail [42]. Not withstanding, assessing the role of Wolbachia during oogenesis is important, given that factors promoting CI rescue are deposited in the egg cytoplasm during oocyte development [43] and bacterial deposition in the oocyte is an essential even for efficient maternal transmission.
Before we could perform crossing experiments to assess for CI, we evaluated the effect of Wolbachia clearance on male reproductive capacity. This evaluation is important given that tetracycline has been shown to negatively affect reproductive fitness in Drosophila simulans [33]. Additionally, the importance of this finding is highlighted by a study of the mosquito A. albopictus system in which the natural Wolbachia strains (wAlbA and wAlbB) were cleared and transinfected with the Wolbachia strain wRi from D. simulans [44]. Their results showed that the wRi transinfected males have a reduced mating capacity compared with the wild type super infected males [44]. In contrast, in our system, no decrease in mating capacity was observed in GmmApo males compared with GmmWt males under the laboratory conditions. Our observation agrees with the evolutionary model proposed by Charlat et al., [45], where Wolbachia is exclusively maternally transmitted therefore males may be considered an evolutionary dead end in terms of Wolbachia infection [46]. Consequently, no direct selection by Wolbachia can be theoretically expected on paternal reproductive fitness.
Loss of fecundity in the cross (♀ GmmApo x ♂ GmmWt) could conceivably arise from loss of Wigglesworthia-mediated nutritional benefits in GmmApo females rather than to Wolbachia mediated CI. To test this possibility, we compared the larval deposition rates in crosses between ♀GmmApo × ♂ GmmApo and ♀GmmWig− × ♂ GmmWig− flies (Table 1). Our results show no statistically significant differences between these crosses indicating that loss of fecundity in the CI cross is not due to loss of Wigglesworthia.
Our empirical results were used to parameterize a population genetic model of the spread of Wolbachia. Our model demonstrated that GmmWt would successfully invade an uninfected natural population with a large enough release given CI rates. Indeed, uninfected natural populations and natural populations with low infection prevalence have recently been identified for multiple tsetse species [47]. This modeling result is consistent with the natural spread of Wolbachia in Drosophila populations [48]–[50]. In addition, the rise to the predicted fixation prevalence of between 86% and 100% is rapid. Apparently, the Wolbachia-mediated CI has the potential to rapidly and effectively drive a desirable phenotype into natural populations. We have previously been able to culture and genetically transform the commensal symbiont of tsetse, Sodalis glossinidius [51]. It has also been possible to reintroduce the transformed Sodalis into tsetse, called a paratransgenic approach [52], [53]. Given that Sodalis resides in close proximity to pathogenic trypanosomes in tsetse’s midgut, products expressed in recSodalis can have an immediate effect on trypanosome biology. The potential paratransgenic strategy in tsetse could harness the Wolbachia mediated CI to drive a recombinant Sodalis strain that would encode parasite resistance genes into natural populations [6], [10]. Our studies on the maternal transmission dynamics of tsetse’s symbionts in the laboratory indicated perfect transmission of both Wolbachia and Sodalis into tsetse’s sequential progeny [54]. This high transmission fidelity of the two symbionts, coupled with strong nearly 100% CI caused by Wolbachia would serve paratransgenic applications favorably.
An alternative control strategy to paratransgenic population replacement strategy would be use CI as part of an incompatible insect technique (IIT), which is analogous to a SIT approach [29], [55]–[58]. In a Wolbachia-based SIT approach female sterility is artificially sustained by repeated releases of cytoplasmically incompatible males. Similar to SIT, the increasing ratio of incompatible matings over time can lead to population suppression. The benefit of an IIT strategy is that it would not require the use of irradiation or chemosterilants to sterilize males prior to release, which often reduces the fitness of released males, but would rely on the naturally induced sterility of an incompatible Wolbachia infection [59]. A Wolbachia-based paratransgenic and IIT control strategy for tsetse would rely upon the introduction of a novel infection type into a population with an existing infection that could result in bi-directional CI or the introduction of a novel infection into an uninfected host population. Typically, in other insect systems novel Wolbachia infections are established by embryonic microinjections [60], [61]. This would be difficult in tsetse given their viviparous reproductive biology, in that adult females carry and nourish their offspring for their entire larval developmental cycle making injections of embryos difficult. Future studies however can focus on the introduction of novel infection types via microinjection in aposymbiotic and naturally infected adult flies [62]. Maternal intrathoracic injections of Wolbachia infection establishment has also been successful in Aedes aegypti [63].
There has been a growing interest in understanding the variety of Wolbachia induced phenotypes in arthropods given the impact that Wolbachia infections could potentially have on genetic variation and host speciation impacting evolution of the species. Our data add to this growing field, as this is the first demonstration of the biological significance of Wolbachia infections in tsetse. Interestingly, CI in tsetse appears to be strong in that by the second gonotrophic cycle 0% of the females in an incompatible cross give rise to progeny. This is an exception given that in many insect systems incomplete CI is observed [27], [64]. Future studies with natural populations would now be important to confirm some of the parameters we report here including maternal transmission rates, infection prevalence and the maternal linkage efficacy between Wolbachia and other maternally transmitted symbionts such as Sodalis, which is being entertained for paratransgenic applications.
Additionally, the aposymbiotic lines generated in this study are currently being used to address the interactive role of trypanosome transmission in tsetse. The importance of which is highlighted by recent studies that have shown that Wolbachia infections may impact host immune biology, limiting pathogen proliferation in insect hosts [65]–[70].
The Glossina morsitans morsitans colony maintained in the insectary at Yale University was originally established from puparia collected in Zimbabwe. Newly emerged flies are separated based on sex and mated at three to four days post eclosion. Flies are maintained at 24±1°C with 50 – 55% relative humidity and fed defibrinated bovine blood (HemoStat Laboratories, CA) every forty eight hours using an artificial membrane system [71]. Selective elimination of natural tsetse endosymbionts was obtained as described below.
Wild type (GmmWt) fertile females were maintained on blood meals supplemented with 10% (w/v) yeast extract (Becton Dickinson) and 20 ug/ml of tetracycline. The yeast extract was briefly boiled in water before being added the blood meal each time. Flies were fed every 48 h using an artificial membrane feeding system (as above) for the duration of their life span. The resulting progeny are aposymbiotic (GmmApo) in that they lack their natural endosymbionts, Wigglesworthia and Wolbachia. These GmmApo lines were maintained on blood meals supplemented with 10% (w/v) yeast extract without tetracycline.
GmmWt fertile females were maintained on blood meals supplemented with 50 ug/ml of ampicillin. The resulting progeny do not have Wigglesworthia (GmmWig−), and were maintained on blood meals supplemented with 10% (w/v) yeast extract without ampicillin.
Newly eclosed aged matched females and males were divided into six groups and copulation observed. Three of these groups were provided with either normal blood meals (control) or blood meals supplemented with ampicillin at 50 ug/ml or tetracycline at 20 ug/ml. Whereas the remaining three groups received blood meals supplemented with 10% (w/v) yeast extract with either ampicillin (50 ug/ml) or tetracycline (20 ug/ml). The cages were monitored daily for pupal deposition and fly mortality over two gonotrophic cycles (40 days). Fecundity was quantified by determining the number of fecund females relative to total number of females alive at the end of the gonotrophic cycle to give an average percent of females depositing pupae. Each group was setup with 100 females per cage.
Total DNA was extracted from adults eight days post eclosion using the Qiagen Blood and Tissue extraction kit under manufacturers conditions (Qiagen Kit #, 69506. CA). The presence of the symbionts Sodalis, Wigglesworthia and Wolbachia was determined by a species-specific PCR amplification assay using the primer sets and conditions described (Table S1). For input DNA quality control, the tsetse gene β-tubulin (GmmTub) specific primer set was used. All PCR reactions were performed in an MJ-Research thermocycler and the amplification products were analyzed by electrophoresis on a 1% agarose gel and visualized using image analysis software.
Dissected reproductive tracts from GmmWt and GmmApo females were fixed in 4% paraformaldehyde (PFA), embedded in paraffin, cut into 5 mm thick sections and mounted on poly L-lysine coated microscopy slides. After dewaxing in methylcyclohexane and rehydration the sections were processed using the FISH protocol previously described in Anselme et al. 2006 [72]. Slides were covered with a drop of 70% acetic acid and incubated at 45°C until drop had dried, followed by dehydration and a 10 min deproteinization step in 0.01N HCl/pepsine at 37°C. Slides were then dehydrated again, prehybridized for 30 min at 45°C and hybridized for 3 h at 45°C with 5′ end rhodamine labeled 16S RNA probes (5′-AAT CCG GCC GAR CCG ACC C -3′) and (5′-CTT CTG TGA GTA CCG TCA TTA TC -3′). Microscopic analyses were conducted using a Zeiss Axioskop2 microscope equipped with an Infinity1 USB 2.0 camera and software (Lumenera Corporation). Fluorescent images were taken using a fluorescent filter set with fluorescein, rhodamine and DAPI specific channels.
GmmApo and GmmWt flies that emerged within a 24-hour period (teneral) were collected, mated with GmmApo males at a ratio of 5∶2 and copulation was observed. After six days males were removed from experimental cages. Six independent cages were set-up for both GmmApo and GmmWt groups, comprising of a total of 169 GmmApo and 170 GmmWt females, respectively. Both the males and females used represented offspring acquired from different gonotrophic cycles (1st and 2nd). All flies were maintained on yeast extract supplemented blood meals and fly mortality was monitored daily over a 40-day period.
To determine the expression of CI, reciprocal crosses were set up between GmmApo, GmmWt and GmmWig− flies, in triplicate. Cages with a minimum of 15 females and 7 males each were set-up in the following combinations: 1) ♀ GmmWt × ♂ GmmWt, 2) ♀ GmmWt × ♂ GmmApo, 3) ♀ GmmApo × ♂ GmmApo, 4) ♀ GmmApo × ♂ GmmWt and 5) ♀ GmmWig− × ♂ GmmWig−. All flies received yeast supplemented blood meal diets. Flies were observed over two-gonotrophic cycles with daily recording of mortality, larval deposition dates, pupal eclosion dates and sex of emergent progeny. Larval deposition rates for each gonotrophic cycle were determined by dividing the number of larvae deposited per day by the number of remaining females in the cage on the day of larviposition and summing the values for each gonotrophic cycle. At the conclusion of the experiment, all females were checked for insemination by examination of dissected spermatheca for the presence of sperm microscopically. Additionally, single line crosses consisting of a single female and male per cage were set up (Table S2). For the ♀ GmmWt × ♂ GmmWt a total of 31 crosses were set up. Also set up were 40 crosses for ♀ GmmWt × ♂ GmmApo, 20 for ♀ GmmApo × ♂ GmmApo and 33 for ♀ GmmApo × ♂ GmmWt. Both the males and females used in these crosses represented offspring acquired from different gonotrophic cycles to rule out batch affects. Spermathecae of females was also dissected to confirm insemination.
Here we will briefly describe the mathematical modeling used in this study; full details are available in the supplementary material (Text S1). The data from mating crosses were modeled as samples from the standard binomial random variable, with probability of larval deposition per mated female per gonotrophic cycle, and using a different probability for each cross. Following the empirical findings regarding Wolbachia -mediated CI in Drosophila [48], the probabilities were then defined in terms of four mechanistic parameters: the probability of reproduction success (larval deposit) from a cross between an GmmApo female and an GmmApo male (), the proportion of Wolbachia-free eggs of Wolbachia-carrying mothers (), the relative benefit to reproduction success of Wolbachia infection to females (), the relative benefit to reproduction success of Wigglesworthia infection to females (), and the proportion of fertilizations of Wolbachia-free eggs by Wolbachia-affected sperm that are not viable (). The larval-deposition probabilities in terms of these parameters arewhere the subscripts refer to the types of the female and male, respectively, with for wild type (GmmWt), for tetracycline treated (GmmApo), and for ampicillin treated (GmmWig−).
In addition to these mechanistic parameters, we also estimated population-genetic quantities fundamental to the invasion of Wolbachia into a novel tsetse population. Again following existing models for Wolbachia-induced CI in Drosophila [38], a mathematical model was developed for the temporal evolution of tsetse abundance with and without Wolbachia infection. We incorporated the Wolbachia-mediated CI trade-off of the fitness cost to male hosts in reducing their mating success with uninfected females versus the fitness benefit to female hosts in allowing them to successfully mate with both infected and uninfected males (in addition to direct effects of Wolbachia on fecundity and mortality).
For some values of the mechanistic parameters, these models exhibit a threshold for Wolbachia invasion into the host population: if, in a novel population, the proportion that is initially Wolbachia infected is above the threshold, Wolbachia will continue to stable fixation in the population at a high level. If the proportion infected is below the threshold, Wolbachia will be driven out of the population over time. This threshold level was calculated, along with the prevalence of Wolbachia at fixation, and the time to fixation. For the population-genetic model, several parameters could not be estimated from the data on mating crosses. Thus, we also performed a sensitivity analysis on these parameters, along with the parameters estimated from the mating-cross data.
To estimate both the mechanistic parameters for CI and the population-genetics quantities derived from these parameters, a Bayesian Markov chain Monte Carlo (MCMC) method was used with uninformative prior distributions for the parameters [49].
|
10.1371/journal.pmed.1002201 | Mutational Profile of Metastatic Breast Cancers: A Retrospective Analysis | Major advances have been achieved in the characterization of early breast cancer (eBC) genomic profiles. Metastatic breast cancer (mBC) is associated with poor outcomes, yet limited information is available on the genomic profile of this disease. This study aims to decipher mutational profiles of mBC using next-generation sequencing.
Whole-exome sequencing was performed on 216 tumor–blood pairs from mBC patients who underwent a biopsy in the context of the SAFIR01, SAFIR02, SHIVA, or Molecular Screening for Cancer Treatment Optimization (MOSCATO) prospective trials. Mutational profiles from 772 primary breast tumors from The Cancer Genome Atlas (TCGA) were used as a reference for comparing primary and mBC mutational profiles. Twelve genes (TP53, PIK3CA, GATA3, ESR1, MAP3K1, CDH1, AKT1, MAP2K4, RB1, PTEN, CBFB, and CDKN2A) were identified as significantly mutated in mBC (false discovery rate [FDR] < 0.1). Eight genes (ESR1, FSIP2, FRAS1, OSBPL3, EDC4, PALB2, IGFN1, and AGRN) were more frequently mutated in mBC as compared to eBC (FDR < 0.01). ESR1 was identified both as a driver and as a metastatic gene (n = 22, odds ratio = 29, 95% CI [9–155], p = 1.2e-12) and also presented with focal amplification (n = 9) for a total of 31 mBCs with either ESR1 mutation or amplification, including 27 hormone receptor positive (HR+) and HER2 negative (HER2−) mBCs (19%). HR+/HER2− mBC presented a high prevalence of mutations on genes located on the mechanistic target of rapamycin (mTOR) pathway (TSC1 and TSC2) as compared to HR+/HER2− eBC (respectively 6% and 0.7%, p = 0.0004). Other actionable genes were more frequently mutated in HR+ mBC, including ERBB4 (n = 8), NOTCH3 (n = 7), and ALK (n = 7). Analysis of mutational signatures revealed a significant increase in APOBEC-mediated mutagenesis in HR+/HER2− metastatic tumors as compared to primary TCGA samples (p < 2e-16). The main limitations of this study include the absence of bone metastases and the size of the cohort, which might not have allowed the identification of rare mutations and their effect on survival.
This work reports the results of the analysis of the first large-scale study on mutation profiles of mBC. This study revealed genomic alterations and mutational signatures involved in the resistance to therapies, including actionable mutations.
| Breast cancer often results in poor outcomes after it has metastasized to distant organs, but, while primary breast tumors have been extensively characterized at the molecular level, metastatic lesions are poorly understood.
This study aims to characterize the mutational landscape of metastatic breast cancer by performing and analyzing whole-exome sequencing of a large collection of metastatic breast tumors and corresponding blood samples.
Understanding of the mutational landscape of metastatic tumors should open new avenues for assessing resistance to therapy and developing better treatments.
The authors generated a large collection of whole-exome sequencing data from the DNA of breast cancer metastases and from each patient’s corresponding unmutated DNA in order to identify mutations and gene copy number alterations specific to the tumors.
The bioinformatics analyses identified recurrently mutated genes in metastatic tumors and revealed the genes specifically involved in metastatic disease by comparing their mutational frequency to those of primary breast tumors.
The study allowed identification of the affected genes and of mutational signatures that were more prevalent in metastatic as compared with primary tumors and that may be involved in the resistance to therapies.
The identification of mutational and copy number alterations specifically involved in breast cancer metastasis demonstrated that tumors evolve under the pressure of therapy.
Characterization of mutations and copy number alterations in metastatic lesions in addition to primary tumors should help to tailor treatment for patients, with the potential for improved clinical outcomes.
| Major efforts have been made to characterize early breast cancer at the genomic level [1,2]. These efforts have led to extensive description of genomic alterations involved in tumorigenesis or tumor progression of early breast cancer. These studies report that early breast cancer includes a large number of rare segments characterized by actionable genomic alterations such as PIK3CA mutations, ERBB2 amplification, FGFR1 amplification, CCND1 amplification, AKT1 mutations, and GATA3 mutations [1,2]. Follow-up studies report that C>T mutations at CpG sites are the major mutational pattern in early breast cancer [3]. Although sequencing of primary breast cancer has provided insight into the biology of early malignancy, around 80% of the patients presenting with such a disease will never relapse after conventional therapy. Therefore, understanding the biology of early breast cancer will not help in deciphering the specificities of the lethal disease or translate into treatment advances. Recent data from different types of cancer have suggested that there is a strong heterogeneity between primary tumors and metastases and that genomic profiles of metastases could dramatically differ from primary tumors. Gerlinger and colleagues have shown that only 30% of the mutations are shared between different tumor sites of kidney cancers [4]. Also, Haffner and colleagues have shown that lethal prostate cancer can derive from a minority subclone of the primary tumor [5]. There is therefore a need to extensively describe the genomic alterations observed in metastatic breast cancers in order to identify pathways involved in drug resistance and metastatic processes and to generate new strategies to treat these patients. To this end, we have performed whole-exome sequencing of 216 pairs of metastatic breast cancers and blood and report on the mutational landscape associated with lethal malignancy.
The following methodology was specifically developed for this analysis and did not follow an established protocol or analysis plan.
Metastatic breast cancer patients who underwent a biopsy in the context of the SAFIR01 [6] (NCT01414933), SAFIR02 (NCT02299999), SHIVA [7] (NCT01771458), and MOSCATO (NCT01566019) prospective trials were potentially eligible for this study. These French multicenter trials used high-throughput genome analysis on fresh frozen tumor biopsies as a therapeutic decision tool for metastatic cancer patients, with solid cancers (SHIVA and MOSCATO) or specifically with breast cancer (SAFIR01 and SAFIR02). SAFIR01 included patients with metastatic breast cancers resistant to therapy, and SHIVA and MOSCATO included patients with metastatic cancers eligible for phase I trial, while SAFIR02 included patients with metastatic breast cancers who were starting first- or second-line chemotherapy. Details of each trial are given in S1 Text. Exclusion criteria for the whole-exome sequencing analysis were defined as follows: small or no quantity of tumoral DNA, <30% cancer cells on the biopsy sample (from frozen tissue), and no blood sample available. With these criteria, we identified 86 tumor-normal pairs from patients included in the SAFIR01 trial, 80 pairs in the SAFIR02 trial, 35 pairs in the SHIVA trial, and 15 pairs in the MOSCATO trial (S1 Table). All patients gave their informed consent for translational research and genetic analyses of their somatic DNA. All the studies were approved by the relevant IRBs. Overall, whole-exome sequencing for a total of 216 pairs of metastatic tumor and unmutated DNA derived from corresponding blood samples was performed using Illumina technology. Estrogen (ER) and progesterone (PR) receptors were considered positive if >1% of the cancer cells were stained or when the case was reported positive in the case report form of the trial. HER2 status was determined locally.
Data were summarized by frequency and percentage for categorical variables and by median and range for continuous variables. Comparisons between groups were performed using the Mann-Whitney rank sum test for continuous variables and Chi square or Fisher's exact test for categorical variables. Overall survival (OS) was estimated by using the Kaplan-Meier method, and univariate analyses were performed using the log-rank test. OS was defined as the delay between the inclusion in the trial and death. Patients who were alive were censored at last follow-up news. The Cox proportional hazard regression model was used for multivariate analysis. All variables associated with p < 0.05 on univariate analysis were included in the model. All statistical tests were two sided, and differences were considered statistically significant when p < 0.05. Stata 13.0 software (StatCorp, College Station, Texas) or R version 3.2.2 were used for the statistical analyses. False discovery rate (FDR), used for correcting p-values for multiple hypothesis testing, was computed using the Benjamini-Hochberg procedure.
Genomic DNA was captured using Agilent in-solution enrichment methodology with their biotinylated oligonucleotides probes library (SureSelect All Exon V5, Agilent, or SureSelect Clinical Research Exome, Agilent), followed by 75-base paired-end massively parallel sequencing on Illumina HiSeq2500, HiSeq4000, or NextSeq500 (S2 Table). For detailed explanations of the process, we refer the reader to the publication by A. Gnirke and colleagues [8]. Sequence capture, enrichment, and elution were performed according to the manufacturer’s instruction and protocols (SureSelect, Agilent) without modification. Briefly, 600 ng of each genomic DNA was fragmented by sonication and purified to yield fragments of 150–200 bp. Paired-end adaptor oligonucleotides from Illumina were ligated on repaired, A-tailed fragments and then purified and enriched by 4–6 PCR cycles. Five hundred ng of these purified libraries was then hybridized to the SureSelect oligo probe capture library for 24 h. After hybridization, washing, and elution, the eluted fraction was PCR amplified for 10–12 cycles, purified, and quantified by qPCR to obtain sufficient DNA template for downstream applications. Each eluted-enriched DNA sample was then sequenced on an Illumina HiSeq2500/4000 or NextSeq500 as paired-end 75 b reads. Image analysis and base calling were performed using Illumina Real Time Analysis Pipeline version 1.12.4.2 with default parameters. Mean coverage was 83 +/− 18X for normal blood samples and 122 +/− 15X for tumor samples, with respectively 87% (77%–93%) and 90% (85%–95%) of the targeted regions covered at 20X or more (S2 Table).
Fastq files were aligned to the reference genome hg19 with the BWA mem algorithm [9]. After alignment, the BAM files were treated for PCR duplicate removal and then sorted and indexed with Picard for further analyses. Base recalibration and local realignment around indels were done with GATK [10]. For defining somatic mutations, we used the Mutect [11] (version 1.1.7) algorithm for identifying substitutions and the Scalpel [12] algorithm (version 0.5.2) for identifying small insertions and deletions (indels). Indels occurring in regions with a high number of point mutations detected by Scalpel were filtered out using the GATK VariantFiltration tool with parameters set to 3 mutations in a window of 35 bp. We kept indels of a size lower than 35 bp. We then merged the output of Mutect and Scalpel and further filtered for mutations organized in a cluster of 3 mutations or more in a window of 35 bp using the GATK VariantFiltration tool. We defined the final list of somatic mutations with the following filters: frequency of the reads with the altered base in the tumor > 10%; number of reads with the altered base in the tumor sample ≥ 5; frequency of the reads with the altered base in the normal sample < 2%; number of reads with the altered base in the normal sample ≤ 3; and total coverage in normal and tumor samples ≥ 10. The resulting somatic mutations were annotated with the snpEff and snpSift algorithms [13], and we selected somatic mutations occurring in coding regions only. We removed variants that were also detected in at least one normal sample in our cohort or annotated as known polymorphisms (reported by 1000 Genomes or the ESP databases) unless the variant was also reported in Catalogue of Somatic Mutations in Cancer (COSMIC) [14] or ClinVar (http://www.ncbi.nlm.nih.gov/clinvar/). In order to control for possible biases due to the difference in bait territories from the two capture kits, we verified the mutations that were unique to one bait territory and found that 50 mutations involving 39 genes were unique to one of the bait but none of these mutations affected significantly mutated genes. We filtered six indels after manual inspection. We manually added 2 hotspot mutations (1 His1047Arg PIK3CA [COSM94986] and 1 Glu349* TP53 [COSM140784]) that were originally identified in the tumor in the clinical trial screenings and that were filtered by the somatic mutation filters because of the high frequency of the altered allele in the blood samples (respectively four supporting reads for an allele frequency of 0.022 and seven supporting reads for an allele frequency of 0.11), probably due to circulating tumor DNA. The list of mutations is reported in S3 Table. We computed the cancer cell fraction (CCF) of each mutation using the following steps: we first estimated the tumor purity with Sequenza [15] as well as the copy number at the mutated locus and the number of mutated alleles, as estimated by the altered reads allelic fraction [15]. We then computed the CCF of each mutation using the predicted tumor cellularity by Sequenza, the reference and variant allele read counts at the corresponding chromosomal position, and the estimated copy number at the locus following the framework previously proposed by Carter and colleagues [16]. Mutations were classified as clonal if the 95% confidence interval of the CCF overlapped 1 and as subclonal otherwise. To identify significantly mutated genes, we used the Mutation Significance (MutSig) [17], Mutational Significance in Cancer (MuSiC) [18], and Driver Genes and Pathways (drGAP) [19] algorithms. We defined significantly mutated genes as those with an FDR < 0.1 according to the MutSig algorithm that takes into account more parameters for identifying drivers than the other two algorithms, including gene size, background mutation rate, and replication timing.
For deriving somatic copy number variations from whole-exome sequence data, we used the following strategy: we first computed the normalized ratio of reads between each tumor and corresponding normal sample using the package ExomeCNV in R and created the segmented profiles with the DNAcopy package. For defining amplifications and deletions, we used the Gistic2 algorithm [20] with the following thresholds for the log2 ratios: amp > 0.3 and del < −0.3. Gistic2 was run including all the samples and specifically for the HR+/HER2− samples and for the HR−/HER2− samples in order to control for disease subtypes. Focal peaks are listed in S4 Table.
De novo mutational signature analysis was done using the Matlab Welcome Trust Sanger Institute’s signature framework. We used the deconstructSigs R package [21] to determine the contribution of the known signatures that explain each sample mutational profile with more than 50 somatic mutations. We considered the 13 signatures (Signatures 1, 2, 3, 5, 6, 8, 10, 13, 17, 18, 20, 26, and 30) operative in breast cancer as defined in COSMIC (http://cancer.sanger.ac.uk/signatures/matrix.png). A signature was defined as operative or predominant if its contribution to the mutational pattern was respectively >25% (or >100 mutations) or >50%.
Somatic mutations for breast cancer TCGA cohort were extracted from the genome.wustl.edu_BRCA.IlluminaGA_DNASeq.Level_2.5.3.0.somatic.maf file available for download on the TCGA data matrix website, with somatic mutations available for primary tumors of 772 patients. We extracted ER, PR, and HER2 status from the clinical file downloaded from the TCGA data matrix, retrieving 419 HR+/HER2−, 100 HR−/HER2−, and 145 HER2+. In order to fairly compare mutational loads between TCGA and the metastatic cohort, we downloaded raw data for 33 randomly selected TCGA patients and processed the BAM files with the same pipeline described in this manuscript. We found that the number of mutations identified by our pipeline and by the TCGA pipeline was very similar (S1 Fig, linear regression R2 = 0.98, p < 2e-16). We also verified the identity of mutations called by the two pipelines and found that 80% of the mutations were common to the two pipelines (S2 Fig). Therefore, we used the somatic mutations as defined in the TCGA maf file for comparing the mutation frequencies of the genes.
The population analyzed in the current study included 216 pairs of tumor and normal blood DNA from patients with metastatic breast cancer. Patients were classified in three subgroups according to hormone receptors (HRs; estrogen and progesterone receptors) and HER2 status (Table 1). One hundred and forty-three patients (66%) presented with HR+/HER2− breast cancer, 51 (24%) with triple-negative breast cancer, and 14 (6%) with HER2-overexpressing breast cancer. Ninety-four percent of the patients had received prior chemotherapy, and 120 (84%) of the patients with HR+/HER2− disease had received prior endocrine therapy.
We identified 12 driver genes using the MutSig algorithm (FDR < 0.1) (Fig 1, S5 Table). Ten of these genes (TP53, PIK3CA, GATA3, MAP3K1, CDH1, AKT1, MAP2K4, PTEN, CBFB, and CDKN2A) have been previously shown to be frequently mutated in primary breast cancers (>2%, TCGA). In particular, TP53 was mutated in 27% of HR+/HER2− metastatic breast cancer (mBC) as compared to 20% in HR+/HER2− early breast cancer (eBC) (Fisher Exact Test p = 0.13) while PIK3CA was mutated in 37% of the HR+/HER2− mBC and in 40% in eBC.
Two of the driver genes observed in mBC (ESR1 and RB1) were infrequently mutated in primary tumors (<1% of HR+/HER2− eBC [TCGA]). Twenty-four mutations of ESR1 were identified (1 synonymous, 2 indels, and 21 missense mutations) for a total of 22 mBCs, and these included 22 mutations in 20 out of 143 HR+/HER2- mBCs (14%). All ESR1 mutations occurred in the hormone receptor domain (S3 Fig) and included mutations in previously reported hotspots [22–24], as well as 2 new mutations (S3 Table). All of these 22 patients had received prior endocrine therapy. RB1 was mutated in 7 out of 143 HR+/HER2− mBCs (5%) and 3 out of 51 HR−/HER2− mBCs (6%). Most of the mutations were disruptive, leading to truncated proteins (5 nonsense mutations, 3 splice sites, 1 indel, and 2 missense mutations, S4 Fig). When considering the estimation of the percentage of tumor cells harboring the mutation, i.e., CCF, we found that ESR1 and RB1 mutations were mostly identified as subclonal (ESR1: 14/21 mutations [67%]; RB1: 5/10 mutations [50%]). In comparison, PIK3CA and TP53 mutations were identified as subclonal for respectively 32% and 37% of their nonsynonymous mutations.
Using a FDR < 0.1, 199 genes out of 1,569 genes tested were more frequently mutated in mBC (n = 216) as compared to eBC (TCGA) (S6 Table). When a FDR < 0.01 was applied, 8 genes (ESR1, FSIP2, AGRN, FRAS1, IGFN1, EDC4, OSBPL3, and PALB2) were found to be more frequently mutated in mBC as compared to eBC (Fig 2). None of these genes, except ESR1, were identified as a driver using MutSig. However, OSBPL3 and PALB2 were both identified as drivers by MuSiC and drGAP at an FDR < 0.1 (S5 Table). PALB2 was mutated in eight (4%) samples, while only one (0.1%) eBC was mutated in TCGA (FDR for enrichment in mBC = 0.006). Out of the 8 PALB2 mutations, 5 were found in HR+/HER2− mBC. None of the cases with PALB2 somatic mutations presented with a PALB2 deleterious germline polymorphism in the other allele. We analyzed outcome data for comparing the OS of patients with metastatic tumors carrying at least one of the mutations enriched in the metastatic setting (n = 76) to the rest of the population (n = 140). Results of the univariate and multivariate analyses are reported in S7 and S8 Tables. In a multivariate analysis, mBC with at least one mutation in the 8 genes enriched in the metastatic setting presented a 2-fold increase in the hazard of death (hazard ratio = 1.97, 95% CI: 1.34–2.89, p = 0.001). Survival curves are reported in Fig 3.
As this analysis might be biased by the difference in distribution of HR and HER2 subtypes between eBC and mBC, we also focused the analysis on the HR+/HER2− subtype (n = 143), in which 278 genes were more frequently mutated in mBC as compared to eBC (FDR < 0.1, S6 Table). Several of these genes were considered actionable. TSC1 and TSC2 were mutated in five (3.5%) and four (2.8%) samples, respectively (Fig 4). Overall, 6.3% of HR+/HER2− mBC presented an alteration in TSC1/2 as opposed to 0.7% of HR+/HER2− eBC (TCGA, p = 0.0004). Other actionable genes were more frequently mutated in HR+ mBC with an FDR < 0.1. These include ERBB4 (nine missense mutations, including the two mutations COSM4764538 and COSM1015992, involving eight mBCs [five HR+/HER2−]), NOTCH3 (eight missense and one splice site mutation(s) involving seven mBCs [four HR+/HER2−]), ALK (five missense and two splice site mutations in seven mBCs [six HR+/HER2−]), EZH2 (two missense and one splice site mutation(s), including COSM220530, involving three HR+/HER2− mBCs) and BRAF (four missense mutations, including one COSM476, involving four mBCs [three HR+/HER2−]). The consequence of these mutations on the activity of the encoded proteins was difficult to assess as, even though ERBB4 and NOTCH3 mutations were all missense mutations (except for one splice site mutation in NOTCH3), they were located in different protein domains with no apparent hotspot (Fig 4).
First, in order to identify a potential metastatic-specific mutational signature, we performed de novo mutational signature analysis that revealed five signatures operative in metastatic and primary breast cancer [25], but none of these signatures were specific to the metastatic setting (S1 Text, S9 Table and S5 and S6 Figs). We then assessed the contribution of 13 mutational signatures [25] in 118 metastatic samples and 278 primary tumors from TCGA presenting >50 mutations (S10 Table). Among the 13 signatures previously identified as operative in primary breast cancer, the most represented signatures in the metastatic samples were signature 1, related to aging; signatures 2 and 13, related to APOBEC3B activity; signature 3, associated with failure of DNA double-strand break-repair by homologous recombination; and signature 6, associated with defective DNA mismatch repair (Fig 5). While the identity of the signatures remained the same between primary and metastatic samples, their contribution dramatically changed, especially in the HR+/HER2− subtype (Fig 6 and S7 and S8 Figs). Of note, the signatures related to the APOBEC3B enzyme (signatures 2 and 13) contributed to 58.8% of the mutations of the HR+/HER2− metastatic tumors as compared to only 31.9% in the primary TCGA samples (p < 2e-16), confirming previous work demonstrating a link between APOBEC-mediated mutagenesis and the acquisition of subclonal mutations [26].
Gistic2 analysis using sequence-based levels reported regions previously described to drive oncogenesis of primary tumors including amplified genes CCND1, ERBB2, and MYC and lost gene PTEN (S4 Table). In addition to these previously reported gene amplifications and deletions, the current study identifies a focal amplification of the ESR1 locus confirming the mutational pattern of the gene. ESR1 was amplified in 7 HR+/HER2− mBCs for a total of 27 HR+/HER2− mBCs (19%) with either ESR1 mutation or amplification (Fig 1). Additionally, RB1 was lost in 2 HR+/HER2− mBCs for a total of 9 samples (6%) with either RB1 mutation or homozygous deletion.
Finally, we computed two indices to describe the chromosomal instability of the metastatic samples based on the copy number analysis as previously described [27]: a global genomics index (GGI) and the number of breakpoints per sample (S1 Table). We found that the number of mutations per tumor did not correlate with either the GGI or the number of break points (S9 Fig). We also verified that the mutational load and the chromosomal instability were not affected by the tumor cell content of the samples. While we found that there was no correlation between cellularity and estimated chromosomal instability, we found a positive correlation between the percentage of tumor cells and the number of mutations (Pearson’s cor = 0.16, p = 0.02). However, among the five samples with no nonsynonymous mutations, only two samples had <50% tumor cells, while the other three had >70% tumor cells.
In the present manuscript, we have described the mutational landscape of 216 mBCs. This study reported genes significantly mutated in mBC and genes significantly more mutated in mBC as compared to eBC. HR+/HER2− mBC presented the most differences with HR+/HER2− eBC, including an increased mutational signature linked to APOBEC3B activity and a higher prevalence of actionable genes that may represent new strategies for mBC treatment.
Using a stringent definition (MutSig, FDR < 0.1), the current study identified ESR1 and RB1 as driver genes that are specific to mBCs. Previous studies [23,24,28] have already reported that ESR1 mutations could be acquired during the disease evolution and could mediate resistance to endocrine therapy. In the present study, we confirm that mutation of ESR1 is the most frequent “metastasis-specific” mutation observed in mBC. As expected, all the 22 patients who presented ESR1 mutations were ER+ and resistant to endocrine therapy. ESR1-mutated mBC could be a genomic segment defining an unmet medical need, for which fast-track approval of new agents is required. Rb1 is a tumor suppressor protein involved in cell cycle and phosphorylated by CDK4. The protein is required for the bioactivity of palbociclib (CDK4 inhibitor), a drug recently approved to treat HR+/HER2− mBC [29]. The present study identifies RB1 mutations, most of them pointing to a loss of function of the protein, as driver alterations in mBCs; while this gene is almost never mutated in HR+/HER2− eBC (<1%), it was found mutated in 5% of the HR+/HER2− mBC (p = 0.008, FDR = 0.09). This finding suggests that a subset of HR+/HER2− mBC is deficient for RB1 and could present a primary resistance to CDK4 inhibitors. If validated, this finding suggests that RB1 mutations should be assessed on metastatic samples before starting CDK4 inhibitors.
Several genes were more frequently mutated in mBC as opposed to eBC but did not meet the criteria for drivers using the MutSig algorithm. PALB2 is a partner of BRCA1/2 and is involved in Fanconi anemia. Heterozygous loss-of-function mutations in PALB2 have been shown to be a risk factor for breast cancer, while PALB2 germline mutations have recently been associated with a poor outcome [30]. Several studies have suggested that PALB2-deficient cancers could be sensitive to PARP inhibitors. In the present study, PALB2 somatic mutations were found in 4% of metastatic samples (n = 8), while the gene is mutated in only 0.1% of eBC (FDR = 0.006). The present results suggest that there is a population of PALB2-deficient mBC in which PARP inhibitors could be evaluated. Genes located on the mTOR pathway (TSC1 and TSC2) were more frequently mutated in HR+/HER2− mBC (6%) as opposed to HR+/HER2− eBC (0.7%). All these mutations were observed in patients previously treated with endocrine therapy, suggesting that it could be a mechanism of resistance. mTOR inhibitors (everolimus) have been approved in HR+/HER2− mBC [31]. While this drug prolongs progression-free survival (PFS) for a majority of patients, only a few percent of them are outlier responders to this drug, and there is currently no molecular alteration that explains such cases. Further studies should evaluate whether the subset of patients with genomic alterations on mTOR pathways (TSC1 and TSC2) could be outlier responders to everolimus. Other actionable genes were more frequently mutated in HR+ mBC, including ERBB4, NOTCH3, and ALK.
Analysis of mutational processes did not identify any signature specific to the metastatic setting but revealed a high increase in APOBEC-mediated mutations in HR+/HER2− mBC as compared to eBC. As for metastatic-specific mutations identified, this might also present a mechanism of resistance to therapy that needs further careful investigation.
The study included 216 sample pairs classified in three different classes based on hormonal receptor expression in the tumors, the largest group being HR+/HER2− (n = 143). Although this is the largest effort for profiling the mutational landscape of mBC to this day, the size of the cohort presents a limitation to the identification of rare events, especially in the triple-negative and HER2+ groups. Our ability to provide an exhaustive picture of mutational profiles of mBC may be limited by two main biases. The first bias comes from the absence of bone metastases in the study, due to the difficulty of extracting DNA from these lesions. A second potential bias may come from our inability to identify those mutations leading to a disease so aggressive that the patients will not be eligible for trial recruitment, preventing an exhaustive picture of the mutational profiles of mBC. However, recent studies on first-line therapies for advanced breast cancer have shown that early death is limited [32,33], and it is therefore unlikely that it has dramatically impacted the analysis. Additionally, there is a chance that mutations enriched in metastasis might be subclonal driver events and therefore might not be such good drug targets. The comparison of gene mutational prevalence between mBC and eBC suffers from two limitations. First, it would have been ideal to directly compare the mBCs with their corresponding primary tumor profiles, but this was not possible because of the obvious reason of sample availability. Second, we used mutational profiles of TCGA tumors that were identified by the TCGA team, whereas it would have been ideal to run the pipeline used for mBC on the TCGA data. Although we controlled for major biases, the use of different bioinformatics pipelines may have some unexpected consequences. It should also be noted that the copy number analysis is limited by the nature of the sequencing data, which does not allow for a uniform coverage of the genome. Finally, the survival analysis based on the mutational status of the eight metastasis-specific genes was independent from any other parameters such as mutational load or mutational signature contribution, making it difficult to establish a causal link between mutated genes and prognosis.
The dataset and the accompanying analysis described in this study provide a better understanding of the genetic basis of mBC and how much it differs from that of primary breast tumors. This study demonstrated that profiling metastatic cancer can be a major step in defining optimal treatments for patients, as new mutation events and processes may arise during cancer treatment. Follow-up studies will be essential for validating resistance mechanisms identified in this study.
|
10.1371/journal.pcbi.1006732 | Identifying the mechanism for superdiffusivity in mouse fibroblast motility | We seek to characterize the motility of mouse fibroblasts on 2D substrates. Utilizing automated tracking techniques, we find that cell trajectories are super-diffusive, where displacements scale faster than t1/2 in all directions. Two mechanisms have been proposed to explain such statistics in other cell types: run and tumble behavior with Lévy-distributed run times, and ensembles of cells with heterogeneous speed and rotational noise. We develop an automated toolkit that directly compares cell trajectories to the predictions of each model and demonstrate that ensemble-averaged quantities such as the mean-squared displacements and velocity autocorrelation functions are equally well-fit by either model. However, neither model correctly captures the short-timescale behavior quantified by the displacement probability distribution or the turning angle distribution. We develop a hybrid model that includes both run and tumble behavior and heterogeneous noise during the runs, which correctly matches the short-timescale behaviors and indicates that the run times are not Lévy distributed. The analysis tools developed here should be broadly useful for distinguishing between mechanisms for superdiffusivity in other cells types and environments.
| Cells must move through their environment in many different biological processes, from wound healing to cancer invasion to the development of an embryo. There are different ways for cells to explore the physical space around them—ranging from moving along a straight path at constant speed to executing a random walk where the cell changes direction at every time point. Understanding what mechanisms are driving motility patterns in different cell types is important for identifying possible treatments for disease. We found that mouse fibroblast cells moving on a two-dimensional substrate were super-diffusive, meaning that they were able to cover distance faster than a random walk but not as fast as a straight walk. Traditional analysis of cell trajectories was not well-suited to distinguish between different possible mechanisms for super-diffusivity, so we developed a new tool to examine cell trajectories and distinguish between mechanisms. We found that mouse fibroblasts were super-diffusive due to a combination of large fluctuations in speed and “run-and-tumble” behavior, where cells move in a straight line for a while before changing direction rapidly. We expect this tool to be useful for analyzing motion in many other cell types.
| Cell motility is an integral part of biological processes such as morphogenesis [1], wound healing [2], and cancer invasion [3]. But what are the rules that govern how cells move? Cell migration involves a multitude of organelles and signaling pathways [4] and therefore a fruitful, bottom-up approach studies correlations between cell motion and sub-cellular processes that govern motility, including surface interactions [5], integrin signaling pathways [6], or formation of focal adhesions [7].
An alternate approach with recent successes is to develop simple models at the cellular scale that can help identify a coarse-grained set of rules that govern cell migration in specific cell types. One such class of models, composed of self-propelled (SPP) or active Brownian particles [8] has been used to make predictions about the motion of biological cells in many contexts, including density fluctuations [9], formation of bacterial colonies [10], and both confined [11], and expanding monolayers [12].
These SPP models represent each cell as a particle that moves by generating active force on a substrate, which acts along a specified vector θ ^. Therefore, the parameters for the model specify both the magnitude of the force as well as how the orientation of the force changes with time. Given the ubiquity and usefulness of these models, one would like to have a standard framework for extracting these parameters from experimental data for all trajectories. In the past this has often been accomplished by analyzing ensemble-averaged features of cell trajectories.
One such quantity is the time averaged mean-squared displacement (MSD), which is the squared displacement between positions r → ( t ) and r → ( t + d t ) averaged over all starting times t and the ensemble of trajectories. This yields the MSD as a function of timescale, 〈(r(t + dt)) − r(t))2〉 ∝ dtα. Ballistic motion, which corresponds to a cell moving in a straight line at constant speed, corresponds to α = 2. Diffusive motion, where a cell executes a random walk with no time correlation in orientation, corresponds to α = 1. In non-active matter at low densities, thermal fluctuations generically induce diffusive behavior at long timescales. In contrast, many cell types, including T-cells [13], Hydra cells [14], breast carcinoma cells [15], and swarming bacteria [16] display super-diffusive dynamics, defined as trajectories with a MSD exponent between 1 < α < 2.
Several authors have proposed explanations for why super-diffusive migration might be beneficial in biological systems. For example, super-diffusive trajectories are well known for being the optimal search strategy for randomly placed sparse targets [17, 18], and have been found in animal foraging and migration patterns in jellyfish [19], albatross, and bumblebees [20]. In the context of cell biology, superdiffusive migration implies that cells are covering new areas more quickly than they would if they were executing a simple random walk.
Although super-diffusive dynamics are commonly observed in in vitro experiments, the fundamental mechanism that generates anomalous diffusion in cell trajectories remains unclear. Pinpointing the mechanism would allow biology researchers to better isolate the signaling pathways that govern these processes.
Although one might think that simply including the effects of persistent active forces generated by cells would change the long-time behavior, it turns out that standard self-propelled particle models exhibit a fairly sharp crossover from ballistic to diffusive motion, with no extended superdiffusive regime. Since SPP models are commonly used to model cells and superdiffusive dynamics are commonly observed in experiments, we would like to identify the mechanism generating superdiffusitivity to improve the ability of these models to capture cellular phenomena.
Standard SPP models include smoothly varying persistent random walkers and standard run-and-tumble particles (RTP) [21]. Persistent random walkers obey the following equations of motion for the cell center of mass ri and the orientation angle θi:
∂ t ri→= v 0 θ ^ i , (1) ∂ t θ i = η ( t ) , (2)
where η(t) is a Gaussian white noise (〈η(t)η(t′)〉 = 2Dr δ(t − t′)). In a standard persistent random walk, the speed v0 and the rotational diffusion coefficient Dr, which controls the strength of fluctuations in orientation, are constant. In a standard run-and-tumble model, particles are ballistic during runs, ∂tθi = 0, followed by tumbling events where large changes in orientation occur. Variations of run-and-tumble models are characterized by the distribution of times particles remain in the run state.
Two different classes of modifications to SPP models have been highlighted as being able to generate super-diffusive behavior on long timescales. The first modification is a heterogeneous speed model, which draws rotational diffusion coefficients and particle speeds from distributions [15, 22]. While persistent random walk models transition from ballistic to diffusive behavior at one characteristic timescale, heterogeneous speed models possess a heterogeneous distribution of crossover timescales, which generates an MSD with a broad superdiffusive regime, though the system becomes diffusive on timescales longer than 1 / D r m i n.
The second modification is a Lévy walk model, which is a run-and-tumble model where particles have power law distributed run times:
P ( τ ) = μ τ o ( 1 + τ / τ o ) 1 + μ , (3) ⟨ τ ⟩ = τ o μ - 1 , (4)
with P(τ) the distribution of run times with mean < τ > for μ > 1. [23]. In contrast to the heterogeneous SPP model, super-diffusivity generated by Lévy walks is not transient, so that the long-time MSD scaling exponent is constant: MSD ∝ dt3−μ.
So which of these models is the “right” one for a given cell type? By analyzing ensemble-averaged statistics such as the MSD and the velocity autocorrelation function (VACF), one group of researchers was able to show that heterogeneous motility models matched data from breast cancer carcinoma cells [15]. This model, based on an autoregressive process (AR-1), uses a Bayesian inference method to extract activity and persistence from cell trajectories. However, these quantities do not directly correspond to physical quantities such as cell speed or rotational diffusion. Effects of cell heterogeneity were also explored in human fibrosarcoma cells by Wu et al., where the authors show that these effects are sufficient to explain non-Gaussian velocity distributions [24], similar to those we observe in mouse fibroblast cells. The authors also investigate anisotropic contributions, modeling 3D human fibrosarcoma trajectories with a 3D anisotropic persistent random walk.
Differentiating between inherently anisotropic behavior and cell response to external cues such as chemotaxis is another difficult problem, investigated in T cells by Banigan et al. using a unique model that features a mix of passive Brownian particles and persistent random walkers [25]. Other efforts evaluated a different ensemble-averaged quantity, the probability displacement distribution, and used that data to suggest that T-cells were undergoing generalized Lévy walks [13]. We would like to better understand whether these ensemble-averaged quantities are in fact a unique identifier of the underlying mechanism for superdiffusivity. Moreover, we also seek to develop a systematic procedure for using experimental data to constrain both the appropriate mechanism and the optimal model parameters for a specific subtype. To this end, we use automated tracking software to analyze over 1000 mouse fibroblast trajectories and, using the work of Metzner and colleagues as inspiration, extract parameters for a generalized model based on persistent random walkers. We demonstrate that some ensemble-averaged statistics, such as the MSD and VACF, can not distinguish between mechanisms for superdiffusivity.
In order to better distinguish, we begin with a very general model for cell dynamics. Although standard SPP models have only two fit parameters, average cell speed v0 and average rotational noise Dr, in principal a generalized SPP model could have arbitrary distributions for cell speed P(v0) and rotational diffusion P(Dr) with arbitrary correlations between them. The heterogeneity motility model from [15] is the limit of such a model with Gaussian-distributed P(v0) and P(Dr), while a standard Lévy walk is the limit with a constant v0 and a specialized bimodal P(Dr). Generalized Lévy walks such as those studied in [13] have additional parameters. Because this is such a large parameter space, we first constrain the functional form of these distributions using specific features of single cell trajectory statistics. We find that the mouse fibroblast data are consistent with run-and-tumble dynamics but the run times are not power-law distributed, confirming that in mouse fibroblasts the mechanism for superdiffusivity is heterogeneous dynamics and not Lévy walk statistics. The toolkit we have developed here should be useful for pinpointing the origin of superdiffusivity in many other cell types.
Cell motility data was collected from C3H10T1/2 mouse fibroblast cells (ATCC). Although the cells were cultured on gold-coated shape memory polymer substrates, which in principle can be programmed to form anisotropic nanowrinkles [26], all of the data in this manuscript is from cells cultured on “control” substrates that remain flat throughout the entire experiment, as our goal is to characterize the origin of superdiffusivity in this most simple case. While the data used in this manuscript are from an experimental protocol with a temperature shift, Baker et al. saw very similar superdiffusive trajectories in systems with no temperature changes [27], indicating that the superdiffusivity we observe here was not generated by or dependent on temperature changes. Future work will analyze behavior on more complicated wrinkled or transitioning substrates. Cell nuclei were labeled with Hoechst dye and cell motility imaged by time-lapse microscopy. The resultant image stacks were analyzed using the ACTIVE image analysis package to track nuclei centers-of-mass [27]. See S1 Text for more information on substrate preparation.
Cell motility was characterized using statistical analysis of cell nuclei trajectories, including MSD, VACF and displacement probability distributions. Tumbling events were identified with a one dimensional Canny edge detection algorithm, as shown in Fig 1. This algorithm takes a time series of changes in orientation and classifies each timestep as either a “run” or “tumble.” Additional details on cell trajectory analysis can be found in S1 Text.
This manuscript focuses on two different models for non-interacting active particles. The first model is a Lévy walk with constant particle speed v0 at all timesteps. Particles execute ballistic runs with zero rotational noise for times τ drawn from the distribution in Eq 3 and a mean run time 〈τ〉 given by Eq 4.
The generalized SPP model has particles which follow the equations of motion seen in Eqs 1 and 2, however the parameters for each model are not constant in time. A particle is initialized with a random orientation and assigned an initial speed v0 and rotational diffusion Dr drawn from distributions P ( v 0 ) = | v 0 | σ v 2 e - ( v 0 - μ v ) 2 σ v 2 and P ( D r ) = 1 π σ D 2 e - ( D r - μ D ) 2 σ D 2. To account for possible correlations between the speed and rotational diffusion variables in our model, we utilize a copula modeling method [28]. First, we sample a bivariate normal distribution with a covariance matrix given by Σ = [ 1 p p 1 ], where p = 0 indicates no correlation and p = ±1 indicates full positive (negative) correlation. Then we use the standard method of inverse cumulative distribution functions to transform the marginal distributions into the distributions P(v0) and P(Dr) listed above. This results in a set of variables with a correlation between them parameterized by p, and also with the desired marginal distributions.
Following sampling v0 and Dr, we evolve the particle position and orientation for a time τ drawn from P ( τ ) = 1 τ 0 e τ / τ 0, where τ0 is the mean run time determined by experimental data. The particle then undergoes a tumbling event across one time step where Dr = 2π, where the value of rotational diffusion is chosen to approximate an event where the orientation is completely randomized. After the tumble a new v0 and Dr are assigned until the next tumbling event. In contrast to a Lévy walk or standard SPP model, motility parameters are varied in time to replicate the variations and changes in a biological environment.
For both models, particle trajectories are constructed by numerically integrating the equations of motion using a simple Euler scheme with a timestep dt = 0.1. For fitting purposes, we choose the natural timescale in our simulations equal to four minutes in experiments, which is the time between frame captures. In addition, we use the averaged goodness-of-fit of model MSD, VACF and displacement probability distributions compared to that of mouse fibroblast trajectories to determine optimal model parameters, shown in Table 1 and discussed later in the text.
Finally, we note the VACF for experimental data shows a sharp dropoff across one frame due to errors in reconstructing the nuclei centers caused by imaging noise and fluctuations in dye intensity. To replicate this feature we incorporate positional noise into both models through small Gaussian fluctutations. After particle trajectories are constructed, each position is changed by a vector δ r → = d r ϕ ^, where dr is drawn from a Gaussian distribution of variable width Δ and the direction ϕ ^ is chosen randomly from the unit circle. This replicates experimental error in reconstructing cell positions, and allows our model trajectories to match the mouse fibroblast data.
Previous reports have compared models to experimental data using ensemble-averaged statistics to confirm model validity such as the MSD and the VACF. Therefore, our first goal is to determine whether one of the existing models for explaining superdiffusive cell trajectories is a better fit to the experimental MSD and VACF data, shown by the red lines in Fig 2.
For comparison, we simulate a Lévy walk model with dynamics given by Eqs 3 and 4, as well as a generalized SPP with no Lévy-walk behavior, described in more detail below. With the best-fit parameters, we find that both models match the data equally well. As shown in Fig 2(B), the velocity autocorrelation function exhibits a sharp decrease after the first frame window, due to errors that we make in reconstructing the nuclei center of mass caused by imaging noise and fluctuations in the dye intensity. Therefore, we add an additional term to the model that shifts the particle position by a Gaussian distributed variable with zero-mean and variance Δ2 to account for this effect.
While the mean-squared displacement and velocity auto-correlation function are standard metrics for characterizing ensembles of trajectories, they may not be ideal for studying systems with superdiffusion. In an investigation of the Lévy walk properties of T-cells, Harris et al. study a quantity that reveals structures on shorter timescales: the probability for a cell to be at a displacement r(t) at time t [13]. They suggest that generalized Levy walks can be distinguished in part by collapsing these probability distributions with rescaled displacements ρ ( t ) = r ( t ) t γ, with γ significantly larger than the value of 1/2 expected for a persistent random walk. In their initial work characterizing Lévy walks, Harris and colleagues considered a wide range of Lévy walks as well as several other random walk processes, and finding the best match for T-cell trajectories was to a generalized Lévy walk. As seen in Fig 3, we find that the mouse-fibroblast data does collapse, with the best fit exponent γ = 0.69 ± 0.02 as shown in Fig 4. The best-fit standard Levy walk model collapses at γ = 0.58 ± 0.03, which is above the value expected for a persistent random walk but still lower than γ for mouse fibroblast cells. Importantly, the best-fit generalized SPP model also collapses at a similar value of γ = 0.67 ± 0.03, suggesting that such a collapse is not sufficient to uniquely identify Lévy walks as a mechanism for superdiffusivity.
Moreover, the functional form of the displacement probability distribution (PDF) P(r(t)) provides additional information. It is well-fit by a Gaussian curve, shown as an offset dashed black line in Fig 3(A) and 3, and a Fig 3(B) shows that non-Lévy version of the generalized SPP model also matches the shape of mouse fibroblast P(r(t)) very well. In contrast, Fig 3(C) shows that P(r(t)) for the best-fit standard Lévy walk model has a very different functional form, due to ballistic runs over relatively large distances. However, this mis-fit in the functional form does not rule out Levy walks as a possible mechanism, as it could be corrected by considering a generalized Lévy Walk with more parameters [13]. To truly distinguish between the two mechanisms, we need access to more granular details about the individual cell trajectories.
We next study single-cell trajectories. A generalized SPP model with arbitrary distributions for P(v0) and P(Dr) has an infinite number of parameters that we could never hope to constrain. As a first step to simplifying our model we constrain functional form of these distributions using experimental data through microscopic statistics, such as velocity and run-time distributions, calculated from single-cell trajectories. This is in contrast to ensemble and time-averaged macroscopic statistics such as MSD and VACF. As shown in Fig 5(A), we first construct a distribution of cell speeds, determined from the magnitudes of the displacement of nuclei centers-of-mass between image capture events. Our experimental data is consistent with a Gaussian distribution of cell velocities, or equivalently, a distribution of cell speeds of the form P ( v 0 ) = | v 0 | σ v 2 e - ( v 0 - μ v ) 2 σ v 2, where μv and σv are the mean and standard deviation, respectively, with estimates shown in Table 1. Therefore, we use this functional form in our generalized model. Next we estimate a distribution P(Dr) of rotational diffusion constants (Dr) from the distribution of turning angles, shown in Fig 5(B). Simple active Brownian systems with a single value of Dr will generate a Gaussian distribution of turning angles [21]. A Gaussian distribution of rotational noise broadens this distribution significantly. One can show the expected turning angle distribution in this case is a modified Bessel function of the second kind with an exponential tail, consistent with the numerical simulation data given by the red line in Fig 5(B). We were unable to match the mouse fibroblast turning angle distribution, which is given by the blue line in Fig 5(B) and has significant weight as the largest values of Δθ, with any Gaussian function for the rotational noise.
This suggests that mouse fibroblast cells may have a strongly bimodal distribution of rotational noises, further supported by intermittent run-and-tumble behavior seen in cell trajectories. We choose to capture this bimodal behavior with a noisy run-and-tumble model, where cells have a distribution P ( D r ) = 1 π σ D 2 e - ( D r - μ D ) 2 σ D 2 during runs, which are punctuated by tumbling events. Distribution parameters μD and σD, shown in Table 1, can be estimated from the distribution P(Dr) used to generate the run and tumble distribution of turning angles seen in Fig 5(B). In our implementation of this model we include possible arbitrary correlations between these distributions through the parameter p, ranging from fully correlated (p = 1) to anti-correlated (p = −1). We use the Canny algorithm described in the methods section to explicitly identify tumbling events, and the data points in Fig 5(C) show the distribution of times between such events. The red line in Fig 5(C) shows this is well-fit by an exponential distribution with τ0 ≈ 1 hour, and so in our model the distribution of run times τ is given by P ( τ ) = 1 τ 0 e - τ / τ 0. We note that this is a strong indication that the mouse fibroblasts are not well-described by a Lévy walk model with power-law distributed run times. Specifically, although we focus here on a standard Lévy walk model with fewer parameters than the generalized model used by Harris et al. [13], it is clear that adding additional parameters to our a Lévy walk will still not generate the P(τ) we observe in mouse fibroblasts. The magenta line in Fig 5(B) shows the distribution of turning angles for a noisy run-and-tumble model with the parameters identified above.
To confirm that the model parameters we have identified are robust, and to quantify their sensitivity, we vary model parameters around the microscopically determined values and quantify how much this changes their displacement probability distributions. Specifically, we use the linear regression goodness-of-fit parameter (R2) between P(r(t)) for mouse fibroblast and generalized model trajectories to characterize each parameter configuration and identify a best-fit between our model and mouse fibroblast statistics [29]. Using this method we are able to capture the functional form of P(r(t)) very well, as shown in Fig 3. We explored several additional methods to parameterize goodness-of-fit, but because the shape of P(r(t)) exhibited significant fluctuations as we swept parameter space, nonlinear fitting approaches were inconsistent. Therefore we focus on the more stable linear regression results here.
Happily, the configuration of parameters that best matches the macroscopic P(r(t)), located at μD = 0.09, σD = 0.002, μv = 1.2, σv = 0.8, p = 0, τ0 = 10, is very similar to those identified from microscopic statistics, indicating that the model is consistent with experimental results. A construction of the dynamical matrix as a Hessian in parameter space around this minima and subsequent analysis of local eigenvectors indicates that our system is most sensitive to perturbations in the mean velocity and mean rotational noise as shown in Fig 6(A), and relatively insensitive to correlations between Dr and v0 parameterized by p (Fig 6B) as well as mean run time τ0 (Fig 6C).
Both Lévy walks and heterogeneous SPP models are capable of generating superdiffusive trajectories. Previous studies have focused on one model or the other in order to identify possible mechanisms for superdiffusive cell trajectories.
We show that while both types of model are equally capable of matching the large-scale ensemble averaged statistics of mouse fibroblast cells, an analysis of single cell trajectories demonstrates that Lévy walks are not consistent with this data set, despite a very good scaling collapse of the probability displacement distribution with scaling exponent γ > 1/2. Instead, a careful analysis of turning angle distributions suggests these mouse fibroblasts exhibit heterogeneous speeds, with noisy run-and-tumble behavior.
Because superdiffusive cells are able to cover distance faster than diffusive counterparts, it would be useful to adapt the tools developed here to study many more cell types. For example, directed cell motion is known to be a signature of invasiveness in cancer cell lines [30], and it would be interesting to know if these cell types are altering the mechanisms or timescales for superdiffusion as they become more malignant. To that end, we have created a MATLAB software package for deploying these analyses on generic data sets [31], which can be used to quantify superdiffusive dynamics and distinguish between different mechanism behavior in cells and active matter.
Another important question is whether the tumbling events seen here are cell-autonomous or generated by cell-cell interactions. On the one hand, It is possible that the run and tumble behavior is at least partially cell-autonomous, although no biochemical mechanisms for such behavior have been identified in fibroblasts. To begin to investigate this question, it would be useful to correlate tumbling events with the dynamics of sub-cellular features such as spatio-temporal distributions of focal adhesions [32], Golgi bodies [33], or actin waves [30]. This would help us to understand which signaling networks and components of motility machinery are involved in generating tumbling behavior or broad distributions of rotational diffusion. Furthermore, it might be useful to study such behavior on structured or controllable substrates [34], to tease apart the influence of environment vs. internal circuitry on controlling these timescales.
On the other hand, many cell types exhibit contact inhibition of locomotion (CIL) [35], where contact with another cell will either halt their motion or cause them to immediately recoil and begin moving in the opposite direction. It is possible that the tumbling events we see in mouse fibroblast cells are CIL events. In this work mouse fibroblast trajectories were identified from nuclei centers-of-mass, and we do not have direct observations of the cell membrane. Due to this imaging limitation, we were not able to confirm which tumbling events are associated with cell-cell contacts. This would be an interesting avenue of future research, as our cells are seeded at intermediate densities and it is possible that a significant fraction of of tumbling events are caused by cell-cell contacts. If so, this would be an interesting mechanism for generating super-diffusive behavior of a group of cells at intermediate densities, which could contribute to enhanced diffusion of moving cell fronts.
In addition to CIL, there could also be additional interactions between cells, such as alignment of motility polarization between neighbors or between cells and the underlying substrate to generate flocking-like behavior [8]. It would be interesting to explore the effect of alignment in a generalized SPP model, to see if heterogeneity causes any significant differences in the flocking transition.
From this discussion, it is obvious that a natural extension of our current work is interacting SPP models. If tumbling events are caused by cell-cell contacts, such a model would also allow us to predict how superdiffusivity changes with cell density. In even higher density cell populations and confluent tissues, cells will be in nearly constant contact and steric cell-cell interactions will play an even larger role in constraining cell positions. The effect of super-diffusion, whether generated by a Lévy walk or heterogeneity based model, could potentially alter the high-density behavior of standard SPP models.
For example, recent work suggests that groups of cells [36] and packings of SPPs undergo jamming transitions [11, 37, 38]. Could the addition of superdiffusive dynamics have an effect on these types of transitions? Persistent motility can alter the jamming transition—higher speeds and more persistent trajectories allows particles to explore areas of the energy landscape that were previously inaccessible [38]. Similar effects are seen in shape-based models for confluent tissues [36]. The inclusion of both run-and-tumble dynamics as well as varying persistence length through broadly distributed rotational diffusion coefficients in a generalized SPP model could offer an interesting mechanism for tuning jamming.
Another emergent feature of self-propelled particle models is motility induced phase separation (MIPS). Persistently moving particles create an inward oriented boundary layer that cage interior particles into a solid phase, while other cells are in a lower density gas phase outside of this boundary [37, 39] and this effect has recently been implicated in generating colony formation in bacteria [40]. MIPS relies on persistence length to generate this behavior. Our generalized SPP model could reinforce this effect due to relatively persistent run phases, destroy the effect due to tumbling, or perhaps alter the nature of the transition due to enhanced fluctuations, and this is an interesting direction for future study.
|
10.1371/journal.pntd.0000068 | Functional and Structural Insights Revealed by Molecular Dynamics Simulations of an Essential RNA Editing Ligase in Trypanosoma brucei | RNA editing ligase 1 (TbREL1) is required for the survival of both the insect and bloodstream forms of Trypanosoma brucei, the parasite responsible for the devastating tropical disease African sleeping sickness. The type of RNA editing that TbREL1 is involved in is unique to the trypanosomes, and no close human homolog is known to exist. In addition, the high-resolution crystal structure revealed several unique features of the active site, making this enzyme a promising target for structure-based drug design. In this work, two 20 ns atomistic molecular dynamics (MD) simulations are employed to investigate the dynamics of TbREL1, both with and without the ATP substrate present. The flexibility of the active site, dynamics of conserved residues and crystallized water molecules, and the interactions between TbREL1 and the ATP substrate are investigated and discussed in the context of TbREL1's function. Differences in local and global motion upon ATP binding suggest that two peripheral loops, unique to the trypanosomes, may be involved in interdomain signaling events. Notably, a significant structural rearrangement of the enzyme's active site occurs during the apo simulations, opening an additional cavity adjacent to the ATP binding site that could be exploited in the development of effective inhibitors directed against this protozoan parasite. Finally, ensemble averaged electrostatics calculations over the MD simulations reveal a novel putative RNA binding site, a discovery that has previously eluded scientists. Ultimately, we use the insights gained through the MD simulations to make several predictions and recommendations, which we anticipate will help direct future experimental studies and structure-based drug discovery efforts against this vital enzyme.
| RNA editing ligase 1 (TbREL1) is required for the survival of both the insect and bloodstream forms of Trypanosoma brucei, the parasite responsible for the devastating tropical disease African sleeping sickness. The type of RNA editing that TbREL1 is involved in is unique to the trypanosomes, and no close human homolog is known to exist. Here we use molecular dynamics simulations to investigate the dynamics of TbREL1, both with and without the ATP substrate present. The flexibility of the active site, dynamics of conserved residues and crystallized water molecules, and the interactions between TbREL1 and the ATP substrate are investigated and discussed. During the apo simulations, a significant structural rearrangement of the enzyme's active site opens an additional cavity adjacent to the ATP binding site that could be exploited in the development of effective inhibitors against this protozoan parasite. State-of-the-art electrostatics calculations reveal a novel putative RNA binding site, a discovery that has previously eluded scientists. Ultimately, we use the insights gained through the MD simulations to make several predictions, which we anticipate will help direct future experimental studies and structure-based drug discovery efforts against this vital enzyme.
| The existence and widespread occurrence of several devastating trypanosomal tropical diseases, such as Chagas' disease and African sleeping sickness, cause an estimated 1 million deaths each year in developing countries [1]. In 2005, the completely sequenced genomes of Trypanosoma brucei, the causative agent of African sleeping sickness, T. cruzi, the causative agent of Chagas disease, and Leishmania major, the causative agent of Leishmaniasis, were published, yet, despite these great genomic successes, the need for effective and suitable drugs still remains [2]. Currently available drugs were developed in the first half of the twentieth century and they are toxic, difficult to deliver and often ineffective [3].
The trypanosome pathogens responsible for these diseases all share unique post-transcriptional mRNA editing features, the discovery of which revealed a rich addition to the central dogma of biology, in which information not only passes from DNA to RNA to protein, but also between different classes of RNA [4]. Through the insertion and deletion of uridylates (U's), the editing process transforms premature mitochondrial RNA (pre-mRNA) to mature mRNA in a multi-protein complex known as the editosome [5],[6]. The exact composition of the editosome complex has yet to be fully characterized, although 20S core complexes have a Mw of 1.6 MDa and appear to be comprised of 16-20 proteins [7]. To further complicate matters, it has recently been demonstrated that at least three different 20S editosomes of heterogeneous composition and distinct specificity are involved in the editing process [8], possibly reflecting compositional changes of this dynamic multicatalyst complex at different stages in the editing process [5].
The remarkable mRNA editing process begins in the trypanosomal mitochondrial (mt) DNA, which consists of a topologically linked network of thousands of minicircles and dozens of maxicircles. It is the transcripts of these maxicircles, encoding components of respiratory complexes and energy transduction systems, which undergo extensive RNA editing. The editing process begins when guide RNAs (gRNAs) are transcribed from the minicircles in the mt genome and subsequently base-pair with pre-mRNA sequences through a conserved “anchor sequence” [9],[10]. Endonucleolytic cleavage of the pre-mRNA strand occurs at a point of mismatch between the trans-acting gRNA and its cognate pre-mRNA, by endonucleolytic enzymes that have yet to be characterized. Depending on the type of RNA mismatch, U's are either added, by terminal uridylyl transferase (TUTase), or deleted, by a U-specific 3′ exonuclease (ExoUase). The processed RNA fragments are then religated by one of two RNA ligases, Kinetoplastid RNA editing ligase 1 (TbREL1) or RNA editing ligase 2 (TbREL2), generally depending on whether the process is deletion or insertion editing, respectively. Religation of the now completely base-paired double stranded RNA (dsRNA) strands occurs in a three-step process (Fig. 1A). In the first step, the catalytic lysine residue acts in concert with a divalent metal cofactor within the ATP binding pocket of TbREL1 and is autoadenylated, forming a protein-AMP intermediate and releasing pyrophosphate. In the next step, a 5′-5′ phosphate linkage is formed when the AMP is transferred to the 5′ end of the nicked 3′ RNA substrate, which is proximal to the active site. Finally, the ligation process is completed when the 3′ hydroxyl group of the other strand displaces the 5′ AMP, resulting in a new phosphodiester bond.
TbREL1 has been shown to play a key role in the vitality of Trypanosoma brucei, as it is required for survival of both the insect and bloodstream forms of the pathogen [11],[12]. TbREL1 is comprised of a catalytic N-terminal adenylation domain and a C-terminal domain that facilitates non-covalent interaction with another editosome protein, KREPA2 [13]. Interestingly, an oligonucleotide binding (OB-fold) domain, usually associated with DNA ligases and capping enzymes in cis, appears to be provided by KREPA2 in trans and has been predicted to act as a conformational switch regulating various steps in the editing process [13]. The recently crystallized 1.2 Å resolution structure of the adenylation domain, TbREL1, from T. brucei with the bound ATP ligand revealed a number of interesting features of the enzyme, including an unusual water-solvated ATP binding pocket [14]. As the N-terminal domain alone has shown to be capable of RNA ligation activity, further studies characterizing its functional dynamics are relevant. Interestingly, although a single Mg2+ ion is clearly coordinated to the protein and ATP in the crystal structure, the protein is not yet adenylated, indicating that the structure is in a pre-catalytic conformation. Furthermore, it has been suggested for this system, and shown experimentally for related systems, that the nucleotidyl transfer reaction proceeds by a two Mg2+ mechanism [15]. It has been hypothesized that the first Mg2+ binds with high affinity between the ATP β− and γ−phosphate groups, as seen in the TbREL1 crystal structure, and the second magnesium coordinates to a lower-affinity site between the ATP α-phosphate group and catalytic lysine, subsequently promoting the bond cleavage between the α and β phosphate and the adenylyl transfer [14].
TbREL1 belongs to the covalent nucleotidyl transferase superfamily of enzymes, along with other RNA ligases, mRNA capping enzymes, and DNA ligases [16]. A thorough bioinformatics and phylogenetic analysis of the RNA ligase family shows five well-conserved structural motifs responsible for the three-step nucleic acid repair and strand-joining reaction, a shared overall protein fold, and common evolutionary traces. At the level of the superfamily, the percent identity among the sequences is less than 10%, which renders traditional sequence alignment measures ineffective. However, a structural alignment of an unbiased and nonredundant set of the members of the superfamily, which includes RNA ligase II, mRNA capping enzyme, NAD+ dependent DNA ligase and the ATP dependent DNA ligase, indicates that there is a well-conserved core structure surrounding the nucleotide binding site (Fig. 2). Furthermore, this analysis reveals eight highly conserved residues that may play key catalytic roles. The structural differences within the nucleotide binding sites are likely due to the subtle differences in their cofactor specificities. A structural phylogenetic analysis based on the multiple structural alignment of the superfamily indicates that the closest known relative to TbREL1 is T4 phage RNA ligase 2 (Fig. S1), and this is consistent with a previous sequence-based evolutionary analysis of the family [16]. Steady-state and pre-steady-state kinetic analysis coupled with strategic mutagenesis of the T4 RNA ligase 2 has established functional roles for the highly conserved binding site residues and mapped many of the important interactions of the RNA ligase active site [17],[18]. The established similarity between these two enzymes is important, as much of this information for the T4 phage system can be used to help interpret, understand, and direct strategic studies for the enzymatic activity of TbREL1.
The existence of a high-resolution ligand-bound crystal structure and unique active site features, coupled with the fact that there is no close human homolog, make TbREL1 an important target for development of inhibitors against these protozoan parasites. In this work, we use all-atom explicit solvent molecular dynamics (MD) simulations to probe the structure, function, and dynamics of TbREL1 on the nanosecond timescale. A comprehensive structural and sequence alignment of all the known superfamily members identifies several key residues that we monitor throughout the simulations. Deeply buried water molecules within the nucleotide binding site and their effect on the mode of ATP-binding are also investigated. A comparison of the principal components for the apo and ATP-bound systems illustrates large local and global differences in the enzyme motion. Ensemble-averaged electrostatics calculations from the MD simulations reveal a putative RNA binding site near the ligase active site, a finding that has previously eluded scientists. Coupled with the crystal structure and currently available experimental information, the dynamics and structural analysis presented here will likely prove to be salient aspects of future successful drug design efforts against this vital enzyme.
Structural coordinates were taken from PDB 1XDN [14]. TbREL1 has a fifty residue N-terminal segment that is believed to be a mitochondrial import signal, which was cleaved before crystallization. The simulated protein contains residues 52–365 of the 469 total residues, thus representing the N-terminal domain. All crystallographically resolved water molecules were retained in the simulations.
In order to prepare the apo-system, the ATP and single Mg2+ ion were removed from the active site and replaced by 6 waters using the program Dowser [19]. The selenomethionines used for crystal structure refinement were replaced with methionines. Histidine protonation states were determined using the WHATIF program and manually double-checked. All other hydrogens were added according to the Charmm27 topology parameters [20] using PSFGEN within NAMD2.5 [21]. The protein was immersed in a rectangular box of TIP3P waters [22] providing a 10 Å buffer from the protein to the periodic boundary in each direction. Four sodium atoms were randomly placed at least 5 Å away from the protein in order to neutralize the system's charge. The apo system is comprised of a total of 35,464 atoms.
In order to remove spurious contacts, a set of 26,000 energy minimization steps were carried out. The first 6,000 steps were performed in three 2,000 step cycles. Hydrogen was relaxed during the first 2000 steps, holding all other atoms fixed. Hydrogen, water and ions were relaxed during the next 2000 steps. In the last cycle, the protein backbone was fixed, minimizing all other atoms. No constraints were applied during the last 20,000 steps, freely minimizing all atoms.
Molecular dynamics simulations were carried out for twenty nanoseconds with no constraints and a 1 fs timestep at 1 atm pressure and a temperature of 298.15 K. The temperature bath was maintained by Langevin dynamics while pressure was maintained with the hybrid Nose Hoover - Langevin piston method [23], using period and decay times of 100 and 50 fs, respectively. The Particle Mesh Ewald algorithm was used to treat long-range electrostatics without a cutoff [24]. A multiple time-stepping algorithm was employed, where bonded interactions were computed at every time step, short-range non-bonded interactions were computed every 2 time steps, and full electrostatics were computed every 4 time steps. All minimization and molecular dynamics were carried out using NAMD 2.5. Simulations were performed on our own local cluster, as well as the San Diego Supercomputer Center's Datastar machine and the National Center for Supercomputing Applications' Cobalt machine. A typical benchmark for the 35,000 atom system on 64 processors on NCSA's Altix platform is 0.12 days per nanosecond of simulation. System configurations were sampled every 500 fs, generating 40,000 coordinate snapshots for subsequent analysis.
The ligand-bound system, with ATP and the single magnesium ion present, was prepared in an identical fashion, except that the crystallized ATP and Mg2+ ion were included. An additional 20 ns simulation was performed on the ligand-bound system with identical simulation parameters as described above.
Trajectory analysis was performed with VMD, Matlab, and customized scripts. Ensemble averaged electrostatics calculations were performed in VMD with the PME Electrostatics Plugin. Snapshots every 5 ps over the course of the 20 ns simulation (4000 snapshots) were used in the calculations. All images were created using VMD.
Two 20 nanosecond simulations were carried out on the apo and ATP-bound TbREL1 systems in order to investigate their stability and dynamical properties. Both systems were simulated without constraints and reached equilibrium after approximately 7.0 ns of constant pressure and temperature equilibration at 1 atm and 298 K. Correspondingly, the trajectory time from 0–7 ns is referred to as the “equilibration phase” and from 7–20 ns is referred to as the “dynamics phase” (Fig 3).
Plots of the time evolution of the root-mean-square-deviation (RMSD), where each trajectory frame was aligned to the initial starting structure in order to remove any rotational or translational motion, indicate that equilibration was achieved for both systems (Fig. 3A). The overall root-mean-square-fluctuations (RMSF) per residue was calculated during the dynamics phase for each system based on Cα positions after alignment to the average equilibrated structure, and the resulting RMSF difference plot shows the differences in RMS fluctuations between the two systems (Fig. 3B). Our results indicate that the presence of ATP appreciably affects the system dynamics, providing stabilization to some regions of the protein structure, while causing increased fluctuations in other areas, relative to the apo-system. Notably, the peripheral loop comprised of residues 213 to 223 exhibits marked increases in flexibility when ATP is bound. The residues with the highest destabilization upon ATP binding are found in the middle of helix-2, A133 and Q134. The increase in fluctuation is due to a significant sidechain and backbone conformational transition of Q134 at 7 ns that allows the amine group of Q134 to change hydrogen bonding partners from the backbone carbonyl of N239 to that of E130. Not surprisingly, the RMSF of adjacent A133 is also affected. Among the most stabilized residues are R107 and D54. The formation of a salt bridge between R153 and D54 at 7 ns prevents excessive motion of the D54 sidechain, which would otherwise make hydrogen bonds to solvent water molecules.
An analysis of the interactions between the bound ATP and conserved TbREL1 residues over the course of the MD simulation can be used to gain insight into the complex dynamics, as well as to predict solution state behavior of the enzyme-substrate complex. Here we describe the conserved ATP-protein interactions, which can be broadly categorized into three groups based on their proximity to the ATP adenosine moiety, ribose, or triphosphate tail.
The adenosine moiety of ATP sits deep within the binding site; residues that interact with the adenosine moiety include E86 and F209. The carbonyl oxygen of E86 accepts a hydrogen bond from ATP's amine and helps lock ATP into position, preventing it from translating fore and aft in the binding pocket. The phenyl group of F209 provides important stabilizing pi-pi interactions with the adenosine moiety, further attenuating lateral movement of ATP within the active site. Both the F209 and the E86 interactions have relatively low standard deviations and average values close to those observed in the crystal structure (Table 2) demonstrating that these interactions are among the most stable in the protein-substrate complex.
Two conserved residues interacting with the ribose moiety are E159, N92. This pair of residues forms a unique hydrogen bonding pattern with the ribose O2' that was conserved throughout the 20 ns simulation (Table 2). The ribose O2' hydroxyl group donates a hydrogen bond to the carboxyl group of E159, while the amine group of N92 donates a hydrogen bond to the ribose O2' (Fig. 4). The slight deviation between the bond distances observed in the crystal structure and the average values observed during the simulation is attributable to conformational fluctuations within the ribose ring from the slightly more compact C-3 and C-2 endo conformations to the more extended envelope conformation.
A number of charged residues near the periphery of the binding site, including R309, K307, K87, R111, interact with the polyphosphate of ATP. K307 and R111 primarily stabilize the conserved triphosphate conformation observed over the course of the 20 ns simulation. The interaction distance between K307 and the O2A of ATP is 0.82 Å shorter than that observed in the crystal structure. This is attributable to rotation about the O5'-Pα bond that brings the O2A closer to the K307 amine. The low standard deviation exhibited throughout the trajectory illustrates the importance of this interaction in maintaining the triphosphate conformation (Table 2).
While it is not immediately apparent from the 4.9 Å average R111 NH–O3G interaction distance, R111 also contributes important triphosphate longer-range stabilizing interactions. For the first 1.2 ns, R111 samples conformations local to that found in the crystal structure, with the R111 NH2 group proximal to O3G. Subsequently, R111 rearranges by rotating through the CD-NE bond whereby the R111 NH1 group donates a hydrogen bond to triphosphates bridging O2B while the NH2 group maintains a hydrogen bond with O3G. Following this rearrangement, a rotation about the NE-CZ bond occurs, which exchanges NH1 and NH2 hydrogen bonding partners while preserving the hydrogen bonding pattern. This second rotation moves the NH2 group 5.2 Å from O3G, and because it occurs frequently, is responsible for the high standard deviation and longer average interaction distance (Table 2). The K87 electrophile remains at an unreactive 4.0 Å average from the alpha phosphate nucleophile. This agrees with crystal structure data from other superfamily members, which suggest that the triphosphate tail must undergo a conformational rearrangement to properly position the alpha phosphate for catalysis [25]. Because of its relatively distal position and larger standard deviation, we predict R309 plays a secondary role in triphosphate stabilization.
The interactions of several conserved residues within the ATP binding site exhibit altered dynamics in the apo and ATP-bound systems (Table 3). Here we provide insight into the stability of the protein complex by examining how these protein interactions are affected by the presence of ATP in the binding pocket.
In the ATP-bound state, D210, R288 and Y58 form a tight hydrogen bonding and electrostatic network at the deep end of the active site (Fig. 4). D210 interacts with R292, but because it is already involved in a stable network of three hydrogen bonds, it makes a poor binding partner. As a result R292 undergoes large fluctuations relative to those observed in the D210, R288, Y58 triad (Table 3). A significantly different behavior is seen in the absence of ATP. In the absence of the pi-pi stacking interactions provided by the adenosine moiety, F209 swings inward toward the active site cavity, disrupting the D210, R288, Y58 triad. Once the network of interactions is destabilized, thermal fluctuations cause R288 to swing away from D210 causing a 1.96 Å increase in average bond distance relative to that observed in the crystal structure (Table 3). D210 subsequently forms a salt bridge with R292 whose average is 0.83 Å shorter than that observed in the crystal structure. This new interaction pattern is conserved throughout more significant active site rearrangements that were observed later in the simulations, as discussed in greater detail below.
TbREL1 has several important loop regions on the periphery of the ATP binding site that may be functionally relevant. One region in particular, a helix-loop segment formed by residues F262-A282, is highly conserved among all trypanosomatids and has been hypothesized to be involved in protein-protein association within the editosome complex (Fig. 2) [14]. Comparing the backbone RMSF values for the apo and ATP-bound system reveals that ATP binding induces a significant increase in motion within this conserved helix-loop region (Table 1). As this region is not directly linked to the ATP binding site, the propagation of this motion must be through cooperative interactions. This is further supported by a principal components analysis (Supplementary Information), which shows that motion in this helix-loop region is primarily accounted for in the most dominant principal components for the ATP bound system, and that the motion is not present in the apo system (Fig. S2). These results suggest that these loop regions may play an important role in interdomain signaling or crosstalk upon ATP binding.
During the apo system simulation, a significant rearrangement of the active site cavity is observed. In the absence of ATP, the active site pocket is initially solvated with water molecules. After 6.5 ns, F209 swings out of the binding pocket towards Y58, making favorable hydrophobic contacts with the sidechain of I305. This structural reorganization forms a hydrophobic barrier to the deep end of the active site, occluding penetration of solvent water molecules (Fig. 5, Dataset S1).
A radius of gyration analysis for motifs IIIa and IV indicates that in the absence of ATP the formation of the hydrophobic barrier leads to a slight contraction of the distance between motifs IIIa and IV that comprise the deep end of the binding pocket (Fig. S3). Further rearrangement occurs after 12.5 ns when Y58 swings away from motif IIIa, towards the bulk solvent. Over the course of the simulation, the distance between motif IIIa and the strand housing residues 58 to 61 systematically increases from 7 Å initially to almost 13 Å, as measured by the distance between the alpha carbons of Y58 and R111.
These results suggest a strong induced fit effect may occur upon ATP binding. The structural reorganization and increased overall flexibility of TbREL1 without ATP-bound (Table 3) may explain the difficulty in crystallizing the apo structure experimentally (personal communication, J. Deng). In addition, the rearrangements observed in the simulations reveal an altered topography near the ATP binding site, which could potentially be exploited in a structure-based inhibitor design scheme in which compounds could be designed that stabilize this inactive conformation. The strategy of targeting a unique inactive conformation was successfully accomplished with Abelson tyrosine kinase and the inhibitory compound Gleevac, which locks the kinase in an inactive conformation with high specificity and affinity [26],[27]. To expedite and assist drug discovery efforts against TbREL1, we are providing a representative reorganized apo structure extracted from the simulations as Dataset S1.
The TbREL1 crystal structure revealed three deeply buried water molecules coordinating interactions between conserved active site residues and ATP [14]. The simulations presented here include all the crystallized water molecules, allowing us to monitor their behavior over the course of the 20 ns trajectory. For continuity between the crystal structure and this work, we adopt the water molecule naming convention as defined in Ref. Deng et al, 2004. These deeply buried water molecules are a unique feature of TbREL1, therefore it may be advantageous to consider them in the inhibitor design process. In particular, insights into their exchange rates and structural features are of interest.
In the ATP-bound system, wat1, which forms bridging hydrogen bonds between conserved R288 and the N1-adenine group of ATP (Fig. 4), remains fairly rigid in its position throughout the equilibration phase. Wat2, which initially forms a hydrogen bond with the backbone carbonyl of conserved V286, immediately changes position to form a hydrogen bond network with wat1, R288 and D210. Once this initial rearrangement occurs, the new configuration remains for the duration of the equilibration phase.
During the dynamics phase of the ATP-bound system, two of the three deeply buried water molecules exchange with bulk solvent. Wat3, which is initially coordinated to wat 1, the backbone carbonyl of F209, and the charged group of R288, is the first of the buried water molecules to exchange. Wat1, the single water molecule that interacts directly with the adenine ring of ATP, is the only buried water molecule that does not exchange. The exchange of these water molecules and subsequent rearrangement of conserved active site residues may indicate that they help modulate the plasticity of the ATP binding site and that it may be possible to advantageously displace one or more of these water molecules with inhibitor functional groups. Furthermore, during the dynamics phase of the ATP-bound system, additional water molecules gain access to the ATP-binding pocket and lubricate interactions between the ligand and the protein. These additional water molecules contribute to the expansion of the active site cavity as seen in the radius of gyration analysis (Fig. S3).
To date, the RNA binding region of TbREL1 has eluded scientists. Despite the high resolution crystal structure, a surface electrostatics calculation performed with DelPhi on the crystal structure did not reveal any large positive patches on the surface of the enzyme where RNA binding might occur [14].
To further investigate the electrostatics of TbREL1, we calculated the electrostatic potential around the ATP-bound protein using ensemble averaging over the 20 ns trajectory. This analysis reveals a large positive electrostatic lobe near the ligase active site, clearly indicating where RNA could favorably bind (Fig. 6A). Furthermore, a structural alignment of TbREL1 with the related human DNA ligase shows nearly perfect overlap of its DNA-binding region with the putative RNA binding region predicted here (Fig. 5B).
Within the center of the putative RNA binding site are two peripheral loops that are unique to the trypanosomes (Y165 - K175 and V190–Y200). These loops had previously been speculated to be involved in RNA recognition due to the presence of a positively charged residue within each, K166 and R194, respectively [14]. The ensemble averaged electrostatic potential maps presented here provide compelling evidence that these loops may be involved in RNA binding, and they are particularly encouraging in light of a recent study that showed extensive MD simulations can accurately reproduce experimentally observed changes in protein electrostatic fields [28]. Interestingly, these putative RNA binding loops are absent in the N-terminal domain of homologous T4 RNA ligase 2, which, in the absence of its C-terminal domain, is not able to catalyze the ligation reaction [18], unlike TbREL1, which is still active in the isolated N-terminal domain [14]. The large positive electrostatic potentials substantiate the idea that TbREL1 is equipped with its own RNA binding motifs, which allows it to retain activity even in the absence of a C-terminal OB-fold domain.
The results presented here allow us to make several predictions that may help motivate future experiments on this vital enzyme. As no single experimental or theoretical approach will answer all questions regarding enzyme function, complex association, and its role in the vitality of the cell, we suggest a variety of approaches to probe the function of TbREL1 and its interactions with the multicomponent editosome complex.
The persistent hydrogen bond interactions formed between N92 and E159 and the ribose groups of ATP (Table 2, Fig. 4) indicate that these residues may play an important role in ATP binding. Mutagenesis studies of the equivalent of N92 in homologous T4 phage RNA ligase 2 substantiates this hypothesis [17]. We predict that mutation of either or both of these residues will decrease ATP binding, most likely through an increase of the Km of ATP for this enzyme. As a negative control, we predict that mutation of nonconserved residues far away from the ATP binding pocket should not affect ATP binding, such as N239A or I98A.
The adenylation of TbREL1 is a key component of the overall reaction catalyzed by this enzyme. K87 has been shown to be the adenylated active site residue, as even the charge-conservative mutation K87R abolishes ligation activity [29]. The triphosphate tail was crystallized in an unreactive conformation relative to K87 that persists throughout the duration of the trajectory (Table 2), which suggests that the second magnesium, the C-terminal domain, RNA substrate, or some other component of the editosome complex needs to bind before the TbREL1 active site will adopt a catalytic conformation. Repositioning of the polyphosphate tail is necessary in order to bring the alpha-phosphate proximal to the K87 electrophile and in-line for attack. This is also suggested by previously reported structural data on other superfamily members [25]. We predict that nearby R309 plays a secondary role in triphosphate stabilization and that mutation of this residue will reduce the rate of catalysis but not abolish it. The interactions between the non-bridging beta and gamma triphosphate oxygens and the charged end of R309 may help stabilize the polyphosphate tail for the in-line attack. Furthermore, E283, one of only eight residues conserved in both structure and sequence for the entire superfamily, is likely the catalytic base in the ligation reaction. It is the only negatively charged residue within reasonable proximity of the catalytic K87, and it remains oriented in a position that would promote catalysis throughout the simulations (Table 2, Fig. 4). This is again substantiated by mutagenesis studies of the equivalent conserved glutamic acid residue in the closely related T4 phage RNA ligase 2 [17].
Based on the RMSF (Table 1) and principal components analysis (Fig. S2), we predict that the conserved peripheral loop comprised of residues F262-A282 may play an important role in interdomain signaling upon ATP binding. It may be possible to test this by engineering a fluorescent probe (e.g. Y275W) in this loop region and then testing whether the fluorescence signal is quenched with the addition of various components of the editosome complex (TUTase, KREPA2, etc), which would indicate that the region is buried and in contact with another editosome protein.
Our simulations and analysis also suggest that the pi-stacking interactions between conserved F209 and the adenosine ring of ATP are of critical importance to stabilizing TbREL1 tertiary structure. Without the ring stacking interactions provided by ATP, the apo protein was greatly destabilized in the vicinity of F209 (Fig. 5). We predict that inhibitors that are competitive with ATP will need to exploit this moiety, providing stabilizing interactions to the core protein structure, either through similar pi-stacking interactions or general hydrophobic interactions. In addition, the exchange of water molecules at the deep end of the ATP binding pocket may indicate that they could be replaced with inhibitor functional groups.
Ensemble averaged electrostatics calculations have allowed us to predict the putative RNA binding region for TbREL1 (Fig. 6). Two unique loops, comprised of residues Y165 - K175 and V190–Y200, are predicted to be near the center of the RNA binding region. We propose that mutation of the positively charged groups in this region, such as K166, K172, K175, or R194, to either neutral groups or negatively charged groups, should alter the electrostatic potential, disrupt the rate of association with the RNA, and reduce the overall rate of catalysis. After the adenylation reaction takes place, the dissociation of the negatively charged pyrophosphate would further increase the positive potential generated in this region.
RNA editing ligase 1 from T. brucei is a promising drug target against African sleeping sickness, and it shares close homology to proteins from the parasites responsible for both Chagas' disease and Leishmaniases, which further underscores its importance to global health. It also presents an ideal system to deepen our understanding of the fundamental and fascinating process of RNA editing. Advancing the knowledge of enzymes and biochemical pathways unique to the trypanosomes, such as mitochondrial RNA editing, will play an important role in the efforts to bring these diseases out of neglected status and into the limelight. |
10.1371/journal.ppat.1001208 | Effector Memory Th1 CD4 T Cells Are Maintained in a Mouse Model of Chronic Malaria | Protection against malaria often decays in the absence of infection, suggesting that protective immunological memory depends on stimulation. Here we have used CD4+ T cells from a transgenic mouse carrying a T cell receptor specific for a malaria protein, Merozoite Surface Protein-1, to investigate memory in a Plasmodium chabaudi infection. CD4+ memory T cells (CD44hiIL-7Rα+) developed during the chronic infection, and were readily distinguishable from effector (CD62LloIL-7Rα−) cells in acute infection. On the basis of cell surface phenotype, we classified memory CD4+ T cells into three subsets: central memory, and early and late effector memory cells, and found that early effector memory cells (CD62LloCD27+) dominated the chronic infection. We demonstrate a linear pathway of differentiation from central memory to early and then late effector memory cells. In adoptive transfer, CD44hi memory cells from chronically infected mice were more effective at delaying and reducing parasitemia and pathology than memory cells from drug-treated mice without chronic infection, and contained a greater proportion of effector cells producing IFN-γ and TNFα, which may have contributed to the enhanced protection. These findings may explain the observation that in humans with chronic malaria, activated effector memory cells are best maintained in conditions of repeated exposure.
| Protective immunity against malaria develops only after several infections and can be lost on leaving an area in which malaria is transmitted. This suggests that the chronic infection may maintain the protective immune response. In this paper we have used a mouse model of a blood-stage malaria infection to examine the memory response of CD4+ T cells during chronic infection. These T cells are required for protective immunity, and also play a part in the inflammatory response that gives rise to malaria disease. Understanding what constitutes a protective CD4+ T cell may help us design more protective vaccines. We show that these memory CD4+ T cells persist in an activated state, produce the inflammatory cytokines TNFα and IFN-γ, and are more protective than “resting” memory CD4+ T cells obtained from mice in which the infection has been eliminated. This may explain why people are better protected against malaria disease when they are infected frequently.
| Protective immunity to malaria develops only after repeated infections; although protection from homologous infection [1] and lethal malaria occurs after one to two infections [2]. Immunity to infection can persist for years; however, clinical immunity can be lost on emigration away from endemic areas, and high levels of exposure lead to lower disease prevalence than lower exposure [3]. Furthermore, human vaccine trials and mouse models have shown that immunity decays both with time after vaccination and that treatment of infection reduces protection [4], [5], [6]. These observations suggest that continuous exposure to the parasite may be required for the maintenance of immunological protection from malaria, as has also been suggested in Leishmania and other chronic infections [7], [8]. Recent work with Plasmodium chabaudi demonstrated that the decay of protection is replicated in mouse models and that this may be determined by a decay in memory T cell (Tmem) function [5].
Adaptive immunity to infection develops by accrual of antigen-experienced memory cells. In the absence of chronic infection, resting, antigen-independent memory T cells reside in secondary lymphoid organs; however, in chronic infection, memory cells and effector cells may be continually generated [9], and may even expand the memory cell pool [10]. Central memory T cells (Tcm, [11]), defined by high levels of expression of CD62L, have been shown to be protective in various infections [12], [13]. However, infection with Plasmodium liver-stages and other chronic infections have been shown to primarily produce effector memory and effector CD8+ T cells [14], [15], which are also protective [12], [16]. In humans, CD8+ effector memory (Tem) cells have been subdivided with activation markers into early and late subsets, with different subsets predominating in different infections, however, it has not yet been determined how they are derived [17], [18], [19]. In some chronic infections where high pathogen loads persist, such as HIV and LCMV, chronic stimulation leads to functional impairment or exhaustion of CD8+ T cells, and production of IL-10, which slows clearance of the pathogen [20], while in other infections, such as HCV, virus-specific CD8+ memory T cells actually accumulate [9], [21].
While there have been relatively few studies of CD4+ T cell memory in malaria, it is known that immunity to the blood stages of Plasmodium is dependent on both CD4+ T cells and B cells [22], and the presence of Plasmodium-specific CD4+ T cells, in some cases, has been shown to correlate with clinical immunity [23]. However, it has been shown that P. yoelii and P. berghei infections can lead to deletion of specific CD4+ T cells generated by vaccination [24], and a recent study showed that protective CD4+ T cell memory decays after 6.5 months in P. chabaudi infection [5], suggesting some impairment of long-lived immunological memory in blood-stage malaria infections. Therefore it is critical to understand more about the generation and maintenance of memory T cells in malaria to improve our capacity to generate long-lived protection by vaccination.
Here we have investigated the development of Plasmodium-specific CD4+ T cells using a transgenic mouse with CD4+ T cells specific for a peptide within Merozoite Surface Protein-1 (MSP1, [25]), the B5 T cell receptor Transgenic (B5 TCR Tg). MSP1 is expressed on the surface of merozoites and is a candidate antigen for inclusion in a blood-stage malaria vaccine. We establish that a heterogeneous memory CD4+ T cell population develops after a primary infection, and is composed predominantly of early effector memory T cells (TemE). We demonstrate a pathway of differentiation from central memory to early and then late effector memory CD4 T cells by adoptively transferring these three subsets into P. chabaudi-infected RAG° mice. We also show that upon adoptive transfer, memory CD4+ T cells from chronically infected mice delay the onset of parasitemia and protect better against pathology than rested memory cells, suggesting that continuous stimulation enhances the effector function and protective capacity of memory T cells in malaria, correlating with their enhanced Th1 cytokine production. While we have previously shown the importance of B cells and antibodies in clearance of primary infection, the current studies suggest a role for cytokine production by memory T cells in long-term protection. This may be one of the reasons why people are better protected against malaria when they are re-infected frequently.
In order to identify malaria-specific CD4+ T memory cells during malaria, we have followed the response of P. chabaudi-specific transgenic CD4+ T cells, which carry a TCR recognising a peptide of Merozoite Surface Protein-1 (B5 Tg, [25]), throughout a blood-stage infection of P. chabaudi in mice. This infection is characterised by an acute phase with a peak parasitemia 8–10 days post-infection of approximately 25% infected erythrocytes. After the acute phase, parasitemia is reduced but not eliminated and can persist as a low-grade chronic infection for up to 3 months [6]. Purified naïve (CD44loCD25−), MSP1-specific B5 TCR Tg CD4+ T cells (Thy1.2+) were labelled with CFSE and injected into BALB/c Thy1.1 mice (2×106 per mouse). The recipient mice were infected with P. chabaudi the following day, and the transferred B5 Tg T cells were followed by flow cytometry for 60 days after infection.
The acute P. chabaudi infection lasting about 20 days, and the subsequent low-level chronic infection may result in the continuous generation of effector cells, as has been reported in other persistent infections [9], [21], [26]. Therefore, we utilized multiparameter flow cytometry and the activation and memory markers, CD44, CD43, CD27, IL-7Rα and CD62L [16], [27], [28] to distinguish between CD4+ effector and memory subsets. At day 9 of infection, the majority of Tg cells had divided (CFSEneg, FACS gating showing level of CD44 expression correlating with cell division is shown in Figure S1), and were defined as effector cells (Teff) CD62Llo, IL-7Rα−, CD43+, and CD27−, and CD44int (Figure 1A). Flow cytometry data gated on CD4+Thy1.2+ CFSEneg divided B5 Tg cells were collected at several time-points during the infection, and subjected to boolean gating analysis, which distributed them into the 32 possible subsets generated by testing every possible combination of the 5 activation markers. This is shown in the histogram in Figure 1B, where the percentage of the divided B5 Tg CD4+ T cells falling into each subset is illustrated. Strikingly, this unbiased analysis identified effector cells as CD62LloIL-7Rα− during the acute parasitemia at day 9, while memory cells were easily distinguished by their upregulation of IL-7Rα, as seen in the majority of divided, CFSEneg B5 Tg cells by day 60, which fell into both the effector (Tem) and central (Tcm) memory categories as indicated by their expression of CD62L. Various subsets could be seen within these major populations, and the other three markers, CD44, CD43, CD27, were used to determine the overall activation status of the cells. By grouping subsets based on the number of activation markers they express, we constructed pie charts to demonstrate graphically how activated the MSP1-specific cells were throughout the P. chabaudi infection (Figure 1C). More than 80% of the Tg T cells expressed markers of activation at day 9, and interestingly, many malaria-specific cells remained activated even at day 21, when the parasite was cleared to below 0.01% of erythrocytes infected. At day 45, the activation status of the cells appeared to have stabilized, so that even at day 60, 25% of the cells maintained three or more markers of activation. This suggests that a large proportion of the MSP1-specific Tg T cells that persisted late into the chronic phase of infection remained considerably activated and did not become resting or central memory cells in this time.
In these studies, when 2×106 B5 Tg CD4+ T cells were transferred into the recipient mice, more than 70% of the specific cells divided, as shown in Figure 1A. We focused on divided cells (CFSEneg) as they represent the specific response, however, it is possible that transfer of this number of cells may have led to non-physiological responses [29]. We therefore transferred fewer T cells (Figure S2A) and were able to detect them by enriching Thy1.2+ cells before analysis and utilizing the enhanced signal to noise ratio of double staining to detect the small numbers of cells. We could see expansion of 30,000 or 5,000 naïve purified T cells at the peak of infection (day 6) in the spleen, with some cells still remaining undivided (Figure S2B), as in the transfer of 2×106 cells. By day 60 more cells divided when fewer cells were transferred, as previously reported, and there was a similar distribution of memory T cell subsets as with more cells. When less than 106 cells were transferred, less than 50 B5 cells were collected at day 60 necessitating concatenation of the FACS data for visualization. Transferring more cells allowed us to analyze the data from each mouse separately, concatenating only for display. Therefore it was not possible to carry out further experiments with lower numbers of cells and attain statistically meaningful data. When few cells are transferred no transferred cells are detectable in uninfected mice after two months, consistent with the data of others [29], and suggesting that the transgenic cells are not expanding in a malaria-antigen-independent manner. The approach of transferring 2×106 cells was further validated by the consistency of data from 5 individual mice in Figure S3. Importantly, transfer of 2×106 cells resulted in a physiological response, in that the small fraction that survives (0.01%) was the same proportion as MSP-1 responsive T cells in wild-type mice, as assessed previously by limiting dilution analysis (0.011% (1/8800) of CD4+ T cells from BALB/c mice respond to B5-containing fragment of MSP-1 after ten weeks of P. chabaudi infection [30]).
Effector and memory cells amongst the transferred Tg CD4 T cells were further analyzed using the expression profile, IL-7Rα−CD62Llo, described above for effector cells (Figure 2A) and CD44hiIL-7Rα+ (Figure 2B, left contour plot) for memory T cells. As shown graphically in Figure 2C, at day 9 of infection, these effector CD4+ T cells accounted for approximately 75–80% of the MSP1-specific Tg T cell population, were still a substantial proportion of the total B5 T cell population at day 21, and despite low parasitemias, some even persist to day 60. CD44hiIL-7Rα+ CD4 memory cells were present by day 21 when parasitemia is very low, but memory cells did not become the dominant population until day 45. This is important as it suggests that the memory phase where memory T cells dominate over effector cells may be delayed by chronic low-level infection.
Human CD8+ effector memory T cells observed in various virus infections have been subdivided further into “early”, “intermediate” and “late” based on differential expression of CD28 and CD27 [18]. We wanted to determine whether the CD44hiIL-7Rα+ Tg CD4+ memory cells in P. chabaudi infection also contained subpopulations suggesting differential stimulation, or stages of differentiation. CD28 could not be used in our experiments as a marker for further subdivision of Tem subsets, as expression of this molecule is not up-regulated on mouse T cells with activation [17], [18]. We therefore defined subsets within the IL-7Rα+ memory Tg cells using CD27 and CD62L. CFSEneg (divided) B5 T cells were gated on CD44hiIL-7Rα+ memory cells and expression of CD27, and CD62L analyzed on these cells to show the distribution of central and effector memory T cells (Figure 2D, left contour plot). There were two effector memory subsets, which correlated with early and late effector memory (TemE, TemL). This analysis on day 60 post-infection showed that TemE B5 Tg cells (CD27+, CD62L−) dominated the response (Figure 2D, graph). Both the proportion of CD44hi IL-7Rα+ memory T cells and the subsets defined by CD27 and CD62L remained similar on days 45 and 60 (data not shown). We also observed the development and trafficking of memory T cells in the lymph nodes and saw both central and effector memory cells, while liver contained only effector memory cells (Figure S4). Memory cells were also observed in the bone marrow in small numbers (data not shown).
In many, but not all, persisting infections, T cells have been shown to express inhibitory receptors such as PD-1 (CD279) and KLRG1. These correlate with dysfunction of the T cells called exhaustion, and can regulate pathogenic responses [31], [32]. However, although P. chabaudi infection in mice has a prolonged acute parasitemia and a long chronic phase, few PD-1+ effector (Figure 2E) Tg CD4+ T cells (all CD62Llo, maximum of 5.7% at day 21) were transiently present, and no KLRG-1+ cells were observed (data not shown). This suggests that terminal differentiation of effector cells is taking place normally, and that exhaustion of CD4+ T cells is not a problem in this malaria infection, but the data show the memory cells remain in an activated state. Therefore, we investigated the pathway of differentiation that leads to this state.
As early (CD27+ CD62L−) and late (CD27− CD62L−) effector memory cell subsets (TemE and TemL) have not previously been studied in mouse CD4+ T cell populations, we investigated the cytokine profiles and differentiation potential of each of the three subsets in this P. chabaudi infection. Using TAPI-2, an inhibitor of the metalloproteases that cleave CD62L and TNFα [33], we were able to perform intracellular cytokine staining on memory T cells at day 60 post-infection whilst maintaining expression of CD62L. MSP1-specific Tcm, TemE and TemL cells were identified using CD44, CD62L and CD27, as described above, and their IFN-γ and IL-10 profiles analyzed (Figure 3A). Interestingly, while Tcm and TemL subsets contained small populations of IFN-γ+ cells and few IL-10+ cells; 52% (+/−4.1 SEM) of TemE cells made either IL-10 or IFN-γ, and 14% (+/−2.9 SEM) made both cytokines. Despite producing little IFN-γ or IL-10, almost 70% CD44hiCD62Lhi Tcm cells were capable of making TNF or IL-2. Tem on the other hand, contained fewer TNF and IL-2 positive cells (Figure 3B). It was not technically possible to analyse three cytokines together and still have sufficient available parameters to subdivide the Tem cell subsets further, or to include IL-7Rα to distinguish effector memory cells from effector cells (<10% on day 60).
In order to determine how the three memory T cell subsets relate to each other and to effector cells, and how they differentiate and survive in the face of constant exposure to infection, transfer experiments with the CD44hiIL-7Rα+ memory cell subsets were undertaken. We could not magnetically sort Thy1.2+ Tg cells from the spleen at day 60 due to hemozoin, an insoluble iron-containing catabolite of hemoglobin that makes phagocytes magnetic. Therefore, to analyze memory T cell differentiation and protection in vivo, we generated memory cells to test by infecting the B5 transgenic mouse. Interestingly, although 80% of the cells in this mouse are malaria-specific [25], a normal percentage of effector cells and memory cells was generated (Figure S5). Therefore, purified Tcm (CD62Lhi), TemE (CD62LloCD27+), or TemL (CD62LloCD27−) from B5 TCR Tg mice (Figure S6 shows the FACS profiles of the sorted cells), were transferred into RAG° mice, that were then infected with P. chabaudi. These experiments have not yet been done in the absence of infection to ascertain the effect of homeostatic proliferation in the RAG° recipients on the differentiation of memory T cells, however, as antigen-induced proliferation is faster, it is likely that it is the dominant pathway in chronic infection. The results showed that not many Tcm cells were present on day 38. Although some did survive and maintain their phenotype, they had largely differentiated into effector memory and effector cells (Figure 3C, Tcm); suggesting that Tcm have the intrinsic capacity to become TemE, TemL and Teff when stimulated to divide. By contrast, TemE had the capacity to expand and generate TemL and Teff cells but not Tcm; while TemL only generated more TemL and Teff cells. These data indicate that there is a linear differentiation pathway from central to early effector to late effector memory, in high-level chronic P.chabaudi infection (represented in Figure 3D), and that all subsets can generate Teff.
We have shown previously that chronically infected C57Bl/6 mice are more resistant to a second infection with the homologous strain of P. chabaudi than mice whose chronic infection had been eliminated with the anti-malarial drug chloroquine (CQ) [6]. This was also the case for P. chabaudi infections in BALB/c mice (Figure S7A). This suggests that memory B and T cells are affected by chronic infection in such a way that they are more effective when continually exposed to antigen.
To determine whether chronically stimulated memory T cells contribute to this enhanced protection, we assessed the ability of MSP1-specific CD4+ T cells to confer protection in adoptive transfer experiments into RAG° mice. As we found that the individual memory T cell subsets in chronically infected mice did not show differential protection, unfractionated Tg memory CD4+ T cells (CD4+CD44hi) from chronically infected or from previously infected and chloroquine-treated B5 transgenic mice, were transferred into RAG° mice together with immune B cells, as described previously [25], [34].
We first verified that chloroquine treatment had resulted in loss of the B5 antigen from the system, as well as eliminating the chronic infection as described [6] (Figure S7B). CFSE-labelled MSP-1-specific Tg CD4 T cells were injected into chronically infected or infected and drug-cured mice at day 45 and 60 post-infection, and recovered and analyzed after 3 days. Cell division of the Tg cells, measured by CFSE levels, indicated whether the MSP1 peptide was still being presented. Indeed, in the chronically infected B5 mice, the B5 MSP1 peptide was still presented on antigen-presenting cells at 45 and 60 days of infection as detected by proliferation of CFSE-labelled MSP1-specific B5 Tg T cells. By contrast, much less proliferation was observed in the Tg T cells recovered from drug-cured mice (Figure S7B), suggesting that after removal of parasites there was insufficient residual MSP1 antigen to stimulate Tg T cells and that we could consider these memory cells as “resting”.
After adoptive transfer of chronically simulated or rested CD4 T cells into recipient mice, there were two clear major measurable effects of these cell transfers on parasitemia (Figure 4A); a delay in appearance of early parasitemia, and a reduction in peak parasitemia. Mice receiving chronically stimulated memory T cells (−CQ) showed a delayed onset of parasitemia (Figure 4A, left graph), showing that these cells were more effective in controlling early parasite growth. Consistent with this, chronically stimulated memory T cells also reduced peak parasitemia (p = 0.02) compared with resting cells, (Figure 4A, right graph) or naïve cells (data not shown). Consistent with reduced parasitemia, mice receiving chronically stimulated cells exhibited less pathology than mice receiving rested memory T cells (Figure 4B) and naïve cells (not shown). Thus chronically stimulated malaria-specific memory T cells show an enhanced early protective effect with reduced parasitemias and reduced pathology compared with rested memory T cells. Furthermore, chronically stimulated B5 memory T cells provided more effective help for a P. chabaudi-specific antibody response than resting memory B5 cells (Figure 4B). RAG° mice receiving chronically stimulated T cells also had higher levels of TNFα, IFN-γ and IL-10, in plasma at day 7 of infection (Figure 4C), than those receiving rested T cells. These inflammatory cytokines may enhance parasite killing, while IL-10, may regulate their pathogenic effects, as has been described previously [35].
As we have shown that the more protective chronically stimulated memory T cells resulted in larger amounts of circulating cytokines upon adoptive transfer into RAG° mice than rested memory T cells, we investigated whether this was also the case in the chronic infection itself, which might explain the greater protective efficacy of the chronically stimulated T cells. Memory cells making multiple cytokines have been proposed to correlate with the protectiveness of vaccines and the effectiveness of memory cells in various chronic infections that require T cells for clearance of the pathogen [36].
In line with the more rapid effect on parasitemia after adoptive transfer, there was a larger population of IL-7Rα− effector T cells (Teff) in the chronically infected mice (Figure 5A, p = 0.038) compared with CQ-treated mice. Conversely, there was a significantly greater proportion of IL-7Rα+CD44hi memory T cells in the B5 CD4+ T cell population from CQ-treated mice (+CQ, Figure 5A, p = 0.0073). The increase of effector cells in chronically infected mice and the decrease in memory T cells was also seen in the endogenous CD4+ population (Figure S8). We could not detect any reproducible differences in the activation markers of effector cells (as defined in Figure 1) between chronically stimulated and “rested” memory Tg CD4 T cells, nor in the composition of the memory cell subsets (as defined in Figure 2C).
To determine whether the proportion of memory T cells making multiple cytokines in P. chabaudi was different in chronically stimulated and “rested” Tg CD4 T cells, we compared their intracellular cytokine profiles on day 60 post-infection, as described above for Figure 3. Chronically stimulated (−CQ) Tg CD4 T cells at day 60 contained a higher proportion of cells producing both TNF and IFN-γ but not IL-2 than memory cells from drug-treated mice (Figure 5B, Figure 5C). The proportions of triple cytokine producing cells however were not significantly different (Figure 5C, left). The difference in the double-producing cells was mainly within the CD62LloCD44hi effector or effector memory cell population (Figure 5D). Our data suggest that the greater capacity of CD4+ T cells in chronic infection to control early parasitemia is due to the maintenance of Th1 effector cytokine potential by the effector or effector memory cells.
There is a real need to understand protective immune mechanisms in malaria in order to improve vaccination. Development of immunological memory to malaria infection develops over several exposures, and protection is short-lived in the absence of exposure to the parasite [37]. CD4+ T cell responses and specifically CD4+ T cell memory, which are of central importance in immunity to blood-stage Plasmodium infections, have been comparatively little investigated in infections in vivo, compared with CD8+ T cell responses to the liver stages of malaria [38], despite the fact that it is the blood stages and CD4+ cells which cause disease.
Here, we have used a mouse model of malaria, P. chabaudi, to investigate whether Plasmodium-specific CD4+ T cell memory develops in a blood-stage infection. We show that CD4+ T cells with phenotypic characteristics of memory cells are indeed generated within two months of an infection. They develop slowly, and the predominant memory cells are early effector memory cells, the majority of which are able to produce cytokines on short-term re-stimulation and to become effector cells on re-infection. Importantly, we showed that chronically stimulated memory T cells protect immunocompromised mice from infection better than rested memory cells.
There was a striking predominance of CD4+CD62Llo effector memory cells remaining after the acute P. chabaudi infection, which remained in a relatively activated state. Such cells have also been shown to predominate in the liver stages of malaria [15], [39] and are present in humans protected against malaria by P. falciparum sporozoite challenge [40] as well as in other chronic infections such as LCMV [26], T. cruzi [13], [14] T. muris [41], gamma herpes virus [42], and human Hepatitis B and C [43], and in some cases are critically important in effector sites such as the lung [44], [45], [46].
The type of memory cells maintained may depend on persistence of specific or non-specific stimulation. Studies of CD8+ T cell memory in human viral infections have correlated memory phenotypes with the level of chronic infection [19], and suggested that memory T cell differentiation progresses through a maturation process dependant on the level and duration of antigenic stimulation [17], [18], [47]. Thus a linear pathway of differentiation (Tcm>TemE>TemL) has been proposed based on surface markers and the length of the telomeres in each subset; where low level chronic infections would lead to maintenance of early effector memory cells, but high levels would enhance late effector memory [18]. Here we show conclusively that in chronic infection central memory CD4+ T cells can produce these other two subsets that are indeed related in a linear manner to one another as suggested by the human studies as well as effector T cells. It is likely that selection for high affinity clones and regulation of cytokine production [48] occurs during this differentiation process.
Both Teff and Tem CD4 T cells can protect in various infectious disease models [12], [16], [28], [41], [49], and for CD8 T cells, in some cases, they are more potent than resting memory cells [41], [50], [51]. Although effector memory cells are likely to be more short-lived than central memory cells [52], there are other reports of long-term persistence of effector-type cells [53]. From our studies it appears that CD4+ Tem survive the acute infection, especially CD27+ cells. This molecule, a member of TNFR family, has been shown to prolong the life of T cells [54]. Extrapolating these findings to human Plasmodium infections, it is possible that TemE may survive and protect for extended periods after an infection.
We compared memory CD4 T cells generated and maintained in chronic infection, with those that survived after drug-clearance of infection. We identified an increase in CD44hiIL-7Rα+ memory CD4 T cells after drug treatment. Although IL-7 can be utilized by memory T cells for survival [55], the observation that in chronically infected mice there were more IL-7Rα− effector cells two months after infection suggests that a sub-population in infected mice may depend instead on antigen or infection for maintenance.
In an adoptive transfer and infection system, chronically stimulated memory CD4+ T cells were more protective; they slowed parasite growth, reduced peak parasitemia and were associated with less pathology compared with resting memory cells obtained from infected and drug-cured mice. Despite the eventual higher antibody titers in mice receiving chronically stimulated CD4+ T cells, it is unlikely that antibody was the mechanism of parasite control, as at this early stage of infection anti-malaria antibodies were not detectable. However, the chronically stimulated CD4+ T cells contained more effector cells and a greater proportion of TNFα+IFN-γ+IL2− double-producing cells than resting memory cells. Furthermore, mice receiving the chronically stimulated CD4+ T cells had significantly greater plasma levels of IFN-γ, and TNFα, cytokines known to play a role in early parasite control [56]. Together, our data suggest that Th1 memory cells can control the acute P. chabaudi parasitemia. It will be important to determine which of the activated memory cell subsets [17], [28] within the chronically activated T cell population containing the IFN-γ+TNFα+IL2− cells are responsible for these anti-parasite effects. The lesser pathology observed in mice receiving chronically stimulated memory T cells compared with rested memory cells, may be the result of the lower parasitemia brought about through the effects of TNFα and IFN-γ, and to the increased levels of IL-10, a cytokine known to down-regulate pathology associated with P. chabaudi infections [35].
Here we have seen that the changes in a memory population on withdrawal of infectious stimuli for a month after infection, have subtle effects on the proportions of individual memory T cell subsets in the spleen in P. chabaudi malaria, but nevertheless reduce the potential for Th1 cytokine production, and protection. Our data suggest that chronically stimulated memory CD4+ T cells are the most protective for controlling early parasitemia and pathology, as seen in Leishmania infections, another low-level chronic infection. However Tem may be short-lived, as they are not thought to be the main component of long-lived memory [57]. This might explain why protective memory is lost in mice over time [5] and in humans who move away from endemic areas [3], as previously activated cells revert to a resting state [11]. A contribution of continually stimulated effector memory CD4+ T cells to parasite control and regulation of pathology are very much in line with observations that the best protection of humans in areas of endemic malaria is a certain level of continuous exposure [2], [37], and with the association of effector memory cells producing multiple cytokines observed in humans experimentally infected with P. falciparum sporozoites [40]. This suggests that vaccination methods that enhance production, and survival of cells which maintain effector functions, may be the most successful in protection against severe malaria, although this must be balanced with the well-tuned production of regulatory cytokines.
Female BALB/c (MRC strain) were maintained in the breeding facilities of the MRC National Institute of Medical Research. BALB/c rag2−/− mice (RAG°) were a gift from Dr Anton Rolink (University of Basel, Basel, Switzerland). Thy1.1 BALB/c congenic mice (N = 15 BALB/c) were a kind gift of Dr. David Tough, (Jenner Institute, Compton, UK) and were further backcrossed four generations to BALB/c (MRC) for adoptive transfers. B5 TCR Transgenic mice were generated as described [25]. The B5 TCR recognizes MSP1 (1157–1171, ISVLKSRLLKRKKYI/I-Ed); B5 TCR Tg mice were typed using correct primers Va2, gaacgttccagattccatgg and atggacaagatcctgacagcatcg; and Vβ8.1, cagagaccctcaggcggctgctcagg and atgggctccaggctgttctttgtggttttgattc.
Mice, 5–8 weeks, were infected with 105 P. chabaudi chabaudi (AS)–infected erythrocytes i.p. and monitored by Giemsa-stained blood films [22]. The presence of sub-patent chronic parasitemias were determined by subinoculation of blood into naïve recipient mice at various times during the chronic infection as described [6]. To eliminate chronic infections, mice were treated three times with 50mg/kg Chloroquine (Sigma, Dorset, UK) in saline (Sigma) alternate days between days 30–34. P. chabaudi (AS) is sensitive to chloroquine at low parasite density, as on day 30 post-infection [6]. Weight (g) was measured on a balance and percent change was calculated as: ((weight - d0 wt)/d0 wt)×100%. Temperature change was recorded from subcutaneously implanted transponders on BioMedic Data Systems DAS5002 (PLEXX, Elst, Netherlands).
All animal experiments were carried out according to UK National guidelines (Scientific Procedures) Act 1986 under licence PPL80/2385 approved by the British Home Office. The procedures used were approved by the NIMR institutional Ethical Review Panel.
Single-cell suspensions of spleen, liver and lymph nodes were made in HBSS, incubated in red blood cell lysis buffer (Sigma), and stained in PBS 2% FBS (PAA Laboratories, Somerset, UK) and 0.1% sodium azide with anti-CD16/32 (24G2) supernatant followed by combinations of FITC–, PE–, PE-TexasRed, peridinin chlorophyll protein (PerCP), PE/Cyanine 7 (Cy7), Pacific Blue (PB), biotin- (-b-), allophycocyanin (APC)–, or APC/Cy7-conjugated antibodies: CD4 (RM4-5, BD Biosciences, Cambridge Biosciences Oxford, UK) or CD127-PE, CD90.2-eFluor450, -PE/C7 (A47E, 53-2.1 eBiosciences, Hatfield, UK) or CD43-PE/Cy7 (1B11 Biolegend/Cambridge Bioscience, Cambridge, UK) or CD4-Pacific Orange (PO, Caltag, Invitrogen Paisley, Scotland). The second-step reagents streptavidin-PerCP; (BDbiosciences) and strepdavidin-PB (Invitrogen) were used. For intracellular cytokine staining, cells were pre-treated for one hour with 80–100uM TAPI-2 (Peptides International) [33], and then stimulated for 5h in complete Iscove's medium (cIMDM, Sigma), 10% FBS, 2 mM L-glutamine, 0.5 mM sodium pyruvate, 100 U penicillin, 100 µg streptomycin, and 50 µM 2-ME (Gibco, Invitrogen) with PMA (50 ng/mL; Sigma), ionomycin (500 ng/mL; Sigma), and BrefeldinA (10 µg/mL; Sigma) for the last two hours. Cells were fixed in 2% paraformaldehyde 20 minutes, and resuspended in staining buffer overnight. Permeabilized in Perm/Wash buffer (BDbiosciences) 25 minutes and washed twice, then incubated for 40 minutes with anti–IFNγ-PE (XMG1.2), IL2-APC (JES6-5H4), TNFα (MP6-XT22), IL10-APC (JES5-16E3; all from BD Biosciences). Cells were washed thrice in Perm/Wash and resuspended in staining buffer. Cells were collected on a FACSCalibur or 9-color CyAn ADP (DAKO, Beckman Coulter) using Summit software (Cytomation) and analyzed in FlowJo (Tree Star, Ashland, OR). Staining combinations were designed through a meticulous optimization strategy as described by Seder et al and Tung et al [36], [58]. Compensation was performed in Flow Jo using single stained splenocytes or bead controls (BD CompBeads Anti-Rat Ig, κ) and results thoroughly checked for accuracy and consistency. For presentation, data from 2–5 mice is concatenated to achieve sufficient cell numbers (5–10,000) for presentation and Boolean gating analysis, only after each mouse was carefully analyzed and averages and standard errors of the mean (SEM) calculated.
For in vitro stimulation, naïve cell transfers, and RAG° transfers, B5 Tg CD4+ cells were purified by positive selection with magnetic micro-beads (Miltenyi Biotec, Surrey, UK) using magnetic-activated cell sorting (autoMACS) and further purified as naïve (CD44lo, CD25−) or memory (CD44hi) T cells by high speed (MoFlo, Cytomation, Beckman Coulter) sorting to >99% purity. In naïve cell transfers, cells were labeled for 10 min at 37°C with 1µM CFSE (Molecular Probes, Invitrogen) after washing thrice in cation-free PBS. Memory subsets were purified on a FACS Aria (BDbiosciences) as shown in Figure S4, as CD44hi IL-7Rα+, and Tcm, CD62Lhi CD27+; TemE, CD62Llo CD27+; and TemL, CD62Llo CD27− to about 95% purity.
Splenic CD4+ T and CD19+ B cells were purified by MACS (95% purity, Miltenyi). B5 Tg CD4+ T cells were transferred (2×106) into uninfected congenic Thy1.1 BALB/c mice. To determine presence of B5 antigen, CFSE-labeled CD4+ B5 T cells were transferred into previously infected mice for 4–5 days. Immune B cells were generated by infecting BALB/c mice with 105 P. chabaudi twice, 2 months apart. In RAG° transfer, 105 naïve, memory, or memory subsets T cells and 107 B cells were sort-purified (MoFlo, cytomation) and transferred. Reconstitution was analyzed day 39 post infection and all mice had 3.5–28% (of lymphocyte gate) T cells (naïve T cells average 19.5±2.4%, Tmem average 11±1.0%) and 0.8–5.3% B cells in the spleen (average 2.3±0.2%). One mouse with less than 0.5% of both B and T cells, and excessive splenomegaly was excluded from the study.
IL-10, TNF and IFN-γ were measured in serum from animals at day 7 post-infection using the Luminex Flexible Beadlyte Immunoassay (BioRad or Upstate/Millipore, Milton Keynes, UK), according to the manufacturers instructions. Plates were read and analyzed using the Luminex 100 machine and software (Upstate).
Malaria-specific antibodies were measured by ELISA, as described. Plates were coated with a soluble fraction of the lysate of P. chabaudi blood-stage parasites, IgG was detected using alkaline-phosphatase linked secondary antibody (Southern Biotech, Cambridge, UK) and PNPP substrate (Sigma). Results are expressed as relative units based on a standard hyperimmune plasma (all described in [59]).
Statistics were performed in Prism (GraphPad, La Jolla, CA) using Student t-test, with P≤.05 was considered significant.
|
10.1371/journal.pntd.0006787 | Efficacy of subcutaneous doses and a new oral amorphous solid dispersion formulation of flubendazole on male jirds (Meriones unguiculatus) infected with the filarial nematode Brugia pahangi | River blindness and lymphatic filariasis are two filarial diseases that globally affect millions of people mostly in impoverished countries. Current mass drug administration programs rely on drugs that primarily target the microfilariae, which are released from adult female worms. The female worms can live for several years, releasing millions of microfilariae throughout the course of infection. Thus, to stop transmission of infection and shorten the time to elimination of these diseases, a safe and effective drug that kills the adult stage is needed. The benzimidazole anthelmintic flubendazole (FBZ) is 100% efficacious as a macrofilaricide in experimental filarial rodent models but it must be administered subcutaneously (SC) due to its low oral bioavailability. Studies were undertaken to assess the efficacy of a new oral amorphous solid dispersion (ASD) formulation of FBZ on Brugia pahangi infected jirds (Meriones unguiculatus) and compare it to a single or multiple doses of FBZ given subcutaneously. Results showed that worm burden was not significantly decreased in animals given oral doses of ASD FBZ (0.2–15 mg/kg). Regardless, doses as low as 1.5 mg/kg caused extensive ultrastructural damage to developing embryos and microfilariae (mf). SC injections of FBZ in suspension (10 mg/kg) given for 5 days however, eliminated all worms in all animals, and a single SC injection reduced worm burden by 63% compared to the control group. In summary, oral doses of ASD formulated FBZ did not significantly reduce total worm burden but longer treatments, extended takedown times or a second dosing regimen, may decrease female fecundity and the number of mf shed by female worms.
| Safe and effective macrofilaricidal drugs are critically needed to treat onchocerciasis and lymphatic filariasis, which affect over 54 million people worldwide. Flubendazole (FBZ) in its current commercial formulations is an effective anthelminthic for intestinal soil transmitted helminth (STH) infections but not for filarial infections due to its low bioavailability. The purpose of this study was to assess the efficacy of a new amorphous solid dispersion (ASD) formulation of FBZ given orally to jirds (Meriones unguiculatus) infected with the filarial nematode Brugia pahangi and compare it to FBZ (in suspension) given subcutaneously as a single or multiple dose. Our results indicated that treatment with ASD FBZ did not significantly reduce the total number of worms. However, doses as low as 1.5 mg/kg caused ultrastructural damage to the early stages of developing embryos. No worms were recovered from jirds given 10 mg/kg SC injections of FBZ (suspension) for 5 days when necropsied 68 days after the first-dose, and the number of mf at necropsy was significantly decreased compared to the control group. Animals given only a single 10 mg/kg SC injection had a 63% decrease in the number of adult worms. These animals also had fewer female worms and mf compared to the control group suggesting that even a single SC injection of FBZ had an effect on female survival and fecundity. TEM of worms recovered from jirds given low doses of the ASD formulated FBZ (1.5 mg/kg) suggested that female fecundity and mf production were reduced, but longer treatments, longer takedown times or a second dosing regimen of ASD FBZ may be needed to significantly decrease the total worm burden.
| River blindness (onchocerciasis) and lymphatic filariasis are two major neglected diseases caused by parasitic nematodes that together affect an estimated 54 million people worldwide in mostly poor, developing countries.
With river blindness, approximately 12 million people suffer from skin disease and 1 million people have vision loss [1]. It is a chronic disease caused by the first larval stage, microfilariae (mf) of Onchocerca volvulus which are released from female worms residing in subcutaneous tissues. Microfilariae migrate throughout the skin causing severe itchiness, and in the eye, they induce an inflammatory response that eventually leads to blindness [2–5].
Lymphatic filariasis (LF) or elephantiasis is caused by Wuchereria bancrofti, Brugia malayi and B. timori whose adult worms infect and damage the lymphatic tissues. This debilitating disease is characterized by pain and severe lymphedema often involving the arms, legs, breasts and genitalia, leading to great economic losses as well as social suffering and the stigma associated with elephantiasis [2, 6].
To date, there are no vaccines for these diseases and international control programs attempt to interrupt transmission of infection in Africa with mass drug administration (MDA) annually or biannually using microfilaricidal drugs: ivermectin for onchocerciasis; albendazole plus ivermectin or albendazole plus diethylcarbamazine (DEC) for LF. Recently a triple-drug therapy with ivermectin, DEC and albendazole was explored for the treatment of LF outside of Africa with the assumption that it will accelerate the elimination of LF if a coverage of >65% of the population is achieved [7]. Notably, this new therapy significantly improves mf clearance and maintenance of amicrofilaremia compared to the two-drug MDA regimen with DEC and albendazole and offers great promise in eliminating LF [7, 8].
The triple-drug therapy however, is not relevant for treatment of onchocerciasis due to the major adverse affects caused by DEC [9]. Although elimination of onchocerciasis has been achieved in a few foci in Africa and in the Americas [10–12], there has been only a 31% reduction in the incidence of onchocerciasis in Africa since 1995 [13]. The African Program for Onchocerciasis Control (APOC) has therefore called for some 1.15 billion treatments by 2045 [14], though many neglected tropical disease experts doubt that onchocerciasis can ever be eliminated through MDA alone [15], especially given that MDA with ivermectin cannot be used in 11 Central African countries co-endemic with Loa loa infections due to the risk of severe adverse events [16, 17]. Given the longevity and high fecundity of the adult worms (macrofilariae) [18–22] and the current lack of macrofilaricidal drugs, it is unlikely that the WHO goal of eliminating onchocerciasis by 2025 will be met when microfilaricidal drugs alone are used [17, 23, 24].
To achieve the ultimate goal of onchocerciasis elimination, drugs that cure infections and thus stop transmission of infection and ultimately shorten the time to elimination, a safe and effective drug that kills adult worms is needed. The need for such alternate treatment strategies is further supported by the occurrence of foci in Africa with suboptimal response to ivermectin [25].
The benzimidazole anthelmintic flubendazole (FBZ) is highly efficacious as a macrofilaricide in experimental filarial rodent models but it must be administered subcutaneously (SC) due to its low oral bioavailability [26–28]. Unfortunately, when administered parenterally to patients with onchocerciasis, severe reactions around the intramuscular injection site were reported [9]. Therefore, efforts have been made to develop a re-formulation of FBZ that would enable oral dosing [26–28]. The purpose of this present study was to assess the efficacy of a new amorphous solid dispersion formulation of FBZ (ASD FBZ; Janssen Pharmaceutica) for the treatment of onchocerciasis. ASD FBZ was given orally (0.2, 0.6, 1.5, 6, 15 mg/kg for 5 days) to jirds (Meriones unguiculatus) infected with the filarial nematode Brugia pahangi and compared to a single and a 5 day SC injection of 10 mg/kg FBZ. The rodent model using jirds as hosts for adult B. pahangi has been used extensively to study efficacy of antifilarial compounds and is one of the surrogate models used to investigate drugs for treatment of onchocerciasis [26, 29–33]. Results of this study showed that worm burden was not significantly decreased in jirds given oral doses of ASD FBZ (0.2–15 mg/kg). However, doses as low as 1.5 mg/kg caused extensive ultrastructural damage to developing embryos and mf.
Male jirds (Meriones unguiculatus) approximately 6 weeks of age (50–60 g) were purchased from Charles River (USA, Kingston K62 jirds) and infected by intraperitoneal injection with Brugia pahangi third-stage larvae (L3). Dosing regimens began a minimum of 12 weeks post-infection following development of the larval stage to the adult stages and secretion of microfilariae. Animals were allowed to eat and drink ad libitum and maintained following the approved IACUC protocol AN109629-03D.
Oral suspensions of ASD flubendazole (Janssen Bend 1/9) contained 10% FBZ-AAA G001, 10% flubendazole:hydroxypropyl methylcellulose acetate succinate, Lot number BREC-1113-036 with a vehicle of aqueous solution of 0.5% w/v methocel A4M (Premium) in demineralized water. For the flubendazole subcutaneous suspension (FBZ-AAA, lot 0020470007), FBZ was purchased from Shaanxi Hanjiang Pharmaceutical Group LTD, Hanzhong City, Shaanxi, China and formulated with aqueous solution of 0.5% w/v HEC (2-hydroxyethyl cellulose, Sigma 434965) in demineralized water and 0.1% Tween 80. Formulations were acclimated for 30 min at room temperature, protected from light and homogenized prior to dosing for at least 30 seconds to ensure no visible sedimentation.
ASD flubendazole was given per os (PO) at 0.2, 0.6, 1.5, 6 or 15 mg/kg for 5 consecutive days, or subcutaneously (SC) at 10 mg/kg one time or for 5 days; control groups were not given any treatment which allowed comparison with both the oral and SC groups rather than having to include 2 different vehicle groups (Table 1). Doses were selected to determine the efficacy of the new ASD formulation of FBZ after oral administration for 5 days, which is considered a dosing regimen feasible for use in patients in the field as well as on the basis of previous pharmacokinetic and toxicological data. For the subcutaneous route and doses, the 5-day 10 mg/kg dose group is used as a positive control group in antifilarial rodent models. A single dose subcutaneous dose of 10 mg/kg was included to test if the same efficacy could be obtained as with the positive control group.
All animals used in the study were lightly anesthetized with isoflurane just to the state of drowsiness prior to dosing to avoid any handling stress. Animals were given PO doses using a metal gavage needle and 1 ml tuberculin syringe. Animals receiving SC doses were injected on the scruff of their necks with a 25 gauge needle in a clockwise fashion to avoid injection into the same site over the 5 day treatment period. The untreated control group was also given light anesthesia as in the case with the treated animals but was not dosed. Animals were allowed to feed ad libitum and dosed according to their body weight determined prior to each dosing. The takedown times were 68 and 72 days post-first dose.
The sparse sampling approach was used to obtain a sufficient time profile while minimizing stress to the animals, e.g. 2–3 animals per group were sampled per time point using the micro-sampling technique. Blood sampling times were chosen based on the route and duration of treatment and previous PK data, and thereby took into account the allowable volume of blood that could be taken from gerbils without causing stressful manipulation of the animals. Approximately 30–50 μl of blood was collected per animal from the vena saphena, using heparin-coated hematocrit capillary tubes. Blood was placed immediately on ice and centrifuged for 1810 g for 15 min at 4°C. 10 μl of plasma was then transferred into a PCR microfuge tube, frozen in a dry ice/ethanol bath and stored at -80°C until shipped.
Plasma samples from jirds were collected at the following times for Experiment 1: For the SC single dose (10 mg/kg): 1, 3, 8 and 24 hrs post-dose; for the SC repeat doses (10 mg/kg): 2 hrs post-dose on Days 1–4; at Day 5: 1, 3, and 8, 24, 48 hrs post-dose; for the SC groups: weekly for 9 weeks post-last dose; for the PO repeat doses: 2 hrs post-dose on Days 1–4; 2 hrs and 24 hrs post-dose on Days 5. Plasma samples from jirds were collected at the following times for Experiment 2: for the SC single dose (10 mg/kg): 1, 3, 8, 24 and 48 hrs post-dose; and 1, 3, 5, 7, and 9 weeks post-dose; for the PO repeat doses (6 mg/kg and 15 mg/kg): 2 hrs post-dose on Days 1–4; 1, 2, 4, 8 and 24 hrs post-dose on Day 5. All flubendazole formulations used in the study were also analyzed at the end of the dosing period.
For Experiment 1, analyses were conducted at Janssen Research and Development (1400 McKean Road, Spring House, PA 19477), and for Experiment 2, analyses were conducted at Janssen Research and Development, Beerse, Belgium. For both studies, plasma samples were analyzed individually for flubendazole (JNJ-161941), hydrolyzed flubendazole (H-FBZ, JNJ-114699) and reduced flubendazole (R-FBZ, JNJ-1809600) using a qualified LC-MS/MS method. 10 μl plasma aliquots in end-to-end capillaries were washed with 100μl of 2% BSA in phosphate buffer pH 7.5. From this diluted sample 44 μl was taken for analysis corresponding to 4 μl of plasma. After addition of 10 μl of internal standard dilution and 200 μl of acetonitrile for the precipitation of the plasma proteins, the samples were mixed and centrifuged. 150 μl of the supernatant was evaporated to dryness under nitrogen flow at 50°C and reconstituted in 150 μl of a mixture of 0.1% formic acid and acetonitrile (90/10, v/v). 20 μl of the extract was injected onto an Acquity UPLC BEH C18 column (50 x 2.1 mm, 1.7μm particles) (Waters, Milford, USA). The chromatographic system consisted of a Shimadzu SIL30ACMP autosampler and Shimadzu LC30 pumps (Shimadzu, Kyoto, Japan). The mobile phase was a mixture of 1% formic acid and acetonitrile with a flow rate of 0.6 ml/min and a 2.5 minute gradient from 20 to 60% acetonitrile followed by a 1-minute step gradient to 98% acetonitrile.
Mass spectrometric detection was performed on an API4000 triple quadrupole mass spectrometer (Sciex, Framingham, MA, USA) with Turbo Ion SprayTM ionization operated in positive ion mode. FBZ, H-FBZ and R-FBZ were quantified against calibration samples and quality control samples, prepared in the same matrix as the study samples by means of a qualified analytical method with a lower limit of quantitation of 0.2, 0.4 and 0.2 ng/ml, respectively and an upper limit of quantitation of 3000 ng/ml for all three analytes.
Animals were euthanized on day 68 (Expt 1) and day 72 (Expt 2) after the first dose. Adult worms and mf were recovered by opening the body cavity and washing the peritoneal cavity with 100 mL of phosphate buffered saline (PBS). Male and female worms were separated and counted using a dissecting microscope. To count the number of mf present in the peritoneal cavity at necropsy, a 100 μl sample of the aforementioned PBS washing fluid was added to 900 μl of 0.04% methylene blue, and then a 50–150 μl sample of the stained mf was streaked onto a glass slide and counted using a compound microscope. The sample mf counts were multiplied by the appropriate dilution factor to calculate the total number of mf from each jird.
Animals were euthanized by carbon dioxide inhalation followed by bilateral thoracotomy, following University of California San Francisco IACUC protocol AN109629-03D.
At necropsy, 7–12 female worms recovered from 2–3 jirds per group were fixed in 2.5% paraformaldehyde, 2% glutaraldehyde in 0.1 M cacodylate buffer, pH 7.4 [34, 35]. Worms were chopped into 1–2 mm long pieces in the fixative. Samples were incubated for 3 hrs at room temperature protected from light and kept at 4°C overnight. Samples were then washed thoroughly in buffer and post-fixed in 1% osmium tetroxide and 1.6% potassium ferricyanide in 0.1 M cacodylate buffer, pH 7.4, for 1 hr. Following washes in buffer and then in distilled water, en bloc staining was performed for 1 hr with 2% uranyl acetate with samples protected from light. Samples were again washed with water and dehydrated through a series of ethanol dilutions: 50%, 70%, 95% and 100% ethanol. Samples were infiltrated with a gradient of acetone-Embed 812 resin and embedded in 100% resin. After sectioning the solidified blocks, 70 nm sections were post-stained with 1% uranyl acetate and Reynolds lead citrate. Images of the sections were collected on a 120kV Microscope (Philips).
To compare adult worm and mf counts in the treated groups versus the control group, raw data were first tested for normality using the Shapiro-Wilk test. When data did not pass the Shapiro-Wilk test, data were then log10 transformed and retested using the Shapiro-Wilk test. In Experiment 2 the log10 transformation of the total number of adult worms per jird passed the Shapiro-Wilk test, so significance was determined by a one-way ANOVA followed by the Holm-Sidak multiple comparisons test. The remaining data did not pass the Shapiro-Wilk test, even after log10 transformation, so significance was determined by the Kruskal-Wallis test, followed by Dunn’s multiple comparison test. All data were analyzed using Prism 6.0f 2014, GraphPad Software, Inc. with 95% confidence limits.
To calculate the geometric means of the number of adult worms, female worms and mf recovered at necropsy, 0.1 was used in place of 0. The percent efficacy was calculated for each treatment group by subtracting the geometric means of treatment groups from the geometric mean of the control group, multiplying by 100%, and dividing the numerator by the geometric mean of the control group. All results written as percentages are given as a geometric mean % (e.g. geometric mean % reduction).
No significant pathology occurred at the site of injection nor was any pathology observed over the course of the experimental period when SC doses of FBZ were injected into the nape area of the jirds. No adult worms were recovered from animals given 10 mg/kg SC injections of FBZ for 5 days, and the number of mf was significantly decreased compared to the control group (Fig 1 and Table 2). In comparison, animals given a single SC injection of 10 mg/kg FBZ had only a 63% (not statistically significant) decrease in the number of adult worms compared to the control group. These animals however, had fewer female worms and mf suggesting that even a single SC injection had an effect on female survival and fecundity. This effect was also observed in Experiment 2 for animals given a single SC injection of FBZ (Fig 1 and Table 2).
There was no statistically significant reduction in the total number of adult Brugia pahangi recovered from jirds given oral doses of ASD FBZ compared to the control group (Fig 1 and Table 2). Reduction in the total number of worms ranged from 38% to 80% and did not correlate with ASD FBZ dose concentrations. There was also no statistical difference in the percent reduction in the number of female worms nor mf in the treatment groups compared to the control group due to the high variability within each group. However, the results showed that all groups given oral ASD FBZ still had a 52%-93% reduction in the number of female worms recovered at necropsy and 59%-96% reduction in the mf count, respectively (Fig 1 and S1 Table).
Individual and mean plasma concentrations and pharmacokinetic (PK) parameters were determined for FBZ and the metabolites, H-FBZ and R-FBZ (Tables 3–6, Fig 2, S2 Table). Results showed that after oral administration, concentrations of FBZ from the 2 hr plasma samples increased with the dose (Table 3). At day 5, the mean concentration at 2 hr post-dose was 0.015 μg/ml for 0.2 mg/kg, 0.039 μg/ml for 0.6 mg/kg, 0.085 μg/ml for 1.5 mg/kg, 0.436 μg/ml for 6 mg/kg and 1.73 μg/ml for 15 mg/kg. At 24 hr on day 5, the plasma concentrations were similar between 0.2 and 0.6 mg/kg/ day and increased with the dose between 0.6 and 15 mg/kg/day.
FBZ exposure (AUC0-24h) at day 5 increased more than dose proportionally to the dose between 6 and 15 mg/kg/day (Table 4). Cmax however was dose proportional. The tmax for repeat oral administration of FBZ at 6 and 15 mg/kg/day for 5 days was 1 hr and 2 hr, respectively, indicating fast absorption. Ratios of H-FBZ/FBZ AUC were 10-fold higher than for R-FBZ/FBZ AUC (Fig 2, Table 4).
For the SC groups, the following PK parameters were calculated: 10 mg/kg SC single dose (Table 5, Fig 2): AUC0-last = 2.3 μg•hr/mL with t last at 1512 hr; Cmax = 0.06 μg/mL at 8 hr. For the 10 mg/kg SC dose repeated for 5 days (Table 6): AUC0-last on day 5 = 15 μg•hr/mL with t last at 1512 hr; Cmax = 0.26 μg/mL at 24 hr. After a single subcutaneous administration of FBZ at 10 mg/kg, peak plasma concentrations after dosing were observed at 8 hr in the first experiment and 840 hr in the second experiment (Table 5). FBZ was measured up to the last sampling time point (1512h).
After repeated subcutaneous administration of FBZ at 10 mg/kg for 5 days, peak plasma concentrations were observed at 24 hr after the last dosing (Fig 2, Table 6). FBZ was measured up to the last sampling time point (1392 h). The H-FBZ/FBZ AUC ratio ranged from 0.2 to 1.4 and ranged from 0.001 to 0.18 for R-FBZ/FBZ AUC ratio across all FBZ dosed groups (Tables 4–6).
Although the treatment with the highest oral dose of ASD FBZ (15 mg/kg) did not result in a significant reduction in worm burden, the ultrastructural analysis of female worms recovered from jirds dosed with 1.5 mg/kg ASD FBZ for 5 days revealed extensive morphological alteration in embryos and developing mf. This observation is similar to what was seen in worms treated with a single SC injection of 10 mg/kg FBZ (Fig 3). Eggshells surrounding embryos were disrupted and disorganized compared to eggshells of developing embryos within the gonads of female worms recovered from control animals.
The benzimidazole anthelmintic flubendazole (FBZ) had been shown to be an excellent macrofilaricidal drug but its use was limited because of its need to be administered subcutaneously due to its low oral bioavailability [26] and adverse reactions at the site of injection [9]. Although efforts were made to reformulate FBZ as an oral drug [26–28], it did not move to clinical development. The purpose of the present study was to assess the efficacy of a new amorphous solid dispersion oral formulation of FBZ (ASD FBZ). ASD FBZ was administered orally to Brugia pahangi infected jirds at doses of 0.2, 0.6, 1.5, 6 and 15 mg/kg for 5 days to assess the effects of this new formulation of FBZ on adult worm burden, number of female worms and microfilarial counts. Although there was no statistically significant decrease in total number of worms, female worms or mf recovered from treated animals compared to untreated animals, ASD FBZ appeared to have some effect on the fitness of female worms and their fecundity as evidenced by the relatively lower number of female worms (60–88% reduction) and mf (59–96% reduction). This may suggest that ASD FBZ had an effect on female worm viability, which in turn, caused the reduction in the number of mf that were shed.
Hübner et al. observed histological damage in female Litomosoides sigmodontis from jirds given oral ASD FBZ doses of 6 and 15 mg/kg for 5 days and with 2 and 6 mg/kg for 10 days [32]. Degenerative changes were seen in the body wall, intestine and uteri, with major structural damage to the developing mf. In the present study, TEMs of ovaries and uteri from B. pahangi females recovered from jirds treated with 1.5 mg/kg ASD FBZ for 5 days showed that this treatment also caused extensive damage to the developing mf. Thus ASD FBZ appears to cause similar effects in the Brugia/jird model as in the Litomosoides/jird model.
TEMS also revealed extensive ultrastructural damage to mf from female worms retrieved from animals given only a single SC injection of FBZ suggesting that a low dose of orally administered ASD FBZ given for 5 days may cause damage to developing mf, similar to when animals are given a single 10 mg/kg FBZ injection. In a study by Franz et al., a single SC dose of 25 mg/kg of FBZ impaired cell division in oogonia and embryonic cells in B. malayi female worms [36]. The damage to developing mf and female fecundity may account for the reduction in the number of mf (albeit not statistically significant reduction) seen in treated jirds in the present study.
With the exception of one jird, all animals treated with oral ASD FBZ had mf at the time of necropsy which suggests ASD FBZ may have inhibited embryogenesis but it did not have a direct-acting effect on the mf that were already shed. In vitro studies by O’Neill et al. showed that concentrations as low as 100 nM of commercially purchased FBZ induced damage to the early embryonic developmental stages of B. malayi but caused little to no damage to the later stages (pretzel and stretched mf) in female worms cultured for 3 days [37]. Similar effects were observed when B. malayi adult worms were first incubated in 100 nM FBZ in vitro for 24 hr and subsequently implanted intraperitoneally into jirds. Female worms removed from jirds 8 weeks later contained fewer embryos, a larger number of degenerating embryos and released fewer mf compared to the controls [38]. Furthermore, a study by Sjoberg et al. found that ASD FBZ given orally to SCID mice at a dosage of 2 or 40 mg/kg was not directly microfilaricidal to B. malayi mf circulating in the blood for 2 days. They concluded that elevated exposures would not likely cause rapid killing of bloodborne mf, a major concern due to the severe adverse events occurring in individuals who had been treated with the microfilaricidal drug ivermectin while infected with high numbers of Loa loa mf [33].
In the present study, only 2 of the 6 animals from the group given SC injections of FBZ for 5 days had mf at the time of takedown. In this group, no female worms (nor male worms) were recovered from any of the jirds at necropsy. Since ASD FBZ does not appear to be microfilaricidal, the lower number of mf that were found in this group compared to the control group, were not due to any direct-acting effect but rather due to the elimination of female worms early in the dosing period.
The PK analyses indicated that FBZ given SC led to a slow FBZ release from the injection site and the low plasma levels remained constant up to the study endpoint. In contrast, after oral administration of ASD FBZ, higher plasma concentrations were observed with a rapid decline after the Cmax. Thus, it appears that the slowly released and sustained levels of FBZ following repeated SC injections for 5 days were highly effective in eliminating worms while the levels of ASD FBZ at the dosages given, were not effective in reducing the adult worm burden, number of female worms or mf.
The ASD formulation was selected based on data obtained in rats after testing at least 8 formulations in order to improve the bioavailability of the drug [39]. The selection of formulation took into account the feasibility and stability of the formulation and exposures. The ASD formulation was also tested in another study with female jirds infected with Litomosoides sigmodontis at the same oral doses of 6 and 15 mg/kg for 5 days [32]. At Day 5, the exposures in the present study were 2–3 fold lower than the Hübner study. Compared to an aqueous hydroxypropyl—β cyclodextrin solution that was administered to jirds at 5 mg/kg in a study by Ceballos et al. the exposure in the present study was slightly lower, most likely because the stability of FBZ in hydroxypropyl—β cyclodextrin solution was not optimal [27]. Flubendazole exposures are quite different across species, and this difference needs to be taken into consideration since both R-FBZ and H-FBZ may be potentially active in humans [40, 41].
Because Mongolian jirds are capable of maintaining Brugia pahangi infections for approximately 2 years, long-term efficacy studies should also be considered when evaluating macrofilaricidal drugs that may require long exposures. ASD FBZ given orally for 5 days did not significantly reduce worm burden but embryogenesis appears to have been affected at a dose as low as 1.5 mg/kg as evidenced by the damage to developing mf at the ultrastructural level. An extended takedown time beyond the 2-month time period may be required to observe a significant reduction in total number of worms and female fecundity or viability. Future studies should include longer takedown times with embryogram analyses to further substantiate the effects on female worm sterilization and fecundity when evaluating oral ASD FBZ as a macrofilaricidal drug.
|
10.1371/journal.pntd.0006185 | Host plant forensics and olfactory-based detection in Afro-tropical mosquito disease vectors | The global spread of vector-borne diseases remains a worrying public health threat, raising the need for development of new combat strategies for vector control. Knowledge of vector ecology can be exploited in this regard, including plant feeding; a critical resource that mosquitoes of both sexes rely on for survival and other metabolic processes. However, the identity of plant species mosquitoes feed on in nature remains largely unknown. By testing the hypothesis about selectivity in plant feeding, we employed a DNA-based approach targeting trnH-psbA and matK genes and identified host plants of field-collected Afro-tropical mosquito vectors of dengue, Rift Valley fever and malaria being among the most important mosquito-borne diseases in East Africa. These included three plant species for Aedes aegypti (dengue), two for both Aedes mcintoshi and Aedes ochraceus (Rift Valley fever) and five for Anopheles gambiae (malaria). Since plant feeding is mediated by olfactory cues, we further sought to identify specific odor signatures that may modulate host plant location. Using coupled gas chromatography (GC)-electroantennographic detection, GC/mass spectrometry and electroantennogram analyses, we identified a total of 21 antennally-active components variably detected by Ae. aegypti, Ae. mcintoshi and An. gambiae from their respective host plants. Whereas Ae. aegypti predominantly detected benzenoids, Ae. mcintoshi detected mainly aldehydes while An. gambiae detected sesquiterpenes and alkenes. Interestingly, the monoterpenes β-myrcene and (E)-β-ocimene were consistently detected by all the mosquito species and present in all the identified host plants, suggesting that they may serve as signature cues in plant location. This study highlights the utility of molecular approaches in identifying specific vector-plant associations, which can be exploited in maximizing control strategies such as such as attractive toxic sugar bait and odor-bait technology.
| Plants play an important role in the fitness of mosquito disease vectors, yet the identity of plant species that they feed on in their natural habitats remains largely unknown. In this study, we employed DNA barcoding to identify the plant species fed upon by Aedes aegypti, Aedes mcintoshi, Aedes ochraceus and Anopheles gambiae in their natural habitats. Since plant feeding is mediated by olfactory cues, with potential application as attractant-based tools for vector surveillance, we identified specific odor signatures that may modulate host plant location. Our findings showed preference in odor detection among the vectors for different compound classes; benzenoids for Ae. aegypti, aldehydes for Ae. mcintoshi and sesquiterpenes and alkenes for An. gambiae. This study highlights the utility of molecular approaches in identifying specific vector-plant associations, a knowledge which can be exploited in maximizing vector control strategies such as attractive toxic sugar bait. Furthermore, the elucidation of potential odor signature lays foundation for development of plant odor-bait technology which is critical for surveillance of different mosquito disease vectors of varying physiological states and the pathogens they transmit.
| There has been an increase in the incidence of vector-borne diseases, key among them arboviral diseases such as dengue, chikungunya, Rift Valley fever (RVF) and zika. While dengue predominantly affects Asian countries, increasing outbreaks in East African coastal regions have become evident in recent times [1–3]. Rift Valley fever which mainly occurs in Africa, with foci in East Africa, is rapidly spreading eastwards into Asia and the Arabian Peninsula [4–6]. The recent upsurge in dengue incidence has been attributed to rapid and unplanned urbanization which creates conducive breeding habitats for the key mosquito vector of the disease, Aedes aegypti [3, 7]. On the other hand, RVF is an epizootic disease mainly associated with devastating outbreaks following widespread elevated rainfall, leading to flooding that creates favorable breeding sites for the primary mosquito vectors Aedes mcintoshi and Aedes ochraceus [5, 8]. In East Africa, in particular Kenya, the public health burden due to these arboviral diseases is further compounded by the endemicity of the parasitic malaria disease transmitted by certain species of the Anopheles mosquito [9]. Globally, vector-borne diseases pose risk of infection to more than half of the world population with more than a million deaths annually [10]. Consequently, there is renewed effort to come up with new disease control measures and vector control forms a key pillar in these efforts. Detailed understanding of the vector ecology is needed in search for novel control strategies.
Plant feeding plays a critical role in the bio-ecology of mosquito disease vectors. Several studies have demonstrated that both sexes of different mosquito species forage on plants to obtain carbohydrates required for metabolic processes vital for their survival [11–13]. Besides providing a ready source of energy for flight, fecundity and cell metabolism [12, 14, 15], plant carbohydrates are also utilized during diapause by mosquitoes such as Culex pipiens to synthesise lipid reserves [16–18]. The availability of host plants has also been shown to extend survival of Anopheles gambiae and An. sergentii, likely allowing for the completion of sporogonic cycle of malaria parasites and thereby increasing disease transmission potential of these vectors [19–22]. On the other hand, abundance of flowering plants has been linked to reduced human biting behavior by mosquito disease vectors, which can impact either positively or negatively on disease transmission potential depending on the infection status of the mosquitoes [13, 20, 23, 24]. In addition, studies have shown that both mosquitoes and sandflies imbibe plant secondary metabolites during plant feeding, some of which reduce parasite load in the vector [21, 25, 26]. This has led to the hypothesis of possible self medication by these disease vectors [15, 26] opening up new avenues for exploiting phytochemicals in development of novel chemotherapeutics against the pathogens they transmit. Thus, beyond the provision of nutrients, understanding plant feeding in disease vectors offers promising opportunities for development of new control strategies against the myriad of vector borne diseases. Despite this, little is known about plant feeding behavior of mosquito vectors of dengue and RVF.
Previous studies have demonstrated that mosquitoes are highly selective in their choice of plants [27–30]. These inferences were drawn from semi-field and field experiments which either involved feeding mosquitoes on randomly selected peri-domestic plants to determine their acceptability [27, 28] or determining the attraction of mosquitoes to randomly selected fruits/seedpods and flowering plants [29–31]. Determination of plant feeding among various mosquito species has mainly been based on analytical techniques such as cold anthrone tests to detect fructose in the crop of field collected mosquitoes, chromatographic methods to detect plant sugars and cellulose staining to detect plant tissue feeding [12, 16, 19, 23, 24, 28, 32], with little to no direct field observations of mosquitoes feeding on plants [33, 34]. While contributing immensely towards our understanding of the role of plant feeding in the vectorial capacity of mosquitoes, these methods are, however, limited by their inadequacy in determining the precise host plants in the natural mosquito habitats. As diverse plants often occur in each habitat, the critical question of which plants, if any, are foraged upon by mosquito disease vectors, remains unanswered.
Recent advances have seen the application of DNA barcoding targeting specific genes to profile plant species fed upon by disease vectors [35, 36]. However, this has not been applied for any Afro-tropical disease vector, thus far. By employing DNA barcoding targeting multiple gene loci, we tested the hypothesis that Afro-tropical disease vectors feed on certain plants in their respective ecologies. We focused on four mosquito species which transmit dengue (Aedes aegypti), Rift Valley fever (Aedes mcintoshi and Aedes ochraceus) and malaria (An. gambiae) [3, 37–39]. These diseases rank among the most important vector-borne diseases in Kenya with some having been associated with large outbreaks affecting humans in the recent past [2, 5, 6, 9]. Given the central role of olfactory cues in locating this key plant resource [40–42], we further used coupled gas chromatography/mass spectrometry and electrophysiological assays to test the hypothesis that these disease vectors use unique odor bouquet to locate their suitable natural host plant. Our results show that the four Afro-tropical mosquito species feed on certain plant species within their ecological range and detect common and specific chemical cues to locate their suitable host plants. This study provides useful insight that can inform vector control strategies targeting plant feeding behavior such as attractive toxic sugar bait and odor bait technology.
Mosquito samples were obtained from three sites in Kenya: Ae. aegypti from Kilifi (3.6333° S and 39.8500° E) in the coastal region with high dengue endemicity [2, 3], Ae. mcintoshi and Ae. ochraceus from Ijara (1.5988° S and 40.5135° E) in north eastern Kenya which is a Rift Valley fever endemic region [5, 6] and An. gambiae s.l. from Ahero (0°10’S, 34°55’E) which is a malaria endemic area in western Kenya [9]. The trapping methods used to collect mosquito samples are described in detail in Nyasembe et al. [42]. Briefly, unlit CDC traps separately baited with linalool oxide (LO), BioGent (BG) lure and HONAD (a mixture of heptanal, octanal, nonanal and decanal) formulated from mammalian odor by Tchouassi et al. [37] were used to trap Ae. mcintoshi, Ae. ochraceus and An. gambiae; while BG sentinel traps separately baited with LO and BG lure, were used to trap Ae. aegypti. The traps were also either baited with or without carbon dioxide in the form of dry ice at all the three sites. All trappings were carried out outdoors. In Ahero and Kilifi, traps were placed in three distinct settings i) close to the homestead, ii) close to breeding sites a priori identified as positive for the specific mosquito larvae, and iii) in vegetation away from human dwelling [3, 39, 43, 44]. In Ijara, the traps were set at two distinct settings i) next to ‘dambos’ which serve as breeding sites for Ae. mcintoshi and Ae. ochraceus and ii) in bushy grasslands where pastoralists graze their livestock [6, 37, 38]. In Kilifi, trappings were carried out both during the day and night (informed by the diurnal nature of Ae. aegypti) while in Ahero and Ijara trappings were carried out during the night only (due to the nocturnal nature of mosquito vectors in these localities). Traps were emptied after 12 h and the collected mosquitoes immobilized by placing on dry ice, immediately frozen in liquid nitrogen and transported to icipe laboratories in Nairobi for further processing.
To prepare the samples for biochemical and molecular analyses, individual mosquitoes were submerged in a solution of 0.5% hypochlorite, agitated gently for 1 min with forceps, and then rinsed in double distilled water (ddH20) for 1 min. This was to remove any plant debris that may have been on the outside of the insect, which could otherwise contaminate the sample. The mosquitoes were then placed individually in a 1.5 ml sterile Eppendorf tube and macerated using round-tipped glass rods sterilized through the flame of a Bunsen burner. One hundred microlitres of absolute ethanol was added and the solution homogenized. Two sets of controls were used as follows: a) laboratory-reared An. gambiae s.s. fed on Parthenium hysterophorus (Asteraceae) overnight, and b) An. gambiae aspirated directly from P. hysterophorus field in Ahero using a backpack aspirator (3” IN-LINE BLOWER, John W. Hock Company, Gainesville, FL, USA). All mosquitoes from the controls were prepared as described above.
This was done using the cold anthrone test as described by van Handel et al., [45] as a quick initial test to detect fructose. Aliquots (50 μl) of the prepared mosquito homogenate were individually placed in the wells of a flat bottomed 96-well microtiter plate followed by 300 μl of the reaction solution comprising 0.15% anthrone (Sigma) (wt/vol) in 71.7% sulphuric acid. This was incubated at 25 oC for 60 min before being examined for color changes. In the presence of fructose, the reaction mixture changed its color from yellow to blue. The remaining aliquot of fructose-positive samples was subjected to plant DNA extraction as described below.
Plant DNA was extracted from the homogenized samples of fructose-positive mosquitoes only using the manufacturer’s protocol described by DNeasy Plant Minikit- (QIAGEN, USA) with a minor modification. The incubation period with lysis buffer AP1 and RNase was extended by 30 min while that with the elution Buffer AE was extended for 3 hr. The extracted DNA was stored at -20°C until use in PCR amplification.
Plant DNA extracted from the fructose-positive mosquito specimens was amplified targeting the trnH-psbA intergenic spacer region and maturase K (matK) gene (Table 1) using established primers. The use of more than one target was to maximize on the detection possibility as individual genes selectively amplify certain plant families [46]. Each PCR reaction (carried out in a volume of 20 μl) consisted of 7 μl template DNA, 10 μl 2x HotStarTaq Master Mix (HotStarTaq Plus Master Mix Kit, Qiagen), 0.5 μM of each primer, and 2 μl of RNase free water. A PCR negative control (RNase-free water) was routinely used. Samples were amplified using Veriti 96-well Thermal Cycler (Singapore). For trnH-psbA, the cycling parameters were 94 oC for 1 min, followed by 45 cycles of 94 oC for 1 min, 55 oC for 40 sec and 72 oC for 1 min, and final extension at 72 oC for 10 min. Similar cycling conditions were used for matK amplification with the annealing temperature set at 48 oC.
Successful amplifications were confirmed by visualizing PCR amplicons in 1% agarose gel electrophoresis. They were purified using the Exo/SAP-IT Kit for PCR product (Affymetrix Inc., USA) as per the manufacturer’s instructions and outsourced for bidirectional sequencing to Inqaba Biotechnological Industries (Pty) Ltd (Pretoria, South Africa).
The obtained plant DNA sequences for each gene were cleaned, edited and compared to reference sequences in the GenBank database [47]. In GenBank, the ‘megablast’ search option of nucleotide Basic Local Alignment Search Tool (BLASTn) [48] algorithm was used with the default search parameters. The hits with sequence identity above 96% were retrieved and added to the original sample query sequences. The sequences were aligned using ClustalW in MEGA 6 [49]. Aligned matrices were used to construct p-distance phylogenetic tree using the Neighbor Joining method for individual genes with 1000 bootstraps. Nodal support was evaluated by bootstrapping with values of 95% or more considered significant.
Further steps to confirm the plant identity included on-site identification within the specific ecologies from where the mosquitoes were sampled by a plant taxonomist (Simon Mathenge, retired from the Herbarium, Department of Botany, University of Nairobi) and comparison to established botanical database by the National Museum of Kenya (http://www.museums.or.ke).
Leaves and flowers (where applicable) of identified plants were sampled from the respective field sites for DNA extraction and sequencing. The samples were cleaned using double distilled water before obtaining approximately 100 mg wet weight of the sample which were placed in sterile 1.5 ml Eppendorf tubes. The plant samples were homogenized and DNA extracted using the DNeasy Plant Mini Kit as described above. The obtained plant DNA was similarly amplified for trnH-psbA and matK genes, processed and sent for sequencing as described above. The sequences were then aligned with those from the mosquitoes and phylogenetic trees obtained. Nodal support was evaluated by bootstrapping with values of 95% or more considered significant.
The haplotypes generated from this study have been deposited in GenBank under accession numbers MG573108, MG573126 –MG573131 (RVF vectors host plant trnH-psbA gene sequences), MG573132 –MG573139 (dengue vector host plant trnH-psbA gene sequences), MG573109 –MG573125 (malaria vector host plant trnH-psbA gene sequences), and KY308115—KY398121 (malaria vector host plant matK gene sequences).
Headspace VOCs were collected from five of the confirmed natural host plants for Ae. aegypti, Ae. mcintoshi, Ae. ochraceus and An. gambiae. The five plants included Pithecellobium dulce (Fabaceae), Opuntia ficus-indica (Cactaceae), Leonotis nepetifolia (Lamiaceae), Senna alata (Fabaceae) and Ricinus communis (Euphorbiaceae). This was done by collecting the headspace volatiles from these plants in situ at their natural habitats using a portable field pump (Analytical Research Systems, Gainesville, Florida, USA). The aerial parts of an intact plant were gently enclosed in an air-tight oven bag (Reynolds, Richmond, VA, USA) and charcoal filtered air passed over the plant at a flow rate of 350 ml/min into a Super-Q adsorbent trap (30 mg, Analytical Research Systems, Gainesville, Florida, USA). The aerial plant parts enclosed in the oven bags included leaves, flowers and pods of P. dulce and S. alata, leaves and flowers of L. nepetifolia, leaves and leaf stalks of R. communis, and leaves, flowers and fruits of O. ficus-indica. For all plant species, volatiles were collected for 12 hr during the day and 12 hr at night and replicated three times using different plants in each replicate. The Super-Q traps were eluted with 200 μl GC/GC-MS-grade dichloromethane (DCM) (Burdick and Jackson, Muskegon, Michigan, USA) and the eluents stored at -80 °C until analysis.
For quantification and identification of the constituent compounds of the plant volatiles, an aliquot (1 μl) of each sample was injected into a gas chromatograph (Agilent technologies-7890) coupled to inert XL EI/CI mass spectrophotometer (5975C, EI, 70eV, Agilent, Palo Alto, Califonia, USA) (GC/MS) in a splitless injection mode. The GC was equipped with an HP-5 column (30 m x 0.25 mm ID x 0.25 μm film thickness, Agilent, Palo Alto, California, USA), with helium as the carrier gas at a flow rate of 1.2 ml/min. The oven temperature was held at 35 °C for 5 min, then programmed to increase at 10 °C/min to 280 °C and maintained at this temperature for 10 min. The volatile organic compounds were identified by comparing their mass spectra with library data (Adams2.L, Chemecol.L and NIST05a.L) and with those of authentic standards where possible (see sources and purity under chemical section below). The absolute areas of each constituent as calculated by the NIST05a.L software was used to estimate their amounts using an external calibration equation generated from known amounts of authentic compounds.
To isolate the specific VOCs that are detected by the different mosquito disease vectors and their preferred natural host plants, wild caught adult Ae. aegypti, Ae. mcintoshi and An. gambiae s.l., were collected from their respective habitats using methods described above and transported alive to the icipe laboratories in Nairobi under high containment level and directly used in electrophysiological assays. The trapped mosquitoes were aspirated into 30 x 30 x 30 cm cages and provided with 10% glucose solution soaked in cotton wool during transportation. The tops of the cages were covered with a moist towel to maintain high humidity. Once at icipe, the mosquitoes were kept in a high containment animal rearing unit at a temperature of 27–31°C and average humidity of 80%. Only female mosquitoes were used for electrophysiological assays and they were starved for 2 hr before experimentation. Anopheles gambiae s.s. antennal responses to R. communis headspace volatiles had been tested in our previous study [40], hence was not repeated in this study. In addition, Ae. ochraceus was not used in these studies as none were collected during this field sampling.
Coupled gas chromatography/electro-antennographic detection (GC/EAD) analyses were performed as described by Nyasembe et al. [40]. Briefly, 5 μl of volatile samples were analyzed using a Hewlett-Packard (HP) 5890 Series II gas chromatograph equipped with an HP-5 column (30 m x 0.25 mm ID x. 0.25 μm film thickness, Agilent, Palo Alto, California, USA) with nitrogen as the carrier gas at 1 ml/min. Volatiles were analyzed in the splitless mode at an injector temperature of 280 °C and a split valve delay of 5 min. The oven temperature was held at 35°C for 3 min, then programmed at 10 °C/min to 280 °C and maintained at this temperature for 10 min. The column effluent was split 1:1 after addition of make-up nitrogen gas for simultaneous detection by flame ionization detector (FID) and EAD. For EAD detection, silver-coated wires in drawn-out glass capillaries (1.5 mm I.D.) filled with Ringer saline solution served as reference and recording electrodes. Live mounting in which the mosquito was restrained with an adhesive tape with the reference electrode connected to the base of the head and the recording electrode connected to the tip of the antennae. The analogue signal was detected through a probe (INR-II, Syntech, Hilversum, the Netherlands), captured and processed with an intelligent data acquisition controller (IDAC-4, Syntech, the Netherlands), and later analyzed with EAG 2000, software (Syntech). FID signals from the respective host plant volatiles that elicited repeated antennal responses in at least three replicates using fresh antennae were designated as EAD-active compounds and identified by matching them with corresponding GC/MS data and those of authentic standards.
EAG puffs were used to confirm the detection of seven EAD-active components which elicited antennal responses using synthetic standards. The seven compounds were selected based on either being detected by more than one mosquito species from the volatiles from their respective host plants, or by the same mosquito species from volatiles of different host plants. The synthetic standards were prepared at a concentration of 1 ng/μl, 2 ng/μl and 4 ng/μl in dichloromethane (Sigma Aldrich, 99.9%) and separately delivered as puffs on 1 cm X 1 cm filter paper placed in Pasteur pipettes. The puffs were delivered at 1 min interval, allowing the antennae to equilibrate post-exposure. To correct for variability in response, responses to blanks (filter paper laced with solvent only) were subtracted from each sample and antennal response values were normalized to a standard stimulus set at 100% (2 ng/μl 1-octen-3-ol, chosen based on its known attractiveness to hematophagous insects [41]. EAG puffs were replicated nine times for each dose of every stimulus.
The synthetic standards of the following EAD-active compounds were used: hexanal (Sigma Aldrich, 99%), (E)-2-hexenol (Aldrich, 96%), benzaldehyde (Sigma Aldrich, 99.5), β-myrcene (Sigma Aldrich, 99%), ocimene (International Flavors and Fragrance, New York, USA, (Z)-β-ocimene = 27%, (E)-β-ocimene = 67% and allo-ocimene = 6%), linalool oxide (Sigma Aldrich, mixture of stereoisomers with furanoid form, 99.5% and 0.5% pyranoid form), indole (Sigma Aldrich, 99%) and 1-octen-3-ol (Fluka Chemica, racemic mixture of R and S 98%).
To determine if there was any significant difference in the volatile profiles of the five plant species, ten most abundant volatile constituents in each plant species were selected. Attempts were then made to retrieve each of these compounds from the VOCs analyzed for the rest of the plant species, yielding a total of 26 different compounds. The absolute areas of these compounds were then measured and converted into a percentage of the total. These percentages were then subjected to Principal Component Analysis (PCA) to determine which ones, if any, are important in explaining the variation in the odor profiles of the five different plant species. Quantitative differences in VOCs of the five different plant species were detected using Univariate analysis of variance and Tukey post hoc test. Differences between the antennal dose responses and between the three different mosquito species, were detected using ANOVA and Tukey post hoc test. All statistical analyses were carried out at 95% confidence interval using R 2.15.1 software [50].
By applying the cold anthrone test to detect fructose as evidence of recent plant feeding, we established the degree of plant feeding among the females of four Afro-tropical mosquito species Aedes aegypti (dengue vector), Aedes mcintoshi, Ae. ochraceus (RVF vectors) and Anopheles gambiae s.l. (malaria vector) trapped from different habitats in Kenya during the long rainy season (April-June, 2014). Since no male mosquitoes were collected for RVF and malaria vectors, this analysis was limited to only female mosquitoes for all the four species. We found evidence of recent plant feeding in Ae. aegypti (17%, n = 245), Ae. mcintoshi (56%, n = 68), Ae. ochraceus (65%, n = 50) and An. gambiae (24%, n = 146).
To determine the identities of the plant species fed upon by these mosquito species in their natural habitats, we subjected aliquots of samples that tested positive for the anthrone test to DNA extraction followed by amplification targeting two plant genes; trnH-psbA and matK; and then sequencing. We observed that the success rates in amplification of plant DNA from the mosquito crop differed significantly between the two gene targets; trnH-psbA (24.5%) and matK (8.8%) (P < 0.05; Table 2). Similarly, sequencing success rates differed significantly between trnH-psbA (16.4%) and matK (1.9%) (P < 0.05; Table 2). The sequenced fragment sizes ranged from 276–617 bp for trnH-psbA and 133–846 bp for matK genes.
Blast searches of the sequences for each target in GenBank and further phylogeny showed strong support (bootstrap values 95% and above) and identified host plants as Pithecellobium dulce (Fabaceae), Senna uniflora (Fabaceae) and Hibiscus heterophyllus (Malvaceae) for Ae. aegypti (Fig 1A); Opuntia ficus-indica (Cactaceae) for Ae. mcintoshi; and O. ficus-indica and an unidentified plant species for Ae. ochraceus (Fig 1A); and Senna alata (Fabaceae), Senna tora (Fabaceae), Ricinus communis (Euphorbiaceae), Parthenium hysterophorus (Asteraceae) and Leonotis nepetifolia (Lamiaceae) for An. gambiae (Fig 1A and 1B). The plant identities were further corroborated by on-site botanical identification to confirm their presence and inclusion of matched sequences of extracted DNA in the analyses. These and putative sequences from the mosquito gut clustered together with strong bootstrap support in the phylogenetic analysis (Fig 1A and 1B).
We analyzed headspace volatiles from five of the identified natural host plants viz P. dulce, O. ficus-indica, L. nepetifolia, S. alata and R. communis. The VOCs of the five different plant species were differentiated by unique chemical constituents of varying abundance (Fig 2A). Principal Component Analysis (PCA) resolved these chemical constituents into three clusters which accounted for more than 90% of the total variation (Fig 2B). PC1 explained 38% of the variation; PC2 explained 32% while PC3 explained 22% of the variation. PC1 was weighed positively by monoterpenoids and benzenoids, predominantly unique to P. dulce, while PC2 was positively contributed to by sesquiterpenes which were characteristically abundant in L. nepotifolia (S1 Table). PC3 was positively characterized by monoterpenes which were the key constituents detected in the VOCs of R. communis (S1 Table). The headspace volatile constituents of S. alata and O. ficus-indica contained benzenoids (S1 Table). (E)-β-Ocimene was present in the VOCs of all the five host plant species, while hexanal, (E)-2-hexen-1-ol, β-myrcene, benzaldehyde, α-pinene, nonanal, linalool oxide, decanal, methyl salicylate, (E)-β-caryophyllene and germacrene D were variably present in the volatiles of two or more plant species (S1 Table). Multivariate analysis of variance revealed significant quantitative differences in the volatile profiles of the five plants (F(4, 1095) = 142.907, P < 0.001; Fig 2C).
To test if Afro-tropical mosquito species detect odors of their natural host plants, we employed coupled gas chromatography/electroantennographic detection (GC/EAD) and GC/mass spectrometry to isolate and identify VOCs that are detected by antennae of Ae. aegypti, Ae. mcintoshi and An. gambiae. Our assays revealed that the antennae of the three different mosquito species detected a total of 21 different VOCs, some of which were unique to their preferred host plants while others were common across two or more of the plant species. Antennae of Ae. aegypti detected 8 components in P. dulce headspace volatiles (Fig 3A), with 12 components detected by Ae. mcintoshi from O. ficus-indica (Fig 3B) while those of An. gambiae s.l. detected 13 and 7 components in the volatiles of L. nepetifolia and S. alata, respectively (Fig 3C and 3D). β-Myrcene and ocimene were detected by antennae of all the three different mosquito species from their respective host plants while hexanal, (E)-2-hexenol, and linalool oxide isomers, and benzaldehyde were variably detected by the three different mosquito species. On the other hand, antennae of the three different mosquito species also detected unique compounds from their respective host plants which included benzenoids (benzyl alcohol and indole) by Ae. aegypti, aldehydes (octanal, nonanal and decanal) by Ae. mcintoshi, and sesquiterpenes (β-cedrene, (E)-β-caryophyllene, α-humulene and δ-cadinene) and C-13, C-18 and C-20 alkenes by An. gambiae (Fig 3A, 3B, 3C and 3D).
We then selected compounds which were detected by two or more mosquito species for further electrophysiological assays to confirm and compare their bioactivity. These included hexanal, (E)-2-hexen-1-ol, benzaldehyde, β-myrcene, (E)-β-ocimene and (E)-linalool oxide (the mass spectra of these six compounds are provided in S1A–S1G Fig). In addition, following isolation of indole as an EAG-active VOC from Ae. aegypti host plant and its known role as an oviposition cue for different mosquito species [51], we also included it in these electrophysiological assays. 1-octen-3-ol was used as reference compound. Antennal responses of the two Aedes species Ae. mcintoshi and Ae. aegypti to the seven compounds tested were dose dependent, while that of An. gambiae was dose-dependent to three of the compounds including β-myrcene, (E)-β-ocimene and indole (Fig 4).
β-Myrcene and (E)-β-ocimene elicited consistent dose-dependent antennal response across the three different mosquito species. In depth analysis revealed significant differences in odor detection intensities between the three mosquito species (F(2, 163) = 6.492, P < 0.01), with linalool oxide, (E)-2-hexenol, hexanal and indole variably detected by the three mosquito species (Fig 5).
Our findings confirm that plant feeding is common among the four Afro-tropical mosquito vectors of dengue, RVF and malaria among other diseases, evidenced by a significant proportion of anthrone-positive mosquitoes. For the first time, we also identify the host plants fed upon in nature from the habitats of these vectors using DNA barcoding. Plants identified included P. dulce, S. uniflora and H. heterophyllus for Ae. aegypti, O. ficus-indica for Ae. mcintoshi and Ae. ochraceus, and L. nepetifolia, S. alata and S. tora for An. gambiae. This study represents the first evidence of plant feeding among RVF vectors. Also, some of the plant species which had been presumed to be potential host plants for malaria vectors due to their presence near human dwellings in malaria endemic regions were confirmed as host plants. These included the highly aggressive invasive plants P. hysterophorus and R. communis [27, 28]. The implications of these findings in the context of control of mosquito-borne diseases include: 1) the precision of attractive toxic sugar baits can be greatly improved by application of the baits on preferred natural host plants as opposed to random selection of plants to be laced with insecticides, 2) some of these plants might have metabolites that impact on pathogen-vector interactions which can be exploited for development of chemotherapeutics and transmission blocking agents, and 3) chemical cues utilized by these mosquito species in locating their preferred natural host plants can be harnessed for development of odor-bait technology to be used in vector surveillance and control.
Previous studies have suggested that mosquitoes mainly feed on plant nectars, extrafloral nectaries and honeydew [41], with limited evidence of tissue feeding [19]. The isolation of plant DNA from field collected mosquitoes points to plant tissue feeding in addition to nectars. Plant nectars are mainly composed of sugars, with some plants having small amounts of amino acids and proteins [52, 53], and trace amounts of DNA [54]. Our results show discrepancy in the number of fructose positive mosquitoes and those from which plant DNA was successfully isolated, with only about 25% and 8% of fructose positive mosquitoes amplified for trnH-psbA and matK gene targets, respectively. Junilla et al. [35] attributed the detection of plant DNA in mosquitoes fed on flowering plants to possible presence of DNA in the nectar or plant tissue feeding. We have observed in our laboratory plant feeding assays that mosquitoes feed not only on plant nectar but also pierce through the stems and leaf stalks (data not included). Similar observations were made by Junilla et al. [35]. These observations strongly suggest that mosquitoes possibly pierce through plant tissue to draw nutrients from plant sap in addition to nectars. In such a scenario, the mosquitoes would be predisposed to a few plant metabolites which may impact on their fitness and pathogen transmission potential, as evidenced in previous studies [21, 25, 26]. Given that these Afro-tropical mosquito species can discriminate their host plants from a plethora of plant species present in their habitats, with more than one host plant identified for three of the mosquito species, it is possible that the mosquitoes forage on different host plants for different fitness-related benefits. This, however, does not rule out the possibility of chance feeding depending on the seasonal availability of a given plant species, an aspect that warrants further research.
The present study further documents both qualitative and quantitative differences in the VOCs of five of the identified host plants. The variable headspace volatile profiles are not surprising; similar observations even within the same plant species from different cultivars, seasons and geographical locations has been made before [55–57]. Besides, it is probable that these plants utilize different metabolic pathways to give them the unique fragrance necessary for a competitive advantage in the event of scarcity of certain shared resources such as pollinators, parasitoids and self-defense [58]. Olfactory cues play a central role in herbivorous insect- host plant interactions, as evidenced by previous studies [40, 59, 60]. In a complex environment permeated with many odor plumes from different plant species, plant feeding insects are expected to evolve mechanisms that allow them to discriminate biologically relevant chemical cues for resources that confer fitness [60]. Consequently, the finding that mosquitoes can discriminate beneficial host plants from a plethora of plant species in their habitats with variable VOCs is intriguing. Noteworthy, however, is the fact that some of the VOCs such as (E)-β-ocimene, β-myrcene, hexanal, (E)-2-hexen-1-ol, benzaldehyde, α-pinene, nonanal, linalool oxide, decanal, methyl salicylate, (E)-caryophyllene and germacrene D were common to more than half of the plants analyzed, albeit in variable amounts. These compounds have been implicated in plant-insect interactions either as pollinator attractants or in plant defense to attract natural enemies of detrimental herbivores [61, 62]. In addition, some of these compounds have also been shown to be utilized by disease vectors to locate either vertebrate host [37, 44] or host plant [40, 41].
The electrophysiological assays revealed a range of specific compounds from different host plants which elicited antennal activity. Among VOCs that were common across two or more plant species, β-myrcene, hexanal, (E)-2-hexen-1-ol, benzaldehyde and the different isomers of ocimene and linalool oxide were detected by two or more mosquito species from the VOCs of their respective host plants. On the other hand, the three different mosquito species also detected unique classes of VOCs from their host plants. These included benzenoids by Ae. aegypti, aldeydes and a benzenoid by Ae. mcintoshi, and sesquiterpenes and alkenes by An. gambiae. These results point to the adaptive nature of plant odor reception by different mosquito species that allows them to discriminate beneficial from non-beneficial plants. (E)-Linalool oxide and (E)-β-ocimene have been shown to be among the VOCs from different plant species that elicit antennal activity and behavioral responses in Ae. aegypti, An. gambiae and Culex pipiens [40, 43, 56, 63]. On the other hand, linalool oxide and benzaldehyde have also been reported as components of human odors that elicit antennal activity in An. gambiae [64] and Ae. aegypti [65], respectively. Interestingly, indole which has been reported as an oviposition site attractant for different Culex and Aedes species [51], was antennally detected by Ae. aegypi from the VOCs of its host plant P. dulce. Indole has been shown to be a by-product of both bacterial degradation of tryptophan [66] and plants [67]. Taken together, these overlaps in odor detection by different mosquito species across different host plants, with some of the compounds having been identified from vertebrate hosts and oviposition sites, point to a conserved nature of receptors for certain biologically relevant chemical cues.
To further assess the potential differences in specificities of plant odor detection between the three different mosquito species, we conducted dose-dependent electrophysiological assays using seven of the identified VOCs. The seven compounds included hexanal, (E)-2-hexen-1-ol, benzaldehyde, β-myrcene, (E)-β-ocimene, (E)-linalool oxide and indole. Our results showed variable dose response detection to the seven compounds by the three different mosquito species in electrophysiological assays, with β-myrcene and (E)-β-ocimene eliciting significant dose response across all the three species. There were no differences in the sensitivity of the three different mosquito species to β-myrcene and (E)-β-ocimene, possibly pointing to the conserved nature of the receptors for these two compounds across different mosquito species. Thus, it is possible that mosquitoes use both β-myrcene and (E)-β-ocimene to detect the presence of potential host plants, with the other volatiles playing a background role to the overall chemical signature in determining the suitability of the plant as a potential nutrient source. Molecules of high biological significance have been suggested to be encoded by narrowly tuned odor receptors [68], which have been shown to be highly conserved both quantitatively and qualitatively across different mosquito species [51]. While considerable effort has been dedicated towards structural elucidation of vertebrate [42, 64] and oviposition [51] odor receptors in different mosquito species, plant odor receptors in these disease vectors is yet to be fully explored. Our results point to a similar odor partitioning in mosquito-plant interactions as is the case in vertebrate host and oviposition site location. While this study presents a significant step in elucidating important plant VOCs mediating mosquito-host plant interactions, additional studies to identify plant odor receptors are necessary to understand their nature and help narrow down on key plant VOCs that can be used in their management. In addition, further studies are needed to elucidate the role these compounds and other identified VOCs play in the behavior of the different mosquito species.
Overall, this study presents a significant milestone in the quest for novel control strategies to either supplement or replace existing ones. For both dengue and RVF, no viable vaccine or treatment exists, at least for the human cases [10]. Consequently, vector control constitutes a key pillar in their eradication/containment efforts. Besides, both diseases are characterized by cyclic patterns of outbreaks with low viral activity during the inter-epidemic periods [2, 3, 6, 38]. Thus, accurate and efficient monitoring tools are needed to predict outbreaks, a task which can be greatly complemented by plant-based odors identified in this study. Similarly, new control tools incorporating vector ecology are needed to sustain the achievements recorded in reducing malaria incidence and further move towards elimination [10].
In conclusion, this study demonstrates that the Afro-tropical mosquito species feed on various plant species available within their ecological ranges. Interestingly, they use specific chemical cues to interact with their natural host plants, some of which are common to all the three different mosquito species while others are species-specific. These findings provide a critical insight into chemical communication that underpins mosquito-plant interactions and present a unique opportunity for advancement of plant-based mosquito control strategies. We, however, take note of the fact that the plants identified in this study might not be entirely representative of the full spectra of plants fed upon by these mosquito species in their respective ecology as these are likely to vary with season. In addition, the low success rates of the two gene targets used in this study is indicative of the likelihood that some plant species might not have been detected, hence the need for further screening using additional gene targets. Nonetheless, these findings provide new insights critical in understanding the ecological drivers of the emergence of vector-borne tropical diseases and a baseline for new control strategies.
|
10.1371/journal.ppat.1002751 | Crystal Structures Reveal the Multi-Ligand Binding Mechanism of Staphylococcus aureus ClfB | Staphylococcus aureus (S. aureus) pathogenesis is a complex process involving a diverse array of extracellular and cell wall components. ClfB, an MSCRAMM (Microbial Surface Components Recognizing Adhesive Matrix Molecules) family surface protein, described as a fibrinogen-binding clumping factor, is a key determinant of S. aureus nasal colonization, but the molecular basis for ClfB-ligand recognition remains unknown. In this study, we solved the crystal structures of apo-ClfB and its complexes with fibrinogen α (Fg α) and cytokeratin 10 (CK10) peptides. Structural comparison revealed a conserved glycine-serine-rich (GSR) ClfB binding motif (GSSGXGXXG) within the ligands, which was also found in other human proteins such as Engrailed protein, TCF20 and Dermokine proteins. Interaction between Dermokine and ClfB was confirmed by subsequent binding assays. The crystal structure of ClfB complexed with a 15-residue peptide derived from Dermokine revealed the same peptide binding mode of ClfB as identified in the crystal structures of ClfB-Fg α and ClfB-CK10. The results presented here highlight the multi-ligand binding property of ClfB, which is very distinct from other characterized MSCRAMMs to-date. The adherence of multiple peptides carrying the GSR motif into the same pocket in ClfB is reminiscent of MHC molecules. Our results provide a template for the identification of other molecules targeted by S. aureus during its colonization and infection. We propose that other MSCRAMMs like ClfA and SdrG also possess multi-ligand binding properties.
| Staphylococcus aureus (S. aureus), an important opportunistic pathogen, is a major threat to humans and animals, causing high morbidity and mortality worldwide. It is responsible for a variety of infections ranging from mild superficial infections to severe infections such as infective endocarditis, septic arthritis, osteomyelitis and sepsis. Such infections are of growing concern due to the increasing antibiotic resistance of S. aureus. In order to understand the mechanism of the S. aureus pathogenesis, we studied one of the bacterial surface proteins clumping factor B (ClfB) bound by the fibrinogen α (Fg α) and cytokeratin 10 (CK10). From analyses of the high resolution crystal structures we found that the ClfB-binding peptides harbor a stretch with consensus sequence (GSSGXGXXG) that is also conserved in Engrailed protein, TCF20 and Dermokines. The interaction between ClfB and a dermokine-derived peptide was demonstrated using binding assays. Consistent with a role of ClfB in the inflammatory responses induced by S. aureus, expression of dermokines is predominant in epithelial tissues and upregulated in inflammatory diseases. The data presented in this study raise a possibility that multiple human proteins are targeted by ClfB during S. aureus infection. The multi-ligand binding feature of ClfB would be valuable for developing new therapeutic strategies.
| Staphylococcus aureus (S. aureus), an important opportunistic pathogen, is a major threat to humans and animals causing high morbidity and mortality worldwide. It is responsible for a variety of infections ranging from mild superficial infections to severe infections such as infective endocarditis, septic arthritis, osteomyelitis and sepsis [1]. Such infections are of growing concern because of the increasing antibiotic resistance of S. aureus [2], [3]. Multiple sites within the body can be colonized, including the perineum and the axilla, but the most frequent site of the carriage is the moist squamous epithelium of the anterior nares. Moreover, the organism can be disseminated from a superficial site via the bloodstream to internal organs where it can set up a metastatic focus of infection. Approximately 80% of invasive S. aureus infections are autologous in that they are caused by strains carried in the patient's nose prior to illness [4], [5].
The ability of S. aureus to cause diseases has been generally attributed to two classes of virulence determinants: cell wall-associated proteins and extracellular protein toxins. The initial step in pathogenesis is often cell adhesion, mediated by surface adhesins called MSCRAMMs (Microbial Surface Components Recognizing Adhesive Matrix Molecules) [6], [7]. To date, S. aureus is known to express more than 20 different potential MSCRAMMs [8], [9].
SD-repeat-containing (Sdr) proteins are members of the MSCRAMM family, including clumping factor A (ClfA), ClfB, SdrC, SdrD and SdrE of S. aureus and SdrF and SdrG of S. epidermidis. The Sdr proteins are characterized by the presence of an R region composed largely of repeated SD dipeptides [10]. They exhibit a comparable structural organization including an N-terminal secretory signal sequence followed by a ligand-binding A region and a dipeptide repeat region (R) composed mainly of aspartate and serine residues. The LPXTG cell wall-anchoring motif (W) immediately follows the SD-repeat region and is followed by a hydrophobic membrane-spanning domain (M) and a short positively charged cytoplasmic tail (C). Despite their conserved structural organization, the Sdr proteins are not closely related in sequence, with only 20 to 30% identical amino acid residues in the ligand-binding A domain. This suggests that different Sdr proteins might play different roles in S. aureus pathogenesis [11].
ClfB is one of the best characterized surface proteins on S. aureus during the past decade [12]–[18]. The multi-functional characteristics are quite unique to this adhesin, unlike ClfA and SdrG that have been shown to bind only to fibrinogen [19]–[21]. ClfB plays a key role in establishing human nasal colonization by binding to the human type I cytokeratin 10 (CK10) expressed on squamous epithelial cells [17], [18], [22], [23]. Consistently, recent studies have shown that the immunization of mice with ClfB reduces nasal colonization [24]. As a bifunctional MSCRAMM, ClfB also binds to fibrinogen α (Fg α), which is assumed to be significant in platelet activation and aggregation and has been shown to contribute to the pathogenesis of experimental endocarditis in rats [7], [17], [25], [26]. Unlike ClfA, FnBPA and FnBPB, which bind to the γ chain of fibrinogen, ClfB binds to repeat 5 (NSGSSGTGSTGNQ) of the flexible region of its α chain [15], [16], [27]. The repeat may form a loop, similar to the Tyr-(Gly/Ser)n Ω loops present in the C-terminus of CK10, to which ClfB also binds [18]. Fg α and CK10 harbor the same or overlapping binding sites on ClfB [18], but the detailed mechanism of ClfB recognition of Fg α and CK10 is unclear.
Structural studies suggest that ClfA and SdrG have different ligand binding characteristics and mechanisms [21], [28], although the structural organizations of the adhesion domains of these two MSCRAMMs are very similar. A “dock, lock, and latch” (DLL) model was proposed for SdrG-ligand recognition, where SdrG adopts an open conformation that allows the Fg ligand to access a binding trench between the N2 and N3 domains [21]. In ClfA, however, the cavity is preformed in a stabilized closed configuration, into which the C-terminal of the γ chain of fibrinogen threads. Therefore, the ClfA-Fg binding mechanism was proposed to be “Latch and Dock” [28].
Here we solved the crystal structures of the apo-ClfB adhesive domain and its complexes with peptides derived from Fg α and CK10. Our structures showed that ClfB recognizes its ligands in a similar manner with the DLL model. A previous study on the structures of ClfB complexed with Fg α and CK10 peptides suggested that the conserved peptide-derived motif (GSSGXG) is required for their binding to ClfB [29]. The data presented in the present study, however, support a minimal nine amino acids Gly-Ser-Rich (GSR) motif that is necessary and sufficient for binding to ClfB. Human genome mining using the motif as a template identified several candidates including Engrailed protein, TCF20 and Dermokine as potential ClfB-binding proteins. Interaction of Dermokine with ClfB was confirmed by biochemical and structural studies, which demonstrate that nearly identical mechanisms are utilized by ClfB to recognize its binding partners. Our data not only provides insights into the ligand binding mechanism of ClfB but also raises the possibility that ClfB targets multiple substrates during S. aureus infections. These results would be valuable for the development of new therapeutic strategies.
Previous studies indicated that a segment of ClfB containing N2 and N3 regions (Figure 1A) is sufficient for recognition of Fg α and CK10 [18], [27], [29]. We therefore cloned the segment encoding the two regions (amino acids 197 to 542) of the ClfB protein from S. aureus and purified the protein from E. coli for our structural studies. The structure of the ClfB(208–540)-Fg α(316–328) complex was solved by a Se-Met derived protein and was used as a starting model for determination of the other structures by the molecular replacement method (Table 1).
The apo-ClfB(197–534) structure was solved at 2.5 Å resolution, consisting of residues Ser197-Ala534 (Figure 1B). No electron density was observed for the C-terminal eight residues in the apo-ClfB structure. The polypeptide chain of apo-ClfB(197–534) is composed of two distinct domains N2 and N3, as previously described for other MSCRAMMs in S. aureus (Figure 1A) [30] . The N-terminal N2-domain contains 146 residues (amino acids 213–358) and the N3 domain 170 residues (amino acids 359–528). In the crystal structure, both N2 and N3 have two layers of β-sheets that pack tightly against each other (Figure 1B). In contrast, packing between the two domains is much looser, resulting in the formation of a large groove between them where presumably ligands bind. In N3 domain, strands A, B, E, and D form one of the two principal sheets, while strands D′, D″, C, F, and G on the opposite face present the other. Similar to the structures of other Fg-binding MSCRAMMs [21], [28], [29], [31], the structures of N2 and N3 display a typical Dev-IgG fold featured by the existence of the additional strands D′ and D″ as compared to the C-type IgG fold [30]. The structures of the N2 and N3 domains can be well superposed with an rms deviation of 0.98 Å for all Cα atoms. One structural difference between them, however, is the three-stranded β-sheet (A, B and E) on one side of N2 in comparison with a four-stranded β-sheet (D, F, C and G″) on its corresponding side in N3, as described in the structures of ClfA, SdrG and ClfB [21], [28], [29].
ClfBSer197 and ClfBLeu198 or even a short N-terminally extended segment such as the unrelated His-tag were shown to be necessary to maintain the Fg binding activity of ClfB [16], though the mechanism of how the N-terminal segment of N2 participates in substrate binding is unclear. In the crystal structure of apo-ClfB, the N-terminus (Ser197-Ala201) of one ClfB(197–542) molecule binds to the N3 domain of a symmetry-related ClfB molecule, forming a β-sheet together with the strand G (Figure 1C and Figure S1) mediated by 2 pairs of main-chain hydrogen bonds. Additional hydrogen bonds involving ClfBGln235 and ClfBVal200 further contribute to the N-terminus-mediated interaction between ClfBs (Figure 1D). These interactions may act together to stabilize the G-strand of the N3 domain, thus maintaining its Dev-IgG fold and mimicking the transition state of ligand binding.
(All structural figures in this paper were generated with PyMOL [32]).
ClfB is a key adhesin mediating S. aureus adherence by binding to CK10 and Fg [18], [27]. To study the molecular mechanisms underlying ClfB-ligand recognition, we solved the crystal structures of ClfB(208–542) in complex with CK10 (amino acids 499–512, referred as CK10(499–512)) or Fg α (amino acids 316–328, referred as Fg α(316–328)) at 2.3 Å and 1.92 Å, respectively (Figure 2A). The electron density unambiguously defines the existence of the peptides in the structures (Figure S2). In both complexes, the peptides adopt an extended conformation and are inserted into the tunnel formed between N2 and N3. Structure comparison revealed that the peptide binding induces an extension of β-strand G at its C-terminal side, which covers the bound peptides (Figure 2A and Figure S2). Similar structural features have also been observed in the structures of ClfA and SdrG complexed with their respective ligands (Figure S3) [21], [28]. Tight contacts between the peptide and the two domains in each complex result in extensive interactions, with a buried surface area of 966.6 Å2 in ClfB-Fg α(316–328) and 1002.6 Å2 in ClfB-CK10(499–512).
Structural comparison of the apo-ClfB and the two complexes shows that the RMSDs of the Cα atoms in ClfB are 0.46 Å and 0.49 Å respectively, indicating that the overall ClfB remains unchanged upon binding of the ligands (Figure 2A). Marked conformational changes, however, occur to the C-terminus of ClfB(499–512) in both complexes. In ClfB-Fg α(316–328), the residues ClfBArg529- Ser542 that are disordered in the structure of apo-ClfB become well defined following Fg α(316–328) binding. The distal C-terminus of ClfB(197–542) forms a short β-strand G′, which forms a parallel β-sheet with the β-strand E from the N2 domain. The formation of the β-sheet is mediated by several main chain and side chain hydrogen bonds (Figure 2B). The ligand-induced stabilization of the C-terminal peptide of ClfB allows it to run across Fg α(316–328) on the top. This binding mode is consistent with the DLL model as demonstrated in SdrG-Fg β complex [21], [28]. In contrast with Fg α(316–328), the peptide CK10(499–512) did not induce formation of the β-strand G′ in ClfB (Figure 2A). Nonetheless, the C-terminal portion of strand G that interacts with Fg α(316–328) also becomes well defined and caps on the CK10(499–512) peptide.
While we were preparing this manuscript, the structures of apo- and ligand binding ClfB were reported by V.Ganesh et. al [29]. Interestingly, the structural features we observed here are noticeably distinct from those of Fg α/CK10-ClfB complexes solved by them [29]. In both of their structures, particularly in the Fg α-ClfB complex, although the peptide adopts a conserved conformation as that in our structure, the C-terminus of the G-strand exhibits a different orientation and is not inserted into the N2 domain to form an extra strand G′ with the strand E, and thus the peptide is not locked in the groove between N2 and N3 (Figure 2C). In this way, their structures do not support the DLL model proposed based on the SdrG protein structure [21]. In addition, on peptide binding no rearrangement occurs to the loop between D and D′ in N2 (Figure 2C). Although the C-terminus of ClfB in the CK10-ClfB complexes has similar conformation as that in our structure, the D D′ loop in N2 domain shows no rearrangement, either (Figure 2D). The differences in the peptide conformations observed between our and Ganesh et al. works, could be attributed to the methodologies adopted in crystallization. While we co-purified the ClfB with the peptides to form a complex prior to crystallization, Ganesh et al. reported that they soaked the peptides into the apo-ClfB crystals [29]. In their structures, the conformational changes observed in our study to accommodate the peptide and then to lock it in place could have been hindered by crystal packing within the crystals.
In all, our structures strongly support the DLL model for ClfB-ligand binding. Briefly, “Dock” of the peptide triggers the rearrangement of the C-terminus of the N3 domain, allowing ClfBArg529 to form a hydrogen bond with the ClfBAsn238 from N2 domain. This would result in “Lock” of the peptide into the substrate binding groove, whereas the strong interaction between G′ and the E strand of N2 can “Latch” the peptide (Figure 2B).
In spite of the low identities in the amino acid sequences, the structures of ClfB, ClfA and SdrG exhibit high similarities (Figure 3). The most conserved residues are mainly located in the loop region of them (Figure 3B). Although the adherence domain organizations of ClfB, ClfA and SdrG and their ligand binding sites are conserved, the ligand binding specificities of the three MSCRAMMSs vary (Figure 3D) [18], [21], [28]. All the bound peptides form into a β-strand paired with the G-strand and pass through the tunnel formed by the N2, N3 and the end of the G-strand (Figure S3). In the ClfB-Fg α(316–328)/CK10(499–512) structures, one peptide is bound to one ClfB, in the same orientation as the Fg γ-chain peptide in ClfA and a reverse orientation compared to the Fg β-chain peptide in SdrG (Figure 3D) [21], [28].
In both ClfA-Fg γ and SdrG-Fg β structures, the C-terminus of the N3 domain forms a β-stand G′ (Figure 3D). ClfATyr338 that is conserved in the structures of SdrE and SdrD (data not shown), forms a hydrogen bond with the amino acid at the end of the G strand (Asn530 in ClfA), thus stabilizing the conformation of the G′ strand (Figure S4). In ClfB, the amino acid at the corresponding position is substituted with phenylalanine (ClfBPhe328) (Figure 3A). Comparison of the apo- and ligand-bound form structures of ClfB indicates that the interactions between the ligands and the G strand of N3 play a vital role in the redirection of the C-terminus of N3. ClfBArg529, the last residue in the C-terminus of the G strand in ClfB, interacts with the ligand peptides in both complex structures. ClfBAsn238 and ClfBArg529 form a stable hydrogen bond to lock the peptides into the GG′ covered tunnel. Interestingly, although in the ClfB-CK10 structure the G′ strand appears disordered, the ClfBAsn238-Arg522 hydrogen bond also exists (Figure 2B), consistent with the DLL model. Taken together, our structures strongly support the DLL model for ClfB-ligand binding.
In the crystal structures of the ClfB-Fg α(316–328)/CK10(499–512) complexes, both peptides lie down into a tunnel between N2 and N3. The peptides are covered by the C-terminal end of β-strand G (Figure S2). The C-termini of the two peptides have nearly identical conformations, with a turn formed at Fg αGly326 and CK10Gly510 (Figure S5). In contrast, the N-termini of the peptides are notably different. A sharp twist at Fg αGly318 allows the N-terminal portion of the peptide to exit the tunnel and point upward. Unlike Fg α(316–328), CK10(499–512) adopts a more extended conformation.
Numerous contacts with distances of less than 4 Å between the protein and the peptides are observed (Figure 4 and Figure S6). The interactions between ClfB with the peptides are primarily mediated through a number of hydrogen bonds. The conserved hydrogen bonds are observed between ClfB and the middle region of the two peptides (Figure 4). Hydrophobic contacts of the middle region of both peptides with the G strand of ClfB(208–542), the loop between β-sheet A, B and the loop between β-sheet C, D of N2 domain also contribute to peptide-protein interactions.
In the ClfB(208–531)-CK10(499–512) complex structure, four pairs of hydrogen bonds were observed between the main chains of the peptide and the G strand of N3 domain, resulting in the formation of a parallel β-sheet. Polar groups in side chains of ClfBTrp522, and Asn524 in N3 domain form two hydrogen bonds with the hydroxyl group of CK10Ser503. The hydroxyl group of CK10Ser503 forms the third hydrogen bond with ClfBSer376 of N3 and side-chain hydroxyl group of CK10Ser504 forms another hydrogen bond with ClfBSer236 of N2. Residues from the middle region of CK10 interact with ClfBSer236, Asp270, and Asn526 via main chain-main chain hydrogen bonds (Figures 4A and Figure S6A). Hydrogen bonds were also observed between the amino groups of CK10Ser504, and Gly506 and side chain hydroxyl or carbonyl groups of ClfBAsn234, and Asp270 in the loop region of N2. The carbonyl groups of the C-terminal residues CK10Ser508, and Ser509 interact with the side chain hydroxyl group of ClfBTyr273 in the CD-loop of N2 (Figure 4A and Figures S6A, B). The aromatic ring of ClfBTrp522 of the G strand of N3 plays an important role in anchoring the N-terminus of CK10 peptide through hydrophobic interactions with CK10Gly501 and CK10Gly502. The C-terminal segment of the peptide lies in the hydrophobic trench formed by residues of the loop region of N2 and is covered by the G strand of N3 (Figure 4A and Figures S6A,B).
Foster's study demonstrated that substitution of CK10Gly507 with the bulky residue tyrosine resulted in loss of interaction of CK10 with ClfB [33]. Structural analysis showed that the space surrounding CK10Gly507 is significantly circumscribed by its neighboring residues ClfBVal528, Gly269, Val271, and Phe328. Modeling studies (data not shown) indicated that any residue with a side chain would generate steric hindrance and cannot be accommodated in the pocket defined by the above four ClfB residues (Figure 4B and Figures S6C,D).
The Fg α C-terminal domain (amino acids 221–610) of human Fg contains ten 13-residue tandem repeats, within which up to eight residues are glycines or serines [34]. Despite the similar sequences among the repeats, only Fg α5 was shown to be recognized by ClfB [18]. The reason for this was proposed to be the presence of proline or arginine residues in the center of the putative Ω loops in the other repeats though the precise underlying mechanism remains unknown [27]. The crystal structures presented here offer an explanation for this observation. Structural comparison of the two complexes revealed that interactions of the peptides with ClfB are primarily mediated through a conserved motif in the peptides: G-S-S-G-S/T-G-S-X-G (Figure 5A). Sequence alignment of the repeats indicates that Fg α5 differs from the other repeats at the 5th, 7th and 9th positions (Figure 5B). The hydroxyl group of S/T at the 5th position is involved in hydrogen bonding interactions. On the other hand, the size of the residue at this position is limited by its neighboring residues. Thus, other residues except S/T at this position are expected to compromise the interactions between the repeat and ClfB either because of loss of hydrogen bonding interaction or generation of steric hindrance. The 7th position appears to play a role in maintaining the local conformation of the peptide by forming a γ-turn with the 9th position. In the structure of CK10-ClfB complex solved by V.Ganesh et al., the 7th position was replaced with a histidine residue, suggesting that the residue at this position can be varied (Figure S6). The G9 residue was headed to the end of the β-sheet D and the ClfBMet280 and ClfBPro281 in N2 limit residues with any side chain which would generate clash against them. In addition, a turn at the G9 is required to permit the peptide out of the tunnel, explaining why the repeat 2 with an alanine at this position cannot bind to ClfB (Figure 5B) [18].
After carefully analyzing the sequences and the peptide binding specificities of ClfB, we propose that a small motif G1-S2-S3-G4-G/S/T5-G6-X7-X8-G9 is responsible for ligand binding to the ClfB adhesive domains. Taking the Fg α(316–328)-ClfB complex as an example, within this motif, the G1 is limited by the side chain of ClfBW522 with the limitation of the space and is also required for the Fg α peptide making a turn thus exiting the tunnel. The S2 is the most critical residue because it not only forms two hydrogen bonds with the side chains of ClfBW522 and ClfBQ377 but also binds to the main chain of ClfBS376. Similar to the S2, the S3 forms two hydrogen bonds with the side chain of ClfBQ235 and ClfBS236 in the N2 domain, which could be replaced by a smaller residue such as alanine. The following residues, especially the G4, G/S/T/5 and G6, are necessary for the formation of the stable protein-peptide complex because they form hydrogen bonds with ClfB and the size of the β-sheet G covering tunnel does not accommodate residues with larger side chains. The S7 might play a role in maintaining the local conformation of the peptide by forming a γ-turn with the G9. The space for the S8 appears to be enough for residues with larger side chains (Figure 5B and Figure S6). Finally, the G9 needs to form a turn to allow the peptide out of the tunnel. Thus, the somewhat soft binding trench of ClfB would be able to bind to a series of peptides with this feature.
To further confirm our hypothesis regarding the importance of the nine-amino-acid GSR motif, we did the alanine scan using the SPR (Surface Plasmon Resonance) system with a synthetic 9-residue peptide derived from the GSR motif (GSSGSGSNG). The results are highly consistent with our structural observation and clearly show that the nine-amino-acid peptide is necessary and sufficient for binding to ClfB in vitro (Figure 6).
Our results suggest that proteins carrying the GSR motif are able to bind to ClfB. To find other potential ligands of ClfB, we searched the NCBI protein database for additional proteins containing the sequence of G1-S2-S3-G4-G/S/T5-G6-X7-X8-G9. Three proteins, TCF20, Engrailed protein and Dermokine (Derm) were found to be the hits, out of which Dermokine was evaluated more in detail in this study (Figures. 5 and 6). Dermokine is expressed in many epithelial tissues, localized to intracellular or pericellular spaces and overexpressed in inflammatory diseases. The two major isoforms α and β are transcribed from different promoters at the same locus. Recently, additional transcript variants γ, δ and ε have been identified [35], [36].
Firstly, Derm was tested for its interaction with ClfB. To this end, we synthesized a 15-amino-acid-peptide (250–264; GQSGSSGSGSNGDNN, designated as Derm15 hereafter) derived from Derm and then characterized its binding to ClfB using the SPR (Surface Plasmon Resonance) assay. In the assay, ClfB bound to the peptide with a dissociation constant of 2.37 µM (Figure 7A). Interestingly, the results also showed that the Derm peptide interacted with ClfB with slow kinetics, further supporting the DLL model (Figure 7A). To understand the molecular mechanism underlying this interaction, we solved the crystal structure of ClfB(208–542) bound to the peptide at 2.5 Å resolution. As expected, Derm15 interacts with ClfB in a nearly identical manner with Fg α(316–328) and CK10(499–512) (Figure 7 B). The ClfBArg529 forms a hydrogen bond with ClfBAsn328 and the C-terminus of N3 forms an extra strand, which is similar as that in the ClfB-Fg α(316–328) and ClfB-CK10(499–512) complexes (Figure 7 C). Mutagenesis studies were conducted to further verify the binding of Derm15 to ClfB. We replaced the residues ClfBS236, W522 that participate in interactions with the peptide with alanine respectively. The mutant proteins were purified to homogeneity and tested for their interaction with the Derm peptide using SPR. While the wild type ClfB bound tightly to Derm15, the mutant proteins ClfB(197–542) S236A or W522A exhibited much lower binding affinities with the peptide in mM range (Figure S8). Interestingly, besides the low binding affinities, both mutant proteins exhibited rapid association and dissociation behaviors in the experiments, as compared to the slow association and scarcely any dissociation behaviors observed for the wild type protein. These results indicated that the residues ClfBS236, W522 are not only involved in the binding with ClfB, but also participate in stabilizing or “locking” the peptide in place. Collectively, our results strongly support the interaction between ClfB and Derm in vitro and suggest that Derm may involve in the infection process and pathogenesis caused by S. aureus in vivo.
The colonization of the host nares by the Gram-positive bacterium S. aureus is mediated by a family of cell surface proteins which promote its adhesion to the extracellular matrix, that is, the MSCRAMMs [13], [18]. ClfB, as a component of this family protein, has been studied for the past decade and was unique in its multi-functional characteristics, as compared to ClfA and SdrG that only bind to fibrinogens [18], [19], [21], [29].
Consistent with the studies of the SdrG-fibrinogen complex [21], data from this study support the DLL binding mechanism of ClfB with the Fg α/CK10-derived peptides, but not the mechanism suggested in the previous study by V. Ganesh et al. [29]. In their work, due to the absence of the “Latch” procedure observed in the crystal structure, the binding mechanism was ascribed to the “DL” model. However, the structures of ClfB-peptide complexes solved in this study, together with the SPR data, indicate that the DLL model should be the mechanism utilized by ClfB to bind to its ligands. Our results also indicate that the DLL model may be the principal mechanism of MSCRAMM-ligand complexes.
In V. Ganesh et al.'s studies of the ClfB complexes, they proposed a common GSSGXG motif constituting the ClfB binding site [29], which is inconsistent with the previous studies on ClfB. For example, within the ten tandem Fg α repeats, repeat 2, 3, 4 and 5 all contain the GSSGXG motif but only the repeat 5 can bind to ClfB (Figure 5B) [18]. Our structural and the alanine screening analyses demonstrate that a 9-residue peptide G1-S2-S3-G4-G/S/T5-G6-X7-X8-G9 is necessary and sufficient for binding to ClfB in vitro. It is therefore predicted that a protein incorporating such a motif is able to interact with ClfB. Indeed, our biochemical assays showed that a Dermokine-derived peptide containing the ClfB binding motif interacted with ClfB (Figures 7B, 7C). Further supporting this prediction, our structural studies revealed that the binding mode of the Dermokine-derived peptide to ClfB is nearly identical with that of the Fg α/CK10-derived peptide (Figure 7B). Collectively, these findings raise a provocative possibility that ClfB might act on multiple targets during S. aureus infections. Given the fact that ClfB acts as a key determinant of S. aureus nasal colonization, this may not be totally surprising.
Interestingly, Dermokine was first identified as a gene expressed in the suprabasal layers of the epidermis, and more recently, other isoforms of this gene besides its α and β isoforms have also been found. This gene is expressed in various cells and epithelial tissues and over-expressed in inflammatory conditions [35], [36], suggesting that Dermokine might play a role in inflammatory processes since the over-expression of the mediators in immune cell activation characterizes many inflammatory diseases. ClfB is involved not only in the S. aureus's colonization of human nares but also in the diseases caused by this bacterium. Additionally, S. aureus has also been implicated in several inflammation processes including corneal inflammation. ClfB's binding to Dermokine raises the possibility that ClfB might play a role in the S. aureus caused inflammation and the Dermokine gene's over-expression might serve as biological markers whose products could bind to ClfB and participate in this process. Obviously more investigations are needed to verify ClfB-Dermokine interaction during S. aureus infections as well as the biological significance of the interaction.
The characterization of ClfB as a multi-ligand binding protein will be meaningful for the identification of putative substrates and for furthering our understanding of the S. aureus infection pathway. Our findings also provide important leads towards the development of new therapeutic agents capable of eradicating S. aureus carriage in individuals and efficiently interfering with staphylococcal infection. This is particularly important since new antibacterial strategies are in urgent need to combat the drug resistant bacteria that continuing to emerge [37], [38].
The fragment of the ClfB gene (corresponding 197–542 aa) was amplified by PCR from the S. aureus Newman genomic DNA. After digestion with BamHI and HindIII (NEB), the amplified genes were cloned into the prokaryotic expression vector pQE32 (GE Healthcare Life Sciences) to produce His-tagged fusion protein and were confirmed by DNA sequencing. The expected protein was expressed in E.coli strain BL21 with a high yield. Recombinant His-tagged protein was purified by Ni-affinity column chromatography and ion exchange chromatography. For the purification of protein-peptide complexes, the synthesized peptides were added into the concentrated protein samples at a 10∶1 ratio and further subjected to gel filtration chromatography (Superdex-75 column) using buffer (10 mM Tris-HCl pH 8.0, 150 mM NaCl, 2 mM DTT) on the FPLC system (GE Healthcare Life Sciences). The proteins from different stages of purification (i.e. affinity and gel filtration chromatography) were monitored by SDS-PAGE. The selenomethionine (Se-Met)-substituted ClfB derivative was expressed and purified similarly.
The apo-ClfB and its complexes with different peptides were concentrated to 30 mg/ml in 10 mM Tris-HCl pH 8.0, 150 mM NaCl and 2 mM DTT. Crystals were produced by the hanging-drop vapor diffusion method [39] using sparse-matrix screen kits from Hampton Research (Crystal Screen reagent kits I and II), followed by a refinement of the conditions through the variation of precipitants, pH, protein concentrations and additives.
Crystals were grown at 18°C by mixing 1.1 µl of protein with 1.1 µl of reservoir solution and equilibrating against 200 µl of reservoir solution. The apo-ClfB crystals are grown in 0.2 M LiSO4, 0.1 M Tris-HCl pH 8.5, 30% polyethylene glycol 4000 and all the complexes with peptides are grown in 0.1 M sodium citrate tribasic dehydrate pH 5.6, 20% 2-propanol and 20% polyethylene glycol 4000. Similar conditions were used for generation of the crystals of Se-Met-substituted ClfB. Native and Se-SAD data were collected at Shanghai Synchrotron Radiation Facility (SSRF) at a wavelength of 0.919 Å and 0.979 Å respectively using a MAR225 (MAR Research, Hamburg) CCD detector at 100 K and processed with HKL2000 [40]. Further processing was carried out using programs from the CCP4 suite (Collaborative Computational Project, 1994).
The selenium sites were located using SHELXs [41] from the Bijvoet differences in the Se-SAD data. Heavy atom positions were refined and phases were calculated with PHASER's SAD experimental phasing module [42]. The real-space constraints were applied to the electron density map in DM [43]. The resulting map was of sufficient quality for model building of the ClfB molecules in COOT [44]. The structures with other peptides were solved with molecular replacement methods in CCP4 and all the structures were refined with the PHENIX [45] packages. Data collection and structure statistics are summarized in table 1.
The synthesis and purification of the peptides were described previously [18], [27]. For the following peptides, the amino acid residue numbers are given and the sequences are as follows: peptide from repeat 5 of the C terminus of the α-chain of Fg (NSGSSGTGSTGNQ); a peptide in the tail region of CK10 (YGGGSSGGGSSGG); peptide 15 from Dermokine protein (SQSGSSGSGSNGDNN); The peptide 9 of GSR motif (GSSGSGSNG) and its mutated forms by alanine scan; The six-amino-acid peptide (GSSGSG).
Binding of ClfB(197–542) to peptide 15 was assessed by SPR using the ProteOn XPR36 equipment (Bio-Rad Laboratories, Inc.). Each SPR experiment used multichannel detection. The system was equilibrated with buffer (10 mM HEPES pH 7.2, 150 mM NaCl). At each channel, peptide was captured to a ProteOn NLC Sensor Chip (BIO-RAD) at 25°C, using a flow rate of 100 µl/min. This resulted in peptide coupled at response levels of 460 RU. For binding measures, ClfB(197–542) was injected simultaneously at different concentrations at a flow rate of 100 µl/min. The experiments were repeated three times.
The binding affinities between ClfB and the ten 9-amino–acid peptides and the 6-amino-acid peptide were determined by surface plasmon resonance (SPR) using BIAcore T200 instrument (GE Healthcare) at 10°C. The ClfB protein was immobilized to about 5300 Response Unit (RU) on a research-grade CM5 sensor chip in 10 mM sodium acetate, pH 5.0 by standard amine coupling method. The flow cell 1 was left blank as a reference. For the collection of data for affinity analyses, the 11 peptides in a buffer of 10 mM HEPES pH 7.4, and 150 mM NaCl, plus 0.005% (v/v) Tween 20, were injected over the flow cells at various concentrations at a 30 µl/min flow rate. The ligands were allowed to associate for 60 s and dissociate for 120 s. Data were analyzed with the BIAcore T200 evaluation software by fitting to a 1∶1 Langmuir binding fitting model.
|
10.1371/journal.pmed.1002232 | Taxes and Subsidies for Improving Diet and Population Health in Australia: A Cost-Effectiveness Modelling Study | An increasing number of countries are implementing taxes on unhealthy foods and drinks to address the growing burden of dietary-related disease, but the cost-effectiveness of combining taxes on unhealthy foods and subsidies on healthy foods is not well understood.
Using a population model of dietary-related diseases and health care costs and food price elasticities, we simulated the effect of taxes on saturated fat, salt, sugar, and sugar-sweetened beverages and a subsidy on fruits and vegetables, over the lifetime of the Australian population. The sizes of the taxes and subsidy were set such that, when combined as a package, there would be a negligible effect on average weekly expenditure on food (<1% change). We evaluated the cost-effectiveness of the interventions individually, then determined the optimal combination based on maximising net monetary benefit at a threshold of AU$50,000 per disability-adjusted life year (DALY). The simulations suggested that the combination of taxes and subsidy might avert as many as 470,000 DALYs (95% uncertainty interval [UI]: 420,000 to 510,000) in the Australian population of 22 million, with a net cost-saving of AU$3.4 billion (95% UI: AU$2.4 billion to AU$4.6 billion; US$2.3 billion) to the health sector. Of the taxes evaluated, the sugar tax produced the biggest estimates of health gain (270,000 [95% UI: 250,000 to 290,000] DALYs averted), followed by the salt tax (130,000 [95% UI: 120,000 to 140,000] DALYs), the saturated fat tax (97,000 [95% UI: 77,000 to 120,000] DALYs), and the sugar-sweetened beverage tax (12,000 [95% UI: 2,100 to 21,000] DALYs). The fruit and vegetable subsidy (−13,000 [95% UI: −44,000 to 18,000] DALYs) was a cost-effective addition to the package of taxes. However, it did not necessarily lead to a net health benefit for the population when modelled as an intervention on its own, because of the possible adverse cross-price elasticity effects on consumption of other foods (e.g., foods high in saturated fat and salt). The study suggests that taxes and subsidies on foods and beverages can potentially be combined to achieve substantial improvements in population health and cost-savings to the health sector. However, the magnitude of health benefits is sensitive to measures of price elasticity, and further work is needed to incorporate potential benefits or harms associated with changes in other foods and nutrients that are not currently modelled, such as red and processed meats and fibre.
With potentially large health benefits for the Australian population and large benefits in reducing health sector spending on the treatment of non-communicable diseases, the formulation of a tax and subsidy package should be given a more prominent role in Australia’s public health nutrition strategy.
| A growing number of countries have implemented or are considering implementing food and drink taxes or subsidies to address the growing burden of dietary-related diseases.
However, the long-term population health impact and cost-effectiveness of combining unhealthy food taxes and healthy food subsidies are not well understood.
We modelled the potential impact for the Australian population of five policy options: taxes on saturated fat, salt, sugar, and sugar-sweetened beverages and a subsidy on fruits and vegetables.
The simulations indicated that the combination of the taxes and subsidy could avert as many as 470,000 disability-adjusted life years (95% uncertainty interval: 420,000 to 510,000) in the Australian population of 22 million, at a net cost-saving of AU$3.4 billion (AU$2.4 billion to AU$4.6 billion; US$2.3 billion) to the health sector.
The sugar tax produced the biggest gains in health, followed by the salt tax, the saturated fat tax, and the sugar-sweetened beverage tax; the fruit and vegetable subsidy, while cost-effective when added to the package of taxes, did not lead to a net health benefit for the population when modelled as an intervention on its own, due to an associated increase in the consumption of other less healthy foods.
The modelling illustrates the potentially large benefits of combining food taxes and subsidies for improving population health and reducing health sector spending.
The Australian government should follow the lead of France, Mexico, the UK, and other countries in developing a policy for taxing and subsiding foods and drinks to improve public health.
While the modelling strongly indicated that there could be benefits with taxing unhealthy products, the health benefits of focusing policy only on subsidising healthy food were more sensitive to modelled data, and we recommend further research to measure and model how populations respond to healthy food subsidies.
| The relationship between non-communicable diseases and consumption of unhealthy foods and drinks is well known, with dietary factors contributing almost 10% of the global disease burden [1]. Price is a key driver of food purchasing [2], and experimental studies in real-world environments (e.g., canteens and vending machines) and virtual supermarket-type environments show that people reduce consumption of unhealthy foods when the price of these products is increased [3].
Countries have been experimenting with taxing foods and drinks since the early 1980s [4]. In most countries, taxes have been applied to food and drink items that are clearly unhealthy or a luxury item, such as sugar-sweetened beverages and confectionery, with taxes often applied as a sales tax (e.g., United States) or an import duty (e.g., Fiji, Nauru, and French Polynesia). More recently, countries such as Denmark and Hungary have implemented taxes on a wider range of foods and drinks, including products such as meat and dairy, with taxes based on the levels of saturated fat, sugar, or salt. Mexico also introduced an 8% tax on nonessential high energy density foods and a 10% tax on sugar-sweetened beverages in 2014. Evaluation of the Mexico taxes found a 5% decrease in purchases of taxed foods in the first year, compared with no change in the purchase of untaxed products [5].
Using observations of fluctuations in product prices and changes in household expenditure on foods and drinks, economists have estimated the consumer price elasticity (the change in consumption with change in price) for a wide range of foods and drinks [6–9]. Cost-effectiveness models of taxation interventions using price elasticity data suggest that a 10% tax on unhealthy foods in Australia [10], a 20% tax on sugar-sweetened beverages in Australia [11], and a tax on salt in the US [12] are all potentially cost-saving, leading to large improvements in population health and savings in disease treatment costs.
However, the scope of the health outcome modelling in the previous cost-effectiveness studies was restricted to single dietary outcomes; the Australian studies only modelled the relationship between energy intake and body mass index (BMI), and the US study only modelled the relationship between salt and blood pressure. While cost-effectiveness models have typically been narrowly focused, health models that take a broader range of dietary and health impacts into account suggest that the balance of benefit and harm for food taxes does not always lead to favourable disease outcomes [13]. While most health outcome studies indicate that taxing sugar-sweetened beverages or foods and drinks high in sugar, fat, or salt is effective in reducing energy intake and consumption of the targeted nutrients [2,13], several high-quality studies have shown that a tax or subsidy may have an unfavourable effect on nutrients that are not the intended target (e.g., a decrease in fibre [9]), and this may lead to a net increase rather than decrease in deaths [14,15].
It is possible that taxes or subsidies could be combined in a way that would achieve a net improvement in health and a net reduction in the costs of disease treatment, for example, by combining a saturated fat tax with a subsidy on fruits and vegetables. But methods for deriving an optimal package of taxes and subsidies that takes long-term population health and cost implications into account are yet to be developed. In this study, we use an Australian cost-effectiveness model to evaluate a range of food and drink taxes and subsidies, implemented individually and in all combinations, to determine an optimally cost-effective package of tax and subsidy options.
In this study, we evaluate combinations of five interventions: (1) taxing saturated fat, (2) taxing excess salt in processed foods, (3) taxing sugar-sweetened beverages, (4) subsidising fruits and vegetables, and (5) taxing processed foods high in sugar. We initially set the taxes on sugar-sweetened beverages and saturated fat to be equivalent to the taxes implemented in Denmark [16,17]. Finding no precedent for a tax on salt or subsidy on fruits and vegetables, we set the tax on excess salt (i.e., sodium in excess of maximum recommended levels for food items [18]) and the subsidy on fruits and vegetables to be equivalent in size to the Danish taxes, with adjustment of units to reflect the different food types (e.g., milligrams of sodium instead of kilograms of fat, kilograms of fruits and vegetables instead of litres of soft drink). We then scaled the magnitude of all taxes so that the combination of all five taxes/subsidies had a negligible effect on current average weekly expenditure on food (defined as a change of <1%) and checked that the expenditure effect of any one tax or subsidy also did not exceed 1%. Table 1 gives a description of the five taxation/subsidy scenarios, and S1 Table provides an illustration of the change in price for some typical purchases.
We estimated the baseline prices of all food and drink products from two surveys of the prices of all product varieties available online from Woolworths Supermarket in June 2011 (winter) and February 2012 (summer). We took the median price from all brands available within each product sub-category (wholemeal bread, white bread, etc.), adjusting prices to 2010 values (baseline year for cost-effectiveness analyses) using the consumer price index for each food [19], and taking the average of the two seasonal surveys.
Using nutrient levels (e.g., sodium, saturated fat) estimated for Australian food and drink products [20], we then determined the change in price of all products with each tax and subsidy scenario.
From Australian Health Survey data on the consumption of foods and drinks [21], and taking product wastage [22,23] into account, we determined the change in daily intake of foods (e.g., servings of fruits and vegetables) and nutrients (e.g., sodium, saturated fat) for each tax and subsidy scenario, using price elasticity data. In economic analyses, an own-price elasticity measures the change in purchase of a product with a 1% increase in its price, while a cross-price elasticity measures the change in product purchase with a 1% increase in the price of another product.
Price elasticities have been measured in a number of countries, including the UK [24], Denmark [9], US [6], New Zealand [7], and Australia [25]. The authors of the Australian elasticity study had concerns with inaccuracies in their price data; further, their study did not include beverages and was limited to a small number of food categories that are poorly matched with nutritional analyses (all products from grains were categorised into bread or rice, rather than distinguishing between breads and cereals, cakes and biscuits, pastry products, etc.). Given the limitations of the Australian data, we chose to use more recent and comprehensive food price elasticities from New Zealand in our primary analyses, even though we were modelling the health implications of food taxes and subsidies in the Australian population. While price elasticities can be influenced by cultural food preferences and levels of wealth [13], Australia and New Zealand have a common history of British colonisation and later migration, which has led to similar dietary patterns and economic systems. In addition, the close proximity of the two island nations, and their relative isolation from other Western countries, has led to a common range of food products and manufacturers. However, we also repeated the analyses with UK elasticities to explore the sensitivity of the modelling to the choice of elasticity values.
From the modelled shifts in dietary patterns, we determined changes in three disease risk factors: fruit and vegetable intake, systolic blood pressure, and BMI. Baseline prevalence of the risk factors was derived by age and sex from the Australian Health Survey 2011–2012. Change in daily fruit and vegetable intake due to each tax/subsidy intervention was determined directly from the change in dietary intake. Change in systolic blood pressure was determined from the net change in sodium intake across the diet, using regression models derived by Law et al. [26]. While there has been much recent debate about a possible J- or U-shaped relationship between sodium intake and health outcomes, we did not think the evidence sufficiently strong to model an increased risk at lower levels of sodium intake [27,28]. Change in BMI was determined from the modelled change in energy intake, using the formula described by Christiansen and Garby [29], assuming constant height and physical activity, which we estimated, by age and sex, from the Australian Health Survey 2011–2012.
We modelled the change in disease-specific incidence by using the population impact fraction equation [30] to quantify the disease risk reduction associated with each change in risk factor exposure. We modelled the effects of changes in daily fruit and vegetable intake on risk of ischaemic heart disease, ischaemic stroke, and cancers of the colon, lung, stomach, and oesophagus [31]; changes in systolic blood pressure on risk of ischaemic heart disease and stroke [32]; and changes in BMI on risk of type 2 diabetes, ischaemic heart disease, hypertensive heart disease, ischaemic stroke, osteoarthritis, and cancers of the breast (in women), colon, lung, stomach, oesophagus, endometrium, kidney, and thyroid [33].
Since type 2 diabetes is itself a risk factor for ischaemic heart disease and stroke, we explicitly modelled the contribution of changing prevalence of type 2 diabetes (due to changing BMI) to the incidence of ischaemic heart disease and stroke, using relative risks of these diseases due to diabetes from the Asia Pacific Cohort Study [34]. To prevent double-counting, we reduced the relative risks of ischaemic heart disease and stroke until the total fraction of ischaemic heart disease and stroke attributable to excess BMI was equal to the sum of the fraction contributed directly from BMI and the fraction contributed from BMI via diabetes.
In addition, we reduced the excess risk of ischaemic heart disease and stroke that is associated with decreasing BMI by 50% [35], to account for mediation via systolic blood pressure.
We modelled the effect of changing incidence of diseases on the future health of the population using proportional multi-state lifetable modelling methods that we previously developed to evaluate the cost-effectiveness of preventive interventions in the Australian health care context [36–38]. Using these methods, we simulated the 2010 Australian population (by 5-y age and sex cohorts) over time until everyone was dead or had reached age 100 y. At each year, we determined incidence, prevalence, and mortality for each dietary-related disease, as well as mortality from all other causes. All model input data are presented in S1 Text.
From the changing incidence in disease, we determined the change in years of life lived by the Australian population due to each tax/subsidy intervention, and adjusted for time spent in ill health using disability weights from the Australian Burden of Disease Study [39], to determine the total disability-adjusted life years (DALYs) averted.
We determined the change in costs of treating disease over the lifetime of the Australian population by applying unit costs (e.g., cost per prevalent case of stroke) that we estimated, by age and sex, from Australian Institute of Health and Welfare data on disease costs and impacts [40]. We included changes in the costs of treating non-dietary-related diseases in the added years of life, but reported on these separately in the cost-effectiveness analysis.
The costs of implementing new taxation/subsidy interventions for foods in Australia was estimated to be $22 million (95% confidence interval: $19 million to $24 million) in 2010 Australian dollars, based on a previous estimate of costs for implementing changes to taxation in Australia [41]. This estimate included costs of basic administration, promotion through the media, and enforcement of the new tax/subsidy interventions. We assumed that these costs would be the same for all combinations of food taxes and subsidies and modelled them as a one-off up-front cost.
Since we were interested in evaluating cost-effectiveness from a health sector perspective, we did not include any costs to food manufacturers or retailers for product reformulation or labelling changes. We did calculate changes in household expenditure due to changes in product prices and consumption patterns with each tax/subsidy option (or combination); however, since these are a transfer payment, we report the changes in household expenditure separately and do not include them in the cost-effectiveness calculations.
We evaluated the cost-effectiveness of all 31 combinations of the five taxation and subsidy options. We simulated total DALYs and costs over the lifetime of the 2010 population, discounting future DALYs and costs back to the 2010 baseline year at a rate of 3%, in line with previous analyses of preventive interventions in Australia [36–38].
Each scenario was run 2,000 times in Monte Carlo analysis, to determine 95% uncertainty intervals (UIs) and probabilities of cost-effectiveness. The uncertainty in model input parameters (food consumption, proportion of purchased food that is wasted, food price elasticities, relative risks of disease, etc.) is presented in S2 Table. We present the probability of each tax and subsidy option being “dominant” (i.e., leading to an increase in health and net cost-savings) and being cost-effective against a threshold of AU$50,000 per DALY averted [36–38].
To determine the optimal package of tax and subsidy options, we calculated net monetary benefit [42] for all tax and subsidy combinations at cost-effectiveness thresholds between AU$0/DALY and AU$1 million/DALY. The optimal combination of the five tax and subsidy options was then determined by selecting intervention options in order of highest probability of maximum net benefit when incrementally added to the package at the AU$50,000 per DALY threshold.
We explored a number of possible demand- and supply-side responses to the imposition of food taxes and subsidies in scenario analyses. These included feasibility constraints on changes in total energy intake and total weight of foods consumed, food industry reformulation of foods to avoid taxes, and under- or over-shifting of price changes on taxed products.
For the constraint on energy intake, we restricted the change in average daily energy to ±250 kJ/d, which is approximately 3% of average daily energy intake, and roughly equivalent to the energy provided by a small apple. We did this by proportionally scaling food intake until the change in total daily energy did not exceed the target amount. Similarly, for the constraint on weight of foods consumed, we restricted the change in total mass of the diet to ±50 g/d, which is approximately 3% of the average weight of the current diet, and roughly equivalent to the weight of half of a medium-sized banana.
For the reformulation scenario, we estimated the impact if product manufacturers reduced the salt content of foods to the maximum recommended level [16], replaced oils high in saturated fat (e.g., palm oil) with oils lower in saturated fat (e.g., canola), replaced sugars in sugar-sweetened beverages with artificial sweeteners, and reduced sugars to no more than 10 g per 100 g in non-fruit-based products high in sugar (by reducing added sugars or replacing sugars with artificial sweeteners).
For the under-shifting of price changes, we assumed that price changes due to taxes would be reduced by 20% on products from the three companies with largest retail volume in Australia [38], assuming that these companies may have the greatest capacity to absorb costs. The proportion of all products from the three largest companies varies by food and drink category, ranging from 0% market share for around 60% of products up to more than 40% market share for sugar-sweetened beverages and some confectionery (S3 Table). We assumed that all other companies passed on the full price change due to taxes to the consumers. Our assumptions were similar for the over-shifting of price changes, but we assumed that price changes due to taxes were instead increased by 20%.
The effects of the tax and subsidy options on daily intake of fruits and vegetables, sodium, and total energy are shown in Table 2. Only the sugar tax led to improvements in all dietary measures—a reduction in sodium and energy intake and an increase in fruit and vegetable intake. The taxes on saturated fat, salt, and sugar-sweetened beverages all led to improvements in sodium and energy intake, but due to cross-price elasticity effects, there was an accompanying decrease in fruit and vegetable intake. The subsidy on fruits and vegetables led to an increase in fruit and vegetable intake, but an undesirable increase in sodium and energy intake.
There was a big difference in outcomes between the subsidy intervention and the four tax interventions. When modelled over the lifetime of the population, the fruit and vegetable subsidy did not lead to an improvement in health or a reduction in disease treatment costs. The intervention was dominated, with only an 11% probability of being cost-effective against a AU$50,000 per DALY threshold.
In contrast to the subsidy intervention, the taxes on saturated fat, salt, sugar-sweetened beverages, and sugar all led to an improvement in population health, which ranged from 12,000 DALYs (95% UI: 2,100 to 21,000) averted for the sugar-sweetened beverage tax up to 270,000 DALYs averted (95% UI: 250,000 to 290,000) for the sugar tax (Table 3). Although the costs of treating dietary-related diseases were reduced with each of the taxes, this was partly countered by an increase in the costs of treating non-dietary-related diseases in the added years of life. Nevertheless, all of the tax interventions were cost-saving (dominant) for the health sector when all disease costs were included in the cost-effectiveness analysis.
The highest net monetary benefit was achieved by implementing the sugar tax first, followed by the salt tax, saturated fat tax, sugar-sweetened beverage tax, and fruit and vegetable subsidy (Fig 1). Interestingly, the fruit and vegetable subsidy, although dominated when implemented on its own, was cost-effective when added to the combination of taxes. The median cost-effectiveness of adding the fruit and vegetable subsidy was AU$18,000/DALY, but the probability of being under the AU$50,000/DALY cost-effectiveness threshold was only 54% (S4 Table). The combination of all five tax and subsidy interventions led to an increase in population health of 470,000 DALYs (95% UI: 420,000 to 510,000) averted and a net cost-saving of AU$3.4 billion (AU$2.4 billion to AU$4.6 billion), and had a 100% probability of cost-savings.
The package of interventions was dominant under all scenarios we evaluated (Table 4). Constraining change in dietary energy intake to no more than 250 kJ/d (approximately 3% of average daily intake) reduced the health impact by 28%, although constraining change in the weight of foods consumed had no impact, because the changes were within the maximum of 50 g/d (approximately 3% of average daily consumption). A 20% over- or under-shifting of prices by the three largest product suppliers had only a ±3% impact on health outcomes. Food reformulation by product manufacturers—to reduce excess salt, replace sugars with artificial sweeteners, and replace unhealthy fats with healthier options—approximately doubled the potential population health gain.
When analyses were run with UK rather than New Zealand elasticities, all of the tax scenarios were still dominant and the subsidy intervention was still dominated (Fig 2). The health gain was around 1.5- to 5-fold greater for the saturated fat, salt, and sugar-sweetened beverage taxes, but around half the size for the sugar tax; the health gain for the fruit and vegetable subsidy was virtually identical. However, although there was variation in the magnitude of health gain with the two different sets of price elasticities, the effects largely cancelled each other out, such that, overall, there was relatively little difference when the tax and subsidy interventions were combined as a package (Fig 2F).
Our modelling using price elasticity data and a proportional multi-state lifetable model to simulate the health of the population into the future suggested that a combination of taxes on saturated fat, salt, sugar, and sugar-sweetened beverages and a subsidy on fruits and vegetables would very likely be cost-saving if implemented in Australia. The benefits to population health could be substantial. If implementing only one intervention, we found that any of the taxes could be a good option, leading to cost-savings for the health sector. Of the options we evaluated, the sugar tax was most likely to bring the biggest gains in health, followed by the salt tax, the saturated fat tax, and the sugar-sweetened beverage tax. However, we would not recommend the fruit and vegetable subsidy as an intervention on its own since our results suggest that it may lead to increased costs and no net health benefit for the population.
It is important to keep in mind that these results have used price elasticities derived in New Zealand, rather than Australia. While there are many similarities between the countries, in terms of geographical location, culture, history, income, diet, and product choices, there are still some differences. For example, Australia is a much larger island, with greater extremes of climate and a more urbanised population. These factors may influence the storage and transport of food, retail practices, product availability, and food quality, all of which may affect people’s preferences and ultimately their sensitivity to changes in prices. However, it is reassuring that when analysed with UK elasticities, where the dietary patterns and products are likely to be even more divergent, the cost-effectiveness of the tax and subsidy interventions and the combined health gain were similar, although there was some variation in health gain with individual taxes.
In addition to variations in diet and price sensitivity by country, the values of own- and cross-price elasticity will depend on the methods and assumptions applied in their derivation. For example, the elasticities derived in New Zealand were “conditional”, where it was assumed that there would be no change in the share of the household budget allocated for foods. While we did not assume there would be no change in food costs with the dietary interventions, we did set the magnitude of the interventions such that there would be less than 1% variation in total food expenditure for the average household. The largest impact of this was seen with the saturated fat tax, which led to a 0.8% increase in the cost of the diet (Table 2), which is equivalent to AU$0.15 per day (95% UI: AU$0.03 to AU$0.27).
While our modelling focused on the health implications of consumer responses to price changes, the imposition of taxes (or subsidies) also sends a signal to those in the food supply chain about a government’s commitment to tackling unhealthy diets [43]. We explored a number of possible supply-side responses, including variations in the pass-through of price changes and product reformulation. Reformulation to healthier products, so that they do not attract a tax (and therefore prices do not change for the consumer) is a good outcome from a health sector perspective. Our scenario reflects the maximum that might be achieved if manufacturers reformulate products that would be taxed (e.g., by reducing salt or replacing unhealthy fats and added sugars with healthier options). However, we do assume that the reformulation has no impact on consumer preferences (e.g., changes in taste or texture are not large enough to influence choices) and that food is not added to or manipulated in other ways that influence health (e.g., by reducing fibre). It is also possible that food manufacturers or retailers could use other strategies, such as marketing or price promotions to reduce or exaggerate the price effects, which we have not modelled.
Like our study, previous Australian analyses of a sugar-sweetened beverage tax [11] and a junk food tax [10] (similar to our combination of saturated fat, salt, sugar, and sugar-sweetened beverage taxes) indicated that the taxes could be cost-saving for the health sector. However, there were differences in the magnitude of the health gains. The sugar-sweetened beverage tax modelled by Veerman et al. [11] was approximately twice the size of our sugar-sweetened beverage tax. Taking this difference in the size of the tax into account, the effects on BMI were comparable in the two analyses; however, the DALY health gain was more than twice the size. Similarly, the health gain associated with the junk food tax modelled by Sacks et al. [10] was around 20% larger than the health gain from our comparable tax analyses (taking population size differences into account), although the effects on BMI were again comparable between the two analyses. There are a number of possible reasons for the differences between the studies. Both our current study and the previous studies used a proportional multi-state lifetable approach to model health outcomes, simulated outcomes over the lifetime of the Australian population, and presented results with a discount rate of 3%. However, there were important differences in the inclusion of cross-price elasticity effects and in the risk factor–disease relationships that were included in the health model. Veerman et al. measured cross-price elasticity effects on artificially sweetened beverages and modelled the health impacts of the changes in BMI that would stem from the beverage effects on energy intake. Sacks et al. measured cross-price elasticity effects on a broader range of foods, but only those that were considered to be unhealthy (e.g., biscuits, cakes, pastries, pies, snack foods, confectionery, and soft drinks), and, using an earlier version of the Veerman et al. model, they also modelled health impacts based on BMI changes stemming from energy intake effects. However, in this study, we measured cross-price elasticity effects across all foods and drinks in the diet (excluding infant foods and alcoholic beverages) and modelled the health impacts of changes in sodium intake and fruit and vegetable intake, in addition to changes in BMI stemming from energy intake effects. In our analyses of changes across the whole diet, the sugar-sweetened beverage tax led to a non-significant effect on sodium; however, it did lead to a reduction in fruit and vegetable intake, which countered the benefits from reduced BMI, lessening the health gain overall, in comparison to the previous studies.
The probability of harm associated with the fruit and vegetable subsidy (when implemented without accompanying unhealthy food taxes) in our modelling may at first appear counter-intuitive. Fruits and vegetables are generally high in fibre and have a low energy density, so if people substitute fruits and vegetables for relatively unhealthier foods, then measures of total fibre and energy intake will improve. However, using price subsidies or discounts as an incentive to purchase more fruits and vegetables may have the effect of increasing real income available to buy food, including unhealthy products, and could therefore lead to an overall increase in dietary measures such as saturated fat, sodium, or total energy intake. There is limited experimental evidence on the effects of price subsidies [44] to help us understand these potentially competing effects. In a review of experimental evidence on subsidies to promote healthy food purchases, An [45] identified only two studies (one in a university cafeteria and one in a high school cafeteria) that examined the effect of price discounts on fruits and/or vegetables. While both of these studies found an increase in the purchase of discounted fruit and vegetable items, neither study fully evaluated the effect on intake of undiscounted foods (e.g., unhealthy foods) to determine an overall impact on diet or energy intake.
The own- and cross-price elasticities reflect the change in the purchase of the products targeted by the subsidy (fruits and vegetables) and the average change in the purchase of all other foods in the diet. Our modelling using elasticities from New Zealand and the UK suggests that although fruit and vegetable purchases do increase in response to the price reduction, there is an accompanying increase in the purchase of other foods that are on average higher in energy and sodium. When we modelled the effects of these changes in energy, sodium, and fruit and vegetable intake on risks of diseases, and simulated the combined impact of all diseases on population health, we found that there was no net improvement in health (i.e., the benefits of increased fruit and vegetable intake were “cancelled out” by the harms from the increased energy and sodium intake). However, there are other factors that we did not include in this modelling, such as the health effects from changes in fibre, whole grains, and red and processed meats, which may alter the balance between benefit and harm.
There is still a great deal of uncertainty about the many possible causal pathways between diet and disease. The relative risks that quantify the dose–response relationships in our modelling are taken from meta-analyses of prospective cohort studies. While trial evidence may be preferred in establishing causality, trials also have drawbacks, notably non-adherence combined with misreporting. While the observational studies we have based our analyses on do adjust for potential confounding (e.g., age, sex, smoking), there is still potentially residual confounding from missing or poorly measured explanatory variables, although this may to some extent be offset by the countervailing regression dilution bias. Dietary measurement is far from perfect; random error (misclassification) in exposure measurements would theoretically lead to an underestimation of the true strength of an association.
The cross-price elasticities that so strongly influence how people spend their money when fruits and vegetables are subsidised will be influenced by a range of factors, such as habit, marketing, and cultural food preferences, and these factors will vary from one country to another. Interestingly, although we found differences in the types of foods and drinks that changed in response to a fruit and vegetable subsidy when using elasticities from two different countries (New Zealand and the UK), the overall effect on dietary parameters (e.g., saturated fat, energy, and sodium intake) was similar, and this led to similar cost-effectiveness outcomes. However, we really need more measurement studies of price elasticities in other countries that comprehensively and more uniformly address all categories of foods and drinks, before patterns can truly be discerned.
While the fruit and vegetable subsidy was cost-effective when added to the package of taxes in Australia, this may not be the case in other countries, where the typical diet and price elasticities may be different (e.g., in Asia) and where the costs of health care treatment may be different. In countries that might be expected to have similar per capita health benefits to those in Australia (e.g., in Western Europe and North America), the cost-effectiveness of adding the fruit and vegetable subsidy is likely to be more favourable in countries with higher health expenditure (and hence larger potential cost-savings from reducing disease) than in countries with lower health expenditure, and vice versa. The modelling methods used in this study could potentially be used in other countries to tailor a package of taxes and subsidies with the best probability of improving population health at the least cost to the government.
The package of taxes and subsidies we evaluated led to an average change in the price of food and drink products of 10%, which is only half the 20% that is often proposed as the minimum needed to have significant effects on obesity and cardiovascular disease [4]. Nevertheless, the outcome of 470,000 DALYs that we found could potentially be averted with the tax and subsidy package is greater than we have modelled for a wide range of other dietary interventions in Australia, including mandatory and voluntary regulation of salt in processed foods (breads, margarines, and cereals) [37], traffic light food labelling [10], community lifestyle programs [36], and individually tailored dietary counselling [38].
A citizens’ jury in Australia was in favour of using revenue from taxes to subsidise healthy food options [46]. It also gave strong support for taxation of sugar-sweetened beverages, but had reservations about applying taxes to processed meats and snack foods, concerned by the potential complexities of distinguishing between healthy and unhealthy options. Although there are concerns that food and drink taxes may unfairly impact poorer people (who spend proportionally more of their income on food and drink), economic modelling from Ireland [47] suggests that while a tax on unhealthy food may be regressive, when combined with a healthy food subsidy (as we have modelled here), the effect is poverty neutral.
While food taxes and subsidies are not currently on the political agenda in Australia, there is recurrent interest in broadening the goods and services tax, which would lead to price changes on a wide range of food and drink products [48], and this could be an excellent opportunity to introduce an accompanying package of taxes and subsidies to encourage healthier eating in Australia.
The food industry is unlikely to welcome policies that steer consumption away from profitable processed foods. Large companies or corporations can use their financial power to promote information that distracts and confuses the public, as well as influence research and lobby politicians [49]. Although such pressures did lead to a repeal of Denmark’s saturated fat tax, the continued implementation of taxes in other countries (e.g., Hungary, France, and Mexico) does indicate a growing political and public motivation to implement health-based food and drink taxes despite industry pressures. While food and beverage industries may seek to dismiss modelling studies such as this and rely only on the results of real-world population trials of tax or subsidy interventions, such studies are not currently available and take many years to complete. By synthesising the best available evidence, modelling studies can give an indication of whether it is worthwhile implementing and evaluating a change in policy around the pricing of foods and beverages, and can provide guidance on the design of the new policy and accompanying monitoring and evaluation strategies.
With large multi-national corporations playing a powerful advisory role in the negotiation of international trade and investment agreements, governments may be less willing to implement public health policies and programs that compel action from food manufacturers [50]. Given the huge global burden of dietary-related disease, and the rapid escalation that is occurring in low- and middle-income countries [1], it could be helpful if there was a legally binding global convention around diet, similar to the World Health Organization’s Framework Convention on Tobacco Control, to support and protect government rights to implement taxes and regulatory measures that will improve public health [50].
This study adds to the growing evidence of large health benefits and cost-effectiveness of using taxes and regulatory measures to influence the consumption of healthy foods [51]. We believe that with such large potential health benefits for the Australian population, and large benefits in reducing health sector spending on the treatment of non-communicable diseases, the formulation of a tax and subsidy package should be given a more prominent role in Australia’s public health nutrition strategy.
|
10.1371/journal.pntd.0002101 | Microbiological, Histological, Immunological, and Toxin Response to Antibiotic Treatment in the Mouse Model of Mycobacterium ulcerans Disease | Mycobacterium ulcerans infection causes a neglected tropical disease known as Buruli ulcer that is now found in poor rural areas of West Africa in numbers that sometimes exceed those reported for another significant mycobacterial disease, leprosy, caused by M. leprae. Unique among mycobacterial diseases, M. ulcerans produces a plasmid-encoded toxin called mycolactone (ML), which is the principal virulence factor and destroys fat cells in subcutaneous tissue. Disease is typically first manifested by the appearance of a nodule that eventually ulcerates and the lesions may continue to spread over limbs or occasionally the trunk. The current standard treatment is 8 weeks of daily rifampin and injections of streptomycin (RS). The treatment kills bacilli and wounds gradually heal. Whether RS treatment actually stops mycolactone production before killing bacilli has been suggested by histopathological analyses of patient lesions. Using a mouse footpad model of M. ulcerans infection where the time of infection and development of lesions can be followed in a controlled manner before and after antibiotic treatment, we have evaluated the progress of infection by assessing bacterial numbers, mycolactone production, the immune response, and lesion histopathology at regular intervals after infection and after antibiotic therapy. We found that RS treatment rapidly reduced gross lesions, bacterial numbers, and ML production as assessed by cytotoxicity assays and mass spectrometric analysis. Histopathological analysis revealed that RS treatment maintained the association of the bacilli with (or within) host cells where they were destroyed whereas lack of treatment resulted in extracellular infection, destruction of host cells, and ultimately lesion ulceration. We propose that RS treatment promotes healing in the host by blocking mycolactone production, which favors the survival of host cells, and by killing M. ulcerans bacilli.
| Mycobacterium ulcerans infection causes Buruli ulcer (BU), a disfiguring skin disease now found principally in poor rural areas of West Africa. M. ulcerans produces a toxin called mycolactone (ML), which destroys fat cells in skin tissue. BU typically first shows as a nodule that eventually ulcerates. The lesions may continue to spread over limbs or occasionally the trunk. The current standard treatment is 8 weeks of daily rifampin and injections of streptomycin (RS). The treatment kills the bacilli and wounds gradually heal. We tried to determine if RS treatment actually stops mycolactone production before killing bacilli. Using a mouse footpad model of M. ulcerans infection where the time of infection and lesion development can be followed in a controlled manner before and after antibiotic treatment, we found that RS treatment rapidly reduced footpad swelling, M. ulcerans numbers, and ML production. Microscopic analysis of footpads revealed that RS treatment resulted in bacilli being destroyed by host cells whereas lack of treatment resulted in extracellular infection, destruction of host cells, and lesion ulceration. We propose that RS treatment promotes healing in the host by blocking mycolactone production, which favors the survival of host cells, and by killing M. ulcerans.
| Mycobacterium ulcerans infection is the cause of the neglected tropical disease, Buruli ulcer, found in poor rural areas of West Africa as well as in beach resorts in Australia [1] principally, although transmission has occurred in every continent except Europe. It is the third most important mycobacterial infection of humans worldwide after tuberculosis and leprosy, although in Benin and Ghana, for example, it is now the second most common mycobacterial disease [2], [3]. Infection begins after exposure in slow-moving fresh water to plants or biting insects or other unknown mechanisms and slowly leads to swelling, manifested as a nodule, plaque, or edema in humans, and in experimental animals [4], [5], [6], [7], [8]. Untreated lesions may progress and involve an entire limb or the trunk. Part of the pathogenic process is the production of the immunosuppressive mycolactone (ML) toxin by M. ulcerans that promotes the development of necrotic ulcers and possibly the initial swelling. How soon ML production begins after infection is unknown [9] as is how soon it stops due to antibiotic treatment, currently rifampin and streptomycin [1], although studies of human lesions have suggested production may be affected soon after treatment [10].
Evidence from the mouse model indicates that BALB/c mice develop swelling accompanied by increasing bacterial burden as measured by CFU followed by a plateau in CFU but progression of swelling [4], [11], [12]. After antibiotic treatment, both swelling and CFU decline [13].
We evaluated the evolution of M. ulcerans (Mu1615, Malaysian strain) infection in the mouse footpad model by assessing not only lesion appearance and CFU number, but also ML production, systemic immune responses, and histopathology at different times after infection and after the onset of antibiotic therapy. We find that the current standard antibiotic therapy started after the appearance of swelling not only reduces bacterial load but also preserves an effective host cell infiltrate leading to loss of acid-fast stain integrity of the bacilli. The host immune response was tested in splenocytes stimulated with mycobacterial antigens and assessed by measuring cytokine and chemokine production. Additionally, there is an early and continuing restraint on ML production that may help the host control and reduce bacillary numbers.
M. ulcerans 1615 (Mu1615), an isolate originally obtained from a patient in Malaysia in the 1960s, [14] was kindly provided by Dr. Pamela Small, University of Tennessee. Previous studies have confirmed that this strain produces mycolactone and kills macrophages and fibroblasts [15], [16]. The strain was passaged in mouse footpads before use in these studies. The bacilli were harvested from swollen footpads at the grade 2 level, i.e., swelling with inflammation of the footpad [4].
BALB/c mice, age 4–6 weeks (Charles River, Wilmington, MA), were inoculated in the right hind footpad with approximately 5.5 log10 (3×105) CFUs of Mu1615 in 0.03 ml PBS. Footpads were harvested from 9 mice (3 for CFU count, 3 for ML detection, 3 for histopathology) at different time points after infection (Table 1) up to ≥grade 3 swelling. After the onset of swelling (∼day 21–23), treatment with rifampin (R, Sigma, St. Louis, MO, 10 mg/kg by gavage) and streptomycin (S, Sigma, 150 mg/kg by subcutaneous injection) began on day 24, and was administered 5 days/week for 8 weeks. Groups of the treated mice were also sacrificed for these analyses. For CFU counts footpad tissue was harvested, minced with fine scissors [17], suspended in 2.0 ml PBS, serially diluted, and plated on Middlebrook selective 7H11 plates (Becton- Dickinson, Sparks, MD). Alternatively, footpads were harvested and placed in 10% buffered formalin for histopathological analysis by hematoxylin and eosin (H&E) or acid-fast (AF) staining. Footpads from a third group of mice at each time point were placed in freezer vials and frozen at −80°C. These latter footpads were shipped on dry ice to St. George's University of London for mycolactone detection experiments. Mice were evaluated for footpad swelling weekly using an established scoring system [4] with grade 1 showing footpad swelling, grade 2 swelling with inflammation, and grade 3 swelling and inflammation of the entire foot [18]. Spleens were harvested to assess responses to mycobacterial culture filtrate proteins (CFP) and to the concanavalin A (ConA) mitogen (Sigma) by multiplex ELISA (Bio-Plex Pro Mouse Cytokine 23-plex Assay, Biorad, Hercules, CA) as described [15]. The 23 analytes included Th1 cytokines: IL-2, IFN-γ, IL-12p40, IL-12p70; Th2 cytokines: IL-4, IL-5, IL-10, and IL-13; proinflammatory cytokines: TNF-α, IL-1α, IL-1β, and IL-6; IL-9; IL-17; colony stimulating factors: GMCSF, GCSF, IL-3; and chemokines: CXCL-1, CCL-2, CCL-3, CCL-4, CCL-5, and CCL-11. All animal procedures were conducted according to relevant national and international guidelines. The study was conducted adhering to the Johns Hopkins University guidelines for animal husbandry and was approved by the Johns Hopkins Animal Care and Use Committee, protocol MO08M240. The Johns Hopkins program is in compliance with the Animal Welfare Act regulations and Public Health Service (PHS) Policy and also maintains accreditation of its program by the private Association for the Assessment and Accreditation of Laboratory Animal Care (AAALAC) International.
GraphPad Prism 4 was used to assess significant differences by student T and analysis of variance.
Three days after infection, mice harbored 3.29±0.33 log10 CFU in the infected footpads. Just before the onset of swelling, at day 20, the footpads contained 5.05±0.19 log10 CFU of M. ulcerans. Footpad swelling first appeared in some mice 3 weeks after infection and averaged grade 1 soon after the initiation of treatment at ∼3.5 weeks after infection. At day 27, there were approximately 6.02±0.13 log10 CFU per footpad. Swelling continued to increase over the following 3 weeks (Figure 1) for an average of grade 3.42±0.43 in untreated mice whereas CFU counts remained essentially unchanged. In contrast, both swelling and CFU counts declined markedly in the RS-treated mice. Swelling after 3 weeks of treatment averaged grade 0.14±0.18 and log10 CFU count per footpad was reduced to 0.52±0.45. By day 55 after infection, untreated mice had 6.37±0.32 log10 CFU whereas RS-treated mice, at this time point after 31 days of treatment, had only 0.20±0.35 log10 CFU (Figure 2a). By day 63 (i.e., 39 days of treatment), the RS-treated mice were culture negative. Swelling was essentially undetectable. Figure 2b shows the contrast in M. ulcerans CFU counts between treated and untreated mice.
Histologically, increased numbers of AFB correlated with increased footpad swelling in untreated mice; bacilli were evident initially in the dermis and eventually in sub-epidermal zones and the epidermis with the onset of ulceration (Figure 3A.1–5). Inflammatory cell infiltrates appeared to be maintained in RS-treated mice but were disrupted and progressively disorganized in untreated, control mice (Figure 3B.1–4). After 2–3 weeks of treatment, AFB were still detectable but acquired a beaded appearance and this morphology persisted even after treatment completion. The beaded appearance of the bacilli correlated with reduced numbers of cultivable bacilli in treated mice (Figure 2) whereas the bacilli in untreated mice appeared solidly stained. Three months after treatment completion, bacilli were detectable in tissue by acid-fast staining but were not cultivable (Figure 3B.5).
ConA and CFP of M. tuberculosis (Mtb) were used as a potent T-cell mitogen and a non-specific mycobacterial antigen, respectively, to both assess the functional capability of T-cells and the evolution of cellular immune responses by splenocytes to mycobacterial antigens during the course of M. ulcerans infection. There is a biphasic pattern to the kinetics of secretion of representative Th-1 (IFN-γ), Th-2 (IL-5) and Th-17 (IL-17) cytokines upon stimulation with the potent T-cell mitogen ConA (Figure 4, left column). There is an initial rise detected at week 2, then a decline during weeks 4 to 6 and then a later surge from week 7 onwards. The cellular responses to a culture filtrate protein of Mtb however is generally a slowly evolving one and gradually builds up (IL-2, IL-5, IFN-γ and TNF-α) over the course of infection with M. ulcerans (Figure 4, right column). Immune responses were, however, generally lower during the course of infection in mice treated with antibiotics compared with untreated mice (Figure 4). Although the amount of TNF-α detected in supernatants of splenocytes from mice treated with RS was much less after 10 days of treatment, it was statistically significant (p<0.001) only at day 55 by two-way ANOVA. A t-test analysis indicated borderline statistical significance (p<0.05) at day 48.
Mycolactone toxin production was assessed by two different methods: cytotoxicity against the HELF cell line and mass spectrometric analysis.
Our results show that RS treatment of experimental M. ulcerans disease results not only in the reduction or elimination of viable bacteria but also in clinical and histopathological improvement of lesions, supporting results obtained in the analysis of human specimens [22], [23] and in RS-treated mice [24], [25], [26]. The inflammatory response is also down-regulated with reduced production of cytokines such as TNF-α during treatment. By mass spectrometric analysis, mycolactone was detectable before the observation of swelling and increased markedly and continuously after swelling onset whereas M. ulcerans CFU counts peaked and remained on a plateau. RS treatment resulted in a parallel loss of cultivable M. ulcerans in the footpad lesions, a decline in footpad swelling and an abrupt decline in the production of mycolactone within days of treatment initiation.
The initial clinical response to infection with M. ulcerans is typically not strikingly different from the course of infection with M. marinum [27], [28], [29], [30]; a nodule develops in both cases but, with M. ulcerans, there is an absence of tenderness and erythema. The absence of pain and an inflammatory response is attributed to the production of mycolactone inhibiting the immune response [31], damaging nerves [32], [33] and/or destroying infiltrating inflammatory cells [34], [35]. At which point in the course of infection sufficient mycolactone is present to mediate these effects cannot be readily ascertained in humans but can be approached in the mouse model. The mouse model is also suitable to monitor the impact on mycolactone production of antibiotic treatment, which is thought to block mycolactone production and to reverse local immunosuppression leading to the restoration of an active inflammatory process [36].
In the experiment reported here, mycolactone was detectable 2 to 3 weeks after inoculating approximately 4.6 log10 CFU of M. ulcerans into mouse footpads and before detectable swelling of the footpads. The numbers of bacilli at the beginning of a human infection are unknown but presumably considerably less. A sensitive assay that would be appropriate for use in endemic settings could conceivably detect infection before the nodular phase or confirm the diagnosis of M. ulcerans infection in nodular or plaque lesions before the onset of ulceration. Treatment with the current gold standard of rifampin and streptomycin appears to result in inhibition of mycolactone production. Whether the inhibition blocks the pathway of enzymes encoded by the M. ulcerans giant plasmid or has an impact on the plasmid itself has not been assessed here but it could be investigated using samples where there are still cultivable bacilli but negligible mycolactone production. Mechanistically, streptomycin and rifampicin inhibit protein synthesis by different mechanisms but they both result in a decrease in bacterial viability and, directly or indirectly, in a decrease of mycolactone production.
By sampling mouse footpads on a weekly basis we have been able to confirm the early phagocytic, intracellular phase of M. ulcerans infection followed by a later extracellular phase [26]. One week after infection (Fig. 3A2), there was moderate infiltration of host cells and bacilli were predominantly intracellular, an observation made by Ruf et al. [25]. The extracellular phase of bacillary growth appeared to coincide with the onset of footpad swelling, beginning about 3 weeks after infection in this experiment. Over the next two weeks, AFBs were increasingly found outside of cells. The shift to an extracellular growth was blocked by treatment with rifampin and streptomycin, consistent with the notion that mycolactone production and phagocyte destruction are inhibited by antibiotic treatment [22], [23], [24]. In the absence of treatment, the location of bacterial clusters also changed from relatively deep in the dermis to sub-epidermal and epidermal zones to the skin surface with the onset of footpad ulceration. In contrast, in mice treated with the antibiotics, M. ulcerans was found in a few clusters associated with phagocytic cells within the dermis. The appearance of the bacteria became beaded with less intense acid-fast staining [22], [23], [24], [25], comparable to the changes observed in the morphological index of leprosy bacilli after the onset of treatment. At 3 months after the completion of 8 weeks of daily treatment, pale-staining bacilli were still detectable and associated with host cells in the dermis but remained uncultivable, suggesting that residual antigenic components may have helped maintain an inflammatory infiltrate. Overall, from the literature [25], [34], [37], we find that early after infection, there is a dynamic picture of infiltrating host cells and bacillary localization within phagocytic cells. When the same time points were examined in mouse footpads on day 7 (this study and [25]) and day 13 (this study and [34]), the histological impressions were strikingly similar.
In this experiment, we observed increases in M. ulcerans CFU counts starting at day 3 (3.29±0.33 log10) and continuing to day 27 (6.02±0.13 log10) after infection. The CFU counts remained at approximately 6 log10 for the duration of the experiment through day 55, after which time the footpads started to ulcerate and could not be readily decontaminated for further analysis, given the prolonged culture time for this organism (8–12 weeks). Swelling was not apparent until day 20 in some mice, but by days 24 and 27, it averaged grade 1. Swelling continued to increase during this time, while CFU counts were stable, and peaked only at day 55 at an average of swelling grade 3.75. The fact that mycolactone concentration continued to rise after the plateau in cultivable organisms suggests that there was equilibrium between living and dying M. ulcerans cells with mycolactone being released by both populations. Alternatively, or in addition, mycolactone may be induced by a quorum sensing mechanism as bacterial numbers reach the plateau stage. Concurrently, or slightly before the manifestation of swelling, mycolactone could be detected by mass spectrometric analysis at day 13 (1.6 ng/ml of homogenized footpad extract) and 4.18 ng/ml on day 20. In untreated mice the concentration of mycolactone in infected footpads continued to rise throughout the experiment. In mice treated with antibiotics from day 24, mycolactone concentration tailed off after day 35 but it did not become undetectable by MS until day 63. This suggests either that viable organisms persisted for more than 5 weeks after antibiotics were started or that mycolactone could still be detected some time after M. ulcerans had been killed. The latter is supported by the observations that no viable M. ulcerans were detected in mouse footpads after antibiotic treatment, similar to that used here, followed by a course of corticosteroid therapy [24], [38] but we did not use the very recently described 16S rRNA assay to determine if viable organisms were still present [39].
Using the cytotoxicity assay, biological activity associated with mycolactone was detectable as early as day 3 and persisted even at day 63 despite antibiotic treatment. Although it is possible that this assay is more sensitive than MS, experiments in vitro using synthetic mycolactone as a standard show that MS can detect mycolactone at a concentration of less than 10 pg/ml so it is more likely that some cytotoxicity was caused by other lipid molecules generated during the course of M. ulcerans infection. Nevertheless this evidence that mycolactone may have persisted in mouse tissue well after antibiotics had started to reduce the burden of infection when footpad swelling was diminishing is important in the context of human infection in which slow healing has been observed in some cases without an obvious reason.
Mycolactone has profound effects on the production of cytokines and chemokines in vitro even at sub-cytotoxic concentration and this may well influence the rate of wound healing. In the present experiments for example, TNF-α production by splenocytes increased in response to mycobacterial antigens during M. ulcerans infection but this reverted to normal during antibiotic treatment. A similar but delayed and relatively attenuated response was observed for other cytokines such as IFN-γ, IL-5, and IL-17. ConA, a potent mitogen, assessed the capacity of splenocytes to produce cytokines upon stimulation, while the Mtb CFP elicited responses to mycobacterial antigens but is not specific for M. ulcerans. The initial rise in the secretion of cytokines coincides with the infiltration of host inflammatory cells and the intracellular stage of M. ulcerans when mycolactone concentrations were low. Subsequently mycolactone concentrations start to build up at the site of infection, M. ulcerans assumes an extracellular localization and prevents antigen-presenting cells from processing mycobacterial antigens for a systemic immune response. The fact that M. ulcerans assumes an extracellular localization and that mycolactone concentrations begin to increase locally could prevent antigen presenting cells from processing M. ulcerans for systemic immune responses to develop [40]. Again Hong et al. [41] have shown that mycolactone A/B could be detected in PBMC's of mice infected with M. ulcerans and indeed recently ML has been detected in sera of some human's infected with M. ulcerans [42]. The later recovery of systemic cytokine secretion in the light of increasing concentrations of mycolactone locally in the footpads is difficult to explain. In the proximal phases of M. ulcerans infection, this initial exposure may be inhibited by increasing bacillary load and the production of mycolactone. Subsequently, the splenocytes could encounter and process mycobacterial antigens released into the circulation or conveyed from the site of infection in the later stages of M. ulcerans infection from degradation of defunct M. ulcerans in mice treated with bactericidal antibiotics (Figure 4, both columns).
Some patients are able to control M. ulcerans infections at the early nodular stage without antibiotics and these data may provide some insight into why this might be the case. It would seem that those with robust immune responses, a lower burden of M. ulcerans and thus mycolactone secretion initially may be able to contain the M. ulcerans infection. Further studies are, however, required to elucidate the correlations between local and systemic mycolactone kinetics in tandem with immune responses in these two compartments in this model.
The finding that cytotoxicity was increased in contra-lateral footpads well away from the site of infection remains difficult to explain. Mycolactone molecules were not detected by MS but we cannot rule out the possibility that cytotoxic breakdown products of mycolactone, or other molecules, circulated to these footpads. Other studies have shown that mycolactone itself was detectable in peripheral blood white cells during mouse infection with M. ulcerans [41].
Further studies may focus on the determination of cytokines in the milieu of the footpad lesion in mice before and after treatment with the current regimen or a new all oral regimen, such as one replacing streptomycin with clarithromycin, as well as the impact of vaccination on mycolactone production and local immunity after footpad challenge. Refinements in calculating the amount of mycolactone obtained after extraction may be made by spiking different concentrations of synthetic mycolactone into uninfected footpads and determining the linearity of the yield.
|
10.1371/journal.pntd.0004790 | Performance Testing of PCR Assay in Blood Samples for the Diagnosis of Toxoplasmic Encephalitis in AIDS Patients from the French Departments of America and Genetic Diversity of Toxoplasma gondii: A Prospective and Multicentric Study | Toxoplasmic encephalitis in patients with AIDS is a life-threatening disease mostly due to reactivation of Toxoplasma gondii cysts in the brain. The main objective of this study was to evaluate the performance of real-time PCR assay in peripheral blood samples for the diagnosis of toxoplasmic encephalitis in AIDS patients in the French West Indies and Guiana.
Adult patients with HIV and suspicion of toxoplasmic encephalitis with start of specific antitoxoplasmic therapy were included in this study during 40 months. The real-time PCR assay targeting the 529 bp repeat region of T. gondii was performed in two different centers for all blood samples. A Neighbor-Joining tree was reconstructed from microsatellite data to examine the relationships between strains from human cases of toxoplasmosis in South America and the Caribbean. A total of 44 cases were validated by a committee of experts, including 36 cases with toxoplasmic encephalitis. The specificity of the PCR assay in blood samples was 100% but the sensitivity was only 25% with moderate agreement between the two centers. Altered level of consciousness and being born in the French West Indies and Guiana were the only two variables that were associated with significantly decreased risk of false negative results with the PCR assay.
Our results showed that PCR sensitivity in blood samples increased with severity of toxoplasmic encephalitis in AIDS patients. Geographic origin of patients was likely to influence PCR sensitivity but there was little evidence that it was caused by differences in T. gondii strains.
ClinicalTrials.gov NCT00803621
| Diagnosis of toxoplasmic encephalitis (TE) in patients with AIDS is not straightforward because clinicians rely initially on an empiric diagnosis based on clinical and radiographic improvement to specific anti-Toxoplasma gondii therapy. There is therefore a need for biological tools to improve the diagnosis of TE, especially in tropical areas where this diagnosis is likely to be underestimated. The use of PCR testing in blood samples for the diagnosis of TE has been limited by its poor sensitivity in the studies conducted in Europe. In tropical South America, the results of PCR sensitivity in blood samples were controversial. Considering that T. gondii strains from tropical South America have substantial genetic and pathogenic differences with those from USA and Europe, it is therefore important to re-evaluate the performance of the PCR assay in blood samples for the diagnosis of TE in AIDS patients from this region. Our results showed that the only interest of PCR would be in the most severe forms of TE with altered consciousness because PCR is more likely to be positive. We also provided important genotyping data on T. gondii strains isolated in human cases of toxoplasmosis in the Caribbean and in South America.
| The protozoan Toxoplasma gondii is a cosmopolitan parasite that virtually infects all warm-blooded animals, including humans who become infected postnatally by ingesting tissue cysts from undercooked meat, consuming food contaminated with oocysts, or by accidentally ingesting oocysts from the environment [1]. The genetic diversity of this parasite is limited to a few successful clonal lineages in North America, Europe, Africa and China but is considerably higher in tropical South America [2]. In non-immunocompromised persons, toxoplasmosis is usually asymptomatic or limited to a mild symptomatology except in tropical South America. The prevalence of acquired and congenital ocular toxoplasmosis is much higher in Brazil and Colombia than in another place in the world and the Amazonian toxoplasmosis is a disseminated infection that requires management in intensive care units even in otherwise healthy adults [3, 4]. There is more and more evidence that the greater severity of toxoplasmosis in South America results from poor host adaptation to the genetically diverse T. gondii strains from this region [5].
Toxoplasmic encephalitis (TE) in patients with AIDS is a life-threatening disease mostly due to reactivation of Toxoplasma gondii cysts in the brain. TE can be inaugural of AIDS in patients who are not aware of their HIV seropositivity, but poor compliance with cotrimoxazole prophylaxis in patients with CD4 cell counts <200/μL is the major event leading to TE [6]. The incidence of TE in AIDS patients has greatly decreased since the introduction of HAART, but HIV-associated toxoplasmosis hospitalizations remain substantial, even in the United States [7, 8]. TE must be treated with specific anti-toxoplasmic therapy as soon as the diagnosis of TE is clinically and radiologically suspected [9]. Cerebral biopsy showing T. gondii tachyzoites is the only way to make a definite diagnosis of TE but is rarely undertaken in AIDS patients at baseline. The clinical and radiological response to specific therapy is still the gold standard for confirming a posteriori the diagnosis of TE in AIDS patients. This presumptive diagnosis has important limitations since up to 40% of AIDS patients with suspected TE and treated with specific therapy could not have in fact TE [10].
Laboratory investigations are considered not helpful in the diagnosis of TE. The majority of patients have positive IgG and negative IgM against T. gondii, simply indicating that they acquired toxoplasmosis in the past, mostly during childhood. However, the negative predictive value of a negative serologic testing for toxoplasmosis is high because it is estimated that < 3% of patients with AIDS have no demonstrable antibodies to T. gondii at the time of diagnosis of TE [11]. The diagnostic performance of PCR tests in various biological samples, mostly CSF and blood, was regularly assessed since the early years of the AIDS pandemic [12]. Blood samples are the only ones that can be easily obtained from the patients without invasive procedures. Although specificity was high, the use of PCR testing in blood samples for the diagnosis of TE has been limited by its poor sensitivity in the studies conducted in Europe [12–16].A few studies have been conducted in tropical South America and the results of sensitivity were highly controversial [17–20]. Considering that T. gondii strains from tropical South America have substantial genetic and pathogenic differences with those from USA and Europe, it is therefore important to re-evaluate the performance of the PCR assay in blood samples for the diagnosis of TE in AIDS patients from this region.
The main objective of this study was to evaluate the performance of real-time PCR assay in peripheral blood samples for the diagnosis of TE in AIDS patients from the French departments of America. The French departments of America (DFA) are French tropical overseas departments that include French Guiana in mainland South America and the French West Indies islands of Martinique and Guadeloupe in the Caribbean. The secondary objective was to collect and genotype T. gondii isolates from these patients.
This study was approved by the French Ethics Committee “Comité de Protection des Personnes du Sud-Ouest et Outre-Mer 4” on April 4, 2008, with the reference number cpp08-006a. The Toxo-DFA study is registered in the ClinicalTrials.gov database (Identifier: NCT00803621). All patients included in the study were adult and provided written informed consent.
The Toxo-DFA study is an epidemiological, prospective, and multicentric study to validate the real-time PCR assay in peripheral blood samples for the diagnosis of TE in AIDS patients from a tropical area. It was conducted between September 16, 2008, and December 30, 2011, in four hospital centers of the French departments of America: two in French Guiana (Cayenne and Saint-Laurent du Maroni) and two in the French West Indies (Fort de France in Martinique and Pointe à Pitre in Guadeloupe).
Study participants had to meet all of the following criteria: age >18 years, informed written consent, positive serologic test for HIV, and clinical and radiological suspicion of TE with start of specific antitoxoplasmic therapy. Patients legally protected or uncovered by social insurance, or with a specific antitoxoplasmic therapy already initiated since 72h or more were excluded from the study.
The collection of data was done in an online secured case report form with CS-ONLINE from CAPTURE SYSTEM software. Patients were assessed clinically at baseline, between days 6–8, 15–21, and 42–56. Neuroradiographic scans by computed tomography (CT) or magnetic resonance imaging (MRI) were performed at baseline, between days 15–21, and 42–56.
The gold standard for diagnosing TE was based on the clinical and radiological responses to specific antitoxoplasmic therapy after suspicion of cerebral toxoplasmosis. A validation committee of independent experts reviewed and classified the cases in 4 categories. TE was considered definite when there was a complete or significant clinical and radiological response to specific therapy (with not necessarily disappearance of radiological and clinical lesions), and no elements for an alternative diagnosis. TE was considered probable when radiological lesions were compatible but only partial improvement was observed with specific therapy (due to non-optimal treatment or incomplete follow-up), and no elements for an alternative diagnosis. Absence of TE was considered definite when there was no improvement or worsening of lesions with specific therapy or absence of T. gondii in cerebral biopsy samples and presence of elements for an alternative diagnosis. Absence of TE was considered probable when there was no response to specific therapy and no elements for an alternative diagnosis.
Two different centers were involved in the laboratory investigations: the Limoges center in metropolitan France and the Cayenne center in French Guiana.
A Neighbor-Joining tree was reconstructed from microsatellite data to examine the relationships between strains collected from human cases of toxoplasmosis in South America and the Caribbean. The tree was constructed with Populations 1.2.32 (http://bioinformatics.org/populations/) based on Cavalli-Sforza and Edwards chord-distance estimator [26] and generated with MEGA 6.05 (http://www.megasoftware.net/history.php) software.
The software used for statistical analyses was SAS 9.3 and the significant threshold of the p value was 0.05 (SAS Institute, Cary, USA). Median and interquartile intervals were given for quantitative variables while qualitative variables were presented as sample size and percentages. Nonparametric tests were used for comparing variables between patients with TE and those without TE: a Fisher exact test was performed for the qualitative variables and a Mann-Whitney test was used for the quantitative variables. To evaluate the diagnostic performance of the PCR assay in detecting TE in peripheral blood samples from AIDS patients in the French West Indies and Guiana, the reliability and the validity of the test were assessed. For these analyses, cases reviewed by the validation committee with definite or probable TE were coded positive and those with definite or probable absence of TE were coded negative. The reliability of the PCR assay was estimated by the Cohen’s Kappa coefficient of agreement and its 95% confidence interval between results of the test in Cayenne and Limoges centers. The validity of the PCR assay was estimated by the sensitivity and the specificity of the test in detecting definite and probable cases of TE that had been identified by the validation committee. The 95% confidence intervals of sensitivity and specificity were estimated by the exact method. A study of the false negative results was carried out to search explanatory factors according to the sample size of this sub-group by using a logistic regression model with the false negative status as the response variable and the potential associated factors as the explicative variables.
A total of 46 patients were included in this study: 17 in French Guiana (8 in Cayenne and 9 in Saint-Laurent du Maroni) and 29 in the French West Indies (23 in Guadeloupe and 6 in Martinique).
The validation committee reviewed the cases as follows: 2 cases were not assessable because of insufficient data (both included in the center of Saint Laurent du Maroni in French Guiana), 36 cases were classified in the TE group and 8 in the non-TE group. In the TE group, 30 were classified as definite TE and 6 as probable TE. In the non-TE group, 6 were classified as definite absence of TE and 2 as probable absence of TE. The demographic, laboratory and clinical baseline characteristics of the 44 patients classified in TE and non-TE groups by the validation committee are available in Table 2. There was no statistical difference between patients with TE and those without TE with respect to the variables listed in Table 2 except for the place of birth and the results of T. gondii serology.
Two different classifications were used for clustering the 44 patients into two groups according to their place of birth. In the first classification, the first group gathered the 29 patients who were born in the French West Indies and Guiana (6 in French Guiana, 17 in Guadeloupe, and 6 in Martinique) while the 15 patients who were born elsewhere were put together in a second group (7 in Haiti, 3 in Brazil, 2 in Suriname, 1 in Dominica, 1 in Dominican Republic, and 1 in Spain). Being born in the French West Indies and Guiana was significantly more common in patients without TE than in those with TE (p = 0.04) because all patients without TE (n = 8) were born in the French West Indies and Guiana (Table 2). The second classification was based on geography with 32 patients born in the Caribbean (17 in Guadeloupe, 7 in Haiti, 6 in Martinique, 1 in Dominica, and 1 in Dominican Republic), 11 in South America (6 in French Guiana, 3 in Brazil, 2 in Suriname), and 1 in Europe. According to this second classification, being born in the Caribbean was not statistically different between patients with TE and those without TE (Table 2).
Of the 41 patients with available data on T. gondii serology, only 2 had negative test results for IgG and IgM against T. gondii, and both of them had definite absence of TE (Table 2). The blood samples of the remaining 39 patients tested positive for IgG and negative for IgM, indicating past immunization against T. gondii. At presentation, only 7 patients were receiving systemic antiprotozoal prophylaxis: trimethoprim-sulfamethoxazole (cotrimoxazole, n = 5), pyrimethamine (n = 1), and atovaquone (n = 1). Of the 5 patients with cotrimoxazole prophylaxis, good compliance was reported in only one patient.
The choice of specific empirical antitoxoplasmic first-line therapy was based on routine practice of each center: all patients from French Guiana (n = 15), five out of six patients from Martinique and only one patient from Guadeloupe were treated with trimethoprim-sulfamethoxazole (cotrimoxazole) whereas 22 out of 23 patients from Guadeloupe were given a pyrimethamine-based combination with either sulphadiazine (n = 16) or clindamycin (n = 6). One patient was treated with a combination of pyrimethamine plus atovaquone in Martinique. Five (11.4%) patients, all with TE, died within the first 12 weeks after antitoxoplasmic therapy was begun.
Of the 44 patients, the PCR assay tested positive in blood samples of 9 patients with TE and tested negative in 35 patients (27 in the TE group and 8 in the non-TE group). The sensitivity was 25.00% (95% CI 12.12–42.20) which is low, and the specificity was 100% (95%CI 63.06–100.00) which is maximal. Of the 9 blood samples with a positive PCR test, only 4 were detected simultaneously in both Limoges (Parasites/mL blood: 0.01–0.13, 0–0.42, 4.67–6.49, 2.30–8.80) and Cayenne (Parasites/mL blood: 1.46–1.83, 0.45–0.53, 15.38–22.25, 2.15–9.83, respectively) laboratories, whereas 4 were detected in the Cayenne laboratory alone (Parasites/mL blood: 0.02–0.24, 0.01–0.18, 0.37–0.98, 4.80–8.30) and 1 in the Limoges laboratory alone (Parasites/mL blood: DNA: 0–1.16). The sensitivity was 11.11% (95% CI 3.11–26.06), 13.89% (95% CI 4.67–29.50), and 22.22% (95% CI 10.12–39.15) when positive PCR tests were observed in both centers, in the Limoges center, and in the Cayenne Center, respectively. The Cohen's kappa coefficient used to estimate the agreement between results of PCR tests performed in both centers was 0.5528 (95% CI 0.2112–0.8945), which indicates moderate agreement [27]. From these data, we can conclude that the majority of PCR results were close to the limits of detection and the difference in detection of T. gondii DNA at the two centers were due to differences in the analytical sensitivity of the PCR assay at each site.
The relationship between the risk of having a false negative result with the PCR assay in the blood for the diagnosis of TE and different baseline variables is shown in Table 3. Altered level of consciousness and being born in the French West Indies and Guiana were the only two variables that were associated with significantly decreased risk of false negative results with the PCR assay according to multivariate logistic regression analysis.
Five blood samples that tested positive with the PCR assay in the Limoges laboratory were inoculated into mice and only one strain was isolated. This strain was isolated from a patient who was living in Guadeloupe but born in Haiti. The blood sample of this patient tested positive with the PCR assay in both laboratories of Limoges and Cayenne, and corresponded to the sample with the highest parasite load (Parasites/mL blood: 4.67–6.49 and 15.38–22.25, respectively). This strain was designated HTI01 in this study and cryopreserved at the Toxoplasma BRC with the denomination code TgH40001. The genotype of this strain with 15 microsatellite markers was compared with the genotypes of the three reference type I, II, and III strains and with those of 43 strains collected in human cases of toxoplasmosis from South America and the Caribbean region (Table 1). The neighbor-joining analysis clustered the HTI01 strain collected in the present study with the type II reference strain in the unrooted tree (Fig 1). The 43 strains collected from human cases of toxoplasmosis in the Caribbean and in South America by the French national reference center for toxoplasmosis were clustered as follows: i) strains collected in patients from the French West Indies were highly structured in only two groups: 2 strains clustered with the HTI01 strain in the type II group and 8 strains were grouped into a separate cluster called Caribbean group; ii) the anthropized strains from French Guiana were also highly structured into three groups: 2 strains were assembled in the type I group, 2 in the type III group, and 1 in the Caribbean group; iii) the 18 wild strains from the Amazonian forest of French Guiana were found on separate long branches and were highly divergent from all the other strains, as reported previously [24, 25]; iv) of the 10 Brazilian strains, six were structured in one separate group called Brazilian group and four strains were divergent and found on separate long branches like wild strains from French Guiana.
There is a need of treatments and diagnostic tools for TE adapted to AIDS patients in the specific context of tropical areas. For example, the standard therapy of TE is the combination of pyrimethamine and sulfadiazine [9, 28]. However this treatment has several limitations, such as its cost, the high frequency of adverse reactions in AIDS patients, the absence of intravenous formulation and its frequent unavailability in poor resource settings. For these reasons, cotrimoxazole is often preferred as a first line therapy of TE in AIDS patients in tropical areas because it is efficacious, cheap, better tolerated, with an intravenous formulation and, most of all, is widely available in developing countries [29, 30]. Diagnosis of TE is not straightforward because the majority of clinicians rely initially on an empiric diagnosis based on clinical and radiographic improvement to specific anti-T. gondii therapy in the absence of a likely alternative diagnosis [28]. In tropical areas, many patients are diagnosed with HIV only after developing opportunistic infections such as TE and the differential diagnosis of focal neurological disease in patients with AIDS can be complex in the context of poor-resource settings [29]. Under-diagnosis is likely to be the consequence of the difficulties with diagnosing TE in tropical areas [31]. In this study, we aimed at evaluating the diagnostic performance of TE with the real-time PCR assay in peripheral blood samples from AIDS patients in a tropical setting. A total of 44 patients, 36 with TE and 8 without TE, from the French West Indies and Guiana in the French departments of America were included in the present study. The standards of healthcare are close to those of mainland France but this region is a crossroads for poor Caribbean and South American people who emigrate there for socio-economic reasons [32]. In this region, the HIV epidemic is a major public health problem and TE is a leading cause of death among HIV-infected adults [33, 34].
All patients without TE tested negative with the PCR assay in blood samples and all patients with a positive PCR result had TE. The 100% specificity of the PCR assay in blood samples for the diagnosis of TE in patients with AIDS in our study confirms the very high specificity of this test reported in the literature (median 99%, IQR 93.1%–100%) [12–18, 20]. However, with a sensitivity of 25%, the capacity of the PCR assay to detect TE in blood samples from patients with AIDS is low in a tropical area like the French departments of America. In fact, the sensitivity of this test to diagnose TE in blood samples from patients with AIDS seems to vary with geography in the literature. According to 5 studies conducted in the 1990s [12–16], the sensitivity ranged from 13.3% to 29.5% in Europe (median 24.3%, IQR 17.9–25.6) which is similar to the result of our study although this latter was conducted in a tropical area. In contrast, 3 studies from tropical South America in the 2000s showed contradictory results with a sensitivity of 1.2% in north-east Brazil, 18.8% in Colombia, and 80% in south-east Brazil [17, 18, 20]. It is difficult to compare these studies from Europe and South America with our study in the French departments of America because the volumes of blood samples (1, 5, or 10 mL), DNA extraction protocols (buffy coat versus whole blood), DNA targets (REP529, B1, TgRE1, and rDNA repetitive gene), and even primers for the same DNA target (B1 or REP529) were different. The 6 oldest studies performed a conventional PCR [12–17] whereas real-time PCR was done in the present work and in the most recent studies [18, 20]. The Brazilian study with the highest PCR sensitivity (80%) used a volume of 10 mL of blood sample and a conventional PCR assay targeting the B1 gene with primers B22 and B23 [17].
The DNA extraction step is essential for detecting T. gondii in blood samples. A recent study in the animal model showed that it was preferable to use buffy coat rather than whole blood, but stressed the importance of the volume of blood sample to increase the sensitivity of PCR assay [35]. The volume of blood sample in most studies that evaluated the performance of the PCR assay in AIDS patients with TE was 5 or 10 mL [12–15, 17] except in 2 studies that used only 1 mL [18, 20]. Using a limited volume of blood sample might explain, in part, the poor sensitivity of the PCR assay reported in one Brazilian study [19]. The choice of the DNA target for the PCR assay is also important because studies that tested different targets showed different sensitivity results according to the target [14, 15]. Experts generally recommend the use of REP529 DNA target and real-time PCR for reaching the highest sensitivity but they stress the importance of the proficiency of the laboratory performing the diagnosis and the need for optimization of PCR conditions [36]. It is therefore better to use a well-optimized PCR assay targeting the B1 gene with a conventional PCR assay in a reference laboratory rather than a non-optimized PCR assay targeting REP529 with a real-time PCR assay in an inexperienced laboratory [37]. The use of different primers for the same target may also lead to different results of sensitivity, as suggested in one study [15]. In our study, the target was REP529 in Limoges and Cayenne laboratories but, because each center independently developed their own laboratory-optimized PCR assay for routine diagnosis of toxoplasmosis, primers were different and the sensitivity in each center was also different (13.89% and 22.22%, respectively). However, it is also true that identical primers can give variable results of sensitivity depending on the laboratory, which underlines, once again, the crucial importance of PCR optimization [36]. What makes consensus is the systematic use of uracyl DNA N-glycosylase (UDG) to avoid false-positive results caused by carry-over contaminations and an internal positive control (IPC) to avoid false-negative results caused by PCR inhibitors of PCR [38]. Such basic precautions were taken in the present study, in 3 studies from Europe, and in 1 study from Colombia [12, 14, 15, 20]. One study in Europe and one in Brazil reported the use of IPC but not UDG with sensitivities of 20% and 1.2%, respectively [13, 18]. The two studies that performed PCR without IPC and UDG reported a sensitivity of 25% in Europe and 80% in Brazil [16, 17].
Altogether, it seems that the methodological issues raised here cannot entirely explain the huge difference between sensitivities of the PCR assay in blood samples for diagnosing TE in AIDS patients from 4 tropical areas: 1.2% in patients from Recife, north-east Brazil [18], 18.8% in Colombia [20], 25% in the French West Indies and Guiana (this study), and 80% in São Paulo, south-east Brazil [17]. In the present study, the geographic origin of patients was likely to influence the sensitivity of the PCR assay because being born in the French West Indies and Guiana was a variable significantly associated with a decreased risk of false negative results according to multivariate logistic regression analysis. The first hypothesis to explain the link between geography and sensitivity of T. gondii DNA detection in blood samples is the strain hypothesis because the hotspot of T. gondii genetic diversity is in tropical South America and because the genotype of T. gondii strains is strongly linked to the presumed geographical origin of infection in immunocompromised patients [2, 39]. Most cases of TE result from local reactivation of brain cysts without parasitemia which explains the absence of detection of T. gondii in most blood samples. If some strains are more likely to disseminate in blood flow than others, this would have a strong effect on sensitivity of PCR in blood samples. For example, the sensitivity of PCR in blood samples is very low for diagnosing ocular toxoplasmosis in immunocompetent patients from France and T. gondii DNA is detectable only in ocular fluid samples [40, 41]. In contrast, T. gondii genotypes involved in ocular toxoplasmosis in south and south-east Brazil were not characterized from ocular fluid samples but from peripheral blood, and this prolonged parasitemia was confirmed by direct microscopic observation of tachyzoites in some blood samples [42–44]. Brazil is a big country with a complex T. gondii population structure and it is possible that such differences also exist at a regional scale in Brazil. If strains from south and south-east Brazil are more likely to disseminate in blood flow than those from the north-east, this could explain the regional variation of PCR sensitivity in blood samples for diagnosing TE in AIDS patients from Brazil [17, 18]. Another example of disseminating disease is the Amazonian toxoplasmosis whose diagnosis is always confirmed by a positive result of the PCR assay in blood samples despite the fact that the patients are not immunocompromised [4].
Little is known about the genetic background that characterizes disseminating strains but, based on what is known from wild strains of the Amazonian rainforest, the genotypes of these atypical strains are found on separate long branches in neighbor-joining trees and are highly divergent from the genotypes of all other strains, especially from the clonal type II and III strains that are common in Europe [24]. However, in the present study, we found little evidence that the effect of geographic origins on PCR sensitivity in blood samples for the diagnosis of TE in AIDS patients was caused by differences in T. gondii strains. Unfortunately, we isolated only one T. gondii strain in a patient who was not born in the French departments of America but in Haiti. The genotype of this strain was not atypical but rather related to type II which represents >95% of strains in Europe where the sensitivity of PCR assay is low in blood samples. The other patient who was not born in the French West Indies and Guiana and who had a positive PCR result was born in Spain and therefore also likely infected by a type II strain. If the strain hypothesis were true in our study, we would have expected positive PCR results in blood samples of the 3 patients from Brazil but none of them tested positive. In fact, the proportion of positive PCR results was higher in patients born in the French West Indies and Guiana (7/29, 24%) than in those born elsewhere (2/15, 13%).
We included in the analysis the genotyping data of 43 T. gondii strains collected by the French national reference center from patients infected in tropical South America and the Caribbean. Strains that infect humans in the French West Indies and anthropized areas of French Guiana were not found on separate long branches in the neighbor-joining tree like wild strains from the Amazonian rainforest or some strains from Brazil but were highly structured like in Europe. Type II and III strains that are common in Europe are also common in the French departments of America. The difference with Europe is the predominance of an endemic lineage called Caribbean group that comprises the Caribbean 1, 2 and 3 genotypes already described in domestic animals from the anthropized area of French Guiana and in immunocompromised patients from the French West Indies [24, 25, 39]. Although we did not isolate strains in patients born in the French departments of America in this study, it is likely that they were also infected by T. gondii strains belonging either to the types II and III lineages or to the Caribbean group but not to highly divergent strains that could have explained the better detection of T. gondii in blood samples for these patients. In the absence of a clear explanation by differences in T. gondii strains, the effect of geography on the sensitivity of PCR in blood samples of AIDS patients remains to be elucidated.
The main result of our study is that the sensitivity of PCR in blood samples increases with the severity of TE. The main severity factors of TE in AIDS patients are profound immunodepression and impaired consciousness. In a study conducted in AIDS patients with TE at admission in intensive care units, the factors independently associated with a poor outcome were a Glasgow coma scale ≤8 and a CD4 cell count <25/μL [45]. In our study, a CD4 cell count <25/μL was not associated with a decreased risk of false negative results with the PCR assay. However, altered level of consciousness was the second variable significantly associated with a decreased risk of false negative results according to multivariate logistic regression analysis. Of the 8 patients with altered level of consciousness and TE in our study, 5 (62.5%) tested positive in blood samples. All patients (n = 3) with a Glasgow coma scale ≤9 had a positive test result with the PCR assay in blood samples. The high PCR sensitivity of 80% in blood samples of the 64 patients from São Paulo, Brazil, could be explained by a high number of severe TE cases in this study but clinical data were not available [17].
In conclusion, the PCR assay in blood samples is not recommended for diagnosing TE in the tropical setting of the French departments of America areas because of a poor sensitivity. The only interest of PCR would be in the most severe forms of TE with altered consciousness because PCR is more likely to be positive. Even in these cases, it seems difficult to reach a good sensitivity with the PCR assay because the concentration of T. gondii DNA is very low. PCR protocols have to be perfectly optimized because positive PCR results rely on high Ct values that are at the limit of the detection of the method which jeopardizes a good agreement between diagnostic laboratories, as showed in our study. There is no argument that the PCR sensitivity could be influenced by the genetic background of T. gondii strains in this area even if the geographic origin of patients is likely to play a role for unclear reasons. We believe that our results can be expanded in any tropical setting with the exception of other parts of tropical South America, especially Brazil where T. gondii strain diversity is far more complex than in the French West Indies and the anthropized areas of French Guiana. Other studies are needed in Brazil to know whether genetic-based differences in the capacity of hematogenous dissemination of locally acquired T. gondii strains are likely to explain the considerable regional variations of the sensitivity of the PCR assay in blood samples of AIDS patients from this country.
|
10.1371/journal.ppat.1000992 | TOPO3α Influences Antigenic Variation by Monitoring Expression-Site-Associated VSG Switching in Trypanosoma brucei | Homologous recombination (HR) mediates one of the major mechanisms of trypanosome antigenic variation by placing a different variant surface glycoprotein (VSG) gene under the control of the active expression site (ES). It is believed that the majority of VSG switching events occur by duplicative gene conversion, but only a few DNA repair genes that are central to HR have been assigned a role in this process. Gene conversion events that are associated with crossover are rarely seen in VSG switching, similar to mitotic HR. In other organisms, TOPO3α (Top3 in yeasts), a type IA topoisomerase, is part of a complex that is involved in the suppression of crossovers. We therefore asked whether a related mechanism might suppress VSG recombination. Using a set of reliable recombination and switching assays that could score individual switching mechanisms, we discovered that TOPO3α function is conserved in Trypanosoma brucei and that TOPO3α plays a critical role in antigenic switching. Switching frequency increased 10–40-fold in the absence of TOPO3α and this hyper-switching phenotype required RAD51. Moreover, the preference of 70-bp repeats for VSG recombination was mitigated, while homology regions elsewhere in ES were highly favored, in the absence of TOPO3α. Our data suggest that TOPO3α may remove undesirable recombination intermediates constantly arising between active and silent ESs, thereby balancing ES integrity against VSG recombination.
| Trypanosoma brucei, the causative agent of African sleeping sickness, escapes the host immune response through a mechanism known as the antigenic variation. Each individual trypanosome expresses a single species of surface antigenic protein at any time yet possesses an infinite potential to express different surface antigens by transcriptional and recombinatorial switching. Periodic switching to a different antigen allows parasites to escape the antibody-mediated host immune response and causes chronic infection, eventually overwhelming the host's immune system and leading to death. DNA recombination factors are critical for the protection of chromosome integrity. One of the major antigen-switching mechanisms exploits particular recombination pathways to achieve its purpose. We have used a new switching assay to study a regulator of recombination and to demonstrate that antigenic variation is a complex mechanism balancing chromosome integrity and antigen diversity by suppressing and promoting particular recombination events. Recombination is used in evasion or virulence mechanisms by several pathogens. Exploring how Trypanosoma brucei manipulates the recombination machinery to gain advantage against their host will help us understand pathogenesis in various organisms and may reveal weaknesses that can be exploited to control infectivity and virulence.
| Trypanosoma brucei proliferates in the bloodstream of its mammalian host and periodically escapes the antibody-mediated immune response. A single species of variant surface glycoprotein (VSG) is expressed at a given time, from among >1,000 VSG genes and pseudogenes [1], [2], and ∼10 million VSG molecules homogenously coat the surface of a parasite. Switching the expressed VSG causes antigenic variation (reviewed in [3]–[5]).
VSG genes are found in 15 expression sites (ESs) — polycistronic transcription units that are transcribed by RNA Polymerase I [3], [6]–[8] — of the Lister 427 strain [9]. These VSGs are located 40–60 kb downstream of their ES promoters and are flanked by 70-bp and telomere repeat sequences. Several expression-site-associated genes (ESAGs) with mostly unknown functions, and ESAG and VSG pseudogenes, are located between the promoter and the 70-bp repeat region. Only one ES is transcriptionally active at any time and the rest are silent. Many VSGs are found upstream of telomere repeats in minichromosomes but most are thought to reside in ‘telomere-distal’ arrays. Minichromosomal and telomere-distal VSGs lack promoters, but small numbers of 70-bp repeats are present upstream of these VSGs.
By analyzing switched variants, two major pathways of antigenic switching have been identified in T. brucei: in situ ES transcription switching and recombination-mediated switching [4], [5], [10]. In situ switching occurs by silencing the active ES and activating a silent ES, without DNA rearrangement [11], [12]. Recombination-mediated switching occurs mainly by gene conversion (GC) and can involve just the VSG or larger regions of the ES. VSG GC can occur by recombination between the active VSG and a silent ES-associated VSG, a minichromosomal VSG, or a telomere-distal VSG [13]–[18]. Gene conversion between larger regions can result in the duplication of an entire ES, including its VSG [12]. Crossover switches, where two VSGs are exchanged, have also been observed infrequently [19]–[22].
Deficiency of RAD51 or RAD51-3 (RAD51-related gene), or BRCA2, a mediator for RAD51 filament formation, decreased switching frequency in T. brucei [23]–[25]. Mre11 is essential for DNA damage response, as a sensor of double strand breaks (DSBs) that can be repaired by homologous recombination (HR) or non-homologous end joining (NHEJ) [26]–[28]. As in yeast and mammals, T. brucei mre11 null mutants exhibited growth defects, hypersensitivity to a DNA damaging agent, and gross chromosomal rearrangements (GCR), but no detectable decrease in VSG switching [29], [30], indicating that, although antigenic variation shares core features with classic HR, specific roles for recombination factors in antigenic variation remain to be determined.
Mitotic crossover can be detrimental, leading to unequal exchanges. Sgs1, a RecQ family helicase in yeast, is one of the major factors that control spontaneous crossovers [31]. Sgs1 forms a complex with Top3 (type IA topoisomerase) and Rmi1 (RecQ-mediated genome instability), and plays major roles in the suppression of genome instability by influencing mitotic and meiotic recombination, replication fork stability, and telomere maintenance [32]–[38]. At least one mechanism of crossover-suppression appears to involve ‘dissolution’ of double Holliday Junction (dHJ) intermediates. Sgs1-Top3-Rmi1, also known as the RTR (RecQ-Top3-Rmi1) complex, is well conserved in humans as the BLM (Bloom mutated)-TOPO3α-BLAP75/18 (Bloom associated protein 75kDa/18kDa, or RMI1/2). Mutations in any member of the RTR complex increase recombination frequency and crossover [31], [32], [39]–[43]. Defects in the BLM pathway are associated with elevated levels of sister chromatid exchanges (SCEs), chromosomal breaks and translocations [40], [41], [44]–[46].
Crossover has rarely been observed in VSG switching. Suppression of crossover is intriguing because, in principle, the outcome of duplicative VSG conversion holds no apparent advantage over crossover events, as re-expressing a VSG, either exchanged or duplicated, will be lethal in vivo. Given the similarities between HR and VSG switching, we hypothesized that certain yeast hyper-recombination mutants could be hyper-switchers in trypanosomes. Using new recombination and VSG switching assays, we took advantage of a potential member of T. brucei Sgs1 pathway, TbTOPO3α (Tb11.01.1280), to get better insights on how trypanosomes employ recombination factors to control antigenic variation.
Type IA topoisomerases cleave DNA by covalent attachment of one of the DNA strands through a 5′phosphodiester bond to a tyrosine residue in their catalytic domains [47]. In many organisms, type IA topoisomerases function in cooperation with helicases, as a combination of Top3-Sgs1 in yeasts and TOPO3α-BLM in humans. T. brucei expresses a 102.5-kDa TOPO3α protein with 918 amino acids. Figure 1 shows an alignment of TbTOPO3α with human TOPO3α and S. cerevisiae and S. pombe Top3. The primary sequences are well aligned at the N-terminal catalytic domain including the active site tyrosine. Both E. coli Top1 and human TOPO3α contain Zn-binding motif(s) in their C-terminal regions. E. coli Top3 and two yeast Top3 lack a Zn-binding domain (reviewed in [47]). TbTOPO3α seems to have a Zn-binding motif in the C-terminus (four cysteine residues written in red), although this region does not align well with human TOPO3α. The sequences of TOPO3α are very well conserved in T. brucei, T. cruzi and Leishmania major (Supporting Figure S1). T. brucei also has a type IA TOPO3β (http://www.genedb.org/genedb/tryp), but its function has not been studied.
To explore the role of TOPO3α, we sequentially deleted both alleles. We used deletion-cassettes containing hygromycin (HYG) or puromycin (PUR) resistance genes fused to Herpes simplex virus thymidine kinase (HSVTK or TK) and flanked by loxP sites, allowing the markers to be removed by transient expression of Cre-recombinase and reused [48]. Deletion of both alleles was confirmed by PCR analyses (Supporting Figure S2).
Loss of Top3 causes a severe growth defect in budding yeast and is lethal in fission yeast [43], [49]. The absence of TOPO3α or TOPO3β results in embryonic lethality or shortened life span in mice [50], [51]. In contrast, TOPO3α null mutants exhibited only a minor growth defect in T. brucei (Figure 2A).
Yeast Top3 is important for the maintenance of genome integrity. top3 mutants are sensitive to DNA-damaging agents and show defects in the activation of the cell-cycle checkpoint kinase Rad53 (CHK2 in mammals), in response to genotoxic stresses [52]–[54]. We therefore asked whether T. brucei TOPO3α is required for the DNA damage response, by assessing sensitivity to the DSB-inducing agent phleomycin or the replication inhibitor hydroxyurea (HU). Cells were treated with phleomycin for 24 hours and single cells were distributed in 96-well plates. The color of the medium turns from red to yellow when the culture becomes saturated. Yellow wells were counted after 7–8 days and the percent viability was calculated by normalizing to the untreated samples. In the null mutants, viability was reduced by 3-fold at 0.3 µg/ml and 10-fold at 0.6 µg/ml phleomycin (Figure 2B). Viability of the HU-treated null mutants was reduced by 3-fold in 0.04 mM HU (Figure 2C). topo3α−/+ was comparable to the wild type in both experiments. We conclude that TOPO3α is required for the response to DNA damage and replication block, similar to the roles of yeast Top3.
top3 was isolated as a hyper-recombination (‘hyper-rec’) mutant in a genetic screen designed to identify mutations that increase recombination frequency at SUP4-o locus in budding yeast [43]. We therefore hypothesized that Tbtopo3α could be a ‘hyper-rec’ mutant and this phenotype could be reflected in the frequency of recombination-mediated antigenic switching.
To test whether TOPO3α deficiency increases recombination frequency, we established a new recombination assay. Thus far, transfection-based recombination assays have been predominantly used, in which trypanosomes are transfected with linear DNA containing a selection marker flanked by targeting sequences, and the recombination frequency is calculated from the number of drug-resistant clones that arise. Although this method can give reliable measurements, it requires a high rate of recombination at the target site and is subject to variations in transfection efficiency. To allow a more convenient, natural and reliable measure of recombination efficiency, we established an assay (Figure 3A) in which HYG-TK can replace one allele of what we will call TbURA3 (the bifunctional orotidine-5-phosphate decarboxylase/orotate phosphoribosyltransferase Tb927.5.3810) on chromosome V. The frequency of loss of either the HYG-TK or TbURA3 allele represents the rate of gene conversion at this locus. The frequency of HYG-TK loss can be measured with gancyclovir (GCV), a nucleoside analog, as only the cells that had lost the TK gene can grow in the presence of GCV. The loss of TbURA3 can be measured with 5-FOA (5-fluoroorotic acid), as only the ura3− cells can grow in the presence of 5-FOA.
To remove the HYG-TK and PUR-TK markers that were used for the deletion of TOPO3α, Cre-recombinase was transiently transfected into the topo3α−/− cells and GCVR HYGS PURS clones were selected (Supporting Figure S2). One allele of TbURA3 was then replaced with HYG-TK and the targeting was confirmed by PCR. Gene-conversion frequencies were determined by counting total GCRR and FOAR cells, in three wild-type and five topo3α−/− independent HYG-TK clones. As shown in Figure 3B, Tbtopo3α gave indeed a hyper-recombination phenotype. Total gene-conversion frequency was increased 6-fold in topo3α−/− (5.12±0.15×10−5) compared to the wild type (0.87±0.70×10−5).
To investigate the roles for TOPO3α in VSG switching, we generated a VSG switching reporter strain in which we could easily measure switching frequency and score different switching mechanisms. As illustrated in Figure 4, the parental strain expresses VSG 427-2 (221) in ES1, which was doubly marked with a blasticidin-resistance gene (BSD) downstream of the promoter and PUR-TK at the 3′ end of the 70-bp repeat region, without disrupting the co-transposed region (CTR), disruption of which has been shown to induce rapid VSG switching [55]. The 5′ boundaries for recombination-mediated VSG switching have been mapped at regions upstream of CTRs that are located between the 70-bp repeats and the VSG. Therefore, the PUR-TK gene will either be lost or repressed in switched cells. This will allow switchers, but not the parental cells, to grow in the presence of GCV.
Doubly marked wild-type and topo3α−/− cells were maintained in media containing blasticidin and puromycin, to exclude switchers from the starting population. The cells were allowed to switch in the absence of drugs for 3–4 days. Un-switched VSG 427-2-expressing cells were depleted by magnetic-activated cell sorting (MACS) [56]. The column flow-through, highly enriched with switchers, was serially diluted in medium containing 4 µg/ml GCV and distributed into 96-well plates. Switching frequency was determined as the ratio of GCV-resistant cells to the total number of cells prepared for the MACS column experiments. We analyzed three independent wild-type cultures and four topo3α−/− cultures. As shown in Figure 5A, TOPO3α deficiency caused a 10–40-fold increase in switching frequency (26±16×10−5) compared to wild type (1.01±0.45×10−5). This is the only known example of an increase in VSG switching frequency when a repair factor is deleted. To confirm that the column-mediated depletion of VSG 427-2-expressing cells was not biasing our results, other batches of cells were directly diluted in GCV-containing media and distributed into 96-well plates. Switching frequency was 10–30-fold increased in the absence of TOPO3α (data not shown).
We have determined the switching frequency in a strain without the TK marker but with a PUR marker inserted downstream of VSG 427-2 and obtained similar frequencies, ∼1×10−5, in wild type. In two different but closely related cell lines, with the same genotype except that one line has PUR-TK inserted at the 70-bp repeat and the other just PUR, again similar switching frequencies, ∼1×10−5, were observed [56] (personal communication with Nina Papavasiliou).
Reintroduction of wild-type TOPO3α complemented the hyper-switching phenotype of topo3α−/− (−/−/+ in Figure 5B), confirming that this phenotype is associated with the TOPO3α deficiency. The results were obtained from three complemented clones (−/−/+) and two cultures each of wild type and topo3α mutant.
RAD51-dependent recombination intermediates accumulate in top3 mutants and the removal of persistent intermediates requires the cleavage activity of Top3 [57], [58]. We examined whether the hyper-switching phenotype of topo3α−/− is dependent on RAD51. Both RAD51 alleles were sequentially deleted in the wild-type and topo3α−/− strains. We analyzed four independent cultures of rad51−/− and two of topo3α−/− rad51−/−. RAD51 deletion reduced the switching frequency of the wild type by 2-fold and abolished the hyper-switching phenotype of topo3α−/− (Figure 5C). Collectively, we concluded that TOPO3α functions as an important regulatory factor for recombination-mediated VSG switching and that, in the absence of TOPO3α, recombinogenic structures may accumulate between the active ES and VSG donors, and could then be resolved to give rise to switched variants.
In other organisms, Top3 defects are associated with elevated crossover as well as hyper-recombination [32]–[34], [45], [46]. To learn how individual switchers had undergone antigenic variation, we analyzed total 296 cloned switchers. The rationales for the double marking of parental cells are as follows (Figure 4). First, switchers can be effectively counter-selected using GCV (Figure 4 and 5). Second, transcription is initiated at silent ESs but elongation is prematurely terminated [59]: genes that are located closer to silent ES promoters are not completely silenced. Therefore, in-situ switchers can be distinguished from recombination-mediated switchers using different concentrations of blasticidin. Based on our titration for blasticidin concentration, in-situ switchers can grow in 5µg/ml blasticidin but not in 100 µg/ml, while ES gene conversion (ES GC) switchers cannot grow in either concentration. VSG gene conversion (VSG GC) and VSG-exchange (crossover) switchers will be resistant to 100 µg/ml blasticidin, and these alternatives can be distinguished by the absence or presence of VSG 427-2, respectively, which can be analyzed by PCR. The strategies to score individual switching mechanisms are summarized in Table 1 and examples of PCR analyses are shown in Figure 6A (right).
We analyzed cloned switchers isolated from six independent cultures and were able to discriminate among the alternative switching mechanisms. The results are summarized in Table 2 and Figure 6A. Switchers from cultures 1 and 2 were isolated by the column method and switchers from culture 3 by directly plating in GCV. Switching occurred largely by gene conversion (Figure 6A). In both wild type and topo3α mutants, 64∼77% of switching exploited VSG GC. Crossovers were rare in wild type (∼3%) but, on average, 20% of switchers exchanged their VSGs in topo3α−/−. These data suggest that, in the absence of TOPO3α, recombination intermediates may be accumulated and these could be repaired mostly by duplicative VSG GC and crossover.
In a previous study designed to examine in-situ switching, using a cell line with TK marker inserted next to the active ES promoter, frequent loss of entire active ES was observed. This could be caused by duplicative transposition of a silent ES (ES GC) or by deletion of the active ES coupled with transcriptional activation of a silent ES [60]. In our experiments, ES GC and ES loss cannot be distinguished, as switchers that lost both BSD and VSG 427-2 genes could arise either by duplicative transposition of a silent ES or by ES breakage coupled with an ES transcriptional switch. The ‘ES GC or ES loss’ events were rather frequently detected in wild-type cells (average ∼30%), while they were either not detected (culture 1 and 2) or detected at a low frequency (4 our of 51 cloned switchers in culture 3) in the absence of TOPO3α (Figure 6A and Table 2). Interestingly, RAD51 deletion significantly decreased ‘ES GC or ES loss’ frequency (unpublished data), indicating that ‘ES GC or ES loss’ events are mainly under the control of RAD51-dependent recombination.
We noticed that some switched variants had growth disadvantages. Depending on how long it took to saturate the medium, wild-type switchers were categorized as ‘fast’, ‘medium’ or ‘slow’. ‘ES GC or ES loss’ switchers were prevalent in clones that grew up more slowly (data not shown). The functions of ESAGs are mostly unknown, but expressing different ESAGs might be advantageous when entering different hosts [61]. The slower-growth phenotype of some of these switchers may reflect impaired function of one or more ESAGs in the bovine serum-containing culture medium, which appears to favor stable transcription of the VSG 427-2-containing ES1.
In-situ switchers were rare in our assay. This phenotype is different from previous reports [24], [62], for reasons we do not understand. In our hands, in-situ switchers generally grew slower than VSG-GC switchers, so VSG-GC switchers would quickly take over the switched population if it was initially mixed, although this is unlikely because our switching population was initiated at 500–1000 cells/ml, while it was at 5,000–10,000 cell/ml in previous assays. Before this seeding, cells were grown in the presence of drugs that prevented switching.
The 70-bp repeat unit has been proposed to be a recombination hot spot, possibly as a potential target for a site-specific endonuclease playing a similar role to that of the HO-endonuclease in yeast. Such an endonuclease has not been identified in trypanosomes. The 70-bp repeats could serve as switching hot-spots because of their structural features [63], rather than require cleavage by a specific endonuclease. Early experiments suggested that the overall VSG switching-frequency was not reduced in the absence of 70-bp repeats or by inversion of the repeats although, when present in the correct orientation, the repeats were used more than 10% of the time [62]. More recently, however, it has been shown that the 70-bp repeats of the actively transcribed ES are prone to break, which could induce recombination-mediated switching, and that the switching frequency was greatly increased when breaks were experimentally induced at the 70-bp repeats, but not when induced elsewhere in the ES or in the absence of 70-bp repeats [56].
We mapped the region where the recombination occurred (or resolved) in the VSG-GC switchers from wild type and topo3α mutants, to learn whether the 70-bp repeat unit is the hot spot of duplicative VSG GC and whether TOPO3α can redirect this preference. ESAG1 genes are located immediately upstream of the 70-bp repeats, and their sequence polymorphisms allowed us to design ES1-specific-ESAG1 oligonucleotides for PCR analysis. PCR results from several VSG-GC switchers were shown in Figure 6A (right). The presence of ES1-specific ESAG1 in VSG-GC switchers indicates that gene conversion occurred at 70-bp repeat regions, and its absence indicates that recombination occurred upstream of ESAG1 (Figure 6B). Crossover and ‘ES GC or ES loss’ switchers were used to verify that the PCR primer set was amplifying only the ES1-specific ESAG1 gene. The ES1-specific ESAG1 was lost in all ‘ES GC or ES loss’ switchers but was detected in all crossover switchers examined, as expected. The ES1-specific ESAG1 gene was amplified in ∼63% of VSG-GC switchers in wild-type cells but ∼81% of VSG-GC switchers lost the ES1-specific ESAG1 gene in topo3α−/−, indicating that, in the absence of TOPO3α, the active ES recombined mostly with silent ESs upstream of ESAG1, rather than within the 70-bp repeats, but not with minichromosomal or telomere-distal VSGs. We concluded that the 70-bp repeat region is an important but not an essential element for recombination-mediated switching. Gene conversion upstream of 70-bp repeats, at ESAG2, has also been reported [64]. The primary function of TOPO3α may be to prevent accumulation of recombination intermediates constantly arising between the active and silent ESs, to maintain the integrity of ESs.
Recombination by a one-strand invasion event could replace VSGs by break-induced replication (BIR) [56]. Alternatively, a second strand invasion at homologous sequences within or downstream of the VSG could generate VSG-GC switchers. Duplication of a telomere-distal VSG into an active ES is a relatively rare event, at least in the modest extent to which switching events have been characterized experimentally, but it appears to serve as an important switching mechanism in later stage of infection and as a mechanism to further expand the expressed VSG repertoire [22], [65], [66]. The few telomere-distal VSG arrays so far characterized contain only short stretches of 70-bp repeats but lack telomeric repeats. To determine how VSG GC occurred, we analyzed the sequences downstream of the 3′ homology region of VSG 427-2 by PCR in all VSG-GC switchers (Supporting Figure S3). If the second strand invaded at this 3′ homology region, downstream sequences should be unchanged. We found, however, that the ES1-specific downstream sequences were lost in all the VSG-GC switchers obtained from wild-type and topo3α cells, indicating that VSG-GC switchers were most likely repaired by BIR, consistent with a recent report [56], and that internal-VSG duplication is extremely rare. PCR results from a selection of VSG-GC switchers were shown in Figure S3.
To confirm the duplicative translocation of newly expressed VSGs to the VSG 427-2 ES and to examine whether minichromosomal VSGs contribute to antigenic switching, 32 VSG-GC switchers from wild-type cells were further analyzed. Minichromosomes terminate with telomeres, VSGs and 70-bp repeats. Gene conversion with minichromosomal VSGs occurs frequently [56], but only when recombination is initiated at the 70-bp repeats. Therefore, we cloned and sequenced newly activated VSGs from VSG-GC switchers that utilized 70-bp repeats. From 32 switchers that had undergone at least one type of switching, VSG GC at the 70-bp repeats, we obtained eight different newly activated VSGs (Supporting Figure S5, left). It is possible that we have underestimated the number of independent switching events as these switchers may have used different sequences within or near the 70-bp repeats, which should be counted as independent. Some switchers might have arisen earlier than others, for examples VSG 427-32, as these were presented more often than others. Among these eight newly expressed VSGs, four were novel VSGs, 427-32, 33, 34 and 35, full or partial sequences of which can be found in the following website (http://tryps.rockefeller.edu). Switchers expressing VSGs 427-3, 11, 32, 33 and 35 were examined by rotating agarose gel electrophoresis (RAGE) and Southern blot [56]. As shown in Supporting Figure S5 (right panel), VSG 427-2 was lost in all the switchers and all newly expressed VSGs were duplicated and translocated to the 427-2 ES, except for 427-33, an intermediate chromosomal (IC) VSG. The original copy of 427-33 may be lost after recombination. VSGs 427-32 and 35 came from megabase chromosomes (MBC). We have not isolated any minichromosomal VSGs in these switchers, indicating that recombination between ES-associated VSGs was the major source for VSG switching.
Repair by recombination serves to preserve genome integrity and can either homogenize or diversify genetic information, occasionally causing detrimental outcomes or benefiting certain organisms by providing adaptation systems to escape lethal situations. African trypanosomes escape the host immune response through a mechanism known as the antigenic variation. Here, we report that T. brucei TOPO3α, a member of a potential T. brucei RecQ-Top3-Rmi1 (RTR) complex, takes an important part in the regulation of recombination-mediated antigenic variation. Our results reveal a complex mechanism that has to balance ES integrity and VSG diversity to maximize the survival of a trypanosome population by suppressing crossovers on one hand and by promoting duplicative VSG gene conversions on the other.
As illustrated in Figure 7, ES structures seem to play a particular role in VSG switching. ES-associated VSG genes are located between the 70-bp and telomeric repeats. ESAGs and some pseudogenes are present upstream of the 70-bp repeats in all ESs, sometimes duplicated and sometimes missing [3], [9]. Strong sequence homologies are present throughout the ESs, with the exception of most of the VSG coding sequence and the immediately upstream ‘co-transposed region’ (CTR). VSG sequences are highly dissimilar except for ∼200-bp encoding the C-terminus and within the 3′ UTR [67]. The reason why every VSG cassette contains a unique CTR is unknown. The purpose of CTR could be to insulate the individuality of VSG cassettes, so that the VSG sequences can evolve separately from other regions in ESs, which maintain their sequences to serve for VSG recombination. When HR occurs, the CTR could block branch migration of HJ or dHJ downstream of the 70-bp repeats.
What roles does TOPO3α play in this scheme? Our study shows that TOPO3α deficiency increases VSG switching, especially VSG GC and crossover, and that the hyper-switching phenotype requires RAD51. The accumulation of toxic recombination intermediates accounts for the slow growth phenotype of yeast top3 mutants, which is suppressed by mutations in SGS1 or in the RAD51-pathway [43], [68], [69]. Recombination intermediates accumulate in cells over-expressing dominant-negative Top3-Y356F in response to methylmethane sulfonate in a RAD51-dependent manner [58]. The function of TOPO3α is not restricted to the 70-bp repeats in antigenic switching, as its absence appears to cause promiscuous recombination throughout the ESs. We therefore propose that TOPO3α removes recombinogenic structures constantly arising between ESs so as to maintain the albeit limited individuality of different ESs. In the absence of TOPO3α, recombination intermediates would accumulate during VSG switching and unresolved intermediates would have to be repaired either by GC associated with crossover or by placing a new duplicated VSG into the active ES by BIR (Figure 7).
Suppression of crossover in recombination-mediated VSG switching is an interesting result, considering that there are probably more than 200 potential VSG donors: ∼20 ESs with extensive sequence homology and ∼200 minichromosomal VSGs. Antigenic variation probably requires balancing preservation and variation of VSG information, but we cannot explain how suppression of crossover would be important for maintaining this balance. However, we think that by favoring duplicative GC over crossover, rather than crossover over GC, trypanosomes could slowly accumulate VSG diversity without abrupt loss of their functionalities, because duplicative GC requires VSG DNA synthesis, during which point mutations could be incorporated into newly synthesized VSGs, but VSG crossover does not require VSG DNA synthesis.
TOPO3α deficiency increased VSG GC far more than GC at the URA3 locus (Figures 3 and 6). GC at these two loci is probably mediated by different pathways. Recombination at URA3 locus would prefer flanking homologies, rather than BIR. In contrast, BIR would present a better option for VSG GC, as only one end homology appears to be involved (supporting Figure S3) [56]. It is possible that a second invasion could occur within the telomere repeats, but this is impossible to determine. The higher VSG GC rate could also be because the active ES is less stable than URA3 locus. Alternatively, TOPO3α may specifically suppress BIR-mediated VSG switching. The role of TOPO3α in BIR has not been extensively characterized elsewhere. Our results show a novel function of TOPO3α in VSG switching, which could be an excellent system to study BIR.
DNA recombination involves many factors, of which only a few have been studied in the context of antigenic variation: RAD51, RAD51-related genes, BRCA2, KU70/80, MRE11, and MSH2/MLH1 [23]–[25], [29], [30], [70], [71]. Among these, only the deletion of RAD51, RAD51-3, and BRCA2 decreased VSG switching, in wild-type cells that already had a very low switching rate.
Our findings on TOPO3α in VSG switching suggest potential roles for numerous DSB-HR response factors in antigenic variation. Two RecQ family helicases are annotated in the T. brucei gene database (http://www.genedb.org/genedb/tryp). Rmi1 is required to load Top3 onto the substrates and stimulate its activity through the physical interaction [72]. We have identified a TbRMI1 homologue. All the phenotypes that we have examined in Tbrmi1 mutants were identical to those in topo3α mutants (unpublished data). Therefore, we believe that RecQ, TOPO3α and RMI1 are likely to function as a complex in antigenic variation in T. brucei.
Synthetic-lethality screens with sgs1 in budding yeast identified three pathways working in parallel with Sgs1 [73]; Mus81-Mms4, Slx1-Slx4, and Slx5-Slx8. Synthetic lethality of sgs1 mus81 or sgs1 mms4 requires HR factors [74]. Mus81-Mms4 is a structure-specific endonuclease that cleaves 3′ flap, replication fork, or HJ substrates [74]–[76]. Resolvase, an endonuclease, symmetrically cleaves HJs and the products can be resolved with crossover or non-crossover. Human and yeast resolvases have recently been characterized [77]. MUS81 appears to be present in T. brucei but a resolvase remains to be identified. Although we do not yet have functional data for these proteins, we propose, based on the studies from other organisms, that the regulation of antigenic variation is similar to that of mitotic HR. When present, TOPO3α could dissolve dHJs to prevent the ES instability, consequently generating non-crossover recombinants (no switching). In the absence of TOPO3α, resolvase (Figure 7a, grey box) or MUS81 may cleave the accumulated recombination intermediates arising between the ESs and generate crossover switchers. Alternatively, stalled replication forks can be cleaved by MUS81 and the broken leading strand can invade a silent ES to generate VSG-GC switchers (Figure 7b, grey box).
Although VSG switching has similarities with mitotic HR, it appears that specific elements are present for its regulation. A hyper-recombination phenotype does not always correlate with hyper-switching phenotype. The mismatch repair (MMR) pathway can abort recombination during strand exchange between non-identical substrates and mmr mutants can increase recombination frequency (reviewed in [78]). Consistent with their roles in repair and recombination, Tbmsh2 or Tbmlh1 mutants increased recombination frequency but did not change switching frequency [71]. Recombination is closely linked with DNA replication and checkpoint pathways as well [32], [57], [58], [79]. Therefore, we believe that roles for DNA replication, checkpoint, and recombination factors and their interactions need to be determined to fully understand the mechanisms of antigenic variation.
Measuring VSG switching has, until now, been time-consuming and not very reproducible. Our new switching assay circumvents previous technical difficulties and can effectively assign specific roles to individual proteins.
It has recently been shown that a DSB introduced at the active 70-bp repeats by the I-SceI endonuclease causes a 250-fold increase in VSG switching and that the DSBs were repaired by BIR [56]. However, it is unknown whether the VSG switching is activated by targeted DSBs or by random chromosomal breaks, or whether recombinogenic ssDNA is a primary cause for the initiation of VSG switching. HR can be instigated by many different sources; random breaks, endonuclease cleavage at specific target sites, replication fork instability, unusual secondary DNA structure, or transcription.
The Mre11 complex, which consists of Mre11, Rad50, and Xrs2 (NBS1 in mammals), plays a central role in the DSB-HR response [26]–[28]. MRE11 deficiency, however, did not change the VSG switching frequency [29], [30], promoting the idea that ssDNA regions may generate recombinogenic structures for the initiation of switching. Uncoupling of leading and lagging strand DNA synthesis caused by DNA lesions can destabilize a replication fork, leaving ssDNA gaps behind the fork, which could be processed into recombinogenic structures. If an ssDNA gap is a major trigger for recombination-mediated switching, switching frequency should increase in cells suffering from replication challenge. To address this issue, we treated cells with aphidicolin, an inhibitor of lagging strand DNA synthesis, and HU, and measured the switching frequency in parallel (Supporting Figure S4). Cells were treated with the drugs at a sub-lethal dose to exclude a possibility of chromosome break-induced switching. No significant correlation was observed between these treatments and switching frequency. Therefore, an ssDNA gap may not be a major initiating factor for VSG switching. Rather, random breaks might be responsible for switching induction, consistent to a previous study [56]. However, it is still difficult to rule out the possibility that an ssDNA gap triggers switching, as ssDNA gaps might not be extensive enough to create recombinogenic structures at the low doses of aphidicolin or HU. The best way to test this hypothesis would be to use conditional mutants associated with replication defects. Unfortunately, we do not yet have such genetic tools, as nuclear DNA replication has not been studied in T. brucei.
A high transcription level can stimulate recombination, a mechanism known as transcription-associated recombination (TAR) (reviewed in [80]). Transcription has been shown to promote recombination in T. brucei [81], [82]. Interestingly, it was shown in budding yeast that transcription- and DSB-induced recombination events were similar, indicating that transcription affects only the initiation of recombination, not the mechanism of recombination [83]. ssDNA regions exposed in the active ES during transcription could be readily accessible by recombination factors. Alternatively, transcription-replication collision causes replication fork stalling, which could also induce switching. Studies of mammalian cells have shown that TAR is dependent on replication [84], and that transcription increases recombination frequency when a replication fork converges with transcription [85]. The active ES is more fragile than silent ESs [56]. The high level of transcription may explain why the active ES breaks more frequently, and this may induce VSG switching.
The 70-bp repeat has been proposed to be a potential endonuclease target site to induce switching, but such an enzyme has not been found. Instability of the 70-bp repeat [63] may play a role in the initiation of switching and could lead to template switching. However, according to our results and previous studies [62], [64], switching is not completely dependent on the 70-bp repeats. With the available data, it would be reasonable to conclude that random breaks may occur throughout the active ES but more frequently at 70-bp repeats, and these could initiate various switching events.
Gene conversion is used by several other pathogens, including Borrelia hermsii and Anaplasma marginale, as an evasion mechanism [10], [86]. Our study suggests that exploring how trypanosomes manipulate the HR machinery to gain advantage against their host's immunity, while successfully preserving their genomes, may reveal weaknesses that can be exploited to control infectivity and virulence.
Trypanosoma brucei bloodstream forms (strain Lister 427 antigenic type MITat1.2 clone 221a (VSG 427-2)) were cultured in HMI-9 at 37°C. The cell lines constructed for this study are listed in Supporting Table S1, and they are of ‘single marker’ (SM) background that expresses T7 RNA polymerase and Tet repressor (TETR) [87]. Stable clones were obtained and maintained in HMI-9 media containing necessary antibiotics at the following concentrations, unless otherwise stated: 2.5µg/ml, G418 (Sigma); 5µg/ml, blasticidin (Invivogen); 5µg/ml, hygromycin (Sigma); 0.1µg/ml, puromycin (Sigma); 1µg/ml, phleomycin (Invivogen). Plasmids used for this study are listed in Supporting Table S2.
TOPO3α genes were sequentially deleted using deletion-cassettes containing either puromycin or hygromycin-resistance gene fused with HSVTK, Herpes simplex virus thymidine kinase (TK), PUR-TK and HYG-TK. These fusion genes are flanked by loxP sites so that the markers can be removed by transient expression of Cre recombinase (pLew100-Cre). The entire open reading frame (ORF) of TOPO3α was deleted by transfecting ‘single marker’ (SM) cells with a deletion-cassette that was amplified with primer 35 and 36 using pHJ18 (PUR-TK) as a template. Primer 35 and 36 contains 70 nt homologies to the target sites. This topo3α ‘single knock-out’ cells (sKO, HSTB-97) were used to PCR amplify a cassette containing a marker (PUR-TK) along with 453 nt upstream and 402 nt downstream sequences of TOPO3α gene. The PCR fragment was inserted into pGEM-easy-T vector by TA cloning to create pHJ63. pHJ64 was constructed by replacing a PUR-TK marker with a HYG-TK from pHJ17. topo3α ‘double knock-out’ (dKO) was generated by transfecting NotI-digested pHJ64 into topo3α sKO, HSTB-97. Deletion of both TOPO3α alleles was confirmed by PCR analyses.
To remove the selection markers, topo3α dKO cells were transfected with pLew100-Cre to transiently express Cre-recombinase, and the cells that lost both HYG-TK and PUR-TK were selected in 50µg/ml ganciclovir (GCV). Loss of markers was confirmed by resistance to puromycin and hygromycin, and by PCR analysis. The sequences of primers used here are available upon request.
pLHTL-pyrFE [48]-linearized by PvuII digestion was transfected into wild-type (HSTB-188) and topo3α−/− (HSTB-328 and HSTB-330) cells, to replace one allele of TbURA3 with HYG-TK. The integration was confirmed by PCR analysis with primers 48 and 49. Three or five independent HYGR clones from wild-type or topo3−/− cells were analyzed. Cells were grown in the absence of hygromycin for 2 days to allow recombination to occur. Approximately 500,000 cells were diluted in HMI-9 media containing 30 µg/ml GCV or 6 µg/ml FOA, and distributed into 96-well plates. Yellow wells (phenol red indicating acidification due to growth) containing GCVR or FOAR cells were counted after 7–8 days of incubation and the GC frequency was determined. The sequences of primers used for genotyping are available upon request.
To create a doubly-marked switching reporter strain (Figure 4), pHJ23 was linearized by KpnI-NotI digestion and integrated downstream of the ES1 promoter, to confer resistance to blasticidin. These cells were then marked with PUR-TK at the 3′ end of 70-bp repeats by transfecting a PCR-amplified PUR-TK cassette. Ten µg/ml of puromycin, 100 times higher than normal usage, was added to select clones targeted specifically at the active ES. When determining switching frequency, the parental cells were maintained in the presence of blasticidin and puromycin to exclude switchers from the starting population. Cells were then allowed to switch in the absence of selection for 3–4 days. Switchers were enriched using a MACS [56]. Flow-through enriched with switchers was collected and serially diluted in media containing 4 µg/ml GCV, and distributed into 96-well plates. The switching frequency was determined by counting GCVR clones. Alternatively, switching frequency was determined without the column enrichment step. Cells were diluted in GCV-containing media and directly distributed into 96-well plates. Non-switchers that carry spontaneous mutation(s) in TK gene but not in PUR were ruled out by examining puromycin resistance. Non-switchers that carry mutations in PUR and TK were ruled out by western blot analysis using antibodies against VSG 427-2.
To determine switching mechanisms, cloned switchers were analyzed for blasticidin sensitivity at 5 µg/ml and 100 µg/ml concentrations. Genomic DNA was prepared from 296 switchers and PCR-analyses were performed at four regions: BSD, VSG 427-2, ESAG1, and VSG 427-2 downstream. The primer set designed for BSD-PCR can also amplify TETR (Tet Repressor) gene, which was used as a control for PCR analyses. The sequences of primers used here are available upon request.
Wild type (SM), topo3α−/+ (HSTB-97), and topo3α−/− (HSTB-226 and HSTB-227) cells were incubated with indicated concentration of phleomycin for 24 hours. The same number of cells was distributed into 96-well plates. All the plating was duplicated. The wells that contain viable cells were counted after 7–8 days of incubation at 37°C and the viability was calculated by normalizing to untreated samples. Sensitivity to HU and aphidicolin was determined similarly. Cells were incubated with HU or aphidicolin for 2 or 3 days. The viability was calculated by normalizing to untreated samples.
Database ID numbers (http://www.genedb.org and http://tritrypdb.org) for TOPO3α discussed in this paper are Tb11.01.1280, LmjF36.3200 and Tc00.1047053511589.120. What we refer to as TbURA3 is the bifunctional orotidine-5-phosphate decarboxylase/orotate phosphoribosyltransferase Tb927.5.3810.
|
10.1371/journal.pntd.0002053 | Latent Infection with Leishmania donovani in Highly Endemic Villages in Bihar, India | Asymptomatic persons infected with the parasites causing visceral leishmaniasis (VL) usually outnumber clinically apparent cases by a ratio of 4–10 to 1. We describe patterns of markers of Leishmania donovani infection and clinical VL in relation to age in Bihar, India.
We selected eleven villages highly endemic for Leishmania donovani. During a 1-year interval we conducted two house to house surveys during which we collected blood samples on filter paper from all consenting individuals aged 2 years and above. Samples were tested for anti-leishmania serology by Direct Agglutination Test (DAT) and rK39 ELISA. Data collected during the surveys included information on episodes of clinical VL among study participants.
We enrolled 13,163 persons; 6.2% were reactive to DAT and 5.9% to rK39. Agreement between the tests was weak (kappa = 0.30). Among those who were negative on both tests at baseline, 3.6% had converted to sero-positive on either of the two tests one year later. Proportions of sero-positives and sero-converters increased steadily with age. Clinical VL occurred mainly among children and young adults (median age 19 years).
Although infection with L. donovani is assumed to be permanent, serological markers revert to negative. Most VL cases occur at younger ages, yet we observed a steady increase with age in the frequency of sero-positivity and sero-conversion. Our findings can be explained by a boosting effect upon repeated exposure to the parasite or by intermittent release of parasites in infected subjects from safe target cells. A certain proportion of sero-negative subjects could have been infected but below the threshold of antibody abundance for our serologic testing.
| In this study we assessed trends with age in the probability of being sero-positive or sero-converting for two serological markers of visceral leishmaniasis (VL) among asymptomatic residents of high incidence villages. As markers we used Direct Agglutination Test (DAT) and rK39 ELISA. We also compared titers among asymptomatic sero-positives with those of known recent VL cases among our study population. Infection with VL is assumed to be permanent, but sero-positivity is a temporary phenomenon. Though clinical VL was most common among children and young adults (median age 19 years), we observed a statistically significant increase with age in the probability of being sero-positive and in the probability of sero-converting. We also observed that the average antibody titers among asymptomatic sero-positives were much lower than those among recent clinical VL cases. The increase with age in the probability of being sero-positive but also in the probability of sero-conversion can be explained by subjects experiencing repeated episodes of sero-positivity. This could be due to a boosting effect upon re-exposure, or to internal release of parasites from safe target cells. The implication of our findings is that in VL endemic areas it will be difficult to reliably distinguish between infected and non-infected subjects.
| Visceral leishmaniasis (VL) or kala-azar is a parasitic infectious disease that is fatal if left untreated. Two Leishmania species are causal agents of VL: L. infantum and L. donovani. The first is a zoonosis and is endemic in countries around the Mediterranean basin and in Latin-America. The second is assumed to be an anthroponosis and is endemic in East-Africa and the Indian subcontinent [1].
India, Nepal and Bangladesh face a very high burden of VL. The ecological conditions for transmission of VL are very favorable in the Gangetic plains [2]. The State of Bihar in North Eastern India contains the biggest focus of VL, and reports about half of the world's annual new cases. In 2005 India, Nepal and Bangladesh launched a joint VL elimination initiative with a target of bringing the incidence down to less than 1 case per 10,000 by 2015. The main strategy to achieve this was early diagnosis and treatment, along with vector control measures. Key assumptions underlying this elimination strategy are that the disease is indeed an anthroponosis and that active cases of VL and post-kala azar dermal leishmaniasis (PKDL) are the only reservoirs maintaining disease transmission. During the past decade these critical assumptions have been questioned, and there are claims that animal reservoirs may play a role in L. donovani transmission [3], [4]. More recently Stauch et al [5] pointed to the possible role that latent carriers of L. donovani infection could play in transmission, if they were infectious for sand flies. One xenodiagnosis experiment in Brazil failed to demonstrate the ability of subclinicallly infected individuals with L. infantum to transmit parasites to the sand fly vector [6]. Nonetheless, given the biological differences between L. infantum and L. donovani, the different host, vector and epidemiological factors and the potential clinical and public health implications, these conclusions cannot necessarily be generalized to L. donovani.
The concept of latent infection with the agents causing VL has been well demonstrated. Indeed many individuals in endemic regions for L. infantum or L. donovani test positive for immunological markers of infection, but lack present or past symptoms of disease. In 1959, Manson-Bahr showed subjects without any symptoms of VL that were positive in the Leishmanin Skin Test (LST) in endemic areas in Kenya [7]. The term “asymptomatic infection” was used for the first time in 1974 by Pampiglione et al. based on LST responses in a region endemic for L. infantum in Italy [8], and later by Badaro et al. in Brazil [9]. Evidence for latent infection with Leishmania spp. has recently been expanded beyond tests of immunological response. As such, Leishmania spp. DNA was detected by PCR in the peripheral blood of asymptomatic human carriers in Brazil [10] and in Nepal [11]; Le Fichoux et al. [12] cultured promastigotes of L. infantum from the buffy coat of 9 out of 76 asymptomatically infected blood donors in southern France.
Several prospective studies have documented the ratio of asymptomatic infection to clinical disease by estimating the number of incident sero-conversions to incident new VL cases due to L. donovani. Bern et al. [13] in Bangladesh used LST and rK39 as markers of infection and found a 4 to 1 ratio between incident infection and disease. Ostyn et al. [14] used the Direct Agglutination Test (DAT) to document a 9 to 1 ratio in Bihar, India. Although ratios may be skewed because some of these “asymptomatically infected” individuals could progress to disease, the majority do not [15].
Proper assessment of the outcome of asymptomatic infection with parasites causing VL requires the identification of incident infections, and follow-up study of cohorts of these persons [16]. Ostyn et al. [14] showed a ten-fold higher risk of developing symptomatic VL among incident sero-convertors in India and Nepal compared to sero-negative controls. Such studies have been hampered by the fact that the routinely used antibody detection tests have shown excellent performance in detecting clinical cases when used in combination with a clinical case definition, but these tests have never properly been validated for the identification of asymptomatically infected persons [17]. The cut offs for positive serological tests have been chosen to separate healthy from diseased individuals, but not to separate infected healthy individuals from uninfected healthy individuals. Studies that document a working definition of asymptomatic infection with L. donovani are therefore needed.
This manuscript describes the results of two rounds of sero-survey which were conducted with the purpose of identifying a cohort of recent sero-convertors in Bihar, India. We use the data from the two baseline surveys to explore patterns in VL sero-positivity and sero-conversion in relation to age. Such baseline information will set the stage for downstream investigations of the spectrum of infected humans that can possibly serve as a reservoir for this pathogenic protozoan.
The study was performed in the context of a larger ongoing longitudinal study in a high VL incidence area of Muzaffarpur district, Bihar State, India. The study site is a rural area comprised of 50 villages with a total population of 85,333, in which 193 VL cases were reported over a 2 ½ year period (March 2007–December 2009). To investigate L. donovani transmission and validate markers of infection, we set up a cohort of recent seroconverters, i.e. persons who were negative in leishmania markers at baseline, but converted during the follow up. To do so, we worked in the subset of villages with the highest VL incidence within the study area. We selected 11 villages with a total population of 19,886 individuals above 2 years of age, from which 144 cases of VL had been reported since March 2007.
Two house to house surveys were conducted with a one year interval in between. The first survey took place between December 2009 and February 2010. All residents above 2 years of age who were present and gave their informed consent were enrolled in the study. Informed consent was obtained from parents or legal guardians of subjects under age 18. Among subjects consenting to participate, a capillary blood sample was collected by finger prick on Whatman 3 filter paper. Samples were dried, after which they were packed in sealed plastic envelopes with silica gel. Filters were stored at −20°C until further processing.
DAT and rK39 ELISA were applied to detect antibodies against L. donovani in the participants sera; tests were performed as detailed elsewhere by Khanal et al. [16]. DAT titers were determined using a kit from the Institute of Tropical Medicine (DAT/VL, ITM) according to the manufacturer's instructions. Briefly, eight serial two-fold dilutions starting from an initial 1 in 200 dilution of serum were made. End titers for samples not reactive at the first step were classified as <1∶200, those still reactive in the final dilution step were classified as >1∶25,600. The recommended threshold for a positive DAT indicating VL is 1∶3,200; using this cutoff Harith et al. [18] arrived at 100% sensitivity and 99.3% specificity in a VL endemic district in Kenya. Joshi at el [19] in Nepal found DAT at a cut off of 1∶800 to be 100% sensitive and 99.2% specific for detection of clinical VL. Davies and Mazloumi Gavgani [20] and Saha et al. [21] used a cut off of 1∶1,600 to detect sub clinical infection, which is also the titer we opted for in this study.
RK39-ELISA results were expressed as the subject optical density (OD) value divided by OD value of a positive control serum sample ×100, and called percentage point positivity (pp) of a positive control. The cutoff chosen to define rK39 positivity was calculated as the mean value for a healthy non-endemic control plus three standard deviations. A log transformation was used to compensate for skewed distribution. The resulting value was 14 percent of the OD of a positive control. When endemic controls were used, the value increased to 23 percent of the OD of a positive control. To validate the cutoffs chosen we conducted a receptor operator curve (ROC) analysis based on comparison with confirmed recent VL cases since March 2007. We determined the cutoff point with the highest Youden index, i.e. the point with the highest combined sensitivity and specificity [22].
Sensitivity analysis was performed with higher cutoff values for both DAT and rK39. For DAT we used 1∶3,200, the conventional cutoff used for diagnosis of clinical cases; for rK39 ELISA we used the mean among non-cases in our (endemic) study population plus 3 standard deviations, i.e. 23 pp.
The study was part of a larger study in which information on episodes of active VL in the period between March 2007 and the time of the second survey had been collected for all subjects; episodes reported had been verified by a study physician [23]. During the survey, subjects were asked whether or not they had suffered from VL before March 2007.
Data were analyzed using Stata/IC V10.1 (Stata Corp., College Station Tx, USA). Agreement between DAT and rK39 on a binary scale was assessed using Cohen's kappa coefficient [24]. Kappa coefficients were interpreted following Landis and Koch [25]: 1.00–0.81 excellent, 0.80–0.61 good, 0.60–0.41 moderate, 0.40–0.21 weak and 0.20–0.00 negligible agreement.
To assess the trends with age for the different markers studied, the study population was subdivided into 8 age groups (2–9; 10–19; 20–29; 30–39; 40–49; 50–59; 60–69; and 70+). For each age group the proportion of sero-positives and the proportion of VL cases were determined with their 95% confidence interval. Statistical significance of trends observed over age groups was assessed by chi squared for trend analysis. Logistic regression models were used to explore the associations between age as independent variable and serological markers or active VL as dependent variables. For this purpose we included in the model age, age squared and age to the power of three and used a backward elimination procedure, probability for removal was set at <0.05. The probabilities of suffering VL, being sero-positive and sero-converting were plotted by age.
We also assessed by age group the amplitude of the response for DAT and rK39, i.e. the median DAT or rK39 titer among sero-positives. For this purpose, DAT results were expressed in titer steps ranging from 1 (<1∶200) to 9 (>1∶25,600), rK39 results were expressed as percentage points as explained earlier. To test for statistical significance of differences in the distribution of DAT or rK39 titers by age group we used a Kruskal-Wallis test.
To assess the association between time elapsed since diagnosis and DAT or rK39 titers among ex-VL cases we used a Kruskal-Wallis test and linear regression on log transformed titers for rK39.
This study forms part of a larger study for which ethical clearance was obtained from the review committee of the U.S. National Institutes of Health (NIH), as well as Institutional Review Boards of the Institute of Medical Sciences, Banaras Hindu University, Varanasi, India, and the University of Iowa. The IRB at Banaras Hindu University is registered with the US National Institutes of Health. Data was anonymized. All subjects provided written informed consent; in case of illiterate subjects a thumb print plus a signature of an independent witness were used. For minors under the age of 18 informed consent was obtained from a parent or guardian.
We enrolled 13,163 subjects, among whom were 118 individuals who had recently been diagnosed with VL (between March 2007 and December 2009), and an additional 411 individuals who reported a diagnosis of VL before March 2007. Individuals with former VL were all excluded from the first stage of the analysis. The DAT titers showed a peak of subjects around the 1∶400 dilution, and a much lower second peak at a much higher titer of 1∶25,600 (figure 1). The distribution of absorbance according to rK39 ELISA was skewed to the right with a steep peak at approximately 7 pp, but no second mode was observed at higher levels of optical density (figure 2).
Based on the pre-specified cutoff criteria for differentiating positive from negative tests, 777 subjects (6.2%) were DAT positive (cutoff of 1∶1,600). Considering rK39 results with a cutoff chosen using serology form non-endemic controls (14 pp cutoff), 741 (5.9%) of the sera were rK39 positive. Agreement between the two assays was weak with a Kappa of 0.30 (95% CI 0.28–0.32). ROC analysis showed that for the rK39 ELISA, 14 percentage points was the value with the highest Youden index.
The alternative cutoff for rK39 ELISA of 23 pp resulted in 206 positive subjects (1.6%). Agreement with results of the DAT improved only marginally when using this higher cut off for rK39 in combination with a cutoff of 1∶3200 for DAT. (Kappa 0.31, 95% CI 0.30–0.33).
Considering the trends in serologic response with age, we observed a steady increase with increasing age according to each of the serologic measures. The proportion of reactors according to DAT increased from 2.6% in the 2–9 year-old group to 15.9% in those aged 70 or older (table 1). Chi squared for trend was highly significant (p<0.001). Findings using rK39 were similar, although the proportion of reactors in the age group of 70 years and above was lower than that of the two preceding age groups. The confidence interval was wide, however, and the decline was not statistically significant. Overall the Chi squared for trend with age was highly significant (p<0.001).
We assessed whether the amplitude of the response among DAT or rK39 reactors was influenced by age (figure 3). The DAT titer appeared to increase in each age group up to age 40, and decrease among age groups over age 59 (figure 3 left). There was not a consistent pattern of rK39 titer (figure 3 right). Differences in median DAT titers between age groups observed among those above 40 years of age were not statistically significant (p = 0.92).
Between the first round sero-survey and the second sero-survey one year later, 2 VL cases were reported. Both were already rK39 positive (15.6 and 45.6 pp) and one was also DAT positive (end titer ≥25,600) at the time of the first survey round. During the second round survey, 252 DAT conversions and 145 rK39-ELISA conversions were documented. Among these, 16 persons had converted according to both tests. Among the DAT convertors, 19 were already rK39-ELISA positive during the first sero-survey, among the rK39-ELISA convertors 11 had already been DAT positive; thus in total there were 351 sero-negative individuals who seroconverted to DAT and/or rK39-ELISA over a year in our endemic study population. This constitutes 3.6% of the susceptible population of 9,873. The agreement between DAT and rK39 conversion was negligible, kappa = 0.071 (95% CI 0.052–0.090). The incidence of sero-conversion increased with age from 1.9% in the 2–9 years old to 7.0% in those aged 70 and above, chi squared for trend was highly significant (p<0.001) (table 2).
Among 741 individuals who had a positive rK39 test identified in the first survey round, 626 were also sampled in the second round. Out of those, 372 (59%) had reverted back to sero-negative for rK39. Similarly, 777 individuals were initially DAT positives. Among these, 664 were sampled in the second survey and 216 of these (33%) had reverted back to DAT sero-negative.
Included in the study population were 118 VL cases diagnosed between March 2007 and December 2009. Median age at time of diagnosis was 19 years. At the time of the sero-survey, 110 (93%) were still DAT positive and 96 (81%) were still rK39-ELISA positive; 116 (98%) were positive on either of the two assays. Average titers among ex-VL cases were clearly higher than those observed among subclinically infected subjects. The median DAT titer was >25,600 among ex-VL cases and 3,200 among subclinically infected; for rK39 the median titers were 33 pp and 18 pp respectively (figure 3 and 4). Whereas rK39 titers appeared to decline over time, DAT titers were fairly stable. The decline in rK39 titers was marginally statistically significant (p = 0.045).
Over the period of March 2007 till December 2009 (assuming a steady population) the highest VL incidence was among young adults in the 20–29 years age group; from this age onwards there was a steady decline, and there were no cases in persons above 70 years of age (table 3).
We modeled DAT and rK-39 positivity and the probability of active VL and sero-conversion against age as independent variable. The probability for being a VL case was highest at 25 years, with a steep decline from that age onwards. For sero-positivity and sero-conversion there was a steady increase with age, until 59 years for sero-conversion, until 67 years for rK39 positivity and beyond that age for DAT positivity (figure 5).
In this large sero-survey in highly VL endemic villages in Bihar, India, we found that at the time of our baseline survey 5.9% of the population aged 2 years and above who had never suffered from VL were rK39-positive; 6.2% were DAT-positive. There was limited overlap between both markers (kappa = 0.30), which is similar to what was observed in Ethiopia by Gadisa et al [26] and by Custodio et al [27].
Despite the observed increase with age, DAT and rK39 sero-positivity are temporary. Fifty nine percent of rK39 positives and 33% of DAT positives in the first survey round had lost their antibodies at the time of the second survey; similar observations were reported in other studies [28], [29], [30], [13], [14]. The discrepancy between the prevalence of the two serologic tests depends on the thresholds chosen but also on the relative rates of acquisition and loss of the markers. Among recent ex-VL cases we observed much higher titers than among subclinically infected but in this group too DAT titers appeared to be more stable than rK39 titers. Whereas the rK39 response appears to decrease over the time followed in this survey, the DAT response remained fairly stable over the 2½ year period of observation (figure 4). Notably, when the interval between disease episode and survey approached 2 ½ years, rK39 titers had fallen to levels similar to subclinical cases, whereas DAT titers remained much higher than subclinical subjects' titers (figures 3 and 4). The data suggest that the polyclonal response to the DAT total parasite antigen lingers longer than the monoclonal response to rK39. As such, one would predict the most likely time to find concordance between these serologic tests would be during acute infection, whether symptomatic or latent.
Although symptomatic VL was most common among young adults (age group 20–29 years) and became increasingly rare in the older age groups in our study, the frequency of both rK39- and DAT-positivity increased with age. We also observed an increase with age of the probability of sero-conversion on either DAT or rK39 between the first and second surveys.
An increase in leishmania sero-prevalence with age has been reported in other studies of L. donovani as well as L. infantum [31], [32], [33]. Although most active VL cases occur in younger age groups, infection with Leishmania spp. is assumed to persist for life [34]. Sero-positivity however is a time-limited phenomenon. The observation that nevertheless sero-prevalence and the frequency of sero-conversion increased with age could be explained by individuals experiencing repeated inoculations with infected sand flies, of which the early ones are more prone to lead to disease.
Hailu et al. [35] in Ethiopia report similar observations for LST, which reflects cellular immunity and is generally assumed to remain positive for many years or for life. The prevalence of LST positivity also increased with age, but individual subjects converted and reconverted in-between 7 rounds of sero-survey. Subjects were also tested with DAT and though prevalence levels were lower, DAT-positivity also fluctuated between surveys. Hailu et al. conclude that maintaining DAT or LST positivity requires continued exposure to L. donovani. Individuals may thus revert back to DAT negativity but rapidly reconvert to DAT positive on renewed exposure [35].
In our study population, the day-to-day intensity of exposure to L. donovani certainly varies during the year and between years [36], but at any given moment there is probably not much difference in the level of exposure between different age groups present in the villages. Older people may also have experienced periods of higher transmission in the past than younger individuals. However, the most likely explanation for the age-pattern we observed is that as people get older, their chances of ever having been infected increase. Upon renewed exposure they will more readily convert to a seropositive state in comparison to those that have never been infected before.
Another possible explanation for fluctuations in sero-positivity in asymptomatic individuals over time would be the occasional proliferation of parasites contained in safe target cells, i.e. cells that are not able to exert anti-parasite activities such as fibroblasts [37], [34]. Intermittent proliferation of parasites from such cells could lead to repetitive antigenic stimulation of the innate and adaptive immune system [38].
Assuming that, as also suggested by the observations of Hailu et al. [35], repeated sero-conversions are mainly due to repeated exposure, sero-conversion is probably an adequate measure of (re)infection. Sero-conversion has been used to evaluate public health interventions in Iran [39] and in India and Nepal where L. donovani sero-conversion and clinical VL results were correlated [40]. Infection is a necessary step in development of clinical VL, but it is not sufficient in itself [41]. Host factors also play a role, and individuals prone to develop disease are probably more likely to develop VL when first infected, at an early age.
Whichever the mechanism behind the increase in sero-prevalence and sero-conversion with age, the consequence is that in cross sectional surveys or in cohort studies that are too widely spaced, a substantial number of subclinically infected persons will turn out (false) negative on serology. This has implications for epidemiological studies and could have implications for control strategies as well. Though so far there has been little evidence for a role of asymptomatic infection in transmission of VL, more definitive measures of asymptomatic infection are needed before any firm conclusions can be drawn.
In highly VL endemic villages in Bihar, India, substantial portions of the population react positive to DAT and/or rK39 suggesting they have asymptomatic infection. The proportion of seropositive individuals increased with increasing age. This pattern could be explained by a boosting effect in asymptomatically infected persons upon repeated exposure to the parasite or by intermittent internal release of parasites. Either mechanism could result in misclassification of asymptomatically infected persons in epidemiological surveys. Further follow-up is required to elucidate the significance of the sero-positive states according to both tests as markers of infection with L. donovani.
|
10.1371/journal.pntd.0006117 | Genotyping of Mycobacterium leprae for better understanding of leprosy transmission in Fortaleza, Northeastern Brazil | Leprosy is endemic in large part of Brazil with 28,761 new patients in 2015, the second largest number worldwide and reaches 9/10.000 in highly endemic regions and 2.7/10.000 in the city of Fortaleza, Ceará, Northeast Brazil. For better understanding of risk factors for leprosy transmission, we conducted an epidemiologic study supplemented by 17 locus VNTR and SNP 1–4 typing of Mycobacterium leprae in skin biopsy samples from new multibacillary (MB) patients diagnosed at a reference center in 2009 and 2010. Among the 1,519 new patients detected during the study period, 998 (65.7%) were MB and we performed DNA extraction and genotyping on 160 skin biopsy samples, resulting in 159 (16%) good multilocus VNTR types. Thirty-eight of these patients also provided VNTR types from M. leprae in nasal swabs. The SNP-Type was obtained for 157 patients and 87% were of type 4. Upon consideration all VNTR markers, 156 different genotypes and three pairs with identical genotypes were observed; no epidemiologic relation could be observed between individuals in these pairs. Considerable variability in differentiating index (DI) was observed between the different markers and the four with highest DI [(AT)15, (TA)18, (AT)17 and (GAA)21] frequently demonstrated differences in copy number when comparing genotypes from both type of samples. Excluding these markers from analysis resulted in 83 genotypes, 20 of which included 96 of the patients (60.3%). These clusters were composed of two (n = 8), three (n = 6), four (n = 1), five (n = 2), six (n = 1), 19 (n = 1) and 23 (n = 23) individuals and suggests that recent transmission is contributing to the maintenance of leprosy in Fortaleza. When comparing epidemiological and clinical variables among patients within clustered or with unique M. leprae genotypes, a positive bacterial index in skin biopsies and knowledge of working with someone with the disease were significantly associated with clustering. A tendency to belong to a cluster was observed with later notification of disease (mean value of 3.4 months) and having disability grade 2. A tendency for lack of clustering was observed for patients who reported to have lived with another leprosy case but this might be due to lack of inclusion of household contacts in the study. Although clusters were spread over the city, kernel analysis revealed that some of the patients belonging to the two major clusters were spatially related to some neighborhoods that report poverty and high disease incidence in children. Finally, inclusion of genotypes from nasal swabs might be warranted. A major limitation of the study is that sample size of 160 patients from a two year period represents only 15% of the new patients and this could have weakened statistical outcomes. This is the first molecular epidemiology study of leprosy in Brazil and although the high clustering level suggests that recent transmission is the major cause of disease in Fortaleza; the existence of two large clusters needs further investigation.
| Leprosy is a transmissible disease that is still endemic in several countries including in Brazil, a country with highly variable region associated incidence of disease. Fortaleza is a city in Northeast Brazil with high incidence and conventional epidemiology studies are suggestive for high levels of recent transmission. Genotyping of M. leprae allows the recognition of individuals that have been infected with the same strain (called a cluster) and therefore being suggestive for belonging to the same transmission network. In the present work, by analyzing genotypes of M. leprae in the skin lesion of multibacillary patients, we made observations that improve our knowledge on interpretation of genotypes and clusters and confirm the high levels of recent transmission in the city of Fortaleza. This is one of the few studies that used molecular epidemiology to look for risk factors for recent transmission of leprosy and to our knowledge, the first in Brazil. Our data support further investigation of the workplace as a source of infection, preferentially by a study designed on a larger number of patients and including analysis of M. leprae present in the nose.
| Leprosy, caused by infection with Mycobacterium leprae remains a significant public health problem in many developing countries. The disease presents a wide spectrum of clinicopathologic forms that ranges from tuberculoid leprosy (TT) to borderline forms and lepromatous leprosy (LL) and lesions involve skin and peripheral nerves. Disease can be paucibacillary (PB) or multibacillary (MB) with the most severe LL form involving organs such as liver, spleen and bone marrow and the bacterial burden in such patients is massive and causes severe deformities when not treated. Multi-drug therapy using dapsone, rifampicin and clofazimine was implemented in the 1980s and has considerably reduced disease prevalence, but that is not the case with incidence, implying that leprosy is still being transmitted to a considerable extent [1]
As M. leprae cannot be cultured on artificial media, molecular techniques have been used for better characterization of the organism [2, 3, 4, 5], including the deciphering of the genome sequence [6]. Single nucleotide polymorphisms (SNPs) analysis allowed studies on phylogeography of leprosy, evolving models for the global spread of M. leprae [7,8]. Besides SNPs, Variable Number Tandem Repeats (VNTRs) are used for genotyping and a certain relation between the number of certain VNTR alleles and SNP-Type has been observed [9], showing that VNTR typing adds to our knowledge on spread of leprosy. Multiple-locus variable number tandem repeats analysis (MLVA) of a set of micro- and mini-satellites of M. leprae is a fingerprinting procedure for differentiation at the strain level [10, 11, 12, 13] and useful during transmission studies, to distinguish reactivation from re-infection [14] and to study bacterial population structure on different levels and countries, as described for Brazil [15,16], China [9, 17, 18, 19], India [20, 21, 22, 23, 24], Philippines [25, 26], Thailand [27, 28], Mexico [29], Colombia [30, 31] and the United States [32].
Fortaleza is the capital of Ceará, a state located in northeastern Brazil. In 2015, 80.5% of the 184 municipalities in the state diagnosed new patients of leprosy and 10% were classified as being hyperendemic, defined by having an incidence of higher that 4/10.000. Ceará is one of the poorest regions of the country, reporting 1.743 new leprosy patients in the same year, including 528 in the capital, representing an incidence rate of 2.7/10,000 inhabitants. MB is detected in two thirds of these patients and 5.9% of the total patients reported in the state are younger than 15 years of age, both of which are indicators of ongoing and recent transmission seems of the disease [33].
Previous genotyping of M. leprae strains in Brazil, from a set of unrelated patients from the Southeast region of the country demonstrated a high VNTR based genetic variability in predomintly SNP-Type 3 background [15]. Later, it was observed that SNP-Type 4 is much more frequent in the North-northeast part of the country [6]. Although preliminary data on use of genotyping to add to transmission studies have been presented in Ceará [34], Mato Grosso [35] and Pará [36], no full reports exist on molecular epidemiology studies of leprosy and the rearch for risk factors for recent transmission in Brazil; therefore our study addresses this gap.
Fortaleza is the capital and also largest city of the state of Ceará, and the fifth largest city (314,930 km2) in Brazil with 2,627,482 inhabitants in 2017. It has 120 neighborhoods and the highest population density among the country's capitals. Although Fortaleza has the tenth highest GDP in the country and the highest in the Northeast region, it has the typical uneven distribution of wealth observed in most of Brazil’s major cities. Besides being an important industrial and commercial center, it is the second most desired tourist destination in Brazil and fourth in number of visitors [37, 38].
This study was designed to better understand the clinical and epidemiological characteristics of leprosy in the city of Fortaleza. A cross-sectional study was conducted from November 2008 to December 2010 and during this period, all new leprosy patients diagnosed by trained dermatologists of the National Reference Center of Dermatology Dona Libânia (CDERM) were invited to participate in the study. This tertiary reference center serves about 80% of the almost 800 new leprosy patients diagnosed annually in Fortaleza and is the most important reference center for skin disease, including leprosy, in that city [39].
Patients were diagnosed by clinical evaluation; microscopic evaluation of bacillary index of acid fast bacteria in slit skin smears (SSS) analysis and histopathological evaluation of biopsy specimens. Patients were classified according to Ridley-Jopling criteria based on histological study and bacterial indices (BI) [40]. All new patients responded to a detailed questionnaire that included demographic, epidemiologic, socioeconomic, environmental and behavioral components. In addition to the questionnaire, data for the patients were introduced and maintained by registered health workers in the SINAN database (http://portalsinan.saude.gov.br). A second skin biopsy and nasal swab was collected for genotyping of M. leprae in a subset of all diagnosed patients.
The skin biopsy samples were collected using a 5 mm punch. Tissue for histopathology was treated with formol and embedded in paraffin while the tissue for genotyping was placed in a sterile 1.5 mL tube and stored at -20°C. The DNA was extracted by using the DNeasy Blood & Tissue kit (Qiagen Biotecnologia do Brasil Ltda, SP, Brazil) following the manufacturer's guidelines.
Nasal swabs were collected from patients who also provided a second skin biopsy for genotyping, by gently rubbing a swab previously wetted with Tris-EDTA buffer (pH 8.0), in one side of each nostril over the lateral conchae. After collection, each swab was immersed in a sterile and labeled tube and stored at -20°C until processing as described by Lima et al. [41]
Genotyping by MLVA of 17 VNTRs was performed as described by Kimura et al. [13] and based on four multiplex PCRs that generated 17 amplicons. The allele for each VNTR locus is the copy number of the repeats which was determined by denaturation of amplicons and capillary gel electrophoresis on the sequencer ABI 3130 Genetic Analyzer, using the internal molecular weight sizing standards (LIZ 500). The copy number of each locus was calculated based on the size of the PCR amplicon using the Peak Scanner software (Applied Biosystems do Brasil) and comparing to previously calibrated M. leprae strain NHDP63. To study reproducibility of the assay, DNA from five M. leprae samples from Brazil was sent to CSU for comparative analysis of the alleles.
For differentiation of four genotypes of M. leprae based on three SNPs, we used a procedure that combined PCR-restriction enzyme analysis (REA) and direct sequencing as described by Sakamuri et al. [26]. Differentiation of genotypes 1/2 from 3/4 was obtained by submitting to BstUI mediated PCR-RFLP analysis of the locus at nucleotide position 2,935,685; digestion occurs in case of genotype 3/4 and lack of digestion for genotype 1/2. Differentiation of genotypes 3 and 4 is obtained by SmlI mediated PCR-RFLP at nucleotide position 14,676; digestion indicates SNP-Type 4 and lack of SNP-Type 3. Differentiation of SNP-Type 1 or 2 was performed by direct PCR sequencing as described by Monot et al. [7].
The copy number of all alleles were introduced into Microsoft Excel files and imported into Bionumerics software (version 7.6; Applied Maths; Sint Martens Latem, Belgium).
Definition of clustering was based on comparison of the copy number of the VNTRs using two different stringencies: either considering those that presented identical copy number for all 17 alleles, or considering those that had identical copy number in 13 alleles, excluding the four most variable loci. A similarity matrix was constructed using the categorical similarity coefficient and the unweighted pair group method with arithmetic mean (UPGMA). This was the basis for a complete linkage tree, a circular top score UPGMA tree and a range of minimum spanning trees (MST).
The cartographic bases and the population used were obtained from the Brazilian Institute of Geography and Statistics (http://www.ibge.gov.br/). The coordinates were obtained using a global positioning system (GPS) and stored in a geographic database (BDGeo). Data were used to generate graphics, satellite imagery processing, to establish topological relations between the graphic elements and their attributes, spatial analysis and visualization through thematic maps. We evaluated the spatial analysis Kernel density estimation (KDE) using a fixed radius of 2 km. Analyses were performed in ArcGis (http://www.esri.com/) and TerraView (http://www.dpi.inpe.br/menu/Projetos/terraview.php). In TerraView it was possible to build a dual Kernel or Kernel ratio, based on the number of patients and the population [42]. We used the interpolator points by Inverse Distance Weighting (IDW) to estimate the cell values using a weighted linear combination of a set of sampling points. The satellite image in Fig 1 was generated using the sensor Sentinel 2 of the European Space Agency (ESA) (https://sentinel.esa.int/web/sentinel/user-guides/sentinel-2-msi) with Open Access CC-BY License (http://open.esa.int/).
For evaluation of the association of the demographic, clinical and environmental/behavior variables and having a clustered or a unique M. leprae genotype, chi squared and Fisher exact tests were used. Mann-WhitneyU test was used for evaluation of differences between a single characteristic in individuals with clustered genotypes or unique patterns.
An informed consent form was signed by the participants of the study, authorizing the collection of clinical samples. The present study was approved by the Ethics Committee of CDERM and the national ethical committee.
At CDERM, 830 (284 PB and 546 MB) and 689 (237 PB and 452 MB) new leprosy patients were diagnosed respectively in 2009 in 2010, totaling 1519 in the study period and among these, 998 were MB patients (65.7%). Recruitment was conducted only on two days per week, which was further reduced in December, January and July and on holidays. This resulted in the collection of a second biopsy specimen for genotyping from 301 MB patients only of whom we received 160 (only 92 from 2009 and 68 from 2010). This resulted in M. leprae genotyping of 16.8% and 15% of the newly diagnosed MB patients respectively in 2009 and 2010. From these 160 patients, 101 also had nasal swab collected.
Because the questionnaire was developed for a larger case-control study evaluating risk factors for leprosy and the patients within the study presented here only partially overlapped with the larger study, we accessed data from the SINAN database (http://portalsinan.saude.gov.br) for 61 of the 159 patients (31%).
Of the 160 patients biopsies submitted to M. leprae genotyping, 159 yielded high quality MLVA-based and 157 SNP-based genotypes and are presented in S1 Table and S1 Fig. Initially, 134 M. leprae were defined as SNP-Type 4 (85%), 15 as SNP-Type 3, three as SNP-Type 1 and six samples could be characterized only to the SNP-Type 1 or 2 level because of insufficient material for sequencing. Four isolates with SNP-Type 3 were grouped within the MLVA-based clusters of isolates with SNP type 4 so we suspected wrong classification due to partial digestion during PCR-RFLP. Two samples had sufficient material left to repeat and both were indeed confirmed as being SNP-Type 4, resulting in 136 SNP-Type 4 (86%) and 13 (8.2%) SNP-Type 3.
The MLVA-based typing results are presented in S1 Table and all but eight strains yielded the complete 17 locus-based genotypes (95%), five isolates failed in the amplification of one locus while another two lacked alleles for five and six alleles respectively. The latter strain was clustered with another but not included in the analysis of clusters at high cluster stringency. Two isolates presented a double peak, one at (TA)18 and another at (AT)15 and these alleles were not considered for analysis. The differentiating power and allele distribution of the satellites is presented in Table 1, and varied between 0 and 0.93, and 1 and 22, respectively. Three markers–[(GGT)5, 6–3 and 21–3] were invariable, four [(AT)15, (TA)18, (AT)17 and (GAA)21] were highly discriminatory with a Hunter-Gaston discriminatory index (HGDI) above 0.8, and the rest had a HGDI of less than 0.8. Interestingly, the copy number distribution pattern of 18–8 is different from that of the other markers and presented a bimodal pattern (Table 1).
Regarding cluster analysis, when using the highest stringency including 17 markers, we observed 157 different genotypes formed by 154 singletons and three clusters of two patients each, resulting in an overall cluster level of 3.8% (6/157). Upon analysis of the data of the patients within each of the three clusters and or of those with unique genotypes, no particular risk factor for belonging to one of or any cluster was identified.
However, when excluding the four markers with HGDI > 0.8, 83 different genotypes were detected, 63 unique ones and another 20 found in 96 patients, resulting in a cluster rate of 60.4% (96/159). The two largest clusters were composed of 23 and 19 patients and the remaining clusters were composed of one of six, two of five, one of four, six of three and eight of two patients. Again, no clear patient characteristic was detected that could explain the formation of an individual cluster but when analyzing the data of those belonging to a genotype cluster or not, some significant associations and tendencies for clustering were observed. We observed a significant association of clustering being BI positive (p = 0.037) or having worked (p = 0.049) with someone with leprosy (note: working together has p = 0.25). Surprisingly, the variable ‘having lived with someone who had leprosy’ demonstrated an inverse relation with clustering (p = 0.065). Although not significant, an association was observed between clustering and disability (p = 0.445), mainly because of a tendency to have more grade 2 disability among clustered patients (13.1% against 5.1%) and longer time between observing first lesion and diagnosis/disease notification (p = 0.14). Another unusual finding was that alcohol consumption was significantly associated with non-clustering, i.e., of having unique genotypes (p = 0.047) (Table 2). Note however that some of these associations occurred with the number of patients for some categories being < 5 and a detailed relation between clustering and variables is presented in S2 Table.
We also performed chi-square analysis of patients and other characteristics of the 18 patients from cluster 12 and 23 patients from cluster 14 (totalling 41 among 159 = 25.8% in cluster) and observed no significant association of the clustered cases when compared to the rest of any of the variables. Additionally, we plotted the date of diagnosis of the patients, clustered cases and those belonging to the two major clusters (12 and 14) on a monthly based time scale of the study period and although some higher frequency of diagnosis was observed between March and June of 2009, no particular independent increase in clustering was observed during the study period (S2 Fig).
Among 38 patients, MLVA patterns were also available from nasal swabs, with the exclusion of alleles 6–3 and 18–8, not performed in this sample type and as described recently, difference in copy number of the alleles with highest DI was observed in the M. leprae genotypes when comparing both samples in a considerable number of patients [43]. Upon inclusion of the genotypes of M. leprae present in nasal swabs in the analysis, we observed that eight genotypes from nasal swab were part of some cluster, increasing the number of clustered patients by 10. One cluster with a genotype shared by M. leprae in skin biopsy of five patients increased to eight when considering genotypes in nasal swab while three new clusters were observed, one composed of the genotype observed in the nasal swab of two patients, and two others composed of a genotype that was observed in the nasal swab and skin biopsy of two patients each (S3 Table).
Data to perform spatial analysis was available for 156 patients and demonstrated clearly a higher density in the western part of Fortaleza (S3A Fig). The Figure also presents the number of patients per neighborhood (S3B Fig), the population estimated by the 2010 census by neighborhood (S3C Fig), KDE using georeferenced homes of patients (S3D Fig), KDE using neighborhood (S3E Fig) and dual Kernel using neighborhood (S3F Fig) in Fortaleza. The neighborhood with the highest number of patients is Granja Lisboa. The result obtained by using KDE in the neighborhoods showed, as expected, the same clusters, both centered in the neighborhoods Bom Jardim, Bonsucesso, Granja Lisboa and Granja Portugal (S3D and S3E Fig). However, when applying dual Kernel analysis, two clusters are observed, one including Granja Lisboa and Siqueira (southwest region), the same as observed using Kernel (neighborhood), and the second centered in the neighborhood Jacareacanga in northern Fortaleza.
The distribution of the patients among the non-clustered patients and for each cluster across the neighborhoods of the city is presented in S4 Table and demonstrates that nine groups with clustered genotypes had at least two patients in the same neighborhood (groups 2, 7, 8, 9, 12, 14, 16, 17 and 19). Overall, patients with the same M. leprae genotype are spread across the city, except for the biggest cluster 14 showing two pairs of two very nearby patients (Fig 1). When performing KDE analysis using a distance of 2 km concentrating on the distribution of the patients from the two largest clusters, association was observed with some neighborhoods. The cluster formed by 19 patients was associated with Jacareacanga, Canindezinho, Conjunto Esperança and Manoel Sátiro, while those of the cluster with 23 isolates with Bonsucesso and Vila Pery (Fig 2). However, when performing the same type of analysis with the 62 patients with unique genotypes, we observed association with Granja Lisboa, Granja Portugal and Bom Jardim. Note however that the number of patients with unique patterns is about three times higher than those in each of the two biggest clusters.
Finally, we plotted distribution of patients and performed spatial analysis according the number of lesions and number of bacilli observed by bacilloscopy (S4 Fig). The number of lesions varied between 0 and 88 (total of 2120, medium value 13.6 and standard deviation of 13.2) and bacterial indices were between 0 and 6+ (total 421, median value 2.7 and SD 1.99). Although we observed that patients with high number of lesions or high bacillary load were spread over the city, two neighborhoods that were associated with cluster 12 (Canindezinho and Conjunto Esperança) and 14 (Bonsucesso) presented patients with high BAAR.
When UPGMA based dendrograms including all 17 satellites and with or without including SNP-Type were constructed, most isolates belonged to two major groups. The isolates that were not of the SNP-Type 4 were observed at the outer limits of the tree (S1 Fig). For evaluation of the bacteriological population structure and the influence of inclusion of loci on cluster formation and tree topography, we constructed a MST including either all 17 satellites or gradually removing the most variable ones. As observed in organizing the allele number in an Excel file, the same three clusters of genotype pairs were observed in the MST when including all 17 loci, and the 20 clusters when omitting the four most variable markers (Fig 3). Depending on the number of VNTRs included for MST construction, we observed either two or three major groups and gradually excluding VNTRs with the highest variability, we observed that AC9 or/and AC8b are the main drivers for maintaining separate groups; omitting these markers resulted in a population with a large central cluster of 88 isolates with 11 branches. Most of the isolates have indeed a 6- or 7-copy number of these alleles and leaving out these markers coincides with the observation of clusters formed by different SNP-Types (S5 Fig).
In 1991, the WHO adopted a resolution for elimination of leprosy by the year 2000 and implementation of MDT resulted in a significant reduction of prevalence. Between 2002 and 2012, a 65% reduction in the prevalence (from 4.33 to 1.51 patients/10,000) was achieved in Brazil but leprosy is unevenly distributed within the country with pockets of incidence levels of more than 10/10,000 [44]. The Northeast region is the poorest of the country reporting a third of the newly diagnosed patients and a detection rate that is twice that of the average in the country and the State of Ceará is one of the poorest states in the region. Over 10% of its municipalities classified as hyperendemic and the capital, Fortaleza, considered a priority for leprosy control, having the highest demographic density in the country and one of the municipalities in the state with the highest detection rates [45]. In addition, 5.9% of the new patients are less than 15 years of age and only half of the contacts are being investigated for disease [33].
Transmission of leprosy is assumed to be from person to person through the respiratory system or damaged skin, with risk for developing disease being higher if a family member had disease and even more when these presenting the LL form [46, 47]. However, new patients often mention lack of contact with other leprosy patients, suggestive of unrecognized transmission routes [48], including exposure to an environmental source such as water, soil, plants and animals [49] but no study unequivocally demonstrated the mechanism of leprosy transmission [50].
Since the report on the existence of genetic variability [4, 5] and of the genome sequence of M. leprae [6], analysis of SNP-Types and micro- and mini-satellites added to our knowledge about genetic variability of M. leprae and its biology, such as existence of geographic or family associated genotypes [18, 19, 23], genetic divergence between bacilli inhabiting different tissue [20] and differentiation between relapse and re-infection [14]. Although studies on genetic variability of M. leprae have been conducted in several regions endemic for leprosy, mostly detailed epidemiologic information is missing except for a study in Qiubei, China, demonstrating intra-familial strain types [19] and regional differences in clustering [18]. No prospective molecular epidemiology study with detailed epidemiologic and clinical data have been reported except for a study reporting transmission of dapsone resistant M. leprae in Cebu, the Phillipines [51].
We hereby confirm the high prevalence of SNP-Type 4 in the northeast of Brazil as reported previously [16] and probably due to introduction of leprosy by slave traffic from West Africa. Isolates with SNP-Type 3 are partly 3I, as defined by the gyrA97 SNP (SNP7614) [52]; and our earlier observation during studies on drug resistance [14, 53]. We also observed a surprising strong correlation between SNP- based and VNTR based genotypes suggesting that in certain populations, microsatellites are also deeply rooted into the bacterial population structure. Only by omitting GTA9 and AC8a from the analysis, the relation between VNTR and SNP-Type was disrupted. Association between certain VNTRs and SNP-Type has been demonstrated before [15, 31] but might be more pronounced here due to the very high level of SNP-Type 4 in our study population. Because of the high level of SNP-Type 4 in the studied population, it would have been interesting to characterize the M. leprae isolates to the sub-SNP-Type level but no DNA was left to perform this.
The influence of stringency of definition of genotype clustering for interpretation of transmission and phylogeny has been clearly demonstrated for tuberculosis [54] but not extensively for leprosy [26]. The difference in clustering level using two stringencies in the present study is remarkable (3.8% vs. 60.4%) and we believe that the high clustering level represents recent transmission and therefore being the major drive for developing leprosy in Fortaleza. Although clusters are generally small, we also observed two larger ones and clusters of considerable size have also been described in China [18], in the Philippines [26, 27] and among those shared between humans and armadillos in the US [32]. The choice of stringency for definition of clustering in the present study is partly based on the fact that the four markers with Simpson Index >0.85 also were mostly presenting allele differences in the genotypes of M. leprae present in the nose and in the skin. Those were also among those disfavoring MLVA analysis for M. leprae genotyping as described by Monot et al. [8]. One weakness of our study is that we have no epidemiologic links that proovef that the 13 VNTR-based clustering is indicative for intense leprosy transmission in the present setting but this is probably due to lack of healthy household contacts (HHC) in the present sampling and low representativity of sampling. Extensive MIRU-VNTR genotyping data from M. tuberculosis show that the most variable MIRUs can be omitted without much loss of transmission links [54]. The number of M. leprae bacilli in the human body can reach 1012 so differences in copy number due to higher number of replication cycles during development of leprosy are imaginable. Finally, the presently used VNTR-based stringency is still higher than that those used by Sharma et al. [32] and Lavania et al. [24]. Sharma et al. related the SNP-VNTR type 3I-2-v1 genotype among 80.3% of the armadillo samples from the South of the US and 22/52 human patients were infected with M. leprae presenting one of two major genotypes. Interestingly, Lavania et al. [24], using an identical typing approach observed 66 different patterns among 70 leprosy patients. Although sample representativity and other variables might strongly influence clustering levels, the difference between both studies is striking and might also be due to differences in transmission dynamics. Some markers that were included [(TA)10 and 18–8] were not used for genotype definition in the before mentioned studies but in the present study had a HGDI of 0.38 and 0.22, respectively. This again suggests the need for regional evaluation of VNTRs for local M. leprae genotyping for developing "lower cost" genotyping in the mostly poorer endemic regions. However, having in mind the huge amount of information obtained from the standardized 24-MIRU-VNTR procedures for phylogenetic studies of M. tuberculosis, we here suggest the use of 17 STRs or even more for better understanding of transmission and phylogeny of M. leprae on a larger scale.
The comparison of M. leprae genotypes present in skin biopsy and nasal secretion is described and discussed in detail elsewhere [43]. While all isolates presently presented four copies of (GGT)5, one nasal swab sample presented six copies of this allele [43] and although other alleles than that of four copies are described with very low frequency in Brazil [16], they are more frequent in countries like Thailand [27] and the Philippines [25]. Contrary to the single allele with two copies of 23–3 described by Lima et al. [43], in 8% of our patients, a single copy of this marker was observed. A further finding by Lima et al. was the observation that some individuals presented differences in copy number in five to seven loci, including less variable ones, being highly suggestive for multiple infection or more extensive intra-patient strain evolution. In addition and more importantly for transmission studies, our data show that inclusion of the genotypes from nasal swabs may have consequences for clustering outcome. Because the hypothesis is that the nose is a port of entry and exit of M. leprae, the genotype in nasal swabs could contribute to the transmission links suggested by genotyping M. leprae in skin biopsies. We therefore suggest that more studies including both samples are needed to understand transmission dynamics. However, as stated elsewhere, there is no guarantee that M. leprae in the nasal swab is representative for disease but very recently, molecular evidence for an important role of the nose in leprosy transmission was presented by Araujo et al. [55].
High levels of recent transmission in Fortaleza is also evidenced by the observation of two large clusters of about 20 patients and may indicate the existence of two main lineages of M. leprae strains differing in four alleles (AC8b, GTA9, AC9 and AC8a) in Fortaleza. This might be related to some undetected factor causing more transmission of these strains but unfortunately, our study did not allow their definition and might depend on a social network approach as demonstrated in molecular epidemiology studies of tuberculosis [56]. Alternatively, these strains might have higher transmissibility, undescribed so far in leprosy but proven for some lineages of M. tuberculosis. Our earlier observation that reinfection or strain selection of M. leprae isolates of SNP-Type 4 was very frequent in relapse patients in Rio de Janeiro, a region predominant for SNP-Type 3 could be an example of that [14].
Identifying behavioral and environmental risk factors for developing leprosy is a difficult task because of the long incubation time of the disease (2–5 years for tuberculoid leprosy and 8–12 years for lepromatous leprosy). It is not easy to determine time and duration of exposure and onset of infection and risk factors for disease might change over time. Among 165 municipalities in the state of Ceará, a 300-fold difference in disease incidence was observed and associated with poverty, inequality, uncontrolled urbanization, population growth and low level of education [57]. The same group [44] also looked for socioeconomic, environmental and behavioral factors associated with leprosy in a case control study in four municipalities including that of Fortaleza; low education level, experience of food shortage at any time in life, frequent contact with natural bodies of water and infrequent changing of bed linen were associated with leprosy. Another study in this city concentrated on infection with M. leprae in the absence of clinical disease and demonstrated that higher levels of anti PGL-1 in patients without known contact with leprosy patients are much higher than reported elsewhere in the literature [58]. More recently, nasal carriage of M. leprae by PCR was observed in 67% of HHC but interestingly, 28% of persons living in richer part of the city were also positive. This is probably due to complex interaction between the populations at high and low risk for infection by leprosy. Domestic service and daily migration of the poor in houses of the upper class and richer parts of the city is still common [41].
An earlier spatial analysis in Ceará showed the highest density of disease is among the most urbanized and economically highest developed [59]. Our spatial analysis on genotype distribution did not demonstrate a distribution of clustering that was different from disease distribution in Fortaleza in general, showing that with the present data, there do not seem to be clear hot spots of (recent) transmission in the city. However, some neighborhoods were associated with the two biggest clusters, being group 12 (Jacareacanga, Canindezinho, Conjunto Esperança and Manoel Sátiro) and group 14 (Bonsucesso and Vila Pery). We also observed that three of these neighborhoods (Bonsucesso, Canindezinho and Conjunto Esperança) presented patients with high BI (note that only MB cases were submitted to genotyping) and in a recent study on the social, educational and economic development of neighborhoods in Fortaleza, both were indicated as being among the poorest in the city (www.ipece.ce.gov.br/publicacoes/Perfil%20Socioeconomico%20Fortaleza%20final-email.pdf). In addition, very recent data also demonstrate that both neighborhoods are hyperendemic (> 4/10.000) for leprosy with high incidence in children less than 15 years of age (0.5-1/10.000) [60].
Some limitations of our study is that our sampling occurred during a relatively short period of time, that genotyping was performed only on 15% of the new MB patients and that PB patients were omitted from analysis. This might mask transmission links due to factors other than contact with MB patients and explain why a considerable proportion of the new patients were not aware of earlier contact with patients. Nonetheless, the most significant association with clustering was having positive bacilloscopy, which is in agreement with the long standing idea that transmission of leprosy is caused by close contact with MB patients. However, significance of this finding is weakened because the mean BI between groups with clustered and unique genotypes is almost the same, but again, only MB patients were submitted to genotyping. Definition of being MB or PB in the present study is based on Ridley-Jopling method and our results are in favor for maintaining this technique as part of the diagnostic procedure, contrary to the current recommendation of WHO to define PB and MB patients only on basis of number of lesions and nerve involvement.
The significant association of clustering with patients having had contact with another case at work but not at time of diagnosis present could be due to the long incubation time for developing leprosy; however, a low number of patients reported contact at work. Although we could not establish a relation of cluster with the nature of the work or localization of the workplace, this needs further investigation because some working places harbor a large number of persons including undetected leprosy cases during long periods and could be hot spots of transmission. Some examples are metallurgic and car assembly factories, areas of civil construction, handicraft fairs and offices. Social interactions and the physical, residential and occupational environments have been suggested to be more conducive to transmission of a community in Qiubei, China [18]. This finding is not in line with our observation that having lived with a leprosy patient is associated with belonging to a non-cluster and to explain this, further research, eventually using whole genome sequencing is warrented.
HHC have been described to be at higher risk for developing leprosy in several conventional epidemiologic studies but also in studies that performed M. leprae genotyping, including China [17], Thailand [27], Colombia [31] and India [24]. Although investigation of HHC is part of the leprosy program in Brazil, this is not always being performed and in Fortaleza in particular, this seems to be the case in about 50% of the patients [43]. The lack of association between clustering and house hold in the present study is probably due to the inclusion of new patients only and without contact investigation and inclusion of patients from the same house hold. Nonetheless, our observation of inversed association of sharing home with a leprosy case and cluster is surprising and needs to be better investigated.
Another puzzling finding was the significant association between alcohol use and having M. leprae with a unique genotype. Several studies associated alcohol (ab)use as a risk factor for leprosy, including a case control study in Mato Grosso state [61], Maranhao state [62] and with treatment abandonment in Tocantins [63]. This finding needs further investigation but again, the low number of patients in some analytical cells due to the paucity of biopsied patients and lack of specific questionnaire data could be partly responsible. Another issue are the different protocols used for collecting information about alcohol (ab)use.
We also observed that some characteristics that are usually associated with higher risk for leprosy also had a tendency to be more pronounced in clustered patients. This was the case of clustering among males and later diagnosis at a later stage due to more reluctance to seek care among men as widely in Brazil. We also observed a tendency to have a higher disability grade in clustered patients. Higher disability grade reflects longer incubation time, bacillary load and time before diagnosis, therefore being able to infect more individuals. This is in concordance with the longer time delay between first observation of lesions and disease diagnosis reported in clustered patients.
We conclude by referring to a very recent study that evaluated temporal trends in leprosy in Fortaleza for the period 2001 to 2012 [59]. Although there was a steady decrease in the number of new patients, from hyperendemic (≥4/10,000) in 2001 to highly endemic (2<4/10,000) in 2012, the number of new patients in children less than 15 years old was steady and there was also noted a steady increase in the number of MB and of lepromatous patients since 2005. Such data indicate both ongoing recent transmission including to children and late diagnosis in adults, reflected also by the rise in grade 2 disability (from 6% to 9% in new patients). Given the chronic nature and natural history of the disease it is unlikely that there will be an improvement of these trends in the near future. Low levels of education, unfavorable socioeconomic conditions, and delayed presentation to the health system are factors that are generally associated with late diagnosis. This is in agreement with our data of high clustering levels and, demonstrating that recent transmission of leprosy is a serious problem in Fortaleza. The realization of a prospective molecular epidemiologic study in a complex setting like Fortaleza is difficult but we hope that a new study of longer duration, with higher intake of patients, collecting both skin biopsy and nasal swabs or biopsy, inclusion of HHC, a more detailed questionnaire including social network studies that might allow definition of risk factors for belonging to the same cluster, and finally investment in DNA extraction and more sensitive genotyping that allows inclusion of PB patients. As a final comment, we believe that, although whole genome sequencing of M. leprae genomes is still challenging because of the need of bacterial DNA enrichment, the technical expertise needed and the considerable cost, inclusion in future studies might be beneficial for better understanding of leprosy transmission.
|
10.1371/journal.pmed.1002887 | Impact of reduced dose of ready-to-use therapeutic foods in children with uncomplicated severe acute malnutrition: A randomised non-inferiority trial in Burkina Faso | Children with uncomplicated severe acute malnutrition (SAM) are treated at home with ready-to-use therapeutic foods (RUTFs). The current RUTF dose is prescribed according to the weight of the child to fulfil 100% of their nutritional needs until discharge. However, there is doubt concerning the dose, as it seems to be shared, resulting in suboptimal cost-efficiency of SAM treatment. We investigated the efficacy of a reduced RUTF dose in community-based treatment of uncomplicated SAM.
We undertook a randomised trial testing the non-inferiority of weight gain velocity of children with SAM receiving (a) a standard RUTF dose for two weeks, followed by a reduced dose thereafter (reduced), compared with (b) a standard RUTF dose throughout the treatment (standard). A mean difference of 0.0 g/kg/day was expected, with a non-inferiority margin fixed at −0.5 g/kg/day. Linear and logistic mixed regression analyses were performed, with study site and team as random effects. Between October 2016 and July 2018, 801 children with uncomplicated SAM aged 6–59 months were enrolled from 10 community health centres in Burkina Faso. At admission, the mean age (± standard deviation [SD]) was 13.4 months (±8.7), 49% were male, and the mean weight was 6.2 kg (±1.3). The mean weight gain velocity from admission to discharge was 3.4 g/kg/day and did not differ between study arms (Δ 0.0 g/kg/day; 95% CI −0.4 to 0.4; p = 0.92) confirming non-inferiority (p = 0.013). However, after two weeks, the weight gain velocity was significantly lower in the reduced dose with a mean of 2.3 g/kg/day compared with 2.7 g/kg/day in the standard dose (Δ −0.4 g/kg/day; 95% CI −0.8 to −0.02; p = 0.041). The length of stay (LoS) was not different (p = 0.73) between groups with a median of 56 days (interquartile range [IQR] 35–91) in both arms. No differences were found between reduced and standard arm in recovery (52.7% and 55.4%; p = 0.45), referral (19.2% and 20.1%; p = 0.80), defaulter (12.2% and 8.5%; p = 0.088), non-response (12.7% and 12.5%; p = 0.95), and relapse (2.4% and 1.8%; p = 0.69) rates, respectively. However, the reduced RUTF dose had a small 0.2 mm/week (95% CI 0.04 to 0.4; p = 0.015) negative effect on height gain velocity with a mean height gain of 2.6 mm/week with reduced and 2.8 mm/week with standard RUTF dose. The impact was more pronounced in children under 12 months of age (interaction, p = 0.019) who gained 2.8 mm/week with reduced and 3.1 mm/week with standard dose (Δ −0.4 mm/week; 95% CI −0.6 to −0.2; p < 0.001). Limitations include not blinding participants to the RUTF dose received and excluding all children with negative appetite test. The results are generalisable for relatively food secure contexts with a young SAM population.
Reducing the RUTF dose provided to children with SAM after two weeks of treatment did not reduce overall weight or mid-upper arm circumference (MUAC) gain velocity nor affect recovery or lengthen treatment time. However, it led to a small but significant negative effect on linear growth, especially among the youngest. The potential effect of reducing the RUTF dose in a routine program on treatment outcomes should be evaluated before scaling up.
ISRCTN registry ISRCTN50039021.
| Severe acute malnutrition (SAM) affects 19 million children worldwide and is treated with ready-to-use therapeutic foods (RUTFs).
The current RUTF formulation and dosage are based on an inpatient treatment model and aim at achieving fast weight gain and recovery.
However, when used in a home-based setting, the RUTF dose prescribed is often considered too large, resulting in sharing and a suboptimal cost-efficiency of SAM treatment.
We sought to investigate the efficacy of reducing the RUTF dosage used in community-based treatment of uncomplicated SAM.
We conducted a randomised controlled trial testing the non-inferiority of a reduced RUTF dose in the management of uncomplicated SAM compared with standard dose. This was called the MANGO trial.
We enrolled 801 children 6–59 months of age and randomised them individually into receiving (1) a standard dose of RUTF throughout the treatment and (2) a standard dose during the first two weeks, followed by a reduced dose from third treatment week onwards.
The trial showed non-inferior weight gain velocity from admission to discharge and similar recovery and length of stay in treatment.
However, the reduced dose also resulted in a significantly slower height gain velocity compared with the standard dose.
Our findings suggest that the reduction of the RUTF dose after the first two weeks results in similar weight gain velocity and recovery rates as with the standard dose given throughout SAM treatment.
However, the reduced RUTF dose seems to slow down the height gain velocity of children and might thus not be fully optimal for children’s healthy growth.
The reduced dose approach should be tested in a routine programmatic setting and in different food security contexts before scale-up.
| Worldwide, 19 million children under 5 years of age suffer from severe acute malnutrition (SAM), contributing to over 500,000 deaths per year [1]. According to the World Health Organization (WHO) guidelines for community-based management of acute malnutrition (CMAM), children without medical complications at admission are treated as outpatients, with weekly checkup visits [2]. Treatment consists of a systematic antibiotic regimen, as well as a ready-to-use therapeutic food (RUTF), prescribed according to the weight of the child and continued until discharge [2].
RUTFs are highly fortified energy dense pastes that are designed to fulfil 100% of the nutritional needs of children during the recovery from SAM [3]. In theory, the prescribed dose should enable weight gains up to 20 g/kg/day, as observed in inpatient treatment of SAM with RUTFs [4]. However, high weight gain rates have never been observed in community settings where the average ranges between 1.0 and 5.5 g/kg/day using RUTFs [5–19], suggesting a lower intake of the therapeutic product in home-based treatment. Several studies have suspected or reported product sharing within and outside the household [8,13,20] as a reason for lower weight gain.
The perceived high cost and large quantity of RUTFs administered [20–25] have sparked attempts to optimise the product formulation and use [11,26–28]. One cluster-randomised trial in Sierra Leone gradually reduced the RUTF dose of children recovering from SAM when they reached moderate acute malnutrition (MAM) criteria [28]. However, the use of different recovery criteria between intervention and control groups limits the interpretation of the results. A retrospective analysis of a CMAM program in Myanmar, where, due to RUTF shortage, the dose was reduced once children with SAM reached MAM status, showed high recovery rates (90.2%). The lack of a control group in the study limits the interpretability of data [18]. To our knowledge, no rigorous clinical trial looking at the efficacy of reducing the RUTF dose among children with SAM has been conducted. Reducing the RUTF dose given to children treated for SAM in a cost restrained setting could enable the management of more malnourished children with the same resources.
The present study aimed to test, in a non-inferiority randomised controlled design, the impact of reducing the RUTF dose, after two weeks, on the weight gain velocity of children treated for uncomplicated SAM in the community. The reduction aimed to support weight gain rates of 5 g/kg/day and simplify the distribution and use of RUTF by children with SAM to 1 or 2 daily sachets for children <7 kg and ≥7 kg, respectively.
The study was performed in accordance with the principles in the Declaration of Helsinki. The research protocol obtained ethical clearance from the national ethics committee (Comité d'éthique pour la recherche en santé [CERS]) and the clinical trials board (Direction Générale de la Pharmacie, du Médicament et des Laboratoires [DGPML]) of Burkina Faso. An independent Data Safety Monitoring Board composed of one paediatrician and one statistician was responsible for monitoring serious adverse events and conducted five complete data reviews during the course of the study. Caregivers provided verbal and written consent prior to enrolment and were made aware of their right to withdraw from the study at any time. Caregivers in both arms were given an instant photo of their child at the end of the treatment period and a bucket with soap at the end of the 3-month post-discharge follow-up period to compensate for the time spent on study procedures.
We conducted a randomised controlled clinical trial (called MANGO) comparing the efficacy of a reduced RUTF dose to a standard RUTF dose in the management of uncomplicated SAM in children 6–59 months of age in a non-inferiority design.
The study was conducted in the Fada N’Gourma health district located in the Eastern region of Burkina Faso. Malaria is endemic, with 69.3% of children presenting a positive rapid test [29]. HIV prevalence is 1.0% among 15–49-year-olds. In 2016, the prevalence of severe wasting (weight-for-height z-score [WHZ] <−3) and moderate wasting (WHZ between −3 and −2) was 2.4% and 8.6%, respectively [30]. There were 42 health centres in the district in 2015, all run by the Ministry of Health and supported by Action Against Hunger; 10 were chosen as study sites based on criteria on minimum SAM caseload (>7 new SAM admissions/month), accessibility, and a suitable schedule to couple study visit days with routine growth monitoring days.
Between October 2016 and July 2018, study participants were selected from children presenting with SAM at the 10 participating health centres for curative and preventive activities. Study staff checked admission criteria: WHZ <−3 and/or mid-upper arm circumference (MUAC) <115 mm, positive appetite test (performed as per the national protocol [31]), no oedema or medical complications, and between 6 and 59 months of age. Exclusion criteria included having received treatment for SAM within 6 months, caregiver planning to travel or unable to comply with the weekly checkup schedule, peanut or milk allergy, or disability affecting food intake. Children with any grade of oedema or medical complications, as defined by the Burkina national protocol for CMAM [31], at any time during the study were referred to inpatient care.
Randomisation was stratified by health centre using varying block sizes from 2 to 8. Randomisation lists were generated using the website www.randomization.com. After confirming eligibility and obtaining consent from the caregiver, children were given a unique study identifier (ID) by a team supervisor and assigned to a treatment group. Only the RUTF distributors had access to the randomisation lists, while staff involved in assessing the eligibility and study outcomes of the child were blinded to the trial arm. Participants could not be blinded to the RUTF dose received. Investigators remained blinded to treatment groups until the final analysis stage.
Upon admission, the child’s caregiver was interviewed regarding household socioeconomic characteristics, care practices, and recent morbidity of the child and encouraged to adhere to weekly visits until recovery. Anthropometric measurements and a clinical examination were performed at each visit from admission to discharge. As per national SAM treatment protocol, seven key messages were delivered to caregivers in both groups, including advice to continue breastfeeding and to offer family foods in addition to RUTF if needed.
Anthropometrics were measured in duplicate at each visit: weight using an electronic scale (SECA 876, SECA, Hamburg, Germany) to the nearest 100 g, height (recumbent for <24 months of age; standing for ≥24 months of age) using a wooden measuring board (locally made) to the nearest 1 mm, and MUAC using a non-stretchable colourless measuring tape to the nearest 1 mm. Using WHO field tables, WHZ was determined and used for admission and discharge. In later analysis, WHZ was calculated using the package ‘zscore06’ [32] in STATA 15 (StataCorp, College Station, TX).
Children were followed up until recovery. Children missing their study visit were contacted either directly by telephone or via a community health worker and encouraged to return. Children referred did not return to trial after inpatient phase, as referral was considered a trial endpoint. Recovered children were followed up fortnightly for 12 weeks and relapses recorded. A supplementary feeding program accompanied the post-discharge follow-up, providing ready-to-use supplementary foods when available.
Treatment followed the Burkina national CMAM guidelines in all aspects except the RUTF dose. Half of the children received a reduced dose from the third treatment week onwards (Table 1). Medical treatment included 7 days of amoxicillin for all children at admission (50–100 mg/kg/day), albendazole at the second treatment visit for children ≥12 months (200 mg to 12–23-month-olds; 400 mg to ≥24-month-olds) and catch-up doses for missed routine vaccinations or vitamin A supplementation (100,000 IU to 6–11-month-olds; 200,000 IU to 12–59-month-olds, every 6 months) at admission. Any illness, such as malaria, respiratory tract infections, or diarrhoea, diagnosed during the study was treated according to national protocol. See S2 Text for the full protocol for the study.
Two study teams were comprised of one nurse, three measurers, one food distributor, and one supervisor per team. All team members were trained on research ethics and processes; standard operating procedures were defined, tested, and applied. Data were collected via tablets using the Open Data Kit (ODK1 software), and continuous data monitoring and cleaning were performed by a data manager under the supervision of the principal investigator. Electronic data were password protected, and field registries were kept in a locked office. Data were de-identified prior to analysis.
The primary outcome was weight gain velocity (g/kg/day) from admission to discharge. Other outcomes included weight gain velocity after two weeks, length of stay (LoS), discharge anthropometrics, linear and MUAC growth, treatment outcome, morbidity, and relapses.
Weight gain velocity from admission to discharge was calculated by dividing the weight gain (weight at discharge − weight at admission) in grams by the weight at admission in kilograms and the LoS in days. Weight gain velocity after two weeks was measured as follows: (weight at discharge − weight at visit 3 [in g]) ÷ (weight at admission [in kg]) ÷ (LoS − 14 [in days]). Missing weights at visit 3 (60 in reduced and 58 in standard arm) were imputed using mean weekly weight gained between an earlier visit (1 or 2) and later visit (4 or 5). The length of the stay was calculated as the number of days spent from admission to either recovery, referral, nonresponse, false discharge, or last visit before defaulting, lost to follow-up, or death. Linear and MUAC growth were defined as gains in millimetres (exit measure − admission measure)/week (LoS/7). A minimum acceptable mean rate of weight gain of 3.0 g/kg/day was defined at the protocol stage as a quality cutoff for evaluating general program performance.
Nutritional recovery was defined as reaching a WHZ of ≥−2 for those admitted with a WHZ <−3 only, or MUAC ≥125 mm for those admitted with a MUAC <115 mm only, or both WHZ ≥−2 and MUAC ≥125 mm for those admitted with both WHZ <−3 and MUAC <115 mm upon two consecutive visits and absence of any illness. Referrals included children referred to inpatient care as a result of medical complications, >5% weight loss within three weeks, or ≤100 g weight gain over four weeks in the absence of apparent illness. Nonresponse included children not reaching anthropometric discharge criteria by 16 weeks of treatment who were referred to inpatient care for further examinations. Defaulters were defined as having missed three consecutive visits, but the child was confirmed to be alive. Transfers to health centres not involved in the study were categorised as defaulters. ‘Lost to follow-up’ was defined as having missed three consecutive visits without a known status of the child. False discharges included children who were erroneously discharged as recovered or referred, but upon analysis did not meet the criteria. Relapses were recorded over 12 weeks following recovery and were defined as presenting a WHZ <−3 and/or a MUAC <115 mm, or any grade of bilateral oedema.
We assumed an expected mean difference in weight gain velocity between the two groups of 0.0 g/kg/day and a standard deviation (SD) of 2.6 g/kg/day with a non-inferiority margin of 0.5 g/kg/day. Assuming a power of 80% and a 5% significance level for a one-sided test, 335 children were needed in each group to demonstrate non-inferiority. To allow for dropout, the total target sample size was 800 children. Applying a 0.22 SD difference as could be observed in the main outcome with the calculated sample would allow us to detect a difference of 12% in recovery, seven days in LoS, 0.4 mm/week in MUAC gain velocity, and 0.3 mm/week in height gain velocity.
Baseline characteristics of the study population were summarised as percentages and means (±SDs). Linear mixed models were used to compare primary and secondary outcomes of weight, linear and MUAC growth velocities, LoS, and anthropometric endpoints. Results were reported as differences of reduced dose from standard dose (reduced minus standard), with positive values meaning greater estimates among reduced dose. For programmatic outcomes, logistic mixed models were used to compare groups. Time to recovery was analysed using the Cox proportional hazards model. Study sites and research teams were included in the mixed models as random effects. Unadjusted models and models adjusted for sex, age, admission measure of weight, height, MUAC, WHZ, wealth, LoS in treatment, and month of admission were fitted. Adjustments were defined in the statistical analysis plan development stage prior to data analysis (S3 Text). Model checking was based on residual plots and normal probability plots, when applicable. All analyses were performed using STATA 15 (StataCorp).
Both intention to treat (ITT) and per protocol (PP) analyses were carried out for the main outcome and key secondary outcomes. ITT analysis included all children admitted to the study for whom an endpoint observation was available. PP analysis included children without missed visits who, according to maternal recall, consumed >50% of the daily dose at all times and excluded those who had received a wrong treatment dose or had been falsely discharged.
Interactions were only investigated in ITT analyses. Interactions between treatment and age group (<12 months versus ≥12 months), sex, MUAC category (<115 versus ≥115 mm), WHZ category (<−3 versus ≥−3), and stunting (height-for-age z-score [HAZ] < −2 versus HAZ ≥ −2) at admission were evaluated for the main outcome of weight gain velocity and the key secondary outcomes of recovery, LoS and height gain velocity, by means of likelihood ratio tests. Only significant interaction terms led to subgroup analyses.
‘Urban’ was defined as those living ≤30 minutes’ return trip from the regional capital city. Low birth weight (<2,500 g) was confirmed from an official birth certificate or health card. Household Food Insecurity Access Scale (HFIAS) was constructed according to FANTA indicator guide [33].
From October 17, 2016, to July 20, 2018, 1,186 children were diagnosed with SAM and assessed for eligibility at 10 study sites. Of these, 802 (68%) children were eligible for the study and randomised to standard or reduced RUTF dose (Fig 1). One child was excluded after randomisation for not meeting the SAM criteria at admission. Therefore, 801 patients were included in the trial: 402 in the reduced dose and 399 in the standard dose arm. Thirteen children defaulted or were referred immediately after admission; four and nine in reduced and standard dose arms, respectively. Three children developed oedema (two in reduced and one in standard dose arm) and were excluded from weight gain calculation. Only recovered children continued to the post-discharge follow-up, contributing to the post-discharge outcome analyses including relapse rate (Fig 1).
Randomisation resulted in baseline equivalence between the reduced and standard dose arms with respect to potential confounders (Table 2). The mean age at admission was 13.4 months, 49% were boys, and the mean admission weight was 6.2 kg. At visit 3, no children were <3.5 kg, 5% were 3.5–4.9 kg, 61% were 5.0–6.9 kg, 30% were 7.0–9.9 kg, and 4% were 10.0–14.9 kg. Caregivers were, on average, 28 years of age, 76% had no formal education, and 88% were categorised as food secure.
The mean weight gain velocity from admission to discharge was 3.4 g/kg/day in both groups in ITT analysis (Δ 0.0 g/kg/day; 95% CI −0.4 to 0.4). Non-inferiority of the reduced dose could be confirmed in both ITT (inferiority rejected: p = 0.013) and PP (inferiority rejected: p = 0.019) for this main outcome (Fig 2). No differences were found in weight gain velocity in PP analysis (Δ 0.2 g/kg/day; 95% CI −0.5 to 0.8), in ITT among recovered only (Δ −0.1 g/kg/day; 95% CI −0.6 to 0.4), referrals (Δ 0.5 g/kg/day; 95% CI −0.6 to 1.5), or defaulters (Δ −0.3 g/kg/day; 95% CI −1.3 to 0.8) (Table 3). No interactions were found between treatment and sex, age, MUAC category, WHZ category, or stunting status at admission. In general, mean weight gain velocity was high at the start of the treatment and decreased rapidly (Fig 3). When entering third treatment week, 27% of children still had SAM (WHZ <−3 and/or MUAC <115 mm): 108 children in the reduced and 106 children in the standard group.
Weight gain velocity after the first two weeks of treatment (in ITT) was significantly different between groups, with a mean of 2.3 g/kg/day with reduced versus 2.7 g/kg/day with standard dose (Δ −0.4 g/kg/day; 95% CI −0.8 to −0.02). Results comparing the MUAC gain velocity between reduced and standard doses mirrored the results obtained with weight gain velocity (Table 3). Adjusted analysis yielded similar results that are found in S1 Table.
No differences were found in anthropometry at discharge between study arms in the unadjusted model (all p > 0.2). However, when using the adjusted model (adjusting for sex, age, admission measure of weight, MUAC, WHZ and height, month of admission, LoS, and wealth index), height at discharge was significantly smaller in the reduced dose arm (Table 4). This difference in height of 0.1 cm could still be observed 3 months post-recovery, although the difference was then no longer significant (p = 0.33). Weight, MUAC, weight-for-age z-score (WAZ), WHZ, and HAZ were not different at three months post-recovery.
The median LoS was 56 days (interquartile range [IQR] 35–91) in both arms. The WHZ category was an effect modifier (interaction, p = 0.028) whereby children who were admitted with a WHZ ≥−3 and treated with the reduced dose had a LoS 6.9 days (95%CI −0.1 to 13.9; p = 0.055) longer than those treated with the standard dose. No difference was found in the LoS of children admitted with WHZ <−3 between reduced and standard doses. No effect modification for LoS was observed between treatment and age, sex, MUAC category, or stunting at admission. The recovery rate was similar in both arms: 52.7% in reduced dose and 55.4% in standard dose (Δ −2.6%; 95% CI −9.5 to 4.3) (Table 5). Cox proportional hazards model showed no difference (p = 0.54) in the time to recovery between the two arms (Fig 4). No significant interactions were found for recovery. No differences were found in the proportion of children referred (19.2% and 20.1%), defaulting (12.2% and 8.5%), nonresponding (12.7% and 12.5%), and relapsed (2.4% and 1.8%) between reduced and standard RUTF dose arms, respectively (Table 5). PP analysis provided similar results and is found in S2 Table. The number or duration of illnesses between study arms did not differ (p > 0.2): 60% of children had a respiratory illness, 38% had malaria, and 52% had diarrhoea at some point after admission. During the intervention, two children died (one in each group), one was lost to follow-up (standard dose), and 24 were falsely discharged (12 in each group).
Height gain velocity was lower in children who received the reduced dose (2.6 mm/week) than the standard dose (2.8 mm/week) (Δ −0.2 mm/week; 95% CI −0.4 to −0.04). Age was an effect modifier (interaction, p = 0.019): the height gain of children under 12 months of age was 2.8 mm/week with reduced dose and 3.1 mm/week with standard dose (Δ −0.4 mm/week; 95% CI −0.6 to −0.2). Similarly, from admission to discharge, HAZ increased by 0.05 SD with reduced and 0.09 SD with standard RUTF dose (Δ −0.04 SD; 95% CI −0.09 to 0.002). Again, age was an effect modifier (interaction, p = 0.016): children under 12 months of age had 0.00 SD catch-up in HAZ with reduced dose compared with 0.09 SD with standard dose (Δ −0.09 SD; 95% CI −0.15 to −0.03) (Table 6). Height gain or HAZ catch-up did not differ between RUTF dose among children ≥12 months. Adjusted analysis provided similar results that are found in S3 Table.
Evidence is needed to inform policy on the optimisation of treatment of uncomplicated SAM. The current trial investigated the efficacy of reducing the RUTF dose after two weeks and showed that there was no effect on the total weight or MUAC gain velocity, recovery, or LoS in treatment. However, the linear growth of children receiving the reduced dose was significantly slower, particularly among the youngest group of <12-month-old children.
The mean weight gain velocity of 3.4 g/kg/day observed from admission to discharge is in line with those reported in earlier CMAM studies [5–20]. While non-inferiority was confirmed for weight gain velocity from admission to discharge, the difference in weight gain after two weeks was 0.4 g/kg/day (p = 0.041) between children receiving the reduced dose of RUTF (mean 2.3 g/kg/day) compared with the standard dose (mean 2.7 g/kg/day). This finding suggests that the rapid weight gain in the first two weeks, when all children received the standard dose, masks the small negative effect the reduced dose has on the subsequent growth of children. The weight gain velocity during treatment shows a quickly decreasing pattern, in which the first weeks represent high catch-up in weight. From the fifth treatment week onward, the weight gain velocity drops to <2 g/kg/day and resembles that observed in MAM treatment programs [34–36]. From the eighth week onward, the weight gain velocity was approximately 1 g/kg/day, similar to normal weight gain velocity for a healthy one-year-old child [37]; this despite continuing to receive RUTF. However, recovery continues throughout treatment until the 16th week, when those not yet recovered were considered ‘nonresponse’ to treatment.
Linear growth observed in the trial (2.6 mm/week with reduced dose and 2.8 mm/week with standard dose) is in line with other CMAM studies [5,6,9,10,16,19,38] and with the 2.8 mm/week growth rates expected in healthy 13-month-old children [39]. However, the reduced dose slowed down the linear growth of children by 0.2 mm/week (95% CI 0.04 to 0.4; p = 0.015). Whether this difference is clinically significant remains questionable. In general, linear growth is considered at least as important, if not more important, than weight gain for the healthy growth of children [40]. A follow-up study of children recovered from SAM found that seven years later, these children had similar weight- and BMI-for-age, but significantly lower HAZ and absolute height compared with their siblings and community controls [41]. HAZ and absolute height are both predictors of chronic disease in later life [42]. In our trial, the mean HAZ was −2.4 at admission, and it increased by 0.05 and 0.09 with the reduced and standard doses, respectively (Δ −0.04 HAZ; 95% CI −0.09 to 0.002; p = 0.063). On the contrary, among children treated for MAM, HAZ decreased by 0.17 during 12 weeks of supplementary feeding in Burkina Faso [34]. Reasons for different linear catch-up growth between children with MAM and SAM could include therapeutic food quality and quantity. Linear growth requires micronutrients, in particular, type 2 micronutrients such as zinc, magnesium, and potassium [43], which are provided by RUTF [3] at much higher levels than are available in local diets [44]. A reduction in the RUTF dose or quality possibly reduces the quantity and density of these nutrients and by consequence may affect linear growth.
The observed recovery rates (52.7% with reduced dose and 55.4% with standard dose) are low but are explained by the strict application of referral criteria in our trial, and that referral was considered an effective study endpoint. When using the SPHERE calculation method [45] excluding referred and false discharge, recovery rates were 68% in the reduced dose and 72% in standard dose, somewhat under the recommended >75%. Up to 20% of children were referred primarily as a result of weight loss or stagnant weight, which may not often be identified in routine programs. Nevertheless, the similar referral rate between study arms suggests these referrals are not related to a dose effect. In a post hoc analysis, weight loss was associated with higher numbers and longer duration of illness episodes. Episodes of infection are known to drive undernutrition via appetite loss, reduced nutrient absorption, nutrient losses, diversion of nutrients to inflammatory responses, and tissue repair [46]. Open defecation was practised by 76% of households, indicating a poorly sanitised home environment with high risk of exposure to pathogens [47]. This could partially explain a proportion of illness episodes and the relatively high proportion of cases with weight loss or stagnant weight. Beyond acute illness, environmental enteric dysfunction is another potential driver of suboptimal recovery [46].
In the current trial, the reduced RUTF dose had no negative effects on children ≥12 months of age but slowed down the height gain of children <12 months of age. Bahwere and colleagues (2016) found that providing milk-free RUTF had no adverse effect among children with SAM ≥24 months of age, while children <24 months of age had a significantly lower recovery rate. Seemingly, younger children are somewhat more sensitive to changes in RUTF quantity and quality, possibly requiring standard treatment in order to gain the full benefit from SAM treatment. Whether separate protocols should be considered for different age groups remains a question requiring operational feasibility, effectiveness, and cost estimations.
In the context of the current global transition from a focus on purely ensuring the survival of children to actually enabling them to thrive, the quest is emerging in the malnutrition community to go beyond recovery and seek to optimise the functional and long-term outcomes of children treated for SAM [41,48]. This will require looking at body composition, micronutrient status, cognitive development, and other long-term outcomes. At the same time, the resources are limited and cost-efficiency should be taken into account and existing and new investments evaluated against the benefits they can bring.
The main strength of the study is the individually randomised design with few dropouts immediately after admission, which reduces confounding and enables a causal analysis of the effect of the reduced RUTF dose on the weight gain of uncomplicated SAM. The field study was implemented by well-trained and experienced research staff who were responsible for the diagnosis, treatment, and discharge of children with SAM. The number of patients per day was also limited to enable thorough care and follow-up of the study children.
As with all studies, there are limitations. First, although we did not reveal the study arm to participants, it was not possible to blind them to the RUTF dose received. However, because the daily dose prescribed to children in both arms depended on the weight of the child, we did not expect caregivers to be fully aware of the treatment allocation. With the exception of the RUTF distributor, the research staff were blinded to the dose. Second, we excluded and referred to inpatient care all children who did not pass the appetite test at admission. While officially a referral criterion, the appetite test is not always implemented in the field. It is possible that children who in routine practice fail the appetite test have a slower weight gain at the beginning of treatment and thereafter an inferior response to a reduced dose after two weeks.
The study findings are only generalisable to a nonemergency context with relatively good food security and where SAM cases are primarily very young (<24 months of age). However, further research is needed to corroborate these findings in a routine program with fewer resources and time per patient, and to translate these finding to an emergency context or to older-aged SAM cases.
In conclusion, reducing the RUTF dose prescribed to children with SAM after two weeks does not appear to affect the total weight or MUAC gain velocity, recovery rate, nor the LoS in treatment. However, the linear growth of children became slower with the RUTF reduction, especially in young children. Before considering a reduction of RUTF during SAM treatment, an effectiveness study in a routine program setting is needed to confirm the results.
|
10.1371/journal.pntd.0004857 | Invasive Non-typhoidal Salmonella Infections in Asia: Clinical Observations, Disease Outcome and Dominant Serovars from an Infectious Disease Hospital in Vietnam | Invasive non-typhoidal Salmonella (iNTS) infections are now a well-described cause of morbidity and mortality in children and HIV-infected adults in sub-Saharan Africa. In contrast, the epidemiology and clinical manifestations of iNTS disease in Asia are not well documented. We retrospectively identified >100 cases of iNTS infections in an infectious disease hospital in Southern Vietnam between 2008 and 2013. Clinical records were accessed to evaluate demographic and clinical factors associated with iNTS infection and to identify risk factors associated with death. Multi-locus sequence typing and antimicrobial susceptibility testing was performed on all organisms. Of 102 iNTS patients, 71% were HIV-infected, >90% were adults, 71% were male and 33% reported intravenous drug use. Twenty-six/92 (28%) patients with a known outcome died; HIV infection was significantly associated with death (p = 0.039). S. Enteritidis (Sequence Types (ST)11) (48%, 43/89) and S. Typhimurium (ST19, 34 and 1544) (26%, 23/89) were the most commonly identified serovars; S. Typhimurium was significantly more common in HIV-infected individuals (p = 0.003). Isolates from HIV-infected patients were more likely to exhibit reduced susceptibility against trimethoprim-sulfamethoxazole than HIV-negative patients (p = 0.037). We conclude that iNTS disease is a severe infection in Vietnam with a high mortality rate. As in sub-Saharan Africa, HIV infection was a risk factor for death, with the majority of the burden in this population found in HIV-infected adult men.
| Invasive non-typhoidal Salmonella (iNTS) infections occur when Salmonella bacteria, which normally cause diarrhea, enter the bloodstream and spread through the body. Invasive NTS infections have become a common cause of infection and death in children with malaria and adults with HIV in sub-Saharan Africa. However, it is unknown whether iNTS is as common or as severe outside sub-Saharan Africa. We evaluated over 100 iNTS cases from an infectious disease hospital in southern Vietnam admitted between 2008–2013. We used hospital records to determine the clinical features of iNTS disease and to identify risk factors associated with death and performed typing of the isolated organisms. The majority of patients were HIV positive (72/102, 71%), >90% of patients were adults, 71% were male and 33% reported intravenous drug use. The mortality rate of iNTS patients was 28% (26/92), and HIV infection was a significant risk factor for fatal outcome (p = 0.039). The serovars most commonly identified were S. Enteritidis and S. Typhimurium; S. Typhimurium was found more frequently in HIV-positive individuals (p = 0.003). We report that iNTS disease is a severe infection in Vietnam with a high mortality rate. Similar to sub-Saharan Africa, HIV infection was a strong risk factor for death.
| Infections with organisms belonging the bacterial genus Salmonella are associated with a range of disease syndromes in humans. A small subset of the >2,500 described serovars of that belong to Salmonella subspecies I are capable of causing typhoidal illness, these include Salmonella enterica serovar Typhi (S. Typhi) and the various S. Paratyphi pathovars [1]. However, the vast majority of the Salmonella subspecies I serovars are not commonly associated with systemic disease in humans and are referred to as non-typhoidal Salmonella (NTS). NTS organisms include S. Typhimurium, S. Dublin and S. Enteritidis, which are characterized by their wide host range and their ability to induce a self-limiting diarrhea [2]. However, in addition to the common diarrheal clinical syndrome induced by NTS organisms in humans, invasive (bloodstream) NTS (iNTS) disease can also occur in specific populations [3,4]. iNTS disease, which is most commonly caused by the Salmonella serovars Typhimurium and Enteritidis [5,6], is associated with an aggressive systemic infection that loosely resembles typhoid fever [2,4,7]. In sub-Saharan Africa, the disease has a high mortality rate (20–25%) and infection is most common in children with malaria, malnourished children and HIV-infected adults [3]. There are an approximately 1.9 million cases of iNTS disease in sub-Saharan Africa annually, with an overall estimated incidence rate of 227 per 100,000 population [8], and 175–388 and 2,000–7,500 per 100,000 population in children 3–5 years of age and HIV-infected individuals, respectively [3,9–14].
NTS are a common cause of diarrhea in Asia, and we have previously shown that NTS are responsible for approximately 4% of pediatric hospitalized diarrhea in Ho Chi Minh City (HCMC), Vietnam [15]. In a retrospective study of blood cultures conducted between 1994 and 2008 at the Hospital for Tropical Diseases (HTD) in HCMC we observed that S. Typhi was the predominant cause of culture positive bacteremia (66%) until 2002 [16]. After this period we detected a significant annual decline in the isolation rate of S. Typhi and a concurrent increase in organisms associated with the HIV epidemic, including NTS. The isolation rate of NTS increased from 1% (n = 47) of total bacteremia cases between 1994–2001 to 4% (n = 146) of cases from 2002–2008. Whilst the increase in burden was modest in comparison to sub-Saharan Africa these data support a longitudinal shift in the etiology of bloodstream infections in southern Vietnam.
There is a paucity of data regarding iNTS infections from Asia, with limited reports from Taiwan [17], India [18], Thailand [19,20] and our aforementioned study in Vietnam [16]. It is apparent that the burden of iNTS in sub-Saharan Africa is not mirrored in Asia. However, iNTS disease is present in Asia but there are no or few data regarding clinical symptoms, disease outcome, patient demographics or the infecting serovars. By accessing available clinical data and bacterial isolates we sought to retrospectively investigate the clinical and microbiological manifestations of iNTS in a major infectious disease hospital in southern Vietnam.
Ethical approval for this study was provided by the institutional review board (IRB) of the HTD. This study was performed retrospectively with routine anonymous laboratory and clinical data; individual patient identifying were not accessed and informed consent was not required.
HTD is a 550-bed hospital that serves as a main primary and secondary facility for the surrounding local population in HCMC and a tertiary referral center for infectious diseases for the southern provinces of Vietnam. Nearly 70% of HTD admissions live in HCMC, with the remainder residing in the surrounding provinces. Neonates, patients without infectious diseases, including those with surgical requirements, tuberculosis, cancer, primary hematological disorders or immunosuppression (other than HIV) are referred to other hospitals within HCMC. HIV-infected children are often referred to local pediatric hospitals.
The study population consisted of all individuals from which an NTS organism was isolated alone or in combination with an additional pathogen in blood culture from January 2008 through June 2013. This source data was collected from routine microbiology laboratory logbooks in which data from positive and negative blood culture are recorded. Patients with multiple positive blood cultures for the same NTS serogroup and antimicrobial susceptibility profile were considered to be a single case.
A patient record form was used to collect clinical and laboratory data from the hospital chart for every patient. Clinical data recorded on admission included sex, HIV status (HIV diagnosed according to the World Health Organization (WHO) guidelines [21]), axillary temperature, presence of co-infection and hospital outcome. Outcome was classified based on clinician notes as follows: (1) recovery or improvement, (2) worsening status on discharge (often deteriorating patients taken from hospital by their relatives to die at home—a common custom in Vietnam), (3) death or (4) transfer to a different hospital (patient’s condition was unchanged but transferred to other hospital for specific treatment or surgery intervention, or patient left against medical advice). Outcome 2 and outcome 3 were considered to be fatal. Laboratory data was comprised of standard hematology and biochemical testing from hospital records on the day of admission.
Clinical and laboratory data were compared between groups using Fisher's exact or Kruskal-Wallis tests for categorical and continuous data, respectively. We performed univariable and multivariable logistic regression to evaluate covariates that were independently associated with fatal outcome. Covariates selected for multivariable analysis a-priori included age, sex and immunosuppression (HIV status, hepatitis), in addition to other fixed demographic or clinical covariates that were significantly associated (p<0.05) with outcome from the univariate analysis. All statistical analyses were performed using Stata version 11 (StataCorp, College Station, TX, USA).
For blood culturing, 5–10 mL of venous blood from adults and 2–5 mL of venous blood from infants and children was inoculated into BACTEC plus aerobic bottles (Becton Dickinson). Inoculated BACTEC bottles were incubated at 37°C in a BACTEC 9050 automated analyzer for up to five days and sub-cultured when the machine indicated a positive signal. Organisms were identified by standard methods including API20E identification kits (Bio-Mérieux, Craponne, France). Specific grouping antisera were used to identify the serogroup of the isolated Salmonella on original culture. Vi antisera (along with 0:9) was used to identify S. Typhi; these were excluded from all analyses. All NTS isolated from blood cultures were stored in Brain Heart Infusion (BHI) glycerol at -70°C. For the purposes of this study all NTS isolated were recovered on MacConkey agar and subjected to re-identification and antimicrobial susceptibility testing. Re-identification of Salmonella serogroups was performed using specific grouping antisera as before. Antimicrobial susceptibility testing was performed on Muller-Hinton agar against ampicillin, amoxicillin/clavulanate, azithromycin, ceftazidime, ceftriaxone, chloramphenicol, ciprofloxacin, gentamicin, nalidixic acid, ofloxacin and trimethoprim-sulfamethoxazole, using the disk diffusion method as recommended by CLSI guidelines [22]. Antimicrobial disks were purchased from Oxoid (Thermo Fisher Scientific, UK) and susceptibility was determined using the Clinical Laboratory Standard Institute (CLSI) guidelines [22].
To further characterize the iNTS isolates, all organisms were genotyped (and molecular serotyped) using multi-locus sequence typing (MLST) following previously described methods [23,24]. Briefly, a set of seven housekeeping genes (aroC, dnaN, hemD, hisD, purE, sucA and thrA–primer sequences accessed at http://mlst.warwick.ac.uk/mlst/dbs/Senterica) were PCR amplified using template DNA extracted from each isolate after boiling bacterial colonies in PBS. PCR amplicons were cleaned using Agentcourt Ampure XP (Beckman Coulter) and were sequenced in both directions using BigDye Terminator v3 (Applied Biosystems, USA) followed by capillary sequencing on a 3130XL Genetic Analyzer (Applied Biosystems, USA). All sequences were manually trimmed to align to a reference sequence and were submitted to the previously mentioned MLST database for allelic profile and molecular serotyping (i.e. inferring serovar from MLST profile). A minimum spanning tree was created using the allelic profiles (variation in number of alleles between isolates of the seven housekeeping genes) using Bionumerics software (Applied Mathematics).
Between January 2008 and June 2013 there were 142 culture confirmed bloodstream infections caused by an NTS bacterium in this single centre. Hospital records were obtainable for 102/142 (72%) iNTS cases. The median patient age was 33 years (IQR: 28 to 41 years) (Table 1). Eight of the 102 (8%) iNTS cases were children (<16 years) of which five (5% of total) were infants (<12 months). The majority of patients (61/102; 60%) were from HCMC, with the remainder residing in the surrounding provinces. The median duration of illness (including fever and other symptoms) prior to hospital admission was 13 days (IQR 1–60 days). Patients were more commonly male (72/102; 71%) and three quarters (71/94; 76%) of adults (>16 years) reported that they were unemployed upon admission. A third (31/102; 33%) of cases reported a history of intravenous drug use, which was more common in men (26/65, 40%) than women (5/30, 17%) (p = 0.019, Fisher’s exact test).
HIV testing was performed for all patients diagnosed with iNTS infections; CD4 counts were not routinely measured. Seventy-two (71%) of the iNTS cases were HIV-infected: 71 adults (76% of 94 adults) and one infant (13% of all 8 children). Only 16/72 (22%) of the adult HIV-infected iNTS patients were on active antiretroviral therapy (ART) prior to this episode of bacteremia, and 6/72 (8%) of the HIV-infected iNTS patients were taking trimethoprim-sulfamethoxazole for Pneumocystis jiroveci pneumonia prophylaxis on admission. A history of long-term steroid use was reported in 4/30 (13%) of the iNTS cases testing negative for HIV infection.
Table 1 describes the clinical characteristics of the patients. The most common clinical features were fever (76/97, 78%) (≥38.0°C) and pallor (56/101, 56%). Almost half of the cases (45/102) had oralpharyngeal lesions, including ulcers and candidiasis; these symptoms were chiefly restricted to the HIV-infected group. Notably, gastrointestinal symptoms such as diarrhea (42/102, 41%) and abdominal pain (23/102, 23%), which are synonymous with the archetypal, non-invasive manifestation of NTS infection, were uncommon. However, comorbidities such as hepatitis (induced by hepatitis B, C or alcohol abuse) and pneumonia (caused by PCP or Mycobacterium tuberculosis) were recorded in 27% (27/102) and 71% (72/102) of patients, respectively. Furthermore, 16/102 (16%) patients had an additional pathogen identified in their bloodstream (BS) or cerebrospinal fluid (CSF): 9 Talaromyces marnefeii (BS), 4 Cryptococcus neoformans (CSF), 2 Escherichia coli (BS) and 1 Streptococcus pneumoniae (BS). None of these additional BS or CSF infections were identified in children. Septic shock was diagnosed in 6/102 (6%) cases; hypovolemic shock (due to fluid loss) was diagnosed in 2/102 (2%); a secondary infection was identified in only 1/6 (17%) patient with septic shock. Furthermore, 2/8 (25%) of the pediatric patients were diagnosed with hand-foot and mouth disease prior to the isolation of an NTS organism from the blood.
Overall 66/102 (65%) patients improved or recovered before hospital discharge; four (4%) died in hospital and 22 (22%) were discharged to die at home; the remaining 10 patients had an outcome that was non-assessable (five left against medical advice, two were unchanged and three transferred hospitals). One child (1/8, 12%), who was HIV-uninfected, died. The overall mortality rate was 26/92 (28%), of which 24 (92%) were HIV-infected. A total of 23% (6/26) of fatal cases had a secondary infection in BS or CSF. The median time to death in hospital was one day (IQR: 1–2 days) while median length of hospital stay for patients not discharged to die at home was 10 days (IQR: 3–15 days).
Hematology parameters for the 102 patients stratified by outcome are shown in Table 2. Notably, total white cell count was generally low (median 5.1 (IQR: 3.1–10.8) x 103 cells/μl) but characterized by a high proportion of neutrophils: 82% (IQR: 66.1–87.5). The platelet count was lower in fatal cases than nonfatal cases but this was not statistically significant. Fatal cases were significantly more likely to have s higher potassium, lower hemoglobin and lower hematocrit levels (Table 2). Additionally, we performed univariable and multivariable logistic regression analyses to assess the clinical and laboratory variables that were associated with death (Table 3). Though HIV positivity, age and infecting serovar were associated with death in the univariable analysis, after controlling for confounding only HIV positivity remained independently associated with an increased risk of fatality (Table 3).
The vast majority of iNTS patients received an antimicrobial (100/102; 98%) (Table 4). The most commonly used antimicrobial was ceftriaxone; 89/100 (89%) patients received this drug in mono or combination-therapy. A fluoroquinolone (levofloxacin, ciprofloxacin or ofloxacin) was used in 22/100 (22%) of cases, again either used in monotherapy or in combination with ceftriaxone (Table 3). Switching to an alternative antimicrobial (imipenem or meropenem) occurred on two occasions, of which one patient had a positive outcome and one was fatal. Trimethoprim-sulfamethoxazole was used in early therapy in 25/101(25%) iNTS; ceftriaxone was later added to this regime. Patients who were additionally diagnosed with Talaromyces marneffei or Cryptococcus neoformans in their BS or CSF were also treated with antifungal drugs. There was no significant difference in disease outcome with differing antimicrobial treatment regimens (Table 4). The median time from hospitalization to the use of an antimicrobial was 2.9 days (IQR 0–3 days); patients with a fatal outcome received an antimicrobial significantly earlier than those with non-fatal disease with a median of two days after hospitalization in the fatal group compared to 3.5 days in the non-fatal group (p = 0.01; Kruskal-wallis test).
We performed MLST on the complete collection of 142 iNTS isolates cultured in this center between January 2008 and June 2013; the resulting minimum spanning tree of these data is shown in Fig 1a. We were able to identify 17 different serovars by MLST that were associated with iNTS disease in this population. The most common serovars causing invasive disease were S. Enteritidis (ST11) and S. Typhimurium (STs 19, 34 and 1544), which were responsible for 63/147 (43%) and 44/147 (30%) of all cases, respectively. S. Typhimurium was identified more frequently in HIV-infected patients (Table 1) (p = 0.003, Fisher’s exact test). The remaining organisms (n = 40) were a combination of less commonly isolated Salmonella serovars including S. Choleraesuis (n = 14), S. Stanley (n = 3) and S. Weltevreden (n = 1) (Fig 1a), which were generally identified in nonfatal cases (Table 1).
We next compared antimicrobial susceptibility profiles between S. Enteritidis, S. Typhimurium and the remaining serovars (Fig 1b). The susceptibility profiles varied between the three different etiological groups and we found that S. Typhimurium were significantly more likely to exhibit resistance against ampicillin, amoxicillin, chloramphenicol and trimethoprim-sulfamethoxazole than S. Enteritidis and the remaining serovars (p<0.05 for all pairwise comparisons, Fisher’s exact test) (Fig 1b). Further, >50% of S. Typhimurium isolates were resistant to 6/10 antimicrobials tested (including ciprofloxacin and gentamicin); the same was true for 2/10 tested antimicrobials with the S. Enteritidis isolates and 5/10 antimicrobials with the other iNTS isolates. The majority of iNTS were susceptible to both azithromycin and third generation cephalosporins, with the exception of a single Extended Spectrum Beta Lactamase (ESBL) producing S. Choleraesuis. After PCR amplification and sequencing we found this particular S. Choleraesuis isolate to harbor a blaCTX-M-55 ESBL gene. Despite differences in AMR profiles we found no significant difference in mortality between those infected with S. Typhimurium and S. Enteritidis (p = 0.431, Fisher’s exact test). Lastly, isolates from HIV–infected patients were significantly more likely to exhibit reduced susceptibility against trimethoprim-sulfamethoxazole (28/62 45%) compared to HIV-uninfected patients (6/28, 21%) (p = 0.037, Fisher’s exact test).
NTS pathogens are a leading cause of community acquired bloodstream infections in parts of sub-Saharan Africa [3,4]. In sub-Saharan Africa the disease is concentrated in children and HIV-infected adults and complicated by the recent emergence and dominance of ST313, a multidrug resistant (MDR) variant of S. Typhimurium [25]. It was not known whether similar epidemiological patterns and iNTS sequence types existed in Asia. We report that iNTS infections are not as common in this setting in comparison to parts of sub-Saharan Africa, but similarly the disease is associated with immunocompromised adults and primarily caused by the serovars S. Enteritidis and S. Typhimurium.
Recent evaluations in sub-Saharan Africa have highlighted that the emergence of iNTS has been largely driven in adults by the HIV epidemic, while malnutrition and malaria infection are heavily associated with iNTS in children [3,4]. The overall incidence of iNTS infections in Southeast Asia is limited compared to that of Africa [26,27], though similar to the African context we confirm that HIV infection is the primary risk factor for iNTS disease in adults in Vietnam. The overall prevalence of HIV infection is low in Vietnam (0.5%) [28], yet it is known that disease is common in adults using intravenous drugs [29]. Indeed, HIV positive individuals in our study were likely to be male between the ages of 28–37 years. Therefore, iNTS disease should be considered as a possible etiology for febrile HIV-infected individuals.
Through MLST testing we found that approximately 75% of iNTS organisms were either S. Enteritidis or S. Typhimurium, which is consistent with the organisms causing iNTS disease in Africa [5,6], and the predominant organisms found in non-invasive NTS infections in this setting [30]. Although S. Typhimurium isolates were more likely to exhibit resistance against antimicrobials than other serovars, we did not identify the MDR S. Typhimurium clone ST313, which appears to have replaced resident NTS strains in sub-Saharan Africa [25]. The sequence types identified in our setting, namely S. Enteritidis ST11 and S. Typhimurium ST19 and ST34, have been found in invasive infections in Africa previously [5,25,31,32] while S. Typhimurium ST1544 has recently been isolated from food samples from China [33]. We additionally identified 14 S. Choleraesuis isolates, a serovar known to be associated with the consumption of pork products [34], and previously shown to be a cause of bacteremia in Taiwan [35]. We surmise that it is likely that organisms causing gastroenteritis in immunocompetent individuals may be comparable to those causing iNTS disease in immunocompromised patients in Vietnam. Continuing efforts to improve food safety and hygiene may have a positive effect on reducing both non-invasive and iNTS disease in our setting, though such interventions are costly and may be difficult to sustain in an industrializing setting like Vietnam [36].
Over one quarter of patients with iNTS disease either died in hospital or were discharged to die at home with family. This mortality rate is similar to the African context and confirms the severity of this infection in an immunocompromised population. The primary risk factor for death in our population was HIV infection, confirming trends identified in adults in sub-Saharan Africa [3,4]. Though we did not have CD4 cell counts available, iNTS disease is known to be a major risk factor for death in patients with advanced HIV disease [37]. As only 22% of HIV-infected patients were on active ART at the time of admission, improving access to ART would likely prevent the number of iNTS cases in Vietnam.
The majority of patients received ceftriaxone either in mono or combination therapy. Current susceptibility profiles confirm this is an appropriate choice, however high existing resistance against a variety of antimicrobials including ampicillin, chloramphenicol and ciprofloxacin signal the propensity for Salmonellae organisms to acquire a variety of resistance mechanisms. High levels of antimicrobial resistance in S. Typhimurium is cause for concern, particularly as HIV-infected patients were most often diagnosed with this serovar and the presence of resistance could further complicate management. Attempting to identify whether such antimicrobial resistance is related to food consumption and the excessive use of antimicrobials in animal husbandry known to occur in Vietnam is warranted [38]. Furthermore, the significant elevation of trimethoprim-sulfamethoxazole resistance amongst HIV-infected patients suggests that pneumocystis prophylaxis with the drug leads to colonization by resistant organisms. These data indicate that reduced antimicrobial susceptibility may not purely arise in animals in zoonotic pathogens, further work regarding the use of specific antimicrobials is animals is justified.
Our study has several limitations. First, children with HIV are generally referred to one of two large local pediatric hospitals so it is likely the burden of iNTS disease in children substantially underestimated. Though HIV is a risk factor for iNTS in children in Kenya [39], malnutrition and malaria infection are also important risks in children in the sub-Saharan African context [9,40]; future work in an Asian context should examine the epidemiology of pediatric iNTS more thoroughly. Secondly, our retrospective analysis for risk of death may be biased by misclassification as we coded patients who were taken home by family members as fatal, though we did not have a confirmed death report from these individuals. Notwithstanding these limitations of a retrospective study our work provides the largest description to date of iNTS patients to date in Southeast Asia and highlights important similarities and differences between the African and Asian settings. We suggest that continued surveillance, including sequence typing/whole genome sequencing, should be performed to monitor for emergence or introduction of MDR strains or strains with any apparent enhanced virulence phenotype, such as ST313 [41].
We conclude that iNTS disease is a severe infection in Vietnam, with a mortality rate (26%) similar to that of sub-Saharan Africa. We also highlight HIV infection as the major risk for both infection and death in this setting. Though the sequence types of iNTS organisms identified in this study are common globally, we suggest continued surveillance to monitor for the presence of MDR sequence types, such as ST313, which has not, as of yet, been identified in Asia.
|
10.1371/journal.ppat.1000901 | Bordetella Adenylate Cyclase Toxin Mobilizes Its β2 Integrin Receptor into Lipid Rafts to Accomplish Translocation across Target Cell Membrane in Two Steps | Bordetella adenylate cyclase toxin (CyaA) binds the αMβ2 integrin (CD11b/CD18, Mac-1, or CR3) of myeloid phagocytes and delivers into their cytosol an adenylate cyclase (AC) enzyme that converts ATP into the key signaling molecule cAMP. We show that penetration of the AC domain across cell membrane proceeds in two steps. It starts by membrane insertion of a toxin ‘translocation intermediate’, which can be ‘locked’ in the membrane by the 3D1 antibody blocking AC domain translocation. Insertion of the ‘intermediate’ permeabilizes cells for influx of extracellular calcium ions and thus activates calpain-mediated cleavage of the talin tether. Recruitment of the integrin-CyaA complex into lipid rafts follows and the cholesterol-rich lipid environment promotes translocation of the AC domain across cell membrane. AC translocation into cells was inhibited upon raft disruption by cholesterol depletion, or when CyaA mobilization into rafts was blocked by inhibition of talin processing. Furthermore, CyaA mutants unable to mobilize calcium into cells failed to relocate into lipid rafts, and failed to translocate the AC domain across cell membrane, unless rescued by Ca2+ influx promoted in trans by ionomycin or another CyaA protein. Hence, by mobilizing calcium ions into phagocytes, the ‘translocation intermediate’ promotes toxin piggybacking on integrin into lipid rafts and enables AC enzyme delivery into host cytosol.
| The adenylate cyclase toxin (CyaA) of pathogenic Bordetellae eliminates the first line of host innate immune defense. It penetrates myeloid phagocytes, such as neutrophils, macrophage or dendritic cells, and subverts their signaling by catalyzing an extremely rapid conversion of intracellular ATP to the key signaling molecule cAMP. This efficiently inhibits the oxidative burst and complement-mediated opsonophagocytic killing of bacteria, thus enabling the pathogen to colonize host airways. We show that translocation of CyaA into phagocyte cytosol occurs in two steps. The toxin first binds the integrin CD11b/CD18 and inserts into phagocyte membrane to mediate influx of calcium ions into cells. This promotes relocation of the toxin-receptor complex into specific lipid microdomains within cell membrane called rafts. The increased concentrations of cholesterol within rafts and their particular lipid organization then support translocation of the adenylate cyclase enzyme directly into the cytoplasmic compartment of cells. The mechanism of CyaA penetration into cells sets a new paradigm for membrane translocation of toxins of the RTX family.
| The secreted adenylate cyclase toxin-hemolysin (CyaA, ACT, or AC-Hly) plays a key role in virulence of Bordetellae. This multifuctional protein binds the αMβ2 integrin (CD11b/CD18, CR3 or Mac-1) of myeloid phagocytic cells and delivers into their cytosol a calmodulin-activated adenylate cyclase enzyme that ablates bactericidal capacities of phagocytes by uncontrolled conversion of cytosolic ATP to the key signaling molecule cAMP [1]–[5]. In parallel, the hemolysin moiety of CyaA forms oligomeric pores that permeabilize cell membrane for monovalent cations and contribute to overall cytoxicity of CyaA towards phagocytes [6]–[10].
The toxin is a 1706 residues-long protein, in which a calmodulin-activated adenylate cyclase (AC) enzyme domain of ∼400 N-terminal residues is fused to a ∼1300 residue-long RTX (Repeats in ToXin) cytolysin moiety [11]. The latter consists itself of three functional domains typical for RTX hemolysins. It harbors, respectively, (i) a hydrophobic pore-forming domain, (ii) a segment recognized by the protein acyltransferase CyaC, activating proCyaA by covalent post-translational palmitoylation at ε-amino groups of Lys860 and Lys983 [12], [13], and (iii) an assembly of five blocks of the characteristic glycine and aspartate-rich nonapeptide RTX repeats that form numerous (∼40) calcium-binding sites [14].
Since no structural information on the RTX cytolysin moiety is available, the mechanistic details of toxin translocation across the lipid bilayer of cell membrane remain poorly understood. Delivery of the AC domain into cells occurs directly across the cytoplasmic membrane, without the need for toxin endocytosis [15] and requires structural integrity of the CyaA molecule [6], unfolding of the AC domain [16] and a negative membrane potential [17]. Recently, we described that CyaA forms a calcium-conductive path in cell membrane and mediates influx of extracellular Ca2+ ions into cell cytosol concomitantly with translocation of the AC domain polypeptide into cells [18].
The current working model predicts that both Ca2+ influx and AC translocation depend on a different membrane-inserted CyaA conformer than the pore-forming activity [19], [20]. The two membrane activities of CyaA, however, appear to use the same essential amphipatic transmembrane segments within the pore-forming domain (α-helix502–522 and α-helix565–591), employing them in an alternative and mutually exclusive way. These segments harbor two pairs of negatively charged glutamate residues (Glu509/Glu516 and Glu570/Glu581) that were found to play a central role in toxin action on cell membrane. These control, respectively, the translocation of the positively charged AC domain, the formation of oligomeric CyaA pores and the cation-selectivity of the CyaA pore. Charge-reversing, neutral or helix-breaking substitutions of these glutamates were found to shift the balance between AC translocating and pore-forming activities of CyaA on cell membrane [8], [10], [19], [20].
The very high specific AC enzyme activity of CyaA allowed previously to detect its capacity to promiscuously bind and penetrate at reduced levels also numerous cell types lacking the CD11b/CD18 receptor [21], [22]. This is likely due to a weak lectin activity of CyaA, which would enable interaction of the toxin with cell surface gangliosides [23] and glycoproteins [24]. Indeed, binding of CyaA to CD11b/CD18 was recently found to depend on initial interaction with the N-linked glycan antenna of the receptor [24], where the specificity of CyaA for CD11b/CD18 appears to be determined by a segment of the stalk domain of the CD11b subunit (Osicka et al., manuscript in preparation).
CD11b/CD18 belongs to the β2 subfamily of polyfunctional integrins playing a major role in leukocyte function. The same β2 subunit (CD18) can, indeed, pair with four distinct α subunits to yield the αLβ2 (CD11a/CD18, LFA-1), αMβ2 (CD11b/CD18, CR3, Mac1), αXβ2 (CD11c/CD18, p150/195) and αDβ2 (CD11d/CD18) receptors, respectively [25]. Among key features of these integrins is their capacity of bi-directional signaling, where the avidity and conformation of the integrins is regulated by intracellular signals in the ‘inside-out’ signaling mode. In turn, binding of ligands or counter-receptors results in ‘outside-in’ signaling [26]. Among other effects, the latter yields actin cytoskeletal rearrangements and can result in lateral segregation of the β2 integrins from the bulk phase of the plasma membrane into distinct lipid assemblies known as lipid rafts [27], [28]. These were first detected as detergent-resistant membrane (DRM), characterized by insolubility in some detergents under certain conditions and enriched in cholesterol, sphingolipids, and glycosylphosphatidylinositol-anchored proteins [29]–[32]. Besides playing an important role in signal transduction, receptor internalization, vesicular sorting or cholesterol transport [33], the components of lipid rafts are often exploited as specific receptors mediating cell entry of toxins, pathogenic bacteria, or viruses [34]–[36].
Here, we show that CyaA-mediated influx of Ca2+ ions into cells induces mobilization of the toxin-receptor complex into lipid rafts, where translocation of the AC domain across cytoplasmic membrane is accomplished.
To examine whether CyaA localizes to lipid rafts, murine J774A.1 monocytes exposed to 1 nM CyaA (176 ng/ml, 37°C, 10 min) were lyzed with ice-cold Triton X-100 and detergent-resistant membrane (DRM) was separated from soluble cell extracts by flotation through sucrose density gradients. As shown in Fig. 1A, while the CD71 marker of bulk membrane phase was exclusively detected in the soluble extract at the bottom of the gradient, up to 30% of total loaded CyaA was found to float in fraction 3 at a lower buoyant density towards the top of the gradient, together with the DRM marker protein NTAL (see Fig. 1E for quantification).
Notably, while over 50% of the full-length CyaA molecules (∼200 kDa) remained in the soluble phase at the bottom of the gradient, the floating DRM fractions were selectively enriched in a processed CyaA form of ∼160 kDa, representing up to 60% of total CyaA in the fraction 3 of the gradient (see Fig. 1F for quantification). This appeared to have the entire AC domain-cleaved off, as it could only be detected by the 9D4 antibody recognizing the C-terminal RTX repeats and not by the 3D1 antibody binding between residues 373 and 399 of the C-terminal end of the AC domain of CyaA [37]. Most CyaA molecules accumulating in DRM appeared, hence, to have the AC domain translocated across cellular membrane and accessible to processing by intracellular proteases [10], [38].
As further documented in Fig. 1B, no CD11b was floating with DRM from mock-treated J774A.1 cells, or when toxin binding occurred at 4°C. In turn, exposure of monocytes to 1 nM CyaA at 37°C, resulted in mobilization of over 30% of total cellular CD11b into the floating DRM, showing that CyaA relocated from the bulk of the membrane to DRM together with its αMβ2 integrin receptor. This was, however, not due to any generalized clustering and mobilization of β2 integrins into rafts resulting from toxin action, as the highly homologous CD11a subunit of the other β2 integrin expressed by J774A.1 cells (LFA-1), was not mobilized into DRM (Fig. 1B). Hence, the relocation of CD11b/CD18 into rafts was specifically due to interaction with CyaA.
To assess whether CyaA association with DRM depended on toxin interaction with CD11b/CD18, we used Chinese hamster ovary (CHO) cells that do not express any β2 integrins unless transfected by genes encoding CD11b and CD18 subunits (CHO-CD11b/CD18). As shown in Fig. 1C, even when the CyaA concentration was raised to 113 nM, to obtain detectable amounts of the toxin associated with mock-transfected CHO cells, CyaA was detected exclusively in the soluble extract at the bottom of the gradient. In contrast, association of CyaA and CD11b/CD18 with DRM was detected already upon treatment of CHO-CD11b/CD18 transfectants with 1 nM CyaA (Fig. 1D). This showed that CyaA depended on binding to CD11b/CD18 for association with DRM and it was able to mobilize CD11b/CD18 into DRM independently of the myeloid cell background.
Since CyaA exerts several activities on cells in parallel, we analyzed which of them enabled mobilization of the CyaA-CD11b/CD18 complex into DRM. Towards this aim, we used a specific set of CyaA variants that retain the capacity to bind CD11b/CD18, while lacking one or more of the other CyaA activities (Table 1). As documented in Fig. 2A, the capacity of CyaA to elevate cellular cAMP concentrations was not required for mobilization of CyaA into DRM. The enzymatically-inactive CyaA-AC− construct, unable to catalyze conversion of ATP into cAMP, was indeed accumulating in DRM with the same efficacy as the intact CyaA (fractions 3–4). Fatty-acylation of CyaA as such was also not essential for association of CyaA with DRM. As further shown in Fig. 2A, the non-acylated proCyaA was detected in DRM despite an importantly reduced capacity to associate with cells. Moreover, the pore-forming activity of CyaA was both insufficient and dispensable for mobilization of CyaA into DRM. The acylated CyaAΔAC construct, lacking the entire AC domain but retaining an intact pore-forming (hemolytic) capacity, was unable to mobilize into DRM (Fig. 2A). In contrast, the CyaA-E570Q+K860R-AC− construct unable to permeabilize cells to any significant extent, but exhibiting an intact capacity to translocate the AC domain across cell membrane was, indeed, recruited into DRM together with CD11b/CD18 as efficiently as intact CyaA. In turn, the CyaA-E570K+E581P double mutant that was unable to form CyaA pores, or to translocate the AC domain across membrane, and retained only the CD11b-binding capacity (Table 1), was also unable to associate with DRM.
To corroborate these observations, we used fluorescence microscopy to examine the distribution of individual CyaA proteins in cell membrane. As documented in Fig. 2B, the intact CyaA, CyaA-AC− and CyaA-E570Q+K860R-AC− proteins were found to induce formation of, and to localize within, patches on cell membrane. Moreover, the same patches were labeled to high extent also with B subunit of cholera toxin (CtxB), which specifically binds the GM1 ganglioside accumulating in lipid rafts. Hence, the three CyaA variants capable of associating with DRM (cf. Fig. 2A) were also found to co-localize with CtxB within membrane patches. In turn, no formation of membrane patches, a diffuse distribution on cell surface, and low if any co-localization with CtxB, were observed for the CyaAΔAC and CyaA-E570K+E581P constructs that were unable to associate with DRM, too.
The pattern of DRM association, processing to the 160 kDa form and co-localization of the different CyaA variants with CtxB, respectively, resembled strongly the pattern of structure-function relationships observed recently for the capacity of CyaA to promote influx of extracellular calcium ions into J774A.1 cells [18]. Indeed, as documented in Fig. 2C by measurements of intracellular calcium concentrations ([Ca2+]i), the CyaA, CyaA-AC− and CyaA-E570Q+K860R-AC−proteins (17 nM) exhibited an expected capacity to promote Ca2+ influx into J774A.1 cells (see [18] for details on different kinetics of Ca2+ entry for AC− and AC+ constructs). In contrast, the CyaA-E570K+E581P and CyaAΔAC constructs, failed to mediate any increase of [Ca2+]i even when used at a 113 nM concentration (Fig. 2D). Collectively, hence, these results show that the capacity of different CyaA variants to associate with DRM and co-localize with CtxB within coalesced lipid rafts was mirrored by the capacity to promote Ca2+ influx into cells.
We showed recently that CyaA-mediated influx of Ca2+ into cells is independent of the AC enzyme or pore-forming (hemolytic) activities of CyaA and occurs concomitantly to translocation of the AC domain across target cell membrane [18]. It remained, however, to assess whether it was the mere insertion of a CyaA translocation precursor into cell membrane, or whether the accomplishment of translocation of the AC domain across cell membrane was required for formation of a calcium conductive path in cell membrane. Towards this aim, we used the 3D1 monoclonal antibody (MAb) that binds to the distal end of the AC domain (residues 373 to 400) and was previously shown to block membrane translocation of the AC domain of cell-associated CyaA [39]. As expected and documented in Fig. 3A, preincubation of CyaA with the 3D1 MAb did not affect the capacity of CyaA to bind J774A.1 cells, while strongly inhibiting AC domain delivery and cAMP concentration elevation in cells. However, as revealed by detection of both CyaA and 3D1 in the fraction 3 of the sucrose gradient shown in Fig. 3B, the membrane-inserted CyaA with bound 3D1 MAb still associated with DRM at the same levels as CyaA alone. Moreover, as also shown in Fig. 3B, due to 3D1-mediated inhibition of AC domain translocation, the relative amount of processed CyaA (∼160 kDa) in the DRM fractions decreased to 10 to 15% of total present CyaA, while over 50% of CyaA in DRM was processed in the presence of isotype control.
As shown in Fig. 3C, however, despite arrested AC domain translocation, the CyaA-3D1 complex promoted elevation of [Ca2+]i in cells with kinetics resembling the Ca2+ influx produced by CyaA-AC− (cf. Fig. 2C). Hence, 3D1 binding uncoupled translocation of the AC domain from membrane insertion of CyaA and ‘locked’ the toxin in the conformation of a ‘translocation intermediate’ that permeabilized cells for Ca2+ ions and associated with DRM.
Next, we aimed to determine whether elevation of [Ca2+]i as such would mobilize into DRM also the CyaA-E570K+E581P protein unable to associate with DRM on its own (cf. Fig. 2). As demonstrated in Fig. 4, upon permeabilization of cells for extracellular Ca2+ ions with the Ca2+ ionophore ionomycin (500 nM), up to 15% of the added CyaA-E570K+E581P was found associated with DRM. In contrast, no association of CyaA-E570K+E581P with DRM was observed upon treatment of cells with 1 µM thapsigargin that increases [Ca2+]i by triggering Ca2+ release from intracellular stores. This showed that entry of extracellular Ca2+ across the cytoplasmic membrane was required for mobilization of CyaA into DRM.
Influx of extracellular Ca2+ during leukocyte activation was reported to induce mobilization of integrins in cell membrane by calpain-mediated cleavage of talin that tethers β2 integrins to actin cytoskeleton [40], [41]. Therefore, we examined whether CyaA-promoted recruitment of the toxin receptor into rafts depended on talin processing. As shown in Fig. 5A, intact talin (∼270 kDa) was largely predominating in lyzates of cells treated with the CyaA-ΔAC or CyaA-E570K-E581P proteins that are unable to promote Ca2+ entry into cells. In contrast exposure of cells to the CyaA, CyaA-AC−, or CyaA-E570Q+K860R-AC− proteins, promoting influx of Ca2+ into cells, increased about seven-fold the detected amounts of the ∼220 kDa C-terminal fragment of processed talin (Fig. 5A, left panel). Concomitantly, increased amounts of the 47-kDa N-terminal fragment of talin (talin head) were detected in cell lyzates. Moreover, tightly associated talin head was found to float together with the CD11b/CD18 heterodimer in DRM (Fig. 5B) and could be co-immunoprecipitated with the integrin on beads coated with anti-CD11b antibody (Fig. 5C). This CyaA-induced processing of talin was clearly due to activation of calpain, as preincubation of cells with 100 µM calpain inhibitor, calpeptin, blocked talin cleavage in CyaA-treated cells (Fig. 5A, right panel). Remarkably, pretreatment of cells with calpeptin strongly inhibited also the association of CyaA with DRM (Fig. 5D) and decreased by at least a factor of two the capacity of cell-associated CyaA to translocate the AC enzyme into target cells (Fig. 5E). In turn, no effect of calpain inhibition was observed for CyaA-mediated Ca2+ influx (Fig. 5F). In line with these results, pretreatment of cells with 100 µM calpeptin blocked effectively also the formation of CyaA-AC− patches in cell membrane and ablated co-localization of CyaA-AC− with CtxB, as documented in Fig. 5G. It can, hence, be concluded that CyaA-mediated influx of Ca2+ into cells activated cleavage of talin by calpain and this was required for mobilization of CyaA-CD11b/CD18 complexes into lipid rafts.
To determine what role does association of CyaA with lipid rafts play in the mechanism of toxin action on cellular membrane, we analyzed the activities of CyaA on cells having the rafts disrupted by depletion of cholesterol. As shown in Table 2, the total cholesterol content of J774A.1 cells could be decreased about two-fold by cholesterol extraction with 10 mM MβCD for 30 min. While the disruption of raft structures did not impact on association of CyaA with cells (Fig. 6A and Fig. S1), the modest decrease of cellular cholesterol content yielded an about five-fold decrease of the capacity of CyaA to translocate the AC domain across cell membrane. This defect was further mirrored by decreased DRM association of CyaA in MβCD-extracted cells, as shown in Fig. 6B. In parallel, the specific capacity of CyaA to promote Ca2+ influx into cholesterol-depleted cells was reduced and the [Ca2+]i increase ensuing toxin addition was delayed by several minutes, reaching a plateau at about a half-maximal [Ca2+]i concentration, as compared to non-depleted cells (Fig. 6C). In line with this, the two-fold decrease of cellular cholesterol level moderately decreased also the co-localization of CyaA with CtxB in lipid rafts (Fig. 6D).
Therefore, the above described experiments were replicated on monocytic U937 histiocytic lymphoma cells (CD11b+) that are defective in endogenous cholesterol synthesis. These cells can be efficiently depleted of cholesterol without losing viability, by them growing for 48 hours in media containing cholesterol-free (delipidated) serum. As shown in Table 2, such treatment reduced the cholesterol content of U937 cells almost 10-times.
As shown in Fig. 6E, a pronounced, over ten-fold decrease of specific AC translocation capacity of CyaA was observed on U937 cells grown in media with delipidated serum, as compared to CyaA activity on cells grown with standard serum. At the same time, however, the total amounts of cell-associated CyaA remained equal, irrespective of cell treatment. However, by difference to well-detectable DRM association of CyaA on cholesterol-replete U937 cells, grown with standard serum, no association of CyaA with DRM was observed in lyzates of cholesterol-depleted U937 cells grown in delipidated serum, respectively (Fig. 6F).
Intriguingly, compared to the Ca2+ influx elicited by equal concentrations of CyaA in J774A.1 cells, about an order of magnitude lower amplitude and delayed kinetics of CyaA-mediated Ca2+ influx was observed for U937 cells grown in media with standard serum (cf. Fig. 6C and Fig. 6G). These cells exhibited a 3-fold lower cholesterol content than the J774A.1 cells (see Table 2), suggesting that the low cholesterol content of U937 cells might have accounted for the poor capacity of CyaA to elicit Ca2+ influx in these cells. Indeed, when cholesterol content of J774A.1 cells was reduced about two-fold by cholesterol extraction with 10 mM MβCD, a delayed kinetics of CyaA-induced influx of Ca2+ into J774A.1 cells and a two-fold lower final [Ca2+]i reached in 20 minutes, were also observed (cf. Fig. 6C). Similarly, a delayed influx of Ca2+ and a lower final level of [Ca2+]i was observed also upon addition of equal CyaA concentrations to U937 cells depleted of cholesterol by growth in delipidated media, as compared to U937 cells grown in standard media, as shown in Fig. 6G. At the same time, however, the respective amounts of CyaA associated per 106 J774A.1 or U937 cells remained the same (∼5 ng of CyaA bound per 106 cells), irrespective of whether the cholesterol content of cells was decreased by the treatments (cf. Fig. 6A and 6E). These results, hence, strongly point towards a close relation between the overall content of cholesterol in cellular membrane and the propensity of the membrane-inserted CyaA to adopt the ‘translocation intermediate’ conformation, which would account for the Ca2+ conducting path across cell membrane (cf. Fig. 3 and [18]).
Finally, a correspondingly reduced CtxB binding and little if any co-localization of CtxB with CyaA were observed on cholesterol-depleted U937 cells, grown in delipidated serum, as compared to binding and some observable co-localization of CyaA with CtxB on cholesterol-replete U937 cells (Fig. 6).
In the light of the above results, we aimed to test the hypothesis that AC translocation across membrane was supported and accomplished upon recruitment of the membrane-associated toxin into the cholesterol-rich environment of lipid rafts. Therefore, we examined whether the inactive CyaA-E570K+E581P construct would gain any capacity to translocate its enzymatically active AC domain across cellular membrane upon mobilization into lipid rafts. Since this mutant is intact for receptor binding but fails to promote Ca2+ influx into cells, we reasoned that mobilizing Ca2+ ions into cells in trans, by co-incubation with a translocating CyaA-AC− toxoid, might promote recruitment of CyaA-E570K+E581P mutant into rafts to some extent.
As shown in Fig. 7A, when biotinylated CyaA-E570K+E581P was added to cells alone, or when it was co-incubated with equal amounts of the enzymatically inactive CyaA-E570K+E581P-AC− toxoid, unable to cause calcium influx, the CyaA-E570K+E581P-biotin failed to associate with DRM. In contrast, upon co-incubation with equal amounts of the translocating CyaA-AC− toxoid (1∶1), a significant fraction of CyaA-E570K+E581P-biotin associated with DRM. Moreover, as shown in Fig. 7B, this mobilization into DRM was paralleled by a doubling of the residual capacity of the CyaA-E570K+E581P variant to deliver the AC domain across cell membrane and elevate cytosolic cAMP concentrations (Fig. 7B). Thus, recruitment into cholesterol-rich lipid rafts enhanced the residual AC translocating activity of this defective CyaA variant.
We show here that membrane translocation of the adenylate cyclase domain of CyaA occurs by a two step mechanism and involves toxin piggybacking on the αMβ2 integrin for relocation into lipid rafts. The present results allow us to propose a new model of CyaA mechanism of action, as summarized in Fig. 8. Upon initial binding of CyaA to the CD11b/CD18 receptor distributed in the bulk phase of cell membrane, a ‘translocation intermediate’ of CyaA would insert into the cytoplasmic membrane. It is assumed that in this ‘translocation intermediate’ a part of the AC domain is already inserted within the membrane and is shielded form the lipids by association with the amphipathic α-helical transmembrane segments of the hydrophobic domain of CyaA (residues 502–522, 529–549, 571–591, 607–627 and 678–698 [19], [20]). This ‘translocation intermediate’ then forms a path conducting external Ca2+ ions across cellular membrane into the submembrane compartment of cells. Incoming calcium ions activate the Ca2+-dependent protease calpain, located in the submembrane compartment, which produces cleavage of the talin tether. This liberates the toxin-receptor complex from association with actin cytoskeleton and mobilizes it for recruitment into lipid rafts. Within the specific liquid-ordered environment of cholesterol-rich lipid rafts, translocation of the positively charged AC domain across the cellular membrane is completed, driven by the negative gradient of membrane potential.
Deciphering this fine-tuned mechanism of toxin action on cell membrane fosters our understanding of the key role played by CyaA in virulence of Bordetellae during the early phases of bacterial colonization of host respiratory mucosa. It allows to propose the following scenario. The produced CyaA targets the CD11b/CD18 receptor of incoming myeloid phagocytic cells, such as neutrophils, macrophages and dendritic cells [42]. As CyaA action does not depend on receptor-mediated endocytosis, the toxin recruited into lipid rafts can rapidly translocate its highly active AC enzyme domain across the cytoplasmic membrane of cells, in a process exhibiting a half-time of only about ∼30 seconds [38]. Mobilization of toxin-receptor complexes into lipid rafts than promotes their clustering and potentially induces recruitment of cellular cAMP-responding elements, such as the protein kinase A anchored to AKAPs, the specific A-kinase anchoring scaffolds [43]–[45]. This would allow maximization of toxin action through subversive cAMP production in close vicinity of components of the cAMP-regulated PKA signaling pathway. This capacity to hijack the spatio-temporal regulation of cellular cAMP/PKA signaling would then endow CyaA with the high potency in paralyzing the central bactericidal mechanisms employed by myeloid phagocytic cells. Indeed, few picomoles of CyaA (1 ng/ml or less) were previously reported to instantaneously suppress the oxidative burst capacity of neutrophils [46], or the phagocytosis of complement-opsonized particles by macrophages [2].
Several other bacterial protein toxins appear, indeed, to utilize lipid rafts as a portal of cell entry, exploiting as specific receptors directly certain raft components, such as cholesterol, sphingolipids or GPI-anchored proteins [47]–[50]. In contrast, we found here that CyaA associates with rafts only upon binding and mobilization (hijacking) of its receptor CD11b/CD18. Unless activated in the process of leukocyte activation, this β2 integrin is distributed diffusely over the entire cellular membrane. As outlined above, we show here that upon binding of CyaA the integrin relocates into lipid rafts, due to toxin-induced and calcium-activated cleavage of talin by calpain. Moreover, the recently discovered capacity of CyaA to bind N-linked oligosaccharides of CD11b/CD18 [24] might also play a role in this process. It is, indeed, plausible to propose that CyaA interaction with terminal sialic acid residues of glycan chains of raft sphingolipids might also be contributing to accumulation of the CyaA-CD11b/CD18 complex in lipid rafts, as well as it may contribute to clustering of lipid rafts containing CyaA later-on. An evidence for CyaA interactions with gangliosides can, indeed, be deduced from the previously observed inhibition of CyaA activity on macrophages by the presence of micromolar concentrations of free gangliosides, such as GT1b [23].
It remains to be addressed in future studies if CyaA can form oligomeric pores also once engaged in interaction with the target cell membrane through binding of the CD11b/CD18 receptor and whether CyaA can form pore-forming oligomers also in phagocyte membrane. We have recently succeeded in demonstrating the presence of the long-predicted CyaA oligomers within the membrane of cells lacking the receptor CD11b/CD18, such as erythrocytes [10]. Vojtova-Vodolanova with co-authors (2009), indeed, showed that formation of CyaA oligomers underlies the pore-forming activity of CyaA towards erythrocytes. However, despite significantly higher amounts of CyaA binding per single phagocyte cell through the CD11b/CD18 receptor, fairly high concentrations (>1 µg/ml) of the recombinant enzymatically inactive but fully pore-forming CyaA-AC− variants are needed to provoke lysis of cells like J774A.1 monocytes in several hours [8]. While this resistance to colloid-osmotic lysis is likely to be to large extent due to membrane recycling mechanisms and pore removal form phagocyte cytoplasmic membrane, it remains to be shown that CyaA can form oligomeric pores in leukocyte membrane as well.
The results presented here do not indicate any role of CyaA oligomers in promoting calcium influx, toxin mobilization into rafts, or AC enzyme translocation into CD11b+ phagocytes. Early dose-dependence studies indicated that the AC domain was delivered across target cell membrane by CyaA monomers. Indeed, toxin molecules with the AC domain cleaved-off by cytosolic proteases, upon AC translocation into cells, were detected exclusively in form of CyaA monomers within erythrocyte membranes and were excluded from the detected CyaA oligomers [10]. Moreover, we used here the CyaA-E570Q+K860R-AC− protein, which essentially lacks any pore-forming activity and fails to permeabilize the membrane of J774A.1 cells, thus being unlikely to form any CyaA oligomers (Table 1 and [51]). On the other hand, this construct is fully capable to translocate the AC domain into cytosol of CD11b-expressing J774A.1 cells, to promote calcium influx and to associate with DRM, or to co-localize with CtxB in coalesced rafts, respectively (cf. Fig. 2). It appears, therefore, unlikely that oligomerization plays a role in DRM association of CyaA.
We also observed here that the levels of binding of CyaA to CD11b-expressing cells were not affected upon cholesterol depletion of cell membrane, while the translocation of the AC domain across the membrane depended strongly on the cholesterol content. This suggests that by modulating the physical properties lipid bilayers, cholesterol was specifically supporting the translocation of the AC domain across cell membrane. Indeed, cholesterol removal was previously found to impair the residual penetration capacity of CyaA on artificial membranes and erythrocytes [52], [53]. This goes well with the impact of cholesterol concentration on membrane fluidity, lateral phase separation, formation of liquid-ordered structures and the propensity of lipids to adopt the inverted hexagonal phase [54], [55]. The same membrane properties would also be expected to support AC domain translocation into cells by lowering the energy barrier for polypeptide penetration into and across the lipid bilayer [56]. It is plausible to speculate that membrane translocation of the AC domain requires the presence of cholesterol-dependent liquid-ordered (lo) phase, in which the acyl chains of lipids are tightly packed, while the individual lipid molecules have a high degree of lateral mobility. The relative mobility of lipids in lo domains represents, indeed, a likely prerequisite for passage of the AC domain across lipid bilayer. A high condensation and immobility of lipids in liquid-disordered (ld)-phase domains would, in turn, be expected to interfere with AC polypeptide translocation. The requirement for sufficient membrane fluidity for AC translocation to occur is also indicated by the block of AC translocation at 4°C [38].
Recently, we demonstrated that AC domain translocation across target cell membrane is accompanied by entry of Ca2+ ions into cells. Moreover, the AC domain polypeptide as such was found to participate in formation of the transiently opened calcium influx path in cell membrane [18]. Here, we used the 3D1 MAb recognizing a distal segment of the AC domain and show that blocking of AC domain translocation across cell membrane can lock CyaA in a ‘translocation intermediate’ conformation that forms a path for Ca2+ influx across cell membrane (cf. Fig. 3). Moreover, this ‘translocation intermediate’ was found to be recruited into lipid rafts (Fig. 3). The sum of the data hence allows us to answer the question what happens first, whether calcium influx precedes toxin mobilization into rafts, or whether recruitment of CyaA into rafts precedes calcium influx and AC translocation.
We showed here that calpeptin-mediated inhibition of calcium-activated processing of talin by calpain yields (i) inhibition of CyaA recruitment into rafts and (ii) it inhibits AC translocation across membrane. Collectively, hence, these results strongly suggest that the transient influx of Ca2+ into cells accompanies the earliest step of membrane insertion of the toxin ‘translocation intermediate’. This would precede and be essential for subsequent recruitment of CyaA into lipid rafts, whereupon AC translocation is accomplished.
It remains, however, to be determined what is the threshold of the calcium signal required for initiation of talin cleavage and mobilization of CyaA into lipid rafts. Two major isoforms of calpain have, indeed, been so far identified in eukaryotic cells. The calpain I (μ-calpain) is activated at µM Ca2+ concentrations, while calpain II (m-calpain) only responds to mM concentrations of Ca2+ [57]. Here we observed that CyaA relocalization into DRM occurred at 1 nM toxin concentration, which is about two-times less than the lowest CyaA concentrations still allowing to elicit a [Ca2+]i increase detectable in cells by the Fura-2/AM probe [18]. Moreover, only influx of extracellular Ca2+ ions into cells, and not the elevation of cytosolic [Ca2+]i due to Ca2+ release from intracellular stores, enabled the accumulation of CyaA in DRM (cf. Fig. 4). This differs importantly from the mechanism reported for localization of the leukotoxin (LtxA) of Actinobacillus actinomycetemcomitans into rafts. LtxA binds yet another β2 integrin of human leukocytes, the LFA-1 or CD11a/CD18 heterodimer. Horeover, LtxA appears to first adsorb on cell membrane of T lymphocytes in a receptor-independent manner, to trigger, somehow the store-operated elevation of cytosolic [Ca2+]i, to induce talin cleavage, and upon relocation into rafts, the Ltx clusters with LFA-1 within rafts to promote cell lysis [58].
With CyaA, all the Ca2+ ions entering macrophage cytoplasm due to toxin action appear to come from extracellular medium [18]. It is generally accepted that there exists a gradient of about four orders of magnitude in Ca2+ concentrations between the external medium (∼2 mM) and cell cytosol (∼100 nM). Therefore, numerous Ca2+-buffering proteins accumulate beneath the inner face of cell membrane, accounting for formation of local Ca2+ gradients and controlling signaling induced by alterations of Ca2+ concentrations in the submembrane compartment. These concentrations can, indeed, be still much higher, and rise more rapidly, than the bulk Ca2+ levels in cell cytosol [59]. Therefore, it is likely that even an importantly lower CyaA concentration than used here (1 nM = 176 ng/ml), may still be generating sufficiently high local Ca2+ signal beneath cell membrane in order to promote activation of μ-calpain at the inner face of cell membrane. It appears, thus, plausible to assume that mobilization of CyaA into rafts in phagocyte membrane, and translocation of the AC domain from rafts directly into the cytosolic compartment of phagocytes, are indeed taking place also during natural Bordetella infections in vivo. This would account for the remarkable efficacy of CyaA in disarming the sentinel cells of the host innate defense.
Intact recombinant CyaA and its mutant variants were expressed and purified as previously described [19]. Except of pro-CyaA, the CyaA proteins were produced in E. coli XL1-Blue in the presence of the co-expressed toxin-activating acyltransferase CyaC, as previously described [20]. Lipopolysaccharide was eliminated by repeated 60% isoporopanol washes of CyaA bound to the Phenyl Sepharose resin [60]. This reduced the final endotoxin content below 50 EU/mg of purified protein, as determined by the Limulus amebocyte lyzate assay (QCL-1000, Cambrex, NJ, USA). For fluorescence microscopy, the CyaA proteins were labeled while bound to Phenyl-Sepharose resin during the final purification step. Briefly, the CyaA eluates from a DEAE-Sepharose columns (GE Healthcare) in 50 mM Tris-HCl (pH 8), 8 M urea, 0.2 mM CaCl2, 200 mM NaCl, were diluted 1∶4 with a buffer containing 50 mM Tris-HCl (pH 8), 1 M NaCl and 1 mg of CyaA was loaded on an 0.5 ml Phenyl-Sepharose column. The columns were extensively washed with 0.1 M sodium bicarbonate (pH 9), 1 M NaCl. Next 10 µg/ml Alexa Fluor 488 succinimidylester solution (Molecular Probes) was loaded and labeling proceeded at 25°C for 1 hour. The columns were washed with 50 mM Tris-HCl (pH 8), 1 M NaCl, and the CyaA-Alexa Fluor 488 conjugates were eluted in a buffer containing 50 mM Tris-HCl (pH 8), 8 M urea and 2 mM EDTA. Unreacted dye was separated from labeled CyaA on Sephadex G-25 columns (GE Healthcare). Efficiency of protein labeling was assessed spectrophotometrically and a molar ratio of about 1∶4 (protein∶dye) was found for all CyaA preparations. It was verified that this extent of labeling did not affect the biological activities of CyaA.
See Protocol S1 for full description.
Detergent-resistant membranes (DRM) were separated by flotation in discontinuous sucrose density gradients. Briefly, J774A.1 cells (2.107) were washed with prewarmed DMEM and incubated with 1 nM CyaA proteins at 37°C for 10 min. Cells were washed with ice-cold phosphate-buffered saline (PBS), scraped from the Petri dish and extracted at 4°C for 60 min using 200 µl of TBS buffer (20 mM Tris-HCl, pH 7.5, 150 mM NaCl) containing 1% Triton X-100, 1 mM EDTA, 10 mM NaF and a Complete Mini proteinase inhibitor cocktail (Roche, Basel, Switzerland). The lyzates were clarified by centrifugation at 250×g for 5 min and the post-nuclear supernatants were mixed with equal volumes of 90% sucrose in TBS. The suspensions were placed at the bottom of centrifuge tubes and overlaid with 2.5 ml of 30% sucrose and 1.5 ml of 5% sucrose in TBS. Membrane flotation according buoyant density was achieved by centrifugation at 150,000×g in a Beckman SW60Ti rotor for 16 h at 4°C. Fractions of 0.5 ml were removed from the top of the gradient.
Calcium influx into J774A.1 and U937 cells was measured as previously described [18]. Briefly, cells were loaded with 3 µM Fura-2/AM (Molecular Probes) at 25°C for 30 min and the time course of calcium entry into cells induced by addition 3 µg/ml of CyaA proteins was determined as ratio of fluorescence intensities (excitation at 340/380 nm, emmision 505 nm), using a FluoroMax-3 spectrofluorometer equipped with DataMax software (Jobin Yvon Horriba, France).
J774A.1 cells were incubated in DMEM supplemented with 10 mM methyl-β-cyclodextrin (MβCD) at 37°C for 30 min. Cholesterol-depleted U937 cells were obtained upon growth in RPMI medium supplemented with 10% of delipidated serum (lipoprotein-deficient serum from fetal calf, Sigma) for 48 h. Cholesterol content was determined using an Amplex Red Cholesterol Assay Kit (Molecular Probes, Invitrogen) according to manufacturer's instructions. Viability of cells was tested by trypan blue staining and no significant cell death occurred upon cholesterol extraction.
U937 cells were grown in media with 10% standard, or delipidated serum, and were mounted on polylysin-coated coverslips prior to incubation with labeled proteins. J774.A1 cells (5.104) were grown directly on coverslips (∅ φ12 mm) and incubated with Alexa Fluor 488-labeled CyaA proteins (6 nM) at 37°C for 10 min, before cells were washed and 5 µg/ml of Alexa Fluor 594-labeled cholera toxin subunit B (CtxB) was added for additional 5 min. The unbound proteins were washed-off with ice-cold PBS, cells were fixed with 4% paraformaldehyde in PBS at 25°C for 20 min, and mounted in Mowiol solution (Sigma). Fluorescence images were taken using a CellR Imaging Station (Olympus, Hamburg, Germany) based on Olympus IX 81 fluorescence microscope, using a 100× oil immersion objective (N.A. 1.3). Digital images were processed using ImageJ software.
J774.A1 cells (106) were incubated with 17 nM CyaA in DMEM for 30 min at 37°C, washed with Hank's Buffered Salt Solution buffer (HBSS), and lyzed at 4°C during 30 min in 500 µl of Tris-buffered saline (pH 7.4) supplemented with 1% Triton X-100 and EDTA-free Complete Mini proteinase inhibitor cocktail (Roche, Basel, Switzerland). The lyzate was centrifuged for 15 min at 10,000×g at 4°C, and the supernatant was incubated with MEM-174 MAb covalently linked to CNBr-activated Sepharose beads (GE Healthcare) at 4°C for 1 h. The beads were washed five times with 1 ml of the lysis buffer and the bound proteins were eluted with SDS-PAGE loading buffer and analyzed by SDS-PAGE followed by Western blotting.
J774A.1 or U937 cells (106) were incubated with 6 nM CyaA proteins at 37°C for 10 min, washed repeatedly in buffer, and the amount of cell-associated adenylate cyclase (AC) activity was determined in cell lyzates as previously described by [61].
J774A.1 or U937 cells (106) were incubated with CyaA proteins at indicated concentrations for 10 min at 37°C and intracelular cAMP concentrations were determined in cell lyzates using a competitive ELISA as previously described [62].
|
10.1371/journal.pntd.0005263 | Cellular Immune Responses to Live Attenuated Japanese Encephalitis (JE) Vaccine SA14-14-2 in Adults in a JE/Dengue Co-Endemic Area | Japanese encephalitis (JE) virus (JEV) causes severe epidemic encephalitis across Asia, for which the live attenuated vaccine SA14-14-2 is being used increasingly. JEV is a flavivirus, and is closely related to dengue virus (DENV), which is co-endemic in many parts of Asia, with clinically relevant interactions. There is no information on the human T cell response to SA14-14-2, or whether responses to SA14-14-2 cross-react with DENV. We used live attenuated JE vaccine SA14-14-2 as a model for studying T cell responses to JEV infection in adults, and to determine whether these T cell responses are cross-reactive with DENV, and other flaviviruses.
We conducted a single arm, open label clinical trial (registration: clinicaltrials.gov NCT01656200) to study T cell responses to SA14-14-2 in adults in South India, an area endemic for JE and dengue.
Ten out of 16 (62.5%) participants seroconverted to JEV SA14-14-2, and geometric mean neutralising antibody (NAb) titre was 18.5. Proliferation responses were commonly present before vaccination in the absence of NAb, indicating a likely high degree of previous flavivirus exposure. Thirteen of 15 (87%) participants made T cell interferon-gamma (IFNγ) responses against JEV proteins. In four subjects tested, at least some T cell epitopes mapped cross-reacted with DENV and other flaviviruses.
JEV SA14-14-2 was more immunogenic for T cell IFNγ than for NAb in adults in this JE/DENV co-endemic area. The proliferation positive, NAb negative combination may represent a new marker of long term immunity/exposure to JE. T cell responses can cross-react between JE vaccine and DENV in a co-endemic area, illustrating a need for greater knowledge on such responses to inform the development of next-generation vaccines effective against both diseases.
clinicaltrials.gov (NCT01656200)
| The Flavivirus genus member Japanese encephalitis (JE) virus (JEV), causes severe brain disease in tens of thousands of children across Asia every year. JE is vaccine preventable, and the immune response to JEV plays a major role in disease outcome. However, the response to JEV is hard to study as JE affects young children in rural areas. Related flaviviruses, such as dengue virus (which has no good vaccine), can influence the outcome of JE, probably due to cross-reactive immune responses. T cells (a subset of white blood cells) respond to virus infections, but we know little about the timing and nature of T cell responses to JEV after infection and whether T cells are protective against JEV.
We used the live JE vaccine SA14-14-2 as a model to study the immune response to JEV. We found T cell responses frequently after JE vaccination. In this small group of volunteers, many of whom were exposed to dengue virus, most of the T cell responses tested cross-reacted between JEV and dengue virus. However, only about two thirds of people made antibody responses to the vaccine. Studying these responses could help design new vaccines for use against JE and dengue in Asia.
| Japanese encephalitis (JE) virus (JEV) is the cause of around 68 000 cases of encephalitis per year in Asia, mostly in children [1]. JEV is a single stranded positive sense RNA virus of the family Flaviviridae, genus Flavivirus. The JEV genome is 11 kb comprising a single 10.3 kb open reading frame encoding three structural proteins (core, C; pre-membrane, prM; envelope, E) and seven non-structural (NS) proteins denoted NS1, NS2a, NS2b, NS3, NS4a, NS4b and NS5 [2]. JEV is transmitted naturally among birds and pigs by Culex mosquitoes, with humans infected coincidentally as dead-end hosts. Ecological control of JE is, therefore, unrealistic: vaccination is the only reasonable prospect of preventing disease in humans [3].
JE vaccines are effective, have been available for many decades [4], and appear to protect through neutralising antibody (NAb) against JEV [5]. Early JE vaccines were inactivated; subsequently an infectious attenuated vaccine (JEV SA14-14-2) has been developed which is safe and immunogenic [6–8]. In JE endemic areas a single dose is 94.5% to 99.3% effective [9, 10] and gives durable protection for up to five years [11, 12]. The vaccine was prequalified by the World Health Organisation in October 2013 [13].
JEV co-circulates in many parts of Asia with the related flavivirus dengue virus (DENV), currently the target of several developmental vaccines. T cell and antibody responses to DENV are cross-reactive, with clinically relevant effects, both potentially beneficial and harmful [14]. The sequence of exposure to JEV and DENV may also be relevant; DENV partially protects against JE [15], whereas JEV may predispose to worse dengue disease [16]. Cross-reactivity between DENV and other flaviviruses is less well studied, though we have recently described highly cross-reactive CD8+ T cell responses between JEV and DENV in South India, associated with asymptomatic exposure to JEV [17].
In addition to their clinical use, live attenuated vaccines may serve as models for viral infection in humans and allow the study of the development of anti-viral immune responses [18, 19]. Greater knowledge of cellular responses to both JEV and JE vaccine was identified as research priority in a JE vaccine Cochrane review [20]. A protective role for T cell responses against JE is not clearly established, but both animal and human studies suggest a role for the cellular response as well as NAb in protection and/or recovery from JE [17, 21–23].
In JE endemic areas most of the population are exposed by adulthood [24]. Therefore, live JE vaccination may mimic repeated exposure to wild type JEV in an immune host, giving information on the T cell response to wild type JEV as well as the vaccine. Although the live JE vaccine is predominantly used in children, the repeated blood sampling required makes such studies impractical in this age group. For these reasons we conducted an exploratory study of T cell responses after vaccination of adults with a single dose of JE vaccine SA14-14-2 in South India, a JE endemic area. Because dengue and JE vaccines will ultimately be used together in much of Asia, and South India is also dengue endemic, we also sought to determine whether T cell responses to JE vaccine could cross-react with DENV (or other flaviviruses), and whether there were JEV-specific T cell responses. Here, we report the first description of T cell responses to live attenuated JE vaccine SA14-14-2 in humans.
Healthy adults aged 18 to 50 years were recruited into the study by advertisement and word of mouth and vaccinated at the Indian Institute of Science (IISc) or National Institute of Mental Health and Neurosciences (NIMHANS), both in Bengaluru, Karnataka State, India. Any laboratory workers who were being vaccinated because of potential occupational exposure to JEV were eligible. Because of concern that recruitment would be insufficient, an interventional protocol was developed to enrol additional participants. Participants who were being vaccinated on this protocol were screened for anti-JEV NAb before trial entry as JE vaccination was deemed more readily justified if NAb was not detectable. Participants with positive NAb or ELISpot screening assays were included in an observational study [17]. Participants on the interventional protocol also had HIV, hepatitis B and C excluded before entry. Apart from the pre-vaccination screening, both sets of participants followed an identical protocol and were analysed as one group. Exclusion criteria were previous administration of JE vaccine, pregnancy, immunosuppression of any cause, allergy or adverse reaction to a vaccine or component of the investigational vaccine, previous episode of encephalitis, use of any other investigational drug or vaccine within 30 days of vaccination.
This was an open label single arm study. The target sample size was 20, chosen to give a reasonable chance of representing common HLA types in the South Indian population. No power calculation was performed and no comparative analysis was pre-specified. The primary endpoint was a description of the timing, magnitude, specificity and cross-reactivity of the T cell response to JE vaccine SA14-14-2 up to 8 weeks after vaccination. The number of participants seroconverting to the vaccine (defined as NAb titre > 1:10 if negative pre-vaccine, or a four-fold increase over baseline titre), geometric mean NAb titre, the number of adverse events occurring one month after vaccination and number of serious adverse events at any time were secondary endpoints. Data were analysed descriptively; statistics were performed using R version 3.1.2 (www.r-project.org).
Pre-vaccine samples were collected before subcutaneous injection of 0.5 ml attenuated JE vaccine SA14-14-2 (Chengdu Biological Products, China) over the deltoid by a study physician (lot numbers 201107C017-1, 201107C021-2, 201103C002-2 or 201206C030-2; derived from primary hamster kidney cells). Blood (40-50ml) was drawn at 1, 2, 4 and 8 weeks and monthly thereafter (in some participants). Peripheral blood mononuclear cells (PBMC) and serum were separated and cryopreserved.
Safety was assessed actively using weekly symptom diaries for the first four weeks and passively thereafter. Participants were asked about any symptoms at each contact up to six months and were telephoned at this point if face-to-face contact was not possible. Adverse events were graded 1 (symptoms but no change in behaviour), 2 (symptoms sufficient to interfere with usual daily activities), 3 (symptoms prompting medical consultation) or 4 (hospital admission).
The study was conducted according to the principles of the Declaration of Helsinki. All participants gave written, informed consent separately for screening and then for administration of the vaccine. The protocol was approved by the IISc Institutional Human Ethics Committee (ref 5/2011). The observational study was also approved by the Liverpool school of Tropical Medicine ethics committee (ref. 10.59). The interventional protocol was registered at clinicaltrials.gov (NCT01656200).
A library of 18 amino acid peptides overlapping by 10 corresponding to the entire JEV SA14-14-2 open reading frame based on the two sequences available in Genbank in 2010, accession numbers AF315119 and D90195 (see S1 JEV peptide library), was synthesised commercially (Mimotopes). Peptides were dissolved in dimethylsulphoxide (DMSO) and pooled according to JEV proteins: C/prM, E (2 pools), NS1, NS2a/NS2b, NS3 (2 pools), NS4a/NS4b, NS5 (3 pools). For proliferation assays adjacent pools (except C/prM) were combined. In cross-reactivity assays, the following peptide sets, obtained through Biodefense and Emerging Infection (BEI) Resources, NIAID, NIH, were used: DENV1 Singapore/S275/1990 E (NR4551), NS1 (NR2751), NS3 (NR2752); DENV2 New Guinea C (NGC) prM (NR506), E (NR507), NS1 (NR508), NS3 (NR509); DENV3 Philippines/H87/1956 NS1 (NR2753), NS3 (NR2754); DENV4 Dominica/814669/1981 E (NR512), DENV4 Singapore/8976/1995 NS1 (NR2755), NS3 (NR2756); West Nile virus NY99-flamingo382-99 prM (NR433), M (NR434), E (NR435), NS1 (NR436), NS3 (NR439). JEV infected cell lysate was prepared from Vero cells which were infected with JEV P20778 (MOI 5), fixed with 0.025% glutaraldehyde (Sigma), washed with phosphate buffered saline, suspended in MEM/10% FCS and sonified in a Branson cup-horn sonifier (Model 450, Branson Ultrasonics, Danbury, CT) as previously described [25]. The antigen preparation was diluted to a stock containing 4 μg/ml of JEV E protein and used at a final concentration of 80 ng/ml.
Interferon-gamma (IFNγ) enzyme linked immunospot (ELISpot) assays were conducted as previously described [17], using 2 x 105 fresh PBMC in triplicate, with peptides at 3 μg/ml and a final DMSO concentration of 0.5% (pools), or 3 μg/ml and DMSO <0.001% (individual peptides). PBMC were cultured in 100 μl RPMI supplemented with 2mM L-glutamine, 100U/ml penicillin, 0.1mg/ml streptomycin (sRPMI) and 10% fetal calf serum (FCS, R10). The cut-off for a positive ELISpot was at least 50 spot forming cells (SFC)/106 PBMC and twice the background count.
Proliferation assays used cryopreserved PBMC which were thawed, then rested overnight before labelling with Carboxyfluorescein succinimidyl ester (CFSE) as previously described [17]. Briefly, PBMC were labelled at 5–10 x 106 cells/ml of pre-warmed phosphate buffered saline (PBS)/1μM CFSE at 37°C for 10 minutes, followed by quenching with five volumes of ice cold R10 and two washes. After eight days in culture with 3 μg/ml JEV peptide pools, cells were stained with near-infra-red (IR) viability dye (molecular probes), anti-CD3-AF700 (clone UCHT1), anti-CD4-PE (clone RPA-T4), anti-CD8-APC (clone RPA-T8) and anti-CD38-PE-Cy7 (clone HIT2) fluorescent antibodies (all from BD biosciences) for flow cytometry.
Peptide epitope mapping was done either by ELISpot, using an additional blood sample if the volunteer was available, or by expanding short term T cell lines (TCL) using left over PBMC. For cross-reactivity assays, short-term T cell lines (TCL) were always used for reasons of consistency. PBMC (2 x 106) were cultured with 5 μg/ml JEV peptide pools, or 10 μg/ml individual peptide, to which responses had been detected in ELISpot assays, in 1 ml sRPMI supplemented with 10% human serum, 10% Natural T cell growth factor/IL2 (“T-stim,” Helvetica Healthcare) and 20 ng/ml recombinant human IL7 (R&D systems). TCLs were expanded for 7–10 days in culture, rested overnight in R10 without peptide, then stimulated with peptides for six hours in the presence of 10 μg/ml brefeldin A (Sigma).
Intracellular cytokine staining (ICS) assays were done using whole blood, TCL (6 hours) or PBMC (overnight), stimulated with JEV peptides (3–10 μg/ml), peptide pools (3 μg/ml), JEV infected cell lysate, or approximately 103.4 to 104.4 plaque forming units (PFU) of JEV SA14-14-2 in the presence of 10 μg/ml brefeldin A during the stimulation. Following stimulation (and red cell lysis in the case of assays using whole blood), cells were stained with near infrared viability dye (Invitrogen) at room temperature in the dark for 20 minutes, fixed with 2% formaldehyde at room temperature for 20 minutes, and cryopreserved at -80°C in PBS/1% bovine serum albumin/10% DMSO. Later, cells were incubated in FACS perm/wash buffer (BD) at room temperature for 20–30 minutes followed by staining in perm/wash buffer for 30 minutes at 4°C. Antibody clones used for anti-CD3, CD4 and CD8 were as above. Antibodies used for TCL ICS were: anti-CD3-FITC, anti-CD4-PerCP-Cy5.5, anti CD8-APC, anti-IFNγ-PE or PE-Cy7 (clone 4S.B3), anti-IL2-PE (clone 5344.111) and anti-TNFα-PE-Cy7 (clone MAb11). Antibodies used for ex-vivo ICS were: anti-CD3-AmCyan, anti-CD4-PerCP-Cy5.5, anti-CD8-Horizon v450, Anti-CD14-APC-Cy7 (clone MφP9), anti-IFNγ-PE-Cy7, anti-TNFα-APC, anti-IL2-PE and anti-MIP-1β-FITC (clone D21-1351). MIP-1β was from R&D systems, all other antibodies were from BD.
Flow cytometry was performed using a BD Canto (TCL ICS and proliferation assays) or Canto II (ex-vivo ICS) cytometer. Ex-vivo ICS responses were considered positive if the responding population was at least 0.02% of the parent gate and double the negative control value. Proliferation responses were considered positive if the CFSElo/CD38hi responding subset were at least 1% of the parent gate and double the negative control value. Analysis of T cells stained for >2 cytokines was done using Simplified Presentation of Incredibly Complex Evaluations (SPICE) software version 5.35 with pre-processing in Pestle version 1.7 [26].
Screening assays before vaccination measured the ability of heat inactivated sera at two fold dilutions from 1:4 to 1:32 to prevent destruction of a monolayer of PS cells infected with 100 plaque forming units of JEV P20778. Fifty percent plaque reduction neutralisation titres (PRNT50) were measured using LLC-MK2 cells for all study samples together at the end of the study using the same batch of cells and JEV stock to minimize variation in the assay. For PRNT50, sera were heat inactivated and assayed according to the method of Russell et al [27]. Viruses used were JEV SA14-14-2 (expanded by three passages in C6/36 cells), DENV1 16007, DENV2 16681, DENV3 16562, DENV4 C0036/06. PRNT50 values were calculated by probit regression.
Seventeen participants were recruited into the study (Table 1); nine participants were vaccinated for occupational reasons and seven were vaccinated on the interventional protocol. Median age was 25 years, range 20–39 years. One participant withdrew after a week, a second donated 5ml per sample so had limited assays performed; both remained in the study for safety. Therefore 17 participants were evaluated for safety, 16 for seroconversion, 15 for T cell immunogenicity over 8 weeks, and nine for immunogenicity beyond 8 weeks. Six adverse events in three participants were reported in total, in the first 4 weeks two grade 1 and two grade 2 adverse events occurred, two further grade 2 events occurred >4 weeks after vaccination (Table 1). Four were febrile illnesses (two within 4 weeks), the others were dizziness with, and without, headache. All adverse events recovered spontaneously and no serious adverse events occurred.
Two participants vaccinated for occupational reasons were JEV seropositive (PRNT50 >1:10) before vaccination. In addition, despite negative screening neutralizing antibody assays using PS cells, two participants on the interventional protocol were found to be seropositive when neutralising antibody measurements were repeated using PRNT50 after vaccination. Because NAb at baseline was allowed in the occupational vaccinees, these participants remained in the analysis. Ten out of 16 participants (62.5%) seroconverted to PRNT50 >1:10 or >4-fold increase over baseline (Fig 1A and S1 Fig). Of the four seropositive participants at baseline, two seroconverted with 7.2 and 9.7 fold increases in PRNT50. Therefore, 12 volunteers were sero-protected after vaccination (75%). Reciprocal geometric mean titre (GMT) at week 4 was 18.5 overall but 51 among those who seroconverted. The reciprocal GMT of maximal responses among seroconverters was 61.5. PRNT50 waned after vaccination and, of the nine participants with data 16–26 weeks after vaccination, only four (44%) had PRNT50 >1:10.
Participant VA001/1 had a positive IFNγ-ELISpot assay (NS3) at baseline (but was vaccinated because of PRNT50 1:6 and laboratory work with JEV), but made additional responses after vaccination. In total, 13 out of 15 participants tested (87%) developed new IFNγ-ELISpot responses, peaking two weeks after vaccination with significant increases over baseline at weeks 1, 2 and 4 (Fig 1B). However, there was variation amongst the participants with some mounting the peak response later (S1 Fig).
Proliferation responses were available for 13 participants for at least two time points. Interestingly, T cell proliferation responses were detected in most participants before vaccination, despite negative ELISpot assays and/or PRNT50. Although five volunteers appeared to make new T cell proliferation responses over the course of the study (S2 Fig), proliferation responses were variable and overall there was no significant difference from baseline values at any time point (Fig 1C). Example flow cytometry data of CFSE assays over the course of the study are shown in supplementary figure S3 Fig.
The function and characteristics of ex-vivo T cell responses were investigated further by intracellular cytokine staining (ICS) and flow cytometry in 13 participants with positive ELISpot assays; responses were detected in five (ELISpot is a more sensitive technique than ICS). CD4+ T cell responses were detected in three participants; CD4+ and CD8+ T cell responses in two. Flow cytometry data showing CD4+ and CD8+ T cell responses throughout the study from participant VA019/3 are shown in Fig 2, though sample limitation meant that we could not perform all these assays on every participant. Cytokine responses were small (Fig 3A) and polyfunctional T cell responses to JE vaccine were rare, in contrast to our recent findings in natural exposure and recovered JE [17]. In CD4+ T cell responses, IFNγ+ or IL2+ only populations dominated (Fig 3B), indicating that IFNγ-ELISpot alone may not detect all such responses. Participant VA001/1 (who showed a CD8+ T cell response at baseline) was an exception; we identified a polyfunctional (IFNγ/TNFα/IL2 triple positive) CD4+ T cell response at week 16 after vaccination.
Most participants made IFNγ responses to >1 peptide pool (median five) and all viral proteins were targeted. The total magnitude of the ELISpot response correlated with the number of responding pools (Spearman’s R = 0.78, p = 0.0005, Fig 4A) indicating that there were no strongly immunodominant pools. Normalised to protein size, NS1 elicited responses most frequently (Fig 4B).
In six participants, further experiments were conducted to identify some of the epitopes recognised after vaccination with JEV SA14-14-2. Responding peptide pools were either mapped ex-vivo by ELISpot, or short-term T cell lines (TCL) were expanded to peptide pools showing responses in ex-vivo assays. The peptide pools were then deconvoluted first into “mini-pools” of 6–10 peptides which were used to stimulate short term TCL in ICS assays (also allowing determination of the responding subset), followed by mapping down to individual peptides. Fifteen peptides were mapped (Fig 4C and Table 2) and a further six 46 to 90 amino acid regions (Fig 4D and Table 2) eliciting IFNγ responses were identified (participant VA023/1 had a response mapped to mini-pools only). In participant VA019/3 an antigenic region corresponding to amino acids 214–303, in prM, overlapped with peptide TRTRHSKRSRRSVSV, amino acids 209–223. The response to amino acids 214–303 was larger than the response to 209–223, making it unlikely the response to 214–303 was accounted for by only the 10 amino acid overlap. The peptides identified were mostly in the prM, NS1 and NS3 proteins (Fig 4C) with one identified in E protein. Table 2 shows all the peptides and antigenic regions identified.
All four DENV serotypes were detected in Karnataka State in the two years prior to this study (S1 Data). Therefore, in some participants, we investigated whether the IFNγ responses identified cross-reacted with DENV. Two participants, VA012/3 and VA020/1, had responses that were not present before vaccination mapped to individual peptides by ex-vivo ELISpot assays. Partial peptide libraries from DENV serotypes 1, 3 and 4 and complete libraries from DENV2 and WNV were available (see methods). To test for cross-reactive responses, short term T cell lines (TCL) were expanded by culturing PBMC from the volunteers in the presence of “T-Stim” (IL2), IL7 and the relevant JEV peptide at 10 μg/ml for 7–10 days. The same TCL were then stimulated with JEV peptide alongside variant peptides from DENV or WNV selected on the basis of a ClustalW protein alignment. Because the cells specific for the JEV peptide had expanded in culture (typically from around 0.1% (Fig 3A) to 5% (Fig 5)), a response to the variant peptide of equivalent magnitude to the JEV peptide indicates that it is likely to be the same cells being triggered by the variant peptide and the response is therefore cross-reactive.
Participant VA012/3 had high levels of NAb to DENV4 and intermediate levels of NAb to DENV2 and DENV3 at baseline, and no detectable JEV NAb. This participant developed a CD8+ T cell response to peptide TAVLAPTRVVAAEMAEVL (NS3), subsequently mapped to APTRVVAAEM (S4 Fig panel A), very similar to a previously described partially mapped HLA B*07 restricted DENV4 epitope, LAPTRVVAAEME [28]. A short-term T cell line recognised the very close DENV1 sequence APTRVVASEM equally well (Fig 5A); the sequence in DENV2-4 is identical to JEV. Interestingly, this subject did not seroconvert to JE vaccine.
Participant VA020/1, who had the highest titre of NAb to DENV1 with lower levels to DENV2 and DENV3 and no JEV NAb at baseline, developed a CD8+ T cell response to GATWVDLVL (E), based on overlapping peptide ELISpot assays and confirmed by flow cytometry with truncated peptides (S4 Fig panel B). This was confirmed by expanding a short-term TCL (Fig 5B); the epitope cross-reacted with a variant peptide conserved in DENV1 and DENV3. In a subsequent experiment variant peptides from DENV2 and DENV4 were not recognised, consistent with the serology assays suggesting DENV1 exposure (S4 Fig panel C). These responses represent either priming by JE vaccine that cross-reacts with DENV, or priming by DENV, boosted by JE vaccine, that cross-reacts with additional DENV serotypes because of close sequence similarity.
Participant VA019/3 was DENV exposed (DENV2 PRNT50 1:538), but also had JEV NAb (titre 1:123) before vaccination. This participant had several CD4+ responses mapped by expanding short term TCL to responding pools in the ELISpot assays using 5 μg/ml equivalent concentration of each peptide. The short term TCL were then stimulated with smaller pools (“mini-pools”) followed by deconvolution down to individual peptides (Table 2), Finally, the same TCL were then tested with variant peptides from DENV and WNV. As before, because TCLs were expanded with JEV peptides, responses to variant peptides indicate cross-reactivity, though in this case all but one of the responses were JEV specific (Table 3).
Participant VA001/1 carries the HLA B*15:01 allele and had high levels of DENV2 NAb and intermediate levels to other serotypes at baseline (Fig 6A). Peptide ALRGLPVRY was mapped before vaccination, and a cross-reactive response was identified to the previously described HLA-B*15 restricted peptide ALRGLPIRY from DENV2/4 and WNV [29]. The corresponding vaccine library peptide VVAAEMAEVLRGLPVRY had a modest Val for Ala substitution corresponding to position 1 of the 9-mer (the same position as the NS3 peptide TAVLAPTRVVAAEMAEVL recognised by participant 012/3), and is predicted to bind the same HLA allele as the wild type peptide with slightly lower affinity (IEBD.com [30, 31]). The ex-vivo IFNγ response to the pool containing this peptide did not change after vaccination or with seroconversion (Fig 6B), nor did the responses to the wild type and DENV peptides at week 16 (Fig 6C), though other responses developed. The SA14-14-2 peptide produced a smaller IFNγ response, though the response was still detectable (Fig 6C left panel). Short-term T cell lines expanded with JEV wild type peptide before and after vaccination (Fig 6D & 6E) confirmed that IFNγ responses were smaller to the vaccine peptide (Fig 6E). However, when analysed for additional cytokines, the total number of responding cells was similar, with the difference mostly accounted for by MIP-1β single positive cells (Fig 7A & 7B). MIP-1β may have a lower triggering threshold than other cytokines [32], suggesting that the vaccine variant epitope is less efficient than the wild type in this case. This represents an example where responses to DENV2/4 and JEV are highly cross-reactive with each other, but were less efficiently cross-reactive with the SA14-14-2 variant.
Together, these data show that T cell responses induced to JEV vaccine SA14-14-2 can recognise wild type JEV, and that responses primed to natural flavivirus infection can cross-react with JEV vaccine SA14-14-2 in some instances. These data are consistent with our recent finding that CD8+ T cell responses to JEV are highly cross-reactive, whereas CD4+ responses are much less so [17], although the small number of participants in the present study prevents generalisation of these findings.
We have shown that, in participants resident in a JE endemic area, T cell IFNγ responses are detectable after vaccination with JEV SA14-14-2, are modest in magnitude, peak within eight weeks of vaccination, and return to baseline levels by 4–6 months. Determination of epitope specificity and cross-reactivity in four participants showed that some responses to JE vaccine can cross-react with DENV, and in one case a variant epitope of SA14-14-2 was recognised less efficiently than the JEV wild type peptide by memory CD8+ T cells from a pre-existing response.
Our study found a seroconversion rate of 62.5%, sero-protection rate of 75% and GMT of 18.5 four weeks after vaccination with a single dose of JEV SA14-14-2. This is lower than studies of single dose JEV SA14-14-2 in children, where the seroconversion rate is 80–99% and GMTs 4 weeks after vaccination range from 56–370 [7, 8, 33]; or even higher after previous JEV exposure or vaccination in a DENV non-endemic area (Korea) [34]. However, our findings are consistent with a recently published clinical trial using SA14-14-2 in India in which 57.7% of participants seroconverted at 4 weeks, including 54.9% of adults aged 18–50, a similar age group to this study [35]. In the study of Singh et al., seroconversion in children aged 1–6 years was 58.8%, suggesting it is the environment and not age driving this effect. The greater number of participants showing T cell responses than NAb, and the presence of proliferation responses at baseline, indicate that measuring NAb alone may not be the only potential test of JE-immunity. NAb titres to JE vaccines can fall below “protective” levels after vaccination in the presence of protection and rapid recall responses [36, 37], a feature also observed following HBV vaccination, where B cell memory pools may be detected [38].
DENV circulates in South India where this study was conducted. Four participants in whom DENV PRNT50 were measured all showed neutralising antibody to DENV, indicative of past infection, as did a larger group recently recruited in Karnataka State, South India [17]. One possibility is that immune interference by DENV may account for the lower seroconversion rate after JEV SA14-14-2 in Indian adults. Experiments are underway to investigate the possibility of interference by DENV exposure. So far, five more participants have been tested for DENV3 NAb post JE vaccine, including two participants who did not seroconvert. Preliminary data suggest our study population is highly DENV exposed, and we have so far not identified any evidence of an increase in DENV NAb titres after JE vaccine, or an original antigenic sin type response.
Cross-reactive T cells against DENV are readily detectable after natural infection or attenuated tetravalent dengue vaccine (TV003) [32, 39] and have been associated with pathology [40] but also protection [41, 42]. One hypothesis for the disappointing results with the new dengue vaccine, Dengvaxia, in some studies [43, 44] is that, because the vaccine is a chimera based on yellow fever 17D, the appropriate T cell responses are not elicited [14]. In this study, as well as in our earlier work [17], we have observed a high degree of T cell cross-reactivity between JEV and dengue viruses in adults in this JE/dengue co-endemic area. This may reflect priming by multiple flavivirus exposures, whereby the most conserved epitopes, which receive the largest number of re-stimulations, are the most readily detectable. This is consistent with observations in a humanised animal model and following tetravalent vaccination [39, 45]. However, this is not always seen in dengue endemic areas, where a sizeable fraction of the T cell response is directed against serotype specific epitopes even in populations with multiple DENV exposures [46]. Conservation between JEV and DENV epitopes and the cross-reactivity observed in most individuals who were tested here suggests that a better understanding of how such responses develop and whether they are protective would be of benefit. There may be strategies that could include such epitopes in next generation vaccines, such as a chimeric JEV/DENV vaccines [47, 48], or heterologous prime/boost immunisation schedules with dengue and JE vaccines.
The ex-vivo T cell cytokine profiles after vaccination in this study were different from those seen in our recent study of circulating memory T cell responses to JEV [17]. In our previous study, 75% of responding CD8+ T cell responses in healthy, JEV exposed people made two or more cytokines; CD4+ T cell responses were very infrequent. Compared with recovered JE patients, where CD4+ T cell responses were much more frequent, CD4+ T cell responses in this study showed many fewer TNFα secreting cells and cells making two or more cytokines; instead IFNγ and IL2 single positive cells dominated the response. High TNFα levels have been linked to mortality in JE [49], and this finding provides further support for our earlier observations linking CD4+ T cell derived TNFα with pathogenesis in JE.
JE vaccination is effective even if the vaccine virus is relatively genetically distant from circulating virus [5] and protection may be long lasting even in the absence of NAb [50]. In our study, 10 individuals were NAb negative at baseline and had proliferation data available. Nine had T cell proliferative responses at baseline, reflective of ‘central’ memory (which presumably also make cytokines, though we did not measure this) rather than IFNγ-ELISpot assays, which, by virtue of their much shorter incubation period, are biased towards ‘effector’ memory [51]. The combination of proliferation in the absence of NAb and ELISpot responses may reflect a common, but hitherto unrecognised, state of long term immunity, mediated by long lived memory T cells [52].
In summary, here we have given the first description of T cell responses to JE vaccine SA14-14-2 in adults in a flavivirus endemic area, most of whom have evidence of prior adaptive immunity against JEV. We have shown that (i) T cell responses were detected in most volunteers after vaccination and cross-react with other flaviviruses; (ii) seroconversion after vaccination of adults with single dose JEV SA14-14-2 in an endemic area is relatively poor and (iii) T cell proliferative responses were detectable before vaccination in most volunteers, even if ELISpot and NAb assays were negative. T cell proliferation is worthy of investigation, in JEV naïve subjects, as an additional immunologic measure of response to vaccine [53]. To what extent prior exposure to related flaviviruses and associated T cell cross reactivity could influence vaccine responsiveness, especially in the B cell compartment, requires further study. Some of the questions surrounding the nature of the response to JEV SA14-14-2 can best be addressed in populations without extensive prior flavivirus exposure.
|
10.1371/journal.pgen.1003072 | Evolution of Minimal Specificity and Promiscuity in Steroid Hormone Receptors | Most proteins are regulated by physical interactions with other molecules; some are highly specific, but others interact with many partners. Despite much speculation, we know little about how and why specificity/promiscuity evolves in natural proteins. It is widely assumed that specific proteins evolved from more promiscuous ancient forms and that most proteins' specificity has been tuned to an optimal state by selection. Here we use ancestral protein reconstruction to trace the evolutionary history of ligand recognition in the steroid hormone receptors (SRs), a family of hormone-regulated animal transcription factors. We resurrected the deepest ancestral proteins in the SR family and characterized the structure-activity relationships by which they distinguished among ligands. We found that that the most ancient split in SR evolution involved a discrete switch from an ancient receptor for aromatized estrogens—including xenobiotics—to a derived receptor that recognized non-aromatized progestagens and corticosteroids. The family's history, viewed in relation to the evolution of their ligands, suggests that SRs evolved according to a principle of minimal specificity: at each point in time, receptors evolved ligand recognition criteria that were just specific enough to parse the set of endogenous substances to which they were exposed. By studying the atomic structures of resurrected SR proteins, we found that their promiscuity evolved because the ancestral binding cavity was larger than the primary ligand and contained excess hydrogen bonding capacity, allowing adventitious recognition of larger molecules with additional functional groups. Our findings provide an historical explanation for the sensitivity of modern SRs to natural and synthetic ligands—including endocrine-disrupting drugs and pollutants—and show that knowledge of history can contribute to ligand prediction. They suggest that SR promiscuity may reflect the limited power of selection within real biological systems to discriminate between perfect and “good enough.”
| The functions of most proteins are defined by their interactions with other biological substances, such as DNA, nutrients, hormones, or other proteins. Some proteins are highly specific, but others are more promiscuous and can interact with a variety of natural substances, as well as drugs and pollutants. Understanding molecular interactions is a key goal in pharmacology and toxicology, but there are few general principles to help explain or predict protein specificity. Because every biological entity is the result of evolution, understanding a protein's history might help explain why it interacts with the substances to which it is sensitive. In this paper, we used ancestral protein reconstruction to experimentally trace how specificity evolved in an ancient group of proteins, the steroid hormone receptors (SRs), a family of proteins that regulate reproduction and other biological processes in animals. We show that SRs evolved according to a principle of minimal specificity: at each point in time, these proteins evolved to be specific enough to distinguish among the substances to which they were naturally exposed, but not more so. Our findings provide an historical explanation for modern SRs' diverse sensitivities to natural and man-made substances; they show that knowledge of history can contribute to predicting the ligands to which a modern protein will respond and indicate that promiscuity reflects the limited power of natural selection to discriminate between perfect and “good enough.”
| Cells, like biological entities at higher levels [1], can be viewed as information processing systems, because they change their state or activity in response to specific internal or external cues. This behavior is mediated by functional interactions among the proteins and other molecules that comprise the system [2]. Some proteins are highly specific [3], [4], but others can be regulated by a broader array of molecular partners, including various endogenous ligands, drugs, and pollutants [5], [6].
There has been much speculation about the evolutionary causes of specificity and promiscuity. It is widely believed that evolution usually proceeds from generalist ancestral proteins to more specific recent forms [5], [7]–[10]. Both narrow and broad specificity are often assumed to be the result of optimization by natural selection; according to this view, the capacity of ancient molecules to interact with many partners allowed species with small protein repertoires to carry out a broad set of biological activities and promoted the future evolvability of new functions, while specialization in more recent proteins provides greater efficiency, finer regulation, or prevention of deleterious interactions (refs. [7], [8], [11]–[14], but see ref. [10]).
These hypotheses are largely untested, because there are few natural protein families for which the historical trajectory of changes in specificity has been carefully dissected, although the proximate mechanisms for promiscuous responses have been studied in some extant and engineered proteins [9], [10], [15]. Further, although promiscuous interactions of proteins with exogenous substances are core issues in pharmacology and toxicology, the lack of strong historical case studies means that there are no general principles that explain why molecules have evolved their present-day ligand-recognition criteria. Without such principles, predicting the ligands to which proteins will be sensitive has proven difficult [5], [16].
Steroid hormone receptors (SRs) are an excellent model for the evolution of specificity. SRs are hormone-activated nuclear transcription factors with distinct specificities for endogenous steroid hormones and exogenous substances. In all SRs, the activating hormone binds in an internal cavity within a well-conserved ligand binding domain (LBD), causing the LBD to change conformation, attract coactivator proteins, and increase transcription of target genes [17]. The SR family diversified through a series of gene duplications that took place during early chordate and vertebrate evolution [18]. Humans have two phylogenetic classes of SRs, which correspond to the chemical classes of endogenous ligands that activate each receptor's LBD. In the first class – the estrogen receptors (ERs) – the endogenous ligands are 18-carbon steroids with an aromatized A-ring and a hydroxyl attached to C3 on the steroid skeleton (Figure 1A). The other class – the nonaromatized steroid receptors (naSRs) – includes receptors for androgens (AR), progestagens (PR), glucocorticoids (GR), and mineralocorticoids (MR); these ligands all contain a nonaromatized A-ring, an additional methyl at C19, and, in most cases, a ketone at C3. Each paralog within the naSR class has distinct specificity based on the size and polarity of the functional groups at C17 and C21 on the steroid's D-ring. Although functional groups at other positions may affect sensitivity, they do not distinguish the classes of ligands recognized by paralogous receptors. SRs also differ in their promiscuous sensitivity to exogenous substances: ERs can be activated by a large set of phenolic drugs and pollutants in diverse chemical classes with highly variant structures, whereas naSRs have far fewer synthetic agonists [19], [20].
Here we characterize in detail the evolutionary trajectory of changes in ligand specificity/promiscuity in the SR protein family, as well as the underlying structural mechanisms for promiscuous responses to non-target ligands. For this purpose, we use ancestral protein resurrection (APR), which uses computational phylogenetic techniques to infer ancestral protein sequences from an alignment of their present-day descendants, followed by gene synthesis, molecular functional assays, and experimental studies of protein structure to directly characterize them. APR represents a powerful strategy for experimentally testing hypotheses about the structure and function of ancient proteins [21], [22]. By dissecting the structure-activity criteria by which ancient receptors distinguished among ligands – and tracing how those criteria changed over time – we sought to gain insight into the evolution of specificity versus promiscuity in the SR family. We also sought to determine whether an understanding of a protein family's history can reveal explanatory principles for understanding and predicting the ligands to which its members will respond.
To understand how and why the differences in ligand specificity between the ERs and naSRs receptors evolved, we used ancestral protein resurrection [21] to experimentally characterize the LBDs of two key ancient members of the protein family. AncSR1 is the last common ancestral protein from which the entire SR family descends by a series of gene duplications; AncSR2 is the ancestral protein of all naSRs (Figure 1B). The family's phylogeny indicates that both proteins are hundreds of millions of years old: AncSR1 predates the divergence of vertebrates from other chordates, and AncSR2 predates the divergence of jawed vertebrates from jawless fishes [18].
From alignments of ∼200 extant receptor proteins, we used likelihood-based phylogenetic methods to infer the best-fitting evolutionary model, phylogeny, and ancestral protein sequences. The sequence of AncSR2 was reconstructed with high confidence (mean posterior probability (PP) = 0.93 per site, Figure S1, Table S1), and even less ambiguity at ligand-contacting sites (mean PP = 0.96). AncSR1 was more ambiguously reconstructed (mean PP = 0.70 overall, Figure S2, Table S2), but at ligand-contacting sites its reconstruction was considerably more robust (mean PP = 0.90).
The AncSR1 sequence is most similar to those of the extant ERs, whereas that of AncSR2 is most similar to the naSRs, and this pattern is most pronounced at sites in the ligand-contacting pockets (Figure 1C, Table S3). These findings suggest that AncSR1 may have been activated by estrogens and AncSR2 by nonaromatized steroids, a scenario also supported by the phylogenetic distribution of ligand specificities among extant receptors – particularly the presence of estrogen-sensitive receptors in invertebrates such as annelids and cephalochordates [18], [23].
To experimentally test these hypotheses, we synthesized cDNAs for the AncSR1 and AncSR2 LBD protein sequences, expressed them as Gal4-DBD fusion constructs, and characterized their sensitivity to hormones using luciferase reporter gene assays. As predicted, we found that AncSR1 is a highly specific estrogen receptor, activating transcription in the presence of nanomolar concentrations of physiological estrogens. It was unresponsive to a broad array of androgens, progestagens, and corticosteroids, as well as cholesterol (Figure 2A, Figure S3). In contrast, AncSR2 was completely unresponsive to estrogens (and cholesterol) but strongly activated by low concentrations of diverse nonaromatized steroid hormones, including progestagens and corticosteroids and – to a lesser extent – androgens (Figure 2A, Figure S4). We also experimentally characterized numerous alternative reconstructions of AncSR1 and AncSR2 and found that these proteins' specificities for aromatized and nonaromatized steroids, respectively, are highly robust to uncertainty in the reconstruction (Figures S5, S6).
We conclude that a fundamental inversion of ligand specificity for endogenous steroid hormones – not a narrowing of specificity from a promiscuous ancestor – took place during the evolutionary interval between AncSR1 and AncSR2. This inversion must have occurred in the lineage leading to vertebrates after they diverged from cephalochordates, because cephalochordates possess a single naSR ortholog, which retains the ancestral specificity for estrogens (Figure 1B, see [24]). Subsequently, the promiscuous responses of AncSR2 to nonaromatized steroids were differentially partitioned among its descendant lineages to yield the more specific PR, GR, MR, and AR. In extant receptors, mutations that make these SRs sensitive to the ligands of other members of the family now now cause deleterious phenotypes [25]–[27].
Our findings, viewed in the context of the ancient pathway for steroid hormone synthesis, suggest that some hormone-receptor pairs were assembled during evolution by a process of molecular exploitation, whereby molecules with a different ancient function are recruited into new signaling partnerships after gene duplication and/or divergence [23], [28]. That the ancient AncSR1 was specific for estrogens implies that progestagens and androgens, which are intermediates in the synthesis of estrogens (Figure 1A), existed before steroid receptors evolved to transduce their signals. When AncSR2 and its descendants evolved the capacity to be activated by nonaromatized steroids, these biochemical steppingstones in estrogen synthesis were recruited into new, bona fide signaling partnerships.
The specificity of a protein can be described by the biochemical criteria by which it distinguishes between functionally relevant binding partners and all other substances. To dissect more precisely how the ligand-recognition criteria of SRs evolved during the interval between AncSR1 and AncSR2, we applied a structure-activity approach. We characterized the specificity of these two ancestral proteins using a library of synthetic and natural steroids that differ from each other only by the aromatization of the A-ring or the functional groups at specific positions that vary among physiological steroids (Figure 2, Table S4).
We found that AncSR1's specificity is determined primarily by a single major criterion: requirement for an aromatized A-ring. All aromatized steroids tested activated AncSR1, but no natural nonaromatized steroids were effective at nanomolar concentrations (Figure 2A, Figure S3). Comparisons using several matched pairs of aromatized/nonaromatized steroids confirm that AncSR1 distinguishes strongly among potential ligands based on its requirement for an aromatized A-ring, with EC50s that increase by orders of magnitude when only this aspect of the ligand is changed (Figure 2B, Table S5). Beyond this major criterion, AncSR1's specificity is rather loose. In particular, it tolerates different functional groups around the D-ring, as shown by its similar sensitivity to estradiol and estrone, which contain a 17-hydroxyl and ketone, respectively (Figure 2A, 2B). Even the “chimeric” steroid 19-nor-1, 3, 5(10)-pregnatriene-3-ol-20-one (NPT) – which has the much larger 17β-acetyl group found on progesterone and corticosteroids – is almost as potent an AncSR1 activator as endogenous estrogens (Figure 2B).
AncSR2's ligand-recognition criteria differ from AncSR1's in two major ways (Figure 2, Table S5). First, AncSR1's A-ring rule is inverted in AncSR2, which is more sensitive to nonaromatized steroids than to otherwise identical aromatized substances by two to three orders of magnitude (Figure 2C). Second, AncSR2 evolved an additional criterion: it prefers steroids with a 17β-acetyl group (such as progestagens and corticosteroids) to those with smaller hydroxyls or ketones (androgens and estrogens), as demonstrated by the 21- to 87-fold difference in EC50 values between pairs of hormones that differ only at this position (Figure 2D).
Beyond these two criteria, AncSR2's specificity is rather loose (Figure 2E–2I). AncSR2 does not distinguish strongly between progestagens and corticosteroids because it has only a weak preference for steroids with a 21-hydroxyl (Figure 2F). The presence/absence of an 11-hydroxyl, present on many corticosteroids, does not strongly affect the receptor's sensitivity (Figure 2G). AncSR2 does not distinguish between 3-hydroxy and 3-ketosteroids, so long as the A-ring is not aromatized (Figure 2E), and it does not require the 19-methyl present on endogenous nonaromatized steroids (Figure 2H). Taken together, these data indicate that the evolution of AncSR2's ligand specificity entailed two major changes: inversion of AncSR1's fundamental ligand-recognition criterion for an aromatized A-ring and acquisition of an additional criterion at the D-ring.
The evolving ligand recognition rules of AncSR1 and AncSR2 can be understood in light of existing knowledge concerning the biosynthesis and evolution of the ligands. Taken together, our findings suggest that the evolution of the SR family has been characterized by minimal specificity, a concept borrowed by analogy from information theory [29]: each receptor evolved to be specific enough to distinguish among the set of contemporaneous endogenous ligands to which it was exposed, but not more so.
The concept of minimal specificity provides an evolutionary explanation for the specificity and promiscuity possessed by each receptor. For example, AncSR1's single criterion – requiring an aromatized A-ring – provided minimally sufficient specificity for estrogens (Figure 3A). Estrogens are the only aromatized steroids produced in animals, because aromatization of the steroid A-ring is the final step in a conserved estrogen synthesis pathway beginning with cholesterol and proceeding via progestagens and androgens as intermediates (Figure 1A). AncSR1's simple criterion therefore allowed it to exclude all other endogenous steroids, including androgens, progestagens, and cholesterol and its metabolites. These hormones are all ancient: synthesis of estrogens via progestagens and androgens is as old as the ancestor of cephalochordates and vertebrates [30], and it may be even older, given the presence of all these hormones in mollusks [31].
Minimal specificity is also apparent in the evolution of AncSR2 and its descendants (Figure 3A, Figure S7). When AncSR2 became sensitive to nonaromatized steroids, it would have excluded estrogens but become sensitive to both progestagens and androgens; acquiring its second ligand-recognition rule restricted AncSR2's sensitivity to progestagens only. AncSR2 did not yet distinguish progestagens from corticosteroids, but endogenous synthesis of these steroids had not yet evolved; only later – during or after the same period of early vertebrate evolution when synthesis of corticosteroids first evolved due to the emergence of 21-hydroxylase activities in the CYP450 family [30], [32] – were AncSR2's promiscuous sensitivities partitioned among the PR, GR, and MR.
These data indicate that each receptor evolved ligand recognition criteria sufficiently complex to parse the repertoire of ligands present during its evolution, but those rules were not sufficient to prevent promiscuous responses to other substances that had not yet evolved. By evolving narrower specificity as the synthesis of new steroids emerged during vertebrate evolution, the various SRs presumably maintained the capacity to transduce specific signals despite the organism's increasing chemical repertoire (Figure 3A).
Predicting ligands that interact with intended and secondary protein targets is an important goal in pharmacology and toxicology, but understanding from first principles which targets will respond more or less promiscuously has proven difficult [5], [16]. The concept of minimal specificity predicts that ER's capacity to be disrupted by exogenous phenolics is inherited from AncSR1. To test this possibility, we characterized the ability of several xenoestrogens to activate AncSR1. As predicted, we found that AncSR1 is activated by the strong nonsteroidal ER agonists diethylstilbestrol and genistein and is competitively inhibited by the ER antagonists 4-hydroxytamoxifen and ICI182780 (Figure 3B).
Our observations provide an historical explanation for the greater susceptibility of ERs than naSRs to activation by pollutants, pharmaceuticals, and dietary compounds. Extant ERs inherited AncSR1's simple ligand-recognition criterion requiring little more than an aromatized A-ring with a 3-hydroxyl (Figure S8). Although this rule provided sufficient specificity throughout virtually all of vertebrate evolution, ERs are now exposed to – and fortuitously activated by – a wide range of aromatized pharmaceutical, industrial, and agricultural substances of the appropriate size and shape that have come into large-scale production only in the last century [19].
In contrast, the more restrictive specificity of AR, PR, GR, and MR – which reflects the greater variety of endogenous potential activators to which they were exposed during evolution – makes them susceptible to activation by fewer synthetic substances than ERs, although they can still be disrupted by some novel substances, such as nonaromatized 19-norsteroids used as synthetic androgens. As predicted, we found that AncSR2, like its descendants, is insensitive to the aromatized xenoestrogens (Figure S9).
Taken together, our findings suggest that analysis of a protein's history and the chemical milieu in which it evolved can provide useful information for predicting the endogenous and exogenous ligands that can interact with it.
Finally, we sought to understand the underlying features of protein structure that caused AncSR1's and AncSR2's promiscuous responses associated with minimal specificity. We first used X-ray crystallography to determine the structures of bacterially expressed AncSR2-LBD in complex with progesterone and with 11-deoxycorticosterone (DOC), at 2.75 and 2.82 Å resolution, respectively (Figure 4A, Table S6). The structures reveal why AncSR2 did not yet distinguish between progestagens and corticosteroids, which differ only in that the latter contain a 21-hydroxyl. The two protein backbones have nearly identical topologies (RMSD = 0.28 Å), and there are virtually no differences in the ways the ligands are bound (Figure 4A, Table S6). The AncSR2-progesterone complex contains ample room to accommodate the additional 21-hydroxyl of corticosteroids (Figure 4B). Further, Asn35 offers a perfectly positioned hydrogen bond partner, which is unpaired in the AncSR2-progesterone complex, for DOC's hydroxyl (Figure 4B). This additional favorable interaction explains why AncSR2 not only accommodates corticosteroids but is even more sensitive to them than progestagens.
To understand the structural causes of AncSR1's inability to distinguish between 17-hydroxyl and 17-acetyl steroids, we used homology modeling/energy minimization based on a human ERα template to predict the AncSR1-LBD structure in complex with estradiol and NPT. Despite differing by 172 amino acids, AncSR1 and AncSR2 have remarkably similar peptide backbone conformations (RMSD = 0.87 Å). AncSR1's capacity to adventitiously accommodate larger 17-acetyl steroids appears to be due to excess volume and hydrogen bonding capacity in AncSR1's cavity near the ligand's D-ring. When NPT is docked in the AncSR1 cavity, virtually no adjustment is required in the position of nearby residues compared to those in the estradiol complex: instead, the long axis of the ligand moves slightly towards H10, allowing NPT's larger acetyl group to slot into space that was unoccupied in the estradiol complex (Figure 4C). Further, the 20-keto of NPT accepts a hydrogen bond from His206, which can serve as a donor (as in the NPT complex), acceptor (as in the estradiol complex), or both, depending on its ionization state.
Taken together, these data indicate that the promiscuous responses of both AncSR1 and AncSR2 to non-target ligands are due in large part to unfilled volume in the internal cavity and untapped potential of polar side chains to form hydrogen bonds with polar atoms on the ligand [5], [9].
The promiscuity we observed during SR evolution appears to reflect the fact that there is no functional difference between a receptor that excludes ligands to which the cell is never exposed and a more promiscuous receptor that does not possess such ligand recognition criteria. Although ancient and extant SRs are only minimally specific, their potential promiscuity would not have caused them to transduce noisy signals in their historical chemical environments, because such signals were not rampantly produced at the time; there would presumably have been be no fitness cost or benefit associated with the specific forms of promiscuity these receptors manifested. Rather than representing an optimum, then, the imperfect specificity of each SR appears to reflect the limited power of selection to distinguish between “perfect” and “good enough,” given the chemical context in which these proteins evolved. Our findings are related to prior work suggesting that other protein properties, such as marginal stability, may not be uniquely adaptive states but may instead reflect the limited power of selection to optimize a property that affects fitness only when the property is near a threshold [33].
We predict that minimal specificity will be apparent in many other protein families. Protein engineering studies have shown that enzymes in the laboratory often neutrally evolve promiscuous responses to substrates not yet present in the system [9], [15]. Further, the limited specificity of natural proteins is what allows them to respond to novel drugs and xenobiotic pollutants. Direct study of historical evolution in other protein families and their ligands is necessary to determine the generality of the principle of minimal specificity and to characterize the dynamics that have shaped proteins' natural specificity and their responses to drugs and pollutants.
A phenomenon similar to minimal specificity is well known in biological information systems at higher levels, such as choice by individuals of conspecific mates [34] and mimics that lure prey or pollinators by exploiting a receiving species' signal recognition capacity [35], [36]. In each case, the “receptor” distinguishes target from nontarget signals in the species' environment but fails to exclude novel signals to which it has not previously been exposed. Minimal specificity, reflecting evolution in the face of the limited set of stimuli present in real environments, may therefore be a general characteristic of signaling and information systems from molecular to community scales.
Annotated protein sequences for nuclear receptors were downloaded from UniPROTKB/TrEMBL, GenBank, the JGI genome browser, and Ensemble (Table S7). For the reconstruction of AncSR2, 184 steroid and related receptor sequences containing both DNA binding and ligand binding domains were aligned using the Multiple Sequence Alignment by Log-Expectation (MUSCLE) program [37]. The alignment was checked to ensure alignment of the nuclear receptor AF-2 domain and manually edited to remove lineage-specific indels. The N-terminal variable region and hinge region were removed from the alignment file, as these areas could not be aligned reliably among sequences. AncSR1 was reconstructed using an expanded alignment (213 sequences), reflecting the deposition of many new SR sequences in public databases since a much earlier study of AncSR1 [23].
Phylogenies (Figures S10, S11) were inferred from these alignments using PHYML v2.4.5 [38] and the Jones-Taylor-Thornton model with gamma-distributed among-site rate variation and empirical state frequencies, which was the best-fit evolutionary model selected using the Akaike Information Criterion implemented in PROTTEST software. Statistical support for each node was evaluated by obtaining the approximate likelihood ratio (the likelihood of the best tree with the node divided by the likelihood of the best tree without the node) and the chi-squared confidence statistic derived from that ratio [39].
AncSR1 and AncSR2 were initially reconstructed by the maximum likelihood method [40] on the ML phylogeny for each alignment using the Codeml module of PAML v3.14 [41] and Lazarus software [42], assuming a free eight-category gamma distribution of among-site rate variation and the Jones-Taylor-Thornton protein model. AncSR2 was also reconstructed on a single-branch rearrangement of the ML phylogeny that requires fewer gene duplications and losses to explain the distribution of SRs in agnathans and jawed vertebrates (Figure S12, Table S8). Average probabilities were calculated across all LBD sites except those containing indels.
cDNAs coding for the maximum likelihood AncSR2 LBD and AncSR1 LBD were synthesized (Genscript) and verified. The LBDs were then cloned into the Gal4-DBD-pSG5 vector; 31 amino acids of the GR hinge containing the nuclear localization signal-1 [43] were inserted between the DBD and LBD to ensure nuclear localization and conformational independence of the two domains. The hinge and ligand-binding domain (LBD) of the human progesterone receptor (hPR; aa 632–933; Swiss-Prot P06401), human estrogen receptor alpha (hERα, aa 435–595; Swiss-Prot P03372, [44]), human glucocorticoid receptor (hGR; aa 485–777; Swiss-Prot P04150, [45]), human mineralocorticoid receptor (hMR, aa 736–984, Swiss-Prot P08235; [45]) were cloned into the Gal4-DBD-pSG5 vector in frame with the Gal4 DBD. The human androgen receptor (hAR) LBD was cloned into the pFN26A (BIND) hRluc-neo Flexi Vector (Promega) without the hinge domain (aa 671–919; Swiss-Prot P10275), as the hinge domain of the hAR inhibits AF-2 dependent activation of the hAR [46].
The hormone-dependent transcriptional activity of resurrected ancestral receptors and their variants as well as the human receptor LBDs was assayed using a luciferase reporter system. CHO-K1 cells were grown in 96-well plates and transfected with 1 ng of receptor plasmid, 100 ng of a UAS-driven firefly luciferase reporter (pFRluc), and 0.1 ng of the constitutive pRLtk Renilla luciferase reporter plasmid, using Lipofectamine and Plus Reagent in OPTIMEM (Invitrogen). After 4 h, transfection medium was replaced with phenol-red-free αMEM supplemented with 10% dextran-charcoal stripped FBS (Hyclone). After overnight recovery, cells were incubated in triplicate with the hormone of interest from 10∧−12 to 10∧−5 M for 24 h, then assayed using Dual-Glo luciferase (Promega). Firefly luciferase activity was normalized by Renilla luciferase activity. Dose-response relationships were estimated using nonlinear regression in Prism4 software (GraphPad Software, Inc.); fold increases in activation were calculated relative to the vehicle-only (ethanol) control.
To determine the robustness of functional inferences to statistical uncertainty in the reconstruction of AncSR1 and AncSR2, we used two approaches. AncSR1 had too many ambiguously reconstructed sites to examine each such residue individually, so we computationally sampled from the posterior probability distribution of reconstructed amino acid states to generate a cloud of possible ancestral sequences, each harboring a large number of alternate states. Specifically, we generated 1,000,000 possible ancestral sequences by sampling from the posterior probability distribution of states at each site. Of this sample, the five sequences with the highest total posterior probability differed from the ML reconstruction at 55 to 59 sites and from each other by 63 to 82 sites; these sequences had total posterior probabilities lower than AncSR1-ML by a factor of 10−23 to 10−24. They differed from each other at several sites in the ligand pocket and included four unique combinations of ligand-contacting residues. We synthesized these five radically alternative ancestral reconstructions de novo and repeated the functional assays. Despite their extreme distance from AncSR1-ML, all five alternative reconstructions were sensitive to estrogens and did not respond to nonaromatized steroids (Figure S5).
For AncSR2, we identified all plausible alternate reconstructions (those with posterior probability >0.20 excluding biochemically similar K/R, D/E, S/T, and I/L differences) and introduced each alternate state individually into the AncSR2 background using the Quikchange Mutagenesis kit (Stratagene), verified clones by sequencing, and repeated the activation assays with each version of AncSR2 (Figure S6). The ML AncSR2 sequence reconstructed on the ML tree had high baseline activation in the absence of ligand; this phenotype is almost certainly an artifact, because constitutive baseline activity is not present in any of AncSR2's extant descendants; it is well-established that some amino acid replacements can cause nuclear receptors to become constitutive by marginally stabilizing the active conformation in the absence of hormone [47]. We therefore introduced all plausible alternate reconstructions into AncSR2-ML and found that one (L79M) eliminated this ligand-independent activity. The “constitutive” Leu79 state is weakly supported on the ML tree (PP = 0.59), and has no support (PP = 0.00) on the phylogeny that is most parsimonious in terms of gene duplications and losses; in contrast, the “non-constitutive” state Met79 has PP = 0.41 on the ML tree and PP = 1.00 on the rearranged gene duplication/loss tree (Figure S12, Table S8). The AncSR2 sequence used for all experiments reported in the text therefore contains state Met79. The other alternate reconstructions were then reintroduced into this AncSR2 sequence: none qualitatively changed the receptor's sensitivity to the various classes of steroid hormones, except for A171V, which conferred constitutive activity (Figure S6).
The AncSR2 ligand binding domain (LBD) cDNA (residues 1–252) was cloned into pLIC-MBP (provided by J. Sondek, Chapel Hill, NC), which contains a hexahistadine tag followed by the maltose binding protein (MBP) and a tobacco etch virus (TEV) protease site N-terminal to the protein. AncSR2 was expressed as a fusion protein in BL21(DE3) pLys cells in the presence of 50 µM ligand using standard methods, and initially purified using affinity chromatography (HisTrap columns, GE Healthcare). Following TEV cleavage, the tagged MBP was removed by an additional nickel affinity column. AncSR2 was purified to homogeneity via gel filtration. Pure AncSR2 LBD was dialyzed against 150 mM sodium chloride, 20 mM Tris HCl (pH 7.4), 5% glycerol, and 50 µM CHAPS and concentrated to 2–5 mg/mL.
Crystals of AncSR2-LBD with ligand were grown by hanging drop vapor diffusion at 22°C from solutions containing 1.0 µL of protein at 2–5 mg/mL protein and 1.0 µL of the following crystallant: 0.8–1.2 M MgSO4, 6–12% glycerol, and 100 mM MES, pH 5.4–6.4. Orthorhombic crystals of the AncSR2 – progesterone and 11-DOC complex grew in P212121 and C2221 spacegroups with either two monomers or one monomer in the asymmetric unit, respectively.
Crystals were cryoprotected in crystallant containing 20% glycerol and were flash-cooled in liquid N2. Data to 2.75 Å and 2.82 Å resolution were collected for the AncSR2-progesterone and AncSR2-deoxycorticosterone complexes, respectively (Table S6). All data were collected at South East Regional Collaborative Access Team (SER-CAT) 22-ID at the Advanced Photon Source at Argonne National Laboratory in Chicago, IL, and were processed and scaled with HKL2000 (HKL Inc.). Initial phases for the AncSR2- progesterone complex were determined using a homology model to the progesterone receptor (1A28) as the initial search model in Phenix (Phenix) [48]. Subsequent structures were solved using the best available AncSR2 structure for initial phases. All structures were refined using standard methods in the CCP4 suite of programs and COOT v0.9 was used for model building [49]. Omit maps were generated by removing coordinates corresponding to the ligand and running 10 rounds of restrained refinement in CCP4. Maps are contoured to 1 σ (Figure S13). Figures were generated using PyMol (Schrödinger, LLC). AncSR2 structures with progesterone and DOC have PDB accessions 4FN9 and 4FNE, respectively. Structures were rendered for display using Pymol software.
The structure of AncSR1-LBD was predicted by homology modeling, based on a human ERα∶estradiol structure (1ERE), the most similar human receptor in sequence and function. We used Modeller software [50] to infer the AncSR1-LBD structure 100 times, chose the lowest-energy iteration from these structures, and verified it using RAMPAGE software [51], which showed only 4/237 Ramachandran outliers, all of which were in surface loops. Cavity volumes were inferred using VOIDOO software [52] by calculating the volume accessible to a probe 1.4 Å in diameter.
|
10.1371/journal.pgen.1002580 | Variation in MSRA Modifies Risk of Neonatal Intestinal Obstruction in Cystic Fibrosis | Meconium ileus (MI), a life-threatening intestinal obstruction due to meconium with abnormal protein content, occurs in approximately 15 percent of neonates with cystic fibrosis (CF). Analysis of twins with CF demonstrates that MI is a highly heritable trait, indicating that genetic modifiers are largely responsible for this complication. Here, we performed regional family-based association analysis of a locus that had previously been linked to MI and found that SNP haplotypes 5′ to and within the MSRA gene were associated with MI (P = 1.99×10−5 to 1.08×10−6; Bonferroni P = 0.057 to 3.1×10−3). The haplotype with the lowest P value showed association with MI in an independent sample of 1,335 unrelated CF patients (OR = 0.72, 95% CI [0.53–0.98], P = 0.04). Intestinal obstruction at the time of weaning was decreased in CF mice with Msra null alleles compared to those with wild-type Msra resulting in significant improvement in survival (P = 1.2×10−4). Similar levels of goblet cell hyperplasia were observed in the ilea of the Cftr−/− and Cftr−/−Msra−/− mice. Modulation of MSRA, an antioxidant shown to preserve the activity of enzymes, may influence proteolysis in the developing intestine of the CF fetus, thereby altering the incidence of obstruction in the newborn period. Identification of MSRA as a modifier of MI provides new insight into the biologic mechanism of neonatal intestinal obstruction caused by loss of CFTR function.
| Cystic fibrosis (CF) is a monogenic disease with considerable phenotypic variability. About 15% of newborns with CF suffer from an intestinal obstruction called meconium ileus (MI), and studies in CF twins have shown that modifier genes play a substantial role in the development of this complication. We used a family-based study design to enrich for genetic modifiers of MI and found that variations in the MSRA gene, represented by combinations of SNPs, or haplotypes, were protective against this manifestation of CF. We investigated association between one of the MSRA haplotypes and MI in an independent sample of CF patients and showed that it had a similar protective effect. Furthermore, CF mice lacking Msra expression had lower mortality due to intestinal obstruction at the time of transitioning to solid food and lived longer than CF mice with normal Msra, thus supporting the protective effect of the haplotype we observed in human CF subjects. The identification of modifiers of MI such as MSRA offers new insight into the mechanism of this life-threatening complication of CF.
| Cystic fibrosis (CF; MIM 219700, http://www.omim.org) is an autosomal recessive condition caused by mutations in the cystic fibrosis transmembrane conductance regulator (CFTR; MIM 602421) gene [1]. The earliest manifestation of CF is meconium ileus (MI), a prenatal obstruction of the small intestine at the ileocecal junction. Meconium, the intestinal contents of the developing gut that form the first bowel movement, has an abnormally high protein content in CF neonates thought to be due to defective proteolysis [2]–[4]. Impaction of the tenacious meconium results in intestinal obstruction in approximately 15% of CF newborns. This complication presents as abdominal distention, failure to pass meconium, and vomiting and was near universally fatal in CF newborns until effective treatment (enema or surgery) was developed. The long term effects of MI have been a matter of debate as some investigators have reported worse outcomes while others observed no significant differences from CF subjects without MI [5]–[7].
Genetic modifiers have been implicated in the development of MI for over 40 years as recurrence risk for this complication in siblings with CF (∼0.25) has consistently been shown to be higher than that in unrelated individuals with CF (∼0.15) [8]–[12]. Concordance analysis in monozygous and dizygous twins demonstrated that the heritability of MI approaches 1.0, confirming that modifier genes play a substantial role in MI [12]. CFTR also contributes to risk as MI almost exclusively manifests in CF subjects with exocrine pancreatic insufficiency (PI), which is highly correlated with CFTR genotype [13], [14]. Furthermore, the incidence of MI appears to vary among CFTR mutations that confer PI. For example, the amino acid substitution p.Gly551Asp (“G551D”) has been associated with reduced risk of MI compared to the most common CFTR mutation that causes a deleterious in-frame deletion of one amino acid (p.Phe508del, “delta F508”) [15], [16].
Initial attempts to identify MI modifier genes in humans utilized localization results from mouse studies. Mice with disruption of Cftr present with an MI-like phenotype; however, it differs from human MI in several respects. First, intestinal obstruction in CF mice causes mortality shortly after birth and at the time of weaning with the introduction of solid food [17]–[19]. Life-threatening obstruction in humans occurs in the perinatal period, while episodes of variable severity termed distal intestinal obstruction syndrome (DIOS) can also occur throughout life (5 to 12 episodes per 1,000 patient-years), especially in adults with PI and as a post-operative complication of surgical intervention, particularly organ transplantation [20], [21]. Second, obstructive lesions in CF mice have been observed in the jejunum, ileum and colon, compared to predominantly ileo-colic localization in humans [22]. Third, pancreatic exocrine disease is much less prominent in mouse models of CF [17], [23], [24]. On the other hand, there are instructive genetic similarities between mice and humans with CF. Cftr alleles influence the rate and severity of murine intestinal obstruction [17], [25]–[28] and strain-specific differences in the penetrance of intestinal obstruction indicate that different modifier genes underlie obstruction at birth and at weaning [19]. Candidate gene approaches in CF mice have revealed that decreased expression of the sodium hydrogen exchanger 3 (Nhe3) or mucin 1 (Muc1) or over-expression of the chloride calcium channel activated 3 protein (Clca3/Gob5) can reduce intestinal obstruction at weaning [29]–[31].
Newer animal models of CF have provided additional clues in the search for modifiers of MI. CFTR knock-out ferrets and pigs have been shown to develop intestinal obstruction that is anatomically and temporally equivalent to that observed in humans; however, the phenotype is highly penetrant in these animals (75% and 100%, respectively) [32], [33]. Animal models of CF and heritability studies in humans suggest that intestinal obstruction due to loss of CFTR function is a consistent feature, and that the incomplete penetrance of this trait in humans with CF is likely due to the presence of genetic modifiers. We present here the results of a regional association analysis of a linked locus on chromosome 8 [12], and report the identification and functional confirmation of a modifier gene for MI in humans and mice with CF.
To investigate a region on chromosome 8 that had previously shown linkage to MI [12], transmission analysis of SNPs was performed using families from the Cystic Fibrosis Twin and Sibling Study (TSS). As MI is associated with an increased recurrence risk among siblings, we enriched for genes that modify this phenotype by analyzing 133 families in which at least one subject was affected with MI (26 MI concordant and 91 discordant pairs, 2 concordant and 8 discordant sets of 3, and 6 singletons; Table 1). Since MI rarely occurs in the absence of PI, individuals were excluded if their CFTR genotypes were associated with exocrine pancreatic sufficiency (PS) (n = 7) or if they clinically demonstrated PS (n = 5). All SNPs on the Illumina 610-Quad genotyping panel that passed rigorous quality control (2,896 SNPs) [34] were included from an approximately 9 Mb region within the linkage locus where the SNP LOD score exceeded 1.0 (boundaries: rs2945913-rs2285274; green shaded area in linkage plot inset, Figure 1). Parental genotypes were utilized to test the transmission of SNP alleles using pedigree-based association testing (PBAT), an extension of the transmission disequilibrium test that is robust against population stratification [35], [36]. A cluster of SNPs proximal to and within MSRA (MIM 601250) showed evidence of association with MI when an additive mode of inheritance was assumed (Figure 1). One SNP within this cluster 5′ to MSRA, rs614197, exceeded the threshold for region-wide significance (P = 8.35×10−6, Bonferroni corrected P = 0.024). However, this SNP was not associated with MI in a sample of unrelated CF patients from the Canadian Consortium for Genetic Studies (CGS; Table 1).
Lack of significant association between MI and rs614197 in the CGS sample led us to question whether the initial observation in the TSS families was spurious, or if detection of association could be confounded by interacting loci [37] or by heterogeneity of effect of alleles at the locus [38], [39]. To test the latter concept, we used haplotype analysis to search for additional genetic variation associated with MI in the region surrounding MSRA. Haplotypes derived from a sliding window of three consecutive SNPs across a 2 Mb region centered at rs614197 (boundaries: rs17700611-rs4240673) were tested for transmission distortion under an additive genetic model (Figure S1). We used combinations of three SNPs as a compromise between the number of haplotypes generated and the penalty incurred by multiple testing. Two haplotypes exceeded the threshold for significant regional association after Bonferroni correction for 2,890 different haplotypes observed in the 133 families studied (Table 2). The combination of rs10903323 T, rs4840475 G and rs17151637 A (T-G-A) spanning a 3.5 kb region in intron 3 of MSRA (Figure 2) was significantly over-transmitted to individuals without MI, or in other words had a protective effect on MI (54 informative families, raw P = 1.08×10−6; Bonferroni P = 3.13×10−3). In the entire TSS cohort, the T-G-A haplotype frequency was 14.9% (13.0% in subjects with MI, 25.0% in those without MI). A second, overlapping haplotype (Figure 2) containing two of the same SNP alleles as the T-G-A haplotype (rs4840475 G, rs17151637 A, plus rs6601427 C; Table 2) similarly demonstrated association with the absence of MI (48 informative families, raw P = 1.17×10−5; Bonferroni P = 0.034). Another haplotype located just 5′ to MSRA showed nearly significant over-transmission to individuals with MI, indicating that it conferred risk for MI (59 informative families, raw P = 1.99×10−5; Bonferroni P = 0.057). Interestingly, this haplotype included rs614197, the SNP that was associated with MI in the initial single marker analysis (rs586123 G, rs614197 G and rs2055729 C; Table 2). The SNPs comprising these three haplotypes displayed weak linkage disequilibrium (LD) with the exception of the second and third SNP in the 5′ haplotype (r2 = 0.55; Figure 2). Thus, the haplotypes are likely detecting additional genetic variation that cannot be assessed by the tagging SNPs on this array platform.
To evaluate whether the MSRA alleles identified here by association could explain the former linkage signal on chromosome 8, the frequency of the T-G-A haplotype that showed the strongest association signal was calculated in siblings with CF that were concordant for MI. Sixteen sib-pairs contributed positively to the LOD score while the remaining 14 sib-pairs had a negative LOD score. The frequency of the T-G-A MSRA haplotype was lower in siblings that contributed to linkage (5 of 64 chromosomes, 7.8%) compared to siblings that did not (12 of 44 chromosomes, 21.4%; P = 0.033). Given that the T-G-A haplotype is associated with a decreased rate of MI, it appears that the observed linkage with risk for MI in our previous study was due to the increased sharing of other MSRA haplotypes that confer a higher risk of MI.
A replication study was performed using CF patients recruited by the CGS. The CGS is representative of the general CF population in Canada and comprises approximately 70% of all Canadian CF patients [40]. The overall rate of MI in this group was 15.9% (n = 220 with MI, 1,163 without MI), which is consistent with the incidence of MI reported in numerous Caucasian CF populations [8], [10], [16]. The frequency of the T-G-A haplotype in the 1,335 CGS subjects in whom haplotypes were ascertained was similar (15.4%) to that observed in the TSS (14.9%; Table 2). The T-G-A haplotype showed association with MI in the CGS sample under an additive model (OR = 0.72, 95% CI [0.53–0.98], P = 0.04), in agreement with the initial finding in the TSS. The incidence of MI in subjects with 0 copies of the T-G-A haplotype was 17.3% (165/954 subjects), 13.1% (46/351) in those with 1 copy and 10% (3/30) in those with 2 copies.
The concept that variation in CFTR influences the risk of MI is evident from the observation that a PI state (primarily determined by CFTR genotype) is required for the development of MI [14]. However, there is a finer correlation between CFTR genotype and MI as two CFTR mutations that are highly correlated with PI, p.Gly551Asp and p.Gly542X, have been shown to decrease or increase MI risk, respectively, from that conferred by the common mutation p.Phe508del [14]–[16]. As p.Gly551Asp is present at a relatively high frequency among European CF alleles (∼2% [41]), we evaluated the association between this mutation and MI in the TSS and CGS cohorts. The incidence of MI in p.Gly551Asp-bearing subjects was 7.8% in the TSS (n = 51) and 5.9% in the CGS (n = 51), compared to 20.5% (n = 507) and 17.9% (n = 851) in p.Phe508del homozygotes (P = 0.026, 0.033), respectively. Combining these two samples of CF subjects demonstrated that the odds of MI in subjects with p.Gly551Asp was about a third of that in p.Phe508del homozygotes (OR = 0.32, 95% CI [0.13–0.76]; P = 0.010), comparable to the report by Hamosh, et al [15]. The finding that variation in the disease-causing gene alters the incidence of MI even among PI subjects suggested that the relationship between MI and the MSRA haplotype could be confounded by variation in CFTR. To control for genetic heterogeneity in CFTR, we tested for association between the T-G-A haplotype and MI by evaluating 1,017 subjects with the same genotype (p.Phe508del homozygotes) drawn from the TSS MI families (n = 166) and CGS studies (n = 851). A meta-analysis conducted using logistic regression coefficients and standard errors [42] from individual TSS and CGS analyses revealed that the T-G-A haplotype retained an additive protective effect (P = 0.001), indicating that MSRA modifies MI independently of variation in CFTR.
Intestinal obstruction at the time of weaning is the primary cause of death in mouse models of CF [22], [43]. As the effect of the MI-associated haplotypes upon MSRA expression was unknown, we elected to introduce null alleles of Msra into mouse models of CF with high and low rates of mortality due to intestinal obstruction to detect whether loss of Msra expression reduced or exacerbated the rate of obstruction. Intestinal obstruction in a null CF mouse model (C57BL/6J Cftr−/−) leads to high mortality (>80% [17]) by 40 days of age while lower rates of mortality [44] occur in a CF mouse model (C57BL/6JR117H/R117H) with a targeted knock-in of a missense mutation (p.Arg117His) associated with residual CFTR function [45], [46] and very low rates of MI in humans with CF [47]. Mice heterozygous for the Cftr null (Cftr+/−) or p.Arg117His allele (Cftr+/R117H) were crossed to mice with one or two Msra null alleles to produce CF mice (Cftr−/− or CftrR117H/R117H) with wild-type (+/+), heterozygous (+/−), and null (−/−) Msra genotypes. As expected, Cftr−/− mice displayed a sharp drop in survival around the time of weaning when solid food is introduced to the diet (Figure 3). The median survival of the Cftr−/− mice was 22 days, consistent with the high rates of mortality due to intestinal obstruction reported in other CF ‘null’ mice [17], [18], [24], [48], all of which would have been homozygous for wild-type Msra (i.e. Msra+/+). However, survival was significantly improved in Msra+/− and Msra−/− CF mice compared to their Msra+/+ littermates (P = 0.022 and P = 1.2×10−4, respectively, by log-rank test; Figure 3A). At the end of the follow-up period, 61% of Msra−/− and 42% of Msra+/− mice were still living compared to 17% of Msra+/+ mice. The increasing trend in survival across genotypes (Ptrend = 1.3×10−4) mirrors the additive effect of the MSRA haplotype observed in humans. As anticipated, CftrR117H/R117H mice displayed reduced mortality, notably through the weaning period, compared to Cftr-null mice (64.3% of Msra+/+ mice alive at 40 days, n = 14; Figure 3B). However, there was no difference in the rate of survival between Msra+/+ mice and mice with one Msra null allele (Msra+/−: 76.5% alive at 40 days, n = 51; P = 0.33, log-rank test) or two null alleles (Msra−/−: 70.6% alive at 40 days, n = 51; P = 0.51). Mortality due to intestinal obstruction was confirmed in all animals for which the carcass was identified intact, and these were primarily animals that succumbed after weaning. Thus, loss of Msra expression increased survival in Cftr−/− mice by reducing the rate of fatal intestinal obstruction.
Excessive mucus accumulation in the crypts and lumen along with goblet cell hyperplasia are characteristic findings in the small and large intestine of Cftr−/− mice [17], [22], [43]. As goblet cells are the primary source for mucus in the intestine, we sought to determine if the goblet cell content of villi observed in the ileum of 15 day old mice is affected by Msra expression (Figure 4A). In wild-type (WT; i.e. non-CF) mice, goblet cell counts per villus ranged from 7.6% to 19.7% with a median of 11.1% (Figure 4B). The fraction of goblet cells in Msra−/− mice was similar to WT, ranging from 5.8% to 19.9% with a median of 13.0%, while the fraction of goblet cells in Cftr−/− mice had a wider range, 12.4% to 90.1%, and higher median (23.7%) than WT or Msra−/− mice (Figure 4B). The fraction of goblet cells in WT and Msra−/− mice and the increased proportion in Cftr−/− mice is comparable to the numbers reported in other studies [29], [31]. Like the Cftr−/− mice, goblet cell fraction in the ileum of Cftr−/−Msra−/− mice varied widely, ranging from 16.3% to 90.1% with a median of 25.9% (Figure 4B). Cftr−/− and Cftr−/−Msra−/− mice displayed considerable heterogeneity in the fraction of goblet cells per section with both groups having a subset of villi where ∼90% of cells were goblet cells (two sections in Cftr−/− and four in Cftr−/−Msra−/−).
As goblet cell fractions were not normally distributed for the Cftr−/− and Cftr−/−Msra−/− mice, we evaluated these differences using the non-parametric Mann-Whitney test. WT and Msra−/− mice had similar distributions (P = 0.97) whereas both differed significantly from Cftr−/− (WT: P = 7.6×10−5; Msra−/−: P = 2.6×10−4). Similarly, Cftr−/−Msra−/− mice differed from WT (P = 2.4×10−5) and Msra−/− mice (P = 3.8×10−5). However, the distributions in Cftr−/− and Cftr−/−Msra−/− did not differ when all observations were included (P = 0.67) or when the sections with goblet cell fractions exceeding 89% were excluded (median 22.2% vs. 22.3%, respectively; P = 0.26). Thus, loss of Msra expression does not appear to affect goblet cell hyperplasia in the ileum of CF mice despite reducing intestinal obstruction and increasing survival.
Neonatal intestinal obstruction (also known as meconium ileus or MI) has incomplete penetrance (15%) and high heritability (∼1.0) suggesting a prominent role for modifier genes in this complication of CF [12]. Both candidate gene and genome-wide studies indicate that multiple genetic modifiers of low effect contribute to this trait in humans and in mice [12], [19], [49], [50]. The polygenic etiology of MI combined with its low incidence in CF present a substantial challenge to identifying the responsible genetic modifiers. However, by employing both linkage and transmission methods in a family-based study followed by replication in an unrelated sample of CF patients, we were able to implicate the MSRA gene on chromosome 8. To test whether manipulation of Msra expression modified intestinal obstruction in the CF mouse, we elected to use a null allele of Msra to avoid temporal or spatial issues that might have complicated a transgenic over-expression strategy. As we did not know if loss of expression would increase or decrease the rate of obstruction, we employed two CF mouse models with different rates of mortality due to intestinal obstruction at the time of weaning. Our hypothesis was that the CF null model (with high rates of obstruction) would reveal whether loss of Msra function decreased obstruction while the p.Arg117His model (with lower rates of obstruction) would reveal whether loss of Msra function increased obstruction. Indeed, reduction of Msra expression in the null CF mouse model resulted in a significant decrease in intestinal obstruction. The lack of effect in the p.Arg117His CF mice suggested that the modifying effect of Msra did not exceed the reduction in obstruction conferred by residual function of CFTR bearing p.Arg117His. Together, the two CF mouse models indicated that loss of Msra afforded protection from intestinal obstruction during the time of weaning.
The Msra null allele was generated from 129-derived embryonic stem cells, and mortality from obstruction is nearly 100% in mice on the 129 background [17]. Thus, it is unlikely that variation in the 129-derived region surrounding Msra is responsible for the reduced intestinal obstruction. Furthermore, the Msra region in mice displays minimal synteny with the region surrounding MSRA on chromosome 8 in humans. For example, tankyrase (TNKS), a gene adjacent to the 5′ end of MSRA in humans is not adjacent to Msra on chromosome 14 in the mouse but is located on mouse chromosome 8. The mouse studies provide compelling evidence supporting the contention that MSRA modulates MI in humans. Interestingly, non-CF Msra+/− mice have previously been shown to have reduced lifespan thought to be the consequence of enhanced vulnerability to oxidative stress compared to wild-type animals [51]. In contrast, we observed longer survival in CF mice lacking Msra. Our opposing finding suggests that in the disease context of CF, having less Msra affords a protective benefit.
The initial starting point in our modifier gene search was a 9 Mb region on chromosome 8 within a region linked to risk for MI in 30 concordant sibling pairs [12]. Association analysis of 133 families using transmission disequilibrium testing identified a single SNP (rs614197) that achieved region-wide significance. Lack of significant association between MI and this SNP in the CGS sample motivated us to search for evidence of untyped alleles using haplotype analysis. The association of both ‘protective’ and ‘risk’ haplotypes with MI led us to surmise that alleles of different effect exist in or near MSRA. We then tested whether the ‘protective’ MSRA haplotype had a similar influence on MI in an independent CF sample. While the TSS sample was potentially subject to ascertainment bias due to its recruitment criteria (i.e. having a sibling with CF), the CGS sample was ideal for replication as recruitment was based only on a diagnosis of CF, and the sample comprises 70% of the patients with CF in Canada. The significant correlation between the ‘protective’ haplotype and reduced incidence of MI in the CGS sample and conformity with an additive model provided reassurance that variants in MSRA modified MI risk. Finally, as several studies, including the present study, have indicated that the CFTR genotype can affect the rate of MI [14]–[16], we evaluated whether CFTR allelic variation contributed to the association between the MSRA haplotype and MI. The T-G-A haplotype associated with MI in an additive fashion in individuals with identical CFTR genotypes (p.Phe508del homozygotes), thereby demonstrating that MSRA modifies MI independently of variation in CFTR, the disease-causing gene.
Heterogeneity of effect appears to explain the observed linkage of risk for MI to the region encompassing MSRA. As noted above, the T-G-A haplotype demonstrating robust association with MI conferred protection from MI rather than risk for MI. However, this is not the most common haplotype (∼15%); thus most siblings carry other haplotypes derived from the alleles of the three SNPs that, by definition, confer higher risk for MI than the protective T-G-A haplotype. As predicted by this assumption, the T-G-A haplotype was significantly underrepresented in siblings concordant for MI who contributed to the linkage signal on chromosome 8. By the same token, the T-G-A haplotype was over represented in siblings who did not contribute to linkage. Hence, the observed modest linkage in siblings concordant for MI was the result of sharing of neutral MSRA haplotypes or haplotypes conferring risk for MI, and lack of sharing of alleles associated with protection from MI. Sequencing of the coding regions of MSRA in three CF subjects with MI and no T-G-A haplotypes and in three subjects without MI and two T-G-A haplotypes did not identify any plausible causative variants (data not shown). Thus, we conclude that the haplotypes are tagging as yet unidentified genetic variation within or near MSRA.
Central roles for luminal hydration and mucus production in intestinal obstruction in CF mice at the time of weaning have been supported by the manipulation of expression of two genes. In one study, reduced expression of the sodium hydrogen exchanger (Nhe3) led to decreased intestinal sodium absorption, thereby increasing the hydration of luminal contents, alleviating obstruction, and improving survival [29]. Similarly, knock-out of the mucin gene Muc1 in CF mice improved survival due to reduced intestinal mucus content and less obstruction at the time of weaning [30]. As noted by many others [22], [43] as well as in this study, goblet cell hyperplasia is a consistent histologic feature in the small and large intestines of CF mice. However, the role of goblet cells in intestinal obstruction in CF mice is not clear [43]. Reduced expression of Nhe3 relieved obstruction and eliminated goblet cell hyperplasia [29] while increased expression of the goblet cell marker Clca3 (Gob5) relieved obstruction and increased goblet cell hyperplasia in CF mice [31]. Rozmahel and colleagues suggested that Clca3 may reduce mucin release from goblet cells, given the observation of increased goblet cell size, thereby reducing luminal mucus content and intestinal obstruction in the CF mice [31]. Evidence of association between MI and the p.Ser357Asn variant in CLCA1, the human ortholog of the murine Clca3, in 682 European CF subjects suggests that this goblet cell marker may also contribute to intestinal obstruction in humans [52]. Thus, alteration of mucus content in the CF intestine, either by reduction in goblet cell numbers or down-regulation of mucus release, appears to affect the rate of intestinal obstruction. Our analysis indicated that loss of Msra expression did not affect goblet cell hyperplasia in the Cftr−/− ileum despite mitigating intestinal obstruction; thus the link between Msra and intestinal obstruction in the context of CF is not immediately clear.
In humans, loss of CFTR function leads to a combination of impaired pancreatic secretion of proteolytic enzymes and deficient luminal hydration of meconium [53], [54]. Consequently, meconium from CF neonates is abnormally proteinaceous compared to that of normal neonates, and it has been proposed that the altered viscoelastic properties of meconium predispose fetuses and neonates with CF to intestinal obstruction [55]. In young CF mice, pancreatic exocrine function is relatively preserved [17]. However, inadequate hydration and excessive mucus secretion leads to distension of the crypts of Lieberkühn in the ileum and colon and formation of concretions that appear to play a role in obstruction [17], [22], [23], [56]. MSRA encodes an antioxidant enzyme that modifies the activity of certain proteins by reducing methionine residues [57]. It is expressed in the intestine and other tissues, particularly the liver, kidney, and brain [58]–[60]. A possible link between MSRA and MI may be that MSRA can modulate the activity of proteolytic enzymes [61]. Maximizing the activity of any residual enzymes produced by the fetal pancreas would likely contribute to the breakdown of intestinal proteins in utero, thereby reducing the risk of obstruction. Evidence that MSRA modifies intestinal obstruction provides new opportunities to investigate the above concepts and the mechanisms underlying intestinal obstruction in mice and in humans lacking CFTR.
This study was approved by the institutional/ethical review boards of all participating institutions. Written, informed consent or assent was obtained from all subjects before enrollment in the study. Experiments on mice were approved by the Case Western Reserve University Institutional Animal Care and Use Committee.
Study subjects were derived from the North American CF Modifier Gene Consortium, which is comprised of three independent collections of CF subjects. Subjects in the discovery sample were part of the Cystic Fibrosis Twin and Sibling Study (TSS) at Johns Hopkins University (n = 1,125 subjects). Enrollment was based on conclusive diagnosis of CF [62]. Methods for isolation of patient DNA [63] and identification of CFTR mutations [64], [65] have been previously described. The diagnosis of MI was based on the presence of the following features in the newborn period: lack of passage of stool within 24 hours after birth, evidence of intestinal obstruction on abdominal radiograph (ground-glass appearance of intestine, air-fluid levels, and/or intra-abdominal calcifications), evidence of colonic abnormality (microcolon on radiograph), and treatment for obstruction (enema or surgery). Individuals with clinically defined PS, a CFTR mutation associated with PS, or unknown pancreatic status were excluded (n = 143), as was one subject with unknown MI status. The primary analysis was conducted in 133 “MI families” in which at least one sibling was affected with MI (270 subjects, 169 parents). Case/control analyses were restricted to persons of self-reported European descent to minimize the potential for spurious associations due to race-related differences in allele frequencies; whereas the primary family-based transmission analysis, which was robust against population stratification, included an additional 23 individuals of non-European or mixed descent.
Findings from the primary analysis were tested in an independent sample which has been described elsewhere [34]. The replication population consisted of 1,573 CF subjects from the CGS [40]. All subjects were defined as having PI. Exclusion of non-whites yielded 1,383 subjects (including 56 sib-ships) for analysis. Rates of MI in subjects carrying the CFTR p.Gly551Asp mutation (c.1652G>A) or who were homozygous for p.Phe508del were evaluated in the entire TSS sample and the CGS sample. Subjects with p.Gly551Asp carried this mutation in trans with another PI-associated mutation: p.Cys343X (c.1029delC), c.1585-1G>A, p.Lys1177SerfsX15 (c.3528delC), c.489+1G>T, p.Glu585X (c.1753G>T), p.Phe508del, p.Gly542X (c.1624G>T), p.Gly551Asp, p.Asn1303Lys (c.3909C>G), p.Arg553X (c.1657C>T), p.Val520Phe (c.1558G>T), or p.Trp1282X (c.3846G>A). Mutation legacy names can be found in the Cystic Fibrosis Mutation Database (http://www.genet.sickkids.on.ca).
Linkage disequilibrium between SNPs was assessed using Haploview (http://www.broad.mit.edu/mpg/haploview) [66]. DNA extracted from either whole blood or transformed lymphocyte cell lines was hybridized to the Illumina Infinium 610-Quad SNP array platform for whole genome genotyping at the McGill University and Génome Québec Innovation Centre. Genotyping was performed and stringent quality control measures were employed simultaneously in both cohorts, and the quality of SNP calls was deemed to be very high (0.004% discordance between replicate samples). SNPs were excluded from analysis in all cohorts if the call rate was <90%, if the minor allele frequency was <1%, or if the Mendelian error rate was >1%. For family-based studies, any marker displaying non-Mendelian inheritance was dropped from analysis for any family with the error. A detailed description of additional quality control measures can be found in Wright, et al [34].
For the primary study, family-based association testing of SNPs and haplotypes was performed using the PBAT module [67] implemented within the Golden Helix HelixTree software package (Golden Helix, Inc. Bozeman, MT, USA; http://www.goldenhelix.com). A sliding-window approach was employed to test the transmission of haplotypes composed of three adjacent SNPs (frequency >1%). A 2 Mb region was selected for haplotype testing as the number of possible unique haplotypes would be nearly equal to the number of SNPs tested in the initial analysis. Therefore, to be considered significant, a haplotype would have to reach the same Bonferroni-corrected threshold (or higher) that was set for SNPs in the initial analysis. An additive genetic model was applied under a null hypothesis of linkage and no association, and standard phenotypic residuals were used as offsets to increase the power of the test statistic. Bonferroni correction was applied by multiplying nominal P values by the total number of SNPs or haplotypes tested. The SNP association plot was generated using LocusZoom 1.0 (http://csg.sph.umich.edu/locuszoom).
For case/control analyses, haplotypes were derived in the primary and replication populations using an expectation-maximization (EM) algorithm implemented in Golden Helix. Statistical analysis was performed in Stata10 (StataCorp, College Station, TX, USA). Comparison of MI status to the number of copies of haplotypes was performed using Fisher's exact test (using only subjects with haplotypes that could be determined with 100% posterior probability). Odds ratios comparing the odds of MI in subjects with 0, 1 or 2 copies of the chr8 haplotype, thus assuming an additive model, were estimated using logistic regression. For subjects with more than one haplotype assignment, EM probability estimates were used to weight haplotypes. For TSS subjects, parental and sibling genotypes were utilized when possible to resolve phase. For studies including siblings (TSS and CGS), empiric standard errors account for the possibility of sib-pair correlation [68].
Mice heterozygous for a Cftr null allele, B6.129P2-Cftrtm1Unc [17] or for the Cftr missense allele p.Arg117His, B6.129S6-Cftrtm2Uth [44], and either homozygous or heterozygous for a null allele of Msra [51] were generated as breeders to produce CF mice carrying the three genotypes of Msra (+/+, +/−, and −/−) used in this study. Mice were housed at constant temperature (22°C) on a 12 hour light/dark cycle. Cages were checked daily for births and for monitoring health and survival of animals to 40 days. Mice were weaned at 21 days of age onto an enriched diet (9F Sterilizable Rodent Diet 7960, Harlan Teklad, Madison, WI) and provided water ad libitum. Mice were of mixed background, but predominantly C57BL/6J (>87%; crossed a minimum of two generations to C57BL/6J and some animals up to ten generations).
Genotyping was carried out on 7 to 10-day old animals and if death occurred before this point, genotypes were determined from carcasses. Cftr genotyping was performed as previously described [44]. Msra genotyping was carried out using two primer sets to generate a 579-bp product for the wild-type allele (forward 5′-GTGTGAGAATAAACAGATGTTCTATGC-3′ and reverse 5′-GGGTTGAGTACACTCCTTTCA-3′) or a 320-bp product for the mutant null allele (forward 5′-AAAGCGCCTCCCTACCCG-3′ and reverse 5′-ACTGTGCCCAGTTTAGTCCGTG-3′). Samples were amplified by an initial denaturation for 5 min at 95°C followed by 35 cycles of 95°C for 30 sec, 59°C for 30 sec, and 72°C for 30 sec. PCR products were fractionated on a 1% agarose gel. Kaplan-Meier survival curves were plotted and differences in survival were analyzed by the non-parametric test of trends and log-rank test of equality using Stata10.
For histology, freshly harvested tissues were fixed in 10% formalin. Tissue was embedded in paraffin, sectioned at 5 µm thickness on a microtome, and mounted on glass slides for microscopy. Deparaffinized slides were stained with Alcian blue and periodic acid Schiff (PAS) stains, then counterstained with acidified Harris hematoxylin. Comparison of goblet cell proportions between groups was performed using the non-parametric Mann-Whitney test.
|
10.1371/journal.pgen.1002577 | Physiological Notch Signaling Maintains Bone Homeostasis via RBPjk and Hey Upstream of NFATc1 | Notch signaling between neighboring cells controls many cell fate decisions in metazoans both during embryogenesis and in postnatal life. Previously, we uncovered a critical role for physiological Notch signaling in suppressing osteoblast differentiation in vivo. However, the contribution of individual Notch receptors and the downstream signaling mechanism have not been elucidated. Here we report that removal of Notch2, but not Notch1, from the embryonic limb mesenchyme markedly increased trabecular bone mass in adolescent mice. Deletion of the transcription factor RBPjk, a mediator of all canonical Notch signaling, in the mesenchymal progenitors but not the more mature osteoblast-lineage cells, caused a dramatic high-bone-mass phenotype characterized by increased osteoblast numbers, diminished bone marrow mesenchymal progenitor pool, and rapid age-dependent bone loss. Moreover, mice deficient in Hey1 and HeyL, two target genes of Notch-RBPjk signaling, exhibited high bone mass. Interestingly, Hey1 bound to and suppressed the NFATc1 promoter, and RBPjk deletion increased NFATc1 expression in bone. Finally, pharmacological inhibition of NFAT alleviated the high-bone-mass phenotype caused by RBPjk deletion. Thus, Notch-RBPjk signaling functions in part through Hey1-mediated inhibition of NFATc1 to suppress osteoblastogenesis, contributing to bone homeostasis in vivo.
| Osteoporosis is a disease caused by disruption of the balance between bone formation and resorption resulting in a net loss of bone mass. Although anti-resorptive agents are the current mainstay of osteoporosis therapy, novel strategies to promote bone formation are critically needed for more effective prevention and treatment of the disease. Notch signaling, an evolutionally conserved mechanism among multi-cellular organisms, was recently shown to control bone formation and therefore represents a potential target pathway for novel bone-promoting therapeutics. In this study we elucidate the intracellular signaling mechanism through which Notch controls bone formation, providing a molecular framework that may guide future drug development.
| Notch signaling mediates communication between neighboring cells to control cell fate decisions in all metazoans [1], [2]. The mammalian genome encodes four Notch receptors (Notch1-4) and at least five ligands (Jagged1, 2 and Delta-like 1, 3, 4) [3]. In the canonical Notch pathway, binding of the ligands to the Notch receptors present on the neighboring cell surface triggers two successive intramembrane proteolytic cleavages of the receptors mediated by the γ-secretase complex and resulting in the release of the Notch intracellular domain (NICD) [4], [5], [6]. Upon its release from the plasma membrane, NICD translocates to the nucleus where it interacts with a transcription factor of the CSL family (RBPjk/CBF-1 in mammals) to activate transcription of target genes [7]. Among the best known targets of Notch/RBPjk signaling are the Hes/Hey family of basic helix-loop-helix (bHLH) transcription repressors [8]. However, the regulation of individual Hes/Hey proteins by Notch and their role in mediating Notch function are highly dependent on cell context. In addition to the canonical pathway, Notch has also been reported to signal through noncanonical, RBPjk-independent mechanisms, but the molecular nature of these mechanisms is not well understood [6], [9], [10], [11].
Notch signaling has emerged as a critical regulator of the mammalian skeleton. Initial mouse genetic studies identified a role for Notch in axial skeletal patterning, as mice lacking either Delta-like 3 (Dll3) [12], presenilin 1 (PS1) [13], [14], a catalytic subunit of the γ-secretase complex, or lunatic fringe, a glycosyltransferase that modifies Notch proteins [15], exhibited defects in the axial skeleton due to deficiency in somite segmentation and maintenance. In addition, mice lacking either Notch1 and 2 specifically in the limb bud ectoderm or Jagged2 globally displayed syndactyly [16], [17]. Consistent with the mouse studies, human mutations in Dll3 [18] were found to cause spondylocostal dysostosis, whereas those in Notch2 [10] and Jagged1 [19], [20] were responsible for Alagille syndrome.
More recent mouse genetic studies have expanded our view of Notch function in the osteoblast lineage. By genetically removing both catalytic subunits of the γ-secretase complex, PS1 and PS2, or both Notch1 and 2 in the embryonic limb mesenchyme, we have shown that Notch critically controls postnatal bone homeostasis: the Notch-deficient long bones exhibited excessive bone formation in adolescent mice with concomitant loss of bone marrow mesenchymal progenitors [21]. Consistent with the negative role of Notch in osteoblast differentiation, Zanotti et al reported that forced-expression of NICD in osteoblastic precursors reduced osteoblast numbers and caused osteopenia [22]. Conversely, forced-expression of NICD at a later stage of the osteoblast lineage led to sclerosis owing to excessive proliferation of the immature osteoblasts, highlighting stage-specific functions of constitutive Notch activation in the osteoblast lineage [23], [24]. The negative role of physiological Notch signaling in osteoblast differentiation uncovered in mice is congruent with the clinical findings that Notch1 haploinsufficiency causes ectopic osteoblast differentiation and calcification in the aortic valves [25], [26], whereas Notch2 stabilizing mutations are responsible for the Hadju-Cheney syndrome, a disorder of severe and progressive bone loss [27], [28]. However, the signaling cascade through which Notch inhibits osteoblastogenesis is not yet well understood.
Here we have genetically assessed the role of RBPjk and Hey proteins, known components of the Notch canonical pathway, in the regulation of osteoblastogenesis. Moreover, we have evaluated the role of NFAT in the high-bone-mass phenotype caused by RBPjk deficiency. The NFAT (nuclear factor of activated T cells) family of transcription factors (NFATc1-c4) [29] have been shown to play important roles in several skeletal cell types, including chondrocytes [30], osteoclasts [31] and osteoblasts [32], [33]. Our results support a model wherein canonical Notch signaling suppresses osteoblastogenesis in part through inhibition of NFATc1 transcription, therefore integrating extracellular signals with transcription factors that control osteoblast differentiation.
Previously, simultaneous removal of both Notch1 and 2 (PNN mice) from the embryonic limb mesenchyme with Prx1-Cre, which targets all of the early limb bud mesenchyme and a subset of the craniofacial mesenchyme [34], caused high bone mass due to increased osteoblast differentiation [21]. To discern the individual contributions of Notch1 versus 2 in the osteogenic progenitors, we employed the same Cre-loxP strategy to delete the two receptors separately. Western analyses confirmed that Notch1 or Notch2 was efficiently deleted in the limb mesenchyme of Prx1-Cre; Notch1f/f (PN1) or Prx1-Cre; Notch2f/f (PN2) mice, respectively (Figure 1A). As expected from our previous study of the PNN mice, PN1 and PN2 mice were viable without gross morphological anomalies. However, X-ray radiography of the limb bones at eight weeks of age revealed a marked increase in mineral content within the trabecular region of the PN2 but not the PN1 mice, when compared with their respective littermate controls (data not shown). Three-dimensional reconstruction using micro computed tomography (μCT) of the proximal tibial trabecular region confirmed this finding (Figure 1B). In particular, PN2 mice exhibited a 130% increase in trabecular bone volume owing to increased trabeculae numbers and decreased trabeculae spacing (Table 1). The PN2 phenotype was less dramatic than that of the PNN mice [21] (Figure 1B), indicating that Notch 1 performed a discernible role in the absence of Notch 2, even though deletion of Notch 1 alone did not cause an effect. Similar to the PNN mice, the high bone mass in PN2 mice was not due to decreased total osteoclast activity, as serum CTX levels, which reflect the amount of cleaved type I collagen by osteoclasts in the whole animal, did not differ significantly from the controls (Figure 1C). In addition, osteoclast number or osteoclast surface per bone perimeter did not change in either PN1 or PN2 mice (Figure 1D–1E). Therefore, physiological signaling from Notch 2, rather than Notch 1, plays a dominant role in suppressing bone formation.
To test the hypothesis that Notch suppresses bone formation through the canonical pathway, we deleted RBPjk with Prx1-Cre from the embryonic limb mesenchyme. Western analyses confirmed that RBPjk was efficiently deleted in the tibia of the Prx1-Cre; RBPjkf/f (PRBP) mouse (Figure 2A). Moreover, Hey1 and HeyL, two Notch target genes previously identified in the PNN bones [21], were markedly reduced in the PRBP tibia (Figure 2B). The PRBP mice were born at mendelian ratio with no gross abnormalities. However, at eight weeks of age, X-ray radiography revealed that the PRBP mice contained much greater mineral content within the presumptive bone marrow cavity than the wild-type littermates (Figure 2C). μCT analysis of the proximal tibia confirmed a marked increase of bone mass in the PRBP mice (Figure 2C), as reflected in a 730%, 220% or 140% increase in BV/TV, trabeculae number or trabeculae thickness, respectively, coupled with a 70% decrease in trabeculae spacing (Table 2). Consistent with the μCT data, both H&E and picrosirius red staining of the tibia detected excessive trabecular bone occluding much of the marrow cavity of the PRBP bones (Figure 2D, 2E). These analyses also revealed an abnormal elongation of the growth plate hypertrophic cartilage in the PRBP bones (Figure 2D–2E); this phenotype was similar to that previously analyzed in the PNN mice and could not be contributed to changes in osteoclast numbers at the chondro-osseous junction (Figure 3A, 3B). The dramatic increase in bone mass in the PRBP mice was very similar to that seen in the presenilin 1- and 2-deficient (PPS) animals, but more severe than the PNN phenotype, likely due to contributions from Notch3 and 4 in the PNN mice [21]. Although the data do not exclude that RBPjk may control bone formation through a yet unknown mechanism independent of Notch, the striking similarity in the bone phenotype among the PPS, the PNN and the PRBP mice indicates that RBPjk is likely the principle mediator of physiological Notch signaling in bone.
We then analyzed the cellular basis for the high bone mass in the eight-week-old PRBP mice. Tartrate-resistant acid phosphatase (TRAP) staining on tibial sections revealed a strikingly uneven distribution of osteoclasts within the trabecular bone region of the PRBP mice: whereas TRAP-positive cells were more abundant than normal within the metaphyseal region, few were detected towards the diaphysis (Figure 3A). The reason for this regional disparity is not certain at present but may be due to uneven compartmentalization of osteoclast precursors within the occluded marrow cavity. Serum CTX assay did not detect any significant difference between the PRBP and the WT littermates (Figure 3C). Further investigation of the metaphyseal region revealed that although osteoclast number per bone perimeter (No. OC./mm) was higher in the PRBP mice, the spreading of individual osteoclasts (µm/OC.) was decreased, resulting in no change in the percentage of bone surface covered by osteoclasts (OC. S./B. S.) (Figure 3D). Thus, the PRBP mice possessed abundant, but apparently less functional osteoclasts within the metaphyseal trabecular bone. Real-time PCR experiments showed that the mRNA levels for both the osteoclastogenic signal Rankl and the anti-osteoclastogenic factor Opg were reduced in the PRBP bone, but the ratio of Rankl over Opg (Rankl/Opg) was 230% higher in the PRBP bone than the control (Figure 3E). Moreover, the mRNA level for M-CSF, a potent mitogen of osteoclast precursors, was 690% higher in the PRBP mice (Figure 3F). The higher level of M-CSF coupled with an increased ratio of Rankl/Opg could explain the supernumerary but dysfunctional osteoclasts populating the metaphyseal trabecular bone in the PRBP mice. Overall, the high bone mass in the PRBP mice was not caused by an overall decrease in bone resorption.
Having ruled out resorption deficiency as the main cause for the high bone mass in PRBP mice, we next focused on bone formation parameters. Static histomorphometry of tibial sections from the eight-week-old PRBP mice revealed a marked increase in the number of cuboidal (active) osteoblasts, when normalized to either bone perimeter (60% increase) or trabecular bone area (400% increase) (Figure 4A). The number of flat (inactive) osteoblasts, when normalized to trabecular bone area, was also increased by 100% in the PRBP mice. Consistent with the increase in osteoblast number, real-time PCR experiments showed that a number of common osteoblast markers were upregulated in bone total RNA (Figure 4B). Dynamic histomorphometry showed that the mineral apposition rate (MAR), which measured osteoblast activity, did not differ significantly between PRBP and the control littermates (Figure 4C). However, the percentage of double-labeled trabcular bone surface was increased by 230% in the PRBP mice, resulting in a significant increase in the bone formation rate (BFR) within the trabecular region (Figure 4C). Thus, the increase in bone mass in the PRBP mice was primarily due to a marked increase in osteoblast numbers.
To explore the mechanism responsible for the increase in osteoblast numbers, we assessed the status of apoptosis and proliferation of osteoblasts in PRBP versus wild-type bones. To this end, osteoblast protein extracts were prepared from the bone surface of the long bones, and subjected to Western analyses for activated caspase 3 and PCNA, markers for apoptosis and cell proliferation, respectively. These assays did not detect a significant difference in either protein between the genotypes (Figure 4D). Therefore, the increase in osteoblast numbers is unlikely to be caused by changes in apoptosis or proliferation, but rather due to enhanced differentiation from the progenitors.
Uncontrolled osteoblast differentiation may lead to loss of bone marrow mesenchymal progenitors and rapid age-dependent boss loss [21]. To test whether this is the case in the PRBP mice, we analyzed bone mass by X-ray (data not shown) and μCT at 26 weeks of age. Indeed, bone mass was drastically reduced in the PRBP mice at 26 weeks when compared with 8 weeks (Figure 5A). When quantified, the trabecular bone mass of the PRBP tibia was no longer significantly different from the wild type at 26 weeks, representing a drastic decline from a level 730% above normal at 8 weeks (Table 2). Similarly, both trabeculae thickness and number were reduced to levels either equivalent or close to the wild type. Interestingly, bone resorption, as measured by serum CTX assays, was significantly higher in the PRBP mice over the control at 26 weeks, likely contributing to the rapid bone loss (Figure 5B). Thus,similar to the PNN mice [21], the PRBP mice, despite their high bone mass when young, rapidly lost bone with age.
We next assessed the status of the mesenchymal progenitors in the bone marrow. To this end, bone marrow stromal cells (BMSC) isolated from PRBP versus wild type littermates were subjected to CFU-F (colony forming unit-fibroblast) assays. These assays were not feasible with adolescent PRBP mice due to the occlusion of the marrow cavity, and therefore performed only after six months of age. Remarkably, no type I CFU-Fs could be detected from the PRBP bone marrow at either 26 weeks (data not shown) or one year (Figure 5C), indicating a severe diminution of the mesenchymal progenitor pool. Moreover, BMSC isolated from the PRBP bone were severely deficient in undergoing osteoblast differentiation when cultured in osteogenic media and monitored by the expression of alkaline phosphatase (AP) (Figure 5D). Therefore, the PRBP animals exhibited a marked deficiency in the bone marrow mesenchymal progenitor pool.
To delineate potential stage-specific requirement of Notch-RBPjk signaling during osteoblast differentiation, we deleted RBPjk with either Osx-GFP::Cre [35] or 2.3ColI-Cre [36], which are believed to target progressively more mature osteoblastic cells. Western analyses confirmed that both Cre lines efficiently deleted RBPjk in the long bones (Figure 6A). However, when assessed by either X-ray radiography or μCT, neither deletion caused any significant changes in bone mass at either 8 or 21 weeks of age (Figure 6B, 6C), a finding confirmed by quantitative analyses (Table 3). Because previous studies have suggested that Notch signaling in the more mature osteoblastic cells regulated osteoclastogenesis through modulation of Rankl and Opg [23], [37], we examined osteoclasts in both the Osx-GFP::Cre; RBPjkf/f (OsxRBP) and the 2.3ColI-Cre; RBPjkf/f (ColIRBP) mice. However, serum CTX assays detected no significant changes in either OsxRBP or ColIRBP mice over controls at either 8 or 21 weeks of age (Figure 6D, 6E), indicating largely normal bone resorption in these animals. Similarly, osteoclast number and osteoclast surface per bone surface were comparable between the mutant strains and their wild-type littermates (Figure 6D, 6E). Thus, RBPjk does not appear to play a major role in the more committed osteoblast-lineage cells.
To assess the role of Hey1 and HeyL in bone formation, we analyzed the bones of mice wherein the two genes have been deleted. Because previous work by others revealed no major bone phenotype in the Hey1−/− mice [38], we focused on the HeyL−/− and the Hey1/HeyL double mutant animals, all in the C57BL6 background. As Hey1−/−; HeyL−/− mice died prematurely due to heart defects [39], we analyzed the bones of the viable HeyL−/−; Hey1+/− animals. μCT analyses showed that the HeyL−/− and the HeyL−/−; Hey1+/− mice possessed progressively more trabecular bone than their wild-type littermates at 8 weeks of age (Figure 7A). In particular, the femoral trabecular bone mass was increased by 80% and 150% over the control in the HeyL−/− and the HeyL−/−; Hey1+/− animals, respectively (Table 4). Moreover, like the PRBP bones, the HeyL−/−; Hey1+/− samples exhibited a significant increase in trabeculae number and thickness with a corresponding decrease in trabeculae spacing. At the cellular level, the HeyL−/−; Hey1+/− bones exhibited more cuboidal osteoblasts than the wild type whereas their number of osteoclasts appeared to be normal (Figure 7B, 7C). Thus, Hey1 and HeyL, like Notch and RBPjk, negatively regulate osteoblast numbers.
We next investigated the mechanism through which Notch-RBPjk-Hey signaling regulates osteoblast differentiation. In a separate effort to identify Hey1 and HeyL target genes, we performed genome-wide ChIP-seq (Chromatin immunoprecipitation followed by high-throughput sequencing) experiments by expressing Flag-tagged Hey1 or HeyL in HEK293 cells. We identified strong binding for both proteins around the P1 promoter of NFATc1 (Figure 8A and data not shown). Importantly, Hey1 was also found to bind to the NFATc1 P1 promoter region in ST2 cells, a mouse bone marrow stromal cell line that can be induced to differentiate into osteoblasts (Figure 8B); the binding is consistent with the presence of a predicted Hey1 binding site “CGCGCG” within the region. In contrast, no binding was detected for the alternative P2 promoter (Figure 8B). We next focused on the functional relevance of Hey1 binding. Full-length Hey1, but not a form missing the HLH domain, suppressed the activity of the NFATc1 P1 promoter in transient transfection assays in both HEK293T and ST2 cells (Figure 8C). Because NFATc1 was previously shown to increase osteoblast numbers [33], we explored the potential involvement of NFATc1 in Notch-RBPjk signaling in bone. Real-time PCR revealed that NFATc1 mRNA was increased by 200% in the PRBP tibia over the control (Figure 8D). Western analyses identified a 70 kD isoform of NFATc1 greatly induced in the PRBP bones, whereas a 77 kD form was less affected (Figure 8E). To gain insight on the induced isoform, we employed semi-quantitative RT-PCR to identify the specific NFATc1 mRNA variant(s) increased in the PRBP bones. The NFATc1 mRNA variants are known to differ both at the 5′ end containing either exon1 or 2, and at the 3′ end that either terminates with exon 9b, or contains exon 9a through exon 11. By using primer pairs spanning exons 1 and 3, 2 and 3, 8 and 9b or 8 and 11, we observed a marked increase of exon 1 in the PRBP samples, whereas exon 2 was unchanged (Figure 8F). Moreover, exon 9b was enriched in PRBP over wild type, whereas exon 11 was not detectable in either genotype (Figure 8F, and data not shown). Thus, an NFATc1 mRNA variant transcribed from the P1 promoter and containing exons 1 and 9b was specifically induced in the PRBP bones.
The finding above raises the possibility that the increase in NFATc1 might contribute to the high bone mass in the PRBP mice, and further that inhibition of NFATc1 activity may be able to alleviate the phenotype. To test this hypothesis, we injected littermate PRBP animals daily with either FK506, a potent inhibitor of NFAT signaling, or vehicle, for one month starting at one month of age. As expected, the vehicle treatment did not alter the high bone mass phenotype of the PRBP animals. However, FK506 markedly reduced bone mass in the PRBP mice, especially in the femur (Figure 8G and 8H, Table 5). This reversal of the high bone mass occurred in the face of decreased bone resorption in the FK506-treated animals, as indicated by a significantly lower serum CTX level (Figure 8I), presumably due to the known role of NFAT in osteoclastogenesis [31]. The suppression of bone resorption may explain the observation that FK506 did not consistently correct the high bone mass in the tibia (data not shown). However, in one case where the serum CTX level was less affected by FK506, both the tibia and the femur were corrected (Figure S1). As controls, the wild-type littermates were subjected to the same inhibitor or vehicle treatment. Similar to a previous report [32], the trabecular bone mass was reduced in the wild-type animals by FK506, but the extent of reduction was modest compared to that seen in the PRBP mice (Figure S2). Overall, the results showed that NFAT inhibition could override the effect of RBPjk deletion on bone mass. Furthermore, because Hey1 directly inhibits NFATc1 expression, Notch-RBPjk-Hey signaling appears to inhibit bone formation in part by down-regulating NFATc1.
The present study establishes canonical Notch signaling as a critical mechanism for maintaining bone homeostasis under normal physiological conditions. In this capacity, Notch appears to function as a gatekeeper to ensure that a proper number of osteoblasts are produced during differentiation. Because removal of Notch signaling from the Osx-positive stage onward did not have an obvious effect, we propose that Notch mainly controls the transition from Runx2- to Osx-positive cells. Mechanistically, Notch signals through RBPjk to induce transcription of Hey1 and HeyL, which in turn inhibit osteoblast differentiation by suppressing both Runx2 activity [21] and NFATc1 expression (this study) (Figure 8J).
The current study not only demonstrates the stage- and receptor-specificity of Notch signaling during osteoblast differentiation, but also sheds light on the intracellular mechanism mediating Notch function. Although RBPjk was previously shown to mediate the effect of NICD overexpression on both chondrogenesis and preosteoblast proliferation [24], [40], this study establishes for the first time the importance of RBPjk in physiological Notch signaling within the osteoblast lineage. Moreover, the present study uncovers a direct regulation of NFATc1 expression by Notch signaling.
The relationship between Notch and NFAT signaling warrants further investigation. The direct suppression of NFATc1 promoter by Hey1, and the dominant effect of FK506 over RBPjk removal support the model wherein NFAT functions downstream of and opposite to Notch signaling in regulating bone formation. However, we cannot rule out that FK506 may have NFAT-independent functions, or that the systemically-delivered FK506 acted on other cell types to affect bone mass indirectly. Our finding that Hey1 binds to and suppresses the NFATc1 promoter is consistent with a recent report that Notch inhibited NFATc1 transcription in ST2 and primary osteoblasts [41]. However, NFATc1 is unlikely to be the sole effector, as simultaneous removal of NFATc1 and RBPjk with Prx1-Cre did not rescue the high bone mass phenotype caused by RBPjk deletion (data not shown). NFATc2 may play a redundant role as it was previously shown to stimulate osteoblast differentiation [32]. Indeed, Western blotting showed that several isoforms of NFATc2 were markedly increased in the PRBP bones over the control (Figure S3). The mechanism for this upregulation however, is currently unknown. Future experiments with simultaneous deletion of NFATc1 and NFATc2 will test the hypothesis that the two proteins redundantly mediate Notch function in bone.
Notch-RBPjk removal from the early limb mesenchyme led to a severe deficit in type I CFU-F in the postnatal bone marrow. This phenotype could reflect either a direct requirement of Notch signaling in the CFU-F cells, or a secondary effect due to changes in osteoblast differentiation. To distinguish these possibilities, we performed lineage-tracing experiments with the Rosa26 reporter mouse, and observed that the type I CFU-F cells were not targeted by Prx1-Cre although some stromal cells in the same culture were (data not shown). In addition, experiments with a transgenic mouse (TNR) that reports Notch-RBPjk signaling [42] revealed that some stromal cells but not the type I CFU-F cells exhibited canonical Notch signaling in vitro (data not shown). Thus, diminution of bone marrow mesenchymal progenitors, as reflected by CFU-F assays in vitro, was likely to be secondary to changes in osteoblast differentiation. In this scenario, assuming mesenchymal progenitors normally exist in equilibrium with Runx2-positive osteogenic precursors, we envision that unchecked differentiation of the latter due to Notch deficiency may lead to exodus of cells from the mesenchymal progenitor pool. Alternatively, the altered bone marrow environment due to the excessive bone mass in Notch-deficient mice may be unfavorable for either establishing or maintaining a normal mesenchymal progenitor pool.
A mechanistic understanding of the dominance of Notch2 over Notch1 awaits further studies. A similar dominance of Notch2 over 1 was observed during nephron formation in the mouse embryo [43], [44], whereas a dominance of Notch1 over 2 was reported in the skin [45], [46] as well as in osteoclasts [37]. The mechanism underlying the differential roles among Notch paralogs is currently unclear, but it could reflect differences in either expression levels or ligand-binding affinities among the different receptors within a given cell type. This model predicts that Notch2 is normally preferably activated in the osteogenic progenitors. Alternatively, Notch1 and2 may be similarly activated but Notch2-NICD is more potent than Notch1-NICD in suppressing osteoblast differentiation. However, Notch 1 deficiency was sufficient to cause ectopic ossification in human aortic valves [25]. Thus, the relative contribution of Notch1 versus 2 in suppressing the osteogenic program appears to be context dependent.
The effect of Notch signaling in osteoblast-lineage cells on osteoclast differentiation is likely to be complex. Although TRAP-positive osteoclasts were more abundant than normal within the metaphyseal trabecular region of the PRBP bones, the total bone resorption activity was relatively normal at 8 weeks of age. However, by 26 weeks bone resorption was more robust in the PRBP mice than the littermate control, and likely contributed to the rapid bone loss seen at this age. The mechanism for the age-dependent bone resorption phenotype is not understood at present, but likely involves additional factors beyond Notch deficiency in the osteoblast lineage. In addition, unlike the PRBP mice, the PN1, PN2, OsxRBP and ColIRBP mice did not display an obvious bone resorption phenotype at either 8 or 21 weeks, even though certain changes in M-CSF, Rankl and Opg were observed in these animals (Figure S4). The lack of bone resorption phenotype in the ColIRBP mice appears to be at odds with the previous report that deletion of presenilin 1 and 2 by 2.3Col1-Cre caused an increase in bone resorption at six months, although not at three months of age [23]. Besides trivial explanations such as slight age differences or genetic background variations between the two studies, the discrepancy could indicate that the previously observed effect was independent of RBPjk. In summary, the predominant function of physiologic canonical Notch signaling in the osteoblast lineage is suppression of osteoblastogenesis from the precursors. Thus, potential pharmaceutical inhibition of this pathway in osteogenic progenitors may be beneficial for bone formation.
The N1f/f [47], N2f/f [48], RBPjkf/f [49], 2.3ColI-Cre [36], Osx-GFP::Cre [35], Prx1-Cre [34], Hey1+/− [50], HeyL−/− [39] and NFATc1f/f [51] mouse strains are as previously described. The Animal Studies Committee at Washington University approved all mouse procedures.
Radiographs of mouse skeleton were generated using a Faxitron X-ray system (Faxitron X-ray Corp) with 20-second exposure under 25 kV. Micro computed tomography (μCT 40, Scanco Medical AG) was used for three-dimensional reconstruction, and quantification of bone parameters (threshold set at 200). Serum CTX assays were conducted with mice without feeding for 6 hours with the RatLaps ELISA kit (Immunodiagnostic Systems Ltd.). H&E, TRAP and picro-sirius red staining were performed on paraffin sections, following decalcification for postnatal samples. For dynamic histomorphometry of postnatal mice, calcein (Sigma) was injected intraparitoneally at 7.5 mg/kg on days 7 and 2 prior to sacrifice, and bones were sectioned in methyl-methacrylate. Bioquant II was used for quantification in both static and dynamic bone histomophometry. FK506 (Sigma) was dissolved in DMSO and was injected subcutaneously into one-month-old mice at 0.30 mg/kg/day for one month before harvest.
The CFU-F and osteoblast differentiation assays were preformed as previously described [21]. Only type I CFU-Fs were scored in the present study.
Transient transfections were performed as follows. ST2 were plated at 3×104/well in a 24-well plate overnight, and transfected with pCS2-Hey1, pCS2-Hey1-ΔHLH or empty pCS2 vector (0.2 µg) [21], pNFATc1-0.8P1 (0.1 µg) [52] and pRL-Renilla (0.01 µg, Promega) for 8 h using Lipofectamine (1 µl/well). HEK293T cells were plated at 4×105/well in a 12-well plate overnight, transfected with pCS2-Hey1, pCS2-Hey1-ΔHLH or empty pCS2 vector (0.4 µg), pNFATc1-0.8P1 (0.2 µg) and pRL-Renilla (0.02 µg) using Fugene (1.8 µl/well). The transfected ST2 or HEK293T cells were harvested at 48 hours after the beginning of transfection and subjected to dual luciferase activity assays (Promega).
Western analyses were performed with bone proteins extracted with RIPA buffer from tibiae and femora that were cut into small pieces after bone marrow cells were flushed out. The Notch 1 monoclonal antibody mN1A was as previously described [53], and the Notch 2 antibody (C651.6DbHN) was from Developmental Studies Hybridoma Bank at the University of Iowa. The antibody against RBPjk was from Cosmobio (Japan).
Real time PCR was performed with SYBR-Green (Roche) in ABI-7500 (Applied Biosystems) using cDNA reverse-transcribed from bone total RNA, extracted with Trizol (Invitrogen) from pulverized tibia and femur after removal of the bone marrow. Sequence information for the real-time PCR primers is listed in Table S1. The exon-specific primers for NFATc1 (Table S2) were as previously described [54], but the exons were renumbered according to the current NCBI nucleotide database. Semi-quantitative RT-PCR for NFATc1 was performed at an annealing temperature of 57°C for 40 cycles. GAPDH used as loading control was amplified for 30 cycles.
ST2 cells were infected with lentivirus to express a doxycycline-inducible Flag-Hey1 transgene. Flag-Hey1 expression was induced with100 ng/mL Doxycycline (Sigma D9891) for 12 hours. Chromatin and protein complexes were crosslinked for 10 minutes in 1% formaldehyde and flash frozen. The chromatin was sonicated to an average size of 200–450 bp using a Sonics Vibracell sonicator (model Vcx 500). Chromatin complexes were immunoprecipitated using an anti-Flag antibody (Sigma F1804). Immunoprecipitated DNA fragments were amplified by PCR using primers adjacent to a predicted Hey1-binding site within the P1 promoter of the mouse NFATc1 gene (5′ - TCTCGGTCTCACTCTGACGCA - 3′ and 5′ - TTCCCTCTTGTACACCTTTGCCCA - 3′), or primers near the P2 promoter approximately 4 Kb downstream (5′ - TCCGGGTTTACATAAACAAGCGGC - 3′ and 5′ - ACTGCACACCACGCTGAACAGGAA - 3′).
HEK293 cells that express a doxycycline-regulated Flag-Hey1 or -HeyL transgene were used for ChIP-seq analysis. Cells were induced with 50 ng/ml doxycycline for 48 hours to ensure a low-level expression and Hey1- or HeyL-containing chromatin was immunoprecipitated with a Flag antibody. Cells carrying the same transgenes but grown without doxycycline were used as control. Preparation of ChIP libraries, Illumina sequencing and data analysis were performed as previously described [55].
|
10.1371/journal.pcbi.1002874 | Cost and Effects of Different Admission Screening Strategies to Control the Spread of Methicillin-resistant Staphylococcus aureus | Nosocomial infection rates due to antibiotic-resistant bacteriae, e.g., methicillin-resistant Staphylococcus aureus (MRSA) remain high in most countries. Screening for MRSA carriage followed by barrier precautions for documented carriers (so-called screen and isolate (S&I)) has been successful in some, but not all settings. Moreover, different strategies have been proposed, but comparative studies determining their relative effects and costs are not available. We, therefore, used a mathematical model to evaluate the effect and costs of different S&I strategies and to identify the critical parameters for this outcome. The dynamic stochastic simulation model consists of 3 hospitals with general wards and intensive care units (ICUs) and incorporates readmission of carriers of MRSA. Patient flow between ICUs and wards was based on real observations. Baseline prevalence of MRSA was set at 20% in ICUs and hospital-wide at 5%; ranges of costs and infection rates were based on published data. Four S&I strategies were compared to a do-nothing scenario: S&I of previously documented carriers (“flagged” patients); S&I of flagged patients and ICU admissions; S&I of flagged and group of “frequent” patients; S&I of all hospital admissions (universal screening). Evaluated levels of efficacy of S&I were 10%, 25%, 50% and 100%. Our model predicts that S&I of flagged and S&I of flagged and ICU patients are the most cost-saving strategies with fastest return of investment. For low isolation efficacy universal screening and S&I of flagged and “frequent” patients may never become cost-saving. Universal screening is predicted to prevent hardly more infections than S&I of flagged and “frequent” patients, albeit at higher costs. Whether an intervention becomes cost-saving within 10 years critically depends on costs per infection in ICU, costs of screening and isolation efficacy.
| Within hospitals antibiotic-resistance of bacteria is common and it complicates treatment of bacterial infections. Screening of patients on admission for carriage of methicillin-resistant Staphylococcus aureus (MRSA) allows for strategies where carriers are treated with barrier precautions, e.g., isolation in single-bedrooms. At least theoretically, this should prevent spread of these bacteria. Several screen-and-isolate studies have been performed. However, the outcome was not unequivocal, possibly because clinical trials to determine the optimal screening strategy would necessitate long periods of follow-up due to stochasticity. In the absence of direct evidence we have used mathematical modelling to quantify the theoretical effectiveness and expenses of different screen-and-isolate strategies in hospitals with a high prevalence of antibiotic-resistant bacteria. We find that a strategy to screen patients who were previously known as carriers, possibly combined with screening of ICU-patients is the most cost-saving strategy for the best estimate of isolation efficacy of 25%. With a high efficacy of isolation all strategies are expected to become cost-saving compared to the do-nothing scenario.
| Methicillin-resistant Staphylococcus aureus (MRSA) may cause severe infections in hospitalized patients, such as bloodstream infections, surgical wound infections and pneumonia. These infections are associated with increased mortality rates, longer length of hospital stay and higher health care costs compared to methicillin-sensitive strains [1]. Typically, such infections are most prevalent in intensive care units (ICUs) [2]. Patient to patient transmission via – temporarily – contaminated hands of health care workers is considered an important mode of spread [3]. Therefore, prevention of nosocomial spread has been focused on reducing transmission opportunities through isolation measures and enhanced adherence to basic infection control practices, such as hand hygiene [4]. Nevertheless, despite multiple guidelines recommending these practices, infection rates due to MRSA remain high in most countries [5], [6].
It has become increasingly clear that rapid identification of carriage of MRSA, followed by implementation of barrier precautions for carriers, could be a powerful tool in controlling nosocomial spread [7]–[10]. However, screening all patients admitted to the hospital (universal screening) imposes a huge (financial) burden on a hospital system, and its benefits have not been unequivocally demonstrated [11]–[13]. Other screening strategies may, therefore, be more cost-beneficial, such as screening of ICU admissions only, screening of certain high-risk patients or screening of patients who were detected as MRSA-carriers at previous admissions. The optimal screening strategy may differ between settings, but evidence for the most cost-effective strategy in each setting is lacking. As a result, screening strategies vary substantially between hospitals, even within countries. Experimental trials to determine the optimal screening strategy for each of those settings would necessitate long periods of follow-up and huge financial investments. For such complex problems in the absence of direct evidence, mathematical modelling might offer the best alternative to quantify theoretical effectiveness and expenses of different screening strategies in different settings [14].
Here we have performed multiple scenario analyses of a mathematical model to compare the effects and costs of different “screen and isolate” (S&I) strategies, with special emphasis of such a strategy in ICU populations.
We have used an extended version of a previously described dynamic stochastic simulation model that contains three hospitals of 693 beds, each with an extramural population of 220,000 subjects [9]. Upon hospitalization, patients are usually admitted to “their own” hospital, but sometimes to one of the other hospitals (ratio 38 to 1). Each hospital comprises two types of wards: 36 18-bed normal wards with five health care workers (HCWs) per ward and five 9-bed Intensive Care Units (ICUs) with nine HCWs per ICU and 80 HCW per hospital with non-restricted patient contacts. After 8-hours, each shift of HCWs is replaced and HCWs are confined to a single ward during each shift. Upon hospitalization patients can be admitted to both types of wards. One of the most important changes of the model [9] is a change in the length of stay and mortality of patients in the different wards. In ICUs, 70% of the patients stay, on average, 1.5 days, with an ICU mortality of 2% per stay. After ICU discharge, these patients stay, on average, seven days in non-ICU wards, before hospital discharge. The remaining 30% of ICU-patients stay, on average, 10 days in ICU and have an ICU-mortality of 25% per stay. The ICU survivors remain hospitalized for, on average, 15 days in non-ICU wards. Patients without ICU admission stay on average 7 days. These parameters are based on patient data from a multi-center ICU study in the Netherlands [15]. Apart from transfer from ICUs to other wards, patients can be transferred between non-ICU wards, from non-ICU wards to ICUs, between ICUs, and between hospitals, all with different rates. Most important model parameters are shown in Table 1.
Individuals are also subdivided into “frequent” patients and “occasional” patients, distinguished by hospitalization rates of once per year (frequent) and once per ten years (occasional) (average sizes in the population being 20,000 belonging to the “frequent” group and 200,000 to the “occasional” group). Patients from either group can be admitted to both non-ICU and ICU wards and the mortality rate during hospitalization is the same for both groups. As a result, on average, 50% of the hospital population consists of “frequent” patients. In this study we use the “frequent” group as a high-risk population for MRSA carriage, and one of the possible screening strategies includes screening of “frequent” patients. All patients are either carrier of MRSA or uncolonized and susceptible for colonization and 1% of the colonized patients is 10 times more infectious (so-called superspreaders).
Infection control interventions are not based on the true colonization status, but on the available documentation of the colonization status only. Patients either (1) have documented carriage, (2) are not suspected of MRSA colonization (but could still be colonized), or (3) are suspected of colonization, e.g., after documented carriage during previous hospitalization or because of risk factors for MRSA carriage. Throughout this paper the latter patient category will be labelled as “flagged” patients. Importantly, we assume that the pathogen predominantly spreads in hospitals through cross-transmission and that there is hardly any spread in the community. Transmission occurs primarily between patients and HCW in the same ward, but occurs also, at a much lower rate, between wards. Transmission parameters are chosen such that the per admission reproduction number RA [16] is around 1.1 and 0.3 for ICU and non-ICU wards respectively, which corresponds in our do-nothing scenario to an endemic prevalence of 5% hospital-wide and of 20% in ICUs. Although most estimates of the prevalence of MRSA in ICUs and hospital wards are slightly lower than our values [17], in some ICUs MRSA-prevalence of 20% is not uncommon even with isolation measures [18].
MRSA blood stream infections may impact LOS, as was shown, by de Kraker et al. [1] and Wolkewitz et al. [19]. However, the attributable mortality and LOS due to MRSA colonization is limited [20]. As most patients colonized with MRSA do not have overt infections, the transmission dynamics of MRSA will be dominated by these patients. We, therefore, have chosen not to explicitly incorporate the additional LOS in patients with overt infections in the model, but to incorporate these additional LOS in the costs associated with an MRSA infection.
Results are based on 1,000 independent runs of the stochastic simulation model for a period of 10 years after implementation of interventions.
The microbiological screening method is, in all simulations, a rapid diagnostic test with turnaround time of 1 day and sensitivity and specificity of 93% and 96%, respectively [21], [22]. This is supplemented with conventional microbiological cultures with a turnaround time of 4 days and an assumed sensitivity and specificity of 100%. The conventional culture results are used as backup to correct false-negative and false-positive results of the rapid diagnostic test. MRSA carriers that are not detected by screening (due to absence of screening, false-negative results or acquisition of MRSA after screening) can be identified as carrier when conventional microbiological cultures, i.e., with a turnaround time of 4 days, are performed for clinical reasons at a rate of 0.03 and 0.3 per patient day for non-ICU wards and ICUs, respectively. The main reason for taking clinical cultures is the presence of fever.
We consider four different S&I strategies that are compared to a do-nothing scenario without any active screening at admission. In all four S&I strategies patients identified as MRSA-carrier will be “flagged” as such. The flagged status will be removed when such a patient has a negative conventional culture. The following S&I strategies are considered:
In all scenarios we assume that 12% of the admission screenings that should be performed according to the strategy are missed. In each scenario patients documented as carrier will be treated in isolation, which reduces the likelihood of transmission by 100% (perfect isolation), 50%, 25% or 10%, with 25% as default value [18], [23]. Screening of flagged patients, e.g., patients with a history of MRSA colonization [24], [25] and screening of ICU patients both are strategies that are used in hospitals across the world [12], [18].
Screening of “frequent” patients is not a strategy that is currently applied. Yet, since previous hospitalization is associated with MRSA colonization [26], we have chosen this strategy as an intermediate between screening flagged patients only and universal screening.
Although no limits to isolation capacity are assumed, we keep track of the number of patients in isolation to determine the volume of isolation capacity needed. The daily probability to develop an infection for a colonized patient is set at 0.7% and 0.2% in ICU and non-ICU wards, respectively, with sensitivity analysis ranges of (0.14%–1.4%) and (0.1%–0.3%) respectively. This implies that on average 3% and 1.4% of all patients in ICU and non-ICU wards will develop an infection in the do-nothing scenario (Table 1) [27], [28].
We estimated the costs of the different S&I strategies for a hospital using a 3% inflation rate per year. The analysis was performed from a hospital perspective and costs are reported in Euros using the price level of 2010. The default incremental costs from a hospital perspective of these infections (including a costs of prolonged due to MRSA-infection length of stay) were €30,000 in ICU and €1,000 in non-ICU wards, with ranges for sensitivity analysis of (€1,000–€40,000) and (€500–€2,500) respectively [29], [30]. The costs of a screening test performed at admission ranged from €2 to €102 with €20 as default value [31]. The incremental costs of treating a patient one day in isolation varied from €2 to €102 with €20 as default value [31].
For every S&I strategy and set of costs we determined (see supplementary Text S1) the time till the mean daily costs with the intervention strategy became lower than the mean daily costs in the do-nothing scenario (denoted as T). The 90% credibility intervals denote the uncertainty due to the inherent stochasticity of the dynamics of MRSA (with 5% of simulations yielding higher and 5% yielding lower results than the credibility interval).
Univariate sensitivity analyses were performed for all costs, the discount rate, and the probability to develop an infection. For the parameters with the highest sensitivity on results in the univariate analysis we investigated the dependence of T on the parameters.
The model predicts a decrease of the mean hospital-wide prevalence of MRSA in five years after the start of the interventions from 5% to, depending on the strategy, a value between 3.7–3.9% when isolation efficacy is 25% and 0.8–1.2%, 2.5–2.9% and 4.3–4.5% when isolation efficacy is 100%, 50% and 10%, respectively (Figure 1). The mean prevalence in ICU is predicted to decrease from 20% to 15.9–17.2% for isolation efficacy of 25% and 3.8–5.6%, 11.6–13.0% and 18.4–19.2% for isolation efficacy 100%, 50% and 10%, respectively. Ten years after the start of the intervention, the hospital-wide prevalence is predicted to be 0.2–0.5%, 1.9–2.3%, 3.3–3.8% and 4.2–4.4% and the mean prevalence in ICU predicted to be 1.0–2.2%, 8.5–10.5%, 14.7–16.5% and 18.2–18.5% for an isolation efficacy 100%, 50%, 25% and 10%, respectively (Table 2). Naturally, universal screening leads to the largest decline in the prevalence, while S&I of flagged patients results in the smallest decline.
Only when isolation efficacy >50% the strategy to screen flagged and ICU patients is predicted to reduce the prevalence in both ICU and non-ICU units more than screening flagged and “frequent” patients (Table 2). Universal screening leads to slightly lower prevalence in 10 years than other strategies. However, when isolation efficacy is low (10%), universal S&I is hardly more effective. S&I flagged patients only is less effective for all considered values of isolation efficacy. Differences in effects of interventions will increase with higher initial prevalence of MRSA (data are not shown).
Naturally, the number of isolation days needed varies considerably with the strategies and the isolation efficacy. The number of isolation beds needed increases immediately after the start of the intervention, most prominently for universal screening (Figure 1). The peak of the mean number of isolation days required for universal screening is 2.5, 2 and 1.4 times higher, as compared to screening flagged patients only, screening of flagged and ICU patients and screening of flagged and “frequent” patients, almost independently of isolation efficacy.
With isolation efficacy of 100%, 50%, 25% and 10%, screening of flagged and ICU patients will prevent on average 310, 175, 88 and 39 infections per hospital (168, 89, 40 and 14 in ICU) in 10 years time at the costs of €845.000, €909.000, €949.000 and €972.000 (Table 2). The costs of intervention measures per infection averted were lowest for screening of flagged patients only, being €632, €1.529, €3.598 and €8.447 for isolation efficacy levels of 100%, 50%, 25% and 10%, respectively. Universal screening was associated with the highest costs per infection averted, i.e., €19.918, €35.056, €67.857 and €164.093 for isolation efficacy levels of 100%, 50%, 25% and 10%, respectively. We have also compared the predicted number of infections prevented in ICUs and hospital, and the financial consequences of different strategies in high-endemicity settings (Figure 2 and Figure S1 in supplementary).
Whether a strategy will become cost-saving from the hospital perspective, as compared to the do-nothing scenario, critically depends on the isolation efficacy and the costs per infection averted. With our default efficacy of isolation of 25%, only two strategies are expected to be cost-saving within 10 years: screening of flagged and ICU patients and screening of flagged patients only. The expected total gain in 10 years time is estimated to be €420.000 and €850.000 respectively (Table 2). When efficacy of isolation is 10%, screening of flagged patients only is the only cost-saving strategy within a time window of 10 years. Universal screening is not expected to be cost-saving within 10 years even if efficacy of isolation is 100%.
In the do-nothing scenario the number of infections caused by MRSA, and, therefore, the costs associated with these infections, will be –more or less– constant in time (Figure 3). The costs associated with the intervention will initially lead to increased hospital costs. Yet, due to prevention of infections the hospital costs per unit of time will decrease and may – at a time T – become lower than in the do-nothing scenario (Figure 3); For all values of the efficacy of isolation, our model indicates that T is minimal for screening of flagged and ICU patients and screening of flagged patients only. Universal screening has the largest value of T, which is only below 10 years when isolation efficacy is 100% (Figure 3 and Table 2).
Univariate sensitivity analyses indicate that the total costs are rather insensitive to the costs of isolation, the costs per MRSA infection in non-ICU wards and the probability to develop an infection in non-ICU wards (Figure 4, supplementary figures S2, S3, S4). However, the total costs are sensitive to the costs associated with an infection in ICU wards and the probability per day for a colonized patient to develop infection in ICU wards. The dependence of total costs on the costs per screening test varies between screening strategies and is highest for universal screening.
Naturally the number of infections and the costs due to infections are more or less proportional to the number of colonized patient days. We, therefore, define a “constant of proportionality” as the “cost of an infection in ICU wards multiplied by the probability per day to develop an infection in ICU wards”. This constant can be interpreted as the costs due to infections per colonized patient day in an ICU ward. The total costs of the interventions are sensitive to “the costs due to an infection per colonized patient day in an ICU ward” divided by the costs of a single screening test performed at admission (Figure 5). We denote the ratio of these two costs by q. When screening is cheap and the costs of infections in ICU are high (q is large), all four strategies will have lower daily costs as compared to the do-nothing scenario within 10 years (T<10 years) for high values of isolation efficacy. With isolation efficacies of 25% or 10% only S&I of flagged and S&I of flagged and ICU patients will reach T within 10 years in the considered range of q. With decreasing values of q, the time T increases and at some critical value of q, the number of infections prevented by a strategy becomes too low to compensate for the costs of the intervention. This critical value of q depends on the strategy and the isolation efficacy (Figure 5).
The value of T is relatively insensitive to the other costs. Only for relatively low infection costs per colonized patient day in ICU and high costs per test, the costs per isolation day will significantly impact T (data not shown).
Changing the discount rate to either 2% or 4% hardly influenced the results.
Using a dynamic stochastic simulation model, we have evaluated four intervention scenarios to control the spread of MRSA under comparable in silico conditions. Although universal screening at hospital admission leads to the fastest decline in both the hospital-wide and ICU prevalence of MRSA, it also requires the highest investment costs and the longest time till return of investment. In our analyses, screening all patients at ICU admission and those previously detected with MRSA (so-called flagged patients) or screening of flagged patients only were almost equally cost-saving in a 10 years period and were both associated with the fastest return of investment. These strategies should, therefore, be seriously considered by hospitals that aim to control the nosocomial spread of MRSA.
Our findings are complimentary to those of two other modelling studies on screening for carriage with antibiotic-resistant bacteria in hospitalized patients. In one study, Hubben and co-workers compared the effects of PCR-based and chromogenic screening tests [32]. Determination of the optimal screening was not investigated in the current study, and we have, therefore, used a fixed time-to-result parameter. In the other modelling study, Robotham and co-workers investigated the effects of different screening tests in ICU patients, in combination with patient isolation and decolonisation [33]. The latter study did not include the effects of ICU-screening on the non-ICU hospital population and did not include the possibility of patients being readmitted while still colonized.
Yet, this is an important aspect of the dynamics of nosocomial MRSA as it explains why control measures may have not only a direct, almost instantaneous, effect on the prevalence of the nosocomial MRSA in the hospital, but also an indirect effect due to interruption of the so-called feedback loop; when less patients acquire colonization during hospitalization, less patients will be colonized upon readmission to the hospital (see supplementary Figure S5). This lower admission prevalence in time ensures that controlling spread of the nosocomial MRSA will become easier in time. Therefore, neglecting these feedback loop dynamics will underestimate the cost-savingness of interventions.
An important assumption of our model is that the pathogen spreads predominantly in health care settings. Interventions in health care settings will not be very effective in prevention of acquisitions in the community. With substantial spread in the community, a smaller fraction of the acquisitions can be prevented and also the fraction of the patients colonized on admission that are flagged will reduce. In the extreme case that transmission almost exclusively occurs outside health care settings, interventions in hospitals are ineffective and the cheapest strategy is the optimal one. For these reasons, our model is not applicable for community-associated MRSA, but is applicable for other pathogens with similar epidemiological characteristics as MRSA.
Although reductions in the occurrence of nosocomial MRSA infections have been reported [7], [10], multi-resistant Gram-negative bacteria, such as those producing extended-spectrum β-lactamases (ESBL) or carbapenemases are emerging in health care settings worldwide [34]. With no new antibiotics on the horizon to treat infections caused by these bacteria, effective transmission control strategies are needed. Yet, identifying the most effective control strategy for every possible setting through clinical trials seems impossible. Well-designed large clinical trials on rapid diagnostic testing of MRSA yielded highly variable results, varying from no effects on infection rates in surgical units [11], [12], [35] to 69.6% reductions in hospital-wide infection rates [10]. Moreover, the stochastic nature of ARB dynamics necessitates long study periods to avoid that conclusions are primarily based on chance events, rather than on true effects. We have, therefore, used mathematical modelling. Of note, mathematical models always are a simplification of real life complexities and cannot produce very precise predictions for a certain situation. For instance, we have assumed that all isolation measures were equally effective in all isolated patients and that all measures were executed with equal efficacy. One can easily think of scenarios in which these assumptions do not hold [36]. Therefore, the main value of modelling is the comparison of different scenario analyses, while keeping other important parameters constant, rather than providing exact values.
In doing so, our analyses identified screening of flagged patients and ICU patients as a very powerful control strategy, even reducing prevalence levels in non-ICU wards. The central role of the ICU in our model follows from two assumptions. First, many patients discharged from ICU are transferred to other wards. Therefore, prevention of spread in ICUs will reduce the frequency at which MRSA is introduced in other wards. Second, the likelihood of cross-transmission is higher in ICUs than in non-ICU wards. This assumption is motivated by the more frequent (and possibly even more intense) contacts between patients and HCWs, allowing HCWs to act as transmission vectors of MRSA. Moreover, antibiotic selective pressure is higher in ICUs than in non-ICU wards, which may increase the likelihood that a HCW will pick up a pathogen during a physical patient contact and that another patient will be successfully colonized after being contacted by a temporarily contaminated HCW. Finally, the severity of disease of critically ill patients in ICU wards makes them more susceptible to acquire colonization with MRSA than patients in non-ICU wards. Several studies indeed support the potential effects of ICU-screening on hospital-wide resistance levels [37].
With regard to the costs of interventions, our analyses were most sensitive to the costs associated with an ICU-acquired infection caused by MRSA. Many studies have quantified the costs of ICU-acquired bacteremia and ventilator-associated pneumonia [30] and these estimates were all in the range of the €30,000 that we used. However, these costs sensitively depend on the additional length of stay that can be ascribed to infections, which is difficult to determine, see e.g. [38], [39]. Another important aspect is the role of the ICU in the patient flow. We have used data on patient admissions to 13 ICUs in the Netherlands. Naturally, patient flow may be different in other hospitals.
One of the simplification of the model is that patients should be colonized with MRSA before they are at risk of getting an infection with MRSA, i.e., we did not explicitly incorporate that some patients may acquire MRSA infection directly without being colonized first, i.e., due to invasive medical procedures. A slight increase in the daily probability for colonized patients to acquire an infection would lead to the same ratio of colonized and infected patients. Therefore, our sensitivity analysis on the daily probability for colonized patients to acquire an infection can also be interpreted as a proxy for a sensitivity analysis to the parameter which determines how often patients acquire an infection without being colonized.
We also assumed that the rates of conventional microbiological cultures performed for clinical reasons are independent of screening on admission (0.03 and 0.3 per patient day in non-ICU and ICU wards). We have assumed that a clinical suspicion of infection is the main reason for obtaining clinical cultures, and that screening for MRSA-carriage on admission reduces the frequency of obtaining clinical cultures in case of a clinical suspicion of infection.
Our model contains many parameters and some parameter values are unknown, whereas others may differ between hospitals and countries. We have based our values on data from the literature and from our own hospital, where possible. To fully capture the effects of parameter uncertainty we would have considered to perform a probabilistic sensitivity analysis (PSA) for all parameters simultaneously, as was performed by Robotham et al. [33]. However, due to the higher complexity of our simulation model, as compared to the model of Robotham et al., this was computationally unfeasible. We, therefore, had to restrict our sensitivity analysis primarily to univariate sensitivity analysis. As a result, there may be more uncertainty in the results as we have presented here.
We did not include decolonization of detected carriers as a measure to control MRSA. Naturally, adding this measure (if successful at low costs) would increase intervention effects and would make the duration till return of investments shorter. Although persistently colonized HCWs were included as potential sources for MRSA transmission, we did not include screening and decolonization of them as intervention measure. This intervention measure would - in most settings – only slightly enhance the control of MRSA transmission, at the cost of significant expenses due to the necessity to replace colonized HCWs.
The (cost)-efficacy of admission screening strategies critically depends on the effectiveness of the infection prevention measures taken when a carrier of MRSA is detected. If these measures are not very effective, it may not be wise to invest lots of efforts in detecting carriers. The effectiveness of barrier precautions has been sufficiently high in the Netherlands and the Scandinavian countries to prevent high prevalence levels of MRSA. However, it is still debated whether patient isolation prevents transmission at all [23], and a recent estimate indicated that the efficacy is in the order of 25% [18], We, therefore, advocate to perform more clinical studies to determine the efficacy of decolonization, isolation or cohorting measures in different settings.
In conclusion, our study demonstrates marked and robust differences in the costs and effects of different infection control measures for MRSA. Because of the central role of ICU wards in patient flow in hospitals, the vulnerability of ICU patients to infections caused by MRSA and the high costs associated with these infections targeted infection control measures in ICU wards are likely to be the most effective and cost-saving from a hospital perspective.
|
10.1371/journal.pcbi.1002175 | Using Structure to Explore the Sequence Alignment Space of Remote Homologs | Protein structure modeling by homology requires an accurate sequence alignment between the query protein and its structural template. However, sequence alignment methods based on dynamic programming (DP) are typically unable to generate accurate alignments for remote sequence homologs, thus limiting the applicability of modeling methods. A central problem is that the alignment that is “optimal” in terms of the DP score does not necessarily correspond to the alignment that produces the most accurate structural model. That is, the correct alignment based on structural superposition will generally have a lower score than the optimal alignment obtained from sequence. Variations of the DP algorithm have been developed that generate alternative alignments that are “suboptimal” in terms of the DP score, but these still encounter difficulties in detecting the correct structural alignment. We present here a new alternative sequence alignment method that relies heavily on the structure of the template. By initially aligning the query sequence to individual fragments in secondary structure elements and combining high-scoring fragments that pass basic tests for “modelability”, we can generate accurate alignments within a small ensemble. Our results suggest that the set of sequences that can currently be modeled by homology can be greatly extended.
| It has been suggested that, for nearly every protein sequence, there is already a protein with a similar structure in current protein structure databases. However, with poor or undetectable sequence relationships, it is expected that accurate alignments and models cannot be generated. Here we show that this is not the case, and that whenever structural relationship exists, there are usually local sequence relationships that can be used to generate an accurate alignment, no matter what the global sequence identity. However, this requires an alternative to the traditional dynamic programming algorithm and the consideration of a small ensemble of alignments. We present an algorithm, S4, and demonstrate that it is capable of generating accurate alignments in nearly all cases where a structural relationship exists between two proteins. Our results thus constitute an important advance in the full exploitation of the information in structural databases. That is, the expectation of an accurate alignment suggests that a meaningful model can be generated for nearly every sequence for which a suitable template exists.
| Most protein sequences do not have an experimentally determined structure and at least 40% do not even have a sequence homolog with a known structure [1]. Nevertheless, the current Protein Data Bank (PDB) [2] is thought to represent structure space nearly exhaustively [3]–[5]. Therefore, for most proteins, a structural homolog that can serve as a “template” for modeling at least part of its structure is likely to exist. However, the degree of sequence similarity will generally be too low to allow a template to be detected or for an accurate sequence alignment to be found [6]. A central problem is that current alignment methods based on dynamic programming (DP) [7] generate the unique “optimal” alignment (the alignment producing the highest score based on a residue-residue similarity score and a gap penalty), while the “correct” alignment (producing the most accurate model) is not guaranteed to be optimal in terms of this score at low sequence identity ranges.
Numerous variations of both the residue-residue similarity score and gap penalty have been developed to address these issues. Individual residue-based scoring functions have been replaced with more complex profile-profile [8]–[10] and environment-dependent methods [11]–[13]. Recognizing that affine gap penalties typically over-penalize long gaps, several studies have described the probability of a gap as a function of its length or location in the structure with the goal of penalizing it appropriately [14]–[19]. Threading methods [20], [21] incorporate an energy term into the alignment procedure, but they face the drawback of not being compatible with the traditional DP algorithm [22].
Even with these more sophisticated approaches, there are still many issues that will confound the generation of an accurate alignment. Moreover, it is generally necessary to consider an ensemble of alternative alignments in order to produce an accurate model at low sequence identity ranges. Such ensembles are frequently called “suboptimal” since by necessity they have lower scores than the optimal alignment produced by DP. A variety of suboptimal sequence alignment schemes have been reported. Waterman [23] produced an ensemble of alternative alignments by changing the dynamic programming algorithm to return all alignments with scores within a small difference, δ, from that of the optimal alignment. However, the difference between the DP scores of the correct alignment and the optimal sequence alignment can be significant, especially for remote homologues. Increasing δ until it encompasses the correct alignment often produces an unmanageably large ensemble. Keeping δ small returns a more reasonable number, but the alignments tend to deviate negligibly from the optimal alignment.
Saqi and Sternberg [24] adapted this approach to return a more diverse ensemble by penalizing an alignment that is similar to one previously determined. John and Sali [25] used genetic algorithm operators to splice and re-combine alignments in order to achieve the same goal. Chivian and Baker [26] produced alternative alignments by systematically varying the parameters in their optimal alignment method. Each alignment in their returned ensemble was therefore “optimal” (ie. highest-scoring) under a different set of conditions. One problem faced by all suboptimal methods is how to adequately sample the gigantic space of possibilities. Jaroszewski et al [27] sought to explore the size of alignment space by examining pairs of small and medium-sized proteins (seven or fewer template secondary structures). Even though only “significantly different” alignments were enumerated by disallowing gaps in template secondary structures and ignoring alignment variations in loop regions, tens of millions of alternative alignments were required in some cases to generate the correct one.
We describe here a new method to generate suboptimal alignments, S4 (Sampling Shifts in Secondary Structures), that takes an approach that is fundamentally different from the standard dynamic programming algorithm. In validation tests that we describe below, we show that S4 is highly effective at producing an accurate alignment within a set of 100 top-ranked alternatives and can almost always produce such an alignment within a set of 1000 alternative alignments. The utility of the S4 approach is most evident when the query/template sequence identities are low, but S4 also improves accuracy when the homology is clear. Our results are shown to constitute a significant improvement over DP-based alternative alignment methods, which we show is due to unique features of the algorithm, in particular to the effective use of the 3-dimensional structure of the template. The ability to generate a small set of alignments likely to contain the correct one suggests that S4 offers the possibility of significantly improving the accuracy of homology models, extending the number of sequences that can currently be modeled based on existing structures in the PDB.
A flowchart for the S4 algorithm is shown in Figure 1. The method starts by searching the DP matrix for a set of short, ungapped alignments bounded by individual template secondary structure elements (SSEs). The rationale is that whatever sequence similarity may exist between query and template will more likely be in SSEs than loop regions. To generate a global alignment, pairs from a high-scoring set of “primary” fragments are connected with lower-scoring “secondary” fragments. This is a crucial feature of S4. In particular, we find that correctly aligned fragments can generally be identified within a very small set of primary fragments, significantly reducing the combinatorial complexity of the alignment problem. This characteristic, combined with the requirement that alignments containing the fragments be structurally plausible (see Materials and Methods), improves accuracy in regions where the relationship between the query and template is less clear. The constraints also allow S4 to remove many alignments from consideration through the application of filters that identify geometrically or energetically unreasonable alignments based on knowledge of the template structure. Filters are also applied to check for redundancy in order to ensure that the alignments represent unique regions of alignment space. (A detailed description of each step of the S4 algorithm and the filters applied is provided in the Materials and Methods.)
S4 was tested on a set of target sequences from the CASP [28] experiments (T0129–T0359). Potential templates for each target/query sequence were identified by structurally aligning the native structure to other proteins in the PDB using the ska program [29], [30]. Templates were then selected based on a set of criteria (see Materials and Methods) to ensure that an alignment existed between the template and query structure that would produce a model with a TM-score [31] >0.5, and also that S4 would not be run on sequences longer than 350 residues. The resulting test set contained 3,342 query sequence/template pairs and was heavily populated by those with low sequence identity: over 90% of all pairs had less than 20% identity and more than 60% had less than 10% identity. Overall, there were 137 queries with an average of 24 templates each that satisfied all the criteria. The queries represented at least 65 different SCOP folds (some targets are not classified in SCOP).
We define the correct alignment to be the structure-based sequence alignment between the query and template and evaluate the performance of S4 by comparing it to three DP-based approaches, HMAP [10], hhalign [32], and SP3 [33]. We also compare against a DP-based suboptimal alignment method [23]. We calculate the accuracy of an alignment in different ways. While an alignment algorithm should ideally be able to reproduce the structure-based sequence alignment residue-by-residue, several issues make this an overly sensitive measure of success. For example, consider a situation in which a template contains a helix with an axis that is at an angle with respect to that of the topologically equivalent helix in the query. Because of such differences between the template and query structures, no alignment in this region can be considered strictly correct even though there may be residues in the query and template that occupy roughly equivalent positions in space. The same difficulty occurs in the alignment of loop regions and also in β-strands, where β-bulges can affect alignment accuracy. However, it is clearly desirable for an alignment algorithm to pair residues in topologically equivalent SSEs, even if this pairing does not exactly correspond to the structure-based sequence alignment because of conformational differences.
Because of these issues, we use three measures that reflect a variety of characteristics. We first use a measure called “inter-alignment distance” (IAD). As described in Materials and Methods, IAD corresponds to the average deviation of the position of residues in a given alignment from their correct position in the structure-based alignment. An IAD of 2 implies that, on average, each residue is shifted by two away from its position in the correct alignment, but also implies that topologically equivalent SSEs in the template and query have been correctly paired. Thus IAD is a measure of overall alignment quality. To calculate how well the correct alignment is generated on a residue-by-residue level, we use a measure that we call FDS2, adapted from the FD measure of Sauder et al. [34]. This measure is simply the percentage of residues that are within 2 from their position in the correct alignment, with the restriction that this is calculated only in regions corresponding to template SSEs. This restriction results in a more informative alignment metric, since measuring accuracy in the structurally equivalent—but conformationally dissimilar—loop regions of remote homologs imposes a correspondence of residues that is not necessarily meaningful. Finally, to determine whether the models produced from the alignments are actually useful, we directly compare models to the native structure using the TM-score [31].
Figure 2 plots IAD for the best alignment generated by S4 and the single optimal alignment produced by HMAP, hhalign, and SP3. Points in the figure represent individual query/template pairs and are ordered according to the IAD of the optimal alignment generated by the different methods (i.e., moving left-to-right in the graph corresponds roughly with query/template pairs that range from higher to lower sequence identity). Figure 2 illustrates a central difficulty with most DP-based alignment methods. That is, at the higher range of sequence identities, most methods produce a reasonably accurate alignment, but there appears to be a threshold beyond which an accurate alignment becomes impossible when considering a single, optimal alignment. On the other hand, S4 generates an alignment with improved accuracy at all sequence identity levels and the improvement is quite dramatic at lower identities when the optimal alignment is severely flawed.
Table 1 presents this explicitly, showing IAD values for the different methods averaged over all pairs in several ranges of sequence identity. In the 0–5% identity range, the average IAD for the optimal alignment is over 13, implying that many topologically equivalent SSEs are not correctly paired. In contrast the average IAD for the best S4 alignment found in the top 1000 is 2.3, indicating that S4 is able to find good alignments even in the low identity regime. We note that this is true whether or not the template is identified as a significant hit by the individual methods (E-value<10 for HMAP, E-value<0.001 for hhalign and Z-score<−0.5 for SP3). In Figure 2, the IAD's for the best S4 alignments are colored in light or dark green if the template for that case was identified as significant by the corresponding alignment method, and in red for those templates that are not considered significant.
Of course, there is an inherent difficulty in comparing the performance of a method which generates an ensemble to a method which generates a single alignment. In fact, the optimal alignment is the most accurate in many cases (about 30% of the time for hhalign and 21% for SP3) and is more often than not in the top 5% in a set of 1,000 alignments ranked by IAD. The average rank is ∼200 however, so there is generally room for improvement, and our main point here is not that the optimal alignment shouldn't be used, but that an ensemble is necessary to generate an alignment that makes an accurate model, especially for highly remote/query template pairs. In practice, the optimal alignment would always be part of such an ensemble.
To determine the extent to which the improvement in alignment quality of S4 relative to the optimal alignment is due simply to the increased number of alignments generated, we also compared S4 to two versions of the conventional DP-based Waterman algorithm for generating alternative alignments [23], which have been implemented in-house as part of HMAP [10]. Figure 3 shows the results. As discussed above, while the IAD is effective at measuring overall alignment accuracy, it does not define the fraction of residues that are within a specified distance from their position in the correct alignment, and thus in Figure 3 we use the FDS2 measure. We also compare to two versions of the DP-based suboptimal alignment algorithm. A problem with the strict implementation of this algorithm is that alternate alignments can be generated that are not meaningfully different because variations in loop regions produce essentially equivalent models. Thus, we also implemented a modified algorithm which ignores such alignment variations. In the Figure the standard implementation of the algorithm is referred to as “unconstrained Waterman” and the modified version is referred to as “constrained Waterman” (see Materials and Methods and Figure S6 in Text S1 for more detail).
Figure 3 depicts the best FDS2 in the ensemble from each method as a function of the FDS2 of the optimal alignment. The vertical distance above the dotted line represents the improvement over optimal for the best alternative alignment generated. S4 is seen to significantly outperform the DP-based optimal and suboptimal algorithms, particularly when the optimal alignment is flawed. Even the best alignment out of the top 100 S4 alignments is significantly better than the best out of 1000 from the other DP-based methods. A further improvement in accuracy can be obtained by modeling the ensemble of 1000 alignments and using the pG score [35], [36], to select the top 100 alignments based on the quality of the models they produce. Figure 3 also shows the same data as a function of sequence identity. Again, we see that S4 offers a significant improvement compared to all DP-based methods for aligning remote homologs, even when using an ensemble one-tenth as large.
The results shown in Figure 2 suggest that S4 generates alignments that are much improved over DP-based optimal methods, but since the IADs of the best S4 alignments are not 0 (i.e., the S4 alignments are not identical to the correct alignment) an important question is whether these improved alignments produce improved 3-dimensional models. To examine this, we made models from the optimal alignment, the correct, structure-based alignment and all alignments in each S4 ensemble for each pair in the data set. The models were then compared to the native structure using the TM-score [31] with results shown in Figure 4. It is evident from the figure that many of the models produced by S4 constitute a significant improvement over the one produced by dynamic programming. The improvement in model quality is most dramatic when the model produced by the optimal alignment is inaccurate. Notably, the best models from S4 are often quite close to the accuracy of the model from the correct alignment. The line labeled “S4 90%” represents the 90th percentile cutoff within each segment, indicating that S4 produced a model for 10% of the pairs that was as accurate as possible, i.e., as good as the model produced by the correct, structure-based alignment.
Figure 4 also shows that evaluating models can significantly reduce the number of models that need to be considered. “S4 100 (pG)” represents the best model of the 100 top-ranked models in the ensemble as determined by the pG score. The proximity of this line to “S4 1000” demonstrates that the pG score consistently ranks the best model from the entire ensemble in the top 100. It is important to be able to reduce the ensemble size in this manner without removing the best models, if further processing of the models is to be carried out (i.e., refinement, minimization, etc.)
Since they use the same scoring function, the improved performance of S4 compared to HMAP seen in Figure 3 is not due to better scoring, but to a broader sampling of alignment space while also avoiding regions that would produce poor alignments. The latter feature is achieved with the rules and filters discussed in Materials and Methods. The ability of S4 to sample broadly should manifest itself in greater sampling at both the residue and whole alignment levels. Indeed, in Figure 5A, we see that S4 samples 3–5 times as many different query residues at each template position compared to the DP-based methods with the same ensemble size.
In Figure 5B, we choose the structure-based sequence alignment as a reference and report the standard deviation of the IAD for all alignments in an ensemble. A low standard deviation indicates that many of the alignments in the ensemble are clustered around a particular distance from the correct alignment, which implies that they are in a narrow region of alignment space. For DP-based methods that region will be centered on the optimal alignment (see Discussion below). We see in Figure 5B that S4 samples broadly within its small ensemble, but can still return an alignment closer to the correct alignment than the DP-based methods (see Figure 3).
A specific example illustrates S4's approach to sampling alignment space. Figure 6 depicts a query/template structure alignment along with a listing of their respective SSEs and several ways they are matched in the alignments produced by different methods. The query is the N-terminal domain of KaiA, a non-enzyme circadian clock protein [37] and the template is a single domain of DXR, which is a reductoisomerase [38]. The two proteins are classified as belonging to different folds in SCOP [39] and have less than 2% sequence identity.
Despite being classified as different folds, these two proteins have high overall structural similarity and thus an alignment exists that would generate an accurate model. The structural alignment for this pair describes the proper correspondence of all eight of the SSEs that are common between the template and query, as depicted in the first two rows of the alignment shown in Figure 6. The DP-based optimal alignment contains major flaws and only four out of eight SSEs are in proper correspondence. The poor performance of the DP-based approach is due more to issues with sampling alignment space than to the absence of a detectible sequence relationship between the two proteins. In fact, the eight fragments representing correct correspondences of query and template SSEs were all highly-ranked fragments (or “primary” fragments, in the terminology used in Materials and Methods) as determined by the same HMAP scoring function. All eight correct fragments were chosen within the first 58 (out of a total of 122 used). This local similarity between the profiles is consistent with other local structural, functional and sequence similarities that have recently been described between proteins that have significantly different topologies [40], [41].
Overall, out of an ensemble of 1,000 alignments, the best alignment from S4 has an IAD of 0.56 and an FDS2 of 97% compared to the correct alignment and the TM-score of the corresponding model is 0.50 (compared to a TM-score of 0.57 for the model built from the structure-based alignment). In contrast, the best alignment generated by the constrained Waterman approach (out of an ensemble of 1,000) had an IAD of 15.4. That the improvement in accuracy of S4 is due to differences in sampling can clearly be seen by calculating average IADs of the alignments in each ensemble, but here with respect to the DP-based optimal alignment instead of the correct alignment. The constrained Waterman approach is “trapped” near this incorrect alignment (average IAD of 0.3 and standard deviation of 0.8). Even though S4 samples the DP-based optimal alignment, it also searches far from this alignment (average IAD of 9.4, standard deviation of 4.6, and a maximum IAD of 25.4).
Though we have shown that S4 generates accurate alignments to almost every template appropriate for a given query sequence, we have not discussed how to identify these templates or how to select the correct alignment from the S4 ensemble. However, the results shown in Figure 2 suggest that S4 can be a valuable component of currently used homology modeling strategies. That is, based on the results in Figure 2, most of the appropriate templates that we identify based only on structural similarity to the native structure are recognized as significant using the scoring function associated with the different methods we compare to in the figure. But for a significant majority of these templates an accurate alignment is not possible, at least considering a single alignment generated based on the techniques and information used in the different alignment strategies. This severely limits the number of templates which can be considered useful even if they are recognized.
By building models from templates selected by other methods, but based on alignments generated by S4, these templates can be exploited assuming an accurate model evaluation procedure can be applied. There is a wide array of such tools that range from measures of the suitability for residues to be in a given environment (e.g., Verify3D [42]), to statistical potentials such as D-FIRE, Prosa, or Anolea [43], to all-atom molecular dynamic simulations to estimate the thermodynamic stability of the model (GROMOS [44]). The choice of the best method of evaluation is a complicated one and goes beyond the scope of the current paper where we have focused on S4 as an alignment tool. Nevertheless, for a third of the cases used in our benchmarking, the model with the lowest pG score differs negligibly from the best possible model available from the ensemble (i.e., the best model and the model selected based on pG have TM-scores with respect to the native structure that are within 0.05 of each other). Further, it has been shown that construction of 3D models followed by evaluation using a statistical potential can be used to distinguish true from false homologs when the sequence relationship is ambiguous [35], [45]. These results suggest that more accurate alignments obtained using S4 should significantly expand the number of good templates and models that can be found.
Since S4 produces accurate alignments in nearly every case where there is a structural similarity that leads to an accurate model, this suggests that, using a model-building and evaluation procedure, templates with scores that are outside the range of what is usually considered significant for a particular method could also be identified. Using the widely used tool PSI-BLAST as an example, about half of the templates in our data set were identified as significant (where we define this loosely as E-value<10). As shown in Table 2, in these cases S4 can generate more accurate alignments, in terms of the FDS2 score, than PSI-BLAST. Even for those templates with E-values that are not typically considered useful, (10−3<E-value<10), S4 is able to find an alignment that is more than twice as accurate and S4's performance decreases only slightly among the pairs that are not detected at all by PSI-BLAST, which comprise over half the benchmark set. The results shown in Figure 2 indicate that the same conclusion holds no matter what the method used to identify templates. Moreover, preliminary work using a protocol in which templates are selected by PSI-BLAST, models are built from every alignment in the S4 ensemble and evaluated using the pG score as well as other criteria suggests that good templates in this E-value range can be identified with high precision.
As shown in Figure 5, the primary difference between S4 and other alternative alignment methods is the manner in which alignment space is sampled. The central advantages of S4's sampling are that it generates enough diversity in a small ensemble so that an accurate alignment can be found, while limiting on the number of potential alignments that need to be considered (<10 million, see Materials and Methods). In contrast, as we show in Figure 5, DP-based sampling is highly local as a result of the fact that DP must start with the optimal alignment and successively generate other alignments in decreasing order based on their score. This severely limits the amount of diversity that DP can generate and ensures that many more alignments would need to be considered (at least an order of magnitude and probably more) when the DP-based score of the correct alignment is far below the optimal. A low DP score is typical for the more remote query/template pairs in our benchmark, since the correct alignments frequently require long indels or pass through low-scoring regions of the alignment matrix. Moreover, application of the structural filters used in S4 would not be expected to improve this situation, since there are a significant number of inaccurate alignments that satisfy them. Again, if an inaccurate alignment had a better DP-score than the correct one, a DP-generated ensemble would be trapped near the inaccurate alignment, since the local sampling inherent in DP would most likely not generate alignments that break the structural rules in any manageably small ensemble.
While it appears necessary based on our results to consider an ensemble in order to find an accurate alignment, especially for highly remote query/template pairs, it is clearly beneficial to consider the optimal alignment as well. As mentioned above, the optimal DP-based alignment is the most accurate (in an ensemble of 1,000 S4 alignments and 1 optimal alignment) for many cases in our benchmark. An ideal modeling strategy then would be one that generates an ensemble with S4 and simply adds the optimal alignment to that ensemble. This would ensure the best of both worlds at no increase in computational cost. Moreover, the S4 algorithm is independent of the underlying residue-residue scoring function employed. In the work presented here, the HMAP profile-profile method was used, but the sampling algorithm used in S4 could be applied using any other residue-residue scoring function. Therefore, if better scoring functions are available or if future improvements to scoring functions are able to raise the level of accuracy of the DP-based methods, S4's performance using the same scoring function should improve as well.
To ensure that a meaningful structural relationship existed within each query-template pair, several conditions had to be met: the protein structural distance (PSD) [30] could not exceed 0.5 (corresponding to a maximum RMSD of about 3.5 for aligned residues); the sequence identity was less than 50%; and a “pseudomodel” of the query built from the aligned portions of the structure-based sequence alignment and based on the template structure had to return a TM-score [31] of 0.5 or greater compared against the native query structure. A pseudomodel is constructed by simply copying the backbone and Cβ coordinates of residues of the template mutated to the identities of the corresponding aligned residues in the query (unaligned residues are ignored). Also, proteins of length greater than 350 residues were not considered.
The S4 algorithm is described in detail as six distinct steps below (see Figure 1). Overall, the algorithm proceeds as follows. First (Steps 1 and 2), short ungapped alignments entirely contained within template SSEs (‘fragments’) are selected based on their sequence similarity. Any subset of fragments, listed in order from N to C-terminal, is called a fragment alignment. Next, all fragment alignments are exhaustively enumerated and those that pass a set of tests for modelability, are kept. Finally, full alignments are constructed from fragment alignments. The full alignments are generated by standard dynamic programming with the constraint that DP is applied only to a narrow region (defined by the fragment alignment) of the dynamic program scoring matrix. A schematic for the different steps in the process is provided in Text S1, as well as a specific example of how S4's features lead to improvement in alignment accuracy.
Figure 1 shows a typical dynamic programming matrix with the query sequence along the side and the template sequence across the top. The template sequence is divided into columns defined by its secondary structure elements. A diagonal contained within a column is called a “fragment” and represents a short ungapped alignment of the query to the template. To start the alignment process, an initial set of “primary” fragments is identified as follows. Each fragment, (i.e., every diagonal in every column) is examined and is assigned a score that is the sum of the residue-residue similarity scores of the aligned pairs it contains, calculated based on the HMAP profiles [10] of the query and template sequences. The fragment from each column with the highest normalized score (the profile-profile similarity score divided by the length of the fragment) is added to the list of “primary” fragments (black lines in Figure 1). Each template SSE will contain at least one primary fragment and usually several more.
For every pair of primary fragments we perform a recursive search for “secondary” fragments to fill in the region defined by the fragment endpoints, if the fragments in the pair belong to non-consecutive SSE's. For example, in Figure 1, two secondary fragments are chosen for being the highest scoring secondary fragments that are “adjacent” to primary fragments PF1 and PF2. (An adjacent fragment is contained in a neighboring SSE and is on the same or a nearby diagonal.) Other secondary fragments are chosen by virtue of being high-scoring or in an SSE whose deletion would violate the alignment rules (e.g., a missing core strand, see below). This process continues recursively until all regions between non-consecutive fragments in a subset have been filled in.
A “fragment alignment” is a list of primary and secondary fragments in order from the N- to C-terminal. Two examples of fragment alignments are shown in Figure 1. The blue and green lines both run alongside two sets of four fragments (which share a common first member). Fragment alignments such as these will later form the basis of full alignments (constructed as described below).
To enumerate all fragment alignments that are possible within our set of primary and secondary fragments, S4 connects the N-terminal pseudo-fragment (upper-left corner of Figure 1) to each downstream primary fragment (either directly or through subalignments of secondary fragments). This process progresses to further downstream fragments until all alignments end at the C-terminal pseudo-fragment (bottom-right corner of Figure 1). After any connection between fragments is established, a set of conditions must be met. If an alignment fails to meet one of these conditions (described below), the enumeration process is discontinued for that particular path. (Some conditions can only be applied when the C-terminal is reached). It should be noted the total number of possible fragment alignment can be calculated efficiently during the above process, and no new fragments are added once the total number of alignments exceeds 10 million.
Some of the conditions placed on the fragment alignments are based on the properties of the alignment itself and some are based on a 3D pseudomodel of the query. The conditions that must be met by each alignment/pseudomodel are described below.
At this stage in the process, no full alignments in the normal sense have been created, only fragment alignments, which are just lists of fragments. A round of alignment sampling using the full sequences of the query and template is used to generate a final alignment from each fragment alignment. In this final step, alignments are restricted to a specific region of the dynamic program matrix. The boundaries of the region extend 3 residues above and below the fragments in each fragment alignment. The loop regions are constrained only by the boundaries of the surrounding fragments (dashed lines in Figure 1). Alignment sampling is carried out using the constrained Waterman approach. That is, we apply this algorithm in regions of alignment space that we expect to be unique based on the structure of the template. Again, a pseudomodel is constructed for each alignment which is scored with DFIRE [47] as described above. The alignment with the best/lowest energy is selected to represent the original fragment alignment.
The S4 algorithm typically generates thousands of fragment alignments. A single, full alignment is generated for each one, starting with the highest-scoring, until N unique alignments have been found, where N is the ensemble size chosen by the user. The score of an alignment is simply the sum of the similarity scores of the paired residues in the original fragment alignment minus a flat penalty for each inserted residue. The insertion penalty only applies to residues inserted between template residues and is therefore used to encourage insertions at the termini. Deletions are not penalized since we found that structural considerations enabled us to disallow unreasonable gaps without an explicit penalty. A worked example illustrating each step is provided in Text S1.
We calculate the distance between any two alignments using a measure similar to the gALD measure developed by Chen and Kihara [48]. If we plot two alignments of the same two sequences on the dynamic programming matrix (Figure 1, blue and green lines) there is a region between them for which we can calculate the area. Dividing this area by the length of the template yields an average height of this region, which can be interpreted as the average distance that a query residue in one alignment is shifted from its position in the other. This average distance we have termed the IAD and it should be considered to have units of residues. This measure is quick to calculate and useful for determining if two alignments occupy the same region of alignment space.
The unconstrained Waterman and constrained Waterman in Figures 3 and 5 are implementations of the method described by Waterman. [23]. The “unconstrained Waterman” approach is an unmodified version of that algorithm that that use the HMAP scoring function and gap penalty and generates alternate alignments by allowing the DP procedure to branch to an alternate path at any point in the DP matrix where doing so will lead to an alignment with a score within δ of optimal. However, in the constrained Waterman approach, branching to alternate paths is allowed only when moving between SSE and loop regions (see Figure S6 in Text S1 for more details). For both methods, it is impossible to know which value of δ will generate an ensemble of a desired size. To generate the alignments for comparison, we started with very small values for δ and increased it until the ensemble size exceeded 1000. We then sorted the alignments by their DP-based score and kept only the top 1000.
Models were built with the program Nest [29] for all S4 alignments, the optimal HMAP alignment and the correct/structure-based alignment. TM-score [31] was used to evaluate the accuracy of the model compared to the native query structure.
|
10.1371/journal.pgen.1002749 | Cohesin Proteins Promote Ribosomal RNA Production and Protein Translation in Yeast and Human Cells | Cohesin is a protein complex known for its essential role in chromosome segregation. However, cohesin and associated factors have additional functions in transcription, DNA damage repair, and chromosome condensation. The human cohesinopathy diseases are thought to stem not from defects in chromosome segregation but from gene expression. The role of cohesin in gene expression is not well understood. We used budding yeast strains bearing mutations analogous to the human cohesinopathy disease alleles under control of their native promoter to study gene expression. These mutations do not significantly affect chromosome segregation. Transcriptional profiling reveals that many targets of the transcriptional activator Gcn4 are induced in the eco1-W216G mutant background. The upregulation of Gcn4 was observed in many cohesin mutants, and this observation suggested protein translation was reduced. We demonstrate that the cohesinopathy mutations eco1-W216G and smc1-Q843Δ are associated with defects in ribosome biogenesis and a reduction in the actively translating fraction of ribosomes, eiF2α-phosphorylation, and 35S-methionine incorporation, all of which indicate a deficit in protein translation. Metabolic labeling shows that the eco1-W216G and smc1-Q843Δ mutants produce less ribosomal RNA, which is expected to constrain ribosome biogenesis. Further analysis shows that the production of rRNA from an individual repeat is reduced while copy number remains unchanged. Similar defects in rRNA production and protein translation are observed in a human Roberts syndrome cell line. In addition, cohesion is defective specifically at the rDNA locus in the eco1-W216G mutant, as has been previously reported for Roberts syndrome. Collectively, our data suggest that cohesin proteins normally facilitate production of ribosomal RNA and protein translation, and this is one way they can influence gene expression. Reduced translational capacity could contribute to the human cohesinopathies.
| Cohesin is a protein complex known for its essential role in chromosome segregation. However, cohesin and associated factors have additional functions in transcription, DNA damage repair, and chromosome condensation. Two human diseases, Cornelia de Lange syndrome and Roberts syndrome, are caused by mutations in cohesin. These “cohesinopathies” are thought to be caused by gene misregulation, although the role of cohesin in transcription has been enigmatic. Here we show that mutations in cohesin are associated with reduced production of the structural RNAs that are components of the ribosome in the budding yeast Saccharomyces cerevisiae. This causes defects in protein translation, which can explain a large fraction of the gene misregulation observed. We further show similar physiology in a human Roberts syndrome cell line. We postulate that reduced translational capacity contributes to the cohesinopathies.
| Cohesin is a protein complex that binds to chromosomes from the time of their replication until their segregation. Cohesin creates cohesion between two sister chromatids in order to ensure their correct segregation upon division at the metaphase to anaphase transition. In addition to its essential role in chromosome segregation, the cohesin complex and its accessory factors have been shown to play roles in chromosome condensation, DNA damage repair and gene regulation. The cohesin complex is composed of four subunits: Smc1, Smc3, Mcd1/Scc1/Rad21, and Scc3/Irr1. The complex is loaded onto chromosomes by the Scc2-Scc4 complex [1], [2], [3]. In order to establish cohesion between sisters, Eco1 acetylates the Smc3 subunit of the complex [4], [5], [6]. Pds5 is required for maintenance of cohesion in G2/M [7], [8]. Cohesion is dissolved at the metaphase to anaphase transition when sisters are separated to opposite poles for inclusion in new daughter cells.
Heterozygous mutations in Smc1, Smc3 and Scc2/Nipped-B/NIPBL have been associated with the human disease Cornelia de Lange syndrome (CdLS) [9], [10], [11], [12]. Homozygous mutation of ESCO2 (yeast ECO1) is associated with the human disease Roberts syndrome [13]. The human diseases, referred to as the cohesinopathies, are perplexing since the developmental defects suggest that the primary dysfunction is transcription, rather than chromosome segregation [14]. Metaphase chromosomes from Roberts syndrome patients show “heterochromatic repulsion,” which refers to regions of “puffing” at heterochromatic regions around the centromeres and nucleolar organizers (rDNA) [15].
In order to better understand the molecular underpinning of the human diseases, and to further explore the cohesin network, we constructed yeast strains bearing mutations analogous to those associated with human disease [16]. Our yeast strains are haploid, so they do not genocopy the disease state. However, characterization of the cellular defects associated with the mutations may still be informative. Previous characterization of these strains revealed very few defects in chromosome segregation or the location of cohesin binding, but interestingly, two mutants (eco1-W216G and scc2-D730V) had defects in nucleolar morphology, induction of the GAL2 gene, and chromosome condensation. Three mutant strains exhibited cohesion defects at 37°C (eco1-W216G, smc1-Q843Δ, and scc2-D730V). The eco1-W216G mutation disrupts the acetyltransferase activity of the protein toward Smc3 and is lethal at 37°C [17], [18]. Scc2 has recently been shown to participate not only in cohesin loading, but also in condensin loading [19]. Despite cohesion defects at 37°C, the growth of the scc2-D730V and smc1-Q843Δ mutants appears nearly normal.
To further characterize the mutants, we carried out gene expression profiling in rich medium and at various times following amino acid starvation. The gene expression pattern of the eco1-W216G mutant showed changes in over 1600 genes while the scc2-D730V mutant had essentially a wild-type gene expression profile. Under rich medium conditions, the gene expression profile of the eco1-W216G mutant suggested that protein translation was inhibited. By directly testing protein synthesis and ribosome biogenesis, we confirmed that translation was reduced. Strikingly, ribosomal RNA (rRNA) transcripts were significantly reduced in eco1-W216G and smc1-Q843Δ mutants. Since ribosome assembly is regulated at the level of rRNA [20], this could affect ribosome biogenesis. Cohesion was specifically reduced at the rDNA in the eco1-W216G mutant, reminiscent of the heterochromatic repulsion observed in Roberts syndrome. Importantly, protein synthesis and ribosomal RNA production were reduced in a human Roberts syndrome cell line, very similar to our yeast mutants. Taken together, our results suggest that cohesin proteins may normally promote production of ribosomal RNAs.
Given the hypothesis that mutations in cohesin can affect gene expression, we undertook gene expression profiling of three strains: 1) wild-type (WT), 2) scc2-D730V, and 3) eco1-W216G. We selected conditions under which we expected many transcriptional changes to maximize the likelihood of finding transcriptional differences in the mutants. Cultures growing in log phase in rich YPD medium (time 0) were transferred to medium lacking amino acids and samples were collected for analysis at 15, 35, and 55 minutes. Three independent cultures were analyzed for each strain background. mRNA was extracted, purified, labeled, and used for hybridization to Affymetrix microarrays (Yeast Genome 2.0) to examine gene expression.
To compare each mutant directly to WT, ratios were formed between each mutant and WT for each time point. Contrasts were created using limma to average replicates and determine p-values for each difference. After adjusting the p-values for multiple hypothesis testing, a set of genes was selected on the basis of adjusted p-values of less than 0.001 from any time point in either mutant/WT comparison. The result was that 1659 genes differed in expression, 1657 for eco1-W216G and 2 for scc2-D730V. Hierarchical clustering of the 1657 genes revealed the expression pattern in the eco1-W216G mutant was highly disrupted relative to the other two strains (Figure 1A). The number of genes up or down regulated in mutant/WT by at least 1.4 fold, with p-values of less than 0.05, for each timepoint is shown in Figure 1B. The lack of disruption in the scc2-D730V mutant background is notable since scc2-D730V and eco1-W216G mutant strains both have similar levels of chromosome decondensation and disrupted nucleolar morphology [16]. These results suggest that the scc2-D730V defects are not sufficient to cause major changes in gene expression.
We have previously reported that the clustering of tDNA adjacent to the nucleolus is disrupted in both the scc2-D730V and eco1-W216G mutant strains [16]. This clustering has been associated with the silencing of genes adjacent to tDNAs, a phenomenon referred to as tDNA gene mediated silencing [21]. We analyzed whether expression of the genes adjacent to tDNAs were misregulated in the mutants relative to WT. We found no evidence that genes adjacent to tDNAs were differentially regulated in the mutants (Figure S1), suggesting that control of gene expression via tDNA clustering is not a wide-spread phenomenon, although there still may be individual cases of gene regulation via this mechanism. Our results are consistent with previous findings showing that mutations in RNA pol III, which disrupt tDNA clustering, do not disrupt the expression of neighboring genes [22].
We performed a GO analysis on the genes differentially expressed (both up and down) with an adjusted p value less than 0.005 at time 0 (639 genes) and 15 minutes (627 genes) in the eco1-W216G mutant as compared to WT [23]. At time 0 we found that many of the differentially expressed genes are involved in glutamate metabolic processes, TCA cycle, cell wall organization, and acetyl-CoA metabolism (Table S1). Glutamate and glutamine are donors of amino groups for the biosynthesis of nucleotides, amino acids, and other nitrogen containing compounds. When the gene expression profile in rich medium for the eco1-W216G strain was compared to a variety of stress response profiles [24], it most closely matched nitrogen starvation. At the 15 minute timepoint, the enriched GO terms are almost all related to ribosome biogenesis, including biogenesis of ribosomal proteins and processing of RNAs needed for ribosome assembly (Table S1).
The gene expression data was further analyzed to determine whether the genes that were misregulated in the eco1-W216G mutant had any enrichment for particular transcription factor binding sites in their promoter regions. In the promoters of genes that were upregulated at the time 0 timepoint, we found a significant enrichment for Gcn4 and Tbp1/Spt15 binding sites (Figure 2A). Gcn4 is a transcriptional activator that activates the expression of many classes of genes, including stress and amino acid biosynthesis genes. Tbp1/Spt15, or TATA binding protein, is an evolutionarily conserved general transcription factor that interacts with other factors to form transcription preinitiation complexes at promoters. SNO1 and SNZ1 have been reported to be upregulated in pol III mutants [22] in a Gcn4-dependent manner [25]. These genes were found to be upregulated in the microarray data. The misregulation of SNO1 and SNZ1 was confirmed by RT-qPCR (SNO1, 3-fold, SNZ1, 9-fold, eco1-W216G at time 0, Figure S2). In the promoters of genes that were downregulated at time 0 there were fewer than average Gcn4 and Tbp1/Spt15 binding sites.
In the promoters of genes that were differentially expressed at the 15 minute timepoint, there was a significant enrichment for Rap1 binding sites. One group of these genes spans two clusters (Figure 1A, green bar); most of these genes are involved in ribosome biogenesis (adj p<0.001). Rap1 (Repressor Activator Protein) regulates the transcription of many ribosomal protein genes [26]. When cells are starved for amino acids, they normally repress genes involved in ribosome biogenesis [24]. While these genes were repressed once amino acid starvation was initiated in all three strain backgrounds, the genes were more weakly repressed in the eco1-W216G background. The reason for this is currently unclear, but may be related to the baseline ribosome defect in this strain (see below).
Gcn4 is a transcriptional activator that is normally translated only when cells encounter stress or nutritional starvation [27]. Surprisingly, the enrichment for Gcn4 binding sites in the promoters of genes upregulated in the eco1-W216G mutant at time 0 suggested that Gcn4 was activating the transcription of its normal target genes in the eco1-W216G mutant when cultures were growing in rich medium, prior to amino acid starvation. Although many Gcn4 target genes were induced in the eco1-W216G mutant background under rich growth conditions, the mRNA corresponding to Gcn4 was unchanged in the mutants relative to WT (see microarray data GEO GSE27235). Gcn4 contains an unusual leader sequence with four short ORFs (uORFs). One level at which Gcn4 is regulated is translation; translation of the Gcn4 mRNA occurs when ribosomes become processive due to limiting pools of GTP. For this reason, Gcn4 has been used extensively as a reporter for ribosome function [27], [28].
We used a Gcn4-lacZ reporter (p180) to determine whether β-galactosidase levels were elevated in the cohesinopathy mutants in the W303a strain background. We found a 4-fold elevation in β-galactosidase activity in the eco1-W216G strain as compared to a WT strain (Figure 2B). The cohesinopathy mutant smc1-Q843Δ also showed elevated β-galactosidase activity in this assay, while the scc2-D730V showed a very mild elevation. We also analyzed the β-galactosidase levels in two additional eco1 alleles. We previously reported that eco1-H53Y, eco1-W216G, and eco1-ack represent an allelic series (strongest to weakest) with respect to both cohesion as measured by a 1 spot-2 spot assay, and DNA damage sensitivity [17]. Stronger cohesion defects and DNA damage sensitivity were correlated with higher levels of β-galactosidase activity. Defects in cohesion have been previously noted at 37°C for the eco1-W216G, smc1-Q843Δ, and scc2-D730V strains [16] and the degree of defect correlates with the β-galactosidase activity observed.
We previously showed that deletion of RAD61/WPL1 rescued the growth of the eco1-W216G mutant at 37°C but did not rescue the X-ray sensitivity [17]. While β-galactosidase levels in the eco1-W216G rad61 double mutant were lower than the eco1-W216G single mutant, they remained higher than WT, suggesting that some defect persists. Deletion of RAD61 has been shown to partially rescue the cohesion defect of an eco1-1 mutant [4].
We further tested whether the increase in β-galactosidase activity was dependent on the presence of uORF4 in the Gcn4 promoter using an additional reporter construct. p226 has only uORF4. The deletion of the first 3 uORFs results in very minimal translational control [29]. The elevation in β-galactosidase activity remained with uORF4 for eco1-W216G (Figure 2C), but the level was reduced compared to the p180 reporter, as expected if translational control is contributing to the elevation.
We also analyzed the β-galactosidase levels in the cohesinopathy mutants in the BY4742/S288C strain background, as well as scc2-4 and pds5-2 mutants (Figure 2D). All mutants except scc2-D730V showed elevated levels of β-galactosidase compared to a WT control. We conclude that mutations in many different cohesin associated genes and in two different strain backgrounds can give rise to elevated levels of β-galactosidase activity expressed from the Gcn4 promoter.
We measured Gcn4 protein levels directly by Western blotting. The eco1-W216G mutant strain has a higher level of Gcn4 than a wild-type strain when grown in rich medium (Figure 2E), consistent with the results from the reporter assay and the gene expression data.
Given that high levels of Gcn4 can indicate a defect in protein translation, we tested whether protein translation was impaired in the cohesinopathy mutants. An evolutionarily conserved indicator of translational inhibition is the phosphorylation of elongation initiation factor 2α (eiF2α) [27], [30]. Phosphorylation of eiF2α inhibits the exchange of GDP for GTP in the ternary complex, blocking translation. We used Western blotting to measure the levels of total eif2α and the phosphorylated fraction. We found a 3-fold, 2.4-fold, and 1.9 fold increase in phosphorylated Eif2α in the eco1-W216G, smc1-Q843Δ, and scc2-D730V lysates, respectively (Figure 3A).
Since defects in translation could slow growth, we monitored growth in our cohesinopathy strains in rich medium (YPD+CSM) at 30°C. The eco1-W216G mutation confers a strong growth defect in the W303a background (p<0.0001). However, the growth of the scc2-D730V and smc1-Q843Δ mutant strains was not significantly different from WT (Figure 3B). Since mutations in cohesin or its regulators could cause chromosomal instability, we verified that our strains (Table S2) are not aneuploid (Figure S3).
Since growth can be affected by many different factors, we decided to analyze protein translation using more direct measures. To analyze ribosomes directly, we used sucrose gradients in combination with fractionation (Figure 3C). The ratio of polyribosomes to 80S indicates the active translating fraction. The 80S peak will consist of ribosomes without an associated mRNA or “vacant” ribosomes as well as some with an mRNA (monosomes). In theory, the polysome to 80S ratio becomes smaller with initiation defects, while it becomes larger with elongation defects [31]. The ratio of polysomes to 80S in WT, smc1-Q843Δ, and eco1-W216G, respectively, was 1.78, 1.23, and 0.89, consistent with a translation initiation defect in the mutants. The decrease in actively translating ribosomes could indicate a defect in protein synthesis.
In order to further measure protein translation, we used 35S-methionine incorporation to quantify protein synthesis. We found ∼50% reduction in incorporation in the eco1-W216G mutant and a ∼20% reduction in the smc1-Q843Δ mutant relative to the WT strain (Figure 3D). Collectively these results are consistent with the idea that the smc1-Q843Δ and eco1-W216G mutants support lower levels of protein synthesis.
The ribosome profiles suggested that initiation was limiting in the mutants. To test whether initiation was defective in the eco1-W216G strain, we transformed it with a plasmid carrying the ternary complex (eif2α, β, and γ, and tRNA-fMet) [32]. Overexpression of the ternary complex could reduce β-galactosidase levels expressed from the Gcn4 promoter if the high levels were due to poor translation initiation. We found that the plasmid reduced β-galactosidase levels in the eco1-W216G strain background (Figure 3E), consistent with a defect in the initiation of translation.
In order to further analyze the production of ribosomes in the scc2-D730V, smc1-Q843Δ and eco1-W216G mutants, we transformed them with plasmids that contain GFP reporters for the assembly of the 40S (Rps2-GFP) and 60S (Rpl25-GFP) components of the ribosome. In WT cells these proteins are mainly found evenly distributed in the cytoplasm. However, if there is an assembly and/or export defect, this is visualized as an accumulation of the GFP protein in the nucleus or nucleolus [33], [34]. We collected images of our mutants transformed with these reporters and we observed the accumulation of both reporter proteins in the smc1-Q843Δ and eco1-W216G mutants (Figure 4A and 4B). To further quantify this effect we developed a cytometric approach that allowed us to monitor at least 10,000 cells per sample. When the peak GFP fluorescence was measured, the smc1-Q843Δ and eco1-W216G mutants had higher mean fluorescence for both the 40S and 60S reporters, while the scc2-D730V mutant showed a mild phenotype for the 40S reporter but no increase in fluorescence for the 60S reporter (Figure 4C and 4D). To further analyze the data we generated the cumulative distribution function for each sample (not shown), and then we calculated the distance between biological replicates and between mutant and WT using a KS test (see Materials and methods). These distances are depicted as a box plot with an associated p value to indicate whether the distance from WT is statistically significant (Figure 4E and 4F). In summary, both the 40S and 60S subunits of the ribosome exhibit assembly/export defects in both the smc1-Q843Δ and eco1-W216G mutants, with a more severe defect observed in the eco1-W216G mutant.
We noticed from the microarray data that RNA polymerase I dependent ribosomal RNA (35S transcript) was downregulated approximately 4-fold in the eco1-W216G mutant (median p value 0.01, median adjusted p = 0.07) and 2-fold in the scc2-D730V mutant (median p value 0.12, median adjusted p = 0.40) in rich medium (Figure 5A). We note that transcripts corresponding to RNA polymerase I subunits appear to be unaffected in the transcription profile of the eco1-W216G mutant, suggesting downregulation of RNA Polymerase I is not causing the reduction in the 35S transcript. Notably, ribosomal RNA has been shown to be a limiting factor for ribosome assembly [20]. Since ribosomal protein genes showed no significant differences in transcription between the eco1-W216G mutant and WT in rich medium (Figure 1), we speculated that the ribosome defect was not due to a lack of proteins needed to make ribosomes, but possibly due to the low levels of rRNA.
Because rRNA constitutes ∼60% of the RNA being made by actively growing cells, 3H-uridine incorporation is commonly used to measure total rRNA synthesis. To further test the new production of rRNA, we pulsed with 3H-uridine and measured incorporation into RNA. In the eco1-W216G and smc1-Q843Δ mutants, there is less incorporation in 5 minutes in an equal number of cells (Figure 5B), indicating that these mutants produce less rRNA in this time frame. These experiments were carried out in the BY4742 background and the growth in SD-ura at 30°C was measured (Figure 5C). In log phase, which is when the labeling is performed, only the eco1-W216G mutant showed slower growth. We carried out a similar labeling experiment with the eco1-W216G mutant in the W303a background and obtained similar levels of incorporation (Figure S4A). In this background, growth is much more severely affected (Figure S4B). Thus, while the eco1-W216G mutation confers different growth defects in different strain backgrounds, the effect on total rRNA production appears to be similar, suggesting growth may not perfectly correlate with rRNA production.
RNA polymerase I produces the 35S transcript that is then processed into the 25S, 18S, and 5.8S transcripts and further modified by methylation and pseudouridylation. To measure the production of methylated rRNA, we used incorporation of 3H-methyl-methionine. Total RNA was isolated from equal numbers of cells following a 5 minute pulse labeling and a chase with cold methionine. Equal amounts of RNA were electrophoresed on a formaldehyde agarose gel and visualized with ethidium bromide (Figure 5D). Following exposure to film, the bands were excised and radioactivity was measured. We found that the eco1-W216G mutant produced 8–10% of WT levels of the methylated 25S and 18S transcripts and the smc1-Q843Δ mutant produced 18–28% of WT levels (Figure 5D). The growth curves for the mutants in SD-met at 30°C are shown (Figure 5E). Thus, while new production of total rRNA appears to be reduced approximately 2-fold in both mutants, the methylated form of the 25S and 18S transcripts is produced at a 10-fold lower level in the eco1-W216G mutant as compared to a 4-fold lower level in the smc1-Q843Δ mutant. The difference in production of total rRNA versus the processed and modified forms suggests that both initial production and subsequent processing are defective in the mutants, with a more severe defect in the eco1-W216G mutant. The fact that both 25S (60S rRNA component) and 18S (40S rRNA component) transcripts are affected in both mutants is consistent with the result that both 40S and 60S biogenesis are affected in both mutants.
A W303a strain bearing the eco1-W216G mutation does not grow at 33°C and cohesion is compromised at 37°C. Cohesion defects have been correlated with growth defects, and so it might be assumed that errors in chromosome segregation cause the lethality associated with mutations in cohesin. However, the scc2-D730V and smc1-Q843Δ strains have cohesion defects at 37°C, but can grow [16] (Figure S4C), suggesting precocious sister separation does not necessarily cause lethality. We tested whether transcription by RNA polymerase II of the 35S transcript from a galactose-inducible promoter would rescue growth of the eco1-W216G strain at 33°C. This plasmid allowed a partial rescue of the growth defect, suggesting some portion of the defect may be due to limiting levels of rRNA (Figure 5F). We further tested how much the rRNA levels increase in galactose medium and we found that the increase was a modest 40–50% (Figure 5G). However, this increase is similar in degree to the decrease in labeling observed with 3H-uridine, suggesting this increase should be sufficient to make up the difference. To explain the partial rescue we point out that 1) the morphology of the nucleolus is disrupted in the mutant, so even with more rRNA ribosome biogenesis may still be impaired, 2) the endogenous rDNA locus may still have defects associated with it, for instance, if there is difficulty with its replication, this defect will not be corrected by providing more rRNA and 3) at the elevated temperature there may be so little Eco1 function that other chromosomal processes such as chromosome segregation have become severely affected. A high copy plasmid with the 35S transcript produced from the normal promoter provides no rescue (data not shown). Overall our results suggest that some mutations in cohesin are associated with defects in 25S and 18S production.
Many different cohesin mutations confer elevation in β-galactosidase levels from the Gcn4 leader sequence, suggesting the elevation is related to defects in chromosome cohesion. However, mutations in the cohesin network have been shown to affect both chromosome condensation [35] and DNA damage repair [36]. Both the eco1-W216G and scc2-D730V mutations confer defects in chromosome condensation and nucleolar morphology, but importantly, the smc1-Q843Δ strain does not share these defects [16]. This suggests that aberrant chromosome condensation and nucleolar morphology are not the primary cause of the reduction in rDNA transcription.
However, since condensation can affect segregation of the rDNA we decided to further examine whether the cohesinopathy mutations disrupted rDNA segregation. At the metaphase to anaphase transition, chromosomes segregate, followed by segregation of the rDNA. The segregation of the rDNA is dependent on condensin and decatenation [37]. Since the rDNA is silenced during anaphase [38], a longer anaphase could potentially account for a reduction in transcription. To measure rDNA segregation, we used yeast strains tagged with Net1-GFP (rDNA marker) and Spc42-mCherry (spindle pole body marker). The duration of rDNA separation can be calculated by the timing of the start of spindle elongation (sudden increase in the distance between the two SPBs) to fully separated Net1-GFP. In wild-type cells, rDNA separation takes an average of 6.5 minutes. We found no significant difference in the kinetics of rDNA segregation in any of the mutants (Figure 6A). Thus, delayed rDNA segregation during anaphase cannot account for the slow growth or the transcriptional defects at the rDNA.
The number of rDNA repeats can expand and contract, controlled by recombination. We considered the possibility that contraction of the rDNA was limiting transcription. We monitored the copy number of the rDNA using qPCR. To demonstrate that our assay can detect differences in copy number, we used strains containing 20, 40, 80, and 110 copies of rDNA, as estimated by pulsed field gel electrophoresis [39]. We found that the number of rDNA repeats was not significantly different from WT in the scc2-D730V and eco1-W216G mutants in either the BY4742 or W303 backgrounds. Copy number was also examined in a smc1-Q843Δ strain and found to be normal (data not shown). This result suggests reduced copy number cannot account for reduced transcription (Figure 6B).
The rDNA is especially susceptible to genotoxic stress. It is estimated that the rDNA incurs several DSBs per S phase which result in an average of 3.6 Holliday junctions [40]. Cohesin is known to bind to the rDNA [41], [42] and the eco1-W216G mutation decreases cohesin binding at the rDNA as measured by ChIP approximately 2-fold [16]. Since cohesion is important for the resolution of DNA damage, we hypothesized that the decrease in transcription at the rDNA in some cohesin mutants might be related to an inability to efficiently resolve recombination intermediates due to defective damage induced cohesion. We used Southern blot analysis to measure whether DSBs accumulate at the rDNA. The level of DSBs in cohesin mutants and a WT strain was similar, indicating unresolved DSBs do not accumulate at the rDNA in the cohesin mutant strains (Figure 6C). Thus, failure to repair the locus cannot account for the transcriptional defect.
A normal yeast cell contains 100–150 copies of the 9.1 kb rDNA repeat, about half of which are actively transcribed and half are inactive. The cell can regulate the number of active repeats and the rate of transcription since in a 20 or 40 copy strain, all the repeats are active and the rRNA is present at normal levels [39]. rDNA repeats can be differentiated by their different chromatin structures and accessibility to cross-linking by psoralen followed by Southern blot [43]. Inactive or closed gene copies contain nucleosomes and are therefore less accessible to psoralen, and migrate faster on a gel following crosslinking whereas active or open gene copies are devoid of nucleosomes and are more accessible to psoralen, and migrate slower following crosslinking [43]. To verify the method, we used a strain with 40 copies and found few closed repeats, as previously reported (data not shown) [39]. We examined whether the mutations in cohesin were affecting the fraction of open repeats. We found no reproducible change in open repeats in the cohesin mutants relative to a WT control strain, at least in asynchronous culture (Figure 6D). Thus, a steady state increase in closed repeats does not appear to account for the decrease in transcription.
Given that the copy number and fraction of open rDNA repeats do not seem to be affected in the cohesin mutants, we sought to further understand the reduction in rRNA we observed by microarray and metabolic labeling. We used a FISH assay in which a unique sequence is inserted into the 5′ end of one 35S gene (Figure 6E) [44]. The transcription of this sequence can be monitored with a fluorescent probe in individual cells to indicate the dynamics of transcription in the population. We integrated three different cohesinopathy mutations into this strain and monitored transcription. We found that transcripts made from this single repeat were present at significantly lower levels in the smc1-Q843Δ and eco1-W216G strains, but not in the scc2-D730V strain. Thus, when a single repeat is monitored, less rRNA is made from this repeat.
Lower production of rRNAs could potentially be explained by 1) reduced copy number, 2) fewer transcriptionally active repeats, or 3) reduced RNA production from active repeats. Collectively our data suggests that mutations in ECO1 and SMC1 can be associated with production of fewer transcripts from the open fraction of rDNA repeats. Interestingly, the smc1-Q843Δ and eco1-W216G mutations were associated with a ∼2-fold reduction in rRNA using either the 3H-uridine labeling method to detect total rRNA or FISH to detect a single repeat. However, the eco1-W216G mutant showed a 10-fold reduction in the production of the methylated rRNA while the smc1-Q843Δ mutant showed a 4-fold reduction. This difference correlates well with the degree of defect in protein synthesis and ribosome biogenesis. We speculate that due to the disruption in nucleolar morphology in the eco1-W216G mutant [16] that processing and modification of the 35S transcript as well as ribosome assembly and export might be more severely affected than in the smc1-Q843Δ mutant, with the outcome that translation and growth are more affected.
We have previously measured cohesion at three loci in the eco1-W216G mutant. We observed a 15% reduction at an arm locus, a 9% reduction at a telomere locus, and an 8% reduction at a pericentric locus relative to a WT strain, and no defect in chromosome transmission [16], [17]. However, when we measured cohesion using strains with lacO repeats integrated adjacent the rDNA [39], cohesion is reduced ∼25% in the eco1-W216G background in the 50 copy strain (Figure 6F). Thus, Eco1 acetyltransferase activity is differentially required for genomic and ribosomal DNA cohesion. The mechanism for this is currently unclear and will require more investigation. However, we speculate that the decrease in cohesion at the rDNA is related to the reduced transcription at this locus.
Furthermore, the specific defect in cohesion at a heterochromatic region is reminiscent of the heterochromatic repulsion observed in cells from Roberts syndrome patients [45].
Given that the eco1-W216G mutation is associated with reduced protein translation and rRNA production in budding yeast, we decided to investigate whether a human cell line bearing the same mutation displays similar physiology. We used 35S-methionine labeling to measure protein synthesis in 1) a Roberts syndrome fibroblast cell line, 2) a version of the cell line that has been corrected with a wild-type copy of ESCO2 [45] and 3) a normal fibroblast line. We found that the Roberts syndrome cells incorporated methionine at about 50% the level as the corrected line or a normal fibroblast line (Figure 7A), very similar to the observations in yeast (Figure 3D). Furthermore, we measured the incorporation of 3H-uridine as an indicator of ribosomal RNA synthesis. We found that the rate of incorporation in the Roberts syndrome cells is about 50% the level as the corrected line or a normal fibroblast line (Figure 7B), very similar to the observation in yeast (Figure 5B). Finally, we examined the polysome profile in the Roberts cells. We find that the polysome to 80S ratio is lower in the Roberts cells relative to the corrected line (Figure 7C), similar to the observation in yeast (Figure 3C). Thus, it appears that protein synthesis and ribosomal RNA production are reduced in human Roberts syndrome fibroblasts, and suggests that the findings in yeast are relevant to the human disease.
Several groups working in fish, flies, mouse, and humans have shown that cohesin associated mutations or reductions in cohesin associated genes result in hundreds of small alterations in gene expression, and a few cases of big changes in gene expression [46], [47], [48], [49]. These changes in gene expression are thought to cause the human cohesinopathies. However, the mechanism by which mutations in cohesin-associated genes alter gene expression has been elusive. Cohesin together with CTCF [50] or mediator [51] may facilitate gene looping and communication between promoters and enhancers [52] to influence transcription by RNA polymerase II. Cohesin may also act directly at certain loci in an activating manner [47], [49] or a repressive manner [53] to regulate transcription by RNA polymerase II . In this report, we suggest a key locus at which cohesin proteins may influence transcription is the ribosomal DNA. Misregulation at this locus can affect the transcription of hundreds of genes as translation is affected.
The elevation in Gcn4 targets suggested that this transcriptional activator was induced in the eco1-W216G strain, and further suggested that translation would be repressed. Analysis using a Gcn4-lacZ transgene revealed that mutations in Pds5, Scc2, Eco1, and Smc1 all showed an increase in expression, consistent with the idea that cohesion defects correlate with reduced protein translation. Also consistent with our findings, inactivation of Mcd1/Rad21 in budding yeast in G1 was shown to affect the expression of 29 genes, including genes involved in rRNA maturation and ribosome biogenesis [54]. The differential effect of the eco1-W216G mutation on cohesion at the rDNA is notable since the rDNA in budding yeast has many properties of heterochromatin and lack of cohesion specifically in heterochromatic regions, including the rDNA, is a hallmark of Roberts syndrome. Thus, a cohesion deficit at the rDNA is common to both our yeast model and Roberts syndrome cells. The local cohesion defect at the rDNA in the eco1-W216G mutant is associated with the production of fewer 35S RNA products and reduced translation. A human Roberts fibroblast line displays similar physiology to our yeast mutant in that both protein synthesis and ribosomal RNA production are impaired, suggesting yeast may provide a good model for these particular defects.
We have characterized three different cohesinopathy mutants in yeast which have overlapping sets of defects. The mutation with the strongest phenotype is eco1-W216G, which confers defects in nucleolar morphology, DNA damage response, growth, condensation, gene expression, ribosome biogenesis and rRNA production (this work, [16], [17]). The smc1-Q843Δ mutant shares the defects in ribosome biogenesis and rRNA production, albeit less severe, and without much effect on growth. If one extrapolates to multicellular organisms, one can imagine that the developmental outcomes for the RBS allele could be more severe compared to the SMC1 CdLS allele. This proposal is consistent with observations made in zebrafish in which ESCO2 and RAD21 depletion were compared and ESCO2 depletion was uniquely associated with poor cell proliferation and cell death [55]. The scc2-D730V allele does not have the same effect on protein synthesis as the SMC1 and ECO1 mutations, instead exhibiting defects in nucleolar morphology and chromosome condensation. These defects could potentially be explained by the requirement for Scc2 for condensin loading [19]. The scc2-D730V mutation in the W303a background does show weak elevation in β-galactosidase activity from the Gcn4 leader sequence and eif2α-phosphorylation, and a weak 40S biogenesis defect. The scc2-D730V mutation may affect some aspect of chromosome biology that we do not currently understand or cannot be fully evaluated in budding yeast. We note that the scc2-4 mutation causes more severe defects in yeast (Figure 2 and data not shown). A future challenge will be to achieve a molecular understanding of how different mutations in different proteins can lead to similar disease outcomes.
In RBS both copies of ESCO2 have lost function, but CdLS is most often caused by a single mutant copy of SCC2/NIPBL. However, SMC1 is on the X chromosome in humans and the cases of CdLS associated with the smc1-Q843Δ allele have been in males with a sole mutant copy [10]. Thus, both our haploid yeast and human patients express only mutant copies of ECO1/ESCO2 or SMC1. In contrast, the evaluation of the scc2-D730V allele in haploid yeast does not genocopy the human disease since there would be an additional WT copy of SCC2/NIPBL present. It may be important to model haploinsufficiency to understand how the SCC2/NIPBL mutations cause disease.
Since defects in protein translation affect cell growth and division, protein translation can affect size. The reports of small size in a mouse model [48] and human CdLS patients [14] are consistent with our hypothesis that mutations in cohesin can generate a deficit in ribosome function. In a report on gene expression in Drosophila cells depleted for Nipped-B or Rad21 (CdLS model), nearly all ribosomal protein and aminoacyl-tRNA synthetase transcripts are reduced [46]. In addition, expression of Myc, p53 and Mdm2 are altered by depletion of Rad21 and Nipped-B in humans [47], mouse [48], flies and zebrafish [49]. These targets are known to be regulated by ribosome biogenesis [56]. When these data are taken in context of our current report, they collectively suggest that reduced translational capacity may contribute to the developmental defects associated with the cohesinopathies.
How do cohesin proteins facilitate rRNA production? Our data suggests that transcription from a given repeat is reduced in the eco1-W216G strain, rather than there being fewer open repeats or reduced copy number. One mechanism by which we can imagine cohesin contributing to transcription is through gene looping, which might facilitate reloading of RNA Polymerase I from the 3′ to 5′ end of the 35S transcript. Loops at the rDNA have been reported [57] and cohesin binds flanking each repeat in a pattern that would enable looping [41]. Other possibilities include cohesin promoting some other aspect of rDNA metabolism such as replication fork speed [58] or nucleolar organization that in turn facilitates the production of the rRNA transcripts. In any case, the lower levels of ribosomal RNA as measured in both yeast and human are associated with decreased protein synthesis. In future work it will be important to further examine the mechanism by which cohesin proteins promote production of rRNA. Coupling protein synthesis capacity to chromosome metabolism might provide the cell with a useful feedback loop for regulating proliferation.
Wild type and cohesin mutant strains transformed with the plasmids p180 (pGCN4 URA3 lacZ CEN) having all four μORFs or p226 (with only the fourth μORF) [29] were grown at 30°C to an A600 of ∼0.8 under repressive conditions (overnight growth in SD-ura then shifted to YPD+CSM till desired absorbance is reached). The cells were pelleted and protein extracts were made. β-galactosidase activity was measured following standardized protocols using ONPG (o-nitrophenyl-β-D-galactopyranoside) as the substrate. We note that the level of β-galactosidase activity is very sensitive to the growth protocol used [27].
Concentration and quality of RNA were determined by spectrophotometer and Agilent bioanalyzer analysis (Agilent Technologies, Inc., Palo Alto, CA). For array analysis, labeled mRNA was prepared from 300 ng of total RNA using the MessageAmp III RNA Amplification kit (Applied Biosystems/Ambion, Austin, TX) according to the manufacturer's specifications. Array analysis was performed using Affymetrix GeneChip Yeast Genome 2.0 Arrays processed with the GeneChip Fluidics Station 450 and scanned with a GeneChip Scanner 3000 7G using standard protocols. Resulting CEL files were analyzed using RMA [59] and limma [60] in the R statistical environment. Affymetrix GeneChip data are available at GEO under accession number GSE27235.
Motif identification and analysis for Gcn4, Tbp1, and Rap1 were based on presence or absence calls for each binding site within the region of the annotated gene start site and 400 bp upstream. Presence of the Gcn4 motif was determined by a match to the sequence TGA(C/G)TC(T/A). The Tbp1 and Rap1 matches were determined using the TRANSFAC [61] matrices F$TBP_Q6 and F$Rap1_C and the MATCH program [62]. The score cut-off profiles for Tbp1 and Rap1 were minFP and minFN, respectively, using TRANSFAC version 2009.3. Sequences, microarray probe mapping, and gene annotations were from Ensembl 56. P-values for the gene set indicated were determined using the hypergeometric test of all protein-coding genes.
100 ml of yeast culture was grown to an OD600 of 0.8 and treated with 100 µg/ml of cycloheximide for 10 mins on ice before centrifugation. After centrifugation the cell pellets were washed twice and resuspended in lysis buffer (10 mM Tris-HCl pH 7.5, 100 mM NaCl, 30 mM MgCl2, 100 ug/ml cycloheximide, 0.2 mg/ml heparin in DEPC). The cells were lysed in the cold by bead beating and the lysate (10 OD units) was loaded on top of an 11 ml 7–47% sucrose gradient in 15 mM Tris-Cl pH 7.4, 140 mM NH4Cl and 7.8 mM MgOAc-4H2O centrifuged at 36,000 rpm for 3 h. The gradients were fractionated and OD254 was monitored using an ISCO UV-6 monitor [63].
Polysome quantitation was done by both Image J and Mathematica, version 7.0. A common baseline was chosen and the area under the peaks was calculated with the Image J software. For Mathematica, TIFF images from the instrument were read using the Import function. The signal intensity was isolated from the image using the ImageSubtract function of the non-signal colors. The signal plot was scaled and shifted along the y-axis to position the baseline (x = 0) at the lowest signal level of the plot. Boundary regions were selected manually by zooming onto the image and recording the x-axis coordinate of extremal point. The total area between the selected boundaries and above the baseline was calculated using the Take and Total functions.
Growth curves were collected in triplicate for each genotype at time intervals of 15 min. We used a Tecan Infinite M200 Pro machine. Due to the non-linearity between optical density (OD) and cell number at higher cell densities, the measured Tecan ODs were converted to ‘real’ ODs using the calibration function ‘real OD’ = −1.0543+12.2716×measured OD [64]. The maximum slope was determined for each curve from 12 consecutive points and the statistical significance between slopes was calculated using a t test.
We used flow cytometry to quantify the peak GFP fluorescence in WT and mutant cells. By measuring the digitized pulse height from the B1 detector (525/50 emission), the maximum GFP intensity of each cell could be ascertained. For each sample approximately 10,000 cells were measured. WT and mutant strains that did not bear the GFP-plasmid showed no significant difference in their maximum fluorescence intensities, indicating they have similar levels of intrinsic fluorescence (autofluorescence); however, some mutant strains expressing a GFP tagged ribosomal subunit had on average a higher maximum GFP intensity than WT cells expressing the same fluorescent tag. Since the distribution of fluorescence intensity among GFP positive cells was non-Gaussian, we used the Kolmogorov-Smirnov (KS) statistic to characterize the distribution differences, which quantifies the distance between empirical cumulative distribution functions of two samples. Using this statistic, we can calculate distances between biological replicates (same genotype) and distances between samples with different genotypes. In this way we can determine whether the average KS-distance between the WT and mutant samples is significantly greater than between replicates (same genotype) using a t test.
Overnight cultures of yeast cells were diluted in YPD+CSM to an OD of 0.1 and grown to an OD of approximately 0.8. Cells were then pelleted by centrifugation and washed in PBS. Cells were lysed with glass beads in buffer containing 10 mM Tris, pH 7.4, 100 mM NaCl, 1 mM EDTA, 1 mM EGTA, 1 mM NaF, 20 mM Na4P2O7, 2 mM Na3VO4, 0.1% SDS, 0.5% sodium deoxycholate, 1% Triton-X 100, 10% glycerol, and 1 mM PMSF protease inhibitor cocktail (Sigma). Equal amounts of protein were loaded for each sample after quantification by Bradford assay. Protein samples were electrophoresed on a 12% SDS-polyacrylamide and transferred to nitrocellulose filters. The immunoblots were probed for phospho-specific eIF2α (Ser-51) (Cell Signalling #9721). Total eIF2α was measured with rabbit polyclonal antibody (a gift from T. Dever). eIF2α was visualized by HRP-conjugated anti-rabbit IgG.
Methods for RNA labeling were derived from a previous report [65]. For the experiment in Figure 5B, triplicate cultures of BY4742, eco1-W216G and smc1-Q843Δ carrying pRS316 were grown in SD-Ura medium to exponential phase (OD600∼0.3). 3H-uridine (5 µCi) was mixed with 500 µL of each culture and incubated at 30°C for 5 min with aeration. Then samples were treated with 2.5 mL of 10% trichloroacetic acid (TCA) with 2.5 mg/ml of uridine. After filtration through a 25 mm glass filter, each membrane was washed with 5% TCA, dried, and counted in a Beckman LS 6500 multipurpose scintillation counter. For labeling with 3H-methylmethionine (Figure 5D), we grew cells in SD-met medium and pulse-labeled for 5 min with 20 µCi/mL 3H-methylmethionine followed by a 5 min chase with cold methionine. RNA was prepared from 1×10∧7 cells. 8 µl of each sample was run on a 1.2% formaldehyde agarose gel, transferred to a Gene Screen membrane, and detected by autoradiography. Individual RNA species (25S and 18S) were excised from the blot (together with nearby regions of the blot for assessment of background) and quantified with a scintillation counter.
Yeast strains were grown to mid-log phase (OD600∼0.5) at 30°C in medium containing dextrose (YPD+CSM). Cells were harvested and washed in PBS and resuspended in a similar volume of prewarmed methionine-minus medium containing dextrose (SD-met). Aliquots were taken from this culture (0.75 ml) for the zero time point. The medium was supplemented with 27.5 µCi of 35S-methionine and unlabeled methionine 1 mg/ml. At 15–20 mins intervals (0.75 ml) samples were withdrawn from an actively growing culture. The amount of 35S-methionine incorporated into proteins was then measured by an adaptation of the method of Kang and Hershey [66]. The cells were lysed in 1.8 N NaOH containing 0.2 M β-mercaptoethanol. Proteins were precipitated by the addition of hot 10% trichloroacetic acid. After centrifugation, the precipitate was washed twice in acetone. The precipitate was dissolved in 100 µl of 1% sodium dodecyl sulfate and heated at 95°C for 10 min. An aliquot of the SDS extract was counted in Ecoscint for 35S radioactivity in a liquid scintillation spectrometer to determine the amount of 35S-methionine incorporated into proteins.
Psoralen cross-linking experiments were carried out as previously described [67] with the following modifications: 1.3% Tris-Taurine-EDTA (TTE) gels were run at 80 volts for 20 hours in 0.5× TTE, processed and transferred to Gene Screen membrane in 6× SSC. Hybridization with a 35S specific probe was carried out at 60°C and the membrane was exposed for 2 hours to a phosphorimager screen (GE/Amersham).
Genomic DNA was isolated from strains and used as a template for qPCR. For each chromosome arm, one locus, usually near the centromere, was monitored according to the method previously described [68].
Yeast cells were grown in CSM-URA at 30°C to an OD600 of 0.4. The cells were then fixed by adding formaldehyde to a final concentration of 4% (v/v) for 45 min at room temperature with shaking. After three washes with wash buffer (1.2 M sorbitol, 0.1 M potassium phosphate, pH 7.5), the cell wall was digested with 0.3 mg/ml zymolase in spheroplast buffer (1.2 M sorbitol, 0.1 M potassium phosphate, 10 mM vanadyl ribonucleoside complex, 0.06 mg/ml PMSF, 28 mM β-mercaptoethanol) at 37°C for 45 min. After digestion, the cells were washed three times with FISH wash buffer (30% formamide, 2×SCC). The cells were then hybridized in 30 µl hybridization solution containing 5 ng/µl DNA probe in 25% (v/v) formamide, 2×SCC, 1 mg/ml BSA (nuclease free), 10 mM vanadyl ribonucleoside complex, 0.5 mg/ml salmon sperm DNA and 0.1 g/ml dextran sulfate overnight at room temperature. Before imaging, cells were washed twice with FISH wash buffer for 30 min and then added to slides pre-coated with poly-L lysine.
The probe used for the FISH experiment is a synthesized DNA oligonucleotide modified from the previous publication [44]. The sequence of the oligonucleotide is 5′-CGGCRGGTAAGGGRTTCCATARAAACTCCTRAGGCCACGA-3′; the ‘R’s indicates an amino-dT replacing a regular dT where a fluorescein molecule was coupled. The probe was further purified by polyacrylamide gel purification to ensure that each amino-dT was coupled with a fluorescein molecule.
For counting RNA levels, it was first necessary to derive a calibration plot that relates intensity observed to RNA levels. This is required due to a wide range of RNA levels observed between different cells. In cells with more than ∼20 RNA, RNA spots overlap, making it impossible to distinguish individual RNAs. Extreme examples of this occur when cells have undergone recent ‘bursts’ in transcription (see Figure 6).
The general method previously developed was followed [44]. To generate a calibration plot, we acquired long-exposure z-stacks of RNA using the widefield module of a Zeiss-200 m that was also equipped with a Yokagawa CSU-10 spinning disc. For cells with few (generally less than 15) RNAs, it was possible to use the long-exposure images to count single RNAs. After counting RNAs in these cells, we switched to the confocal set-up and acquired a confocal z-stack as described below. This iteration allowed for the generation of a calibration plot that related overall intensity of the sum projection of the confocal z-stack to the number of counted RNA per cell. We obtained a linear plot, with an intercept at ∼0, demonstrating that RNA per cell is linearly proportional to total RNA, and thus total intensity per cell from the confocal data can be used to measure total RNA per cell even in cells where density is too high to distinguish single RNAs.
To acquire RNA per cell for groups of where the range is between 0 and ∼300, it was necessary to develop a system where it was possible to obtain fluorescence from cells with few (1 to 10) RNA, but yet not saturate the camera with cells possessing up to hundreds of RNA. We acquired 30 z-slices with spacing 0.3 microns. A background was subtracted for each slice, and then a sum projection was applied. Total intensity per cell was compared to the linear, extrapolated calibration plot to generate RNA per cell. A sum-projection of a non-background subtracted z-series was able to detect the location of cells where RNA levels were very low, eliminating the risk of missing low RNA-possessing cells with the background-subtracted analysis. We note that this method generated a distribution of RNA per cell that matched very closely the published result for the same strain [44].
Emission from the confocal z-slices was collected through a 500–550 nm bp filter onto a Hamamatsu C9100-13 EMCCD. A 488 nm laser line was used to excite the flourescein tagged FISH probe.
|
10.1371/journal.ppat.1006835 | Alphavirus-induced hyperactivation of PI3K/AKT directs pro-viral metabolic changes | Virus reprogramming of cellular metabolism is recognised as a critical determinant for viral growth. While most viruses appear to activate central energy metabolism, different viruses have been shown to rely on alternative mechanisms of metabolic activation. Whether related viruses exploit conserved mechanisms and induce similar metabolic changes is currently unclear. In this work we investigate how two alphaviruses, Semliki Forest virus and Ross River virus, reprogram host metabolism and define the molecular mechanisms responsible. We demonstrate that in both cases the presence of a YXXM motif in the viral protein nsP3 is necessary for binding to the PI3K regulatory subunit p85 and for activating AKT. This leads to an increase in glucose metabolism towards the synthesis of fatty acids, although additional mechanisms of metabolic activation appear to be involved in Ross River virus infection. Importantly, a Ross River virus mutant that fails to activate AKT has an attenuated phenotype in vivo, suggesting that viral activation of PI3K/AKT contributes to virulence and disease.
| In order to replicate to high titres, viruses need a sufficient supply of building blocks from the cell in the form of amino acids, nucleotides and lipids. To this end, viruses have evolved different mechanisms to reprogram and exploit host metabolism towards the synthesis of new biomass. Pharmacological inhibition of pathways involved in metabolic activation can represent an opportunity for therapeutic intervention. In this study, we explore how members of the alphavirus group, including Semliki Forest virus (SFV) and Ross River virus (RRV), modify host metabolism. We discover that both viruses activate cellular glycolysis by activating the PI3K/AKT pathway. This activation is mediated by a YXXM motif in the viral protein nsP3, which binds to PI3K and initiates AKT signalling. In SFV, mutations in this motif prevent AKT activation, reduce virus-induced metabolic activity, and lower viral replication. In the human pathogen RRV the same mutation causes less severe disease in vivo. We therefore conclude that AKT activation during infection with these viruses is an important determinant of pathogenicity.
| Many cellular functions have evolved in close interdependence with cell metabolism and nutrient availability. The immune system itself, central for host defence from pathogens, is highly dependent on the metabolic activity of the organism for critical processes such as cell proliferation and differentiation [1]. Importantly, many viral pathogens significantly alter host cell metabolism in order to fulfil their energy requirements, but their impact on metabolic balance has only recently begun to be investigated. The emerging picture suggests that viruses use a range of strategies to activate glycolysis [2–6] and modulate glutamine metabolism [7–9]. However it is less clear how viral proteins drive metabolic changes and whether these mechanisms are conserved across related viruses. Understanding the effect of virus replication on cellular metabolism is therefore needed not only to define virus requirements, but also to understand the impact of viral replication on the energy status of an infected host, and its role in pathogenesis.
Alphaviruses are enveloped, positive sense, single stranded RNA viruses, transmitted to humans and a variety of mammals by insects, often mosquitos. There is currently no available vaccine or treatment for most alphaviruses and this contributes to their rapid spread and to several recent epidemics [10]. Causing approximately 5000 cases each year, Ross River virus (RRV) is the most prominent human alphavirus in the South Pacific region, responsible for a debilitating musculoskeletal disease [11]. The link between infection and pathogenicity remains unclear. In contrast, Semliki Forest virus (SFV) generally does not cause disease in humans, but is neuropathogenic in mice and has often been used as a model system for the study of alphavirus biology [12]. In this study, we apply nuclear magnetic resonance spectroscopy (NMR) and gas chromatography-mass spectrometry (GC-MS) to analyse the metabolic alterations that accompany cellular infection with SFV and RRV. In SFV infected cells we measured sustained activation of glycolysis and of the pentose phosphate pathway (PPP) towards higher synthesis of new metabolites, and demonstrated that this is dependent on virus-induced activation of PI3K/AKT via a YXXM motif in the viral non-structural protein (nsP) 3. To test whether the presence of this motif is sufficient to predict changes in host metabolism during alphavirus infection, we examined the metabolic profile of cells infected with RRV, an alphavirus carrying the same motif in nsP3, whose metabolic profile has never been characterised. Although infection with RRV also triggers additional mechanisms of metabolic activation, disruption of the YXXM motif causes lower viraemia and reduced pathogenicity in vivo, thus demonstrating the importance of AKT activation in viral pathogenesis.
SFV has been used extensively as a model to study the cell biology of alphavirus infection. Although SFV infects a variety of cells in tissue culture, neurons are key targets in vivo [12]. Therefore we chose to study the metabolic alterations induced by SFV in the neuroblastoma cell line SH-SY5Y, a well-known model for neuronal function and differentiation [13], after 5 days differentiation with trans-retinoic acid (S1A Fig). In contrast to rapidly dividing cells, these post-mitotic neurons better reflect the less metabolically active environment that SFV would encounter in vivo, and therefore represent a good model to study how SFV infection might alter cell metabolism. Metabolic changes were profiled by 1H-NMR and GC-MS at 8 hours post infection (hpi). The complete list of assigned metabolites is shown in S1 Table (1H-NMR) and S2 Table (GC-MS). SFV infection doubled lactate concentration in the media (p = 1.1*10−6) and increased it by 43% in the cells (p = 3.97*10−4) (Fig 1A and 1B), while it decreased glucose concentration in the media (Fig 1A) by 10% (p = 1.93*10−3) and reduced choline, phosphocholine (PS), glycerophosphocholine (GPC), and adenosine monophosphate (AMP, an intermediate of nucleotide metabolism) levels in the cells (Fig 1B). It also markedly increased palmitic (C16:0) and stearic (C18:0) acids (Fig 1C). These data suggest increased glycolytic activity and fatty acid synthesis upon SFV infection. An increase by 40% in lactate accumulation (p = 3.98*10−4) was also observed in the media of infected primary rat cortical neurons (Fig 1D), validating the SH-SY5Y model. The kinetics of this metabolic reorganisation mirrored the kinetics of viral protein expression (S1B Fig), with a progressive accumulation of lactate with time and a simultaneous decrease of glucose in the media and glycerophosphocholine in the cells (S1C and S1D Fig).
To understand the consequences of this increased glycolytic activity, we followed the fate of uniformly labelled 13C glucose ([U-13C]glucose) 8 h after SFV infection. In infected cells we observed increased concentrations of [4-13C]glutamate (p = 0.02), [3-13C]aspartate (p = 0.03), and labelled palmitate (Fig 1E), suggesting an increased flux of glucose through the TCA cycle and a consequent increased export of citrate to the cytoplasm for de novo synthesis of fatty acids. No changes in the total level of succinate (S1E Fig) were found in control unlabelled samples, suggesting no accumulation of this intermediate during infection. Increased concentrations of labelled UMP (p = 0.04) and the constant levels of labelled AMP (Fig 1E), together with the marked decrease in the total concentration of this nucleotide in unlabelled samples (S1E Fig), suggest increased synthesis (and use) of nucleotides, supporting a key role for the PPP.
The glucose analogue 2-deoxyglucose (2DG), an inhibitor of glycolysis, has been shown to decrease SFV replication when added 16 h before infection [14]; however to confirm the importance of glycolysis and the PPP during the course of viral replication only, we treated cells with 2DG or the glucose-6-phosphate dehydrogenase inhibitor (dehydroepiandrosterone, DHEA, an inhibitor of the PPP) at the same time of infection with SFV. At 16 hpi, 2DG reduced production of new infectious virions by almost 2 logs, and DHEA by almost 1 log (Fig 1F). Profiling of SH-SY5Y cells treated for 16 hours with either inhibitor in the absence of infection showed the anticipated effects on glycolytic metabolites (S1G Fig). In both cases, AMP concentrations were significantly reduced (p = 0.001 for 2DG and for DHEA). In cells treated with 2DG (which replaces glucose in the first step of glycolysis), glucose was present at higher concentration in both media and cells, while a significant reduction in lactate production was observed. In cells treated with DHEA (which inhibits glucose entrance into the PPP), we observed an increased consumption of glucose in both media and cells and a simultaneous increase in lactate production, likely a compensatory effect triggered by the PPP inhibition. During infection, inhibition of the first step of glycolysis with 2DG, which is detrimental for both glycolysis and PPP, had a more dramatic effect on virus production (Fig 1F). Importantly, no significant toxicity was observed following treatment with either drug (S1F Fig), and no effect was observed on the early stages of viral replication (S1H Fig), indicating that blocking glycolysis or the PPP does not affect SFV infectivity.
The extensive and rapid increase in glycolysis upon SFV infection is reminiscent of the dramatic metabolic reprogramming typical of cancer cells [15], suggesting that SFV might activate a metabolic “master switch”, able to rapidly reprogram cellular metabolism. The PI3K/AKT signalling pathway has been shown to be activated upon SFV infection in a very strong and sustained manner, here referred to as “hyperactivation”. This PI3K/AKT hyperactivation overrides inhibition by growth factor depletion and requires the viral protein nsP3 [16,17]. However, no link to metabolism was made in these previous studies. PI3K/AKT hyperactivation in SH-SY5Y cells upon SFV infection was analysed by western blot at various times post infection. Mirroring the kinetics of viral replication and increased glycolysis, we observed phosphorylation of AKT from 5 hpi in the SH-SY5Y cells, with even higher levels at 8 hpi (Fig 2A). In agreement with a role for PI3K/AKT regulation of cell metabolism, we also observed phosphorylation of the downstream targets phosphofructokinase 2 (PFK2), the Rab GTPase-activating protein AS160, which increases trafficking of glucose transporters to the plasma membrane, and ATP citrate lyase (ACL), the enzyme responsible for cytosolic acetyl-coA synthesis from citrate. Phosphorylation of AKT was also observed in primary rat cortical neurons (S2A Fig and S2B Fig). Consistent with the activation of AKT, a kinase that modulates glycolysis primarily by phosphorylating key glycolytic enzymes, we did not observe any increase in the mRNA levels of glycolytic genes (S2C Fig).
The PI3K inhibitor Wortmannin almost entirely abrogated infection-mediated phosphorylation of AKT and of downstream targets without substantially changing the expression of SFV E1-E2 proteins (Fig 2A), indicating that reduced AKT activation is not due to reduced infection. No inhibition was detected upon Wortmannin treatment at 8 hpi (S2E Fig), suggesting that viral entry, early RNA replication and protein synthesis are not affected. Conversely, Wortmannin decreased the release of progeny virus by ~60% compared with the untreated control (Fig 2B), implying that activation of AKT signalling is more important for late stages of viral replication. Indeed, adding Wortmannin either at the same time or two hours after virus infection resulted in similar levels of viral decrease (S2F Fig). An even more striking decrease was observed in primary neurons (Fig 2C). Similar levels of inhibition were measured using a different PI3K inhibitor, LY294002 (S2G Fig). Cell viability was not markedly compromised by 16 h treatment with either drug (S2D Fig).
We next tested whether Wortmannin affected SFV-induced glycolysis. In mock-infected samples, Wortmannin caused a marginal reduction of cellular glycolytic activity and did not affect myristic, palmitic or stearic acid content. In SFV infected cells, Wortmannin induced a significant decrease in lactate levels in both media (by 40%, p = 0.027) and cells (by 73%, p = 0.003) (Fig 2E and 2F, respectively), a decreased use of glucose (by 13%, p = 0.013) (Fig 2E), and reduced fatty acids levels (Fig 2G). The dramatic decrease of glycolysis in infected cells treated with Wortmannin suggests that PI3K/AKT is indeed responsible for activating this pathway upon SFV infection, and for the increase in glycolytic products and fatty acid synthesis. Production of new infectious virions, inhibited by Wortmannin, could be rescued by 25% by providing ribose, a nucleotide precursor generated primarily through the PPP branch of glycolysis, and by 35% by palmitate, a precursor for the more complex lipids generated from glycolysis (Fig 2D). This suggests that, although other factors are also likely to restrict SFV replication upon Wortmannin treatment, activation of glycolysis is important for the generation of extra metabolic building blocks, needed for maximal production of new virus.
We next tried to understand how SFV nsP3 activates PI3K by searching for the presence of eukaryotic linear interaction motifs in this protein. Using the database ELM (http://elm.eu.org/), we identified the well-characterised PI3K activation motif YXXM in the amino acid stretch Y369-E370-P371-M372, situated in the C-terminal hypervariable domain (S3A Fig). To test whether this motif is relevant for PI3K/AKT activation, we generated viral mutants where Y369 was replaced with an alanine (SFV-YA) or with a phenylalanine (SFV-YF). In BHK cells, both viruses, although replicating in a similar manner as indicated by nsP3 levels, failed to activate AKT to the same extent as the wild type virus, confirming the importance of this motif in PI3K/AKT activation (Fig 3A). Consistent with low PI3K/AKT activation, the viral replication complexes (RC) of both mutant viruses were mainly localised at the cell periphery, while wt SFV showed efficient RC internalisation (S3B Fig), as previously reported [16]. Disruption of the YEPM sequence by leaving the tyrosine unchanged and mutating the three downstream residues to alanines (SFV-YAAA) also reduced AKT phosphorylation (S3C Fig), indicating that a complete YXXM motif is required for AKT activation. To minimise potential unrelated structural changes, SFV-YF was selected for further studies. Anchorage of nsP3 to the plasma membrane is required for PI3K/AKT activation, however membrane anchored Myr-Pal-tagged nsP3-YF failed to activate AKT, further confirming the relevance of the YXXM motif for activation of the pathway (Fig 3B).
Activation of class IA PI3Ks is typically mediated through the interaction between the phosphorylated tyrosine in a YXXM motif and the arginines in the FLVRD/E motifs of the SH2 domains of the p85 subunit of PI3K [18]. We therefore hypothesized that the SFV-nsP3 YXXM motif binds the SH2 domain of p85, and that either mutation at Y369 in nsP3 or mutations in the SH2 domains of p85 would abolish the interaction. Consistent with this scenario, Myr-Pal-nsP3-wt, but not Myr-Pal-nsP3-YF or biotin acceptor peptide (BAP)-tagged nsP3-wt without a membrane anchor, was pulled-down by EYFP tagged p85α (EYFP-p85-α) (Fig 3C) and by EYFP tagged p85β (EYFP-p85-β) (Fig 3D). Equally, nsP3 co-immunoprecipitated with wild type EYFP tagged p85 (EYFP-p85-α-wt), but not with a mutated form of p85 where the arginines in the FLVRD/E motifs in both SH2 domains were mutated to alanines (Fig 3E) (“RARA mutant”) [19,20]. Importantly, we also show by co-immunoprecipitation that nsP3-wt but not nsP3-YF interacts with endogenous p85 in the context of SFV infection (Fig 3F).
The effect of the Y369F mutation in nsP3 on viral replication was tested by measuring viral release over time after infection of BHK cells at low (0.1) and high (10) MOI, compared to wt SFV and SFV-Δ50 carrying a larger deletion in nsP3, also affecting PI3K activation (S3C Fig). Consistent with a role for PI3K/AKT activation on late stages of the virus life cycle, viral titres were similar at early times post-infection but differed at later time points, with wt SFV titres exceeding SFV-YF by approximately one (MOI 0.1; Fig 3G) or more (MOI 10; Fig 3H) orders of magnitude by 12 hpi.
Taken together, these data show that during infection, the YXXM motif of SFV nsP3 binds the SH2 domains of p85 at the plasma membrane and activates the PI3K/AKT pathway leading to internalisation of replication complexes and efficient virus replication (S3D Fig).
To establish the role of nsP3 Y369 in reprogramming cell metabolism, we compared the metabolic profiles of SH-SY5Y upon mock infection or infection with either wt SFV or SFV-YF. The inability of SFV-YF to activate AKT and its downstream targets over the course of infection was confirmed in SH-SY5Y (S4A Fig). In addition, the production of new viral progeny was lower after infection with SFV-YF compared to wt virus (S4B Fig). After 8 hours of infection at MOI 1, lactate levels in the media were higher for wt SFV compared to both mock controls and SFV-YF (Fig 4A), suggesting activation of the glycolytic pathway in cells infected with the wt virus, but not the mutant. As observed previously, AMP was decreased in the wt SFV infected samples compared to mock controls, but a larger decrease was seen in the SFV-YF infected cells. UMP did not differ between samples infected with wt SFV and mock controls, but was significantly decreased in SFV-YF, suggesting the possibility that in samples infected with SFV-YF nucleotide usage exceeded new synthesis (Fig 4B). Finally, a similar decrease in glycerophosphocholine levels between wt SFV and SFV-YF samples (Fig 4C) showed that the mutation only affected the activation of the glycolytic pathway.
With higher viral load (MOI 5), increased levels of lactate were detected at 8 hpi both in the media (Fig 4D) and in cells (Fig 4E) infected with wt SFV compared to mock and SFV-YF (by 27% in the media, p = 0.0046, and by 29% in the cells, p = 0.00012), with lower concentrations of glucose in the media compared to the mutant (by 2%, p = 0.017). These results also suggest an important contribution of the TCA cycle: SFV-YF-infected cells contained significantly lower concentrations of glutamate (a source of 2-oxoglutarate) and succinate (S4C Fig and S4D Fig), suggesting depletion of these TCA intermediates, presumably as citrate is consumed to feed fatty acid synthesis, as suggested by the labelling data (Fig 1E). Overall, these data mirror the results obtained following treatment with Wortmannin, and emphasise the crucial role of the YXXM motif in SFV nsP3 in the hyperactivation of AKT and in host cell metabolic reprogramming.
SFV-induced PI3K/AKT hyperactivation likely affects other downstream features in addition to cell metabolism. We assessed one such feature, the mammalian target of rapamycin (mTOR), involved in translational control, using phosphorylation of rpS6 as a readout. Indeed, mTOR was activated in SFV-wt infected cells even under nutrient and growth factor depletion, which was not the case in SFV-YF or mock infected cells (S4F Fig). However, despite sustained activation of mTOR, treatment of cells with rapamycin to specifically inhibit the SFV-induced mTOR hyperactivation had no major effect on wt SFV growth kinetics (S4G Fig). We hence conclude that the activation of mTOR seen during WT SFV infection as a downstream effect of infection-induced PI3K/AKT hyperactivation is not critical for virus growth in vitro.
AKT inhibitors have been studied as anticancer treatments for nearly two decades, resulting in the development of a number of compounds currently undergoing clinical trials [21]. MK-2206 is a diphenylquinoxaline analogue that inhibits AKT by locking the kinase in a closed conformation and it has been tested alone or in combination against several malignancies [22,23]. Treatment of differentiated SH-SY5Y with MK-2206 caused a reduced release of infectious virions even more striking than Wortmannin (~84% vs. ~63%, respectively) (Fig 4F). As for Wortmannin and LY294002, we observed an even more dramatic inhibition in rat primary neurons (~2 logs, Fig 4G). Cell viability following MK-2206 treatment is shown in S4H Fig.
Interestingly, Sindbis virus (SINV), an alphavirus lacking the YXXM motif in nsP3, failed to induce sustained AKT activation (Fig 5A) and only caused a ~30% increase in glycolytic activity (measured as lactate over glucose ratio (Fig 5B)), compared to the ~100% increase observed for SFV. Importantly, while we measured a decrease in nucleotide concentrations (Fig 5C), no decrease in choline and phosphocholine levels was observed upon SINV infection (Fig 5D) and no changes were measured in the concentrations of myristic, palmitic or stearic acids in infected cells (Fig 5E). Thus SINV, which does not rely on AKT hyperactivation during replication, may have evolved different mechanisms of metabolic activation and lipid use/synthesis. Inhibition of glycolysis (with 2DG) or of the PPP (with DHEA) caused similar degrees of inhibition of almost 1 log, suggesting that SINV relies on these two metabolic pathways (Fig 5F).
As indicated above, the sequence YXXM is not present in the nsP3 of all alphaviruses. Systematic sequence analysis however reveals the presence of YXXM in nsP3 of the human alphavirus pathogen Ross River virus (RRV), located in a similar position as in SFV nsP3, despite overall low conservation of the nsP3 hypervariable domain between the two viruses (S5A Fig).
As with SFV, wild type RRV (RRV-wt) induced hyperactivation of AKT both in BHK cells (S5B Fig) and in differentiated C2C12 (Fig 6A), a murine cell line commonly used as a model of muscle myotubes, a RRV target in vivo. RRV in which the relevant tyrosine (Y356) was mutated to phenylalanine (RRV-YF) induced lower levels of AKT activation (Fig 6A and S5B Fig) and prevented internalisation of RC (S5C Fig). However, when we measured release of RRV-wt and -YF in BHK cells (S5D Fig, left and middle panels), no significant differences were observed between the two viruses, neither in a multi-step (MOI 0.1) nor a single-step growth curve (MOI 5). Similar results were obtained in C2C12 cells (MOI 5, S5D Fig, right panel). To understand whether this lack of difference could be explained by a metabolic phenotype somewhat different from the one observed for SFV, we examined the metabolic profile of differentiated C2C12 infected with RRV-wt or RRV-YF. Metabolic analysis of RRV-wt infected cells showed activation of glycolysis with an 18% increase in lactate in the media (p = 0.0008), and higher levels of alanine in media and cells (Fig 6B and 6C, respectively). Interestingly, no significant decrease in glycolysis was measured in RRV-YF infected cells compared to RRV-wt, as we also found a 20% increase in lactate levels in the media (p = 0.00001, compared to mock) and higher concentrations of alanine in cells infected with RRV-YF (Fig 6B and 6C). However, significantly higher levels of fatty acids were measured upon infection with RRV-wt than with RRV-YF (Fig 6D). The concomitant differences in glutamine and glutamate levels in cells infected with either virus (Fig 6C), which were not observed upon infection with SFV (S4C Fig and S4E Fig), suggest the possibility that RRV may rely on an additional, AKT-independent mechanism of metabolic activation that stimulates both glucose and glutamine metabolism. Despite the differences between SFV and RRV, activation of AKT by either virus was linked to increased fatty acid synthesis.
We next tested whether hyperactivation of the AKT pathway by RRV-wt but not RRV-YF resulted in differences in pathogenesis in vivo in a murine model of infection. C57BL/6 mice were subcutaneously infected with 104 pfu of RRV-wt or RRV-YF and monitored daily for disease signs. Limb weakness and loss of gripping ability, which in RRV-wt infected mice reached a disease score of 5.5 on day 10, were significantly milder in mice infected with RRV-YF, with disease scores not exceeding 4 (Fig 7A). Also, the average weight gain of RRV-wt infected mice was considerably lower than mice infected with RRV-YF (Fig 7B), while histological analysis of quadriceps samples at day 10 pi showed much more prominent lesions in RRV-wt than RRV-YF infected mice (Fig 7C and S6A Fig). In addition, viraemia for RRV-YF was slightly reduced compared to RRV-wt, reaching significance at 2 days pi, which corresponded to the viraemia peak (Fig 7D). Equally, although not significant, a trend suggesting lower viral titres at day 1 pi in the spleen, quadriceps and ankle, and at 6 and 10 days pi in quadriceps and ankle was observed (Fig 7E). Taken together, these results indicate that a mutation in the YXXM motif that reduces AKT hyperactivation results in attenuated infection compared to RRV-wt, and that AKT activation contributes to pathogenesis.
The importance of cellular energy metabolism in determining the outcome of a viral infection is increasingly recognised. However, the impact of a sustained metabolic activation on viral pathogenesis remains unclear. Primarily using SFV as a model alphavirus, we show changes in central metabolism that accompany viral infection, reveal the molecular mechanisms responsible for these changes, and, for the first time, explore their impact on viral pathogenesis in vivo using the relevant human pathogen RRV.
Using a combination of 1H- and 13C-NMR spectroscopy and GC-MS, we show that SFV infection increases and redirects glucose metabolism into macromolecular synthesis, by activating both glycolysis and the PPP. This metabolic switch leads to higher synthesis of nucleotides and fatty acids, critical building blocks for the formation of new SFV virions. Indeed, pharmacological inhibition of either of these pathways during virus replication results in reduced viral titres. As previously shown for cancer cells, and more recently for a number of viruses [24], increased glycolysis maximises the processing of glucose into macromolecules, and this provides a higher proliferative advantage than generation of ATP alone. Our work further emphasises the role of the PPP, which not only generates nucleotides for viral nucleic acid synthesis, but also contributes to lipid synthesis by providing NADPH. Interestingly, the kinetics of the metabolic shift observed in SFV infected cells mirrors the kinetics of viral replication, suggesting the importance of continued synthesis of metabolic intermediates during the production of new virions.
Mechanistically, we discovered that the metabolic changes induced by SFV infection are triggered through hyperactivation of the PI3K/AKT pathway. Moreover, we showed that a YXXM motif in SFV nsP3 is responsible for this activation through interaction with p85, the regulatory subunit of PI3K. Mutations in this motif in SFV nsP3 abolish hyperactivation of the pathway and lead to a metabolic profile consistent with increased consumption of metabolic intermediates, in the absence of a major increase in new metabolic synthesis. Moreover, fewer infectious particles are released upon infection with the SFV-YF mutant, suggesting that activation of glucose metabolism is important for maximal viral replication. In this study, AKT activation was seen in a variety of mammalian cells, including human SH-SY5Y, murine C2C12 and MEFs (upon transfection with Myr-Pal-nsP3), BHKs, and rat primary neurons, suggesting that this mechanism is not cell-specific and is conserved in different mammalian species. Whether the same is true in arthropod vectors and whether it represents an advantage for viral replication in the different species involved in virus transmission remains an interesting question. SINV, a related alphavirus that lacks the YXXM motif, has been shown to replicate largely independently of the PI3K/AKT/mTOR pathway in human cells [25], but to activate the pathway to some extent in arthropod cells [26]. We show that infection of differentiated SH-SY5Y cells with SINV does not lead to detectable activation of AKT, increases cellular glycolysis to a much smaller extent than SFV and does not alter fatty acids levels. This suggests that these two viruses have evolved distinct mechanisms for regulating lipid synthesis in mammalian cells, but are both dependent on glycolysis and the PPP.
Sequence analysis of nsP3 of all alphaviruses shows that YXXM motifs are present in nsP3 of SFV, RRV and also Getah virus and Sagiyama virus, which are equine pathogens in Asia [27], as well as Middelburg virus, which mainly infects sheep, goats and horses in Africa [28]. We found the YXXM motif to be highly conserved in nearly all the sequenced clinical isolates of the human pathogen RRV, despite its localisation within the hypervariable region of nsP3, in which sequence features are poorly conserved. Consistent with a critical role for AKT activation, mice infected with RRV-YF showed lower viraemia at the infection peak (2 days pi) and milder disease outcome compared with RRV-wt. Interestingly, failure to activate AKT in vivo attenuates viral replication even in the presence of an additional regulatory mechanism that appears to activate both glucose and glutamine metabolism. It is likely that while a compensatory mechanism is sufficient to maintain equal levels of viral replication in vitro, AKT activation is required for maximal production of new viral progeny in vivo, likely through increased fatty acids synthesis. It is possible that the trend of lower levels of viral replication is also responsible for a less sustained and aggressive immune activation, as indicated by lower levels of IFN-β in the lymph nodes and spleen (S6B Fig) and, as a consequence, milder disease progression after viral clearance (day 6 pi).
Intriguingly, while not conserved in all alphaviruses, activation of AKT via a YXXM motif has been reported for viruses in different families, including influenza virus [29] and herpes simplex virus [30], suggesting that different viral pathogens have evolved to use the same cellular mechanism of PI3K activation. While a number of viruses have been shown to activate PI3K, this is the first study that demonstrates a link between virus-induced activation of PI3K, cellular energy metabolism, viral replication, and in vivo pathogenesis, and explores the conservation of this mechanism of metabolic activation across different members of the alphavirus family.
As an association between high blood virus titres and the severity of arboviral infections has been demonstrated [31], inhibition of virus-induced pathways of metabolic regulation may be a viable antiviral strategy. If rapidly replicating viruses are hyper-dependent on generating new biomass, inhibition of the pathways responsible for metabolic activity might suffice to reduce viral titres to non-pathogenic levels without compromising the basal metabolic activity of the host. This scenario is particularly attractive because, by targeting common pathways of metabolic activation, a single drug or cocktail of drugs might be used to target a wide range of viral pathogens. Additionally, this study also reveals that even closely related viruses may develop different or redundant mechanisms of metabolic activation. Whether these differences arise from or are responsible for differences in virus replication or tropism, or what their impact is on the development of broad-spectrum antivirals remain compelling questions calling for a systematic study of the metabolic changes induced by different viruses and of the signalling pathways involved.
All animal experiments were approved by the Animal Ethics Committee of Griffith University (Gly/01/14/AEC). All procedures conformed to the National Health and Medical Research Council of Australia.
SH-SY5Y cells (ATCC CRL-2266) were grown in 45% F-12 media and 45% DMEM with 10% (v/v) foetal calf serum (FCS, PAA), 1X non-essential amino acids, 1X sodium pyruvate, 20 mM HEPES, 1 mM L-glutamine, and 1X penicillin/streptomycin (p/s) (all from Gibco). For differentiation, 10 μM retinoic acid (Calbiochem) was added to the culture medium every 2–3 days for 6 days. C2C12 cells (ATCC CRL-1772) were grown in 10% DMEM GlutaMAX with 10% (v/v) FCS. For differentiation cells were plated on collagen coated dishes and cultured with DMEM GlutaMAX supplemented with 10% Horse Serum (Sigma) for 5 days. Immortalized mouse embryonic fibroblasts (MEF, kind gift from Nancy Kedersha, Harvard Medical School) and HEK293 cells (ATCC) were grown in DMEM, 10% (v/v) FBS, 2 mM L-glutamine, and 1X p/s. BHK-21 [C-13] (ATCC CCL-10) were grown in GMEM supplemented with 10% (v/v) FCS, 10% (v/v) tryptose phosphate broth (TPB, Gibco), and 1X p/s. Rat cortical neurons were isolated from P0 pups. Cortices were dissected in HBSS (Gibco) and trypsinized for 30 min at 37°C. Cells were collected by centrifugation, resuspended in Neurobasal media (ThermoFisher Scientific) supplemented with B27 (Gibco), triturated, and plated on glass coverslips coated with 1mg/ml poly-L-lysine (Sigma Aldrich). Neurons were incubated at 37°C with 5% CO2 for 3 days. For nutrient and growth factor depletion (starvation), cells were treated with Earle’s balanced salt solution (EBSS; Sigma). as previously described [32].
SFV and SFV-GFP (gift of Dr Giuseppe Balistreri, University of Helsinki) were expanded and titrated on BHK-21 cells. The stock used for the metabolomic experiments was concentrated and purified by centrifugation through a sucrose cushion. The SFV-Δ50 mutant was described previously [33]. SFV-wt, SFV-YA, and SFV-YF were rescued by transfection of BHK-21 cells with the infectious plasmid pCMV-SFV4 [34] and harvesting the cell culture supernatant upon appearance of cytopathic effect as described [16]. The mutations described in the text were introduced using Gibson assembly technology (NEBuilder HiFi DNA Assembly Master Mix, New England BioLabs, according to the manufacturer’s instructions) and verified by sequencing. SINV (kind gift of Penny Powell, University of East Anglia) was expanded and titrated in BHK-21 cells. Wild type virulent RRV (strain T48, here referred to as RRV-wt) was rescued from pRR64 [35] as described [36]. To generate RRV-YF, the codon for tyrosine 356 of RRV nsP3 was mutated to phenylalanine in pRR64 as above.
Construction of Myr-Pal-nsP3-wt-FLAG was described previously [16]. To generate the Myr-Pal-nsP3-YF-FLAG, the nsP3-wt sequence was replaced with that of nsP3-YF, amplified from pCMV-SFV4-YF by PCR. Plasmid pEBB/PP-nsP3-wt, encoding SFV nsP3 with a biotin acceptor peptide [37] was used as a control without Myr-Pal signal. For p85, the following expression plasmids were employed: pEYFP-p85-α and the mutant pEYFP-p85-α-RARA [19] (kind gift of Ji Luo [National Cancer Institute, USA]); EYFP indicates the enhanced yellow-fluorescent protein); pEYFP-C1-p85-β (Addgene plasmid # 1408) and empty vector pEGFP-C1 (Clontech, EGFP indicating the enhanced green-fluorescent protein). All plasmids were verified by sequencing (Eurofins). Cells were transiently transfected using Lipofectamine 2000 (Life Technologies) according to the manufacturer’s instructions.
Metabolomic analysis was performed as previously described [8,38,39] and as more extensively outlined in the Supplementary information.
Differentiated SH-SY5Y cells were infected with SFV at MOI 3 in complete media with the indicated concentrations of inhibitors. Sixteen hours later, media was harvested and 500 μl of 10-fold serial dilutions of media were used to infect monolayers of BHK-21 cells for 2 h at 37°C. Infectious media were then replaced by a semisolid carboxymethylcellulose overlay (Rectapur, low viscosity, VWR; 1.5% in MEM). 48 h later the overlay was removed, cells washed in PBS, and plaques revealed by crystal violet staining. 2DG, DHEA, and Wortmannin were purchased from Sigma Aldrich, LY294002 from Cell Signaling Technology, and MK-2206 from Selleckchem.
SH-SY5Y cells were seeded in 96 well format, differentiated and infected with SFV-GFP in culture media containing the indicated dilutions of inhibitors (MOI 3). 8 h later cells were washed and fixed in 4% formaldehyde, and the nuclei stained with Hoechst (5 μg/ml ThermoFisher Scientific). Images were acquired using an Opera high-content spinning-disk confocal microscope (PerkinElmer) and the percentage of infected cells quantified using the Columbus Image Data Management and Analysis Software (PerkinElmer).
Cells were washed with PBS prior to lysis in lysis buffer (20 mM 4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid [HEPES], pH 7.4, 110 mM potassium acetate, 2 mM magnesium chloride, 0.1% [vol/vol] Tween 20, 1% [vol/vol] Triton X-100, 0.5% [wt/vol] sodium deoxycholate, and 500 mM sodium chloride), supplemented with Complete protease inhibitor and PhosSTOP phosphatase inhibitor cocktails (Roche) on ice. After centrifugation (20,000×g for 10 min, 4°C), cell lysates were incubated with mouse-anti-GFP (Abcam, ab1218) or rabbit-anti-p85 (Cell Signaling Technology, catalog numbers 4257 and 4292, used at 1:1 ratio) antibodies for 20 min at room temperature by using a rotator, followed by incubation with protein G magnetic beads (GE Healthcare) at 4°C overnight. Immunoprecipitation samples were washed with lysis buffer, eluted with 4× reducing NuPAGE LDS sample buffer (Life Technologies), heated at 95°C for 5 min and analyzed by SDS-PAGE and western blotting.
Detailed procedure, antibodies, and primers are listed in the supporting file.
C57BL/6 mice were obtained from the Animal Resources Centre (Perth, Australia). Twenty-day old C57BL/6 mice were inoculated subcutaneously (s.c.) in the thorax below the right fore limb with 104 PFU of RRV-wt or RRV-YF diluted in PBS to a volume of 50 μL. Mock-infected mice were inoculated with PBS only. Mice were weighed and scored for disease daily. RRV disease scores were assessed based on strength and hind-leg dysfunction using the scale described in the supplementary file. Mouse quadriceps were collected and fixed in 4% paraformaldehyde (PFA), followed by paraffin embedding. Samples were cut into 5 μm-thick sections and stained with hematoxylin and eosin. Images were taken using a Nikon microscope.
|
10.1371/journal.pcbi.1003288 | A Neurocomputational Model of the Mismatch Negativity | The mismatch negativity (MMN) is an event related potential evoked by violations of regularity. Here, we present a model of the underlying neuronal dynamics based upon the idea that auditory cortex continuously updates a generative model to predict its sensory inputs. The MMN is then modelled as the superposition of the electric fields evoked by neuronal activity reporting prediction errors. The process by which auditory cortex generates predictions and resolves prediction errors was simulated using generalised (Bayesian) filtering – a biologically plausible scheme for probabilistic inference on the hidden states of hierarchical dynamical models. The resulting scheme generates realistic MMN waveforms, explains the qualitative effects of deviant probability and magnitude on the MMN – in terms of latency and amplitude – and makes quantitative predictions about the interactions between deviant probability and magnitude. This work advances a formal understanding of the MMN and – more generally – illustrates the potential for developing computationally informed dynamic causal models of empirical electromagnetic responses.
| Computational neuroimaging enables quantitative inferences from non-invasive measures of brain activity on the underlying mechanisms. Ultimately, we would like to understand these mechanisms not only in terms of physiology but also in terms of computation. So far, this has not been addressed by mathematical models of neuroimaging data (e.g., dynamic causal models), which have rather focused on ever more detailed inferences about physiology. Here we present the first instance of a dynamic causal model that explains electrophysiological data in terms of computation rather than physiology. Concretely, we predict the mismatch negativity – an event-related potential elicited by regularity violation – from the dynamics of perceptual inference as prescribed by the free energy principle. The resulting model explains the waveform of the mismatch negativity and some of its phenomenological properties at a level of precision that has not been attempted before. This highlights the potential of neurocomputational dynamic causal models to enable inferences from neuroimaging data on neurocomputational mechanisms.
| Recent advances in computational neuroimaging [1] have enabled inferences about the neurophysiological mechanisms that generate non-invasive measures of task or stimulus-evoked neuronal responses; as measured by functional magnetic resonance imaging (fMRI) or electroencephalography (EEG). One such approach is dynamic causal modelling [2] that tries to explain EEG data in terms of synaptic coupling within a network of interacting neuronal populations or sources. However, this description is at the level of physiological processes that do not have a direct interpretation in terms of information processing. Cognitive scientists have been using formal models of cognitive processes to infer on information processing from behaviour for decades [3], but it has remained largely unclear how such inferences should be informed by neurophysiological data. We argue that one may overcome the limitations of both approaches by integrating normative models of information processing (e.g., [4], [5]) with physiologically grounded models of neuroimaging data [4], [5]. This approach may produce computationally informed neuronal models – or neurocomputational models – enabling one to test hypotheses about how the brain processes information to generate adaptive behaviour. Here, we provide a proof-of-concept for this approach by jointly modelling a cognitive process – perceptual inference – and the event related potential (ERP) that it may generate – the mismatch negativity (MMN). Specifically, we ask whether the MMN can be modelled by a neuronal system performing perceptual inference, as prescribed by predictive coding [4], [5].
The MMN is an event-related potential that is evoked by the violation of a regular stream of sensory events. By convention, the MMN is estimated by subtracting the ERP elicited by standards, i.e. events that established the regularity, from the ERP elicited by deviants, i.e. events violating this regularity. Depending on the specific type of regularity, the MMN is usually expressed most strongly at fronto-central electrodes, with a peak latency between 100 and 250 milliseconds after deviant onset [1]. More precisely, the MMN has been shown to depend upon deviant probability and magnitude. Deviant probability is the relative frequency of tones that violate an established regularity. In studies of the MMN evoked by changes in sound frequency, deviance magnitude is the (proportional) difference between the deviant frequency and the standard frequency. The effects of these factors are usually summarized in terms of changes in the MMN peak amplitude and its latency (see Table 1). While increasing the deviance magnitude makes the MMN peak earlier and with a larger amplitude [4], [6], [7], decreasing deviant probability only increases the MMN peak amplitude [8] but does not change its latency [9].
The question as to which neurophysiological mechanisms generate the MMN remains controversial (cf. [10] vs. [11]), even though this issue has been addressed by a large number of studies over the last thirty years [12]. One reason for an enduring controversy could be that the MMN's latency and amplitude contain insufficient information to disambiguate between competing hypotheses (but see [13]). While the MMN is the sum of overlapping subcomponents that are generated in temporal and frontal brain areas [12], [14] – and are differentially affected by experimental manipulations [15] – it is a continuous function of time. This means that the underlying ERP waveforms may contain valuable information about MMN subcomponents, the physiological mechanisms that generate them and, critically, their functional correlates (see e.g. [16]). Predictive coding offers a unique and unified explanation of the MMN's neurophysiological features. In brief, predictive coding is a computational mechanism that formally links perception and learning processes to neural activity and synaptic plasticity, respectively [17]. More precisely, event-related electrophysiological responses are thought to arise from the brain's attempt to minimize prediction errors (i.e. differences between actual and predicted sensory input) through hierarchical Bayesian inference. In this context, the MMN simply reflects neuronal activity reporting these prediction errors in hierarchically organized network of auditory cortical sources. If this is true, then the rise and fall of the MMN may reflect the appearance of a discrepancy between sensory input and top-down predictions – and its resolution through perceptual inference. These ideas have been used to interpret the results of experimental studies of the MMN [8], [18] and computational treatments of trial-wise changes in amplitude [6]. However, no attempt has been made to quantitatively relate predictive coding models to empirical MMN waveforms. Here, we extend these efforts by explicitly modelling the physiological mechanisms underlying the MMN in terms of a computational mechanism: predictive coding. In other words, our model is both an extension to dynamic causal models of observed electrophysiological responses [18], [19] to information processing, and a neurophysiological view on meta-Bayesian approaches to cognitive process [15]. We establish the face validity of this neurocomputational model in terms of its ability to explain the observed MMN and its dependence on deviant frequency and deviance magnitude.
This paper comprises two sections. In the first section, we summarize mathematical models of predictive coding (as derived from the free energy principle), and describe the particular perceptual model that we assume the brain uses in the context of a predictable stream of auditory stimuli. The resulting scheme provides a model of neuronal responses in auditory oddball paradigms. In line with the DCM framework, we then augment this model with a mapping from (hidden) neuronal dynamics to (observed) scalp electrophysiological data. In the second section, we use empirical ERPs acquired during an oddball paradigm to tune the parameters of the observation model. Equipped with these parameters, we then simulate MMN waveforms under different levels of deviant probability and deviance magnitude – and compare the resulting latency and amplitude changes with findings reported in the literature. This serves to provide a proof of principle that dynamic causal models can have a computational form – and establish the face validity of predictive coding theories brain function.
To simulate the MMN under the predictive coding hypothesis, we simulated the processing of standard and deviant stimuli using established Bayesian filtering (or predictive coding) – under a hierarchical dynamic model of repeated stimuli. This generates time-continuous trajectories, encoding beliefs (posterior expectations and predictions) and prediction errors. These prediction errors were then used to explain the MMN, via a forward model of the mapping between neuronal representations of prediction error and observed scalp potentials. In this section, we describe the steps entailed by this sort of modelling. See Figure 1 for an overview.
Perception estimates the causes () of the sensory inputs () that the brain receives. In other words, to recognise causal structure in the world, the brain has to invert the process by which its sensory consequences are generated from causes in the environment. This view of perception as unconscious inference was introduced by Helmholtz [2] in the 19th century. More recently, it has been formalized as the inversion of a generative model of sensory inputs [20]. In the language of probability theory, this means that the percept corresponds to the posterior belief about the putative causes of sensory input and any hidden states that mediate their effect. This means that any perceptual experience depends on the model of how sensory input is generated. To capture the rich structure of natural sounds, the model has to be dynamic, hierarchical, and nonlinear. Hierarchical dynamic models (HDMs) [21] accommodate these attributes and can be used to model sounds as complex as birdsong [22].
HDMs generate time-continuous data as noisy observations of a nonlinear transformation of hidden states and hidden causes :(1)where the temporal evolution of hidden states is given by the differential equation:(2)This equation models the change in as a nonlinear function of the hidden states and hidden causes plus state noise . The hidden causes of the change in are modelled as the outputs of a hidden process at the second level. This second process is modelled in the same way as the hidden process at the first level, but with new nonlinear functions and :(3)As in the first level, the hidden dynamics of the second level are driven by hidden causes that are modelled as the output of a hidden process at the next higher level, and so forth. This composition can be repeated as often as necessary to model the system under consideration – up to the last level, whose input is usually modelled as a known function of time plus noise:(4)The (Bayesian) inversion of HDMs is a difficult issue, which calls for appropriate approximation schemes. To explain how the brain is nevertheless able to recognise the causes of natural sounds, we assume that it performs approximate Bayesian inference by minimising variational free energy [23]. More generally, the free-energy principle is a mathematical framework for modelling how organisms perceive, learn, and make decisions in a parsimonious and biologically plausible fashion. In brief, it assumes that biological systems like the brain solve complex inference problems by adopting a parametric approximation to a posterior belief over hidden causes and states . It then optimises this approximation by minimizing the variational free-energy:(5)One can think of this free-energy as an information theoretic measure of the discrepancy between the brain's approximate belief about the causes of sensory input and the true posterior density. According to the free-energy principle, cognitive processes and their neurophysiological mechanisms serve to minimize free-energy [24] – generally by a gradient descent with respect to the sufficient statistics of the brain's approximate posterior [5]:(6)This idea that the brain implements perceptual inference by free-energy minimization is supported by a substantial amount of anatomical, physiological, and neuroimaging evidence [4]. Algorithms that invert HDMs by minimizing free-energy, such as dynamic expectation maximization [25], [26] and generalized filtering (GF) [4], [5], [23], [27], [28], are therefore attractive candidates for simulating and understanding perceptual inference in the brain.
Importantly, algorithmic implementations of this gradient descent are formally equivalent to predictive coding schemes. In brief, representations (sufficient statistics encoding approximate posterior expectations) generate top-down predictions to produce prediction errors. These prediction errors are then passed up the hierarchy in the reverse direction, to update posterior expectations. This ensures an accurate prediction of sensory input and all its intermediate representations. This hierarchal message passing can be expressed mathematically as a gradient descent on the (sum of squared) prediction errors which are weighted by their precisions (inverse variances) :(6b)where are prediction errors and are their precisions (inverse variances). Here and below, the ∼ notation denotes generalised variables (state, velocity, acceleration and so on). The first pair of equalities just says that posterior expectations about hidden causes and states change according to a mixture of prior prediction– the first term – and an update term in the direction of the gradient of (precision-weighted) prediction error. The second pair of equations expresses precision weighted prediction error as the difference between posterior expectations about hidden causes and (the changes in) hidden states and their predicted values (,), weighed by their precisions . The predictions are nonlinear functions of expectations at each level of the hierarchy and the level above. In what follows, this predictive coding formulation will serve to simulate perceptual recognition. We will then use prediction errors as a proxy for neuronal activity producing ERPs. To simulate neuronal processing using Equation 6, we need to specify the form of the functions that constitute the generative model:
To model auditory cortical responses, we assume that cortical sources embody a hierarchical model of repeated stimuli. In other words, the hierarchical structure of the auditory cortex recapitulates the hierarchical structure of sound generation (cf. [25]). This hierarchical structure was modelled using the HDM illustrated in Figure 2. Note that this model was used to both generate stimuli and simulate predictive coding – assuming the brain is using the same model. The model's sensory prediction took the form of a vector of loudness modulated frequency channels (spectrogram) at the lowest level. The level above models temporal fluctuations in instantaneous loudness () and frequency (). The hidden causes and of these fluctuations are produced by the highest level. These three levels of representation can be mapped onto three hierarchically organized areas of auditory cortex: primary auditory cortex (A1), lateral Heschl's gyrus, and inferior frontal gyrus.
A1 and lateral Heschl's gyrus contain neuronal units encoding posterior expectations and prediction errors, respectively. The activity of the expectation units encodes the time course of for A1 and expectations about hidden states for Heschl's gyrus. Error units encode prediction error, i.e. the difference between posterior expectations and top-down predictions. Top-down connections therefore convey predictions, whereas bottom-up connections convey prediction errors. The hidden causes are the expectations of , providing top-down projections from units in inferior frontal gyrus.
Our model respects the tonotopic organization of primary auditory cortex (see e.g. [26]) by considering 50 frequency channels . It also captures the fact that, while most neurons in A1 have a preferred frequency, their response also increases with loudness [29]–[31]. Specifically, we assume that the activity of neurons selective for frequency is given by:(7)We can rewrite this equation in terms of the loudness and a tuning function that measures how close the log-frequency is to the neuron's preferred log-frequency :(8)This is our (perceptual) model of how the frequency and loudness is encoded by frequency-selective neurons in primary auditory cortex. We use it to simulate the activity of A1 neurons.
Note that a neuronal representation of depends only on frequency. In the brain, frequency representations that are invariant to the sound level (and other sound attributes) are found in higher auditory areas; for instance in marmoset auditory cortex [32]. Neuroimaging in humans suggests that periodicity is represented in lateral Heschl's gyrus and planum temporale [33], and LFP recordings from patients again implicate lateral Heschl's gyrus [34]. We therefore assume that is represented in lateral Heschl's gyrus. The dynamics of the instantaneous frequency is given by(9)This equation says that the instantaneous frequency converges towards the current target frequency at a rate of . In the context of communication, one can think of the target frequency as the frequency that an agent intends to generate, where the instantaneous frequency is the frequency that is currently being produced. The motivation for this is that deviations from the target frequency will be corrected dynamically over time. The agent's belief about reflects its expectation about the frequency of the perceived tone and its subjective certainty or confidence about that expectation. Therefore, the effect of the deviant probability – in an oddball paradigm – can be modelled via the precision of this prior belief.
The temporal evolution of the hidden states and (encoding loudness) was modelled with the following linear dynamical system:(10)In this equation the first hidden cause drives the drives the dynamics of hidden states, which spiral (decay) towards zero in its absence. Finally, our model makes the realistic assumption that the stochastic perturbations are smooth functions of time. This is achieved by assuming that the derivatives of the stochastic perturbations are drawn from a multivariate Gaussian with zero mean:(11)The parameters of this model were chosen according to the biological and psychological considerations explained in Supplementary Text S1.
Having posited the relevant part of the generative model embodied by auditory cortex, one can now proceed to its inversion by the Bayesian generalized filtering scheme described in Equation 6. This is the focus of the next section, which recapitulates how auditory cortex might perceive sound frequency and amplitude using predictive coding mechanisms, given the above hierarchal dynamic model.
The production of the MMN from prediction errors was modelled as a two stage process: the generation of scalp potentials from neuronal responses and subsequent data processing (see Figure 1). We modelled the scalp potentials (at one fronto-central electrode) as the linear superposition of electromagnetic fields caused by the activity of prediction error units in the three simulated cortical sources – plus background activity. Specifically, prediction error units in the A1 source are assumed to encode – the precision weighted sensory error; error units in lateral Heschl's gyrus were assumed to encode – the precision weighted errors in the motion of hidden (log-frequency and amplitude) states; and prediction error units in the inferior frontal gyrus were assumed to encode – the precision weighted errors in their inferred causes. The prediction errors were transformed into event related potentials by three transformations. First, the time axis was shifted (to accommodate conduction delays from the ear) and scaled so that the simulated stimulus duration was 70 ms. Second, a sigmoidal transformation was applied to capture the presumably non-linear mapping from signed precision-weighted prediction error to neural activity (i.e. the firing rate cannot be negative and saturates for high prediction error) and in the mapping from neuronal activity to equivalent current dipole activity; these first two steps are summarized by(12)Finally, the scalp potential is simulated with a linear combination of the three local field potentials plus a constant:(13)Data processing was simulated by the application of down-sampling to 200 Hz and a 3rd order Butterworth low-pass filtering with a cut-off frequency of 40 Hz, cf. [6], [23], [28], [39]. We performed two simulations for each condition. In the first simulation the subject expected stimulus A but was presented with stimulus B (deviant). In the second simulation, the subject expected stimulus B and was presented with stimulus B (standard). The MMN was estimated by the difference wave (deviant ERP – standard ERP). This procedure reproduces the analysis used in electrophysiology [7], [40].
This completes the specification of our computationally informed dynamic causal model of the MMN.
To explore the predictions of this model under different levels of deviant probability and magnitude, we first estimated the biophysical parameters (i.e. the slope parameters in (12) and the lead field in (13)) from the empirical ERPs described in [19], using standard nonlinear least-squares techniques (i.e. the GlobalSearch algorithm [41] from the Matlab Global Optimization toolbox). We then used the estimated parameters to predict the MMN under different combinations of deviant probability and magnitude.
In particular, the simulated MMN waveforms were used to reproduce the descriptive statistics typically reported in MMN experiments, i.e. MMN amplitude and latency. MMN latency was estimated by the fractional area technique [19], because it is regarded as one of the most robust methods for measuring ERP latencies [42]. Specifically, we estimated the MMN latency as the time point at which 50% of the area of the MMN trough lies on either side. This analysis was performed on the difference wave between the first and last point at which the amplitude was at least half the MMN amplitude. This analysis was performed on the unfiltered MMN waveforms as recommended by [43]. MMN amplitude was estimated by the average voltage of the low-pass filtered MMN difference wave within a ±10 ms window around the estimated latency.
Figure 4 shows that the waveforms generated by our model reproduce the characteristic shape of the MMN, the positivity evoked by the standard and the negativity evoked by the deviant. The latency of the simulated MMN (164 ms) was almost identical to the latency of the empirical MMN (166 ms). Its peak amplitude (−2.71 µV) was slightly higher than for the empirically measured MMN (), and its width at half-maximum amplitude (106 ms) was also very similar to the width of the empirical MMN waveform (96 ms). In short, having optimised the parameters mapping from the simulated neuronal activity to empirically observed responses, we were able to reproduce empirical MMNs remarkably accurately. This is nontrivial because the underlying neuronal dynamics are effectively solving a very difficult Bayesian model inversion or filtering problem. Using these optimised parameters, we proceeded to quantify how the MMN waveform would change with deviance magnitude and probability.
To simulate the effect of deviant probability, we simulated the responses to a deviant under different degrees of prior certainty. To simulate the effect of deviance magnitude, we varied the discrepancy between the expected and observed frequency, while keeping the deviant probability constant. Finally, we investigated potential interactions between deviance magnitude and deviant probability by simulating the effect of magnitude under different prior certainties and vice versa.
We have described a process model of the MMN and its dependence on deviant stimulus (deviance magnitude) and context (deviant probability). Together with the study presented in [9], this work demonstrates the potential of predictive coding to provide a comprehensive explanation of MMN phenomenology. More precisely, our model explains the effects of deviant probability and magnitude on the MMN amplitude under the assumption that evoked responses reflect the neuronal encoding of (precision weighted) prediction errors. The simulated MMN was a superposition of the electrical fields generated by prediction errors at different hierarchical levels of representation (see Figure 2), where their relative contributions (i.e. the coefficients in equation (13)) differed: the errors in the predictions at the highest level of representation (inferior frontal gyrus) were weighted most strongly, followed by prediction error at the sensory level (A1) and prediction errors at the intermediate level (lateral Heschl's gyrus). As a result, the simulated MMN primarily reflected prediction errors on the hidden causes (attributes), rather than prediction errors on their physical features.
Our model offers a simple explanation as to why the MMN amplitude decreases with deviant probability and increases with deviance magnitude. Precision weighted prediction errors are the product of a prediction error and the precision of the top-down prediction. Hence, according to our model, deviance magnitude increases MMN amplitude, because it increases prediction errors. Similarly decreasing the probability of the deviant increases the MMN amplitude by increasing the precision of (learned) top-down predictions. Furthermore, since precision and prediction error interact multiplicatively, the precision determines the gain of the effect of prediction error and vice versa.
This model explains the shortening of the MMN latency with deviance magnitude by a selective amplification of frequency-related prediction errors that are only transiently expressed – because they are explained away quickly by top-down predictions. These prediction errors increase with deviance magnitude. However, there are also prediction errors that are not explained away by perceptual inference. These errors are sustained throughout the duration of the stimulus (as the stimulus amplitude fluctuates) and do not depend on the difference between the standard and the deviant event. Hence, according to our model, deviance magnitude selectively increases the early prediction error component, but not sustained errors. In effect, as deviance magnitude increases, an early trough emerges within the MMN, so that the MMN latency shortens (see Figure 5a and Figure 6). By contrast, increasing the precision of high-level beliefs increases all precision weighted frequency prediction errors – the transient and the sustained – equally. Thus the MMN deepens uniformly, and no early trough emerges. This is why – according to the model – the deviant probability has no effect on the MMN latency for moderate deviance magnitudes. However, if the deviance magnitude is so large that the transient component dominates the frequency-related prediction error, the situation is different. In this case, increasing the weight of the frequency-related prediction errors relative to loudness-related prediction errors can shorten the latency, because the frequency-related prediction error predominates at the beginning of perception – whereas the amplitude related prediction error is constant throughout perception. This is why our model predicts that the MMN latency becomes dependent on deviant probability at higher levels of deviance magnitude.
Our MMN simulations predict a nonlinear interaction between the effects of deviant probability and magnitude. The upper plot in Figure 6 suggests that the effect of deviant probability on MMN peak amplitude increases with increasing deviance magnitude. Conversely, the effect of deviance magnitude increases with decreasing deviant probability. Furthermore, the lower plot in Figure 6 suggests, that the effect of deviant probability on MMN latency depends on deviance magnitude: If deviance magnitude is at most 12.7%, the MMN latency does not depend on deviant probability, but when deviance magnitude is as large as 32%, the MMN latency increases with deviant probability. Conversely, the size of the effect of deviance magnitude on MMN latency depends on deviant probability. Hence, our simulations predict a number of interaction effects that can be tested empirically.
Although the physiological mechanisms generating the MMN have been modelled previously [9], the model presented here is the first to bridge the gap between the computations implicit in perceptual inference and the neurophysiology of ERP waveforms. In terms of Marr's levels of analysis [53], our model provides an explanation at both the algorithmic and implementational levels of analysis – and represents a step towards full meta-Bayesian inference – namely inferring from measurements of brain activity on how the brain computes (cf. [13], [19], [51]–[55]).
Our model builds upon the proposal that the brain inverts hierarchical dynamic models of its sensory inputs by minimizing free-energy in a hierarchy of predictive coding circuits [56]. Specifically, we asked whether the computational principles proposed in [15], [20] are sufficient to generate realistic MMN waveforms and account for their dependence on deviant probability and deviance magnitude. In doing so, we have provided a more realistic account of the algorithmic nature of the brain's implementation of these computational principles: While previous simulations have explored the dynamics of perceptual inference prescribed by the free-energy principle using dynamic expectation maximization (DEM) [23], [39], the simulations presented here are based on GF [23], [39]. Arguably, GF provides a more realistic model of learning and inference in the brain than DEM, because it is an online algorithm that can be run in real-time to simultaneously infer hidden states and learn the model; i.e., as sensory inputs arrive. In contrast to DEM it does not have to iterate between inferring hidden states, learning parameters, and learning hyperparameters. This is possible, because GF dispenses the mean-field assumption made by DEM. Another difference to previous work is that we have modelled the neural representation of precision weighted prediction error by sigmoidal activation functions, whereas previous simulations ignored potential nonlinear effects by assuming that the activity of prediction error units is a linear function of precision weighted prediction error [6], [24], [27], [39]. Most importantly, the model presented here connects the theory of free-energy minimisation and predictive coding to empirical measurements of the MMN in human subjects.
To our knowledge, our model is the first to provide a computational explanation of the MMN's dependence on deviance magnitude, deviant probability, and their interaction. While [26] modelled the effect of deviance magnitude, they did not consider the effect of deviant probability. Although [6], [24] modelled the effect of deviant probability, they did not simulate the effect of deviance magnitude, nor did they make quantitative predictions of MMN latency or amplitude. Mill et al. [13], [55] simulated the effects of deviance magnitude and deviant probability on the firing rate of single auditory neurons in anaesthetized rats. While their simulations captured the qualitative effects of deviance magnitude and deviant probability on response amplitude, they did not capture the shortening of the MMN latency with decreasing deviant probability. By contrast, our model generates realistic MMN waveforms and explains the qualitative effects of deviant probability and magnitude on the amplitude and latency of the MMN. Beyond this, our model makes remarkably accurate quantitative predictions of the MMN amplitude across two experiments [53] examining several combinations of deviance magnitude and deviant probability.
The simulations reported in this paper demonstrate that predictive coding can explain the MMN and certain aspects of its dependence on the deviant stimulus and its context. However, they do not imply that the assumptions of predictive coding are necessary to explain the MMN. Instead, the simulations are a proof-of-concept that it is possible to relate the MMN to a process model of how prediction errors are encoded dynamically by superficial pyramidal cells during perceptual inference. For parsimony, our model includes only those three intermediate levels of the auditory hierarchy that are assumed to be the primary sources of the MMN. In particular, we do not model the subcortical levels of the auditory system. However, our model does not assume that predictive coding starts in primary auditory cortex. To the contrary, the input to A1 is assumed to be the prediction error from auditory thalamus. This is consistent with the recent discovery of subcortical precursors of the MMN [52]. Since MMN waveforms were simulated using the parameters estimated from the average ERPs reported in [9], [10], the waveforms shown in Figure 4 are merely a demonstration that our model can fit empirical data. However, the model's ability to predict how the MMN waveform changes as a function of deviance magnitude and deviant probability speaks to its face validity.
Our model's most severe failure was that while our model correctly predicted that MMN latency shortens with deviance magnitude, it failed to predict that this shortening occurs gradually for deviance magnitudes between 2.5% and 7.5%. In principle, the model predicts that the latency shortens gradually within a certain range of deviance magnitudes, but this range did not coincide with the one observed empirically.
There are clearly many explanations for this failure – for example, an inappropriate generative model or incorrect forms for the mapping between prediction errors and local field potentials. Perhaps the more important point here is that these failures generally represent opportunities. This is because one can revise or extend the model and compare the evidence for an alternative model with the evidence for the original model using Bayesian model comparison of dynamic causal models in the usual way [57]–[59]. Indeed, this is one of the primary motivations for developing dynamic causal models that are computationally informed or constrained. In other words, one can test competing hypotheses or models about both the computational (and biophysical) processes underlying observed brain responses.
This work is a proof-of-principle that important aspects of evoked responses in general – and the MMN in particular – can be explained by formal (Bayesian) models of the predictive coding mechanism [19]. Our model explains the dynamics of the MMN in continuous time and some of its phenomenology at a precision level that has not been attempted before. By placing normative models of computation within the framework of dynamic causal models one has the opportunity to use Bayesian model comparison to adjudicate between competing computational theories. Future studies might compare predictive coding to competing accounts such as the fresh-afferent theory [60]–[62]. In addition, the approach presented here could be extended to a range of potentials evoked by sensory stimuli, including the N1 and the P300, in order to generalise the explanatory scope of predictive coding or free energy formulations.
This sort of modelling approach might be used to infer how perceptual inference changes with learning, attention, and context. This is an attractive prospect, given that the MMN is elicited not only in simple oddball paradigms, but also in more complex paradigms involving the processing of speech, language, music, and abstract features [7], [53], [63]. Furthermore, a computational anatomy of the MMN might be useful for probing disturbances of perceptual inference and learning in psychiatric conditions, such as schizophrenia [13], [55]. Similarly, extensions of this model could also be used to better understand the effects of drugs, such as ketamine [12], [64]–[66], or neuromodulators, such as acetylcholine [67]–[69], on the MMN. We hope to pursue this avenue of research in future work.
|
10.1371/journal.pcbi.1000648 | A Multidimensional Strategy to Detect Polypharmacological Targets in the Absence of Structural and Sequence Homology | Conventional drug design embraces the “one gene, one drug, one disease” philosophy. Polypharmacology, which focuses on multi-target drugs, has emerged as a new paradigm in drug discovery. The rational design of drugs that act via polypharmacological mechanisms can produce compounds that exhibit increased therapeutic potency and against which resistance is less likely to develop. Additionally, identifying multiple protein targets is also critical for side-effect prediction. One third of potential therapeutic compounds fail in clinical trials or are later removed from the market due to unacceptable side effects often caused by off-target binding. In the current work, we introduce a multidimensional strategy for the identification of secondary targets of known small-molecule inhibitors in the absence of global structural and sequence homology with the primary target protein. To demonstrate the utility of the strategy, we identify several targets of 4,5-dihydroxy-3-(1-naphthyldiazenyl)-2,7-naphthalenedisulfonic acid, a known micromolar inhibitor of Trypanosoma brucei RNA editing ligase 1. As it is capable of identifying potential secondary targets, the strategy described here may play a useful role in future efforts to reduce drug side effects and/or to increase polypharmacology.
| Proteins play a critical role in human disease; bacteria, viruses, and parasites have unique proteins that can interfere with human health, and dysfunctional human proteins can likewise lead to illness. In order to find cures, scientists often try to identify small molecules (drugs) that can inhibit disease-causing proteins. The goal is to identify a molecule that can fit snugly into the pockets and grooves, or “active sites,” on the protein's surface. Unfortunately, drugs that inhibit a single disease-causing protein are problematic. A single protein can evolve to evade drug action. Additionally, when only one protein is targeted, drug potency is often diminished. Single drugs that simultaneously target multiple disease-causing proteins are much more effective. On the other hand, if scientists are not careful, the drugs they design might inhibit essential human proteins in addition to inhibiting their intended targets, leading to unexpected side effects. In our current work, we have developed a computer-based procedure that can be used to identify proteins with similar active sites. Once unexpected protein targets have been identified, scientists can modify drugs under development in order to increase the simultaneous inhibition of multiple disease-causing proteins while avoiding potential side effects by decreasing the inhibition of useful human proteins.
| Researchers have traditionally focused in silico efforts on designing inhibitors of specific protein targets, giving less attention to the computational identification of unpredicted secondary targets. This tendency is surprising given the frequency with which secondary receptors are responsible for both detrimental and beneficial pharmacological effects. The cost of developing a novel therapeutic ranges from $500 million to $2 billion dollars [1]. Millions of dollars are typically invested to advance a compound through clinical trials, but one third of these compounds fail or are later removed from the market due to unacceptable, medically harmful side effects [2] often caused by binding to off-target receptors. The detrimental effects caused by drug binding to unknown secondary targets can be financially and clinically devastating.
In other cases, compound binding to multiple therapeutic targets (polypharmacology) is beneficial. Conventional drug discovery embraces the “one gene, one drug, one disease” philosophy; however, drugs that target only one protein are susceptible to resistance, as a single amino-acid mutation in the target active site often substantially reduces compound binding affinity. Resistance to multi-target drugs, on the other hand, requires simultaneous mutations in multiple protein targets. Furthermore, drugs with polypharmacological mechanisms often have greater therapeutic potency. Some serotonergic drugs, for example, bind both 5-HT G-protein coupled receptors as well as the 5-HT3A ion channel to achieve their therapeutic benefits, despite the fact that these two target proteins are not related by sequence or structure [3].
Identifying secondary targets in neglected tropical diseases, diseases for which drug development is neither profitable nor prevalent, allows researchers and doctors to retool approved drugs as novel treatments for the otherwise abandoned infections of the developing world. For instance, eflornithine, initially developed as an anti-cancer compound, was found to be a potent inhibitor of Trypanosoma brucei ornithine decarboxylase and is now a critical therapeutic in the fight against human African trypanosomiasis [4]. Examples like these motivate the need to develop new tools and algorithms to predict potential protein targets of candidate compounds.
A barrier to the development of these tools is the frequent absence of apparent evolutionary relationships among the multiple protein targets of a given compound, requiring that any potential method be capable of identifying target receptors independently of global sequence or structural homology. One approach is chemo-centric [5]; similar ligands are more likely to have similar properties and therefore often bind proteins with similar active sites. A number of studies have successfully identified secondary receptors by comparing their known small-molecule ligands [3], [6]–[8], leading to probabilistic models that can in some cases successfully predict polypharmacology. Despite these successes, however, chemo-centric approaches have their limitations. Chemically similar small molecules do not always inhibit proteins with similar active sites; indeed, even small changes in the chemical structures of some small-molecule inhibitors can drastically alter binding affinity [3] and the broad profile of binding to pharmacological targets.
A second approach is protein-centric. As the evolutionary relationships between secondary targets are not always apparent [3], receptor active-site geometries and pharmacophores must be compared directly, independently of global sequence or structural homology. Geometric constraints have been used extensively to identify binding sites and to assess binding-site similarity [9]–[16]. Of these methods, the sequence order independent profile-profile alignment (SOIPPA) algorithm is largely insensitive to both conformational changes in protein structure as well as the uncertainty inherent in homology models and low-resolution structures [17].
In the current work, we present a novel multidimensional strategy to identify the multiple protein receptors of a given compound that incorporates three levels of information: sequence-based homology clustering, the SOIPPA algorithm in conjunction with a geometric-potential metric [17]–[19], and in silico ligand docking. To demonstrate the utility of the strategy, we identify several human and pathogen secondary targets of compound 1 (NSC-45208), 4,5-dihydroxy-3-(1-naphthyldiazenyl)-2,7-naphthalenedisulfonic acid, a recently discovered micromolar inhibitor of T. brucei RNA editing ligase 1 (TbREL1) [20]. TbREL1 is a confirmed drug target in T. brucei, the causative agent of human African trypanosomiasis, a disease for which drug development has been largely neglected [21].
Sequence homology clustering was used to identify 12,646 protein-chain clusters from among the 110,000 protein chains present in the Protein Data Bank (PDB) as of late 2007 (Figure 1A, B). A representative protein chain was chosen from each cluster, creating a smaller set of chains called the PDB30 (Figure 1C). The SOIPPA algorithm in conjunction with a geometric-potential metric [17]–[19] revealed that the active sites of 12,428 of the representative PDB30 protein chains (98.3%) were dissimilar to that of TbREL1 (p>0.05), the known target. After discarding these dissimilar protein chains, 218 representative PDB30 chains remained (Figure 1D). The remaining PDB30 proteins and the clusters they represented were merged into a single list containing 2,897 chains, a list enriched with possible secondary targets (Figure 1E). By considering only proteins from human or known human-pathogen species and eliminating PDBs with formatting errors, the number of chains was reduced from 2,897 to 645 (22.3%). This new list of protein chains was called the PDBr (PDBreduced).
Compound 1, a known inhibitor of the primary target, TbREL1, was docked into each of the 645 potential secondary targets of the PDBr with AutoDock 4.0 (Figure 1F). After docking, both protein chains of unknown function and chains belonging to proteins with duplicate names were deleted without regard for species. This pruning produced a list of 87 non-redundant predicted secondary targets. The docked pose of 1 into each of these 87 predicted secondary targets was analyzed to confirm binding to a pocket of known biochemical or pharmacological activity. In two instances, SITE data was included in the published PDB file, allowing us to identify one active-site and one off-site docking. In 25 instances, the docked ligand bound at the same location as a co-crystallized ligand, typically superimposed on top of it. Homology modeling revealed that an additional nine docked ligands bound at active sites of known biochemical or pharmacological activity, while 34 bound at alternate sites on the receptor. Though these alternate sites may be involved in allosteric regulation of protein function, we chose not to pursue them further. In all, the list of 87 protein chains contained 35 chains whose known active sites contained docked ligands, 35 chains whose alternate sites contained docked ligands, and 17 chains that could not be classified (Text S1).
The predicted secondary targets that gave the best docking scores, H. sapiens mitochondrial 2-enoyl thioester reductase (HsETR1), T. brucei UDP-galactose 4′ epimerase (TbGalE), H. sapiens phosphodiesterase 9A (HsPDE9A2), and Streptococcus pneumoniae teichoic acid phosphorylcholine esterase (SpPce), were subsequently tested experimentally.
Compound 1 inhibited HsETR1 with a measured IC50 of approximately 33.5 µM and a Hill slope of 1.06. Neither the FATCAT algorithm nor CLUSTALW2 judged HsETR1 to be significantly homologous in sequence or structure to the primary target, TbREL1 (p = 0.754; identity = 4%). In contrast, the SOIPPA algorithm judged the active sites of these two proteins to be significantly similar (p<1×10−5) (Table 1). HsETR1 aggregates were not detected, as measured by both spectrometry and centrifugation. A mixture of compound 1 and HsETR1 was run through a gel filtration column, thereby removing any unbound ligand. Spectroscopic analysis subsequently demonstrated that 1 was not covalently bound to HsETR1.
Compound 1 inhibited TbGalE with a measured IC50 of 0.7±0.2 µM and a Hill slope of 1.13+/−0.36. Again, the FATCAT algorithm did not judge the structure of TbGalE to be significantly similar to that of the primary target, TbREL1 (p = 0.627), and the CLUSTALW2 algorithm did not identify significant sequence homology (identity = 1%) (Table 2). TbGalE inhibition was unaffected by the presence of detergent, and activity could be restored by dialysis of the protein, demonstrating that the TbGalE inhibition was not due to aggregation of the compound or chemical modification of the protein.
Two of the predicted secondary targets, HsPDE9A2 and SpPce, were uninhibited by 1 at 200 µM and 10 mM, respectively. AutoDock predicted that 1 would bind HsPDE9A2 and SpPce with −18.19 and −28.00 kcal/mol, respectively (Tables 1 and 2).
In this work, we attempt to further the study of polypharmacological and side-effect prediction by presenting a multidimensional strategy for identifying secondary targets of known enzyme inhibitors in the absence of global structure and sequence homology. To demonstrate the utility of the strategy, we identify secondary targets of 1, a recently discovered inhibitor of TbREL1 [20] from T. brucei, the causative agent of African sleeping sickness. TbREL1 plays a critical role in the editing of trypanosomatid mitochondrial RNA transcripts and is required for the survival of both the T. brucei insect and bloodstream forms [22],[23]. Additionally, TbREL1 is a particularly attractive drug target because there are no known close human homologs [20].
Compound 1 was chosen to illustrate how the current strategy can be applied early in the drug-discovery process. The compound inhibits a known drug target, satisfies Lipinski's rule of five, and is structurally similar to surinam, a drug currently approved for the treatment of human African trypanosomiasis. In this sense, 1 is drug like. Compound 1 has not yet been optimized to bind TbREL1 in the nanomolar regime, however, and does contain some undesirable functional groups, and so is still very much under development. By incorporating the identification of secondary targets early in the drug-design process, we hope to eventually make modifications to compound 1 that will increase the binding affinity to the primary target while decreasing binding to undesirable secondary targets.
Of the 35 predicted secondary targets of compound 1 identified, twelve were human proteins. Potential side effects of 1 can be predicted by considering the physiological role of these targets. For instance, a number of the predicted secondary human targets regulate metabolism, including the experimentally confirmed secondary target HsETR1. Neither FATCAT nor CLUSTALW2 judged HsETR1to be homologous to the primary target, TbREL1 (Table 1). The current strategy, which is not dependent on sequence or global structural homology, was able to identify this secondary target where identification by homology would have failed. HsETR1 is thought to be essential for fatty acid synthesis (FAS) type II [24]. In the process of optimizing 1 to make it more druglike, modifications that reduce binding to human HsETR1 may diminish unforeseen side effects. Interestingly, AutoDock predicted that 1 partially occupies a co-factor (NADPH) binding site, suggesting that the compound may function as a competitive inhibitor for NADPH (Figure S1A).
H. sapiens UDP-galactose 4-epimerase (HsGalE), a second protein involved in human metabolism, was also identified as a potential secondary target. Though HsGalE shares little homology with the primary target, TbREL1 (FATCAT p-value: 0.610; CLUSTALW2 identity: 2%), it is highly homologous with TbGalE (FATPAT p-value: 0.00; CLUSTALW2 identity: 37%; Figure S2A), which we show to be a secondary target of 1 (IC50: 0.7±0.2 µM). Mutations in the HsGalE gene cause type 3 galactosemia in humans. As toxic levels of galactose build up in patients' blood, vomiting, hepatomegaly, jaundice, renal failure, and cataracts typically follow [25]. Chronic administration of 1 is thus ill advised, though short-term treatment may be acceptable if patients avoid dairy and other sources of galactose.
In addition to metabolism, a number of the predicted human secondary targets are involved in polynucleotide synthesis, repair, and replication. H. sapiens DNA ligase I (HsLigI), a protein that belongs to the same enzyme superfamily as TbREL1, is one notable example. HsLigI catalyzes the ultimate, essential step in DNA replication, repair, and recombination [26]. Studies have demonstrated that HsLigI is defective in at least one representative lymphoid cell line of Bloom's syndrome origin [27], and a mouse model with a mutant HsLigI allele exhibits an increased incidence of spontaneous cancer [28].
Though not tested explicitly, additional evidence does suggest that 1 binds HsLigI. First, 1 is known to inhibit H. sapiens DNA ligase IIIβ (HsLigIIIβ; IC50: 27.49±6.40 µM) [20], a HsLigI homolog (Table 3). Second, the FATCAT algorithm [29] judged the structure of HsLigI to be significantly similar to that of the primary target, TbREL1, and the SOIPPA algorithm judged the active sites of these two proteins to be similar (Table 1). Finally, the AutoDock-predicted binding energy of 1 to HsLigI was −9.70 kcal/mol (Table 1).
Of the 35 predicted secondary targets identified, 23 belonged to bacterial and parasitic species. Among these predicted pathogenic secondary targets, enzymatic assays demonstrated that 1 inhibits TbGalE [30] (Table 2) with high nanomolar affinity. Interestingly, the FATCAT algorithm did not judge the structure of TbGalE to be significantly similar to that of the primary target, TbREL1 (Table 2), nor did CLUSTALW2 suggest sequence homology (identity = 1%). Thus, had we attempted to identify secondary targets by protein sequence or structural homology alone, TbGalE would not have been detected. This result again illustrates the power of the current strategy.
Like TbREL1, TbGalE is essential for T. brucei survival [31]. Thus, 1 inhibits two essential T. brucei enzymes, an example of potential polypharmacology. In the process of optimizing 1 to make it more druglike, modifications that increase binding to both TbREL1 and TbGalE would likely improve drug efficacy and decrease the chances of resistance through mutation. Interestingly, AutoDock again predicted that 1 would bind partially in the TbGalE NAD+ pocket, suggesting that the compound may be a competitive inhibitor for NAD+ rather than UDP-galactose (Figure S1B).
Though not tested explicitly, another predicted secondary target in pathogens, Salmonella enterica dTDP-D-glucose 4,6 dehydratase (SeRmlB), shares sequence and structural homology with TbGalE and may also bind 1 (Table 2, Figure S2A). The FATCAT algorithm judged SeRmlB to be significantly structurally similar to TbGalE (p value = 0.00), and CLUSTALW2 identified some sequence homology (identity = 21%). Additionally, AutoDock predicted that SeRmlB would bind 1 with relatively high affinity (Table 2).
SeRmlB is the second enzyme in the dTDP-L-rhamnose biosynthetic pathway; L-rhamnose is part of the LPS endotoxin (the O antigen) in many serotypes and serovars of Gram-negative bacteria. As L-rhamnose is common in the cell walls and envelopes of some pathogenic bacteria but absent in humans, SeRmlB is thought to be a potential drug target [32]. These findings support the hypothesis that compounds similar to 1 may have anti-bacterial properties.
Aside from inhibiting an essential protein in the bacterial dTDP-L-rhamnose biosynthetic pathway, the current strategy also identified two bacterial DNA ligases as potential secondary targets. Enterococcus faecalis v583 NAD-dependent DNA ligase (Ef ligase) and Mycobacterium tuberculosis DNA ligase (Mt ligase) are structurally homologous to HsLigI (Table 3, Figure S2B). Human DNA ligases require ATP as a co-factor, but bacterial ligases require NAD+ [33]. Because of this important biochemical difference, bacterial ligases are thought to be good drug targets.
A limitation of the current method is the prediction of false positives. Two of the predicted secondary targets, SpPce and HsPDE9A2, were uninhibited by 1. Closer inspection of the docking results revealed electrostatic energy components of −30.89 kcal/mol and −12.29 kcal/mol, respectively. In both cases, the partial charges of several active-site metal cations had been manually set to the corresponding formal charge. Subsequent analysis of the docked poses revealed that for both receptors, one of the sulfonate groups of 1 was juxtaposed against these highly charged metal cations (Figure S3). Clearly, a more careful treatment of the electrostatic interactions that accounts for electron polarization is warranted when docking into active sites that include metal ions.
An additional limitation of the current method is its dependence on sequence-homology clustering. Because of computational limitations, the number of potential secondary targets analyzed with SOIPPA had to be reduced; consequently, rather than analyzing all protein chains in the PDB, representative chains were selected from clusters of homologous proteins based on the supposition that sequence-homologous proteins would likely have similar active sites. This supposition, however, is hardly a universal truth. Through convergent evolution, two proteins with very different primary sequences may have evolved to bind similar ligands, and so may have similar active sites despite a lack of sequence homology.
Fortunately, recent advances in the SOIPPA algorithm now make sequence-homology clustering unnecessary. The version of SOIPPA used in the current study estimated the statistical significance of each active-site comparison using a non-parametric statistics method that required at least several hundred additional comparisons to derive a background distribution. Recently, an extreme-value distribution model has been developed to compute the statistical significance of SOIPPA scores [34]; this model improves the speed of the algorithm by at least two orders of multitude, so that each active-site comparison can be performed in mere seconds. Ligand binding-site similarity searches can now be performed on a genome-wide scale without the need for sequence-homology clustering [34],[35]. The new statistical model has been implemented in SMAP v2.0, available at http://funsite.sdsc.edu.
The manual verification of predicted compound binding to active sites of known biochemical or pharmacological activity also presented a limiting challenge. This step was very time consuming; had multiple compounds been tested, manual verification of all docked poses may have been impossible. To automate the process, active-site binding can be confirmed in many cases based on the proximity of the docked ligand to known catalytic residues annotated in the Catalytic Site Atlas [36] or to active-site residues specified in the site records of the PDB. Additionally, the SOIPPA implementation in SMAP v2.0, not available until recently, suggests an initial ligand binding pose for each predicted secondary target. Docked poses could be compared to this initial suggestion using an automated script and rejected in the absence of proximity.
As experimental validation has confirmed that 1 inhibits multiple protein targets, the possibility of nonspecific, promiscuous binding must be eliminated. Previous work has demonstrated that compound 1 inhibition of several ATP-dependent proteins depends on the degree of homology with the primary target, TbREL1, suggesting that binding occurs via a specific rather than promiscuous mechanism [20]. Additionally, the lack of SpPce or HsPDE9A2 inhibition further suggests that indiscriminant inhibition is unlikely.
Promiscuous inhibition can occur when a compound forms colloidal aggregates that inhibit indiscriminately. In one recent study, 95% of the inhibitors identified in a high-throughput screen were subsequently found to inhibit via a nonspecific aggregation-based mechanism [37]. In theory, the chances of aggregation are reduced in the case of 1 because of its negative charge; individual molecules should repel each other, preventing aggregation. To confirm this theory, Amaro et. al assayed 1 against TbREL1 in the presence of a non-ionic detergent as well as an additional protein (BSA) known to prevent aggregation. The presence of the detergent and the separate test with BSA did not significantly influence TbREL1 inhibition, demonstrating that 1 does not aggregate [20].
The current work likewise suggests no aggregation. The presence of detergent had no effect on TbGalE inhibition. Additionally, no HsETR1 aggregation was observed, as measured by both spectrometry and centrifugation (see the Text S1). The Hill slopes corresponding to the inhibition of HsETR1 and TbGalE were 1.06 and 1.13, respectively. As these slopes are approximately equal to one, the inhibition of these two proteins likely occurs via ligand binding to a single site, as predicted. One recent study suggested that aggregation-based inhibition typically produces Hill slopes that are much steeper, with average values of about 2.2 [37].
Promiscuous inhibition can also occur when compounds chemically modify the proteins they inhibit. Several experiments were performed in order to rule out chemical modification. To test for chemical modification of TbGalE, TbGalE was incubated with 1, and subsequent dialysis was used to remove any unbound ligand. TbGalE activity was unaffected following dialysis, as compared to treatment with DMSO alone. Had 1 been covalently linked to TbGalE, the compound would not have been washed away, and the enzyme would have shown little activity.
To test for chemical modification of HsETR1, a mixture of compound 1 and HsETR1 was run through a gel filtration column, which likewise removed any unbound ligand. The absorption spectrum of the fraction containing HsETR1 was subsequently analyzed and did not demonstrate the peaks characteristic of 1 at 530 nm and 320 nm, likewise demonstrating that 1 was not covalently bound to HsETR1.
A number of the predicted secondary targets identified belong to the same or similar biochemical pathways (i.e. metabolic pathways; DNA synthesis, repair, and replication pathways; and DNA ligase pathways) (Table 1). This result is encouraging, as proteins of the same pathway often act on similar substrates and so have similar active sites. A cursory chemo-centric look at the native substrates of the identified secondary targets in many instances corroborates our findings. For instance, the human proteins HsLigI, 3-methyl-adenine DNA glycosylase, and thymidylate synthase, all involved in DNA synthesis, repair, and replication, were identified as potential target receptors. HsLigI is a TbREL1 homolog (Table 1) that ligates DNA in a way analogous to TbREL1 dsRNA ligation. In contrast, 3-methyl-adenine DNA glycosylase is not a TbREL1 homolog (Table 1), but an examination of its structure nevertheless reveals that it also binds DNA. One of the nucleotides of the bound DNA, an alkylated base generated endogenously by lipid peroxidation, protrudes into a deep binding pocket in a way analogous to TbREL1-ATP binding [38]. Thymidylate synthase, another predicted secondary target involved in nucleotide synthesis, does not bind double-stranded nucleic acid, but rather binds deoxyuridine monophosphate, a compound with a nucleotide-ribose-phosphate substructure like that of ATP, a known substrate of the primary target, TbREL1.
Notably, two experimentally validated secondary targets contain NAD+ or NADPH co-factors, and in both cases 1 is predicted to bind at least partially in one of the co-factor binding pockets. Similar to ATP, NAD+ and NADPH both contain adenine-ribose-biphosphate substructures. Of the 23 proteins listed in Tables 1 and 2, eight contain NAD+-like co-factors. We expect that there are even more true positives among the predicted secondary targets; to this end, we provide the entire list of predicted targets in Text S1.
Traditionally, researchers have devoted their computational efforts to designing inhibitors of specific protein targets while paying less attention to the in silico prediction of secondary targets. Because adverse side effects are often discovered late in the drug-development process, often after the investment of many millions of dollars, we recommend using the current strategy to help bad drugs “fail early,” or, better yet, to guide the drug-discovery process towards more selective inhibitors. Additionally, the current methodology can help medicinal chemists overcome the conventional “one gene, one drug, one disease” paradigm. The rational design of drugs that act via polypharmacological mechanisms can produce compounds that exhibit increased therapeutic potency and against which resistance is less likely to develop.
Compound 1 was provided by the NCI/DTP Open Chemical Repository (http://dtp.cancer.gov). Compound identity was confirmed by high-resolution mass spectrometry, and no impurities were detected by 1H-NMR.
The computational strategy presented here utilizes three distinct components in order to identify secondary pharmacological targets.
Identifying potential secondary targets from among the 110,000 protein chains deposited in the RCSB PDB [39] as of late 2007 was judged computationally intractable. In order to reduce the number of protein chains, redundancies in the RCSB PDB were eliminated by clustering all protein chains by sequence using the NCBI blastclust program, with a sequence identity threshold of 30% and an overlap threshold of 0.9 (Figure 1A, B). A representative protein chain was then chosen at random from each cluster, thus creating a smaller set of chains called the PDB30 (Figure 1C).
In order to eliminate those members of the PDB30 whose active-site alpha-carbon configurations were different enough from that of the primary target, TbREL1, so as to likely preclude NSC4520 binding, we used the SOIPPA algorithm in conjunction with a geometric potential [17]–[19], which computes general binding-site similarity based on shape (alpha-carbon tessellation), physical properties, and the evolutionary profiles of the active-site residues, without regard for specific side-chain positions or global sequence or structure (Figure 1D). Using SOIPPA and the geometric potential, we eliminated all protein chains in the PDB30 with active sites that were dissimilar to that of the primary target, TbREL1 (p-value>0.05) (Figure 1D). The p-value was calculated from a non-parametric density function generated from 980 PDB chains with unique SCOP folds [18]. To derive the background distribution, a Gaussian function was placed at each observation. The mean of the Gaussian of the observed binding-site similarity score and its variance were fixed. The final probability density function was the sum of all these Gaussian functions. The optimal bandwidth was estimated from the data by using a least square cross-validation approach [40].
Each representative protein chain corresponded to a PDB cluster containing multiple homologous chains. A new set of protein-chain structures called the PDBr (PDBreduced) was created by taking the union of all those clusters whose representative PDB30 protein chains had active sites that were not dissimilar to that of TbREL1 (p-value<0.05) (Figure 1E). By considering only proteins from human or known human-pathogen species, the number of chains was significantly reduced. An additional protein, 1GJJ, was eliminated because of apparent PDB formatting errors. 1S31 was retained despite having a malformed GLU residue (561) to which Gasteiger partial charges could not be assigned.
AutoDock 4.0 [41] was used to dock 1 into the protein chains of the PDBr (Figure 1F). In previous work, AutoDock was validated against TbREL1 [20]. To define the distinct docking grid associated with each protein chain, SOIPPA was used to identify the most probable active site, and the grid box was set to the smallest possible X-Y-Z cube encompassing all the atoms of the SOIPPA-reported alignable alpha carbons. Additional details regarding receptor and ligand preparation and grid and docking parameters can be found in Text S1.
All dockings were sorted by the predicted binding energy of their most-populated AutoDock clusters. After eliminating protein chains of unknown function from the list, the data was grouped according to species and protein, revealing significant redundancy in the PDBr. We selected the ligand-protein pair from each group with the best AutoDock score, producing a list of non-redundant ligand-protein pairs, one corresponding to each protein/species group (Text S1).
For each of these ligand-protein pairs, we used one of several methods to determine if AutoDock placed 1 in an active site of known biochemical or pharmacological activity. First, SITE data included in the published PDB file identified several active sites. Second, co-crystallized ligands bound in native active sites were examined. Finally, homology modeling was used to determine the locations of active sites for the remaining protein chains, when possible. A protein chain was considered to be a “hit” if 1 had a high predicted energy of binding and if 1 was predicted to bind in an identified active site of known biochemical or pharmacological activity. A description of the assays used to experimentally validate several of the predicted secondary targets is included in Text S1.
|
10.1371/journal.ppat.1006670 | Comparative genomics of geographically distant Fusarium fujikuroi isolates revealed two distinct pathotypes correlating with secondary metabolite profiles | Fusarium fujikuroi causes bakanae (“foolish seedling”) disease of rice which is characterized by hyper-elongation of seedlings resulting from production of gibberellic acids (GAs) by the fungus. This plant pathogen is also known for production of harmful mycotoxins, such as fusarins, fusaric acid, apicidin F and beauvericin. Recently, we generated the first de novo genome sequence of F. fujikuroi strain IMI 58289 combined with extensive transcriptional, epigenetic, proteomic and chemical product analyses. GA production was shown to provide a selective advantage during infection of the preferred host plant rice. Here, we provide genome sequences of eight additional F. fujikuroi isolates from distant geographic regions. The isolates differ in the size of chromosomes, most likely due to variability of subtelomeric regions, the type of asexual spores (microconidia and/or macroconidia), and the number and expression of secondary metabolite gene clusters. Whilst most of the isolates caused the typical bakanae symptoms, one isolate, B14, caused stunting and early withering of infected seedlings. In contrast to the other isolates, B14 produced no GAs but high amounts of fumonisins during infection on rice. Furthermore, it differed from the other isolates by the presence of three additional polyketide synthase (PKS) genes (PKS40, PKS43, PKS51) and the absence of the F. fujikuroi-specific apicidin F (NRPS31) gene cluster. Analysis of additional field isolates confirmed the strong correlation between the pathotype (bakanae or stunting/withering), and the ability to produce either GAs or fumonisins. Deletion of the fumonisin and fusaric acid-specific PKS genes in B14 reduced the stunting/withering symptoms, whereas deletion of the PKS51 gene resulted in elevated symptom development. Phylogenetic analyses revealed two subclades of F. fujikuroi strains according to their pathotype and secondary metabolite profiles.
| Fusarium fujikuroi causes bakanae disease of rice. Infected seedlings appear to be taller and more slender when compared to healthy seedlings due to its ability to produce gibberellic acids (GAs). The disease is responsible for high yield losses, and its incidence varies with regions, rice cultivars grown and the aggressiveness of the fungal isolates. However, not all infected seedlings show bakanae symptoms: one of the isolates, B14, causes stunting and early withering of infected seedlings. The reason for the two pathotypes is not well understood. Researchers thought that the stunting phenotype was mostly caused by fungal-derived secondary metabolites such as fusaric acid, but there is no experimental evidence yet. B14 differs from the other strains by the presence of more PKS gene clusters, low expression of GA genes, lack of detectable levels of GAs and the production of high amounts of fumonisins in rice. Analysis of additional field isolates revealed a strong correlation between the pathotype (bakanae or stunting) and either GA or fumonisin production. Based on phylogenetic analyses, F. fujikuroi strains can be divided into two phylogenetically distinct subclades according to their pathotype and secondary metabolite profiles. This study provides new insights into the genomic variations and the population structure inside the species F. fujikuroi which will help to develop disease control strategies for this rice pathogen.
| The heterothallic ascomycete Fusarium fujikuroi Nirenberg is a member of the Fusarium fujikuroi species complex (FFC), a monophyletic lineage which includes at least eleven mating populations (MPs A-K) that are sexually infertile with one another, and numerous distinct anamorphic species [1].
F. fujikuroi (MP-C) is the causal agent of the rice disease bakanae (“foolish seedlings”), one of the most notorious seed-borne diseases with increasing economic importance in the major rice-growing countries in the world, including all rice-growing Asian and African countries, California, and more recently, Italy [1–3]. The fungus was one of the first fungal pathogens to be described, and bakanae is one of the oldest known diseases of rice being reported more than 100 years ago by Japanese scientists [4,5]. The most prominent symptoms of the disease are chlorotic, elongated and thin seedlings that are often several inches taller than healthy plants, and empty panicles leading to yield losses ranging from ca. 30–95% [6–9]. However, not all infected seedlings show the bakanae symptoms: sometimes they may be stunted or appear symptomless [10]. The incidence and severity of the bakanae or stunting disease symptoms varies with regions and isolate. The pathogen is dispersed predominantly with infected seeds, infected crop residues from the previous season in the soil, or by conidia on diseased stems which can be transmitted by rain and wind [6]. Disease control has become increasingly difficult due to rapidly developing fungicide resistance in the fungal population [4].
The enormous elongation of infected plants is caused by the ability of the pathogen to produce gibberellic acids (GAs), a family of plant hormones [11]. Fungal GAs are structurally identical to those synthesized by higher plants, but the respective biosynthetic pathways, genes and enzymes differ [12–14]. Previously we have shown that the ability of the fungus to produce GAs contributes to the efficient colonization in the rice roots [15]. In contrast to typical bakanae symptoms, it is unknown how the stunting phenotype of infected rice seedlings is triggered by F. fujikuroi. In addition, stunting of infected rice plants can also be caused by other Fusarium species, such as F. proliferatum, which can also be isolated from rice, though less abundantly than F. fujikuroi [16].
Recently, the first high-quality draft genome sequence of F. fujikuroi IMI 58289 has been published and the genetic capacity for biosynthesis of a whole arsenal of natural compounds has been demonstrated [15]. The genomes of additional F. fujikuroi isolates revealed some diversity regarding genome composition and virulence [17,18].
Here we present the genome sequences of eight additional F. fujikuroi strains, all but one isolated from infected rice from different geographic regions. The isolate FSU48 was obtained from maize. We provide a comparative analysis of the genome features, chromosome polymorphism, the ability to produce micro- and macroconidia, virulence, metabolome and transcriptome analyses under in vitro and in planta conditions. By the use of high-performance liquid chromatography coupled to a Fourier transform mass spectrometer (HPLC-FTMS) and genome-wide RNA-sequencing (RNA-seq), we demonstrate that in addition to species-specific common features there are differences between the isolates in all these aspects. Most importantly, we describe two pathotypes of F. fujikuroi on rice at the genomic, transcriptomic, and phylogenetic levels. Whereas most of the F. fujikuroi isolates caused typical bakanae symptoms with elongated chlorotic internodes, the isolate B14 caused stunting and withering of rice seedlings. We show that variations in the production of secondary metabolites (SMs), such as GAs and fumonisins, are crucial factors for the development of the bakanae or stunting pathotype, and that these two pathotypes are phylogenetically distinct groups among the field population of F. fujikuroi.
F. fujikuroi is broadly distributed world-wide in all rice-growing countries. To gain insight to the level of variation regarding genome structure, morphology, asexual spore formation, virulence, expression profiles and secondary metabolism under laboratory conditions as well as on rice, we chose nine isolates from different areas of the world for comparative analysis of all these parameters (Table 1, Fig 1A).
The high quality genome sequence of strain IMI 58289 that was assembled into twelve scaffolds corresponding to the twelve chromosomes [15] was used as master genome for structural annotation and for comparative analysis. The rice isolate V64-1 from Ruanda appeared to be a F. oxysporum strain when the genome was sequenced and analyzed. Therefore, this strain was used as outgroup in this study. A phylogenetic tree including all so far sequenced F. fujikuroi isolates and other FFC members, as well as more distantly related Fusarium species is shown in Fig 1B. The tree was generated based on the protein sequences of 5,181 single copy genes by the fast approximate likelihood ratio test to calculate branch support (aLRT) [19], which is a fast and accurate alternative to the time-consuming bootstrap analysis. Strain KSUX10626 [17] seems to be outside the F. fujikuroi clade.
Although all newly sequenced strains except for V64-1 clearly belong to the species F. fujikuroi, visualization of chromosome content of the ten strains by pulse field gel electrophoresis (PFGE) combined with clamped homogeneous electric fields (CHEF) indicated that all strains contained multiple chromosomes of varying sizes (Fig 1C). However, the precise number of chromosomes in each strain could not be determined because several chromosomes had a similar size and could not be distinguished.
Table 2 summarizes physical genome features of the newly and previously (IMI 58289) sequenced strains. The genome size for the F. fujikuroi strains is in the range of 43.9 Mb (IMI 58289) to 46.1 Mb (E282 and FSU48). The outgroup genome F. oxysporum V64-1 is smaller (49.1 Mb) than the reference F. oxysporum strain 4287 (61.4 Mb). Most of these differences are likely due to a different read coverage and different completeness of assemblies for repetitive regions. Despite the varying number of protein-encoding genes (14,817 for IMI 58289 to 16,088 for E282) which partially may be due to a manual gene structure validation, most key genome features of the newly sequenced strains are similar to those of F. fujikuroi IMI 58289 [15].
The completeness of the new draft genomes was explored by comparing each predicted proteome to two different, highly conserved eukaryote protein sets by BLAST [20,21]. Orthologs for all conserved proteins were found in all proteomes with the exception of a missing ortholog to ‘T-complex protein 1 subunit theta’ and an ortholog to ‘translation factor eIF6’ in F. fujikuroi B20. In F. oxysporum V64-1, an ortholog to ‘MTO1 –mitochondrial translation optimization’ is missing suggesting that also these genomes are more than 99% complete. In addition, the protein sets were subjected to BUSCO analysis and subsequently compared to published Fusarium proteome sets [22]. Of the 3,725 single-copy orthologs searched (library Sordariomyceta_odb9), 97.9%–99.2% were detected as complete and single-copy in the F. fujikuroi strains and F. oxysporum V64-1 which even exceeds the completeness of F. graminearum PH1 (94.7%) and F. oxysporum (94.2%) (S1 Table).
Species of the FFC are heterothallic [23]. To determine the mating type of the nine F. fujikuroi isolates, we searched the genomes for the presence of genes either of the MAT1-1 or MAT1-2 mating types. Five of the strains contain the genes MAT1-1-1, MAT1-1-2 and MAT1-1-3, all belonging to the MAT1-1 mating type, while the other four contain the MAT1-2 mating type locus with the HMG-box type transcription factor (TF)-encoding gene MAT1-2-1 and an additional gene, MAT1-2-3, that has been recently identified as specific to the Hypocreales [23] (S2 Table).
Differences between the isolates were found regarding the size of gene families prevalent in the genome of B14, which is the only one of the ten examined strains causing stunting of rice seedlings (see below). The number of transporter-encoding genes (954), Zn(II)2Cys6 fungal type TFs (567), cytochrome P450- (173) and dehydrogenase-encoding genes (404) were higher when compared to the other F. fujikuroi strains. Also the number of polyketide synthase (PKS) genes (18 plus two truncated pseudogenes) was larger than in the genomes of the other strains (Table 3).
The strain NCIM 1100 encodes less PKS (eight plus one truncated pseudogene) and nine terpene cyclase (TC) genes which results in the inability to produce gibepyrones, fujikurin and certain terpenes besides other unknown products (S3 Table and below).
To gain a deeper insight into the variation at the chromosome ends between the ten analyzed strains, we used a PCR approach with primer pairs from the two ends of each chromosome based on the genome sequence of strain IMI 58289 [15]. This analysis revealed differences between the isolates either on one side or on both sides of the chromosomes (S1 Fig). The most variations were found for subtelomeric regions of chromosomes 3, 5, 7, 10 and 11. Two of the key enzyme-encoding genes of SM gene clusters, NRPS31 (apicidin F) and PKS16 (unknown product), are located at the far end of chromosome 1 and 11, respectively, and the PCR analysis clearly showed differences in the presence of these SM genes (S1 Fig).
All isolates showed differences in colony morphology and pigmentation on solidified and in liquid media, respectively, indicating metabolic variation between them (Fig 2A–2E). There were differences in the ability to produce the red pigments, bikaverin and fusarubins, despite the presence of the respective gene clusters in all F. fujikuroi isolates (Fig 2C–2E). Previously, we have shown that both PKS-derived pigments are only produced under low nitrogen conditions. However, while bikaverin biosynthesis is induced at acidic pH, the perithecial pigments fusarubins are produced only under alkaline pH conditions [24,25]. Strain C1995 showed no coloration under bikaverin-production conditions (Fig 2C), and C1995, B14, FSU48 and NCIM 1100 showed no pigmentation under fusarubin production conditions (Fig 2E). To verify that the biosynthetic genes are not expressed in these isolates, we performed Northern blot analyses using the bikaverin and fusarubin biosynthetic genes as probes (S2 Fig). FSR2 (encoding an O-methyltransferase in the fusarubin gene cluster) was expressed in IMI 58289, but not in the remaining strains suggesting that the fusarubin genes are only slightly expressed in m567, MRC2276, B20, E282 and F. oxysporum V64-1, or alternatively that the red pigmentation in these strains under fusarubin production conditions (Fig 2F) might be due to the expression of the bikaverin biosynthetic genes.
The species F. fujikuroi is described to produce slender macroconidia with three to five septa and oval or club-shaped microconidia, mostly without or with one septum [1]. However, the strains used in this work differ in their ability to produce asexual spores, i.e. micro- and macroconidia (Fig 2F). While most of the strains produce both types of conidia, others produce predominantly (B14 and NCIM 1100) or exclusively (IMI 58289) microconidia or predominantly macroconidia (E282), respectively, on V8 agar under light conditions. Strain m567 hardly forms any spores. In A. nidulans, asexual reproduction has been extensively studied for several decades [26–28]. Sequential activation of three major regulators, BrlA, AbaA and WetA, is necessary for the fungus to undergo asexual development. In addition, there are a number of fluffy regulatory genes (FLBA–FLBE) which regulate BRLA expression. Recently, it has been shown that an AbaA-WetA pathway is conserved in the distantly related species Fusarium graminearum [29–31]. There are highly conserved homologs for FlbB, FlbC, FlbD, FlbE and the central regulators WetA and AbaA also in the genomes of the F. fujikuroi isolates. However, no close relative to BrlA, which is responsible for the vesicle formation during conidiogenesis in A. nidulans, has been found in any Fusarium genome.
In recent years, several new SMs have been identified in F. fujikuroi due to the deciphering of the fungal genome and the application of molecular techniques to activate silent gene clusters, e.g. those for apicidin F, fujikurins, beauvericin, trichosetin, and reversely N-prenylated tryptophan (r-N-DMAT) [32–36] as well as for the sesquiterpenes eremophilene and guaia-6,10(14)-diene [37,38]. In addition, well known SMs such as fusaric acid, fusarins, fusarubins and gibepyrones have been linked to the respective biosynthetic gene clusters [39–42].
Bioinformatic analysis of the nine F. fujikuroi strains revealed several differences in the presence of PKS, NRPS, dimethylallyltryptophan synthase (DMATS) and TC gene clusters between them and also compared to closely related FFC members, such as F. mangiferae and F. proliferatum [43] (S3 Table). Altogether, there are 17 NRPS-, 23 PKS-, three DMATS-, and twelve TC-encoding genes present in the nine F. fujikuroi strains, i.e. 55 unique core-enzyme-encoding genes that could give rise to 54 distinct SMs (the fusaric acid cluster encodes two core enzymes, PKS6 and NRPS34). Only twelve of the 23 PKS-, 13 of the 18 NRPS-, eight of the twelve TC- and two of the three DMATS-encoding genes are present in the genomes of all F. fujikuroi isolates. The Korean strain B14 has three additional PKS genes (PKS40, 43 and 51) not present in any other analyzed F. fujikuroi strain (S3 Table, Fig 3A and 3B). PKS51 is also not present in any other member of the FFC, while PKS40 is present in F. proliferatum ET1 and F. verticillioides, and PKS43 in F. mangiferae, respectively (S3 Table). In addition, B14 is the only F. fujikuroi isolate with a complete PKS5 cluster. This yet uncharacterized cluster is either absent or contains pseudogenes in the other isolates of this species, but seems to be functional in F. proliferatum and F. mangiferae (Fig 4A). The fujikurin gene cluster (PKS19) has been previously described as F. fujikuroi-specific [15], but was recently discovered in two newly sequenced strains of F. proliferatum [43]. However, the fujikurin cluster is absent in three of the nine analyzed F. fujikuroi isolates (S3 Table, Fig 4B). The apicidin F gene cluster (NRPS31) which is located at the far end of chromosome 1 in strain IMI 58289 [15] was shown to be unique for F. fujikuroi [32]. This cluster is present in all isolates but one, B14, most likely due to chromosome rearrangements in the subtelomere regions (S1 and S3A Figs). While the entire fumonisin gene cluster (PKS11) is present in most members of the FFC except for F. mangiferae (S3 Table), nine genes of the cluster, the homologs of FFUJ_09248 –FFUJ_09256, are missing in F. fujikuroi B20 (S3B Fig) and C1995.
The Indian isolate NCIM 1100 has only 14 PKS-encoding genes (S3 Table). The absence of five PKS genes/clusters is due to large parts of chromosome 11 missing at both ends in this strain compared to IMI 58289 (Figs 3C and 4C, S1 Fig). The right border of chromosome 11 contains PKS8 (FFUJ_12090), PKS13 (FFUJ_12020), PKS17 (FFUJ_12066), and PKS18 (FFUJ_12074), while PKS16 (FFUJ_11199) and adjacent genes are absent at the left border of this chromosome (Figs 3C and 4C, S3 Table). Also the discontinuous distribution of some other SM gene clusters among the F. fujikuroi isolates is likely the result of gene loss and gene gain events, respectively, at the chromosome ends. However, gene clusters located in central parts of chromosomes, e.g. PKS19 (fujikurin) and PKS11 (fumonisin) clusters, are also only present in some and absent in other strains, while the genes at the right and left borders of the clusters show colinearity between all isolates. It is not clear whether these gene clusters are the result of horizontal gene transfer [44] or cluster duplication and loss (birth and death) as shown for the fumonisin gene cluster [45].
Thanks to the genome sequencing and subsequent identification of putative gene clusters [15], SM products have now been assigned to 22 of the 54 predicted PKS-, NRPS-, DMATS-, and TC-derived SM gene clusters in isolates of F. fujikuroi, and an additional two (ferrirhodin and depudecin) in members of the FFC. One reason for the inability to identify more products is probably due to the fact that many of the SM genes are not or only minimally expressed under laboratory conditions [46]. Only 15 of these predicted clusters with known products are present in the genomes of all nine F. fujikuroi isolates indicating the genetic diversity among isolates of one species. The genes which are commonly present in all F. fujikuroi genomes are four NRPS genes required for synthesis of the siderophores ferricrocin and fusarinine and the mycotoxins beauvericin and fusaric acid (together with PKS6); six PKS genes required for synthesis of fusarubins, bikaverin, fusaric acid (together with NRPS8), fusarins, trichosetin, and fumonisins; six TCs required for synthesis of GAs, phytoene, eremophilene, (–)-α-acorenol, (–)-guaia-6,10(14)-diene, and (+)-koraiol, and one DMATS responsible for biosynthesis of r-N-DMAT (S3 Table).
To study the variability in the production of the most prominent SMs, we cultivated the strains under three standardized culture conditions (6 mM and 60 mM glutamine, 6 mM NaNO3). The optimal conditions for the production of the different metabolites were previously shown to vary considerably regarding nitrogen availability and pH in strain IMI 58289 [15]. Therefore, we analysed these SMs in the three media by high-performance liquid chromatography coupled to a Fourier transform mass spectrometer (HPLC-FTMS) or by HPLC with a diode array detector (HPLC-DAD) in the case of GAs. We compared the production levels (Table 4) with transcriptome profiles of the SM biosynthetic genes generated by RNA-seq for two of these conditions (6 mM and 60 mM glutamine) (Table 5A and 5B). For most of the SMs, we observed similar regulatory mechanisms regarding nitrogen availability as previously described for strain IMI 58289. However, there were strain-specific differences.
The most prominent and species-specific SMs are the GAs causing the F. fujikuroi-specific bakanae disease of rice. Although some recently sequenced species of the FFC such as F. mangiferae and F. proliferatum contain one or even two GA gene clusters, these species produce either no or only very small amounts of GAs [43]. In this work, we examined whether all analyzed F. fujikuroi strains produce GAs, and whether the GA levels can be correlated to the virulence on rice.
Previously, the regulation of GA biosynthesis has been extensively studied for strain IMI 58289. It has been shown that GA gene expression is strictly regulated by nitrogen availability in an AreA- and AreB-dependent manner [47–49]. To examine whether low nitrogen conditions are optimal for GA production also in the other isolates, we performed HPLC-DAD analysis of all strains and analysed the expression by Northern blot analysis in addition to RNA-seq data. Accordingly, we observed the highest GA yields and GA gene expression levels under low nitrogen conditions for eight of the nine F. fujikuroi strains. The only exception among the F. fujikuroi isolates is strain B14 which showed no production of GAs and no visible expression of the GA genes despite the presence of the complete GA gene cluster in the genome (Table 5A and 5B; Fig 5A and 5B). Besides the very low expression of GA genes, strain B14 differs from the other strains by high expression of fumonisin biosynthetic genes under low nitrogen conditions (Table 5A, Fig 5A and 5C). Recently, we have shown that the fumonisin gene cluster in F. fujikuroi IMI 58289 is almost silent. Consequently, only very low amounts of fumonisins are produced in comparison to F. verticillioides. In F. fujikuroi IMI58289, the genes are only expressed and fumonisins are only produced, when the cluster-specific TF gene FUM21 is constitutively and strongly expressed [50]. Here, similar results were observed for eight of the nine F. fujikuroi isolates (except for B14) which all produce either no or only very small amounts of fumonisins and showed very low expression levels.
In addition, B14 is the only isolate producing fujikurins under alkaline and low nitrogen conditions (Table 4). Previously, the fujikurin gene cluster (PKS19) was shown to be silent under all conditions tested in strain IMI 58289, and the products have been identified only after simultaneous over-expression of both the cluster-specific TF gene and the PKS19 gene itself [15,34].
In most of the strains, fusarins, fusaric acid, and apicidin F (except for B14) were only produced under high nitrogen conditions (60 mM glutamine) in accordance with the higher expression levels of the corresponding biosynthetic genes. The genes of the beauvericin cluster were recently shown to be silent under all conditions tested and activated only after deletion of the histone deacetylase gene HDA1 and knock-down of the histone methyltransferase gene KMT6 in strain IMI 58289 [33,38]. Most of the other strains analyzed here showed no or very low expression levels and no beauvericin production. The only exception was strain B20 that highly expressed the biosynthetic genes under both nitrogen conditions (low and high amounts of glutamine) and produced high levels of beauvericin at high nitrogen (Table 4; Table 5A and 5B) suggesting that the chromatin status around the beauvericin cluster differs in B20.
To examine whether the isolates differ in their virulence on rice, and whether the different levels of GAs or the strain-specific production of other SMs (e.g. fumonisins in B14) correleate with the extent of symptom development, we performed assays with both germinating seeds and rice seedlings.
To determine the ability of all ten strains to impair the seed germination, rice seeds were soaked for 18 h in spore suspensions of the respective strains. The percentage of germinated seeds was counted after 14 days post inoculation (dpi). Strains B14 and B20 induced a high percentage of seedling death indicating a high potential to kill the host seedlings (S4 Table). As B14 does not produce much GAs, the aggressiveness of this strain cannot be caused by these phytohormones.
Next, we performed a rice seedling assay to compare the virulence of the different isolates to assess their ability to induce bakanae symptoms. The surface-sterilized seeds were first germinated, and the young healthy seedlings were then inoculated with spores. All strains except for B14 and F. oxysporum V64-1 (outgroup) induced the formation of slender, elongated and yellowish stems (Fig 6). Wheras V64-1-infected seedlings behaved like the water control, B14-infected seedlings were stunted and showed withering instead of the typical bakanae symptoms. The total seedling length and the length of the internodes were smaller than, or comparable to uninfected seedlings (water control). Previously, it has been reported that heavily infected seedlings can also be stunted and can show severe crown and root rot [51]. The type of symptoms and severity of disease depends on the fungal isolate and is thought to be affected by the proportions of GA and fusaric acid produced by the fungus, which potentially cause elongation of the plants or stunting, respectively [52,53]. However, this assumption has never been proved experimentally.
To examine which of the SM gene clusters are similarly expressed in rice in all isolates, and which of them are specifically expressed or not expressed in strain B14, and therefore could be relevant for disease symptoms (especially stunting), we performed RNA-seq for all isolates also from rice seedlings at 7 dpi (Table 5C). The expression patterns were compared with those under in vitro conditions at high (60 mM glutamine) and low (6 mM glutamine) nitrogen (Table 5A and 5B).
In general, the GA biosynthetic genes (shown for the key enzyme gene CPS/KS) are the most highly expressed genes in rice roots except for B14 in which they are hardly expressed (Table 5C).
To check whether the expression profiles of SM genes correspond to the in planta production levels, we also performed SM analysis of infected rice plants by HPLS-FTMS at 7 dpi. In accordance with gene expression, the in planta GA production levels also differed between the isolates: B14-infected rice plants contained no detectable GA amounts in rice compared to the bakanae strains (Table 6). The most obvious difference of strain B14 compared to the other isolates was the high expression of the fumonisin genes resulting in significant levels of fumonisins (FB1 and FB2) in planta, similar to those observed in vitro (Table 5C; Table 6). Furthermore, while all isolates showed similar expression patterns for fusaric acid biosynthetic genes (PKS6; NRPS34) and similar fusaric acid production levels (Table 5C, Table 6), B14 was the only strain with low expression of the fusarin C genes. As expected, no fusarins were detectable in rice roots inoculated with this strain (Table 6).
In addition, the yet uncharacterized PKS51 gene cluster, which is only present in B14, was exclusively expressed in planta, suggesting that the unknown product of this gene cluster might play a role during infection (Table 5C). Besides B20, B14 also gave high expression for beauvericin genes, but only low expression of gibepyrone (PKS13) and acorenol (STC6) biosynthetic genes.
In conclusion, the very low expression of GA genes and the lack of detectable GA levels in B14-infected rice seedlings after 7 dpi are most likely responsible for the absence of bakanae symptoms in this isolate. Instead, the high levels of fumonisins which are exclusively produced only in this isolate, may overrule the growth-stimulating effect of the GAs and cause stunting/withering.
To find out whether one of the gene clusters that are specifically expressed in B14 during infection on rice (Table 5C) might indeed cause the stunting effect of this isolate, the key enzyme-encoding genes for fumonisins (FUM1 = PKS11), and the yet uncharacterized B14-specific PKS51 gene were deleted in this strain. In addition, we also deleted the key gene for fusaric acid biosynthesis (FUB1 = PKS6) due to the speculation that fusaric acid production might cause the stunting pathotype [52].
The 5-day-old healthy rice seedlings were soaked in the conidial suspension of B14 or the mutant strains. At 5 dpi, the B14-infected seedlings were already stunted compared to the water control and showed withering symptoms which were even more obvious at 7 dpi and 9 dpi (Fig 7A). In contrast, the Δfum1 and Δfub1 strains of B14 appeared healthy at 7 dpi and caused delayed withering symptoms at 9 dpi only. The roots of all infected seedlings were heavily colonized with fungal mycelia at 9 dpi (Fig 7D). The double deletion mutants of B14 lacking both FUM1 and FUB1 behaved like the mock control and did not induce stunting at 5 dpi (Fig 7A) or occasionally upto 9 dpi. However, in some cases, they caused a similar delayed disease development as the single deletion strains at 9 dpi (Fig 7B and 7C). The add-back strains, which were generated by introducing the native copy of FUM1 and FUB1 into Δfum1 and Δfub1 mutants, respectively, caused typical stunting/withering symptoms similar to B14 (S4A Fig). The delay of stunting/withering symptom development by Δfum1 and Δfub1 strains indicates that the production of fumonisins and/or fusaric acid, in combination with the non-detectable levels of GAs, play an important role for the development of this specific pathotype in B14.
For comparison, the genes FUM1 and FUB1 were also deleted in strain B20 (bakanae strain). Unlike the B14-derived mutants, all of the gene deletion strains of B20 we examined caused typical bakanae symptoms (elongated, slender and chlorotic shoots), as did B20. No withering of the seedlings or mycelial colonization on roots were visible suggesting that the absence of fumonisin production by B20 and B20-1 derived mutants protect the rice seedlings from development of B14-like symptoms (S5 Fig).
Surprisingly, rice seedlings inoculated with the ΔPKS51 mutant strains showed an earlier and more severe symptom development compared to those with B14. The stunting/withering symptoms were clearly shown already at 5 dpi (Fig 7A). The more severe symptoms caused by the ΔPKS51 strains may suggest a possible role of the PKS51 product as an avirulence determinant. A similar role was described for the product of the Magnaporthe grisea PKS-NRPS gene ACE1 (Avirulence Conferring Enzyme1) which is also specifically expressed only on rice. Its yet unknown product is probably recognized by rice cultivars carrying a specific resistance gene [54,55]. However, further investigations will be needed to show whether the product of PKS51 acts in a similar way in F. fujikuroi strain B14.
To further study the impact of these SMs on disease symptom development, fumonisin FB1, fusaric acid, or GA3 were exogenously supplied to rice seedlings infected with wild-type or mutant strains (Fig 7A). Addition of FB1 and fusaric acid to rice seedlings infected with Δfum1 and Δfub1, respectively, restored the WT phenotype and resulted in stronger withering symptoms compared to the deletion strains without the toxins. However, the even more reduced virulence of the double deletion strain (Δfum1/Δfub1) was not clearly complemented by exogenous supplies of both fumonisin and fusaric acid (Fig 7A). Interestingly, exogenous supply of GA3 did not cause bakanae symptoms on rice seedlings inoculated with the wild-type B14 strain, while addition of GA3 to the Δfum1/Δfub1 double deletion strain caused stem elongation of rice seedlings at 5 dpi, similar to the bakanae symptom caused by B20 (Fig 7A). Therefore, conversion of B14 into a bakanae pathotype by addition of GAs was only possible after deleting the key genes for the production of fumonisins and fusaric acid.
Furthermore, exogenous supply of culture filtrate from strain B14 to seedlings infected with B14, B20 or B20 Δcps/ks caused stunting and withering symptoms while the culture filtrate of the B14 Δfum1/Δfub1 mutant caused milder symptoms (S4B and S4C Fig). Addition of culture fluid to B14-infected seedlings resulted in even more severe stunting than B14 alone (S4B Fig). Exogenous supply of B14 culture filtrate to seedlings inoculated with B20 or the GA-deficient B20 Δcps/ks mutant revealed severe stunting symptoms in both cases irrespective of the ability to produce GAs (S4C and S4D Fig). These data support our suggestion that fumonisins and fusaric acid play an important role for symptom development.
To further investigate the role of fumonisins for causing stunting and withering, we performed the rice seedling pathotest with two F. verticillioides strains from corn which were shown to produce high amounts (more than 3,000 μg/g) of fumonisins and no GAs in rice seedlings. Both F. verticilloides isolates (Os35 and Os40) [56] caused withering at 7 dpi although the plants were not stunted compared to those of the mock control (S6A Fig). In addition, mycelia of both F. verticillioides strains colonized the roots of infected rice seedling as much as those of B14 (S6B Fig). This strong root colonization was not observed for roots infected with B20. Previously, it has been already reported that B14 triggered severe inhibition of root growth, and that its own growth rate in rice roots was more than 4 times higher compared to that of B20.
Taking together the results of FUM1 and FUB1 deletion, exogenous addition of pure toxins or culture filtrate of B14 to rice seedlings inoculated with wild-type or mutant strains, and the pathotests with GA-deficient, highly fumonisin-producing F. verticillioides isolates provide strong indications that the biosynthesis of both fumonisins and fusaric acid and the lack of GA biosynthesis in B14 play cruicial roles for causing the stunting/withering phenotype on rice seedlings. However, additional factors are probably involved in inducing the stunting symptoms.
Besides SMs, stunting/withering might be caused also by the different sets of TFs present in the genome of B14 and the bakanae strains, or by different expression levels of TF-encoding genes in rice. Therefore, we compared the expression levels of TF-encoding genes between B14 and the other eight F. fujikuroi isolates (S5A Table). There are 37 genes which are present in most of the nine genomes and which were specifically up-regulated during infection of rice (S5B Table). B14 had slightly higher expression levels only for three of them: FFB14_03090, FFB14_05980 and FFB14_01631. In addition, B14 has 28 strain-specific TFs which are not present in the genomes of the other strains (S5C Table). The most highly expressed gene in planta was FFB14_06367 encoding the pathway-specific TF of the putative PKS51 gene cluster. The high expression of PKS51 and the adjacent genes, including the TF-encoding gene, supports our assumption that this unique gene cluster is involved in determining the severity of disease symptom development.
Because B14 was the only isolate causing the stunting pathotype among the analyzed ten strains, we attempted to determine how often this pathotype can be found in rice fields. Therefore, we inoculated rice seedlings with 15 field isolates, which were collected from rice grains and air above rice paddy fields in Korea between 2014 and 2016. Among the 15 isolates, we identified additional nine field isolates causing stunting and early withering symptoms similar to B14 while six isolates caused typical bakanae symptoms (Fig 8A).
To determine whether these isolates can be phylogenetically distinguished from each other, we generated a phylogenetic tree using the nucleotide sequences of the RPB2 and EF1A genes [57] from the new field isolates, the ten isolates used in this study, and some closely related Fusarium species of the FFC. The F. fujikuroi clade, which was clearly separated from those of other closely related species such as F. proliferatum and F. verticillioides, contained two strongly supported subclades (with 76% bootstrap support: BS). Interestingly, the subclade with B14 consists of all of the field isolates causing stunting symptoms, while the other subclade contained B20 and the other F. fujikuroi strains used in this study and all bakanae-type field isolates (Fig 8B). This result indicates that the two pathotypes of F. fujikuroi exist as phylogenetically distinct groups within the population.
To examine if there is a correlation between the pathotype and the presence of one of the SM clusters specific to B14 or B20 (PKS51 for B14, unknown product, and NRPS31 for B20, apicidin F) among the field isolates of each pathotype, we performed a PCR amplification using primer sets derived from PKS51 (unknown product) and NRPS31 (apicidin F), respectively. The PKS51-specific primer set amplified a fragment from all of the stunting-type isolates examined including B14, but not from the bakanae-type isolates. Similarly, the NRPS31-specific primer set amplified a fragment only from bakanae-type isolates examined (S7A Fig).
Based on the correlation between the phylogenetically distinct pathotypes on the one hand, and the presence of either the PKS51 or the apicidin F (NRPS31) gene clusters, we examined additional field isolates from Korea using these primer sets. Among a total of 151 isolates examined, the B14-specific PCR amplification pattern was found in 69% of rice isolates, 100% of corn isolates, and 85% of airspora isolates. A phylogenetic tree generated from all of these isolates revealed a clustering of the stunting-type and bakanae-type isolates similar to that for the field isolates examined by pathogenicity tests (S7B Fig). However, it is currently unclear why the stunting-type population is predominant among the field isolates in Korea.
To gain more information on the SM profile of these new stunting-type isolates, we examined the expression levels of the key genes of GA (CPS/KS) and fumonisin (FUM1) biosynthesis in the field isolates grown in liquid culture with 6 mM glutamine by use of quantitative real-time PCR (qPCR). The CPS/KS transcript levels from all the stunting-type isolates examined were similar to or even lower than that in B14, while those from all the bakanae-type isolates, including B20, were 8- to 24-fold higher than those in the stunting-type strains (S8A Fig). Furthermore, all stunting-type isolates showed clear expression of FUM1 in contrast to the bakanae isolates (shown for B20) (S8B Fig). These data indicate that both subgroups of isolates differ in a whole set of characteristic features (presence of either the PKS51 or NRPS31 genes, expression of either GA or fumonisin genes) which correlate with either the stunting or the bakanae pathotype. We also examined the production levels for both SMs in some of the new B14-like strains in vitro and in planta. As expected, the stunting-type isolates produced no or 10 to 15 times less GAs than the bakanae strain IMI 58289, while only the stunting-type strains produced fumonisins in submers cultures (S9A and S9B Fig). The analysis of GAs and fumonisins in rice roots and shoots revealed no GAs at all in seedlings infected with stunting-type isolates, while only the latter produced significant amounts of fumonisins in planta (S9C and S9D Fig). A similar correlation between low GA and high fumonisin levels on the one hand, and a pathotype called “dwarfism” on the other hand, was also described for the Italian F. fujikuroi isolate CSV1 [58].
In conclusion, we provide genome sequences of eight new F. fujikuroi isolates and one F. oxysporum isolate, all collected in different rice growing regions worldwide. We show that all these strains differ in genome and chromosome size, number of genes in major gene families such as TFs, transporters, SM biosynthetic genes and others. In addition, the isolates differ in colony morphology, pigmentation and the number and type of asexual spores (micro- and/or macroconidia). Major differences were identified in subtelomeric regions of the chromosomes where several SM gene clusters are located. Besides the differences in the presence of gene clusters, we observed variations in the ability to express SM genes and to produce the respective metabolites under in vitro and in planta (rice) conditions. Among the nine F. fujikuroi isolates analyzed, eight cause typical bakanae symptoms on rice seedlings due to their ability to produce GAs. Only one isolate, B14, does not cause elongation of infected plants, but instead causes stunting combined with early withering of rice seedlings. This isolate is the only one which does not produce GAs under in vitro and in planta conditions. Instead, B14 produces high amounts of fumonisins both under in vitro and in planta conditions. Furthermore, it is the only strain containing a putative gene cluster with PKS51 as key gene which is highly expressed in rice. To examine the determinant(s) of the stunting/withering pathotype, several key enzyme-encoding candidate genes were deleted in B14. The data demonstrate that the formation of fumonisins, and probably also fusaric acid, on the one hand, and the lack of GA production on the other hand, contribute to the stunting/withering pathotype, because the deletion of the respective key genes resulted in reduced virulence. Furthermore, the unknown PKS51 gene cluster seems to produce a SM which acts as an attenuator of disease, because the deletion of the PKS51 gene caused early withering of infected seedlings. Examination of more field isolates from Korean rice fields revealed a correlation between the pathotype and the ability to produce either fumonisins or GAs which is supported by the clear separation into two distinct phylogenetic clades.
The fungal strains used in this work and their origin are shown in Table 1. Strain IMI 58289 was derived from Commonwealth Mycological Institute, Kew, United Kingdom. Strains m567 and FSU48 were provided by the Fungal Stock Center at the University Jena, Germany, C1995 by J.F. Leslie, Kansas State University, E282 by S. Tonti, University Bologna, Italy, MRC2276 by W.C.A. Gelderblom, South Africa, and NCIM 1100 was provided by the National Collection of Industrial Microorganisms, India. Strains B14 and B20 were provided by S.-H. Yun, Korea. Strain V64-1 was kindly provided by T. Kyndt from the University Ghent, Belgium.
Escherichia coli strain Top10 F’ (Invitrogen, Groningen, The Netherlands) was used for plasmid propagation. The uracil-auxotrophic Saccharomyces cerevisiae FGSC 9721 (FY 834) was provided by the Fungal Genetics Stock Center (Kansas State University) and used for yeast recombination cloning.
F. fujikuroi B20: Illumina TrueSeq genome sequencing by TheragenEtex, Suwon, Korea. The assembly was performed using Celera Assembler version 7.0 [59], ‘overlap minimum length’ set to 150 bases. The assembly resulted in 318 scaffolds with a 21-fold coverage of the TrueSeq large reads.
For all other strains sequencing was carried out by shot gun sequencing of an 8 kb library with paired end 100 bp read length using Illumina HiSeq 2000 by Eurofins MWG Operon, Germany. The assemblies were performed by ALLPATHS-LG [60] and the scaffolds were error corrected by mapping all Illumina shotgun paired-end data and further scaffolded using SSPACE [61] (Table 2). The data on the new genomes, including annotation, was submitted to the European Nucleotide Archive, study accession PRJEB14872 available at: http://www.ebi.ac.uk/ena/data/view/PRJEB14872. Sample accession numbers are listed in Table 1. RNA-seq data are available at: https://www.ncbi.nlm.nih.gov/gds/?term=GSE89480.
Draft gene models for all genomes were generated by three de novo prediction programs: 1) Fgenesh [62] with different matrices (trained on Aspergillus nidulans, Neurospora crassa and a mixed matrix based on different species); 2) GeneMark-ES [63] and 3) Augustus [64] with Fusarium ESTs and RNA-seq based transcripts as training sets. Annotation was aided by exonerate [65] hits of protein sequences from F. fujikuroi IMI 58289 and F. oxysporum 4287 to uncover gene annotation gaps and to validate de novo predictions. Transcripts were assembled on the RNA-seq data sets using Trinity [66]. The different gene structures and evidences (exonerate mapping, RNA-seq reads and transcripts) were visualized in GBrowse [67] allowing manual validation of coding sequences with a focus on SM cluster genes and other genes of interest. The best fitting model per locus was selected manually and gene structures were adjusted by splitting or fusion of gene models and redefining exon-intron boundaries if necessary. tRNAs were predicted using tRNAscan-SE [68]. The predicted protein sets were searched for highly conserved single (low) copy genes to assess the completeness of the genomic sequences and gene predictions. Orthologous genes to all 246 single copy genes were searched for all proteomes by BLASTp comparisons (eVal: 10−3) against the single-copy families from all 21 species available from the FunyBASE [21]. Additionally, the proteomes were searched for the 248 core-genes commonly present in higher eukaryotes (CEGs) by BLASTp comparisons (eVal: 10−3) [20]. We also used BUSCO Version 3.0.1 in the Ubuntu virtual machine with the lineage specific profile library Sordariomyceta_odb9 (3.725 BUSCO groups), downloaded from http://busco.ezlab.org. The analysis was performed in gene set (protein) assessment mode running the python script run_BUSCO.py [22]. All genomes were analyzed using the PEDANT system [69]. To avoid misleading ortholog information based on similarity and bi-directional best hits, QuartetS [70] was applied to retrieve a reliable ortholog matrix which was used for all comparative representations.
The phylogenetic tree of Fusarium species was calculated based on the protein sequences of 5,181 single copy genes that are shared among all analyzed species. Orthologs of the sequences were aligned separately using MAFFT [71]. After that, we concatenated the alignments and removed columns with gaps using Gblocks [72]. The evolutionary history was inferred using the Maximum Likelihood method PhyML [73] with default parameters and the amino acid substitution model LG. Branch support was tested using approximate likelihood ratio test (aLRT) based on the Shimodaira-Hasegawa-like (SH-like) procedure [74]. The tree is drawn to scale, with branch lengths measured in the number of substitutions per site.
We calculated single copy genes in clustering proteins of all genomes and selecting clusters with exactly one representative from each genome. Protein clusters were calculated using usearch [75] (e-value cutoff: 0.01) and mcl [76] (inflation value: 2).
For phylogenetic analysis of field isolates of F. fujikuroi in Korea, TEF1α and and RPB2 were amplified from fungal genomic DNAs using the primer sets fRPB2-7cF/fRPB2-11aR [77] and EF1/EF2 [78], respectively (S6 Table). All nucleotide sequences from PCR products were edited with Lasergene (ver. 6.0; DNASTAR, Madison, WI, USA) and aligned using ClustalW [79]. Maximum parsimony (MP), neighbor-joining (NJ), and unweighted pair group method with arithmetic mean (UPGMA) analyses were performed using MEGA (ver. 4.02) with 1,000 bootstrap replications.
Protoplasts from Fusarium strains were prepared as described previously. The protoplast suspension was mixed with 1.2% InCert agarose (Lonza Group AG, Basel, Switzerland) and then loaded on a CHEF gel as described in [80]. Chromosomes of Schizosaccharomyces pombe and S. cerevisiae were used as a molecular size marker (Bio-Rad, Munich, Germany).
To identify SM clusters in each genome, the InterPro scan results of the PEDANT analysis were used as described [81]. Essentially, predicted proteins with homology to a domain of a signature SM core enzyme (e.g. PKS, NRPS, TC or DMATS) were considered a marker for a gene cluster. A cluster was verified if any neighboring genes with homology to typical SM cluster enzymes, like P450 monooxygenases, oxidases, methyltransferases, MFS or ABC transporters or TFs, were identified. The extent of each putative gene cluster was then adjusted by comparison to previously published data and to homologous clusters in other Fusarium species.
For SM production experiments and RNA preparation under in vitro conditions, strains were first cultivated for 3 days in 300 mL Erlenmeyer flasks with 100 mL Darken medium [82] on a rotary shaker at 180 rpm at 28°C. 500 μL of this culture were then used to inoculate 100 mL of ICI (Imperial Chemical Industries, UK) media [83] containing either 6 mM glutamine, 60 mM glutamine, or 6 mM NaNO3 for 3 days (RNA extraction) or 7 days (SM analysis), respectively. For RNA extraction, mycelia were flash-frozen with liquid nitrogen prior to lyophilization. For in planta expression studies by RNA-seq, lyophilized roots of infected rice plants were used for RNA preparation.
For the generation of infected rice samples for transcriptome analysis by RNA-seq, rice seeds of the cultivar GSOR 100, Nipponbare, and Dongjin were used. The seeds of the former two were provided by Genetic Stocks-Oryza (GSOR) Collection, USDA ARS Dale Bumpers National Rice Research Center, Hwy, Arkansas, USA.
Total RNA was extracted from mycelia grown for 3 days in liquid ICI media (containing either 6 or 60 mM glutamine) and from infected rice seedlings after 7 dpi using TRIzol Reagent (Life Technology, Karlsruhe, Germany) and purified using an RNeasy Plant MinElute Cleanup Kit (Qiagen, Hilden, Germany). The quality of DNase-treated RNA (28S:18S > 1.0; RIN≥ 6.5; OD260/280 ≥1.8; OD260/230 ≥ 1.8) was determined using an Agilent Bioanalyzer. The high quality RNA was sent to BGI Tech Solutions Co., Limited (Hong Kong) for library construction and sequencing by Illumina HiSeq2000 technology.
RNA-seq reads were mapped on the reference genome using tophat2 (v2.0.8). The interval for allowed intron lengths was set to min 20 nt and max 1 kb [84–86]. We used cufflinks to determine the abundance of transcripts in FPKM (fragments per kilobase of exon per million fragments mapped) and calculated differentially expressed genes using cuffdiff [85,86]. The gene models were included as raw junctions. Genes with a minimum of fourfold increase or decrease in expression (|log2 of the FPKM values +1| ≥ 2) between the two experimental conditions were considered as regulated. The RNA-seq data has been deposited in NCBI's Gene Expression Omnibus [87] and are accessible through GEO Series accession number GSE89480.
The DNA constructs for deletion of FUM1 (FFB14_08440, FFB20_01984), FUB1 (FFB14_01651FFB20_13404), and PKS51 (FFB14_06372) from the genomes of F. fujikuroi strains B14 or B20 were created using a split-marker recombination procedure as previously described [88]. To delete FUM1, the 5' and 3' flanking regions of the FUM1 ORF were amplified with the primer pairs JFUM1f5/JFUM1rt5 and JFUM1ft3/JFUM1r3, respectively (in the first round of PCR), fused to the hygromycin B resistance gene (hygB) cassette, which was amplified from pBCATPH [88] with the primers HygB-for and HygB-rev (in the second round of PCR), and used as a template to generate split markers with the new nested primer sets, JFUM1fn/pUH-BC/H3 and JFUM1rn/pUH-BC/H2, respectively (in the third round of PCR) (S6 Table). Similarly, DNA constructs for targeted deletions of the other genes were created using the strategy described above. For the deletion of FUB1, the primer pairs JFUMB1f5/JFUM1Brt5 and JFUB1ft3/JFUB1r3 were used for amplification of 5' and 3' flanking regions of the FUB1 ORF, respectively, and JFUB1fn/pUH-BC/H3 and JFUB1rn/pUH-BC/H2 were used as the nested primer sets, respectively. For the deletion of PKS51, the primer pairs B14_6372For5/B14_6372rev5t and B14_6372for3t/B14_6372rev3 were used in the first round of PCR, and B14_6372forN/ pUH-BC/H3 and B14_6372revN/ pUH-BC/H2 for the third round of PCR. Additionally, for double deletion of FUM1 and FUB1, we generated a knock-out construct through which the 5′- and 3′-flanking regions of the FUM1 ORF were fused to a geneticin resistance gene cassette (gen) amplified from pII99 using the primer pair Gen-for and Gen-rev, as described above. The resulting construct was transformed into the deletion strain of FUB1.
For the complementation of each deletion mutant of FUM1 or FUB1 derived from B14, intact copies of each gene were amplified from the genome of B14 using the primers JFUMf5/JFUMr3 and JFUB1f5/JFUBr3, respectively, and directly added into protoplasts of each deletion strain along with pSK660 including the geneticin resistance (gen) gene for co-transformation as previously described [89]. All primer sequences are listed in S6 Table.
The effect of the single isolates on rice seedling germination was studied by infecting the seeds with conidia of each isolate. For spore formation, the fungal strains were cultured on PDA plates for 7 days at 28°C under light/dark (12 h/12 h) conditions. The plates were flooded with sterile water to obtain a conidial suspension (1 x 10−6). Seeds were soaked in the suspension for 18 h. Inoculated and non-inoculated (control) seeds were sown into 100 mL plastic pots. Eight days after sowing the number of germinated seeds was assessed. The number of dead, chlorotic and elongated seedlings was measured 15 days after sowing.
For pathogenicity assays, healthy rice seeds were surface-sterilized by submersion in 70% of ethanol followed by 1% of sodium hypochlorite. Sterilized seeds were germinated in Murashige and Skoog (MS) agar [7] at 26°C for 5 days. The Fusarium isolates were first grown on oatmeal agar for one week. The pathogenicity assays were performed as previously described [7]. Agar plugs from the oatmeal plates were placed on top of 3 cm of sterilized vermiculite in a glass tube (18 cm x 1.6 cm). The agar plugs were then covered with 3 cm of vermiculite. Five-day-old seedlings were transferred to the surface of the vermiculite layer to avoid the direct contact between seedlings and fungal inocula. Before covering the tubes with a cap, 4 mL of Yoshida solution was gently added to each test tube to help retain high humidity [90] and placed at 26°C for 3, 5, 7 or 9 days. Their heights and internode lengths were measured and photographs of the seedlings and infected roots were taken. The symptom development caused by wild-type and mutant strains was examined in five independent pathogenicity tests. For pathotests with exogenous supply of culture filtrates, the wild-type B14 strain or its mutant lacking FUM1 and FUB1 (Δfum1/Δfub1) was inoculated into 50 mL of PDB (potato dextrose broth) and incubated for 5 days. The fungal liquid cultures were filtered through 2 × cheesecloth followed by filtration through 0.25 μm membranes. The culture fluid was dried to 5 ml by lyophilization (10-fold concentration). For inoculation assay, 500 μl of the concentrated culture filtrate was exogenously supplied to a single rice seedling.
F. fujikuroi B14 and B20 strains were transformed as previously described for F. graminearum [91]. Vector integration events were confirmed by diagnostic PCR (S10 Fig) using specific primers as indicated (S6 Table).
PCR mixtures contained 25 ng of template DNA, 50 ng of each primer (S6 Table), 0.2 mM deoxynucleoside triphosphates, and 1 U of Biotherm Taq polymerase (Genecraft, Lüdinghausen, Germany). The cDNA synthesis was performed using Superscript II (Invitrogen, Groningen, The Netherlands) and 1.5 μg of total RNA as the template, according to the manufacturer's instructions. The qPCR was performed using iTaq Universal SYBR Green Supermix (BioRad) and Superscript II cDNA as template, in a Biorad thermocycler iTaq. In all cases, the qPCR efficiency was between 90–110% and the annealing temperature was 60°C. Gene expression was measured in three biological replicates from each time point, and the relative expression levels were calculated using the ΔΔCt method [92]. The expression of a translation elongation factor α gene (EF1A), amplified by a primer pair (EF1-PS1 and EF1-PS2) (S6 Table), was used as an endogenous reference for data normalization.
For analyses of the SMs, the strains were grown in submerged cultures as described above. After 7 days, mycelia were removed from the culture by filtration through Mirachloth (Calbiochem, Merck KGaA, Darmstadt, Germany). The culture filtrates were filter-sterilized using 0.45 μm syringe filters (BGB, Schloßböckelheim, Germany).
Fusaric acid and beauvericin were obtained from Sigma-Aldrich (Deisenhofen, Germany), GAs from Serva (Heidelberg, Germany) and methylparaben (MePa) was obtained from Fluka (Steinheim, Neu-Ulm, Germany) in analytical grade. The remaining standard substances were obtained as described in previous work [24,25,34,35,42,93–95]. All solvents were obtained in gradient or analytical grade from Sigma-Aldrich, VWR (Darmstadt, Germany) or Merck Schuchardt (Hohenbrunn, Germany). Water was purified by a Milli-Q Gradient A 10 system (Millipore, Schwalbach, Germany).
Liquid culture samples were prepared as following: 10 μL of the culture filtrate and 10 μL of MePa (100 μg/mL) as internal standard were added to 80 μL of water. For in planta analysis, ten rice samples were combined and freeze-dried. The dried samples were treated with liquid nitrogen and pestled simultaneously, larger pieces were cut first with a scalpel. The samples were weighed and extracted with 1.5 mL of the following mixture: ethyl acetate:methanol (MeOH):dichloromethane, 3:2:1. Precellys ceramic beads 1.4/2.8 (Peqlab, Erlangen, Germany) were added to the samples, and the mixture was vortexed for 1 min. Afterwards, the samples were shaken for 1 h on a rotary shaker with 150 rpm. After a short centrifugation step (3 min, 2900 g), 500 μL of the supernatant were transferred to a new vial and the solvent was evaporated to dryness under constant nitrogen flow. The residue was dissolved in 100 μL MeOH and put to an ultrasonic bath for 10 min. After vortexing the samples, they were centrifuged again with 5000 g, and 50 μL of the supernatant were collected. Afterwards, the samples were dried again under nitrogen flow and 1.5 mL of MeOH/water, 3/1 (v/v), + 0.1% formic acid (FA) were added. The extraction procedure described above was repeated. 10 μL of each extract were mixed with 10 μL MePa and 80 μL water for analysis. Some extracts were diluted again 1:10; the corresponding values are labelled.
The liquid culture samples were analyzed as following: A Shimadzu LC-20AD HPLC (Shimadzu, Kyoto, Japan) with a SIL-20ACXR autosampler coupled to a Sciex QTRAP 5500 (Sciex, Darmstadt, Germany) mass spectrometer was used. Separation was carried out on a Synergi Hydro-RP column from Phenomenex (Aschaffenburg, Germany) with 50 × 2.0 mm and 2.5 μm particle size, protected by a KrudKatcher classic filter (Phenomenex), and on a Nucleodur C18 Gravity-SB column from Macherey-Nagel (Düren, Germany) with 50 × 2.0 mm and 3 μm particle size, protected by a KrudKatcher classic filter (Phenomenex), at 50°C. MeOH + 1% FA + 5 mM NH4Ac was used as eluent A, water + 1% FA + 5 mM NH4Ac was used as eluent B. A flow volume of 450 μL/min was applied, and the gradient started at 5% A. This condition was held for 1.5 min. The gradient rose up to 98% A in 10.5 min, and finally the column was rinsed for 3 min with 98%. After that, the column was equilibrated for 2.5 min with 5% A. The integrated valco valve was used, discarding the first 2 min of the run, and the samples were cooled to 7°C. 5 μL of each sample was injected.
Advanced scheduled multiple reaction monitoring (MRM) mode was used for acquisition. Both positive and negative ionization mode were applied. The curtain gas (CUR) was set to 35 psi, the collision gas was set to medium. The temperature of the heater gas (TEM) in the ion source was set to 450°C. Nebulizer gas (GS1) and auxiliary gas (GS2) were adjusted to 35 and 45 psi, respectively.
In the positive MRM mode, the ion spray voltage (IS) was set to 4500 V, and the target scan time of this experiment was adjusted to 0.3 s, resulting in a cycle time of 0.5 s. The positive ionization mode was applied for the relative quantification of O-methyl-fusarubin, fusaric acid, gibepyrone A, apicidin F, beauvericin, fumonisins (FB1+FB2), fusarins, trichosetin and fujikurin A. The cell exit potential (CXP) was set to 11 V, the entrance potential (EP) was set to 10 V. For the negative MRM mode, the IS was set to -4500 V and the target scan time was adjusted to 0.2 s, resulting in a cycle time of 0.5 s. The negative ionization mode was applied for the relative quantification of MePa, and the gibberellins GA3, GA4 and GA7. The CXP was set to 11 V, the EP was set to 10 V.
MRM transition for the quantification were as follows: FB1−722.4–223.1; FB2−706.4–336.3; GA3−345.2–143.0; GA4−331.3–243.1; GA7−331.3–243.1. The calibration curve for all the standards was prepared in a concentration range of 0.0001–1 μg/ml.
For bikaverin determination in liquid culture samples, HPLC-UV measurements were carried out on a Shimadzu LC-20AT pump system with a Shimadzu SIL autosampler and a photodiode array (PDA). A Gemini 5 u C6-Phenyl 110A, 250 × 4.60 mm, 5 μm column (Phenomenex) was used, with water + 1% FA as eluent A and acetonitril + 1% FA as eluent B. The column oven was set to 40°C. The gradient started with 10% B with 1.35 μL/min. After 3 min, the gradient rose up to 100% B during 17 min. The column was washed with 100% B for 6 min, and afterwards the column was equilibrated with 10% B again for 4 min. The wavelength for PDA analysis ranged from 220–600 nm. 100 μL of the sterile culture filtrate were injected. Peak areas were determined at 508 nm, and bikaverin obtained from Sigma Aldrich was used as standard substance.
In planta sample analysis was performed with a different HPLC-system but the same mass spectrum. An Agilent 1260 HPLC system (Santa Clara, USA) was used, and the expected retention time for scheduled MRM analysis needed to be adjusted (S7–S10 Tables). Furthermore, bikaverin was analyzed in positive ionization mode.
|
10.1371/journal.pgen.0030106 | Sex-Specific Crossover Distributions and Variations in Interference Level along Arabidopsis thaliana Chromosome 4 | In many species, sex-related differences in crossover (CO) rates have been described at chromosomal and regional levels. In this study, we determined the CO distribution along the entire Arabidopsis thaliana Chromosome 4 (18 Mb) in male and female meiosis, using high density genetic maps built on large backcross populations (44 markers, >1,300 plants). We observed dramatic differences between male and female map lengths that were calculated as 88 cM and 52 cM, respectively. This difference is remarkably parallel to that between the total synaptonemal complex lengths measured in male and female meiocytes by immunolabeling of ZYP1 (a component of the synaptonemal complex). Moreover, CO landscapes were clearly different: in particular, at both ends of the map, male CO rates were higher (up to 4-fold the mean value), whereas female CO rates were equal or even below the chromosomal average. This unique material gave us the opportunity to perform a detailed analysis of CO interference on Chromosome 4 in male and female meiosis. The number of COs per chromosome and the distances between them clearly departs from randomness. Strikingly, the interference level (measured by coincidence) varied significantly along the chromosome in male meiosis and was correlated to the physical distance between COs. The significance of this finding on the relevance of current CO interference models is discussed.
| Meiotic crossovers between homologous chromosomes ensure their proper segregation to generate ultimately gametes. They also create new allelic combinations which contribute to the diversity of traits among individuals. In all eukaryotes, the number and the localization of crossovers along chromosomes are not random. In addition, crossovers are not independent of each other: the occurrence of a crossover lowers the probability that another crossover arises in its vicinity. The mechanism of this phenomenon, called “crossover interference,” is one of the most challenging puzzles that geneticists have been faced with in the last century. In this paper, we precisely described the distribution of crossovers along Chromosome 4 of the model plant species Arabidopsis thaliana, separately in male and female meiosis. Interestingly, we observed that crossovers are 1.7 more numerous in male than in female meiosis, and this increase is especially marked at the ends of the chromosome. Moreover, our results provide the first evidence that the level of interference along a chromosome is not a constant and is correlated with the physical distance between crossovers. These results shed new light on the determinism of crossover localization and could have important outcomes on the relevance of current models of crossover interference.
| One prominent feature of the eukaryotic life cycle is the segregation of homologous chromosomes to two different cells during the first, also known as reductional, meiotic division. The proper completion of this segregation relies on the formation of stable physical connections between homologous chromosomes. In most eukaryotic species, these connections are mediated by crossovers (COs). These are sites where large (megabase scale) segments of homologous (nonsister) chromatids are exchanged. Consequently, COs are essential to the ploidy reduction process, as well as to play a role in the creation of allelic combinations.
CO number and distribution along chromosomes differ between male and female meiosis in many plant and animal taxa (for review see [1]). This widespread phenomenon is called heterochiasmy. Both the direction and magnitude of these differences are highly variable. For example, depending on the species, CO number may be higher in female (F) meiosis (most eutherian mammals), or male (M) meiosis (some metatherian mammals), or there may be no significant difference between sexes (goat, dog, barley). This difference may be small or moderate, but sometimes it is huge (e.g., teleostean fishes). Even closely related species can exhibit different M/F CO ratios. In the Brassicaceae, for example, this ratio reaches 1.2 in Sinapis alba [2], whereas in Brassica oleracea it is inversed (0.6) [3], and there is no significant difference in Brassica napus (0.98) [4]. Therefore, the nature of evolutionary forces driving heterochiasmy is a puzzling issue. In addition, the underlying molecular and cellular mechanisms are currently unknown. Yet sex-related differences in CO number per chromosome are paralleled by sex-related differences in the length of synaptonemal complex (SC) in human [5] and mouse [6]. The SC is a proteic structure scaffolded along synapsed homologous chromosomes at pachytene stage [7].
COs can be localized along chromosomes by analyzing genetic recombination data. They can also be visualized cytologically either as chiasma, or as immunolabeled MLH1 foci that mark most CO sites [8], or as late recombination nodules [9], which are electron-dense structures located on SCs [9–11]. COs originate from programmed double-strand breaks (DSBs) that occur early in prophase of the first meiotic division [12]. Only a part of these DSBs give rise to COs; the remaining DSBs are repaired as “noncrossovers” (NCOs), without exchange of large DNA segments between homologous chromosomes.
Numerous studies showed that CO formation is tightly controlled at both chromosomal and local scales [13,14]. Indeed, COs are not uniformly distributed and inter-CO distances are not random. The former feature is well illustrated by numerous datasets in mammals [15–17] and higher plants [11,18]. Several studies have tried to correlate CO rates along chromosomes with various sequence features, such as gene or transposable element density, GC nucleotides %, CpG ratio, simple repeats, etc. However, even if some weak correlations were found, it seems that none holds true in all species [15,16,19–21], suggesting that other constraints act on CO distribution.
One of these constraints is CO interference. This phenomenon was originally described as a lower frequency of double-COs in disjoint chromosomal segments than expected if they occur independently of each other [22]. The existence of interference has been confirmed in most species tested [10]. As a consequence of interference, COs tend to be more evenly spaced than expected if CO positions were random [23]. In addition, in many species, which have a limited number of COs per chromosome, interference tends to increase physical distances between adjacent COs. This is well illustrated by recombination nodules or MLH1 foci maps produced in various species [5,17,24–26].
The mechanisms of interference setup are still poorly understood. Several models of meiotic CO interference have been proposed over years (see [14] for a comprehensive review). The two main contenders are currently the “counting” model [27] and the mechanical stress model [28]. The basic postulate of the counting model is that the CO designation process among recombination precursors occurs in such a way that any two adjacent COs are separated by a fixed number of NCOs. Alternatively, the mechanical stress model hypothesizes that COs originate from a mechanical stress imposed on the chromosome. CO designation would promote a stress relief that would (i) inhibit CO designation among nearby recombination intermediates and (ii) attenuate in a distance-dependent manner. Neither of these two models is presently strongly supported by experimental data.
In a previous study, we produced a high resolution map (at around the 210-kb scale) of meiotic crossovers on Arabidopsis thaliana Chromosome 4 [18]. We showed that CO rates vary greatly along the chromosome from 0 to 20 cM/Mb, and that COs displayed interference. However, CO rates on this map were sex-averaged because we used the selfed progeny of F1 hybrids for the mapping population. Given that the existence of heterochiasmy in A. thaliana had been previously suggested by several studies [29–32], we decided to investigate the relative contributions of male and female meiosis in the distribution of COs on Chromosome 4. We observed dramatic differences between male and female genetic maps. Strikingly, we found a good correlation between the sex-ratio of mean CO number per Chromosome 4 on one hand and the sex-ratio of total SC length on the other hand. Moreover, we were able to detect significant variations in interference strength along Chromosome 4. Stunningly, it turned out that interference strength covaries with the physical distance between COs. These results could have important upshots on the reliability of current interference models.
A. thaliana “Columbia” (Col) and “Landsberg erecta” (Ler) accessions were crossed to obtain F1 hybrids. Col plants were then crossed with an F1 hybrid used either as the male (Col × (Col × Ler)) or as the female ((Col × Ler) × Col) parent. Seeds from these crosses were sowed in vitro, and then seedlings were grown in short-day conditions at 21 °C.
After 2 wk, 1,476 whole seedlings of each population were collected and their DNA was extracted as described previously [33].
In a previous experiment, F2 plants from a Col × Ler cross were genotyped with a set of 70 SNP markers spanning A. thaliana Chromosome 4 [18]. In the present study, 46 SNPs out of these 70 and two additional SNPs were chosen so that the mean sex-averaged genetic distance between adjacent markers was 1.9 cM. SNPs are listed in Table S1.
Genotyping was performed using SNPlex technology (Applied Biosystems, http://www.appliedbiosystems.com) following the supplier protocols. After quality scoring of genotyping data, four markers were dismissed from the whole dataset. In some cases it was not possible to assess the genotype of remaining markers in some plants so these were also removed from the dataset. The resulting populations comprised 1,305 and 1,419 plants for female and male meiosis, respectively.
The genetic size of intervals was computed as the ratio between the number of recombined chromosomes and the number of analyzed meioses, which in the case of a backcross progeny is equal to the number of analyzed plants. CO rates, physical and genetic sizes are listed in Table S2.
In order to calculate single-interval, sex-averaged CO rates in the pool of M and F populations, the sex-specific CO rates were weighted according to the respective population size.
All pair-wise comparisons between CO rates were performed using a chi-square homogeneity test. For multiple testing, p-values were subsequently corrected using the false discovery rate procedure [34].
Predicted Poisson distributions of CO number per chromosome were calculated using the following formula:
where S(k) is the number of chromosomes harboring exactly k CO, e is the neperian logarithm base, m is the observed mean number of CO per chromosome, and N is the total number of chromosomes.
Whole comparisons between observed and Poisson distributions were performed using a chi-square goodness-of-fit test.
For both M and F datasets, the genetic “width” of inter-CO distance classes was chosen to be 17.5 cM (± 5%) in order to: (i) provide distance classes spanning the whole chromosome genetic length, (ii) ensure a common denominator in both M and F maps, and (iii) prevent small class sizes, in order to maintain moderate sampling variances, thus allowing conclusive statistical testing.
The continuous probability distribution function of inter-CO distances on chromosomes with exactly a independent COs is:
, where L is the genetic size of the chromosome and d is the distance between successive COs. The derivation of this formula is as follows:
a independent CO points are randomly placed on a chromosome of length L. Then a CO point is added at one end to bring the chromosome to the shape of a ring. a + 1 points are thus randomly and independently positioned on the circle of perimeter L. The statistics of distances between successive COs is the same for all pairs; to compute this for the first pair, we need to find the distribution of the smallest of a random variables, representing the positions of COs along the interval, which are uniformly distributed in [0, L] (the remaining point is by definition at position zero). The probability that this smallest value, which is the distance between the last and the first CO, is greater or equal to X is (1 − X/L)a. The minus derivative of this cumulated distribution then gives the desired probability distribution.
The following formula, which is easily deduced from the formula above, allows convenient calculation of discrete distributions of inter-CO distances on finite-size chromosomes with exactly two independent COs:
, where n is the number of classes, k is the rank of the class (increasing with distance), N is the population size, and S(k) is the size of the kth′ class.
Whole comparisons between observed and calculated distributions were performed using a chi-square goodness-of-fit test.
For both M and F datasets, the genetic size of intervals used for coincidence analyses was chosen to be the same as the genetic “width” of distance classes used for inter-CO distance comparisons, for the same reasons (see above). Given two intervals, the coefficient of coincidence between them is calculated as follows:
where C is the coincidence and rij is the chance of i CO across the first interval and j CO across the second interval. In most cases i and j values are either 0 or 1; 2 COs were rarely found in one of the intervals and were considered as no CO, while 3 COs were considered as 1 only, accordingly to what would have been observed if the intervals would have not contained internal markers.
The standard deviation of coincidence was calculated according to [35].
We have developed a procedure which computes the p-value for the hypothesis H0 that two coefficients of coincidence cu and cv estimated from quadruplets or triplets of markers are in fact generated from the same theoretical coincidence value cth. Under that hypothesis, H0, cu, and cv are actually not expected to differ from each other. A small p-value for the difference between cu and cv then indicates that H0 is unlikely to be true given the genotype data of the mapping population.
Let a quadruplet have markers A, B, C, and D, assumed to be in the order in which they appear on the chromosome. If the quadruplet is instead a triplet, this formalism can be applied by setting B = C.
In a first phase, we compute cth by the maximum likelihood method. Consider the first quadruplet: the probability (likelihood) that N gametes lead to a measured coincidence value of cu is
where nnn, nrn, nnr, nrr are the number of gametes that are respectively recombinant between (i) neither A and B nor C and D, (ii) A and B but not C and D, (iii) C and D but not A and B, (iv) both A and B and C and D. N is the total number of gametes with valid data at the four markers, namely nnn+ nrn+ nnr+ nrr. The dependence on cth is through the probabilities:
where rAB and rCD are recombination fractions between A and B and B and C, respectively.
Next, we consider the two quadruplets of interest. L(cv) is calculated as for L(cu). The joint likelihood of both observations is the product L(cu) × L(cv), and we numerically determine the cth which maximizes this joint likelihood. The result is a cth lying somewhere between cu and cv.
In a second phase, we compute a p-value for the hypothesis H0 given cth. We do this by determining the probability that | cu − cv | is at least as large as measured from the experimental data. But if cu and cv are estimated using shared gametes, the recombination events in the four intervals are a priori correlated. Thus, when measuring cu and cv we need to use independent sets of gametes by using half (N/2) of the gametes for cu and the other half for cv. So that the value | cu − cv | is not dependent on the data order, | cu − cv | is computed for 105 random order combinations and the median value taken.
The p-value is obtained by simulating interference events within H0 given cth: we generate N/2 realizations of gametes for each quadruplet; for each realization, we choose among the four possibilities of recombinants or not in each interval according to the probabilities prr, prn, pnr, pnn. For this set of N gametes, we extract the two associated coincidence coefficients cu′ and cv′. Repeating this 105 times, we get a probability distribution for | cu′ − cv′ |; the desired p-value is then the frequency with which | cu′ − cv′ | is larger than the experimental value.
Cytological observations were carried out on Col × Ler F1 plants.
The anti-ASY1 polyclonal antibody has been described elsewhere [36]. It was used at a dilution of 1:500. The anti-ZYP1 polyclonal antibody was described by [37]. It was used at a dilution of 1:500.
Preparation of prophase stage spreads for immunocytology was performed according to [36] with the modifications described in [38].
All observations were made using a Leica (http://www.leica.com) DM RXA2 microscope; photographs were taken using a CoolSNAP HQ (Roper, http://www.roperscientific.com) camera driven by Open LAB 4.0.4 software; all images were further processed with Open LAB 4.0.4 or AdobePhotoshop 7.0 (http://www.adobe.com). SC length measurement was performed using Optimas (Bioscan Incorporated, http://www.bioscan.com) software.
Plants from a Col × Ler F1 population were backcrossed with Col plants using the F1, either as the male or the female parent, in order to create two populations subsequently referred to as M and F, in which the observed recombination events occurred either in male or female meiosis of the parental F1 hybrid. 1,419 M plants and 1,305 F plants were genotyped with 44 SNP markers spanning Chromosome 4 at a density of 1.9 cM (calculated from sex-averaged data, see Materials and Methods). Given that interval sizes are small, we calculated genetic distances simply by dividing the number of recombinant chromosomes by the total number of plants analyzed.
We first compared CO rates in the F2 population previously described to those in pooled M and F populations, in each of the same 43 intervals spanning Chromosome 4 (Figure 1A). As expected, the “averaged” (see Materials and Methods) CO rates observed in the pool of the M and F backcross progenies (corresponding respectively to male and female meiosis) were not significantly different from those observed in the F2 progeny (resulting half from male meiosis, half from female meiosis) generated from the same parental accessions (lowest p-value is 0.35; Figure 1A).
This implies that there is no significant variation in meiotic recombination over time for a given genetic background, thus enabling direct comparisons of data.
At first glance, the difference between male and female recombination rates is obvious when comparing total genetic size of both maps (Figure 1B). The M map is 87.9 cM long and the F map is 52.3 cM long. This indicates that a Chromosome 4 bivalent experiences on average 1.76 CO in male meiosis, but only 1.05 CO in female meiosis (M/F ratio 1.68). This M/F difference is highly significant (χ2 p-value < 0.001)
Next, we compared recombination rates in male and female meiosis interval-by-interval (Figure 1C). For a majority of intervals (36/43) the M/F ratio was above 1, with the most notable differences in the last telomeric third of the long arm. However, only the distal interval on the short arm and the five distal intervals on the long arm were highly significantly different in male and female (mean M/F ratio for these six intervals is 6.1, χ2 p-value < 0.05). The remaining central intervals were not significantly different in male and female, when compared one-by-one. However, if these were grouped and considered as a single interval, there was still a significant difference between male and female (M/F ratio 1.37, χ2 p-value < 0.001).
In summary, male and female meiotic CO landscapes along Chromosome 4 are strikingly different. The difference is high close to the telomere on the long arm and to the nucleolar organizer region on the short arm and modest in the median region of the chromosome (see Figure 1B and 1C).
Meiotic chromosomes at pachytene stage were immunolabeled with antibodies against ZYP1. This protein is a major component of the central element of the SC, which ties homologous chromosomes together. We used ASY1 immunolabeling to visualize the axial element, which is a proteinaceous axis formed along pairs of sister chromatids [39] (Figure 2). At pachytene stage, ZYP1 labeling extends continuously along the entire SC, hence allowing total SC length measurement. We found that SC length in male meiocytes is 166 ± 24 μm (n = 22) compared to only 98 ± 20 μm (n = 25) in female. Our estimate of male SC length is in good agreement with that obtained in a previous study (147 ± 28 μm; n = 19) using electron microscopy [40]. The value we obtained for the M/F ratio of total SC lengths is very close to that for the M/F ratio of mean CO numbers per chromosome (1.70 versus 1.76). This suggests that sex-related differences in CO number and total SC length are correlated.
We next looked at the distribution of CO number per Chromosome 4 in the M and F populations (Figure 3). According to the hypothesis that CO placements are random and independent events, the distribution of CO number per chromosome should fit a Poisson distribution. Thus, we calculated the Poisson distributions expected for the observed average number of COs per Chromosome 4 and compared these to the observed ones (see Figure 3A and 3B).
In the F population, about half of chromosomes had no CO or only one CO, while very few had two COs or more (Figure 3B). This distribution is highly significantly different (p-value < 0.001) from the theoretical Poisson distribution, in which the “0 CO” group was the main class (59%) and multiple CO classes accounted for 10%. Hence, in female meiosis almost all bivalents experienced only the “obligate CO” required for the proper segregation of homologous chromosomes at anaphase I.
In the M population, only one third of chromosomes had no CO and about half had one CO (Figure 3A). Consequently, chromosomes with multiple COs were more frequent than in the F population. Conversely, in the corresponding Poisson distribution “0 CO” and “1 CO” chromosomes were represented at 42% and 36%, respectively. Observed and expected distributions were clearly different from each other (p-value < 0.001).
As a consequence of interference, inter-CO distances are less variable and greater (when CO number is limited) than expected under the assumption that COs are distributed randomly and independently. Positions of double-COs (on chromosomes with exactly two COs) were represented in two-dimensional plots in Figure 4. x and y axis coordinates correspond to the positions of the first and second CO on the genetic map, respectively. Under the assumption of no interference, points should be uniformly distributed over the triangle. For both M and F double-CO populations, the observed points were clearly heterogeneously distributed: they were underrepresented next to the diagonal line, which corresponds to low inter-CO distances.
In order to test this deviation from independence between COs, inter-CO distances on chromosomes with two COs only (see Materials and Methods) were grouped into size classes, and the observed distributions were compared to the “random” (no CO interference) distributions (Figure 5). For both M and F datasets, the genetic length (17.5 cM ± 5%) of the intervals was chosen to optimize the number of double-COs per interval, in order to avoid high sampling variance and thus allow statistically significant differences to be detected.
In the F population, we found opposing observed and expected distributions: for the expected distribution the minor class was 35–52.5 cM and the major class 0–17.5 cM, whereas in the observed distribution the majority of inter-CO distances were long, and short distances were the minority (Figure 5A). This difference was highly significant (p-value < 0.001).
The observed M distribution was rather symmetrical, with the mode between 35 and 53 cM. It was strikingly different from the theoretical distribution, in which the class size decreased with increasing genetic length (p-value < 0.001; Figure 5B).
For both M and F distributions the mean observed inter-CO distance, respectively 51% and 63% of the total map size, exceeded the expected one, which is exactly one third of the total map size.
Therefore, in male and female meiosis, widely spaced COs were overrepresented, whereas closely spaced COs were underrepresented. This difference between expected and observed distributions of distances between COs is fully consistent with interference.
Besides altered inter-CO distances, another expected consequence of interference is a lowered chance of finding close double-COs than expected from randomness. More precisely speaking, given two intervals, double-COs (one CO in each interval) will occur at a lower frequency than two independent COs (one CO in the first or in the second interval, both being not exclusive). This departure is called coincidence and can be calculated as follows:
where rij is the chance of i CO across the first interval and j CO across the second interval. The value of C is 1 if there is no interference and 0 if interference is absolute (meaning that double-COs are completely absent). Coincidence is widely used as a measure of interference from genetic data. Moreover, most mathematical models of CO interference assume a covariation between coincidence at a given genetic distance and the level of interference (see for example [27], reviewed in [14] ).
Hence, plotting coincidence for pairs of adjacent intervals (three-point coincidence: C3; [27]) all along a chromosome gives access to local variations of interference level, provided that the genetic size of intervals remains constant. We thus performed all coincidence analyses on every possible pairs of 17.5 cM (± 5%) adjacent intervals (30 and 21 pairs fit these requirements in M and F datasets, respectively). This means that we “moved” a 2 × 17.5-cM window along the genetic map. The interference level measured by C3 was clearly variable across Chromosome 4 for both maps. In male meiosis, starting from the short-arm end, interference strength was high until ∼30 cM (C3 < 0.1), then it decreased from ∼30 cM to ∼45 cM (C3 ∼0.3), to reach a minimum at ∼52 cM (C3 ∼0.75), and finally increased again from ∼65 cM to the end of the map (C3 ∼0.3; Figure 6A). Most of these variations in C3 were found to be significant (see Figure 6, Table 1, and Materials and Methods). We can thus conclude that local interference level varied significantly along Chromosome 4 in male meiosis. In the F plot, all observed C3 values are very low (≤0.1). We could not observe any significant variation in coincidence among the few points of the plot (Figure 6 and Table 1). Given the very small number of double-COs, it seems likely that many more plants would be needed to detect any possible coincidence variation.
Another way of analyzing interference is to plot coincidence between one fixed interval and a series of increasingly distant intervals (four-point coincidence: C4; [27]). This means that we fixed a 17.5 cM (± 5%) window and moved a second 17.5 cM (± 5%) window all along the chromosome. This method provides a global description of interference depending on genetic distance (see Materials and Methods). For each M and F population, two different C4 plots were made, using either the terminal interval of the short arm (Figures 7A and 8A) or the telomeric interval of the long arm as the fixed interval (Figures 7B and 8B). For the M population, regardless of whether the fixed interval was located at the end of the long or short arm, plots had globally the same shape. This kind of shape has been consistently observed in various species (for review see [41]), showing that interference decreases with genetic distance. Nevertheless, C4 values from the short-arm end were systematically lower (from 0.05 to 1.11) than those from the long-arm end (from 0.3 to 1.23). This confirms that the strength of interference is weaker at the distal region of the long arm than on the short arm. In both plots, coincidence increased up to 1 at ∼45–50 cM, peaked above 1 at ∼55–60 cM, and then decreased toward 1. The shape of the C4 plots was determined by two factors: (i) the genetic distance between intervals and (ii) fluctuations of interference strength as described above. The occurrence of a peak of coincidence above 1 means that at some genetic distances, double-COs are more frequent than expected if COs were randomly placed: this corresponds to what is called negative interference. At short distances from the terminal short-arm interval (which includes the centromere) coincidence was very low. Interestingly, this shows that the presence of the centromere did not block interference.
Regarding the F population (Figure 8), because the F map was very short (52.4 cM) and double-COs were rare, C4 plots were less informative. From both ends, coincidence increased up to ∼0.4 at ∼35 cM. Strikingly, C never reached 1, showing that in female meiosis interference acted across the whole chromosome.
Given that significant variations of interference level along Chromosome 4 were detected in male meiosis, we addressed the issue of possible correlations between interference level and physical distance. We thus calculated the physical size of the pairs of intervals considered for the C3 coincidence analysis described above, which have all the same genetic size (2 × 17.5 cM). The coincidence values were next plotted against these sizes (see Figure 9A). We could clearly observe two scatters of points. The smallest one contains all pairs of intervals encompassing the heterochromatic knob located on the small arm and the centromere, while the largest scatter comprises all the pairs of intervals located on the long arm only. For the large scatter, we could note a striking positive correlation between the C3 coincidence and the physical size (r2 = 0.91 ± 0.04).
The small scatter was shifted by about 4.5 Mb relative to the large scatter. This is a direct consequence of the presence of a large CO-free region, which does not contribute to the genetic size of the considered pairs of intervals, but hugely increases their physical size. Indeed, when subtracting the cumulated size of the knob and the centromere (4.7 Mb) from the size of the pairs of intervals encompassing these chromosome parts and making a new plot (see Figure 9B), the two scatters of Figure 9A grouped into a single scatter that showed this time a chromosome-wide tight correlation between C3 coincidence and physical size (r2 = 0.93 ± 0.04).
It means that interference level—measured by coincidence analysis on adjacent intervals of constant genetic size—decreases as the physical size increases. In other words, for a given genetic size (i.e., a given CO frequency), a greater physical size enhances the opportunity for double-COs to occur.
In this study, we compared male and female meiotic CO distributions along A. thaliana Chromosome 4. We depicted strong sex-related differences in CO distribution (heterochiasmy) and unequivocal variations in CO interference level along the male chromosome.
One major finding in this study was that the genetic sizes of male and female maps were strikingly different (M map 88 cM, F map 52 cM). This means that Chromosome 4 harbors an average of 1.8 CO in male meiosis, but only 1 in female meiosis. Noticeably, the size of our male map was very close to that previously reported for the Col accession [42] (87.9 versus 83.6 cM). Moreover, we described a marked difference in total SC length between sexes. Remarkably, this difference was very close to the M/F CO ratio observed for Chromosome 4. By using a high density of markers, as well as a large-sized population, we could describe the sex-specific fine scale distribution of COs for the first time in A. thaliana. Thus, we could detect highly significant variations of M/F CO ratio along the chromosome. We observed large differences (up to 18.7-fold) at both ends of the genetic map and less pronounced—though significant—differences on the median part of the chromosome (from 5 Mb to 13 Mb, see Figure 1C). In this median region both curves presented the same overall pattern of “peaks” and “valleys,” even if the local male CO rate was higher than the female CO rate for most intervals (30/37). In summary, we detected heterochiasmy, not only at the whole chromosome level, but also at a regional scale.
Heterochiasmy in A. thaliana has been suggested by several previous studies. Indeed, chiasma counts showed that COs are more frequent in male meiosis (9.7) than in female (8.5) [32], but the result of this count, carried out on only ten meiocytes, was not highly significant and did not provide any precise information about CO location. In two other studies, a comparison of male and female recombination rates on all five A. thaliana chromosomes was performed [29,31], but in each case only a few markers were scored, covering only part of the genome. Nevertheless, these results also suggested that recombination is higher in male than female meiosis and that there are variations in the M/F CO ratio along chromosomes. At the whole genome scale, heterochiasmy is widespread (reviewed in [1]), but this observation remains largely unexplained from an evolutionary point of view. Nevertheless, SC length was shown to differ significantly between male and female meiocytes in human [5] and mouse [6], paralleling variations in MLH1 foci number. Similar results were reported in other species, such as zebrafish [43,44], Dendrocoelum lacteum [45], and others (reviewed in [46]). This statement can now be extended to higher plants, since we obtained comparable results in A. thaliana. Such correlated variations in SC length and CO number were also reported among chromosomes in one sex only, among individuals in the same species, and even among meiocytes in a single organism [5,17,47–50]. However, based on current knowledge it is difficult to claim that SC length determines CO number, if the reverse is true, or even if another unidentified factor determines both.
Besides global differences, there is also compelling evidence that the distribution of COs along chromosomes is contrasted in both sexes. Similarly to what we observe in A. thaliana, enhancement of the M/F ratio close to telomeres was reported in other Brassicaceae species [2,3]. In vertebrates, such a conservation in heterochiasmy patterns along chromosomes was also observed: in mouse, human, and several teleostean fishes, the M/F ratio decreases around the centromeres and tends to increase close to the telomeres [15,16,43,51].
At the present time, the molecular and cellular bases of regional heterochiasmy remain elusive. And, generally speaking, the mechanisms ruling CO distribution along chromosomes are also poorly characterized. Data from various model organisms show that CO distribution results from the integration of several levels of control [14]: (i) the density of meiotic DSBs initiating recombination between homologous chromosomes varies along chromosomes; (ii) the propensity of a DSB to be repaired as a CO or a NCO probably varies; (iii) interference shapes the final CO distribution; (iv) only a part (variable among species) of COs are sensitive to interference (type I CO), the remaining are insensitive (type II CO). Each of these layers of control could act on observed heterochiasmy patterns along chromosomes. The DSB distribution, CO/NCO ratio, interference strength/interference strength variations, and the proportion of type II COs could vary between male and female meiosis, even if no experimental data presently support these hypotheses.
Other factors were also suggested to effect differences between CO distributions in male and female meiosis. Several studies suggested that in human, parental imprinting in a few regions could explain at least part of local heterochiasmy [52,53]. Additionally, it was proposed that synapsis initiation sites colocalize with COs [54]. For example, in human, synapsis initiation occurs in sub-telomeric regions in male [55] whereas it is rather interstitial in female (reviewed in [56]), which seems compatible with the observed pattern of heterochiasmy. Due to the availability of whole genome sequences, correlations between CO rates along chromosomes and various genomic features could be examined [15,16,19–21], but all the resulting correlations were weak. This could be explained by the fact that only sex-averaged recombination rates were used in these studies. A priori, there is no evidence that genomic features correlated to CO rates have the same weight in male and female meiosis. Thus, when possible, correlation analyses should be done separately on data from both sexes. Presumably, this could disclose previously unidentified relationships or reinforce existing ones and reveal differences in correlations between sexes.
Altogether, it is likely that multiple constraints act synergistically to shape CO distribution along male and female chromosomes in meiosis, some of which remain to be elucidated.
In this paper, we have presented the first detailed study on the effect of interference on CO distribution along a whole chromosome in male and female meiosis of A. thaliana. Both CO number per chromosome and inter-CO distances clearly show that COs are not independent of each other. Interestingly, we unequivocally show that the centromere is not a barrier to interference, in accordance with previous reports [18,57–59]. The coincidence plots also clearly show the existence of negative interference at some genetic distance (55–60 cM), which corresponds to a greater chance of another CO than expected from random. This phenomenon has been repeatedly observed from various genetic datasets (see, for instance, [58]) and is also predicted by various models of CO interference [27,28]. Furthermore, we provide unambiguous evidence that interference strength varies significantly along A. thaliana Chromosome 4 in male meiosis.
It has recently been shown that in most eukaryotes, a part of meiotic COs arising from a distinct pathway are not sensitive to interference [60]. Such COs account for about 15% of the total in A. thaliana [37,61]. Thus, the observed variation of interference level, measured on the whole population of COs, can be explained by two nonexclusive hypotheses: (i) the interference level between interfering COs is actually variable, or (ii) this interference level is constant, but the relative proportions of the two kinds of COs are variable along chromosome, so that locally, a high density of noninterfering COs leads to a decrease of the interference level that is measured on the whole population of COs.
In female meiosis, observed variations are not significant because double-COs are rare, hence sampling variance is high, causing an increase in p-values from statistical testing.
Such variations in interference strength along chromosomes were previously suggested from analysis of human pedigree data [62]. Moreover, the level of interference was reported to vary among chromosomes in humans in several studies [17,62–64]. Sex-linked variations of interference have also been reported along human Chromosome 21 [64]. Even fluctuations of interference level among human individuals have been described [17,26].
The molecular bases of these variations are currently poorly documented. Our results provide clear evidence that across a chromosome segment displaying a given CO frequency, a greater physical size enhances the opportunity for double-COs to occur. In other words, interference level between COs separated by a fixed genetic distance is a function of physical distance. Interestingly, cytogenetic data collected in humans demonstrate a negative correlation among chromosomes between SC length and the global (chromosomal) level of interference [17]. At the present time, the molecular bases of these variations are totally unknown.
The mechanisms of interference itself are still elusive. Several models have been proposed, but no experimental data directly support them. One of the most widely used is the “counting” model. Its basic postulate is that a fixed number of NCOs occurs between any two adjacent COs. As a consequence, interference strength is supposed to be constant at the chromosome scale [27,65]. Our data and those cited above strongly argue against such constancy and also call into question the concept of an unchanging “count” itself. Moreover, a recent study in yeast showed that CO number is maintained at the expense of NCOs when the DSB number is reduced, without affecting interference [66]. Other models propose that an interference signal results either from the progressive polymerization of a hypothetical structure along the chromosome [67], or from a mechanical stress imposed on the chromosome axis [28]. Our data are compatible with these two models, in which the interference level is not explicitly intended to be constant and an interference signal propagates along the chromosomes. However, as data in the field of meiotic recombination continue to accumulate exponentially, it is likely that new CO interference models supported by experimental evidences will emerge in the near future.
Our study provides the first detailed analysis of heterochiasmy and CO interference at the whole chromosome scale in a plant species. It provides the basis for future investigations on the determinism of CO distribution at the whole genome scale in A. thaliana and other species. |
10.1371/journal.ppat.1002244 | Haemophilus influenzae Infection Drives IL-17-Mediated Neutrophilic Allergic Airways Disease | A subset of patients with stable asthma has prominent neutrophilic and reduced eosinophilic inflammation, which is associated with attenuated airways hyper-responsiveness (AHR). Haemophilus influenzae has been isolated from the airways of neutrophilic asthmatics; however, the nature of the association between infection and the development of neutrophilic asthma is not understood. Our aim was to investigate the effects of H. influenzae respiratory infection on the development of hallmark features of asthma in a mouse model of allergic airways disease (AAD). BALB/c mice were intraperitoneally sensitized to ovalbumin (OVA) and intranasally challenged with OVA 12–15 days later to induce AAD. Mice were infected with non-typeable H. influenzae during or 10 days after sensitization, and the effects of infection on the development of key features of AAD were assessed on day 16. T-helper 17 cells were enumerated by fluorescent-activated cell sorting and depleted with anti-IL-17 neutralizing antibody. We show that infection in AAD significantly reduced eosinophilic inflammation, OVA-induced IL-5, IL-13 and IFN-γ responses and AHR; however, infection increased airway neutrophil influx in response to OVA challenge. Augmented neutrophilic inflammation correlated with increased IL-17 responses and IL-17 expressing macrophages and neutrophils (early, innate) and T lymphocytes (late, adaptive) in the lung. Significantly, depletion of IL-17 completely abrogated infection-induced neutrophilic inflammation during AAD. In conclusion, H. influenzae infection synergizes with AAD to induce Th17 immune responses that drive the development of neutrophilic and suppress eosinophilic inflammation during AAD. This results in a phenotype that is similar to neutrophilic asthma. Infection-induced neutrophilic inflammation in AAD is mediated by IL-17 responses.
| Approximately 50% of asthmatics have non-eosinophilic inflammation, and 20% of these patients have severe neutrophilic inflammation and increased IL-8 levels. These so-called neutrophilic asthmatics have persistent airway colonization with bacteria, and Haemophilus influenzae is one of the bacteria most commonly isolated. However, how H. influenzae is associated with the pathogenesis of neutrophilic asthma is unknown. In this study we used mouse models to investigate the relationship between H. influenzae infection and allergic airways disease (AAD). We showed that infection promoted the development of hallmark features of neutrophilic asthma. Infection suppressed Th2 cytokines, eosinophilic inflammation, and AHR in AAD, while increasing neutrophilic inflammation and IL-17 responses. Importantly, inhibition of IL-17 during AAD reduced airway neutrophils and neutrophil chemokines, suggesting that infection drives the development of neutrophilic inflammation through an IL-17-mediated mechanism. This provides novel insights into the mechanisms that may underpin infection-induced neutrophilic asthma. These data also suggest that treatments targeting infection may lead to improved management of neutrophilic asthma.
| Asthma is a complex disease of the airways that is generally characterized by symptoms of wheeze, cough, breathlessness and airway inflammation. While eosinophilic inflammation has been considered to be the hallmark of airway inflammation in asthma [1], [2], it is present in only 50% of asthmatics [3]. Non-eosinophilic asthma has now been described in persistent [4], [5] and severe asthma, [6] as well as in steroid naïve asthma [7]. Further investigation of the non-eosinophilic subtype has identified a subgroup with an intense neutrophilic bronchitis [5], [8] with increased interleukin (IL)-8 [4]. Compared to eosinophilic asthmatics, neutrophilic asthmatics have reduced eosinophilic inflammation and AHR. Furthermore, they are frequently resistant to corticosteroid treatment, which results in a significant proportion of asthma-related health care costs [5], [8], [9], [10], [11], [12]. IL-17 is also elevated in asthma and other obstructive airway diseases that are characterized by increased neutrophils [13], [14], [15], [16].
IL-8 and IL-17 are important mediators of neutrophilic inflammation during infection and in disease states [4], [12], [13], [17], [18] and their elevated expression in neutrophilic asthma correlates with increased levels of neutrophils in sputum [15]. IL-8 is a potent neutrophil chemoattractant, produced by macrophages, lymphocytes, epithelial cells and neutrophils [19], [20]. IL-17 is produced by several cells including Th17 cells [21], [22], [23], γδ T cells [24], [25], neutrophils [26], and macrophages [27], [28]. IL-17 has critical roles in host defence against bacterial infections [29], [30], [31], [32], suggesting a potential role in the pathogenesis of bacterial-induced neutrophilic asthma.
Chronic bacterial colonization is evident in the airways of patients with neutrophilic asthma [12] and is also associated with an intense neutrophilic bronchitis in asthma [33]. H. influenzae is a common bacterium of the respiratory tract, is one of the bacteria most frequently isolated from the airways of neutrophilic asthmatics [12], [33], and often causes recurrent respiratory disease [34], [35], [36] in those with compromised airways. Nevertheless, how H. influenzae is associated with the pathogenesis of neutrophilic asthma is unknown. Specifically, whether infection promotes the pathogenesis of neutrophilic asthma, or if neutrophilic asthmatics are predisposed to infection is not known.
In this study we used murine models of H. influenzae infection and OVA-induced allergic airways disease, which mimics hallmark features of human asthma, to elucidate the potential association between infection and the development of neutrophilic asthma.
In order to investigate the association between H. influenzae lung infection and asthma we first established and characterized a murine model of NTHi lung infection alone. Inoculation intratracheally (i.t.) with 5x105 CFU of NTHi resulted in a mild respiratory infection that induced inflammatory responses but did not significantly affect lung function (Figure 1).
Bacterial numbers in bronchoalveolar lavage fluid (BALF) and lung homogenates peaked at 5 days and were cleared 16 days after inoculation (Figure 1A). NTHi infection induced airway inflammation. Neutrophil influx into the airways peaked at 24 hours post-infection while lymphocytes and eosinophils were significantly increased at 5 days. Neutrophil numbers returned to baseline after 5 days, while lymphocyte and eosinophil numbers returned to baseline after 10 days post-infection (Figure 1B). Infection also induced significant but low level increases in NTHi-induced IL-5, IL-13, IL-17 and IL-22, and higher levels of IFN-γ release from mediastinal lymph nodes (MLN) cultures after 5 days, which returned to baseline levels after 26 days (Figure 1C). Infection did not affect lung function, with no changes in AHR (dynamic compliance or transpulmonary resistance in response to increasing doses of methacholine) compared to sham infected (Saline) controls 5, 16 and 26 days after inoculation (Figure 1D–E).
To investigate the effect of infection on AAD in sensitized animals, groups were infected during (d0 NTHi+OVA) or 10 days after (d10 NTHi+OVA) OVA sensitization (Figure 2A), and AAD assessed on day 16. Infection suppressed OVA-induced T cell cytokine responses, inflammatory cell influx and AHR in AAD (Figure 2).
The development of AAD (OVA groups) resulted in increased OVA-induced release of IL-5, IL-13 and IFN-γ from MLN and splenic T cells, eosinophilic inflammation and AHR (decreased compliance and increased resistance in response to methacholine), compared to uninfected, nonallergic (Saline) controls (Figure 2B–F and Figure 3). Infection during (d0 NTHi+OVA) or after (d10 NTHi+OVA) sensitization resulted in significant reductions in OVA-induced IL-5, IL-13 and IFN-γ release from MLN T cells (Figure 2B), compared to uninfected, allergic (OVA) controls. Infection also significantly reduced the numbers of total cells and eosinophils in the airways and blood (Figure 2C-D). The reduction in eosinophils correlated with the reduced release of IL-5 from MLN T cells. Infection significantly suppressed, but did not abolish AHR in AAD. Infection during sensitization (d0 NTHi+OVA) had no effect on compliance, but significantly reduced resistance of the lungs. However, infection after sensitization (d10 NTHi+OVA), significantly suppressed both compliance and resistance (Figure 2E–F). Notably, compliance remained decreased in infected, allergic (NTHi+OVA) groups compared to uninfected, nonallergic (Saline) controls.
Infection during (d0 NTHi+OVA) or after (d10 NTHi+OVA) sensitization had no effect on systemic IL-5 (Figure 3A) but significantly reduced systemic IL-13 and IFN-γ release from splenocytes (Figure 3B–C), compared to uninfected, allergic (OVA) controls.
To determine if Tregs were involved in the suppression of AAD, Tregs, TGF-β and IL-10 were quantified on day 16 of the model. TGF-β and IL-10 are critical immunosuppressive factors that are produced by Tregs. NTHi infection during and after sensitization did not alter the numbers of Tregs (Figure 4A) in the lung, compared to uninfected allergic controls. Notably, infection decreased the expression of TGF-β (Figure 4B) and IL-10 (Figure 4C) in lung tissue in infected, allergic groups compared to uninfected, allergic controls.
To determine if alterations in antigen-presenting cells were involved in the suppression of AAD, the effect of infection on MHCII and CD86 expressing DCs was also investigated (on day 16). The development of AAD resulted in increases in the numbers and proportions of MHCII expressing plasmacytoid DCs (pDCs) and myeloid DCs (mDCs), and CD86 expressing MHCII+ pDCs and mDCs in MLNs and lungs (Figure 5A–H), compared to uninfected, nonallergic controls. Infection during or after sensitization resulted in significant decreases in MHCII and CD86 expressing pDCs and mDCs compared to uninfected, allergic controls (Figure 5A–H).
We then assessed the effects of infection on other features of AAD (on day 16). Significantly, whilst eosinophilic inflammatory responses were suppressed by infection during AAD, NTHi infection induced AAD with an enhanced neutrophilic inflammatory profile.
The development of AAD resulted in an increase in neutrophil influx into the airways (Figure 6). Infection during or after sensitization resulted in a two-fold increase in neutrophil recruitment in BALF compared to uninfected, allergic controls. Moreover, infected allergic groups had a four-fold increase in neutrophil recruitment compared to groups with infection alone (i.e. infected, nonallergic groups) at the same time point after infection (i.e. 16d and 5d after infection, Figure 6). These results demonstrate that the combination of infection with AAD results in enhanced neutrophilic inflammation. Collectively, our results show that NTHi infection in AAD may induce a phenotype of neutrophilic AAD that resembles neutrophilic asthma in humans.
Neutrophilic inflammation in asthma has been linked with increased IL-17 expression and IL-17 has been shown to be involved in neutrophil recruitment in response to bacterial infection. Therefore, the effects of infection on IL-17 responses during infection-induced neutrophilic AAD were further investigated. The experiments described hereafter were performed with infection during (d0 NTHi+OVA) OVA sensitization.
The profile of neutrophil influx into the airways and IL-17 production was determined in lung tissue and MLNs during the development of infection-induced neutrophilic AAD (1, 5, 12 and 16d, Figure 7A). The development of AAD resulted in increases in neutrophilic influx into the airways on day 12 and 16, and had minimal effects on IL-17 responses. Significantly greater numbers of neutrophils were recruited into the airways during infection and OVA sensitization (1d) and after OVA challenge (16d) in infected, allergic compared to uninfected, allergic groups (Figure 7B).
Importantly, increases in the infection-induced neutrophil influx were accompanied by significant increases in IL-17 responses in pulmonary tissue and MLN T cells. The expression of IL-17 mRNA in lung tissue was significantly elevated 1 day after infection in infected, allergic groups (Figure 7C) and returned to baseline levels by day 12, immediately prior to OVA challenge. Expression again significantly increased on day 16, after OVA challenges.
NTHi-induced IL-17 release from MLN T cells was also increased from days 5 to 16 in infected, allergic compared to uninfected, allergic groups (Figure 7D). Interestingly, infection did not affect OVA-induced IL-17 release (Figure 7E). Taken together, these data demonstrate that infection induces increased IL-17 responses in lung tissue and MLNs that correlate with elevated airway neutrophil numbers in infection-induced neutrophilic AAD.
IL-17 can induce neutrophilic inflammation by enhancing the expression of the chemotactic factor IL-8. Therefore, the mRNA expression and protein levels of KC and MIP2, the mouse orthologs of IL-8, in lung tissue were also investigated. KC and MIP2 mRNA and protein were elevated 1 day after infection and OVA sensitization in the lungs of infected, allergic, compared to uninfected, allergic groups, (Figure 8A–B). There were no differences in mRNA expression between groups at later time points. Therefore, the early induction of neutrophil influx into the lung during the development of infection-induced neutrophilic AAD is associated with early increases in neutrophil chemokine responses during infection.
To investigate the mechanisms that underpin infection-induced neutrophilic AAD, we assessed the potential cellular sources of IL-17 and the role of adaptive and innate immune cells in its release.
The development of AAD resulted in modest increases in IL-17 factors and responses (Figure 9A–E). ROR-γt, the Th17 differentiation factor was assessed, and expression was significantly elevated after 12 and 16 days in the lungs of infected, allergic, compared to uninfected, allergic groups (Figure 9A), suggesting that there was enhanced Th17 polarization. The numbers and proportions of T cells that were Th17 cells in lung tissue and MLNs were then determined by flow cytometry. CD3+CD4+IL-17+ (Th17) cells were significantly increased in the lungs after 12 and 16 days in infected, allergic groups (Figure 9B–C), but in the MLNs were increased only at day 5 (Figure 9D–E). These results indicate that Th17 cells in the lungs and MLNs may be the potential adaptive immune source of IL-17 after day 5.
We then assessed which cells were the early innate sources of IL-17 on day 1. Increased numbers and proportions of pulmonary macrophages (Figure 10A–B) and to a lesser extent neutrophils (Figure 10C–D) produced increased amounts of IL-17 at early but not other time-points in infected, allergic groups compared to uninfected, allergic controls. Lung macrophages and neutrophils isolated on day 1 also had increased levels of IL-17 mRNA transcripts (Figure 10E–F).
Taken together these results demonstrate that infection induces early IL-17 responses from lung macrophages and neutrophils and later responses from Th17 cells in lungs and MLNs that are associated with neutrophil influx into the airways.
We have shown that neutrophilic inflammation in infection-induced neutrophilic AAD correlates with increased expression of IL-17 during OVA challenge. To determine whether infection-induced neutrophilic inflammation is mediated by IL-17, IL-17 was depleted in infected, allergic groups during AAD, by administration of anti-IL-17 monoclonal antibody during OVA challenge on days 11 and 13 (Figure 11A), and AAD assessed (on day 16). This approach has previously been shown to deplete IL-17 in vivo [21]. Importantly, IL-17 depletion significantly reduced the numbers of neutrophils in the BALF compared to isotype treatment of infected, allergic groups (Figure 11B). Significantly, neutrophil numbers were not different to those observed in uninfected, allergic groups, which were unaffected by treatment. IL-17 depletion also significantly reduced KC mRNA expression levels in the lung, but had no effect on MIP2 mRNA (Figure 11C). Anti-IL-17 treatment of infected, allergic groups also partially restored IL-5 (3.198±0.679 ng/ml in anti-IL-17 compared to 1.576±0.238 ng/ml in isotype-treated infected allergic groups p<0.01) and IL-13 (18.240±1.533 ng/ml in anti-IL-17 compared to 11.988±0.938 ng/ml in isotype-treated infected allergic groups p<0.01) responses, but had no affect on IFN-γ release. These results demonstrate that infection-induced IL-17 release is responsible for neutrophil influx into the airways and the induction of neutrophilic AAD.
In this study we have demonstrated for the first time that H. influenzae respiratory infection drives IL-17-mediated development of neutrophilic AAD. NTHi infection suppressed pulmonary and systemic eosinophilic inflammation and reduced Th2 cytokine responses and AHR in AAD. However, infection induced neutrophilic inflammation during AAD, by promoting early (innate) and late (adaptive) IL-17 responses from pulmonary macrophages and Th17 cells, respectively. This indicates that H. influenzae infection may modulate immune responses in asthmatics that promote the development of neutrophilic asthma.
NTHi is commonly isolated from the nasopharynx of healthy individuals, but is also associated with chronic airway diseases such as bronchiectasis [37], COPD [38], and chronic bronchitis [39]. NTHi is the bacterium most commonly isolated during COPD exacerbations, and NTHi strains isolated during these exacerbations induce higher levels of IL-8, and subsequent neutrophil recruitment to the airways, than colonizing strains [40]. Simpson and colleagues have recently demonstrated that a large proportion of neutrophilic asthmatics are colonized with H. influenzae, have increased innate immune activation, and 6-8 fold higher endotoxin levels compared to other asthma subtypes and healthy controls [12]. A more recent study has demonstrated that 41% of neutrophilic asthmatics assessed had a significant load of potentially pathogenic bacteria, and H. influenzae was identified in 60% of patients that tested positive for these bacteria [33]. We have extended these studies to show that H. influenzae may promote neutrophilic asthma by suppressing Th2-mediated responses that are associated with alterations in antigen-presenting cells, and by inducing potent neutrophilic inflammation that is driven by Th17 responses.
We show that infection during and after sensitization inhibits characteristic features of eosinophilic asthma. Irrespective of the time of inoculation, infection significantly reduced both local and systemic allergen-induced cytokine release from MLNs and splenocytes, as well as airway and blood eosinophil recruitment. All of these effects may lead to the suppression of AHR. Tregs are an important cell involved in immune tolerance and the suppression of inflammation [41]. We show that Treg numbers were not changed by infection, and TGF-β and IL-10 expression in the lung, which are involved in the suppression of inflammation by Tregs, were reduced by infection in AAD. These results suggest that Tregs are not involved in the suppression of cytokines or cellular inflammation. The role of infection on DC function was investigated as DCs play an integral role in the uptake and presentation of antigen to naïve T cells, and as a result direct immune responses [42]. Infection significantly decreased markers associated with antigen presentation and co-stimulation of DCs. Therefore, infection is able to alter the phenotype of antigen-presenting cells, which may affect the interaction between APCs and T cells, and result in attenuated adaptive responses to allergen.
By contrast, NTHi infection induced potent neutrophilic inflammation in the airways. Persistent airway neutrophilia is also a feature common to chronic airway diseases, such as COPD [38], chronic bronchitis [39] and bronchiectasis [43], where recurrent infection is known to play an important role in pathogenesis. Neutrophilic inflammation is often associated with acute asthma exacerbations, and in particular infection-mediated exacerbations. Indeed, several studies have shown increased neutrophilic inflammation in both viral and bacterial infection-induced exacerbations [44], [45], [46]. Here we show that NTHi induces strong neutrophilic inflammatory responses, and may be involved in the development of neutrophilic asthma through the induction of neutrophilic inflammation. We demonstrate that immune responses that lead to the development of NTHi-induced neutrophilic AAD occur in two phases. The first involves innate immune activation during infection that is likely to result in neutrophil chemoattraction to the airways. NTHi infection during OVA sensitization resulted in a significant neutrophil influx to the airways 1 day after infection, a two-fold increase compared to NTHi infection alone. This correlated with the early production of KC, MIP2 and IL-17 in the lungs. These neutrophils and to a greater extent macrophages were able to produce significantly more IL-17 than those from infected, nonallergic and uninfected allergic controls. Therefore these cells, particularly macrophages, may be sources of early (innate) IL-17 release. This observation may have important implications for other diseases where the innate source of IL-17 has not yet been identified.
The second phase involves adaptive immune responses during allergen challenge resulting in increased infection-mediated Th17 responses. During the challenge phase, days 12-15, there was a significant upregulation of ROR-γt and IL-17 mRNA in the lungs of infected allergic groups compared to infected, nonallergic and uninfected allergic controls. These results directly correlated with increases in Th17 cells in the lungs of infected, allergic groups. The increased production of IL-17 from T cells in conjunction with increases in ROR-γt expression suggest that infection drives Th17 responses that preferentially induce IL-17 production and neutrophilic inflammation during subsequent allergen challenge.
Collectively, our data suggest that infection induces early responses, involving neutrophils, KC, MIP2 and IL-17 expression that may prime the host for enhanced Th17-mediated neutrophilic responses upon later allergen challenge, which subsequently induces neutrophilic AAD. Our findings are consistent with data from a recent study by Bullens et al., which showed that increased IL-17 responses in asthmatics correlate with increased neutrophil numbers in sputum [15]. Hellings et al., [21] demonstrated that IL-17 is important in lung neutrophil recruitment in response to an allergen, while Ye et al., [29] showed that IL-17 responses and signalling through the IL-17R is vital for neutrophil recruitment and host defence against Klebsiella pneumoniae infection. Here we extend these findings by demonstrating that H. influenzae infection-induced IL-17 responses in AAD may play a role in driving neutrophilic inflammation in asthma. We recently demonstrated that chlamydial respiratory infection is also able to drive neutrophilic asthma [47]. Chlamydial infection suppressed Th2-mediated eosinophilic inflammation and promoted neutrophilic inflammation and AAD. Neutrophilic asthmatics are resistant to corticosteroid treatment, which is the mainstay of asthma therapy [10], and with evidence that asthmatics with infection are more resistant to steroids than asthmatics with no infection [48], alternative therapies are needed for infection-induced neutrophilic asthma.
It is possible that either infection alone or the synergistic effects of infection and AAD are required for the induction of the neutrophilic AAD phenotype. NTHi infection alone did increase neutrophil influx into the airways (p<0.01), IL-17 mRNA expression in the lung (p<0.05) and the percentages of lung macrophages and neutrophils producing IL-17 (both p<0.001) 1 day after infection, compared to uninfected, nonallergic (Saline) controls. These effects would be expected as they are normal responses to infection; however, they may contribute to the establishment of a pro-neutrophilic environment and the subsequent development of neutrophilic AAD upon allergen challenge. However, NTHi infection does not persist past 10 days, and yet is able to modify AAD after 16 days. It is, therefore, likely that the major affect of the infection is to induce persistent immune changes, that continue even after the clearance of the infection, that synergize with allergen exposure to drive neutrophilic AAD. This mechanism has recently been proposed in humans [49].
OVA sensitization at the time of, or prior to, infection, and subsequent OVA challenge does not affect NTHi load. When infection occurs during sensitization, bacteria are still cleared by day 16, and; when infection occurs 10 days after sensitization, bacterial recovery at the end of the protocol (i.e. 5 days later), is the same as that in the infected, non-allergic group (data not shown).
It is likely that the induction of a neutrophilic phenotype is specific to a subset of infectious agents, particularly NTHi and Chlamydia [12], [33], [50]. In other studies we have investigated the impact on AAD of an unrelated respiratory pathogen and commensal bacterium, Streptococcus pneumoniae, which is not associated with neutrophilic asthma. We show that S. pneumoniae infection or components do not induce changes in IL-17 responses or increase neutrophilic inflammation with infection or component administration either before, during or after the induction of AAD [51], [52]. Both NTHi- and Chlamydia-induced neutrophilic AAD, may potentially be driven by conserved pathogen-associated molecular patterns (PAMPs), such as lipopolysaccharide (LPS) or CpGs. Numerous studies have investigated the effects of LPS and CPG in AAD, and the interactions are complex. LPS is highly variable and different types and levels of LPS have different effects in AAD. Low dose LPS administration during sensitization promotes Th2 responses and is a risk factor for severe asthma [53], [54], while high doses decrease Th2 responses and induce non-eosinophilic inflammation [55], [56], [57], [58]. Chlamydia-derived LPS is atypical and is 1,000 fold less immunogenic compared to other bacterial-derived LPS [59], and therefore, is unlikely to be the cause of Chlamydia-induced neutrophilic AAD. CpGs when given together with antigen in established disease induce Th1 and regulatory T cell responses that suppress the features of AAD including Th2 responses and AHR [60], [61], [62]. We have shown that depleting neutrophils during a Chlamydia infection inhibits the development of neutrophilic AAD, by contrast IL-17 responses drive NTHi-induced neutrophilic AAD. Therefore, we suggest that it is specific immune responses to these, as well as potentially other infections, which are driven by as yet unidentified factors in the infections, that are the mechanisms that drive neutrophilic AAD [47]. Our studies do not rule out LPS, CpGs, or other PAMPs as drivers of the neutrophilic phenotype and further research is required to elucidate this possibility.
Importantly we have shown that infection-induced neutrophilic AAD and the suppression of Th2 responses are dependent upon IL-17. Depletion of IL-17 with anti-IL-17 monoclonal antibody during AAD prevented the development of infection-induced neutrophilic AAD. This suggests that IL-17 is critical in the recruitment of neutrophils and may suppress Th2 responses in infection-induced neutrophilic inflammation and neutrophilic asthma. Wakashin et al., demonstrated that adoptive transfer of antigen-specific Th17 cells induced airway neutrophil recruitment, which supports our data [63]. Little is known about how Th17 and Th2 cells interact with or regulate each other. We demonstrate that infection inhibits cytokine release compared to uninfected allergic controls, and that anti-IL-17 treatment partially restored these effects, while having no effect on eosinophil recruitment (data not shown). This suggests that other mechanisms are also involved in the suppression of Th2 responses by NTHi infection, which requires further investigation. Several recent studies have investigated the relationship between Th17 and Th2 cells. Schnyder-Candrian et al., showed that the administration of rIL-17 in a murine model of AAD significantly reduced allergen-induced eotaxin, thymus and activation-regulated chemokine (TARC) and IL-5, thereby reducing eosinophilic inflammation [64], while another study showed that inhibiting IL-17 in AAD also reduced airway eosinophils, neutrophils, AHR, and Th2 cytokines [65]. These data suggest that IL-17 may suppress or promote eosinophilic inflammation, but the mechanisms that drive these different effects remain unknown. Our data is in agreement with Schnyder-Candrian et al., who suggest that IL-17 interferes with DC activation and antigen uptake, which prevents T cell activation and reduced IL-4, -5 and -13 production, leading to suppressed allergic responses. Interestingly, a recent study has shown a CD4+ T cell subtype that expresses both Th17 and Th2 cytokines, including IL-4, IL-5, IL-13, IL-17 and IL-22, and this subset is increased in asthmatics compared to healthy controls [66]. However, these cells have thus far only been found in the periphery, and confirmation of their presence is needed in BAL, sputum and/or bronchial biopsies.
In conclusion, we show that H. influenzae infection may be involved in the development of neutrophilic asthma. Infection suppressed features of Th2-mediated eosinophilic AAD, while inducing features of neutrophilic asthma that are mediated by infection-induced IL-17. Therefore, infection-induced IL-17 responses may play a major role in the pathogenesis of neutrophilic asthma. Our studies indicate the important role of infection in driving neutrophilic asthma-like disease, and identify new areas of investigation that may enhance the understanding of disease progression. Developing new treatments targeting infection may lead to better management of individuals with this disease phenotype.
This study was carried out in strict accordance with the recommendations in the NSW Animal Research Regulation 2005, and the Australian Code of Practice for the care and use of animals for scientific purposes (National Health and Medical Research Council). All protocols were approved by the Animal Care and Ethics Committee of the University of Newcastle (permit number 987/0111). All surgery was performed under sodium pentobarbital anaesthesia, and all efforts made to minimize pain and suffering.
Six to eight week old female BALB/c mice were used. Mice were sensitized by intraperitoneal (i.p.) injection, with OVA (50 µg, Sigma-Aldrich, Castle Hill, NSW, Australia) with the Th2-inducing adjuvant Rehydrogel (1mg, in 200 µl sterile saline, Reheis, Berkeley Heights, USA). On days 12 to 15 mice were challenged intranasally (i.n.) with OVA (10 µg, 50 µl) and AAD was assessed on day 16 [67]. Controls were sham sensitized to saline.
NTHi (NTHi-289) glycerol stocks were plated onto chocolate agar plates (Oxoid, SA, Australia), grown overnight (37°C, 5% CO2), then washed off the plate and suspended in sterile PBS. To determine the effects of infection, mice were inoculated i.t. with 5x105 CFU NTHi (in 30 µl PBS) during (Day 0) or after (Day 10) OVA sensitization. Controls were infected but not exposed to OVA. In preliminary studies we determined that this inoculum induced an infection from which the mice recovered and could be used to study the effects of infection on AAD.
BALF was collected and processed as previously described [68]. Briefly, the left lung was tied off and the right lung was washed twice with Hank's buffered salt solution (700 µl, HBSS; Trace Scientific, Noble Park, Vic, Australia). Cells were pelleted and resuspended in red blood cell lysis buffer, washed and resuspended in HBSS, then cytocentrifuged (300g, 5 min, ThermoFisher Scientific, Scoresby, Vic, Australia) onto microscope slides. Blood smears were prepared from a drop of whole blood. BALF and blood cells were stained with May-Grunwald-Giemsa, and differential leukocyte counts were enumerated using light microscopy [68].
Right lobes of lungs, from which BALF had been obtained, were aseptically removed and homogenized in 1ml of sterile PBS. Serial dilutions of BALF and lung homogenates were prepared in sterile PBS, plated onto chocolate agar plates and incubated overnight (37°C, 5% CO2). Colonies were enumerated and bacterial numbers per right lung calculated.
AHR was measured in response to increasing doses of aerosolized methacholine, by whole body invasive plethysmography as previously described [47]. Briefly, mice were anaesthetized and tracheas were cannulated and attached to a ventilator. Peak dynamic compliance and transpulmonary resistance were assessed by analysis of pressure and flow waveforms following challenge with increasing doses of aerosolized methacholine (Sigma-Aldrich).
Supernatants from lung draining MLN T cells were restimulated with OVA (200 µg/ml) or ethanol-killed NTHi (2×107 CFU/ml) and cultured for six days (5% CO2, 37°C, 1x106 cells per well). After culture supernatants containing soluble factors were recovered and analyzed for IL-5, IFN-γ, (BD Biosciences, North Ride, NSW, Australia), IL-13, IL-17A and IL-22 (R&D Systems, Minneapolis, MN, USA) by ELISA, according to manufacturer's instructions [69].
Whole lungs were homogenized in RIPA buffer (1ml, Sigma-Aldrich) and incubated on ice for 5 mins. Cells were pelleted and the supernatant recovered and analyzed for KC and MIP2 by ELISA, (R&D Systems), according to manufacturer's instructions.
RNA was TRIZOL extracted from whole lung homogenates according to manufacturer's instructions (Invitrogen, Mount Waverly, Vic, Australia). Target gene expression was determined relative to the reference gene hypoxanthine-guanine phosoribosyltransferase (HPRT) [69]. Primers used were IL-17, Fwd 5′-aaacatgagtccagggagagcttt-3′, Rev 5′-actgagcttcccagatcacagagg-3′; ROR-γt, Fwd 5′- ccgctgagagggcttcac-3′, Rev 5′- tgcaggagtaggccacattaca-3′; MIP2, Fwd 5′- ctagctgcctgcctcattctac-3′, Rev 5′- caacagtgtacyyacgcagacg-3′; KC Fwd 5′- cttggggacaccttttagca-3′, Rev 5′- gctgggattcacctcaagaa-3′; TGF-β Fwd 5′- cccgaagcggactactatgctaaa-3′, Rev 5′- ggtaacgccaggaattgttgctat-3′; IL-10, Fwd 5′-catttgaattccctgggtgagaag-3′, Rev 5′- gccttgtagacaccttggtcttgg-3′; and HPRT Fwd 5′- aggccagactttgttggatttgaa-3′, Rev 5′- caacttgcgctcatcttaggcttt-3′.
Single cell suspensions of MLNs and collagenase-D digested lungs were prepared. IL-17 producing cells were identified by stimulation with phorbol 12-myristate 13-acetate (PMA, 0.1 µg/ml) and ionomycin (1 µg/ml, Sigma-Aldrich) in the presence of Brefeldin A (8 µg/ml, Sigma-Aldrich) for 4 hours [70]. Cells were incubated with Fc block for 15mins, then stained for surface markers CD4, CD3, CD11b, Gr-1 (BD Bioscience), or F4/80 (eBioscience, San Diego, CA, USA), fixed with 4% paraformaldehyde (PFA), permeabilized with 0.1% saponin, and stained for intracellular IL-17 (or isotype control rat IgG2a, eBioscience). Tregs were identified using surface markers CD4, CD25 and a staining kit for intracellular FoxP3 (or IgG2a isotype control) according to manufacturer's instructions (eBioscience). pDCs were characterized as CD11clowCD11b-B220+, and mDCs characterized as CD11c+CD11b+B220− (BD Bioscience); using MHCII and CD86 (R&D Systems) for activation and co-stimulation status. All cells were analyzed using a FACS Canto (BD Bioscience) [47].
Single cell suspensions of collagenase-D digested lung tissue were prepared, and resuspended in red blood cell lysis buffer. Resuspended cells were either incubated overnight (5% CO2, 37°C, 1x106 cells per well) and macrophages isolated by adherence to culture plates; or were put through a mouse neutrophil enrichment kit (Stemcell Technologies, Melbourne, Vic, Australia) and neutrophils isolated by negative selection following manufacturer's instructions. mRNA was purified from these macrophages and neutrophils using a PureLink RNA mini kit (Invitrogen) according to manufacturer's instructions.
Monoclonal anti-IL-17A neutralizing antibody (clone 50104, rat IgG2a) was administered by i.p. injection (100 µg/mouse, eBioscience) on days 11 and 13, and features of AAD were assessed on day 16. Control groups were uninfected and treated with anti-IL-17 or treated with IgG2a isotype control antibody [47].
Results are presented as mean ± standard error of the mean (SEM) from 6–8 mice, in duplicate. Significance was determined by one-way ANOVA or Student t-test (GraphPad Software, CA, USA).
|
10.1371/journal.pmed.1002125 | The Policy Dystopia Model: An Interpretive Analysis of Tobacco Industry Political Activity | Tobacco industry interference has been identified as the greatest obstacle to the implementation of evidence-based measures to reduce tobacco use. Understanding and addressing industry interference in public health policy-making is therefore crucial. Existing conceptualisations of corporate political activity (CPA) are embedded in a business perspective and do not attend to CPA’s social and public health costs; most have not drawn on the unique resource represented by internal tobacco industry documents. Building on this literature, including systematic reviews, we develop a critically informed conceptual model of tobacco industry political activity.
We thematically analysed published papers included in two systematic reviews examining tobacco industry influence on taxation and marketing of tobacco; we included 45 of 46 papers in the former category and 20 of 48 papers in the latter (n = 65). We used a grounded theory approach to build taxonomies of “discursive” (argument-based) and “instrumental” (action-based) industry strategies and from these devised the Policy Dystopia Model, which shows that the industry, working through different constituencies, constructs a metanarrative to argue that proposed policies will lead to a dysfunctional future of policy failure and widely dispersed adverse social and economic consequences. Simultaneously, it uses diverse, interlocking insider and outsider instrumental strategies to disseminate this narrative and enhance its persuasiveness in order to secure its preferred policy outcomes. Limitations are that many papers were historical (some dating back to the 1970s) and focused on high-income regions.
The model provides an evidence-based, accessible way of understanding diverse corporate political strategies. It should enable public health actors and officials to preempt these strategies and develop realistic assessments of the industry’s claims.
| Interference by the tobacco industry in government policy-making is known to be an important reason for governments’ failure to adopt proven measures to reduce tobacco consumption.
Our study aimed to systematically review tobacco industry political activity from a critical societal perspective using in-depth empirical evidence.
We set out to build a widely applicable model that governments in different countries could use to identify and preempt industry interference in policy.
We analysed 65 papers included in previous systematic reviews that examined tobacco industry political activity in two policy areas: taxation and marketing of tobacco products.
Using constructivist grounded theory, we identified industry arguments and techniques and then grouped these under the more general heading of strategies, finally developing a taxonomy and model of tobacco industry political activity.
According to our Policy Dystopia Model, the industry produces an alarmist narrative that proposed policies will fail and lead to a great number of undesirable social and economic consequences (outlined in our taxonomy of discursive strategies).
The industry also uses different methods (outlined in our taxonomy of instrumental strategies) to disseminate this narrative and persuade decision-makers in order to block policies.
Public health actors and policy-makers can use our model and taxonomies to better promote tobacco-related policies and to render ineffective tobacco industry opposition and political activity.
In particular, they need to pay attention to alliance formation, information provision, interdisciplinary working, and transparency in policy-making.
| Globally, tobacco kills 6 million people annually, potentially rising to 8 million by 2030 [1]. Of the 193 member states of the United Nations, 180 have now ratified the Framework Convention on Tobacco Control (FCTC), which outlines the evidence-based policies required to reduce tobacco use. The Convention has driven policy implementation internationally [2], but progress remains slow [3], with parties to the treaty identifying industry interference as the greatest impediment to progress [4]. It is increasingly recognised, therefore, that understanding, exposing, and addressing tobacco industry interference is key to progressing tobacco control [4,5]. The archive of more than 14 million internal tobacco industry documents disclosed as a result of litigation in the United States [6,7] has generated a unique evidence base for understanding the conduct of transnational tobacco companies (TTCs). However, given that there are now over 800 research publications based on these documents [8], evidence syntheses and conceptual models are required to more effectively use this evidence base to inform policy and augment social scientific understanding of TTC efforts to influence policy.
To date, only three systematic reviews—on TTC efforts to influence taxation [9], marketing [10], and policies in lower- and middle-income countries [11]—and two conceptual frameworks of tobacco industry political activity have been published [10,12]. In the latter category, only one was based on industry documents and used a systematic review of studies on marketing policy to begin to develop a taxonomy of tobacco industry political activity [10], drawing on Hillman and Hitt’s widely cited taxonomy of corporate political activity [13]. This work highlighted major shortcomings in Hillman and Hitt’s exchange-theory–based representation of corporate political activity as a mutually beneficial process through which corporate involvement in policy-making enables governments to develop optimal public policies.
Our current research builds on this initial work by incorporating evidence on the tobacco industry’s attempts to influence two key policy areas, taxation and marketing, and by taking a critical approach in order to develop a more comprehensive and grounded understanding of tobacco industry political activity. Our research questions were: What does the tobacco industry aim to achieve through its political activity? What would a critical taxonomy of tobacco industry political activity look like? By answering these questions, we also aimed to explore the value and limitations of Hillman and Hitt’s taxonomy in the context of the tobacco industry. We conducted in-depth interpretive analysis of papers included in two systematic reviews of tobacco industry political activity using grounded theory methods. Our analysis led to the development of two critical taxonomies that we hope will be of use to policy-makers and public health groups and an overall model that we present as an alternative conceptualisation of corporate political activity.
Our data comprised the papers included in two systematic reviews that we had previously conducted on tobacco industry political activity, taxation [9], and marketing policies [10]. We based our analysis on these two systematic reviews, the methodological details of which are published elsewhere [9,10], because they provide the best quality of evidence in the area of inquiry. The two reviews used comprehensive searches (database searches, hand searches, internet searches, expert contact) to identify all relevant academic and grey literature, yielding 2,678 taxation and 1,754 marketing sources. Relevance and quality criteria were applied to identify the best quality evidence in the field; 46 and 48 papers were included, respectively. The database for the current analysis comprised 65 papers (Table 1). For taxation, we used all but one of 46 papers in the original review (we excluded the interim version of a report). For marketing, because this literature had only recently been reviewed by two of the same authors to develop a taxonomy [10], we selected half (24 of 48) of the papers in the review, using the following criteria: papers covering the last ten years (2003–2013) of the original review period and an even representation (within the constraints of the original sample) of geographic location and specific marketing policies. Four of the 24 were also in the taxation systematic review (with different sections analysed for each topic), making a total of 20 marketing papers. The included papers are listed in S1 Table (taxation) and S2 Table (marketing).
Our approach to analysis was critical; drawing on the findings of our previous systematic review [10], we rejected Hillman and Hitt’s assumption that corporate political activity is a transparent and cooperative endeavour based on mutuality and honesty between public and private actors. We used Sarah Pralle’s concepts of expansion and containment of issues, actors, and spaces in advocacy work [14]. We also added the extra dimension of “voice” because we found that diverse voices were instrumental in framing, expanding, and containing issues and arguments. Additionally, we delineated stylistic features of tobacco industry political activity.
We used the techniques of constructivist grounded theory [15,16]: conceptual coding, systematic conceptual comparison, discourse sensitivity, attention to divergent data, and conceptual explanatory conclusions. Starting with the taxation literature, all the papers were entered into the ATLAS ti software and SU microcoded them for the smallest meaningful conceptual idea either as “argument” or “technique”; these were subsequently grouped under discursive and instrumental strategies, respectively. During this initial microcoding, we did not follow the coding schema of the two systematic reviews but conducted inductive and emergent coding, although it quickly became clear that most of the three strategies identified in Hillman and Hitt [12] and the three additional ones in Savell et al. [10] as well as the frames and arguments in the latter were relevant. However, we identified many new strategies and conceptually revised and refined others.
We included in the analysis all strategies identified in the dataset regardless of frequency but recorded the number of instances each was used (shown in our results tables). When the coding of the taxation papers was completed, a draft taxonomy was developed. Next, we inductively coded the marketing papers, regularly comparing the emergent coding frame with the draft taxonomy developed from the taxation papers and revising the latter as necessary. When all microcoding was completed, the taxonomy was reexamined for conceptual coherence and clarity and further modified. Next, a dynamic model was developed that accounted for not only the categories in the taxonomy but the relationships between them and the directions of influence. The study team (SU, GJF, ABG) met regularly (every 4–6 wk) throughout the study period to discuss microcodes, taxonomy categories, and the model, discussing and reaching consensus on divergent views.
Our data showed that, faced with policy proposals aimed at reducing tobacco consumption, the tobacco industry attempts to secure a range of preferred outcomes that eliminate or limit the likely impact on its business (Table 2). Defeat—scrapping or shelving the policy—is the optimal outcome. Delay and weakening are sought if defeat is not possible. Foreclosing the legislative space is a future-facing strategy designed to make subsequent initiation and enactment of tobacco regulation more difficult. Once policy or regulation is in place, the industry may seek to overturn it. Alternatively, it attempts regulatory/policy avoidance through noncompliance, circumventing the rules or, for earmarked taxes, diverting earmarked funds.
Early on in our analysis, we recognised that industry political activity was performed through both arguments and actions and that these should be conceptualised as synergistic components of a dynamic model of political influence. We therefore distinguish between discursive (argument-based) and instrumental (action-based) strategies. We use the term “argument” as a subcategory of discursive strategies and the more neutral term “technique” instead of the often-used “tactic” as a subcategory of instrumental strategies. This terminology builds on that in our previous work [10]. We now present our taxonomies for each.
The industry’s overall discursive strategy is to exaggerate—expand by argument—potential costs of proposed policy while simultaneously dismissing—containing by argument—potential benefits or denying these altogether. It seeks to build a comprehensive and credible narrative of undesirability for the policy by generating tailored arguments covering many different social domains (Table 3). Collectively, these mutually reinforcing arguments build the impression that the proposed public health policy will be detrimental to public health, the economy, and society. A key feature of this narrative is that it spans diverse sociopolitical domains and communities and is articulated not only by tobacco companies but also through a plurality of third-party voices, including those of law enforcer, concerned citizen, and public health policy analyst (Table 3). In this way, opposition to public health policies is represented not as the self-interested response of a profit-oriented business but the genuine concerns of different sectors of the public. A 1989 Philip Morris document explains:
The industry uses both insider (legislative and governmental) and outsider (public domain) strategies [73,74] in order to persuade the public and decision-makers of the plausibility of its predictions encapsulated in the arguments outlined above (Table 4). Coalition management represents a key outsider strategy. The main insider strategy—“lobbying”—is not consistently defined in the literature, being used by US scholars to refer to the provision of information to policy-makers [13] and by others as a combination of information provision and pressure techniques [75]. We use the term in the non-US sense and further suggest that actions normally subsumed under “lobbying” can be disaggregated into the categories of information management and direct involvement/influence (Table 4), with information management straddling both insider and outsider domains as we explain below. Litigation and illicit trade are both outsider strategies.
Our analysis suggests that the tobacco industry’s overall approach to opposing tobacco control policies is to construct an overarching narrative of a dysfunctional future that will ensue if the proposed policy is implemented and to widely disseminate this narrative in order to enhance its persuasiveness. We term this the Policy Dystopia Model. Embedded within a cost–benefit paradigm, the central narrative asserts that the policy will undermine public welfare because its costs will be large and will fall indiscriminately on a wide range of stakeholders, damaging the economy and society as a whole, while its benefits will be limited, non-existent, or enjoyed by the wrong stakeholders. This dystopian narrative is processed through three main (coalition management, information management, direct involvement/influence in policy) and two subsidiary (illicit trade, litigation) instrumental strategies (Fig 1). Within the model, there is fluidity and strong interdependence within and between discursive and instrumental strategies. For example, the strategies of information and coalition management facilitate each other: the support of a variety of constituencies allows diverse (tailored) dystopian arguments to be shaped and disseminated, and the voicing of these arguments by constituencies render them more credible and persuasive [91,92]. Furthermore, these diverse arguments help secure the support of a variety of groups whose interests might not otherwise align with those of the industry. Similarly, the strategy of facilitating or engaging in illicit trade feeds into the information strategy by providing “evidence” for the industry’s arguments concerning smuggling and reduced government revenues. The model represents a highly dynamic process in which different strategies are accentuated at a given time in line with prevailing political/economic contingencies.
The tobacco industry’s core premise is that policy-makers fail to consider, or underestimate, the potentially disastrous consequences of proposed public health policies, which, translated as costs, far outweigh any (marginal) benefits of the policy, making it unfeasible and damaging to legislators. To impress this alternative reality on the public and decision-makers and create an unwarranted impression of wide public and business support, it sets into motion an interlocking ensemble of activities: producing inaccurate and biased information, forming (unlikely) coalitions with diverse groups, and exerting direct influence on decision-makers.
The timeframe and geographic focus of the papers reviewed may limit the model’s applicability: although 32 out of 46 of the taxation papers and all the included marketing papers were published between 2003 and 2010, some covered historical events, and both were heavily dominated by North American/European/Australasian sources. However, our previous review and other work have shown that similar strategies are used in developed and developing economies and repeated over time [10,11], while the sociopolitical conditions in the emerging markets where the industry is currently establishing itself may not be that different from the historical contexts represented in our source literature [45–47]. Furthermore, it is apparent that TTCs are currently using the strategies and techniques we identify [5,93,94], although the weight given to different elements may vary over time. For example, the industry appears to be shifting the emphasis of its political strategies towards those that centre on illicit trade [68], litigation [95], and international trade agreements [96], with the use of third parties increasing in apparent response to the industry’s declining insider status [97,98]. Finally, we note that the tobacco industry uses similar strategies to oppose policies and regulation in areas other than marketing and taxation, such as smoke-free policies [11,98].
Our model and taxonomies take a critical perspective that recognises the fundamental conflict between corporate interests and public health. It was to address this conflict that the WHO FCTC’s Article 5.3 [99] and its implementing guidelines [100] suggested measures aimed at protecting public health policy-making from tobacco industry influence. A recent FCTC report [4] shows that much progress is still needed to address this interference. Our work, by identifying TTCs’ key strategies, can be used by public health advocates and policy-makers to direct efforts to effectively implement Article 5.3; it highlights, for example, the importance of full transparency. Second, on the basis that TTCs’ strategies and arguments are repeated over time and place, it can potentially be used to anticipate and counter industry opposition. For example, counterarguments and media strategies could be prepared in advance based on the details given above. Finally, the taxonomy can enable advocates and policy-makers to recognise and label contemporary industry strategies without using resources to investigate each incident.
Public policy formulation is a collective process, with corporations constituting only one part of an ensemble of governmental and nongovernmental actors and institutions. Whether corporations secure favourable policy outcomes depends not only on their actions but also on how those promoting the policy position their case and respond to corporate strategising, and how readily policy-makers accept corporations’ dystopian cost-based projections. Studies in other areas [101–103] indicate that governments attach considerable emphasis to corporate claims when contemplating health related policies. Furthermore, corporations’ emphasis on projected costs of public health policies dovetails with the “Better Regulation” agenda that increasingly dominates European, Australasian, and North American policy landscapes [104–106] and is often accompanied by mandatory impact assessments of costs and benefits. Both are expressions of prevailing neoliberal norms that posit state intervention in markets as inefficient and illiberal and promote minimal (“light-touch”) regulation of industries/business. Our analysis suggests that this convergence between corporate and government interests and the embedding of cost–benefit analysis within policy-making is likely to have major implications for public health. It raises the possibility of corporate annexation of public policy where corporate interests are better represented than broader public interests and where alternative approaches such as the precautionary principle [107] are squeezed out. A sustained public debate is needed on whether it is ethically defensible for the values of market economics, competition, and profit maximisation to guide deliberations on health- and welfare-oriented policies.
Both our previous work [10] and the wider analysis we offer here point to the inadequacy of exchange-based conceptualisations of corporate political activity as underpinning socially optimal policy-making, an approach encapsulated in the Hillman and Hitt taxonomy [13] and other work [108,109]. In the tobacco industry’s case, the failure of this theoretical approach to account for the industry’s political aims and actions as well as the costs of these for policy and public welfare have been demonstrated empirically. The industry has failed to act “responsibly” [110], systematically misused and misrepresented information and scientific evidence [69,111–113], exaggerated the costs of policies [94,114], obscured its involvement in the production of evidence designed to favour its case [114–117], and has been extensively involved in tobacco smuggling while opposing policies on the basis that they will increase illicit trade [39,90,118–122].
Although further research is needed to confirm this, our model and taxonomies are likely to be applicable to political action by nontobacco industries that also threaten the public’s health: for example, the oil and gas, ultra-processed food/soft drinks, and alcohol industries. There is growing evidence that nontobacco sectors use key strategies of the dystopia model in opposing public health policies: problem reframing and introducing “unintended” consequences [123]; exaggerating economic costs [124]; and constituency recruitment/fabrication and information management [125,126]. Our earlier work shows that the alcohol industry, for example, uses remarkably similar strategies to tobacco in opposing marketing policies [127]. Therefore, it is likely that management-focused theoretical positions are equally inadequate in studying and understanding political activity in other sectors. There is urgent need for more critical social science scholarship in this field.
The Policy Dystopia Model and taxonomies can be a useful resource to the public health community and to policy-makers at national, regional, and international levels. First, this work will enable systematic focus on the types of dystopian narratives the industry is likely to produce for specific policies and the voices it is likely to use; these can then be preempted through effective counterarguments [128]. The health and societal benefits of the policies, in particular, need to be foregrounded to counter industry attempts to background them. Second, the taxonomies will enable health actors to anticipate and identify the kinds of coalitions the industry may attempt to build. Third, our work highlights that a crucial informational task for health actors is to deconstruct industry deconstructions of public science as well as to produce and disseminate information on the policy, paying attention to audience and language. Here, two strategies emerge as important: involving supportive organisations and individuals in producing and cascading information and enabling interdisciplinary networking by lawyers, public policy experts, economists, and scientists [129]. Fourth, the work highlights that government must ensure and the public health community must insist on transparency in official interactions with the industry and lobbyist access to legislatures as well as funding disclosure for all individuals and organisations acting as stakeholders. Finally, the model and taxonomies can place public health advocates and policy-makers in an advantageous position in which they are able to proactively plan narratives and strategies and not merely react to those of the industry. Further empirical work is required to examine whether the policy dystopia model is applicable to other tobacco control policy areas such as smoke-free environments as well as to nontobacco public health policies; how public and elected officials make sense of and respond to industry strategies; and what works in countering corporate political activity.
|
10.1371/journal.pcbi.1005357 | Iterative sure independence screening EM-Bayesian LASSO algorithm for multi-locus genome-wide association studies | Genome-wide association study (GWAS) entails examining a large number of single nucleotide polymorphisms (SNPs) in a limited sample with hundreds of individuals, implying a variable selection problem in the high dimensional dataset. Although many single-locus GWAS approaches under polygenic background and population structure controls have been widely used, some significant loci fail to be detected. In this study, we used an iterative modified-sure independence screening (ISIS) approach in reducing the number of SNPs to a moderate size. Expectation-Maximization (EM)-Bayesian least absolute shrinkage and selection operator (BLASSO) was used to estimate all the selected SNP effects for true quantitative trait nucleotide (QTN) detection. This method is referred to as ISIS EM-BLASSO algorithm. Monte Carlo simulation studies validated the new method, which has the highest empirical power in QTN detection and the highest accuracy in QTN effect estimation, and it is the fastest, as compared with efficient mixed-model association (EMMA), smoothly clipped absolute deviation (SCAD), fixed and random model circulating probability unification (FarmCPU), and multi-locus random-SNP-effect mixed linear model (mrMLM). To further demonstrate the new method, six flowering time traits in Arabidopsis thaliana were re-analyzed by four methods (New method, EMMA, FarmCPU, and mrMLM). As a result, the new method identified most previously reported genes. Therefore, the new method is a good alternative for multi-locus GWAS.
| Genome-wide association study is concerned with the associations between markers and traits of interest so as to identify all the significantly associated markers. In genome-wide association studies, hundreds of thousands of markers are genotyped for several hundreds of individuals. Usually, only a very minor subset of these markers is associated with the trait. Most penalization methods fail when the number of markers is much larger than the sample size. Based on this fact, we have developed an algorithm that proceeds in two stages. In the first stage (screening), we reduced the number of markers via correlation learning to a moderate size. We then used a moderate-scale variable selection method to select variables in the reduced model. Conditional on the selected variables, we repeated the screening procedure and chose another set of variables. In the second stage (estimation), all the above-selected variables are accurately estimated in a multi-locus model. Our approach is simple, accurate in estimation, fast and shows high statistical power of detecting relevant markers on simulated data. We have also used this method to identify relevant genes in real data analysis. We recommend our approach for conducting a multi-locus genome-wide association study.
| Genome-wide association study (GWAS) focuses on associations between single nucleotide polymorphism (SNP) and traits of interest in order to investigate the genetic foundation of these traits [1, 2]. In GWAS, hundreds of thousands of SNPs are genotyped for several hundreds of individuals. In this case, statistical estimation and detection of the relationship between these SNPs and the traits become challenging. Although the single variant analysis in standard GWAS methods has succeeded in identifying thousands of genetic variants associated with hundreds of various traits, this approach fails to consider the joint effect of multiple genetic markers on traits. Another problem with this approach is the issue of multiple test correction for the threshold value of significance test. The Bonferroni correction is too stringent, and many relevant loci are missed out.
In genetics, only a small subset of SNPs is associated with the phenotype of a trait. This is an example of a variable selection problem for high-dimensional data, where the number of SNPs (p) is several times larger than the number of individuals (n) [3]. To solve this issue, many penalization methods have been developed in statistics, for example, bridge [4], nonnegative garotte [5], least absolute shrinkage and selection operator (LASSO) [6], smoothly clipped absolute deviation (SCAD) [7], elastic net [8], fused LASSO [9], adaptive LASSO [10] and minimax concave penalty [11]. Among these methods, Bayesian LASSO [12], penalized logistic regression [13], sure independence screening [14], adaptive mixed LASSO [15], elastic net [16], LASSO [17], empirical Bayes [18] and empirical Bayes LASSO [19] have been adopted in GWAS. These methods are multi-locus in nature, hence a less stringent significance criterion can be adopted [20]. Despite these methods being able to shrink some markers to zero, they will fail when the number of markers is several times larger than the sample size. In this case, the solution lies in reducing the number of markers before employing a shrinkage method in the multi-locus genetic model. For example, a Bayesian sparse linear mixed model [21] and Bayesian mixture models [22]. However, the computing time becomes a major concern for these Bayesian approaches. An alternative is to integrate single-marker scanning with the multi-locus models, such as a model-free approach [23], multi-locus random-SNP-effect mixed linear model (mrMLM) [20], and fixed and random model circulating probability unification (FarmCPU) [24].
In this study, we developed an approach that reduced the number of markers, p, via correlation learning (i.e. Iterative modified-Sure Independence Screening) to a moderate number. A moderate-scale variable selection method, SCAD, was then employed to select variables in the reduced model. We chose SCAD because of its nice oracle property. Conditional on the selected variables, we repeated the screening procedure and chose another set of variables. All the effects of the above-selected variables were estimated by Expectation-Maximization (EM)-Bayesian LASSO algorithm [25] and tested by likelihood ratio statistic for true quantitative trait nucleotide (QTN) detection. We call this approach Iterative modified-Sure Independence Screening EM-Bayesian LASSO (ISIS EM-BLASSO) algorithm. A series of simulated and real datasets were used to validate this new method. We compared our new method with single-locus methods: efficient mixed-model association (EMMA) [26] and FarmCPU [24], multi-locus methods: SCAD [7] and mrMLM [20]. Several reasons guided the choice of these comparison methods: EMMA [26] has been a standard gold method for GWAS, FarmCPU [24] reduces the number of markers used in GWAS just like ISIS EM-BLASSO, SCAD [7] is used in the screening method of ISIS EM-BLASSO hence the need to compare it independently and lastly, mrMLM [20] integrates single locus with multi-locus approach.
Three Monte-Carlo simulation experiments were carried out to measure the effectiveness of our new method. We used statistical power to evaluate the effectiveness of ISIS EM-BLASSO method alongside the three methods for comparison purposes. For each QTN, we defined its statistical power as the fraction of the samples in which the QTN was detected (see Methods for significant testing). Fig 1A, 1B and 1C represents the results of the statistical power of detecting each QTN from the three simulation experiments respectively. In the first simulation experiment in which no polygenic variance was simulated, ISIS EM-BLASSO method has the highest power for detecting almost all the six simulated QTNs except QTN two. mrMLM has a high power of detecting the second QTN. EMMA and FarmCPU have a low power of detecting the second and fourth simulated QTNs (Fig 1A and S1 Table). Indeed, Bonferroni correction is too stringent, and it may cause many significant loci to be missed out. SCAD has a moderately higher power than EMMA and FarmCPU for the second, third and sixth QTNs. SCAD lacks consistency in detecting the simulated QTNs hence it cannot be relied upon especially when the QTN size is small. The same trends were observed in the second simulation experiment (Fig 1B and S2 Table) when an additive polygenic variance was added to the polygenic background. In the third simulation experiment (Fig 1C and S3 Table) where three pairs of epistatic effects (collectively contributing 0.15 to the phenotypic variance) were added to the genetic background, ISIS EM-BLASSO is still powerful in detecting almost all the six simulated QTNs. We presented a paired t- test for the differences in statistical power (Table 1). We observe that there are significance differences (at the 0.05 level) in statistical power between ISIS EM-BLASSO and the other three methods (SCAD, EMMA, FarmCPU). There are no significance differences in statistical power between ISIS EM-BLASSO and mrMLM except in the third simulation (at the 0.1 level). Based on these findings, it implies that the simulated QTNs are mostly likely to be identified when ISIS EM-BLASSO method is used.
Mean squared error (MSE) was used to measure the accuracy of each estimated QTN effect for all the methods. We evaluated the accuracies of all the six simulated QTNs effects in the three simulation experiments. The results of all methods considered are presented in Fig 2 and S1, S2 and S3 Tables. The ISIS EM-BLASSO method is consistently more accurate in estimating the QTN effects than the other methods (EMMA, SCAD, and FarmCPU). From these results, EMMA has the highest MSEs for each of six simulated QTNs, implying it is inaccurate in estimating the QTN effect. ISIS EM-BLASSO has lower MSEs than mrMLM for simulated QTNs 4, 5 and 6 in all the three simulations. This implies that it is reliable for estimating the QTN effects. At the 0.05 significance level, the differences of MSEs between ISIS EM-BLASSO and other methods (EMMA and SCAD) were significant (Table 1). Applying EM-Bayesian LASSO to SNPs selected from the iterative procedure will not only remove unimportant SNPs but also improves the effect estimation.
Type 1 errors for all the methods in the three simulation experiments were calculated (Fig 3). The values for Type 1 errors for each method in each simulation are recorded in S1, S2 and S3 Tables. Despite having high power in the detection of QTNs, ISIS EM-BLASSO had slightly higher Type 1 errors compared with SCAD, EMMA, FarmCPU, and mrMLM. Note that all the Type 1 errors were less than 0.04%. Bonferroni correction eliminates many un-associated loci hence reducing the false positive rate in FarmCPU and EMMA at the expense of some associated SNPs. Conversely, this study reveals that ISIS EM-BLASSO method may slightly include some un-important SNPs in the model than SCAD, EMMA, FarmCPU and mrMLM though with a higher power in detecting associated QTNs.
Receiver operating characteristic (ROC) curve is used to compare different methods for their efficiencies in the detection of significant effects. ROC curve is a plot of the statistical power against the controlled Type 1 error. A method with the highest ROC curve is deemed the best. We simulated various 67 probability levels of significance between 1e-8 to 1e-2, with these values we calculated the corresponding powers in the first simulation experiment. Fig 4 shows a comparison of the ROC curves from the four methods for the first experiment for each of the six QTNs simulated. ISIS EM-BLASSO performs best among all the other methods considered for all the simulated QTNs.
Comparing the computing times (Fig 5 and S1, S2 and S3 Tables) of these methods in the three simulation experiments respectively, we observed that ISIS EM-BLASSO has the lowest computing time whereas EMMA takes a longer computing time (Intel Core i5-4570 CPU 3.20GHz, Memory 7.88G). ISIS EM-BLASSO is computationally efficient and can be used in GWAS in a few hours to obtain the associated genes. ISIS EM-BLASSO in itself reduces the number of SNPs to only those that are significantly correlated with the phenotype and hence this reduces the problem to a moderate high dimensional data setting problem saving on the computational time.
Six Arabidopsis flowering time traits in Atwell et al. [27] have been re-analyzed by ISIS EM-BLASSO, EMMA, FarmCPU, and mrMLM. These traits are LD, LDV, SD, 0W, 2W, and 4W. ISIS EM-BLASSO detected 14, 11, 23, 21, 9 and 11 SNPs to be significantly associated respectively with the six traits above. These detected SNPs for each trait were used to conduct a multiple linear regression analysis, and the corresponding AIC and BIC were calculated. The ISIS EM-BLASSO method showed low AIC and BIC values for nearly all traits (S4 Table). The only method that compares almost equally with ISIS EM-BLASSO is mrMLM. This indicates that SNPs detected by ISIS EM-BLASSO fit the data better than the other methods.
The numbers of known genes in the proximity of SNPs for the above six traits were in total 67 genes from ISIS EM-BLASSO, 22 genes from mrMLM, 15 genes from FarmCPU, and 13 genes from EMMA. ISIS EM-BLASSO detected more known genes than the other methods (S5 Table). Among these known genes, 50 were identified only by ISIS EM-BLASSO (Table 2).
It is interesting to note that for the trait SD, EMMA was not able to determine any significant gene whereas the new approach identified 21 genes. A similar trend is also observed when we considered trait 0W where FarmCPU did not detect any gene. Based on these results, we observe that the new approach can capture the genes associated with the trait under study. We also noted that some genes tested were found significant in nearly all the traits considered. For example, gene DOGI was discovered to be associated with LDV, SD, 0W, and 4W, gene SVP was found to be related to LD and LDV, gene ETC3 was found to be associated with LD, 0W, and 2W, and gene ABCB19 was discovered to be related to SD, 2W, and 4W. These results are consistent with previous studies related to these traits as outlined in the references presented in S5 Table.
Based on the results obtained from this study, we observe that correlation learning can be used as a screening tool to reduce the number of markers in GWAS study. As already noted, most methods used in variable selection in linear regression fail in high and ultra-high dimensional settings. This study has presented a simple yet powerful tool to solve this problem. Therefore, the Arabidopsis thaliana GWAS results of this study are reliable.
The single locus tests in standard GWAS methods have been used successfully in identifying thousands of genetic variants associated with hundreds of various traits. As noted by Segura et al. [28], when we carry out single-locus tests of association, we risk using the wrong model unless the trait is actually due to a single locus. Single locus tests fail to consider the joint effect of multiple genetic markers on traits. They also suffer from the issue of multiple test correction for the threshold value of significance test. The Bonferroni correction is too stringent, and many significant loci are missed out. This calls for multi-locus testing in GWAS. Because only a subset of SNPs is associated with the phenotype, penalized variable selection methods are appropriate in GWAS. Several penalized methods have been used in solving this problem although they fail when the number of variables, p, is several times larger than the sample size (n).
In this article, we developed an algorithm, ISIS EM-BLASSO that screens and significantly reduces the number of SNPs to a moderate number. A moderate-scale variable selection method was used to select variables in the reduced set. We chose SCAD since it is asymptotically oracle efficient. Parameter estimation and significance testing were done in the last stage by applying EM-Bayesian LASSO [25] and likelihood ratio test. This algorithm is based on correlation learning in the screening stage. Our approach lays emphasis on the significance of the correlation between SNPs and the trait of interest in the screening stage. It is only the SNPs that are significantly correlated with the trait that are selected in the screening stage. Hence we do not need to subjectively fix the number of SNPs in the screening stage as in Fan and Lv [14] method. ISIS EM-BLASSO differs with the original sure independence screening method by how screening is done. Secondly, it integrates EM- Bayesian LASSO algorithm [25] in the final stage to select and estimate effects. We compared the results of our new approach with results obtained from EMMA [26], FarmCPU [24] and SCAD [7] methods. The reasons are varied: EMMA [26] has been a standard gold method for GWAS, FarmCPU [24] just like ISIS EM-BLASSO also reduces the number of markers used in GWAS, SCAD [7] is actually used in the screening method of ISIS EM-BLASSO hence the need to compare it independently and mrMLM [20] integrates single locus with multi-locus approach. ISIS EM-BLASSO is the fastest as compared to these other methods. ISIS EM-BLASSO only takes 3% of the computing time needed by the EMMA method, 16% of the time taken by mrMLM, 20% of the time taken by SCAD and 50% of the time taken by FarmCPU methods. Screening stage reduces the number of SNPs to an average number hence less time taken in the overall process. More importantly, ISIS EM-BLASSO performs generally the best; it has high statistical power, low Type 1 error and low MSE of estimated QTN effects (S1, S2 and S3 Tables). This algorithm improves the estimation of parameters because even in the last stage, EM-Bayesian LASSO still performs variable selection shrinking other SNPs to zero hence significantly improving the estimates and empirical power. Notice that EM-Bayesian LASSO [25] performs effectively in the last stage because the number of SNPs has been reduced considerably. Just like SCAD, EM-Bayesian LASSO will fail when the number of SNPs runs into hundreds of thousands. For this reason we did not perform tests for EM-Bayesian LASSO independently. Combining sure independence screening, SCAD, and EM-Bayesian LASSO improves the results regarding of empirical power, accuracy, and computational efficiency.
In the screening stage of ISIS EM-BLASSO, iterative sure independence screening can be performed several times. In the article, we only performed a single iteration since we chose a high level of significance, 0.01, for identifying predictors that are significantly correlated with the response. At this level of significance, we expect to have moderately more variables in the screening stage so that SCAD can be applied to shrink some of these variables to zero. A lower level of significance and/or multiple test correction in the screening stage might be too stringent and may result in missing significant SNPs at this juncture. Most shrinkage methods will still perform effectively even when the number of variables, p, is moderately greater than or equal to the number of observations, n. Therefore the choice of a level of significance of 0.01 without any multiple test correction is justified. Note that if several iterations are performed, then many unimportant variables are selected. Even so, this algorithm is still valid because the extension of this method with EM-Bayesian Lasso [25] in the last stage eliminates un-associated. Other multi-locus GWAS approaches such as multi-locus mixed model (MLMM) [28] have been studied before. MLMM approach of Segura et al. [28] is inadequate since its greedy forward-backward method inclusion of SNPs is clearly limited in exploring the huge model space.
ISIS EM-BLASSO considers the joint effect of all SNPs that passes the screening criterion. Unlike EMMA and FarmCPU, we do not apply the Bonferroni correction for multiple testing hence ISIS EM-BLASSO performs better than these methods regarding statistical power. Bonferroni correction is indeed too stringent hence it removes some significant SNPs in the final results when EMMA and FarmCPU are used. Although we still used a slightly stringent criterion of LOD value 3 in our final stage, ISIS EM-BLASSO still has high statistical power and low false positive rate, indicating a better performance of the new algorithm over EMMA and FarmCPU. mrMLM compares almost equally with ISIS EM-BLASSO regarding power and MSE (S1, S2 and S3 Tables). mrMLM is also a two stage method hence the similarity. Nevertheless, ISIS EM-BLASSO is still faster than mrMLM. The results obtained here also demonstrate that many penalized methods fail when the number of SNPs is many times larger than sample size as seen from results obtained when SCAD is used to analyze data.
ISIS EM-BLASSO was used to analyze six flowering time traits in this study. As a result, 67 known genes were detected. Among these known genes, 50 were identified only by ISIS EM-BLASSO (Table 2). Many genes obtained by our approach are in the neighborhood of the 89 SNPs detected (S5 Table). These results are consistent with those previously reported, as shown in the database (http://www.arabidopsis.org/) the work of Atwell et al. [27] and other related references in S5 Table. Atwell et al. [27] listed many significantly associated SNPs to these traits though some of them were not significant at the 0.05/m criterion. Therefore, the Arabidopsis thaliana GWAS results presented by our algorithm in this study are reliable.
We considered an iterative modified sure independence screening (ISIS) and extended it by applying EM Bayesian LASSO algorithm herein referred to as ISIS EM-BLASSO. This approach was used to identify relevant genes in GWAS study in real data. The new approach is simple, fast and shows high statistical power of detecting relevant SNPs on simulated data. Mean squared errors of the estimated effects are also minimal. We recommend this approach as an accurate and fast alternative for carrying out multi-locus GWAS study especially in high dimensional settings. ISIS EM-BLASSO reduces the search to a moderate number of SNPs that are significantly correlated with the trait of interest. As a result, we reduce the computing time for GWAS, and also ensure that GWAS can be carried out on a small computer. Our extension by EM-Bayesian LASSO ensures that the parameter estimates are reliable.
In this study, we considered the regression model,
y=μ+∑j=1qQjαj+∑i=1pXiβi+ε
(1)
with y being a n×1 vector of phenotypic quantitative trait, μ is the overall average, Qj is the jth fixed effect which must be included in the model, for example, the population structure, Xi is a n×1 vector of the ith SNP values and ε∼MVNn(0,σ2I) is the residual error.
Given the model in Eq (1) we corrected y for the fixed effects Qj, j = 1,2,…,q before applying the screening procedure. The effects α^j are obtained by ordinary least squares. Eq (1) then becomes, yc=y−∑j=1qQjαj=μ+∑i=1pXiβi+ε and without loss of generality, this can just be denoted as
y=μ+∑i=1pXiβi+ε
(2)
ISIS-EM-BLASSO gives the marker effects estimates, γk which must be tested for their significance to the phenotype under study. We propose that all γ^k≠0 k = 1,2,⋯,p could be viewed to be associated with the trait under study. With the model (10) above at hand, and all γ^k≠0 let’s say of size q, we consider the model’s likelihood functions. Let L0=L(θ^−k) and L1=L(θ^1) be short expressions of the natural logarithms of the likelihood functions under the null model and the full model, respectively, with θ^−k={γ(1),⋯,γ(k−1),γ(k+1),γ(q)} and θ^1={γ(1),⋯,γ(q)}.
In essence, we test the null hypothesis, H0:γ(k) = 0 that there is no QTN linked to the marker k. We use log of odds (LOD) score as the test statistic. The original likelihood functions (before taking the natural log) are l0=eL0 and l1=eL1, respectively. The LOD score is defined as
LOD=log10(l0l1)=−2(L0−L1)4.6052
(11)
The LOD score is easy to interpret because of base 10. We select all markers with a score LOD ≥ 3 and regard them as significant. Note that LOD value 3 is a slightly stringent criterion and is equivalent to p-value: Pr(χ2 > 3×4.6052) = 0.0002. If the null hypothesis is true, LOD × 4.6052 follows a Chi-square distribution with one degree of freedom. A similar procedure is used to test the significance of the estimates obtained from SCAD and mrMLM.
It is important to point out that, for the EMMA and FarmCPU methods we select significant markers based on Bonferroni correction for multiple tests by setting a threshold for P-value at 0.05/m, where m is the number of markers.
|
10.1371/journal.pgen.1007757 | Evidence for gene-environment correlation in child feeding: Links between common genetic variation for BMI in children and parental feeding practices | The parental feeding practices (PFPs) of excessive restriction of food intake (‘restriction’) and pressure to increase food consumption (‘pressure’) have been argued to causally influence child weight in opposite directions (high restriction causing overweight; high pressure causing underweight). However child weight could also ‘elicit’ PFPs. A novel approach is to investigate gene-environment correlation between child genetic influences on BMI and PFPs. Genome-wide polygenic scores (GPS) combining BMI-associated variants were created for 10,346 children (including 3,320 DZ twin pairs) from the Twins Early Development Study using results from an independent genome-wide association study meta-analysis. Parental ‘restriction’ and ‘pressure’ were assessed using the Child Feeding Questionnaire. Child BMI standard deviation scores (BMI-SDS) were calculated from children’s height and weight at age 10. Linear regression and fixed family effect models were used to test between- (n = 4,445 individuals) and within-family (n = 2,164 DZ pairs) associations between the GPS and PFPs. In addition, we performed multivariate twin analyses (n = 4,375 twin pairs) to estimate the heritabilities of PFPs and the genetic correlations between BMI-SDS and PFPs. The GPS was correlated with BMI-SDS (β = 0.20, p = 2.41x10-38). Consistent with the gene-environment correlation hypothesis, child BMI GPS was positively associated with ‘restriction’ (β = 0.05, p = 4.19x10-4), and negatively associated with ‘pressure’ (β = -0.08, p = 2.70x10-7). These results remained consistent after controlling for parental BMI, and after controlling for overall family contributions (within-family analyses). Heritabilities for ‘restriction’ (43% [40–47%]) and ‘pressure’ (54% [50–59%]) were moderate-to-high. Twin-based genetic correlations were moderate and positive between BMI-SDS and ‘restriction’ (rA = 0.28 [0.23–0.32]), and substantial and negative between BMI-SDS and ‘pressure’ (rA = -0.48 [-0.52 - -0.44]. Results suggest that the degree to which parents limit or encourage children’s food intake is partly influenced by children’s genetic predispositions to higher or lower BMI. These findings point to an evocative gene-environment correlation in which heritable characteristics in the child elicit parental feeding behaviour.
| It is widely believed that parents influence their child’s BMI via certain feeding practices. For example, rigid restriction has been argued to cause overweight, and pressuring to eat to cause underweight. However, recent longitudinal research has not supported this model. An alternative hypothesis is that child BMI, which has a strong genetic basis, evokes parental feeding practices (‘gene-environment correlation’). To test this, we applied two genetic methods in a large sample of 10-year-old children from the Twins Early Development Study: a polygenic score analysis (DNA-based score of common genetic variants associated with BMI in genome-wide meta-analyses), and a twin analysis (comparing resemblance between identical and non-identical twin pairs). Polygenic scores correlated positively with parental restriction of food intake (‘restriction’; β = 0.05, p = 4.19x10-4), and negatively with parental pressure to increase food intake (‘pressure’; β = -0.08, p = 2.70x10-7). Associations were unchanged after controlling for all genetic and environmental effects shared within families. Results from twin analyses were consistent. ‘Restriction’ (43%) and ‘pressure’ (54%) were substantially heritable, and a positive genetic correlation between child BMI and ‘restriction’ (rA = 0.28), and negative genetic correlation between child BMI and ‘pressure’ (rA = -0.48) emerged. These findings challenge the prevailing view that parental behaviours are the sole cause of child BMI by supporting an alternate hypothesis that child BMI also causes parental feeding behaviour.
| The home and family environment has been studied for decades with the assumption that it is a crucial determinant of children’s health and development. Since the onset of the childhood obesity crisis at the turn of the century, the spotlight has turned onto environmental factors associated with variation in adiposity, in the hope that modifiable elements may be identified as intervention targets. Perhaps unsurprisingly, parental behaviours have received a great deal of attention. Parents are widely considered to be the ‘gatekeepers’ to their children’s food, and powerful shapers of their developing eating behaviour[1–3]. Two types of parental feeding practices (PFPs) in particular have been hypothesised to play a causal role in children’s ability to develop good self-regulation of food intake and consequently determine their weight. Excessive restriction of the type and amount of food a child is allowed to eat (‘restriction’) has been hypothesised to lead to overeating when parental restriction is no longer in place, because the child will potentially then hanker after the foods he or she is not usually allowed to eat–the so-called ‘forbidden fruit effect’[1,4,5]. On the other hand, overly pressuring a child to eat, or to finish everything on the plate (‘pressure’), is thought to be anxiety-provoking for a child with a poor appetite, and serves only to increase undereating further, and compromise weight gain[6,7].
A wealth of cross-sectional findings are consistent with these hypotheses[8], but another plausible explanation for the observed correlations is that parents are responding to their child’s emerging characteristics, not simply causing them. Parents may only adopt restrictive strategies when a child shows a tendency toward overeating, or gains excessive weight; and they may pressure their child to eat only if he or she is a poor eater, or has underweight. The few longitudinal studies testing bidirectionality have shown that children’s weight prospectively predicts PFPs[9–13]. Furthermore, three studies showed no prospective association from PFPs to child weight[10], and the studies reporting bidirectional relationships found stronger associations from child weight to parental behaviour than the reverse direction[9,11]. Although these findings point towards children’s weight eliciting PFPs, the possibility of residual confounding in observational studies hinders conclusions about causation–temporality does not necessarily mean causality.
Testing whether children genuinely cause their parents’ behaviour presents challenges. It is not possible–practically or ethically–to randomise children to have overweight or underweight, and examine how parents respond. Genetic approaches provide a powerful alternative method of interrogating the role of children in causing their parents’ behaviour towards them, especially for child characteristics with an established genetic basis. To date, no study has applied genetically sensitive methods to test for gene-environment correlation in parental feeding behaviour. Family and twin studies have shown that Body Mass Index (BMI), is highly heritable in both adulthood and late childhood (~70%)[14–16]. Twin designs can also be used to test if parental behaviour has a heritable component, by comparing within-pair resemblance for identical and fraternal twin pairs in childhood. If found, this indicates that parental behaviour is explained to some extent by variation in children’s genotype–termed evocative gene-environment correlation[17]. Twin designs can also be extended to the analysis of multiple variables to establish if genetic influence on a particular child characteristic (e.g. weight) also predicts the parental behaviour of interest (e.g. PFPs). If such analyses show that a child characteristic is genetically correlated with parenting traits, it indicates that these child characteristics influence parenting behaviours. A meta-analysis of 32 twin studies of different types of parenting behaviour reported an average heritability estimate of 24%, indicating that children’s genotype is predictive of a moderate amount of variation in parental behaviour[18].
Children’s DNA can also be used to test for gene-environment correlation. Genome-wide meta-analyses have made great progress in identifying common single nucleotide polymorphisms (SNPs) that are associated with body mass index (BMI) in adults and children[19]. These can be combined to calculate a genome-wide polygenic score (GPS) that indexes individual-specific propensity to higher or lower BMI, along a continuum, although in the aggregate the GPS explains only a small proportion of variance in BMI (approximately 3%)[20]. Nevertheless, children’s BMI GPS can therefore be used to test the hypothesis that parents develop their feeding practices specifically in response to their child’s weight, as indicated by a correlation between child BMI GPS and PFPs. A caveat to this is that a parent’s feeding practices may reflect their own genetic predisposition to be of a higher or lower BMI, rather than that of their children. In this way, a correlation between child BMI GPS and PFPs may simply reflect a child’s genetic predisposition to be of a higher or lower BMI, which they inherit from their parent with whom they share 50% of their DNA. In addition, genetic effects related to adult BMI discovered in genome-wide association studies could potentially incorporate effects of PFPs if they were to causally influence child BMI, and its trajectory into adulthood. However, within-family designs can circumvent both of these limitations to some extent. Studying variation in PFPs according to variation in BMI GPS within non-identical co-twins accounts for both genetic and environmental shared effects within families (e.g. parental genetic predisposition to be of higher or lower BMI). By applying both quantitative and molecular genetic methods, and utilising statistical approaches to account for shared family effects, we intended to address the various limitations presented by the individual methods.
The goals of this study were to test for gene-environment correlation between children’s BMI and PFPs, using a twin design and a BMI GPS. We hypothesised that: (i) children’s BMI GPS would be positively associated with parental restriction and negatively associated with parental pressure, even after accounting for shared genetic and environmental family influences; and (ii) parental restriction and parental pressure would be moderately heritable, and that genetic influence on PFPs would be partly explained by genetic influence on children’s BMI.
Child BMI-SDS was significantly positively correlated with ‘restriction’ (β = 0.19, t(4004) = 12.09, p = 4.45x10-33, R2 = 0.035), such that parents were more restrictive over their child’s food intake if the child had a higher BMI. In contrast, child BMI-SDS was significantly negatively correlated with ‘pressure’ (β = -0.24, t(4058) = -15.59, p = 3.14x10-53, R2 = 0.056), such that parents exerted higher amounts of pressure on their child to eat, if their child was leaner. ‘Restriction’ and ‘pressure’ were significantly positively correlated (β = 0.15, t(4207) = 9.51, p = 3.08x10-21, R2 = 0.021), suggesting that parents who tend to exert higher levels of ‘restriction’ also exert a more pressuring feeding style, to some extent.
In our sample of unrelated individuals, child BMI GPS was positively correlated with child BMI-SDS (β = 0.20, t(4226) = 13.08, p = 2.41x10-38, R2 = 0.039). Mirroring phenotypic results for child BMI-SDS, children’s BMI GPS was significantly positively correlated with ‘restriction’ (β = 0.05, t(4255) = 3.53, p = 4.19x10-4, R2 = 0.003), and significantly negatively correlated with ‘pressure’ (β = -0.08, t(4315) = -5.15, p = 2.70x10-7, R2 = 0.006) (Fig 1). These findings indicate that children’s genetic predisposition to higher BMI, elicits, to some extent, restrictive feeding behaviours in the parent; whereas children’s genetic predisposition to lower BMI elicits greater pressure to eat by parents.
Parental BMI correlated positively with child BMI-SDS (β = 0.26, t(3761) = 17.00, p = 1.57x10-62, R2 = 0.071) and ‘restriction’ (β = 0.08, t(3711) = 4.64, p = 3.65x10-6, R2 = 0.005), but was not significantly associated with ‘pressure’ (β = -0.03, t(3757) = -1.68, p = 0.09, R2 < 0.001). The magnitude and direction of effects remained identical after controlling for parental BMI in ‘restriction’ (β = 0.05, t(3711) = 2.92, p = 3.48x10-3, R2 = 0.003) and in ‘pressure’ (β = -0.08, t(3757) = -4.62, p = 3.97x10-6, R2 = 0.005).
To establish the association between children’s BMI GPS and PFPs entirely without confounding by genetic and environmental family factors shared by twin pairs, we performed family fixed-effect analyses in dizygotic (DZ) co-twins. This analysis examined the extent to which parents vary their ‘restriction’ and ‘pressure’ across twin pairs in response to differences in their BMI GPS. As shown in Fig 2, beta coefficients for BMI GPS predicting PFPs remained largely stable when comparing unrelated individuals (Model 1) and DZ twin pairs (Model 2). For unrelated individuals (Model 1) child BMI-SDS significantly positively predicted ‘restriction’ and significantly negatively predicted ‘pressure’, as previously reported. The magnitude of the within-family estimates for the combined (same-sex and opposite-sex) DZ co-twins (Model 2) were virtually the same as those for the unrelated individuals for the relationships between BMI GPS and ‘restriction’ (t(2054) = 3.50, p = 7.10x10-3, Adj. R2model = 0.724) and BMI GPS and ‘pressure’ (t(2103) = -4.82, p = 1.52x10-6, Adj. R2model = 0.641) (R2 magnitudes for Model 2 are large because all shared factors among family members, including genetic and environmental influences, are accounted for). These findings indicate that even when shared family effects are completely accounted for, children’s BMI GPS is significantly associated with PFPs, providing additional evidence that children’s genetic predisposition to BMI evokes certain parental feeding responses. When repeating Model 2 analyses separately for same-sex and opposite-sex DZs, magnitudes of effect sizes (Fig 2) remained consistent for the prediction of ‘pressure’ in same-sex DZ pairs (t(1118) = -3.36, p = 8.02x10-4, Adj. R2model = 0.607) and opposite-sex DZ pairs (t(984) = -3.49, p = 5.12x10-4, Adj. R2model = 0.678). Although BMI GPS in opposite-sex DZs was a significant predictor of within-family differences in ‘restriction’ (t(966) = 3.76, p = 1.82x10-4, Adj. R2model = 0.731), same-sex DZ data did not show a significant within-family association (t(1087) = 1.21, p = 0.23, Adj. R2model = 0.719), indicating that within a family environment, GPS differences in BMI between same-sex DZ twins are not related to differences in parental ‘restriction’.
We performed multivariate genetic analyses (a correlated factors model) to establish the heritability of ‘restriction’ and ‘pressure’ and to test the extent to which genetic influence on child BMI-SDS elicited PFPs as indicated by the magnitude of genetic correlations between BMI, ‘restriction’, and ‘pressure’. Fig 3 shows the variance components (A, C and E) for each measured phenotype, as well as the genetic, shared environmental and non-shared environmental correlations between phenotypes derived from the correlated factors model (see Supplementary S1 Table for fit statistics and model comparisons, and Supplementary S2 Table for intra-class correlations). Heritability estimates (A) were moderate to high for parental ‘restriction’ (43%, 95% CI [40%, 47%]) and parental ‘pressure’ (54%, 95% CI [50%, 59%]); heritability of child BMI-SDS was high (78%, 95% CI [72%, 84%]). Consistent with the findings from the GPS analyses, there was a significant, positive moderately sized genetic correlation between child BMI-SDS and parental ‘restriction’ (rA = 0.28, 95% CI [0.23, 0.32]), indicating that some of the genetic effects that predispose a child to a higher BMI also elicit more food restriction by their parent. A sizeable significant negative genetic correlation was observed between child BMI-SDS and parental ‘pressure’ (rA = -0.48, 95% CI [-0.52, -0.44]), indicating that many of the genetic effects that predispose a child to a lower BMI elicit greater parental pressure on the child to eat.
As shown in the twin analyses (Fig 3 and Supplementary S3 Table), variation in child BMI-SDS is partly caused by non-shared environmental influences, which correlate significantly with non-shared environmental influences for ‘restriction’ (rE = 0.20) and ‘pressure’ (rE = -0.29). We therefore performed MZ twin difference analyses to examine these relationships more closely. In contrast to child BMI-SDS MZ difference scores, most twins did not differ in their PFP (Supplementary S1 Fig). Nevertheless, we found that child BMI-SDS difference scores predicted both differences in ‘restriction’ (β = 0.14, t(1484) = 7.98, p = 2.88x10-15, R2 = 0.041) and ‘pressure’ (β = -0.25, t(1498) = -12.26, p = 5.12x10-33, R2 = 0.09). These findings suggest that there are common non-shared environmental sources of variance for both PFP and child BMI; within identical twin pairs who share 100% of their genetic and shared environmental influence, parents apply more restrictive feeding practices on the twin with the higher BMI, and more pressuring feeding practices on the twin with the lower BMI score.
We describe the first study to test for gene-environment correlation for parental feeding behaviour in relation to child weight, using a twin design and children’s DNA. Results support our hypothesis that parents’ feeding practices are evoked, in part, by their children. Parental ‘restriction’ and ‘pressure’ were positively and negatively associated with child BMI respectively, in keeping with many previous cross-sectional studies[8]. We applied novel genetic methods to show, for the first time, that children’s BMI GPS was significantly positively associated with ‘restriction’ and negatively associated with ‘pressure’, even after accounting for the potentially confounding shared familial effects (both genetic and environmental). This suggests that children’s genetic influence on weight explains part of the observed phenotypic associations. Our twin analysis provided quantitative estimates of the total variance in parental feeding practices explained by children’s genotype. Heritability was substantial for both ‘restriction’ (43%) and ‘pressure’ (54%), indicating that children’s genes explain about half of the variation in parental feeding behaviour. Multivariate twin analysis established the extent to which parental feeding behaviour was determined by children’s genetic influence on BMI specifically. The genetic correlations between children’s BMI and both ‘restriction’ (rA = 0.28) and pressure (rA = -0.48) were moderate, indicating overlap between the genes that influence parental feeding behaviour and children’s BMI.
A potential confounder of the association between child GPS and parental feeding behaviour, was the parent’s own genetic propensity to a higher or lower BMI. Children inherit half of each of their parents’ genetic material, so the expected correlation between a child’s GPS with that of their parent’s is 0.50. A parent’s genetic predisposition to be of a higher or lower BMI may also influence the way they feed their children, which could introduce a passive (rather than ‘evocative’) gene-environment correlation. For example, a parent with a higher BMI may be more restrictive over their child’s food intake, but their child also inherits their parent’s susceptibility to be of a higher BMI. Restrictive feeding may therefore simply be a marker for a child’s genetic predisposition to be of a higher BMI that is transmitted to them by their parent, rather than a causal risk factor (the same could be true for a more pressuring feeding style and lower BMI). In line with this, parental BMI (indexing parental GPS) was significantly positively associated with parental restriction indicating that parents of a higher weight exert greater restriction over their children’s food intake (β = 0.08); although the association with parental pressure was not significant. Adjustment for parental BMI did not attenuate the associations between child GPS and either restriction or pressure, suggesting it was not confounding the relationship between parental feeding behaviour and child BMI GPS. Nevertheless, adjustment for parental BMI cannot completely remove confounding from parental BMI, nor can it account for the potential effect of longer-term BMI on parental feeding behaviours. However, in order to rule out confounding by any parental characteristics (both genetic and environmental), we took advantage of a family fixed-effect design, which held the effects of family constant while testing the association between the child BMI GPS and parental feeding practices in DZ co-twins. The within-family analysis allowed us to demonstrate that even after accounting for all genetic and environmental familial effects, parents vary their feeding behaviour for each child depending on their GPS–larger GPS differences between pairs were associated with more pronounced differences in parental feeding behaviour. The magnitudes of the between- and within-family associations between parental feeding behaviour and child GPS were virtually the same, with the exception of the relationship between child GPS and ‘restriction’ in same-sex twins, strengthening the evidence that children evoke parental responses based on their genetic predispositions for BMI. Nevertheless, as expected, and consistent with the small amount of variance explained in BMI by the GPS, the size of the associations between the BMI GPS and PFPs were small.
The findings from this study accord with those from twin studies of many other types of parenting behaviours that have also tended to show moderate heritability. A meta-analysis of 32 child twin studies on maternal positivity, negativity, affect and control in relation to parenting showed an average heritability of 24%[18], indicating widespread, child-driven genetic influences on parental behaviour. The heritability estimates for ‘restriction’ (43%) and ‘pressure’ (54%) were somewhat higher than the average heritability estimate for the parenting styles considered in the meta-analysis (24%), but in keeping with the magnitude of the heritability of negative parenting styles observed across early childhood (~55%)[21].
In addition to providing evidence for gene-environment correlation, results from the MZ discordance design also indicated that non-shared environmental influences for child BMI and PFPs are correlated as well. This suggests that child BMI and PFPs are also related due to common non-shared environmental influences. However, the MZ discordance design was not able to shed light on the causal direction–i.e. if child BMI causes PFPs or if PFPs cause child BMI–because our variables were measured at the same time. The few prospective studies that have attempted to establish the cause-effect relationship in the parent-child dynamic using bidirectional analyses have suggested either only a small effect of restriction and/or pressure on child weight, or none[9–11,13]. Prospective studies therefore suggest that PFPs may be less important than is commonly assumed. The well-established strong genetic influence on children’s weight–in the order of 70–80%[15,16]–also supports the hypothesis that parents influence child weight via genetic inheritance more than by creating an ‘obesogenic’ family environment. However, it cannot be ruled out that genetic effects related to BMI in the parents also contribute to an obesogenic environment if gene-environment correlation was at play, further passively reinforcing the child’s inherited genetic propensities. The shared environmental influence on BMI in late childhood is also low[15,16]. In the current study, the shared environmental influence on parental feeding behaviour was the proportion of variance that was common to both twins in a pair (invariant within families). It therefore likely reflects variation in feeding behaviour that was parent-driven rather than child-directed. These estimates indicated that a substantial proportion of variation in both ‘restriction’ (C = 43%) and ‘pressure’ (C = 37%) also originated in the parent.
Experimental studies in the form of large well-designed randomised controlled trials (RCTs) are needed to truly test the hypothesis that PFPs causally modify children’s weight gain trajectories. Very few of these have been conducted to date, and they have focused on the preschool years. Nevertheless, two landmark studies have indicated that parental behaviour may, in fact, be influential in early life. NOURISH[22] was an Australian RCT that randomised 352 parents and infants to receive a feeding intervention (including using low amounts of pressure, and employing child-responsive methods of food restriction) during the period of complementary feeding; 346 families were randomised to the standard care control group. At three to four years of age, children in the intervention group had better appetite control than those in the control group, and there were fewer children with overweight; although this did not reach statistical significance[23]. INSIGHT[24], a US RCT, randomised 145 new mothers to a responsive parenting intervention that focused on feeding infants only in response to their hunger and satiety signals (but neither pressuring nor restricting their milk and food intake), during milk-feeding and complementary feeding; 145 mothers were randomised to a control group. At one year significantly fewer infants in the intervention group had overweight (6%) compared to the control group (13%). These RCTs indicate that parental feeding behaviour can modify young children’s eating behaviour and weight gain. However, these studies were conducted in infants and young preschool children so it is unclear whether these findings are generalisable to older children.
The genetic correlations between children’s BMI and parental feeding behaviour were modest, and were far from complete (i.e. less than 1.0), indicating that other genetically-determined child characteristics are also influencing parental feeding behaviour. Children’s appetite is under strong genetic control; twin studies–including this sample–have shown high heritability for appetite[25,26] and shared heritability with BMI[27]. Appetite is associated with the BMI GPS in this sample and has been shown to mediate part of the GPS-BMI association[28]. It is therefore likely that child appetite also influences parental feeding behaviour[25,26]. In support of this, prospective and within-family studies have provided evidence that within the context of parental feeding, parents respond not only to their child’s weight but also to their eating styles. A large prospective population-based study used bidirectional analyses to show that parents whose children were excessively fussy at baseline increased their pressure over time[29]. A reverse relationship also pertained, but the temporal association from child to parent was stronger. A large within-family study of preschool twins showed that parents varied their pressuring feeding style when their twins were discordant for food fussiness[30]. The fussier twin was pressured more than their co-twin, also in support of a child-driven model of parental feeding behaviour. It stands to reason that a child who is a picky eater is pressured to try some of their vegetables or to eat more overall. Along the same lines, a natural response from a parent who has a child who shows a tendency toward excess intake and a relatively pronounced preference for foods rich in sugar or fat, is to enforce some restriction.
We also found a positive phenotypic correlation between ‘restriction’ and ‘pressure’ (β = 0.15), indicating that parents who exert higher levels of restriction on their children also tend to pressure them more. This suggests that some parents have a more controlling feeding style in general.
The relationship between parental behaviour and children’s emerging characteristics appears to be reciprocal and complex. The current findings suggest that parents’ natural feeding responses to child weight are to exert greater restriction of food intake on children with a higher BMI, and to pressure a thinner child to eat. However, these strategies may not be effective in the long run. RCTs have suggested that PFPs can have a lasting and important impact on children’s weight and eating behaviour in the early years, although whether or not these findings apply to older children has yet to be determined. It is well established that genetic influence on BMI in younger children is lower, and the shared environmental effect is higher, than it is in older children[15,16]. This suggests that parental influence diminishes as children grow older, gain independence and spend increasing time outside the home with peers rather than parents[31]. Large RCTs that follow children from early life to later childhood are needed to establish if PFPs influence the weight of older children.
A strength of this study is that we used several genetically sensitive methodological approaches to explore the directionality of relationships between child BMI and PFPs, yielding consistent results. PFPs were measured using the Child Feeding Questionnaire, which has well established criterion and construct validity, as well as good internal and test-retest reliability[32]. This instrument has been used widely in previous research into child weight, allowing the findings from this study to be directly compared to a wealth of existing results.
A potential limitation is that heritability estimates from twin studies rely on the assumption that MZs and DZs share their environment in terms of the trait in question to the same extent, so-called the ‘equal environments assumption’; if this is violated, the findings are invalid. Therefore if parents feed MZs more similarly than DZs simply because they are identical, this would artificially inflate the MZ correlation and, consequently, heritability. However, if MZs are fed more similarly than DZs because parents are responding to their genetically determined BMI or traits that share genetic influence with BMI such as appetite, differences in feeding experience across MZs and DZs do not constitute a violation of the equal environments assumption because these differences in feeding practices are being driven by greater genetic similarity between MZs than DZs. In addition, if parents’ reports of how similarly they fed their twins were biased by their perceived zygosity (i.e. reported treatment was not a true reflection of actual treatment, but related to the twins being MZ or DZ), this would also render the heritability estimates unreliable. However, this seems unlikely given previous findings that parents’ reports about their twins’ are not biased by their beliefs about their zygosity, using the ‘mistaken zygosity’ design[33].
Another limitation was the lack of parental genotypes assessments. Parental BMI is by no means a perfect proxy for their genotypic predisposition to higher or lower BMI; the most powerful approach would be to have parental genotypes whereby the non-transmitted alleles from the parents (which relate to their own BMI and behaviour, but not to that of their child) can be entirely separated from the child’s genotype[34]. Nevertheless, the within-family analysis controlled for all family-level genetic and environmental effects, and the magnitudes of the relationships between child BMI and PFPs were unaffected. A further limitation is that we were unable to validate self-reported parental BMI, which may have been inaccurate and could potentially bias our results. Additionally, it may be possible that PFPs are largely explained by environmental factors that influence children’s BMI. As the BMI GPS is not yet strong enough to be a sufficient proxy to separate genetic and environmental effects on child BMI, we were unable to test this question empirically. However, considerable genetic correlations between child BMI and PFPs derived from the twin model renders this explanation unlikely. Lastly, BMI was only reported at one time point, but PFPs are likely to be driven by the child’s emerging BMI throughout the developmental years. However, BMI-associated SNPs and BMI GPS are associated with weight gain trajectories from infancy throughout childhood, so the BMI GPS in fact captures a long window of child BMI[14,35].
This study provides new evidence for gene-environment correlation in parental feeding practices. We have shown that parental feeding practices are substantially heritable and appear to be partly elicited by the common genetic variants that influence children’s BMI. Genome-wide polygenic scores that index children’s genetic propensities for their BMI significantly predicted their parents’ feeding practices, even after potentially confounding shared family effects were taken into account. The findings of this study provide a new perspective on the nature of the associations between parental feeding practices and child BMI.
Participants were drawn from the Twins Early Development Study (TEDS). Between 1994–1996 TEDS recruited over 15,000 twin pairs born in England and Wales, who have been assessed in multiple waves across their development up until the present date. Despite some attrition, about 10,000 twin pairs still actively contribute to TEDS, providing genetic, cognitive, psychological and behavioural data. TEDS participants and their families are representative of families in the UK[36]. Written informed consent was obtained from parents prior to data collection. Project approval was granted by King’s College London’s ethics committee for the Institute of Psychiatry, Psychology and Neuroscience (05.Q0706/228). This study included 4,445 unrelated individuals with genotyping for the GPS analysis, 2,164 genotyped dizygotic (DZ) twin pairs (1,151 same-sex DZ pairs, 1,013 opposite-sex DZ pairs), and 4,375 twin pairs for the twin analysis (1,636 monozygotic (MZ) pairs, 1,441 same-sex DZ pairs, and 1,298 opposite-sex DZ pairs).
Two different genotyping platforms were used because genotyping was undertaken in two separate waves, five years apart. AffymetrixGeneChip 6.0 SNP arrays were used to genotype 3,665 individuals at Affymetrix, Santa Clara (California, USA) based on buccal cell DNA samples. Genotypes were generated at the Wellcome Trust Sanger Institute (Hinxton, UK) as part of the Wellcome Trust Case Control Consortium 2 (https://www.wtccc.org.uk/ccc2/). Additionally, 8,122 individuals (including 3,607 dizygotic co-twin samples) were genotyped on HumanOmniExpressExome-8v1.2 arrays at the Molecular Genetics Laboratories of the Medical Research Council Social, Genetic Developmental Psychiatry Centre, using DNA that was extracted from saliva samples. After quality control, 635,269 SNPs remained for AffymetrixGeneChip 6.0 genotypes, and 559,772 SNPs for HumanOmniExpressExome genotypes.
Genotypes from the two platforms were separately phased using EAGLE2[37], and imputed into the Haplotype Reference Consortium (release 1.1) through the Sanger Imputation Service[38] before merging genotype data from both platforms. Genotypes from a total of 10,346 samples (including 3,320 dizygotic twin pairs and 7,026 unrelated individuals) passed quality control, including 3,057 individuals genotyped on Affymetrix and 7,289 individuals genotyped on Illumina. The final data contained 7,363,646 genotyped or well imputed SNPs (for more details, see Supplementary S1 Methods).
We performed principal component analysis on a subset of 39,353 common (MAF > 5%), perfectly imputed (info = 1) autosomal SNPs, after stringent pruning to remove markers in linkage disequilibrium (r2 > 0.1) and excluding high linkage disequilibrium genomic regions so as to ensure that only genome-wide effects were detected.
The samples used for the analyses differed by necessity in order to accommodate the different methodological approaches: unrelated genotyped individuals (UG); dizygotic genotyped co-twins (DG); twin sample (TS) for quantitative genetic analysis. For the classical twin model approach, only phenotypic data from genotyped twins and their co-twins were selected for comparability across the study samples. Descriptive statistics for all phenotypic measures are reported in Supplementary S4A Table for unrelated genotyped individuals, in Supplementary S4B Table for genotyped DZ twin pairs and in Supplementary S4C Table for samples used for twin modelling.
Children’s body mass index (BMI) was calculated from parent-reported weight (kg) divided by the square of parent-reported height (metres): kg/m2. The 1990 UK growth reference data[39] were used to create BMI standard deviation scores (BMI-SDS) which take account of the child’s age and sex, and represent the difference between a child’s BMI and the mean BMI of the reference children of the same age and sex. BMI-SDS are used rather than BMI itself because BMI varies substantially by age and sex until early adulthood. Reference BMI-SDS have a mean of 0 and a SD of 1: a value greater than 0 indicates a higher BMI than the mean in 1990; a value less than 0 indicates a lower BMI than the mean in 1990. The validity of parent-reported height and weight was tested through home-visits of researchers in a subset of 228 families. Correlations between measurements taken by parents and researchers were high (height: r = 0.90; weight: r = 0.83)[40]. BMI-SDS were available for 4,259 (UG), 4,134 (DG), and 8,406 (TS) individuals. Children had a mean age of 9.91 years (SD = 0.87) when anthropometric measures were assessed.
Parental BMI was calculated for 4,112 individuals using self-reported weight (kg) and height (metres) of the responding parent (kg/m2), which was assessed at the same time as childhood height and weight. To account for the gender of the responding parent (97% mothers, 3% fathers), we used the z-standardized residuals of gender-corrected BMI in analyses.
To assess PFPs, we used the Child Feeding Questionnaire[41], which parents completed when their twins were approximately 10 years old (mean = 9.91 years, SD = 0.87). To measure the degree to which parents restricted their children’s food intake (‘restriction’), we calculated a mean composite score based on 6 items (Cronbach’s alpha = 0.78), such as “I intentionally keep some foods out of my child’s reach“, or “If I did not guide my child’s eating, he/she would eat too many junk foods”. Data were available for 4,386 (UG), 4,228 (DG) and 8,582 (TS) children. Similarly, we created a mean composite score to assess the amount of pressure parents exerted on their children to increase their food intake (‘pressure’), including 4 items (Cronbach’s alpha = 0.61) such as “If my child says “I’m not hungry”, I try to get him/her to eat anyway”, or “I have to be especially careful to make sure my child eats enough”. Data were available for 4,445 (UG), 4,328 (DG) and 8,750 (TS) children. All items were scored on a 5-point Likert scale (Disagree, Slightly disagree, Neutral, Slightly agree, Agree).
For child and parent anthropometrics we removed extreme outliers with implausible values that were deemed to be errors. For children we excluded values based on the following criteria: -/+ 5 standard deviations above or below the mean of height SDS, weight SDS or BMI-SDS; shorter than 105 cm or taller than 180cm; lighter than 12 kg or heavier than 80 kg. After removing outliers, child BMI-SDS had a mean of 0 and a standard deviation of 0.99, showing that the sample is representative of the UK reference population for BMI in 1990 (mean = 0; SD = 1). For parental BMI, we removed individuals with values that fell -/+ 3.5 standard deviations above or below the mean, as well as individuals that weighed below 35 kg. To account for the positive skew, we log transformed this variable. As all variables showed age or sex effects (described in Supplementary S4A, S4B and S4C Table), we controlled for these variables by applying the regression method, using z-standardized residuals for all further analyses. Supplementary S5A, S5B and S5C Table show descriptive statistics for all clean measures (regressed onto age and sex) in unrelated samples, for DZ twin pair samples, and individuals used for twin modelling, respectively.
We created Genome-wide Polygenic Scores (GPS) for BMI, using summary statistics from a genome-wide meta-analysis of BMI including 339,224 participants[19]. We calculated a GPS for each individual as the sum of the weighted count of BMI-increasing alleles:
GPSBMI=∑i=1kβiSNPi
where i ∈ {1,2,..,k} and indexes SNPi and the i number of the k BMI increasing alleles included in the score is determined by the p-value threshold of the SNP–phenotype association in the discovery GWAS, the β-coefficients for each respective genetic variant is used as a weight, and the count of each reference allele is represented by genotype dosage (0,1, or 2 alleles) of SNPi.
We used the software PRSice[42] to calculate GPS in our sample. To account for multicollinearity among SNPs in Linkage Disequilibrium (LD), which can upwardly bias GPS predictions[43], genome-wide clumping was performed (r2 = 0.1, kb = 250). Using the clumped, independent SNPs, we created eight GPS for 10,346 individuals (7,026 unrelated individuals; 3,320 DZ twin pairs) using increasingly liberal GWAS p-value thresholds (pT: 0.001,0.05,0.1,0.2,0.3,0.4,0.5,1). Diagonals in Supplementary S2 Fig show the number of SNPs included in each respective GPS. As all thresholds performed similarly well (Supplementary S2 Fig), we used a GPS based on the smallest p-value threshold of 0.001 for all further analyses. Potential effects due to population stratification and genotyping were accounted for by regressing the first ten principal components, and factors capturing genotyping information (microarray, batch and plate) onto the child BMI GPS, subsequently using the z-standardised residuals in our analyses.
To obtain broad estimates of the extent to which individual differences in PFPs are determined by children’s genotypes, we used a multivariate ‘correlated factors’ twin model. This allowed us to estimate: (1) the heritability of PFPs, which provided an indication of the extent to which PFPs are caused by children’s genotypes in general; and (2) the extent of common genetic influence on both child BMI-SDS and PFPs, which provided an indication of the extent to which PFPs are caused by children’s genetic propensity to higher or lower BMI, specifically.
Based on biometrical genetics theory[44], it is possible to decompose variance in a single trait into three components: additive genetic (A; heritability), shared environmental (C; all environmental effects that make family members more similar) and non-shared environmental (E; all environmental effects that contribute to dissimilarities across family members, including random error measurement). The basis of the method is to compare resemblance for a single trait between monozygotic (MZ) and dizygotic (DZ) twin pairs, who share 100% and 50% (on average) of their segregating genetic material, respectively, while both types of twins are correlated 100% for their shared environmental influence. The observed covariation between MZ and DZ pairs is compared with the expected covariation, based on the knowledge of different degrees of allele sharing (or identity by descent (IBD)) of MZ (IBD = 1.0) and DZ pairs (IBD = 0.5 on average). The twin method therefore assumes that MZ and DZ twins share their environments in terms of the trait in question to the same extent (so-called the ‘equal environments assumption’), and the only difference between the two types of twins is the extent of their genetic relatedness.
Using the same principles, comparison of MZ and DZ covariation across traits (so-called cross-twin cross-trait covariance, e.g. the covariation between twin 1 BMI-SDS and twin 2 ‘restriction’) provides an indication of the extent to which the genetic and environmental influences on multiple traits are the same. The key pieces of information provided are the aetiological correlations, which indicate the extent to which child BMI and PFPs are caused by the same additive genetic (genetic correlation; rA), shared environmental (shared environmental correlation; rC), and non-shared environmental influences (non-shared environmental correlation; rE). In this analysis we were primarily interested in the genetic correlation, which indicates the extent to which the additive genetic influences on child BMI cause PFPs. The aetiological correlations range from -1 to 1 and can be interpreted similarly to Pearson’s correlations. For example, a high positive genetic correlation between ‘restriction’ and BMI would indicate that many of the DNA variants that cause higher child BMI are the same as those cause higher levels of ‘restriction’, while a high negative genetic correlation would indicate that many of the DNA variants causing higher child BMI are the same as those causing lower levels of ‘restriction’.
Maximum likelihood structural equation modelling was used to estimate intra-class correlations across the zygosities, the A, C and E parameter estimates and aetiological correlations (with 95% confidence intervals), and goodness-of-fit statistics. Sex differences in the parameter estimates were also tested for using a sex-limitation model. Analyses were implemented in the R package OpenMx[45].
|
10.1371/journal.ppat.1000095 | APOBEC3G and APOBEC3F Require an Endogenous Cofactor to Block HIV-1 Replication | APOBEC3G (A3G)/APOBEC3F (A3F) are two members of APOBEC3 cytidine deaminase subfamily. Although they potently inhibit the replication of vif-deficient HIV-1, this mechanism is still poorly understood. Initially, A3G/A3F were thought to catalyze C-to-U transitions on the minus-strand viral cDNAs during reverse transcription to disrupt the viral life cycle. Recently, it was found more likely that A3G/A3F directly interrupts viral reverse transcription or integration. In addition, A3G/A3F are both found in the high-molecular-mass complex in immortalized cell lines, where they interact with a number of different cellular proteins. However, there has been no evidence to prove that these interactions are required for A3G/A3F function. Here, we studied A3G/A3F-restricted HIV-1 replication in six different human T cell lines by infecting them with wild-type or vif-deficient HIV-1. Interestingly, in a CEM-derived cell line CEM-T4, which expresses high levels of A3G/A3F proteins, the vif-deficient virus replicated as equally well as the wild-type virus, suggesting that these endogenous antiretroviral genes lost anti-HIV activities. It was confirmed that these A3G/A3F genes do not contain any mutation and are functionally normal. Consistently, overexpression of exogenous A3G/A3F in CEM-T4 cells still failed to restore their anti-HIV activities. However, this activity could be restored if CEM-T4 cells were fused to 293T cells to form heterokaryons. These results demonstrate that CEM-T4 cells lack a cellular cofactor, which is critical for A3G/A3F anti-HIV activity. We propose that a further study of this novel factor will provide another strategy for a complete understanding of the A3G/A3F antiretroviral mechanism.
| Cytidine deaminases are host enzymes that remove the amino group from the cytidine base on single-stranded DNA or RNA, resulting in a replacement of the cytidine with a uracil. Such replacement may alter the amino acid–coding sequence of the gene and change protein function. It has been well documented that APOBEC1 and AID play very important roles in protein metabolism and immune response via this mechanism. Interestingly, recent advances in retroviral researches have discovered that the seven cytidine deaminases (APOBEC3A to 3H) on human Chromosome 22 can restrict retrovirus replication. In particular, APOBEC3G and APOBEC3F have the most powerful anti–HIV-1 activity and also inhibit other retroviruses, including retrotransposons. They could inhibit viral replication in either a cytidine deamination-dependent or -independent manner, but the precise mechanism remains to be defined. In this report, we found that in a particular human T cell line, APOBEC3G and APOBEC3F failed to block HIV-1 replication. Further analyses indicated that this cell line lacks a cellular factor, which is very critical for APOBEC3G and APOBEC3F antiviral activity. Thus, APOBEC3G and APOBEC3F require a cofactor to inhibit viral replication, and identification of this cofactor will provide an important strategy to decipher this poorly defined antiretroviral mechanism.
| Cytidine deaminases are RNA-editing enzymes that target cytosines for conversion to uracils (C-to-U). They belong to the apolipoprotein B mRNA-editing enzyme catalytic polypeptide-like (APOBEC) family, which includes activation-induced deaminase (AID), APOBEC1 (A1), APOBEC2 (A2), a group of APOBEC3 (A3), and APOBEC4 (A4) in humans [1]. A1 is the original member of this family and remains the best characterized. It has the capability to introduce a premature termination codon on apolipoprotein B100 (apoB) mRNA by C-to-U editing to produce a truncated form of this protein [2]. AID is the second member identified. It edits specific “hotspots” on immunoglobulin gene loci in activated B cells to direct somatic hypermutation and isotype class switching to generate different antibodies [3].
The human A3 subgroup contains seven members: A3A, A3B, A3C, A3DE, A3F, A3G, and A3H. All these proteins have antiretroviral activities against different targets, including exogenous retroviruses and endogenous retroelements [1]. The replication of human immunodeficiency virus type 1 (HIV-1) is inhibited by A3B, A3DE, A3F, A3G, and A3H [4]–[11], and A3G shows the most powerful anti-HIV-1 activity [9]. A3G/A3F also blocks various retroelements, including LTR retrotransposons and non-LTR retrotransposons [12]–[18]. Nevertheless, HIV-1 is able to elude this defense mechanism and cause disease in humans for two reasons. First, A3B and A3H are poorly expressed in vivo [5],[7],[19],[20]. Second, HIV-1 produces a viral infectivity factor (Vif) that binds to and mediates the destruction of A3DE, A3F, and A3G in 26S proteasomes via recruitment of the Cullin5 ubiquitin E3 ligase [6], [10], [11], [21]–[25]. Recently, a protein degradation–independent mechanism was also reported [26].
The antiretroviral mechanism of these A3 proteins has been extensively studied. Initially, it was found that A3G proteins deaminate deoxycytidines (dCs) to form deoxyuridines (dUs) on viral minus-strand cDNAs during viral reverse transcription [27]–[30]. These C-to-U mutations could either cause the degradation of the viral minus-strand cDNAs or result in G-to-A hypermutations in the plus-strand viral cDNAs, which create havoc in viral transcripts and produce noninfectious virions. Although this model has been favored for a while, recent investigations suggest that the cytidine deamination may not be absolutely required for A3 antiretroviral activity. First, it was found that anti-HIV activity of A3G/A3F does not correlate with hypermutations, but correlates with the reduction of viral reverse transcripts [31]. In addition, A3F and A3H were shown to inhibit HIV-1 replication in the absence of hypermutations [5],[32]. Second, A3G/A3F and other A3 proteins were also shown to inhibit the replication of some other retroviruses or retrotransposons in the absence of hypermutation [14], [15], [18], [33]–[37]. Further investigations demonstrated that A3G/A3F could interrupt viral reverse transcription by reducing the efficiency of tRNAlys3 priming to the viral RNA template, elongation, and DNA strand transfer [38]–[43]. Moreover, they could also block viral integration [42],[44]. However, how these viral enzymatic reactions are inhibited by A3G/A3F is still not clear.
To understand the mechanism of A3G/A3F antiretroviral activity, efforts have made in another direction by isolating their cellular binding proteins, although there is no evidence to prove that they are functionally required. Both A3G and A3F were found in a ∼700 kDa high-molecular-mass (HMM) complex in immortalized cell lines [45],[46]. Unlike A3F, the A3G HMM complexes are RNase-sensitive; when treated with RNase A, they fall apart into 100 kDa low-molecular-mass (LMM) complexes. Biochemical isolations of A3G/A3F-binding proteins from these HMM complexes have generated a list that contains almost 100 different proteins, some of which are shared by A3G and A3F [47]–[50]. However, it is still unclear whether these interactions are essential for A3G/A3F antiviral activities.
In this report, we studied the anti-HIV-1 activity of A3G/A3F in six different human T cell lines. We found that A3G/A3F lost anti–HIV-1 activity in the CEM-derived cell line CEM-T4. Further investigation demonstrated that CEM-T4 cells lacked a cellular cofactor. These results lead to a new direction of investigation on the mechanism of how A3G/A3F inhibits retroviral replication.
To study A3 protein antiretroviral activity, six human T cell lines, including HUT 78, HUT 78–derived H9 and PM1, and CEM-derived CEM-SS, CEM-T4, and A3.01, were selected for HIV-1 infection. HUT 78 and H9 cells are nonpermissive cells because they express A3G and restrict HIV-1ΔVif virus replication; A3.01 cells are semipermissive cells because the HIV-1ΔVif virus is not completely restricted although it expresses A3G; and CEM-SS cells are permissive cells because they do not express A3G and no longer restrict HIV-1ΔVif replication [51]. CEM-T4 is a natural subclone of CEM and isolated by its relatively high CD4 expression (Paul Clapham, personal communication). The permissiveness of PM1 and CEM-T4 for the HIV-1ΔVif virus has not been determined. To understand whether there is a significant difference in CD4 levels in these cells, the surface expression of CD4 and CXCR4 was determined by flow cytometry. No significant difference in CD4 and CXCR4 expression was observed, although CEM-T4 cells did show a slight high CD4 expression (Figure 1A).
Next, these cells were infected with wild-type or vif-defective HIV-1, and viral replication curves were determined for 8 d. Although a robust replication of the wild-type virus was observed in all six cell lines, a significant variation in vif-defective virus replication was found (Figure 1B). The replication of HIV-1ΔVif was completely restricted in HUT 78, H9, and PM1 cells; less severely restricted in A3.01 cells; and not restricted at all in CEM-SS and CEM-T4 cells. This result indicated that CEM-T4 should belong to the permissive cell type with no A3G/A3F expression.
To confirm this, A3G and A3F expressions were determined by Western blotting. High levels of A3G were detected in CEM, A3.01, H9, HUT 78, and PM1 cells, and no A3G expression was detected in CEM-SS cells (Figure 1C), which is consistent with previous observations. Strikingly, a high-level of A3G expression was also detected in CEM-T4 cells. In addition, A3F expression was also detected in CEMT4 as well as H9, A3.01, and HUT 78 cells, but not in CEM-SS cells (Figure 1C). These unexpected results caused us to conclude that although CEM-T4 cells express A3G/A3F, they are unable to block HIV-1 replication.
Because A3G has much more potent anti–HIV-1 activity than A3F, we decided to further characterize the A3G protein from CEM-T4 cells to understand why A3G could not inhibit HIV-1 replication. First, we cloned and sequenced the A3G gene. No mutation was found in this gene from this cell line (unpublished data). Second, we determined whether a defect at a post-translational level was present that could disrupt A3G interaction with other cellular partners. The A3G protein complex was isolated from CEM-T4 cells by sedimentation of cell lysates in sucrose gradients as described previously [46]. A3G was found in the HMM complexes in CEM-T4 cells as it was in 293T and A3.01 cells and, importantly, they were sensitive to conversion to LMM complexes upon RNase A treatment (Figure 2A). This result indicated an intact ability of A3G to interact with cellular RNAs and proteins. Third, we determined A3G cytidine deaminase activity by a scintillation proximity-based assay using an A3G-specific template. As presented in Figure 2B, cells such as 293T and CEM-SS that do not express A3G had marginal deaminase activity; CEM-T4 and A3.01 cells that express A3G had higher levels of activity, which were significantly increased after RNase A treatment. These results confirmed that A3G is present in HMM complexes in these cells (Figure 2A) and further indicated that A3G in CEM-T4 cells is enzymatically active. Thus, no apparent defect on A3G was detected from CEM-T4 cells.
Since the antiretroviral activity of A3G is associated with its presence in virions, we decided to further determine whether A3G could be packaged into HIV-1 virions from CEM-T4 cells. To produce sufficient amounts of virions, cell lines stably producing HIV-1 virions were generated. HIV-1 viruses carrying a neomycin-resistant gene were initially produced from 293T cells after transfection with pNL-Neo or pNL-NeoΔVif. A3.01, H9, CEM-SS, and CEM-T4 cells were then infected with these viruses, and stably infected cell lines were created by G418 selection. Virions were then purified by ultra-centrifugation, and viral proteins were determined by Western blotting. All the cell lines were able to produce viruses as evidenced by the detection of p24Gag; Vif was only detectable in samples of the wild-type virus but not the vif-deficient virus (Figure 2C). A3G was consistently detected in virions from H9, A3.01, and CEM-T4 cells, but not CEM-SS cells, and more A3G proteins were found in the vif-defective than the wild-type virions. The level of A3G encapsidation from CEM-T4 cells was lower than that from H9, but was at least comparable to those from A3.01 cells (Figure 2C; compare lanes 2, 4, 6). Thus, A3G was effectively encapsidated by HIV-1 from CEM-T4 cells. We further compared the infectivity of viruses from CEM-SS, A3.01, and CEM-T4. We found that HIV-1ΔVif infectivity was significantly reduced only in A3.01 cells, not CEM-SS and CEM-T4 cells (Figure 2D). This result indicated that the vif-deficient virions produced from CEM-T4 cells were still infectious, which explained why CEM-T4 cells were permissive for HIV-1ΔVif replication.
We developed two hypotheses to explain why CEM-T4 cells are permissive for HIV-1ΔVif replication: 1) CEM-T4 cells lack a cofactor essential for A3G/A3F anti-HIV activity; and 2) CEM-T4 cells express a dominant inhibitor that blocks A3G/A3F activity. A previously described trans-complementation assay was used to test these hypotheses [52],[53]. In this assay, CEM-T4 cells were fused with 293T cells to form heterokaryons. Because 293T can support A3G/A3F antiviral activity, if A3G/A3F antiviral activity is restored in heterokaryons, it would indicate that a cofactor is missing in CEM-T4 cells; otherwise, CEM-T4 cells should express an inhibitor. To ensure that infectious virions are produced exclusively from heterokaryons, we expressed env-deficient viral particles (HIV-1ΔEnv) in CEM-T4 cells and HIV-1 Env protein in 293T cells. No infectious particle could be produced from these two cell lines unless they formed heterokaryons by HIV-1 Env and CD4/CXCR4-mediated cell fusion and trans-complemented for the missing viral components (Figure 3A). The infectious viral particles were then detected by infecting the HIV-reporter cell line TZM-bI, which contains an integrated firefly luciferase gene under the control of the HIV LTR.
Initial control experiments were performed with T cells that support A3G/A3F anti–HIV-1 activity. A2.01, a CEM-derived human T cell line that does not express CD4, was used as a negative control, whereas A3.01, HUT 78, and H9 were used as positive controls. Since A2.01 cells should not fuse to 293T cells, no infectious particles should be recovered. Indeed, very low luciferase activity was detected from TZM-bI cells inoculated with culture supernatant from A2.01 and 293T coculture (Figure 3B; lane 1). In sharp contrast, when A2.01 was replaced by A3.01, HUT 78, or H9, 10- to 30-fold higher luciferase activities were detected from TZM-bI, indicating a high efficiency of heterokaryon formation and release of infectious particles from these heterokaryons (Figure 3B; lanes 2–4).
Since CEM-SS cells do not express A3G/A3F, they were further used to test the sensitivity of this system to A3G/A3F and Vif activities. When A3G or A3F was not expressed in 293T cells, infectious particles were recovered from heterokaryons regardless of whether Vif was expressed (Figure 3B; lanes 5 and 8). However, when A3G or A3F was expressed, infectious particles were only recovered from heterokaryons in the presence of Vif (Figure 3B; lanes 6 and 7), not in the absence of Vif (Figure 3B; lanes 9 and 10). These results not only demonstrated the efficiency and accuracy of this trans-complementation assay, but also confirmed the expression of corresponding proteins from different constructs.
Finally, we fused CEM-T4 cells with 293T cells. When Vif was expressed in heterokaryons, infectious particles were recovered regardless of whether A3G or A3F was expressed (Figure 3B; lanes 11–13). In sharp contrast, when Vif was not expressed and A3G or A3F was expressed either from CEM-T4 or 293T, no infectious particles were recovered (Figure 3B; lanes 14–16). These results indicated that A3G/A3F in the heterokaryon between CEM-T4 and 293T could block HIV-1 replication. Thus, we concluded that CEM-T4 cells lack a cellular cofactor required for A3G/A3F anti-HIV activities.
Although we found that the endogenously expressed A3G/A3F in CEM-T4 cells lost anti-HIV activity, we wanted to know whether this defect is still present when these proteins are overly expressed. We therefore attempted to express A3G or A3F genes transiently in an HIV-based vector. A3F, A3G, or the noncatalytic A3G (A3GE259Q) gene with a 3′ HA tag was inserted into the Nef open reading frame in either pNL4-3 or pNL4-3ΔVif vector so that these A3 proteins could be expressed during HIV-1 replication (Figure 4A; top panel). In total, six different HIV-1 proviral constructs were generated: pNLA3F, pNLA3FΔVif, pNLA3G, pNLA3GΔVif, pNLA3GE259Q, and pNLA3GE259QΔVif.
To test their activities, recombinant viruses were first produced by transfection of 293T cells and viral infectivity was determined. To further confirm A3G/A3F gene function in these vectors, we cotransfected the vif-defective version of these vectors with pNL-A1. pNL-A1 is a Vif expression vector created from pNL4-3 with most of the viral genes deleted, including the viral RNA packaging signal, and can only provide Vif expression in trans for a single round. High levels of A3G or A3GE259Q expression were detected in transfected 293T cells in the absence of Vif by Western blotting (Figure 4A; middle panels; lanes 5 and 8), and their expressions were decreased by Vif expressed either in cis or in trans (Figure 4A; lanes 4, 6, 7, 9). The expression of A3F was relatively low (Figure 4A; lane 2), and Vif further decreased this expression (Figure 4A; lanes 1 and 3). Next, recombinant viruses were collected to infect TZM-bI cells for a single-round replication. Overall, these viruses had a very similar infectivity in the presence of Vif (Figure 4A; bottom panel; lanes 1, 3, 4, 6, 7, 9). However, A3 proteins decreased viral infectivity when Vif was absent. The wild-type A3G had the most powerful anti-HIV activity, which reduced viral infectivity by around 10-fold (Figure 4A; lane 5). Both A3F and A3GE259Q mutant reduced viral infectivity by around 2-fold (Figure 4A; lanes 2 and 8). The low anti-HIV activity of A3F could be due to its low expression, and the low activity of A3GE259Q is consistent with previous reports [16], [54]–[56]. Nevertheless, this result confirmed that these constructs expressed functional A3G or A3F proteins.
Next, we infected CEM-SS and CEM-T4 cells with these recombinant HIV viruses. In this experiment, pNLA3 indicates HIV-1 expressing both Vif and A3 proteins, pNLA3ΔVif indicates viruses expressing only A3 protein, and pNLA3ΔVif+Vif indicates viruses expressing Vif provided in trans by pNL-A1 and A3 proteins. As presented in Figure 4B, the replication of both pNLA3FΔVif and pNLA3FΔVif+Vif were slightly delayed in CEM-SS cells when compared with pNLA3F, and this result was reversed in CEM-T4 cells; the replication of both pNLA3GE259QΔVif and pNLA3GE259QΔVif+Vif were also slightly delayed in CEM-SS cells during the first 3 d of infection, and all three A3GE259Q-expressing viruses replicated equally well in CEM-T4 cells; the replication of pNLA3GΔVif and pNLA3GΔVif+Vif but not pNLA3G viruses was strictly restricted in CEM-SS cells, and all 3 viruses replicated almost equally well in CEM-T4 cells. The growth curve of these viruses in CEM-SS cells was consistent with their infectivity data in Figure 4A, confirming that A3F only weakly inhibited viral replication due to low expression, A3G potently inhibited viral replication, and A3GE259Q had very low antiviral activity. Nonetheless, since all vif-deficient viruses expressing A3G or A3F replicated very well in CEM-T4 cells, these results indicated that A3G/A3F lost their antiviral activity, which further supported that CEM-T4 cells lack a cofactor. The slight delay of pNLA3GΔVif virus replication in CEM-T4 cells could be due to an incorporation of this cofactor from 293T cells that compensated A3G activity during the first round of infection, which provides another piece of evidence that CEM-T4 cells do not express this cofactor.
To further confirm these observations, we stably transduced CEM-T4 or CEM-SS cells with an A3G, A3GE259Q, A3F, or GFP gene by the murine leukemia virus (MuLV)-based vector pMSCVneo. These genes, containing a 3′ HA-tag, were inserted into pMSCVneo, and recombinant MuLV viruses were created by transfection of Phoenix-AMPHO cell line. Viruses were then used to infect CEM-T4 or CEM-SS cells, and 8 stable cell lines were created by G418 selection. All transduced genes were expressed although the expression of A3F and GFP was lower than that of A3G and A3GE259Q (Figure 5A). It is known that human A3G inhibits MuLV replication, but one group reported that this inhibition might not depend on cytidine deamination [57]. Nevertheless, we wanted to make sure that the tranduced gene did not contain any mutation. The exogenous A3G and A3F genes in CEM-T4 cells were cloned and sequenced, and no mutation was found (unpublished data). Thus, these cell lines should express functional exogenous A3G/A3F proteins.
Another possibility that A3G/A3F lost anti-HIV activity is that they are mislocalized in CEM-T4 cells. To exclude this possibility, we compared A3G subcellular localization in CEM-T4 and CEM-SS cells by confocal microscopy. A3G and A3F are both known as cytoplasmic proteins, and can be found in the mRNA processing (P) bodies [58],[59]. A recent report showed that A3F is colocalized with cellular protein MOV10 [48], which is also a P-body protein [60]. When stable CEM-T4 and CEM-SS cells expressing A3G were stained with anti-A3G and anti-MOV10 antibodies, A3G protein was found in the cytoplasm of both cell lines and was colocalized with MOV10 (Figure 5B). Thus, A3G/A3F should not be mislocalized in CEM-T4 cells.
Finally, we determined HIV-1 replication in these cell lines. As presented in Figure 5C, both wild-type and vif-deficient HIV-1 replicated equally well in CEM-T4 cells expressing A3G, A3GE259Q, A3F, or GFP, suggesting that the vif-deficient virus is not restricted by any of these genes. However, in CEM-SS cells, although the replication of vif-deficient virus was not restricted by GFP or A3GE259Q, it was severely restricted by A3G or A3F. The anti-HIV activity of A3F in CEM-SS cells was relatively lower than that of A3G, which could be due to its low expression as shown in Figure 5A. In addition, the poor activity of A3GE259Q further confirmed that this noncatalytic A3G mutant has poor antiviral activity. Thus, we concluded that A3G/A3F failed to inhibit HIV-1 replication in CEM-T4 cells even though they were overexpressed.
In this report, we studied A3G/A3F anti–HIV-1 activity in 6 different human T cell lines. We found that in one cell line, CEM-T4, A3G/A3F lost anti–HIV-1 activity due to the absence of a cellular factor, which is very critical for A3G/A3F anti-HIV activity.
Although A3G/A3F potently inhibits HIV-1 replication, this mechanism is still poorly defined. A3G/A3F has two conserved zinc-binding domains. In the process of blocking HIV-1 replication, these two domains have different functions. The N-terminal domain has a high affinity for RNAs, which normally serves as a virion-packaging signal, and the C-terminal domain has cytidine deamination activity, which is the real catalytic domain [61]–[63]. Initially, it was found that the noncatalytic A3G mutant E259Q still retained intact anti-HIV activity, suggesting that the cytidine deaminase activity is not required for antiviral activity [64]. However, this result could not be reproduced by the other investigators [16], [54]–[56], and we also found that the E259Q mutant had very marginal anti-HIV activity (Figures 4B and 5C). Nevertheless, it is clear that A3G could inhibit the replication of hepatitis B virus and human T cell leukemia virus type 1 in the absence of cytidine deamination [36],[37], and many similar cases have been found in blocking HIV-1, adeno-associated virus, and retrotransposon replications by different A3 proteins [5], [14], [15], [18], [32]–[35]. Thus, although cytidine deamination is not required, the cytidine deaminase activity is absolutely required for A3 antiretroviral activities.
As introduced before, A3G/A3F can reduce the accumulation of HIV-1 cDNAs, which correlates well with their anti-HIV activity. Previously, two groups reported that uracil DNA glycosylases-2 (UNG), a host DNA repair enzyme, is involved in the degradation of viral cDNAs containing uracils [65],[66]. However, another two groups obtained conflicting results and dismissed the role of this enzyme in A3G antiviral activity [16],[67]. In addition, A3G/A3F were shown to inhibit tRNAlys3 priming, elongation, or DNA strand transfer during reverse transcription and viral integration [38]–[44]. How A3G/A3G can virtually disrupt these critical reactions in the viral life cycle needs to be understood. Notably, the presence of an A3G/A3F antiretroviral cofactor may help decipher this poorly defined mechanism. One possibility is that this cofactor has nuclease activity that directly degrades viral reverse transcripts. Alternatively, it may increase the affinity of A3G/A3F to viral RNAs or cDNAs so that they can compete with viral reverse transcriptase and integrase for their substrates. Although it was expected that A3G/A3F would specifically interact with viral RNAs or cDNAs to block HIV-1 replication, in vitro study with recombinant A3G proteins failed to demonstrate such specificity [68],[69]. In the case of another deaminase, A1, it has been shown that its editing activity is a highly sequence-specific process dependent upon the primary, secondary, and perhaps even tertiary structure of the RNA substrate. Further investigations have identified two host factors that are required to complement A1 for apoB mRNA editing: ACF (APOBEC1 complementation factor) and its splice variant ASP (APOBEC1 stimulating protein) [70],[71]. ACF is very homologous to the RNA-binding protein GRY-RBP and binds to the U-rich “mooring” sequence of apoB mRNA [71]. Thus, similar cofactors may also be required for A3G/A3F.
Two groups have reported that human A3G/A3F could inhibit yeast LTR retrotransposon Ty1 in Saccharomyces cerevisiae [12],[17]. Whether a similar cofactor is required for A3G/A3F antiretroviral activity in yeast therefore becomes an interesting question. If it is required, it would imply that a highly conserved antiviral ortholog gene exists in different organisms. Otherwise, it may indicate that this cofactor is very specific for HIV-1. The latter possibility could also suggest that A3G/A3F uses different antiviral mechanisms to target different retroviruses. Nevertheless, knowledge of the cofactor involved in the process of blocking HIV-1 replication by A3G/A3F is critical to our understanding of HIV pathogenesis. Further characterization of this cofactor will lead to a complete understanding of A3G/A3F anti-HIV activity.
HIV-1 proviral constructs pNL4-3 and pNL4-3ΔVif and human A3G or A3F expression pcDNA3.1-V5-6XHis vectors have been described previously [6],[11]. The noncatalytic A3G mutant (A3GE259Q) was created by site-directed mutagenesis in pcDNA3.1 vector. pNL-Neo, pNL-NeoΔVif, pNLA3G, pNLA3GΔVif, pNLA3F, pNLA3FΔVif, pNLA3GE259Q, and pNLA3GE259QΔVif were created by replacing the firefly luciferase gene in pNL-Luc and pNL-LucΔVif with a neomycin-resistant gene or an A3G, A3F, or A3GE259Q gene containing a 3′ HA-tag by NotI/XhoI digestion, respectively. In addition, an A3F, A3G, A3GE259Q, or GFP gene with a 3′ HA tag was inserted into the pMSCVneo vector by EcoRI/XhoI digestion. pNL4-3ΔGag and pNL4-3ΔEnv were created by SphI/AgeI double digestion or NheI single digestion of pNL4-3, followed by large Klenow fragment treatment before T4 ligation. The NheI site is still active in pNL4-3ΔEnv although the env gene was inactivated by frame-shift. To create pNL4-3ΔGagΔVif and pNL4-3ΔEnvΔVif, the parental plasmids were digested with PfiMI and filled in with a linker from annealing two DNA oligonucleotides (5′-CTAGCTAGCTAGCCGGCAGA-3′, 5′-GCCGGCTAGCTAGCTAGTCT-3′).
The HIV indicator cell line TZM-bI and human T cell lines HUT 78, H9, PM1, CEM-SS, CEM-T4, A3.01, and A2.01 were from the National Institutes of Health (NIH) AIDS Research and Reference Reagent Program. The Phoenix-AMPHO cell line was from Dr. G. Nolan (Stanford University). T cell lines were cultured in RPMI 1640 with 10% fetal bovine serum (HyClone). Phoenix-AMPHO, 293T, and TZM-bI were cultured in DMEM with 10% bovine calf serum (HyClone).
HIV-1 or MuLV viruses were produced from 293T or Phoenix-AMPHO cells by the standard calcium phosphate transfection.
A total of 1×105 cells were incubated with 100 ng wild-type or Vif-defective HIV viruses at 37°C for 3 h. After removal of the inocula, followed by 3 extensive washings, cells were cultured in 24-well plates for 8 d. Culture supernatants were then collected daily for measurement of p24Gag by ELISA.
A scintillation proximity-based assay was used as described previously [46],[72].
The rabbit anti-human A3G polyclonal antibody was from the NIH AIDS Research and Reference Reagent Program. The mouse anti-human A3F polyclonal antibody was from Abnova, Taiwan. Actin was detected by a polyclonal antibody (C-11; Santa Cruz Biotechnology). HIV-1 p24Gag and Vif were detected by antibodies (nos. 3537 and 6459) from the NIH AIDS Research and Reference Reagent Program. HRP-conjugated anti-rabbit or mouse IgG secondary antibodies were from PIERCE. Detection of the HRP-conjugated antibody was performed using Supersignal Wetpico Chemiluminescence Substrate kit (PIERCE).
A previously established protocol was adopted [52],[53]. Briefly, 293T cells were seeded in 6-well plates at 8×105/well in 2 ml medium. After 12 h, cells were transfected with 6 μg of HIV Env expression vector pNL4-3ΔGag or pNL4-3ΔGagΔVif in the presence or absence of A3G or A3F expression vector and washed with PBS 4 h later. Simultaneously, 8×105 T cells were infected with 500 ng of VSV-pseudotyped Env-defective HIV-1 from pNL4-3ΔEnv– or pNL4-3ΔEnvΔVif–transfected 293T cells at 37°C for 3 h. After removal of the inocula and extensive washing, infected T cells were added to the Env-expressing 293T cell culture. After 48 h, supernatants from these cocultures were collected to infect TZM-bI cells. Viral infectivity was finally determined by measuring cellular luciferase activities after another 48 h.
CEM-T4 cells stably expressing exogenous A3G from the pMSCVneo vector were fixed in a buffer (5% formaldehyde+2% sucrose in PBS). Fixed samples were permeabilized for 30 min at room temperature in a permeabilization buffer (1% Triton X-100, 10% sucrose in PBS) prior to incubation with antibodies. Cells were then stained with a mouse anti-A3G monoclonal antibody at 1∶100 (ImmunoDiagnostics, obtained from the NIH AIDS Research and Reference Reagent Program) and a rabbit anti-MOV10 polyclonal antibody at 1∶100 (Proteintech Group). Cover slips were washed once in PBS (5 min at room temperature) and incubated with secondary antibodies, including goat anti-mouse IgG Alexa Fluor 488 and goat anti-rabbit IgG Alexa Fluor 594 (Invitrogen). Cells were further stained with 1 μg/ml Hoechst 33342 (Sigma-Aldrich, St. Louis, Missouri, United States of America). Cover slips were then washed twice with PBS and mounted onto microscope slides with glycerol gelatin (Sigma-Aldrich) and were stored at 4°C in the dark until analyzed by a confocal microscope Olympus Fluoview 1000.
The GenBank accession numbers for human APOBEC3G and APOBEC3F are BC024268 and BC038808. |
10.1371/journal.pgen.1000414 | A Human Protein Interaction Network Shows Conservation of Aging Processes between Human and Invertebrate Species | We have mapped a protein interaction network of human homologs of proteins that modify longevity in invertebrate species. This network is derived from a proteome-scale human protein interaction Core Network generated through unbiased high-throughput yeast two-hybrid searches. The longevity network is composed of 175 human homologs of proteins known to confer increased longevity through loss of function in yeast, nematode, or fly, and 2,163 additional human proteins that interact with these homologs. Overall, the network consists of 3,271 binary interactions among 2,338 unique proteins. A comparison of the average node degree of the human longevity homologs with random sets of proteins in the Core Network indicates that human homologs of longevity proteins are highly connected hubs with a mean node degree of 18.8 partners. Shortest path length analysis shows that proteins in this network are significantly more connected than would be expected by chance. To examine the relationship of this network to human aging phenotypes, we compared the genes encoding longevity network proteins to genes known to be changed transcriptionally during aging in human muscle. In the case of both the longevity protein homologs and their interactors, we observed enrichments for differentially expressed genes in the network. To determine whether homologs of human longevity interacting proteins can modulate life span in invertebrates, homologs of 18 human FRAP1 interacting proteins showing significant changes in human aging muscle were tested for effects on nematode life span using RNAi. Of 18 genes tested, 33% extended life span when knocked-down in Caenorhabditis elegans. These observations indicate that a broad class of longevity genes identified in invertebrate models of aging have relevance to human aging. They also indicate that the longevity protein interaction network presented here is enriched for novel conserved longevity proteins.
| Studies of longevity in model organisms such as baker's yeast, roundworm, and fruit fly have clearly demonstrated that a diverse array of genetic mutations can result in increased life span. In fact, large-scale genetic screens have identified hundreds of genes that when mutated, knocked down, or deleted will significantly enhance longevity in these organisms. Despite great progress in understanding genetic and genomic determinants of life span in model organisms, the general relevance of invertebrate longevity genes to human aging and longevity has yet to be fully established. In this study, we show that human homologs of invertebrate longevity genes change in their expression levels during aging in human tissue. We also show that human genes encoding proteins that interact with human longevity homolog proteins are also changed in expression during human aging. These observations taken together indicate that the broad patterns underlying genetic control of life span in invertebrates is highly relevant to human aging and longevity. We also present a collection of novel candidate genes and proteins that may influence human life span.
| Genetic modulation of life span is ultimately mediated through proteins, and the mechanisms that allow this control must necessarily involve the interaction of multiple proteins. As a biological pathway, aging is a pleiotropic process, and many of the proteins identified as influencing this process have a proportionate pleiotropy of function. Modulations of the levels in a single protein have been found that provide robust increases in life-span for an organism [1],[2], but contributions from many genes are expected to dictate longevity in all organisms. This idea is supported by an investigation of yeast protein-protein interaction networks that found that proteins related to aging have a significantly higher connectivity than expected by chance [3]. Similarly, a second group found that their computational model suggested aging genes have more connections in interaction networks, and that this may be useful in identifying new aging genes [4]. Therefore, a useful way to identify novel genes with roles that affect life span is to identify their gene product's interactions with known aging-associated proteins.
A role for protein interactions in processes is most apparent at the level of protein complexes that assemble to carry out a particular function. Likewise, protein interactions that mediate signaling cascades demonstrate how interactions functionally translate into a biological pathway. Indeed, biological processes are built of hierarchical protein-protein interaction assemblies that together carry out the overall physiological process. Therefore, the identification of interactions that a protein participates in can be an informative way to pursue an understanding of the protein's function. A common method for identifying protein interactions is the yeast two-hybrid system (Y2H), which uses the interaction of two proteins to reconstitute a transcription factor that then activates expression of a reporter gene [5]. An important development in the Y2H approach was the introduction of the screening of libraries of potential interacting proteins [6]. This development made it possible to identify novel protein interactions. Novel interactions impart a suggested role in a physiological process for proteins based on the established involvement of their interaction partner in that process.
Recently, high throughput approaches have expanded this idea to a systems-scale level: investigators can identify the network of interactions that occur among a large set of proteins, and from this infer the relationships of those proteins in, as well as their contribution to, the system. Such an approach has been used to interrogate the protein interaction networks that underlie model organisms [6]–[12], human cells [13],[14], and organisms responsible for infectious diseases [15]–[17]. Biological processes such as vulval development in nematodes [18], and familial neurodegenerative diseases [19]–[21] have also been the subject of large-scale Y2H interaction mapping. From these studies, many hypotheses for new participants in biological pathways have emerged.
The results from high-throughput protein interaction studies are known to contain false-positive (i.e. biologically irrelevant) interactions intermingled with the biologically relevant interactions. Independent large-scale studies of the same system may not necessarily distinguish the two [22], although detection of an interaction in more than one study is strong evidence for the authenticity of the interaction. An additional approach to address interaction validity is to use features of the network itself to provide evidence for the physiological relevance of the identified interactions. Protein interaction networks behave as scale-free networks, and the resultant properties such as path length and clustering features can be mined with bioinformatic methods to evaluate the properties of a given interaction within the network [23],[24]. Comparisons with other phenotypic data can provide further support. An observation of similar regulation using gene expression analysis has been used to establish confidence in protein interactions by a number of groups [8],[11],[15],[25],[26]. Shared gene ontology annotations [27] can also be used to identify characteristics of proteins that support the link(s) suggested by the interactions [15].
We performed a comprehensive survey of the published literature on the genetics of aging as studied in model systems (yeast, fly, nematode and mouse) and identified 363 genes that have been reported to increase life span when mutated. Most of these genes were curated in the SAGE KE Genes/Interventions Database (http://sageke.sciencemag.org). The remainder were culled from published large-scale genetic screens for longevity phenotypes [28]–[32]. In order to characterize these longevity genes/proteins in the context of a human protein interaction network we sought to analyze their protein interaction partners in a large human protein interaction database. We have used high-throughput yeast two-hybrid methods to construct a large network comprised of 114,689 unique binary interactions between fragments of human proteins. This network was generated using results from ∼345,000 individual yeast two hybrid screens. Aspects of the Prolexys human protein interaction network and methods used to generate it have been described previously [15],[21],[33]. The 114,689 interaction network was filtered to create a Core Network with 70,358 unique binary interactions between protein fragments representing 10,425 unique genes curated as NCBI RefSeq entries. The Core Network was generated by removal of “sticky” proteins identified using a K-means clustering method [15]. Exclusion of bait proteins with >87 interactions and prey proteins with >231 interactions resulted in removal of 44,331 interactions and 855 nodes (i.e. unique genes) from the unfiltered network. Figure 1A shows a log-log graph of node degree distributions of the unfiltered network (black circles) and the Core Network (red circles). The fact that the degree distribution appears as a straight line on a logarithmic plot indicates that the Core Network is scale-free [23],[34]. This Core Network was queried to determine the interaction properties of human protein homologous to proteins experimentally implicated regulation of life span. A masked version of the complete Core Network is shown in Table S6.
The majority of genes and proteins identified as having a role in modulation of life span were discovered in yeast, fly and nematode. We therefore identified the human orthologs and homologs of these invertebrate longevity genes according to definitions used in NCBI's Homologene (http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?DB=homologene) and the Karolinska Institute's Inparanoid Database (http://inparanoid.cgb.ki.se/). Of the 363 invertebrate longevity genes identified, 252 have human homologs and 175 of these homologs are represented in the Core Network of the Prolexys protein interaction database (Table S1). The proteins encoded by the 175 human homologs of invertebrate longevity genes were observed to interact with 2,163 additional human proteins in the yeast two-hybrid assays. This longevity protein interaction network ultimately consists of a total of 3,271 binary connections between the 2,338 proteins (Table S2). When the longevity network was derived from the Core Network it was immediately apparent that the longevity homologs were unusually highly connected with an average node degree of 18.8 and a median node degree of 7.0 (see Table S1 for individual node degrees). These values are notably higher that average and median node degrees of 13.5 and 5.0 observed for the entire Core Network (Table S6). Figure 1B shows a box plot comparing the distribution of node degrees for the 175 human longevity protein homologs and the Core Network from which the longevity sub-graph was derived. This indicates that human homologs of longevity proteins comprise a group of highly connected hubs in the Core Network. The increase in the median node degree for the longevity proteins indicates that this distribution is not due to the effect of outliers.
A path length analysis was performed to determine whether the network of longevity protein homologs were more closely connected to each other than would be expected by chance. Figure 2A shows the average mean shortest path length in 1,000 sets of 175 proteins selected at random from the Core Network is 4.61 as compared to 4.15 for the longevity network (p = 0.004). This result is consistent the prediction that proteins with shared functions (in this case the modification of life span) are more likely to be closely associated in the network than would be expected by chance. To determine whether this path length difference is a trivial result of the high average node degree in the longevity network, we performed a path length analysis using networks with randomized connections. In order to do this, the edges in the Core Network of 70,358 binary interactions were randomly reassigned while preserving the node degree of each individual protein. The average path length between the longevity protein homologs present in 100 randomized core interactomes was then determined. As shown in Figure 2B, we found that the average shortest distance between any two longevity proteins (4.15) is significantly less than the average distance of 4.73 (±0.13) between these proteins in the 100 networks with randomly assigned interactions (p<0.01). This result shows that the decreased path length observed in the longevity network is not simply a feature of high node degrees but is in fact dependent on the connections between the specific interacting proteins included in the longevity network.
The 2,163 human proteins that interact with the invertebrate longevity homologs are not known to be involved directly in aging or longevity phenotypes. However, because of their ability to bind directly to known longevity proteins in the yeast two-hybrid assay, these can be considered as candidate longevity proteins. To validate potential roles for the interacting proteins in human longevity we looked for evidence that the expression of genes encoding these proteins might be changed during the aging process. To do this, we compared the network to DNA microarray datasets comparing gene expression in human skeletal muscle from cohorts of young and old healthy volunteers [35]. In this microarray study, skeletal (vastus lateralis) muscle biopsies from healthy older and younger adult men and women were compared using gene expression profiling. After quantile normalization, the number of genes significantly differentially expressed with age was determined by performing, on a probe-by-probe basis, 24,354 two-sample t-tests. To control the family-wise error rate (FWER), the significant genes were chosen at 5% using Holm's step-down method. FWER was used to insure a low probability of any false positives among this list. Using a false discovery rate cut-off of 5%, a large number of genes were found to be differentially expressed as a consequence of age [35].
To integrate the longevity interactome with the gene expression data, we asked whether any of the genes encoding longevity proteins or their interactors (“1° interactors”) were significantly changed in the transcript profiles from old vs. young human cohorts. Of the 175 longevity proteins, 169 were represented on the microarray used in this study by 210 probes. We determined how many of the 210 probes had a significant association of expression and age using analyses based on loess normalized intensities converted to log scale. HOPACH (Hierarchical Ordered Partitioning and Collapsing Hybrid) was then used to cluster the resulting genes and generate plots of similarly expressed genes. This analysis identified 54 of the 210 probes (52 of 169 unique genes) as being differentially expressed between the old and young cohorts (FDR q-value<0.05). The differentially expressed aging gene homologs are listed in Table 1.
To see whether this was unusual, we included an additional test to determine whether this set of probes is more enriched in genes associated with age than one would expect by pure chance. We drew randomly from the original list of genes probes (24,354 probes genes) 210 at a time and for each of these random draws, examined the number of genes probes significantly associated with age at the same level of significance. However, among only 1 of the 1,000 random draws we performed, did that many or more significant genes probes come up, implying a significant enrichment among this set (p-value = 0.001; see Figure 3B). A permutation test for all 236 gene longevity gene homologs present on the microarray (represented by 291 probes) is shown in Figure 3A. We found among the longevity gene homologs (regardless of whether they were present in the interaction network data), 66 out of 291 probes were significantly associated with age. However, among only 4 of the 1000 random draws we performed, did that many significant genes come up, implying a significant enrichment among this set as well (p = 0.004).
We next evaluated the 2,507 probes that correspond to genes encoding 2,036 of the 2,163 1° interactors in the longevity interactome network. We repeated the analyses described above for the longevity proteins. Among the 1° interactors, 611 out of the 2,507 probes (581 of 2,036 genes) were significantly associated with age. In 1,000 random draws of 2,507 probes, none contained 611 (or more) significant genes, demonstrating significant enrichment among the set of 1° interactors (p<0.001; see Figure 3C).
These statistical analyses clearly demonstrate that genes encoding human homologs of invertebrate longevity genes and genes encoding their interacting proteins are highly enriched among genes with a statistically significant change in expression between young and old muscle tissue in human. This result is somewhat surprising in that these genes are derived primarily from experiments done in invertebrate models, and thus one might not expect a priori to see age-dependent changes in expression levels in human tissue. Two preliminary conclusions are suggested by these observations: 1) longevity genes discovered in invertebrate models are likely to play some roles in human longevity and 2) cells and tissues appear to modulate expression levels of such longevity genes during the aging process in human. A list of human homologs of invertebrate aging genes and genes encoding interacting proteins that show significant expression changes in aging human muscle are shown in Table S3.
Figure 4 shows a subnetwork of the longevity interactome. This subnetwork includes only those genes whose expression is significantly changed in the aging microarray data. This subnetwork contains 339 interactions among 325 proteins, roughly 10% of the interactions in the larger network. We consider proteins in this network to be of high interest for further studies. An example of one group of interest is FRAP1 (mTOR) and its interacting proteins. FRAP1 has total of 63 interacting protein interactions in the longevity network.
FRAP1 has a well-established role in longevity, with loss of function mutations in the FRAP1 orthologs in both nematodes [36] and yeast [30],[32] leading to increased life span. Our results suggest that FRAP1 may also have a role in human longevity. Human FRAP1 interacts with 63 proteins that have not previously been shown to be involved in longevity. Some of these have functions that are consistent with known FRAP1 functions of FRAP1, e.g. an interaction with RPS27, a component of the small ribosomal subunit may be related to the function of FRAP1 in translational control; similarly, nuclear import of FRAP1 is necessary for signaling through S6K and an interaction with TPR supports the idea that mTOR associates with the nuclear pore [37]. Interestingly, mRNA levels for 24 of the 63 partners (38%) of FRAP1 are significantly different between young and old patient samples. Proteins that can interact with FRAP1 are thus frequently expressed differentially with age in human. FRAP1 interacting proteins that show significant changes in gene expression during aging in human muscle are shown in Table 2.
To determine whether there is a relationship between protein interaction and a correlation in gene expression between protein pairs in this network, we compared the distribution of both negative and positive gene expression correlations with binary interactions. Figure 5 shows the distribution of gene expression correlations for the experimentally derived longevity network as compared to simulated networks of genes with randomly assigned binary connections. Both positively and negatively co-regulated protein pairs are enriched in the longevity interaction network relative to that observed in randomized networks. This observation supports the idea that interacting proteins are transcriptionally co-regulated [38]. A list of the binary pairs with significant age-dependent transcriptional co-regulation is shown in Table S4.
In order to test the hypothesis that interacting partners of human longevity homologs might themselves be longevity proteins we tested a group of these for effects on life span in C. elegans. The 24 FRAP1 interacting proteins with significant gene expression changes in aging human muscle are listed in Table 2. Of these 18 were tested for their ability to modulate life span in C elegans using RNAi mediated knock-down (six of 24 were not tested because reagents were not available in our RNAi library). Wild-type N2 C. elegans were fed E. coli expressing double-stranded RNA corresponding to genes encoding 18 FRAP1 interacting proteins and life spans were determined in two independent experiments. Of the 18 genes tested in this way, six reproducibly extended the life span of C. elegans by >10% (Figure 6). These genes are listed in Table 3. The gene showing the greatest effect on life span after RNAi treatment is RPS27. Knock-down of rps-27 expression in nematode resulted in 50% and 44% increases in life span in two independent experiments. Mammalian RPS27 encodes a zinc finger-containing protein component of the 40S ribosomal subunit [39]. Several studies have established that TOR signaling can modulate life span in yeast [30],[32] and fly [40]. It has been demonstrated further that inhibition of translation can also extend life span indicating that loss-of-function in TOR signaling modulates aging through an effect on rates of translation [41]–[43]. Since RPS27 is a component of the ribosome and interacts with FRAP1 (Tor), it is likely that the life span extension seen in the rps-27 knock-down is due to an effect on rates of translation either through TOR signaling, direct effects on ribosome structure, or a combination of the two.
The fact that 33% of the candidates tested had a significant effect on life span extension is noteworthy. Previous genome wide screens in C. elegans using RNAi have reported that less than 1% of the nematode genome may encode genes that can extend life span when knocked-down [28],[29].
We present here a large protein interaction network comprised of human homologs of genes known to influence longevity in invertebrate systems and their interacting proteins. To compile this list of homologs, we selected genes that confer increased life span when mutated, deleted or knocked down in yeast, flies or nematodes. The longevity homolog sub-network (3,271 interactions) is derived from a much larger Core Network (70,358 interactions) that was generated in an unbiased fashion using a random high throughput yeast two hybrid process. The Core Network was generated from larger network after removal of sticky proteins with very high node degrees [15],[21],[33]. Analysis of the human longevity interactome presented here show that the 175 human longevity homologs are more closely connected that would be expected by chance, with a mean path length of 4.15 as compared to and average of 4.61 in the Core Network. Another striking feature of human homologs of invertebrate longevity proteins is their exceptionally high average node degree of 18.8 (as compared to an average of 13.5 in the Core Network). This observation indicates that human longevity protein homologs may function as hub proteins in the human interactome [44],[45]. The fact that longevity proteins are hubs may be indicative of their having a central role in cellular function. They may also function as nodes that connect and/or integrate functionally diverse cellular components and systems. It is interesting to consider the possibility that knock-down of these longevity genes may extend life span through a mechanism that involves uncoupling connections between cellular components of diverse function.
A striking conclusion of this study is dramatic degree of enrichment for genes encoding network proteins among genes that are transcriptionally modulated during aging in human muscle tissue. This correlation indicates that the network is enriched for proteins involved in human aging. This conclusion is consistent with the observation that human proteins interacting with the longevity homolog FRAP1 can increase life span when knocked-down in C.elegans. Overall these results provide evidence that the broad class of longevity proteins identified in invertebrates have a conserved role in processes of human aging and longevity.
Complementary DNA was generated from poly(A)+ RNA isolated from multiple human tissues (including adult brain, fetal brain and liver) and inserted between the Gal4 transcriptional activation domain and the Schizosaccharomyces pombe URA4 coding region of pOAD.102 (prey plasmid) or the Gal4 DNA-binding domain and the S. cerevisiae MET2 coding region of pOBD.111 (bait plasmid). Yeast transformed with bait or prey plasmids were plated on medium lacking uracil (prey) or methionine (bait) to select for transformants expressing the markers fused to the cDNA inserts. Additional information about the plasmids, yeast strains and library construction can be found in Supplementary Information.
The two-hybrid expression plasmids, pOBD.111 and pOAD.102 used in this study have been described [15]. pOBD.111 and pOAD.102 are modifications of pOBD and pOAD [46]. The bait and prey yeast strains used were respectively, R2HMet (MATα ura3-52 ade2-101 trp1-901 leu2-3, his3-200 met2Δ::hisG gal4Δ gal80Δ) and BK100 (MATa ura3-52 ade2-101 trp1-901 leu2-3,112 his3-200 gal4Δ gal80Δ GAL2-ADE2 LYS2::GAL1-HIS3 met2::GAL7-lacZ), a derivative of PJ69-4A [47]. Bait and prey cDNA libraries were made using poly(A)+ RNA prepared from human tissues (see Table S5) by random primed cDNA synthesis followed by the PCR addition of yeast recombination tails. Both bait and prey cDNAs are cloned as a double fusion between the two-hybrid domain on the 5′ end of the insert and an ORF-selection marker on the 3′ end. Specifically, bait cDNA inserts were cloned between the GAL4 DNA binding domain and the TRP1 or MET2 coding regions, and prey inserts between the GAL4 transcriptional activation domain and URA3 [15]. These cDNAs were then cloned into linearized expression vectors by recombination in yeast [46]. Yeast transformed with bait were plated on medium lacking tryptophan or methionine to select for in-frame TRP1 or MET2 fusions, respectively, and prey were selected without uracil for in-frame URA3 fusions.
Y2H screens were performed in 96-well plates by mating in each well 5×106 cells of a yeast clone expressing a single bait with 5×106 clonally diverse cells from a prey library. After mating overnight, the Matings were plated using a Genesis Workstation 150 liquid handling robot (Tecan) onto medium that selected simultaneously for the mating event, the expression of the ORF-selection markers, and the activity of the metabolic reporter genes, ADE2 and HIS3. Yeast that grew on this selection medium (“positives”) were counted and transferred into liquid medium in a 96-well format using a MegaPix colony picking robot (Genetix). A maximum of 48 colonies per mating were picked. Searches that yielded more than 200 positives (∼2% of all searches) were considered to result from bait plasmids that activated transcription in the absence of specific protein-protein interactions, and were not analyzed further. Cloned inserts were amplified from plasmid PCR. Liquid cultures grown from positive yeast colonies were used as templates in PCR reactions that amplified either both bait and prey cDNA inserts, or prey inserts only in screens in which the baits had been sequenced before the matings. The PCR reactions were assembled in 384-well format using the Genesis Workstation 150 or a custom built (Zymark) PCR workstation that included a SciClone ALH 500 liquid handling robot (Zymark). PCR amplification took place in Primus-HT thermocyclers (MWG Biotech). The amplicons served as templates in DNA sequencing reactions. Identities of insert fragments were established by querying against the NCBI RefSeq database. The Y2H protein-protein interaction database is the result of two distinct workflow modes referred to as random and directed. In the random mode individual bait clones are picked randomly from a library and mated with a library of prey cDNAs. Directed searches, on the other hand, are matings of prey libraries with a single intentionally constructed bait cDNA clone whose identity is known a priori. In random searches, moreover, the identity of the bait is discovered – depending, again, on a particular workflow – either before or after the mating has been performed. The alternatives are to sequence both the bait and prey from Y2H positives (called positive-derived sequence) or to sequence the bait plasmid before mating (called pre-sequencing) requiring only the prey to be sequenced from positive diploids. All Y2H search data and DNA sequences used to determine interaction pairs reported in this study are included in Table S5.
A total of 363 genes that had been reported to increase life span when mutated yeast, fly, nematode and mouse species were compiled from SAGE KE and the published literature. We then screened for their respective clusters in Homologene and Inparanoid databases. The human genes among those clusters were deemed to be the orthologs of the respective invertebrate genes. Any additional human paralogs were also taken into consideration. The 363 invertebrate genes have homology to genes had human ortholog/paralog which resulted in a total of 252 human genes.
k-means clustering (k = 2) was applied sequentially to prey and baits in the core protein interaction database to define two populations of genes based on their number of partners [15]. Those interactions involving genes (i.e. baits with >87 interactions and preys with >231 interactions) were deemed promiscuous by this analysis and removed from the final dataset. The remaining interactions were referred to as the “Core Network”. The unfiltered core interactome had a total of 120,779 interactions involving 11,327 genes curated as NCBI Gene entries. The Core Network after filtering comprised of 71,814 interactions from 10,430 genes. The aging interactome reported here includes only interactions from the Core Network.
To establish the basis for suitable null hypotheses, the process of deriving subnetworks from the large interaction network was performed 1000 times with sets of 175 genes randomly selected from one of two sources: 1) any gene contained in the Y2H PPI database or 2) genes in either Homologene or InParanoid having homologs of C. elegans, D. melanogaster or S. cerevisiae. Because the latter set corresponds to genes conserved from phylogenetically distant organisms it is referred to as “ancient.” In each iteration of the process, the 175 genes were used to query the Y2H PPI database and create subnetworks in a manner otherwise identical to that of the procedure for longevity homologs.
The mean shortest path length between any two aging genes in the actual longevity network was calculated. We simulated the Core Network 100 times, by rewiring the edges, preserving the node degree of each protein. The aging related human genes were then screened through 100 randomized networks, to generate 100 simulated longevity networks. We then calculated the mean shortest path length between any two aging genes in the 100 randomized networks. A one sided t-test was used to compare mean shortest path lengths of the experimentally derived data to those of 100 randomizations.
No background correction was performed given the very low levels of background intensity, however we performed loess normalization [48] on the entire set of probes to account for differences in the distribution of intensities among arrays. To select the genes that are differentially expressed with regards to age among the probes that matched our set of longevity proteins we performed, gene by gene, simple two-sample t-tests and used the Benjamini-Hochberg procedure [49] to derive adjusted q-values for the list of genes ranked by statistical significance. After deriving the number of significantly differentially expressed genes (based on an FDR cut-off of 5%), we wished to determine if this set of probes was significantly enriched with genes whose expression changes related to age, which motivated a permutation test to find whether the identification of a gene is related to life span extension was independent of differential expression with regards to the microarray data on muscle tissue in old and young subjects. We simply performed a large number of permutations on the longevity protein label for the total set of probes, each permutation randomly designated genes as either longevity protein genes or not and then among this random set, we performed the same procedure to find the number of significantly differentially expressed genes. After 1000 permutations, we have 1000 randomly generated numbers of significantly differentially expressed genes and we can compare our observed number to this null distribution to find the p-value of the test that these genes (related to life extension) or unrelated to age in human muscle. We performed an identical analysis for the 1° interactor genes.
To examine whether probes for genes encoding binary interaction pairs had more evidence of co-regulation in the microarray data, we examined correlation of log2 expression of probes of pairs of genes that were 1) connected directly and randomly chosen equal number of pairs of probes for pairs of genes unconnected in the network from the total list of probes on the Illumina array. For genes connected in the interactome represented by more than one probe, the correlation of all relevant pairs of probes were estimated (i.e., if there were 3 probes in one gene matched with 2 probes in another, this generated a total of 6 correlations). The purpose of this was to determine whether genes connected in the interactome were more related in expression than randomly drawn pairs of genes.
Animals were grown on NGM agar plates seeded with OP50 E. coli at 20°C. RNAi bacteria strains were cultured as previously described [50]. Wild-type N2 animals at the late L4 larval stage were fed with E. coli expressing different double-stranded RNAs and incubated at 25°C for life span experiments. 5-fluorodeoxyuridine (0.05 mg/ml) was added onto plates during the reproductive phase to eliminate progeny. Animals were transferred onto fresh plates every 3–6 days. The first day of adulthood is Day 1 in survival curves. Animals were scored as alive, dead or lost every other day. Animals that did not move in response to touching were scored as dead. Animals that died from causes other than aging, such as sticking to the plate walls, internal hatching or bursting in the vulval region, were scored as lost. In all life span assays, E. coli carrying the empty RNAi vector L4440 was fed to animals as controls. Statistical analyses were performed using the Prism 4 software (Graphpad Software, Inc., San Diego, CA, USA). Kaplan–Meier survival curves were plotted for each life span experiment and p values were calculated using the log-rank test [50].
|
10.1371/journal.pgen.1004227 | Epigenetic Control of Effector Gene Expression in the Plant Pathogenic Fungus Leptosphaeria maculans | Plant pathogens secrete an arsenal of small secreted proteins (SSPs) acting as effectors that modulate host immunity to facilitate infection. SSP-encoding genes are often located in particular genomic environments and show waves of concerted expression at diverse stages of plant infection. To date, little is known about the regulation of their expression. The genome of the Ascomycete Leptosphaeria maculans comprises alternating gene-rich GC-isochores and gene-poor AT-isochores. The AT-isochores harbor mosaics of transposable elements, encompassing one-third of the genome, and are enriched in putative effector genes that present similar expression patterns, namely no expression or low-level expression during axenic cultures compared to strong induction of expression during primary infection of oilseed rape (Brassica napus). Here, we investigated the involvement of one specific histone modification, histone H3 lysine 9 methylation (H3K9me3), in epigenetic regulation of concerted effector gene expression in L. maculans. For this purpose, we silenced the expression of two key players in heterochromatin assembly and maintenance, HP1 and DIM-5 by RNAi. By using HP1-GFP as a heterochromatin marker, we observed that almost no chromatin condensation is visible in strains in which LmDIM5 was silenced by RNAi. By whole genome oligoarrays we observed overexpression of 369 or 390 genes, respectively, in the silenced-LmHP1 and -LmDIM5 transformants during growth in axenic culture, clearly favouring expression of SSP-encoding genes within AT-isochores. The ectopic integration of four effector genes in GC-isochores led to their overexpression during growth in axenic culture. These data strongly suggest that epigenetic control, mediated by HP1 and DIM-5, represses the expression of at least part of the effector genes located in AT-isochores during growth in axenic culture. Our hypothesis is that changes of lifestyle and a switch toward pathogenesis lift chromatin-mediated repression, allowing a rapid response to new environmental conditions.
| Effectors are key players in pathogenicity of microbes toward plants. Effector genes usually show concerted expression during plant infection but how this concerted expression is generated remains a largely unexplored research topic. Epigenetic mechanisms are involved in genome maintenance and integrity but are increasingly considered as important for regulation of gene expression in numerous and diverse organisms. Here we show that the genomic environment has impact on expression of Leptosphaeria maculans effector genes, and that an epigenetic mechanism that relies on two proteins involved in heterochromatin formation and maintenance, HP1 and DIM-5, modulates this expression, leading to repression during growth in axenic culture. Chromatin decondensation by removal of histone H3 lysine 9 methylation and/or HP1 is presumably a prerequisite for effector gene expression during primary infection of oilseed rape. Thus we show chromatin-based transcriptional regulation that can act on effector gene expression in fungi. Our study highlights the importance of heterochromatic landscapes, not only for genome maintenance but also in rapid and efficient adaptation of organisms to changing environmental situations.
| During infection, plant pathogenic microbes secrete a set of molecules collectively known as effectors, pathogenicity determinants that modulate plant innate immunity and enable parasitic infection [1]. Effectors are either targeted to the apoplast or the cytoplasm of the plant where they are mainly involved in overcoming the host immune defence system, nutrient uptake and, eventually, symptom development [2]. In many plant pathogenic microbes, effectors show common features. They are small proteins, potentially secreted (as many possess signal peptide sequences), generally cysteine-rich and usually have no homology to known proteins in databases [3]. Thus, these proteins are referred to as Small Secreted Proteins (SSPs). Bioinformatic analyses based on common features of effectors aim to identify the complete repertoire of candidate effector genes of several oomycetes [4], [5] and filamentous fungi [6]–[8]. These studies highlighted another common trait of effector genes: their concerted up-regulation upon plant infection which suggests co-operation between a subset of effectors at different time points [9], [10].
While high throughput functional analyses of effectors are in progress in several research groups (e.g. in Ustilago maydis [6], in Magnaporthe oryzae [11] and in Phytophthora infestans [12]), less effort has focused on the regulation of effector gene expression. Effector genes of filamentous plant pathogens are often located close to dispersed transposable elements (TEs) or in TE-rich regions of the genome, like dispensable chromosomes or telomeres [13]. For example, in Leptosphaeria maculans, the effector gene encoding AvrLm11 was found to be located on a conditionally dispensable chromosome (CDC) [14], while the avirulence gene of M. oryzae Avr-Pita is located in a sub-telomeric region [15]. In Fusarium oxysporum, all known effector genes are located in one of its four dispensable chromosomes [16], and in P. infestans effector genes are located in highly plastic genomic regions, enriched in TEs [5]. The location of effector genes in dynamic genomic regions, enriched in TEs is suggested to have an impact on adaptability to new host plants [17], [18]. Here we postulate that it could also influence their expression.
L. maculans is an ascomycete fungus belonging to the Dothideomycete class that causes stem canker of oilseed rape (Brassica napus). Sequencing of the L. maculans genome has revealed an unusual genome structure compared to other fungi as it contains alternating GC-equilibrated (∼51% GC) and AT-rich (∼33.9% GC) blocks, which we call “GC-” and “AT-isochores”. The GC-isochores are enriched in housekeeping genes whereas the AT-isochores are gene-poor, enriched in TEs truncated and degenerated by Repeat-Induced Point mutation (RIP) and comprise 36% of the genome but only 5% of the genes [7]. One hundred and twenty-two (∼20%) of all genes encoding SSPs have been identified in the AT-isochores [7] including the CDC of L. maculans. These genes include five with experimentally demonstrated effector activity (AvrLm1, AvrLm6, AvrLm4-7, LmCys2 and AvrLm11; [14], [19]–[21], and our unpublished data). These effector genes share similar expression profiles, namely very low expression in axenic culture and a drastic increase in expression during primary leaf infection, with a peak of expression 7 days post inoculation (dpi). This expression pattern is a common characteristic of SSP-encoding genes located in AT-isochores, as 73% of these genes are overexpressed at 7 dpi (compared to 19% of SSP-encoding genes in GC-isochores) when compared to growth in axenic culture [7]. AT-isochores in L. maculans were postulated to be heterochromatic regions because they share characteristics of heterochromatin in many organisms: (i) they are rich in TEs affected by RIP [7], [22], [23]), (ii) they are gene poor and (iii) they show lower rates of recombination compared to GC-isochores [7]. Based on this, we speculated that AT-isochores may be targets of reversible epigenetic modifications that affect the regulation of effector genes and would thus be instrumental in concerted expression at the onset of plant infection.
Most eukaryotic DNA is associated with histones and non-histone proteins to form chromatin. At least two different states of chromatin are commonly distinguished: gene-rich euchromatin and gene-poor heterochromatin [24]. Euchromatin is less condensed, more accessible and generally more easily transcribed than heterochromatin. The latter is highly condensed and less accessible to the transcriptional machinery. Additionally, heterochromatin is discriminated into “constitutive heterochromatin”, mainly found at centromeres, sub–telomeres and rDNA clusters, and is thought to be involved in genome stability and integrity, and “facultative heterochromatin” that is dispersed throughout the genome. Facultative heterochromatin has attracted considerable attention, as it allows for epigenetic regulation of gene expression by changing chromatin states. Chromatin states can be classified by distinct combinations of post-translational modifications targeted to histones, commonly referred to as the “histone code” [25], [26]. Heterochromatin is enriched in specific modifications, which are thought to be heritable and reversible, and thus called “epigenetic”. These modifications include lysine hypoacetylation of histone H3 and H4, as well as trimethylation of H3 lysine 9 (H3K9me3) and lysine 27 (H3K27me3) [27]. Other hallmarks are presence of Heterochromatin Protein 1 (HP1) and cytosine DNA methylation [24]. In fungi, mechanisms of heterochromatin formation have been investigated mostly in Neurospora crassa, where H3K9me3 is catalysed by DIM-5 [28], [29]. The HP1 ortholog of N. crassa binds to H3K9me3 via its chromo domain and recruits the DNA methyltransferase DIM-2 via HP1 chromo-shadow domain, resulting in DNA methylation in essentially all AT-rich heterochromatic regions [28]–[32].
On the basis of the specific location of effector genes in AT-isochores of the genome of L. maculans, we investigated whether the concerted expression of these effector genes is influenced by their genomic environment. We generated L. maculans transformants in which expression of the L. maculans orthologs of DIM-5 and HP1 (LmDIM5 and LmHP1) was decreased, analysed chromatin states in the transformants by cytology and performed oligoarray experiments to study the influence of LmHP1 and LmDIM5 on gene expression, with a focus on effector genes. We also analysed effects of changing genomic environment from AT- to GC-isochores on effector gene expression. Our data showed HP1- and DIM-5-mediated repression of effector genes during growth in axenic culture and strongly suggest an epigenetic mechanism is the basis of concerted expression of effector genes during plant infection.
Orthologs of N. crassa DIM-5 and HP1 were identified in L. maculans by bidirectional best hits with BLASTp [28], [30]. LmHP1 encodes a protein of 237 amino acids (37% of all residues are identical to N. crassa HP1), containing the typical N-terminal chromo domain (IPR000953) and the C-terminal chromo-shadow domain (IPR008251; Figure 1) [33]. LmHP1 was further annotated by RACE-PCR. Five introns, a 5′UTR of 140 bp and a 3′UTR of 444 bp were identified (GenBank accession number CBX96122). LmDIM5 encodes a protein of 516 amino acids (36% identity with N. crassa DIM-5). This family of Su(var)3–9 histone methyltransferases was first identified in Drosophila melanogaster [34] and is evolutionarily conserved. LmDIM5 exhibits the two typical domains of DIM-5 proteins described so far, a central Pre-SET (or CXC) domain (IPR007728) and a C-terminal SET domain (IPR001214) (Figure 2) [35]. The N-terminal region of DIM-5 is less conserved (Figure 2), as some proteins (i.e. Su[var]3–9 from D. Melanogaster, Homo sapiens and Clr4 from Schizosaccharomyces pombe [34], [36], [37]) have a chromo domain, which is absent from the protein in filamentous ascomycetes (Figure 2). Six introns, a 5′UTR of 259 bp and a 3′UTR of 262 bp were identified in LmDIM5 by RACE-PCR (GenBank accession number CBX92341).
LmHP1 and LmDIM5 were silenced by RNAi. While all primary LmDIM5 transformants grew, only one-third (30/89) of the LmHP1 transformants were able to grow after primary transformants were plated on a second round of selection medium. Expression of LmHP1 and LmDIM5 during mycelial growth was determined by qRT-PCR in a subset of transformants (Figure 3). Six transformants showed only 12–20% of LmDIM5 expression compared to the wild type strain, while expression of LmHP1 was never lower than 33% of wild type expression (Figure 3). Five transformants per construct were selected for further characterisation, four transformants with significant levels of silencing and one control transformant each for which LmHP1 or LmDIM5 expression was similar to that of wild type.
Transformants in which either LmHP1 or LmDIM5 were silenced showed an average linear growth reduction of 34% in axenic culture when compared to wild type and non-silenced transformants (Figures 4A and 4B). These transformants were not affected in sexual reproduction when crossed with strain v24.1.2 (data not shown). No pathogenicity defect was associated with silencing of LmHP1 (Figure 4C) but silencing of LmDIM5 resulted in reduced pathogenicity (Figure 4D).
We used LmHP1 as a convenient cytological marker for heterochromatin. As the nuclear localisation of HP1 depends on the action of N. crassa DIM-5 [30], we investigated localisation of LmHP1 in nuclei of wild type and in silenced-LmDIM5 transformants. A LmHP1-GFP fusion under the control of the LmHP1 endogenous promoter was introduced in the wild type strain and in two different silenced-LmDIM5 transformants (.1 and .4, with residual LmDIM5 expression of 20 and 16%, respectively). Nuclei were visualised by DAPI staining. In the wild type strain, LmHP1-GFP was located within strongly fluorescent foci throughout the nucleus and the same kind of dense foci were revealed by DAPI staining demonstrating coexistence of condensed and relatively open chromatin states along the chromosomes, and the preferential localisation of LmHP1-GFP to densely DAPI-stained chromatin regions (Figure 5). In the two silenced-LmDIM5 transformants, DAPI staining in nuclei was more diffuse with fewer foci, and LmHP1-GFP was also seen as more diffuse staining throughout nuclei, with only the silenced-LmDIM5.1 showing a few clear foci in the nuclei (12 nuclei out of 100). The LmHP1-GFP localisation observed in the wild type strain and in the silenced-LmDIM5 transformants is consistent with changes of HP1 localisation observed in N. crassa dim-5 mutants [30]. Thus, LmHP1 localises in an LmDIM5-dependent manner to heterochromatic regions in L. maculans, and the silencing of LmDIM5 triggers at least partial chromatin decondensation.
We compared expression of all L. maculans gene models during growth in axenic culture of a wild type isolate, the silenced-LmDIM5.4 transformant (with 16% residual expression of LmDIM5, and for which chromatin decondensation was more pronounced; Figure 5) and the silenced-LmHP1.5 transformant (with 35% residual expression of LmHP1) using oligoarrays. The expression level of eight or six percent of the L. maculans gene models were respectively influenced by the silencing of LmDIM5 or LmHP1 (978 or 746 out of 12,012 genes with a transcriptomic support). Regarding genes down-regulated compared to the wild type strain, 72% and 59% encode proteins with no homology or hypothetical proteins with unknown functions in respectively the silenced-LmHP1 and the silenced-LmDIM5 transformants. Among the genes encoding proteins with predicted functions, we identified respectively six and four percent of putative SSP-encoding genes (Tables S1 and S2). We also found a gene encoding a protein putatively related to pathogenesis down-regulated in the silenced-LmDIM5 transformant (Table S2). In the silenced-LmDIM5 transformant, among the 390 up-regulated genes compared to the wild type isolate (Table S3), 283 encoded predicted proteins with no homology detected by BLAST or hypothetical proteins with unknown functions. The remaining up-regulated genes were mainly putative SSP-encoding genes, transporters and enzymes known to be accessory enzymes involved in the biosynthesis and transport of secondary metabolites (such as oxidases, MFS transporters, dehydrogenases and cytochrome P450 monooxygenases), CAZymes and the ortholog of the sterigmatocystin cluster regulatory gene in Aspergilli, AflR [38] (Table S3). Overall, the same functional categories of up-regulated genes were observed in the silenced-LmHP1 background, where we found 369 up-regulated genes, including 206 genes with predicted functions (Table S4). In addition, we also identified seven genes involved in DNA repair, including three genes encoding DNA repair proteins of the Rad family (Table S4), suggesting that silencing of LmHP1 may somehow trigger DNA damage. We further investigated the 71 genes that were up-regulated in both silenced-LmDIM5 and -LmHP1 transformants (Table S5). About half of these genes encode predicted or hypothetical proteins. Among genes encoding proteins with predicted functions most encode putative SSPs, cytochrome P450, transporters and CAZymes (Table S5). Notably, of the 31 highest up-regulated genes in both silenced-LmDIM5 and -LmHP1 transformants (fold-change>4; Table S5), 52% were putative SSP-encoding genes, while they only represent 5.1% of the genes in the genome [7]. This suggests a specific and strong effect of LmHP1 and LmDIM5 silencing on the expression of effector-encoding genes. We observed a stronger de-repression of gene expression in the silenced-LmDIM5 transformant compared to the silenced-LmHP1 transformant (Tables S3 and S4) that could result from the lower residual expression of LmDIM5 compared to that of LmHP1 in the transformants selected.
We next investigated the effect of LmDIM5 and LmHP1 silencing on the transcriptional behaviour of genes according to their genomic environment (Table 1). Whereas silencing of LmDIM5 and LmHP1 led to an up-regulation of ∼3% of genes located either in GC-isochores or in GC-equilibrated islands located in the middle of AT-isochores, 34% and 28% of genes located in AT-isochores were up-regulated in silenced-LmDIM5 or -LmHP1 backgrounds (Table 1). Remarkably, this effect was even more pronounced for the SSP-encoding genes as 41% of these genes, including AvrLm1, AvrLm4-7 and AvrLm11 were up-regulated when compared to the wild type strain (Table 1, Figures 6A and 6B). These observations support the idea that expression is regulated directly by alteration of H3K9me3 in AT-isochores of transformants in which LmDIM5 and LmHP1 are silenced. Moreover, genes located in AT-isochores that were up-regulated in both silenced-LmHP1 and -LmDIM5 backgrounds were all found up-regulated at 7 dpi during plant infection in the wild type isolate (Table 2). Taken together, these data support the idea that LmDIM5 and LmHP1 exert an epigenetic control on specific regions of the genome, and mainly on the AT-isochores that are expected to be associated with heterochromatic nucleosomes.
We analysed the expression of three effector genes located in AT-isochores (AvrLm1, AvrLm4-7, LmCys2) and one effector gene located in a GC-equilibrated island (LmCys1; I. Fudal and B. Profotova, unpublished data) by qRT-PCR during mycelial growth. Consistent with results from our oligoarray study, we observed a strong overexpression of AvrLm1 and AvrLm4-7 in the silenced-LmHP1 (∼80- and 148-fold respectively; Figure 7A) and -LmDIM5 transformants (∼25- and 63-fold respectively; Figure 7B), and less overexpression for LmCys2 (∼10- and 2-fold in the silenced-LmHP1 and -LmDIM5, respectively; data not shown). In contrast, LmCys1, the gene located in a GC-equilibrated island, was repressed (∼0.25- and 0.1-fold expression in the silenced-LmHP1 and -LmDIM5, respectively; Tables S1 and S2). Thus, we found derepression of LmCys2 in both silenced-LmHP1 and -LmDIM5 by qRT-PCR while no derepression was found using oligoarrays. These results confirm the oligoarray data but also suggest that oligoarrays likely underestimate the number of genes affected by the silencing, as well as the magnitude of derepression.
To investigate whether induction of effector gene expression in axenic culture after silencing of LmDIM5 or LmHP1 was associated with changes of the H3K9me3 level, we performed chromatin immunoprecipitation (ChIP) experiments on strain v23.1.3, and the silenced-LmHP1.5 and -LmDIM5.4 transformants. We assayed AvrLm1 and AvrLm4-7 coding regions, their promoters and used the histone H2A gene as a control. To precipitate nucleosomes associated with H3K9me3, we used an antibody that had been used successfully in studies with N. crassa [39], [40], Trichoderma reesei [41], [42], Fusarium fujikuroi [43] and Fusarium graminearum [44]. In the wild type isolate, we observed higher levels of H3K9me3 in the promoters of AvrLm1 and AvrLm4-7 compared to their coding regions (Figures 8A and 8B). H3K9me3 levels were strongly reduced in AvrLm1 and AvrLm4-7 for all regions tested in both the silenced-LmHP1 and -LmDIM5 transformants when compared to the wild type strain (Figures 8A and 8B); as expected, silencing of LmDIM5 or LmHP1 had no effect on the low H3K9me3 level in the H2A gene. Thus, overexpression of at least AvrLm1 and AvrLm4-7 observed in axenic culture in the silenced-LmHP1 and -LmDIM5 transformants is correlated to a measurable decrease in H3K9me3 levels in these genomic regions. Precipitations with H3K4me2 antibody were done in parallel as controls for ChIP efficiency (Figure S1). This modification has been described as associated with active regions of chromatin [45]. Accordingly, levels of H3K4me2 in the constitutively expressed gene, H2A, are high either in the WT strain or in both silenced-LmHP1 and -LmDIM5 transformants. In the WT strain, levels of H3K4me2 are reduced by 12 to 614 times in the coding sequences and promoters of AvrLm1 or AvrLm4-7 compared to the coding sequence of H2A (Figure S1), consistently with a repression of effector gene during axenic culture. However, in L. maculans, and based on levels of H3K4me2 observed in the silenced-LmDIM5 and -LmHP1 transformants, no correlation seems to exist between induction of effector gene expression and a modification in H3K4me2 levels (Figure S1).
Expression of AvrLm1, AvrLm4-7, LmCys1 and LmCys2 was monitored during oilseed rape infection (3, 7 and 14 dpi) by qRT-PCR in the silenced-LmHP1 and -LmDIM5 transformants (Figures 9A and 9B; data not shown) and compared to the wild type strain or to non-silenced transformants. Expression profiles were similar in all isolates, with a peak of expression at 7 dpi. Thus, silencing of either LmHP1 or LmDIM5 did not alter the profiles of effector gene expression during primary infection.
As previously described, Agrobacterium tumefaciens-mediated transformation (ATMT) of L. maculans mainly results in integration of transgenes in GC-isochores [46]. Isolates naturally lacking the effector genes AvrLm1, AvrLm6, AvrLm4-7 and LmCys2 were subjected to ATMT complementation [19]–[21]. Thermal Asymmetric InterLaced (TAIL)-PCRs were performed on 66 transformants in order to check that the integrated T-DNA was indeed located into GC-isochores. T-DNA flanking sequences were recovered for 54% of the transformants (Table S6) and were identified in the L. maculans genome using BLAST. After screening, 16 transformants with identified T-DNA sequences were kept for further study (24% of the sequences; see Table S6 for details), corresponding to insertions in GC-isochores of 12 different supercontigs; ten were inserted into coding sequences, two into promoters and four into intergenic regions (see Table S7 for details on insertions). In all cases, expression of AvrLm1, AvrLm6, LmCys2 and AvrLm4-7 in axenic culture was greatly increased, varying from 8- to 1270-fold compared to that of the wild type isolate (Table 3). Expression profiles in planta were similar to that of the wild type isolate with a peak of expression at 7 dpi (Figure 10), except for transformants NzT4-AvrLm4-7-16 and -18. These data corroborated the effect of LmHP1 and LmDIM5 silencing, which led to overexpression of effector genes in axenic culture with no altered expression pattern during primary infection. Our data strongly suggest that chromatin structure of AT-isochores represses expression of effector genes embedded in these regions, at least during growth in axenic culture. In contrast, during primary leaf infection, expression of effector genes located in AT-isochores did not depend on the chromatin context alone.
AT-isochores of the L. maculans genome were found to be enriched in effector genes sharing common expression profiles, with repression of expression during mycelial growth but drastic induction during the early stages of oilseed rape leaf infection [7]. We recently showed that the location of effector genes in this genomic environment has selective advantages by allowing extremely rapid responses to resistance gene selection generated by mutation and/or recombination [47] as it was also reported for Phytophthora infestans [17]. Here, we show that the genomic environment is important for control of gene regulation, and thus may confer a second adaptive advantage as it allows repression of genes specifically needed for pathogenicity during vegetative growth and possibly allows rapid induction of genes required for infection. To elucidate the mechanism involved, we investigated possible chromatin-mediated epigenetic repression through histone H3K9 trimethylation in strains where key players involved in heterochromatin formation (LmHP1 and LmDIM5) were silenced by RNAi.
The domain structures of HP1 and DIM-5 proteins and their functions are well conserved in various eukaryotes, from plants to mammals. The homologue of the N. crassa histone H3 methyltransferase DIM-5 identified in L. maculans revealed functional domains typical of Suvar39/Clr4/DIM-5 orthologs. Notably, nine cysteines shaping the zinc cluster of the pre-SET domain [35], as well as three residues (N, H, Y) with catalytic roles are well conserved [48] (Figure 2). Consistent with data obtained from N. crassa [28] and Aspergillus fumigatus [49], the silenced-LmDIM5 transformants of L. maculans showed growth defects. In contrast to DIM-5 mutants of A. fumigatus [49], silenced-LmDIM5 isolates showed pathogenicity defects. HP1 is an adapter protein that recognises and binds H3K9 trimethylation [50], though the closest Arabidopsis thaliana homologue, LHP1, also binds histone H3 lysine 27 methylation (H3K27me3) [51]–[53]. In N. crassa, HP1 recruits the DNA methyltransferase DIM-2, and has been shown to be essential for all DNA methylation in heterochromatic regions [30]–[32]. The HP1 protein of L. maculans revealed functional domains typical of all HP1 orthologs. Notably, the three aromatic residues of the chromo domain, forming a binding pocket for the N-methyl groups of the H3 tail [54] are conserved from N. crassa to H. sapiens (Figure 1). Besides heterochromatin formation, multiple roles are attributed to HP1, including involvement in centromere maintenance, genome integrity, sister-chromatin cohesion, and DNA repair [40], [55]–[57], but defects observed depend on the organism studied to date. Growth defects are observed in all HP1 mutants of N. crassa [30], and we found here similar defects in transformants in which LmHP1 was only partially silenced by RNAi. No such growth defects have been reported in HP1 mutans of F. graminearum or A. nidulans [58], [59]. HP1 mutations trigger larval death in D. melanogaster [60], [61]. In L. maculans, we observed that only one-third of primary LmHP1 transformants could be recovered following the first step of selection. In addition, the residual expression of LmHP1 in silenced isolates remained high (always more than 30% of the control), in contrast to what we observed for the silenced-LmDIM5 transformants. These data suggest that, as for D. melanogaster, LmHP1 deletion may be lethal. In contrast to transformants with silenced LmDIM5, transformants with silenced LmHP1 showed no pathogenicity defects. Whether maintenance of pathogenicity of silenced-LmHP1 transformants is due to the lack of drastic silencing of LmHP1 is an open question, as the involvement of HP1 in pathogenicity has not been analysed in any other plant pathogen. The silencing of LmDIM5 was associated with a decrease in pathogenicity ability that did not depend on the level of effector gene expression during infection. As LmHP1 and LmDIM5 silencing resulted in a deregulation of expression of a considerable number of genes, we can assume that the pathogenicity defect could also be linked to another physiological defect that would be important for infection processes that would be affected. As HP1 localisation in N. crassa relies on H3K9me3, catalysed by DIM-5, we investigated LmHP1 localisation in the wild type strain and in silenced-LmDIM5 transformants. As in N. crassa, LmHP1 location in L. maculans relies on proper function of LmDIM5, and there seems to be at least some defect in chromatin structure in the silenced-LmDIM5 background. Our ChIP experiments showed that LmHP1 and LmDIM5 are responsible for generating heterochromatic regions at two L. maculans avirulence gene loci, and that induction of effector gene expression in axenic cultures in either silenced-LmHP1 or -LmDIM5 transformants relied on reduction of H3K9me3 levels in these regions. Interestingly, the level of H3K9me3 is slightly higher in promoters of AvrLm1 and AvrLm4-7 compared to the coding regions, suggesting that promoters are the preferred targets for epigenetic regulation of gene expression. Overall, our native ChIP experiments tend to support the idea that AT-isochores with effector genes represent epigenetically controlled facultative heterochromatin. The reduction of H3K9me3 in a silenced-LmDIM5 background is consistent with DIM-5 function. We also showed here that levels of H3K9me3 is reduced in coding sequences and promoters of AvrLm1 and AvrLm4-7 while HP1 acts downstream of DIM-5. In S. pombe, the HP1 ortholog Swi6, through self-association, is required for interaction with H3K9me3 but also for the recruitment of the DIM-5 ortholog, Clr4 [62] which is in accordance with the reduction of H3K9me3 levels observed here in a silenced-LmHP1 background. Regarding H3K4me2, our results suggest that this modification does not replace H3K9me3 when genes located in facultative heterochromatin are actively transcribed. Such behaviour for this modification was also observed in F. graminearum [44].
In our oligoarray analyses, in addition to genes located in AT-isochores largely affected in both the silenced-LmHP1 and the silenced-LmDIM5 transformants, we also found 49 genes located in GC-isochores up-regulated, such as genes encoding for nine SSPs and a cytochrome P450 protein. This suggests that hotspots in GC-isochores may also be subject to chromatin-based regulation; alternatively, this up-regulation may be indirect and a consequence of expression of genes located in AT-isochores. Additional ChIP and ChIP-sequencing studies are underway to address this question. To our knowledge, up-regulation of numerous genes involved in DNA repair in transformants with silenced-LmHP1 is a novel finding. Previous results suggest that some heterochromatin mutants, including dim-5 but not hpo (N. crassa HP1) are mutagen sensitive [63]. Silencing or deletion of HP1 may trigger DNA damage, which would explain pleiotropic effects that have been also observed in other organisms as previously mentioned.
We show here that an epigenetic mechanism efficiently tunes effector gene expression and that by removal of H3K9me3, expression of these genes can be rapidly activated in response to changing environmental conditions. It is known that an intimate link sometimes exists between epigenetic regulation of gene expression and transcription factors, the latter triggering chromatin condensation or decondensation, and the three dimensional structure of the chromatin governs promoter access to regulatory proteins such as transcription factors. There is precedence for this idea, as transcription factors can be recruited to promote the reversion of epigenetic silencing, as for example, the BZLF1 transcription factor of the Epstein-Barr virus initiates virus synthesis [64]. In contrast, in Plasmodium falciparum, PfSIP2 initiates the formation of heterochromatin leading to epigenetic silencing of genes involved in pathogenesis [65]. Transcription factors are recruited after chromatin decondensation to specifically regulate certain genes; for example, expression of secondary metabolite gene clusters in Aspergilli requires modulation of H3K9me3 levels and action of AflR, a specific transcription factor [38], [66], [67], thus generating multi-layered control. Based on this, we could suspect that such a multiple control may regulate effector gene expression in L. maculans. Nevertheless, whether one, or several, transcription factor(s) is involved in the concerted expression of effector genes in L. maculans remains to be elucidated. A few transcriptional regulators involved in effector gene expression have already been identified, such as SGE1 of F. oxysporum, which is essential for parasitic growth and Fox1 of U. maydis, a forkhead transcription factor which is also required for pathogenic development into the plant [68], [69].
While all the cells in an organism contain the same genetic information, epigenetic modifications enable a specialisation of function during development, stress or infection mechanisms. In 1942, Conrad Waddington named these different states “epigenotypes”, to indicate that conditions set up under one particular situation to switch gene expression on or off can efficiently and quickly adapt to a new situation, independently of the genotype [70]. If genetic mutations can help an organism to adapt and survive along evolution, the presumed epigenetic code provides a more sensitive and rapid way to respond to environmental stress than DNA sequences themselves. For example, in viruses like human Herpes Simplex Virus 1 and 2, epigenetic modifications determine either latency of this virus, i.e. its quiescent state when there is no de novo production of the virus or lytic infection, i.e. its active state when a large number of viral genes are expressed leading to cell death [71]. As mentioned above, effector genes show common features, notably they remain globally silenced during mycelial growth and are strongly up-regulated upon infection. Producing effectors implies an energy cost for the organism. Thus, it seems plausible that effector production is tightly regulated to avoid wasting energy under conditions when the gene products are not needed. Epigenetic repression is lifted at the precise stage where effectors are required, allowing for finely tuned regulation.
The work presented here is, to our knowledge, the first report of chromatin-mediated epigenetic control of fungal effector gene expression. In organisms as phylogenetically distant and diverse as Apicomplexa, Euglenozoa, Eozoa or Ascomycota (including Plasmodium spp., Trypanosoma spp., Leishmania spp., Giardia lamblia and Candida spp.), families of genes involved in pathogenicity are frequently located in subtelomeric regions and are variantly expressed, to escape the host immune system, through an epigenetic control [72]–[74].
In fungi, particularly Aspergilli, epigenetic control is involved in the regulation of secondary metabolite genes. Genes involved in secondary metabolites production are often organised in clusters [75], including their respective transcriptional activators (e.g. AflR for aflatoxin and sterigmatocystin production; [38], [66], [67]). Clusters are often near the chromosome ends (“subtelomeric”), which may allow for regional regulation mediated by chromatin modifications. These modifications appear to be controlled by a global regulator, LaeA [76], a putative histone methyltransferase, as well as HP1 and DIM-5 homologues [59], [77]. A similar system has been described in F. graminearum [58]. In L. maculans, we found that LmHP1 modulates the expression of three polyketide synthase and nonribosomal peptide synthetase genes, and both LmHP1 and LmDIM5 influence transcription of genes encoding accessory enzymes involved in the modifications of secondary metabolites.
To date, L. maculans is the only fungus for which an isochore-like structure of the genome has been published [7] but there is evidence for the existence of this genome structure from other fungi. In the hemibiotrophic dothideomycete Venturia inaequalis, the region containing the avirulence gene AvrVg is located in isochores with widely different GC content, which may be consistent with an isochore-like genome structure [78]. This organisation is also identifiable in the genomes of two additional dothideomycetes, Mycosphaerella fijiensis (http://genome.jgi.doe.gov/Mycfi2/Mycfi2.home.html) and Cladosporium fulvum, which contain effector-encoding genes in repeat-rich regions, which were presumably mutated by RIP [79]. Thus, similar epigenetic mechanisms to control expression of genes located in these highly dynamic regions may occur in these organisms. Moreover, the location of effector genes in subtelomeric regions, regions enriched in repeats or on CDC is a common feature of plant pathogenic fungi. Like Nectria haematococca [80] and F. oxysporum [16], L. maculans genome contains a CDC, invaded by repeats. This CDC harbours an effector gene, AvrLm11 [14], which is up-regulated in axenic culture in a silenced-LmHP1 background. This suggests that genes located on supernumerary chromosomes in N. haematococca or F. oxysporum may be likewise regulated. In combination, these data strongly suggest that in the course of evolution some epigenetic mechanisms to repress and control expression of genes involved in pathogenicity have been conserved across kingdoms.
Mechanisms controlling gene activation or repression at the chromatin level are under intense scrutiny but signals triggering the switch between two chromatin states remain largely unknown. Here, we show that effector genes are repressed under axenic conditions and that repression is lifted upon leaf infection. This suggests that a particular signal is recognised by L. maculans to abolish epigenetic silencing of pathogenicity-related genes. While signals produced by plant roots begin to be characterised [81], nothing is known about the plant signals produced by leaves that could induce effector gene expression. As mentioned above, production of fungal secondary metabolites is under both general chromatin-mediated as well as specific transcription factor-mediated control, both of which likely respond to various environmental factors, for example nitrogen [43], [82]. Considering the induction of effector gene expression, the Avr9 gene is the only effector gene induced upon nitrogen starvation in C. fulvum [83] but so far no effects of nitrogen levels on the expression of effector genes in L. maculans [19] or other fungi have been found. Identification of environmental factors triggering effector gene expression as well as plant genotypes that will be less favourable for the expression of effector genes is thus a promising field for research to prevent or limit disease development.
The isolates v23.1.3 (AvrLm1-AvrLm4-AvrLm6-AvrLm7) and v29.3.1 (avrLm1-AvrLm4-avrLm6-AvrLm7), NzT4 (AvrLm1-avrLm4-avrLm6-avrLm7) of L. maculans [84] were used as hosts for genetic transformations. Transformants used here to assess effect of genomic context on effector gene expression were previously described [19]–[21] and our unpublished data, and generated during the positional cloning of AvrLm1, AvrLm6 and AvrLm4-7: v29.3.1-AvrLm1, v29.3.1-AvrLm6, v29.3.1-LmCys2 and NzT4-AvrLm4-7. v29.3.1-AvrLm1, v29.3.1-AvrLm6 and v29.3.1-LmCys2 transformants corresponded to v29.3.1 isolates in which AvrLm1, AvrLm6 or LmCys2 allele of v23.1.3 were introduced using vector pBBH [19] or pBHt2 [85] (11, 13 and 15 transformants, respectively, were tested). NzT4-AvrLm4-7 transformants corresponded to NzT4 isolate in which the AvrLm4-7 allele of v23.1.3 was introduced using vector pPZPnat1 [86] (27 transformants tested). Fungal cultures, sporulating cultures and conidia production were maintained or collected as previously described [87]. For chromatin immunoprecipitation, tissue was grown on V8 agar medium at room temperature in the dark for 10 days. Mycelium was inoculated into 75 ml of Fries liquid medium in 250 ml Erlenmeyer flasks. Tissue was harvested after growing for 7 to 12 days in the dark at 27°C.
Pathogenicity assays were performed as described [88] on cotyledons of 15-day-old plantlets of cultivar Westar, a highly susceptible cultivar of Brassica napus. Plants were incubated in a growth chamber at 16/24°C (night/day) with a 12 h photoperiod. Symptoms were scored on 10–12 plants, at 14 days after inoculation using the IMASCORE rating scale comprising six infection classes (IC), where IC1 to IC3 correspond to various level of resistance of the plant and IC4 to IC6 to full susceptibility [89], with two biological replicates. Growth assays were performed by deposition of a 5-mm plug at the centre of 90-mm Petri dishes (containing 25 ml of V8 juice agar medium). Radial growth was measured at 10 days after incubation in a growth chamber, with three biological replicates.
Assessment of sexual reproduction of silenced-LmHP1 and silenced-LmDIM5 transformants compared to non silenced transformants was performed by crossing with the v24.1.2 strain, a near isogenic isolate of v23.1.3, but of opposite mating type [90]. Crosses of v23.1.3 with v24.1.2 were performed as a control.
For PCR, genomic DNA was extracted from conidia with the DNAeasy 96 plant Kit (Qiagen S.A., Courtaboeuf, France) and PCR amplifications were done as previously described [20]. Sequencing was performed using a Beckman Coulter CEQ 8000 automated sequencer (Beckman Coulter, CA, U.S.A) according to the manufacturer's instruction. Procedures for gel electrophoresis have been reported [91] and were adapted from procedures described by Sambrook and Russell [92]. Total RNA was extracted from mycelium grown for one week in Fries liquid medium, and from infected leaf tissues as previously described [20].
We first attempted to adapt the ChIP protocol developed for N. crassa and Fusarium species [28], [43], but found that sonication or bead-beating and formaldehyde crosslinking resulted in very poor shearing or yields of precipitated DNA. We thus subjected L. maculans to “native ChIP” (no crosslinking) and isolated mono- and dinucleosomes after digesting ∼300 mg of mycelium per sample with microccocal nuclease (MNase) for 25 min at 37°C (L.R. Connolly and M. Freitag, unpublished data). Input (40 µl of the whole cell lysate) was stored (−20°C) for each sample and used to normalise data from qPCR and qualitative PCR. For immunoprecipitation, each sample was split into two replicates of 250 µl lysate each and 3.5 µl H3K9me3 antibody (Active Motif 39161) was added for a total of two replicates per sample. Precipitations with H3K4me2 antibody (Millipore 07-030) were done in parallel as controls for ChIP efficiency. Because we used a native ChIP protocol, yields were consistently lower than from ChIP experiments with crosslinked chromatin.
To recover DNA sequences flanking AvrLm1, AvrLm6, LmCys2 or AvrLm4-7 in transformants v29.3.1-AvrLm1, v29.3.1-AvrLm6, v29.3.1-LmCys2 and NzT4-AvrLm4-7 respectively, TAIL-PCR [85], [93] was used. pPZPnat1 sequence-specific primers were designed to match primer characteristics as described [85] (Table S8). For pBBH and pBHt2 vectors, primers used were as designed previously [85], [88] (Table S8). Arbitrary Degenerated (AD) primers used in association with the vector-specific TAIL-PCR primers were identical to primer AD2 [93]. For v29.3.1-AvrLm1 transformants, TAIL-PCR was performed as described [93], while for v29.3.1-AvrLm6, v29.3.1-LmCys2 and NzT4-AvrLm4-7 transformants, TAIL-PCR was performed as described [85]. Tertiary TAIL-PCR products were purified using the Nucleospin Extract II purification Kit (Macherey-Nagel, Hoerd, France) and were used as template for DNA sequencing using the specific tertiary border primer (LB3) as a sequencing primer (Table S8).
Quantitative RT-PCR was performed using a model 7900 real-time PCR machine (Applied Biosystems) and Absolute SYBR Green ROX dUTP Mix (ABgene, Courtaboeuf, France) as previously described [20]. For each condition tested, two different RNA extractions from two different biological samples and two reverse transcriptions for each biological repeat were performed. Primers used for qRT-PCR are described in Table S8. Ct values were analysed as described [94] for analyses of expression profiles. β -Lmtubulin was used as a constitutively expressed reference gene. Expression of Lmactin relative to β-Lmtubulin was used as control.
For qPCR on ChIP DNA, primers were designed to amplify products between 50 to 150 bp (Table S8). For each reaction, we used 1 µl of the sample and performed reaction in duplicates. The “percent of input” method was used to calculate the immunoprecipitated fraction for each primer pair according to the following formula: %of input = 100×(1+E)∧(Ctinput−Ctbound), where E is the primer efficiency of each primer pair designed to amplify the amplicon, Ct input is the Ct of the fraction recovered after digestion by MNase and Ct bound is the Ct of the immunoprecipitated sample.
To compare expression of effector genes located in AT- and GC-isochores, oligoarray data previously obtained [7] and deposited in the Gene Expression Omnibus (GEO) under accession code GSE27152 were used. Additional L. maculans whole-genome expression arrays were manufactured by NimbleGen Systems Limited (Madison, WI) in order to compare whole genome expression in mycelium of wild type isolate v23.1.3 and transformants silenced for LmHP1 or LmDIM5. These arrays contain 5 independent, non-identical, 60-mer probes per gene model, each duplicated on the array. Gene models included were 12,457 EuGene-predicted gene models and 467 additional genes, 2,008 random 60-mer control probes and labelling controls. The oligoarray data are available from the NCBI GEO under the accession number GSE50616. Total RNA was extracted using TRIzol reagent (Invitrogen) according to the manufacturer's protocol from mycelium grown for one week in Fries liquid medium. Total RNA was treated with RNase-Free DNase I (New England Biolabs). Total RNA preparations (three biological replicates for each sample) were amplified by PartnerChip (Evry, France) using the SMART PCR cDNA Synthesis Kit (Invitrogen) according to the manufacturer's instructions. Single dye labeling of samples, hybridisation procedures, data acquisition, background correction and normalisation were performed at the PartnerChip facilities following the standard protocol defined by NimbleGen [95], [96]. Average expression levels were calculated for each gene from the independent probes on the array and were further analysed.
Gene-normalised data were subjected to Analysis of NimbleGen Array Interface Suite (ANAIS; http://anais.versailles.inra.fr; [97]). ANAIS performs an ANOVA test on log10 transformed data to identify statistically differentially expressed genes. This test uses the observed variance of gene measurements across the three replicated experiments. To account for multiple tests, the ANOVA p values are further subjected to Bonferroni correction. Transcripts with p values<0.05 and >1.5-fold change in transcript levels were considered as significantly differentially expressed in the silenced-LmHP1 or in the silenced-LmDIM5 compared to the wild type strain v23.1.3. To estimate the signal-to-noise threshold (signal background), ANAIS calculates the median of the intensity of all of the random probes present on the oligoarray, and provides adjustable cut-off levels relative to that value. Gene models with an expression higher than three-times the median of random probe intensities in at least two of three biological replicates were considered as transcribed.
Vectors pPZPnat1-LmHP1 and pPZPnat1-LmDIM5 for RNAi-mediated silencing of LmHP1 and LmDIM5 genes were constructed as described [98] with some modifications [20]. pJK11 vector contains a Glomerella cingulata gpdA promoter fragment and an A. nidulans trpC terminator fragment, separated by a multiple locus site where inverted repeats of coding sequence of LmHP1 and LmDIM5 have been cloned using the primers described in Table S8. The LmHP1 sense fragment of the inverted repeat was amplified from cDNA of v23.1.3 mycelium using primers SilentLmHP1-HindIII+ and SilentLmHP1-BamHI− and digested with HindIII and BamHI. The antisense region was amplified with primers SilentLmHP1-BamHI+ and SilentLmHP1-EcoRI- and digested with BamHI and EcoRI. Sense and antisense fragments were ligated into an EcoRI-HindIII digested pJK11. The expression cassette was excised by digestion with SpeI and XhoI and inserted into the binary vector pPZPnat1 digested with SpeI and XhoI, creating the vector pPZPnat1-LmHP1. The pPZPnat1 vector, that contains the nourseothricin resistant gene NAT1 (nourseothricin acetyltransferase gene), is used for gene silencing in L. maculans via agro-transformation. The same strategy was used to obtain the pPZPnat1-LmDIM5 vector; the sense fragment was obtained using primers SilentLmDIM5-HindIII+ and SilentLmDIM5-BamHI−; the antisense fragment using primers SilentLmDIM5-BamHI+ and SilentLmDIM5-XmaI-.
Vector pBHt2-LmHP1-GFP was created by fusing the GFP coding sequence (p-EGFP-1, Clontech) downstream the coding sequence of LmHP1 under the control of its own promoter in the binary vector pBHt2 containing the hygromycin B resistance gene (hph). The LmHP1 entire sequence containing 5′ and 3′ UTR was amplified from cDNA of v23.1.3 mycelium using primers LmHP1-EcoRI and LmHP1-XbaI digested with EcoRI and XbaI and ligated into an EcoRI-XbaI digested pBHt2. The stop codon was excised by amplification of the pBHt2-LmHP1 vector with C-LmHP1-SpeI and C-LmHP1-ClaI. The GFP coding sequence was amplified eGFP-ClaI and C-eGFP-SpeI from the p-EGFP-1 plasmid, digested with ClaI and SpeI and ligated into a ClaI-SpeI pBHt2-LmHP1 digested vector, thus creating pBHt2-LmHP1-GFP vector. Primers used are detailed in Table S8.
The constructs were introduced into A. tumefaciens strain C58 by electroporation (1.5 kV, 200 ohms and 25 µF). A. tumefaciens-mediated transformation (ATMT) of L. maculans was performed as previously described [99] with minor modifications [19]. Transformants were plated on minimal media complemented with nourseothricin (50 mg/l) for pPZPnat1-LmHP1 and pPZPnat1-LmDIM5 or hygromycin (50 mg/l) for pBHt2-LmHP1-GFP and cefotaxime (250 mg/l).
Axenic mycelia from a Petri dish with minimal medium were stained by incubating for 15 min in 10 µg/µl DAPI solution (Sigma-Aldrich) at room temperature to observe nuclei. Analyses were performed on a Leica TCS SPE laser scanning confocal microscope. Excitation of GFP was at 488 nm and emission was captured with a 505–530 nm broad pass filter. DAPI was excited at 543 nm and emission was captured with a 580–615 nm broad pass filter. The detector gain was 800 with an amplifier offset from −0.2. All images represent the average of 4 scans.
HP1 and DIM-5 orthologs were identified with the NCBI BLAST program http://blast.ncbi.nlm.nih.gov/Blast.cgi [100]; functional domains have been identified using InterProScan http://www.ebi.ac.uk/Tools/pfa/iprscan. Annotation of untranslated regions (UTR), transcriptional start and stop sites and intron positions were performed following PCR amplification and sequencing of 3′- and 5′-ends of cDNA using Creator SMART cDNA Library Construction Kit (Clontech, Palo Alto, CA) according to manufacturer's recommendation and using HP1-5UTRL1, HP1-5UTRL2, HP1-3UTRU1, HP1-3UTRU2 and DIM5-5UTRL1, DIM5-5UTRL2, DIM5-3UTRU2; DIM5-3UTRU1 as specific primers (Table S8). Hierarchical clustering analyses were carried out using GENESIS [101], multiple alignements with COBALT [102].
Disease scorings and radial growth of each isolate were analysed using ANOVA. For qRT-PCR analyses, Cttarget gene-CtβLmtubuline was calculated for each isolate and analysed using ANOVA. Isolates were compared using a Fisher test (α = 0.05). Transformants were compared to v23.1.3 by a Dunnet multiple comparison test (α = 0.05). All statistical analyses were performed using XLStat 7.5 software.
|
10.1371/journal.pgen.1002229 | Glutamine Synthetase Is a Genetic Determinant of Cell Type–Specific Glutamine Independence in Breast Epithelia | Although significant variations in the metabolic profiles exist among different cells, little is understood in terms of genetic regulations of such cell type–specific metabolic phenotypes and nutrient requirements. While many cancer cells depend on exogenous glutamine for survival to justify the therapeutic targeting of glutamine metabolism, the mechanisms of glutamine dependence and likely response and resistance of such glutamine-targeting strategies among cancers are largely unknown. In this study, we have found a systematic variation in the glutamine dependence among breast tumor subtypes associated with mammary differentiation: basal- but not luminal-type breast cells are more glutamine-dependent and may be susceptible to glutamine-targeting therapeutics. Glutamine independence of luminal-type cells is associated mechanistically with lineage-specific expression of glutamine synthetase (GS). Luminal cells can also rescue basal cells in co-culture without glutamine, indicating a potential for glutamine symbiosis within breast ducts. The luminal-specific expression of GS is directly induced by GATA3 and represses glutaminase expression. Such distinct glutamine dependency and metabolic symbiosis is coupled with the acquisition of the GS and glutamine independence during the mammary differentiation program. Understanding the genetic circuitry governing distinct metabolic patterns is relevant to many symbiotic relationships among different cells and organisms. In addition, the ability of GS to predict patterns of glutamine metabolism and dependency among tumors is also crucial in the rational design and application of glutamine and other metabolic pathway targeted therapies.
| Different types of cells have distinct ways of utilizing nutrients and generating energy, thus resulting in distinct nutrient needs. Such cell type–specific metabolic differences are associated with many biological processes and force the symbiosis between different cells and organisms. For example, glutamine symbiosis is a well-recognized phenomenon due to different glutamine synthesis ability. In human cancers, glutamine is also recognized as an important and essential nutrient, termed glutamine addiction. But very little is known about how glutamine addiction varies among different tumors of diverse cellular origins, which hinders personalized therapeutic strategies. Here, we found that basal-type breast cancer cells were sensitive to glutamine deprivation while luminal-type breast cancer cells were not. Luminal cell–specific glutamine independence results from expression of glutamine synthetase conferring the ability to synthesize glutamine. Glutamine synthetase also represses glutaminase and contributes to the maintenance of the polarized expression of glutamine synthetase and glutaminase among breast cancer cells. Collectively, these data illustrate cross-talk between mammary differentiation programs and unique nutrient requirements, which may offer novel therapeutics for basal-type breast cancers.
| There are a large number of differentiated cell types in the human body. Even among the cells collectively known as fibroblasts [1], endothelial [2] and smooth muscle cells [3], gene expression analysis has identified an unexpected level of positional memory and topographic differentiation. Such functional specialization contributes to the phenotypic variations of many human diseases, including cancer. For example, gene expression analysis of breast cancers has identified five intrinsic subtypes (luminal A, luminal B, basal, HER2+, and normal-like) with unique clinical and histological properties [4], [5]. The classification nomenclature is based on the putative progenitor cell(s) for breast carcinogenesis with properties consistent with derivation from the basal and luminal epithelia arrested at specific differentiation stages or from different mature epithelial cells [4]–[7]. Importantly, these subtype-specific gene expression and phenotypic variations are also observed in many breast cancer cell lines with similar molecular phenotypes [8]–[11]. A number of studies have isolated the different populations of primary epithelial cells to investigate their relevant cellular origins and metabolic features for different breast cancer types [7], [12], [13]. Although the cellular origin of luminal and basal-like breast tumor has not been resolved [14], [15], cell lineage still appears to confer an important source of patterned heterogeneity to the disease.
Although gene expression analysis has yielded important insights into the cellular differentiation and various properties associated with tumors from different cell types, very little is known about the corresponding metabolic phenotypes and nutrient requirements. The processes of oncogenic transformation place energy demands on cancer cells to support proliferation, expansion, and invasion. Dysregulated tumor metabolism is a critical part of oncogenesis and may be targeted for therapeutic benefits [16], [17]. One prominent example of dysregulated tumor metabolism is “aerobic glycolysis” as recognized by Otto Warburg [18]. Most normal mammalian cells shift to glycolysis for energy generation when oxygen is inadequate for effective oxidative phosphorylation under hypoxia. But tumor cells tend to favor glycolysis even with the availability of oxygen, hence termed “aerobic glycolysis” [19]. Such preferential use of glycolysis leads to vigorous glucose uptake and explains the ability of the tracer glucose analog Fluorine-18 (F-18) FDG to image human cancers in FDG-PET. Such understanding of altered metabolism and nutrient requirement in cancer cells may allow us to exploit these differences for diagnostic and therapeutic benefits.
Another aspect of dysregulated tumor metabolism is manifested as altered requirements for amino acids. For example, patients with acute lymphocytic leukemia (ALL) benefit from asparaginase treatment as the leukemic cells require large amounts of exogenous asparagine due to a deficiency in this metabolic pathway [20]. Recently, evidence is also accumulating for the essential role of glutamine for cancer cells as a building block for protein synthesis, to supply cellular ATP, as a metabolic intermediate for nucleotide synthesis, and for its anti-oxidative capacity [21], [22]. Such glutamine dependence or addiction is reflected in the growth restriction and cell death in glutamine limiting conditions. The glutamine addiction is also critical for c-myc-mediated oncogenesis [23]–[25], linked with glucose requirement [26], and proposed as an attractive target for therapeutic intervention [22], [27].
The catabolism of glutamine is initiated by glutaminolysis mediated by two different subtypes of mitochondrial glutaminase (kidney or liver-type encoded by GLS or GLS2 respectively) to become glutamate [28]. The intracellular pool of glutamate is a versatile metabolic intermediate that connects with a wide variety of distinct biological processes including synthesis of the anti-oxidant glutathione, amino acid catabolism through transamination, and conversion to α-ketoglutarate as a substrate for the TCA cycle. This process of glutaminolysis by glutaminases has been shown to mediate signaling events [29], to be coupled with c-myc oncogenesis [25], and proposed as a critical step in targeting glutamine metabolism [24], [27]. In some cell types, glutamine can be generated from intracellular glutamate through glutamine synthetase (GS, encoded by GLUL, glutamate-ammonia ligase) catalyzing the reverse reaction of the glutaminases. This process is important for removal of ammonia or glutamate depending on the cellular context [30]. While glutaminase is known as an important regulator of glutamine requirement, few studies have focused on glutamine synthetase as a potential determinant of glutamine requirement. Although normal glutamine metabolism is well understood, the genetic parameters and mechanisms of variation in this key nutrient pathway among tumors are largely unknown.
Deprivation of glutamine and other amino acids triggers a canonical amino acid response (AAR) in most mammalian cells that is measurable by gene expression changes [31]. The free and uncharged t-RNA associated with glutamine deprivation activates a serine/threonine-protein kinase GCN2 which phosphorylates eIF2α and inhibits cap-dependent translation [32]. While reducing the global translation rate, eIF2α phosphorylation also preferentially increases the translation of ATF4 and other mRNAs [31]. The increased level of ATF4 protein triggers the AAR gene expression program characterized by the induction of XBP1 and DDIT3 as an adaptive response to amino acid deprivation. The importance of the AAR is demonstrated by the fact that deficiency of ATF4 compromises the AAR and renders cells susceptible to amino acid deprivation and oxidative stresses [33].
Through the analysis of how different breast cancer cells respond to glutamine deprivation, we have found a dramatic difference in the glutamine requirement among different breast cancer cells which tracks with the luminal versus basal type. These metabolic differences can be explained by cell-type specific expression of glutamine-metabolizing genes and enzymes likely acting in concert with cell type specific oncogenic programs. Therefore, we have provided a series of fundamental building blocks to understand how differentiation is coupled with distinct glutamine utilization in normal and neoplastic breast epithelia. Such an understanding will be relevant to both the mechanistic understanding of metabolic phenotypes and present insights into how best to select subsets of breast cancer patients most likely to benefit from glutamine-targeting therapies.
Many cancer cells require glutamine for survival and proliferation and thus exhibit a phenotype of “glutamine dependence” or “addiction” [22]. To determine whether such phenotypes could be also found in breast cancer cells, we tested how glutamine deprivation affected seven different breast cancer cell lines. Consistent with the idea of glutamine dependence, three cell lines (BT20, MDAMB231, and MDAMB157) had significantly reduced growth (MTT assay, Figure 1A) and prominent cell death (trypan blue exclusion assay, Figure 1B) upon glutamine deprivation for 48 h. Unexpectedly, glutamine deprivation had only modest effects on the growth and viability (Figure 1A, 1B) of the other four cell lines (T47D, BT474, MCF7, and MDAMB361) indicating relative glutamine independence. When we examined the properties associated with the distinct need for glutamine, we found the cell lines that exhibit glutamine dependence are all of the basal-type whereas the four lines that are more glutamine independent are luminal-type cells (Figure 1A, 1B) [34].
As glucose and glutamine are two important energy sources for cancer cells we compared how deprivation of glutamine and glucose affected the growth of these breast cell lines. In the three basal-type cell lines, glutamine depletion had a stronger effect on cell growth than glucose depletion (Figure 1C). In contrast, glucose depletion had a more dramatic influence on cell growth than glutamine depletion in the four luminal cell lines (Figure 1C). These results suggested that there is a consistent variation in glutamine phenotype associated with cell lineage in breast cancers.
One important function of glutamine is to serve as an energy source in generating cellular ATP. To determine the relative importance of glutamine to ATP generation in the breast cancer cell lines, we measured ATP in cells grown in media containing either normal levels of glutamine (4mM) or no glutamine for 12 hours. Glutamine deprivation led to a much more significant reduction in ATP generation in the basal-type cells than the luminal-type breast cancer cell lines (Figure 1D). These results further support the concept that glutamine is a more important energy source in basal than luminal breast cell lines.
To further analyze glutamine metabolism among different cell types, we measured the consumption of glutamine in the medium and intracellular glutamine levels. When compared with luminal-type cells, the basal cell lines had significantly higher levels of glutamine consumption (Figure 1E) and lower intracellular glutamine concentrations (Figure 1F). Collectively, these data strongly support the concept of distinct glutamine metabolism and varying dependence for external glutamine between basal and luminal type breast cancer cells.
We hypothesized that such distinct glutamine dependence among basal and luminal breast cancer cell lines may be caused by variable expression of key enzymes involved in glutamine metabolism. Glutamine synthetase (GS encoded by GLUL – glutamate-ammonia ligase) and glutaminase (GLS – kidney form or GLS2 – liver form) mediate the opposite reaction in the reversible conversion between glutamate and glutamine. GS mediates the capture of an ammonia group by glutamate to synthesize glutamine, while glutaminase catalyzes the breakdown of glutamine to glutamate. We first examined the RNA expression of these genes in a microarray expression set [34] and found that the expression of GLUL (GS) was significantly higher in the luminal cell lines. In contrast, the expression of GLS (glutaminase, kidney) was higher in the basal lines. While lacking GLS expression, the luminal breast cell lines have a higher level of GLS2 (Figure 2A). We confirmed this cell-type specific differential mRNA expression of GLUL, GLS and GLS2 with real-time PCR (Figure 2B, 2C, and 2D). Differential expression was also found at the protein level as shown by the western blots for GLUL (GS) and GLS2 (in luminal cells) and GLS (in basal cells) (Figure 2E).
We next examined whether the expression patterns of GLUL, GLS and GLS2 found in luminal and basal cell lines were also reflected in the respective subtypes of primary human breast cancers. In a breast tumor expression dataset [35], we found significantly different expression levels of GLUL, GLS and GLS2 in the corresponding luminal (luminal A and B) and basal-types of breast tumors (Figure 3A). We also examined the expression levels of these three genes in the same dataset within the 5 intrinsic subtypes [35] and found significantly different expression between the luminal A and basal tumors (Figure S1). This concordance indicates the differential expression of GLUL, GLS and GLS2 in the luminal and basal-type cancer cell lines reflects similar systematic differences in primary breast tumors.
To determine whether differential expression of genes driving glutamine metabolism is an intrinsic cell-lineage phenomenon in the breast, we examined their expression levels in normal non-transformed basal and luminal epithelial cells. Primary luminal and basal breast epithelial cells were separated based on surface expression of EPCAM (TACSTD1) from reduction mammoplasty specimens and gene expression levels were determined by microarray analysis [12]. Analysis of isogenic basal and luminal epithelial cells showed that the mRNA levels of GLUL, GLS and GLS2 exhibited similar cell-type specific expression in normal breast cells (Figure 3B). These findings were also confirmed by real-time PCR (Figure 3C–3E). In addition, expression of the GLUL (GS) and GLS proteins showed corresponding luminal and basal-specific expression patterns (Figure 3F). The level of GLS2 protein was below detection levels in both primary epithelial cells (Figure 3F). These results suggest that differential expression of glutamine metabolizing enzymes in cancers may be ascribed to systematic differences in cell lineage observed in normal basal and luminal epithelial cells.
Given the well-recognized glutamine dependency of many cancer cells, we investigated the roles of GLUL and GLS2 in the relative glutamine independence of the luminal-type cells. We first treated cells with a GS inhibitor (L-MS [36]) for 48 h and measured cell viability under glutamine deprivation. We found that L-MS reduced the survival of the luminal cell lines but had no statistically significant effect over glutamine starvation on all three tested basal cell lines (Figure 4A) indicating that GS is involved in the glutamine independence of the luminal cells.
Next, we performed genetic experiments to examine the role of specific genes in the glutamine phenotype. Silencing of GLUL (encoding GS) in the luminal MCF7 line significantly reduced the RNA and protein expression of GS (Figure S2A and S2B) and led to a significant reduction in glutamine independence (Figure 4B). In contrast, similar silencing of GLS2 did not affect survival under glutamine deprivation (Figure S3). In addition, the ectopic overexpression of GLUL (verified in Figure S4A, S4B) in the basal MDAMB231 cells conferred partial glutamine independence by significantly increasing the cell survival under glutamine deprivation (Figure 4C). Taken together, these data suggest that GS expression significantly contributes to the differential glutamine phenotypes observed in breast cancer cell lines.
We next investigated potential regulatory mechanisms for the subtype-specific expression of glutamine metabolizing enzymes. During the differentiation of luminal epithelial cells, GATA3 is an important master regulatory transcription factor [37], [38]. The expression of GATA3 in luminal and basal cells is systematically different as previously noted [11], [39]. Using real-time PCR, we also demonstrated the cell-type specific expression of GATA3 mRNA in MCF7 (luminal) and MDAMB231 (basal) cells (Figure S5A). Re-analysis of microarray data of the overexpression of GATA3 in mouse breast epithelial cells [38] shows induction of GLUL and repression of GLS and GLS2 (Figure 4D). These data suggested a role for the lineage factor GATA3 in regulating the luminal and basal-specific expression of GLUL and GLS.
We directly tested the role of GATA3 in regulating the glutamine phenotype in breast cancer cell lines. The mRNA and protein levels of GATA3 could be effectively reduced by gene silencing through siRNAs (Figure S5B, S5C). Silencing of GATA3 in MCF7 cells led to significant reduction in GLUL at both the RNA and protein levels (Figure 4E, Figure S5C). Conversely, overexpression of GATA3 in the basal MDAMB231 line (Figure S5D) led to a significant upregulation of GLUL (Figure 4F, Figure S5E). Furthermore, the silencing of GATA3 in MCF7 cells reduced the survival under glutamine deprivation (Figure 4G, Figure S3), and overexpression of GATA3 in MDAMB231 cells increased the resistance to glutamine deprivation (Figure 4H), consistent with a direct role for GATA3 mediated GLUL expression in the glutamine independence of luminal breast cells. In addition, the glutamine independence caused by GLUL (Figure S6A) or GATA3 (Figure S6B) overexpression in MDAMB231 cells was also abolished with treatment of L-MS (GS inhibitor), indicating of the importance of the catalytic activities of GS.
Given the ability of GATA3 to increase the expression of GLUL, we examined the promoter region of GLUL and found two potential GATA3 binding sites at −524 to −518 bp (region A) and −200 to −194 bp (region B) upstream of the transcriptional start site (Figure 4I). We used chromatin immunoprecipitation (ChIP) to test whether GLUL may be a direct downstream target of GATA3 transactivation. Consistent with previous data [40], the promoters of ESR1 (estrogen receptor alpha), but not albumin, were enriched in the GATA3 ChIP samples. Of the two putative GATA3 binding sites in the GLUL promoter, the distal region A but not the more promoter proximal region B, was significantly enriched in the GATA3 ChIP samples (Figure 4J) indicating that GATA3 protein can directly bind to a regulatory region of GLUL suggesting that this gene is a target of the luminal transcription factor and further serving to explain the lineage specific requirement for glutamine.
The deprivation of amino acids in mammalian cells leads to the stabilization of the ATF4 (activating transcription factor 4) protein and resulting induction of a canonical gene expression program known as the amino acid response (AAR) [41]. The response includes the induction of XBP1 (X-box binding protein 1) and DDIT3 (DNA-damage-inducible transcript 3) which are essential for survival under amino acid deprivation [41]. Given the distinct growth and survival response of luminal and basal breast cells to glutamine deprivation, we used microarrays to compare their transcriptional responses on a global scale. Triplicate plates of MCF7 and MDAMB231 cells were cultured under both control (4 mM glutamine/Q4) and glutamine-depleted (no glutamine/Q0) conditions for 24 hours. RNA from each plate was interrogated with Affymetrix GeneChip U133-A2 arrays (results deposited in Gene Expression Omnibus (GSE26370)). Gene expression profiles of the 12 arrays were normalized by RMA and the transcriptional changes of glutamine deprivation in both cell types were derived by zero-transformation against the average expression levels of the control samples as performed previously [42]–[44]. Probes sets showing at least two fold changes in at least two samples (n = 405) were selected and arranged by hierarchical clustering according to similarities in expression patterns (Figure 5A). This analysis showed that glutamine deprivation induced a strong gene expression response in MDAMB231 (MB231) but less so in MCF7 cells (Figure 5A). We found that the canonical AAR genes were induced by glutamine deprivation only in MDAMB231 cells (Figure 5A). A previous study showed that glutamine deprivation inhibits cell growth by inducing the tumor suppressor gene TXNIP [29]. This gene was also induced only in the MDAMB231 line. We applied a published gene expression study of histidine deprivation [45] as training data and estimated the degree of AAR using a binary regression model. MDAMB231 but not the MCF7 line exhibited a significantly higher probability of AAR after glutamine deprivation using this approach (Figure 5B and 5C). The stronger amino acid response in the MDAMB231 cells was also confirmed by real-time PCR for XBP1 (Figure 5D) and DDIT3 (Figure 5E). These data provide further evidence that glutamine deprivation induces a much dramatic response in the basal cells and a weak response correlating with glutamine independence of the luminal cells.
We examined how glutamine deprivation affected different glutamine-metabolizing enzymes and found that GS protein (Figure 6A), but not mRNA (Figure S7A), were significantly induced in MCF7 cells in a dosage-dependent manner. This translational regulation may be an adaptive response to compensate for reduced environmental levels of glutamine. To examine the role of GATA3 in the induction of GS during glutamine deprivation, we compared the GS protein levels under different glutamine levels in MCF-7 transfected with control or GATA3-targeting siRNA. We found that while the silencing of GATA3 reduced the GS levels, there was still significant protein induction during glutamine deprivation (Figure S7B). We also measured glutamine concentrations in glutamine deficient media used to culture MCF7 and MDAMB231 cells and found a significant increase in glutamine levels in medium cultured with MCF7 but not MDAMB231 cells (Figure 6B). Similarly, intracellular glutamine concentrations were increased only in MCF7 but not MDAMB231 cells under glutamine deprivation (Figure 6C). Therefore, the glutamine independence phenotype of luminal cells may be due to the capacity of these cells to synthesize glutamine from intracellular glutamate and other sources in the absence of external glutamine.
In normal breast ducts, luminal and basal cells are in close physical proximity. Because of the ability of luminal cells to synthesize glutamine and the requirement of basal cells for glutamine, we next tested the potential for glutamine symbiosis between these two cell types with transwell co-culturing experiments (Figure 6D–6F). We found that the viability of MDAMB231 cells under glutamine deficient media was significantly increased when MCF7 cells were used as a feeder layer (Figure 6E), consistent with observed higher extracellular glutamine levels (Figure 6F). Furthermore, conditioned medium from MCF7 cells was also able to support significantly the growth and viability of the MDAMB231 cells (Figure 6G–6I).
We showed above that increased levels of GLUL either by transfection with GLUL or GATA3 makes the MDAMB231 line more resistant to glutamine deprivation (Figure 4C and 4H). We next asked whether this was due to increased synthesis of the nutrient. Intracellular glutamine levels increased dramatically in MDAMB231 cells expressing either GLUL or GATA3 (5×104 cells in the upper well) (Figure 6J, 6K). The effects of GLUL and GATA3 overexpression in MDAMB231 cells on intracellular glutamine levels were blocked with L-MS treatment (Figure S8A). We also showed that the intracellular glutamine levels were reduced in MCF7 with siRNAs targeted to GLUL or GATA3 in medium with normal glutamine level (Q4) or no glutamine (Q0) (Figure S8B). Further, in the co-culture system (Figure 6L), MDAMB231 cells demonstrated increased viability when co-cultured with either GLUL or GATA3 expressing MDAMB231 cells (Figure 6M) and this correlated with both increased glutamine levels in the medium (Figure 6N) and intracellularly (Figure 6O). These data provide a consistent mechanistic picture of a gene expression program related to nutrient requirements and potential glutamine symbiosis.
The expression of GLUL and GLS are inversely correlated in the luminal and basal types of primary breast cancers, cancer cell lines, and primary epithelial cells. This pattern of expression made us investigate whether cross-regulation exists between these two genes that encode enzymes mediating directly opposite chemical reactions. We first used siRNA to silence GLUL in MCF7 cells and observed an increase in GLS mRNA expression (Figure 7A). Further, ectopic over-expression of GLUL in MDAMB231 cells reduced GLS mRNA (Figure 7B). In contrast, similar silencing of GLS did not show any effect on GLUL levels (Figure 7C, 7D). The ability of GLUL overexpression in MDAMB231 to repress GLS was also seen at the protein level with a dose dependent decrease in GLS protein observed with increasing amounts of GS protein from varying levels of transfected GLUL (Figure 7E). These results indicated the ability of GLUL to repress the expression of GLS while GLS had no detectable effect on the level of GLUL.
Since GATA3 regulated the expression of GLUL, we tested the role of GATA3 in regulating GLS by silencing and overexpressing GATA3 in MCF7 and MDAMB231 cells, respectively. Silencing of GATA3 in MCF7 cells increased GLS expression (Figure 7F) and GATA3 overexpression in MDAMB231 significantly reduced the level of GLS (Figure 7G). These changes in GLS expression regulated by GATA3 were also detectable at the protein level compared with GLUL (Figure 7H). These results are also consistent with GATA3 overexpression in the mouse epithelial cells (Figure 4D) [38].
Based on the data presented, we propose that basal and luminal breast epithelial cells exhibit different patterns of glutamine metabolism (Figure 8). In the luminal cells, GATA3 triggers expression of GLUL and contributes to glutamine independence. Furthermore, GLUL has the ability to repress GLS which would also help to maintain the cell-type specific expression pattern and phenotype. Basal-specific expression of GLS may be maintained by the absence of GATA3 and higher activities of c-myc in the basal type cells [4], [46]. These findings suggest that glutamine deprivation may be a viable treatment strategy for basal-type breast cancers. In addition, the expression of GLUL in luminal type cancers correlates with the ability to synthesize glutamine from ammonia and glutamate, and therefore describes at the molecular level a type of cancer that is predicted to be more resistant to glutamine deprivation treatment.
While glutamine has been shown to be critical in many cancer types, its importance for breast cancers is not well defined. In this study, we found a cell lineage-specific variation in the response of basal and luminal breast cancer cells to glutamine deprivation. The basal-type breast cancer cells are dependent on glutamine and exhibit a phenotype of glutamine addiction. Such a phenotype of basal cells was previously reported to be sensitive to inhibitors of glutaminase [27], trans-amination by aspartate aminotransferase [47] and selective estrogen receptor modulators [48]. In contrast, the luminal-type breast cancer cells are much more glutamine-independent. We present a series of data which strongly suggest that this phenotypic difference is related to the luminal-specific expression of glutamine synthetase (GS encoded by the GLUL gene) which is in turn regulated by one of the key luminal transcription factors, GATA3. Further, GS itself represses the expression of glutaminase (GLS) to reinforce the metabolic pathway in the direction of glutamine synthesis in luminal breast cells and the potential for glutamine symbiosis with basal breast cells.
While variations in tumor metabolism can be caused by oncogenic events, our results highlight the importance of the differentiation status and cellular origins as a source of distinct metabolic patterns. Since differentiation state constitutes an important part of tumor heterogeneity, similar investigation into distinct metabolic needs may yield important information on how best to target tumor metabolism. As the induction of differentiation is an important component of some cancer therapeutics [49], such treatment-associated differentiation may also lead to changes in metabolic needs and may be exploited to enhance treatment efficacy. Similarly, the distinct nutrient requirements of tumor stem cells [50] may be used to target these unique populations which are more resistant to conventional cancer therapeutics.
The distinct glutamine requirement and differential expression of glutamine-metabolizing enzymes among luminal and basal breast cancers are consistent with our understanding of the genetic circuitry governing breast cancer subtypes and regulation of these glutamine-metabolizing enzymes. For example, the higher GLS level and sensitivity to glutamine deprivation of basal-type breast cancer cells are consistent with a high level of c-myc activity in basal cells [51], [52] and the recently described role of c-myc in regulating GLS [24], [25]. The higher levels of GLS are also consistent with the susceptibility to growth inhibition by targeting this enzyme [27] and indicate the essential nature of this metabolic pathway in the basal cells. These results indicate that triple-negative basal-like breast tumors, with few current therapeutic options, are addicted to glutamine and may benefit from glutamine-targeting therapies [22], [27]. In contrast, the luminal specific expression of GLS2 may reflect the higher p53 (wild type) activity in luminal cells [4], [52] given the ability of p53 to regulate GLS2 [53], [54]. Our results suggest that GATA3 is directly involved in the transcriptional regulation of GLUL in luminal cells.
The spatial and cell type specific expression of GLUL (GS) seen in our studies on breast epithelial cells is also observed in several other cellular contexts. In the brain, GS is expressed mainly in glial cells [55] and is responsible for the synthesis of glutamine from the uptake of glutamate secreted by adjacent neurons. Similar spatial division of glutamine degradation and synthesis also occurs in distinct patterns of GS and GLS expression in the liver [56]. Glutamine degradation by GLS occurs in the periportal cells where there is a high glutamine level from the digested nutrients in the gastrointestinal tract. In contrast, the expression of GLUL (GS) is restricted to zones of hepatocytes surrounding the central lobular vein with lower glutamine levels [56]. In the renal nephron, GLUL (GS) expression is restricted to the straight portion of the proximal tubules and plays an important role in the removal of ammonia [57]. Such physical separation of glutaminase and glutamine synthetase associated with differentiation and nutrient availability coordinate the glutamine synthesis and effective detoxification of ammonia and glutamate. Similar distinct glutamine metabolism in luminal and basal breast epithelial cells also appears to impact tumors derived from these different lineages and opens an additional window into the metabolic phenotypes of this heterogeneous disease. Therapeutic interventions based upon metabolic targets will need to incorporate these systematic differences between tumor subtypes.
Under glutamine deprivation, the high mRNA levels of GLUL (GS) in luminal cells undergo further protein upregulation to provide glutamine and may also support the glutamine requirement in basal cells in physical proximity. Similar nutritional and metabolic interaction underlies many symbiotic relationships among different organisms and cell types, including the symbiotic nitrogen-fixing root nodules on legumes [58], the mutualistic symbiosis between bacteria and insects [59], and the glutamate-glutamine shuttle between neurons and astroglial cells in the brain [55], [60]. Interestingly, GS plays a critical role in all these diverse examples of metabolic symbiosis. In addition to the inter-cellular exchange of nutrients, the synthesis of glutamine from glutamate and ammonia by GS can also remove the potential toxicity from the accumulation of glutamate (neurotransmitter) and ammonia (nitrogen waste). Ammonia from glutaminolysis has been shown to act as a diffusible autocrine- and paracrine substance inducing autophagy [61]. Given the physical proximity between basal and luminal cells in breast ducts, such a reciprocal metabolic relationship may also be relevant under different environmental or growth conditions. When the tissue organization is disrupted in malignancy, glutamine dependence of the basal-type tumors may be exploited to treat this type of aggressive cancers.
Our findings strongly suggest that there will be significant variation in response to glutamine-targeting therapies. Among breast cancers, systematic variation in the glutamine consumptive vs. synthetic behaviors seen in the basal and luminal tumors will directly influence this response. Similar heterogeneity may be important in other tumor types as well. Our data also provide evidence that glutamine-targeting therapeutics may be of special clinical utility for the triple-negative basal-like breast tumors with few therapeutic options. Many current glutamine-targeting therapeutics focus on glutaminase [25], [27], but the cell-type specific expression of GLS and GLS2 in different tumors indicates the importance of choosing compounds with intended specificity for particular glutaminase activities in the treated tumors. Since GS is a key genetic determinant of glutamine independence in luminal cells, the evaluation of the GS levels in tumors may be helpful in predicting response. In addition to the cell-autonomous variations in the GS expression and response to glutamine deprivation, the efficacy of glutamine-targeting therapies may also be affected by the ability of adjacent non-transformed cells to provide glutamine. It is important to note that GS activities have been reported in fibroblasts [62] and macrophages [63]. The availability of glutamine from other non-tumor cells or blood may reduce the efficacy of glutamine-targeting therapies. Thus, GS inhibition may be combined with glutamine-targeting therapies to further enhance efficacy and reduce resistance, similar to the use of GS inhibitors to sensitize cancer cells to L-asparaginase [64].
With the explosion of genomic data, we have obtained significant knowledge on how genetic dysregulation contributes to tumor heterogeneity in human cancers. Since dysregulated metabolism is an essential part of oncogenesis, similarly detailed knowledge of metabolic profiles may be of equal or greater importance in understanding and treating the disease [65], [66].
All breast cancer cells were cultured in DMEM with 4.5 g/L glucose, supplemented with 10% fetal bovine serum and 1% penicillin/streptomycin in 5% CO2.
Primary luminal and basal cells were obtained from women undergoing breast reduction for non-malignant conditions and were separated by cell surface binding to the TACSD1 protein of the Ber-Ep4 antibody as described [12].
For the MTT assay, 2.5×103 cells in 100 µl of medium were seeded in a 96-well culture plate. After treatments, cell number was evaluated. In brief, 10 µl of MTT (Sigma M5655) (0.5 mg/ml) was added to each well, and then the plates were incubated at 37°C for 3 h. The formazan product was dissolved in DMSO, and the absorbance at 570 nm was measured using a microplate reader. To measure viability by direct counting, 2×104 cells were seeded in 12-well dishes and treated with medium containing different concentrations of glutamine for 48 h; the cells were collected and stained with 0.4% Trypan Blue. Cells excluding and taking up dye were counted on a hemocytometer under phase contrast microscopy. For glutamine synthetase inhibition, L-MS (L-Methionine-Sulfoximine, 5 mM, Sigma-Aldrich) was administered to cells for 48 h.
Cells (5×103/well of a 96 well dish) were treated with or without glutamine for 12 h and ATP content was measured in accordance with the protocol of the ATP-Lite luminescent ATP detection assay kit (Perkin-Elmer). Briefly, 100 µl of assay reagent was added to the wells and mixed for 10 min in the dark; intracellular ATP content was measured using a luminescence multi-label counter. The ATP levels were normalized based on cell counts measured by the MTT assay.
Cells (1×104/well) in a 24 well plate were cultured for 24 h in medium without phenol-red, medium was collected, and cells were lysed with RIPA buffer (Sigma-Aldrich). Concentrations of glutamine in the medium and in the cell lysate were determined with the glutamine/glutamate determination kit (GLN-1; Sigma-Aldrich). Each sample was divided into two parts; part 1 was measured with glutaminase for transferring the glutamine into glutamate, part 2 was measured directly. Samples were then dehydrogenized to α-ketoglutarate accompanied by reduction of NAD+ to NADH. The amount of NADH is proportional to the amount of glutamate and was measured using a spectrophotometer at 340 nm. A standard curve was determined for each experiment to calculate the concentration of glutamate in samples. Glutamine levels were calculated (part 1 minus part 2) and normalized to total protein levels. The glutamine level of normal culture medium was also measured, and the glutamine consumption was calculated as (glutamine in normal medium-glutamine in medium after culturing cells) and normalized to protein level.
Proteins were separated by 10–12% SDS–PAGE and transferred to Immobilon-P membranes (Millipore). Membranes were blocked with 5% skim milk, incubated with primary antibodies (GLUL, G2781, Sigma; GLS, ab60709, Abcam; GATA3, sc269, Santa Cruz; GLS2, ab91073, Abcam; tubulin, 2128, cell signaling), HRP-conjugated secondary antibody (Perkin-Elmer), and detected with the ECL Western blotting reagent (Amersham).
MCF7 and MDAMB231 cells were cultured in medium with or without glutamine for 24 h in triplicate. RNAs were collected with MirVana kit (Ambion) and hybridized to Affymetrix U133A2 arrays. Probe intensities were normalized by RMA and then the changes of expression by glutamine deprivation (0 mM glutamine/Q0) were derived by zero-transformation against the corresponding cells grown in glutamine containing medium (4 mM glutamine/Q4).
Cells were transfected with non-targeting control or synthetic siRNAs targeting GLUL, GLS, GLS2 and GATA3 (Applied Biosystems) with lipofectamine 2000 (Invitrogen). For overexpression experiments, empty vector or overexpression constructs for GLUL or GATA3 (Origene) were transfected into cells with lipofectamine 2000 for 48 hours before the levels of indicated transcripts and proteins were examined by real-time RT-PCR and western blot.
Total RNA was reverse-transcribed to cDNA with SuperScript II reverse transcription kit, then used for real-time PCR with Power SYBR Green PCR Mix (Applied Biosystems) and indicated primers (Table S1), and normalized to β-actin mRNA levels measured in parallel.
10% formaldehyde solution was added to cells to crosslink DNA-protein complexes. Isolated nuclear chromatin extracts were sonicated and incubated overnight at 4°C with either anti-GATA3 (SC269, Santa Cruz) or normal mouse IgG (SC3878, Santa Cruz). This was followed by incubation with 20 ml of Protein G agarose beads (Roche) for 4 hours at 4°C. After extensive washing, DNA fragments were harvested by de-crosslinking the immunoprecipitates. Real time-PCR utilizing SYBR Green master mix (Applied Biosystems) was performed to check the enrichment of indicated promoter regions in pull-down samples using primers listed in Table S1 and normalized with albumin.
For co-culture experiments, MDAMB231, MCF7, or transfected MDAMB231 cells were seeded in minicells (upper well/5×104 cells) with 0.4 µm pores (Millipore) and co-cultured with MDAMB231 (lower well/1×104 cells) for 12, 24 or 48 h in 24-well plates. For conditioned medium experiments, MDAMB231 or MCF7 cells (5×104) were seeded in a 24-well plate and incubated in medium with or without glutamine for 24 hours and then medium was transferred to new wells containing MDAMB231 cells (1×104). In 12 and 24 hours experiments, medium was collected; cells were washed by PBS and then lysed with 100 µl RIPA buffer. Glutamine concentration was measured with GLN-1 (Sigma). In 48 h experiments, cell numbers were counted by trypan blue exclusion assay.
All experiments were expressed as mean ± standard deviation (SD) with t-test. Statistical significance was calculated by t test, considering p<0.05 (*) and p<0.01(**) as statistically significant.
|
10.1371/journal.pntd.0006710 | Aberrant plasma MMP and TIMP dynamics in Schistosoma - Immune reconstitution inflammatory syndrome (IRIS) | Among the different faces of immune reconstitution inflammatory syndrome (IRIS) developing in HIV-patients, no clinical definition has been reported for Schistosomiasis-IRIS (Schisto-IRIS). Although Schisto-IRIS remains largely uninvestigated, matrix metalloproteinases (MMP) and tissue inhibitors of metalloproteinases (TIMP) have previously been associated with S. mansoni infection and tuberculosis-IRIS. Here, we aimed to investigate the relevance of these markers in Schisto-IRIS.
Patients were diagnosed with IRIS related to S. mansoni within a cohort of patients with Schistosomiasis-HIV co-infection, using a clinical working definition of Schisto-IRIS. We compared 9 patients who developed Schisto-IRIS to 9 Schisto+HIV+ controls who did not, and 9 Schisto-HIV+ controls. Plasma levels of MMP-1, MMP-7, MMP-10, TIMP-1, TIMP-2, sCD14, intestinal fatty-acid binding protein, C-reactive protein, and 8 anti-nuclear antibodies (ANA) were analyzed prior to and during 3 months of ART.
Although no differences were observed for MMP-1 and -7, MMP-10 levels decreased significantly in Schisto+HIV+ controls during 3 months of ART (p = 0.005) while persisting in Schisto-IRIS patients at significantly higher levels compared to Schisto-HIV+ controls (p≤0.030). In contrast TIMP-1 levels only decreased significantly in Schisto-IRIS patients (p = 0.012), while TIMP-2 levels were lower compared to Schisto+HIV+ controls at 2 weeks (p = 0.007), 1 month (p = 0.005) and 3 months (p = 0.031) of ART. Five out of 8 ANAs studied decreased significantly in Schisto-IRIS patients after 1 month of ART(p≤0.039), whereas only 1 ANA decreased for Schisto+HIV+ controls (p = 0.027).
In this study, we propose a working definition for the diagnosis of Schisto-IRIS in resource limited settings. We report persistent plasma levels of MMP-10, along with a more pronounced decrease in TIMP-1 and ANA-levels, and low levels of TIMP-2 during 3 months of ART. Corresponding to the clinical symptoms, these data suggest that Schisto-IRIS is marked by unbalanced MMP/TIMP dynamics which favor inflammation.
| A subset of HIV-patients starting antiretroviral therapy are at risk of developing immune-driven worsening symptoms of a previously treated opportunistic infection. This paradoxical immune reconstitution inflammatory syndrome (IRIS) has been abundantly described in common co-infections such as M. tuberculosis (TB-IRIS), whereas IRIS associated with Schistosoma mansoni (Schisto-IRIS) is less well studied. Nonetheless, HIV and S. mansoni are highly co-endemic in Sub-Saharan Africa and the considerable clinical burden of Schisto-IRIS in the field should not be underestimated. Moreover, no clinical definition exists to help diagnose this complication. Although little is known about the immune dysregulation in Schisto-IRIS, matrix metalloproteinases (MMPs) and tissue inhibitor of metalloproteinases (TIMPs) have been linked to schistosomiasis and TB-IRIS on account of their role in tissue-destructive inflammation. The current study is nested within a three-month case-control study in schistosomiasis/HIV co-infected fishermen starting ART in Kenya. We propose a clinical working definition for Schisto-IRIS, based on critical evaluation of symptoms developing during ART. Our study now links aberrant dynamics of MMPs and TIMPs to Schisto-IRIS as well. Given the role of MMPs and TIMPs in tissue remodeling and inflammation, our findings suggest that Schisto-IRIS is marked by unbalanced MMP/TIMP dynamics that favor inflammation.
| HIV-patients initiating antiretroviral therapy (ART) while dealing with a co-infection are at risk of developing immune reconstitution inflammatory syndrome (IRIS). IRIS is described as a clinical deterioration of HIV patients in the first weeks or months after starting ART, often marked by tissue-destructive inflammation [1–3]. At least three conditions need to be present for an HIV patient to be at risk of IRIS; severe immune suppression, a treated (paradoxical IRIS) -or- undiagnosed (unmasking IRIS) opportunistic infection, and initiation of ART as the trigger. Despite these shared features, IRIS embodies a heterogeneous collection of clinical manifestations [1,2], associated with a plethora of pathogens [3,4].
Among these pathogens, IRIS associated with Schistosoma mansoni (Schisto-IRIS) has received only limited attention in research, with few cases reported prior to 2010 [5–7]. Nonetheless, HIV and Schistosomiasis are highly co-endemic in Sub-Saharan Africa [8]. This is especially true in regions with frequent human-water interaction [9,10], e.g. fishing villages along the shores of Lake Victoria. Areas such as these could form focal-points of IRIS development, as demonstrated previously in a Kenyan cohort with ~36% of Schistosomiasis-HIV patients on ART developing worsening schistosomiasis symptoms consistent with IRIS [11]. Thus, while perhaps not as prevalent as IRIS associated with tuberculosis (TB-IRIS) [3,12], the clinical burden of Schisto-IRIS in the field should not be underestimated.
Little is known on the immune dysregulation in patients who develop Schisto-IRIS, nor has the relevance of findings in TB-IRIS for Schisto-IRIS been studied before. Most findings in TB-IRIS can be directly linked to an acute inflammatory response to TB-antigens, coinciding with monocyte activation and a cytokine storm [13–15]. Elevated levels of matrix metalloproteinases (MMPs) have also been reported, with MMP-1, -3, -7, -8, and -10 being of particular interest [16–18]. MMPs hydrolyze various components of the extracellular matrix, allowing classification as collagenases (e.g. MMP-1), gelatinases (e.g. MMP-2), matrilysins (e.g. MMP-7) or stromelysins (e.g. MMP-10) [16]. Along with tissue inhibitors of metalloproteinases (TIMPs), MMPs are functionally involved in inflammation, granuloma formation and tissue remodeling, thus illustrating their involvement in TB lung pathology and symptoms seen in TB-IRIS [17,18]. Cells of the innate immune system have the capacity to produce MMPs and TIMPs [19,20], further suggesting a major contribution of the innate immune system to TB-IRIS development.
In schistosomiasis, MMP-1 and -2 and TIMP-1 and -2 have been linked to active periovular granulomas in humans [21], whereas increased levels of MMP-8 and -10, and TIMP-1 and -2 were observed in mice [22]. Conversely, mice treated with praziquantel (PZQ) showed a decrease in MMP and TIMP levels, although MMP-10 levels increased consistent with ongoing resorption of fibrous tissue [23]. PZQ treatment has been reported to influence schistosome-specific immune responses [24,25]. Indeed, a pro-inflammatory shift in schistosome-specific cytokine production has been observed in patients treated with PZQ, which may contribute to resistance to re-infection [26]. Moreover, one study previously described an inverse relationship between S. haematobium infection intensity and anti-nuclear antibody (ANA) levels, which rise upon PZQ treatment [27]. Together, these findings highlight protective, but potentially pathogenic immune responses to schistosome antigens following PZQ treatment.
Drawing parallels with TB-IRIS, we hypothesized that Schisto-IRIS is characterized by an over-representation of pro-inflammatory factors after PZQ treatment that might otherwise improve resistance to re-infection. Schisto-IRIS could thus be associated with; increased plasma levels of MMPs or ANAs; decreased plasma levels of TIMPs; and increased monocyte activation (measured by soluble CD14). In addition, we previously reported lower levels of intestinal fatty-acid binding protein (I-FABP) in TB-IRIS patients [13]. Since I-FABP is used as a marker for damage to the intestinal epithelium, we further hypothesized that persisting levels of I-FABP could be associated with intestinal symptoms associated with Schisto-IRIS. Using samples collected from Schistosoma mansoni-HIV co-infected fishermen starting ART in Kenya [11], we conducted a nested 3-month case-control analysis of plasma MMPs, TIMPs, ANAs, sCD14, and I-FABP among those who developed Schisto-IRIS.
Patients from a prospective case-control study at the fishing community in Uyoma, Rarieda District, Kenya, were studied as described previously [10,11]. The study focused on a group of permanent residents that are occupationally-exposed to water infested with the infective stage of the Schistosoma mansoni parasite (S1 Fig). All participating individuals were screened for schistosomiasis and underwent voluntary counseling and testing (VCT) for HIV. All HIV-patients were given lamivudine, stavudine and nevirapine based combination ART shortly after screening, according to Kenya national guidelines at the time. When eggs were identified in stool samples, co-infected patients were also given a single dose of 40mg/kg PZQ treatment according to standard local clinical practice. Seventy-one ART naïve HIV-patients with a history of treated schistosomiasis infection (Schisto+HIV+) were included in the study, of whom 26 developed Schisto-IRIS. In addition, a group of ART naïve HIV-patients without a history of schistosomiasis infection (Schisto-HIV+) were recruited as controls (Fig 1). Patients were followed up for 3 months and blood-plasma was collected prior to initiating ART (baseline), at 2 weeks, at 1 and 3 months after starting ART. ART adherence and efficacy were monitored by oral questioning, CD4 counts and viral load (VL) measurements at each time point (Fig 2). In the current study, only patients with available clinical data were included, who had samples of sufficient quality available at all 4 time points to allow longitudinal analysis.
Diagnosis of schistosomiasis was performed at baseline and all subsequent time points by Kato Katz thick stool smear as described previously [28]. At each timepoint, 1 stool sample was taken, and 2 smears were analyzed by experienced lab technicians. S. Haematobium infection was excluded using a parallel urine filtration method [29]. Patients were further excluded when presenting with eggs belonging to other helminth infections; Ascaris lumbricoides, Trichuris trichuria and hookworm. Other exclusion criteria included; being <18 years of age and having any of the most common co-infections such as malaria, tuberculosis or hepatitis B, respectively detected by microscopy, skin test or serology.
For the purpose of the current study, a working case definition of Schisto-IRIS was refined to match consensus definitions used for the diagnosis of other forms of IRIS (Table 1) [2,30]. Antecedent requirements for paradoxical Schisto-IRIS were; diagnosis of S. mansoni, followed by successful PZQ treatment in close proximity to ART initiation (and eggs were no longer detectable during the next visit) [7]. Based on a clinical Schisto-IRIS description used previously [11], symptoms were re-evaluated according to frequency and relevance, and classified as “major” or “minor” symptoms. Schisto+HIV+ patients were diagnosed with paradoxical Schisto-IRIS when presenting with at least 2 major, or 1 major and 3 minor re-emerging symptoms of otherwise successfully treated schistosomiasis during ART. Major symptoms include; new or worsening bloody diarrhea, new or worsening hematuria, and new or worsening hepatomegaly (HMG), splenomegaly (SMG), or portal vein enlargement (PVE) within 3 months of starting ART. Minor symptoms include; constitutional symptoms (fever, night sweats), focal inflammation (skin infection, skin rash, swollen glands, mouth sores), algetic symptoms (chest pain, joint pain), watery diarrhea, and new eggs in stool (whose presence could not be explained by other possible causes). Finally, other possible explanations, such as re-infection, other co-infections and treatment failure were excluded. In accordance with previous IRIS studies, we propose to use the term “ART-associated schistosomiasis” to refer to PZQ naïve individuals who start de novo production of schistosome eggs during ART. Diagnosis of unmasking Schisto-IRIS in patients with ART-associated schistosomiasis required; heightened intensity of clinical manifestations and a clinical course consistent with paradoxical Schisto-IRIS once PZQ treatment was initiated (Table 2).
Venous blood was drawn into EDTA tubes and plasma was separated from cells by centrifugation and stored at -80°C. Plasma concentrations of 3 MMPs and 2 TIMPs were determined in duplicate using Bio-Plex human MMP and TIMP assay kits (Bio-Rad Laboratories NV-SA, Nazareth, Belgium) according to the manufacturer’s instructions. We thus measured plasma levels of MMP-1, -7, -10 and TIMP-1 & -2. The Bio-Plex array reader and Manager 5.0 software were used to analyse concentrations using a weighted five-parameter logistic curve-fitting method. In addition, plasma concentrations of sCD14 and intestinal fatty-acid-binding protein (I-FABP) were determined by ELISA (HyCult biotechnology BV, Uden, The Netherlands), and C-reactive protein (CRP) was determined using VITROS Chemistry Products CRP Slides (Ortho-Clinical Diagnostics, NY, USA). In addition, a semi-quantitive determination of 8 ANAs (directed against; Smith antigen (Sm), U1 small nuclear ribonucleoprotein (U1 snRNP), snRNP/Sm complex, Sjögren’s-syndrome-related antigen A & B (SS-A & SS-B), topoisomerase I (Scl-70), centromere protein B (CenpB) and Jo-1) was performed by ELISA using index values, calculated by ratio of sample to cut-off calibrator (AESKU.DIAGNOSTICS GmbH & Co., Wendelsheim, Germany).
The study was approved by the Scientific Steering Committee (SSC number 1763) and Ethical Review Committee (ERC) at the Kenya Medical Research Institute (KEMRI) and written informed consent was obtained from all study participants. The use of plasma samples in the current study was approved by the institutional review board of the Institute of Tropical Medicine of Antwerp.
Statistics were performed using SPSS software (version 17.0) or GraphPad Prism (version 7) with significance level set at p < 0.05. Differences between patient groups were analyzed using Mann-Whitney U tests, or Pearson Chi-square tests for dichotomous values. The significant change over time of variables for each patient group was calculated using the Friedman test (p-values shown in graphs). When Friedman tests showed global significance, Dunn’s multiple comparison post-hoc tests and multiplicity adjusted p-values were used to indicate differences between specific time points, (indicated in graphs by horizontal bars with an asterisk). Significant differences between 2 individual time points (baseline vs. month 1 for ANAs) within groups were determined using a Wilcoxon signed-rank test. Correlations were performed using Spearman's rank-order correlation. Because of the hypothesis driven nature of this study, no other correction for multiple testing was applied [31,32].
A subset of plasma samples collected within a prospective case-control study at the fishing community in Uyoma, Kenya [11] were selected for further analysis. Plasma samples from a total of 9 Schisto-IRIS patients, 9 Schisto+HIV+ and 9 Schisto-HIV+ patients were thus analyzed. All 3 groups did not differ significantly in age, gender or baseline viral load (Table 3). No significant differences could be observed between any of the groups for CD4 counts across all 4 time points, except for Schisto-HIV+ patients who had lower counts compared to Schisto-IRIS patients after 3 months on ART (p = 0.047). Time analysis showed a significant increase in CD4 counts for Schisto-IRIS patients (p = 0.008) and Schisto+HIV+ patients (p = 0.007) during 3 months of ART, whereas Schisto-HIV+ participants did not have a significant increase in CD4 counts (p = 0.214) (Fig 3). Dunn’s post-hoc test subsequently highlighted a significant increase from baseline to month 3 in Schisto-IRIS patients (p = 0.004) and Schisto+HIV+ patients (p = 0.003). Schisto-IRIS patients did not differ from Schisto+HIV+ patients for baseline egg-count [eggs per gram (EGP)] and EPG were reduced in both groups after PZQ treatment. Due to patients being referred to government facilities for ART, treatment intervals between PZQ and ART varied from patient to patient (Fig 4). Overall, the treatment interval between PZQ and ART did not differ between selected Schisto-IRIS and Schisto+HIV+ patients (p = 0.287). All patients received PZQ before the standard planned visit at week 2, except for 2 Schisto+HIV+ patients (who received PZQ 35 and 81 days after ART initiation).
All co-infected patients reported in this study experienced clinical signs consistent with S. mansoni infection prior to starting PZQ and ART including; hepato-/splenomegaly, bloody/watery diarrhea, abdominal pains, etc. (S1 Table). Patients were retrospectively classified as Schisto-IRIS or Schisto+HIV+ patients, according to symptoms presented during ART. Both groups showed similar distribution of abdominal symptoms at baseline, while neither group experienced PVE at this time. SMG at baseline was diagnosed in 6/9 (67%) Schisto-IRIS patients but only in 1/9 (11%) Schisto+HIV+ patients, though 2/9 (22%) Schisto+HIV+ patients experienced HMG. Following PZQ/ART treatment, EPG declined in all patients. Symptoms subsided within 3 months of ART in patients who were not diagnosed with IRIS. In contrast, 6/9 (67%) Schisto-IRIS patients developed new or worsening bloody diarrhea, 3/9 (33%) developed new HMG or SMG and 5/9 (56%) developed new PVE (Table 4). All Schisto-IRIS patients developed at least 3 minor symptoms during ART, except for one who developed 2 (in addition to 2 major symptoms). This included 7 patients with paradoxical IRIS and 2 patients with unmasking IRIS, whom did not differ in baseline characteristics or CD4 count. Unmasking Schisto-IRIS patients showed zero EPG prior to ART, but experienced de novo egg production at 2 weeks after starting ART. These patients initiated PZQ treatment only after egg production was diagnosed (at 13 and 15 days post ART). Since patients were followed up at pre-determined visits, IRIS was diagnosed at the closest visit. The median time to IRIS diagnosis was 29 days (IQR 14–91) for paradoxical Schisto-IRIS, and 126 days (IQR 119–132) for unmasking Schisto-IRIS (p = 0.053). We observed a significant correlation between treatment interval and time to IRIS diagnosis (R = 0.759; p = 0.021) (Fig 4).
Increased plasma levels of MMPs have previously been associated with Schistosomiasis, TB infection and TB-IRIS [22,23,33–35]. In order to investigate the role of MMPs in Schisto-IRIS, we evaluated 3 different MMPs classified as either collagenase (MMP-1), matrilysin (MMP-7) or stromelysin (MMP-10) [16]. We thus evaluated plasma MMP levels in 3 patient groups at baseline and after 2 weeks, 1 month and 3 months on ART (Table 5). We could not observe significantly different MMP-1, MMP-7 or MMP-10 levels in Schisto-IRIS patients when directly compared to Schisto+HIV+ patients. However, Schisto-IRIS patients showed significantly higher MMP-10 levels during ART compared to Schisto-HIV+ patients (p ≤ 0.030). Conversely, MMP-10 levels were similar between Schisto+HIV+ and Schisto-HIV+ patients during ART, except for month 1 (p = 0.041). Subsequent analysis over time revealed a significant overall change in plasma MMP-10 levels for Schisto+HIV+ patients (p = 0.005), with Dunn’s post-hoc test highlighting a significant decrease from baseline to month 3 specifically (p = 0.006). In contrast, Schisto-IRIS and Schisto-HIV+ patients showed no significant change in MMP-10 levels (Fig 5A–5C).
We next investigated plasma levels of TIMP-1 and TIMP-2 to evaluate their role in Schisto-IRIS (Table 5). TIMP-1 levels did not differ significantly between the 3 patient groups at any time point. Although TIMP-1 levels declined over time in each group, this change over time only reached significance in Schisto-IRIS patients (p = 0.012, Fig 5D and 5E). Dunn’s post-hoc test subsequently highlighted a significant decline from baseline to month 3 in these patients (p = 0.012). Conversely, Schisto-IRIS patients had markedly lower TIMP-2 levels across the board compared to Schisto+HIV+ patients, reaching significance at week 2 (p = 0.007), month 1 (p = 0.005) and month 3 (p = 0.031). No differences over time could be observed in TIMP2 levels within the Schisto+HIV+ group.
In order to evaluate the balance between MMPs and TIMPs, we next analyzed ratios of MMP-1, MMP-7 and MMP-10 to TIMP-1 and TIMP-2 in patients and controls (S2 Fig). The ratio of MMP-10/TIMP-2 decreased over time in Schisto+HIV+ patients (p = 0.012), whereas the ratio of TIMP-1/TIMP-2 decreased in Schisto-IRIS patients only (p = 0.028).
I-FABP is released into the bloodstream upon damage to the intestinal epithelium, sCD14 is shed upon monocyte activation, and CRP is a well-known acute phase protein. In theory, these markers could therefore be used for monitoring tissue damage and inflammation in Schistosomiasis and/or Schisto-IRIS. We thus evaluated plasma levels of these markers in all 3 patient groups (Table 6 & Fig 6). However, all patients showed comparable I-FABP & sCD14 levels at every time point. No differences were observed for CRP levels between Schisto-IRIS and Schisto+HIV+ patients either. Compared to Schisto-HIV+ patients, Schisto-IRIS and Schisto+HIV+ patients showed higher CRP levels after 1 month (p = 0.021) and 2 weeks (p = 0.012) of ART respectively. Only Schisto+HIV+ patients showed significant overall variation in CRP over time (p = 0.028), while Dunn’s post-hoc test showed no significance.
Anti-nuclear antibodies are associated with auto-immune diseases and are reported to rise in S. haematobium patients following PZQ treatment [27]. We thus performed a semi-quantitative analysis of 8 ANAs in Schisto-IRIS and Schisto+HIV+ patients at baseline and after 1 month of ART (Fig 7). No significant differences could be observed between patient groups at either time point. Nonetheless, Schisto-IRIS showed an overall decrease in plasma levels of 5 ANAs after 1 month of ART; U1-RNP (p = 0.012), SnRNP/Sm (p = 0.027), Sm (p = 0.020), Scl-70 (p = 0.039), SS-A (p = 0.020), SS-B (p = 0.074), CenpB (p = 0.570), Jo-1 (p = 0.313), whereas Schisto+HIV+ patients only showed decreased levels of Scl-70 (p = 0.027). Comparison of change over time (delta-values calculated by subtracting baseline from month 1) showed a significantly stronger decline of SnRNP/Sm (p = 0.046) and SS-A (p = 0.008) in Schisto-IRIS patients compared to Schisto+HIV+ patients.
Since the number of CD4+ T cells directly influence immunological processes, we next correlated CD4 counts to our observations. Schisto+HIV+ patients showed a significant negative correlation between CD4 counts and MMP-1, pre-ART (R = -0,683; p = 0.042) and at week 2 (R = -0,717; p = 0.030), whereas Schisto-IRIS and Schisto-HIV+ patients did not. Next, we evaluated whether excretion of S. mansoni eggs could be associated with damage to the intestinal epithelium. However, I-FABP levels showed no correlation with EPG at any time point in any group.
To evaluate whether MMP-10 levels could be maintained by inflammatory factors, we then performed a correlation with CRP levels. However, no correlations were observed between CRP and MMP-10 levels at any time point for any group (S3 Fig). We next correlated CRP levels with ANA levels to evaluate a potential link between ANA-levels and systemic inflammation (S4 Fig and S5 Fig). Schisto-IRIS patients showed moderate to strong correlations at baseline between CRP levels and levels of U1-RNP, SS-A, Jo-1, and SS-B (R ≥ 0.669; p ≤ 0.043). Except for U1-RNP, these correlations were preserved at month 1 (R ≥ 0.695, p ≤ 0.043). Schisto+HIV+ patients showed a similar pattern of correlations at baseline, although CenpB correlated to CRP instead of SS-B (R = 0.783, p = 0.017). At month 1, CRP correlated with SS-A, SS-B, Sm, and CenpB (R ≥ 0.722, p ≤ 0.031) in these patients. Nonetheless, CPR and ANA levels did not show a similar change over time, as no correlation could be observed between delta values (baseline subtracted from month 1) of CRP and any of the ANAs determined. Finally, we evaluated a potential link between the decreasing TIMP-1 and ANA levels in Schisto-IRIS patients (S6 Fig and S7 Fig). However, only Schisto+HIV+ patients showed correlations between TIMP-1 and 6 of the 8 ANAs at both time points (R ≥ 0.700; p ≤ 0.043).
Among the different faces of IRIS developing in HIV-patients, Schistosomiasis-associated IRIS remains largely uninvestigated despite both infections being highly co-endemic [8]. Unlike clinical definitions used for TB-IRIS [2] and Cryptococcal IRIS [30], no consensus definition for Schisto-IRIS exists. Within a previously described cohort of HIV-patients with Schistosomiasis co-infection, 36.6% developed Schisto-IRIS [11]. Here, we report clinical characteristics of patients who developed Schisto-IRIS related to S. mansoni and propose a working definition for the diagnosis of paradoxical and unmasking Schisto-IRIS in resource limited settings. In addition, we explored the immunopathogenesis of Schisto-IRIS by measuring plasma levels of 3 MMPs, 2 TIMPs, sCD14, and 8 anti-nuclear antibodies, which were previously suggested to have a role in TB-IRIS and/or schistosomiasis [22,23,27,33–37]. We hypothesized that Schisto-IRIS could be associated with increased plasma levels of MMPs, sCD14, and ANAs; or decreased plasma levels of TIMPs. To that end, we compared plasma levels of these markers during 3 months of ART between Schisto-IRIS patients and Schisto+HIV+, as well as Schisto-HIV+ controls. In addition, we explored I-FABP levels as a marker of intestinal damage. We report a significant decline in MMP-10 levels following ART initiation in Schisto+HIV+ controls, but not in Schisto-IRIS patients. Conversely, plasma levels of TIMP-1 decreased in Schisto-IRIS patients, and TIMP-2 levels were significantly lower shortly after starting ART. In line with our hypothesis, these findings suggest that Schisto-IRIS patients in our cohort experience a MMP/TIMP profile that favors inflammation and tissue damage. Contrary to our hypothesis, plasma levels of ANAs decreased upon ART in Schisto-IRIS patients, possibly reflecting the release of S. mansoni antigens upon PZQ treatment [27].
The functions ascribed to MMPs range from tissue remodeling and angiogenesis to regulation of immune responses and inflammation [38]. In Schistosomiasis, MMPs regulate the granulomatous response to schistosome-eggs [22]. Upon praziquantel treatment, MMP levels decline in parallel with diminishing inflammatory responses [23]. In line with this, Schistosomiasis-HIV patients in our cohort who did not develop Schisto-IRIS showed a significant decline in MMP-10 levels during ART. Schisto-IRIS patients, however, experienced persistent levels of MMP-10 throughout follow-up. Although we observed no significant differences in MMP-1, -7 or -10 in direct comparison between these 2 groups, Schisto-IRIS patients retained significantly higher MMP-10 levels compared to Schisto-HIV+ controls, whereas Schisto+HIV+ controls did not. Overall, these findings indicate that Schisto-IRIS patients do not readily normalize MMP-10 levels within the first months of ART. The spread in data likely masked this effect in direct comparison between groups. Interestingly, MMP-10 gene expression has previously been observed to be paradoxically elevated in praziquantel-treated mice, matching declining collagen gene expression [23]. The persistent levels of MMP-10 in Schisto-IRIS patients observed here could therefore indicate an ongoing immune response to S. mansoni antigens [39], or a continuous resorption of fibrous tissue surrounding schistosome-eggs [23]. Alternatively, MMP-10 levels could have been maintained by inflammatory factors such as CRP, interferon-gamma, interleukin-6, and/or monocyte activation [40,41], which have previously been associated with TB-IRIS [13,42]. However, we could not observe significant correlations between CRP levels and MMP-10 in our current study. Moreover, no differences were observed for sCD14, which is consistent with our previous observations on TB-IRIS [13].
Counterbalancing the effects of MMPs, TIMPs are natural inhibitors of MMP activity [43]. The balance between MMPs and TIMPs thus influences the level of tissue degradation and inflammation [38]. As expected after praziquantel treatment [23], TIMP-1 levels declined during ART in both Schistosoma-infected groups. However, this decline was much more pronounced in Schisto-IRIS patients. Conversely, TIMP-2 levels remained stable after ART initiation in all groups, but were significantly lower in Schisto-IRIS patients compared to Schisto+HIV+ patients. These findings suggest that Schisto-IRIS patients have a lowered capacity to compensate for the inflammatory effects of MMP-10 which seem to persist during ART. Interestingly, Schisto-HIV+ controls showed similarly low levels of TIMP-2 during ART as Schisto-IRIS patients. However, MMP-10 levels were lower still, leading to significantly decreased MMP-10/TIMP-2 ratios at week 2 compared to Schisto-IRIS patients. Although the difference in MMP-10/TIMP-2 ratios between Schisto-IRIS and Schisto+HIV+ patients did not reach significance, only Schisto+HIV+ patients showed a significant decrease of this ratio during ART. Overall, these findings suggest diverging TIMP & MMP dynamics in Schistosomiasis-HIV patients who develop IRIS, favoring inflammation and/or tissue damage which becomes clinically apparent with the onset of IRIS symptoms.
All co-infected patients reported in this study experienced clinical signs consistent with S. mansoni infection prior to starting PZQ and ART [44,45]. Following effective PZQ/ART treatment, most symptoms gradually subsided within 3 months of ART in patients who did not develop IRIS, in line with decreasing levels of MMP-10. Using the working definition proposed in this study, Schisto-IRIS patients could be distinguished from Schisto+HIV+ controls, as the clinical condition worsened during ART. Compared to baseline, all 9 Schisto-IRIS patients in our study developed new or worsening symptoms of schistosomiasis during this period, despite successful PZQ treatment. In line with persistent MMP-10 levels, a steeper decline in TIMP-1 and consistently low TIMP-2 levels suggest that these patients experience a more vigorous and extended period of tissue damage and/or inflammation. Indeed, the majority of Schisto-IRIS patients developed bloody diarrhea (>50%), new PVE (>50%), and/or new HMG/SMG (>30%) during ART, often accompanied by minor symptoms such as fever, etc. Nonetheless, this did not coincide with altered damage to the intestinal epithelium, as we observed no differences in I-FABP. Overall, our observations on MMP/TIMP dynamics during ART thus correspond to the clinical spectrum of Schisto-IRIS patients as defined here, and may drive or be driven by the ‘major’ symptoms which occurred more rarely in Schisto+HIV+ controls [46–48]. Nonetheless, the distribution of symptoms among IRIS patients during ART remains somewhat heterogeneous, as is to be expected in IRIS [2,30]. Moreover, the interactions of MMPs and TIMPs are very complex. Consequently, it is difficult to predict the specific roles these factors may have, and what other factors (e.g. cytokines) may be involved. The immune responses and cytokine profiles of Schisto-IRIS patients should therefore be further explored.
Despite the heterogeneity that surrounds IRIS, a high pre-ART antigen burden is commonly recognized as a risk factor, as is a short interval between treatment for the opportunistic infection and ART [49,50]. Although the PZQ-ART interval was similar between our patients and controls, we observed a correlation between PZQ-ART interval and onset of IRIS symptoms. Patients who received PZQ treatment sooner relative to ART (i.e. before ART) showed earlier onset of IRIS symptoms than patients who received PZQ later (i.e. during the first 2 weeks of ART), suggesting a time-dependent association between PZQ treatment, ART initiation, and Schisto-IRIS. Interestingly, studies in HIV-negative patients have demonstrated a rise in pro-inflammatory responses to S. haematobium [26,51] and S. mansoni [52,53] antigens within the first months following PZQ treatment. These responses may reflect either the removal of worm-induced immunosuppression or the release of adult worm antigens as a result of PZQ treatment [51]. Although our study did not directly determine antigen levels, we determined relative plasma levels of 8 anti-nuclear antibodies, which have previously been reported to correlate inversely with intensity of S. haematobium infection and rise during PZQ treatment [27]. In contrast to Schisto+HIV+ controls, our Schisto-IRIS patients showed a significant decrease in plasma levels of several ANAs after 1 month of ART. Given the inverse relationship of ANA-levels with infection intensity, one could argue that this decrease reflects a sudden release of antigens. Alternatively, a stronger S. mansoni specific immune response to these antigens could be present which indirectly downregulates ANA levels. However, this decrease was not mirrored in CRP levels, since no correlation could be observed in change over time (delta value) with ANAs. Considering the time-dependent association with PZQ treatment, it is thus plausible that Schisto-IRIS manifests itself as an aggravated reaction to antigens released by worms that were killed as a result of PZQ treatment. Still, additional studies are needed to fully explore the relationship between ANAs and S. mansoni antigen loads in Schisto-IRIS.
Although this work is nested in one of the first Schisto-IRIS specific studies to date, our strict selection of patients with complete samples and follow-up data lead to a relatively small population. As we did not include a Schisto+HIV- population, our study also cannot provide additional insights in the effects of PZQ without ART. Since this study was focused on S. mansoni co-infection, the occurrence of IRIS with other species (e.g. S. haematobium) has also not been investigated. The results reported here should thus inspire larger studies to fully explore MMP profiles in Schisto-IRIS, associated with different Schistosoma species. Our study accounted for re-infection with S. mansoni by monitoring patients at intervals shorter than 6 weeks. Although exposure between study visits cannot be fully accounted for, all Schisto-IRIS patients developed symptoms before any de novo eggs were documented in stool. We identified onset of IRIS symptoms at the nearest pre-planned visit, with 5 patients showing onset of IRIS symptoms earlier than 45 days on ART, and 4 later than 89 days on ART. Thus, our sample collection did not include IRIS-specific time points, but instead spans the timeframe in which IRIS developed. Nonetheless, as our observations persist across multiple time points, the lack of an IRIS time point did not affect our conclusions. Moreover, no differences could be observed when comparing early- and late-onset IRIS cases for any of the parameters analyzed. Finally, our patient selection included 2 unmasking IRIS patients, making our population more heterogeneous. These patients showed no significant variation compared to paradoxical IRIS patients for any of the clinical variables tested (apart from EPG). However, unmasking Schisto-IRIS patients showed some modest differences in laboratory markers at month 3. Since both cases initiated PZQ treatment after starting ART, this timing could have altered MMP/TIMP dynamics we observed at month 3. Nonetheless, these patients also developed IRIS symptoms at this time point, which likely explains these differences. A larger unmasking Schisto-IRIS population is required to fully explore differences with paradoxical Schisto-IRIS.
In conclusion, we describe characteristics of patients who developed IRIS related to S. mansoni in one of the first Schisto-IRIS cohorts to date. Therefore, we propose a refined working definition for the diagnosis of paradoxical and unmasking Schisto-IRIS in resource limited settings. Schisto-IRIS patients in our study showed persistent plasma levels of MMP-10, along with a steep decline in TIMP-1 and low levels of TIMP-2 during 3 months of ART. Consistent with the IRIS symptoms reported in the definition, these aberrant MMP and TIMP dynamics suggest the presence of prolonged inflammation and/or tissue damage. Although further research is required, decreasing levels of anti-nuclear antibodies in Schisto-IRIS patients following PZQ/ART may reflect a PZQ-induced release of S. mansoni antigens, which in turn may drive Schisto-IRIS inflammation. Elucidating the immune pathogenesis behind this complication could lead to treatment strategies for Schisto-IRIS, as well as provide insight in the heterogeneous disease that is IRIS in general.
|
10.1371/journal.pgen.1002801 | The Rad4TopBP1 ATR-Activation Domain Functions in G1/S Phase in a Chromatin-Dependent Manner | DNA damage checkpoint activation can be subdivided in two steps: initial activation and signal amplification. The events distinguishing these two phases and their genetic determinants remain obscure. TopBP1, a mediator protein containing multiple BRCT domains, binds to and activates the ATR/ATRIP complex through its ATR-Activation Domain (AAD). We show that Schizosaccharomyces pombe Rad4TopBP1 AAD–defective strains are DNA damage sensitive during G1/S-phase, but not during G2. Using lacO-LacI tethering, we developed a DNA damage–independent assay for checkpoint activation that is Rad4TopBP1 AAD–dependent. In this assay, checkpoint activation requires histone H2A phosphorylation, the interaction between TopBP1 and the 9-1-1 complex, and is mediated by the phospho-binding activity of Crb253BP1. Consistent with a model where Rad4TopBP1 AAD–dependent checkpoint activation is ssDNA/RPA–independent and functions to amplify otherwise weak checkpoint signals, we demonstrate that the Rad4TopBP1 AAD is important for Chk1 phosphorylation when resection is limited in G2 by ablation of the resecting nuclease, Exo1. We also show that the Rad4TopBP1 AAD acts additively with a Rad9 AAD in G1/S phase but not G2. We propose that AAD–dependent Rad3ATR checkpoint amplification is particularly important when DNA resection is limiting. In S. pombe, this manifests in G1/S phase and relies on protein–chromatin interactions.
| DNA structure–dependent checkpoint activation and the amplification of checkpoint signals are carefully modulated to allow the checkpoint kinases to delay mitosis and regulate DNA metabolism. While much work has gone into understanding how this checkpoint functions, the mechanism by which the checkpoint signal is amplified is less clear. We have characterised a conserved domain in the Schizosaccharomyces pombe TopBP1 homolog, Rad4TopBP1 (also known as Cut5) that is capable of activating the ATR homolog Rad3ATR. We demonstrate that this domain is not required for initial checkpoint activation, but functions to amplify the checkpoint signal, likely when the presence of single-stranded DNA is limiting. Our data suggest that the function of the Rad4TopBP1 ATR-Activation Domain (AAD) is mediated by interactions between checkpoint proteins and phosphorylated histone H2A, which is itself promoted by Rad3ATR. We propose that the resulting amplification of the checkpoint signal is particularly important in G1-S phase, when resection is limited.
| The DNA damage checkpoint is an elaborate signal transduction pathway that monitors the integrity of the DNA, prevents cell cycle progression and promotes appropriate DNA metabolism [1] reviewed in [2]. The DNA damage sensors associated with checkpoint activation define two separate DNA structure-dependent signal transduction cascades. Each pathway engages a phospho-inositol-3 kinase-like protein kinase (PIKK); either the Ataxia Telangiectasia Mutated (ATM) or the Ataxia Telangiectasia and Rad3 related (ATR) kinase [3]. ATM detects DNA double strand breaks (DSBs) by interaction with the Mre11-Rad50-Nbs1 repair complex, while ATR primarily senses single stranded-DNA (ss-DNA) through interactions with RPA. Both ATM and ATR are conserved in the model organisms S. pombe and S. cerevisiae.
For ATR to recognise a DNA lesion, single-stranded DNA (ssDNA) needs to be formed - for example by DNA repair-dependent DNA processing [4] or following the replication machinery encountering the unrepaired lesion [5]. Once ssDNA is generated, it is immediately coated by replication protein A (RPA) (Reviewed in: [6]). Multiple ATR molecules are initially recruited to ssDNA regions via ATRs obligate binding partner, ATRIP, which binds directly to RPA [7], [8]. ATR-ATRIP recruitment to ssDNA-RPA is necessary for “basal” ATR activation, but is insufficient for full checkpoint activation: co-recruitment of a second complex consisting of three PCNA-like proteins, Rad9, Hus1 and Rad1 (known as the 9-1-1 clamp) is also necessary. 9-1-1 is loaded in parallel to ATR recruitment at 5′ ssDNA/dsDNA junctions by the checkpoint clamp loader Rad17-RFC[2–5] [9], [10], [11]. (Figure 1A).
When ATR-ATRIP is first loaded at the site of ssDNA, its “basal” kinase activity promotes phosphorylation of its immediate neighbours, including ATRIP [12], [13], an in trans phosphorylation of a residue within ATR itself, T1989 [14], and the subunits of the 9-1-1 clamp [15], [16]. Dependent on the concomitant recruitment of 9-1-1, a further protein, TopBP1, is recruited [17]. TopBP1 is recruited via an interaction between its BRCT (1+2) domains and a constitutive phosphosphorylation on the C-terminus of Rad9 [18], [19]. Similarly in both yeast systems, Saccharomyces cerevisiae and Schizosaccharomyces pombe, the TopBP1 homologs, Dpb11 and Rad4 respectively, are recruited by the phosphorylation of the C-terminus of Rad9Ddc1 creating a binding site for a pair of BRCT domains (Figure 1A). In S. pombe, the C-terminal phosphorylations occur on Rad9 at residues T412 and S423 [20]. This subsequently recruits Rad4TopBP1 via interaction with BRCT pair (3+4). However, unlike in mammalian cells, T412 and S423 in S. pombe are directly targeted by Rad3ATR in response to its ssDNA/RPA binding and concomitant 9-1-1 loading [16], [20], [21]. Despite these differences, in both S. pombe [20] and mammalian cells [14], Rad4TopBP1 recruitment promotes the formation of a Rad3ATR/9-1-1/Rad4TopBP1 complex (Figure 1A). However, the mode of interaction of Rad3ATR and Rad4TopBP1 within this complex has not been defined in S. pombe.
Mammalian TopBP1 can directly activate ATR-ATRIP both in vitro in the absence of ssDNA/RPA and when over-expressed in cells. TopBP1-dependent ATR activation requires an ATR activation domain (AAD) situated between the 6th and 7th BRCT domains [17] and mutation of a conserved aromatic residue within this unstructured region, W1147, prevents this mode of ATR activation. The AAD contacts a region within the C-terminus of ATR, between the kinase and FATC domains [22], which has been termed the PIKK Regulatory Domain (PRD). Mutation of a conserved PRD residue, K2598, similarly abolishes TopBP1-dependent ATR activation. In both mammalian cells and Xenopus extracts the interaction between TopBP1 and ATR-ATRIP appears to be essential for checkpoint activation in response to replication stress [17], [22], although the initial in trans phosphorylation of ATR on T1989, reportedly essential for full ATR activation, is TopBP1-independent [14].
In the budding yeast model system, the intrinsically disordered C-terminal extension of the TopBP1 homolog, Dpb11TopBP1, contains an AAD which interacts with and activates Mec1ATR via a pair of aromatic residues, W700 and Y735 [22], [23], [24], [25]. Interestingly, in S. cerevisiae, at least two distinct Mec1ATR activation domains have been identified: in addition to the Dbp11TopBP1 AAD, the C-terminal tail of Ddc1Rad9 (S. cerevisiae homolog of the 9-1-1 subunit Rad9) contains an AAD that can directly stimulate Mec1ATR activity in vitro and contributes to checkpoint activation in vivo [11], [26]. The key residues in Ddc1Rad9 required for Mec1ATR activation are W352 and W544. W352 resides on the surface of the PCNA-like domain, while W544 lies within the intrinsically disordered C-terminus. In vivo the Ddc1Rad9 AAD is essential for Mec1ATR activation when S. cerevisiae cells are in G1 [24], [26], while the Ddc1Rad9 AAD acts redundantly with the C-terminal AAD of Dpb11TopBP1 during checkpoint activation in G2. It is proposed that, at least in S. cerevisiae, a minimum of one other protein contains an equivalent AAD (Reviewed in [27]).
We have previously shown that S. pombe Rad4TopBP1 is not required for the Rad3ATR-dependent and DNA damage-dependent phosphorylation of Rad26ATRIP or the 9-1-1 clamp subunits [16], [20], demonstrating that Rad3ATR is active at sites of DNA damage in the absence of activation by the Rad4TopBP1 AAD. However, the presence of Rad4TopBP1 is clearly required to form a robust Rad3ATR/9-1-1/Rad4TopBP1 complex [20], to recruit the Crb253BP1 mediator protein [28], [29] and for Rad3ATR to phosphorylate downstream substrates such as Chk1-S345 [30] and promote robust checkpoint activation.
To further explore the role of Rad4TopBP1 in checkpoint activation in S. pombe we identified and characterised the Rad4TopBP1 AAD. We show that Rad4TopBP1 can interact with Rad3ATR via its AAD and that the AAD contributes to Rad3ATR activation in vivo. We observe that the biological function of the Rad4TopBP1 AAD is most important in G1/S phase, when resection is limited, and that reducing DSB resection in G2 following ionising radiation results in compromised Chk1 phosphorylation in the absence of Rad4TopBP1 AAD function. In order to separate out and study Rad4TopBP1 AAD-dependent Rad3ATR activation we developed a Rad4TopBP1 AAD-dependent lacO-LacI checkpoint activation system for S. pombe and used this to show that Rad4TopBP1 AAD-dependent Rad3ATR activation is also dependent on histone H2A phosphorylation. Consistent with a role for this chromatin modification, mutations in Crb2 that interfere with phospho-binding by Crb2 also decrease Rad4TopBP1 AAD-dependent checkpoint activation. Thus, the Rad4TopBP1 AAD-dependent Rad3ATR activation pathway is chromatin dependent, implying a role in checkpoint amplification and maintenance.
In S. cerevisiae, two aromatic residues, W700 and Y735 were identified as critical for Dbp11TopBP1 AAD activity [24], [25]. We similarly created an alignment of Rad4TopBP1 with the C-terminal tails of a variety of TopBP1 homologs and identified a short sequence encompassing Y599 in S. pombe (Figure 1B, 1C), which we subsequently confirmed as defining the AAD activity of Rad4TopBP1 (see below). To characterise the function of the Rad4TopBP1 AAD, we separately mutated the conserved aromatic amino acid to create strain rad4-Y599R or deleted the minimal conserved motif to create rad4-Δ[595–601].
To establish if the Rad4TopBP1 AAD interacts with Rad3ATR we expressed and purified recombinant Rad4[288–648] (that includes BRCT (3+4) and the unstructured C-terminal tail) as a fusion with GST. Constructs containing a mutation in either the AAD (Y599R), deletion of residues 643–645 encompassing a putative cyclin binding motif (RxL), or mutation of a CDK phosphorylation site (S641A) were similarly purified. Wild-type and all the mutated recombinant proteins, when incubated with yeast extracts, bound to and co-purified the control interacting partner of Rad4TopPB1, Cdc13CyclinB (unpublished data). Wild-type Rad4-GST also co-purified Rad3ATR-Myc from extracts prepared from 3myc-rad3 cells (Figure 1D). In contrast, the amount of Rad3ATR-Myc pulled-down with recombinant AAD-mutated Rad4TopBP1-GST was reproducibly reduced (n = 4). Neither Rad3ATR-myc nor Cdc13CyclinB were pulled-down with a GST-only control. We conclude that, in vitro, this residue of Rad4TopBP1 is part of a Rad4TopBP1 interaction domain for Rad3ATR, consistent with it having the properties of an AAD. We also note that the interaction with Cdc13CyclinB is not abolished by loss of the RxL motif and that mutation of S641 affects the Rad4TopBP1:Rad3ATR interaction in vitro (Figure 1D, lane 4).
Both rad4-Y599R and rad4-Δ[595–601] displayed normal cell cycle progression (data not shown), indicating that, as is observed in S. cerevisiae [24], the Rad4TopBP1 AAD is not required for unperturbed DNA replication. In response to UV, MMS and HU treatment, rad4-Y599R and rad4-Δ[595–601] cells showed intermediate sensitivity when compared to rad4+ and checkpoint defective rad3Δ (Figure 1E and Figure S1A). To establish if the sensitivity to DNA damage correlated with a defective G2 DNA damage checkpoint, we monitored cell cycle progression after cells were synchronised in G2 and UV-irradiated. Following exposure to 50 Jm-2 (Figure 2A), rad4-Y599R cells displayed premature release from cell cycle arrest (∼20 min earlier than rad4+ after 50 Jm-2). We next monitored the sensitivity and checkpoint response to ionising radiation. rad4-Y599R mutant cells displayed only very mild sensitivity to IR (Figure 2B) and the checkpoint was mildly extended (Figure 2C). We do not know the reason for the slight extension of the G2 delay: no obvious increase in numbers or duration of Rad22Rad52 foci were observed, indicating no significant delay to DSB repair (Figure S1B).
We next examined Chk1 phosphorylation status in rad4+ and rad4-Y599R cells as a surrogate for checkpoint activation (Figure 2D). In response to 200 Jm-2 UV irradiation, asynchronously growing rad4-Y599R cells displayed reduced Chk1 phosphorylation when compared to rad4+ cells, consistent with the partial checkpoint defect observed. Conversely, in response to IR, no significant difference is seen between rad4-Y599R and rad4+. An upstream target of Rad3ATR is the C-terminus of histone H2A [31], [32]. To establish if the UV-specific defect in Rad3ATR-dependent phosphorylation is specific to Chk1, we monitored γH2A formation following either UV or IR treatment (Figure 2E). As was seen for Chk1 phosphorylation, a significant decrease in γH2A is observed in rad4-Y599R cells following UV but not IR treatment when compared to rad4+ cells.
The pattern of DNA damage sensitivity seen for rad4-Y599R cells is consistent with a specific sensitivity within S phase. >70% of fission yeast cells in an asynchronous culture are in G2 and mitosis is followed rapidly by S phase: G1 is extremely short. In response to IR, the G2 DNA damage checkpoint is robustly activated and DNA repair completed before cells pass through mitosis and into S phase [33]. Thus, following IR, relatively few cells replicate damaged DNA. Conversely, the G2 checkpoint is not robustly activated following UV [34] and the majority of UV-irradiated cells pass through mitosis and enter S phase with damaged DNA. To monitor S phase-specific events, we thus examined γH2A induction in cells treated with either hydroxyurea (HU), an inhibitor of ribonucleotide reductase, or Camptothecin (CPT), an inhibitor of topisomerase I (Figure 2F). Consistent with both agents manifesting cytotoxicity in S phase, γH2A levels were significantly reduced when comparing rad4-Y599R with rad4+ cells. Finally, since replication of UV damaged DNA induces Cds1Chk2 activity [35], we monitored the kinase activity of immuno-precipitated Cds1Chk2 following UV irradiation of rad4-Y599R and rad4+ cells (Figure 2G). Cds1Chk2 activation was reproducibly lower for rad4-Y599R, indicating an impaired S phase checkpoint activation (n = 3).
If the rad4-Y599R mutant is deficient in activation of Rad3ATR in S phase, we would anticipate increased sensitivity to IR within S phase when compared to rad4+ cells. To test this possibility, rad4-Y599R mutant and rad4+ cells where either synchronised in G2 cells using cdc25-22 or in G1 using a cdc10-m17. Following the block, cells were released by reducing the temperature and cell cycle progression was monitored by FACS analysis (Figure 3A, 3B). Cells were irradiated with 50 Gy IR at the times indicated. rad4-Y599R cells showed significant increased sensitivity when compared to rad4+ when irradiated in S phase, but not when irradiated in G2, when S phase is complete (i.e. see Figure 3B). In an equivalent cdc25-22 block and release experiment, we monitored Chk1 phosphorylation and γH2A induction (Figure 3C). Unlike when asynchronous rad4-Y599R cells are irradiated (>70% of such cells are in G2), when rad4-Y599R cells were irradiated in early-mid S phase, Chk1 phosphorylation was moderately reduced for the first 40 minutes after irradiation and γH2A levels were similarly decreased when compared to rad4+. Interestingly, following progression through S phase and into G2 (150 minute time point), Chk1 phosphorylation levels increased significantly in rad4-Y599R cells, although the same was not seen for γH2A levels.
To determine that the use of cdc25-22 synchronisation was not generating an artefact (Cdc25 is an activator of Cdc2-Cyclin B, which itself is required for normal DNA damage responses in G2 [36]), we used the alternative method of synchronisation where cells were arrested in G1 using cdc10-M17 and released directly into S phase (Figure 3D). Unlike IR treatment of asynchronous cultures where equivalent levels of γH2A were observed (Async IR), treatment of rad4-Y599R cells at 30, 60 or 90 minutes after release from arrest resulted in decreased γH2A levels and Chk1 phosphorylation when compared to rad4+ control cells.
In S. cerevisiae, co-localisation of two or more checkpoint proteins to arrays of lacO repeats bypasses the requirement for DNA damage in Mec1-mediated checkpoint activation [37]. To establish the role of the Rad4TopBP1 AAD in a what has previously been characterised as an RPA-ssDNA independent system, rad4TopBP1, rad9 and rad3ATR were each fused to a construct encoding GFP, the E. coli lac-repressor (LacI) and a nuclear localization signal (NLS); GFP/LN (Figure 4A). The resulting plasmids express the fusion construct under the control of a thiamine-repressible (nmt41) promoter. We established that each of the fusion constructs were functional by expressing them individually in the corresponding null mutants. Each was able to suppress the DNA damage sensitivity (and for rad4TopBP1, the thermosensitivity) of the appropriate mutant, although for rad4-GFP/LN genotoxin resistance was not restored to wild-type levels (Figure S1C–S1E). When expressed in cells harbouring 256 repeats of the lac operator sequence (lacO) integrated at the ura4+ locus, each fusion protein formed a single nuclear focus. No foci were detected in cells devoid of lacO arrays (Figure 4B).
We used Chk1 phosphorylation as a readout for DNA damage checkpoint activation (Figure 4C). Following thiamine removal (induction takes between 12 and 16 hours [38]), Chk1 became phosphorylated in lacO containing cells, but not in lacO-negative control cells, when either Rad3ATR, Rad4TopBP1 or Rad9 LacI fusion proteins were expressed. Similar results were obtained when each pair-wise combinations of two fusion proteins were expressed (Figure 5B). In S. pombe, DNA damage checkpoint activation results in cell cycle arrest and cell elongation. Elongated cells were observed upon expression of single fusion proteins (data not shown), confirming checkpoint activation. From these data we conclude that, in S. pombe, as in mammals [39] tethering of any of these single checkpoint proteins to a lacO array is sufficient to activate the DNA damage checkpoint and that, in contrast to the analogous experiments reported for S. cerevisiae, forced co-localisation of two checkpoint proteins is not required [37].
To establish if the Rad4TopBP1 AAD is involved in this damage-independent mode of checkpoint activation, we tested if Rad3ATR tethering could result in Chk1 phosphorylation in a rad4-Y599R mutant background (Figure 5A). While Rad3-GFP/LN expression resulted in induced Chk1 phosphorylation in rad4+ cells, Rad3-GFP/LN expression did not increase Chk1 phosphorylation in rad4-Y599R cells, demonstrating a role for the Rad4TopBP1 AAD. Next we established if expression and tethering of the AAD-defective Rad4-Y599R protein to lacO arrays was able to activate the checkpoint (Figure 5B). No induction of Chk1 phosphorylation was observed. Furthermore, while co-expression and tethering Rad3ATR and Rad9, or of Rad3ATR and Rad4TopBP1 resulted in checkpoint activation (Figure 5B), we observed that co-expression of Rad3ATR with Rad4TopBP1-Y599R mutant protein did not result in Chk1 phosphorylation. This data suggests that the AAD-defective mutant protein can act as a dominant negative, at least in this specific situation, preventing the endogenous wild-type Rad4TopBP1 from functioning with the tethered Rad3ATR to activate the checkpoint. It also supports the idea that the Rad4TopBP1 AAD domain is required for the activation of Rad3ATR and not simply recruiting it.
While Rad3ATR kinase activity is essential for Chk1 phosphorylation in response to DNA damage [40], it also depends on the 9-1-1 clamp, the Rad17 clamp loader and the Crb253BP1 mediator. To characterise the dependencies for lacO-dependent checkpoint activation we examined which checkpoint genes were required for Chk1 phosphorylation during Rad3ATR tethering (Figure 5C). Rad3ATR-GFP/LN was expressed in lacO-positive strains deleted for rad1, rad9 (encoding 9-1-1 components), rad17 (clamp loader) and crb2. Each was required for Chk1 phosphorylation. Thus, Rad3ATR tethering is not sufficient for checkpoint activation: the clamp loader, the 9-1-1 clamp complex and the Crb2 mediator are all required and this artificial checkpoint activation system does not bypass the usual requirements. However, Brc1, the proposed MDC1/PTIP ortholog is not required for Chk1 phosphorylation in this system (Figure S1F)
In both S. pombe and S. cerevisiae recruitment of the 53BP1 ortholog (Crb2 and Rad9 respectively) to chromatin in response to IR requires prior phosphorylation of histone H2A [32], [41], [42]. In addition to H2A phosphorylation, recruitment also requires the largely constitutive methylation of a further histone residue, H3K79 in S. cerevisiae or H4K20 in S. pombe. These modifications are effected by distinct methylransferases in the two yeasts: Dot1 methylates H3K79 in S. cerevisiae [43] while Set9 methylates H3K20 in S. pombe [44], [45]. In S. pombe it has been demonstrated that the C-terminal BRCT domains of Crb253BP1 binds directly to γH2A [42] while the Tudor domain binds directly to di-methylated H3K20 [46]. Both interactions are required for Crb253BP1 chromatin association and show an epistatic relationship [45].
Using our Chk1 phosphorylation assay in response to Rad3ATR tethering, we tested two strains harbouring charge reversal mutations of residues within the phospho-acceptor site of the C-terminal Crb253BP1 BRCT domains, crb2-K617E and crb2-K619E (Figure 5C) that disrupt the interaction with γH2A [42]. Chk1 phosphorylation was reduced in both mutants. We next established if checkpoint activation by Rad3ATR tethering was affected in cells containing mutants in the two H2A genes that replace the phosphorylated residue with alanine, hta1-S129A hta2-S128A [32]. Chk1 phosphorylation was not observed in this background (Figure 5D), indicating that the Rad3ATR tethering-dependent and Rad4TopBP1 AAD-dependent checkpoint activation acts in the context of chromatin modification.
Upon activation of the DNA damage checkpoint in S. pombe, Rad3ATR phosphorylates the Rad9 C-terminus on T412 and this is required to recruit Rad4TopBP1 [20]. Recruitment of Rad4TopBP1 allows subsequent recruitment of Crb253BP1 and consequent Chk1 activation [29]. A similar requirement for TopBP1 recruitment via Rad9 C-terminal phosphorylation is also evident in S. cerevisiae and mammalian cells [19], [47], [48]. As expected, expression of Rad3ATR-LacI in cells harbouring a rad9-T412A mutation did not result in Chk1 phosphorylation (Figure 5E) implying that lacO-recruited Rad3ATR must phosphorylate endogenous Rad9 to promote Rad4TopBP1 recruitment to activate the checkpoint.
We reasoned that the requirement for Rad9-T412 phosphorylation during activation by Rad3ATR tethering may solely be to bring the Rad4TopBP1 AAD into proximity of Rad3ATR. In this case, we should be able to bypass the requirement for Rad9-T412 phosphorylation specifically for checkpoint activation by Rad3ATR tethering by recruiting both Rad3ATR and Rad4TopBP1 at the same time. Indeed, Chk1 phosphorylation was restored when we co-expressed Rad3ATR-LacI and Rad4TopBP1-LacI in a rad9-T412A mutant background (Figure 5E). Since checkpoint activation by co-expression of Rad3ATR-LacI and Rad4TopBP1-LacI remains lacO dependent (Figure S1G), these data suggest that Rad4TopBP1 AAD can activate the Rad3ATR-dependent checkpoint cascade in the absence of the recruitment activity of the Rad9 C-terminal tail.
We have shown that the Rad4TopBP1 AAD functions to protect cells from insult during S phase, but is not required for G2 checkpoint activation after IR. Further, we demonstrated that when rad4-Y599R (AAD-defective) mutant cells were synchronised in S phase, phosphorylation of both Chk1 and H2A in response to IR treatment was reduced when compared to rad4+ cells. The increase in Cdc2-Cdc13ClyclinB (CDK) activity as cells progress from S phase into G2 [49] is known to establish conditions conducive to HR by regulating factors required for DNA resection [36], [50], [51]. A consequence of this is that, in response to IR treatment but not in response to UV treatment, ssDNA RPA is predicted to be more prevalent in G2 cells when compared to G1/S phase cells. We thus predicted that reducing resection rates associated with IR treatment in G2 cells would create a dependency for full Chk1 phosphorylation on the Rad4TopBP1 AAD and thus that Chk1 phosphorylation would be reduced in rad4-Y599R strains compared to rad4+ strains in response to an equal dose of IR.
To test this prediction we examined the induction of Chk1 phosphorylation in response to 100 Gy IR in exo1Δ rad4+ and exo1Δ rad4-Y599R cells (Figure 6). First we established that, when exo1 was deleted, Rad11RPA foci were reduced in number, consistent with the expectationt hat resection is decreased in this background (Figure 6A). In contrast to rad4+ cells, where exo1 deletion did not reduce Chk1 phosphorylation levels, Chk1 phosphorylation was reduced to approximately 50% when exo1 was deleted in AAD-defective cells. Previous work in both budding and fission yeasts has indicated that, in the absence of resection, Chk1 phosphorylation can occur through an alternative double strand break end-dependent Tel1ATM pathway, as opposed to the canonical resection and ssDNA/RPA-dependent Rad3ATR pathway [52]. Such a response could potentially mask some aspect of defects seen in the exo1Δ background. Thus, we first confirmed that loss of tel1 alone does not influence Chk1 phosphorylation in our assay (Figure S1H) and then concomitantly deleted Tel1ATM in both rad4+ and rad4-Y599R strains (Figure 6B, 6C). In the tel1Δ exo1Δ background, Chk1 phosphorylation was decreased by approximately 50% for both rad4+ and rad4-Y599R when compared to the exo1Δ alone background. These data are consistent with a general increase in Tel1-dependent checkpoint signalling when resection is reduced by exo1 deletion. In the background of the rad4-Y599R mutation, this is superimposed on a decrease in Rad3ATR-dependent signalling caused by the reduced resection.
A second ATR activation domain has recently been identified in the S. cerevisiae Ddc1Rad9 C-terminal tail [26]. Mutations in this domain define a function in Mec1ATR activation during G1, complementary to the function of the Dpb11TopBP1 AAD in promoting robust checkpoint activation in G2 in this organism [24]. Sequence alignments show that the two key Ddc1Rad9 AAD aromatic residues are conserved in S. pombe as Y271 (equating to W352Sc within the PCNA-like domain) and W348 (equating to W544Sc in the intrinsically disordered C-terminal tail) ([26] and Figure 7A, 7B).We thus created an rad9-AAD mutant by mutating both aromatic residues to alanine.
Analysis of the resulting rad9-AAD strain demonstrates no clear sensitivity to DNA damaging agents that create problems during S phase, including CPT, MMS and UV (Figure 7C) or in G2 to IR. However, some increased sensitivity is evident to CPT and MMS when the Rad4TopBP1 AAD mutant is present in the same strain. We next assayed the ability of rad9-AAD mutants to activate Chk1 in response to either IR or UV treatment. Consistent with the lack of sensitivity, the level of Chk1 phosphorylation after IR was not reduced (Figure 7D), either in rad9-AAD mutant alone or in the rad9-AAD rad4-Y599R double mutant when compared to the rad4-Y599R single. In response to UV treatment, there was again no decrease observed for the single rad9-AAD mutant, but a further and reproducible decrease was seen for the rad9-AAD rad4-Y559R double mutant when compared to rad4-Y599R alone (Figure 7E). Thus, the putative Rad9-AAD domain in S. pombe plays, at most, only a minor role in activating Rad3ATR in response to DNA damage and this is only revealed in the absence of the Rad4TopBP1 AAD.
Understanding the mechanism of ATR activation is an important facet of gaining insight into how cells respond to unwanted DNA structures, itself a key aspect in maintaining genomic integrity. TopBP1 was initially implicated in the ATR-dependent checkpoint in fission yeast and later this was extended to higher eukaryotes [53]. TopBP1 is a multi-BRCT-domain containing protein that acts to scaffold proteins during both the initiation of DNA replication and in response to DNA damage, a function dependent of the phospho-binding ability of the BRCT-domain pairs within TopBP1 [25]. In addition to scaffolding phospho-proteins, TopBP1 was shown to be able to directly activate ATR in Xenopus and human cells through a small domain of TopBP1 which is not part of any BRCT pair [17]. The ATR activating domain is sufficient, both in vitro and in vivo, to activate ATR - although it is not always necessary for ATR activation and the pathway in which this TopBP1 AAD domain functions is yet to be fully understood. It has recently been shown in S. cerevisiae that the AAD of the TopBP1 homolog (Dpb11) is also able to activate the ATR homolg (Mec1). However, the Dpb11 AAD plays a relatively minor role in checkpoint activation which is specific to G2 phase [24]. In S. cerevisiae, a second ATR activation domain within the C-terminal tail of the 9-1-1 subunit, Ddc1Rad9 acts to help activate ATR in G1 and G2 and it is only when the function of this AAD is ablated a role for the Dpb11 AAD becomes apparent [26]. However, loss of both domains does not prevent checkpoint activation entirely, suggesting other AADs or modes of activation. Conversely, in the Xenopus system, ATR activation via the TopBP1 AAD is evident in S phase.
Here we show that, in S. pombe, the activation of the ATR homolog (Rad3) by a Rad4TopBP1 AAD is conserved. We demonstrate that the Rad4TopBP1 AAD makes a contribution to checkpoint activation and that this is specific to G1/S phase and is not evident in G2. Note that log phase S. pombe spend little, if any, time in G1 and thus, while we can arrest cells before the onset of replication with cell cycle mutants, we cannot make a clear physiological distinction between G1 and S phase. We go on to demonstrate that, when DNA resection was limited in G2 by ablation of the Exo1 nuclease, checkpoint activation in response to DNA damage during G2 becomes partially dependent on the Rad4TopBP1 AAD, mimicking what we observed in G1/S cells. This leads us to propose that there is a threshold of ssDNA required for activation of the DNA damage checkpoint and that the Rad4TopBP1 AAD serves to amplify checkpoint signals when ssDNA is limiting.
We next used a genetic system to separate Rad3ATR activation from the production of DNA damage and therefore ssDNA, thus allowing us to assess the pathway of Rad3ATR activation dependent on the Rad4TopBP1 AAD. In this system, specific checkpoint proteins are recruited to a defined chromatin locus through dsDNA:protein binding [37], [39]. Interestingly, recruitment of any one of the three checkpoint proteins (Rad3ATR, Rad4TopBP1 and Rad9) tested was sufficient to generate a checkpoint response and these responses followed the expected dependencies. This suggests that the recruitment of multiple copies of a single checkpoint protein results in the formation of active checkpoint complexes that utilise the endogenous proteins. Using this system, we observed that the ability of the Rad4TopBP1 AAD to activate Rad3ATR is fully dependent on phosphorylation of H2A (γH2A) and requires the ability of Crb2 to bind γH2A. This leads us to conclude that, in the absence of ssDNA, the ATR activation domain of Rad4TopBP1 is particularly important for Rad3ATR activation and acts in a chromatin:protein interaction dependent manner. Taking these data together with the requirement of the Rad4TopBP1 AAD to amplify checkpoint signals in either G1/S or G2 when resection was limited, we propose that Rad4TopBP1 acts to amplify the checkpoint in a chromatin-dependent manner when single-stranded DNA levels are limiting. We can therefore hypothesise that there is a threshold level for the amount of active Rad3ATR required for a full checkpoint response. When ss-DNA is limited, such as in S-phase, the chromatin-dependent Rad4TopBP1 AAD-dependent pathway for Rad3ATR activation becomes important to amplify the levels of activated Rad3ATR to obtain a full checkpoint response (Figure 8).
In addition to analysing the Rad4TopBP1 AAD, we also created a mutant predicted to disable the Rad9 equivalent of the S. cerevisiae Ddc1Rad9 AAD and analysed the effect of this mutant in checkpoint activation. Unlike in S. cerevisiae, we observed no significant effect on DNA damage-induced checkpoint activation either in G1/S phase or G2. Although when combined with a Rad4TopBP1 AAD mutant, an additive effect to S-phase but not G2 DNA damage can be seen. This suggests that the Rad9 AAD acts in a separate but redundant pathway for Rad3ATR activation in G1/S with the Rad4TopBP1 AAD. It appears that, during evolution, the mechanism of activating the ATR pathway has diverged significantly with the roles of different ATR activating domains being of more or less importance in different organisms.
It will be interesting to establish if the ATR activating domain of TopBP1 in metazoan systems is particularly important in the context of low levels of ssDNA and whether its function is dependent on γH2AX, especially as a 53BP1(Crb2) and TopBP1 pathway for checkpoint activation in G1 has been previously reported in the mammalian system [54]. The differences in the dependencies of the specific ATR activators in different cell cycle phases between S. pombe and S. cerevisiae is not surprising as the checkpoint mechanism between these organisms has diverged. For example, in S. cerevisiae, the S phase checkpoint is activated independently of the 9-1-1 complex, whereas in S. pombe and mammalian cells ATR activation appears to be largely - if not entirely - dependent on 9-1-1 loading. Such distinctions are likely a result of evolutionary adaptation to the different cell cycle profiles of the two yeasts and it is interesting to note that significant evolutionary plasticity surrounds the interface between TopBP1 and the checkpoint apparatus. These distinctions will have to be considered when extrapolating mechanistic data from yeast to human systems. None the less, we believe that our findings shed light on the role of TopBP1 AAD in DNA damage responses and offer useful insights into metazoan mechanisms of DNA damage signalling.
Standard S. pombe protocols were carried out as previously described [55]. rad4 and rad9 mutant strains were created using PCR site directed mutagenesis and integrated at their endogenous locus using Cre recombinase-mediated cassette exchange [56] In brief, this system uses a “base strain” which is engineered so that the gene of interest is either replaced with the ura4 marker (i.e. rad9), or in the case of essential genes (i.e. rad4) has the marker integrated immediately after the stop codon. In both cases the gene/marker and loci's promoter region are flanked by loxP and loxM sites. These two variant lox sites are incompatible with each other. The marker (and, for essential genes, the actual gene also) is then replaced by transforming in either the wild type (as a control: rad+) or the various mutated copies on a plasmid. These are flanked by the equivalent loxP and loxM sites and the plasmid expresses Cre recombinase, which results in loxP:loxP and loxM:loxM recombination. For cdc10-M17 synchronisation cells were grown to log phase at the permissive temperature (25°C) and shifted to the restrictive temperature of 36°C for 3.5 hours. Cells were then either irradiated with the indicated dose of gamma irradiation at 36°C and released at 25°C, or directly released at 25°C and irradiated at the given time points after release. cdc25-22 block and release [57] and lactose gradient synchronisation [12] were performed as described previously. For FACS analysis cells were resuspended in 50 mM tri-sodium citrate, 1 mg/ml final concentration RNAseA [Sigma], stained with 5 µg/ml Propidium iodide [Sigma] and analysed on FacsCalibur [Becton Dickinson].
For live cell imaging concentrated culture was mounted onto a 2.5% agar patch in standard YE medim [Microworks] and imaged on a Deltavsion Microscope. Septation index was counted as previously described [12]. lacO::NAT chk1-HA strains were created by inserting the 10 Kb lacO repeats into the PUC19 plasmid containing the NAT marker and homology to ura4. This was integrated into the genomic ura4 locus. The appropriate strains were transformed [58] with pRep41-GFP-LacI-NLS (GFP/LN) into which either rad3 or rad9 had been cloned in frame for N-terminal tagging or rad4 cloned in frame for C-terminal tagging. Transformants were grown and expression of the fusion protein induced by the removal of thiamine. All lacO repeats were checked by Southern hybridisation.
Protein extracts for western were prepared by TCA (trichloro-acetic acid) extraction from 1×108 cells and resuspended in SDS sample buffer [59]. Crude extracts for affinity analysis were prepared by mechanical disruption in liquid nitrogen. Antibodies used: α-HA [Santa cruz] 1∶2500, α-Myc [Santa cruz] 1∶2000, α-GFP [Roche] 1∶2500, α-H2ApS129 [Abcam] 1∶2500 or 1∶1000, α-Tubulin [Sigma] 1∶5000, α-Cdc2 Sc-53 [Santa cruz] 1∶2500. α-Cdc13 [Jacky Hayles] 1∶500. α-Cds1 1∶5000 [35]. The secondary antibodies used were Hrp rabbit α mouse [Dako] 1∶2500 or Hrp swine α Rabbit [Dako]1∶2500. Chk1-HA phosphorylation was quantified as a percentage of total signal minus back ground on a ImageQuant LAS 4000 [GE Healthcare]. Cds1 kinase assay was carried out as described [35].
|
10.1371/journal.pntd.0002972 | Predictive Tools for Severe Dengue Conforming to World Health Organization 2009 Criteria | Dengue causes 50 million infections per year, posing a large disease and economic burden in tropical and subtropical regions. Only a proportion of dengue cases require hospitalization, and predictive tools to triage dengue patients at greater risk of complications may optimize usage of limited healthcare resources. For severe dengue (SD), proposed by the World Health Organization (WHO) 2009 dengue guidelines, predictive tools are lacking.
We undertook a retrospective study of adult dengue patients in Tan Tock Seng Hospital, Singapore, from 2006 to 2008. Demographic, clinical and laboratory variables at presentation from dengue polymerase chain reaction-positive and serology-positive patients were used to predict the development of SD after hospitalization using generalized linear models (GLMs).
Predictive tools compatible with well-resourced and resource-limited settings – not requiring laboratory measurements – performed acceptably with optimism-corrected specificities of 29% and 27% respectively for 90% sensitivity. Higher risk of severe dengue (SD) was associated with female gender, lower than normal hematocrit level, abdominal distension, vomiting and fever on admission. Lower risk of SD was associated with more years of age (in a cohort with an interquartile range of 27–47 years of age), leucopenia and fever duration on admission. Among the warning signs proposed by WHO 2009, we found support for abdominal pain or tenderness and vomiting as predictors of combined forms of SD.
The application of these predictive tools in the clinical setting may reduce unnecessary admissions by 19% allowing the allocation of scarce public health resources to patients according to the severity of outcomes.
| Dengue is a mosquito-borne tropical disease that poses a large economic and health burden to 40% of the world's population. Some dengue patients can evolve into a more severe form of the illness (severe dengue) and ideally should be hospitalized when this occurs. However, while only a small proportion of patients have severe dengue, many other patients are unnecessarily hospitalized. Being able to identify those patients at the highest risk of developing severe dengue can be very useful to reduce costs to hospitals and patients. In this study, we develop predictive tools to identify those patients that should be hospitalized. We find that some common features such as age, gender, fever duration, fever on admission, vomiting or abdominal distension can be good predictors of severe dengue. This is important because these features are easy to identify even in resource-limited settings where laboratory tests are not widely available.
| After successful eradication programs of the dengue vector Aedes mosquitoes in Latin America in the 1940s, both disease and vector have resurged and disseminated globally since the late 1970s [1], posing substantial disease and economic burdens on tropical and subtropical regions [2], [3]. It is estimated that currently about 40% of the world's population live at risk of dengue infection and 50–100 million infections occur annually [4], [5].
In Singapore, an equatorial city state in South East Asia, intensive vector control programs initiated in the 1970s led to a substantial decline in dengue notifications [6]. Possibly as a result of reduced herd immunity [7], however, cyclical epidemics have occurred since the 1990s, leading to an estimated average annual economic impacts of US $115 million [8]. Dengue epidemics in Singapore lead to a strong demand for hospital beds and manpower, diverting resources for elective surgery and non-emergency admissions [9] and inflicting large public and private hospitalization costs (e.g. $50 US million in 2005 alone [8]). Despite this, only a small fraction of dengue infections in Singapore develop complications requiring hospitalization – for instance 6% and 21% of patients admitted for dengue into Tan Tock Seng hospital, Singapore, developed dengue hemorrhagic fever [DHF] in 2004 and 2007 respectively [10].
Under the widely adopted and longstanding World Health Organization (WHO) 1997 classification system, some dengue infections with complications were classified as dengue fever (DF), while not all patients classified as having DHF actually required hospitalization. Alternative classification systems were shown to be more sensitive than the WHO 1997 classification system in predicting severity of dengue in Indonesian children [11], showing the need to develop improved classification criteria. In addition, the classification into DF and DHF has been noted to lead to the false perception of low disease burden in the Americas [12]. The need to classify dengue cases according to their clinical severity [11]–[13] led the WHO to issue new classification criteria in 2009 for probable dengue with and without warning signs, and severe dengue (SD) [14]. The new classification criteria has been shown to facilitate dengue case management and surveillance [15] through better classification of severity [16], [17]. This revision makes it urgent to develop predictive tools conforming to the WHO 2009 dengue classification criteria to distinguish patients not likely to develop SD, for whom outpatient management may be appropriate, from those at greatest risk of SD and who require close monitoring and early therapeutic interventions in hospital.
Predictive tools to triage dengue patients into those that are most and least likely to develop complications allow better allocation of limited healthcare resources. By identifying risk factors at presentation that are predictive of complications after hospitalization, admissions may be reduced, alleviating pressure on the healthcare system and avoiding unnecessary costs to patients. Statistical procedures have been used to distinguish dengue severity predictors [18], [19] and predictive tools using logistic regression or classification and regression tree models have been developed to triage patients between dengue fever (DF) and other febrile illnesses [20]–[22], between DF and DHF patients [9], [23], between DF and dengue shock syndrome (DSS) or other severe outcomes [22], [24], [25], and, recently, between DF and subtypes of SD including plasma leakage [25] or internal bleeding and plasma leakage [26].
This study addresses the need to identify predictors for the new WHO 2009 dengue classification system by reporting the results of a retrospective analysis of adult dengue cases in Tan Tock Seng hospital, responsible for the treatment of approximately 40% of notified cases in Singapore [27]. The demographic, clinical and laboratory characteristics of confirmed adult dengue patients at presentation to hospital were employed to develop predictive tools of SD after hospital admission.
We conducted a retrospective cohort study in which we extracted from chart review demographic, epidemiological, co-morbidity, serial clinical and laboratory, radiological, treatment and outcome data from all confirmed adult dengue patients treated using a standardized dengue care path in the Department of Infectious Diseases, Tan Tock Seng Hospital, from 2006 to 2008. Prior to admission, patients were mainly referred from primary care centers where the standard is to monitor with full blood count, give symptomatic medication for fever, pain, nausea or diarrhea, and tell the patients to rest and drink copious fluid. Two groups of laboratory-confirmed dengue cases were considered: (i) had positive dengue polymerase chain reaction (PCR) (definite dengue) [28]; (ii) were positive to dengue immunoglobulin-M (IgM) or immunoglobulin-G (IgG) (probable dengue) (Dengue Duo IgM & IgG Rapid Strip, Panbio Diagnostic, Queensland, Australia). Patients who were diagnosed with severe dengue at the time of hospital presentation were excluded from further analysis; patients without SD but with warning signs were included (Figure 1). SD cases fulfill at least one of three criteria: (1) severe plasma leakage: associated with shock (tachycardia >100 or narrow pulse pressure <20 or blood pressure <90 mmHg) or respiratory distress, (2) severe hemorrhage: gastrointestinal tract bleeding such as haemetemesis or melaena or bleeding per rectum, pack cell transfusion or blood transfusion requirement or menorrhagia (3) severe organ impairment: defined as aspartate or alanine transaminase ≥1000units/l, acute renal impairment (creatinine ×2 upper limit of normal for age and sex (based on modification of diet in renal disease equation for Glomerular filtration rate = 75 ml/min) or baseline estimated from minimum creatinine recorded), encephalopathy and myocarditis [14]. In addition, the WHO proposed warning signs suggestive of high risk for SD were recorded: abdominal pain or tenderness, persistent vomiting (vomiting during two or more consecutive days), clinical fluid accumulation, mucosal bleed, lethargy or restlessness, hepatomegaly and concurrent increase in hematocrit and rapid decrease in platelet count (hematocrit change > = 20% concurrent with platelet <50×109/l occurring in the same day) [14].
Variables presenting missing data in a proportion greater than 10% were removed from the analysis. Those variables with fewer missing data had their missing values replaced by an imputed value being the mean over all individuals in the study. For dichotomous variables, missing values were imputed by noes, the assumption being that if the sign or symptom were present, it would have been noted by the attending clinician. Variables that were not clinically relevant to severe dengue (e.g. runny nose, sore throat) or that conveyed similar information or information that was a linear combination of other variables were further removed from the analysis. Variables corresponded to values at presentation with the exception of hematocrit change > = 20% and drop in platelet count was obtained comparing the value at presentation with values during the course of the disease. In total, 66 explanatory variables extracted from chart review at presentation were employed and classified in groups: demographic and epidemiological characteristics; co-morbidities; symptoms and signs at presentation; laboratory results at presentation; and miscellaneous variables (see the Supporting Information Table S1 for the list of variables considered). Co-morbidities were scored for: congestive heart failure, cerebrovascular disease, chronic pulmonary disease, liver disease, diabetes, hemiplegia, renal disease, malignancy and AIDS. The scores were combined into the Charlson score and two variables indicating whether the individual had any comorbidity and whether individuals had Charlson score greater than three were respectively created. The large number of variables allowed us to minimize potential confounders while retaining variables that were clinically relevant. Prior to the analysis, we further tested the presence of collinear variables using variance inflation factors. Variables with a variance inflation factor greater than 3 were removed from the analysis stepwise using the package car in the R environment [29].
Predictive tools for SD development after presentation were constructed using logistic regression or its equivalent generalized linear models (GLM) with a logit link function and binomial errors [30]. Logistic regression or the GLMs used relate a group of explanatory variables to the log of the odds of a patient developing severe dengue versus not developing it. The GLMs were constructed and simplified using a stepwise procedure that started with the full model and removed one variable at a time using a chi-square test [31], [32].
The receiver operating characteristic (ROC) curve, the area under the ROC curve, sensitivity and specificity of the GLM models were estimated. Because our aim was to minimize the number of SD cases triaged for outpatient care, while allowing practitioners to decide on the tradeoff between sensitivity and specificity, we reported the specificity of the models given high sensitivity rates of 90%, 95% and 100% (see Figure 2, for a comparison of the ROC curves). Given the large number of potential predictors and to avoid the risk of overfitting or finding relationships by chance alone, we corrected the estimates of sensitivity and specificity for optimism—decrease in model performance in new patients compared to those in the sample studied [33]. The bootstrapping method was used to create samples with replacements from the dataset. A new model repeating all the stepwise model simplification process was constructed. The performance of the model on the prediction of the bootstrapped sample and the original sample was estimated. The optimism estimate was the difference between both performances and deducted to the original estimates of sensitivity and specificity [33]. 100 simulations were conducted per model to obtain stable estimates of optimism. In addition, the models were further validated using 5-fold validation by which the data are partitioned in 5 subsets, 4 subsets – the training dataset – are used to fit the GLM model and then employed to predict the remaining subset of the data – validation dataset. The procedure is then repeated changing the subset used as the validation dataset. The 5-fold validation process was repeated 1000 times and the mean of the root mean squared prediction error (RMSPE) was estimated for each model. The estimation of RMSPE was also subjected to optimism correction. Lower RMSPE indicated a better performance in the prediction of new cases and robustness of the model to potential associations found by chance alone. The R statistical environment was used for all analyses [34].
This study received ethical approval from the Domain Specific Review Board, National Healthcare Group, Singapore (DSRB-E/08/567) and all data analyzed were anonymized.
Dengue cases were analyzed with and without laboratory information so that appropriate tools could be developed for both well-resourced and resource-limited settings. GLMs were fitted to six combinations of data and variables:
Our cohort comprised 596 PCR-positive confirmed dengue cases of whom 96 [16.1%] developed SD (Figure 1). After hospital admission, severe plasma leakage developed in 59 (10%), severe hemorrhage in 37 (6%) and severe organ impairment in 5 (0.7%). Severe plasma leakage and severe hemorrhage occurred simultaneously only in 3 cases, showing little overlap between these two forms of SD. Severe organ impairment, on the other hand, occurred jointly with severe bleeding or severe plasma leakage in 4 cases out of 5.
The median age was 37 years (interquartile range [IQR], 27–43 years) and 450 individuals (75%) were male. Any co-morbidity was noted in 20% and only 5 patients had Charlson's co-morbidity score ≥3. Warning signs were present on admission to hospital in 65% of the patients. The median duration of illness prior to hospital presentation was 4.2 days (IQR, 3–5 days). Intravenous fluid therapy was administered to 91%, platelet transfusion to 9% of the patients and blood transfusion to two patients. Two patients were admitted to intensive care and one patient died. The median length of hospitalization was 5 days (IQR, 4–6 days).
For descriptive purposes we performed univariate tests, however, these results have no predictive power and should not be over interpreted because they do not control for the rest of variables and should not be used to triage patients. We also present results of the final outcomes of hospitalization (e.g. length of hospitalization, total platelet transfusion) that cannot be used to predict SD. Univariate analysis indicated that, at presentation, the cases that subsequently developed SD were more likely to be female and to present with shorter duration of fever and more likely to be febrile at presentation (Table 1). Fewer had leukopenia, and more had hypoproteinemia (Table 1). These patients received significantly more platelet transfusion during the course of the disease and stayed longer in hospital.
We fitted GLMs to all types of SD combined as a single explanatory variable. The GLMs obtained optimism-corrected 29% specificity for 90% sensitivity (21% specificity for 95% sensitivity) for well-resourced settings with clinical and laboratory variables (Table 2, Figure 2).
The GLM to predict SD in well-resourced settings contained 8 explanatory variables (Table 3). Higher risk of SD development was associated with female gender, fever on admission, abdominal distension and vomiting (Table 3). Lower risk of SD development was associated with more years or age, leucopenia and normal levels (36.1–44.3% for females and 40.7–50.3% for males) of the minimum reading for hematocrit (Table 3). For instance, the mean of minimum hematocrit levels in females that did and did not develop SD was 35.5 and 36.6% and for males it was 39.4 and 41.6% respectively.
Because laboratory tests are not available in some resource-limited settings, we constructed GLMs that excluded laboratory variables. The resulting drop in the specificity of the predictive tools was small and they even obtained lower optimism-corrected prediction errors (RMSPE of 5.19 compared to 5.34 in the case of GLMs for resourced settings, Table 2). This could indicate that this model had fewer tendencies to overfitting than a model using laboratory variables and made them comparable to tools reliant on laboratory information. The GLMs obtained 27% specificity for 90% sensitivity (15% specificity for 95% sensitivity. Table 2, Figure 2, implying a reduction of 8–12% in specificity).
The GLM to predict SD in resource-limited settings contained 6 explanatory variables (Table 4). Similar to the GLM fitted with laboratory data, higher risk of SD development was associated with abdominal distension, female gender, vomiting and fever on admission; lower risk was associated to older age and fever duration on admission (Table 4).
When fitted to data combining PCR and serology confirmed cases, the GLMs obtained increases in specificity compared to GLMs fitted only to PCR data: e.g. 30% optimism corrected specificity for 90% sensitivity for resourced and resource-limited settings (Table 2, Figure 2, increase of 1–3% in specificity with respect to resourced and resource-limited settings using only PCR confirmed data). These models presented an optimism-corrected predictive error lower than the models fitted to PCR data alone (RMSPE of 4.03–4.25 versus 5.19–5.34).
The GLMs to predict SD in resourced and resource-limited settings contained 9 and 8 explanatory variables respectively (Tables S2 and S3). In both models, higher risk of SD development was associated with abdominal distension, female gender, breathlessness, vomiting and fever on admission (Table 4). More years of age and longer fever duration on admission were associated with lower risk of SD development.
We fitted GLMs using severe plasma leakage and severe hemorrhage as dependent variables. We could not find robust GLMs for severe organ impairment because of the small number of cases (5 cases). Fitting to severe hemorrhage led to higher specificity than fitting to the development of SD in general (Tables 2). Despite their increase in specificity, GLMs fitted to single forms of SD would not be able to reduce hospital admissions substantially further: cases with lower risk of severe plasma leakage may be at risk of severe hemorrhage and vice-versa, impairing the identification of cases with overall low risk of SD that could be managed as outpatients. These models presented however the highest optimism-corrected predictive errors compared to those predicting SD in general, indicating low robustness in the prediction of new cases (Table 2).
Fitting to specific forms of SD was nonetheless useful for comparison with other studies and to increase our understanding of the development of specific forms of SD. We found 6 predictors of severe hemorrhage (Table S4). Less years of age and female gender were common predictors with development of SD in general. High serum urea levels and hemoglobin count were new predictors of severe hemorrhage. Concurrent increase in hematocrit and rapid decrease in platelet count is a warning sign proposed by WHO for SD which was found to be good predictor of severe hemorrhage (Table S4). The prediction of severe plasma leakage presented 7 predictors. Shorter fever duration, vomiting, abdominal distension and fever on admission were predictors shared with any form of SD (Table 3 compared to S5).
Several warning signs proposed by WHO 2009 for SD were identified as significant predictors. Abdominal distension – which was a good predictor of SD development – and could be considered related to abdominal pain and vomiting that was significant in models combining PCR and serology data. Concurrent increase in hematocrit and rapid decrease in platelet were found to be statistically significant in predicting severe hemorrhage and severe plasma leakage [14]. Vomiting and abdominal distension were good predictors of severe plasma leakage (Table S5).
We used a mean length of hospitalization of 5 days from our data, a mean cost per hospitalized case per day of US 2010 $431 [35], a mean cost per ambulatory visit of $62.1 and 4.33 visits per episode [8]. We estimated that the savings per case managed as outpatient in a well-resourced setting like Singapore would be $1886. Using the optimism-corrected sensitivity and specificity of the GLM in a well-resourced setting fitted to PRC confirmed data, reduction in admissions of 19% could be obtained. In the case of Tan Tock Seng hospital in Singapore, incorporating the diagnostic testing costs also to all probable dengue cases from 2006 to 2008 (1920 cases), a total cost to patients of US $0.69 million could be avoided ($0.23 million per year). Extrapolating to national level using the proportion of admissions at Tan Tock Seng hospital, US $0.58 million could be averted annually in Singapore.
The new disease severity classification for dengue issued by the WHO in 2009 has necessitated a reassessment of clinical care, management and diagnosis of dengue in many parts of the world. We developed predictive tools to optimize the triage of adult dengue patients in Singapore according to the new guidelines. In the Singapore context, applying these tools would reduce unnecessary admissions by 19%, alleviating demand on scarce hospital beds and healthcare costs to patients. However, these results are based on adopting models with a sensitivity rate of 90%, i.e. 10% of the SD patients would not be hospitalized at presentation. For sensitivities of 95%, reductions of 16% could be instead obtained.
It is difficult to compare our current results with studies dealing with the prediction of DHF and DSS according to WHO 1997 classification, given the different WHO 2009 classification that we employ. Good predictors of DHF were not necessarily good predictors of severe dengue. Carlos et al. [19] found a relationship between DHF and restlessness, nose bleeding, abdominal pain and low platelet count but, with the exception of abdominal pain, related to abdominal distension, we did not find these as significant predictors of severe dengue. Lee et al. [23] found that clinical bleeding, serum urea, and serum total protein were good predictors of DHF. Our results concur with respect to serum urea (only in the case of prediction of severe hemorrhage).
It was noteworthy that some of the warning signs (abdominal distension, vomiting) proposed by WHO were prognostic of SD when predicted jointly and for specific forms of SD. The apparent low predictive power of other WHO warning signs might respond to the application of intravenous fluid. This application might have eventually prevented the development of SD, particularly shock due to severe plasma leakage and could confound the results. The predictive models are hence trained to predict SD under current fluid therapy conditions and might underperform in resource-limited settings where fluid therapy might not be so prevalent. Another potential reason for the low performance of warning signs is that warning signs tend to appear on the day of defervescence. Given the objective of creating triaging tools, the data were collected at presentation and thus warning signs had not yet been observed in some cases. We further note that two warning signs with slightly different definitions to those provided by the WHO were employed: vomiting during two or more consecutive days and hematocrit change greater than 20%. Using the flexibility to exercise clinical judgments of the WHO guidelines, these warning signs were chosen because they have been shown to be good predictors in the local context [36], [37].
Comparison with other studies focusing on dengue severity prediction, on the other hand, showed some common features. Comparison was not straightforward because several of the studies, contrasting with ours, were based on univariate analyses or specific types of severity only. In relation to abdominal distension as a strong predictor, SD was associated with abdominal pain [22], [25], [26]. We identified below normal serum hematocrit as good predictor of SD. This has been found in previous studies in Thai children [24] and agrees with the prediction of internal bleeding [26] and severe outcomes [22] respectively. Further research would however be needed to unveil the actual mechanisms that could explain this result.
In predicting plasma leakage alone, we found that frequent vomiting was associated with plasma leakage, which agrees with previous studies [24], [26], [38], [39]. In contrast with our results, plasma leakage has been associated with older patients and male gender [25] but we could not find such association in our severe plasma leakage specific model. We found however that SD in general was associated with female gender and less years of age. The differences between studies might be due to the further inclusion of laboratory variables and by controlling for the presence of any co-morbidity and by using the Charlson score. This could avoid confounding effects with age in our study. Another potential reason for differences is that our cohort is relatively young (interquartile age of 27–43 years old and only 7% of individuals above 60 years old). Hence the lower risk of SD effect with an extra year of age that we find is mostly representative of young adults and middle age individuals. A precautionary approach should be taken when using the model to predict SD for patients above, for instance, 60 years old as this would involve extrapolating beyond an age for which the cohort was not as representative. We thus contend that further investigation on the relationship between age and SD outcome are thus necessary, preferably with cohorts mainly composed of old age individuals.
Interestingly, the variables necessary to predict any form of SD without laboratory variables were easier to measure than those aimed at predicting specific forms of SD such as plasma leakage. These results might derive from the heterogeneity of the different SD types. The predictive tools identify those less exclusive variables that are common to severe plasma leakage, severe bleeding and severe organ impairment. Among these variables, fever on admission, fever duration and vomiting are easy to measure and can lead to rapid and cost-effective triaging of patients.
Our study has some limitations. (i) Due to its retrospective nature, it is difficult to ascertain the homogeneity in the level of training of doctors and nurses with regards to the identification of clinical variables that could be subjective (e.g. persistent vomiting, abdominal pain) and how these could be applied to other settings. We have attempted to describe the clinical variables used to minimize ambiguities but different training in other hospitals could lead to variations in their identification. (ii) The GLMs were developed in an overwhelmingly adult cohort, and their performance in pediatric patients more common to other countries in the region needs to be validated. (iii) Not including other febrile illnesses in our study prevented the models from developing discriminatory power against them. This is both one of the strengths of the analysis, because we obtain highly fine-tuned predictive tools for SD, and a caveat, because our predictive tools may need to be supported by dengue diagnostic tests, which might not be easily available in resource-limited settings. Alternatively, published predictive tools to discriminate between DF and other febrile illnesses [20]–[22] can be used as a prior filter to the application of our predictive tools. (iv) Although the dengue epidemics in Singapore in 2006 were predominantly caused by dengue serotype 1, the epidemics in 2007 and 2008 were predominantly caused by dengue serotype 2 [10]. We controlled for serotype via a proxy (year), which was not statistically significant, but further studies in other settings would be needed to verify the validity of our predictive tools for epidemics driven by dengue serotypes 3 and 4. (v) Although serological tests for dengue are less accurate than PCR [28], for comparison we decided to include the results both with PCR and serology tests and PCR combined. Choosing only PCR confirmed cases was initially not preferred as there could be a risk that PCR confirmed cases favored patients that present earlier in the course of the disease and were febrile. If the bias had existed, we would expect fever on admission to be a good predictor in the case of only PCR confirmed cases and not in PCR and serology confirmed cases. This bias appears not relevant and consequently our results on fever on admission are also robust given that fever on admission is significant for both types of datasets (PCR and PCR and serology combined). We note that models fitted to both datasets agreed in a large number of predictors such as age, gender, fever duration, fever on admission , vomiting or abdominal distension. (vi) Our analysis does not control for the exact day of illness. Future research could focus on the prediction of SD one day ahead using the time series of the values of the explanatory variables through time. Although such models would not allow for patient triaging at admission, they will be useful to understand the development of SD and allow for the preparation of treatment at the hospital setting.
The easily implementable predictive tools developed in this study do not require laboratory-based variables and may be beneficial in resource-limited settings, where pressure on scarce health resources is especially high. In addition, no considerable reductions in predictive power were observed in resource-limited tools with respect to resourced tools. Their application will allow redistribution of resources to those patients and conditions that need them more urgently without incurring extra screening costs. In conclusion, we have shown a combination of clinical and laboratory data performed well in identifying potential patients with severe dengue.
|
10.1371/journal.pcbi.0040030 | Stimulus Design for Model Selection and Validation in Cell Signaling | Mechanism-based chemical kinetic models are increasingly being used to describe biological signaling. Such models serve to encapsulate current understanding of pathways and to enable insight into complex biological processes. One challenge in model development is that, with limited experimental data, multiple models can be consistent with known mechanisms and existing data. Here, we address the problem of model ambiguity by providing a method for designing dynamic stimuli that, in stimulus–response experiments, distinguish among parameterized models with different topologies, i.e., reaction mechanisms, in which only some of the species can be measured. We develop the approach by presenting two formulations of a model-based controller that is used to design the dynamic stimulus. In both formulations, an input signal is designed for each candidate model and parameterization so as to drive the model outputs through a target trajectory. The quality of a model is then assessed by the ability of the corresponding controller, informed by that model, to drive the experimental system. We evaluated our method on models of antibody–ligand binding, mitogen-activated protein kinase (MAPK) phosphorylation and de-phosphorylation, and larger models of the epidermal growth factor receptor (EGFR) pathway. For each of these systems, the controller informed by the correct model is the most successful at designing a stimulus to produce the desired behavior. Using these stimuli we were able to distinguish between models with subtle mechanistic differences or where input and outputs were multiple reactions removed from the model differences. An advantage of this method of model discrimination is that it does not require novel reagents, or altered measurement techniques; the only change to the experiment is the time course of stimulation. Taken together, these results provide a strong basis for using designed input stimuli as a tool for the development of cell signaling models.
| A major focus of systems biology is the development of mechanism-based models of cell signaling pathways. These models hold the promise of encapsulating our understanding of complex biological processes while also predicting new behavior. However, as these models become more complex, it can be difficult to distinguish between model alternatives. One means of improved model discrimination involves making measurements of additional components in the biological system to provide more detailed data. Here we present an alternative, which is to apply a time-varying input while monitoring the same network components. This new method was able to discriminate among models with subtle mechanistic differences. A particular advantage is that for many cases, time-varying input stimulation is fairly easy to apply experimentally, whereas measuring additional network components can involve the creation of new reagents or measurement assays. Thus, we believe that the application of time-varying input stimulation will become a powerful tool in the field of systems biology as the community places increased emphasis on the development of quantitative, mechanistic, and predictive models of biological network behavior.
| One goal of systems biology is to develop detailed models of complex biological systems that quantitatively capture known mechanisms and behaviors, and also make useful predictions. Such models serve as a basis for understanding, for the design of experiments, and for the development of clinical intervention. In support of this goal, there has been a strong push to build mechanistically correct kinetic models, often based on systems of ordinary differential equations (ODEs), that are capable of recapitulating the dynamic behavior of a signaling network. These models hold the promise of connecting biological and medical research to a class of computational analysis and design tools that could revolutionize how we understand biological processes and develop clinical therapies [1,2].
One type of experiment for model validation involves stimulating a system with a step change in the input (typically by adding a high concentration of ligand) and then measuring the change of network readouts (the concentrations or activities of various downstream species) as a function of time. Candidate models are fit to the data and the best model is selected based on criteria such as the quality of the fit, the simplicity of the model, and other factors. While it is tempting to select a simple model consistent with the known biochemical mechanisms that fits all available data, future experimentation may prove this choice incorrect. Rather, it may be preferable to collect “all” models consistent with known mechanisms and data, and to design follow-on experiments capable of distinguishing among the model candidates. In support of this less-biased approach, here we develop an approach for designing these follow-on experiments using dynamic stimuli.
While the step-response experiment is attractive for its ease of implementation, dynamic stimuli have the potential to uncover more subtle system dynamics and to improve model selection in the cases where step-response experiments are not sufficiently discriminating. One example that illustrates the use of a dynamic stimulus to distinguish between two models is the work by Smith-Gill and co-workers on the detailed mechanism of antibody–antigen binding [3]. Initial step-response experiments were compatible with either a one-step or two-step binding mechanism, in which the ligand and antibody first come together in a loose encounter complex before forming a fully bound complex. To resolve this ambiguity, the authors applied a series of rectangular pulses of ligand concentration to their system. The resulting binding curves produced by this dynamic stimulus were inconsistent with the one-step model but were consistent with a two-step model and suggested the existence of an encounter complex, even though such a complex could not be measured directly by the assay.
These results show that time varying inputs have the potential to distinguish closely related models of biochemical systems. For the relatively simple antibody–antigen system, an appropriate dynamic input was deduced intuitively. However this sort of intuitive design is difficult, especially in the case of more complex cell signaling pathway models, which may be described by hundreds or thousands of differential equations. An automated approach that could design experiments to test these complex systems has the potential to expand the scope of model selection experiments.
Previous work in designing dynamic stimuli for the purpose of model discrimination in systems biology has focused on choosing input trajectories that maximize the expected difference in the output trajectories of competing models [4–10]. In addition to model discrimination, a rich literature exists on experimental design in systems biology for the purpose of estimating model parameters [2,11–13]. These optimization approaches for model discrimination have been applied to small biological systems, but the nonlinearity of the models combined with the presence of many local minima has thus far limited their application [8].
There is a need to extend these methods to design experiments that may not be optimal but are capable of discriminating between large pathway models. Instead of trying to design an input signal that maximizes the predicted difference between two model readouts, we recast the problem as a control problem (Figure 1). We choose a target trajectory, and then challenge a model-based controller to drive the system to follow the target trajectory. The extent to which the controller based upon a given model is able to drive the physical system is a measure of the fitness of that model.
We demonstrate our methodology by applying it to the epidermal growth factor receptor (EGFR) pathway. This pathway has been extensively studied and modeled [14–18]. EGFR and its family members (Erb2, Erb3, and Erb4) are known to mediate cell–cell interactions in organogenesis and adult tissues [19]. Overexpression of EGFR family members is a marker of certain types of cancer, including head, neck, breast, bladder, and kidney [20]. Because of their clinical importance, the EGFRs themselves, as well as various downstream proteins, are targets of therapeutic intervention [21,22]. Despite clinical interest in the EGFR pathway and over 40 y of intense study, there is still much about the pathway that is not known. For example, in three recent studies [23–25], a number of proteins that changed phosphorylation state in response to EGF stimulation were found that were not previously known to be part of the pathway; in addition, many of the known pathway proteins are not part of any computational model [26].
The ordinary differential equation model of Hornberg et al. is a widely used mechanistic model of EGFR signaling [16]. This model is a refinement of earlier models of the pathway [17,18,27]. It describes signal transduction initiated at the cell surface by EGF binding to EGFR, leading eventually to the dual phosphorylation of ERK as the most downstream outcome, which then participates in a negative feedback to the top of the pathway. The elementary molecular processes modeled include bimolecular association and dissociation, phosphorylation and de-phosphorylation, synthesis and degradation, as well as endocytosis and trafficking all described with mass-action kinetics. The model contains 103 chemical species, 148 reactions, 97 independent reaction rates, and 103 initial conditions.
We applied our computational methods initially to a small portion of the EGFR model for development and demonstration purposes, and then to the full model. In both cases, we formulated a set of closely related models that exhibit similar step-response behavior. We built a controller capable of controlling each candidate model and asked the controller to drive the system output (doubly phosphorylated ERK) to a predetermined value. Finally, by applying these designed inputs based on the reference and perturbed models, we showed that it is possible to discriminate between the various model alternatives.
In this work, we consider mass-action kinetic models consisting of zeroth-, first-, and second-order reactions described by ordinary differential equations. In the equations below, k signifies a rate constant; A, B, and C represent species or concentrations of species, depending on the context; and ∅︀ is the empty set or nothing.
Zeroth-order reaction:
First-order reaction:
Second-order reaction:
Large systems of reactions of this form can be represented compactly using Equation 4.
The state vector x describes the chemical species concentrations that are free to evolve in time according to the kinetics of the system. The input vector u represents the chemical species concentrations controlled by the experimenter. Matrices A1 and B1 represent first-order reactions, matrices A2 and B2 represent second-order reactions, and k represents constitutive (zeroth-order) reactions. The symbol ⊗ denotes the Kronecker product (also known as the matrix direct product) [28]. For vectors, this operator generates a vector of all quadratic products.
The output of the model y is a linear combination of the state variables represented by the matrix C.
A controller was developed to solve for the input signal u(t) that best achieves a particular objective in the output. We formulate this objective as a cost function G(u) that measures the distance between the model output and the desired output.
Here, G(u) is the sum of squares error between y(u,t), the model output for a given input u(t), and ydesign(t), the target output the controller is trying to match. T is the length of time of the experiment. The control problem is then to find an input function u(t) that minimizes G(u).
Equation 6 depends on models of the form of Equation 4, which are nonlinear and potentially high order. This prevents us from solving the minimization problem directly. To address this issue, we implement two different approximations. The first is based on controlling a model formed from successive linearizations of Equation 4 (henceforth referred to as the tangent linear controller), and the second is based on a local search of the input space (henceforth referred to as the dynamic optimization controller) [29].
A first-order approximation to Equation 4 at time t was computed by taking the Taylor series expansion about the current value of the state and input vectors (xt and ut).
Equation 7 is a linear differential equation with state variable Δx and time varying forcing term Δu, which has both numerical and analytical solutions. However, this approximation would tend to diverge from the solution to Equation 4 with increasing Δt, the time beyond the linearization point t, and (Δx, Δu), the distance from the linearization point (xt,ut). To mitigate this problem the true system (Equation 4) was propagated, and successive linearizations were applied to improve the controller performance. Effectively, the linearization point is allowed to slide along with the exact simulation.
Operationally, each time step was solved in three stages. First, the current state of the nonlinear simulation was used to derive a linear approximation about the current time point. Second, the linear system was solved to get the best input Δu. The linear system was solved numerically by discretizing the input as a series of scaled and shifted boxcar functions [30] of width τ. Numerical integration with the MATLAB routine ode15s [31] was used to compute the system response to a unit boxcar input. The output of a linear time invariant system can be expressed as a linear combination of scaled and shifted impulse response functions. Thus, solving for the input was achieved by computing the weights to apply to the input pulses that gave the optimal output. This was solved as a linear system of equations with box constraints on the input to limit the maximum and minimum concentration using the MATLAB routine lsqlin. Third, the computed input signal was applied to the full nonlinear system for a short time step τ. The process was then repeated for the next time interval. Effectively, each step the algorithm solves for an input signal Δu that is piecewise constant. The width of the intervals τ as well as the number of intervals is a parameter of the optimization and should be chosen based on the accuracy of the linear system.
In this controller formulation, rather than exactly solving the tangent linear system, we solved the full nonlinear problem iteratively using a gradient optimization method. Application of this method requires computation of the sensitivities of the least squares objective function (Equation 6) with respect to the input parameterization p. An efficient way to compute this quantity is to first solve for the adjoint sensitivities λ [29]. For the dynamical system (Equation 4) and the objective function (Equation 6), the adjoint equations are given by Equation 8.
Here, λ* indicates the conjugate transpose. We use piecewise linear input functions described by parameters pi, which are the input function value at Ti, u(Ti); u(t) is then linearly interpolated between the control points at Ti. For these piecewise linear input signals the ith component of the gradient
is given by:
The adjoint equations were solved in MATLAB using ode15s [31] and the optimization was implemented using fmincon configured to use Quasi-Newton [32] with BFGS [33,34] in the MATLAB Optimization Toolbox Version 3.1.1.
Thus far the input signals have been unconstrained, except by the choice of the discretization. However, in practice it may be desirable to restrict the space of input signals to those that could be feasibly achieved by a given experimental setup. For example, in many experimental setups it is easy to add material but difficult to take material away. Likewise, there may be a maximum and minimum concentration for the input signals, or a maximum rate of change for the input signal. We implemented these experimental constraints as linear inequality constraints of the form of Equation 10.
The matrix A and the vector b are passed as arguments to lsqlin in the case of the tangent linear controller, or to fmincon in the case of the dynamic optimization controller. An example of a linear constraint that might be applied is that the input increase monotonically. In this case, A and b are given by Equation 11.
We based our model of EGFR signaling on that of Hornberg et al. [16], which itself is a refinement of earlier work [17,18,27]. The model contains 103 chemical species, and 148 elementary reactions; these reactions are of the type given by Equations 1, 2, and 3 and may be reversible. The model is parameterized by 97 distinct reaction rate values and 103 initial conditions. The details of this model are given in Dataset S3.
Here we also introduced a modified model of EGFR signaling, which contained six additional production/degradation reactions of the form of Equation 12, where X is one of {GAP, GRB2, SOS, RAS-GDP, SHC, or GRB2-SOS}.
The degradation rate kdeg was set such that the steady-state value of the species was the same as the steady-state value in the unmodified model computed using Equation 13.
In addition to the protein synthesis and degradation reactions, a GAP-catalyzed turnover of RAS-GTP was implemented.
The rate constants (kon, koff, and kcat) are 5 × 10−7 cell molecules−1 s−1, 0.4 s−1, and 0.023 s−1, respectively. The rate constants kon and koff are taken from the analogous reaction where GAP is part of the receptor complex and the kcat was fit so that the half–life of RAS-GTP in the absence of EGF matched literature values [35].
Finally, a first-order turnover of internalized SOS was implemented with a rate constant of 10−7 s−1 based on the turnover rate of EGFR.
This augmented model has the additional property that if the input is removed (set to zero) it will return to its initial condition.
The mitogen activated protein kinase cascade is a signaling motif found repeated throughout biology [27]. In each step of the cascade a substrate is multiply phosphorylated by a kinase, which in turn is the input to the next layer in the cascade. The off signal, present in each layer, is a phosphatase that removes the phosphate groups. Despite knowing all of the species involved, the detailed mechanism of the enzymatic steps had been difficult to determine [27]. In particular, it was unclear if the kinase acted in two distinct enzymatic steps, whereby it released the substrate between phosphorylation steps (distributive mechanism) or if it performed both phosphorylation steps before releasing the substrate (processive mechanism).
A MAP kinase cascade consisting or RAF, MEK, and ERK is contained in the Hornberg EGFR pathway model. We extracted a tier of this cascade consisting of a single kinase, phosphatase, and substrate. The four reversible bimolecular reactions representing the phosphorylation of ERK by doubly phosphorylated MEK (MEKpp) and the de-phosphorylation by a phosphatase were used as the basis of a new model. The model contains a distributive dual phosphorylation step catalyzed by MEKpp and a distributive dual de-phosphorylation step catalyzed by a phosphatase. MEKpp is the system input; doubly phosphorylated ERK (ERKpp) is the output.
In addition to this basic model, three alternative models were constructed that differed in their mechanism of phosphorylation and de-phosphorylation (processive or distributive). The set of four models (distributive-kinase/distributive-phosphatase, processive-kinase/distributive-phosphatase, distributive-kinase/processive-phosphatase, processive-kinase/processive-phosphatase) represents all possible combinations of processive and distributive phosphorylation and de-phosphorylation mechanisms. The alternative models, which contain some rate parameters not included in the distributive-kinase/distributive-phosphatase base model, were parameterized by fitting the parameters to the step response of the double distributive model, which included both a step-up and a step-down experiment. The details of these four models are given in Dataset S2.
We have developed a method for designing an input signal capable of controlling the output of a candidate model. In practice, these input signals are useful for distinguishing among sets of candidate models.
The dynamic optimization controller was applied to design input stimuli for each of the two alternative antibody binding reactions studied by Smith-Gill and co-workers [3]. For both the one-step and the two-step model (Dataset S1), the objective applied was to produce a constant output of antibody–ligand complex from time zero onwards. In the experiment performed by Smith-Gill and co-workers the measurement was a change in mass due to ligand binding as measured by surface plasmon resonance. While the fully bound complex is more stable than the postulated encounter complex, both have the same mass and would produce the same output signal. Therefore, in the case of the two-step model, the output is the sum of the encounter and fully bound complexes, whereas in the one-step model it is simply the fully bound complex. The basis set for the input was a 50-point piecewise-linear function with linear spacing. In the two-step model points were distributed evenly over the entire interval. In the one-step model points were placed evenly from 500 s to 600 s to accommodate the sharp transition.
The results are shown in Figure 2. Both controllers designed an input signal that starts at high concentration to form complex quickly and then drops to a lower concentration to keep the complex from overshooting the desired value. However, the controller for the one-step model drops abruptly while the controller for the two-step model drops more gradually. The desired outputs were not recovered when the stimuli from the wrong models were applied. When the one-step input was applied to the two-step system, the output produced an undershoot followed by an overshoot. When the input designed for the two-step model was applied to the one-step system, the complex concentration also produced an overshoot, but one that persisted. In both cases, accounting for the presence or absence of the encounter complex was critical for controlling the output correctly.
It is interesting to note that this method allows for the selection of both the more complex model (if it is correct) as well as the simpler model. This is not possible using standard a posteriori metrics, such as least squares, which will always favor the more complex model. While there are methods that try to correct for this bias [36], properly accounting for model complexity in large nonlinear systems remains an open problem [37]. Comparing our results to the Smith-Gill pulse method (Figure 2B), it is clear that both computational experiments permit the two models to be distinguished in favor of the two-step method. However, for larger and more complex cases, it is unclear whether intuitive approaches or square pulse inputs will be sufficient to design distinguishing experiments. Another feature of the simulations is that the designed pulse produces a level output that does not require fine time resolution to accurately measure. This can be a significant advantage for more complex experimental systems, such as cell signaling measurements, where limitations on experimental observations are even more severe, whether in terms of numbers of species, time points or other factors.
Mitogen-activated protein kinase cascades have been extensively studied experimentally and modeled computationally. While many variants exist, the canonical pathway consists of three layers of kinases and phosphatases. For each layer, the kinase activates the downstream kinase by dual-phosphorylation and the phosphatase deactivates the downstream kinase by removing the phosphate groups. Knowing the general structure of this pathway, it was still difficult to determine the details of the enzymatic steps. In particular, it was unknown if the kinase acted in a processive mechanism (adding both phosphate groups in a single step), or if it acted in a distributive mechanism (adding the phosphates in two distinct enzymatic steps). The difficulty arose from the fact that, without measuring all of the phosphorylation forms, both mechanisms could fit the step response data. The issue was eventually resolved by devising an experiment that could separate all of the phosphorylation forms [27]. Here we show that, in principle, the mechanisms could have been distinguished using our method, without adding additional measurements.
To address this problem we generated four candidate models of a MAPK dual phosphorylation reaction. All four models contained forward phosphorylation and reverse de-phosphorylation steps, but differed in the detailed mechanisms. For both the forward and the reverse reactions we considered a processive (one-step) and a distributive (two-step) mechanism (Figure 3A). Taking all combinations of distributive and processive reactions produced four models. For each model the free kinase concentration was the input variable and the concentration of doubly phosphorylated substrate was the output.
For each of the four models, a stimulus was developed using the tangent linear controller. The objective was to drive the output to a fixed value that remained constant with time. Each of the four designed signals was used to stimulate each of the four models, and the resulting 16 experiments are shown in Figure 3C. Along the diagonal, one can see that the input signal derived from the correct model was able to effectively control the system. However, looking at each off-diagonal entry shows that inputs from each wrong model did a poor job controlling each system. In any real experiment, there is only one true system, which corresponds to performing the experiments from a single row of the figure.
As with the antibody models, the algorithm was able to find a set of signals that distinguished amongst multiple models. It is worth noting that these solutions were generated automatically from the candidate models and did not require explicit user supervision.
A popular ordinary differential equation model of the EGFR pathway is that of Hornberg and co-workers [16]. This model consists of 103 differential equations and includes ligand binding, receptor dimerization and activation, adaptor protein binding, trafficking of the receptor complex, and activation of the MAPK cascade terminating with ERK dual phosphorylation (Figure 4). This model was built as a set of successive refinements of earlier models [17,18,27], with each refinement adding a new level of detail to the model. In its most recent formulation an additional negative feedback loop was added whereby activated ERK phosphorylates SOS and deactivates it. This model has been shown to agree with time course data collected in cell based assays as well as literature values for parameters measured in vitro [18]. We compare the original Hornberg model to a version with additional changes. We continue this model evolution by modifying the Hornberg model so that, when the input (EGF) is removed, the model returns to its initial conditions. This reset behavior is observed experimentally. Cells cultured in media containing EGF but switched to serum and EGF free media 12 h before stimulation, are able to respond to a dose of EGF added to the media [23]. This indicates that after EGF has been removed, the pathway returns to an EGF responsive state.
In the Hornberg model, the dominant mechanism for desensitization and adaptation of the pathway to EGF is endocytosis and degradation of the receptor complex. Opposing this process are constitutive production and degradation reactions for the receptor, which allow the receptor level to return back to steady state after stimulation. This same process degrades other proteins in the receptor complex GAP, GRB2, SOS, and RAS, but the current model does not contain synthesis terms for these proteins. As a result, prolonged stimulation depletes these proteins and prevents the activation of RAF, MEK, and ERK. We added production and degradation reactions analogous to the reactions for the receptor for all of the proteins in the receptor complex. Rate constants were chosen such that the steady-state levels in the absence of stimulation were the same as the initial conditions for the model and the exponential time constant for the approach to steady state was the same as for EGFR.
The second modification to the model was in the RAS-GDP/RAS-GTP cycle. In the Hornberg model, activated receptor is needed to catalyze the recycling of RAS-GTP* (a molecule of RAS-GTP that has already activated a molecule of RAF) that is waiting to be recycled to RAS-GDP. If EGF is removed, RAS can be trapped in the RAS-GTP* form, preventing the system from returning to steady state. We addressed this by adding an additional enzymatic step to recycle RAS-GTP* back to RAS-GDP catalyzed by GAP and parameterized using literature rate constants [35].
With the addition of these new reactions, the modified model returns to its initial conditions after stimulation. For the remaining model parameters (the parameters shared with the original model) we fit the modified model to the original using data from a simulated step-response experiment (Figure 5A) constraining them to be within 10% of their original value. Despite the introduction of these new mechanisms and the tight constraints on the parameters, the step responses of the six molecular species modeling those presented in the original paper [18] (Figure 5B) are very similar in the original model (blue curves) and the modified model (red curves). The largest difference is in the SHC* time course, which has a very similar shape and varies by at most 11%. While significant, this difference would be very difficult to detect in a standard biological experiment. As such, the modified model is a reasonable alternative to the original model, and it would be hard to reject either mechanism using the step-response data alone.
From this starting point we used our methodology to design an experiment that could distinguish between the current model and the modified model of the EGFR pathway. For each model we tasked the dynamic optimization controller with driving the concentration of doubly phosphorylated ERK to a constant level of 104 molecules per cell. The input basis set was 25 points linearly spaced over the interval. To model the experimental condition where it is easy to add EGF to the dish of cells but difficult to remove, we implemented a monotonicity constraint. Figure 5B shows the inputs designed for each of the two models applied to each system with the resulting ERKpp time courses. Due to the negative feedback loops, both models required a steadily increasing concentration of EGF to maintain a constant level of ERKpp. However, the original model was much more difficult to control; as time progressed increasingly high doses of EGFR were required to maintain a constant output. The modified model required a much gentler increase in EGF concentration to maintain its level and was able to keep the concentration of ERKpp high to the end of the time period. Trial calculations showed that this result was robust to order of magnitude changes in the new rate parameters introduced in the modified model. Applying these two signals in an experiment could be used to distinguish between these two models, as demonstrated by the simulations.
The most common stimulus-response protocol involves applying a step change in one or more input concentrations and following the evolution of one or more downstream molecules. For a linear system, this type of experiment can provide enough information to fully identify the system [38]. However, even simple biochemical systems are nonlinear, and as such there is no a priori reason to believe that a step-response experiment will be sufficient to uncover the relevant dynamics of the system and allow for the selection of a unique model. As a result, it is often possible, if not probable, that multiple mechanisms fit the same set of step-response data. We have shown here that using dynamic stimulation can improve stimulus-response experiments. Even in the context of complex pathways with limited numbers of inputs and outputs, experiments can be designed that are capable of distinguishing amongst alternative mechanisms. Moreover, for the EGFR pathway studied, the differences detected were in the middle of the pathway, far from the location of the stimulus or the readouts.
One possible explanation for the results presented here is that we have stimulated the systems with high-frequency signals, and it is this fact that allows for model discrimination. While the high-frequency content almost certainly plays a part, the fact that differences between models are observed at low frequency distinguishes our results from other standard test signals. For example, in linear systems it is common to use random or pseudorandom signals to discriminate among models. Figure S1 shows such an experiment. While the signal is discriminating, the observed differences are high frequency and would be difficult to distinguish in a standard biological assay, which is usually sampled sparsely in time.
Formulating experimental design as a control problem yielded a relatively straightforward numerical solution, which allowed us to apply our method to large pathway models. While the method does not yield optimal experiments, in the sense of maximizing the least squares error between model ouputs, the results are still of practical benefit and appear sufficient to distinguish amongst model candidates. In the systems studied here, the designed inputs were able to substantially increase the differences observed between competing models when compared to the corresponding step-response experiment. By prescribing the target output trajectory, it should be possible to tailor the experiments to the available measurement methods, thereby achieving the most benefit from existing assays. It is worth noting that in all of the examples presented here, the target function was a constant output concentration. This was chosen for simplicity rather than for any special property of these targets. The problem of the best target function is an interesting one but is beyond the scope of this work. However, in Figure S2 we show calculations for the antibody–ligand system using other simple target functions, lines of constant slope, and find that designed inputs based on these signals have similar discriminating power.
In each of the cases presented here, the dynamic stimuli allowed us to select the correct mechanism from a set of plausible candidates. However, it is possible that for a particular system and set of constraints, the algorithms presented here may fail to find a signal that is sufficiently discriminating. In this case a different choice of target function or a more sophisticated optimization approach may yield better results. However, it is worth noting that in the systems studied here both methods were able to find very good solutions in all cases. In general, the tangent linear controller was more computationally efficient and yielded smoother signals, whereas the dynamic optimization controller was slower but did not require tuning of parameters such as τ.
One potential limitation of our method comes from our reliance on parameterized models. The accuracy of the parameterizations will affect the quality of the predictions made by the controllers and thus the ability to distinguish between models. To demonstrate this, we generated 100 different parameterizations of the one-step and two-step antibody models and then applied the control signals designed using the nominal parameter set (Figure S3). The parameter variation resulted in output trajectories that were quantitatively different from the predicted output trajectories. However, the overall shape of the output trajectories was preserved.
All of the results presented here were in simulation. In practice, experimental error and measurement noise will make it more difficult to distinguish between models. As a result, one may only be able to effectively discard some candidate models, and reduce the pool of hypotheses. However, these experimental challenges also motivate our method, as it has the potential to increase the experimental observability of model differences when compared to a more traditional experiment, such as a step response. Moreover, the fact that potential mechanisms can be evaluated without having to resort to additional inputs or outputs is especially valuable in laboratory experiments, where adding additional inputs or outputs may require significant effort, such as developing new experimental reagents. |
10.1371/journal.pgen.1006022 | BOD1 Is Required for Cognitive Function in Humans and Drosophila | Here we report a stop-mutation in the BOD1 (Biorientation Defective 1) gene, which co-segregates with intellectual disability in a large consanguineous family, where individuals that are homozygous for the mutation have no detectable BOD1 mRNA or protein. The BOD1 protein is required for proper chromosome segregation, regulating phosphorylation of PLK1 substrates by modulating Protein Phosphatase 2A (PP2A) activity during mitosis. We report that fibroblast cell lines derived from homozygous BOD1 mutation carriers show aberrant localisation of the cell cycle kinase PLK1 and its phosphatase PP2A at mitotic kinetochores. However, in contrast to the mitotic arrest observed in BOD1-siRNA treated HeLa cells, patient-derived cells progressed through mitosis with no apparent segregation defects but at an accelerated rate compared to controls. The relatively normal cell cycle progression observed in cultured cells is in line with the absence of gross structural brain abnormalities in the affected individuals. Moreover, we found that in normal adult brain tissues BOD1 expression is maintained at considerable levels, in contrast to PLK1 expression, and provide evidence for synaptic localization of Bod1 in murine neurons. These observations suggest that BOD1 plays a cell cycle-independent role in the nervous system. To address this possibility, we established two Drosophila models, where neuron-specific knockdown of BOD1 caused pronounced learning deficits and significant abnormalities in synapse morphology. Together our results reveal novel postmitotic functions of BOD1 as well as pathogenic mechanisms that strongly support a causative role of BOD1 deficiency in the aetiology of intellectual disability. Moreover, by demonstrating its requirement for cognitive function in humans and Drosophila we provide evidence for a conserved role of BOD1 in the development and maintenance of cognitive features.
| Intellectual disability (ID) is a form of cognitive impairment characterized by limitations in cognitive functions that manifest as an intelligence quotient (IQ) below 70. ID has a prevalence of 1–3% in the general population and represents a major health-care problem. To understand the functional consequences of causative mutations we study the disease-causing mechanisms of hereditary acquired mutations that result in ID. Here we describe a large family that has a mutation affecting a gene called BOD1. Family members who are homozygous for the mutation (i.e. both maternal and paternal copies of the gene carry the mutation) produce no detectable BOD1 protein and suffer from intellectual disability. We have previously shown that BOD1 is a crucial regulator of an important signalling molecule called Protein Phosphatase 2A (PP2A) during cell division. PP2A also has diverse but poorly understood roles in neuronal function. We demonstrate here that Bod1 can regulate PP2A function throughout the cell cycle and also localises to synapses in neurons. To determine if BOD1 deficiency directly affects the structure and function of neurons we targeted the gene in the model fly organism Drosophila Melanogaster. Neuron-specific knockdown caused pronounced learning difficulties and significant abnormalities in synaptic morphology indicating that BOD1 is involved in an evolutionary conserved mechanism crucial for the development of cognitive features.
| Intellectual disability (ID) [1] is a form of cognitive impairment, characterized by limitations in mental functioning that manifest as an intelligence quotient (IQ) below 70. ID has an estimated prevalence ranging between 1% and 3% in the general population [2–4]. Mutations in more than 750 genes have been identified that cause ID when mutated [5], but particularly autosomal recessive forms of ID (ARID) are poorly understood. Even though the count of genes known to carry ARID causing mutations is now increasing at an accelerated rate because of the recent broad implementation of high throughput sequencing technologies, there are to date still less than 50 genes reported [4]. In order to increase the knowledge about the molecular basis of ARID, we have previously performed autozygosity mapping and mutation screening in a large cohort of Iranian families with a high percentage of consanguinity and identified several loci implicated in non-syndromic ARID [6,7]. In addition, we have identified a diversity of ARID genes involved in several physiological pathways, emphasising the genetic heterogeneity of the non-syndromic ARID phenotype [8–15].The diversity of genes involved in the aetiology of ARID also reflects the complexity of the affected organ, i.e. the brain, so that an increase in knowledge about monogenic causes of ID and their functional implications can greatly contribute to a better understanding of the processes involved in the development and maintenance of the brain and its higher cognitive functions.
We report here on a family with 4 female individuals presenting with ID where we found a single homozygous mutation disrupting the BOD1 (Biorientation Defective 1) gene. BOD1 encodes a highly conserved 22 kDa protein required for proper chromosome biorientation [16]. According to the GTEx Portal (http://www.gtexportal.org/home/gene/BOD1; accessed on 10/02/16) BOD1 mRNA is expressed in the vast majority of investigated tissues. During mitosis BOD1 regulates Protein Phosphatase 2A (PP2A) activity at the kinetochore [17] by specifically binding to and inhibiting PP2A complexes containing the B56 regulatory subunit. PP2A-B56 localises to mitotic kinetochores during mitosis and controls both kinetochore microtubule attachment and checkpoint signalling [18–22]. Depletion of BOD1 from HeLa cells results in a loss of inhibition of PP2A-B56 and subsequent increase of phosphatase activity at the kinetochore. In particular, BOD1 depletion leads to reduced phosphorylation of PBIP/CENP-U, which results in a failure to recruit the mitotic Polo-Like Kinase 1 (PLK1) [MIM 602098] to kinetochores [8].
Additionally, BOD1 may have other functions in cell and organism physiology. For example, somatic deletions in BOD1 were previously found in non-pyramidal neurons and cells in white matter from patients with Schizophrenia [23]. Moreover, it has recently been described to interact with the SET1/MLL (SET Domain Containing 1A/Mixed-Lineage Leukemia) complex, a member of the COMPASS-like H3K3 histone methyltransferase multi-subunit complexes. To date, no defects in histone methylation have been linked to BOD1. However, SET1/MLL also contains HCFC1 (Host Cell Factor C1) [MIM 309541] [24], a protein previously implicated in X-linked ID [25,26].
In this report, we describe the consequences of BOD1 deficiency using cell lines derived from fibroblasts of affected individuals. We found that these show changes in PLK1 protein levels, function and mislocalization of PLK1 and PP2A but, unexpectedly, with no associated mitotic impairments. This observation, which is in agreement with an absence of microcephaly in individuals with BOD1 mutations, raised the possibility of a so far unidentified, cell cycle-independent role for BOD1. In support of this hypothesis we provide evidence for a presynaptic localization of BOD1 in mammalian neurons and show that neuron-specific knockdown of the Drosophila ortholog of BOD1 leads to abnormal learning and affects synaptic morphology. Taken together, our findings strongly support the causative role of the BOD1 mutation in the individuals affected by ID, uncover novel aspects of BOD1 function and pathogenic mechanisms and highlight an evolutionarily conserved role of BOD1 in cognition.
In a family with 4 female individuals with ID (Fig 1A) we performed multipoint linkage analysis based on the assumption of an autosomal recessive pattern of inheritance and a disease allele frequency of 0.001. We identified a single 4.3 Mbp interval on chromosome 5q (5q35.1–35.2) with a LOD score of 4.4 (S1 Fig) and sequenced the coding regions of all protein coding genes within the interval. This revealed a homozygous point mutation (NM_138369.2:c.334C>T; p.R112X) in the second exon of the BOD1 gene, which co-segregated with the disease (Fig 1A). The mutation was not found in 380 Iranian and 340 German control chromosomes and was absent in 200 Danish exomes [27]. In addition, the NM_138369.2:c.334C>T mutation was not found in the current data release (ESP6500SI-V2) of the Exome Variant Server (http://evs.gs.washington.edu/EVS/), NHLBI GO Exome Sequencing Project (ESP), Seattle, WA (accessed June 2015), containing exome sequencing results from 6503 individuals, nor in data from the 1000 Genomes Project [28], nor in the Exome sequencing Results from 60,706 unrelated individuals compiled by the Exome Aggregation Consortium (ExAC), Cambridge, MA (http://exac.broadinstitute.org, accessed February 2016). Moreover, our sequencing of controls and database search also revealed no other homozygous deleterious mutations in other parts of the BOD1 coding region.
The three affected females of the left branch of the family pedigree (V:2; V:3; V:6) suffered from moderate ID with an IQ of 50–55 (determined by Wechsler’s scale) in all three cases. In addition, these individuals presented with either primary (V:3) or secondary (V:2; V:6) amenorrhoea of unknown cause. Endocrinological tests and ultrasound investigations of the ovaries revealed no abnormalities. The affected individual in the right branch of the family pedigree (VI:1) presented with mild ID (IQ: 70–75). Brain MRI scans were performed on all four affected individuals, but revealed no consistent morphological abnormalities. All four individuals were obese or overweight but this phenotype did not co-segregate with the BOD1 mutation. From patient V:2 we obtained a lymphoblast cell line (LCL). In addition we were able to establish fibroblast cell lines from affected individuals V:2 and V:3. Cells derived from homozygous mutation carriers will be referred to as BOD1-/- cells throughout the manuscript.
RT-PCR and sequencing analyses of control fibroblasts showed the presence of the full length BOD1 transcript (NM_138369.2), comprising all exons (1–4), represents the main isoform of the protein, which is 185 amino acids long and most likely its dominant functional form. In addition we detected three comparatively weakly expressed additional transcripts (Fig 1B and 1C), composed of the exons 1+3+4 (isoform b, NM_001159651.1, comprising 129 amino acids), 1+2+4 (ENST00000285908.5, encoding 129 amino acids) and 1+4 (ENST00000480951, encoding 85 amino acids). Quantitative PCR analyses of RNA preparations from control and BOD1-/- cell lines further revealed that BOD1 mRNA was absent in cell lines from both affected individuals (Fig 1D). This loss of BOD1 mRNA is likely caused by nonsense mediated decay (NMD) as mRNA levels of 3 of the 4 detected splice variants were increased to near-control levels following treatment with cycloheximide (Fig 1E). To confirm the loss of the main BOD1 isoform at the protein level, we investigated the patient cell lines by western blot, using a rabbit polyclonal antibody raised against recombinant full length GST-BOD1 [16]. In keeping with our quantitative PCR-results, this isoform was not detected in either BOD1-/- cell line (Fig 1F).
We have previously reported that siRNA-mediated knockdown of BOD1 in HeLa cells produces profound chromosome biorientation defects and a block in mitotic segregation [16]. We therefore set out to determine whether abnormalities in cell cycle progression occur in cell lines derived from BOD1-/- individuals. We first examined cell cycle progression in WT and BOD1-/- fibroblast cells (Fig 2A). This revealed a significant increase in the G1 population in BOD1-/- cells compared to WT. Treating WT fibroblasts with BOD1 siRNA resulted in a similar accumulation of cells in G1 (Fig 1A and 1B) suggesting this change in cell cycle distribution is directly due to loss of BOD1. We observed no mitotic figures in WT fibroblast cells depleted of BOD1 by siRNA, suggesting that any cells with fully replicated DNA in Fig 2A were in G2. In contrast, mitotic cells could be observed in BOD1-/- cells suggesting these cells are capable of progressing through the cell cycle, but with a delay in progressing out of G1.
To detect any gross defects in mitotic chromosome segregation, we next examined the mitotic timing of BOD1-/-and control WT fibroblasts using DIC time-lapse microscopy, measuring the time from nuclear envelope breakdown (NEB) to anaphase onset (Fig 2C; see Methods). BOD1-/- fibroblasts progressed through mitosis rapidly, with 50% of cells completing mitosis in 20 min compared to 30 min for the control cells (Fig 2D). Similar results were obtained for BOD1-/- LCL cells (S2A Fig).
Examination of fixed BOD1-/- fibroblasts by immunofluorescence revealed no significant increase in cells with unaligned or malformed spindles (S2B Fig), suggesting only a subtle disturbance of mitotic regulation in these cells. This is surprising given the profound biorientation defects observed in BOD1-depleted HeLa cells. We therefore explored the properties of mitotic BOD1-/- cells in more detail.
Bod1 is required for proper chromosome alignment during mitosis and the proper phosphorylation of several substrates of the PLK1 and Aurora B (AURKB [MIM 604970]) protein kinases. This effect occurs through the modulation of PP2A-B56 activity [16]. To determine if these pathways were affected in BOD1-/- cell lines, we investigated the localisation of PP2A-B56 and PLK1. In agreement with previously published observations in BOD1-depleted HeLa cells, BOD1-/- fibroblasts had increased levels of PP2A-B56 at kinetochores (Fig 2E and 2F). Furthermore, PLK1 levels were reduced at the kinetochores of BOD1-/- cells compared to control fibroblasts (Fig 2G–2I), just as in Bod1-depleted HeLa cells. However, unlike BOD1 depleted HeLa cells, BOD1-/- fibroblasts had increased PLK1 concentrations at centrosomes.
We also observed that total PLK1 protein levels were reduced in asynchronous BOD1-/- fibroblasts (Fig 2J). However, when these cells were synchronised in mitosis using the Eg5 inhibitor Monastrol they exhibited PLK1 levels that were comparable to control fibroblasts (Fig 2K).
To determine if the changes in protein stability of PLK1 were limited to BOD1-/- cell lines, we performed siRNA-mediated knock down of BOD1 in wild-type primary fibroblasts. The reduction in PLK1 levels was even more pronounced than that observed in BOD1-/- cells with little or no detectable PLK1 protein (Fig 2L). PLK1 protein levels were partially recovered by co-depletion of BOD1 and PP2A-B56 (Fig 2L) suggesting that Bod1 inhibits PP2A mediated PLK1 destabilisation and that PP2A regulates PLK1 function throughout the cell cycle.
Since WT fibroblasts depleted of BOD1 do not enter mitosis, we next tested whether BOD1 has any role in mitosis in WT primary fibroblasts. We used immunofluorescence to localise endogenous Bod1 and observed prominent localisation to mitotic centrosomes and kinetochores (Fig 2M), suggesting that a mitotic function for Bod1 is present in these cells, but is masked in siRNA studies by an additional pathway in primary fibroblasts that results in G1 arrest. This G1 pathway is likely lost in highly transformed HeLa cells. To ensure the arrest was not limited to fibroblast cells, we depleted Bod1 from immortalised, non-transformed RPE1 cells and also observed a G1 arrest (S2C Fig). Once more, localisation of BOD1 to kinetochores and centrosomes was observed in untransfected RPE1 cells (S2D Fig), suggesting multiple roles for BOD1 throughout the cell cycle.
To check the specificity of the Bod1 siRNA, we rescued Bod1 depletion with plasmids expressing WT or constitutively active Bod1 [17] and observed rescue of PLK1 protein levels (Fig 2N) confirming the specificity of the effect on PLK1 by Bod1 siRNA. We conclude that BOD1 depletion in primary cells causes loss of PLK1 and that BOD1-/- fibroblasts stabilise sufficient Plk1 to successfully progress through the cell cycle.
Since BOD1-/- cells are able to propagate and progress through the cell cycle, we hypothesised that they must have modulated the function of PLK1 to adapt to the presence of reduced PLK1 levels during interphase (Fig 2J). To test this hypothesis, we arrested cells at the G2/M boundary, where PLK1 is required for progression into M phase using RO3306, a CDK1 inhibitor. We then released cells into fresh medium containing increasing amounts of the specific PLK1 inhibitor BI2536. This assay uses the formation of bipolar mitotic spindles as a reporter of PLK1 function [17]. Control WT fibroblasts exposed to BI2536 showed a reduced number of bipolar metaphase spindles and an increased number of monopolar spindles as the concentration of BI2536 was increased (Fig 2O). By contrast, BOD1-/- cells showed fewer monopolar spindles and only a small reduction in the frequency of bipolar spindles even at the highest concentrations of BI2536. This result suggests that BOD1-/- cells are hyposensitive to PLK1 inhibition relative to WT controls and that they have adapted to loss of BOD1 by either increasing or bypassing PLK1 function. Consistent with the former hypothesis, BOD1-/- cells have increased levels of PLK1 at centrosomes (Fig 2G and 2I), a critical location for PLK1 function in centrosome and spindle maturation.
These results suggest that despite an overall reduction in PLK1 levels in BOD1-/- cells, they have nonetheless adapted and progress through mitosis in a relatively normal manner. This conclusion is in line with our observation that homozygous mutation carriers do not show any gross developmental or structural brain abnormalities, as are often found in individuals affected by ID with mutations in proteins required for centrosome positioning and localisation (see e.g. Chavali et al. [29] or Kuijpers & Hoogenraad [30]). We therefore hypothesized that cognitive impairment caused by BOD1 deficiency might result from the disturbance of a different, possibly brain-specific functional aspect of BOD1.
Quantitative PCR of RNA from different regions of normal human fetal and adult brain showed that all four BOD1 splice variants detected in lymphocytes and fibroblasts are present throughout the brain, at both investigated developmental stages (Fig 3A).
To address whether BOD1 might exert functions independent from cell-cycle regulation of PLK1, we compared BOD1 and PLK1 expression in different areas of the human brain by RNA-sequencing. In agreement with the low rates of dividing cells in these terminally differentiated tissues our results show that PLK1 expression drops to very low levels (Fig 3B) as compared to the expression rates observed in pluripotent cells. BOD1 expression, however, is maintained in most investigated brain tissues at about 20% of the expression levels observed in pluripotent cells, strongly suggesting that BOD1 plays a PLK1- and cell cycle-independent role in postmitotic brain cells.
Continuing on from the gene expression analysis we were interested in determining BOD1 localisation in mammalian neurons. We analysed murine corticoneuronal cells that were transfected with BOD1-GFP,at short incubation time (7h, to avoid overexpression artefacts). We observed a striking punctuate pattern with BOD1-GFP that co-localised with the synaptic marker Bassoon (Fig 4), which suggests a potential role for Bod1 in synaptic signalling.
To experimentally address whether BOD1 is required directly in neurons for cognitive processes, we took advantage of established animal model for ID disorders, the fruit fly Drosophila melanogaster. The Drosophila genome encodes a single gene representing the human BOD1 protein family (BOD1, BOD1L1 and BOD1L2), termed CG5514. A multiple sequence alignment is shown as supplemental S3 Fig. From now on we refer to this so far uncharacterized Drosophila gene as Bod1. We used the UAS-Gal4 system [31], the panneuronal promoter elav-Gal4 and three inducible RNAi lines carrying different constructs (Bod1vdrc101981 Bod1vdrc4542, and Bod1HMS00720) to knockdown Bod1 specifically in postmitotic neurons. We subjected Bod1 panneuronal knockdown flies to a simple non-associative learning assay: the light-off jump reflex habituation paradigm. In this paradigm, which has previously uncovered learning defects in several Drosophila models of Intellectual Disability [32–34] and in classic learning and memory mutants [35], flies are exposed to a repeated light off stimulus at 1 second intervals. Wildtype (wt) flies quickly adapt to the repeated stimulus and gradually suppress their initial jump response as a result of non-associative learning. In our experiments, flies of all tested genetic conditions showed high jump response towards the initial light-off stimulus, demonstrating that they properly perceived the stimulus and that their startle response was not compromised. We found that Bod1 knockdown flies failed to habituate to the presented light-off stimuli compared to their genetic background controls with normal Bod1 levels, and kept on jumping at high levels throughout the entire course of the experiments (Fig 5). This learning defect was consistent for two Vienna Drosophila Resource Center (VDRC) RNAi constructs, as well as for a non-overlapping independent short siRNA TriP construct (Fig 5A, 5B and 5C). To assess the significance of the habituation defects in both Bod1 knockdown models, flies were considered to have habituated once they failed to jump in five consecutive trials (no-jump criterion). Habituation was quantified as the number of trials required to reach the no-jump criterion (TTC, Trials to criterion). TTC of the Bod1 knockdown flies was 2.38–fold (Bod1vdrc1105166, n = 143, p<0.001), 2.14-fold (Bod1vdrc27445, n = 93, p<0.001) and 1.95-fold (Bod1HMS00720, n = 70, p<0,001) increased over their respective controls (Fig 5A’, 5B’ and 5C), revealing significant habituation defects in all three Bod1 Drosophila models.
The efficacy of the utilised RNAi constructs was confirmed by qPCR on RNA isolated from Drosophila 3rd instar larvae upon ubiquitously-induced knockdown. As Bod1vdrc27445 and Bod1vdrc105166 contain almost identical siRNA hairpins, only Bod1vdrc105166-mediated knockdown was assessed and revealed 23% remaining gene expression; p = 0.0012, student’s t-test; supplemental S1 Table. Bod1HMS00720-mediated RNAi reduced levels of Bod1 to a lesser but still very significant extend, resulting in 37% remaining gene expression (p = 0.0086, student’s t-test; S1 Table).
We conclude that conditional knockdown of Bod1, induced specifically in neurons, does not affect the overall startle response/acute fitness of the flies but leads to specific and highly consistent learning defects.
Synapse biology is crucial for learning and has been proposed to play a central role in ID [36–38]. Therefore, and because of the observed Bod1 localisation to synapses in cultured neurons, we investigated the consequences of Bod1 knockdown on synapse development. The Drosophila larval Neuromuscular junction (NMJ) was investigated in elav-Gal4 induced panneuronal Bod1 knockdown larvae using an antibody against dlg1 that visualizes the overall morphology of synaptic terminals. Knockdown of Bod1 with both Bod1vdrc27445 and Bod1vdrc105166 lines induces modest but highly significant abnormalities in synapse branching (Fig 5D)The average number of branching points was 2.7 versus 1.6 in the appropriate genetic background control (Bod1vdrc105166 versus control); p<0.00001) and 2.5 versus 1.3 (Bod1vdrc27445 versus control; p<0.005). Other synaptic parameters such as length or perimeter of the synaptic terminal were not consistently changed. There was no significant NMJ defect in the weaker Bod1HMS00720 knockdown condition (1.5 in Bod1HMS00720 versus 1.4 in control). We conclude that Bod1 is required for habituation and may control synaptic branching.
We report that BOD1, a protein that has previously been linked to mitotic cell division, is required for cognitive functions in humans and Drosophila. This is in line with a previous report that described somatic deletions in BOD1 within non-pyramidal neurons and cells in white matter from patients with Schizophrenia, implicating BOD1 as a player in brain development and, more importantly, in a neuropsychiatric context [23].
In agreement with the absence of gross morphological brain abnormalities in homozygous mutation carriers, our experiments, using neuronal cultures and Drosophila as a model organism, support a novel postmitotic function for Bod1 and raise the possibility that abnormal synaptic development or function contributes to cognitive impairment in individuals with BOD1 mutations.
In the consanguineous family we describe here, a nonsense mutation in exon 2 of BOD1 co-segregates with moderate-to-severe intellectual disability. This mutation is absent in ethnically matched controls and in all presently accessible genome or exome sequencing data repositories. Furthermore, it is of note that in the ExAC database, which contains exome sequencing data from 60,706 unrelated individuals, only 6 coding positions within BOD1 showed deleterious alterations, of which none was observed in a homozygous state and four were found only once. This shows that homozygous deleterious mutations affecting BOD1 are not tolerated in intellectually healthy individuals and strongly supports our conclusion that loss of BOD1 function is detrimental to molecular pathways involving this protein and is thus causative for ID.
All four homozygous mutation carriers lack detectable levels of BOD1, presumably because the mutation leads to NMD of the corresponding mRNA. Originally BOD1 was identified as a centrosomal and kinetochore protein that is required for proper chromosome biorientation in HeLa cells [16]. It has previously been shown that BOD1 is involved in the regulation of AuroraB kinase dependent phosphorylation of Mitotic Centromere Associated Kinesin alias Kinesin Family Member 2C (MCAK alias KIF2C [MIM 604538]) [16]. Recently BOD1 has also been shown to act as an inhibitor of PP2A-B56 function at the kinetochore and to regulate the recruitment of a number of proteins to the kinetochore and centrosome, including PLK1 and PP2A [17].
A considerable number of other centrosomal proteins or proteins required for centrosome positioning and localisation have already been associated with ID, including ASPM [39], the human ortholog of the abnormal spindle gene (asp) in Drosophila [40], which is essential for the normal functioning of mitotic spindles [41]. Also defects in MCPH1 [MIM 607117], CENPJ [MIM 609279] and CDK5RAP2 [MIM 608201] have been shown to cause ID (for review see Chavali et al. or Bond et al. [29,42]). However, mutations in these genes usually induce not only cognitive defects but also entail severe brain abnormalities such as primary microcephaly. It is therefore remarkable that homozygous mutation carriers (despite the apparent absence of the main BOD1 transcript) do not show microcephaly or any other gross structural brain abnormalities. Therefore, as BOD1-/- fibroblasts still proliferate and pass through mitosis, they must have in some way adapted to the loss of BOD1 and to BOD1-mediated PLK1 function, demonstrating a surprising adaptability in the cell cycle control machinery and thus providing an explanation for the absence of structural defects in the patient’s brains. In this context it is interesting that BOD1-/- fibroblasts show an increased tolerance to the PLK1 inhibitor BI2536, which might be explained by the increased localisation of PLK1 to the centrosomes that we observed. Increased PLK1 at centrosomes may stabilize spindles allowing mitosis to proceed and preventing structural deficits in the affected tissues.
BOD1 is still expressed throughout adult human brain tissues, in contrast to the cell cycle kinase PLK1 (Fig 3). This suggests that BOD1 exerts additional functions in mature neurons, and our observation that Bod1 localises to synapses further supports this notion.
In order to further investigate whether and how loss of BOD1 may affect cognitive function independent from its mitotic role in cell cycle progression we established Drosophila models of neuronal BOD1 deficiency by knocking down the fly ortholog of BOD1, CG5514, specifically in postmitotic neurons. All three models show specific deficits in habituation, an evolutionary conserved form of adaptive learning. Habituation is an important neuronal filtering mechanism, preventing information overload, and a prerequisite for higher cognitive functioning [43–45]. The Drosophila data thus provide independent evidence for a role of BOD1 in cognitive processes. It is also interesting to note that expression of the PLK1 fly ortholog polo compares well with our finding of very low PLK1 expression in human brain tissues. While Drosophila polo is expressed in larval neuroblasts [46], polo transcripts seem to be absent in adult brain [47], supporting an evolutionarily conserved, PLK1-independent postmitotic function of BOD1 in humans and Drosophila.
The synapse is a key compartment of postmitotic, terminally differentiated neurons that plays a pivotal role in the maturation and maintenance of cognitive abilities. While the precise localisation of Bod1 at synapses warrants further investigation, synaptic localisation in mammalian neurons and the Drosophila synaptic phenotype upon strong knockdown suggest a novel role for Bod1 in synapse biology. This assumption is supported by the fact that Drosophila BOD1 (CG5514) is one of 893 Drosophila genes that were predicted to be involved in synapse assembly and function [48]. Interestingly, it has been shown that functional PP2A holoenzymes are required for synaptic growth and synaptic function at the Drosophila NMJ [49]. In view of our finding that BOD1 is required for proper PP2A function in human cells, this could mean that BOD1 is necessary for appropriate PP2A holoenzyme function and might thus be involved in the normal development and maintenance of cognitive features. PP2A has already been implicated in the aetiology of ID. Increased levels of the PP2A catalytic subunit were found in a cellular model of Fragile X syndrome [50] and PP2A was observed to dephosphorylate the Fragile X protein, FMRP, in immediate response to immediate group I metabotropic glutamate receptor (mGluR) stimulation [51]. What is more, putatively causative missense changes in PP2A subunits were found in several affected individuals from a large scale study of the genetics of developmental disorders [52] and most recently de novo missense mutations were identified in B56δ—and Aα—subunit of PP2A in individuals with ID [53].
BOD1 can also be linked to Intellectual Disability through a potential involvement in chromatin modification. The protein was recently found associated with SET1B complexes, COMPASS-like H3K3 histone methyltransferase multisubunit complexes, also containing HCFC1 [24], implicated in X-linked Intellectual Disability [25,26]. Like homozygous carriers of BOD1 mutations, individuals affected by HCFC1 mutations show no microcephaly phenotype. Furthermore MLL1, MLL2 and MLL3, the core subunits of other COMPASS complexes that share a number of subunits with SET1B, and might thus be affected indirectly by loss of BOD1, have been implicated in ID [54–57].
Although our data strongly indicate a post-mitotic role of Bod1, subtle mitotic defects in neurogenesis may also contribute to the cognitive impairment of individuals carrying BOD1 mutations. For example, orientation of the cleavage plane is known to be important for cell fate choice during neurogenesis and development of the neural tube [58–60]. Accelerated progression into anaphase, analogous to those observed in BOD1-/- Fibroblasts, may interfere with the correct positioning of the cleavage plane and lead to abnormalities in the generation of the correct proportions of neurons and progenitors.
Taken together our results identified homozygous loss of BOD1 as a novel cause of ID and revealed a so far unappreciated postmitotic, likely synaptic function of BOD1. Thus our work opens up interesting avenues of research into the function of centrosomal proteins in fully differentiated neurons.
The ID study was performed in agreement with the approval (2100) of the ethics committee of the Charité University Medicine Berlin ("Ethikkommission der Charité—Universitätsmedizin Berlin"), Germany. Written consent for the intellectually disabled individuals was provided by their parents.
The pedigree of the family reported here is shown in Fig 1A. Blood samples for DNA preparation were collected from the parents and all children of both branches and genomic DNA was extracted using a standard method. For the index patient (V:2), Fragile X was excluded by PCR and Southern blot analysis. Filter-dried blood of the index patient was screened by tandem mass spectrometry to exclude disorders of the amino acid, fatty acid (e.g. phenylketonuria) or organic acid metabolism [61,62]. Standard 450 G-band karyotyping was performed in order to exclude cytogenetically visible chromosomal aberrations.
An additional blood sample from patient V:2 was used to establish an EBV transformed lymphoblastoid cell line and skin biopsies were taken from individuals V:2 and V:3 to isolate and culture fibroblast cells.
Genotyping was performed with Human Mapping 10K Array Version 2 (Affymetrix). Multipoint parametric linkage analysis was performed using Allegro [63] applying an autosomal recessive pattern of inheritance and disease allele frequency of 0.001. Sanger sequencing was performed and the whole coding and exon-flanking regions in the interval were screened (The primer sequences are available upon request). The resulting sequences were analyzed using the CodonCode aligner software.
Total RNA from the fibroblast and EBV transformed lymphoblast cultures was isolated by using TRIzol reagent RNA extraction protocol (Invitrogen). In addition, commercially available RNA (BioCat) was used from various tissues as indicated in Fig 3.
Semiquantitative RT-PCR was performed in two steps, after Dnase treatment of the RNA with RQ1 RNase-Free DNase (Promega), cDNA was synthesized with SuperScriptIII reverse transcriptase kit (Invitrogen) together with random hexamers and followed by Polymerase Chain Reaction for 35 cycles with primers in both sides of the gene (Forward primer:CATCGTGGAGCAGCTCAAG and Reverse primer:GCACTCTTATGTAACCGAATC) to amplify all the possible splice variants. Subsequently each of the splice variants were verified by isoform specific RT-PCR (The primer sequences are available upon request) as well as Sanger sequencing.
Isoform specific quantitative PCR was performed in 20 μl volumes and carried out in the ABI PRISM 7900 HT Sequence Detection System using a 96-well format. Cycling parameters: 10 min at 95oC followed by 40 cycles of 15s 95oC and 1 min 55oC. Amplification plot and predicted threshold cycle (Ct) values were obtained with the sequence Detection Software (SDS 2.1, PE Applied Biosystems).
We used absolute quantification to quantify unknown samples by interpolating their quantity from a standard curve. To construct the standard curve four-fold dilutions (1, 0.5, 0.25, 0.125) of a total RNA preparation from control cDNA were used. Negative controls (“no template control”, NTC) were used to verify amplification quality and to exclude contamination or primer-dimer artifacts. The gene used for normalising expression was GAPDH.
The fibroblast cells from two individuals affected by ID (V:2 and V:3) and two controls were cultured into two sister flasks, one treated with cycloheximide (CHX) (Sigma) at the concentration of 500 μg/ml and the other one non-treated (just added DMSO). After 7.5 hrs of incubation in 37C / 5%CO2, the cells were harvested, RNA was extracted and isoform specific quantitative PCRs were carried out.
HeLa S3 cells were maintained in EMEM (Lonza), Primary Fibroblasts were maintained in Quantum 333 media (PAA) or FibroPlus 333 (Capricorn-Scientific) and EBV-transformed lymphoblasts were maintained in RPMI 1640 media (Gibco). All media was supplemented with 10% FCS, 2 mM L-glutamine, 100 U/ml penicillin and 100 ug/ml streptomycin. Quantum 333 and FibroPlus 333 media was also supplemented with 10 ng/ml bFGF (Cell Signalling Technology). Cell lines were maintained at 37°C with 5% CO2 in a humidified incubator.
Fibroblast cell lines were electroporated using the Neon Transfection system (Life Technologies) as per the manufacturer’s instructions using a single pulse of 20 ms and 1600 V. Medium GC content control siRNA and BOD1 Stealth siRNA (5’-GCCACAAAUAGAACGAGCAAUUCAU-3’) were supplied by Life Technologies. The specificity of these reagents and absence of off-target effects were reported previously [16,17]. Briefly, BOD1 Stealth siRNA does not target Bod1L or Bod1L2 mRNA and all phenotypes associated with the BOD1 Stealth siRNA can be rescued using exogenous siRNA-resistant BOD1 expression constructs.
Unless otherwise indicated RO3306 was used at 9 μM, Monastrol at 100 nM and BI 2536 at 12.5 to 200 nM.
Time-lapse imaging and fluorescence microscopy was performed as described [16]. Non-adherent lymphoblast cells were gently centrifuged onto the surface of a Labtek imaging chamber and constricted close to the coverslip surface by the addition of 50% Matrigel (BD Biosciences). During DIC timelapse imaging of lymphoblast and fibroblast cell lines 10 z sections, 3 μm apart were taken every 4 min for 18 hr on a DeltaVision Spectris (Applied Precision) fitted with an environment chamber (Solent) maintained at 37°C. All images were stored and manipulated using OMERO [64] or Photoshop (Adobe). Images were quantified using OMERO and m-tools as described [17]. Statistical significance was determined using Mann-Whitney Rank Sum Tests. Nuclear envelope breakdown (NEB) was defined as the time when a clear delineation of the nuclear envelope was no longer visible and the volume occupied by chromosomes began to expand. While not an exact measure of NEB, this approach combined with the 4 min interval between images provided sufficient accuracy to reveal differences in mitotic timing in these cell lines.
Rabbit anti-BOD1 [16] and mouse anti-B56α (BD Biosciences) were used at 1:100. Mouse anti-PLK1 (Upstate), and alpha-tubulin (Sigma) were used at 1:500. Human ACA (CREST) autoantisera (a kind gift from Sara Marshall, Ninewells Hospital, Dundee), were used at 1:1000. All fluorescently labelled secondary antibodies were obtained from Jackson ImmunoResearch Laboratories.
RNA Expression profiling was carried out on a SOLiD5500XL Sequencing platform as previously described [65], using RNA preparations form human embryonic stem cells (hES), induced pluripotent stem cells (iPSC) and commercially available RNA (BioCat) from various adult brain tissues as indicated in Fig 3.
The cortex was extracted from mouse embryonic brain at an age of E14.5. One day prior to sacrificing the mother plates were prepared as follows: Sterilized cover slips were placed tissue culture wells (12-well plates). The wells were then coated by incubating them for 30 min at room temperature or over night at 4°C with poly-D-Lysine/Laminin (poly-D-lysine 1:500 and laminin 1:50 in PBS, 1 ml per well). The wells were then washed twice with 1 ml PBS before adding 2 ml neurobasal medium to each well and then incubating at 37°C (8% CO2).
The pregnant mother was sacrificed by neck fracture, then the embryos were removed from the womb and placed in a sterile Petri dish containing PBS.
The preparation of corticoneuronal cells was carried out on a sterile flow bench. Embryos were briefly washed with PBS and then transferred in a fresh dry Petridish. Next the heads were removed and the brains excised. The brains were suspended in 1–2 ml DMEM, each. The Brain stem and brain lobes were separated under a stereoscope and the cortex removed, taking care to separate the cortex from the hippocampus. Cortex halves were then transferred in a chilled DMEM-Tube on ice. Next, single cells from isolated embryonic cortices were prepared: 5 ml trypsin was added and incubated for 7 min at 37°C (5%CO2), swirling once in between. Trypsin activity was then stopped by adding DMEM + 10% FCS and centrifugation (200xg for 3 min). This step was repeated twice. Finally the DMEM was carefully removed and 2 ml neurobasal medium were added, before decollating the cells using a narrowed pasteur pipette. Cells were then diluted by adding 28 ml neurobasal medium, counted and seeded to a density of 1.5x105 cells per cm2 in the prepared 12 well plates. After approximately 1 hr of incubation at 37°C (8% CO2) the medium was changed in order to remove unwanted cells. The next change of medium was after day 3–4. Subsequently the medium was changed every 7th day. On the 7th day after seeding (or later) cells were used for experimental applications.
Cells were transfected by pEGFP-BOD1 (wt) or pEGFP-N1 vector (Control) at days 7 and 9 after preparation. A solution-A (containing 0.1 μg of plasmid DNA and 100 μl OptiMEM), and a solution-B (with 2 μl lipofectamine and 100 μl OptiMEM) were prepared and kept for 5 min at room temperature. Subsequently, solutions were mixed and incubated at room temperature for 20 min. Meanwhile, cell culture medium was replaced by transfection medium (antibiotic free culture medium). The transfection mixture was slowly added to the cells which then were incubated at 37°C (8% CO2). On both days cells were fixed with PFA 4% at two time points (6 and 8 hr).
Fly stocks were kept on standard Drosophila diet (cornmeal/sugar/yeast) at 25°C, in a12h:12h light/dark cycle. For the habituation experiments, flies were reared and tested at 25°C and 70% humidity, 28°C and 60% humidity was used during synapse morphology experiments and to generate ubiquitous RNAi-mediated knockdown for quantitative PCR (qPCR). Inducible RNAi lines against the BOD1 Drosophila ortholog CG5514 (vdrc105166, vdrc27445) and their corresponding genetic background control lines (vdrc60100, vdrc60000) were obtained from Vienna Drosophila RNAi Center [66]. A nonsense RNAi line, targeting a C.elegans-specific gene, was obtained from K. Keleman. Inducible short-hairpin RNAi line HMS00720 (BL32928) and its corresponding genetic background control line y v; attP2, y+ (BL36303), generated by Transgenic RNAi Project (TRiP), were obtained from Bloomington Drosophila Stock Center. The ubiquitous actin-Gal4 driver w1118; P(w[+mC] = Act5c-Gal4)/CyO, also obtained from Bloomington Drosophila Stock Center, was used to generate RNAi-mediated knockdown for qPCR. In-house assembled panneuronal elav-Gal4 driver lines were used for habituation (2xGMR-wIR; elav-Gal4, UAS-Dicer2)) and synapse morphology experiments (UAS-Dicer2; elav-Gal4). In all experiments, progeny of a cross between these Gal4-driver lines and the genetic background of the respective RNAi line was used as a control.
The light-off jump habituation assay was performed as previously described [32] with minor adaptations. Briefly, 3–7 day old individual male flies (Bod1vdrc105166, Bod1HMS00720) or female virgin flies (Bod1vdrc24445) were tested for jump responses in two independent 16-unit light-off jump habituation systems. 32 flies (16-flies/system) were simultaneously exposed to series of 100 short (15ms) light-off pulses with a 1s interval between the pulses. The noise amplitude of wing vibration following every jump response was recorded for 500ms after the start of pulse and a carefully determined threshold was applied to annotate jump responses. Data were collected and analyzed by custom-made Labview Software (National Instruments). High initial jumping response to light-off pulse decreased with growing number of pulses and flies were considered habituated when they failed to jump in 5 consecutive trials (non-jump criterion). Habituation was scored as the number of trials required to reach the non-jump criterion (Trials To Criterion, TTC). Main effects of genotype (mutant vs control), day and system on log transformed TTC values were tested using linear model regression analysis (lm) in R statistical software (R version 3.0.0 (2013-04-03)). A nonsense RNAi line targeting a C.elegans-specific gene (KK library construct) was also tested and did not affect habituation.
Total RNA from 3rd instar larvae (3 biological replicates) was isolated using RNeasy Lipid Tissue Mini Kit (Qiagen). During the isolation procedure, samples were treated with RNase-free DNase Set (Qiagen). RNA isolated from Bod1HMS00720 and its respective genetic background control was used directly for cDNA synthesis. RNA isolated from Bod1vdrc105166 and its respective genetic background control was further processed by Oligotex mRNA Mini Kit and purified PolyA+ RNA was used for cDNA synthesis. First strand cDNA synthesis was performed using iScript cDNA Synthesis Kit (Biorad). Gene expression was analysed by real-time PCR (7900HT Fast Real-Time PCR system, Applied Biosystems). PCR reactions were performed in a volume of 25μl containing 150 nM primers and GoTag Green Mastermix (Promega). Primer sequences used for amplification of CG5514: 5’-ACAACAGTGGGGAGCCAG-3’ and 5’- CCTGTGCTAGTCGTCTCCG-3’. The amplicon spans the HMS00720 cleave site and should effectively detect the initial cleavage step of RNA interference. To determine the efficiency of the initial cleavage step of long vdrc105166 siRNA hairpin, the primers were designed to detect the region within the 5’ cleavage fragment, and qPCR was perfomed on PolyA+ mRNA. This ensures that in case of successful initial cleavage, regions 5’ to the cleavage site would not be amplified in qPCR reaction [67]. Pol II was used as reference gene, primer sequences: 5’- TCAGAGTCCGCGTAACACC-3’, 5’- TGGTCACAAGTGGCTTCATC-3’.
Wandering L3 larvae were dissected and fixed in 3.7% paraformaldehyde for 30 minutes. Males were selected for Bod1vdrc105166 and female larvae for the X-linked Bod1vdrc24445. Type 1b neuromuscular junctions (NMJs) at muscle 4 were visualized using the primary antibody anti-dlg1 (1:25, Developmental Studies Hybridoma Bank) in combination with the Zenon Alexa Fluor 568 Mouse IgG1 labelling kit (Invitrogen). NMJ images were acquired using a Leica automated high-content microscope. Individual synapses of segments A2 –A5 were imaged and quantified using an in-house developed Fiji-compatible macro. Synaptic parameters were compared between the RNAi line and the corresponding genetic background control line using a student’s T-test.
|
10.1371/journal.pgen.1005724 | Dynamic Roles for Small RNAs and DNA Methylation during Ovule and Fiber Development in Allotetraploid Cotton | DNA methylation is essential for plant and animal development. In plants, methylation occurs at CG, CHG, and CHH (H = A, C or T) sites via distinct pathways. Cotton is an allotetraploid consisting of two progenitor genomes. Each cotton fiber is a rapidly-elongating cell derived from the ovule epidermis, but the molecular basis for this developmental transition is unknown. Here we analyzed methylome, transcriptome, and small RNAome and revealed distinct changes in CHH methylation during ovule and fiber development. In ovules, CHH hypermethylation in promoters correlated positively with siRNAs, inducing RNA-dependent DNA methylation (RdDM), and up-regulation of ovule-preferred genes. In fibers, the ovule-derived cells generated additional heterochromatic CHH hypermethylation independent of RdDM, which repressed transposable elements (TEs) and nearby genes including fiber-related genes. Furthermore, CHG and CHH methylation in genic regions contributed to homoeolog expression bias in ovules and fibers. Inhibiting DNA methylation using 5-aza-2'-deoxycytidine in cultured ovules has reduced fiber cell number and length, suggesting a potential role for DNA methylation in fiber development. Thus, RdDM-dependent methylation in promoters and RdDM-independent methylation in TEs and nearby genes could act as a double-lock feedback mechanism to mediate gene and TE expression, potentiating the transition from epidermal to fiber cells during ovule and seed development.
| Cotton is the world’s largest source of renewable textile fiber and is an allotetraploid crop consisting of two progenitor genomes. In plants, de novo CHH (H = A, T, or C) methylation depends on RNA-directed DNA methylation (RdDM) and CHROMOMETHYLASE2 (CMT2)-mediated pathways. The biological significance of the two pathways is largely unknown. Here we show dynamic roles of these two pathways in ovule and fiber development. RdDM-dependent CHH methylation is linked to gene activation in ovules, and additional CMT2-dependent methylation leads to silencing of transposons and nearby genes in fibers. Moreover, DNA methylation affects expression bias of homoeologous genes and fiber development. These findings provide novel insights into epigenetic regulation of organ development and polyploid evolution.
| DNA methylation, a conserved epigenetic mark in most eukaryotes, is essential for growth and development and is associated with many epigenetic phenomena, including imprinting and transposon silencing [1–5]. In plants, DNA is methylated in CG, CHG and CHH (H = A, T, or C) sites through distinct pathways. In Arabidopsis, CG methylation is maintained by METHYLTRANSFERASE1 (MET1), a homolog of mammalian DNMT1 [6]. Plant-specific CHROMOMETHYLASE3 (CMT3) is primarily responsible for CHG methylation, which is coupled with H3K9 dimethylation [7]. CHH methylation is established de novo by DOMAINS REARRANGED METHYLTRANSFERASE1 and 2 (DRM1 and DRM2) [8] through the RNA-directed DNA methylation (RdDM) pathway [9], involving 24-nt small interfering RNAs (siRNAs) [1, 2]. Recent studies found that CHH methylation could also be established by CMT2 [10, 11], through histone H1 and DECREASE-IN-DNA-METHYLATION1 (DDM1) activities [12], which is independent of the RdDM. The methylome data indicate that CMT2 and RdDM pathways preferentially function in heterochromatic and euchromatic regions, respectively [10, 11]. However, the role for DNA methylation in developmental regulation is poorly understood.
Cotton is the largest source of renewable textile fiber and an excellent model for studying the developmental transition from ovule epidermal cells to rapidly-elongating singular fiber cells. The most widely-cultivated cotton (Gossypium hirsutum L., AADD) is an allotetraploid species, which originated 1–2 million years ago from interspecific hybridization between A-genome species, resembling Gossypium herbaceum or Gossypium arboretum, and D-genome species, resembling Gossypium raimondii [13]. The intergenomic interaction in allotetraploid cottons induces longer fiber and higher yield, coincident with expression bias of fiber-related homoeologous genes [14, 15], which provides the basis of selection and domestication for agronomic traits in cotton and many other polyploid crops [13, 16]. Each cotton fiber is a single cell derived from the ovule epidermis, undergoing rapid cell elongation and cellulose biosynthesis, and ~100,000 fiber cells develop semi-synchronically in each ovule (seed) and can reach six centimeters in length [17, 18]. In early stages of fiber development, rapid cell growth is associated with a dramatic increase of DNA content by endoreduplication [19, 20] and dynamic changes in gene expression and small RNAs [15, 18, 21]. Interestingly, DNA methylation changes are related to seasonal variation of fiber development in cotton [22] and is also shown to change among different tissues including fibers based on the methylation-sensitive high-performance liquid chromatography (HPLC) assay [23]. Moreover, over-expressing fiber-related transgenes often leads to the unexpected outcome of fiber phenotypes [24]. These data indicate a potential role for DNA methylation in gene expression and phenotypic traits such as cotton fiber, which could be selected and domesticated.
Although genome-wide DNA methylation has been examined in Arabidopsis [25], soybean [26, 27], maize [28, 29], and other plants and animals [30, 31], the roles of RdDM and CMT2-dependent methylation pathways in organogenesis and development remain elusive. In this study, we employed cotton ovule and fiber cells as a model to test the role of DNA methylation in developmental regulation. Using methylcytosine-sequencing (MethylC-seq) [32, 33], RNA-seq, and small RNA-seq analyses, we examined CG, CHG, and CHH methylation patterns in fibers, ovules, and leaves and analyzed differentially methylated regions (DMRs) between the ovule and leaf (OL) and between the fiber and ovule (FO). The methylation patterns in the gene body and 5’ and 3’ flanking sequences were comparatively analyzed with TE densities and expression levels of genes and small RNA loci. The results support unique roles of CG, CHG, and CHH methylation in ovule and fiber development and expression bias of homoeologous genes in the allotetraploid cotton.
DNA methylation affects growth and development in plants and animals [3, 4]. To investigate genome-wide DNA methylation changes during ovule and fiber development, we used allotetraploid cotton (G. hirsutum L. acc. TM-1) to perform whole-genome bisulfite sequencing in leaves, ovules at 0 DPA, and fibers at 14 DPA with two biological replicates and ovules at 14 DPA with one replicate. The TM-1 sequence of A and D subgenomes [34] was used as the reference for data analysis. The bisulfite-conversion rates were over 99% (S1 Table). Approximately 80% of cytosines were covered by at least one uniquely mapped read, and the mean coverage of cytosines was 8.2-fold or higher for all tissues. Methylation levels between biological replicates were highly correlated among symmetric sites (Pearson r>0.9) and all sites (r>0.8), indicating reproducibility of the data (Figs 1A and S1A). Overall, CG and CHG methylation levels were similar among fibers, ovules and leaves but were lower in the D subgenome than in the A subgenome for all tissues (Fig 1A). However, CHH methylation levels were much higher in fibers (~14%) than in ovules (~8.1%) and leaves (~7.8%) (Fig 1A). This high methylation level in fibers was not related to the late developmental stage because the methylation level in ovules was slightly lower at 14 DPA than at 0 DPA (S1B Fig). CHH methylation is induced by small RNAs through RdDM [2, 35]. To test a link of DNA methylation changes in cotton tissues with small RNAs, we generated small RNA-seq data with three replicates in leaves, ovules at 0 DPA, and fibers at 14 DPA (S1 Table). The data showed 24-nt siRNAs were much higher in fibers and ovules than in leaves (Fig 1B) [15].
On each chromosome, DNA methylation was more abundant in transposable element (TE)-rich than gene-rich regions in all tissues (Figs 1C and S1C). However, 24-nt siRNAs were highly enriched in gene-rich regions (Figs 1C and S1C), suggesting a role for these siRNAs in genic methylation. Because each fiber cell is expanded from an epidermal cell of the ovule, further comparison was made between the ovule and leaf (ovule-leaf, OL) and between the fiber and ovule (fiber-ovule, FO) methylation levels, which represented methylation changes in the ovule and fiber, respectively. OL CHH hypermethylation correlated with gene-rich regions, showing the same trend as that of the siRNAs (Fig 1C). On the contrary, FO CHH hypermethylation was enriched in TE- and repeat-rich regions, characteristic of heterochromatin. The data indicate that OL CHH hypermethylation preferentially occurs in euchromatic regions, whereas FO CHH hypermethylation predominates in heterochromatic regions (Figs 1C and S1D).
CHH methylation differences between the ovule and leaf occurred mainly in 5’ upstream and 3’ downstream sequences of the genes (Fig 1D), while CG and CHG methylation levels in genes and TEs were similar among different tissues tested (S2 Fig). Among all TEs, mean CHH methylation levels were higher in the fiber but indistinguishable between the ovule and leaf (Fig 1E). The higher CHH methylation levels of TEs in fibers and ovules were correlated with the closer distances of TEs to the gene (Fig 1F). When TEs were more than 4-6-kb away from the gene, CHH methylation levels became similar between the ovule and leaf. Consistently, the CHH methylation levels mirrored the 24-nt siRNA distribution patterns, which were high near the genes and decreased as they were further away from the gene, indicating the role of the RdDM pathway in CHH methylation of the gene nearby TEs (Fig 1G).
As CG and CHG methylation levels were similar among different tissues examined, further analysis was focused on the CHH methylation changes among these tissues. We predict that differentially methylated regions (DMRs) between tissues play a role in biological function. To test this, we identified 39,668 CHH-hypermethylated DMRs in the ovule relative to leaf (ovule-leaf) (OL CHH-hyper DMRs) and 124,681 CHH-hypermethylated DMRs in the fiber relative to ovule (fiber-ovule) (FO CHH-hyper DMRs) (S2 Table). A subset of these DMRs was validated by bisulfite-sequencing individually cloned genomic fragments (S3 Fig), confirming the results of these DMRs based on the genome-wide analysis. Compared to distributions of exons and introns in genomic distribution, both OL and FO CHH-hyper DMRs were underrepresented in exons and introns (Z-score <-5) (Fig 2A). However, OL CHH-hyper DMRs were overrepresented in intergenic regions (62%) relative to the average genomic distribution (40%), whereas FO CHH-hyper DMRs were enriched in TEs (80% vs. 50%). Among TEs, short TEs (less than 1-kb) were more abundant in OL CHH-hyper DMRs (74%) than in the genomic distribution (30%), whereas the percentage of short TEs in FO CHH-hyper DMR distribution was similar to the genomic distribution (Fig 2B). This suggests that OL hyper-CHH regions were located nearby genes (between genes with short TEs), whereas FO hyper-CHH regions are associated with TEs (genome-wide phenomenon).
Methylation levels of most OL CHH-hyper DMRs were higher in fibers than in ovules, indicating that OL CHH-hyper DMRs continue to be hypermethylated in elongating fibers after they developed from ovule epidermal cells (P < 1e-200, Wilcoxon rank-sum test) (Fig 2C). Interestingly, among the OL CHH-hyper DMRs, siRNA expression levels were significantly higher in ovules than in leaves (P < 1e-100, Wilcoxon rank-sum test), and this difference was not obvious between the fiber and ovule in FO CHH-hyper DMRs (Fig 2D). These data suggest that siRNAs induce CHH hypermethylation in the ovule, whereas the CHH methylation increase in the fiber was independent of the RdDM pathway. In addition to CHH methylation, OL CHH-hyper DMRs had slightly higher CG and CHG methylation levels in ovules (CG: 0.95; CHG: 0.77) than in leaves (CG: 0.94, CHG: 0.73), FO CHH-hyper DMRs also had higher CG and CHG methylation levels in fibers (CG: 0.96; CHG: 0.89) than in ovules (CG: 0.94; CHG: 0.85) (Fig 2E).
To test the role of methylation changes in gene expression, we compared RNA-seq data with DMRs between the ovule and leaf and/or between the fiber and ovule (S1 Table). Down-regulated genes in the ovule relative to the leaf were significantly enriched in the genes overlapping with OL CHH-hyper DMRs in the gene body (Fig 3A and S3 Table), which is consistent with the notion that CHH methylation in the gene body correlates negatively with gene expression [27, 36]. However, up-regulated genes in the ovule were significantly enriched in the genes overlapped with OL CHH-hyper DMRs in the upstream sequences (1 kb) (Fig 3A and S3 Table), indicating a positive correlation of CHH methylation in the upstream sequences with gene expression. For the genes that were preferentially expressed in the ovule (ovule-preferred genes), the CHH methylation level in the flanking sequences was higher in the ovule than in the leaf, whereas CHH methylation in flanking sequences of the leaf-preferred genes was similar between the ovule and leaf (Fig 3B). To further test the relationship between CHH methylation and gene expression, we divided genes into quartiles based on their expression levels in each tissue (from low to high) and compared them with DNA methylation changes. CHH and CHG methylation levels in the gene body were negatively associated with gene expression levels, and moderately transcribed genes showed the highest CG methylation level in the gene body (Figs 3C and S4), consistent with previous findings [27, 36]. Moreover, CHH methylation in the 5’ upstream coincided with siRNA abundance, which was positively associated with gene expression (Fig 3C and 3D). The siRNA levels in the promoter regions of ovule-preferred genes were significantly higher in the ovule than in the leaf (P < 1e-4, Wilcoxon rank-sum test) (Fig 3E). This positive correlation between CHH methylation and gene expression was unexpected but was also reported in other plants including maize [28], soybean [26], and rice [37]. Active transcription could induce hypermethylation, as shown in rice [37]. Alternatively, hypermethylation may induce gene expression.
The higher methylation levels of most OL CHH-hyper DMRs in fibers than in ovules may suggest a role for increased methylation in the gene expression of fibers. To test this, we extracted 26,245 OL CHH-hyper DMRs that showed higher methylation levels (cut-off value >0.05) in fibers than in ovules to examine effects of these DMRs on gene expression in fibers. In contrast to the up-regulated genes in the ovule, down-regulated genes (352) in the fiber relative to the ovule were significantly (Hypergeometric test, P < 0.05) enriched in those that overlapped with the OL CHH-hyper DMRs in the upstream sequences (Fig 4A), suggesting that further increasing methylation levels of some OL CHH-hyper DMRs in the fiber could repress nearby genes. For example, expression of GhMYB25L_D (Gh_D12G1628), a key player for fiber development [24], was induced at the fiber initiation stage (0 DPA) but repressed in the elongation stage (Fig 4B). GhMyb25L_D is flanked by three short TEs corresponding to CHH methylation (boxed). The methylation levels in the flanking TEs of GhMyb25L_D increased in ovules at 0 DPA and continued to increase in fibers at 14 DPA. These data indicate that additional CHH methylation in elongating fibers may be correlated with inhibition of GhMYB25L_D expression during fiber development (Fig 4B).
To our surprise, FO CHH-hyper DMRs were not significantly correlated with expression changes in the fiber (Fig 4C and S4 Table). This was partly because FO CHH-hyper DMRs tended to be excluded from the gene body (Fig 2A) and from the flanking sequences (1-kb) of the gene (Fig 4D). Instead, FO CHH-hyper DMRs were enriched in TEs (Fig 2A), and FO CHH hypermethylation correlated positively with the TE density (S1D Fig). In plants, TE transcripts are generally repressed through DNA methylation. We examined whether CHH hypermethylation in heterchromatin in fiber inhibited TEs. Indeed, RNA-seq data showed that more TEs were expressed in fibers than in ovules or leaves (Fig 4E). These data suggest that endoreduplication in early fiber cells may loosen chromatin structure, leading to activation of TEs. More transcription could induce hypermethylation. As a result, CHH hypermethylation in fibers may serve as a feedback mechanism to repress TEs. Although both ovule and fiber had more 24-nt siRNAs than the leaf, only fiber showed the CHH methylation increase in heterochromatin compared to the leaf, suggesting that higher heterochromatic CHH methylation in the fiber is partially related to siRNA abundance.
This CHH methylation increase in the heterochromatin could be catalyzed by CMT2 through a DDM1-mediated process [10, 11]. To test this, we examined expression of GhCMT2 (Gh_D08G1665, Gh_A08G1371), GhDDM1 (Gh_A12G0603, Gh_D12G2782) and other genes related to DNA methylation in leaves, ovules and fibers. Compared with leaves, GhCMT2, GhCMT3 (Gh_A07G0385, Gh_D07G0449) and GhMET1 (Gh_A05G3224, Gh_D04G0381) were upregulated in ovules and fibers, and the genes involved in the RdDM pathway, including DRM1 (Gh_A09G0264, Gh_D04G0527), DMR2 (Gh_D09G0266, Gh_A05G3114), RDR2 (Gh_D12G2624, Gh_A12G2496), NRPD1 (Gh_A08G1605, Gh_D08G1916), and NRPE1 (Gh_A06G1549, Gh_D06G1919), with an exception of AGO4 (Gh_A08G1752, Gh_D08G2707) and DLC3 (Gh_D06G0845), were up-regulated in ovules but not in fibers (Fig 4F). However, HISTONE H1 (Gh_D08G0660, Gh_A08G0565), which is shown to impede the access of DNA methyltransferases to the heterochromatin [10, 38], was repressed in fibers but not in ovules (Fig 4F). These data suggest that up-regulation of GhCMT2 and repression of cotton HISTONE H1 in fibers may induce CHH hypermethylation in TEs and heterochromatin in fibers. To investigate the biological role of DNA methylation in fiber development, we harvested ovules at -3 to 0 days post anthesis (DPA) from the allotetraploid cotton Gossypium hirsutum L. acc. TM-1 and cultured them in vitro with or without treatments of the DNA methyltransferase inhibitor, 5-aza-deoxycytidine (5-aza-dC) [39]. Consistent with the hypermethylation observed in ovules and fibers, the 5-aza-dC treatment suppressed fiber length and ovule size in the initiation and elongation stages (Fig 4G, 4H and 4I). The data suggest that DNA methylation might be important for normal fiber and ovule development, although we cannot rule out possible side effects of growth inhibition by 5-aza-dC.
In the gene body, CG methlylation levels of most homoeologous genes were relatively equal (Fig 5A) and did not correlate with expression levels of homoeologous genes (Fig 5B). However, in fibers, we identified 539 pairs of A and D homoeologs that were differentially methylated at CHG and CHH sites in the gene body (Fig 5A). The methylation levels of CHG or CHH methylation in the gene body were significantly anti-correlated with expression levels of A and D homoeologs (Fig 5B and S5 Table). Among the genes that were hyper-methylated in the A relative to D homoeologs, nearly three times of D homoeologs were expressed at higher levels than the A homoeologs. Likewise, among the genes that were hyper-methylated in the D relative to A homoeologs, more than twice of A homoeologs were expressed at higher levels than the D homoeologs. For example, A02G0163 was methylated more than D02g0204, and D02G0204 was expressed more than A02G0163 (Fig 5C). In the promoter regions, CG, CHG and CHH methylation levels of most homoeologous genes were relatively equal and did not show a significant correlation with gene expression (S5 Fig). Together, these data suggest a repressive role of CHG and CHH methylation in the gene body for the expression changes of homoeologous genes, which may contribute to fiber selection and improvement in the allotetraploid cotton.
We found little variation of CG and CHG methylation among cotton tissues, which is consistent with the data in maize, showing more variable CG and CHG methylation levels among genotypes than among tissues [40]. Although the overall level of CHH methylation is low compared with that of CG and CHG methylation in the genome, CHH methylation is associated with plant growth and development [41]. In Arabidopsis intraspecific hybrids, CHH methylation changes are associated with the parent-of-origin effect on circadian clock gene expression and biomass heterosis [41]. CHH methylation is mainly produced by 24-nt siRNAs through the RdDM pathway [1, 2]. In cotton, 24-nt siRNAs were enriched in gene-rich regions, which is similar to that in maize [28] and sugar beet [42] but different from that in Arabidopsis [43] and soybean [26]. Enrichment of siRNAs in euchromatic regions is inconsistent with the overall repeat abundance in the genome because the repeat amount is higher in soybean (~59%) [44] than in sugar beet (~42%) [42], but much lower than in maize (~85%) [45]. It is the TE property and location not the TE density that determines location of the siRNA abundance and CHH methylation distribution. These 24-nt siRNAs are derived from short TEs close to the genes, inducing CHH methylation flanking the genes, which is known as “CHH methylation island” in maize [28]. However, in spite of similar siRNA distribution patterns between maize and cotton, CHH methylation patterns are different. In maize, more siRNAs in gene-rich regions than TE-rich regions are associated with higher CHH methylation levels in gene-rich regions. In cotton, although gene rich regions also showed higher siRNA levels, CHH methylation level is lower in gene-rich regions than TE-rich regions (Fig 1C). Furthermore, the percentage of methylcytosines in the CHH context is higher in cotton (~7.8% in leaves) than in maize (~5% in ears and shoots), which is in contrast to a higher TE density in maize than in cotton. These results indicate that in addition to CHH methylation that is produced by the RdDM pathway in cotton as in maize [10], cotton has another active pathway, which is absent in maize, to generate CHH methylation as shown in Arabidopsis [10].
This additional CHH hypermethylation in fibers is likely mediated by CMT2 and DDM1 [10]. The genes overlapping with FO CHH-hyper DMRs were not enriched in the differentially expressed genes, which is consistent with the finding that non-CG methylation generated by CMT2 does not regulate protein-coding genes [11]. The CHH hypermethylation in cotton fibers is probably promoted by concerted induction of GhCMT2 and repression of HISTONE H1. In ovules, GhCMT2 is also induced but HISTONE H1 is not repressed. As a result, CHH methylation was not enhanced in the heterochromatin of ovules, suggesting that chromatin changes by histone H1 is required for promoting CHH hypermethylation in heterochromatin between the fiber and ovule. Fibers undergo endoreduplication in early stages and rapid cell elongation and cellulose synthesis in late stages, which may change chromatin structures in some regions [19, 20]. Indeed, more TEs are active in fibers than in ovules and leaves (Fig 4E), and more siRNAs were generated in fibers than other tissues [15]. The CHH hypermethylation induced in the fiber could function as a feedback mechanism to repress TEs. However, TEs are more expressed in fibers than in leaves, suggesting a paradox of the requirement of transcription for silencing, as previously noted [46]. Consistent with the requirement of DNA methylation for cotton fiber development, inhibiting DNA methylation by aza-dC not only reduces fiber cell initials but also slows down fiber cell elongation (Fig 4G, 4H and 4I). A confirmatory experiment is to generate the transgenic plants that suppress CHH methylation and directly test the CHH methylation effect on fiber development.
Notably, the highly expressed genes are correlated with higher CHH methylation levels in promoter regions close (1-kb or less) to the transcription start site (TSS), which is unexpected but has been documented in maize [28], soybean [26], rice [37], and now in cotton (Fig 3). One possibility is that TE activation enhances expression of nearby genes. Transcription initiates from TEs through Pol II or Pol IV (a homolog of Pol II) and then spreads to nearby genes [2, 47]. While most TEs are transcriptionally silenced in plant genomes, some TEs could activate nearby genes, as reported in Arabidopsis [46, 48] and rice [49]. If TE activation occurs prior to the transcription of nearby genes, TE expression should not be correlated with the distance between TEs and genes. However, more 24-nt siRNAs are present in the TEs closer to the genes (Fig 1G), suggesting another possibility that gene expression induces activation of nearby TEs. This may contribute to positive correlation between CHH methylation in the promoters and gene expression (Fig 6). The regions near the TSS are probably in open chromatin formation to allow active transcription of both genes by RNA polymerase II and short TEs by RNA polymerase IV [2, 47]. This leads to high abundance of small RNAs near the promoters and transcripts from corresponding genes, as observed in cotton, maize [28], and soybean [26]. The siRNAs can induce CHH methylation through the RdDM pathway. A recent study in rice showed that phosphate starvation induces DNA methylation of the TEs close to highly induced genes [37]. The methylation changes occur in the CHH context but are largely independent of the canonical RdDM pathway. Moreover, the methylation is increased after nearby genes are induced, suggesting that stress-induced gene expression promotes DNA methylation in the nearby TEs.
Consistent with this notion, ovule up-regulated genes are enriched in those overlapping with OL CHH-hyper DMRs in the flanking sequences. However, further CHH methylation of OL CHH-hyper DMRs in fibers represses nearby genes. This suggests that CHH methylation in promoters may act as a feedback mechanism to regulate these genes during ovule and fiber development (Fig 6). The spatiotemporal role of DNA methylation in expression changes of ovule- and fiber-related genes could also explain why overexpressing these genes may result in the unexpected outcome of fiber traits [24] because appropriate methylation patterns of the transgenes are not established. Together, the results suggest a functional role of CHH methylation during ovular and fiber development.
Finally, non-CG methylation in the gene body is associated with the expression bias of homoeologous genes in the allotetraploid cotton, providing the unique evidence for epigenetic regulation of nonadditive expression of homoeologous genes in polyploid species [50, 51]. Together, the spatiotemporal role of DNA methylation in developmental regulation and intergenomic interactions provides a conceptual advance, which could be translated into genomic improvement of polyploid plants, including most important crops that provide us with food (wheat), fiber (cotton), and oil (canola).
Gossypium hirsutum acc. TM-1 grew in the greenhouse at The University of Texas at Austin. Leaves and bolls (0 and 14 DPA) were harvested with three biological replications. Ovules and fibers were carefully dissected from cotton bolls at 14 DPA and immediately frozen in liquid nitrogen for RNA and DNA preparation.
Total RNA was isolated from cotton leaves, ovules at 0 DPA and fibers at 14 DPA using Plant RNA Reagent (Life Technologies). By polyacrylamide gel electrophoresis, RNAs corresponding to ~15 to 30-nt in length from 20 μg total RNA were excised and eluted from the gel using 0.3 M NaCl. Small RNAs were precipitated using ethanol and dissolved in 6 μl RNase free water. Small RNA-seq libraries with three biological replicates were constructed using NEBNext® Multiplex Small RNA Library Prep Set (NEB, Ipswich, Massachusetts) according to the manufacturer’s instructions. Small RNA-seq libraries were single-end sequenced for 50 cycles.
After DNase treatment, total RNA (~5 μg) was subjected to construct strand-specific mRNA-seq libraries with two biological replications using NEBNext® Ultra™ Directional RNA Library Prep Kit (NEB, Ipswich, Massachusetts) according to the manufacturer’s instructions. mRNA-seq libraries were single-end sequenced for 150 cycles.
Genomic DNA was isolated from cotton leaves, ovules at 0 DPA and 14 DPA, and fibers at 14 DPA using CTAB method [52]. Total genomic DNA (~5 μg) was fragmented to 100–1000 bp using Bioruptor (Diagenode, Denville, New Jersey). End repair (NEBNext® End Repair Module) was performed on the DNA fragment by adding an 'A' base to the 3'end (NEBNext® dA-Tailing Module), and the resulting DNA fragment was ligated to the methylated DNA adapter (NEXTflex™ DNA Barcodes, Bioo Scientific, Austin, Texas). The adapter-ligated DNA of 200–400 bp was purified using AMPure beads (Beckman Coulter, Brea, California), followed by sodium bisulfite conversion using MethylCode™ Bisulfite Conversion Kit (Life Technologies, Foster City, California). The bisulfite-converted DNA was amplified by 12 cycles of PCR using LongAmp® Taq DNA Polymerase (NEB, Ipswich, Massachusetts) and purified using AMPure beads (Beckman Coulter, Brea, California). The Paired-End sequencing of the MethylC-seq libraries was performed for 101 cycles.
After DNase treatment, total RNA (2 μg) was used to produce first-strand cDNA with the Omniscript RT Kit (Qiagen, Valencia, California). The cDNA was used as the template for qRT–PCR using FastStart Universal SYBR Green Master (Roche, Indianapolis, Indiana). The reaction was run on the LightCycler® 96 System (Roche, Pleasanton, California). The relative expression level was quantified using internal control cotton HISTONE H3 [53].
After adapter clipping, small RNA-seq reads (18–30 nt) were mapped to the TM-1 genome sequence [34] using Bowtie2 settings (-k 100—score-min L,0,0), allowing no mismatches. Multi-mapped small RNA-seq reads were evenly weighted and assigned to all locations as previously described method [54].
mRNA-seq reads were mapped to the TM-1 genome sequence using Tophat settings (—library-type fr-firststrand —b2-score-min L,0,-0.2) [34, 55]. Uniquely mapped reads were extracted and analyzed by Cufflinks to determine gene and TE abundance using annotated genes and TEs [56]. The differentially expressed genes (DEGs) were identified using both the fold-change (>2-fold) and ANOVA tests (P < 0.01).
We applied Bismarck software to align reads to the TM-1 genome sequence with default parameters (—score_min L,0,-0.2 -X 1000—no-mixed—no-discordant) [34, 57]. In brief, the first 75 bases of unmapped reads were extracted and realigned to the genome. Only reads mapped to the unique sites were retained for further analysis. Reads mapped to the same site were collapsed into a single consensus read to reduce clonal bias. For each cytosine, the binomial distribution was used to identify whether this cytosine was methylated. The probability p in binomial distribution B (n, p) was referred to bisulfite conversion failure rate. The number of trials (n) in the binomial distribution was the read depth. Only the cytosines covered by at least three reads in all compared tissues were considered for further analysis. Cytosines with P-values below 1e–5 were identified as methylcytosines.
Differentially methylated regions (DMRs) for CHH methylation were identified using 100-bp sliding-windows. Mean methylation level was calculated for each window [5]. Windows containing at least eight cytosines in the CHH context covered by at least three reads were selected for identifying DMRs. ANOVA test using 2 biological replications (P < 0.05) and difference of methylation levels (cut-off value > 0.1 between two compared samples) were used to determine CHH DMRs between two compared tissues. The cut-off value was set at 0.05 for the comparison between ovule and fiber methylation levels in OL CHH-hyper DMRs.
The fold enrichment of DEGs in DMR-overlapping genes was calculated as (#DEGsDMR-overlapping /#DEGstotal) / (#Genes DMR-overlapping /#Genestotal), and P-values were generated using the hypergeometric test.
Protein sequences of A homoeologs were aligned to protein sequences of D homoeologs using blastp with an E-value less than 1e-10. Alignment information was used by MCScanx to identify homoeologous genes [58] (Score > 2000 and E-value < 1e-10).
We first randomly extracted 1000 of 1-kb regions in the genome for 500 times. Percentage of DMR-overlapping regions in 1000 random regions was calculated at each time. Z-scores were calculated as (x-μ)/σ where x was percentage of DMR-flanking genes in all genes, μ and σ were mean and standard deviation of percentages of DMR-overlapping regions in random loci.
Ovules were removed from flower buds at -3 or 0 DPA. The ovules were sterilized with 75% alcohol, washed with autoclaved water, and cultured in Beasley-Ting (BT) liquid media (20 ml) containing 5 μM indole-3-acetic acid (IAA) and 0.5 μM gibberellic acid3 (GA3) at 30°C for 4–14 days in dark [17, 59]. Treatment of 5-aza-2’-deoxycytidine (aza-dC, 10 mg/L) was applied in the BT liquid media.
The data reported in this paper have been deposited in the Gene Expression Omnibus (GEO) database, www.ncbi.nlm.nih.gov/geo (accession no. GSE61774).
|
10.1371/journal.pntd.0002137 | Chikungunya Virus-associated Long-term Arthralgia: A 36-month Prospective Longitudinal Study | Arthritogenic alphaviruses, including Chikungunya virus (CHIKV), are responsible for acute fever and arthralgia, but can also lead to chronic symptoms. In 2006, a Chikungunya outbreak occurred in La Réunion Island, during which we constituted a prospective cohort of viremic patients (n = 180) and defined the clinical and biological features of acute infection. Individuals were followed as part of a longitudinal study to investigate in details the long-term outcome of Chikungunya.
Patients were submitted to clinical investigations 4, 6, 14 and 36 months after presentation with acute CHIKV infection. At 36 months, 22 patients with arthralgia and 20 patients without arthralgia were randomly selected from the cohort and consented for blood sampling. During the 3 years following acute infection, 60% of patients had experienced symptoms of arthralgia, with most reporting episodic relapse and recovery periods. Long-term arthralgias were typically polyarthralgia (70%), that were usually symmetrical (90%) and highly incapacitating (77%). They were often associated with local swelling (63%), asthenia (77%) or depression (56%). The age over 35 years and the presence of arthralgia 4 months after the disease onset are risk factors of long-term arthralgia. Patients with long-term arthralgia did not display biological markers typically found in autoimmune or rheumatoid diseases. These data helped define the features of CHIKV-associated chronic arthralgia and permitted an estimation of the economic burden associated with arthralgia.
This study demonstrates that chronic arthralgia is a frequent complication of acute Chikungunya disease and suggests that it results from a local rather than systemic inflammation.
| Chikungunya virus (CHIKV) is transmitted to human by mosquitoes. It is a re-emerging virus that has a risk to spread globally, given the expanding dissemination of its mosquito vectors. Chikungunya disease is characterized by acute transient febrile arthralgic illness, but can also lead to chronic incapacitating arthralgia. We have conducted a prospective longitudinal study to investigate in details long-term outcome of CHIKV infection. We found that 60% of patients experienced arthralgia 36 months after the onset of acute disease. Arthralgia affected most often multiple sites and were usually incapacitating. In addition to arthralgia, many patients suffered from myalgia and cutaneous lesions and several cognitive dysfunctions. We also showed that age over 35 years and the presence of arthralgia 4 months after the onset of disease are risk factors for long-term arthralgia. Patients with long-term arthralgia did not display biological markers typically found in autoimmune or rheumatoid diseases. This study demonstrates that chronic arthralgia is a frequent complication of acute Chikungunya disease and suggests that it results from a local rather than systemic inflammation.
| Chikungunya virus (CHIKV) is an arthropod-borne virus that belongs to the Alphavirus genus. Chikungunya disease is characterized by polyarthralgia, sometimes associated with rash. The articular symptoms, often debilitating, usually resolve within weeks, but have been reported to last for months, even though the natural history of this infection has not been precisely studied in prospective studies [1], [2], [3].
In 2005, CHIKV emerged in islands of Indian Ocean including La Réunion, a French overseas department, and approximately one third of the inhabitants (i.e. ∼300,000) was infected at the end of the outbreak in 2006 [4], [5]. Compared to earlier outbreaks, this episode occurred in a highly medicalized area. Moreover previously unreported severe forms of Chikungunya were observed, such as encephalopathy [6], [7], and mother-to-child CHIKV transmission was demonstrated, leading to severe neonatal CHIKV infection [5]. In the wake of this outbreak, CHIKV also re-emerged in India with over 1 million cases [8], [9]. In 2007, CHIKV emerged for the first time in Europe, causing an outbreak in Italy [10].
We have described the clinical and biological features of acute CHIKV infection in a prospective cohort of patients with positive blood CHIKV RT-PCR [11]. It included all patients referred to the Emergency Department in Saint-Pierre de la Réunion with febrile arthralgia between March and May 2006. As little is known about long-term outcome of CHIKV infection, we conducted a prospective longitudinal study to describe in details the specific clinical and biological features of chronic arthralgia, as well as clinical signs associated to this pathology. We evaluated the consequences of long-term arthralgia on patients' daily and social life, looked for risk factors associated with them and estimated their economic impact.
Ethical clearance was obtained from the «Comité de Protection des Personnes Sud-Ouest et Outre-Mer III» of Saint-Pierre, La Réunion, Paris (CCP 2008/65, n° 2008-A00999-46). CHIK-IMMUNOPATH received approval from the ethical committee for studies with human subjects (CPP) of Bordeaux and the National Commission for Informatics and Liberty (CNIL). Written informed consent was obtained from patients included in the CHIK-IMMUNOPATH study. The study respects the STROBE statement (supporting information S1).
We studied a cohort of patients (n = 180) enrolled for febrile arthralgia to the Emergency Department of the Groupe Hospitalier Sud Réunion between March 2005 and May 2006 [11]. Patients were interviewed by telephone 4, 6, 14 and 36 months (M4, M6, M14 and M36) after the viremic phase, using the same questionnaire as that used at day 0 (D0) (supporting information S2). At M36 after the acute phase, all patients who agreed to participate to a complementary study (CHIK-IMMUNOPATH) and who were arthralgic were interviewed and underwent clinical examination. Among them, 22 patients with arthralgia (ART+) and 20 patients without persisting arthralgia (ART−) were randomly selected from the cohort. They signed a written consent and a blood sample was collected.
Blood cell count was performed and viremia was tested by qRT-PCR [12]. Both serum anti-CHIKV IgG and IgM specific antibodies were screened. An enzyme-linked immunosorbent assay (ELISA) was performed with CHIKV antigen [12]. The avidity of anti-CHIKV IgG was tested by ELISA in the presence or absence of urea 8 M [13]. Geometric Mean Antibody Titer (GMAT) was calculated as previously described [14].
Plasmatic protein electrophoresis was performed and C-reactive protein (CRP) concentration was measured. The presence of anti-nuclear, anti-dsDNA, anti-endomysium autoantibodies, anti-cyclic citrullinated peptide antibody (ACCP) and cryoglobulinemia were investigated. Samples were transported at 37°C to research cryoglobulins. Sera were sent to Myriad RBM (Austin,Texas) and analyzed by Luminex using the inflammation MAP. Assays are run according to CLIA guidelines and in all cases, >100 beads per analyte were measured with CV <10% for values that are above the limit of quantification for the given assay.
For each time point, the proportion of patients with monoarthralgia (1 site), oligoarthralgia (2–3 sites) or polyarthralgia (4 sites or more) were compared using a Chi-2 test.
A logistic regression model was used to identify factors associated with long-term arthralgia, defined as presence of arthralgia at M14. We studied factors at D0, including demographic factors (gender and age), biological markers, hospitalization and comorbidities, and factors measured at M4 (arthralgia, treatment, and quality of life). All factors associated with long-term arthralgia with a p-value<0.15 in univariate analysis were entered in the multivariate model. A step-by-step backward procedure was then used to identify factors significantly associated with long-term arthralgia. A sensitivity analysis was also conducted, following the same procedure, defining long-term persistence of arthralgia as the presence of arthralgia at M36. To address the lost of follow up at M36, we looked for parameters that differentiate the patients who were lost between M14 and M36 and these who were followed.
Statistical analyses were performed using the STATA software (Stata Corporation, College Station, Texas, USA); all significance tests were two-sided and p-values<0.05 were considered significant. Luminex data were mined using the Omniwiz software (Biowisdom) and Mann-Whitney analysis is reported. False discovery rate (FDR or q-values) were calculated as correction for multiple analyte testing.
To characterize the spatiotemporal evolution of arthralgia, we considered different types of arthralgia. For a given site at a given time point, arthralgia was defined either as a “persistent symptom” if the affected site was the same as that reported during the previous time point, as “a relapse of an the acute symptom” if the site affected was the same as that at the acute phase, or as “a new symptom” if this site was not affected at the acute phase, or as a “migrating symptom” if the arthralgia was localized at a site distinct from that previously reported. For a given patient, these categories were not mutually exclusive. For modeling migratory arthralgia, we divided joints into three groups: upper limb joints (hand, wrist, elbow), mid body joints, (shoulder, spine, hip) and lower limb joints (foot, ankle, knee), and considered two migratory probabilities for migration to sites within to the same group (e.g., hand to wrist) or to a different group (e.g., hand to foot) (Supporting information S3 and Table S1).
At M4, M6, M14 and M36 after their inclusion as acute CHIKV-infected patients, all patients were interviewed using a questionnaire to monitor persistence of arthralgia, other clinical signs and treatments. The number of patients that participated is provided in Figure 1. There is an important lost of follow up between M14 and M36 however we did not identify a bias associated with it. Among the 180 patients, 76 patients were followed at all time points of the study.
The percentage of patients suffering from long-term arthralgia decreased after CHIKV acute infection and stabilized around 60% (Figure 2A). Of note, all patients suffered from arthralgia at D0. Among them, only 5 on 180 (2.8%) suffered from joint pain prior to CHIKV infection.
Most patients had intermittent arthralgia, with recovery and relapse. For each time point, 25 to 40% of patients complained of permanent arthralgia. Among the 76 patients that could be followed at each time point, 45% had arthralgia at all time, 24% experienced partial recovery at M4, M6 or M14 then relapses, and 31% fully recovered from acute symptoms. Among patients who experienced chronic symptoms at M36, 43.5% reported arthralgia triggered by a change in ambient temperature, 8% by physical effort. At M36, arthralgia caused stiffness in 75.5% of patients with symptoms, and 67.7% of the patients reported a need of morning stretching (time of 32 minutes, standard deviation (SD) 37 minutes, range 5–180 minutes).
We monitored arthralgia in 9 anatomical sites (Figure 2B). Arthralgia in upper limbs mostly affected fingers and wrists, while lower limbs arthralgia mostly affected knees and ankles. At each time point, these locations remain significantly the most affected (Mac Nemar test for matched pairs of subjects). Importantly, arthralgia were typically symmetrical (90%).
We then investigated whether the number of arthralgic sites diminished in patients still suffering from arthralgia among the 76 patients followed at all time points. The number of arthralgia sites decreased until M14, with only 30% of patients suffering from polyarthralgia (number of arthralgia sites >2) (Figure 2C). Despite an increase of arthralgic sites at M36, there is an overall significant decrease of the number of painful joints during the study period (p<0.01).
We attempted to model the spatiotemporal evolution of arthralgia, as defined in the Materials and Methods section. We found that “persistent” symptoms had the strongest effect, as its probability of occurrence was three times higher than a “new” symptom (Figure 2D). The probability of relapse of an “acute” symptom was twice the appearance of a “new” symptom. Finally, “migratory” symptoms tended to be intra-group as compared to inter-group migrations.
Patients with arthralgia at M36 showed other clinical symptoms, including local swelling, cutaneous symptoms, myalgia and osteoligamentaous pain (Table 1). Local swelling localized to affected joints for 63% of patients. Moreover, sleep, memory or concentration disorders and asthenia or depression are significantly associated with arthralgic patients.
The proportion of patients with arthralgia who attended a physician or received a treatment significantly increased between M4 and M36 (p = 0.01) (data not shown), and reached 80% (Table 1). Similarly, the number of patients receiving a treatment increased and these treatments are statistically associated with the arthralgic status of the patient (p<0.001).
Arthralgia in patients at M36 were highly incapacitating for daily life tasks, professional life and spare-time activities (Table 2).
To identify risk factors associated with long-term arthralgia, we performed univariate and multivariate statistical analyses at M14, as the participation was higher than at M36 (Table 3). Gender was not associated with long-term arthralgia, age less than 35 years was protective. Risk of arthralgia was not associated with indicators of the disease severity during the acute phase (viral load, duration of hospitalization or number of sites of arthralgia at D0) [11], however it was weakly associated with C-reactive protein (CRP) level at D0. Diabetes was the only comorbidity found to be a risk factor for long-term arthralgia. Interestingly, arthralgia at M14 was strongly associated with arthralgia at M4, and even more if arthralgia was permanent at M4. Memory and concentration disorders at M4 were also identified as risk factors for developing long-term arthralgia. Arthralgia, memory disorders and concentration disorders at M4 were the only risk factors significantly and independently associated with long-term arthralgia. When long-term arthralgia was assessed at M36, results were very similar.
At M36, 22 patients with arthralgia (ART+) and 20 patients without arthralgia (ART−) were randomly selected from the cohort to participate to the CHIK IMMUNOPATH study. Its aim was to titrate anti-CHIKV antibodies and identify a serum inflammatory or autoimmune signature associated with the arthralgia phenotype.
All patients were negative for CHIKV RT-PCR, and exhibited anti-CHIKV IgGs in serum, while a minority (9.5%) harbored measurable levels of anti-CHIKV IgM. The activity of CHIKV IgG (GMAT) was significantly higher in ART+ patients (30) than in ART− patients (20), but antibody avidity was comparable in both groups (mean ±SD: 31,6±20,4 in ART + patients and 33,7±19,8 in ART− patients). Although lymphopenia is a defining feature of acute CHIKV disease [11], it was a rare finding at M36 (data not shown). Plasma protein levels measured by electrophoresis and CRP concentration were within normal ranges. However, CRP levels were significantly higher in the ART+ group than in the ART− group (mean ±SD: 3.35±3.00 mg/ml and 1.85±2.49 mg/ml, respectively; p = 0.04).
We used Luminex xMAP technology to assay analytes in the serum of patients. Most analytes were undetectable in both groups of patients (Table 4). Five inflammation markers were significantly elevated in ART+ patients: factor VII, C3 complement component, IL1α, IL15 and CRP (Figure 3). Ferritin level was significantly lower in ART+ patients than in ART− patients. These markers did not allow for the identification of a subgroup within the ART+ group, nor did they correlate one with another. No autoimmune marker and no anti-DNA antibody in the serum of ART+ patients were detected, although anti-nuclear antibodies were detected at low level in four ART+ patients. Three patients had elevated anti-nuclear antibodies, one in the ART+ group and two in the ART- group.
As it has been reported that CHIKV could evolve into rheumatoid arthritis [15], we screened for cyclic citrullinated protein antibodies. We also assayed for cryoglobulinemia and anti-endomysium IgA antibodies. All patient were found to be negative.
We estimated the annual economic burden of long-term arthralgia by taking into account the cost of medical visits, therapeutic treatment and the cost for lost work time due to injury or pain (using the population of La Réunion Island as a reference) (Table S2). We found that arthralgia secondary to the CHIKV outbreak in La Réunion in 2005-06 has resulted so far in an estimated total cost of up to 34 millions euros per year. This corresponds to 250€ per year and per patient with long-term arthralgia. However, it should be noted that this sum might be overestimated due to the bias in our cohort selection, as our cohort is likely composed of the most severely affected patients who were referred to the hospital during the acute phase.
Our study is the first prospective cohort study on CHIKV long-term arthralgia that is based on the follow-up of patients who presented with acute CHIKV infection as the inclusion criterion. This study is also the first to define the evolution of CHIKV-induced arthralgia, mapping the frequency and location of arthralgic sites during a three year time period. We have also investigated the impact of CHIKV-chronic arthralgia on daily life of patients, identified clinical signs associated with arthralgia, and analyze biologic markers. Moreover we have evaluated associated risk factors and estimated the economic burden of this disease. Together, these data allow us to define the features of CHIKV-induced chronic arthralgia (Table 5), as compared to other viral arthritis [16], and to establish a detailed understanding of the public health problem resulting from CHIKV-chronic arthralgia.
Our data reveal that more than 60% of CHIKV-infected patients suffer from arthralgia, 36 months after acute infection. This high percentage of patients with long-term symptoms was also reported by other studies of Italian cohorts and French cohorts of La Réunion Island or metropolitan France [17], [18], [19], [20], [21] but is dramatically higher than documented in India and Senegal [1], [22], [23], [24], [25]. While this discrepancy may result from particular features of the CHIKV strain responsible for the La Réunion outbreak, data from Italy following the 2007 outbreak resulted from a CHIKV strain more closely related to the viral strain present in India [10], with more than 60% of CHIKV patients in Italy having reported myalgia, asthenia or arthralgia 12–13 months after the acute disease [20]. Alternatively, reported differences may be a result of different genetic backgrounds of these populations. As joint pain is considered a subjective symptom, it might also reflect a difference in pain threshold of patients or reporting from physicians, thus reflecting differences in health care practices.
Long-term CHIKV-associated arthralgia were mainly symmetrical, involving more than 2 different joints. Hand, wrist, ankle and knee were found to be the most affected, consistent with other studies [17], [18]. Moreover, 60–80% of patients had relapsing arthralgia, while 20–40% had unremitting arthralgia. While some patients reported “migrating” arthralgia, most disease symptoms mapped to joints that were most painful during acute Chikungunya disease. Thus, symptoms at the chronic phase may be indirectly associated to virus replication at the time of acute infection [26], [27], [28].
In addition to arthralgia, many patients suffered from myalgia and cutaneous lesions and several cognitive dysfunctions. Although study patients did not display neurological symptoms at the acute phase of disease, we cannot exclude that cognitive dysfunctions result from CHIKV spread in the CNS, as it has been reported that CHIKV disseminates to the CNS in humans and in animal models [12], [26], [27], [29]. Similar to other studies, chronic arthralgia are considered incapacitating for daily life tasks and impacted professional activities and quality of life [20], [21]. Beside this impact on patient, the economic burden of this long-term pathology is also very significant, independently of the cost of the acute disease [30].
The longitudinal design of our study enabled us to identify risk factors for development of long-term arthralgia. Individuals over the age of 35 years or with diabetes were more likely to suffer from chronic arthralgia. The age has been reported to be a risk factor with some cohorts [21], [31], [32], but not in others [18], [33]. None of our available parameters to measure the severity of acute disease were associated with long-term arthralgia. This may be explained by differences in the way to measure disease severity in other studies [18]. Importantly, we show that the presence and intensity of arthralgia at M4 after the onset of the acute disease is a good predictor of long-term arthralgia.
Our study did not identify positive markers for autoimmune or rheumatoid arthritis. Additionally, we failed to identify systemic biomarkers associated with the arthralgic phenotype. Nevertheless, a slightly more elevated inflammatory status is found in a subset of arthralgic patients who have detectable serum level of IL1α, IL15 and slight elevation in Factor VII, C3 and CRP. This signature differs from that observed at the onset of the infection, when circulating virus is detectable and type I interferon, IP10, MCP1, ISG15 are highly elevated [34], [35], [36]. Others have identified IL6 and GM-CSF or IL12 as being specifically associated with long-term arthralgia [32], [33]. However, these studies were performed much earlier in the chronic phase (2–3 months and one year after disease onset).
Our study shows that anti-CHIKV antibody titers were more elevated in ART+ patients than in ART- patients. This is in agreement with a recent study [37]. This higher level of antibodies could be associated with a more severe acute infection [37]. However, in our study, the level of antibody at M36 did not correlate with acute disease severity. Alternatively, this could reflect a persisting antigenic stimulation in ART+ patients (see below). Interestingly, it has been reported that viremia level at the acute phase correlates with a faster appearance of neutralizing antibodies and a better recovery 2–3 months after the acute phase [38].
Similarly to CHIKV, other so called “arthritogenic” alphaviruses, notably Ross River virus (RRV), are known to cause acute as well as chronic arthralgia [39]. Our data indicates that chronic symptoms are linked to the initial local joint inflammation and are not associated with markers of systemic inflammation or autoimmunity. A local inflammation of the joint could be maintained by the local persistence or delayed clearance of viral antigens. This is consistent with report of Hoarau et al. [32] who detected persistent CHIKV antigens within the synovial fluid of a patient suffering from chronic arthralgia. Moreover, experimental studies in CHIKV infected animal indicate that the joint is the most highly infected tissue, making it plausible that incomplete viral antigen clearance in this anatomical site may account for the long-term symptoms [26]. RRV has been shown to persist in vitro in mouse macrophages, and a model of RRV chronic arthritis suggests that viral persistence may account for chronic disease [40]. RRV and CHIKV have been shown to be weakly tropic for macrophages in vitro [41], [42] but the presence of antibodies dramatically increased RRV entry into macrophage [42]. Further studies will be required to assess the role of macrophages, as well the role of persistent infection and antibodies in chronic arthritis caused by CHIKV.
In sum, this study furthers our understanding of the pathophysiology of CHIKV chronic arthralgia, a prerequisite for the development of efficient therapeutic strategies and for assessing the burden of disease inflicted upon populations affected by epidemic Chikungunya disease.
|
10.1371/journal.pntd.0002095 | Twenty Years of DENV-2 Activity in Brazil: Molecular Characterization and Phylogeny of Strains Isolated from 1990 to 2010 | In Brazil, dengue has been a major public health problem since its introduction in the 1980s. Phylogenetic studies constitute a valuable tool to monitor the introduction and spread of viruses as well as to predict the potential epidemiological consequences of such events. Aiming to perform the molecular characterization and phylogenetic analysis of DENV-2 during twenty years of viral activity in the country, viral strains isolated from patients presenting different disease manifestations (n = 34), representing six states of the country, from 1990 to 2010, were sequenced. Partial genome sequencing (genes C/prM/M/E) was performed in 25 DENV-2 strains and full-length genome sequencing (coding region) was performed in 9 strains. The percentage of similarity among the DENV-2 strains in this study and reference strains available in Genbank identified two groups epidemiologically distinct: one represented by strains isolated from 1990 to 2003 and one from strains isolated from 2007 to 2010. No consistent differences were observed on the E gene from strains isolated from cases with different clinical manifestations analyzed, suggesting that if the disease severity has a genetic origin, it is not only due to the differences observed on the E gene. The results obtained by the DENV-2 full-length genome sequencing did not point out consistent differences related to a more severe disease either. The analysis based on the partial and/or complete genome sequencing has characterized the Brazilian DENV-2 strains as belonging to the Southeast Asian genotype, however a distinction of two Lineages within this genotype has been identified. It was established that strains circulating prior DENV-2 emergence (1990–2003) belong to Southeast Asian genotype, Lineage I and strains isolated after DENV-2 emergence in 2007 belong to Southeast Asian genotype, Lineage II. Furthermore, all DENV-2 strains analyzed presented an asparagine (N) in E390, previously identified as a probable genetic marker of virulence observed in DHF strains from Asian origin. The percentage of identity of the latter with the Dominican Republic strain isolated in 2001 combined to the percentage of divergence with the strains first introduced in the country in the 1990s suggests that those viruses did not evolve locally but were due to a new viral Lineage introduction in the country from the Caribbean.
| In Brazil, the first dengue haemorrhagic cases were reported after the DENV-2 introduction in Rio de Janeiro, which spread to other states in the country. Aiming to perform the molecular characterization and phylogenetic analysis of DENV-2 during twenty years of viral activity in the country, strains isolated from patients presenting different disease manifestations were sequenced. Phylogeny characterized the DENV-2 as belonging to the Southeast Asian genotype, however a distinction of two Lineages within this genotype has been identified. Furthermore, all strains presented an asparagine in E390, previously identified as a probable genetic marker of virulence. The results show a temporal circulation of genetically different viruses in Brazil, probably due to the introduction of a new viral lineage from the Caribbean, which lead to the re-emergence of this serotype after 2007, causing the most severe epidemic already described in the country.
| Dengue viruses (DENV) are the most important human arboviruses worldwide, transmitted by mosquitoes of the genus Aedes, Aedes aegypti is the main vector. Explosive epidemics have become a public health problem, economic impact, socially and politically significant [1], [2].
Currently it is estimated that 70 to 500 millions dengue infections occur annually in 124 endemic countries. Nearly 3.6 billion people (55% of world population) are at risk of contracting the disease (DVI). The rapid global spread of DENV in the last 50 years resulted in the dispersal of genotypes associated with increased severity [3].
The four serotypes (DENV-1, DENV-2, DENV-3 and DENV-4) are closely related yet antigenically distinct and contain a positive-sense RNA genome that is translated as a single polyprotein and post-translationally cleaved into three structural proteins, capsid (C), premembrane (prM) and envelope (E), and seven nonstructural proteins, NS1, NS2A, NS2B, NS3, NS4A, NS4B and NS5. The RNA genome is packaged in an icosahedral capsid, and the nucleocapsid is surrounded by a lipid bilayer containing the E and M proteins [4], [5].
DENV infection causes a spectrum of clinical disease ranging from an acute debilitating, self-limited febrile illness - dengue fever (DF) - to a life-threatening syndrome - dengue hemorrhagic fever/dengue shock syndrome (DHF/DSS) [6]. Despite the similar disease manifestations, the DENV are genetically diverse with approximately 40% of amino acid sequence divergence. Distinct DENV genotypes can be characterized when the genetic divergence are higher than to 6% [7].
A recent analysis of 1,827 complete E gene sequences supported the existence of six genotypes for DENV-2: Asian genotype I, Asian genotype II, Southeast Asian/American genotype, Cosmopolitan genotype, American genotype and the Sylvatic genotype, the most genetically distinct genotype. Furthermore, the Southeast Asian/American genotype's topologies suggested a spatial division of this genotype into two major subclades [8].
In the Americas, the first DHF epidemics in the 80's were due to the introduction of the Southeast Asian/American genotype which replaced the American genotype and more severe cases with higher viremia were reported [9]–[11].
In Brazil, the disease has become a public health problem with explosive epidemics after the introduction of DENV-1 in 1986 in Rio de Janeiro [12]. However, the first DHF/DSS cases were only reported after the DENV-2 introduction in 1990 in the country [13], [14]. From 1990 until the 26th epidemiological week of 2010, a total of 5,481,921 cases, including 17,203 cases of dengue hemorrhagic fever (DHF) and 1954 deaths were reported in the country [15].
Aiming to perform the phylogeny of the DENV-2 and its impact in the disease severity during 20 years of viral activity in Brazil, strains isolated from DF, DHF/DSS and fatal cases occurred since its introduction in 1990 until 2010, were analyzed. In this scenario, the partial sequencing (C/prM/M/E genes) of 25 DENV-2 strains was performed. To determine whether the evolutionary relationships observed for the C/prM/M/E genes were applicable to the complete genome, we further fully sequenced the coding regions of nine DENV-2 strains. In order to avoid mutations introduced by in vitro passages of the virus in cell cultures we used DENV-2 strains extracted directly from serum or originally isolated from cell cultures.
The strains analyzed in this study belong to a previously-gathered collection from the Laboratory of Flavivirus, IOC/FIOCRUZ, Rio de Janeiro, Brazil, obtained from human serum through the passive surveillance system performed by the Laboratory from an ongoing Project approved by resolution number CSN196/96 from the Oswaldo Cruz Foundation Ethical Committee in Research (CEP 274/05), Ministry of Health-Brazil. Samples were chosen anonymously, based on the laboratorial results and clinical manifestations input on the Laboratory database.
Viral strains consisted of DENV- 2 (n = 34) isolated during epidemics occurred from 1990 to 2010 in six states in Brazil (Table 1). Each sample was accompanied by identification form containing clinical and epidemiological data. All strains were determined as DENV-2 serotype by reverse transcriptase polymerase chain reaction (RT-PCR) and or/virus isolation from DF (n = 19), DHF (n = 3), DSS (n = 1) and fatal cases (n = 4; 1 from DF, 2 from DHF and 1 with no classification available). Seven cases were not classified due to data unavailability.
Viral RNA was extracted from infected cell culture supernatant or directly from the patients' serum using QIAamp Viral RNA Mini kit (Qiagen) following the manufacturer's instructions and stored at −70°C for DENV typing and sequencing.
RT—PCR for detecting and typing DENV was performed as described previously [16]. 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 size for DENV-2 (119 bp) were generated.
Virus isolation was performed by inoculation into C6/36 Aedes albopictus cell line [17] and isolates were identified by indirect fluorescent antibody test (IFAT) using serotype-specific monoclonal antibodies [18]. Briefly, patients' sera were inoculated into C6/36 Aedes albopictus cell monolayers in L-15 Medium (Leibovitz, Sigma) supplemented with 2% fetal calf serum (FCS, Invitrogen) and 0.2 mM of nonessential amino acids (Invitrogen). Cells were incubated at 28°C for 5 to 7 days and observed for cytopathic effects. Infected supernatant was clarified by centrifugation and virus stocks stored in 1-mL aliquots at −70°C until use.
Reverse transcription (RT) was performed using 5 µL of extracted RNA in 25 µL of AccessQuick RT-PCR System (Promega Corporation) and specific oligonucleotides primers (Table 1). To amplify the C/prM/M/E region of 2,325 bp, specific primers (1 to 4) were used to produce 4 overlapping amplicons of approximately 900 bp and to amplify the complete coding region (10,173 bp), 15 overlapping amplicons of approximately 900 bp (1 to 15). Thermocycling conditions consisted of a single step of 42°C for 60 minutes and 40 cycles of denaturation at 94°C (30 seconds), annealing at 56°or 63°C (60 seconds) depending on the set of primers, extension at 72°C (2 minutes) and a final extension at 72°C (10 minutes). Amplification was conducted using a Model 9700 thermal cycler (Applied Biosystems). PCR products were purified from 1.0% agarose gels using QIAquick Gel extraction Kit or QIAquick PCR purification Kit (Qiagen) and used as template for cycle sequencing. Sequencing reactions were performed as recommended in the BigDye Dideoxy Terminator sequencing kit (Applied Biosystems) and the products were analyzed using an automated 3130 DNA Sequencer (Applied Biosystems). Partial sequences (C/prM/M/E) and complete coding sequences for the unprocessed polyprotein (5′ and 3′ noncoding regions excluded) were deposited in GenBank (Table 2).
The analysis of similarities, percentage of identity and divergence among the strains analyzed were performed using Megalin Program (DNAstar, www.dnastar.com). The multiple alignment was performed using CLUSTAL W (http://www.ebi.ac.uk/clustalw/) and the phylogenetic analysis by MEGA 4 software (www.megasoftware.net), using the Maximum Likelihood method (ML), according to the Tamura-Nei model, with a bootstrap of 1,000 replications. Strains representative from the five genotypes available in Genbank (www.ncbi.nlm.nih.gov) were used for the comparison, DENV-1 (GenBank accession number GU370049), DENV-3 (accession number EF629369), and DENV- 4 (accession number AF289029) strains were used as outgroup to root the trees (Table 3).
In this study, the strains BR64022/98 isolated in the 90's and Jamaica 1983 were considered as reference strains for comparison purposes. The percentage of similarity among the 25 DENV-2 strains ranged from 80.3 to 99.9% when those compared to each other and to strains representative of the different genotypes available on GenBank. The partial genome sequencing analysis characterized the Brazilian DENV-2 strains from this study as belonging to the Southeast Asian genotype, however a distinction of two Lineages within this genotype has been identified. It was observed that strains circulating prior DENV-2 emergence (1990–2003) belong to Southeast Asian genotype, Lineage I and strains isolated after DENV-2 emergence in 2007 belong to Southeast Asian genotype, Lineage II (Figures 1 and 2). Furthermore, the latter were more closely related to strains from the Dominican Republic (DR59/01), representative from the Southeast Asian genotype, Lineage II.
When the 25 DENV-2 strains were compared to the strain BR64022/98, amino acid substitutions leading to change in the biochemical properties were observed on the C and prM genes. On the E gene, a total of twelve substitutions were observed, with nine resulting in a change on the amino acid change of biochemical property (Supplementary material 1). No consistent differences were observed on the E gene from strains isolated from cases with different clinical manifestations analyzed, suggesting that if the disease's severity has a genetic origin, it is not only due to the differences observed on the E gene.
To determine whether possible amino acids differences on other genes were related to disease severity, we fully analyzed (coding region) DENV-2 strains (n = 9), representative of DF cases isolated from 1990 to 1999 and strains isolated from fatal cases occurred after the DENV-2 re-emergence after 2007 until 2010. The strain 0450/2008, representative of the DENV-2 re-emergence isolated from a DF secondary case who evolved to death was fully sequenced and its comparison to the strain from the Dominican Republic (DR59/2001), representative of the DENV-2 re-emergence, showed 22 amino acid substitutions. Likewise, the strain 0690/2008 isolated from a DHF case occurred also during the re-emergence of DENV-2 had nine had amino acid substitutions when compared to the strain DR59/2001, with seven of those leading to amino acid biochemical property change (Table S1).
The DENV-2 strain 0337/2008 isolated from a newborn presenting a high anti-DENV IgG titer who evolved to death, infected probably due transplacental transmission as his mother was diagnosed with acute DENV infection, showed substitutions on NS2A, NS4A and NS5, which were shared with the other two strains isolated from fatal cases (Table S2). The results obtained by the DENV-2 full-length genome sequencing did not point out consistent differences related to a more severe disease.
A substitution on E390 (N→D) was reported as resulting in a reduction in viral replication in macrophages and dendritic cells [19] whereas E390 (D→N) resulted in enhanced replication, maturation and activation of macrophages, enhancement of the immune response with an increased production of cytokines, increased vascular permeability and consequently a greater chance of developing DHF [20].All DENV-2 strains analyzed presented an asparagine (N) in E390, previously identified as a probable genetic marker of virulence observed in DHF strains from Asian origin.
The percentage of identity of the re-emergent DENV-2 with the Dominican Republic strain isolated in 2001 combined to the percentage of divergence with the strains first introduced in the country in the 90's suggests that those viruses did not evolved locally but were due to a new viral Lineage introduction in the country from the Caribbean.
In the Americas, the first DENV-2 was isolated in 1953 in Trinidad [21] and the first DHF epidemic caused by this serotype occurred in Cuba in 1981 after the introduction of DENV-2 genotype originated in Southeast Asia [10], [22]. Epidemics studies showed that the DENV-2 introduced in Brazil, Colombia, Venezuela and Mexico had a common ancestor with isolates from Southeast Asia, suggesting the direct transmission from that region to the Americas [23].
In Brazil, the first DHF/DSS cases were reported after the DENV-2 introduction in Rio de Janeiro [13], [24], [25], which spread to other states in the country. Phylogenetic analysis of DENV-2 strains circulating at that time confirmed the genotype circulating in Southeast Asia [26], [27]. This observation was further corroborated in an extensive analysis of viruses from the states of Rio de Janeiro (1990 and 1995), Ceará (1994), Bahia (1994 and 1999), Maranhão (1996 and 1998), Mato Grosso (1997), Pará (1998), Rio Grande do Norte (1998), Paraíba (1999) Sergipe (1999), Espiríto Santo (1995 and 2000) and forty strains isolated in Pernambuco (1995–2002) [28], [29].
After seven years without activity in Brazil, DENV-2 re-emerged in April of 2007 in the state of Rio de Janeiro causing the more severe dengue epidemic in the country in 2008 [30], [31]. Phylogenetic analysis of DENV-2 circulating in 90's and after its re-emergence identified two distinct lineages within the Southeast Asian genotype [32].
In the present study, the analysis based on the sequencing of the C/prM/M/E genes (2,325 bp) from 25 DENV-2 Brazilian isolates divided those strains in two distinct groups, one formed by DENV-2 isolated from 1991 to 2003 and another with strains isolated from 2007 to 2010 following the re-emergence of this serotype in the country. Corroborating previous phylogeny [26]–[29] strains isolated from 1991to 2003 were classified as Southeast Asian genotype, Lineage I and presenting similarities with the Brazilian strain BR64022/98 and the strain Jamaica/83. However, the strains isolated between 2007 and 2010, showed higher similarity with the strain DR59/01, from the Dominican Republic, representing the Southeast Asian genotype, Lineage II, corroborating the analysis by Oliveira et al [32]. A study by Aquino et al [33] demonstrated that DENV-2 strains from Paraguay could also be grouped into two distinct lineages within the Southeast Asian genotype and suggested the introduction of a new lineage possibly associated a serotype shift from DENV-3 to DENV-2, as observed in Brazil in 2007 and 2008 [31].
The absence of DENV-2 circulation in the years prior to its re-emergence and the high similarity observed between those viruses and the strain isolated in the Dominican Republic in 2001, suggests the introduction of a new lineage of DENV-2 causing the 2008 epidemic in Brazil. Romano et al [34] also demonstrated that DENV-2 strains isolated in Sao Paulo State in 2010 were in a monophyletic group with the strains circulating in Rio de Janeiro in 2007 and 2008 and that those were closely related to strains isolated in Cuba and Dominican Republic, with a small genetic distance, suggesting that this new lineage of DENV-2 re-emerged in of Brazil may have been imported the Caribbean. Although genetic variants of DENV have been implicated in disease severity in the past [35], [36], it was with the advance of evolutionary studies based on phylogenetic analysis combined to epidemiological data that genotypes within the distinct serotypes were associated with a greater or lesser disease severity [11], [37]–[40].
The strain isolated from a DHF case in 2000 (strain RJ/67922/2000) presented an exclusive substitution on prM143 (T→I) when compared to the other strains analyzed in this study. However, substitutions related to DHF/DSS cases were identified on prM16 and prM81 [41].
Substitutions were found on the residues E129 (V→I) and E131 (L→Q), and these are related to the division of the Southeast Asian genotype in two distinct clades, corroborating the observations that amino acids on E129 and E131 are in critical markers for genetic classification of DENV [33], [42].
All 34 strains analyzed in this study presented an asparagine (N) on E390, previously characterized as a probable trigger for DHF detected in strains of Asian origin [43]. Mutations on the flaviviruses domain III of E protein can induce virulence or attenuation of the virus to escape from the immune system [44], [45] and in this study, changes were observed throughout this domain (aa 297 to 394). The DHF case, which culminated in death (59382/1997) showed amino acid differences only in the E gene, but those differences were shared with other DF cases strains, when they were compared to the strain BR64022/98.
In this study, a substitution on prM39 was observed on the strain 0690/2008 isolated from a DHF case with a fatal outcome, on the strain 55769/1996 from a DF case and on the strain 0199/2010.. Catteau et al [46] demonstrated that the intracellular production of M ectodomain of all four DENV serotypes of DENV induce apoptosis in host cells. The carboxy terminus of prM protein with nine amino acids (aa 32–40) of some flaviviruses was designated as Apopto M [46] and appears to play an important role in inducing apoptosis and cytopathic effects [46]–[48].
Several changes were observed along the NS protein genes. Studies conducted by Yábar, [49] show that mutations in NS1 are related to the development of DHF/DSS cases when they were compared to patients with DF.
Despite the functional importance of mutations in NS genes remains unknown, future studies can elucidate their role in the emergence of strains and/or pathogenesis of the disease. It was not possible to correlate the role of Lineage II emergence with an increased severity of cases observed in the period between the years 2007–2010. Furthermore, the occurrence of secondary infection may have been the risk factor for the development of more severe cases.
In conclusion, this result shows a temporal circulation of genetically different viruses in Brazil probably due to the introduction of a new viral lineage from the Caribbean which lead to the re-emergence of this serotype after 2007. In 2007–2008, DENV-2 was responsible for most severe epidemic already described in the country, with 787,726 cases reported and 491 deaths [31]. Moreover, the Caribbean has been suggested as an important region for the circulation of DENV-2, importation and exportation of strains from and to Central America and South America [42], [50], [51].
In the past 20 years, DENV-2 activity in Brazil has contributed significantly to changes in the disease morbidity and sudden age shift [30]. In dengue endemic countries, displacement of DENV serotypes, genotypes and lineages have been reported previously and have been associated with changes in the disease severity [40], [52]–[55]. This emphasizes the need of straightening virological surveillance to monitor the emergence or re-emergence of DENV strains with pathogenic potential to cause epidemics.
|
10.1371/journal.ppat.1003672 | Molecular Basis for Oligomeric-DNA Binding and Episome Maintenance by KSHV LANA | LANA is the KSHV-encoded terminal repeat binding protein essential for viral replication and episome maintenance during latency. We have determined the X-ray crystal structure of LANA C-terminal DNA binding domain (LANADBD) to reveal its capacity to form a decameric ring with an exterior DNA binding surface. The dimeric core is structurally similar to EBV EBNA1 with an N-terminal arm that regulates DNA binding and is required for replication function. The oligomeric interface between LANA dimers is dispensable for single site DNA binding, but is required for cooperative DNA binding, replication function, and episome maintenance. We also identify a basic patch opposite of the DNA binding surface that is responsible for the interaction with BRD proteins and contributes to episome maintenance function. The structural features of LANADBD suggest a novel mechanism of episome maintenance through DNA-binding induced oligomeric assembly.
| Kaposi's sarcoma-associated herpesvirus (KSHV) establishes latent infections that are associated with several cancers including Kaposi's sarcoma, pleural effusion lymphoma, and multicentric Caslteman's disease. One of the major viral proteins required for establishment and maintenance of the latent state is the latency-associated nuclear antigen (LANA). LANA binds to DNA sequences within the terminal repeats (TR) of the viral genome and stimulates both DNA replication and episome maintenance during latency. Here we present the X-ray crystal structure of the DNA binding domain of LANA (LANADBD) and show that it has the capacity to form oligomeric complexes upon DNA binding. We characterize structural features of LANADBD that are required for oligomerization, DNA binding, and interaction with host cell BET proteins, BRD2 and BRD4, which are important for mediating multiple functions of LANA, including episome maintenance.
| Kaposi's sarcoma-associated herpesvirus (KSHV) is a human gammaherpesvirus that was first identified as the etiological agent of Kaposi's sarcoma and is also associated with pleural effusion lymphomas and multicentric Castleman's disease [1]–[3]. KSHV-associated tumors harbor latent viral genomes that persist as multicopy episomes [4], [5] (reviewed in [6], [7]). During latency the genome is circularized at the terminal repeats (TR), which function as an origin of DNA replication and as sites for tethering the episome to the host cell's metaphase chromosomes [8]–[11]. During latency, the viral episome replicates during S phase using host-cell replication machinery and expresses a limited set of viral proteins and non-coding RNAs responsible for viral genome maintenance and host cell survival [12]–[15].
Latency associated nuclear antigen (LANA) is a 130 kDa multifunctional protein required for TR-dependent DNA replication and episome maintenance during latency [5], [7], [16]–[20]. LANA also maintains latency by suppressing transcription and activity of the lytic trigger Rta [21]–[23]. Additionally, LANA interacts with numerous host cell proteins that mediate viral replication, episome maintenance, transcription regulation, and host-cell survival [15], [18], [24]–[33]. LANA binds to TR DNA through a conserved carboxy-terminal DNA binding domain (DBD) [15], [34]–[38]. LANADBD shares some common features with the functional orthologs Epstein-Barr virus nuclear antigen 1 (EBNA1) and human papillomavirus E2 [39], [40]. Each of these proteins binds to specific semi-palindromic 16–18 bp viral DNA sequences as an obligate dimer [41]–[44]. LANA binds to two adjacent sites in the 800 bp GC-rich terminal repeats, referred to as LANA binding site 1 (LBS1) and LBS2 [42]. Binding to LBS2 is highly cooperative with binding to LBS1 and precisely phased binding to both LBS1/LBS2 is essential for DNA replication function. Episome maintenance requires at least two LBS1/2 binding sites and the viral genome consists of 30–40 terminal repeats [4], [45]. The precise mechanism of DNA binding and how DNA binding and spacing confers replication and episome maintenance remains poorly understood.
There are at least two distinct mechanisms by which LANA can tether viral episomes to metaphase chromosomes. The extreme N-terminus of LANA can interact with host chromosomes through a direct interaction with histones H2A and H2B [46], [47]. A second independent mechanism involves the interaction of the LANADBD with host chromatin-associated protein [30], [48]–[50]. Prominent among these are the BET proteins BRD2 and BRD4, which contain two bromodomains that bind to the acetylated tails of histones H3 and H4, and a conserved C-terminal extraterminal (ET) domain [51], [52]. The BRD ET domains interact directly with the DBD of LANA, providing a linkage between LANA and acetylated histone tails [53], [54]. In both mechanisms, tethering requires a sequence-specific interaction between the LANADBD and the viral episome at LBS1/2.
To better understand the mechanism of LANA function through DNA binding we determined the crystal structure of LANADBD. This structure reveals several remarkable features that provide new insight into the regulation and function of LANA. We found that LANA can form a higher-order decameric ring structure. Mutagenesis studies demonstrate that the hydrophobic interface between LANA dimers is important for cooperative DNA binding, DNA replication, and episome maintenance. We also show that an amino terminal arm is important for DNA binding and replication function. Finally, we demonstrate that the BRD ET domain interacts with a basic patch located opposite to the DNA binding face that is crucial for episome maintenance.
To characterize the structural properties of LANADBD, we crystallized a construct encompassing residues 1011–1153 and determined its X-ray crystal structure at 2.05 Å resolution. A comparison with the DBD of EBNA1 shows that the two proteins share the same general fold (RMSD of Cα atoms = 5.1 Å) despite sharing very little amino acid sequence homology [55] (Figures 1A, 1B, and S1). A central, anti-parallel beta-sheet forms a hydrophobic core through which two proteins subunits form a dimer. Three helices, which contain the key residues involved in DNA binding, flank this core region [56]. The most notable differences between the homologs are in the intervening loop regions, with LANADBD exhibiting a more compact structure.
Remarkably, the crystal structure of LANADBD reveals a higher-ordered assembly comprised of five dimers interacting end-to-end, forming a decameric ring with an exterior diameter of 110 Å and an interior diameter of 50 Å (Figure 1C). The dimers are arranged such that the DNA binding surface faces the exterior of the decameric ring. The interface between the dimers is small and hydrophobic, consisting of residues Phe1037, Phe1041, Met1117, Ala1121, and Ala1124, and buries a total solvent excluded surface area of 963 Å2 (Figure 1D).
Cooperative DNA binding has been described for LANA at the KSHV TR LBS1/2 binding sites [42]. To assess the role of the tetramer interface of LANADBD in cooperative DNA binding, we used a fluorescence polarization (FP) assay with LBS1 and fit the data using a single binding site model with a Hill coefficient (h). We then created LANA mutants F1037A/F1041A and M1117A that lie within the tetramer interface. We found that wild-type LANADBD exhibits cooperativity in DNA binding (h >1). However, F1037A/F1041A and M1117A had reduced DNA binding affinity, with 38% and 12% of wild-type affinity, respectively (Figure 2). Moreover, the F1037A/F1041A mutation resulted in a complete loss of cooperativity (h = 1). As these residues are not positioned near the DNA binding face, it is likely that the reduced binding affinity observed is due to reduced cooperativity. Since only a single site was used in this assay, there is most likely a structural change that occurs upon DNA binding that primes a second LANADBD molecule for DNA interaction.
To determine if the oligomeric states of LANA that are observed in the crystals exist in solution, we performed chemical crosslinking experiments both with and without DNA. In the absence of DNA, we observed the formation of two crosslinked complexes with a molecular weight corresponding to a dimer and a tetramer (Figure 3A). However, when LBS1 DNA was added, a laddering of crosslinked complexes occurred with a maximum molecular weight of ∼150 kDa, corresponding to approximately a decamer. This suggests that DNA binding induces oligomerization of LANADBD.
To further investigate the role of the oligomeric interface in DNA binding-induced oligomerization of LANADBD, we assayed LANA mutants by electrophoretic mobility shift assay (EMSA) using full length LANA derived from human cells (Fig. 3B). In the presence of LBS1 DNA, we found that full-length LANA formed multiple oligomeric nucleoprotein complexes, indicative of DNA binding-induced oligomerization (Figure 3B lanes 1–9). When LBS1/2 DNA was used, we observed the formation of tetramer/DNA complexes as well as additional higher-ordered oligomeric species (Figure 3B lanes 10–18). Mutations predicted to disrupt the tetramer interface, M1117A and F1037A/F1041A, were capable of binding LBS1 but showed greatly reduced capacity to form oligomeric species. Furthermore, M1117A exhibited reduced ability to form fully occupied LBS1/2 complexes and F1037A/F1041A was severely impaired in forming higher order LBS1/2 DNA complexes (Figure 3B, lanes 12–13). These results indicate that hydrophobic residues at the tetramer interface are essential for higher order protein-DNA complex formation and cooperative DNA binding.
To determine if mutations that disrupt oligomerization and cooperative DNA binding are important for LANA function in vivo, we tested these and additional mutations in the context of full-length LANA using plasmid replication and plasmid maintenance assays (Figures 4 and 5). The plasmid contained eight consecutive terminal repeats (p8xTR) and replication was assessed 72 hours post-transfection by measuring the levels of DpnI resistant plasmid DNA (Figure 4). Mutations that disrupted the oligomerization interface (F1037A/F1041A) completely abrogated DNA replication function (Figure 4). All of these proteins were expressed at similar levels as measured by Western blot analysis (Figure 4D and G).
LANA mutations were also assessed for their ability to support long-term plasmid maintenance in B-lymphoid cell lines (Figure 5). Full length LANA and LANA mutant proteins were selected for stable expression in BJAB B cell-lymphoma cell lines, and then co-selected with plasmids containing p8xTR and a G418-resistance marker. While most LANA mutations were expressed at similar protein levels to wild-type LANA, their ability to support plasmid maintenance varied substantially. Single substitution mutations F1037A and M1117A supported nearly wild-type levels of plasmid replication. In contrast, the F1041A mutant showed impaired replication and maintenance activity and the double mutant F1037A/F1041A, which disrupts oligomeric interactions, was nearly completely void of episome maintenance function.
In the crystal structure of EBNA1DBD bound to DNA, the N-terminus of the domain acts as an arm that wraps around the minor groove of the DNA. In our DNA-free structure of LANADBD the N-terminal portion of this construct lies across the DNA binding face, based on homology with the EBNA1 co-crystal structure (Figure 6A and 6B). We anticipate that, when bound to DNA, this arm undergoes a conformational change to allow DNA recognition with the conserved DNA contact residues, and wrap around the DNA similarly to that observed in EBNA1DBD. To determine if this arm plays a role in DNA binding we prepared several mutants, R1013A, Y1014A, P1017G, P1018G, Y1021A, Y10141A/Y1021A, and Y1014F/Y1021F, and an N-terminal deletion construct 1021–1153.
As measured by FP and EMSA, mutation of P1017 or P1018 to glycine caused a slight decrease in affinity to about 75% compared to wild-type (Figure 2). In FP assays, Tyr1014 and Tyr1021 could be singly mutated to alanine with no detriment to DNA binding affinity. Furthermore, the Y1014F/Y1021F double mutant had nearly the same binding affinity as wild-type. However, the Y1014A/Y1021A double mutant had approximately 40% binding affinity compared to wild-type. This effect was comparable to the N-terminal deletion construct 1021–1153. Interestingly, single alanine substitutions of Tyr1014 and Arg1013 completely disrupted DNA binding in EMSA (Fig. 3). Although the differences between FP and EMSA are several, including different protein sources and different biophysical parameters, the results substantiate the importance of the N-terminal arm in stabilizing LANA-DNA binding.
Generally, the results of the in vivo assays agreed with the FP data. Mutants R1013A, Y1014A, and Y1014A/Y1021A showed greatly reduced levels of replicated plasmid with Y1021A also showing some reduction (Figure 4). These same mutants also had a reduced capacity to support the persistence of p8xTR in plasmid maintenance assays (Figure 5). Based on this data, we conclude that the NTA of LANADBD participates in DNA binding and is required for replication and episome maintenance function in vivo.
The surface opposite to the DNA binding face of LANADBD is a dense basic patch composed mostly of lysine residues (Figure 6C and 6D). This feature appears to be unique to LANA, as the analogous part of EBNA1 is composed mainly of acidic residues (Figure 6D). Previous studies have shown that a region of LANADBD encompassing a portion of this patch is important for interacting with the ET domain of BRD2 and BRD4 [53], [54].
To further identify which residues are important for this interaction we made mutations of these lysine residues by subdividing the basic patch into two regions, Lys1138/1140/1141, near the dimer interface, and Lys1109/1113/1114, near the tetramer interface. Mutation to alanines resulted in decreased plasmid replication and maintenance functions with K1109/1113/1114A being most impaired (Figures 4 and 5). Further reduction of these activities was achieved by charge reversal mutations to glutamate.
To understand the role that these mutations play in interacting with the ET domain of BRD2 and BRD4, we performed in vivo co-immunoprecipitation and in vitro pulldowns. Using a FLAG-tagged version of wild-type or mutant full-length LANA, we performed co-IPs looking for the presence of BRD4 (Figure 7A). While wild-type LANADBD showed binding to BRD4, we found that the K1138/1140/1141A and K1138/1140/1141E mutants exhibited decreased levels of BRD4 interaction. The effect of mutations at Lys1109/1113/1114 was minimal. Other mutations in both the N-terminal arm and the tetramer interface did not show any impairment in interactions with BRD4. These findings indicate that amino acid residues within the basic patch near the dimer interface (Lys1138/1140/1141) are principally responsible for BRD interaction.
To confirm that the co-immunoprecipitation was due to a direct interaction between LANA and the ET domain of BRD we performed in vitro pulldowns. For these assays we prepared His-SMT3-tagged versions of BRD2 and BRD4 ET domains and untagged forms of LANADBD and EBNA1DBD (as a negative control). The long constructs (BRD2L and BRD4L) comprised the ET domain with the serine-rich C-terminal tail (Figure S2). The short constructs (BRD2S and BRD4S) contained the portion of the ET domain exhibiting the highest sequence identity between the two proteins. The results show that LANADBD is capable of interacting directly with all four BRD constructs (Figure 7B and 7C), whereas EBNA1DBD did not show any binding to these constructs (Figure 7D and 7E). This data demonstrates that LANADBD directly binds to the BRD ET domain.
LANA is the major viral protein involved in maintaining the latent state of KSHV infection and ensuring that the viral episome is replicated and passaged with cell division. In this capacity, LANA acts to form a site-specific origin of replication and as a molecular tether between the viral episome and the host chromosome. The C-terminal DNA binding domain of LANA (LANADBD) recognizes specific sites in the viral genome and is essential for each of these activities. Here, we determined the X-ray crystal structure of LANADBD and formally demonstrated that LANA is a structural ortholog of EBV EBNA1 and papillomavirus E2 DNA binding proteins. Importantly, the structure of LANADBD presented here reveals several unique features that we have verified are essential for cooperative DNA binding, DNA replication, episome maintenance, and direct interaction with BRD proteins.
The cooperative nature of DNA binding suggests that LANA may form a tetramer prior to DNA binding or that DNA binding induces a conformational change in either the protein or the DNA, which facilitates binding to the lower affinity site, LBS2. We have shown using chemical crosslinking that we can capture LANADBD tetramers in the absence of DNA and that DNA binding induces the formation of high molecular weight oligomers (Figure 3A). This same behavior was observed using full-length LANA, indicating that oligomerization is not an artifact of the truncated protein (Figure 3B) [37]. Furthermore, cooperativity was observed in FP assays that utilized a single site LBS1 probe, and this cooperativity was reduced by mutations in the tetramer interface (Figure 2). This suggests that tetramerization enhances the interaction with lower affinity sites, such as LBS2.
Cooperative binding has also been observed for LANA's functional homolog in EBV, EBNA1 [57]. EBNA1 binding sites exist in two locations in the latent origin of replication (oriP), the family of repeats (FR) and the dyad symmetry (DS) element. Similar to LBS1/2, DS is composed of two sets of tandem EBNA1 binding sites, positioned 21 bp apart, center-to-center [58]. The available structures of EBNA1DBD present a dimeric structure either alone or in the presence of DNA [55], [59]. The residues at the location of the anticipated tetramer interface of EBNA1DBD are generally acidic and, would be unable to participate in a homotypic interaction similar to that seen in LANA. Thus it is likely that a structural change, post-translational modification, or additional cellular protein would be required for the cooperative tetrameric DNA binding observed with EBNA1.
In the crystal structure presented here, LANADBD dimers interact to form a novel decameric ring structure (Figure 1C). While we have not validated that a decamer forms in vivo, the molecular details of the interactions between dimers and the geometric organization of the dimers provide insight into the means by which cooperative binding occurs. The tetramer interface is relatively small and composed of hydrophobic residues Phe1037, Phe1041, Met1117, Ala1021, and Ala1024. We and others have shown that mutation of these residues has adverse effects on DNA binding activity [56]. In particular, the F1037A/F1041A mutant fails to produce DNA binding induced oligomers in solution (Figures 2 and 3). This mutant also fails to form the higher molecular weight complexes with LBS1/2 that are characteristic of full-length wild-type LANA. Most notably, F1037A/F1041A is severely impaired for plasmid replication and maintenance activity (Figures 4 and 5), indicating the importance of the tetramer interface in cooperative DNA binding and LANA functionality in vivo.
The geometric arrangement of the dimers in the crystal structure may indicate the manner in which two dimers interact when bound to LBS1/2. Wong and Wilson demonstrated that the binding of LANA to LBS1 induces a 57° bend in the DNA and that binding to LBS1/2 additively increases the bend angle to 110° [60]. Consistent with this, the angle between dimers in the decameric ring is approximately 110° (Figure 1C). Mutational analysis of residues near the tetramer interface located distal to the DNA binding surface further demonstrates that the angle between dimers is important in mediating cooperativity. Met1117 may act to maintain the angle between dimers at approximately 110°. We observed a decrease in the level of oligomeric species formation by the M1117A mutant in EMSA analysis using an LBS1/2 probe (Fig. 3). Additionally, in FP assays (Fig. 2) the Kd of DNA binding is reduced to about 10% and the Hill coefficient is reduced to less than 1 for M1117A, indicating negative cooperativity. This implies that the angle between dimers in the preexisting tetramers of these mutants exceeds the optimal angle for cooperative binding and the formation of higher ordered complexes. In the context of full-length LANA this geometry may be maintained by additional regions not determined in our crystal structure, as the effects of M1117A are less pronounced in EMSA, replication, and maintenance assays (Fig. 3B, 4, and 5). Taken together, our data supports the role of the tetramer interface as the basis for cooperative DNA binding to TR and functionality in KSHV DNA replication and episome maintenance.
Several other studies support the role of LANA oligomerization in KSHV biology. LANA has been shown to bind TR DNA as an oligomer in EMSA and mutations that disrupt oligomerization were found to block DNA replication and episome maintenance [37]. The oligomerization domain was mapped to the N-terminal arm, which we have also found is required for cooperative DNA binding. Some evidence for oligomeric binding in vivo may be inferred from nucleosome mapping studies of the TR in latently infected BCBL1 cells [17]. This study revealed that four nucleosomes are positioned at regular intervals within the 809 bp TR, and a nucleosome-free region of ∼350 bp surrounds the LBS1/2 binding site [17]. This nucleosome-free region could accommodate two LANA dimers and replication factors, and this extended region is essential for efficient DNA replication [38]. It is also possible that one or two LANA decamers could occupy this nucleosome free region, assuming that the each decamer wrapped ∼120 bp of TR DNA. In this model, only one or two dimers of the decamer would have high-affinity interactions with LBS1 and LBS2, and the remaining three dimers would interact non-specifically with adjacent TR DNA. Whether LANA oligomers mediate additional long-distance interactions between tandem TRs with ∼800 bp of intervening DNA is not yet known. However, long-distance interactions have been described for EBNA1, which can form DNA loops between the family of repeats and the dyad symmetry elements of OriP [61]. Higher order oligomeric conformations have also been described for other viral origin binding proteins, including SV40 T antigen and papillomavirus E1, which undergo conformational changes after DNA and ATP-binding [62], [63]. More recently, a complex oligomeric structure has been described for the adenovirus-encoded E4-ORF3, which forms an intracellular network that is important for compartmentalization during viral DNA replication [64]. Interestingly, the structure of E4-ORF3 was shown to share a similar fold to EBNA1, suggesting that LANA, EBNA1, E2, and E4-ORF3 may share some similarities in the self-assembly of larger structures. Thus, it is possible that higher-order oligomerization of LANA plays an important functional role in KSHV biology.
Another interesting feature of the crystal structure of LANADBD is the N-terminal arm (NTA) of this domain. Mutagenesis studies revealed that the LANADBD NTA is essential for stable DNA binding, replication, and episome maintenance function. In the LANADBD crystal structure, the NTA occupies the exact position where DNA would be located based on superposition with the EBNA1 co-crystal structure (Figure 6B). In the EBNA1DBD-DNA co-crystal structure, the N-terminal arm of the domain wraps around the minor groove of the DNA (Figure 6B) [59]. We have shown that mutation of residues located within LANA NTA (specifically Arg1013, Tyr1014, and Tyr1021) decrease DNA binding affinity and causes a loss of plasmid replication and maintenance. It is possible that the NTA occupies this position due to the effects of crystal packing or because the body of the protein presents an entropically favorable environment. However, in this position the arm would occlude DNA from interacting with LANA (Figure 6B). Therefore, it is most likely that the NTA undergoes a conformational change prior to DNA binding, wrapping around the minor groove in a manner similar to that observed in EBNA1. This would also help to explain some of the cooperative oligomerization induced by a single LBS1 DNA biding site. We suggest that DNA binding induces a change in the conformation of the NTA that facilitates higher order oligomerization and cooperative DNA binding by LANA.
A novel feature of LANADBD observed in the crystal structure is the presence of a lysine-rich basic patch located opposite to the DNA binding. The only prior evidence for a function of the residues within this patch was the observation that deletion of the last 23 amino acids of LANA abrogates interaction with BRD2 and BRD4. These same studies showed that the ET domain of the BRD proteins mediates the interaction with LANA. The sequence of the ET domains of BRD2 and BRD4 reveals two regions that may be important, the N-terminal portion contains a large number of glutamates and the C-terminal tail is serine-rich. By dividing the basic patch on LANADBD into two parts we were able to show that the internal portion of the patch (Lys1138, Lys1140, and Lys1141) contains the key residues involved in BRD binding (Figure 7A). These lysines appear to interact with the acidic residues in the N-terminal portion of the ET domains (Figures 7B and 7C). We did not observe any interaction with either BRD ET domain with EBNA1DBD, as would be expected since the analogous surface of EBNA1 is mostly acidic (Figure 6D).
BRD interactions have been shown previously to mediate LANA function in chromosome binding [53], [54]. Our data is consistent with a role of BRD binding in episome maintenance and DNA replication. However, BRD proteins have also been implicated in other LANA functions, including transcription regulation and cell cycle control. BRD proteins have been shown to mediate multiple functions of E2 family members, including metaphase chromosome tethering, transcriptional repression, and DNA replication. Thus, it is possible that BRDs contribute to multiple functions of LANA.
The structure and associated biochemical and cell-based studies of LANADBD reported here reveal new insights into the higher-order oligomerization, cooperative DNA binding, DNA replication, episome maintenance, and BRD interaction interface of LANA. We identified the N-terminal arm, which is crucial for DNA interaction and, based on homology to EBNA1, likely wraps around the minor groove of the cognate DNA, providing for high affinity binding. We have demonstrated that LANA has the capacity to oligomerize upon binding its cognate DNA. The decameric ring observed in the crystal structure provides one possibility for the arrangement of oligomeric LANA, however it is not known for certain if this is the state of oligomerization in vivo. We also determined that a basic patch located opposite to the DNA binding face of LANADBD serves as the interaction site with host cell BRD proteins. These activities are critical for LANA's function in viral DNA replication and tethering the KSHV episome to host cell chromosomes to allow for passage of the genome upon cell division during latent viral infection.
A construct comprising residues 1011–1153 of LANA (LANADBD) was expressed as a His-SMT3 tagged protein in E. coli. Cells were sonicated in 2 M NaCl, 25 mM HEPES, 25 mM imidazole, 5 mM beta-mercaptoethanol (BME), pH 8.0 and purified over Ni-NTA resin. The His-SMT3 tag was removed by overnight cleavage with ULP1 protease and the cleaved product was further purified by a second run over Ni-NTA. The protein was then run on a Superdex 75 column equilibrated with 2 M NaCl, 20 mM HEPES, 0.5 mM TCEP, pH 7.4. Mutants were prepared using site-directed mutagenesis and purified as described. The protein was concentrated to 30 mg/mL and crystallization screening was performed. Initial crystallization hits did not provide diffraction beyond 4 Å. To improve crystal quality, crystals grown in 20% polyethylene glycol monomethyl ether 2000, 100 mM Tris pH 8.0, 150 mM calcium acetate were crushed, diluted 1∶10000 in reservoir solution, and used to re-seed entire crystallization screens. This yielded crystals in 1 M ammonium formate and 100 mM HEPES pH 8.0 that diffracted to 2 Å and formed in the monoclinic space group P21 (α = 51.4 Å, β = 175.2 Å, γ = 97.1 Å, b = 95.3°).
Initial phasing was unsuccessful using molecular replacement with the crystal structure of EBNA1DBD (PDB 1vhi). Crystals were then soaked with a variety of heavy atom salts and potassium osmate was identified as a successful derivatization. Diffraction data were collected to 3 Å at beamline X29a at the National Synchrotron Light Source using wavelengths at the peak (1.1401 Å) and inflection (1.1404 Å). Ten initial osmium sites were located using AutoSol in Phenix. The phases obtained were of adequate quality to generate electron density maps in which the secondary structure elements could be modeled manually. Once the majority of the protein was built, molecular replacement was performed using a higher resolution native dataset. Model building and refinement were completed using Phenix. The model was refined to convergence with Rwork = 18.18% and Rfree = 22.62%. Complete data reduction and refinement statistics are given in Table 1. The final structure has been submitted to the Protein Data Bank with accession code 4J2K.
To determine dissociation constants for DNA binding, fluorescence polarization assays were performed. Protein was serially diluted 2-fold starting at 100 µM in 1 M NaCl, 20 mM HEPES, 1 mM DTT, pH 7.4. These protein samples were then diluted 10-fold into reactions resulting in a final condition of 200 mM NaCl, 20 mM HEPES, 1 mM DTT, 1 µg/mL BSA, 0.1 µg/mL poly(dA:dT) (Invivogen, San Diego, CA), 0.001% Tween 20, and 1 nM LBS1 probe (Integrated DNA Technologies, Cedar Rapids, IA). This probe was a 21-mer (AGCGGCCCCATGCCCGGGCGG) centered on the LBS1 site and was 5′ labeled with 6-carboxyfluorescein. The reactions were incubated at room temperature for 30 minutes and then read using a Beacon 2000 Fluorescence Polarization reader. Data were collected in triplicate and analyzed using a one-site specific binding with Hill slope model in GraphPad Prism (version 5.0a; GraphPad Software, La Jolla, CA).
Crosslinking reactions were performed using ethylene glycol bis[succinimidyl succinate] (EGS; Thermo Scientific). The DNA was an unlabeled version of the probe used in the FP experiments. Protein was incubated with a 1.2 molar excess of DNA, where applicable, in a reaction buffer of 200 mM NaCl, 20 mM HEPES pH 7.4, and 1 mM DTT for 30 minutes at room temperature. EGS was dissolved and diluted in DMSO and added to the reaction at 5% of the reaction volume. This was incubated at room temperature and then quenched with 5% volume of 1 M Tris, pH 7.5 for 15 minutes. Samples were boiled with loading dye before loading into gels.
FLAG-LANA proteins were expressed by transient transfection in 293T cells, and isolated by FLAG-affinity purification after multiple wash steps with extraction buffer (20 mM Tris, 400 mM NaCl, 10% glycerol, 0.5 mM EDTA, 0.05% NP-40, 0.5 mM DTT, and protease inhibitors (Sigma)) to remove weakly associated cellular proteins and nucleic acids. Binding reactions and gel conditions were described previously [65] DNA binding reactions were resolved on a horizontal 1.5% agarose gel in 0.5× TBE (45 mM Tris-borate, 1 mM EDTA) at 10 V/cm for 2 hours and then dried on DE80 paper prior to PhosphorImager exposure. Binding reactions were in a final of 10 µL containing 10 mM HEPES (pH7.9), 10% glycerol, 100 mM KCl, 5 mM MgCl2, 0.1 U/µl poly-dIdC, 5 mM β-mercaptoethanol with LANA protein at ∼200 nM and radiolabelled LBS1 or LBS1/2 probe at 2.5 nM.
BJAB (uninfected B cell lymphoma) cells or BJAB cells stably expressing triplicate FLAG-epitope-tagged LANA were grown in RPMI medium supplemented with 10% fetal bovine serum (FBS) and maintained at concentrations of 0.2 to 0.8×106/mL. Stable LANA expression was maintained with puromycin selection (2 µg/ml) (Sigma). After cotransfection of p8TR plasmids, LANA stable cells were maintained with both puromycin (2 µg/ml) (Sigma) and G418 (600 µg/ml) (Mediatech) selection.
293T cells were maintained in Dulbecco's modified Eagle's medium (DMEM) containing 10% v/v FBS. For transient transfection experiments, transfection of actively growing 293T cells was processed with Lipofectamine reagent (Invitrogen), and cells were harvested 72 hours post transfection.
For creating stable cells, 10×106 of actively growing BJAB cells were resuspended in 450 µl 10% FBS RPMI media without antibiotics. 30 µg DNA of interest were mixed together with cells in a microcentrifuge tube and incubated at RT for 10 min. All transfections were carried out with the Gene Pulser Xcell (Bio-Rad) setting at 0.22 kv and 960 µF as external capacitor. The transfected cells were incubated at RT for 10 min post electroporation, and then maintained as described above.
Plasmid p8TR contains eight copies of the TR unit cloned into pREP9 (Invitrogen) (gift of K. Kaye, Harvard Medical School). KSHV LANA was cloned into p3XFLAG-CMV-24 (Sigma) as described previously. All the LANA mutants created described in this study are based on this plasmid. Human BRD4 expression construct was a gift from Dr. Jianxin You (University of Pennsylvania, School of Medicine).
For immunoprecipitation (IP), co-transfected 293T cells expressing FLAG-tagged LANA wild-type, mutant, or vector only control and human BRD4 expression vector were lysed in 300 µL of IP buffer (50 mM Tris pH 7.6, 60 mM NaCl, 1% glycerol, 0.5 mM EDTA, 0.2% NP- 40, and protease inhibitor (Sigma)) at 4°C with sonication. FLAG-tagged proteins were precipitated with anti-FLAG rabbit serum (Sigma) followed by protein A/G bead (Thermo Scientific) capture.
For immunoblot assays, proteins were resolved in 8–16% Novex Tris-Glycine gels, LANA was detected using an HRP conjugated anti-FLAG antibody (Sigma), and BRD4 was detected using Anti-BRD4 antibody (Bethyl Laboratories, Inc.) at 1∶2000 dilution in conjunction with HRP-conjugated secondary antibodies (GE Life Sciences) and ECL reagents (Invitrogen).
At day 7 post-transfection and selection, BJAB cells were lysed in 1 mL lysis buffer (0.6% SDS, 10 mM EDTA, 10 mM, Tris-HCl pH 7.5, 50 µg/ml RnaseA) per 5×106 cells and incubated at 37°C for 2 hours. NaCl was then added to 1 M final concentration and incubated overnight at 4°C. After a 30 minute centrifugation at 8,500 rpm at 4°C, DNA was extracted once with phenol∶chloform (1∶1), twice with chloroform∶isoamyl alcohol (24∶1), ethanol precipitated, and the DNA pellet was washed with 70% ethanol, air-dried and resuspended in TE buffer.
FLAG- or LANA-expressing BJAB cells were transfected with p8TR DNA. After 48 h, cells were maintained in medium containing G418 (600 µg/ml) (Mediatech). Hirt DNA extraction was performed about 40 days post transfection. 30 µg DNA was digested with BglII to linearize the p8TR DNA. Double digestion of 30 µg of Hirt DNA with BglII and DpnI was also performed and electrophoresis in a 0.8% agarose gel in Tris-borate-EDTA buffer. DNA was then transferred to a nylon membrane. KSHV DNA was detected with a 32P-labeled TR probes. Quantitation of the linearized BglII- or BglII/DpnI-digested p8TR was performed using a PhosphorImager SI (Molecular Dynamics).
BRD2 and BRD4 were expressed in two forms, both as His-SMT3 tagged proteins. The long form encompassed the extraterminal domain including the serine-rich C-terminus (BRD2 633–801, BRD4 601–722). The short form included the portion of the ET domain that is most conserved between BRD2 and BRD4 (BRD2 633–714, BRD4 601–681). The tagged BRD proteins were incubated with Ni-NTA resin (Thermo Scientific) equilibrated with 200 mM NaCl, 20 mM HEPES, 5 mM BME, at pH 8.0 and washed with 10 column volumes (CV) of this buffer. Untagged LANADBD (1011–1153) or EBNA1DBD (461–642) was then added and the resin was washed again with 10 CV of wash buffer. The protein was then eluted with wash buffer supplemented with 300 mM imidazole. As a negative control the untagged LANADBD or EBNA1DBD was incubated with the Ni-NTA resin and washed and eluted as described above.
|
10.1371/journal.pcbi.1005322 | Modelling Systemic Iron Regulation during Dietary Iron Overload and Acute Inflammation: Role of Hepcidin-Independent Mechanisms | Systemic iron levels must be maintained in physiological concentrations to prevent diseases associated with iron deficiency or iron overload. A key role in this process plays ferroportin, the only known mammalian transmembrane iron exporter, which releases iron from duodenal enterocytes, hepatocytes, or iron-recycling macrophages into the blood stream. Ferroportin expression is tightly controlled by transcriptional and post-transcriptional mechanisms in response to hypoxia, iron deficiency, heme iron and inflammatory cues by cell-autonomous and systemic mechanisms. At the systemic level, the iron-regulatory hormone hepcidin is released from the liver in response to these cues, binds to ferroportin and triggers its degradation. The relative importance of individual ferroportin control mechanisms and their interplay at the systemic level is incompletely understood. Here, we built a mathematical model of systemic iron regulation. It incorporates the dynamics of organ iron pools as well as regulation by the hepcidin/ferroportin system. We calibrated and validated the model with time-resolved measurements of iron responses in mice challenged with dietary iron overload and/or inflammation. The model demonstrates that inflammation mainly reduces the amount of iron in the blood stream by reducing intracellular ferroportin transcription, and not by hepcidin-dependent ferroportin protein destabilization. In contrast, ferroportin regulation by hepcidin is the predominant mechanism of iron homeostasis in response to changing iron diets for a big range of dietary iron contents. The model further reveals that additional homeostasis mechanisms must be taken into account at very high dietary iron levels, including the saturation of intestinal uptake of nutritional iron and the uptake of circulating, non-transferrin-bound iron, into liver. Taken together, our model quantitatively describes systemic iron metabolism and generated experimentally testable predictions for additional ferroportin-independent homeostasis mechanisms.
| The importance of iron in many physiological processes relies on its ability to participate in reduction-oxidation reactions. This property also leads to potential toxicity if concentrations of free iron are not properly managed by cells and tissues. Multicellular organisms therefore evolved intricate regulatory mechanisms to control systemic iron levels. A central regulatory mechanism is the binding of the hormone hepcidin to the iron exporter ferroportin, which controls the major fluxes of iron into blood plasma. Here, we present a mathematical model that is fitted and validated against experimental data to simulate the iron content in different organs following dietary changes and/or inflammatory states, or genetic perturbation of the hepcidin/ferroportin regulatory system. We find that hepcidin mediated ferroportin control is essential, but not sufficient to quantitatively explain several of our experimental findings. Thus, further regulatory mechanisms had to be included in the model to reproduce reduced serum iron levels in response to inflammation, the preferential accumulation of iron in the liver in the case of iron overload, or the maintenance of physiological serum iron concentrations if dietary iron levels are very high. We conclude that hepcidin-independent mechanisms play an important role in maintaining systemic iron homeostasis.
| Iron is an essential element for the organism. It plays a critical role in oxygen transport, DNA synthesis, mitochondrial energy metabolism and as a cofactor of numerous enzymes [1, 2]. However, excess free iron catalyzes reactions that result in the formation of reactive oxygen species and oxidative stress. Hence, iron homeostasis must be maintained within a narrow range to provide sufficient iron for cellular function while preventing the generation of oxidative stress [3]. Systemic iron homeostasis is predominantly controlled by the interaction of the liver produced hormone hepcidin with its receptor, the iron transporter ferroportin (Fpn), resulting in the degradation of Fpn [4–7].
Fpn is the only known cellular iron exporter [8, 9]. It controls iron export from duodenal enterocytes that take up dietary iron, from iron-recycling macrophages, and from hepatocytes that store iron. Iron release from cells through Fpn requires the ferroxidases ceruloplasmin and/or hephaestin [10–12].
Hepcidin is produced in response to iron availability (via the BMP6/SMAD signaling pathway), erythropoetic demand (via erythroferrone), hypoxia and inflammatory mediators (via JAK/STAT signaling) [13–17]. Binding of hepcidin triggers Fpn internalization, ubiquitination and subsequent lysosomal degradation and thus causes iron retention in iron exporting cell types [18]. In addition, Fpn expression is regulated at the transcriptional level by hypoxia-inducible factor-2alpha (HIF2α) in response to hypoxia and iron deficiency [19, 20] as well as by BACH1 and Nrf2 in response to excess heme or oxidative stress [21]. At the translational level its expression is controlled by iron regulatory proteins (IRP), which bind to an iron responsive element (IRE) located in its 5’UTR [22, 23]. Furthermore, Fpn expression is controlled by miRNAs [24, 25].
Under physiological conditions serum iron is bound to the transport glycoprotein transferrin. Transferrin saturation is used as a measure for the serum iron level. The fact that hepcidin and Fpn expression are tightly regulated by numerous control mechanisms assures that physiological concentrations of transferrin-bound iron are maintained. Increased transferrin saturation increases hepcidin expression [26–28], leading to enhanced Fpn degradation and reduced duodenal iron absorption and iron mobilization from storage iron. In the case of hereditary hemochromatosis, where genetic perturbations cause hepcidin deficiency, chronic iron overload and organ damage will develop [29, 30]. On the other end of the spectrum, mutations in genes causing hepcidin overexpression will induce chronic iron deficiency and anemia [31].
Similarly, in an inflammatory setting, Fpn expression is decreased by hepcidin-dependent and hepcidin-independent mechanisms. While hepcidin expression is increased by inflammatory cytokines, such as IL6, causing Fpn degradation, Fpn transcription is additionally reduced [32–35]. Both control mechanisms reduce serum iron levels (hypoferremia), acting as an effective innate immunity mechanism that restricts access of microbes to iron [36]. However, if inflammation persists, the lack of iron results in a reduced iron supply for erythropoiesis, causing anemia of inflammation, which is frequently observed in hospitalized patients.
So far it is not clear how the different control mechanisms quantitatively impinge on Fpn to maintain cellular or systemic iron homeostasis. Studies in both, hepcidin knockout mice and mice with an engineered point mutation in Fpn that render it resistant to hepcidin binding (Fpn C326S mice) clearly demonstrate that LPS treatment results in a similar reduction of serum iron level compared to wild type mice [37, 38]. This indicates that hepcidin is dispensable for the generation of inflammation-induced hypoferremia under some conditions.
To dissect the quantitative contributions of hepcidin dependent post-translational and transcription-mediated Fpn control mechanisms under inflammatory conditions and to identify network components that contribute to the establishment of organ-specific iron pools, we generated a multi-scale model describing systemic iron homeostasis at the organ level. We extended our previously established model for hepcidin regulation via the BMP6/SMAD and the IL6/STAT3 signaling pathways [39] by now considering organ iron pools, organ-specific Fpn mRNA and protein synthesis/degradation rates and the impact of Fpn levels on the iron export from organs into blood (see Fig 1). The model was fit to own experimental data obtained in mice maintained on different iron diets or exposed to inflammatory stimuli (peritoneal LPS injection) as well as to previously published data by Lopes et al. [40], Daba et al. [41], Deschemin et al. [38] and Lesbordes et al. [42]. The model was validated by correctly predicting the responses of iron pools and iron-related proteins to a combined stimulus of dietary iron-loading and LPS treatment. Finally, we applied the model to assess the individual contributions of Fpn regulatory mechanisms to the response to dietary iron perturbations and inflammation and to analyze mechanisms for the establishment of organ specific iron-pools. Fig 2 gives an overview of the different steps of the study.
Iron homeostasis in the body is maintained by intricate regulatory mechanisms of iron uptake, release and flux between compartments. Systems biological approaches are required to fully understand the dynamics of this complex network. So far mathematical models only described subsystems of iron metabolism, including liver iron metabolism [43, 44], intestinal iron absorption [45], iron release from macrophages [46] or storage of iron in ferritin [47]. To describe iron metabolism on a systemic level, Lopes et al. derived a whole-body model, which integrated iron fluxes between blood and different organs [40]. This model revealed how iron fluxes and distribution of iron pools between organs change upon alterations in dietary iron levels. However, the model by Lopes et al. lacked the underlying regulatory mechanisms responsible for maintenance of iron homeostasis and therefore it only described flux changes phenomenologically (by fitting each condition separately).
Motivated by this, we set out to develop a whole-body model of iron homeostasis, which explicitly describes intra- and extracellular regulatory loops that sense and modulate iron fluxes between different compartments. As a major means of regulation, we focused on the systemic regulation of iron metabolism by the hepcidin-Fpn regulatory axis (see Introduction). Our model allowed us to simultaneously fit time-resolved data for multiple experimental perturbations, and to dissect the impact of individual regulatory loops on systemic iron homeostasis.
The structure of our model is outlined in Fig 1. The concentration of iron in the non-cellular part of the blood stream (Fe serum) is the central hub. From the serum, iron can be imported into various organs. This raises the total intracellular iron pool in each organ (Fe liver, Fe spleen, Fe bone marrow, Fe red blood cells, Fe duodenum). Besides these major iron-containing organs, a substantial amount of iron can be found in the remaining body. We modeled this by including a lumped iron pool that sums up all remaining intracellular iron pools in the body (Fe other organs). In several reverse reactions, iron can also be exported from peripheral organs into the blood. Additionally, there is uptake of dietary iron in the duodenum and loss of iron from the duodenum and from the compartment ‘other organs’, which represent loss via the shedding of enterocytes and skin cells desquamation, respectively. Moreover, red blood cells receive iron from plasma through the bone marrow compartment and deliver iron into the spleen, e.g. into splenic macrophages that recycle iron from aging red blood cells. Additionally, iron uptake by spleen macrophages due to ineffective erythropoiesis is included [48].
The iron flux model initially described by Lopes et al. corresponds to the organs and iron fluxes (dark red arrows) depicted in Fig 1. In this study, we additionally considered that the export of iron from peripheral organs is controlled by the iron exporter Fpn which is predominantly located at the plasma membrane of three cell types: duodenal enterocytes, macrophages and hepatocytes. Fpn expression is described separately for each organ and is controlled by three regulatory mechanisms (see Introduction): (i) inflammatory cues (e.g. LPS) reduce the transcription of Fpn mRNA; (ii) intracellular iron levels enhance the translation of Fpn mRNA into protein; (iii) the turnover of Fpn protein is enhanced by the soluble polypeptide hepcidin.
Hepcidin expression is activated by the iron-sensing BMP6/SMAD pathway and by an inflammatory signaling cascade, which involves production of cytokines (primarily IL6 [49]) and the subsequent phosphorylation of STAT3 transcription factor in hepatocytes. Taken together, our model describes several auto-regulatory loops controlling iron homeostasis as well as multi-layered regulation by inflammatory signals, all of which converge on the modulation of Fpn expression levels.
To describe these phenomena, we derived a system of 20 coupled ordinary differential equations (see S1 Text). For example, the dynamics of the bone marrow iron pool derived from the model topology in Fig 1 is given by
d [ F e b m ] d t = F s e r → b m - F b m → R B C - F b m → s p l ,
with Fser → bm, Fbm → RBC and Fbm → spl quantifying the inflow from serum into bone marrow and the outflows from the bone marrow into the RBCs and spleen compartments, respectively. The individual reaction rates were mainly described using mass action kinetics. For instance, it was assumed that most iron fluxes are proportional to the iron concentration in the originating compartment, e.g.
F s e r → b m = v b m [ F e s e r ] , F b m → R B C = v R B C [ F e b m ] F b m → s p l = v s p l [ F e b m ] , (1)
were the rate constants vbm, vRBC and vspl are model parameters (determined by fitting to the data, see next section). Iron export from peripheral organs into blood is additionally assumed to be proportional to the respective Fpn level, e.g.
F d u o → s e r = u d u o [ F e d u o ] [ F p n d u o ]
for the flow of iron from the duodenum into the serum. Thereby, we denote by vorgan the rate constants of iron import into the organs and by uorgan the rate constants of iron export into the serum. Fpn expression is described using a standard model of mRNA transcription and protein translation. Details about the Fpn regulation by LPS, iron and hepcidin can be found in S1 Text. Hepcidin induction by IL6 and BMP6 signaling pathways was described by a previously calibrated model [39], which considers signal integration on the hepcidin promoter using thermodynamic state ensemble approach (in the following referred to as ‘hepcidin promoter model’).
The computational model in SBML format is available in the Supplement (see S1 Model).
The model contained many unknown parameters (e.g. rate constants of the iron flows between compartments vbm, vRBC, vspl in Eq 1, see previous section), which had to be estimated by fitting the simulation results to experimental data. To reduce the complexity of this fitting problem, we assumed in some cases that homologous reactions in different compartments proceed with the same kinetic rate constants (see S1 Text). For instance, the degradation rate of Fpn mRNA and the Michaelis-Menten constant corresponding to Fpn mRNA inhibition by LPS is considered to be equal in all organs. Furthermore, the kinetic parameters of the hepcidin promoter model were fixed to the values that we had previously determined using systematic perturbations in the HuH7 cell culture system [39]. Nevertheless, 48 kinetic parameters remained to be estimated from the data. Additionally, 20 scaling parameters were fitted to match a model formulated in absolute concentration units to experimental data given in arbitrary units.
The model was calibrated based on time-resolved data from male C57BL/6-mice, which were either measured as a part of this study or taken from the literature (in total 344 data points). As summarized in Table 1 the calibration dataset comprised three experimental perturbations, which were either applied alone or in combination:
These experiments were mostly performed in wild type animals, but data from hepcidin knockout mice were also included. Model simulations were performed by analytically calculating a steady state before perturbing the system by a change in diet or in inflammatory status. The different experimental perturbations were mimicked in the model by changing parameter values or the initial conditions for the solving of the ordinary differential equations that describe the temporal dynamics of iron pools and regulatory proteins (see section ‘Model derivation and implementation’). For example, a change in dietary iron corresponds in the model to a change in the parameter value describing the influx on iron into the duodenum. Injection of LPS corresponds to a change in the initial condition for the LPS concentration to a nonzero value.
Iron pool sizes of compartments are given as μg per animal, with experimental data from individual mice scaled to a standard mouse weight of 25 g at 10 weeks of age. Details on the simulation of tracer iron distribution are given in S1 Text. Fitting of the simulated trajectories to the experimental measurements was done using a multi-start local optimizer and by minimizing the χ2 metric which is a weighed difference between model and data (S1 Text). The robustness of the model predictions was assessed by simulating them for multiple parameter combinations with a similar goodness-of-fit.
We describe the dynamics of iron pools and regulatory proteins using a set of ordinary differential equations (see previous section and S1 Text). The kinetic parameters in this model were unknown and had to be estimated by fitting the model simulations to experimental data. To generate the data necessary for model calibration and for the testing of the fitted model, we performed time-resolved experimental measurements in mice. We subjected 9–10 weeks old male C57BL/6-mice to an intra-peritoneal LPS injection (1μg/g body weight) and followed the dynamics of iron-related parameters. At defined time points post injection (0.5/1/2/4/6/8/24/36/48/72 hours in a first experiment and 6/18/48 hours in a second experiment), we measured the levels of iron in serum, liver, spleen, duodenum and red blood cells. Complementary, we also fitted data of previous studies that analyzed temporal dynamics of whole-body iron metabolism in response to dietary iron content changes or upon injection of LPS or iron tracer [38, 40, 41].
LPS injection (1μg/g body weight) led to a transient drop in serum iron levels which was accompanied by iron accumulation in liver and spleen, but not in duodenum or red blood cells (see blue lines in Fig 3A, 3B, 3D, 3J and 3K and S1 Fig). In order to further characterize iron re-distribution, we quantified iron transport regulators in peripheral organs using qPCR and western blotting. The iron exporter Fpn was downregulated at the protein level in liver and spleen which likely explains iron accumulation in these compartments (blue lines in Fig 3G and 3L). Fpn downregulation is expected to be controlled by hepcidin, as we found inflammatory signaling pathways controlling hepcidin expression (serum IL6, liver pSTAT3) to be activated upon LPS injection (see Fig 3H and S1 Fig). Accordingly, we found liver hepcidin mRNA expression to be upregulated upon LPS injection (Fig 3C and S1 Fig). This most likely translates into increased levels of bioactive hepcidin in the circulation, as hepatic mRNA levels typically closely reflect the amount of released hepcidin peptides [50–52]. In addition, we also observed a strong downregulation of Fpn mRNA expression in liver and spleen (blue lines in Fig 3F and S1 Fig), as has previously been reported [33]. Hence, two independent pathways exist for Fpn regulation, a hepcidin-dependent post-transcriptional mechanism and a transcriptional mechanism that may be independent of hepcidin.
We fitted our mathematical model to the time-resolved measurements summarized in Table 1 using the procedure described in the Model section and S1 Text. The final model fits (Fig 3, blue lines and S1–S4 Figs) show a good quantitative match to the data: When averaged over all 344 data points (characterized by the means and standard deviation of 2–6 individual replicates), the best fits were about one standard deviation from the experimental data (χ2/N = 1.08). Furthermore, the model qualitatively reproduced the following key features of iron metabolism and accurately described their dynamics:
From this we concluded that our mechanistic model accurately describes iron homeostasis under various experimental conditions.
We next challenged the model by testing its accuracy in predicting conditions not used for model calibration (data summarized in Table 2). First, we tested how our model would predict the LPS-response in iron-loaded animals, a condition where both STAT3 and SMAD1/5/8-signaling are altered. This was done by changing the value of the parameter that describes dietary iron uptake ([Fefood], see S1 Text) by the same fold-change as the experimental increase in dietary iron. Our model predicted that the LPS-induced dynamics of iron-related parameters for mice maintained on an iron-rich diet should be comparable to those maintained on a regular diet, albeit starting from a higher set point (red lines in Fig 3). Experimentally we maintained male C57BL/6-mice on an iron-rich diet (20000 ppm iron) containing 100 times more iron than the normal diet for 4 weeks, and subsequently injected a single dose of LPS (1μg/g body weight). The experimental data confirm the predictions of the model: the measured dynamics matched the model prediction for most variables (serum, liver and duodenum iron, liver hepcidin, BMP6 mRNA, liver pStat and pSmad, liver Fpn mRNA and protein). Deviations of the model fit from the experimental data were observed for diet-induced changes in the initial concentrations of spleen iron, RBC iron, spleen Fpn protein (Fig 3D, 3K and 3G). We concluded that our model reflects the main features of the response of iron metabolism towards LPS stimulations under iron overload conditions.
To further validate the model, we challenged it with perturbations inside the regulatory network and compared the results to previously published data. Two experimental strategies have been performed to block the hepcidin negative feedback loop: First, disruption of Bmp6 signaling specifically in liver was achieved by using data derived from conditional SMAD4 knockout mice [53]. Second, a knockin mouse was established (Slc40a1C326S), which harbors a Fpn mutation that can no longer bind to hepcidin [37]. We reproduced these new conditions in the model by setting the SMAD expression level to zero (this has an influence on the hepcidin expression level calculated via the hepcidin promoter model, see Model section) or simultaneously setting the parameter values corresponding to the hepcidin effect on ferroportin degradation to zero (k 2 l i v e r, k 2 d u o, k 2 s p l e e n, see Eq. 7 in S1 Text), respectively. In both cases, the model predictions, increased serum and liver iron pools and decreased spleen iron content (see Fig 4A), were confirmed by the experimental data [37, 53]. Moreover, as in the experimental data, Fpn resistance to hepcidin resulted in elevated hepcidin expression, whereas loss of SMAD-signaling resulted in a marked drop of hepcidin expression (Fig 4A).
We next explored the organ-specific role of Fpn regulation by hepcidin in more detail, by simulating the response of the system under the assumption that hepcidin-resistant Fpn is expressed in a tissue-specific manner. To this end, the parameter values corresponding to the hepcidin effect on ferroportin degradation (k 2 l i v e r, k 2 d u o, k 2 s p l e e n, see Eq. 7 in S1 Text) were separately set to zero. Only the elimination of hepcidin-mediated Fpn regulation in the duodenum had a systemic effect on iron levels by increasing iron levels in serum, spleen, liver and the total body. By contrast, the simulation of hepcidin-resistant Fpn in liver or spleen resulted only in a local effect with a decreased iron pool in the respective organ and minimal changes in the other organs (Fig 4B). We conclude that hepcidin shows its strongest effect on steady state iron pools by regulating Fpn in the duodenum, where iron is taken up and lost. By contrast, Fpn level in liver and spleen predominantly affects the fraction of iron that circulates through these organs. Mouse models with tissue specific hepcidin-resistance have not been described so far. However, tissue-specific deletion of the FPN gene has been studied in Fpnflox/flox mice carrying an intestine-restricted villin-Cre transgene that is inducible by tamoxifen [54]. Upon tamoxifen-induced expression, the Cre recombinase cuts out the intestinal FPN gene, and this leads to severe deficiency in the blood, liver and spleen [54]. This also indicates a systemic effect of duodenal Fpn levels on the body iron pools.
We further compared changes in the dynamics of iron-related parameters in response to chronic inflammation predicted by the model to qualitative knowledge from the literature. Chronic inflammation can cause anemia of inflammation, which is characterized by decreased serum iron and hemoglobin levels [55]. In this disease, inflammatory cytokines induce hepcidin expression, which results in reduced Fpn-mediated iron export, thus limiting iron availability in the blood stream for erythropoiesis in the bone marrow. If the inflammation persists anemia will develop as a consequence [56]. Chronic inflammation was mimicked in the model by assuming a constant source term in the equation that describes the time derivative of the LPS concentration (see Model section and Eq. 3 in S1 Text). Our model reflects the known iron-related alterations of chronic inflammation: during persistent inflammatory stimulation by LPS, serum iron levels decreased by about 85% within 2 days, while RBC iron decreased over a longer time scale until it finally stabilized at about 10% of the normal level after two months (Fig 4C). In agreement with data reported in [57], we find that the liver iron content increases in response to chronic inflammation (Fig 4C).
The above described model analyses, comprising both fitting and prediction, show that our mechanistic model is able to accurately describe a broad spectrum of perturbations on the quantitative level in many cases and on the qualitative level in most cases. Hence, our model allows us to mechanistically dissect systemic iron homeostasis and to evaluate to which extent hepcidin or Fpn contribute to the establishment of hypoferremia or maintenance of organ iron pools.
Fpn is regulated on the transcript and protein level in response to inflammation [9], with hepcidin-dependent regulation considered to be the major contributor to hypoferremia. As our model encompasses both mechanisms of Fpn regulation (cell-autonomous transcriptional inhibition and hepcidin-mediated regulation), it allowed us to analyze the relative contribution of each mechanism separately by simulating the LPS response when either the transcriptional or the post-translational LPS effect on Fpn protein levels was eliminated. The initial condition of the simulations was the non-perturbed steady state of the system with both regulatory mechanisms included. Starting from this initial state, the reaction to LPS injection was simulated while keeping either hepcidin levels or Fpn mRNA levels constant (the time derivatives of these model variables were set to zero, see Eqs. 1 and 6 in S1 Text). Interestingly, the lack of the LPS mediated hepcidin induction showed an almost normal drop in serum iron levels (75% of the complete model, see Fig 5A, red line). By contrast, the removal of the Fpn transcriptional control in response to LPS had a stronger effect and alleviated hypoferremia to 50% of the control (see Fig 5A, green line). The strong effect of inflammation-mediated transcriptional regulation of Fpn became even more evident in animals with dietary iron overload. Removal of hepcidin regulation in this case led to a near to normal level of inflammation-induced hypoferremia, while the serum iron drop induced by hepcidin regulation alone was reduced (Fig 5A). This shows that hypoferremia arises by a combination of hepcidin-dependent and independent mechanisms with the total effect on the serum iron level being less than additive.
Even though the relative contribution of transcriptional control of Fpn expression seems dominant over the hepcidin mediated control of Fpn in the high-iron setting, hepcidin does play a critical role. The increased basal level of hepcidin in iron loaded animals reduces the Fpn protein half-life (via the terms k2[hep], see Eq. 7 in S1 Text and S6 Fig). This in turn couples Fpn protein levels more tightly to Fpn mRNA levels (compare red and blue lines in Fig 5B). As a result, induction of hypoferremia by transcriptional inhibition of Fpn and the recovery of serum iron levels are faster in iron loading conditions, with and without inflammatory control of hepcidin (Fig 5A).
From this, we conclude that the transcriptional control of Fpn expression is the major determinant of the degree of hypoferremia. It amplifies the effect of hepcidin-mediated protein degradation in an acute inflammatory setting.
We next focused our analysis on iron distribution among body compartments under conditions when mice were maintained either on an iron-enriched diet or on a regular diet. Our data reflect the role of the liver as the main iron storage organ of the body, which is well described in the literature [58, 59]. Fig 6A shows the measured and fitted distributions of iron between the different pools in mice fed a normal or iron-enriched diet, respectively (same experiment/data as in Fig 3). The iron content in all compartments except for the red blood cells increased after 4 weeks of dietary iron overload. The liver shows both the highest relative and absolute change of iron content: a 10-fold increase in liver iron levels corresponding to an additional ∼600μg iron per mouse (arrows in Fig 6A). The model fits these data quantitatively.
We experimentally checked whether liver iron accumulation correlates with a low expression of the iron exporter Fpn. Unexpectedly, hepatic Fpn protein levels were increased in mice fed with an iron-enriched diet, even though the BMP6/Hepcidin pathway was activated as expected (S5A Fig). Increased Fpn protein levels can be explained only partially by transcriptional regulation, as we observed only a modest increase in FPN mRNA levels (S5A Fig). While the macrophage marker F4/80 was increased in livers of LPS treated animals, this was not observed for iron-loaded animals compared to animals on a normal diet (S5B Fig), suggesting that increased hepatic Fpn protein levels in iron-loaded animals are not explained by the infiltration of macrophages in the liver that express higher levels of Fpn compared to hepatocytes. Thus, our data may suggest that the IRE/IRP-mediated translational regulation of Fpn expression is more pronounced than hepcidin-mediated Fpn regulation in the case of sustained dietary iron overload.
These experimental findings indicate that hepatic iron accumulation cannot be explained by hepcidin controlled Fpn degradation. As a matter of fact, the increase in Fpn levels should counteract iron accumulation. Hence, liver specific iron uptake must be assumed to be considerably higher to make up for the increased iron export. To investigate the cause for the increase in liver iron levels, we tested additional Fpn-independent regulatory mechanisms for the iron exchange between liver and serum and performed model selection based on their fitting performance. Two mechanisms have been included in the final model variant (these were also present in all simulations shown above).
Fig 6B shows for the full model and models lacking one or both of these mechanisms the fitting performance of liver iron accumulation upon dietary iron overload and in HAMP-KO mice. We find that the full model is the only one that fits all liver iron data. We conclude that the hepcidin-Fpn axis alone is not sufficient to explain liver iron accumulation during dietary iron overload.
Chronic iron overload causes irreversible organ damage as exemplified by disease conditions such as hereditary hemochromatosis [67]. Our own data and published data [14, 41] show that even pronounced increases in dietary iron levels translate into a comparably small accumulation of iron in organs (iron homeostasis). Specifically, we observed that a 100-fold increase in dietary iron content results in a less than two-fold increase in most body iron pools. The most dramatic change with a ∼10-fold increase occurs in the liver (see arrows in Fig 6A). Thus, all pools respond to dietary changes in a (much) less than proportional manner, both in vivo and in the model (see Fig 6A).
To better understand the mechanisms that maintain iron homeostasis, we systematically perturbed each parameter of the model (e.g. protein synthesis/degradation rates, iron import/export rates) and analyzed which one shows the most pronounced impact on serum iron levels and induces the measured response of body iron pools to changes in the diet (see S7 Fig). This analysis reveals that limited iron uptake in the duodenum (see below) is the most critical mechanism to buffer all body iron pools against increases in dietary iron.
Apical uptake of iron from the lumen into duodenal enterocytes occurs via the divalent metal iron transporter 1 (DMT-1). This process is regulated locally by cellular iron levels and hypoxia [68]. Under iron rich conditions, iron absorption into enterocytes decreases due to downregulation of iron transporters by the IRE/IRP system and hypoxia-inducible factor 2 [59, 69–71]. Based on this, we assumed in the model that iron uptake from the diet into duodenal enterocytes shows saturation (this mechanism was also present in all simulations shown above). This was implemented by using a Michaelis-Menten type equation to describe the duodenal iron uptake rate as a function of the dietary iron content (see Eq. 14 in S1 Text). The Michaelis-Menten constant of this equation (parameter Kduo, see S1 Text) was fitted to approximately five times the normal iron diet content which results in strong saturation for the 100-times iron enriched diet. Eliminating this saturation by assuming a linear dependence of the absorption rate on the dietary iron content led to a loss of homeostasis for increasing dietary iron content (compare blue and red lines in Fig 7).
The model predicted that a 100-fold increase in dietary iron should result in a 2.5-fold increase in dietary iron uptake (40 and 100 μg iron per day in mice under normal and iron overload conditions, respectively). We have indirectly tested this prediction by determining the iron contents in individual organs and the remaining carcass and summing up the individual pools for mice on normal and high iron diets (Fig 6A). In line with the hypothesis of iron-uptake saturation, we found that raising the iron diet content by 100-fold caused an only 3-fold increase in total body iron levels.
Our modeling analysis showed that the saturated uptake mechanism does not explain homeostasis upon a reduction of dietary iron content (red and blue lines in Fig 7). We therefore analyzed the impact of other mechanisms and found that hepcidin was required for homeostasis at low and normal dietary iron. Elimination of hepcidin from the model was implemented by setting the hepcidin effect on ferroportin protein degradation to zero (parameters k 2 l i v e r, k 2 d u o, k 2 s p l e e n and k 2 r e s t in Eq. 7 of S1 Text). Homeostasis in the range of low and normal dietary iron content is lost upon elimination of hepcidin from the model (green line in Fig 7). Furthermore, upon combined elimination of both mechanisms (uptake saturation and hepcidin regulation) homeostasis is lost over the full range of iron concentrations.
Thus, we conclude that both, regulated iron uptake by mechanisms operational in the duodenal enterocyte, and hepcidin-controlled responses that integrate information about systemic iron availability, are critical for stabilizing whole-body iron levels over a broad range of dietary iron concentrations.
Even though the experiments used for model training comprise measurements perturbations in several parts of the assumed regulatory network, uncertainties exist in the parameter estimation and thus in the predictive power of the model. The model parameters could be better constrained by directly determining additional parameters or by increasing the sampling rate of the time course measurements, thereby reducing the degrees of freedom during the fitting process. However, direct measurements of iron transport rates or synthesis/degradation rates of regulatory proteins (e.g. FPN, DMT1 or ZIP14) or IRE/IRP binding activities are complex and would have gone beyond the purpose of this study.
For ethical reasons, measurements were restricted to a minimum number of time points that allow an evaluation of the time dynamics, since additional sampling would have required the sacrifice of more animals (e.g. 12 more animals for each time point for the diet+LPS experiment).
Due to these limitations, not all model parameters could by determined with the same precision (see S1 Text for the uncertainty interval of each parameter value). For example, the available data allow a good estimation of the hepcidin degradation rate, with values between 0.067 and 0.07 h−1 within the 30 best fitting parameter sets obtained using a local multi-start optimization strategy (see Calibration in S1 Text). In contrast, only the magnitude order can be estimated for the degradation rate of BMP6 mRNA, since its value varies between 1 and 9.5 h−1 within the same 30 best fitting parameter sets. To avoid uncertainty, we have ensured the robustness of the model predictions by assessing the simulation results for these 30 parameter values combinations that fit best the data, and did not restrict ourselves to the best fitting parameter set.
The analysis of the quantitative impact of different mechanisms in the model simulations showed that some regulations (e.g., hepcidin effect on ferroportin) are crucial for the systems response in certain conditions, but play only a supporting role for other experimental conditions (see previous sections). However, all variables and regulatory interactions included in the model were selected because they significantly improved the overall fitting of the calibration data set. Nevertheless, the model topology was chosen as simple as possible in view of the studied experimental conditions, since our aim was not to build a complete, but rather a minimal model that is able to describe these conditions. Thus, several regulatory mechanisms of iron metabolism described in the literature have been omitted (e.g. regulation of hepcidin during inflammation via activin B [33, 72]; ceruloplasmin/hephaestin for iron export by ferroportin [11, 12]; negative regulation of hepcidin via erythroferrone [73]). Even though they were apparently not critical for the reproduction of the experimental measurements, some of these mechanisms could also have been active in the experimental perturbations included in the study. Not including them in the model may explain some of the quantitative differences still present between model fits/predictions and measurements. The same holds for the explicit modeling of sub-organ compartments (such as Kupffer cells, hepatocytes, endothelial cells and stellate cells in liver), that have been omitted to keep the model as simple as possible and because their experimental acquisition was not feasible.
Our model predicted that hypoferremia during acute inflammation requires transcriptional downregulation of ferroportin mRNA, whereas post-transcriptional regulation via hepcidin plays a lesser role. This finding could be directly validated experimentally by ablating the inflammatory signaling branch controlling hepcidin expression. For instance, LPS injection could be performed in hepatocyte-specific STAT3 knockout mice [74] or in a knock-in mouse model in which hepcidin is expressed under the control of a heterologous promoter [75]. Additionally, transcriptional regulation of ferroportin could be studied in more detail using bioinformatics analyses of promoter sequence motifs and reporter gene assays in cell culture. Several transcription factors have been implicated in transcriptional regulation of ferroportin, including hypoxia inducible factor 2 (HIF2α), nuclear factor erythroid (Nrf2), metal-regulatory transcription factor 1 (MTF-1) and estrogen receptor [76]. Identification of the regulatory mechanisms that predominate in vivo would allow interfering with transcriptional regulation of ferroportin during inflammation and hence lead to a better understanding of its importance in inflammation, host-pathogen interactions and cancer.
According to our model, changes in dietary iron are mainly compensated by the BMP/SMAD/hepcidin negative feedback loop. This mechanism of iron homeostasis is well-known and the model predictions could be validated quantitatively by testing how changing dietary iron affect circulating iron levels in transgenic mice which express hepcidin under control of a heterologous promoter [75]. Taken together, our model represents a tool that may help to guide the design and interpretation of future experiments.
We succeeded in establishing the first mechanistic model that explains the distribution of iron between pools, predicts accurately the responses of iron-loaded mice to stimulation by LPS, and evaluates quantitatively the roles of Fpn and hepcidin during the establishment of hypoferremia. The model supports the important role of hepcidin for the inflammatory response but shows that transcriptional regulation of Fpn after an inflammatory insult is required additionally to efficiently establish hypoferremia.
We also provided insights into the mechanisms that maintain iron homeostasis in response to alterations in the dietary iron content. Our results suggest that non-transferrin bound iron uptake by the liver (presumably through ZIP14) and hepatic iron storage in ferritin are the critical determinants for liver-specific iron accumulation during iron loading conditions. Finally, the model shows that regulation of duodenal iron uptake in an enterocyte intrinsic manner is as important as hepcidin-mediated regulation for the establishment of serum iron homeostasis under dietary iron overload conditions.
All mouse breeding and animal experiments were approved by and conducted in observation of the guidelines of the EMBL Institutional Animal Care and Use Committee.
The following data from male mice of the “food iron treatment” group (high, normal and low iron) from [41] were used: organ iron (liver & spleen), serum iron, transferrin saturation, gene expression in liver (Hamp1, Bmp6, Smad7, liver Id1). The data from this article was supplemented with western blot analysis (liver: beta-Actin, pSMAD1/5/8, pSTAT3; spleen: beta-Actin, Fpn1, Ferritin L) and qPCR (spleen: Ireg1/Fpn1, Tfr1) carried out in our lab.
Tracer iron levels in C57BL6 wild-type mice maintained on an iron-deficient, iron-adequate, or iron-loaded diet were taken from [40] (Organ Fe59 contents scaled to the whole mouse body). The data from all tissues not explicitly considered here (intestine, muscle, heart, fur, lungs, testis, kidneys, fat, stomach and brain) was summed up and used to fit the compartment ‘other organs’ in our model. Data corresponding to one time point t was rescaled with the same factor for all compartments, such that the overall tracer sum is equal to 100exp(−rt). In [40], a fixed iron lost rate r of 0.5% per day was used as derived from literature. Here, the iron lost rate r was assumed to be a model parameter, whose value has been determined by fitting the data.
Hematological parameters were determined from blood collected in heparin vials (Sarstedt) using the ABC ScilVet analyzer (ABX Diagnostics). Tissue non-heme iron was measured in dried tissue samples using acid extraction and bathophenantroline (Sigma) as chromogen as described by [77]. Iron content of carcasses was determined in the same way with the following modifications: carcasses were ground in a mortar after drying, acid extraction was performed with the complete material from each carcass and iron measurement was performed on duplicate subsamples of the cleared extract. Plasma/serum iron was measured using the bathophenantroline-based method of the SFBC kit 80008 (BIOLABO, France) adapted to the 96-well format by scaling to 40 μL serum/plasma sample volumes [78]. Unsaturated iron binding capacity (UIBC) in plasma/serum was determined using the UIBC kit 97408 (BIOLABO, France) adapted to 96-well format by scaling to 20 μL serum/plasma sample volumes [37].
RNA extraction cDNA synthesis and qPCR RNA was purified from tissue samples using Trizol reagent (life technologies) according to the manufacturers protocols with one additional ethanol-wash of RNA prepared from liver to reduce high A230nm. 2 ug total RNA were reverse transcribed using random hexamers and RevertAid (Fermentas). qPCR was performed on a StepOne thermocycler (Applied Biosystems) using the SYBR-green Master Mix (Applied Biosystems) and primers listed in S1 Text. Relative mRNA expression was calculated by the delta Ct method and normalized to the reference genes Gapdh or beta-Actin as indicated.
Snap-frozen tissues were homogenized in RIPA buffer supplemented with protease inhibitors (Roche) and protein concentrations were determined using the DC protein assay (BioRad). Total protein (60 ug) was subjected to western blot analysis with the antibodies listed in S1 Text. Western blots were imaged and analyzed using the Fusion-FX system (Vilber Lourmat). Protein levels were normalized by the level of beta-Actin.
|
10.1371/journal.pmed.1002693 | Machine learning assessment of myocardial ischemia using angiography: Development and retrospective validation | Invasive fractional flow reserve (FFR) is a standard tool for identifying ischemia-producing coronary stenosis. However, in clinical practice, over 70% of treatment decisions still rely on visual estimation of angiographic stenosis, which has limited accuracy (about 60%–65%) for the prediction of FFR < 0.80. One of the reasons for the visual–functional mismatch is that myocardial ischemia can be affected by the supplied myocardial size, which is not always evident by coronary angiography. The aims of this study were to develop an angiography-based machine learning (ML) algorithm for predicting the supplied myocardial volume for a stenosis, as measured using coronary computed tomography angiography (CCTA), and then to build an angiography-based classifier for the lesions with an FFR < 0.80 versus ≥ 0.80.
A retrospective study was conducted using data from 1,132 stable and unstable angina patients with 1,132 intermediate lesions who underwent invasive coronary angiography, FFR, and CCTA at the Asan Medical Center, Seoul, Korea, between 1 May 2012 and 30 November 2015. The mean age was 63 ± 10 years, 76% were men, and 72% of the patients presented with stable angina. Of these, 932 patients (assessed before 31 January 2015) constituted the training set for the algorithm, and 200 patients (assessed after 1 February 2015) served as a test cohort to validate its diagnostic performance. Additionally, external validation with 79 patients from two centers (CHA University, Seongnam, Korea, and Ajou University, Suwon, Korea) was conducted. After automatic contour calibration using the caliber of guiding catheter, quantitative coronary angiography was performed using the edge-detection algorithms (CAAS-5, Pie-Medical). Clinical information was provided by the Asan BiomedicaL Research Environment (ABLE) system. The CCTA-based myocardial segmentation (CAMS)-derived myocardial volume supplied by each vessel (right coronary artery [RCA], left anterior descending [LAD], left circumflex [LCX]) and the myocardial volume subtended to a stenotic segment (CAMS-%Vsub) were measured for labeling. The ML for (1) predicting vessel territories (CAMS-%LAD, CAMS-%LCX, and CAMS-%RCA) and CAMS-%Vsub and (2) identifying the lesions with an FFR < 0.80 was constructed. Angiography-based ML, employing a light gradient boosting machine (GBM), showed mean absolute errors (MAEs) of 5.42%, 8.57%, and 4.54% for predicting CAMS-%LAD, CAMS-%LCX, and CAMS-%RCA, respectively. The percent myocardial volumes predicted by ML were used to predict the CAMS-%Vsub. With 5-fold cross validation, the MAEs between ML-predicted percent myocardial volume subtended to a stenotic segment (ML-%Vsub) and CAMS-%Vsub were minimized by the elastic net (6.26% ± 0.55% for LAD, 5.79% ± 0.68% for LCX, and 2.95% ± 0.14% for RCA lesions). Using all attributes (age, sex, involved vessel segment, and angiographic features affecting the myocardial territory and stenosis degree), the ML classifiers (L2 penalized logistic regression, support vector machine, and random forest) predicted an FFR < 0.80 with an accuracy of approximately 80% (area under the curve [AUC] = 0.84–0.87, 95% confidence intervals 0.71–0.94) in the test set, which was greater than that of diameter stenosis (DS) > 53% (66%, AUC = 0.71, 95% confidence intervals 0.65–0.78). The external validation showed 84% accuracy (AUC = 0.89, 95% confidence intervals 0.83–0.95). The retrospective design, single ethnicity, and the lack of clinical outcomes may limit this prediction model’s generalized application.
We found that angiography-based ML is useful to predict subtended myocardial territories and ischemia-producing lesions by mitigating the visual–functional mismatch between angiographic and FFR. Assessment of clinical utility requires further validation in a large, prospective cohort study.
| Invasive fractional flow reserve (FFR, defined as the ratio of maximum flow in a diseased artery to the proximal normal maximum flow) has been a standard tool to detect ischemia-producing lesions with FFR < 0.80.
Although the current guidelines recommend the routine use of FFR for identifying ischemia-producing lesions, the majority of treatment decisions still rely on visual assessment of the degree of angiographic stenosis because of the time and expense associated with FFR-guided decision-making.
Conventional angiographic parameters cannot predict the presence of ischemia in cases in which this is affected by the size of subtended myocardium.
Integration and optimization of information about both myocardial territory and stenosis degree are expected to improve the performance of angiographic prediction of low FFR (FFR < 0.80).
A retrospective study was conducted using data from 1,132 angina patients with 1,132 intermediate coronary lesions (932 in the training dataset and 200 in the internal test dataset), who underwent coronary angiography, coronary computed tomography angiography (CCTA), and FFR to evaluate the lesion morphology, subtended myocardial territories, and inducible ischemia, respectively.
The mean absolute errors between angiography-based machine learning (ML)-derived versus CCTA-derived subtended myocardial volume were 6.26% ± 0.55% for left anterior descending artery, 5.79% ± 0.68% for left circumflex artery, and 2.95% ± 0.14% for right coronary artery lesions.
Using all clinical and angiographic features, the ML models predicted an FFR < 0.80 with an overall accuracy of approximately 80% in the test set. In the external validation, the overall accuracy for predicting FFR < 0.80 was 84%.
Angiography-based supervised ML is useful to predict subtended myocardial territories and to identify ischemia-producing lesions by mitigating the visual–functional mismatch between angiography and FFR.
The data-driven approach may support clinicians in identifying clinically relevant coronary lesions without FFR measurement and in making clinical decisions.
Performance and clinical utility require further validation in a large, prospective cohort study.
| Stratification of cardiovascular risk in patients with stable coronary artery disease is a key to identify high-risk patients who will benefit from percutaneous coronary intervention (PCI). The appropriateness of revascularization has been determined by the presence and extent of myocardial ischemia. A myocardial perfusion imaging study previously suggested that revascularization has a greater survival benefit in patients with a moderate to large degree of ischemic myocardium (≥10% of the total myocardium) [1]. Invasive fractional flow reserve (FFR, defined as the ratio of maximum flow in a diseased artery to the proximal normal maximum flow) has been a standard tool for lesion-specific hemodynamic assessment and treatment decision-making [2–4]. With abundant clinical evidence showing a significant reduction in major adverse cardiac events using FFR-guided PCI (versus angiography-guided PCI), current guidelines recommend FFR measurement when assessing intermediate coronary stenosis. However, in clinical practice, over 70% of treatment decisions still rely on a visual estimation of angiographic stenosis. This may be due to the prolonged procedure time and high short-term costs associated with FFR-guided diagnosis, as well as the need for adenosine-induced hyperemia and the fact that reimbursement systems do not favor this approach [5,6].
Although invasive coronary angiography and intravascular ultrasound (IVUS) are commonly utilized for evaluating coronary anatomy and optimizing PCI, the subjective nature of visual estimation limits the accurate estimation of stenosis severity [7]. In addition, the integration of morphologic and physiologic parameters and the identification of clinically relevant coronary lesions remain challenging [8–10]. In previous studies, the overall diagnostic accuracy of quantitative angiography for predicting FFR < 0.80 was shown to be only 60%–65% [10,11]. One of the reasons for the visual–functional mismatch is that myocardial ischemia is primarily determined by the variable size of the supplied myocardium at risk, as well as by the degree of stenosis [11]. Our previous data suggested that the application of coronary computed tomography angiography (CCTA)-based myocardial segmentation (CAMS)-derived percent myocardial volume subtended to a stenotic segment (CAMS-%Vsub) improves the diagnostic accuracy of angiographic indices used to identify ischemia-producing lesions [12,13]. Nonetheless, the necessity of concurrently performing noninvasive CCTA and invasive angiography limited the clinical utility of the mathematical model.
Machine learning (ML) techniques have emerged as highly effective computer algorithms for the identification of patterns in large datasets with a multitude of variables, facilitating the construction of models for data-driven prediction or classification [14–17]. The aims of this study were to develop an angiography-based supervised ML algorithm for predicting the CAMS-%Vsub and to build an angiography-based supervised ML model to classify lesions into those with an FFR < 0.80 and those ≥ 0.80.
Between 1 May 2012 and 31 January 2015, 5,378 consecutive patients with stable or unstable angina underwent invasive coronary angiography at the Asan Medical Center, Seoul, Korea. Preprocedural FFR and CCTA data for assessing an intermediate coronary lesion (defined as an angiographic stenosis diameter of 30%–80% on visual estimation) were available for 1,143 patients. Among them, 10 patients with tandem lesions, 10 with stented lesions, 17 with in-stent restenosis, 22 with chronic total occlusion, 10 with side branch evaluation, 145 with significant left main coronary artery stenosis, and 5 with scarred myocardium and regional wall motion abnormality on echocardiography were excluded. When FFR was measured in multiple lesions, the lesion with the lowest FFR value was selected. Following exclusions, 932 patients (932 lesions) were used for model training (the training sample). In addition, data from a nonoverlapping population of 200 stable and unstable angina patients (200 lesions) who underwent preprocedural angiography, IVUS, and FFR in a different phase (between 1 February 2015 and 30 November 2015) were used as a test sample to validate the diagnostic performance of the ML models for the prediction of FFR < 0.80 (Table 1). De-identified clinical information, including patient age and sex, was supported by the Asan BiomedicaL Research Environment (ABLE) system. All patients provided written informed consent for the procedures. The protocol of retrospective data analysis (1 January 2017 to approximately 30 November 2017) was approved by the institutional review board of the Asan Medical Center (S1 file), and a waiver for informed consent was granted. This study is reported as per the transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD) guidelines (S1 Checklist).
The external validation of the ML models was conducted in 79 angina patients (64 patients from CHA University, Seongnam, Korea, and 15 patients from Ajou University, Suwon, Korea) who underwent invasive coronary angiography and FFR to assess an intermediate coronary lesion.
Computed tomography imaging, including CCTA, was performed using first- or second-generation dual-source computed tomography (Definition or Definition Flash, Siemens, Germany). The CCTA data with the fewest motion artifacts and clearest demarcation of the coronary artery were transferred to customized software for CAMS analysis (A-View Cardiac, Asan Medical Center, Korea). After extracting the centerline of each coronary artery and the left ventricular myocardium on the computed tomographic images, the 3D Voronoi algorithm was used to assign the myocardial territories of the major epicardial coronary arteries, including the left anterior descending artery (LAD), left circumflex artery (LCX), and right coronary artery (RCA). In brief, the Voronoi algorithm is a mathematical algorithm that divides the area or space between predetermined points or lines according to the shortest distances from those points or lines [18–20]. The left ventricular myocardial volume was divided into three major epicardial coronary artery territories based on the shortest distance from the coronary artery. The CAMS-%RCA, CAMS-%LCX, and CAMS-%LAD were defined as the percentage ratios of the myocardial volumes supplied by the RCA, LCX, and LAD to the total left ventricular myocardial volume. CAMS-%Vsub was defined as the percentage ratio of the myocardial volume subtended to a stenotic coronary segment to the total left ventricular myocardial volume. Fig 1 shows an example of the CAMS analysis.
Quantitative coronary angiography was performed using standard techniques with automated edge-detection algorithms (CAAS-5, Pie-Medical, the Netherlands). After automatic contour calibration by using the known caliber of guiding catheter, angiographic diameter stenosis (DS), minimal lumen diameter (MLD), lesion length, and the proximal and distal reference lumen diameters (RLDs) were measured. Definitions of the angiographic features used for ML training are summarized in Table 2.
FFR is defined as the ratio of the mean distal coronary pressure (Pd, measured with the pressure wire) to the mean aortic pressure (Pa, measured simultaneously with the guiding catheter) at maximum hyperemia. First, “Equalizing” was performed with the guidewire sensor positioned at the guiding catheter tip. A 0.014-inch FFR pressure guidewire (Radi, St. Jude Medical, Uppsala, Sweden) was then advanced distal to the stenosis. The FFR was measured at the maximum hyperemia induced by an intravenous infusion of adenosine administered through a central vein at 140 μg/kg/min increasing to 200 μg/kg/min, to enhance detection of hemodynamically relevant stenoses. Hyperemic pressure pullback recordings were performed. A stenosis was considered functionally significant when the FFR was <0.80 [3,4].
After intracoronary administration of 0.2 mg nitroglycerin, IVUS imaging was routinely performed using motorized transducer pullback (0.5 mm/s) and a commercial scanner (Boston Scientific Scimed, Minneapolis, MN, United States) with a rotating 40-MHz transducer within a 3.2-French imaging sheath. For the 630 patients for whom preprocedural IVUS data were available, the IVUS-derived minimum lumen area (IVUS-MLA) within a stenotic segment was obtained using computerized software (EchoPlaque 3.0, Indec Systems, Mountain View, CA, USA).
The overall flow of the supervised ML models using the angiographic features is shown in Fig 2. A theoretical overview and summary of the ML algorithms and technical details are described in the supporting information (S1 Text).
First, a light gradient boosting machine (GBM) with leave-one-out cross-validation was applied to predict the CAMS-derived percent myocardial volume supplied by each coronary artery (CAMS-%RCA, CAMS-%LCX, and CAMS-%LAD). The angiographic attributes affecting each vessel territory are summarized in Table 2. The variables estimated in our pilot data on the basis of the lumen diameters of LAD, RCA, and LCX (calculated %RCA, %LCX, and %LAD) were also included as attributes (see method in S1 Text). Then, the percent myocardial volumes supplied by each coronary artery (ML-%RCA, ML-%LCX, and ML-%LAD), as predicted by the algorithm, were used for the next step.
The second step was to build a model to predict the CAMS-%Vsub values. The ML algorithms evaluated were ordinary least squares (OLSs), ridge and lasso regressions, elastic net, random forests, extra trees, GBM, light GBM, CatBoost, and multilayer perceptrons. A 5-fold cross-validation scheme divided the training sample into five nonoverlapping partitions (method described in S1 Text). Each partition was rotated to be the validation set, with the remaining partitions being used as the training set (S1 Fig). As attributes, the ML-%RCA, ML-%LCX, and ML-%LAD were added to the angiographic features affecting the CAMS-%Vsub (Table 2). The mean absolute errors (MAEs) and mean squared errors (MSEs) were the metrics used to evaluate the performance of the models for predicting the CAMS-%Vsub values.
To develop the binary classifiers to separate the lesions with an FFR < 0.80 from those ≥0.80, 43 clinical and angiographic features including age, sex, involved segment (proximal LAD, mid LAD, distal LAD, proximal RCA, mid RCA, distal RCA, proximal LCX, distal LCX, first and second obtus marginalis), and the angiographic features affecting vessel territories, CAMS-%Vsub, and lesion severity were used and summarized in S2 Table. The evaluated algorithms were K-nearest neighbor, binary class L2 penalized logistic regression, support vector machine, random forest, extra tree, AdaBoost, light GBM, CatBoost, Gaussian naïve Bayes, and multilayer perceptron (S1 Text). The receiver operating curve (ROC), which was based on the relative performances considering the whole range of possible probability thresholds (from 0 to 1), has an area that ranges from 0.5 for classifiers without any prediction capability to 1 for perfectly classifying algorithms. Analyses based on precision-recall curves were also conducted. Using a 5-fold cross-validation scheme (S1 Fig), the accuracy was calculated by averaging the accuracies over the five tests performed in the multiple rounds of cross-validation. For a nonbiased assessment of the performance for identifying lesions with an FFR of <0.80, the classifiers that had been previously built on the training samples were applied to a completely independent test set of 200 lesions enrolled in the different phase.
In the training set, the algorithms were independently trained on the 200 train-validation random splits with a 3:1 ratio by bootstrap, and the average performances and 95% confidence intervals were calculated. In the 200 bootstrap replicates obtained by random sampling of 50 out of the 200 test samples, the average performance and bootstrap confidence intervals were also calculated.
The statistical analyses for evaluating patient and lesion characteristics at baseline were performed using SPSS (version 10.0, SPSS, Chicago, IL, USA). All values are expressed as means ± 1 standard deviation (continuous variables) or as counts and percentages (categorical variables). Continuous variables were compared using unpaired t tests; categorical variables were compared using χ2 statistics. A p-value < 0.05 was considered statistically significant. ROCs were analyzed using MedCalc Software (Mariakerke, Belgium) to assess the best cutoff for angiographic DS or IVUS-measured lumen area to predict FFR < 0.80 with maximal accuracy.
The clinical characteristics and angiographic data of the patients in the training and test sets are summarized in Table 1. The mean age was 63 ± 10 years, and 76% were men. The evaluated vessels were LAD in 63%. The overall CAMS-%RCA, CAMS-%LAD, and CAMS-%LCX were 27.2% ± 9.3%, 42.4% ± 7.2%, and 27.7% ± 9.6%, respectively. The CAMS-%Vsub was 31.1% ± 10.2%. FFR < 0.80 was shown in 41.6% of the lesions.
By applying the light GBM to the training sample, the feature importance metrics for determining the CAMS-derived percent myocardial volume subtended to each coronary artery were determined, with the values being summarized in S1 Table. The estimated percent myocardial volumes (calculated %RCA, %LCX, and %LAD) based on the proximal vessel diameters were the most important features for predicting the CAMS-%RCA, CAMS-%LCX, and CAMS-%LAD, respectively. Additionally, the presence of ramus intermedius, a diminutive RCA, and an apical LAD curve affected the vessel territories. With the light GBM and leave-one-out cross-validation, the MAEs and MSEs were 5.42% and 7.10%, respectively, for predicting CAMS-%LAD, 8.57% and 11.28% for predicting CAMS-%LCX, and 4.54% and 6.51% for predicting CAMS-%RCA.
Table 3 summarizes the diagnostic performances of the various ML models used to predict CAMS-%Vsub. Among the models, the MAEs between the ML-predicted percent myocardial volume subtended to a stenotic segment (ML-%Vsub) and the CAMS-%Vsub were minimal with the use of the elastic net algorithm (6.26% ± 0.55% for LAD lesions, 5.79% ± 0.68% for LCX lesions, and 2.95% ± 0.14% for RCA lesions). When the elastic net algorithm was applied to all cases, the overall MAE was 5.39%. Table 4 shows the feature importance metrics by elastic net for the prediction of CAMS-%Vsub.
To classify the lesions into those with an FFR < 0.80 versus ≥ 0.80, 43 clinical and angiographic features affecting vessel territories, CAMS-%Vsub, and lesion severity were used for ML (S2 Table). Based on the feature importance metrics by CatBoost, the top-12 features for determining the FFR were identified (Table 4). The ROC-based diagnostic performances of the ML algorithms are shown in Table 5, S3 Table, and Fig 3. In addition, the diagnostic performances based on the precision-recall curves are shown in S4 Table.
In the subgroup that included the 630 patients with available preprocedural IVUS data, the IVUS-MLA was 2.77 ± 1.32 mm2. When the IVUS-MLA was added as an attribute, the classifiers using L2 penalized logistic regression, random forest, and support vector machine showed an overall accuracy of 78%–80% (area under the curve [AUC] = 0.87) to predict an FFR < 0.80 that is used as a hemodynamic index requiring revascularization (Fig 3 and S5 Table).
The test samples including the 200 lesions that were not utilized during the training showed no significant differences in clinical and lesion characteristics in comparison with the training sample (Table 1). In the identification of lesions with an FFR < 0.80, angiographic DS > 53% as the cutoff derived from an ROC analysis showed a sensitivity of 74%, a specificity of 61%, and an overall accuracy of 66% (AUC = 0.71). In addition, an IVUS-MLA < 2.34 mm2 had a sensitivity of 53%, a specificity of 79%, and an overall accuracy of 67% (AUC = 0.72).
Using clinical and angiographic features, the overall diagnostic accuracies of the ML classifiers (L2 penalized logistic regression, support vector machine, and random forest) in the test set were approximately 80% for predicting an FFR < 0.80 (AUC = 0.84–0.87, Table 5 and S6 Table). Table 6 summarizes the performances with bootstrap confidence intervals in the 200 bootstrap replicates for each of the training and test sets.
By adding the IVUS-MLA, the classifiers using L2 penalized logistic regression and support vector machine achieved an overall accuracy of 78%–79% in the test set (AUC = 0.86–0.87, Fig 4 and S5 Table).
In the external validation cohort including 79 patients, the age was 59.6 ± 9.0 years, and 58 (73.4%) were men. An FFR < 0.8 was seen in 25 (31.6%) lesions. The angiographic DS and MLD were 48.3% ± 8.0% and 1.64 ± 0.39 mm, respectively. The performances of the ML models for the prediction of FFR < 0.8 were shown in Table 5.
This study demonstrated that (1) angiography-based ML predicted the CAMS-%Vsub with an MAE of 6.26%, 5.79%, and 2.95% for LAD, LCX, and RCA lesions, respectively, and (2) for the identification of ischemia-producing lesions with a FFR < 0.80, the ML classifiers (L2 penalized logistic regression, support vector machine, and random forest) using the angiographic features showed an overall diagnostic accuracy of 80% (maximal AUC = 0.87), which was greater than that of angiographic DS criterion (66%, AUC = 0.71) or even that of the IVUS-MLA threshold (67%, AUC = 0.72).
Assessment of the myocardial mass at risk is of great importance because the presence and extent of ischemic myocardium determines the clinical relevance of revascularization [21–22]. A recent meta-analysis suggested that, in comparison with medical therapy, PCI significantly reduces mortality in patients with objective ischemia documented by functional tests [21]. Moreover, myocardial perfusion imaging suggests that revascularization has a greater survival benefit in patients with a moderate to large degree of ischemic myocardium [1]. These data have provided an insight into a higher-risk population that may benefit from an approach that incorporates ischemia-guided revascularization.
In daily practice, lesion-specific FFR is used to identify ischemia-producing lesions and to decide whether or not to treat it [2–4]. Although coronary angiography and IVUS have been commonly utilized to assess lesion severity, the diagnostic accuracy for predicting an FFR < 0.80 by angiographic DS or IVUS-MLA alone is <60%–70%, which restricts their clinical utility in treatment decision-making [8–10]. Similarly, the current study showed poor diagnostic accuracies for the detection of FFR < 0.80 using angiographic DS > 53% and IVUS-MLA < 2.34 mm2 based on the ROC analysis (66% and 67%, respectively). One of the reasons for the visual–functional mismatch is that myocardial ischemia is also determined by the variable size of the supplied myocardium, as well as the degree of stenosis [11].
In our previous study, the use of CAMS-Vsub improved the diagnostic performance of angiographic MLD and/or IVUS-MLA for the prediction of FFR < 0.80 [12,13]. Although a mathematical model using Vsub/MLD4 > 6.26 increased the accuracy to 82%, it could be applied only when the patient underwent noninvasive CCTA prior to catheterization. The current angiography-based ML model showed an overall MAE of 5.39% for predicting the CCTA-measured %Vsub. During the procedure, the angiographic prediction of the amount of supplied myocardium supports clinicians by confirming the clinical relevance of revascularization treatment in lesions with a large area of myocardium at risk and by precisely identifying the ischemia-producing lesions by reducing the discrepancy between anatomical and functional severity.
Several approaches for an FFR approximation of FFR using angiography-based models have recently been introduced [23–26]. A virtual functional assessment index and quantitative flow ratio based on computational fluid dynamics have shown the overall accuracies of 80%–86% in predicting an FFR < 0.80. Those approaches require a 3D reconstruction of at least two angiographic projections without foreshortening or overlapping vessels and the subsequent computational analyses. Using the clinical and 2D angiographic features affecting the subtended myocardial mass and degree of stenosis, our current ML classifiers predicted an FFR < 0.80 with an overall diagnostic accuracy of 80%. Therefore, the ML models not only reduce procedural expense by avoiding FFR testing but also provide information on the subtended myocardial territory that cannot be predicted by the FFR value. Ultimately, this data-driven approach extends the role of angiography in decision-making for the management of intermediate coronary stenosis.
Although traditional statistical methods validate the association between specific features and an endpoint, the development of a prediction model remains challenging, particularly in the setting of a nonlinear relationship between a factor and an outcome, interactions among variables, and the presence of many predictor variables. ML, an application of artificial intelligence, provides the ability to automatically learn a task without being explicitly programmed [14–17]. The algorithms attempt to balance two competing interests, “bias and variance,” which are summarized by loss functions to optimize a prediction model. Using angiographic features, both regression and decision tree models showed good performance in the prediction of CAMS-%Vsub, which led to greater detection of ischemia-producing lesions with reduced FFR.
The current study demonstrated the impact of the individual variables according to metrics (feature importance). For the prediction of CAMS-%Vsub, the important features were seen to be proximal segment involvement, RLD, ML-predicted territory of each vessel, the sum of the distal branch diameters, and the distance between the ostium and the narrowest site. Moreover, the key features for predicting FFR < 0.80 were MLD; %DS; age, proximal vessel size of LAD, LCX, and RCA; lesion length; distance between the ostium and MLD site; and the involvement of the proximal LAD, which suggested the importance of the impact of the angiographic determinant for stenosis degree and vascular territory on the FFR value. Although the rank in each algorithm is specific to the ML model, the approach may be hypothesis generating, suggesting which features are valuable for inclusion in future studies.
This study may be subject to selection bias. As the analysis included the single ethnicity and excluded significant left main disease, side branch, and diffuse and tandem lesions, the model cannot be applied generally. Although the developed models were validated in the historical test set and the external validation cohort, the possibility of overfitting cannot be completely excluded. This model did not include the computational fluid dynamics for estimating the anatomical severity. A large prospective trial is required to validate whether the models allow clinicians to dispense with FFR measurement and therefore change the current clinical practice. Finally, prespecified angiographic features were used for ML; an image-based deep learning strategy using big data is worthy of investigation to achieve optimal diagnostic performance for clinical use.
Angiography-based ML models were useful for the prediction of CAMS-%Vsub and for improving the detection of ischemia-producing lesions. The data-driven approach may support clinicians in the identification of clinically relevant coronary lesions and in treatment decision-making.
|
10.1371/journal.pgen.1005610 | Canonical Poly(A) Polymerase Activity Promotes the Decay of a Wide Variety of Mammalian Nuclear RNAs | The human nuclear poly(A)-binding protein PABPN1 has been implicated in the decay of nuclear noncoding RNAs (ncRNAs). In addition, PABPN1 promotes hyperadenylation by stimulating poly(A)-polymerases (PAPα/γ), but this activity has not previously been linked to the decay of endogenous transcripts. Moreover, the mechanisms underlying target specificity have remained elusive. Here, we inactivated PAP-dependent hyperadenylation in cells by two independent mechanisms and used an RNA-seq approach to identify endogenous targets. We observed the upregulation of various ncRNAs, including snoRNA host genes, primary miRNA transcripts, and promoter upstream antisense RNAs, confirming that hyperadenylation is broadly required for the degradation of PABPN1-targets. In addition, we found that mRNAs with retained introns are susceptible to PABPN1 and PAPα/γ-mediated decay (PPD). Transcripts are targeted for degradation due to inefficient export, which is a consequence of reduced intron number or incomplete splicing. Additional investigation showed that a genetically-encoded poly(A) tail is sufficient to drive decay, suggesting that degradation occurs independently of the canonical cleavage and polyadenylation reaction. Surprisingly, treatment with transcription inhibitors uncouples polyadenylation from decay, leading to runaway hyperadenylation of nuclear decay targets. We conclude that PPD is an important mammalian nuclear RNA decay pathway for the removal of poorly spliced and nuclear-retained transcripts.
| Cells control gene expression by balancing the rates of RNA synthesis and decay. While the mechanisms of transcription regulation are extensively studied, the parameters that control nuclear RNA stability remain largely unknown. Previously, we and others reported that poly(A) tails may stimulate RNA decay in mammalian nuclei. This function is mediated by the concerted actions of the nuclear poly(A) binding protein PABPN1, poly(A) polymerase (PAP), and the nuclear exosome complex, a pathway we have named PABPN1 and PAP-mediated RNA decay (PPD). Because nearly all mRNAs possess a poly(A) tail, it remains unclear how PPD targets specific transcripts. Here, we inactivated PPD by two distinct mechanisms and examined global gene expression. We identified a number of potential target genes, including snoRNA host genes, promoter antisense RNAs, and mRNAs. Interestingly, target transcripts tend to be incompletely spliced or possess fewer introns than non-target transcripts, suggesting that efficient splicing allows normal mRNAs to escape decay. We suggest that PPD plays an important role in gene expression by limiting the accumulation of inefficiently processed RNAs. In addition, our results highlight the complex relationship between (pre-)mRNA splicing and nuclear RNA decay.
| Eukaryotic messenger RNAs (mRNAs) undergo a series of maturation events before they are exported to the cytoplasm and translated. The complexity of alternative processing increases the likelihood of mistakes that produce aberrant mRNAs encoding defective proteins. In addition, pervasive transcription occurs across nearly the entire mammalian genome resulting in the generation of nonfunctional RNAs. Consequently, cells have evolved RNA quality control (QC) pathways to eliminate these RNAs [1,2].
The best-characterized RNA QC pathway is nonsense-mediated mRNA decay (NMD)[3]. NMD targets cytoplasmic mRNAs with premature termination codons (PTCs), a potentially dangerous class of RNAs that produce truncated and possibly dominant-negative proteins. NMD is limited in at least three important ways. First, NMD recognizes PTC-containing transcripts upon translation, so each defective transcript still produces one polypeptide. This could be harmful to cells for highly transcribed NMD targets or particularly toxic polypeptides. Second, NMD is stimulated by the presence of a splice junction to identify PTCs, so transcripts from intronless genes will generally not be recognized. Third, pervasive transcription produces nuclear transcripts that would not be targeted by the cytoplasmic NMD machinery.
Cells have additional nuclear RNA QC pathways to degrade RNAs not targeted by NMD, but the mechanisms involved remain unclear. Recently, functions for the nuclear poly(A) binding protein PABPN1 in RNA decay has been reported [4–6]. An RNA-seq study showed that knockdown of PABPN1 increases the accumulation of endogenous long noncoding RNAs (lncRNAs), several noncoding snoRNA host genes (ncSNHGs) and transcripts upstream of mRNA gene promoters [4]. In addition, the Kaposi’s sarcoma-associated herpesvirus (KSHV) produces an abundant polyadenylated nuclear (PAN) RNA during the lytic phase of viral infection. A cis-acting element, called the ENE, protects PAN RNA from PABPN1-mediated decay by forming a triple helix with the poly(A) tail [5,7,8]. PABPN1 additionally promotes the degradation of a poorly exported intronless β-globin mRNA, but not its spliced and efficiently exported counterpart, suggesting it serves a QC function for non-exportable polyadenylated RNAs. PABPN1-mediated decay has been observed in S. pombe and humans suggesting an important conserved function [9–12].
The canonical mammalian poly(A) polymerases PAPα and PAPγ (PAP), and the nuclear exosome are involved in PABPN1-mediated decay of intronless β-globin and PANΔENE reporters [5]. Several observations demonstrate that hyperadenylation by PAP promotes decay. First, knockdown of either PABPN1 or PAP stabilizes RNAs with shorter poly(A) tails. Second, knockdown of the exosome leads to the accumulation of hyperadenylated products. Third, inhibition of polyadenylation by cordycepin inhibits RNA decay. Fourth, expression of a dominant-negative PABPN1 double point mutant (L119A/L136A or LALA) that binds RNA but cannot stimulate PAP [13] stabilizes target RNAs. A global decay function for PAP is validated by the analyses reported here, so we now refer to this pathway as PABPN1 and PAPα/γ-mediated RNA decay (PPD).
PABPN1 and PAP have been extensively characterized for their roles in mRNA 3´-end formation [14]. Polyadenylation is initiated by co-transcriptional recruitment of the cleavage and polyadenylation specificity factor (CPSF) to the AAUAAA polyadenylation signal (PAS) through the CPSF30 and WDR33 subunits [15,16]. Extensive in vitro studies defined the roles of PAP, PABPN1, and CPSF in the normal polyadenylation of mRNA 3´-ends [13,17]. Without CPSF, PAP has low binding affinity for RNA, but the CPSF-PAP interaction drives binding and generation of an oligo(A) tail. PABPN1 binds the oligo(A) tail and forms a complex with PAP-CPSF-oligo(A). PAP becomes tightly tethered to the RNA, and polyadenylation is highly processive to ~200–300 nt poly(A) length. At this point, the interaction between PAP and CPSF is lost and polyadenylation becomes distributive, but this distributive polyadenylation continues to be stimulated by PABPN1.
We proposed that PABPN1-dependent and CPSF-independent stimulation of distributive PAP activity provides the polyadenylation associated with PPD [5]. Here, we refer to this as “hyperadenylation” as it occurs after the initial 3´-end formation step. To explore this globally, we performed RNA-seq following inactivation of hyperadenylation by two distinct methods. Consistent with the PABPN1 knockdown studies, we found that several classes of lncRNAs, including ncSNHGs, primary microRNA transcripts, and upstream antisense RNAs, are susceptible to PPD. In addition, we identified mRNAs and (pre-)mRNAs with retained introns that are PPD targets. Surprisingly, transcription inhibition led to a robust PABPN1-dependent hyperadenylation of a large pool of nuclear RNAs apparently due to the uncoupling of hyperadenylation from decay. Finally, we observed that a CPSF-independent poly(A) tail initiates PPD, but hyperadenylation was not sufficient for PPD in the absence of PABPN1. From these studies, we conclude that PPD is a major human nuclear RNA decay pathway.
We aimed to generate a high-confidence list of PPD targets by performing RNA-seq on polyadenylated RNA from cells in which PPD-associated hyperadenylation had been inactivated by two independent methods. For one treatment, we prepared RNA from cells after a three-day co-depletion of PAPα and PAPγ by siRNAs (siPAP)(S1A Fig). For the second treatment, we created a stable cell line expressing myc-tagged LALA under control of a tetracycline-responsive promoter (TetRP). Following a three-day induction of LALA, we collected RNA in preparation for high-throughput sequencing. Under these conditions, LALA was expressed at levels only slightly greater than endogenous wild-type PABPN1 (S1A Fig). We examined polyadenylated RNAs from detergent-insoluble nuclear fractions of control, LALA, and siPAP-treated cells and on total polyadenylated RNA from control and siPAP-treated cells. Our fractionation procedure enriches for chromatin and nuclear speckle-associated RNAs [18–20]. Admittedly, the protocol results in the loss of some detergent-soluble nuclear material, but the fractions have little cytoplasmic contamination.
We identified 1339 differentially expressed genes (DEGs) with increased (upregulated) and 1576 DEGs with decreased (downregulated) levels in at least one PPD inactivation condition (Fig 1A)(S1 Table). We defined high-confidence PPD targets to be those DEGs upregulated in all three datasets (Fig 1B and 1C). Interestingly, 39% (138/353) of the high-confidence transcripts mapped to unannotated loci in the reference genome, while only one of the 131 overlapping downregulated genes (0.8%) was unannotated. We visually inspected the sequence traces of all high-confidence transcripts and categorized them as mRNAs or one of several classes of ncRNA: promoter upstream transcripts (PROMPTs, also known as TSSa-RNAs)[21,22], antisense RNAs (AS), primary miRNA (pri-miRNA), ncSNHG, or lncRNA (S2 Table). Most of the RNAs were ncRNAs (80%, Fig 1D). We additionally performed an independent bioinformatic analysis utilizing a dataset including nearly 14,000 known and novel annotated lncRNAs (GENCODE)(see Materials and Methods). For this analysis, we observed 1178 upregulated lncRNA DEGs in at least one PPD inactivation condition and 408 of these were identified in all three data sets (S1C–S1E Fig and S3 Table). Thus, a considerable number of noncoding polyadenylated nuclear RNAs accumulate upon LALA overexpression and PAP knockdown, suggesting that these transcripts are PPD substrates.
Eukaryotic promoters produce bidirectional transcripts, but generally only one direction produces a stable RNA [22–25]. With respect to number and fold change, PROMPTs were the most responsive class of PPD targets (Fig 1D–1F). Importantly, composite RNA profiles confirmed that our visual assignment of PROMPT was accurate (Fig 1G and S1B Fig). Interestingly, we observed a small peak upstream of the transcription start site (TSS) when the entire genome was used for the composite (dotted lines), suggesting an effect beyond our high-confidence targets (solid lines). We validated the response of six PROMPTs to several PPD inactivation strategies (Fig 1H). In addition to LALA expression and PAP knockdown, we knocked down PABPN1 (siPABPN1), or co-depleted the two catalytic components of the exosome, DIS3 and RRP6 (siExo)(S1A Fig). We also inhibited poly(A) tail extension using cordycepin, an adenosine analog that acts as a chain terminator for poly(A) polymerase due to the absence of a 3´ hydroxyl group. As expected, the levels of the PROMPTs increased upon PPD inactivation, but in some cases PABPN1 knockdown did not have an effect. This is likely due to a general impairment of transcription upon PABPN1 depletion (see below).
We previously reported that an intronless β-globin reporter RNA is degraded by PPD, but its spliced counterpart is stable [5]. Therefore, we tested whether there was a correlation between number of exons and PPD susceptibility. We found that upregulated genes had significantly fewer exons (median 2) than genes from the reference list (median 7), or downregulated DEGs (median 8)(Fig 1I). Noncoding RNAs tend to have fewer exons than protein-coding mRNAs, so our results could be explained by the high proportion of ncRNAs in our dataset, rather than a direct consequence of reduced number of exons. However, even mRNA targets had significantly fewer exons than genes from the reference list (median 3 vs median of 7, p<0.0001). Moreover, the fold changes upon PPD inactivation inversely correlated with the number of exons (S2 Fig). We conclude that PPD substrates have on average fewer exons than transcripts that are not targeted by PPD. Nonetheless, a number of decay targets are spliced, demonstrating that a single splicing event is not always sufficient to confer resistance to PPD.
While mRNA targets had significantly fewer exons than the reference genes, the mRNA targets had more exons than other PPD target categories except ncSNHGs (Fig 1J). Interestingly, mRNAs also had a significantly lower fold-change upon PPD inactivation (Fig 1F) and mRNAs were expressed at higher basal levels than all other classes except ncSNHGs (S1F and S1G Fig). These data suggest within the cellular pool of the specific PPD-susceptible mRNAs, a subset is exported and thereby escapes PPD. As a result, the mRNAs are less affected by PPD inactivation than PROMPTs, which are presumably not exported.
Most mammalian snoRNAs are excised from introns, but the host genes can produce either coding or noncoding RNAs [26]. We identified several ncSNHGs in our RNA-seq analysis and additional ncSNHGs were upregulated that did not meet our stringent cutoffs. In order to obtain a more complete list of ncSNHG PPD targets, we performed qRT-PCR on 24 ncSNHGs expressed in our cell line following inactivation of PPD by several independent methods. In addition, we inactivated NMD by cycloheximide treatment, which indirectly inhibits NMD by inhibiting translation, or by knocking down the NMD factor UPF1 (S1A Fig). Strikingly, we observed largely non-overlapping clusters of ncSNHGs targeted by NMD or PPD (Fig 2A). No upregulation was observed when we used primers that detect the intron-containing transcripts (Fig 2B), so PPD targets the spliced product. We next examined the effects of inactivating both pathways simultaneously. We reasoned that ncSNHGs that evade PPD in the nucleus may be exported and degraded by NMD in the cytoplasm. However, simultaneous PAP knockdown and cycloheximide treatment did not lead to additive accumulation of PPD targets (Fig 2C), suggesting that NMD does not simply degrade ncSNHGs that escape PPD. Instead, each ncSNHG is targeted by a specific pathway.
Consistent with our observation that the number of exons inversely correlates with PPD susceptibility, intron-poor ncSNHGs were more likely to be targeted by PPD (Fig 2D). Because NMD and PPD function in the cytoplasm and nucleus, respectively, and splicing promotes mRNA export [27], we reasoned that differences in ncSNHG localization may contribute to PPD-sensitivity. To test this hypothesis, we calculated a nuclear enrichment score (NES) by dividing the fragments per kilobase of exon per million reads mapped (FPKM) in the nuclear dataset by the FPKM value in the total dataset for each expressed gene. Plotting the NESs confirmed that the nuclear lncRNA MALAT1 had a high NES (blue), while ACTB and RPL30 mRNAs received lower scores (red)(Fig 2E). Next, we compared the NES to the fold changes observed upon PPD inactivation and found that PPD targets were typically more nuclear, while non-PPD targets were more cytoplasmic (Fig 2F). Thus, the differences in nuclear retention and number of exons influence susceptibility to PPD. The simplest interpretation of these results is that fewer splicing events lead to less efficient nuclear export, which in turn increases PPD-susceptibility.
MAT2A is a high-confidence PPD target and inspection of its sequence traces revealed retention of the 3´-most intron (Fig 3A). Recent studies have established that intron retention is significantly more common in mammals than previously appreciated [28–31]. Retained intron-containing RNAs (RI-RNAs) can be degraded by NMD, but most are degraded in the nucleus by an unknown pathway [28,29,31]. We tested whether PPD affects RI-RNA decay more generally by examining MAT2A and two other RI-RNAs, OGT and ARGLU1. Each gene produced highly expressed nuclear RI-RNAs and fully spliced cytoplasmic mRNAs (Fig 3A and 3B). The presence of the retained intron is verified below (Fig 4A).
Neither ARGLU1 nor OGT was identified as a high-confidence target, but ARGLU1 was upregulated in the siPAP-total and siPAP-nuclear datasets. Similarly, cordycepin treatment increased MAT2A-RI levels ~2-fold, but this effect did not reach statistical significance (p = 0.10) and cordycepin did not affect OGT-RI or ARGLU1-RI levels (Fig 3D). While these data suggest little PPD sensitivity, none of the RI-RNAs responded to UPF1 depletion and only OGT-RI increased in response to cycloheximide, consistent with previous reports that NMD is not the general mode of decay for these RNAs [28,29,31]. To further probe a potential role of PPD in RI-RNA decay, we tested whether timing of the knockdown experiments influenced our results. When we increased siPAP treatment from three to four days, we observed statistically significant upregulation of MAT2A-RI (4.2-fold), OGT-RI (2.5-fold), and ARGLU1-RI (2.5-fold) supporting the conclusion that PPD targets RI-containing RNAs (Fig 3C and 3D).
PABPN1 knockdown increased ARGLU1-RI levels ~1.8-fold, but neither MAT2A-RI nor OGT-RI increased (Fig 3C and 3D). Unlike siPAP treatments, extended knockdown of PABPN1 did not increase RI-RNAs. Moreover, the cell morphology was generally worse for PABPN1 knockdowns compared to PAP knockdowns suggesting greater toxicity. Therefore, we hypothesized that decreases in transcription prevent accumulation of RI-RNAs upon PABPN1 depletion. To test this idea, we performed nuclear run-on (NRO) assays using the modified nucleotide, 4-thiouridine triphosphate (4SUTP), to detect nascent transcripts. We observed a general decrease in Pol II density on several genes after PABPN1 knockdown (Fig 3E). We conclude that steady-state levels of some PPD targets do not increase upon PABPN1 knockdown due to concomitant decreases in RNA synthesis rates. Importantly, we detected no change in transcription upon PAP knockdown (S3C Fig), consistent with our observation that RI-RNAs accumulate after PAP knockdown. We further corroborated the NRO results by examining nascent transcripts from live cells using a metabolic labeling protocol (S3E Fig). These results support a role for PPD in degradation of nuclear RI-RNAs but suggest that the relative rates of transcription and decay of RI-RNAs may differ from the more robustly upregulated ncRNAs such as PROMPTs. We also examined the mRNA isoform of MAT2A, OGT, or ARGLU1, and observed no general trends (S3B Fig). We suggest this is due to distinct half-lives, translation efficiencies, and/or the precursor-product relationship between a specific RI transcript and its cognate mRNA.
Initially, we attempted to examine MAT2A-RI stability by treating cells with the general transcription inhibitor actinomycin D (ActD). As expected, the mRNA degraded over time (Fig 4A). Surprisingly, the MAT2A-RI isoform was robustly hyperadenylated upon ActD treatment and the transcript persisted. We verified that this transcript corresponded to the MAT2A-RI by stripping and re-probing with a retained-intron specific probe (lanes 7–12). In addition, ARGLU1-RI and the OGT-RI transcripts were stable and hyperadenylated after ActD treatment (Fig 4A). Because these transcripts are longer than MAT2A-RI, the hyperadenylation was not as obvious as for MAT2A. Therefore, we cleaved the transcripts ~500 nt from their 3´ ends using RNase H and a specific targeting DNA oligonucleotide and examined the 3´ fragment prior to and after ActD treatment (Fig 4B). Hyperadenylated and shorter poly(A) tails were readily detected, reflecting the RI and mRNA isoforms, respectively. After ActD treatment, the hyperadenylated tails ranged from ~300–800 nt, while mRNAs were ~50–200 nt (S4 Table). NEAT1, a known ncRNA PPD target [4], was also hyperadenylated after ActD treatment (Fig 4A, lanes 13–18). In contrast, neither β-actin nor GAPDH mRNAs displayed poly(A) tail extension upon ActD treatment (lanes 13–18). Moreover, the nuclear ncRNA MALAT1, which does not have a poly(A) tail [32], was not extended upon ActD treatment.
MAT2A and ARGLU1 RNAs of intermediate lengths were hyperadenylated after ActD treatment (Fig 4A, asterisks). We observed only two bands corresponding to fully spliced and RI-RNAs after RNase H/oligo(dT) treatment, so we conclude that these RNAs are spliced, but still subject to hyperadenylation and nuclear retention. (S3D Fig). We discuss possible mechanisms of production of these RNAs in the Discussion section.
PABPN1 knockdown prevents the hyperadenylation of RI-RNAs after ActD treatment (Fig 4C, compare lanes 2 with 4). PABPN1 depletion also decreased the length of MAT2A-RI in the untreated samples (lanes 1 and 3), but the MAT2A mRNA lengths were largely unaffected. Similar results were observed with PAP knockdown (Fig 3C). Thus, PABPN1 and PAP hyperadenylate MAT2A-RI even in control cells and similar results were observed with ARGLU1-RI and OGT-RI isoforms (Fig 4C). If PABPN1 knockdown released RI-RNAs from the nucleus, the shorter poly(A) tails could be due to cytoplasmic deadenylation. However, the RI-RNAs remained predominantly nuclear upon PABPN1 depletion (S3A Fig). We conclude that RI-containing transcripts have longer poly(A) tails due to PABPN1 and PAP activity, and that this effect is exacerbated following treatment with ActD.
MAT2A-RI is targeted by PPD, but upon ActD treatment the poly(A) tail is extended and the RNA is relatively stable. One interpretation of this finding is that ActD treatment decouples hyperadenylation from decay. To test this with a different PPD target, we compared the half-lives of SNHG19 after ActD treatment with a 4SU metabolic pulse-chase assay that does not require general transcription inhibition (Fig 4D). The apparent half-life of SNHG19 in ActD was >3hr, while the pulse-chase method yielded a <30 min half-life (Fig 4E). These observations show that some PPD targets are stabilized by general transcription inhibition and highlight the potential caveats of using general transcription inhibitors to monitor nuclear RNA half-lives.
To explore the generality of the ActD-induced hyperadenylation, we collected RNA from cells treated with ActD over a 6-hr time course and digested them with RNase T1, a G-specific endonuclease, to degrade transcripts but leave poly(A) tails intact. We then detected bulk poly(A) tails by northern blot with an oligo(dT)40 probe (Fig 5A). After ActD treatment, one subset of poly(A) tails lengthened, while another population shortened over time. We observed similar effects with 5,6-dichloro-1-β-D-ribofuranosylbenzimidazole (DRB), flavopiridol, and triptolide, which inhibit transcription by mechanisms distinct from ActD (S4A Fig) [33]. Moreover, this hyperadenylation was observed in HeLa cells and primary mouse macrophages, so the effect is neither cell-type nor species-specific (S4B Fig). Admittedly, the fraction of RNAs hyperadenylated is lower than its appearance on the northern blots (Fig 5A) because more oligo(dT)40 probes will hybridize to the longer tails to increase the signal, but the hyperadenylated transcript pool nonetheless comprises a large fraction of the total poly(A) RNA.
The two bulk poly(A) pools closely mimicked our observations with RI-RNA and mRNA isoforms. For example, the shorter population was primarily cytoplasmic whereas the hyperadenylated RNAs were nuclear (Fig 5B). Moreover, the poly(A) tails were longer in the nuclear pool even in the absence of ActD and hyperadenylation was diminished in PABPN1-depleted cells (Fig 5C). Next, we used a metabolic pulse-chase assay to examine bulk poly(A) tail dynamics (Fig 5D). As expected, the cytoplasmic poly(A) tails shortened over time and ActD did not appreciably change this pattern (Fig 5E). In the absence of ActD, the nuclear poly(A) tails grew longer but disappeared over time. In contrast, in the presence of ActD, the nuclear poly(A) tails persisted and were continually extended, thereby mirroring the hyperadenylation and lack of nuclear decay observed with specific PPD substrates (Fig 4). We conclude that a large fraction of nuclear polyadenylated RNA is subject to hyperadenylation and stabilization upon general transcription inhibition.
PABPN1 and PAPα/γ are components of the 3´-end formation machinery, but whether other components, like CPSF, are involved in PPD is unknown. Even though hyperadenylation occurs after the initial polyadenylation event, CPSF may remain bound to the PAS and influence hyperadenylation or decay. To test this, we took advantage of the unusual processing of the MALAT1 lncRNA. The MALAT1 3´ end is generated by RNase P, which cleaves directly upstream of a tRNA-like element in the RNA [32]. We cloned the tRNA-like element into a TetRP-driven ENE-lacking PAN RNA reporter immediately downstream of a 35-nt A stretch (Fig 6A)(PANΔENE-A35). The processing at the MALAT1 cleavage site is efficient, with ~85% of the RNAs being cleaved by RNase P after a 2-hr transcription pulse (S5A Fig). In cells, the A35 tail was extended to ~100–500 nt (Fig 6B). Importantly, the cleaved transcript lacks an AAUAAA site, so this extension was independent of CPSF. To examine PANΔENE-A35 stability, we used a TetRP-based transcription pulse-chase strategy. After a 2-hr transcription pulse, we monitored stability of PANΔENE-A35 and PANΔENE with its natural PAS (PANΔENE-AAUAAA) and observed indistinguishable decay kinetics (Fig 6C and 6D). Moreover, knockdown of PABPN1 (Fig 6E and 6F) or LALA expression (S5B Fig) stabilized PANΔENE-A35. Thus, PPD does not strictly require CPSF or a PAS.
PABPN1, but not CPSF, stimulates polyadenylation after the initial processive polyadenylation step by increasing PAP association with RNA [13]. We previously proposed that this in vitro activity reflects the hyperadenylation required for PPD, which is further supported by the demonstration that PPD can occur in a CPSF-independent fashion (Fig 6A–6F). In principle, stimulation of hyperadenylation could be the sole requirement for PABPN1 in PPD. To test this hypothesis, we bypassed the requirement for PABPN1 in hyperadenylation by tethering PAP directly to PANΔENE RNA. We inserted six bacteriophage MS2 coat protein binding sites into PANΔENE upstream of the poly(A) tail, which allows us to tether an MS2-PAP fusion protein to PAN RNA in cells (PANΔENE-6MS2)(Fig 6G). When MS2-binding protein was expressed, PANΔENE-6MS2 was rapidly degraded in control cells (Fig 6H, lanes 5–8), but stabilized upon PABPN1 knockdown (Fig 6H, lanes 13–16). When we co-expressed PANΔENE-6MS2 with MS2-PAP, PANΔENE-6MS2 was rapidly degraded in control cells as expected (Fig 6H, lanes 1–4). Importantly, MS2-PAP was unable to rescue decay after PABPN1 depletion, despite the fact that PANΔENE-6MS2 was hyperadenylated (Fig 6H, lanes 9–12). Therefore, hyperadenylation is not sufficient to stimulate PPD in the absence of PABPN1, suggesting that PABPN1 serves multiple functions in PPD by promoting hyperadenylation and an additional step in RNA decay.
The mechanisms and regulation of nuclear RNA decay remain poorly defined, particularly in mammalian cells. Here we show that several classes of nuclear noncoding RNAs are subject to degradation by PPD including upstream antisense RNAs, ncSNHGs, pri-miRNAs, lncRNAs, and antisense transcripts. Our observations are consistent with global analyses reported by Bachand and colleagues demonstrating that PABPN1 knockdown leads to the stabilization of nuclear lncRNAs [4]. In addition, our RNA-seq and knockdown analyses revealed that specific canonical mRNAs and RI-containing RNAs are PPD targets. By using PAP knockdown and PAP-stimulation deficient PABPN1 mutant LALA as the basis of our RNA-seq experiments, these data confirm that PAP activity is necessary for the degradation of a large collection of nuclear RNAs. Given the parameters used in the RNA-seq analysis, it is likely that our high-stringency dataset is an underestimate of the number of RNAs subject to PPD. For example, a subset of ncSNHGs and the RI-RNAs were confirmed to be PPD substrates by qRT-PCR (Fig 2A) and northern blot (Fig 3C and 3D) even though these RNAs were not identified in our RNA-seq study. Based on these global and mechanistic studies we conclude that PPD is a major RNA decay pathway for nuclear polyadenylated transcripts.
The PROMPTs were the most PPD sensitive transcripts based on their fold changes upon PPD inactivation (Fig 1F) and their overrepresentation among DEGs (Fig 1D). Pervasive transcription from bidirectional promoter firing is a common feature in eukaryotes [1,22,23,25,34,35]. In S. cerevisiae, the resulting divergent transcripts are terminated by the Nrd1-Nab3-Sen1 (NNS) pathway due to an over-representation of binding sites for the Nrd1p and Nab3p proteins upstream of yeast promoters [36,37]. The multisubunit Trf4-Air2-Mtr4 polyadenylation (TRAMP) complex then targets the NNS-terminated fragments to the nuclear exosome [38–40]. In contrast, promoter directionality in mammalian cells is achieved by an enrichment in canonical PASs in the upstream antisense direction and depletion of U1 snRNP binding sites [41,42]. At least some PROMPTs are terminated by the combined actions of the canonical cleavage and polyadenylation machinery, the cap-binding complex and its associated protein ARS2 [41–44]. After termination, the trimeric NEXT complex targets PROMPTS for decay by the exosome [24,43,45,46]. In addition, bidirectional transcripts can be terminated and degraded by co-transcriptional decapping and 5´→3´ decay by Xrn2 [47]. Three studies, including this one, report that specific PROMPTs are degraded in a PABPN1-dependent fashion [4,48]. Visual inspection of the sequence traces of previously published NEXT-sensitive PROMPTS is ambiguous regarding their susceptibility to PPD (S6A Fig) [4], suggesting that specific PROMPTs are targeted by distinct nuclear decay pathways. Further experimentation is required to determine whether the PPD, Xrn2 and NEXT pathways target independent subsets of upstream antisense transcripts, or are largely redundant pathways for bidirectional transcript degradation.
U1 snRNP is a core component of the spliceosome that recognizes 5´ splice sites, but it also suppresses the use of premature PASs [49,50]. This latter function contributes to promoter directionality in that U1 snRNP binding sites are depleted in upstream antisense regions and overrepresented in coding regions [41,42]. As a result, antisense transcription normally produces shorter, unspliced transcripts, whereas coding genes produce longer spliced pre-mRNAs. Interestingly, five of our high-confidence PPD substrates classified as mRNAs had increased sequence coverage at the 5´ end of the genes (APOLD1, MTHFD2L, AGBL3, TEX22, and FAM120C)(S6B Fig). We speculate that these transcripts result from a failure of U1 snRNP to protect from premature PAS usage. The resulting RNAs resemble promoter antisense RNAs and are therefore subject to degradation by PPD. This speculation is supported by a recent global analysis demonstrating that PABPN1 depletion increased the levels of similar sense proximal RNAs [48].
We previously demonstrated that an intronless β-globin reporter is rapidly degraded by PPD, but insertion of a single intron into that reporter is sufficient to protect the resulting mRNA from PPD [5]. Consistent with this idea, 174/353 (49%) of the high-confidence RNAs identified are single-exon RNAs (S2 Fig). The simplest explanation for this observation is that splicing promotes the formation of an export-competent mRNP leading to export and escape from PPD [27]. However, a single splicing event is not always sufficient to promote escape from PPD. By definition, all PPD-targeted ncSNHGs are spliced at least once (Fig 2) and only 5/74 PPD-sensitive mRNAs are single exon genes (S2 Table). Because ncSNHGs targeted by PPD had higher nuclear enrichment (Fig 2), we conclude that PPD susceptibility stems from nuclear retention of the spliced transcript. This could be due to nuclear retention signals in the exons or due to variations in recruitment of splicing-dependent export factors.
We also found that RI-RNAs are subject to PPD (Figs 3 and 4). Recent studies point out the importance of intron retention in mammalian cells [28–31]. The efficiency of splicing of these retained (“detained” in [31]) introns can be modulated by developmental or environmental cues supporting an essential role for these RNAs in posttranscriptional gene regulation. These previous studies showed that a subset of RI-RNAs is degraded by NMD while others are retained in the nucleus and degraded by a previously unknown nuclear RNA decay pathway. Our data now show that that nuclear retained RI-RNAs are subject to PPD. Thus, there is a parallel between RI-RNAs and ncSNHGs in that both produce spliced RNAs that are either exported and subject to NMD or retained in the nucleus and subject to PPD. Importantly, the RI-RNAs are not strongly upregulated by PPD inactivation. We had to increase the lengths of time for PAP knockdown to observe increases in ARGLU1 and OGT and cordycepin treatment had no effect on their abundance (Fig 3D). This may be due to the biology of the RI-RNAs. For example, if they serve as precursors to pre-mRNAs as proposed [31,51], the half-lives of these RNAs may be longer than the nonfunctional ncSNHGs or PROMPTs. Thus cells may regulate PPD to control the accumulation of RI-RNAs. Given the widespread use of intron retention in mammals, PPD regulation may have important consequences for gene expression. Interestingly, PABPN1 was recently shown to autoregulate its mRNA levels by intron retention [52].
Testing the half-lives of the nuclear RNAs identified herein is complicated by the unusual behavior of nuclear RNAs upon general transcription inhibition (Figs 4 and 5). We do not understand how transcription inhibition leads to the accumulation of hyperadenylated nuclear RNAs, but the simplest explanation for this striking phenomenology is that PABPN1-dependent hyperadenylation occurs, but is uncoupled from the decay step of PPD. We stress that this is not the result of a specific transcription inhibitor or concentration as four different transcription inhibitors, which utilize at least three distinct mechanisms of transcription inhibition yielded a similar result (S4 Fig). Interestingly, we observed that a portion of completely spliced MAT2A and ARGLU1 RNAs was hyperadenylated after ActD treatment (Fig 4A and S3D Fig). Because there is little fully spliced RNA in the nuclear fraction prior to ActD treatment (S3D Fig), it seems likely that the retained intron is posttranscriptionally spliced. However, this splicing is not sufficient to release the RNA for export, at least in the presence of ActD. Perhaps transcription inhibitors indirectly produce a general block in mRNA export. Alternatively, the RI-RNAs may be fated for the discard pathway, so they are subject to nuclear retention and PPD even after splicing. Another explanation is that the RI-RNAs are normally degraded, but ActD-induced stabilization (Figs 4 and 5) allows sufficient time for the RNAs to be fully spliced. Given the prevalence of intron retention in mammals, the interrelationships between PPD, splicing, and transcription warrant deeper investigation.
In yeast, the TRAMP complex component Trf4, a noncanonical poly(A) polymerase, marks nuclear RNAs for decay by the exosome. While Trf4 is essential for decay, its polyadenylation activity is not necessary [53–55]. In contrast, our studies are consistent with the conclusion that hyperadenylation of PPD targets is linked to their decay. Transcripts that are upregulated following PABPN1-depletion are also increased following depletion of PAP or expression of a polyadenylation defective PABPN1 allele (Figs 1 and 2). Three lines of evidence suggest that distributive rather than processive polyadenylation is the primary driver of decay. First, CPSF is necessary for processive polyadenylation in vitro so the CPSF-independent PANΔENE-A35 is unlikely to undergo processive polyadenylation. Nevertheless, PANΔENE-A35 was degraded by PPD (Fig 6), suggesting that processive polyadenylation is dispensable for decay. Second, a distributive process should be more sensitive to relative concentrations of PPD factors in the cell because of the requirement for re-binding after dissociation. Indeed, our siPAP knockdowns decrease PAP levels such that hyperadenylation is affected, but there appears to be little effect on the initial polyadenylation reaction [5]. Third, upon transcription inhibition, poly(A) tails gradually increased in length as a group over several hours, consistent with PAP disassociating and re-associating with transcripts stochastically (Figs 4 and 5). In contrast, processive polyadenylation that forms the initial poly(A) tail occurs rapidly in vitro and in cells with ~200–250 nucleotides being added in less than one minute [56,57]. Interestingly, even though PABPN1 stimulates CPSF-independent distributive hyperadenylation, hyperadenylation was not sufficient to rescue PPD sensitivity in the absence of PABPN1 (Fig 6H). Thus, PABPN1 likely plays multiple roles in PPD. In fact, Pab2 and PABPN1 co-immunoprecipitate with the exosome [4,58], suggesting PABPN1 may directly recruit the exosome. Alternatively, PABPN1 may compete with poly(A) binding proteins that stabilize RNAs. Thus, upon PABPN1 depletion, these proteins preferentially associate to increase RNA half-lives [59,60].
In summary, our data show that PPD modulates the levels of functional lncRNAs and mRNAs as well as presumably nonfunctional PROMPTs and the spliced byproducts of snoRNA and pri-miRNA processing. We conclude that PPD is an important nuclear RNA decay pathway that lies at the interface of transcription, splicing, 3´-end formation and mRNA export.
RNA-seq and sequencing was performed at the McDermott Center Next Generation Sequencing Core and Bioinformatics Core. Libraries were prepared using the TruSeq Stranded mRNA preparation kit and run on an Illumina HiSeq 2500 (paired-end 100 bp reads). The reads were mapped, aligned and assembled using TopHat2 and Cufflinks2.2 [61,62]. Transcriptome assembly was guided by iGenomes (hg19, UCSC build) and GENCODE (release 19) annotation files. Differential gene expression was analyzed by Cuffdiff using the iGenomes annotations and EdgeR was employed to determine differential expression of the 13,853 known and novel lncRNAs in the GENCODE annotation [63]. Integrative genomics viewer (IGV) was used to visualize sequence coverage and generate figures [64]. DEGs were identified from the Cuffdiff output by removing those transcripts with an FPKM of <1 in the treatment sample and the remaining transcripts with p-value <0.05 and a false discovery rate (FDR) less than 5% were defined as DEGs (S1 Table). DEGs in the EdgeR data were defined as those with log(counts per million) >3.5 and an FDR <5% (S3 Table). Heat maps were generated using the GENE-E software (http://www.broadinstitute.org/cancer/software/GENE-E/index.html).
We categorized each of the 353 high-confidence upregulated DEGs by visual assessment of IGV traces (S2 Table). Any DEG found upstream and antisense to an annotated gene was defined as a PROMPT. Antisense orientation was confirmed in IGV using strand-specific bigWig files generated by HOMER [65]. AS transcripts, on the other hand, were those with considerable overlap within an annotated gene. Pri-miRNA and ncSNHG transcripts were inferred by the presence of an overlapping miRNA/snoRNA or corresponded to annotated genes. We assigned the category lncRNA to any transcript that was from an annotated lncRNA gene or from an unannotated genomic region that did not fall into any of the other categories.
All plasmids were constructed using standard molecular biology techniques. The details of the construction are given in the Expanded View. Transfections and TetRP pulse-chase assays were performed as previously described [5].
Detection of newly made bulk poly(A) tails was performed essentially as previously described [66,67].
Bulk poly(A) tails were detected on 1.8% agarose-formaldehyde gel, and detected with a dT40 probe end-labeled with T4 polynucleotide kinase. Northern blots for specific transcripts were performed using standard techniques with RNA probes. Stripping and re-probing of the membranes were performed as previously described [5]. The RNA probes were generated from PCR products with a T7 RNA polymerase promoter; primers are listed in S5 Table. For some northern blots, 35–80 mg of total RNA were selected on oligo(dT) cellulose to enrich for polyadenylated RNAs prior to gel electrophoresis. In addition, we degraded residual rRNA after oligo(dT)-cellulose selection with Terminator exonuclease (EpiCentre).
To collect cytoplasmic RNA, cells were resuspended in Buffer I (0.32 M sucrose, 3mM CaCl2, 2 mM MgCl2, 0.1 mM EDTA, 10 mM Tris-HCl (pH 8.0), 1 mM DTT, 0.04 U/ml RNase Inhibitor, 0.5% Triton X-100), incubated on ice for 5 min, centrifuged at 500 x g for 3 min at 4°. RNA in the supernatant was extracted using TriReagent (Molecular Research Center) followed by an additional phenol-chloroform extraction. The pellet was then washed in Buffer I with 150 mM NaCl and once again centrifuged at 500 x g for 3 min at 4°. The resulting supernatant was discarded. The RNA from the remaining pellet was then extracted in TriReagent. We note that in cases in which we analyzed RNA from the wash step, we observed both long and short poly(A) tails; whether this is due to cross contamination of cellular compartments and/or is due to a distinct biological fraction is unclear. This fractionation procedure results in the loss of Triton X100-soluble nuclear material, but it enriches for chromatin and nuclear speckle-associated RNAs [18–20].
RNA was harvested using TriReagent according to the manufacturer’s protocol. Following extraction, RNA was treated with RQ1 DNase (Promega). Random hexamers were used to prime cDNA synthesis with MuLV reverse transcriptase (NEB). Real-time reactions used iTaq Universal SYBR Green Supermix (Biorad).
Biotinylation reactions were carried out in a 200μL mixture consisting of 40μg RNA, 20mM NaOAc (pH 5.2), 1mM EDTA, 0.1% SDS, 0.2mg/mL Biotin-HPDP (Pierce), and 50% N,N-dimethylformamide (DMF) for 3 hours at 25°C. Unconjugated biotin-HPDP was removed with three chloroform extractions. After extraction of the aqueous phase, 20μL (10% v/v) of 10M NH4OAc was added to each tube, and the RNA was precipitated in 70% ethanol.
Streptavidin selection was carried out using magnetic Streptavidin T1 beads (Invitrogen). Prior to use, the 20 μl bead slurry was washed three times in a 0.1X MPG solution (1X MPG was 1M NaCl, 10mM EDTA, and 100mM Tris 7.5) supplemented with 0.1% igepal. After the final wash, the beads were resuspended in a 1mL solution consisting of 0.1X MPG supplemented with 0.1% igepal, 0.1μg/μL poly(A) (Sigma-Aldrich), 0.1μg/μL ssDNA, 0.1 μg/μL cRNA, and 0.1% SDS, and blocked for one hour. RNA was precipitated, resuspended in a volume of 63μL water, and denatured at 65°C for 5 minutes. Next, RNA was incubated together with beads for one hour while nutating at room temperature. Beads were sequentially washed in: 0.1X MPG, 0.1X MPG at 55°C, 0.1X MPG, 1X MPG, 1X MPG, 0.1X MPG, 1X MPG without NaCl, 0.1X MPG. With the exception of the 55°C wash, each solution included 0.1% igepal. Biotinylated RNAs were eluted twice for 5 minutes each in a 200μL solution of 0.1X MPG containing 5% β-mercaptoethanol. The first elution step was at 25°C and the second was at 65°C. The two eluted fractions were combined and extracted with PCA once and chloroform twice. After extraction, 40μL of 10M NH4OAc was added to each tube, and the RNA was precipitated in 70% ethanol.
Nuclear run-ons were performed essentially as previously described [67]. The details are provided in the Expanded View.
Following knockdown, cells were treated with 2μM of 4SU for one hour. Afterwards, cells were washed twice with phosphate buffered saline (PBS) containing calcium and magnesium (Sigma-Aldrich), and grown in media lacking 4SU for an additional hour. After the one-hour washout step, we collected 0, 30, 60, and 120 min time points. 40μg of RNA was used as input for a biotinylation and streptavidin selection as described above. Selected RNA was reverse transcribed prior to qRT-PCR analysis. β-actin was used as a loading control for qPCR analysis.
The cells were given fresh media 4.5 hours prior to the 4SU treatment, which was necessary for consistent results. Cells were treated with 100 μM of 4SU for five minutes and incorporation was quickly stopped by addition of TriReagent. Sixty micrograms of total RNA was used for biotinylation and streptavidin selection as described above except one additional 1X MPG and one additional no salt wash was performed and both elution steps were done at room temperature.
|
10.1371/journal.ppat.1005209 | Carcinogenic Parasite Secretes Growth Factor That Accelerates Wound Healing and Potentially Promotes Neoplasia | Infection with the human liver fluke Opisthorchis viverrini induces cancer of the bile ducts, cholangiocarcinoma (CCA). Injury from feeding activities of this parasite within the human biliary tree causes extensive lesions, wounds that undergo protracted cycles of healing, and re-injury over years of chronic infection. We show that O. viverrini secreted proteins accelerated wound resolution in human cholangiocytes, an outcome that was compromised following silencing of expression of the fluke-derived gene encoding the granulin-like growth factor, Ov-GRN-1. Recombinant Ov-GRN-1 induced angiogenesis and accelerated mouse wound healing. Ov-GRN-1 was internalized by human cholangiocytes and induced gene and protein expression changes associated with wound healing and cancer pathways. Given the notable but seemingly paradoxical properties of liver fluke granulin in promoting not only wound healing but also a carcinogenic microenvironment, Ov-GRN-1 likely holds marked potential as a therapeutic wound-healing agent and as a vaccine against an infection-induced cancer of major public health significance in the developing world.
| The oriental liver fluke Opisthorchis viverrini infects millions of people in SE-Asia and kills 26,000 people each year due to parasite-induced liver cancer. The mechanisms by which the parasite causes cancer are complex, but a role for excessive wound healing in response to feeding parasites in the bile ducts has been proposed. We show that a growth factor (granulin) secreted by the worm gets into bile duct cells and drives wound healing and blood vessel growth. We delve into this “supercharged” wound healing process and uncover a range of signaling molecules that initiate healing, but when dysregulated, can result in a deadly liver cancer. On the upside, this liver fluke growth factor is now a candidate drug for the development of novel wound healing therapeutics to treat chronic wounds, such as diabetic ulcers. Understanding this process is another step on the road to developing a vaccine to reduce both parasite burdens and the incidence of the most prevalent and fatal cancer in Thailand and surrounding countries.
| Approximately 10 million people in Thailand and Laos are infected with the South East Asian liver fluke Opisthorchis viverrini [1,2]. Infection with O. viverrini, a one-centimeter long flatworm that inhabits the bile ducts, is strongly associated with the induction of cholangiocarcinoma (CCA), cancer of the bile ducts [3]. The World Health Organization’s International Agency for Research on Cancer classifies infection with O. viverrini as a ‘group 1 carcinogen [1,3,4,5]. In Thailand and neighboring countries, cyprinid fish that are intermediate hosts for O. viverrini are eaten raw as a staple of the diet [1,2]. Infected individuals in endemic areas suffer the world’s highest incidence of CCA, 65 times that experienced in non-endemic regions, and accounting for up to 81% of liver cancers in this region [3,4]. CCA is a primary cancer originating in cholangiocytes, the epithelial cells that line the biliary tree. It has long latency, is invasive, metastasizes, is relatively non-responsive to anti-tumor agents and has a dismal prognosis.
How opisthorchiasis induces cholangiocarcinogenesis is likely multi-factorial, involving immunopathogenesis, increased consumption of dietary carcinogens, and the secretion of parasite proteins mitogenic for cholangiocytes [2]. We described a liver fluke-derived homologue of the human growth factor granulin, termed Ov-GRN-1, from the excretory/secretory (ES) products of O. viverrini [2,6,7]. Ov-GRN-1 binds to cholangiocytes in experimentally infected hamsters and stimulates proliferation of fibroblasts and CCA cell lines. Here we sought to determine whether Ov-GRN-1 possesses wound healing capacity and might therefore function to repair the chronic damage it causes in the bile ducts during feeding activity and the ensuing chronic inflammation. Moreover, given the physiologic and genetic similarities between chronically healing wounds and cancer [8], we sought to address whether Ov-GRN-1 promotes cellular changes that are conducive to the establishment of a tumorigenic environment.
Using fluorescence microscopy we report that recombinant Ov-GRN-1 (rOv-GRN-1) labeled with Alexa Fluor 488 (AF) was putatively internalized by ~75% of cells from an immortalized human cholangiocyte cell line, H69 (Fig 1A and 1B, S1A–S1D Fig). Cholangiocytes co-cultured with rOv-GRN-1-AF exhibited significantly higher (P < 0.001) per cell fluorescence intensity (6.4-fold, or 15.3-fold RFU/mole) than cholangiocytes co-cultured with a control recombinant protein (thioredoxin-AF, rTRX-AF) that had been expressed and purified under identical conditions (S1E Fig). Using 3D-structured illumination microscopy, rOv-GRN-1-AF was detected between the apical and basal actin filaments of cells in monolayer, confirming internalization in cholangiocytes of the liver fluke granulin (Fig 1C and 1D, S1 Movie). The precursor of human granulin is expressed as a seven-domain granulin unit, known as progranulin (PGRN), and initiates context-dependent autocrine and paracrine signaling cascades [9,10,11,12]. PGRN is internalized by cells and targeted to a specific organelle, commonly lysosomes, when bound to co-factors such as sortilin or CpG nucleic acid motifs [9,10,11,13]. Attempts to identify the sub-cellular location of rOv-GRN-1 after internalization by cholangiocytes using a range of organelle-specific markers suggested a cytosolic location, as specific co-localization to organelles was not apparent (S2 Fig). The lack of involvement of an organelle suggested direct cell entry followed by interactions with signaling cascades, rather than the more conventional growth factor receptor-based signal initiation. While unusual, direct cell entry and interaction with signaling molecules is known for small growth factors with alkaline tails, such as basic FGF [14,15]; the C-terminus of Ov-GRN-1 is highly basic [7] with a predicted pI of 12, characteristics that also support this mode of cell entry.
Previously, we silenced expression of the Ov-grn-1 gene using RNA interference (RNAi) that reduced cell proliferation of cholangiocytes co-cultured with the liver flukes [16]. To address the role of Ov-GRN-1 in wound repair we silenced expression of Ov-grn-1 using RNAi and assessed the ability of ES products from dsRNA-treated flukes to accelerate cell proliferation and wound repair. Levels of mRNA encoding Ov-GRN-1 were depleted by 97% in worms transduced with dsRNA specific for Ov-grn-1 but not affected by control dsRNA specific for luciferease (luc) (S3 Fig). ES products were collected from culture supernatants of dsRNA-treated flukes and effects of the ES on proliferation of cholangiocytes assessed. ES products collected on days 1, 5 and 7 from Ov-grn-1dsRNA-treated flukes reduced cell proliferation by ~48% (P < 0.01; F(DFn, DFd) = 24.27 (3,7)) compared to ES from luc-treated flukes (Fig 2A and S3 Fig). To ensure that Ov-grn-1-dsRNA treatment did not have a major impact on the ES composition of the flukes, we compared ES profiles from Ov-grn-1- and luc-dsRNA treated flukes by SDS-PAGE, and did not detect obvious differences in protein yield or composition (S4 Fig).
At the outset, we assessed the role of Ov-GRN-1 in wound repair using in vitro scratch assays given that the procedure is a facile surrogate of cell migration and wound closure [17]. dsRNA-treated flukes were co-cultured in Transwell plates such that they were separated from the underlying cells by a porous membrane, but ES products could traverse the inner membrane of the chamber. Firstly, we showed that ES products from luc dsRNA-treated flukes substantially accelerated wound healing compared to both cholangiocytes and CCA cell lines that were not co-cultured with flukes (Fig 2C and 2D). Secondly, and pivotal to this study, significantly less wound healing/closure was induced by Ov-grn-1 dsRNA-treated flukes in both cholangiocytes over 36 hours (P < 0.01–0.0001; Fig 2B and 2C) and CCA cells over 18 hours (P < 0.001–0.0001; Fig 2C) than with control luc dsRNA-treated flukes. Fewer cells crossed the margin of the wound of the scratched monolayers cultured with ES products from Ov-grn-1 dsRNA-treated flukes at the early time points (6–12 h, Fig 2B–2D), suggesting the involvement of cell migration in scratch closure rather than closure due simply to cell proliferation [17].
To confirm the role of Ov-GRN-1 in in vitro wound healing 20 nM rOv-GRN-1 was shown to be sufficient to significantly accelerate healing of a cholangiocyte monolayer compared to cells exposed to control protein (rTRX) (F(DFn, DFd) = 16.32(2,33); P < 0.01) (Fig 2E).
To determine whether rOv-GRN-1 could accelerate wound repair in vivo, sub-cutaneous deep lesions were surgically inflicted between the ears on laboratory mice, treatment applied and the injury covered with spray plaster, after which the rate of wound healing was quantified at intervals of 24 hours for four days [18] (Fig 3A). This method is considered to be superior to the conventional abdomen wound protocol when investigating growth factors, since it quantifies healing primarily from epithelial re-growth rather than skin contraction [18,19]. Daily application of 56 pMoles of rOv-GRN-1 significantly accelerated wound healing within 2–4 days compared wound closure in response to application of a control protein (rTRX) (F(DFn, DFd) = 32.08 (2,16); P < 0.01–0.001) or PBS (Fig 3B).
Angiogenesis is an integral aspect of wound healing, is essential for the vascularization of new tissue, and is a cardinal hallmark of carcinogenesis. The chorioallantoic membrane (CAM) assay is a commonly accepted in vivo model of vertebrate angiogenesis [7,20,21]; moreover, the ancestral lineage of the granulin family of growth factors [22] made us conclude that the CAM assay was a suitable mean by which to assess angiogenic properties of Ov-GRN-1. Quail eggs were implanted with rOv-GRN-1- or PBS-soaked membranes. Membranes with two picomoles (P < 0.05) or 20 picomoles (P < 0.0001) of rOv-GRN-1 induced angiogenesis (F(DFn, DFd) = 108.4(2,9)) (Fig 3C) in the embryo developing within the egg.
We employed isobaric tags for relative and absolute quantitation (iTRAQ) of changes in expression of cholangiocyte proteins induced by rOv-GRN-1. Using the Scaffold program, we reliably validated 215 proteins in cholangiocytes identified by Mascot compared to cells at baseline and at subsequent intervals (S1 Table). rOv-GRN-1 induced >50% change in detectable expression levels (P < 0.05) of 70 cholangiocyte proteins at ≥1 time point compared to control cells (Fig 4A and S2 Table). During co-culture of up to eight hours there was substantial up-regulation of protein expression, after which moderation or down regulation of the proteins ensued beyond 16 hours from the start of the analysis (Fig 4A). Three KEGG pathways with 12 significantly regulated proteins each—the spliceosome, endoplasmic reticulum protein processing and metabolic pathways (Fig 4B) were revealed by protein ontology analysis in the cholangiocytes cultured with the parasite granulin. Euclidean distance clustering revealed the internal patterning of temporal translational changes (Fig 4A), where group X proteins underwent a short-term up-regulation (0.5–8 h) followed by a lessening of expression. Group Y proteins also underwent a short-term up-regulation followed by a substantial down-regulation. Group Z proteins were distinct due to their high and rapid short-term up-regulation. Notably, six of the 13 group Z proteins are associated with the spliceosome (Fig 4A and 4B). The dysregulated proteins were subjected to a network analysis (Fig 4C). When the top-25 most highly up-regulated proteins were considered, proteins involved in the spliceosome pathway were most highly represented (Fig 4D), and included the top three (HNRNPA3, THOC4 and NONO) and nine of the top 25 most highly up-regulated proteins.
Mass spectrometry is constrained in its ability to characterize changes in low abundance proteins such as growth factors and cytokines. We therefore assessed the changes in cholangiocyte gene expression after one and 24 hours of co-culture with rOv-GRN-1 using gene arrays targeting epithelial to mesenchymal transition (EMT), oncogenesis, wound healing and Toll-like receptor signaling (S3 Table). Thirty genes underwent an Ov-GRN-1-induced change (P < 0.05) in regulation (Fig 4E and S4 Table), including four which exhibited >50% change in expression levels. Three of the four upregulated genes encoded proteins from the C-X-C ligand chemokine family of cytokines: cxcl1, cxcl2 and cxcl8 (also known asinterleukin-8); the fourth gene encoded for serine/threonine kinase 11 (stk11), also known as liver kinase B1. Another member of the cxcl family, cxcl5, was significantly upregulated, but fell below the 50% cutoff (43%).
We report for the first time the secretion of a growth factor from a metazoan pathogen that promotes wound healing of mammalian host tissue in vivo. The implications of the findings are multi-fold and significant. Firstly, the instrumental role described here for Ov-GRN-1 in orchestrating wound repair implies that this protein represents an attractive target for the development of a vaccine that thwarts regulation of the microenvironment within the biliary tract parasitized by the liver fluke. Indeed we previously showed that antibodies to rOv-GRN-1 block proliferation of cholangiocytes [7], which further bolsters the proposition of a vaccine with both anti-infection and anti-cancer properties. One potential caveat of a vaccine that blocks wound repair however is the consequences of an aggressive inflammatory response in the absence of wound resolution, including uncontrolled sepsis or other complications, so appropriate consideration is warranted. Second, the findings highlight the potential therapeutic application of Ov-GRN-1 as a novel biologic for treating both acute and chronic wounds, such as recalcitrant ulcers on the extremities of diabetic patients [23].
Mammalian granulins play diverse roles continuously during development from the embryo into adult life, including key roles in tissue remodeling and inflammation [22]. Mutations in the human granulin gene result in a spectrum of conditions, including neurodegenerative disorders [24] and malignant growth and metastasis [25]. Indeed, granulin has a central role in carcinogenesis of a range of malignancies [22]; pertinent to our findings, granulin is over-expressed in hepatocellular carcinoma (HCC) [26] and renders HCC cells resistant to Natural Killer cell-mediated cytotoxicity by modulating expression of MHC-associated genes [27]. By contrast, granulins of pathogens have received little attention. We detected O. viverrini granulin (Ov-GRN-1) in the ES products of adult flukes and provided the first evidence of a parasite growth factor that drove proliferation of host cells [6,7]. The recent report of the O. viverrini genome revealed additional members of the granulin family–a single granulin domain protein (Ov-GRN-2) and a pro-granulin (PGRN) containing eight granulin subunits [28]. Products of either of these genes were not evident within the ES proteome [6] and their role in the host-parasite relationship is unclear.
The mechanisms by which vertebrate or liver fluke granulins drive cell proliferation and wound repair are poorly understood. Vertebrate PGRN contains seven individual granulin subunits that are post-translationally processed. Mouse PGRN but not the individual subunits of granulin binds to TNF receptors (TNFR), and antagonizes TNF signaling [29]. Dissimilar to PGRN, Ov-GRN-1 consists of a signal peptide and a single granulin motif [7]. Although the ability of rOv-GRN-1 to bind to TNFR has not been investigated, probing a microarray of the human proteome microarray [30] with labeled rOv-GRN-1 failed to reveal binding to any isoforms of TNFR, or indeed to any other obvious cell surface receptors, on the array.
Cholangiocyte proteins involved in the spliceosome pathway were significantly regulated after exposure to rOv-GRN-1 in vitro. The majority of intron removal from pre-RNA molecules is catalysed by the spliceosome, a large ribonucleo-protein complex that consists of five small nuclear ribonucleo-protein particles (snRNP, U1-6) and >150 other proteins [31]. One critical component of the wound healing process that is heavily regulated by RNA binding and splicing is the epithelial to mesenchymal transition (EMT), which increases the migratory and invasive properties of cells and thereby promotes wound closure [32,33,34]. However, cancerization also is an untoward consequence of EMT, and aggressive tumours often display dysregulated expression of spliceosome proteins [31,35].
Liver fluke granulin stimulated expression of genes encoding the chemokines CXCL1, 2, 5 and 8 (also known as IL-8). These chemokines signal through the receptor CXCR2 [36,37] by internal transactivation of the epidermal growth receptor (EGFR) and EGFR signaling through the mitogen activated protein kinase (MAPK) pathways [38]. Chemokines play central roles in wound repair, angiogenesis and recruitment of immune cells [36,39,40]. Inhibitors of MAPK signaling block rOv-GRN-1-induced cell proliferation [7], and the increased expression of cxcl genes induced in cholangiocytes by rOv-GRN-1 may underlie this observation. In addition, expression levels of transcripts encoding several kinases including stk11 and irak1 were markedly stimulated by the parasite granulin. Both STK11 (liver kinase B1) and IRAK1 (Interleukin-1 receptor associated kinase 1) control signaling in inflammatory pathways and regulate chemotaxis in diverse processes including wound healing [41,42]. Moreover, somatic mutations in stk11 [43] and irak1 [44] associated with malignancy. Upregulation of these kinases during proliferation of cholangiocytes within the liver fluke-parasitized biliary tree may, therefore, increase the likelihood of these mutations.
Topical application of picomoles of rOv-GRN-1 significantly accelerated repair of wounds in the skin of mice. Although liver fluke granulin triggers changes in the cellular proteome that establish a pre-tumorigenic environment, short-term therapy would reduce the likelihood of inducing cancer in patients. Whereas advances in understanding the impaired angiogenesis in non-healing wounds have been reported, few effective agents that promote or expedite wound healing and closure are yet available [45]. The ability of rOv-GRN-1 to accelerate wound healing in mice and promote angiogenesis in vivo revealed that this growth factor holds noteworthy promise for a new category of medicines for non-healing wounds and related indications.
Other growth factors are of interest for their therapeutic properties, notably human PGRN due to its ability to bind to TNFR. Indeed, recombinant human PGRN inhibits TNF-activated signaling and protected against inflammation in rodent models of arthritis [29]. PGRN further exerts its anti-inflammatory influence by inducing naïve T cells to transform into FOXP3-expressing regulatory T cells (Tregs) [46], a lymphocyte type that is underrepresented in inflammatory diseases but the presence of which is a hallmark of helminth infections [47,48]. Indeed we speculate now that Ov-GRN-1 may be the major inducer of Tregs during opisthorchiasis, but this hypothesis clearly warrants testing.
In conclusion, we have shown using gene silencing and recombinant protein technologies that the most carcinogenic of parasitic helminths, the liver fluke O. viverrini, secretes a growth factor which in isolation is sufficient to repair wounds both in monolayers of cultured human cholangiocytes and in the skin of mice. While our mouse cutaneous wound healing studies are informative and shed light on the potential therapeutic application of Ov-GRN-1 for chronic wounds, they do not directly address the role of the protein in host-fluke interactions in the biliary tree. With recent advances in genome editing using CRISPR-Cas9, we will soon be well placed to knock out the Ov-grn-1 gene, facilitating in vivo studies that will specifically address the role of the protein in healing parasite-induced wounds in the bile ducts. Ov-GRN-1 therefore is a worthy candidate at which to target novel interventions—drugs and/or vaccines with both anti-helminth and anti-cancer activity. Moreover, Ov-GRN-1 offers potential as a novel biologic for treating acute and chronic wounds where normal tissue repair mechanisms are insufficient. Now more than ever, there is acute need for new therapeutics to combat the epidemic of inflammatory diseases, particularly diabetes and associated chronic ulceration. The therapeutic efficacy of parasitic helminths and their secreted products in treating inflammatory diseases is clear-cut [49]. The present findings indicate that parasite growth factors, which by their very nature have evolved to repair damaged tissues within their hosts, offer great promise as a novel therapeutic modality informed by millennia of host-parasite coevolution.
Excretory/secretory (ES) and somatic proteins were harvested from adult O. viverrini grown in laboratory hamsters as described [7,50]. Briefly, O. viverrini metacercariae harvested from naturally infected cyprinoid fish were used to infect hamsters (Mesocricetus auratus) by stomach intubation. Hamsters were euthanized three months after infection, when adult O. viverrini flukes were removed from the biliary tract. The flukes were washed and cultured in modified RPMI-1640 (Life Technologies) containing penicillin and streptomycin at 37°C/5% CO2 for three days. Culture supernatant was retained as ES products of the parasites, and stored at -80°C [50].
Ov-grn-1 pET41a or thioredoxin (trx) cDNAs contained within the pET32a (Novagen) plasmid were transfected into BL21 Escherichia coli cells (Life Technologies) and used to create recombinant protein with auto-induction as previously described [7,51]. Briefly, ZYM-5052 culture media was supplemented with 100 μM Fe(III)Cl3 and 100 μg L-1 kanamycin to produce recombinant protein (rOv-GRN-1) or 50 μg L-1 ampicillin to produce rTRX [51]. Two hundred ml of inoculated media in a 1L baffled Erlenmeyer flask was incubated overnight at 37°C with 300 rpm rotation to induce expression with auto-induction.
Purification of rOv-GRN-1 was achieved using an AKTA10 purification system at 4°C (GE Healthcare) [52]. The BL21 E. coli pellet was lysed with 3 freeze/thaw cycles followed by sonication (Q4000 sonicator, Qsonix) on ice. Twenty g of the resulting pellet was solubilized in 400 ml urea-containing nickel binding buffer (8 M urea/300 mM NaCl/50 mM imidazole/50 mM sodium phosphate pH 8 [Sigma]) at 4°C for 24 h with slow agitation. After filtration through 0.22 μM membranes, supernatants were incubated in nickel chelate resin on 2× 5 ml Histrap IMAC columns (GE Healthcare). The columns were washed in increasing concentration of imidazole (two column volumes [CV] at 50 mM/5 CV at 100 mM) after which bound material was eluted in 500 mM imidazole in binding buffer. The control rTRX protein was expressed and affinity purified similarly, but under native conditions (without chaotropes), as described [52].
Refolding of urea-denatured rOv-GRN-1 was performed with 28 mL of G10 Sephadex (GE) resin on a XK16/20 column (GE) [52]. A 120 ml Superdex 30 XK16/60 column (GE) was used to fractionate three ml of refolded rOv-GRN-1 into 150 mM NaCl, 50 mM sodium phosphate, pH 6, at a flow rate of 1 ml min-1. Fractions containing rOv-GRN-1 monomer eluting at a size equivalent of ~1 kDa were pooled. Protein concentration was established using a combination of the Bradford assay (Bio-Rad) and absorbance at 280 nm.
The cholangiocyte cell line H69 is a SV40-transformed bile duct epithelial cell line derived from a non-cancerous human liver [53] and was obtained in 2010 from Dr. Gregory J. Gores, Mayo Clinic, Rochester, Minnesota. H69 cells and cells of the human cholangiocarcinoma (CCA) cell line KKU-M214 were maintained in T75 cm2 vented flasks (Corning) as monolayers as described [52,53,54,55] with minor modifications. KKU-M214 cells were maintained with regular splits using 0.25% trypsin (Life Technologies) every 2–5 days in complete media (RPMI with 10% fetal calf serum [FCS] and 1× antibiotic/antimycotic) at 37°C under 5% CO2. Cell proliferation assays were performed with low nutrient media containing 0.5% FCS. H69 cells were maintained under similar conditions with growth factor supplemented media [54] (DMEM/F12 with high glucose, 10% FCS, 1×antibiotic/antimycotic, 25 μg ml-1 adenine, 5 μg ml-1 insulin, 1 μg ml-1 epinephrine, 8.3 μg ml-1 holo-transferrin, 0.62 μg ml-1 hydrocortisone, 13.6 ng ml-1 T3 and 10 ng ml-1 EGF–Life Technologies). Low nutrient media for H69 cells was 5% complete media, i.e. 0.5% FCS and 5% of the growth factor concentrations for complete media. The identities as human-derived of both cell lines were confirmed with single tandem repeat (STR) analysis (15/15 positive loci across two alleles) and mycoplasma free at the DNA diagnostics centre (U.S.A.), accredited/certified by CAP, ISO/IEC 17025:2005 through ACLASS.
Cells were seeded at 1500 cells per well in 200 μl of complete media as described above in E-plates (ACEA Biosciences) and grown overnight while monitored with an xCELLigence SP system (ACEA Biosciences) which monitors cellular events in real time by measuring electrical impedance across interdigitated gold micro-electrodes integrated on the bottom of tissue culture plates [56]. Cells were washed three times with PBS and replaced with 180 μl of low nutrient media as described above and incubated for a minimum of 6 h before further treatments. Treatments were prepared at 10× concentrations and added to each well in a total volume of 20 μl. The xCELLigence system recorded cell index readings hourly for 5–6 days after treatment. Cell index readings were normalized before treatment and cell proliferation ratios were determined from biological triplicates and represent the relative numbers of cells compared to control cells.
H69 cells in complete media (see above) that were grown to confluence in 6 well plates (Falcon) were wounded by scratching the cell monolayer with a disposable 200 μl pipette tip, as described [17]. The wound in the monolayer was photographed regularly and closure was assessed using ImageJ software (National Institute of Health, U.S.A.). Wound widths over time were plotted and compared to controls with matched 2-way ANOVA and Dunnett’s correction for multiple comparisons. For cell scratch assays performed in co-culture with liver flukes, wounded monolayers of cells in 6 well plates were co-cultured with 10 adult liver flukes that had been subjected to RNA interference to silence expression of Ov-grn-1 (below) in the upper chamber of Transwell (4 μm pore size) inserts (Corning, USA).
Recombinant rOv-GRN-1 and rTRX (control) proteins were amine labeled with Alexa Fluor 488 (AF488—Life Technologies) [57]. H69 cholangiocytes were grown to 50% confluence on optical quality glass bottomed culture dishes containing a 0.17 mm thick cover glass (World Precision Instruments). AF488-labeled proteins were added to cells at a final concentration of 3 μM and incubated for 18 h at 37°C under 5% CO2. Cells were fixed in 4% paraformaldehyde/PBS for 20 min at room temperature. Cells were permeabilized in 0.1% Triton X-100/PBS and stained with 10 μM DAPI and 165 nm Alexa Fluor 568 Phalloidin (Life Technologies). Specimens were mounted in 5% N-propyl-Gallate (Sigma) in 80% glycerol/PBS. For localization studies, cells were fixed in 4% paraformaldehyde in PBS for 20 min at room temperature and then permeabilized in 0.1% Triton X-100/PBS. Fixed and permeabilized cells were probed with either LAMP1 (lysosomes), Rab5, EEA1 (early endosome), Rab7 (late endosomes), GRP78 BiP (endoplasmic reticulum) or anti-golgin97 (Golgi) antibodies at a 1:200 dilution, followed by incubation with Alexa Fluor 568 goat anti-mouse or Alexa Fluor 568 goat anti-rabbit antibodies at a 1:1000 dilution. Conventional fluorescence imaging was performed with a 60× (NA1.4) objective using an A1 confocal research microscope (Nikon) or a DeltaVision personal research microscope (Applied Precision, GE Healthcare). Super Resolution imaging was performed using a DeltaVision OMX 3D-Structured Illumination Imaging system (Applied Precision, GE Healthcare) as previously described [58] and images were processed as described elsewhere [59].
The chorioallantoic membrane assay (CAM) assay was established based on previous studies using quail eggs [21,60]. Briefly, fertilized eggs of the quail Cortunix cortunix were incubated at 37°C in a humidified incubator for five days. The surface of the eggshell was sanitized by wiping with 70% ethanol. Subsequently, a 0.5-cm square window of shell was surgically resected. The CAM with visible blood vessels was gently pulled down after which the window was sealed with clear tape. Eggs were incubated at 37°C for 18 h. Subsequently, filter paper presoaked in 20 μl of 2 or 20 pMoles of rOv-GRN-1 was implanted. The surgical window was resealed, and the eggs incubated at 37°C for 18 h. Eggs were chilled and the surgical window was fixed with 25% glutaraldehyde. Implanted filter papers were trimmed and washed with PBS prior to counting the blood vessels using an Olympus SZX12 dissecting microscope with a light box using 32× magnification.
A head biopsy model was employed, as recommended for assessment of growth factors in wound healing [18,19]. Briefly, five female BALB/c mice per group (rTRX, PBS and rOv-GRN-1) were anesthetized (intraperitoneal xylazine 16 mg kg-1; ketamine 80 mg kg-1), after which the crown of the head was shaved with an electric razor. Mice were anesthetized three days later and the surgical site was sterilized with 70% ethanol wipes. A skin-deep wound of 5 mm in diameter was inflicted on the crown of the head using biopsy punch (Zivic instruments). The lesion was rinsed with antiseptic (Betadine, Sanofi), after which 56 pMoles of rOv-GRN-1, rTRX or PBS suspended in 1.5% methyl cellulose (Sigma) in 50 μl was applied. Thereafter, the lesion was covered with Elastoplast Spray Plaster (Beiersdorf). Progress of the wound, and wound closure, was documented with photographs taken at cumulative 1.6× magnification using a dissection microscope (Olympus) fitted with a Nikon D200 camera, each day for five days. Wound closure was ascertained in an unblinded fashion by comparison of the surface area of the lesion with the size as documented immediately after the wound was inflicted, with the assistance of ImageJ software.
H69 cholangiocytes were cultured in complete medium until ~50% confluence was reached in T75cm2 flasks. Cells were washed three times in PBS, 13.5 ml of low nutrient medium was added and cells were grown overnight at 37°C in 5% CO2. rOv-GRN-1 or rTRX (500 nM) were prepared in pre-warmed low nutrient media and 1.5 ml was added to each flask for a final concentration of 50 nM recombinant protein in media. Cells were grown for 0.5, 1, 4, 8, 16, 24 and 48 h, washed 3× in PBS and snap frozen then stored at -80°C. Cells were lysed in three ml of 0.2% SDS with 3× freeze/thaw cycles and centrifuged at 4000 g to remove cell debris. The protein in the supernatant was precipitated with methanol [61]. Precipitated protein was prepared as per manufacturer’s instructions from the 8-plex iTRAQ [62] kit (AB SCIEX) as previously described [63]. Briefly, 100 μg of protein samples for each time-point were digested with 2 μg of trypsin (Sigma-Aldrich) at 37°C for 16 h. Each sample was labeled with different iTRAQ labels and was subsequently combined into one tube for OFFGEL fractionation and LC-MS/MS analysis.
A 3100 OFFGEL Fractionator (Agilent Technologies) with a 24 well setup was used for peptide separation based on isoelectric point (pI), as described [64]. Sample clean up and desalting were performed using a HiTrap SP HP column (GE Healthcare) and a Sep-Pak C18 cartridge (Waters). Samples were separated with the OFFGEL Fractionator and collected fractions were desalted using ZipTip (Millipore) followed by evaporation by centrifugation under vacuum. The sample was reconstituted, desalted and separated with an analytical nano-HPLC column (150 mm x 75 μm 300SBC18, 3.5 μm, Agilent Technologies) before being applied to a Triple TOF 5600 mass spectrometer (Applied Biosystems); the results were analyzed as described [64].
Database searches were performed on the SwissProt database (version September 2013) using MASCOT search engine v4.0 (Matrix- Science) with parameters as previously described [64]. Findings from Mascot searches were validated using the program Scaffold v.4.2.1 (Proteome Software Inc.) [65]. Peptides and proteins were identified using the Peptide Prophet algorithm [66], using a probability cut-off of 95% (peptides) or 99% probability (proteins), and contained at least two identified peptides (proteins) [67]. Proteins containing similar peptides that could not be differentiated based on tandem mass spectrometry (MS/MS) analysis were grouped to satisfy the principles of parsimony. A false discovery rate (FDR) of <0.1% was calculated using protein identifications validated using Scaffold v.4.2.1. Furthermore, a FDR of <0.4% for the proteins identified was calculated using protein identifications validated by Scaffold. Proteins sharing significant peptide evidence were grouped into clusters. Channels were corrected in all samples according to the algorithm described in i-Tracker [68]. Acquired intensities in the experiment were globally normalized across all acquisition runs. Individual quantitative samples were normalized within each acquisition run, and intensities for each peptide identification normalized within the assigned protein. The reference channels were normalized to produce a 1:1 fold change. Normalization calculations were performed using medians to multiplicatively normalize data. A protein-protein interaction analysis was performed using the String software (http://string-db.org/) based on compiled available experimental evidence [69].
Adult flukes from hamsters were transformed with Ov-grn-1 targeted dsRNA (residues 49–333 of the 444 nucleotide transcript [7]) by square wave electroporation [70]. Briefly, 20 flukes in 100 μl of RPMI 1640 medium were dispensed into a 4 mm gap electroporation cuvette containing 5 μg dsRNA followed by a square wave pulse of 125 volts of 20 milliseconds duration. Transformed parasites were cultured for 1, 2, 3, 5 and 7 days after treatment. Total RNA was isolated from parasites and Ov-grn-1 expression measures using qRT-PCR with SYBR Green (TAKARA Perfect Real-time kit, Japan) and O. viverrini actin (GenBank EL620294.1) as a reference transcript [70]. The mRNA levels of Ov-grn-1 were normalized to actin mRNA and are presented as the unit value of 2-ΔΔCt where ΔΔCt = ΔCt (treated worms)– ΔCt (control, luciferase dsRNA-treated worms) [70,71]. ES products from treated and control worms were collected and tested for cell proliferation activity (above). The time point at which maximum cell proliferation was attained with ES products from Ov-grn-1 ds-RNA-treated flukes was used to calculate the percent reduction in cell proliferation relative to ES products from luc dsRNA-treated flukes. ES products from dsRNA-treated (Ov-grn-1 and luc) worms were assessed by SDS-PAGE with silver staining to ensure that the protein profiles were consistent between treatments.
Specific gene pathways in H69 cholangiocytes exposed to rOv-GRN-1 as described above were investigated by qRT-PCR. Cells from 6-well plates were harvested employing a cell scraper after 1 h (“early” time point) or 24 h (“late” time point) after the addition of recombinant proteins, and total RNA was isolated using the miRNeasy Mini Kit (Qiagen). The concentration, purity and integrity of the RNA were evaluated using spectrophotometry (Nanodrop 1000) and an Agilent 2100 Bioanalyzer. The RNAs were stored at -80°C until processed for cDNA synthesis and qPCR following the RT2 Profiler PCR Array protocol (Qiagen). Four RT2 Profiler PCR Arrays (Qiagen) were screened—Wound healing (PAHS-121Z); Oncogenes and Tumor Suppressor genes (PAHS-502Z); Epithelial-Mesenchymal Transition (EMT) (PAHS-090Z); Toll-like Receptors (TLR) (PAHS-018Z). Ct values were exported and analyzed for significance using RT2 Profiler PCR Array Data Analysis software version 3.5 (http://pcrdataanalysis.sabiosciences.com/pcr/arrayanalysis.php). The relative quantitation, included in the software, was performed using the 2-ΔΔCt method employing a panel of 5 house keeping genes as follows: beta actin (NM_001101), beta-2-microglobulin (NM_004048), glyceraldehyde-3-phosphate dehydrogenase (NM_002046), hypoxanthine phosphoribosyltransferase 1 (NM_000194), and ribosomal protein, large, P0 (NM_001002). Control groups (cells exposed to media alone) were used as calibrator samples. Three biological replicates were assessed and included in the analysis. The qPCR experiments were performed using a Bio-Rad iCycler iQ5 with an initial activation step of 95°C for 10 min followed by 40 cycles of 95°C for 10 sec and 60°C for 1 min. A melting curve analysis from 55°C to 95°C and 0.5°C temperature increment every 30 sec was included at the end of the run.
Statistical analyses were conducted using GraphPad Prism 6.02 software. For cell proliferation studies, two-way ANOVA with Sidak’s multiple comparison tests were used to compare the changes in proliferation induction of ES products from Ov-grn-1- compared to luc-dsRNA treated flukes. Degrees of freedom for the F-test output of the ANOVA were calculated with DFn and DFd representing the degrees of freedom of the numerator and denominator, respectively. For CAM studies, statistical analysis compared treatment (rOv-GRN-1) and media alone controls using one-way ANOVA with Dunnett’s correction for multiple comparisons. For wound healing studies, closure rate of wounds was compared by 2-way ANOVA with Dunnett’s correction for multiple comparisons. For proteomics studies with cell lines, differentially expressed proteins were determined using Kruskal-Wallis Test and results were expressed in log2 ratios. Proteins with a P-value < 0.05 and a significant log2 fold-change >0.6 or <-0.6 (for upregulated and downregulated proteins respectively) were considered in subsequent analyses. For gene expression studies, the fold change values of the genes from the four analyzed gene arrays were exported to GraphPad Prism 6.02, pooled and plotted in a volcano plot and the significantly dysregulated genes (P ≤ 0.05) plotted as a gene expression heatmap using Microsoft Excel.
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.