Sequence conservation analysis and in silico human leukocyte antigen-peptide binding predictions for the Mtb72F and M72 tuberculosis candidate vaccine antigens

This in silico study explored the sequence conservation of Mtb32A, Mtb39A, Mtb72F and the current candidate vaccine antigen M72, and predicted MHC-I and/or MHC-II-binding peptides for these proteins. Our results led to two main conclusions. First, the Mtb32A and Mtb39A proteins (and thus also the Mtb72F and M72 constructs) appeared to be well conserved (with at least 98 % identity) among the strains representative for the major part of the Mtb geographical distribution (including MDR, XDR and pre-XDR strains), with for Mtb39A a higher sequence conservation than previously reported in [26, 27]. Second, the putative CD4⁺ T-cell epitopes predicted for Mtb72F and M72 were shown to cover a broad range of HLA Class II DRB1, DRB3/4/5, DQ and DP alleles, and the putative CD8⁺ T-cell epitopes predicted for Mtb72F covered each of the HLA-A or -B alleles assessed.

Hyperconservation has been reported for the large majority of human T-cell epitopes in the Mtb complex, although pe/ppe genes were excluded from that study [40]. Previously, mtb32a sequences of H37Ra, H37Rv, Erdman and the clinical isolate CSU93 were found to be identical [13], and, extending these studies, Mtb32A was shown to be relatively well conserved among several clinical isolates, with amino acid changes in only 6 % of the 225 investigated strains [28]. Consistently, our data revealed a high level of conservation for Mtb32A. For Mtb39A, our findings contrast with previous studies which suggested a substantial genetic variation for this protein, potentially associated with the homologous recombination with ppe19 and ppe60 [28, 29]. Nonetheless, several of the minor insertions and/or deletions we observed in the Mtb39A sequence (the 2-aa insertion at the 162nd base and the 1-aa deletions at the 274th base, in repeat regions in 1 and 6 strains, respectively) may be identical to those reported for Arkansas-derived samples in the above study [28]. Of the frameshifts leading to major amino acid changes, as we observed in 3 strains, the one in the avirulent H37Ra strain was consistent with earlier findings [41] and of little clinical relevance, while the one in EAS054 has not been reported previously to the best of our knowledge. Possibly, the number and origin of the strains evaluated here reflect the worldwide prevalence of Mtb more accurately than those used in the referenced studies, which included 225 US and Turkey-derived clinical strains [28], or 16 clinical isolate sequences from a public database [29]. As we have not assessed whether the minor insertions and/or deletions coincide with locations of predicted putative epitopes, the impact of the associated residue changes on the recognition of clinical Mtb strains by T-cell responses induced by Mtb72F or M72 candidate vaccines, is not known.

MHC-II binding peptides in Mtb72F were predicted to cover all DRB1, DRB3/4/5, DQ and DP alleles assessed, except for 2 DRB1 alleles (03:02 and 16:04), 1 DQ and 1 DP allele. Based on the used database, this DP allele (DPA1*02:02-DPB1*05:01) does not occur in the populations assessed, while the DQ allele (DQA1*02:01-DQB1*02:01) only prevails in these populations in haplotype with DRB1*07:01, which was predicted to be well covered with binding peptides. Several aspects are important to consider with respect to the implications of the absence of predicted binding peptides for the two DRB1 alleles for the candidate vaccines’ potential efficacy in the populations concerned. First, except for the higher frequency of DRB1*03:02 in certain Sub-Saharan African populations (≤17.2 %) as compared with the worldwide genotypic frequency of 1.1 % [37], the associated allele frequencies were generally low in these populations (≤1.5 %). Moreover, a potential absence of binding peptides for the two DRB1 alleles in these populations may be compensated by peptides present on other HLA-DR, −DP or -DQ alleles, and vice versa. For instance, for the South-African Limpopo Venda population, DRB1*13:02 and DQB1*03:01–03:04 have been associated with TB disease [42]. In this population, the frequency of DRB1*03:02 is relatively high (9 %). The absence of binding peptides for DRB1*03:02 may in this case be compensated by several other alleles prevailing at high frequencies in this population (e.g., DRB1*11, DRB1*03:01, DRB1*13:01, DRB*01 and DQB1*06 and DQB1*05; data retrieved from [36]), for which we predicted high numbers of binding peptides per allele in Mtb72F. It may also be reassuring that in a recent clinical trial, the candidate vaccine M72/AS01 was able to induce robust CD4⁺ T-cell responses in healthy adults in South Africa [22]. Furthermore, it is noted that for other antigens (meningococcal serogroup B proteins) it was shown that a set of only 2 predicted T-cell epitopes could theoretically cover large proportions of over 11 populations worldwide [43]. A confounding factor is that the allele frequencies sourced from a public database may have only partial or local coverage (e.g. cities or ethnic groups), since they are usually based on relatively small sample sizes rather than population statistics. Similarly, since certain alleles specific to susceptible populations have likely been the prime focus of several studies, the database may contain disproportionally high numbers of such alleles.

Although our data showed that the changes introduced in Mtb32A in order to construct Mtb72F may result in an overall reduction of Mtb binding peptides, only two alleles present in the populations assessed (DRB1*03:02 and 16:04) were predicted to have no binding peptide in Mtb72F. The other modifications (addition of hinge sequences and a poly-His tag) were predicted to jointly generate 19 binding peptides, of which one peptide was predicted to be absent in M72. Since non-Mtb epitopes do not contribute to the vaccine’s ability to induce Mtb-specific responses, their relatively low frequency in Mtb72F and M72 is reassuring.

Previously, in silico MHC-II binding predictions for 9-mer peptides in Mtb72F have been generated for 34 DRB1 alleles [26, 27]. Comparison of the previous data with the present predictions for 15-mer peptides and DRB1 alleles revealed up to 37-fold differences in the allele-matched binding peptide numbers, which can likely for a great part be explained by the different binding peptide lengths assumed by the prediction algorithm. HLA-I and HLA-II binding peptides are composed of 8–10 and 13–17 amino acids respectively [44]. The peptide-binding groove of MCH-II molecules is open-ended, which complicates identification of the 9-mer core peptides and the flanking residues (which may also contain predictive information [45]). MHC-II prediction algorithms assuming 9-mer binding peptides do not use information present outside the core, leading to lower accuracy [46]. Consistently, it was shown that including flanking residues among inputs improved the performance of MHC-II prediction methods [47, 48]. However, even predictions generated by algorithms assuming 15-mer binding peptides may have limited reproducibility, as suggested by the present results. Indeed, comparison of our results for Mtb32A and Mtb39A with allele-matched experimental binding data [18] revealed that the algorithm’s sensitivity and specificity were variable and occasionally rather low for Mtb72F (e.g., 28 %). It is noted though that due to the lack of additional suitable in vitro data, this comparison covered only 3 of the 158 HLA-DRB1 alleles, and none of the DRB3/4/5, DQ and DP alleles assessed in our study. While the observed variability in the performance of the algorithm may be a consequence of the limited extent of this evaluation, it could therefore also be inherent to the limitations of the used prediction algorithms for peptide binding in general. Similarly, most algorithms predicting T-cell activation for class II binding peptides display low sensitivity, as less than half of the defined T-cell activating peptides were predicted to be binders [46]. Thus, epitope identification requires an integrated approach, in which in silico binding predictions are complemented with confirmatory biochemical verification of MHC processing and binding, and of the ability of the predicted peptides to induce T-cell responses [49]. Other useful tools are in silico methods that enable linking binding affinity patterns to topological characteristics [50].

Due to the capped MHC-I peptide binding groove, this binding is more specific than binding to MHC-II, and HLA-I binding prediction algorithms are considered relatively accurate [46, 49]. Indeed, there was only at most a five-fold difference between the previously reported numbers of HLA A- and B-binding peptides in Mtb72F [27] and the numbers per allele computed here.

In the present study, the majority of predicted peptides were presented by A*02:01, listed in the IEDB among the most frequent HLA class I alleles in regions with high TB incidence [4]. The predicted high affinity-binders were mainly presented by A*02:01 and B*07:02 rather than A*03:01, consistent with data obtained for a panel of 432 H37Rv-derived peptides [5], while other Mtb proteins (e.g. Mtb8.4, CFP10 and Ag85B) were shown, experimentally or in silico, to be primarily presented by various HLA-B molecules [51, 52]. Furthermore, our data showing the predicted binding of the Mtb72F peptides APINSATAM, APAAAAQAV, LLGQNTPAI and SSKLGGLWK to B*07:02, B*07:02, A*02:01 and A*03:01 respectively corroborates earlier predictions for these peptides [5]. Of note, several previously predicted HLA I-restricted Mtb peptides were experimentally shown to induce CD4⁺ rather than CD8⁺ T cell responses [53]. Similarly, the HLA-B44-restricted epitope MWAQDAAAMF [16], was also found in two 15-mer Mtb39A peptides that were experimentally defined [18] and predicted here to bind to DRB1*04:01 and/or 01:01.

Last, for cytomegalovirus antigens, it has been suggested that less than 20 uniquely defined HLA-restricted CD8⁺ T-cell epitopes could provide 90 % coverage of the three major ethnic groups [54]. Possibly, the putative class I Mtb epitopes predicted for Mtb72F could also have a wide coverage of populations worldwide, and, given that only minor changes were introduced into Mtb72F to generate M72, this would also apply to M72.