Defining epitope coverage requirements for T cell-based HIV vaccines: Theoretical considerations and practical applications

Despite advances in therapeutic and prevention campaigns an estimated 33.4 million people currently live with HIV/AIDS and in 2008 alone, an estimated 2.7 million new infections occurred [1]. It is therefore generally accepted that controlling the global HIV pandemic will require a successful HIV/AIDS vaccine [2–5]. The design of an effective HIV vaccine is a multifaceted challenge involving the development of novel vectors, adjuvants and antigen design and formulation strategies. The elicitation of HIV-specific CD4⁺ and/or CD8⁺ T cells is considered an essential component of any potentially efficacious vaccination strategy [4, 6–8]. Such cells could act as direct effectors in limiting virus replication, or as helper cells facilitating other effector mechanisms [9–13]. While a comprehensive review of the evidence supporting a T cell based HIV vaccine is beyond the scope of this introduction, a T cell based vaccine should preferably elicit an immune response capable of preventing infection or, at a minimum, result in control of viral replication after infection [14–16]. The recent Thai phase III trial (RV144) remains an important example of a vaccine that may require appropriate T cell help its protective effect through adequate antibody generation [17].

T cells recognize virally infected target cells or antigen-presenting cells via T cell receptor binding to short peptides (typically 8-12 amino acids in length) presented in the context of HLA class I or class II alleles [18–21]. It is the exquisite specificity of this interaction that determines activation and subsequent function of the T cell [22–24]. Since even a single conservative amino acid substitution within an epitope may alter or ablate appropriate T cell function, it becomes readily apparent why viral genetic diversity is foremost of the many daunting obstacles that have impeded the successful development of an HIV vaccine. The ability of HIV-1 to mutate and rapidly evolve to evade host immune pressure is mediated by an error-prone reverse transcriptase, genetic recombination and the host-derived APOBEC editing enzymes [25–28]. Due to the vast diversity of potential challenge viruses that constitute the global HIV-1 pandemic much effort recently has been focused on designing vaccine insert sequences to reduce the impact of viral diversity. These approaches include, but are not limited to, synthetic constructs that result in conserved minimal epitope (polyepitope) and conserved element concatenation into pseudoprotein strings, computationally derived constructs such as consensus, ancestral, center-of-tree and mosaic sequences, as well as cocktails of natural sequences derived directly from HIV-1 [26, 29–41]. Regardless of which of these insert design strategies ultimately proves to be the most successful, the critical question of how best to assess and impartially compare these products in pre-clinical studies or the setting of a vaccine trial remains open. While several reviews, studies and experimental approaches have focused on how to deal with the question of viral diversity in the context of immunogen design, to date no study has focused upon how diversity coverage may be meaningfully assessed in the setting of an HIV vaccine trial [26, 32, 34, 39, 42].

Mounting data suggests that the breadth (number of discrete epitopes) and depth (amount of epitope variants recognized) of the T cell response is an important determinant of both HIV pathogenesis and primate model SIV vaccine efficacy. Multiple studies have recently indicated that it is the breadth of the T cell response rather than the magnitude, which correlates best with viral load set point in chronic HIV infection [43–50]. In addition a seminal study in the rhesus macaque SIV model demonstrated that immune control of an SIV challenge by a T cell-based vaccine was correlated with the breadth of vaccine generated T cell responses [51]. The fundamental premise of this study was to determine the extent of coverage a particular vaccine product would provide for either the global HIV pandemic or for the homologous subtype from which the product was derived, and to assess the relationship between epitope breadth and depth, and vaccine coverage. In this study, it is shown that by incorporating epitope breadth and depth analysis into vaccine coverage assessment, metrics can be used to determine the suitability of a particular vaccine formulation for a given setting. Parameters such as the level of coverage of possible incoming HIV-1 strains (75%, 90% or 95%), the sample set of possible strains (single subtype versus global group M) and the number of desired exact epitopes matches with an incoming strain (1-, 2-, or 3-hit models) are all shown to place dramatically different requirements upon epitope breadth. Furthermore, increasing the valency (number of included HIV-derived inset sequences) of the vaccine formulation is shown to dramatically increase the breadth of coverage and that for even natural sequence-based products, coping with the enormous genetic diversity of HIV may not be a totally intractable problem. While this study has focused upon extant natural sequence-based vaccine products that are currently under development, the analysis presented could easily be performed on new generation computationally derived vaccine products. Not only does this approach provide an empirical assessment of vaccine coverage potential based upon the formulation, but it also allows direct theoretical comparison of vaccine products as they might apply to a particular epidemic of HIV-1. Finally, using example data from a vaccine trial it is shown that the theoretical calculations compare well with the experimentally generated data.

In this study it has been demonstrated that measurement and analysis of the breadth of the T cell response - the number of discrete T cell epitopes - generated by a vaccine provides an important metric for HIV vaccine assessment. Analysis of the breadth of T cell coverage not only illustrates the problem that HIV genetic diversity presents to the immune system, but it also demonstrates that increasing the breadth of coverage can amplify the capacity of even natural sequence-based vaccine products to cope with viral diversity. Practical metrics can be set prior to the commencement of a vaccine trial that take into account both the HIV sequence content in the vaccine product and the diversity of HIV in particular target settings. Importantly, the relative capacities of different multi-valent formulations to generate sufficient epitope coverage can be compared directly. While the analyses presented here are limited to exact 10-mer matches as epitope hits this was considered to be a reasonable compromise between the median CD8⁺ T cell epitope length (9-10 amino acids) and the minimal core of a CD4⁺ T cell epitope (10-12 amino acids). However, shorter or longer epitope windows (8, 9, 10, 11 or 12 amino acids) and more relaxed matching requirements (1 or 2 tolerated amino acid mismatches) could be incorporated as desired into this analysis. It was felt that the requirements used here represent a more conservative approach and hence may in fact over-estimate the epitope requirements since no mismatches are tolerated. The studies of Korber et al. [26], using computationally generated mosaic vaccine insert sequences clearly demonstrate a progression to increased vaccine coverage as matching requirements are relaxed and as smaller epitope windows are considered [26, 32]. Importantly, the practical application of epitope mapping in vaccine evaluation demonstrates that understanding the depth of individual epitope coverage provides an added layer of sophistication to vaccine coverage assessment. Preclinical studies investigating the cellular immunogenicity of mosaic constructs and comparing them with natural sequence constructs unequivocally demonstrates the importance of breadth and analysis as a metric of the T cell response [55, 56].

Empirical data has begun to emerge that implicates T cell epitope breadth as a critical parameter in both the natural control of HIV-1 infection and in pre-clinical testing of different vaccine strategies and modalities. In cross-sectional studies of naturally HIV-1 infected, treatment naïve human subjects, investigators have identified the breadth of the cellular immune response, rather than the magnitude of the response, as the primary discriminator of effective immune control as measured by set-point viral load [43, 45, 49, 50, 57, 58]. Furthermore, the transmission of escape variants for Gag-specific CD8⁺ T cell epitopes is associated with reduced viral load in linked subjects, and if multiple epitopes are targeted the likelihood of epitope reversion is diminished [49, 50, 59–61]. Computational analysis and epitope prediction based on HLA type and HIV proteome sequence in a large cohort of 302 subjects has also shown that epitope repertoire and breadth is critical to viral load control [48]. The seminal study by Liu and co-workers [51] demonstrated immune control of an SIV challenge by a T cell-based vaccine in rhesus macaques and linked breadth of vaccine generated T cell responses with relative efficacy of different immunogens. This study established for the first time a correlate of cellular immunity and mitigation of disease outcome in a realistic SIV challenge model. To date there is no clear indication of the exact breadth of the T cell response that the human immune system is capable of generating against the Gag and Env proteins considered here. However, pre-clinical data from the rhesus macaque model suggest that 20 or more T cell epitopes can readily be generated by vaccination with a non-replicating viral vector [55, 56]. In natural HIV infection it has been shown that 20 or more epitopes can be detected within a single individual when the entire HIV proteome is considered [47, 54]. Whether such breadth is practically attainable in the setting of a human vaccine study remains to be proven. In this study the generation of greater than 20 epitopes per protein per vaccinee was considered an intractable problem for the immune system. Hence, 20 epitopes was the limit of calculation for epitope breadth in all analyses presented here.

It is therefore not surprising that increasing the breadth of T cell recognition might be associated with better control of viremia, since the more epitopes recognized the more difficult it would be for the virus to find multiple escape mutations of relatively low fitness costs. A less obvious and perhaps more biologically cogent reason behind the breadth association with viral control is that certain T cell epitopes may be more effective than others. Prior studies have implicated T cells targeting certain gene products of the virus (e.g. Gag) as being associated with better control of viremia [43, 45, 48–50, 62]. Further studies have refined these analyses to regions of certain gene products and even particular epitopes. The reasons for this phenomenon could include any or all of the time and timing of gene product expression, the HLA background of the infected individual, the availability of generated epitopes for antigen processing and presentation, and the TCR repertoire. Aside from certain epitopes (particularly in the Gag and Nef proteins) presented by alleles frequently associated with control of viremia (HLA-B2705, -B5701/03, -B5801; [49, 50, 62, 63]) there is a paucity of knowledge relating to exactly which epitopes provide the best protection. Without a priori knowledge of epitopes that should be targeted by any given individual, the multi-hit models of vaccine assessment presented here provides a potential alternative. While the single-hit model assumes that all epitopes may be equally effective at mediating infected cell recognition, the 2-hit and 3-hit models assume that only one-half and one-third of all possible T cell epitopes respectively are actually effective in mediating T cell recognition of virally infected target cells and inhibiting viral replication. The data presented here show how the consideration of multiple hit models places a greater onus upon any given vaccine product to provide appropriate coverage of all possible incoming strains of HIV-1. Using the natural sequence based vaccine constructs developed in the MHRP it is shown that if used in mono-valent formulations there is sufficient coverage of the homologous subtype, even within the Env protein if sufficient breadth of the immune response (4-6 epitopes) is generated. Furthermore, multi-valent formulation of the products potentially provides ample coverage (> 90% of strains) even for the global HIV-1 group M epidemic for both Gag and Env when considering the single-hit model. The greater diversity within the Env protein compared with Gag is reflected in the generalized requirement for nearly twice the number of epitopes in Env to reach any level of equivalent coverage. However, if a multi-hit model proves to be an accurate description of T cell requirements for an effective vaccine, then the requirements to meet any level of coverage increase dramatically. For mono-valent formulations both the 2- and 3-hit models place such high requirements upon both Gag (8-12 epitopes for 75% coverage using a 2-hit model) and Env (17-19 epitopes for 75% coverage using a 2-hit model), that generating sufficient breadth would be a difficult proposition. Considering a 3-hit model and 90% coverage of the global epidemic the epitope breadth requirements present what is most likely an insurmountable challenge for the immune system (> 20 epitopes required per protein). Multi-valent formulation of the natural sequences does provide increased benefit in terms of generating potentially more cross-reactive cells, especially when considering subtype-specific vaccine coverage. However, as far as global group M coverage is concerned multi-valent formulation and delivery of the natural sequence based products is the most realistic means of ensuring sufficient coverage if multiple exact T cell epitope matches are required for vaccine efficacy.

While quite sobering, these data serve to underscore the practical requirements for a T cell-based vaccine to be efficacious in a broader setting and also provides meaningful tools for assessing those requirements. This kind of analysis could also be directly applied to any kind of computationally derived vaccine insert sequences such as consensus, center-of-tree, mosaic, conserved element or string epitope concatenated sequences [26, 29–41]. Such approaches offer a means of maximizing vaccine coverage at the single epitope level and by direct inference would also improve coverage for multiple epitope and multiple-hit requirement models of vaccine coverage. The only caveat that needs to be fulfilled for the computational approaches is proof-of-concept studies in human trials that verify the products can indeed be immunogenic. Epitopes targeting sites of high-depth rather low depth of coverage is embodied in the mosaic, center-of-tree and conserved elements vaccine approaches. Since, the processing of antigens that ultimately results in the presentation of epitopes is not fully understood, a critical issue for the center-of-tree and conserved elements vaccine constructs in particular is a strategy for assembling the peptide fragments for operational vaccine delivery. Simple concatenation of peptide fragments to produce embedded epitopes in the context of the assembled fragments is not sufficient as was demonstrated by the poor immunogenicity observed in two recent trials of such an approach [35, 36]. Mosaic sequence based products have at least been shown to be immunogenic for very broad T cell responses in pre-clinical studies in primates [55, 56]. The present study has been restricted to the inclusion of natural sequence-based products in the MHRP product development plan (MVA vector-based). These products have all been shown to be immunogenic for T cell responses in pre-clinical studies and in particular, the MVA-CMDR product has been shown to be immunogenic for T cell responses as both a prime and a boost in human phase I clinical trials [31, 53, 64]. An added rationale for at least maintaining the native Env protein is conformation for antibody production. A critical assumption that the multi-valent analyses have been predicated upon is the notion that each unique variant of an epitope within the cocktail is either immunogenic or cross-reactive with other variants of the same epitope in the cocktail. In cases where this assumption does not hold, the multi-valent analysis presented here may overestimate epitope depth of recognition. However, as discussed previously this possible over-estimation may be counter-acted numerically by the high stringency of exact epitope recognition that the model is based upon. In the setting of a vaccine trial mapping of the response permits analysis not only of the number of T cell epitopes, but also the epitope repertoire and depth of recognition. As shown in this study accurately mapping T cell responses adds the important dimension of depth of epitope recognition to the analysis. Knowledge of at least the approximate minimal epitope (down to 15-mer resolution), allows both a more accurate theoretical estimation of potential depth of recognition and further experimental validation of the cross-reactivity of a response using designed peptide variants.

The T cell epitope breadth analysis presented here also highlights the issue of formulation and delivery of multi-valent vaccine products. While the simplest approach for delivery would be co-formulation for both priming and boosting vaccinations, a multi-valent product may benefit from priming with a single valency product and sequential boosting with the heterologous products. Such an approach might focus the T cell response upon the boosting of cross-reactive T cell responses at the expense of boosting potentially immunodominant responses with limited cross-reactivity. The heterologous prime-boost approach has already been shown to be effective in generating high magnitude T cell responses in several human vaccine trials [64–66]. The recent successful Thai phase III trial (RV144) provides an important example of how the role of T cells and their cross-reactive capacity must be taken into account even in the setting of a heterologous prime-boost approach involving a recombinant protein as one of the formulations [17]. A vaccine that generates antibodies as the likely mediator of its efficacy would still require appropriate T cell help for formation of an adequate antibody repertoire [67–70]. The HIV vaccine field is currently moving forward from the pre-clinical phase into the clinical testing phase of vaccines with increasing complexity in terms of both formulation and the mechanism(s) by which they mediate protective effects. The role of T cells in all of these scenarios must be addressed whether it is indirectly as help for antibody production against multimeric envelope proteins, or directly as effector or effector-memory cells generated by recombinant viral vectors [2, 6, 51, 71, 72].

There are several further refinements and caveats that need to be kept in mind when such analysis is performed. An implicitly important variable in this analysis is the query database. This study has made use of the LANL HIV sequence database without bias for the prevalence of any given subtype. However, this database is heavily biased towards subtypes B and C in terms of raw sequence constituency. Hence, more nuanced analyses for vaccines targeted for East Africa, West Africa, Asia and eastern Europe, would benefit from more expansive query datasets that include better sampling of subtypes A, D, G, CRF01_AE and CRF01_AG. This highlights the importance of sequence sampling and the dataset used as the basis for the coverage requirement calculations. Experimentally, there is also a need to discern the relative contributions of CD4⁺ versus CD8⁺ T cell responses. CD8⁺ T cells are generally considered the effector cells responsible for viral control and clearance and therefore the specific contribution of these cells to the overall should be specifically assessed. The most robust and practical assay for epitope enumeration and identification in a vaccine trial is the IFNγ Elispot assay. However, this assay is not readily adaptable to discerning the phenotype of the responding cells without significant sample usage. Therefore, high throughput flow cytometry-based assays that can measure epitope depth and breadth as well as provide high content phenotypic analysis of the responding cell populations are required.

In summary, we have shown here that appraisal of epitope breadth and depth of recognition should be an important metric that is considered when laboratory assessment of HIV vaccine trials is performed. Furthermore, the incorporation of a multi-hit model into the theoretical breadth analysis helps to set the metrics for broadly specific T cell responses in the absence of mechanistic data that explains the biological underpinnings of why high breadth responses may be advantageous. The suitability for a particular vaccine and multi-valent formulation can be also assessed during the design phase of a clinical trial as well as during the execution phase. Moreover, this study shows that multi-valent formulation, or heterologous prime-boost strategies, may extend the utility of vaccine products based upon natural sequences. While it is clearly shown that mono-valent natural sequence based products face a potentially intractable problem for providing appropriate coverage for the global HIV-1 pandemic, multi-valent formulation can improve the utility of these extant vaccine products, especially in an epidemic consisting of a single subtype. Given that computationally designed vaccine products optimized for maximizing coverage, such as mosaic products, will soon be tested in human clinical trials it remains paramount that robust and practical breadth assessment is included in the testing schema for comparing and contrasting these and other approaches for successful HIV vaccine implementation.