Our analyses provide an important series of observations about the outcome of the Step trial and future immunogen design. It is clear that while immunogenic in most individuals, the Merck gag/pol/nef vaccine tended to elicit responses to fewer highly conserved epitopes and more less-conserved epitopes than would be predicted from the vaccine sequences alone. An interesting question is raised as to how generic this bias towards less-conserved epitopes may be. The mechanism for these observations are unclear; potential explanations include that such highly conserved but relatively immunologically silent regions could be the consequence of retroviruses evolving to evade epitope processing in regions where mutations have a very high fitness cost, or could be due to very conserved domains also being conserved in HERVs (Human Endogenous Retroviruses), causing such regions to be seen as “self”. To our knowledge, this is the first study to analyze this question in the context of high-resolution epitope mapping with a high number of patients and a clinical HIV-1 candidate. The extent to which the bias may be related to this vaccine candidate, choice of viral vector delivery, or vaccination vs. natural infection is an important question to guide further vaccine development.
We consider the implications of these results upon vaccine efficacy in general, and the Step trial (which employed the same vaccine as studied here) in particular. First, patients became infected in both placebo and vaccine arms of the Step trial. Mutation of HIV-1 to escape effective T cell recognition has been frequently observed in humans and non-human primates. The probability of escape is a complicated function of viral load, conservation at the amino acid and nucleotide level, and HLA association. Analyses of these complex interactions through HIV-1 sequencing of infected patients are underway. We merely point here out that eliciting responses to highly-conserved epitopes would, in general, be expected to reduce the overall frequency of viral escape. Second, the number of persons who become infected may be greater for a T cell vaccine that elicits responses to less-conserved epitopes. Less conserved epitopes are less likely to be shared between vaccine and infection virus strains. Assuming a worst-case complete immunodominance model such that only a single epitope amongst Gag, Pol, or Nef is recognized (despite our data to the contrary in ), and if failure to match this single epitope leads to infection, what would be the potential implications on predicted infection rates?
The experiments and analyses are based upon epitopes of only 9 amino acids in length, thereby omitting CD8 epitopes of 8, 10 or 11 amino acids. In this and subsequent analyses, we may undercount these less frequent non-9-mer epitopes. For example, epitope mapping of select vaccinees with B57/B58 alleles and B27 alleles has shown that key 10 mer Gag epitopes such as TW10 and KK10, respectively, can be detected with peptides 10 or 15 amino acids in length but not with 9-mer peptides. Undercounting of vaccine-induced epitopes may also occur due to the ELISpot assay method. Positivity criteria were established to limit false positive responses to 5% or less, yet some real but weaker responses may be missed at these cut-offs, particularly as responses are narrowed down to the single peptide level. A second issue may complicate the interpretation of the results is TCR recognition degeneracy; A study of 9-mer ELISpot mapping data and multi-isolate sequencing from HIV-infected individuals 
indicated that one mismatch per 9-mer can still be cross-reactive (although this is subject to HLA restriction). However, even with this cross-reactivity suggested by experiment, the requirement that multiple amino acids match elsewhere in the 9-mer leads to a nonlinear relationship between the number of 9-mers matching and the number of amino acids 
. Specifically, as homologies move from more to less conserved, epitope overlap will fall off faster than a traditional sequence/protein homology score would suggest, and this disparity is apparent in actual epitope mapping data. This is a further consideration in the types of inserts that should be designed for candidate HIV-1 vaccines 
. Another finding worth noting from this study is that vaccinated subjects with similar HLA types exhibited markedly different epitope-specific responses, which emphasizes the importance of other factors other than HLA restriction in immune recognition. Indeed, we observed a strikingly different response pattern on the vaccine participants of STEP trial (data not shown).
Our analysis illustrates the need to explore several new avenues for eliciting effective T cell-based immunity for HIV-1 vaccines. One approach to achieving the immune focusing required to elicit responses to conserved regions of the genome is to construct artificial immunogens by linking together highly-conserved epitopes or regions 
; however, such strategies have proved poorly immunogenic in human phase I immunogenicity trials 
. Alternatively, one could attempt to develop immunogens with variable regions masked against immune recognition. With either of these approaches, a vaccine that stimulated responses primarily to these most-conserved epitopes may be less effective if these epitopes are infrequently presented in natural infection. Alternatively, algorithms could be designed to choose optimal natural sequences, or sets of natural sequences, to maximize epitope breadth and conservation. These could be extended to heterologous inserts for prime and boost vaccination to successively boost the conserved epitope determinants. This strategy, especially if utilized with one or more viral vectors, may generate responses focused on more highly conserved epitopes with each successive inoculation, while shared conserved epitopes would always be presented in a natural protein context.
The Step trial has raised the minimum threshold for the level of T cell immunity that must be elicited in human subjects for an effective outcome. The importance of a high responder frequency and the magnitude, quality, and breadth of response are well appreciated. The result presented here is that both HIV infected subjects and healthy volunteers immunized with the same vaccine as in the Step trial exhibited a response biased towards less-conserved epitopes. Developing an effective T-cell based vaccine as part of an overall HIV-1 vaccine will require eliciting robust responses to well-conserved epitopes. Overcoming this bias will require both an improved understanding of HIV and its interaction with the presentation and processing machinery, probing the underlying causes for potential bias towards less-conserved epitopes, and the refinement of computational design strategies to overcome this bias and refocus the immune response on a broad set of highly conserved epitopes.