This study found clear evidence of vaccine-induced anamnestic responses for CD8+ but not CD4+ T-cells. However the T-cell response data only partially explained the vaccine-induced sequence changes identified by Rolland et al.
. Pre-infection T-cell responses did not predict divergence of the breakthrough sequences from the vaccine insert, yet there was some evidence of differences between vaccine and placebo breakthrough sequences in regions with observed pre-infection T-cell responses. This evidence was concentrated in Nef which has been shown to be highly immunodominant during acute infection 
Analyses of the T-cell data were challenged by methodological and data availability issues. T-cell responses were sparse such that few epitopes were recognized by multiple individuals; hence we had limited power to detect a sieve effect in immunofocused regions and to correlate the T-cell responses and sequence data. The T-cell response-based sieve analysis also had limited power since only a fraction of epitopes may experience selection pressure due to an immune response. We were limited by sample availability and minimal overlap between the viral sequence and immunogenicity data as well. In addition, methodological improvements, such as refinement of epitope-mapping algorithms and novel assays, are needed to more directly measure vaccine-induced T-cell responses. Finally, the immune responses we measured in the blood do not necessarily reflect responses in the mucosal tissue which play a key role in early infection.
The vaccine did not significantly reduce acute log viral load, although the point estimate of a 0.4 log10
reduction suggests the possibility of weak suppression that would need confirmation in a study with more acute viral load data. Separate analyses of viral load data in Step from HIV diagnosis to 4 years post-infection (median follow-up two years post-infection) demonstrated that the vaccine did not reduce post-acute viral load 
. Therefore any impact of the vaccine on viral load was early and transient.
We tested the following hypotheses regarding the impact of the viral sequence changes on acute viral load. If vaccine-induced changes in breakthrough sequences led to decreased viral fitness in terms of viral load, we would expect higher epitope-based distances to be associated with lower acute viral load, differentially so in the vaccine group. There are a few examples in the literature of CTL escape being associated with reduced viral fitness 
, although the SLYNTVATL epitope that encompasses the Gag-84 signature site is well known to escape without fitness cost to the virus. Alternatively, if vaccine-induced T-cells recognized certain epitopes and were therefore able to suppress viral replication, we would expect lower epitope-based distances to be associated with lower acute viral load among vaccine recipients, but no association in placebo recipients. Our analyses linking acute viral load with viral sequence data were designed to be sensitive to either hypothesis. We found no detectable evidence of a fitness cost as measured by acute viral load, or of reduced viral replication due to vaccine-induced epitope recognition.
The viral load analyses were also limited by low power. Acute viral load was missing for 70% of subjects, largely due to the Step study’s six-monthly HIV testing from Week 30 on (four weeks after the last vaccination). This highlights the importance of more frequent HIV testing in efficacy trials. Had HIV testing been more frequent, we would likely have been adequately powered to detect a vaccine effect on acute viral load of the size estimated here, a reduction of 0·4 log.
Sieve analysis plays a key role in the assessment of immune correlates. Specifically, it can be useful for identifying the specific immune responses responsible for any observed vaccine efficacy. This study emphasizes the need for immune correlates assessments to be based on assays that are epitope-specific and that are rigorously validated and have high reproducibility.
Despite detecting vaccine-induced anamnestic responses, this study found that the observed T-cell responses did not adequately explain the vaccine effect on founding virus populations identified by Rolland et al.
. Importantly, this implies that the measures of T-cell response that were employed did not adequately explain some immune functions that were nonetheless able to put pressure on the virus. Possible explanations include a lack of sensitivity of our immune assays or use of inappropriate assays. Neither was there evidence of a consequence of the viral sequence changes in terms of acute viral load. Nevertheless, the trend toward modest viral load suppression in vaccine recipients suggests a hypothesis that vaccines with improved CD8+ T cell responses may be able to exert stronger pressure with greater fitness cost and improved viral load suppression.