On modeling the effects of T-cell vaccines on HIV acquisition and disease.

A 'T-cell vaccine' aims at generating cytotoxic T-lymphocytes (CTLs; the so-called 'killer' T-cells) rather than antibodies (as for traditional vaccines). The first (phase IIb) trials of this concept against HIV/AIDS began in 2004. What can mechanistic modeling contribute to understanding the biological action of this class of vaccines, if any? Models are appropriate in any discussion of three potential vaccine effects: on acquisition of infection; on state of disease ('viral load', VL) after infection; and on preventing escape from immune control. Concerning the first two, P. Gilbert, S. Self and I introduced new stochastic models of early HIV infection and the CTL response, and, making use of recent estimates (derived in collaboration with O. Yang and L. Corey) of the rate that CTLs can kill HIV-infected cells, made the (surprising?) discovery that CTLs might prevent some infections--as the trial designers implicitly acknowledged when they chose the dual end points of the study. On sustaining control, we have derived a theoretical formula for the rate of escape by stepwise mutation and a new method of simulating HIV and CTL dynamics in vivo (permitting new mutant strains a stochastic evolution--essential, in our view). These quantitative models and simulation techniques can also prove useful to biostatisticians. For example, in preparation for the STEP trials, Gilbert, Bosch, and Hudgens developed a novel technique for estimating a causal effect of a vaccine on VL while accounting for post-randomization selection bias. By simulating thousands of trials, we demonstrated that GBH's method can correctly identify efficacy while protecting against falsely concluding that the vaccine exacerbates disease. When trial data becomes available, the models may also be exploited to make complementary analyses which, while not relevant to vaccine licensure, may suggest new biological hypotheses.