Vaccines have helped eliminate or control many public health scourges—smallpox, measles and hepatitis B among them—so it's not unreasonable to think that HIV might be similarly kept in check. Yet in spite of a surfeit of discussion in recent years of the best approach to developing AIDS vaccines, that goal remains out of reach. Is it a question of biology? Or policy? Or both?

HIV has thwarted researchers in every way. It mutates in response to drugs, it mutates in response to immune pressure, it co-opts multiple cellular pathways to evade detection or destruction, it hides in reservoirs, and, once it has killed one reservoir cell source, it mutates in order to infect another. It's a basic researcher's dream and a translational researcher's nightmare.

Yet scientists have been persistent. Viral Achilles heels have been identified and targeted, vaccines designed and tested. 1987 heralded the first trial of an HIV vaccine—the gp160 subunit vaccine. Since then, more than 50 candidates for preventive HIV vaccines and more than 30 candidates for therapeutic AIDS vaccines have been tested by the US National Institute of Allergy and Infectious Diseases (NIAID). Yet in 20 years of clinical testing, no AIDS vaccine has been approved.

Still, some of the failed approaches continue to command time, money and resources. The vaccine ALVAC-HIV was tested in phase 1 and 2 trials for its ability to stimulate cytotoxic T cell responses. Because the results suggested that the vector lacked significant immunogenicity, the vaccine did not proceed to phase 3 trials in the US. In 2003, AIDSVax, a vaccine designed to induce neutralizing antibodies, proved to neither elicit detectable antibody titers nor affect viral load when tested in phase 3 trials. And yet, in late 2003, a 16,000-person trial of a combination vaccine of ALVAC-HIV plus AIDSVax was initiated in Thailand with the support of the US Food and Drug Administration. The study will be completed in 2009 at an estimated cost of $119 million. Yet many experts are highly skeptical that it will show any efficacy whatsoever.

In contrast, the phase 2b testing of Merck's vaccine candidate, V520, was regarded with much optimism. Using a modified common cold virus—adenovirus type 5 (Ad5)—as a vector to carry three HIV genes, the vaccine was designed to induce a T cell response that would probably not prevent infection, but might mitigate its severity by decreasing viral load. But on November 9, Merck reported that the vaccine neither prevented infection nor reduced viral load in trial participants and that the number of infections in the vaccine arm exceeded that in the placebo arm of the trial.

Moreover, preliminary stratification of the participants showed that, in those individuals with high levels of preexisting immunity to Ad5, more infections occurred in recipients of the vaccine than in recipients of the placebo, raising the question of whether the vaccine increased susceptibility to HIV in individuals previously exposed to Ad5. More extensive analyses are needed to determine whether these results are simply due to chance, and so what they mean for related approaches in clinical development remains unclear.

So where are we in trying to develop a successful AIDS vaccine? Efforts to induce neutralizing antibody responses have not been successful, and we still don't know the correlates of protection against HIV infection in humans. Although these and other biological obstacles remain, the results of these trials point to additional problems in how vaccine candidates are pursued.

An enormous investment of time, money and effort has been spent developing clinical approaches that do not adequately reflect advances in preclinical HIV research. A vaccine that induces solely antibodies or solely cytotoxic T cells has long been thought to be incapable of controlling HIV. Rather, a combination of the two is probably essential for making headway against AIDS. And yet large trials, such as the Merck V520 and the failed AIDSVax trials, continue to test approaches designed to harness only one arm of the adaptive immune response.

Studies in animals also clearly demonstrate the benefit of a heterologous prime-boost approach, in which the boost administers the immunogen in a different form (often through the use of a different vector). Certainly, research with adenovirus-based vectors shows the need for this approach, because preexisting immunity can hinder the induction of a robust response if the same vector is used in both the prime and the boost, as was the case in the V520 trial.

Although the ALVAC-HIV-plus-AIDSVax trial seems to address these concerns by using a prime-boost approach intended to induce both B and T cell immunity, initiating a costly phase 3 trial of vaccines that failed in earlier testing is not an efficient use of resources.

Studies in humans are certainly necessary to answer fundamental questions in vaccine development. Establishing discrete go/no-go decision-making criteria might better help guide regulatory bodies and the NIAID in evaluating large trial proposals and might allow for faster incorporation of research findings to improve existing approaches. Favoring smaller phase 2b trials for 'proof of concept' over large phase 3 trials might also more rapidly determine the efficacy of any one approach. And increasing responsiveness of trial investigators to the recommendations of advisory panels to improve or revise planned clinical trials—even if underway—can only enhance the value of the information derived from a trial.

Clinical development of HIV vaccines has been underway for more than 20 years. Although the results of the Merck trial are disappointing, the outcome is not a reason to forgo vaccine efforts altogether. Development of the polio, measles and hepatitis B vaccines took 47, 42 and 16 years, respectively (New Engl. J. Med. 353, 753–757, 2005). The problems facing HIV vaccine development aren't new, and the solutions are in no way obvious or straightforward. But the clinical trial machine could perhaps learn from HIV—and evolve a little faster.