As someone accustomed to persevering on a long-term project in which repeated periods of hard work lead to modest progress, only to be followed by setbacks, Sisyphus would be well suited to a career in AIDS-vaccine research. Papers on pages 331 and 335 of this issue1,2 illustrate the point. In the first paper, Shiver and colleagues1 describe how they immunized groups of rhesus macaques against a protein from a monkey AIDS virus, the simian immunodeficiency virus (SIV), effectively stimulating antiviral cellular immune responses. The immunizations did not prevent, but did help to control, subsequent infection with a related virus. But Barouch and colleagues2 show that AIDS viruses can mutate to evade such vaccine-induced, virus-controlling cellular immunity, calling into question a vaccination strategy based solely on such responses.

The aim of most vaccines that offer protection against viruses is to stimulate antibody molecules that can neutralize the virus or otherwise help clear the infection, and cellular immune responses, particularly by cytotoxic T lymphocytes (CTLs) that bear the surface marker CD8 and can kill virus-infected cells. But for the human AIDS virus HIV-1 it has proved difficult to generate vaccines capable of inducing antibody responses that neutralize the broad range of virus strains found in patients. CTL responses may be able to cope with a wider range of virus strains, and appear to help control HIV-1 in infected people, so much current attention in AIDS-vaccine research is focused on vaccines that stimulate CTL responses.

Pursuing this approach, Shiver et al.1 systematically compared vaccination strategies using DNA molecules or engineered non-SIV viruses (a vaccinia virus or an adenovirus known as Ad5) to express a single SIV protein, Gag. After immunizing macaques with these potential vaccines, the authors detected strong and sustained responses by T cells that express CD8, most notably in animals immunized with the Ad5 vaccine — either alone or after 'priming' immunization with DNA.

To assess the efficacy of the vaccines, the authors1 then challenged the macaques with a highly pathogenic chimaeric simian–human immunodeficiency virus (SHIV), SHIV 89.6P (ref. 3). This virus contains a gene encoding the viral glycoproteins from the HIV-1 outer 'envelope', as well as several SIV genes, including a Gag-encoding gene matching that used in the vaccines. Like HIV-1 and SIV, SHIV 89.6P infects and kills 'helper' T lymphocytes bearing the cell-surface marker CD4 (ref. 3).

Shiver et al. found that their vaccination strategies did not prevent infection with SHIV 89.6P, but did modulate its course, most strikingly in animals immunized with the vaccine based on Ad5. Early on, peak levels of virus in the macaques' blood were only slightly lower than in controls. But by 70 days after challenge, vaccinated animals showed markedly lower levels of virus and preservation of CD4-expressing T lymphocytes. This 'partial protection' is similar to that obtained with other vaccine approaches4,5 against challenge with SHIV 89.6P. Used clinically, a vaccine that controlled HIV infection might result in a longer period of infection without symptoms, and a decreased risk of transmission.

However, Barouch et al.'s work2 provides a cautionary counterpoint to these encouraging results, and raises questions about the strategy of controlling infection by using vaccines that stimulate CTLs alone. These authors describe the course of infection in a rhesus macaque that was at first partially protected from SHIV 89.6P by a DNA-based immunization approach5. This vaccinated animal showed high peak levels of virus after challenge, but viral levels eventually declined to below the limit of detection, with no depletion of CD4-expressing T cells5. However, during a year of follow-up2, virus replication increased to measurable, albeit modest, levels, in association with the emergence of mutations that allowed the virus to elude a dominant CTL response. An acute loss of CD4-expressing T cells and progressive disease ensued.

This is perhaps more disappointing than surprising. Such escape mutations have been detected before in SIV-infected animals6, and there is also evidence of HIV-1 evolution in the face of 'effective' anti-retroviral therapy7. So we might have expected, in the setting of similarly 'controlled' viral replication in a vaccinated animal, the emergence of mutations that allow the evasion of immune responses. But this nonetheless represents a fundamental and potentially ominous challenge to the idea of containing, rather than preventing, AIDS-virus infections. The frequency and kinetics of such escape will affect the ultimate usefulness of vaccines that are based on this approach.

Moreover, interpretation of the studies1,2 is complicated by the fact that the SHIV 89.6P challenge virus produces a course of infection that is very different from that typical of HIV-1 and SIV. Infection with HIV-1 and most pathogenic SIVs is generally characterized by the gradual, progressive destruction of CD4-expressing T cells. In contrast, infection with pathogenic SHIVs such as SHIV 89.6P typically results in the rapid, systemic and irreversible destruction of nearly all of these cells3.

SHIVs that include the HIV-1 envelope gene were originally developed to evaluate how well monkeys are protected by vaccines that target the HIV-1 envelope glycoproteins8. But pathogenic SHIVs, particularly SHIV 89.6P, have been more widely used of late, even for studies in which the HIV-1 envelope glycoprotein is not a vaccine component. Although the vaccines have not completely blocked infection, they have generally lowered the post-peak viral load and prevented the rapid, almost complete destruction of CD4-expressing T cells that is characteristic of SHIV 89.6P infection4,5. This alteration of such a dramatic consequence of infection has been interpreted as evidence of an important advance in the development of an effective HIV-1 vaccine.

However, accumulating data indicate that it is comparatively easy to prevent the loss of CD4-expressing cells and to bring about the sustained control of infection through partial inhibition of SHIV replication early in an infection, whether by vaccination4,5,9, passive immunotherapy10, short-term drug treatment11, or limiting the amount of challenge virus used12. Moreover, although rigorous head-to-head comparative vaccine studies have not yet been reported, the results obtained with pathogenic SHIV challenges contrast markedly with those obtained when pathogenic SIV strains are used to challenge similarly vaccinated monkeys. In such studies, sustained control of SIV infection and its consequences has rarely been achieved9,13,14. So we should be cautious when using alteration of the unusual course of disease caused by pathogenic SHIVs, rather than prevention of infection, as a major test of vaccine efficacy.

All in all, the two studies1,2 provide both an encouraging step along the road towards an effective AIDS vaccine and a sobering reminder of how long and difficult the road is likely to be. They do provide some direction along the road, suggesting that successful vaccines will need to induce immune responses that are both durable and broad, recruiting both antibodies and cells15. Unfortunately for sisyphean AIDS-vaccine researchers, it is likely that the signposts on this road will continue to indicate an uphill grade for the foreseeable future.