Unravelling the mystery of the vaccine trials

There have been thousands of small vaccine trials in the history of HIV – the first one took place in 1986 – but only six large efficacy trials, and the results of these have been puzzling. Many presentations at the 13th AIDS Vaccine conference were dedicated to understanding what went on in them and how we can use their findings to guide better vaccine design.

The efficacy trials

The history of HIV vaccine research has been one of both disappointment and one apparent success.

The first trials were the two VAX trials, which used AIDSVAX, a protein from the envelope (surface) of HIV, gp120 (which forms the ‘knobs’ on HIV that act as its cell-contacting mechanism) with the aim of stimulating the humoral branch of the immune system to make antibodies against HIV. The protein did stimulate antibody responses in both trials, but in neither one were they effective in reducing the rate of HIV infection in participants. It became clear that HIV can quickly mutate to shrug off antibodies formed against it.

The two second-generation trials were STEP and Phambili. These used a quite different form of vaccine, HIV genes (respectively from the subtypes most common in the USA and southern Africa) carried into cells inside the shell of an infectious but non-reproducing adenovirus, a common cold virus, called Ad5. Three shots of this Ad5 vector vaccine were given. This method, it was hoped, would stimulate the other, cellular branch of the adaptive immune system to produce CD8 cells that would kill off HIV-infected cells. STEP didn't work and Phambili, which had started later, was stopped when the STEP results were announced.

The failure of these trials was a low point in vaccine development, as not only did they fail to stop HIV, but there was some evidence at the point the trials were unblinded that they may have increased the risk of HIV infection to participants who had pre-occurring immunity to Ad5, and also to uncircumcised men.

Meanwhile two other vaccine trials had started. The first one, RV144, combined the methods seen in the first three studies. It ferried HIV genes into cells inside the shell of a different virus, a canarypox virus called MVA, and then finished off with two doses of the AIDSVAX gp120 vaccine. Expectations for it were not high as it involved one component, AIDSVAX, that had already shown no efficacy, and one group of vaccine researchers even wrote a letter saying its scientific rationale was weak and it should be discontinued. So it was a surprise when, in 2009, it was announced that it reduced HIV infections in recipients by 31%, compared to placebo.         

The one further efficacy trial was another disappointment, however. HVTN 505 consisted of two shots of ‘naked’ HIV DNA followed by a shot of HIV DNA inside an Ad5 shell – the same vector as in STEP and Phambili. It was conducted in the US with gay men and transgender women and because of the STEP trial result, participants had to have no pre-existing immunity to Ad5 and to be circumcised. The study was stopped in April 2013 when the vaccine was clearly having no efficacy and full results were published just before the AIDS Vaccine conference opened, on 7 October.

What worked in RV144?

Merlin Robb of the US Military Research Program presented findings on the correlates of protection (factors associated with efficacy) in RV144. Initially it was hard to find anything that could explain the vaccine’s apparent efficacy: it did not produce a CD8 cell response, and it produced fewer of the supposedly essential broadly neutralising antibodies to HIV – antibodies that block infection by a wide variety of HIV viruses – than AIDSVAX. It did produce weaker binding antibodies, but these, it had been supposed, were too weak and too specific to have a protective effect.

However, it became clear that the vaccine generated two types of binding antibodies to HIV. One type was called immunoglobulin A (IgA) – a type of antibody that mainly lurks in mucous membranes. High levels of anti-HIV IgA had an unhelpful effect: they were associated with increased susceptibility to HIV. The other type was immunoglobulin G (IgG) – a type that mainly circulates in the blood. High levels of IgG were helpful – they were associated with protection against HIV.

It had been thought that antibodies protective against HIV would most likely be formed against a region of the gp120 protein called the V3 loop, but the helpful antibodies were formed in reaction to two other parts, the V1 and V2 loops – indeed antibodies to V3 seemed to be associated with risk.  The participants with the highest ratio of IgG to IgA had a 60% reduced risk of HIV infection.

If RV144 had been stopped after a year, 60% would also have been the efficacy observed: however the antibody protection appeared to fade quite quickly and by 3.5 years it was the observed 31%.

Why did IgA and IgG matter? The first blocked, and the second facilitated, a process called ADCC – antibody-dependent cellular cytotoxicity. Antibodies are Y-shaped molecules and it is the two short branches of the Y that are their ‘keys’ – they are produced in an amazing variety of shapes, a few of which are able to stick to specific components of foreign invaders in the body and proliferate when they encounter them. However, the stem of the Y also has a purpose in some antibodies. When the short branches encounter the foreign material they are keys to, the long stem can send out signals to other parts of the immune system and alert them to come over and kill the foreign invaders.

In this case, the anti-HIV IgGs sent out their signal to the third and oldest part of the immune system, the innate immune system, an unspecific but fast force of cells and chemicals, including the powerful natural killer cells. These rushed to infection sites in response to the IgG antibodies’ ‘distress call’ and quelled infection. In contrast, the IgAs specifically blocked off natural killer cells from reacting to the invading HIV.

The researchers found that anti-V1 and V2 IgG antibodies in the RV144 recipients were reactive to a broad variety of HIV viruses whereas IgG antibodies from VAX trial volunteers were reactive to a much narrower group and belonged to a different subgroup that did not send a strong ADCC signal.

Why didn’t HVTN 505 work?

The above findings also seem to explain why HVTN 505 vaccine didn’t work. It produced high levels of antibodies to the V3 loop and virtually none to V1 and V2. Furthermore, while it did produce high levels of IgG antibodies, most were not in reaction to any part of the gp120 protein. Instead, 90% were antibodies to gp41, the ‘stem’ of the knobs on the surface of HIV, and the part which actually fuses with the cell.

These antibodies were very non-specific: in fact they reacted to many common bacteria, including the ones most people carry around in their gut. The 10% of antibodies to gp120 were mainly IgA antibodies. What this meant was that while the RV144 vaccine was capable of stimulating an immune reaction that could neutralise 20 to 90% of a selection of less exotic and immune-resistant HIV viruses in the test tube (the type that normally get transmitted), the HVTN 505 immune response could only neutralise 3 to 28% of the same viruses.

Taken together, this meant that the antibody response generated by HVTN 505 actually disabled the ADCC process that might have helped to protect against HIV.

Did the Ad5 vaccines increase susceptibility to HIV?

One of the most worrying questions about the vaccine trials is whether some of them really did increase participants’ susceptibility to HIV. A new meta-analysis by Peter Gilbert of the Fred Hutchinson Cancer Research Center in Seattle, USA attempted to shed some light on this. It looked at all 8500 participants in the STEP, Phambili and HVTN 505 trials and established that, over the whole group, participants who were given a vaccine were 33% more likely to become infected with HIV than placebo recipients. However, the increased risk in HVTN 505 was only 9% and this was not remotely statistically significant (p = 0.7).

There was a 41% raised risk of acquiring HIV for vaccine recipients over placebo recipients in STEP and Phambili taken together. This was statistically significant (p = 0.005). However, if STEP was taken by itself, the raised risk for all participants was only 22% and this was not significant. So in terms of whole trials, it was Phambili, where vaccine recipients were 74% more likely to acquire HIV than placebo recipients, which remained significant.

Based on the results at the end of the blinded phase of vaccine trials, when participants did not know whether they were on vaccine or placebo (which is what has been published before) – pre-existing adenovirus immunity, especially in uncircumcised men, seemed to be implicated (being male and having ad5 immunity raised your risk by 68% in both trials taken together).

Both trials, however, continued after unblinding: participants were followed for three years after they found out whether or not they had actually been on the vaccine.

During the unblinded phase, a very odd thing started to happen: while the increased risk of vaccine over placebo remained, especially in Phambili, and the risk for men over women persisted in that trial too, the risks associated with adenovirus immunity and circumcision actually reversed. People without immunity to Ad5 and circumcised men became more likely to acquire HIV than Ad5-immune and uncircumcised participants.In fact this actually started happening before unblinding, after 18 months from the first vaccine or placebo shot.

So it appeared that, whatever was causing the excess infections in vaccine recipients, it either wasn’t Ad5 immunity after all, or the Ad5 risk only contributed to the overall risk in the first 18 months after people were given the vaccine, for some unknown biological reason. This doesn't explain why the risk reversed, though – and it would be very unusual for a vaccine to suddenly start producing a biological risk that hadn’t happened before as long as 18 months after the vaccine was given.

It may have been behaviour in the unblinded phase. Trials are blinded so that participants'  knowledge of whether they are on the treatment or the placebo does not influence their behaviour. As a result, it is much more difficult to tease out influences on outcomes in the unblinded phase.

The continuing difference in infection rates in the unblinded phase of STEP and more markedly in Phambili was characterised by another odd thing: the difference later on was not so much driven by vaccine recipients acquiring more HIV so much as placebo recipients ceasing to acquire it. In Phambili in particular, in the last year of follow-up no infections were seen in the placebo group.

What was happening? Did placebo recipients reduce their risk behaviour after unblinding, or vaccine recipients increase theirs? Conversely, what about people who dropped out of the study? Somewhat more placebo recipients dropped out of the study after it had been unblinded than vaccine recipients, a common phenomenon in vaccine trials. What if the high-risk placebo recipients had been the ones that dropped out? That would also keep infections in the remaining placebo group down. Conversely, high-risk vaccine recipients might have been more likely to stay in the study.

Glenda Gray, Principal Investigator of the Phambili trial, said that self-reported behaviour did not differ between vaccine and placebo recipients and that even if there was differential drop-out in high-risk placebo recipients, it  would only make a single infection's difference to the Phambili result.

Peter Gilbert said that based on the drop-out rates in all three trials, there would have to be 33% more infections in placebo drop-outs than those who stayed in the trials, and 28% fewer in vaccine drop-outs than in those who stayed, to even out the vaccine/placebo difference seen during the unblinded phases of the trials.

For now, then, the explanation for the higher rates of infection seen in some groups in the STEP and Phambili trials remains mysterious.