A post-hoc analysis of an HIV vaccine trial that was recently stopped because it made no difference to the rate of infection has uncovered a more alarming possibility: the vaccine, though it could not possibly have caused HIV infection in itself, may have made some participants more vulnerable to HIV. Final figures show that nearly 50% more people given the vaccine caught HIV than people given a placebo.
A round table of experts, convened at the HIV Vaccines Trial Network (HVTN) Conference on 7th November to analyse the STEP trial, heard that the higher rates of infection in vaccine recipients were not statistically significant: in other words, they could be a random fluctuation due to chance alone.
But the higher rates of HIV infection in at least some of the volunteers given the vaccine are nonetheless more likely to be real than random, the analysis suggests.
If there is a real cause, we don’t yet know whether it lies in the vaccine or is due to behavioural or demographic differences in the trial population.
The STEP trial
Two large trials of the Merck V520 vaccine were stopped in September (see this report). It was eventually found that in one of them, the STEP trial, there had been 49 infections in patients receiving the vaccine and 33 in those receiving placebo. Soon afterwards the Phambili trial of a similar vaccine in South Africa was stopped.
In the STEP trial 1,500 patients at high risk of HIV in the Americas and Australia were originally recruited, starting in December 2004. They were largely gay men and female sex workers. Sixty-two per cent were male and the average age was 29.
This was a population with high levels of risk for HIV acquisition through sex. Half of the women and 21% of the men had had at least 20 sex partners in the previous six months; 75% of the women and nearly 60% of the men had had unprotected anal or vaginal sex in the same period; and one in five of the women and one in three of the men had had sex with someone whose HIV status they didn’t know.
There was only one infection in female volunteers in STEP. While this looks like a startling finding at first, it has not been highlighted in the analysis, and probably reflects two facts; firstly, women lagged behind men in trial recruitment and didn’t have so much time to catch HIV; secondly, it illustrates that anal sex transmits HIV better. The HIV infection figures below are of the men in the study and exclude the one infected woman.
The V520 vaccine consisted of sequences from three HIV genes – the gag, pol and nef genes – wrapped inside the shell of the Ad5 adenovirus, a virus that only causes common cold symptoms, if it causes any at all. The idea behind the vaccine was to infect cells with the adenovirus, which would then, as infected cells do, display epitopes – sections of HIV viral protein – on their surface. These in theory would have stimulated an anti-HIV cellular immune response which, if a real HIV virus came along, would have at least contained and possibly stopped infection.
The original news released on September 21st, then, was already doubly disappointing: not only was there no difference in infection rates (at that time, with incomplete figures, it was known that 24 vaccine recipients and 21 placebo recipients had been infected), but the HIV viral load in those infected was unchanged by the vaccine.
It had been hoped that at the very least the CD8 response created by the vaccine – a real and measurable effect that had already been seen in previous studies – would moderate the course of infection. The fact that it did not do so has caused researchers to ask, in the words of HVTN’s Judy McElrath, “Was the concept of a T-cell-based HIV-1 vaccine clinically not valid, or is the lack of efficacy specific to the Ad5 vaccine candidate?”
The role of pre-existing immunity
However a post-hoc analysis of the trial results brought worse news. An apparently slight excess of HIV infection in those receiving vaccine started looking more significant when the results were stratified according to the level of pre-existing immunity volunteers had to the Ad5 virus.
Originally, people with high levels of pre-existing immunity to the Ad5 adenovirus had been excluded from the STEP trial, but a decision to double the trial’s size had been taken in August 2005 when reasonable immune responses were observed in this group. The trial had reached its target of 3,000 volunteers by March 2007, and the analyses that follow are based on 1,503 volunteers who received at least one shot of vaccine or placebo and 1,363 volunteers who received at least two shots (the full schedule was three injections at zero, one and six months).
Volunteers were stratified into four groups according to ad5 antibody levels in their blood:
- In those with the lowest antibody levels, 4% of volunteers caught HIV regardless of whether they had vaccine or placebo.
- In those with medium-low levels, 4.4% of those given vaccine caught HIV but only 2.2% given placebo, and in those with medium-high levels, 6.1% given vaccine caught HIV and 3.0% given placebo, so twice as many given the vaccine caught HIV in both of these groups.
- In those with the highest levels of ad5 antibodies, 4.4% given vaccine versus 1.2% given placebo became infected - 3.5 times as many, though in this case numbers were very small (seven vaccinees and two on placebo).
In all cases the confidence intervals overlapped, and no result in any category was near statistical significance. But the direction of change was consistent – and alarming.
The classical definition of ‘statistical significance’ is that there is a less than one-in-20 probability that the effects observed could be due to chance. The round table heard that there was a one-in-17 probability that the higher rates of HIV infection seen in male volunteers who had high levels of pre-existing ad5 immunity could have been a chance effect, as opposed to an effect of the vaccine. In this group, 21 volunteers given vaccine acquired HIV versus nine given placebo. Over the volunteer group as a whole the chance of this happening was one in 6.5.
Another possibility to be considered is that high ad5 immunity conferred some kind of pre-existing resistance to HIV infection in the placebo group – which was abrogated by the vaccine. If this is taken into account it means that there is some characteristic in both the vaccine and the placebo groups that has to be taken into account, and this then creates a “doubled sided” possibility- meaning that the probability that observed effects are due to chance is halved.
This means that there is then one chance in 13 that the difference in infection rates observed in those with high ad5 immunity is a random effect, and one in 35 if only the men are considered – within the boundary of statistical significance.
Steve Self of HVTN said that classical statistical significance test was useful when testing to see if something worked. But when evaluating a possibly harmful effect, “probability values much smaller than the usual threshold values still provide some indication of strength of evidence.”
Saying that chance is the explanation for differences in infection rates, he added, was only an “explanation of exclusion” – in other words, it excludes other variables that researchers can measure but doesn’t exclude ones they can’t measure or haven’t thought of.
Keith Gottesdiener of Merck added: "This difference is clinically important for at least one subgroup, the high Ad5. I don't really need any statistics to make a declaration that it's an important factor to take into consideration."
Behavioural and demographic differences
What differences were there between the volunteer groups? There were none between the vaccine and placebo groups: this was a very well-randomised trial. But there were significant differences between those with low and high levels of ad5 immunity. People with higher levels of ad5 immunity, recruited in the second wave of enrolment, were twice as likely to be female, nearly half as likely to be white and gay, and somewhat younger. We know there is already a gender bias in the infections, but was this due to gender, because more women were black, or for some other reason?
One thing can be ruled out: there were no behavioural differences between those receiving placebo and those receiving vaccine. As this was a double-blinded trial, those receiving vaccine (or those administering it) would have to guess correctly what they had received, and then change their behaviour accordingly. Susan Buchbinder of the University of California in San Francisco said there was no evidence of either of these things happening.
In terms of behavioural differences between those with low and high levels of pre-existing ad5 immunity, the few differences there were, such as lower levels of unprotected anal sex amongst men with high ad5 immunity, might have been expected to protect them from HIV, not expose them to it.
The men with high pre-existing immunity to ad5 were also half as likely to be circumcised. This is a potentially significant finding. Nearly two-thirds of the men in the trial (64%) were circumcised. Excluding men with uncertain circumcision status, there were 26 infections each in circumcised vaccine and placebo recipients, but in uncircumcised recipients only six receiving placebo caught HIV versus 26 receiving vaccine – the latter figure representing a fourfold greater risk of HIV acquisition for vaccine over placebo, and a 56% greater risk for uncircumcised over circumcised. The lower figure in men receiving placebo is also intriguing. But all these effects need to be further investigated and, once again, no single difference is statistically significant.
Were there any differences in immune response between vaccinees who contracted HIV and ones who didn’t? There is a suggestion this may be the case.
As expected, volunteers with high levels of pre-existing ad5 immunity had lower immune responses to the HIV vaccine: about 72% of low-immunity volunteers and 54% of high-immunity volunteers had a significant immune response to any one of the three HIV proteins.
But in high-immunity patients who caught HIV, only 43% had an immune response. However it’s difficult to see any consistent pattern in these immune responses – the inter-individual variability is far higher than that between groups and again, no one difference has statistical significance.
Another possibility being looked at is that the immune responses generated to HIV were too narrow to deal with most viruses. Follow-up studies will look in much more detail at the immune responses generated in vaccine recipients and why they seemed to have no effect on HIV.
It may not be participants’ immune responses to the vaccine, but their pre-existing immune characteristics, that made the difference. One hypothesis concerns HIV co-receptor levels.
Both vaccine and placebo recipients had higher levels of CD4 cells expressing the CCR5 co-receptor for HIV on their surface, and thus presenting themselves as targets for infection, if they had high levels of ad5 immunity.
About a third of vaccinees with low ad5 immunity had activated CCR5-expressing CD4 cells compared with 40-45% in those with high ad5 immunity. There were no differences between those who caught HIV and ones who didn’t, except among vaccinees with high ad5 immunity – but those investigated (18 people) actually had lower levels of activated CD4 cells expressing CCR5 than those who weren’t infected – the opposite of what you might expect.
The role of CCR5 is still being pursued as a hypothesis, though, because these are levels of CCR5-expressing cells in the blood, and it’s the ones at the site of infection – the genital tract and rectum – that would be important.
Certainly CD4 and CD8 cells that responded to the vaccine did get into semen – in a group of twelve male vaccinees studied, four had HIV-specific T-cells in their semen 26 weeks after vaccination.
Could the vaccine have stimulated the immune system in an unhelpful way – one that sensitised, rather than immunised, people to HIV?
If this turns out to be the case, it would be of great concern, because it could mean that recipients of the V520 vaccine could be super-receptive to HIV for a long time – possibly for the rest of their life. Immune responses to the vaccine do appear to be long-lasting and had faded only slightly in a group of early vaccinees two years after injection.
Unblinding the trials, and the aftermath
It’s for this reason that there is some urgency directed to the process of ‘unblinding’ the STEP trial – informing participants whether they in fact received the vaccine or the placebo. This isn’t easy to do in a hurry as participant records are anonymised precisely to avoid accidental or unauthorised unblinding.
Glenda Gray, Phambili’s national Lead Investigator in South Africa, described the process of informing the 800 volunteers in the Phambili trial at the HVTN meeting – a process that took two weeks to complete. Initially some of the STEP investigators wanted to keep their trial blinded to get ‘clean’ data but after several days of debate, a decision to unblind the trial was taken and is being started now.
Given that the STEP trial failure is one of the most significant setbacks for years in the field of prevention research and that controversy has surrounded other HIV prevention trials, press reaction has been relatively muted. Glenda Gray commented that there had been one inaccurate report in the Washington Post but in South Africa, where controversy might have been expected, the results were overshadowed by South Africa’s victory in the rugby World Cup.
South Africa’s Health Minister, Manto Tshabalala-Msimang, announced that future HIV vaccine trials in South Africa have been put on hold, but given that most forthcoming ones of any size use similar vector technology to the Merck vaccine, a moratorium on vaccine trials till we understand what failed to work in the STEP trial makes sense.
The STEP trial failure concludes a year of disappointing results in prevention trials, with several other trials, including ones of the female diaphragm and of the microbicide cellulose sulphate, being stopped due to lack of efficacy. In the CONRAD cellulose sulphate trial there were also more infections in the intervention than in the control arm of the trial, though again this was not statistically significant – see this report.
“The Step Trial is not the first one that has shown this trend,” Mitchell Warren of the AIDS Vaccine Advocacy Coalition told aidsmap. “These are figures that trend in an unpleasant direction.”
He agreed that it would be a good idea to take stock and press the pause button on forthcoming efficacy trials of prevention technologies until better correlates of protection had been found. “We need to create a very high standard before investing the kind of resources we need in another efficacy trial. I don’t know how high that bar should be: but we need to put one there.”
At the same time he does not believe that the STEP trial in itself should be seen as a failure. “It essentially took three years to get an answer for a vaccine efficacy trial, which is fast for one this size, and is a tribute to its design,” he says. “We got an answer. It was one we don’t like, but it was an answer.”
All the presentations from the HVTN session on the STEP Study can be downloaded from www.hvtn.org/science/1107.html