It is tempting to say that the story of how Elan/Wyeth’s antibody therapy fared at the 11th International Conference on Alzheimer’s Disease, held last month in Chicago, is one of the stranger tales in present-day AD drug development. (Then, again, weren’t all the recent ones a little strange? Take Alzhemed’s inglorious demise and subsequent resurrection as a nutraceutical, or Flurizan’s sudden death just weeks after its sponsor had announced a European commercialization agreement.)
Bapineuzumab’s story is different, of course. This humanized monoclonal anti-Aβ antibody is very much alive as an experimental treatment. Four separate Phase 3 trials in the U.S., Canada, and Europe are presently enrolling up to 4,100 Alzheimer disease patients. Elan/Wyeth are running an open-label extension of the present Phase 2 trial and are gearing up to test subcutaneous bapineuzumab delivery in a Phase 2 trial. But like the previous two anti-amyloid drugs, bapineuzumab rode to ICAD on high expectations, and its luster has dimmed somewhat after the company’s presentation there. This is not only because investors promptly dumped Elan stock but also because scientists felt that the presentation could have been more straightforward.
What happened? The company presented an overall negative result balanced against a positive effect on a subgroup of patients who don’t carry the AD risk allele ApoE4. Many listeners did not buy it but interpreted the presentation as having massaged the data. In conversations with scientists, most of whom spoke privately, reactions ranged from “they stratified until they got the result they wanted” to “the effect on the ApoE4 negatives makes sense; it’s reassuring,” to “told you so, anti-amyloid drugs lead nowhere” to “I am cautiously optimistic.” Several agreed, however, that the way in which Elan/Wyeth sought to portray the Phase 2 data in the best possible light backfired. In essence, emphasizing a post-hoc pharmacogenomic interpretation shortchanged the trial’s overall decent results, they said. Here’s the lowdown:
At ICAD, Sid Gilman of the University of Michigan, Ann Arbor, presented a more detailed version of the previous top-line data Elan had released on June 16 (see ARF related news story). In this talk, Gilman said that the trial had randomized 234 patients to receive six infusions, one every 13 weeks, of one of four different treatment doses or placebo. He reported that 26 percent of patients dropped out of the treatment groups versus 21 percent in the placebo group.
As a Phase 2, this was first and foremost a safety trial. The safety data Gilman showed received less attention than the efficacy data summarized below, but several scientists noted afterward that some of the reported side effects might bear watching. The most widely discussed side effect is vasogenic edema. Gilman showed that 12 patients developed this, all in the treatment group. Ten occurred in ApoE4 carriers and two in non-carriers, though the trial had twice as many people with ApoE4 than without, meaning that this pharmacogenomic distinction works out to roughly 10/4 rather than 10/2 in this trial. The edemas showed up on MRI and all resolved. Most caused few clinical symptoms, but one patient needed steroid treatment. Six people dropped out of the study; the other six resumed treatment. It is unclear at present how troublesome these edemas really are, and whether they may eventually be deemed an acceptable side effect for treating a deadly disease such as Alzheimer’s. This question generated animated discussion at ICAD and is the subject of ongoing research. Some scientists felt that vasogenic edemas may reflect the underlying mechanism of the drug, as Aβ peptide gets released from plaques, protein concentrations in the interstitial space and in capillary beds rise, osmotic pressure rises along with that, and clearance mechanisms, perhaps some involving ApoE, gradually dispose of the surplus protein. Others doubted whether such a therapy would get the FDA’s blessing.
Concern over vasogenic edema overshadowed the fact that Gilman’s presentation included additional side effects that were more common in the treated patients. The list of unwanted effects that occurred twice as often as in placebo, and in more than 5 percent of patients on active drug, included back pain, anxiety, vomiting, hypertension, weight loss, paranoia, skin lacerations, gait disturbance, and muscle spasms. Syncope and cataracts, among some other side effects, occurred in fewer than 5 percent of patients on active drug. Adverse events were generally mild or moderate and transient, Gilman said. Three men in treatment groups died. According to Gilman, one died of an aortic aneurysm, one succumbed to a fall and Alzheimer’s, and one to infection and Alzheimer’s. None of these deaths was attributed to study drug.
Moving on to efficacy signals, Gilman then briefly noted that the companies had decided to change the originally planned intent-to-treat analysis away from a slope analysis, which shows how the patients do at every time point, to a different analysis he called “MITT repeated measures model without assumption of linearity.” This mouthful essentially compares baseline to the final assessment at 78 weeks/19.5 months and incorporates the time points in between in ways that are not directly shown. That left many in the audience puzzled. To some, the decision to show the data this way reflects a biological argument that a plaque-clearing antibody, unlike a conventional drug, cannot be expected to show a linear effect over time as it might make a patient temporarily worse before it makes him/her better in the long run. To others, this change in the statistical model to a complex one they don’t fully understand implied that Elan/Wyeth might not want to show outcome data over all seven time points because those data looked bad. These scientists would have welcomed trend lines, i.e., graphs plotting the slope of outcome measures along the time axis, because that is how they have come to expect clinical trial data to be shown.
Gilman then showed efficacy data according to the modified analysis. In the total population, the trial posted a 2.3-point improvement on the ADAS-Cog, with a p value of 0.078. On the NTB battery, the improvement was 0.13 with a p value of 0.068. Thus, the trial failed to achieve statistical significance on pre-specified endpoints. Gilman stated this clearly and then moved right on to present a pharmacogenomic stratification of the data. But wait, not so fast! Commentators noted that this overall finding is important. “This is a respectable efficacy signal for a Phase 2 trial. It is a consistent effect, the trial is powered to detect it, and you cannot expect a larger effect from a trial of 230 patients,” Lon Schneider of the University of Southern California told this reporter. Other scientists agreed with this assessment.
In post-hoc analysis, Elan/Wyeth split the patients into those who received all six injections, i.e., the “completers,” versus those who missed one or more doses. Analyzed in this way, the patients who completed the trial (N = 78, ApoE4 carriers and non-carriers combined), showed a 4.3-point improvement over placebo on the ADAS-Cog with a p value of 0.003. “My gut feeling is that [bapineuzumab] is probably modestly effective. But my other gut feeling is that they are chasing their tail by emphasizing efficacy in the E4 non-carriers,” said Schneider.
What, then, is this pharmacogenomic result? Prior research suggests biologic differences between ApoE4 carriers and non-carriers, which might be expected to affect the course of an amyloid-removing therapeutic approach. ApoE4 carriers tend to get the disease earlier than people with ApoE3 or the rare allele ApoE2, and they develop more amyloid in their brain (Rowe et al., 2007). Numerous studies have implicated ApoE in amyloid deposition and removal, and a recent one showed that ApoE4 is less effective at clearing Aβ than the other two alleles (Jiang et al., 2008). Hence, Elan/Wyeth split trial participants into ApoE4 carriers and non-carriers, and then analyzed these groups separately. This created some imbalance between the groups because two-thirds of the patients in the trial happened to have an ApoE4 allele. At 78 weeks, the ApoE4 carriers showed a respectable 4.5-point improvement in ADAS-Cog on the lowest dose but a worsening at the highest dose. Results were inconsistent across the other outcome measures (DAD, NTB, CDR-SB). Some 70 ApoE carriers were in the four treatment groups combined. Of these ApoE4 carriers, a total of 42 across all treatment groups completed the study, i.e., about 10 per group on average. They showed a larger and more consistent benefit. ApoE non-carriers (N = 47) showed a benefit primarily with the second-lowest dose. The completers among this group numbered only 36 patients, i.e., on average less than 10 per dose comparison. But they had larger effects, some reaching a whopping 20 points on ADAS-Cog. (This was relative to an 11-point decline in the placebo group.)
Looking at the best result of all those different groupings, it seemed that a small subset of patients, i.e., the ApoE non-carriers who received the second-lowest of the four doses six times, responded truly well by 78 weeks. The presentation did not state how many people were in this group; it could have been 30, it could have been one. At the same time, the overall appearance of the various bar graphs of these subgroups was one of considerable variation across doses and across the different cognitive/clinical assessments. Subgroup analysis gets statistically weaker the smaller the group sizes become, opening it to criticism. “It could be noise,” Schneider cautioned.
Gilman did show slope curves for one subgroup, i.e., the ApoE4 non-carriers. They suggest that the curves diverged for ADAS-Cog after 36 weeks, on the NTB and CDR-SB right after baseline, but not until 63 weeks on the DAD.
Furthermore, the trial featured three kinds of biomarker, two imaging, and one fluid. Scanned at 71 weeks, neither brain volume nor ventricular volume (both measured by the boundary shift integral) changed in the total population, but again, stratification by ApoE appeared to yield a signal. ApoE carriers on bapineuzumab showed an increase in ventricular brain volume over placebo. Whether this is good or bad is unclear, Gilman said. It could reflect ongoing protein clearance or something else. ApoE4 non-carriers on bapineuzumab suffered less brain atrophy than those on placebo, and here, too, a slope diagram of diverging curves was shown. The CSF biomarker measurements, gleaned from spinal taps at baseline and 52 weeks in 20 antibody- and 15 placebo-treated patients, showed a trend toward lower phospho-tau in treated versus placebo with a p value of 0.056. CSF Aβ and total tau were the same between treated and placebo groups.
“This was a relatively small Phase 2 trial, and it was not realistic that a trial of this size with multiple doses would yield a definitive proof of efficacy. The trial was clearly a success, however, in demonstrating the feasibility of completing multiple IV infusions with intensive safety monitoring in AD patients,” Reisa Sperling of Brigham and Women’s Hospital in Boston, commented by e-mail. “There are some potential safety issues that will need to be followed closely in the larger trials. There may be some encouraging signals that we are on the right track with immunotherapy, but it is important not to over-interpret post-hoc analyses with small sample sizes. We will just have to wait and see."
It is notoriously difficult to extract a solid efficacy signal from Phase 2 trials for AD drugs (see ARF ICAD story). Interestingly, Elan/Wyeth’s immunotherapy competitor Eli Lilly and Company dealt with this dilemma in the opposite way. Rather than slice the data to find efficacy in Phase 2, Lilly placed its bets on biomarker characterization alone and decided to move to Phase 3 just based on that and safety (see upcoming ICAD story). It’s fitting, perhaps, to end with a philosophical remark from a leading basic researcher in the field. “We are at a turning point with regard to developing therapeutics. Everybody is very nervous about the outcome of current anti-amyloid trials because so much is at stake,” said Bart de Strooper at K.U. Leuven and the Flanders Institute for Biotechnology (VIB) in Belgium. “The basis for this test of the amyloid hypothesis is strong, but clearly amyloid is not everything that is going wrong in AD. We should interpret any negative results in this context and try to learn from it. We also need to think about prevention. We need to develop safer medicines that we can test for prevention.”—Gabrielle Strobel.