This is Part 2 of a seven-part series. See also Part 1, Part 3, Part 4, Part 5, Part 6, and Part 7. Read a PDF of the entire series.
9 November 2012. Held 29-31 October in the Principality of Monaco, the 5th Clinical Trials in Alzheimer’s Disease conference offered new data from ongoing analyses of the Phase 3 programs on solanezumab and bapineuzumab. Both programs missed their primary endpoints. The news added up to a curious situation where, at present, it looks as if the biomarkers that are showing predictable and strongly convergent trajectories across natural history studies are not, so far, behaving as expected in therapeutic trials. “We see a dissociation of biomarker and clinical outcome,” said Reisa Sperling of Brigham and Women’s Hospital in Boston. Solanezumab slowed cognitive decline without affecting CSF markers of neurodegeneration. Bapineuzumab did improve those downstream markers, but flopped clinically. On volumetric MRI, both antibodies delivered a head-scratcher that is clouding this marker’s prospects as a potential surrogate and focusing renewed interest on more sensitive cognitive outcome measures. It is unclear at this point whether any of the known biomarkers are theragnostic, i.e., track or predict a drug effect. “We need better markers of synaptic response to fill this gap,” Sperling added.
When considering the data’s nuances—and there was a lot of nuance at CTAD—this disconnect begins to blur a bit, and the data may make more sense. Solanezumab mobilized soluble Aβ in plasma and appeared to nudge CSF concentrations. The finding recalled earlier studies with this antibody and led some attendees to wonder whether a peripheral sink effect might have been at play. Bapineuzumab, on the other hand, appears with more analysis to perhaps have had a hint of functional benefit in the mild subgroup after all, but that came at the cost of more white matter changes than was initially thought. Whatever benefit there was appeared to be disease modifying, not symptomatic. At the same time, a growing number of researchers are downplaying this once-prominent issue as a distinction without a difference, joining an argument one presenter has been making for some time (Doody, 2008). Confused? You are not alone. At CTAD, the program featured five talks by Sperling, Paul Aisen, Rachelle Doody, Nick Fox, and Steve Salloway, as well as a panel discussion to grapple with the data. Small group conversations could be overheard outside the auditorium throughout the conference, and no doubt continued over dinners and nightcaps. Read on for summaries of the data presentations and discussions.
First, a word about how information on this story is coming out. CTAD continued a data trickle that will go on at least through next spring. After the companies disclosed topline news last summer that their programs had missed primary endpoints, the actual results have been presented in bits and pieces at conferences last September, October, and most recently at CTAD. Yet more is to come at the AAN meeting in San Diego in March 2013. Some researchers bemoaned this gradual dissemination. Others were further puzzled that Lilly released its solanezumab data to investors but not to scientists on 9 October on the day of a major neurology conference in Boston, where an independent academic analysis of Lilly’s data was first presented (see ARF related conference story).
When asked about this, scientists at Janssen AIP, Lilly, and at various academic institutions pointed to the multiple forces that drive when data are released. All insisted that the drip-drip-drip was not strategic but a function of huge datasets being continually analyzed for successive questions. “As soon as we have new results, we present them at the next opportunity,” said Salloway of Brown University, Providence, Rhode Island. Companies are required to release topline information as soon as they get it, hence, the summer press releases. The Securities and Exchange Commission requires disclosure to shareholders, hence, Lilly’s posting of slides during the October investor call. Separate from that, Lilly and the Alzheimer’s Disease Cooperative Study have negotiated a unique agreement whereby the Alzheimer's Disease Cooperative Study (ADCS) analyzes Lilly’s raw data independently. Initiated by Richard Mohs and Eric Siemers at Lilly, this serves the field’s collective desire to learn as much as possible from these enormous datasets and keep a version in the public domain. It also was a response to a demand by journal editors, said Paul Aisen of the University of California, San Diego, who leads the ADCS. The agreement stipulates that the ADCS presents its analysis at conferences and takes the lead on publication. This means, in effect, that the ADCS communicates Lilly's Phase 3 results through its lens with the scientific community, while Lilly uses its internal analysis for guidance from the Food and Drug Administration about what to do next.
Solanezumab: The New Data
Before presenting results, Aisen explained how the ADCS handled this industry dataset. Three ADCS statisticians—Ron Thomas, Rema Raman, and Mike Donahue—each conducted his or her own analyses of Lilly’s raw solanezumab Phase 3 data. Each coded, input missing data, and programmed the models to reveal the implication of each statistical decision made along the way. Having been collected by Lilly through contract research organizations, with different forms and procedures than the ADCS has developed over time for its own trials, the solanezumab data needed to be transferred, and in places restructured, to fit the ADCS system. The statisticians handled out-of-window visits, missing scores, and at times had to deviate from Lilly’s statistical analysis plan (SAP). Why is this important? Because when the effect of a drug is small, as is true here, each decision the statistician has to make along the way can affect the p value of statistical significance, Aisen told Alzforum. At CTAD, Donahue gave a whole talk just on how ADCS statisticians treat missing clinical trial data. “We have analyzed Phase 3 data going back to tacrine,” Aisen said, “We have our own approach to imputation and analysis. We opted to follow the spirit of Lilly’s, yet apply some of the specific approaches we have gleaned from our experience.”
Despite the statistical challenges, a coherent picture of a cognitive benefit emerged, Aisen said. The effect is very small but statistically robust in pooled analyses, holding up across variations of the data. The benefit amounts to a 30 to 35 percent slowing of cognitive decline in the mild subgroup. This number fits with the data Lilly released to investors on 8 October, which claimed a 34 percent slowing. The ADCS analysis showed 1.5 points less decline in ADAS-cog 11, and 1 point less in MMSE for solanezumab than placebo in patients with mild AD. The ADL functional score showed 1.5 points less decline in the mild group, though this result has less statistical confidence. Neither the neuropsychiatric inventory nor the CDR Sum of Boxes showed any difference.
On biomarkers, the main result came from plasma Aβ. Solanezumab caused a sharp and sustained increase that is consistent with the antibody’s tight binding to monomeric Aβ. The CSF showed a more complex picture. The concentration of both Aβ40 and 42 rose in the mild to moderate AD group. The much smaller pool of free Aβ40 showed a decrease in treated patients, while free Aβ42 did not. The CSF numbers are based on smaller group sizes (13 to 66) than those for plasma (700s) or MRI (600s to 900s). To Aisen, the rise in plasma and CSF levels of total Aβ constitutes target engagement; many others agreed.
Florbetapir scans in the solanezumab PET imaging substudy showed no significant change between the groups. The mild subgroup had a trend toward less amyloid in the treated group, but it missed statistical significance. There was no effect at all on CSF tau or phospho-tau overall or in any of the subgroups. Whole brain volume was the same between groups. On hippocampal volume, too, the analysis showed no overall treatment effect. Of the subgroups the ADCS analyzed separately, one looking at people with confirmed amyloid positivity showed a trend. This is merely a hint in a subgroup of a subgroup. Still, its direction was toward more shrinkage in the treated group (for more on MRI atrophy, see below).
Solanezumab’s clinical effect may be related to Aβ reduction at synapses as opposed to protection of cells, said Rachelle Doody of Baylor College of Medicine in Houston, Texas. Other scientists at CTAD agreed with this assessment. For example, David Morgan of the USF Health Byrd Alzheimer's Institute in Tampa, Florida, noted that the solanezumab data jibe with what is known from mice. APP mouse models of amyloid deposition have cognitive deficits independent of neurodegeneration, tau pathology, or atrophy (e.g., Kotilinek et al., 2002). The antibody used to engineer solanezumab, m266, was reported to restore memory in mice without a measurable effect on amyloid (Dodart et al., 2002). A large and rapid increase in plasma Aβ soon after m266 injection is known from mice (Demattos et al., 2002). Lars Lannfelt of Uppsala University, Sweden, said that solanezumab prolongs the half-life of Aβ in plasma and delays its degradation in the liver and kidneys. This might explain the sustained elevation in plasma, though how much of this plasma Aβ might come from brain as opposed to from peripheral cells was not shown.
In toto, the Phase 3 solanezumab data support a small cognitive benefit and may call for a confirmatory study in mild AD, Aisen said. Researchers hope that studies in prodromal or preclinical AD (see ARF related news story on DIAN) will yield a larger effect. At CTAD, speculation was rampant about whether these data alone might sway regulators to approve the therapy.
Bapineuzumab: The New Data
Steve Salloway presented a prespecified subgroup analysis of the bapineuzumab patients with mild AD. It differed from a previous analysis in that it used a cutoff of 20, not 21, on the MMSE, because 20 is what the solanezumab analysis used. “We thought it would be informative to see how the data compare,” Salloway told Alzforum. On most measures, this analysis, like the first one, found no difference between antibody and placebo for either the ApoE-carrying or non-carrying patients. On one, however, it sighted a bit of a signal, that is, a statistically significant treatment effect on the DAD functional measure. Non-carriers showed a benefit of about four points on both doses, while the pooled analysis of carriers and non-carriers showed a similar benefit only for the higher dose. “We are not making very much of this effect,” Salloway said, and Janssen scientists reiterated that clinical development of intravenous bapineuzumab has ended. Also new in Salloway’s presentation was the analysis of the secondary clinical endpoint consisting of the CDR-SB, the NTB and the MMSE: no differences in the total study population in either study; subgroup analysis is ongoing.
Most of the new bapineuzumab data at CTAD came on biomarkers. Nick Fox of University College London toggled between MRI scans taken of the same subject at baseline and weeks 19, 45, and 71 to show how the brain’s ventricles subtly expanded, and the cortex and left hippocampus ever so slightly contracted.
Can you see it? Slight ventricular expansion in bapi-treated subject. Image courtesy of Nick Fox/Janssen AIP, from CTAD website
Pooled analysis of both Phase 3 studies confirmed a mildly increased rate of brain volume loss in the bapineuzumab group at the higher dose, and in the mild subgroup at both doses. The difference came to about 2 milliliters per year for whole-brain volume, Fox said. In the left hippocampus, the difference was significant in non-carriers at the higher dose only. The ventricular effect was most pronounced, showing up as significant in the carriers, non-carriers, and pooled analyses.
Taken together, then, biomarker data on intravenous bapineuzumab suggest that the treatment nudged amyloid deposition but not soluble Aβ, clearly decreased CSF tau and phospho-tau (this was presented at a previous conference), and led the brain to shrink a bit faster. This means the antibody engaged its target, but it hardly helped at all clinically, Fox said.
If one includes the faint hint on solanezumab, this bapineuzumab MRI finding would be the third demonstration that anti-amyloid immunotherapy can subtly shrink the brain. (The first came in 2004 with AN1792, a discontinued Aβ42 vaccine; see ARF related news story). Why? The first instinct—that the shrinkage equals more neuronal death—is probably wrong because CSF tau went down with bapineuzumab. Is it amyloid removal? Maybe not alone, because the extent of amyloid removal was small. Does immunotherapy reduce the inflammation and debris associated with amyloid deposition? Does it change CSF absorption, or cause other fluid shifts in the brain? These notions warrant active study, Fox said. “We have to understand this.”
A propos fluid shifts, bapineuzumab has gained some notoriety for causing white matter changes, perhaps indicating leakage of fluid, proteins, or cells from vessel walls after amyloid drainage. Called amyloid-related imaging abnormality (ARIA), this finding has since been reported for other anti-amyloid therapies such as Bristol-Myers Squibb’s γ-secretase inhibitor BMS-708163/avagacestat and Roche’s antibody gantenerumab as well. The FDA requires close monitoring of this poorly understood finding, and the massive amounts of MRI data subsequently generated from this surveillance were the subject of a CTAD talk by Sperling.
There are two kinds of ARIA: E for vasogenic edema, indicating fluid seeping out of blood vessel walls or collecting in sulci, and H for hemosiderin deposit, indicating a microhemorrhage. At CTAD, Sperling presented preliminary results of a large study that determined the incidence of ARIA-E by giving a final read to every one of the 15,000 MRI scans taken during the bapineuzumab Phase 3 program. Two radiologists independently viewed the scans. Every time they found an ARIA-E, they pulled up that person’s prior scan to see if the abnormality had been there before. In this way, they called some 30 percent more cases than had been noted during the conducting of the trial.
The incidence of ARIA-E rises with ApoE status and with dose. About 1 percent of patients on placebo had ARIA-E, but so did one in five ApoE4 carriers on 0.5 mg/kg bapineuzumab. Among ApoE homozygote patients, it was one in three. Among ApoE4 non-carriers, the ARIA-E incidence at that dose was 5.6 percent, but the highest (and discontinued) dose of 2.0 mg/kg brought it to 19.9 percent. Most ARIA-E developed after the first three infusions.
Why were some ARIA-E incidences overlooked? Awareness of this side effect is new; indeed, it emerged during bapineuzumab’s development. Local radiologists missed some edemas on the first read. In many cases, the first ARIA-E was small, Salloway told Alzforum, and grew by the time of the subsequent scan. Midway through the study, a central read was instituted, but there, too, not all radiologists picked up every finding right away. Some site PIs did not acknowledge all findings, Sperling told the audience.
Sperling told Alzforum that the people with ARIA-E tend to be prone to ARIA-H as well, but this analysis is still underway.
How bad is ARIA? It is too early to say, Sperling said. A majority were asymptomatic, but about one in six were linked to symptoms such as headache, confusion, and cognitive changes. When patients were taken off the drug, the abnormality resolved over a period of months. A preliminary analysis of the Phase 3 final-read data shows that, clinically, ARIA-E appeared to make no difference on the course of a person’s Alzheimer’s disease. People with ARIA-E did not decline any faster or slower on the ADAS-cog or DAD than those without, though the scientists need to dig deeper into the data to check if differential dropout might have biased this finding. Some scientists believe that ARIA-E is a sign of amyloid clearance. Others noted that it will be important to watch what becomes of these patients in the longer term. In ApoE4 carriers but not non-carriers, the treatment group had more seizures than the placebo group.
Finally, on imaging, Sperling emphasized that 36 percent of ApoE4 non-carriers who joined the bapineuzumab amyloid PET substudy fell below the preset threshold for amyloid positivity. At CTAD, Sperling told the audience that a look at amyloid PET data of the solanezumab Phase 3 trials, which Lilly had shared with her, showed the same thing for the ApoE-negative participants in those trials. Both large Phase 3 programs appear to have enrolled a third of patients who arguably may not have Alzheimer’s disease but something else. Florbetapir is clinically approved to rule out Alzheimer’s. “This is a large number, and it is a problem,” Sperling said.
In discussion, researchers expressed a consensus that bapineuzumab failed over dose. It could not be safely given at high enough doses to make a difference. Others noted that the picture was still incomplete. For example, critical data on drug exposure antibody levels in plasma and CSF are still missing for both programs, as is a full analysis of the amyloid-positive versus negative participants. Several researchers cautioned that when a drug effect is as small as solanezumab’s, type 1 error, aka false positive, is always a concern, even when the statistics look internally consistent.
“These are valuable data, but I can’t overstate how disappointing that was to our patients and families,” said Fox. Other site leaders agreed, but noted that their patients felt they did contribute to science and want to participate again.
Some slide sets on Bapineuzumab, plus a one-hour audio recording on solanezumab, have been uploaded to the CTAD website. The recording features an attempt by Sperling to synthesize the data available thus far on both antibodies, as well as a subsequent panel discussion. Play it to get a sense of where the field stands at this point.—Gabrielle Strobel.
This is Part 2 of a seven-part series. See also Part 1, Part 3, Part 4, Part 5, Part 6, and Part 7. Read a PDF of the entire series.