In the aftermath of solanezumab’s failure to stem cognitive decline in mild Alzheimer’s disease, Eli Lilly and Company are closing down a Phase 3 trial in prodromal AD and laying off employees who had been hired in anticipation of bringing this anti-Aβ antibody to market. While Lilly retrenches, its senior scientists are also traveling to small gatherings around the country and abroad to share additional Phase 3 data with leaders in academia and industry. Their goal is to inform, and to learn how to prepare for better success next time.
One such meeting was hosted by Britain’s AD research charity Alzheimer’s Research UK on January 19 in London. Lilly’s Eric Siemers disclosed results of solanezumab’s CSF biomarker measurements in the Expedition 3 trial, in addition to data shown previously at CTAD. Debating the data with fellow scientists and competitors at Biogen, Janssen, and Merck, Siemers, along with his colleague Mike Hutton, explored possible explanations for solanezumab’s disheartening performance. Also taking part in the conversation were Bart De Strooper, who leads the U.K.’s nascent Dementia Research Institute; Paul Whiting, John Skidmore, and John Davis of the U.K.’s tripartite Drug Discovery Institutes in London, Cambridge, and Oxford, respectively; as well as 10 academic leaders at universities throughout the United Kingdom, France, Canada, Sweden, and the United States. Research funders included Hilary Evans and colleagues from ARUK, Tetsuyuki Maruyama from the U.K.’s Dementia Discovery Fund, Rob Buckle from the country’s Medical Research Council, Martin Rossor from the National Institute of Health Research, among others. Mike Hobday of the World Dementia Council was there, as well as Elena Becker Barroso, who edits The Lancet Neurology.
Before the assembled group dug into the data, John Hardy of University College London applauded the openness to data-sharing and academic collaboration Lilly has practiced in its Alzheimer’s drug development. “We appreciate the frankness and the commitment, and we feel the genuine disappointment on the part of so many people involved,” Hardy said. He asked the group to come to a conclusion by the end of the day about whether the solanezumab program thus far had, in fact, tested the amyloid hypothesis. Hardy co-organized the roundtable with Evans and other leaders at ARUK. Lilly scientists will formally publish the Expedition 3 data; an expert discussion held in the meantime, to help the field at large learn in a speedy fashion, is summarized below.
First, a brief summary of this humanized antibody as presented by Siemers. Its clinical program dates back 13 years, to 2004, when Phase 1 data showed that solanezumab bound Aβ40 monomer in the plasma of AD patients much like the original mouse version, m266, had done in transgenic mice. Lilly also determined that 0.1 percent of injected solanezumab enters the CSF. The scientists took these two findings as indications that solanezumab would engage its target, and from there calculated that monthly injections of 400 mg of antibody would reduce free plasma Aβ by more than 90 percent. That became the Phase 3 dose, which corresponds to 5 to 7 mg/kg. Initially in Phase 2, Lilly was running a dosing arm of weekly 400 mg injections, i.e., four times higher and closer to what current antibody programs are using in 2016; however, the company ended it after deciding that the lower dose would max out the antibody’s efficacy. After all, the operative idea for solanezumab at the time was the peripheral sink hypothesis, which holds that drawing down free Aβ in blood would gradually reduce it in brain, as well, by way of dynamic equilibria connected across the blood-brain barrier.
It did not work out that way. In discussing why not, the ARUK meeting attendees kept returning to an observation from the early days of solanezumab research: When the plasma concentration of Aβ bound to solanezumab shot up as it did—about 8,000-fold after the antibody was infused—the half-life of Aβ shot up along with it. The half-life of free Aβ in plasma is on the order of hours, but that of the complex was 28 days. In other words, binding to solanezumab appears to have blocked the natural clearance pathway of Aβ.
The importance of this finding was unclear at the time. Moreover, in a subsequent Phase 2 trial, solanezumab treatment caused an increase in the CSF concentrations of total Aβ (i.e. bound and free), a decrease in free CSF Aβ1-40, and an increase in free CSF Aβ1-42. This led Lilly scientists to think that their antibody was in fact shifting the Aβ equilibrium in the brain somewhat. That interpretation won out over the fact that there was no evidence of plaque reduction, or cognitive efficacy, through Phase 2. The hope was that longer, larger trials would show both.
At the ARUK roundtable in London, scientists questioned whether free CSF Aβ42 should have gone up if solanezumab was working as expected. Some argued that any Aβ coming off plaques in response to a shifting equilibrium should have promptly stuck to antibody. Alas, in 2008, when Lilly was weighing the data it had, dose-dependent target engagement in blood, combined with Lilly’s active interest in the peripheral sink hypothesis, won out, and the company headed into Phase 3.
Fast-forward to January 2017, after three Phase 3 trials in some 4,100 patients. This news story will skip over the 33 percent slowing in cognitive decline in mild cases pooled from Expedition 1 and 2, and the cognitive and prior biomarker data in Expedition 3. These results were previously reported (Aug 2012 news; Dec 2016 conference news). What happened to CSF Aβ—i.e. target engagement, where the rubber hits the road—in Expedition 3? In London, Siemers reported that CSF total Aβ1-40 and Aβ1-42 rose in the treatment groups in Expedition 3, replicating the result of Expedition 1 and 2. In contrast, free Aβ1-40 and free Aβ1-42 in the CSF showed minor decreases, some not statistically significant, and some similar to the placebo group. To boot, the CSF free Aβ data did not consistently replicate results from Expedition 1 and 2. By and large, CSF free Aβ looked flat, contrasting with the movement seen in Phase 2.
What could this mean? At the ARUK roundtable, some scientists said that they would have expected a robust, 18-month-long therapeutic response to reduce CSF free Aβ. In discussion, academic and industry scientists honed in on the long half-life of solanezumab-bound plasma Aβ. Some recalled that more than 90 percent of peripheral solanezumab was bound to Aβ. To their minds, this implied that greater than 90 percent of the 0.1 percent of the infused solanezumab that made it into the brain might have carried Aβ from the periphery along with it. This could account for at least some of the total Aβ measured in the CSF.
“The biomarkers may have misled us here,” said Henrik Zetterberg of the University of Gothenburg and UCL. It is unclear whether the solanezumab-Aβ complex measured in CSF was formed in the brain. Dominic Walsh, of Boston’s Brigham and Women’s Hospital and UCL, said this matches preclinical data showing that m266 does not alter free Aβ in mouse brain either; rather, the antibody simply increases levels of Aβ bound to it (Mably et al., 2015). In toto, the CSF data in Expedition 3 provide no objective support for target engagement, agreed Zetterberg, adding, “The huge effects in plasma and the very minor effects in CSF make the CSF data uncertain.”
In London, further discussion elaborated on the equally unexpected florbetapir PET results, which were first presented at CTAD in San Diego. In Expedition 3, Lilly made a positive amyloid PET scan an inclusion criterion after concluding that many participants in Expedition 1 and 2 had likely been amyloid negative. This conclusion was extrapolated from a small amyloid PET sub-study done in those earlier trials, but it resonated with what had been seen in ADNI, AIBL, and indeed postmortem series. The decision to ascertain amyloid positivity in Expedition 3 was fueled by doubt in the accuracy of a clinical-only diagnosis as much as by growing enthusiasm for amyloid PET at the time. After all, Lilly had bought its own tracer, Avid’s florbetapir, and was in a leading position to use it for a large multicenter trial. Surely, weeding out amyloid-negative “non-ADs” would lift what was otherwise meant to be a confirmatory trial up above the small 33 percent effect size netted by the analysis of the pooled Expedition 1 and 2? Moreover, the Expedition 1 and 2 PET sub-study had suggested a statistically significant reduction in aggregated amyloid with solanezumab. Even though plaque reduction was not seen in mouse models, this first Phase 3 PET result was now fanning hope of a peripheral sink effect in people.
Again, it was not to be. In Expedition 3, plaque load nudged down just a smidgen, about 10 times less than seen with aducanumab. What happened? In London, Siemers explained that Expedition 3 participants had to have a florbetapir PET scan that was judged to be positive by way of a visual read. Researchers agree that visual reads tend to equate to a slightly higher amyloid burden than if amyloid positivity is determined quantitatively against a numerical threshold on the PET standard uptake volume ratio (SUVR). Indeed, in Expedition 3, the visual reads correspond to baseline SUVRs well above that threshold, Siemers said. Quite possibly, Expedition 1 and 2 may have included people whose amyloid burden was growing but was still below the SUVR threshold; they would have been screened out in Expedition 3. Site leaders for mild and prodromal-stage trials note that the cutoffs for biomarker requirements—which are still being refined as sponsors gain experience using them—sometimes force them to exclude participants who otherwise seem well-suited to a trial. “Our requirement for a visual read of the scan meant people had a relatively high amyloid load at baseline,” Siemers said.
This indication that neuropathology was more advanced in Expedition 3 than in Expedition 1 and 2 participants draws indirect support from the trial’s demographic data. They show that its groups were well-matched across categories, except that Expedition 3 enrolled about 67 percent ApoE4 carriers whereas Expedition 1 and Expedition 2 enrolled about 57 percent. ApoE4 is associated with amyloid deposition starting earlier in life. “When you require amyloid positivity in a trial, your ApoE4 carrier rate goes up,” Siemers said. In essence, changing the inclusion criteria meant the trial enrolled a different population, in this case slightly more advanced disease that may have been out of reach of the therapeutic oomph of solanezumab.
Overall, solanezumab had little effect on downstream markers of pathology or neurodegeneration. The CSF tau measurement showed so much heterogeneity among people that scientists at the ARUK roundtable considered them mostly noise. Commenting on the MRI findings, Nick Fox of UCL noted that the trial’s brain volume data showing no significant difference with solanezumab made sense. “In every study where there has been engagement with amyloid plaque, for example ARIA cases in immunotherapies, there has also been short-term increased brain volume loss. That we did not see this fits with amyloid PET not changing. We can conclude that solanezumab has not changed the established amyloid plaque load, with its associated inflammation,” Fox said.
Given those difficulties, was the 11 percent cognitive benefit reported in Expedition 3 true, or noise? The ARUK roundtable leaned toward the former view, because all cognitive curves across Expeditions 1, 2, and 3 separated in the same direction. Four of six cognitive readouts, each assessed by a different rater, showed a benefit. Siemers readily agreed that the 11 percent benefit was smaller than donepezil’s. “At Lilly, we agree that a benefit smaller than 15 percent is not clinically meaningful,” Siemers said. “It is not enough, and we cannot be sure it’s real.”
Did the 33 percent slowing of cognitive decline in Expedition 1 and 2, followed by 11 percent in trial 3, represent a “winner’s curse”? In other words, was it the regression toward the mean that often happens in GWAS and epidemiology, when the first test of a given sample nets a larger effect size than the replication study? Or was it a true biological effect of the 0.1 percent of infused antibody that reached the CNS, even if biomarker data does not support a clear mechanism of action? For his part, Hardy told Alzforum that given the measurement variability in clinical science, both Expedition1 and 2’s 33 percent and Expedition 3’s 11 percent cognition effect sizes could be within the 95 percent confidence interval for such studies. “It could be essentially the same result,” Hardy said. He compared solanezumab to the first airplane flight by the Wright brothers at Kitty Hawk in 1903. Their plane was in the air for all of 3.5 seconds and landed on its nose, yet the attempt was critical for success soon after.
The scientists assembled in London were curious about the dosing of solanezumab. Would it make sense to jack it up in the two remaining trials of solanezumab, DIAN-TU and A4? After all, a strong positive result in those two earlier-stage populations would be a tremendous boost for participants and the field at large. Some scientists cited dose increases in current trials of crenezumab and gantenerumab, though Biogen’s Samantha Budd Haeberlein argued against viewing anti-Aβ antibodies as a class. Each is sufficiently different in its binding characteristics and clearance mechanisms that they are no more similar to each other than one small-molecule drug is to the next. Others said that without a clear mechanistic rationale, they would not increase the dose mid-stream in a trial.
Lilly scientists explained that their early clinical studies were limited. For one, they never determined a maximally tolerated dose—a measure that is usually developed as a ceiling at which a drug becomes intolerable. In fact, in 2009, when Expedition 1 and 2 began, this international Phase 3 program of monthly antibody administration was straining Lilly’s manufacturing capacity. Manufacturing yields have risen since then, but at the time the constraint was real. Moreover, the saturation of plasma Aβ by 400 mg/kg of a biologic drug thought to work via peripheral sink made the dosing data seem adequate. In a recent Alzforum comment, Steve Paul, now at Voyager Therapeutics, noted the limitations of solanezumab dose finding. His point drew wholehearted agreement among scientists in London, who noted wryly that Paul was at Lilly at the time.
Be that as it may, industry and academic scientists in London strongly agreed that it appears the peripheral sink hypothesis of amyloid reduction has been refuted. Future antibody programs will approach dose-ranging differently, Siemers said. For example, Lilly and AstraZeneca are co-developing a new antibody to monomeric Aβ42 based on its CSF exposure and pharmacokinetics, not plasma.
Typically, after suffering a stinging loss, pharma companies consign the drug at hand to the scrap heap of history, pick another candidate from the pipeline, and soldier on developing that. But wait! Couldn’t much more use be made from at least a billion dollars’ worth of human research? After all, this program—which De Strooper could not help but mention cost multiple times the budget he gets to build a nationwide dementia research institute—could be argued to have conducted essentially a single experiment. Surely, more can be learned from its rich trove of samples and data.
At the ARUK meeting, scientists volunteered ideas and their collaboration. For one, Zetterberg noted that the CSF measurements thus far reveal but a sliver of Aβ biology in the brain. The major Aβ species in human AD brain is Aβ 4-42. It forms on plaques and can be measured after depleting Aβ 1-42 from the sample. “Looking at more species would be very important,” Zetterberg said.
The PET data warrant close dissection, as well. “This was the first time a global Phase 3 trial used PET as an inclusion criterion. It’s wonderful that there is so much PET data available in Expedition 3,” said Budd, to heads nodding around the room. Budd recommended Lilly drill down to individual-level data and compare people with higher or lower baseline amyloid, or indeed tau. Did they have lower cognitive scores within the baseline distribution? Did they progress faster? Did people respond differently to solanezumab based on their amyloid or tau burden at baseline? Did baseline amyloid or tau load rightly correlate with any other measures of the trial?
Luc Buee of the University of Lille, France, suggested stratifying Expedition 3 data in many more ways than shown thus far. “We’ve only seen ApoE4. We could stratify by other polymorphisms identified in GWAS, additional biomarkers including cytokines, chemokines, and other immunity/inflammation proteins, or even cell analyses such as macrophage, and T cell measures,” Buee said. “There is much going on in AD besides amyloid. We should learn about those things from a large trial like this to help us answer the question of whether solanezumab has completely failed, or too little of it got into the brain.”
About this question, De Strooper urged Lilly to conduct small exploratory studies pumping solanezumab directly into the brain’s intrathecal space. “Do not throw solanezumab away. It is a great tool. Put it directly into the brain, so you get much higher exposure. We know it is safe, we know it binds Aβ, we have a small hope it has an effect,” De Strooper said. Intrathecal infusion is considered impractical to deliver repeat doses to millions of patients, but it can be scientifically informative. It is, in fact, being used in clinical trials, for example in an international, multicenter Phase 1 study of an antisense oligonucleotide, aka ASO, to knock down huntingtin in people with early Huntington’s disease (see clinicaltrials.gov); preclinical research is showing promise for ASOs in tauopathies, as well (see Jan 2017 news).
The Expedition 3 discussion concluded with broad consensus around the room that solanezumab had not fully tested the amyloid hypothesis. Where next? “I worry that we keep making half-finished pointillist paintings: a dot here and a dot there, but we never see the picture,” said Hardy. He argued for systematic efforts to integrate more information into what is still an overly simplistic, amyloid-only view of Alzheimer’s disease. For more on that, see Part 2 of this series.—Gabrielle Strobel
- CTAD: Solanezumab Seen to Nudge AD Ever so Slightly
- Phase 3 Solanezumab Trials "Fail"—Is There a Silver Lining?
- Post-Brexit U.K. Picks Belgian to Run Its Flagship Dementia Research Institute
- Antisense Oligos Tango with Tau Transcripts to Reverse Tauopathy
- After Solanezumab: Where Should Alzheimer’s Research Go?
- Mably AJ, Liu W, Mc Donald JM, Dodart JC, Bard F, Lemere CA, O'Nuallain B, Walsh DM. Anti-Aβ antibodies incapable of reducing cerebral Aβ oligomers fail to attenuate spatial reference memory deficits in J20 mice. Neurobiol Dis. 2015 Oct;82:372-84. Epub 2015 Jul 26 PubMed.
No Available Further Reading