New Data from Past BACE Inhibitor Trials Shed Light on Side Effects

16 Dec 2020

BACE inhibitors succeeded in stemming the tide of Aβ production in the brain, but the drugs also slightly dulled cognition and shrank the brain, bringing the trials to a screeching halt. Rather than washing their hands of a failure and moving on, researchers continue to dig through the trial data to find out exactly what happened, and perhaps even to chart a way forward for the drugs. Three recent studies—all led by Michael Egan at Merck—add new insight.

Verubecestat triggered a rapid atrophy that did not get worse.
This happened in amyloid-rich regions, and did not track with neurodegeneration.
Exploratory cognitive measures from verubecestat and lanabecestat trials indicate opposite effects on memory and verbal fluency.

One, published November 30 in Brain, described regional patterns of brain volume loss in Merck’s EPOCH trial of verubecestat. The shrinkage came on fast but did not progress. It happened mostly in areas burdened by amyloid, uncoupled from cognitive impairment or neurodegeneration. A study in the November 21 Journal of Alzheimer’s Disease reported that, in the same trial, retinal biomarkers and brain atrophy were not linked. Another, published October 13 in Alzheimer’s and Dementia, was a collaboration between Merck, Eli Lilly, AstraZeneca, and academic scientists. It reported exploratory cognitive outcome measures from two trials of the BACE inhibitors verubecestat and lanabecestat in prodromal and early AD, respectively. Both drugs took an early toll on episodic memory but, strangely perhaps, boosted verbal fluency.

The findings do not amount to a complete explanation. Rather, they add to a growing, if largely unspoken, sense that the cognitive impairment and brain volume loss caused by BACE inhibitors is mild, non-progressive, and potentially reversible. They also flag specific changes in cognition or brain volume that could be used to quickly find a safe dose in future trials.

“I congratulate both companies on sharing more detailed data from the terminated trials and on their collaborative effort to compare two trials and inhibitors,” wrote Stefan Lichtenthaler of the German Center for Neurodegenerative Diseases in Munich. “Although the molecular basis of the cognitive side effects is not yet clear, the three studies together provide very helpful guidance for possible future BACE inhibitor trials.”

Over a brutal two-year period, five Phase 3 trials of BACE inhibitors ran into the ground. In the end, the whole class of drug—including Merck’s verubecestat, Janssen’s atabecestat, Novartis’ umibecestat, Eli Lilly/AstraZeneca’s lanabecestat, and Eisai/Biogen’s elenbecestat—were tied to cognitive deficits, brain atrophy, or both (Dec 2017 conference news; Nov 2018 conference news; Jul 2019 news).

Since then, scientists have continued to pore over the data, hoping to understand how this class of drugs veered off course (Dec 2019 conference news). In the Generation trials of umibecestat, researchers continued to monitor cognition and brain volume in trial participants after treatment was stopped, and reported that cognitive deficits and, to some extent, brain volume loss, were reversible (Aug 2020 conference news). Similarly, cognitive deficits triggered by atabecestat largely abated three months after treatment stopped (Novak et al., 2020).

Verubecestat was the first to fall. The Phase 3 EPOCH trial, which enrolled people with mild to moderate AD, stopped in 2017, and the APECS trial of people with prodromal AD stopped soon after (Feb 2017 conference news; Dec 2017 conference news; Nov 2018 news). The drug was associated with hippocampal volume loss in EPOCH, but not with cognitive worsening beyond what was caused by AD. In APECS, however, the drug appeared to make people’s cognition worse (May 2018 news; Apr 2019 news).

In their follow-up study, first author Cyrille Sur and colleagues dove deeper into the imaging data collected during EPOCH. The trial included more than 2,000 participants who took either 12 mg or 40 mg of verubecestat, or placebo, for 78 weeks. More than 450 in each dose group had MRI scans at baseline, while a smaller number did so again at 13, 52, and 78 weeks. Brain volume—as gauged by cortical thickness, whole brain volume, or hippocampal volume—slipped for all groups throughout the trial, but it slipped by more in the two groups on drug. This difference emerged at 13 weeks, and it held steady throughout the trial.

The researchers also measured volume changes in 31 brain regions between baseline and 78 weeks. They found that in people on placebo, regions known to degenerate in AD, such as the entorhinal cortex, amygdala, fusiform cortex, and hippocampus, shrank the most throughout the trial. A similar atrophy pattern happened in the treatment groups, with additional verubecestat-related volume loss to boot. However, the drug did not appear to accelerate AD-related neurodegeneration.

How did the atrophy attributable to verubecestat relate to amyloid? The researchers addressed this question in several ways. Looking at the pattern of verubecestat-related brain volume loss, they found that it coincided with regions most prone to amyloid accumulation. This shrinkage was prominent in the cuneus, precuneus, lateral occipital, and supramarginal cortex, which are among regions burdened by amyloid early in disease. They found no significant treatment effect in white-matter regions, or in regions known to be largely free of amyloid deposition, such as the cerebellum. Alas, despite this apparent link to amyloid, there was no correlation between a person’s baseline amyloid load in a given region and his or her subsequent drug-related volume loss.

Is this volume loss due to plaque reduction? It’s not that simple. According to longitudinal amyloid-PET imaging data collected from a subset of 42 participants, a person’s amyloid reduction over 78 weeks had nothing to do with how much his or her hippocampus shrank. If anything, the relationship trended in the opposite direction: Patients with less amyloid reduction tended to lose more hippocampal volume triggered by verubecestat.

These regional MRI findings mesh with a more fine-grained analysis of the imaging data that David Scott of Merck presented at the Clinical Trials in Alzheimer’s Disease (CTAD) meeting held virtually in November. Tracking change by way of voxel-based morphometry, Scott found similar patterns of verubecestat-related volume loss, which were aligned with amyloid-rich regions.

A more sinister explanation might be that verubecestat shrank the brain by killing neurons. That’s unlikely because the drug-induced atrophy stayed stable throughout the trial, as does CSF biomarker data collected from a subset of 118 participants. The researchers reported that this mysterious volume loss did not correlate with the neurodegeneration marker neurofilament light (NfL). The researchers also found no relationship between the degree of verubecestat-related brain volume loss at week 13 and scores on the ADAS-Cog11. Because participants in this trial, who had mild to moderate AD, already had significant cognitive deficits that worsened throughout the trial, it may have been difficult to detect any deficits triggered by the drug, the researchers noted. This is distinct from the situation in the APECS trial, in which participants were still in the earliest stages of the disease and the drug did worsen their cognition.

In a separate study led by Egan, first author Robert Sergott and colleagues checked whether the brain shrinkage in EPOCH tracked with biomarkers in the eye. They did this because mouse studies had raised the specter of retinal toxicity with BACE1 inhibition. Separately, changes in retinal structure have been proposed as potential AD biomarkers (Aug 2018 news; Apr 2019 conference news), and EPOCH included several retinal biomarkers as exploratory outcome measures. In short, the researchers found no evidence of any retinal changes relating to verubecestat treatment, and the thickness of the retina tracked at best weakly with AD-related neurodegeneration throughout the 78-week trial.

If it wasn’t neurodegeneration or amyloid reduction, then why did the brain get a little smaller on verubecestat? For all his searching, Egan still does not know, he told Alzforum. He speculated that it could relate to microglia around Aβ plaques. Perhaps with Aβ production being way down around plaques, immune cells weren’t being recruited and riled any more, dampening inflammation. Robert Vassar of Northwestern University in Chicago had a similar take, noting that BACE activity is highest in dystrophic neurites around plaques. Shutting down production of Aβ from these hubs of pathology may have doused inflammatory responses enough to ease local swelling, leading to a non-progressive volume loss. This would explain why the brain shrank in regions flush with plaques, he said.

Postmortem neuropathological data presented at CTAD support the idea that verubecestat changes the plaque environment. Alberto Lleó of the Hospital de Sant Pau in Barcelona, Spain, used immunohistochemistry and immunoblotting to detect different forms of Aβ and synaptic markers in the brain of a 63-year-old man who had received 12 mg verubecestat for 38 months. Compared to samples from a group of untreated patients with early onset sporadic AD, the sample from the treated patient contained fewer non-fibrillar, soluble forms of Aβ, despite harboring similar levels of insoluble, fibrillar Aβ. Furthermore, the verubecestat-treated patient had fewer dystrophic neurites around plaques than did the untreated patients.

Jochen Herms of the German Center for Neurodegenerative Diseases, Munich, favors a different explanation. Animal studies have shown that BACE inhibitors affect the density of excitatory synapses, regardless of amyloid (Nov 2014 news; Zhu et al., 2018). “The acute and non-progressing loss in brain volume can easily be the consequence of a loss of a subset of synapses that are sensitive to BACE inhibition,” he wrote. “It is much less likely that this acute volume loss is an amyloid-related process.”

The distinction is important, according to Eric McDade of Washington University in St. Louis. McDade wrote that understanding the relationship between amyloid burden and the detrimental effects of BACE1 inhibition will be helpful for primary versus secondary prevention trials with lower doses of the inhibitors. For example, would amyloid-negative participants be at greater or lesser risk of faster atrophy than amyloid-positive people? Knowing this would be important in the design of any future BACE inhibitor trials in the Dominantly Inherited Alzheimer’s Network (DIAN).

Curious Cognitive Effects Line Up
Perhaps the BACE inhibitors’ most unsettling problem was that they worsened cognition. At conferences and in conversation, the size of this effect has been compared to drinking a glass of wine, or taking a Benadryl. What do the data say? To take a closer look at how the drugs alter cognition, the sponsors of two Phase 3 trials joined forces to report secondary and exploratory cognitive outcome measures. APECS tested verubecestat in prodromal AD and, after failing an interim futility analysis, was cut short by a year. Egan subsequently reported that participants on the 40 mg dose fared worse on the CDR-SB, its primary outcome, and slipped faster into dementia than did those on 12mg or placebo. ADCS-ADL_MCI scores, a secondary outcome, also worsened more on treatment (Apr 2019 news). AMARANTH, a trial of lanabecestat in MCI due to AD or mild AD, halted 16 months early when it failed its futility analysis. Lanabecestat did not appear to influence scores on the ADAS-Cog13, that trial’s primary outcome measure (Wessels et al., 2019).

For the current study, investigators jointly analyzed secondary cognitive outcomes from APECS and exploratory cognitive outcomes from AMARANTH. Because they rely on many statistical tests and are therefore more likely to produce a spurious chance result, exploratory outcome results are rightly met with a hefty dose of skepticism. Egan told Alzforum that the joint analysis was meant to lend power and credibility to the findings.

APECS used the Alzheimer’s Disease Cooperative Study–Activities of Daily Living for Mild Cognitive Impairment scale (ADCS-ADL_MCI) and the 3-domain Composite Cognition Score (CCS-3D) as secondary outcomes. The CCS-3D encompasses multiple tests to evaluate episodic memory, attention/processing speed, and executive function. Co-first authors Alette Wessels of Eli Lilly and Christopher Lines of Merck reported that CCS-3D scores in the treatment groups slipped relative to placebo starting at week 13. This worsening was driven by episodic memory and attention/processing speed. In particular, treatment-related deficits were largest and most consistent for digit symbol coding, a test of memory and processing speed. In contrast, participants on drug outperformed those on placebo in the third domain—executive function. That benefit was driven by scores on letter and category fluency tests.

Domains Dissociated. In the EPOCH trial, people on 12 mg (red) or 40 mg verubecestat (blue) scored worse than those on placebo (grey) in digit symbol coding (left), but better in letter fluency (right).

The same trends emerged in AMARANTH, whose exploratory outcomes were the Repeatable Battery for the Assessment of Neuropsychological Status, Letter Fluency, Category Fluency, and WAIS-III Digit Symbol Coding. For the RBANS, which measures attention, language, visuospatial/constructional abilities, and immediate and delayed memory, participants in both the 20 mg and 50 mg lanabecestat groups slipped relative to placebo at multiple time points across the trial. They also underperformed on Digit Symbol Coding, with those in the 50 mg dose group scoring lower than placebo at all time points, and those in the 20 mg group underperforming at weeks 26 and 52. But here, too, participants in both drug groups scored higher on letter and category fluency than did those in the placebo groups.

Scopalamine, an anti-cholinergic drug used to model cognitive impairment, also blurs episodic memory while improving letter fluency. Researchers have hypothesized that its contrasting effects are caused by a disinhibition that dampens performance on some tasks while boosting it for others (Pompéia et al., 2002).

To Egan, the data imply that BACE inhibition might exert unique influences in different regions of the brain. For example, mouse studies suggest that BACE inhibition can cause synaptic dysfunction in the hippocampus, which would explain the episodic memory effect. In prefrontal and frontal regions needed for verbal fluency, perhaps BACE performs different functions that aren’t affected by the inhibitor. Perhaps the benefit of BACE inhibition—i.e., stemming the production of Aβ monomers and, from there, oligomers—can shine through in these centers, he hypothesized.

It is also plausible that exactly which among its many substrates BACE1 primarily cleaves—and therefore the effect of inhibition—differs between brain regions, Vassar noted. In the hippocampus, where synaptic plasticity is paramount, insufficient processing of substrates like close homolog of L1 (CHL1) or seizure-6, which function in plasticity, could be detrimental (Jun 2012 news; Oct 2016 conference news).

Colin Masters, University of Melbourne, Australia, believes that these effects likely result from on-target inhibition of APP processing. “If the normal cleavage of APP releases the extracellular and intracellular domains, and if the normal function of APP is for heterotypic interactions at the peri-synaptic zones that are used for synaptic plasticity (for learning and memory functions), why then should we not expect 80 percent inhibition to result in deficits in learning and memory?” Masters noted that only 5 percent of Aβ produced every hour accumulates. He believes that much smaller degrees of inhibition, i.e. much lower inhibitor doses, might be sufficient to keep Aβ accumulation in check—and be safer.

“The comparative paper demonstrates impressively that a detailed analysis of the diverse cognitive tests is important, as it shows previously unanticipated cognitive worsening of lanabecestat—not globally but on a subset of tests, whereas improvement in other tests can now also be appreciated,” wrote Lichtenthaler. To his mind, the findings imply that specific cognitive tests—such as digit symbol coding—may help trialists find the right dose of BACE inhibition within a short trial.

What do BACE inhibitor follow-up studies portend for the future of these drugs? To Paul Aisen of the University of Southern California in San Diego, a co-author on this latest batch of papers, now is not time to throw in the towel. “Secretase inhibition remains the most direct method of addressing the accumulation of amyloid in the AD brain, and may be ideal to treat amyloid dysregulation in individuals at high risk for amyloid accumulation (i.e., primary prevention of AD),” he wrote. “It is very important that we investigate the safety of lower doses to consider further early intervention studies; we know well how to monitor the toxicity, so we have a feasible path forward.”—Jessica Shugart