17 May 2019

At the AD/PD 2019 conference held recently in Lisbon, the amyloid hypothesis was reeling from repeated gut punches in the clinic. Both antibodies and BACE inhibitors are facing relentless criticism from academic scientists and biopharma analysts (e.g., May 2019 news). Lost in the cacophony can be the voices of patients and their families urging researchers to push on, and of site leaders who swear they “saw something” among their participants. A little lost as well, amid all the opinion, can be the data itself. Some anti-amyloid investigational antibodies (see Part 15 of this series) and the active vaccine CAD106 are still in Phase 3 or 2/3, as are elenbecestat and umibecestat. In Lisbon, scientists presented and debated data on both these extant BACE inhibitors as well as an extinct one, lanabecestat.

First, the one that doesn’t work, lanabecestat. In June 2018, a futility analysis of its AMARANTH Phase 2/3 trial of 2,219 people with MCI due to AD or mild AD predicted it would fall short of its goal. At the time, Lilly and AstraZeneca stopped not only AMARANTH, which was fully enrolled and 16 months from completion, but also its sister trial DAYBREAK in people with mild AD (June 2018 news). In Lisbon, scientists presented full results on both studies, picking through the wreckage to see what they could learn.

Pierre Tariot of the Banner Alzheimer Institute in Phoenix started off with AMARANTH, the larger of the two. Baseline demographics and clinical assessment showed its groups, gathered at hundreds of sites and dozens of countries around the world, were well balanced. This is a small step forward in the sense that the trials themselves are increasingly poised to pick up a drug benefit once an effective drug comes along. “We know from these presentations that the lanabecestat trials were well done,” said Eric Karran of Abbvie.

Alas, lanabecestat is not that drug, at least not in symptomatic AD. In slide after slide of nearly perfectly superimposed curves, Tariot showed that neither of the two doses tested, 20 or 50 mg per day, moved the primary outcome, or any of the secondary outcomes, even one bit. Lars Lannfelt of Sweden’s Uppsala University captured the mood of many when he said, “Seeing how there was no effect at all was depressing.”

On all of those outcome scales, participants on placebo declined, after both one and two years, as much as one would expect, hence the trial’s negative result can’t be blamed on an aberrant placebo response, Tariot noted. The placebo group’s decline has been a point of contention in trials in years past, so here, too, a small advance was noted. The lanabecestat groups were sizable, boasting more than 700 at baseline and 171–198 who had completed the two-year treatment period when the trial got cut short. In essence, Tariot showed, the trial worked, the drug didn’t.

How safe was lanabecestat? AMARANTH generated a total drug exposure of 1,200 person-years. Lanabecestat was well tolerated, without ARIA, and, most importantly, without the cognitive worsening that had beset Merck’s BACE inhibitor verubecestat and Janssen’s atabecestat. Lanabecestat’s side effects did not warrant stopping, Tariot said. To optimists, this data means that cognitive worsening is not a class effect of BACE inhibition, contrary to previous concerns (Nov 2018 conference news). 

That said, the high dose did come with more adverse events and withdrawals. The most common ones considered serious were psychiatric, weight loss of nearly 2 kg, and hair discoloration. Patchy depigmentation of the skin and hair has been reported in mice, rats, rabbits, and dogs for those BACE inhibitors that also block BACE2’s cleavage of pigment cell-specific melanocyte protein (Shimshek et al. 2016; Cebers et al., 2016; Nov 2016 news). Lanabecestat is such a non-selective BACE1/2 inhibitor.

Next up in Lisbon, Lilly’s John Sims presented DAYBREAK data. They matched those of AMARANTH. The data set is smaller because DAYBREAK started after AMARANTH and was still enrolling when both trials ended. Even though DAYBREAK sought only people with mild AD and AMARANTH also enrolled MCI due to AD, their respective populations turned out to be nearly identical, except the former were a tad more impaired on the ADAS-Cog, MMSE, and the RBANS delayed-memory index, Sims showed. And indeed, the parade of “no benefit” slides on the primary and secondary outcomes was nearly the same, as well.

To the ever-hopeful eye, a few of the drug-placebo curves did appear to separate. However, lest anyone read a hint of a treatment benefit into that, Sims noted that the separations were inconsistent across dose and mostly insignificant. He thinks they are noise. “Once you get below 250 people for a longitudinal study, you have a lot of baseline variation bias and survival bias, so you have to be very cautious,” Sims said. When DAYBREAK was halted last summer, it had around 560 people in each group at baseline, but only 18–24 of the earliest enrollees had reached the 78-week time point, so those groups were too small to support much interpretation.

DAYBREAK’s side effect profile was unsurprising, too. The trial did not replicate AMARANTH’s higher dropout rate due to adverse effects on the high dose, but it did replicate the overall pattern of side effects, especially the weight loss and hair discoloration.

Sims and Tariot both emphasized that the full data showed that the futility analysis had made the right call to end this drug program.

What about biomarker data? Lilly presented those, too. Brian Willis and colleagues combined plasma data from both trials and showed that both lanabecestat doses docked blood Aβ40 and 42 levels by 70 to 80 percent. For CSF, the researchers showed AMARANTH data. Aβ40 and 42 levels dropped by 50 to 73 percent, respectively; the alternative cleavage product sAPPα rose and the BACE cleavage metabolite sAPPβ plummeted. In other words, lanabecestat robustly engaged its target. This means the chosen doses tested—and unfortunately disproved—the hypothesis that lanabecestat slows progression of symptomatic AD.

Markers of downstream consequences of target engagement were inconclusive. CSF tau went down but not in a dose-dependent way, making this result hard to interpret, and a substudy of CSF and plasma neurofilament light showed no change with treatment or placebo.

The trials’ substantial stack of brain scans has its own story to tell. In Lisbon, Mark Mintun of Lilly/Avid Radiopharmaceuticals said that the lanabecestat trials had gathered the most imaging data of an Alzheimer’s clinical program so far. It offers lessons not only on what this BACE inhibitor did not do, but also more broadly on how longitudinal biomarkers change relative to disease progression at this early symptomatic stage. Both trials had scheduled florbetapir, flortaucipir, FDG-PET and volumetric MRI scans at two to four time points. Despite the early termination, researchers were able to analyze serial florbetapir scans from 400 participants, serial flortaucipir scans from 354, FDG PET on 298, and MRI on 2,351 participants.

What did they find? First, about the drug. Lanabecestat dose-dependently reduced amyloid, with the high dose drawing out 20 centiloids’ worth of plaques. This result by itself set tongues wagging at AD/PD. There were numerical hints of neurofibrillary tangle reduction, but they fell short of statistical significance. On both amyloid and tau PET, the placebo groups behaved as expected for this disease stage, i.e., florbetapir uptake stayed unchanged and flortaucipir uptake rose. Measured by MRI, the whole brain and the hippocampus area shrank faster in people on lanabecestat than placebo, Mintun showed.

Second, about disease progression. In correlating two-year biomarker change of all groups with change on their clinical outcomes, Mintun tied decreasing FDG PET to worsening across all clinical measures, from ADAS-Cog13 to ADLs to MMSE, FAQ, and CDR-SB. This adds recent, and sizable, multicenter trial data to the existing consensus that FDG PET might serve as an outcome measure in future trials of other drugs. These correlations were not increased by treatment, i.e., unrelated to lanabecestat.

Other biomarker-outcome correlations made less intuitive sense to the audience. For example, Mintun showed that a reduction in florbetapir uptake correlated with worsening on the MMSE. The ADAS-Cog, too, showed a trend toward worse scores with less amyloid. Less amyloid, worse cognition? Several commentators took this to mean that by the time a person has reached the MCI/mild dementia stage of Alzheimer’s, their disease has become independent of amyloid plaques. In other words, the trial confirmed that a person’s amyloid burden tends to level off and then nudge downward as his or her dementia deepens, but the drivers at this stage are factors beyond Aβ, and removing it does not help (Karran et al, 2011). 

What about tangles and atrophy for tracking disease progression? In this data set, flortaucipir uptake did not correlate with outcome measures, as was the case for CSF tau. But MRI did. A shrinking brain was bad for people, strongly correlating with worsening across all efficacy measures. As with FDG-PET, this atrophy correlation refers to disease progression itself, not to lanabecestat treatment.

In toto, Mintun showed, two years of high-dose lanabecestat slightly reduced a person’s neuritic plaque load, giving yet more evidence of target engagement. In mildly symptomatic people, this yielded no clinical benefit, and the more distal markers did not respond to lanabecestat treatment. With regard to the field’s broader goal of incrementally validating outcome biomarkers for use in future treatment trials, FDG PET and brain volume both tracked with clinical worsening.

How About the Other BACE Inhibitors?
Given this resounding defeat of lanabecestat in mild AD, do BACE inhibitors have a future in Alzheimer’s treatment? Scientists are not ready to throw in the towel, though compounds that block both BACE1 and 2 equally appear to be a thing of the past. Two compounds currently left standing are Eisai/Biogen’s elenbecestat and Novartis/Amgen’s umibecestat/CNP520. Both claim to be selective for BACE 1 over BACE 2.

Elenbecestat is in the midst of a Phase 3 program called MISSION AD. Its two trials both evaluate 50 mg per day in 1,330 people with early AD, and are expected to read out in 2021. Meanwhile, three posters at AD/PD showed early data on screening tools and tau PET. Eisai’s Michelle Gee and colleagues are evaluating how well a Cogstate test that uses a grocery list to measure immediate and delayed recall predicts brain amyloid positivity among 2,746 people who were screened for participation in MISSION trials. People who scored low on remembering the food items, particularly after a half-hour delay, were more likely to have brain amyloid deposition than those who aced the list. Similar data were shown for a brief Cogstate battery of cognitive tests. This kind of evaluation helps trial sponsors find tools to cut back the number of expensive amyloid scans during screening.

Also at AD/PD, Andrew Stephens and colleagues at Life Molecular Imaging, previously Piramal, showed preliminary results of a tau PET substudy in the MISSION AD trials. Among 45 amyloid-positive participants scanned thus far with LMI’s tau tracer PI-2620, 27 had a brain-wide positive scan. Of those, 13 had overt neocortical tangles outside the mesial-temporal cortex; 14 had subtler deposition only in their hippocampal region, or diffuse or focal uptake. Eighteen participants had a negative PI-2620 scan. While the study is still enrolling, this early data suggest that the amyloid-positive, mildly symptomatic MISSION AD participants span a range of tau pathology as they are starting on elenbecestat.

Among AD researchers, BACE1 selectivity has become a touchpoint for justifying effort on a new generation of inhibitors. What about it for elenbecestat and umibecestat?

Elenbecestat is said to preferentially block BACE1 over BACE2 (Nov 2018 conference news). According to Teiji Kimura of Eisai, elenbecestat is not specific for BACE1; however, the dose used in Eisai’s clinical trials, which was guided in part by the 40 percent reduced Aβ production seen for the protective APP mutation, might allow for a preferential effect on BACE1. In a presentation made to investors and media on April 23, Eisai reported that elenbecestat binds BACE1 with an affinity of 19 nanomolar and BACE2, 67 nM, amounting to 3.53-fold selectivity. Pharma companies frequently synthesize their competitors’ compounds to compare them with their own, and this same Eisai presentation posts lanabecestat’s selectivity as being twofold, and verubecestat’s as being 0.17-fold, i.e., as preferring BACE2 over BACE1.

According to Eisai, depigmentation cropped up neither in animal studies nor clinic trials of elenbecestat thus far. At its most recent, eighth, meeting, MISSION’s data-safety-monitoring board reviewed safety and cognition data gathered thus far and recommended that the trials continue.

Of the BACE inhibitors currently in clinical trials, Novartis/Amgen’s umibecestat has a peer-reviewed publication to show it prefers BACE1 over BACE2 threefold. Moreover, it is present in skin only at low concentrations, meaning it has little chance to encounter targets that would lead to depigmentation (Neumann et al., 2018; Dobrowolska Zakaria and Vassar 2018). In Lisbon, Derya Shimshek of Novartis told Alzforum that Novartis, too, has seen no hair discoloration in preclinical studies. This side effect is not dangerous but nettlesome because it breaks the blind.

The API GENERATION program comprises two global Phase 2/3 secondary prevention trials of umibecestat/CNP520 in people whose risk of developing AD symptoms is high because they carry the APOE4 allele and are between 60 and 75 years old. Both studies are enrolling, and so far, there has been no statement of cognitive harm, according to Tariot. By design, one GENERATION study includes people without elevated brain amyloid. In other words, this trial, as well as the API ADAD crenezumab trial in Colombia, will be the first to generate data on how anti-amyloid treatment affects people with preclinical AD, at an earlier stage than all other trials.

Meanwhile, Novartis researchers at AD/PD in Lisbon presented animal data suggesting their compound may not cause the hippocampal mossy fiber damage scientists believe might partly account for cognitive worsening. Novartis’ Carine Kolly and colleagues picked up recent work by Robert Vassar’s group at Northwestern University in Chicago, which had described this phenotype in adult conditional BACE knockout mice. Their hippocampal mossy fibers were shorter and disorganized, just as in knockouts of the axon guidance gene, and BACE1 substrate, CHL1 (Sep 2018 news). 

Kolly and colleagues first confirmed that Novartis’ in-house BACE1 knockout mice had this phenotype. Then they examined brain tissue from a long-term rat toxicology study to see if CNP520 produced it, too. The rats had received a six-month course of up to 200 mg/kg/day of umibecestat, a 10-fold excess over the highest dose used in people. This slashed Aβ levels yet left the length and organization of mossy fibers in the hippocampus intact.

Why would CNP520 not inhibit CHL1 processing? Kolly noted that cells harbor two pools of BACE1, one in the trans-Golgi network and one in early endosomes. CHL1 is cleaved in the trans-Golgi; APP in endosomes. While the genetic knockouts lack BACE1 in both pools, CNP520 preferentially accumulates in endosomes. This subcellular localization, Kolly hypothesized, may explain why the drug leaves CHL1 alone.

CHL1 is but one of about 45 known BACE substrates, and sorting out what BACE inhibition does to them is important. It is a complicated task, because changes in the processing of some of those substrates might mask a cognitive benefit, besides causing outright side effects. “We need more specific inhibitors that avoid other substrates. Ideally, you want to block only BACE1’s interaction with APP,” Lannfelt told Alzforum.

Separately at AD/PD, Shimshek and colleagues at Novartis showed preclinical data hinting that umibecestat could potentially treat cerebral amyloid angiopathy (CAA), a common pathological feature of Alzheimer’s and by itself a cause of stroke and dementia. APP23 mice who ate umibecestat with their chow for six months deposited less amyloid in their cerebral small blood vessels. The effect was dose-dependent.

What Next?
Opinion at AD/PD about the future of BACE inhibition diverged widely. While some scientists believe the approach should be scrapped, others insist that lower doses ought to be tried at earlier disease stages in a preventive mode.

“Lanabecestat’s toxicity was not bad for a terminal disease. In that way, I see Alzheimer’s like cancer. We now know BACE inhibition reduces amyloid, by 20 percent as per PET. We have to find a way to test these drugs earlier,” said an undeterred Colin Masters of the University of Melbourne, Australia. He and others envision future trials that enroll people whose amyloid deposition just crosses the PET threshold, push it below the threshold with an Aβ antibody, and then keep it there with a low dose of a BACE1 inhibitor. Vassar, who cloned BACE but also described several side effects of knocking it out, remains hopeful, too. “Let’s not give up on BACE inhibitors. We still do not know when to treat and how much to inhibit. Every failure teaches us something,” Vassar said.

On May 10, Eisai announced that the Alzheimer’s Clinical Trials Consortium (Dec 2017 news) chose elenbecestat and BAN2401 for primary and secondary prevention trials starting up early next year (company press release).—Gabrielle Strobel

Comments

1. • Heinz Hillen
     
   • Posted: 21 May 2019

   There is now substantial clinical data indicating that inhibition of Aβ generation via γ- or β-secretase inhibition is not effective and, at least in some cases, is associated with cognitive impairment. Over the last years, we have been good at defining compound- or enzyme- related explanations, e. g., Notch or BACE1/2 selectivity.

   Rather than continuing clinical research with new BACE inhibitors, we should conclude that underlying normal Aβ turnover in non-familial AD is not part of the pathology. BACE elevation in AD alternatively can be regarded as a compensatory replenishment of loss of proper Aβ monomer substrate by misfolding processes. Thus therapy and prevention for Aβ-related strategies in AD should be designed to precisely neutralize exclusively early, misfolded Aβ species, which is expected to reverse pathology and result in normalization of elevated BACE levels.

   To demonstrate involvement of Aβ pathology in AD, we need highly potent and highly selective immunotherapies that neutralize the earliest pathological Aβ species, the soluble oligomers at the earliest point in time possible. These antibodies will have the potential to reverse pathology without unwanted plaque dissolution and Aβ monomer interaction.

   View all comments by Heinz Hillen

Make a Comment

To make a comment you must login or register.

References

News Citations

1. Keep Your Enthusiasm? Scientists Process Brutal Trial Data 16 May 2019
2. Scratch Lanabecestat: This BACE Inhibitor Doesn’t Work in Symptomatic AD, Either 12 Jun 2018
3. Bump in the Road or Disaster? BACE Inhibitors Worsen Cognition 2 Nov 2018
4. Will Next-Gen BACE Inhibitors Dodge Side Effects? 3 Nov 2016
5. In Conditional BACE1 Knockouts, Hippocampal Axons Compromised 21 Sep 2018
6. Clinical Trials Consortium Succeeds ADCS, Focuses on Prevention 20 Dec 2017

Therapeutics Citations

1. Elenbecestat
2. Umibecestat
3. Verubecestat
4. Atabecestat

Mutations Citations

1. APP A673T (Icelandic)

Paper Citations

1. Shimshek DR, Jacobson LH, Kolly C, Zamurovic N, Balavenkatraman KK, Morawiec L, Kreutzer R, Schelle J, Jucker M, Bertschi B, Theil D, Heier A, Bigot K, Beltz K, Machauer R, Brzak I, Perrot L, Neumann U. Pharmacological BACE1 and BACE2 inhibition induces hair depigmentation by inhibiting PMEL17 processing in mice. Sci Rep. 2016 Feb 25;6:21917. PubMed.
2. Cebers G, Lejeune T, Attalla B, Soderberg M, Alexander RC, Budd Haeberlein S, Kugler AR, Ingersoll EW, Platz S, Scott CW. Reversible and Species-Specific Depigmentation Effects of AZD3293, a BACE Inhibitor for the Treatment of Alzheimer's Disease, Are Related to BACE2 Inhibition and Confined to Epidermis and Hair. J Prev Alzheimers Dis. 2016;3(4):202-218. PubMed.
3. Karran E, Mercken M, De Strooper B. The amyloid cascade hypothesis for Alzheimer's disease: an appraisal for the development of therapeutics. Nat Rev Drug Discov. 2011 Sep;10(9):698-712. PubMed.
4. Neumann U, Ufer M, Jacobson LH, Rouzade-Dominguez ML, Huledal G, Kolly C, Lüönd RM, Machauer R, Veenstra SJ, Hurth K, Rueeger H, Tintelnot-Blomley M, Staufenbiel M, Shimshek DR, Perrot L, Frieauff W, Dubost V, Schiller H, Vogg B, Beltz K, Avrameas A, Kretz S, Pezous N, Rondeau JM, Beckmann N, Hartmann A, Vormfelde S, David OJ, Galli B, Ramos R, Graf A, Lopez Lopez C. The BACE-1 inhibitor CNP520 for prevention trials in Alzheimer's disease. EMBO Mol Med. 2018 Nov;10(11) PubMed.
5. Dobrowolska Zakaria JA, Vassar RJ. A promising, novel, and unique BACE1 inhibitor emerges in the quest to prevent Alzheimer's disease. EMBO Mol Med. 2018 Nov;10(11) PubMed.

Other Citations

1. CAD106

External Citations

1. May 2019 news
2. company press release

Further Reading

No Available Further Reading