1 April 2011. Is it better to handcuff the bad guy or to reform his ways? The Alzheimer’s research version of this question—whether to inhibit γ-secretase or shift its M.O.—generated a lot of buzz at the 10th International Conference on Alzheimer’s and Parkinson’s Diseases, held 9-13 March in Barcelona, Spain. The question has been around for some time, but AD/PD 2011 galvanized debate about what scientists were to make of the latest thunder clouds gathering on the γ-secretase inhibitor front.

In a nutshell, scientists appear to be losing faith in the concept of Notch-sparing γ-secretase inhibition. That notion had appeared on the scene as the potential savior when Notch first proved to be a particularly troublesome one among the secretase’s many substrates. The enzyme complex is involved in the maturation of some 50 proteins. In his talk, Bart De Strooper of the Flanders Interuniversity Institute of Biotechnology in Leuven, Belgium, made quite clear his view that γ-secretase inhibition may not succeed because the safety margin between lowering Aβ42 and affecting Notch is too small. In his talk, Edward Koo of the University of California, San Diego, essentially concurred. Even some company scientists will say as much. “γ-secretase inhibition is hitting some pretty hard skids,” Dale Schenk of Elan Pharmaceuticals in South San Francisco told this reporter. Instead, scientists said, tweaking the enzyme complex without affecting its total output of Aβ—or that of APP’s C-terminal fragment, for that matter—might be more felicitous. Such compounds would be the γ-secretase modulators, or GSMs for short.

This concept is not exactly new. Koo cautioned that its potential remains theoretical because GSMs have not worked yet or even shown target engagement in humans. Flurizan failed because it may not have engaged its target sufficiently in the brain (see ARF related news story). Even so, excitement about GSMs was palpable at AD/PD 2011. Partly, that’s because scientists are gaining a clearer idea that these might be allosteric modulators and hence fall into an established mechanistic area of enzymology that is amenable to standard medicinal chemistry and drug development. In contrast, exactly how a Notch-sparing inhibitor would preclude processing of one substrate but not another has always remained nebulous. And partly, scientists now have second-generation GSM compounds that penetrate the blood-brain barrier better, and are currently wending their way toward the clinic and through Phase 1 and 2 trials.

What’s Eating γ Inhibitors?
Worry about γ-secretase inhibition could be heard for many years, but the loss of confidence accelerated with the humbling fate of Eli Lilly’s Phase 3 trial of semagacestat. It nosedived last August when the safety monitoring board unblinded the data midway through to discover that people on drug were worsening cognitively and functionally while coming down with unacceptable side effects (see ARF related news story). That trial is still blinded, Lilly’s Patrick May said at AD/PD 2011. Dosing is over, but clinical observation and biomarker measurements are continuing, and only when the data come to analysis this June will the scientists have a chance to figure out what happened.

Reviewing the status of γ-secretase inhibition in a symposium talk, Koo noted the uncertain path ahead. In late September 2010, Script Pipeline Watch reported that Pfizer discontinued development of its γ-secretase inhibitor begacestat after several Phase 1 trials (see ARF related news story). The reasons are unclear. Inherited from Wyeth after the Pfizer merger with Wyeth, this drug had been presented as having a larger therapeutic index than semagacestat, that is, as inhibiting the enzyme’s ability to process APP at much lower doses than are needed to block cleavage of Notch. This difference, the Wyeth/Pfizer scientists had thought, afforded enough space to make γ-secretase inhibition with this compound feasible.

Furthermore, Koo noted that a third compound, which is said to have an even larger window between APP and Notch, may be encountering similar problems. This is the BMS-708163 γ-secretase inhibitor. Bristol-Myers Squibb has completed a Phase 2 trial with this compound in mild to moderate AD, and is currently conducting a Phase 2 trial in prodromal AD. This trial is being closely watched for its novelty. It enriches for patients on the basis of their CSF Aβ42 levels and an objective memory impairment, making it a test case for earlier-stage treatment than the standard mild to moderate trial. According to ClinicalTrials.gov, this trial last October dropped the higher 125 mg dose down to 50 mg. According to a February 17, 2011 article by BioPharm Insight, an industry intelligence company based in Norwood, Massachusetts, dose-related toxicity in the previous mild to moderate trial included skin rash, gastrointestinal symptoms, and a “suggestion of rate of decline not favoring treatment.”

This profile is eerily reminiscent of semagacestat, an earlier-generation γ-secretase inhibitor that does not discriminate well between APP and Notch. BMS-708163 is thought to do so with a much larger margin, and in Barcelona, BMS scientists led by Jere Meredith presented in-vitro data for a selectivity between these two proteins of 193-fold versus 13-fold for semagacestat. Yet when asked about this, scientists at a handful of other biopharma companies and academic labs noted that they had synthesized the BMS compound and discovered that, at least in their hands, it distinguished less strongly between APP and Notch.

While this is true in her lab as well, the field should not put too much weight on in-vitro assays of therapeutic indices of these compounds, commented Barbara Tate of Satori Pharmaceuticals, a Cambridge, Massachusetts-based biotech company. The real answer will come in the clinic as investigators gain experience with drug exposure and deal with the pharmacokinetic and pharmacodynamic balancing act of systemic versus central activity. Human pharmacokinetics could influence dose selection. For instance, the 125 mg dose could have proved too high because the drug accumulates in humans when taken chronically, said Tate, boosting over time the body’s de-facto exposure to the drug. Lowering the dose might solve the problem. Finding the highest tolerable dose, after all, is part of the purpose of Phase 2 trials. Other scientists questioned whether the remaining 50 mg dose in the prodromal BMS trial lowers Aβ enough. Again, if the drug accumulates, both its toxicity and its Aβ-lowering effect would change with its particular human pharmacodynamics and, in theory, both can be brought into range by adjusting the dose, Tate said.

Is Notch the Only Problem?
Recent basic science on the role of Notch signaling in the adult brain has only heightened researchers’ collective awareness that it is important to stay away from this substrate (for a recent review, see Pierfelice et al., 2011). But it’s not just Notch. A poster presented at AD/PD 2011 by Yasuyuki Mitani and colleagues at Astellas Pharma in Tsukuba, Japan, pointed an accusing finger to an intermediate product of APP cleavage that had been the center of debate a decade ago but then faded from view. This is the γ-secretase substrate β-CTF, otherwise known as C99 or C100.

The Japanese group directly compared semagacestat (aka LY 450139), BMS-708163, and GSM-1, a modulator originally made by Mark Shearman and colleagues at Merck, which is not being clinically developed. GSM-1 does not affect Notch cleavage. They fed the three compounds to wild-type and to Tg2576 APP transgenic mice for one day to model acute dosing, and for eight days to model sub-chronic dosing. They measured what each drug did to the mice’s working memory in Y maze tests, and to APP cleavage biochemically. The scientists found that at one day, all three compounds reversed the memory deficits of the Tg2576 mice. By eight days of dosing, however, neither of the two inhibitors helped any longer, whereas the modulator still did. In the wild-type mice, sub-chronic dosing with both inhibitors—but not the modulator—actually impaired working memory. The Lilly and BMS compounds behaved about the same in this experiment. All three compounds lowered Aβ42 in the mice’s hippocampus by a similar amount. Importantly, in the eyes of the Japanese authors, however, only the inhibitors led to the expected dose-dependent rise of the γ-secretase substrate β-CTF. In contrast, with GSM-1, Aβ38 shot up as Aβ42 came down, but β-CTF remained absent.

These scientists concluded that when taken sub-chronically, both γ-secretase inhibitors (GSIs) but not the GSM, worsened cognition. The GSIs affect Notch with different therapeutic indices; both increase β-CTF. Partly for this reason, the Japanese group suggested that the cognitive impairment could be due to the β-CTF. Citing a German study in which Jochen Herms and colleagues had shown that LY 450139 decreases spine density in wild-type but not APP knockout mice (Bittner et al., 2009), the Japanese scientists implied that β-CTF elevation might have contributed to the worsening cognition seen in Lilly’s Phase 3 trial of this discontinued compound.

Debating this study, De Strooper pointed out a limitation. The poster only showed data on Aβ42, 40, and 38, not on other Aβ isoforms such as the shorter Aβ1-16, which has repeatedly been shown to increase in response to γ-secretase inhibition (see Portelius et al., 2010; Mustafiz et al., 2011; Beher et al., 2002) as well. Shearman mentioned a study in which three to seven days of treatment with MRK-560, a GSI originally developed in Shearman’s group at Merck, reversed an LTP deficit in Tg2576 mice (Townsend et al., 2010). Shearman is now at EMD Serono in Cambridge, Massachusetts. Schenk of Elan noted that the β-CTF could be at fault, as could a variety of other factors including other substrates γ-secretase is known to process. Lilly’s May agreed, saying of the Astellas poster, “This is interesting; we certainly see the increase in β-CTF, too.”

β-CTF is the product of BACE cleavage of APP. If this fragment were to turn out to be at fault, some scientists who toiled in the field a decade ago might well say, “Told you!” As early as 1997 and for some years thereafter, Rachael Neve, then at McLean Hospital in Belmont, Massachusetts, published a series of papers claiming that APP’s C-terminal fragment is toxic to neurons (e.g., Neve et al., 1992; Kammesheidt et al., 1992). Neve now studies mood disorders at MIT’s Picower Institute in Cambridge, Massachusetts.

Some scientists suggested that targeting inhibitors to presenilin-1 might be a way to make GSIs work. For example, Santiago Parpal Tamburini and colleagues at AstraZeneca presented on the GSI MRK-560. According to their poster, this inhibitor inhibits presenilin-1 much more potently than its cousin, presenilin-2. The compound caused severe side effects in PS2 knockout mice, but not in wild-type mice, the idea being that the latter deploy PS2 to provide whatever Notch and other substrate cleavage is necessary to stay healthy in the presence of a PS1-specific γ-secretase inhibitor. Other scientists were skeptical of this approach pending further data, and instead gave a nod toward γ-secretase modulation. For AD/PD 2011 reporting on GSMs, see Part 2 of this series.—Gabrielle Strobel.

This is Part 1 of a two-part series. See also Part 2.