Eli Lilly and Company announced yesterday that it has halted its Semagacestat γ-secretase inhibitor program. Preliminary results from two ongoing Phase 3 trials (the IDENTITY and IDENTITY-2 trials) revealed that the drug not only failed to slow cognitive decline in people with mild to moderate Alzheimer disease, but that it actually made them worse (see company press release). Cognition as measured by the ADAS-cog and Activities of Daily Living Scale declined faster in volunteers in the treatment arms compared to those on placebo. “Obviously, we are clearly disappointed about the results for patients, their caregivers, and everyone else,” said Eric Siemers, Medical Director, Alzheimer's Disease Team at Lilly, in an interview with ARF.

“The billion-dollar questions on everyone's mind are whether this is a body blow to the amyloid theory and what this means for all the planned prevention trials using similar drugs,” suggested Murali Doraiswamy, Duke University Medical Center, Durham, North Carolina (see full comment below). Siemers said that there are numerous explanations for the outcome, among them being the possibility that lowering Aβ in the brain makes cognition worse. That would be a difficult pill to swallow for supporters of the amyloid hypothesis, and would raise questions about pursuing other γ-secretase inhibitors and BACE inhibitors as therapies for AD. “I would be surprised if that was the explanation, but at this point we really don’t know,” said Siemers.

Other possible explanations for the accelerated deterioration envisage roles for C-terminal fragments of APP in cognition. Blocking γ-secretase would lead to an increase in the β-secretase product C99, and a decrease in the APP intracellular domain (AICD), which is cleaved from C99 by γ-secretase. “Perhaps the increase in C99 may have some deleterious effect, or maybe decreasing AICD has an adverse effect,” suggested Siemers. Either scenario would raise questions about pursuing any type of γ-secretase inhibitor, trials of which are currently ongoing.

Alternatively, the disappointing trial outcome may have nothing to do with APP or Aβ. “The general concern about γ-secretase inhibitors has been mechanism-based toxicity,” according one of the trial site investigators, who wished to remain anonymous. “The factor(s) that led to faster clinical decline need to be understood in order to guide progress with secretase inhibition or modulation, or, more broadly, anti-amyloid treatment in the future.”

Toxicity concerns stem from γ-secretase’s penchant for processing a multitude of substrates. In particular, researchers in the field have tried to develop modulators that, while lowering production of the more toxic Aβ42, do not block cleavage of Notch, a crucial signaling molecule that regulates the fate of a wide variety of cell types. Semagacestat is not “Notch sparing,” and even though the doses used in the IDENTITY trials (140 mg per day) were conservatively chosen to reduce Aβ production by about 25 percent daily (see ARF related news story), Notch signaling may have suffered. “This needs close investigation, especially since the increase in skin cancer in the treated patients also suggests Notch signaling inhibition,” suggested Bart De Strooper, KU Leuven, Belgium. “The implication is that we should explore now, even more, Notch- (and other substrate-) sparing modulation of γ-secretase activity (e.g., blocking selectively Aph1B-γ-secretase).” (See full comment below.)

Siemers suggested that one positive thing that came from the trials was a justification for the rationale of using biomarker analysis to determine adequate dosing. In Phase 2 trials, the company used CSF analysis to determine whether the drug got into the brain and had an effect (see ARF related news story and ARF news story). “The fact that people were worse shows that we did get [the drug] into the brain and have an effect,” said Siemers. “Obviously it is not the effect that we wanted, but in a sense it tells you that this biomarker strategy does help you make some conclusions.”

Siemers said that Lilly will eventually release more data from the trials, which together enrolled more than 2,600 patients worldwide. Dosing has stopped, but the company plans to follow patients for at least another six months. Lilly has also been going through a major restructuring that has led to some downsizing (see related IndyStar.com news).—Tom Fagan.


  1. First of all, kudos to Lilly and DSMB—they did the right thing to make this public as soon they knew. As a disclosure, I have been an advisor to and received grants from many companies, including Lilly.

    The IDENTITY findings are not entirely unexpected, since I thought the Phase 1 studies didn't have a clear signal, and adverse effects on skin cells and the immune system were always a consideration with this class of drugs. The interim analyses suggest the trials likely would have been not only clearly negative, but also have potentially harmed patients, both in terms of progression and side effects. Since risks exceeded benefits, stopping the trial is the right thing to do. I hope Lilly plans to offer free follow-up to all subjects to fully understand the side effects and ensure proper aftercare and safety.

    The billion-dollar questions on everyone's mind are whether this is a body blow to the amyloid theory and what this means for all the planned prevention trials using similar drugs. I think the amyloid theory is still valid, but this clearly tells us that our current views may be too simple—clearing amyloid at a late stage without affecting tau might not suffice as a cognitive enhancing treatment. Other trials have told us this before, but this one really drives it home, since the results seem pretty unambiguous. One thing that worries me about these findings is that we don't know if this was a toxicity unique to Lilly's drug given to a late stage population or whether it also applies to similar anti-amyloid therapies give at earlier stages of the disease. There are several other anti-amyloid trials that are set to report in the next few months. These data are going to help us better understand how useful biomarkers are as surrogate endpoints in treatment trials. But in terms of biomarkers for accurate diagnosis, I don't think these data will have as much of an impact. Since we define AD (and preclinical AD) on the basis of pathology, we will continue to need amyloid and tau biomarkers for enhancing diagnostic accuracy, staging disease, and predicting future risk.

    Negative studies can be very informative as long as people put aside their preconceived notions and actually listen to what the data are telling them. The IDENTITY dataset has the potential to really help the field as it moves to design studies targeting people with MCI or preclinical AD. So it will be a win-win if Lilly makes this entire trial dataset public (not just placebo data, but all data in the drug-treated arms) so others don't repeat the same mistakes over and over. I would also like to see the field pull together a conference (with experts from other areas such as decision science) to examine all the data from the dozen or so anti-amyloid trials done to date to ensure we are all seeing the correct big picture and to optimize the well-being of participants in future trials.

    I look forward to seeing these data more fully, and my comments are to be interpreted in that context.

  2. Obviously, this is a serious disappointment. The negative effects on cognition and on the ability to complete activities of daily living are very bad news, and should raise questions about the usefulness of targeting γ-secretase for AD. However, this compound is not Notch sparing. The question is, To what extent Notch inhibition can explain the phenotype (Notch has been implicated in memory formation in mice and Drosophila). This needs close investigation, especially since the increase in skin cancer in the treated patients also suggests Notch signaling inhibition. The main conclusion is that there is apparently no such thing as a therapeutic window for classical γ-secretase inhibitors. The implication is that we should explore now, even more, Notch- (and other substrate-) sparing modulation of γ-secretase activity (e.g., blocking selectively Aph1B-γ-secretase). The other message, given the possibility that once Aβ has triggered tau pathology the disease becomes Aβ independent, is that we probably need to treat patients early, before Aβ gets the chance to do its destructive work in the brain. Probably a more healthy brain will be more able to cope with the possible side effects of anti-amyloid drug as well.

  3. There is now a fair bit of evidence that γ-secretase inhibitors do cleave a number of additional growth factors and cause related side effects, not just on Notch signaling. I agree with the colleagues who point out that the negative effects of this “non-Notch sparing” γ-secretase inhibitor is not unexpected at all.

    One issue is that we most likely do not even know all the substrates for γ-secretase, so that the development of “safe” inhibitors is questionable. It also would assume the existence of a wide range of different γ-secretases that are specific for only some of the substrates (e.g., only for APP) and have separate pharmacological profiles and can be inhibited individually. This does not seem very likely.

  4. Semagacestat continues the streak of late-phase trial failures. All failed for lack of efficacy, but as enzyme inhibitors and antibodies are advanced, serious toxicity in the trials may become more frequent as well. Sequential failures discovered late in the drug development process are demoralizing, enervating, and may cause broad swaths of the community to lose interest. We should not have to enroll 2,000 to 4,000 patients in Phase 3 programs in order to discover that a drug is ineffective, the dose range is wrong, or that it is too toxic. In traditional drug development, large Phase 2b and 3 trials are considered confirmatory trials of efficacy and safety that were previously established in earlier proof-of-concept and dose-finding phases. Proof-of-concept studies sometimes are designed to demonstrate target engagement when there are validated drug targets for the illnesses. We don’t have validated targets in AD; rather we have lots of potential targets.

    Current Phase 2a AD trials are either designed as mini-Phase 3 trials, with too few patients to meaningfully assess efficacy, or as even smaller “pharmacodynamic” studies intended to show a particular, expected CNS effect. Semagacestat development used the latter pharmacodynamic model to justify its push to Phase 3, demonstrating effects on plasma Aβ, but not quite clearly on CSF Aβ, in a placebo-controlled, 14-week trial with 51 patients. (Its effect on Aβ might be considered evidence of target engagement if γ-secretase is a valid target for AD.)

    By comparison, and as examples, tarenflurbil (Myriad and Lundbeck), bapineuzumab (Elan, Johnson & Johnson, and Pfizer), and scyllo-inositol (Elan and Transition) development programs used more of a “mini-Phase 3” approach, recruiting about 210 to 350 patients and treating from 1.0 to 1.5 years. None of these studies showed overall efficacy, but they were interpreted as showing potential efficacy in “subgroups,” and showing sufficient safety and enough dosing information that large Phase 3 programs involving 2,600 to 4,000 patients were undertaken or planned. So far, tarenflurbil failed Phase 3; the bapineuzumab program is ongoing but was modified; and a scyllo-insotitol trial is being planned. There are other examples as well.

    Current development programs for AD drugs are not very efficient or effective. AD drug development is too consumptive of human capital to continue this way. There are too many plausible, candidate drugs for us to carry out Phase 3 after Phase 3 trial based on wisps of clinical evidence. How we do early human phase development, decide whether or not to advance drugs to later phases in MCI, AD or prevention, and how we do the later phase trials need to be re-conceptualized, but this is a hugely formidable endeavor.

    Lilly and the ADCS, who conducted the early semagacestat study, understood the dilemma and the risks, as they presciently wrote in 2008:

    “Given the slow rate of clinical progression in AD, we did not expect to see drug effects on measures of cognition or ADLs in this 14-week trial. Without this, a full risk-benefit assessment cannot be made. This trial sufficiently demonstrates that LY450139 can be tolerated, although not without risk. Given the potential for disease-modifying effects of this Aβ-lowering agent, and the arguably acceptable tolerance and safety profile of LY450139 demonstrated in this study, further large-scale efficacy trials are justified. Based, in part, on the results of this Phase 2 study, Eli Lilly & Co. launched a multinational Phase 3 trial in the second quarter of 2008, with an enrollment goal of 1,500 patients with AD.”


    . Phase 2 safety trial targeting amyloid beta production with a gamma-secretase inhibitor in Alzheimer disease. Arch Neurol. 2008 Aug;65(8):1031-8. PubMed.

  5. The observed accelerated cognitive deterioration following γ-secretase inhibition could be due to toxic effects on synapses. By performing in vivo two-photon imaging, we have recently shown that dendritic spines (i.e., the post-synaptic side of excitatory synapses) get irreversibly lost in the cerebral cortex of wild-type mice after applying the Lilly drug (Bittner et al., 2009). By repeating the experiments in APP-deficient mice, we revealed evidence that APP-cleavage products (probably an accumulation of C-terminal fragments, as also suggested by Tom Fagan above) are critically involved.


    . Gamma-secretase inhibition reduces spine density in vivo via an amyloid precursor protein-dependent pathway. J Neurosci. 2009 Aug 19;29(33):10405-9. PubMed.

  6. Negative outcomes in trials are always bad news, since patients don’t get access to new therapeutic interventions and we, as researchers, are forced to rethink our basic assumptions. Nevertheless, trials are an immensely rich resource that can lead to new knowledge about this disease. Congratulations to Lilly for committing to make these data public.

    It will be of great interest to observe whether the patients can partially reverse their cognitive deficit after halting the medication in this trial. This could be a first indication whether the unintended effects of γ-secretase inhibition are real, irreversible toxicity issues (such as affecting other substrates, as suggested by other researchers), or merely a lowering of amyloid-β levels, which drives normal cognitive processes out of their optimal dose window, or a combination of the two. In this regard, it is interesting to consider the inverse U-shape response of glutamatergic functioning to amyloid-β (Abramov, 2009), which might modulate the balance between excitatory and inhibitory activity of brain networks involved in cognitive aspects, such as working memory and executive control. The glutamatergic physiology is extremely tightly regulated, and too much reduction of amyloid-β could drive the system beyond its optimal point and change the emergent properties of the network sufficiently to be detected clinically.

    This theory would predict that, after halting the medication, the cognitive deficit could be partially reversed and the degree of cognitive change would be correlated in a complex way to the change in amyloid-β level for each individual patient. This can readily be investigated in the database, although confounding parameters such as the baseline cognitive performance, different co-medications, and unknown genotypes need to be taken into account. Using well-calibrated, computer-based models of biophysically realistic, complex brain networks could be a way to address these issues, so that the valuable information from this trial can increase our knowledge on this devastating disease.


    . Amyloid-beta as a positive endogenous regulator of release probability at hippocampal synapses. Nat Neurosci. 2009 Dec;12(12):1567-76. PubMed.

  7. Comment by George Perry, Rudy J. Castellani, and Mark A. Smith

    Failure of Amyloid-β-based Therapeutics: Is Amyloid the Culprit or the Savior?
    The failure of the amyloid-β-based therapeutic, semagacestat, to benefit patients suffering from Alzheimer disease (Lilly, 17 August 2010), when seen together with poor results from amyloid-β vaccine trials, is not a reason for further refinement of the amyloid cascade hypothesis. Instead, it is a reason to discard and replace the amyloid hypothesis with one that can lead to effective treatment. Failure of the amyloid-β-targeted therapeutics is not due to a lack of effectiveness, since amyloid-β is cleared from the brain by the AN1792 Aβ vaccine (Homes et al., 2008). Further, the lack of clinical efficacy is unlikely a consequence of antibodies only removing visible deposits, because amyloid, like all known bio-polymers, exists in equilibrium with soluble and oligomeric intermediates, meaning that all are reduced by the vaccine. Instead of being the culprit of Alzheimer disease, we believe amyloid-β marks cellular stress, and its deposits are secondary to cellular death.

    Amyloid-β increases during brain injury; in fact, it is one of the best markers of axonal damage (Cochran et al., 1991). During Alzheimer disease, and in every other condition studied, amyloid-β increase is correlated directly with a decrease in oxidative stress (Nunomura et al., 2001, 2010). Redox silencing of copper may be the mechanism for the reduction of oxidative stress (Hayashi et al., 2007), and there is strong data supporting amyloid-β as the dominant antioxidant of cerebrospinal fluid (Kontush et al., 2001).

    The amyloid cascade hypothesis is fundamentally an anti-Darwinian idea, because it suggests that humans have injury responses that, on balance, cause further injury. What selection pressure would create such an elaborate system of self-destruction in the best-adapted organism on Earth? This is why we have long argued that the vaccine and other amyloid-targeted drugs were likely to fail due to a poor understanding of the basics of biology (Perry et al., 2000; Smith et al., 2002). The argument is not that amyloid-β is unimportant to Alzheimer disease, for it is central, but that causality and centrality are not equivalent (Castellani et al., 2009). Confusion on this point has plagued the meaning of familial Alzheimer disease for almost two decades, ever since genetic mutations associated with alteration in amyloid-β production were associated with Alzheimer disease. Quite simply, the same association would be made whether amyloid-β was critical in stopping the disease or causing it. Unfortunately, the field went “all in” and bet the house on the latter. Mendelian genetic linkage is not equivalent to biological causality. Pharmacological modulation of the amyloid-β response may benefit patients; eradication, as proposed for over a decade, will not. Understanding the biology of amyloid-β offers multiple new avenues for treatment of those devastated by this disease and new hope for a cure.

    See also: Lilly, 17 August 2010


    . Amyloid precursor protein and ubiquitin immunoreactivity in dystrophic axons is not unique to Alzheimer's disease. Am J Pathol. 1991 Sep;139(3):485-9. PubMed.

    . Intraneuronal amyloid beta accumulation and oxidative damage to nucleic acids in Alzheimer disease. Neurobiol Dis. 2010 Mar;37(3):731-7. PubMed.

    . Lipid peroxidation and 4-hydroxy-2-nonenal formation by copper ion bound to amyloid-beta peptide. Free Radic Biol Med. 2007 Dec 1;43(11):1552-9. PubMed.

    . Amyloid-beta junkies. Lancet. 2000 Feb 26;355(9205):757. PubMed.

    . Long-term effects of Abeta42 immunisation in Alzheimer's disease: follow-up of a randomised, placebo-controlled phase I trial. Lancet. 2008 Jul 19;372(9634):216-23. PubMed.

    . Amyloid-beta is an antioxidant for lipoproteins in cerebrospinal fluid and plasma. Free Radic Biol Med. 2001 Jan 1;30(1):119-28. PubMed.

    . Oxidative damage is the earliest event in Alzheimer disease. J Neuropathol Exp Neurol. 2001 Aug;60(8):759-67. PubMed.

    . Predicting the failure of amyloid-beta vaccine. Lancet. 2002 May 25;359(9320):1864-5. PubMed.

    . Reexamining Alzheimer's disease: evidence for a protective role for amyloid-beta protein precursor and amyloid-beta. J Alzheimers Dis. 2009;18(2):447-52. PubMed.

  8. Much was talked about in the past month after the announcement of the apparent failure of the IDENTITY clinical trials. The comments on this website tried to explain why the drug Semagacestat, designed to inhibit the proteolytic activity of γ-secretase, thus diminishing the production of Aβ, had no beneficial effect on patients with Alzheimer disease (AD). The lack of selectivity of inhibition of γ-cleavage caused by this drug was often cited, especially since the publication of a report from the Greengard lab (1) on a novel protein that specifically regulates the cleavage of amyloid-β precursor protein (APP) by γ-secretase.

    We propose an alternative explanation, which—unfortunately—dampens the enthusiasm even for highly specific blockers of APP cleavage (such as the recently announced small molecule modulators of γ-secretase [2]) as therapeutic agents. We point to the possibility that the simple prevention of cleavage of APP by γ-secretase could be detrimental to neurons. From a less accepted point of view, a major contributing factor to AD could be the loss of function of APP and/or of its fragments derived from cleavage by secretases. It is possible that the physiological function of APP could require its cleavage (3), and by preventing this cleavage, the function of APP could be hindered, thus leading to neuronal pathology. In this case, the inhibition of secretase—be it β- or γ-secretase—would not correct the pathology of AD, although it would diminish the load of Aβ.


    . Gamma-secretase activating protein is a therapeutic target for Alzheimer's disease. Nature. 2010 Sep 2;467(7311):95-8. PubMed.

    . Modulation of gamma-secretase reduces beta-amyloid deposition in a transgenic mouse model of Alzheimer's disease. Neuron. 2010 Sep 9;67(5):769-80. PubMed.

    . The cleavage products of amyloid-beta precursor protein are sorted to distinct carrier vesicles that are independently transported within neurites. J Neurosci. 2009 Mar 18;29(11):3565-78. PubMed.

Make a Comment

To make a comment you must login or register.


News Citations

  1. Studies Reveal New Hope, Old Problems With AD Biomarkers
  2. Sweet 16: Novel APP Processing Pathway and a New Biomarker?

External Citations

  1. Eli Lilly and Company
  3. company press release
  4. related IndyStar.com news

Further Reading