Seven years after semagacestat failed in Phase 3 clinical trials for Alzheimer’s disease, researchers led by Masayasu Okochi at Osaka University, Japan, offer a new reason why. In the October 3 Cell Reports, the scientists claim that the γ-secretase inhibitor, developed by Eli Lilly & Co. to reduce Aβ in the brain, traps accumulating Aβ peptides within cells. Because γ-secretase activity typically correlates with release of Aβ42 into the extracellular space, researchers had assumed that semagacestat’s ability to drive down Aβ secretion reflected its block of the secretase. Now the authors suggest the drug instead alters a previously unknown function of γ-secretase, i.e., translocation of Aβ across cell membranes. They caution against using secreted Aβ to measure γ-secretase activity and suggest semagacestat failed because it did not work as expected. Bart De Strooper, director of the U.K. Dementia Research Institute, considers the paper a wake-up call for the field (see full comment below). Researchers at Lilly declined to comment for this article.

  • Researchers call semagacestat a “pseudo” γ-secretase inhibitor.
  • The drug allows Aβ and other APP peptides to accumulate in cells.
  • Semagacestat traps peptides in cell membranes.

“The study is important because it gives us new clues about the mechanism of γ-secretase activities,” said Lucía Chávez Gutiérrez at VIB/KU Leuven, Belgium. “To understand Alzheimer’s disease we need to look at all Aβ products generated, not just those that are released by cells.”

First co-authors Shinji Tagami and Kanta Yanagida decided to look inside and outside cells for γ-secretase products. They exposed neuron-like cells derived from human stem cells to 2 μM semagacestat—the highest concentration of the drug reported in spinal fluid in a Phase 1 clinical trial. As expected, secreted Aβ levels fell, while the β-secretase carboxy-terminal fragment of Aβ precursor protein (βAPP-CTF) rose, as determined by immunoprecipitation and western blot. Surprisingly, however, various Aβ species accumulated within the cells, including Aβ40, Aβ43, and Aβ46. “At first we thought we made a mistake,” said Okochi. “But no matter how many times we repeated it, we got the same result.” 

Sema Surprise. Cleaving βAPP-CTFs (top), γ-secretase releases Aβ and peptides 3-6 amino acids long. Transition-state analogs, called true-GSIs by the authors, reduce production and secretion of these γ-byproducts (bottom left). Paradoxically, semagacestat drives up Aβ within cells (bottom right). (Courtesy of Tagami et al., Cell Reports.)

The researchers then surveyed the collection of Aβ peptides, and the 3-6 amino-acid peptides γ-secretase clips off, in carboxypeptidase fashion, as it trims longer Aβ fragments (Okochi et al., 2013Takami et al., 2009Sep 2016 news). They used high-performance liquid chromatography and mass spectrometry. The scientists also examined neuroblastoma SH-SY5Y cells, human embryonic kidney (HEK) cells, and HEK cells whose Aβ levels were jacked up due to expression of APP KM670/671NL carrying the Swedish mutation, which readily undergoes β-secretase cleavage.

The researchers found many 3-6 amino acid peptides within the various cells, seven of which shot up after semagacestat treatment: VVI (Aβ38-41), IAT (Aβ40-43), TVI (Aβ42-45), VIV (Aβ43-46), VIT (Aβ45-48), ITL (Aβ46-49), and VITL (Aβ45-49) (see image above). Because these small peptides could, in theory, come from hundreds of other proteins by chance, Tagami and Yanagida confirmed they were products of γ-secretase by testing for them in HEK/APP KM670/671NL cells lacking presenilins 1 and 2, key components of the γ-secretase complex. As expected, these cells harbored very low levels of the tiny peptides, and their concentrations remained unchanged after semagacestat treatment.

The researchers also found that related γ-secretase inhibitors RO4929097, MK-0752, and avagacestat increased the small peptides as well as intracellular Aβ40, Aβ42/43, and Aβ45/46. These types of inhibitor may bind allosteric sites on presenilin 1 (Svedružić et al., 2013). However, yet another compound, L685,458, mimics the catalytic transition state, and it decreased both the tiny peptides and intracellular Aβ40-46. The authors concluded that whereas L685,458 acts as a true γ-secretase inhibitor (GSI), semagacestat and related compounds are pseudo inhibitors.

To test their findings in vivo, the researchers fed three 30 mg/kg of semagacestat at 12-hour intervals to PS1 I213T knock-in mice (Nakano et al., 1999) overexpressing APP KM670/671NL. They then measured Aβ in whole brain extracts by immunoprecipitation followed by western blot, and also measured the tiny peptides by mass spec. The researchers chose these knock-in mice, which produce more Aβ42 than wild-type, because they were readily available. Although most small peptides showed up at similar levels in the brains of treated versus untreated knock-ins, the two major ones, VIV (Aβ43-46) and ITL (Aβ46-49), increased in semagacestat-treated mice, as did Aβ1-x peptides ranging in size from 43 to 46 amino acids.

Next, the researchers examined γ-secretase activity in cell-free assays. When they mixed affinity-purified γ-secretase and βAPP-CTF, then added 10 μM of semagacestat, the drug blocked generation of the small, 3-6 amino acid peptides almost completely. But when they included a crude membrane fraction, activity fell by only 50 percent. Chávez Gutiérrez said the conformation and activity of γ-secretase, as well as its interaction with semagacestat, may vary depending on its association with cell membranes. In addition, the authors discovered more small peptides trapped in the membranes in the semagacestat-treated assays than in the controls, suggesting the drug retards the release of γ-products from membranes.

Okochi and Tagami acknowledge they do not fully understand their findings. They think γ-secretase executes two functions: its well-known protease activity and a newly proposed translocator function that ferries Aβ peptides from the membrane to the extracellular space. How semagacestat interferes with each of these remains unclear, they noted, but translocation inhibition could be particularly harmful since it results in accumulation of Aβ peptides inside cells. “This study should be taken seriously. Longer Aβ peptides are potentially more toxic than shorter ones,” said Chávez Gutiérrez.

De Strooper wrote that the work is a technical tour de force. “They show that semagacestat exerts unexpected and paradoxical effects on these intermediary peptides that are different from the effects of real loss of function of presenilin or from a γ-secretase inhibitor that targets the active site of the enzyme,” he wrote. Still, he agreed that the data are difficult to interpret. “As the effect of semagacestat on the initial epsilon endopeptidase cleavage and on the accumulation of these peptides is divergent, the explanation for the semagacestat effect must be complex. … I would have loved to see dose-response curves and time-course experiments to understand better where this pool of peptides is coming from and how they are degraded.”

Pinpointing where in the cell γ-secretase cuts APP and facilitates the exit of Aβ peptides may shed light on this question, he added. A recent study indicates the location of γ-secretase within cells, which varies depending on the presenilin protein associated with the γ-secretase complex, determines the ratio of Aβ42 to Aβ40 peptides produced and their intracellular versus extracellular fate (Jun 2016 news). 

Lili Zhang at Aquinnah Pharmaceuticals in Cambridge, Massachusetts, also said the findings were difficult to explain. Zhang led a group at Schering-Plough (before it merged with Merck) that profiled semagacestat extensively, because it was a major competitor in her efforts to develop γ-secretase inhibitors and modulators. “An obvious alternative interpretation of their data is that the processing of intracellular Aβ involves additional protease activit(ies),” she wrote to Alzforum (full comment below). When semagacestat blocks γ-secretase, other proteases may take over the task of proteolyzing the accumulated βAPP-CTF substrate, she hypothesized. Semagacestat would have no effect on presenilin knockout cells, as the authors observed, because, with no γ-secretase to inhibit, no acute accumulation of substrate would occur.

Zhang also noted the transition-state analogs that appear to act as “true” GSIs may be blocking not only γ-secretase, but other proteases that might also generate Aβ peptides. Indeed, pharmacological data from at least one such inhibitor, L685,458, suggests it targets aspartyl proteases more broadly than do non-transition analog inhibitors (Clarke et al., 2006). “I disagree with the conclusion of this paper and believe semagacestat is a potent, selective GSI that blocks both APP and Notch processing in vitro and in vivo,” Zhang wrote. She also noted that others have reported γ-secretase-independent processing of intracellular Aβ (Wilson et al., 2002). 

What do the findings say about the failed semagacestat Phase 3 IDENTITY trials (Aug 2010 news)? The authors suggested the semagacestat-induced intracellular buildup of Aβ peptides helps explain the worsening of dementia, but Alex Roher, Banner Health System, Phoenix, who reported enhanced accumulation of Aβ in the brain of a patient treated with semagacestat, was not so sure (Roher et al., 2014). “The results should be seriously considered, but the paradigm is far too distant from the complexity of the human brain,” he said.

The IDENTITY trials had other shortcomings (De Strooper, 2014Feb 2015 news). Semagacestat likely never reached levels high or steady enough within the brain to significantly engage γ-secretase, whereas in the periphery, levels were sufficient to block cleavage of potentially dozens of γ-secretase substrates, including Notch, causing side effects. “I do agree that the semagacestat trial did not truly test the amyloid hypothesis, but for reasons different from those suggested by this paper,” wrote Zhang.

What are the study’s implications going forward? “It should revive interest in the γ-secretases,” wrote De Strooper, who believes they are still the best validated drug target for treating AD (Voytyuk et al., 2017). 

Okochi and Chávez Gutiérrez, who share De Strooper’s enthusiasm about γ-secretases as therapeutic targets, noted that the field has been looking beyond inhibitors. “When semagacestat and other GSIs were developed, we had very limited knowledge of γ-secretase mechanisms,” said Chávez Gutiérrez. “I’m now more in favor of stabilizers of γ-secretase.” Based on her recent findings, she thinks chemical chaperones that steady the enzyme’s interaction with amyloid peptides could promote the generation of shorter fragments that are less toxic (Szaruga et al., 2017July 2017 news). Okochi is also interested in modifiers that would enhance the complete digestion of Aβ peptides.

Okochi and Tagami want to test additional secretase inhibitors for unexpected intracellular effects. “If pharma asks us to check, we are very willing to do so,” Okochi said. Their ultra-performance liquid chromatography/mass spec system can measure about 20 small peptides at once with high resolution and sensitivity. They also want to characterize γ-secretase’s translocator function. “No one has proposed before that a translocator malfunction could be involved in AD, but it might be a good target,” he said.—Marina Chicurel


  1. This paper in Cell Reports is a wake-up call for the field. It should revive interest in the γ-secretases, a field in Alzheimer’s research that got a severe blow some years ago when the failure of the semagacestat trial was announced (Doody et al., 2013). That trial marked the end of γ-secretase research in companies. As a consequence, fundamental research on these fascinating enzymes also fell out of fashion and it became very hard to get research funds to study them. This is short-sighted and a mistake (De Strooper, 2014). Luckily, a few stubborn academic groups continued to try to better understand the biology, structure, function, and pharmacological properties of this fascinating family of enzymes. The research group of Okochi and colleagues is one of them, and their new paper should revive interest in the γ-secretases/presenilins also from a pharmaceutical point of view. They expose flaws in the underlying pharmacology of semagacestat and suggest that the field jumped to conclusions with regard to the implications of the failed trial for the amyloid hypothesis. Arguably, γ-secretases remain by far the clinically best-validated targets for drug development in the fight against Alzheimer’s disease available at this moment (see our recent review where we explain why γ-secretases are better-validated drug targets than β-secretase: Voytuck et al., 2017). 

    Okochi and collaborators’ report is a technical tour de force. They provide a systematic analysis in different experimental conditions of all the intermediary peptides that are generated during the consecutive cleavages of APP by γ-secretase and provide a unique tool to assess this important, and still not fully understood, aspect of γ-secretase function. They show that semagacestat exerts unexpected and paradoxical effects on these intermediary peptides that are different from the effects of real loss of function of presenilin or from a real γ-secretase inhibitor that targets the active site of the enzyme. As the effect of semagacestat on the initial epsilon endopeptidase cleavage and on the accumulation of these peptides is divergent, the explanation for the semagacestat effect must be complex.

    The authors suggest that γ-secretase is responsible for the turnover/removal of the tri- and pentapeptides and that semagacestat affects that clearance mechanism, resulting in the paradoxical accumulation of these peptides in the cell. They find also in mouse models that semagacestat causes the accumulation of Aβ43-x peptides. I would have loved to see dose-response curves and time-course experiments to understand better where this pool of peptides is coming from and how they are degraded. One would expect that the high doses of semagacestat (as used in the paper) would block the generation of these peptides since the initial cleavage at the epsilon site is blocked. It is very hard to see how effects on γ-secretase alone could explain this accumulation. Can the authors exclude cellular effects on γ-secretase distribution (Sannerud et al., 2016) that are involved in the accumulation of these peptide pools? In follow-up work we certainly need to get a better understanding of where those peptides are located in the cell: Are they in the γ-secretase complex or in the membrane compartment around the complex or in specific subcellular compartments? How is γ-secretase involved in the removal of these peptides from that compartment?

    An important message of the paper is that the field has jumped to conclusions when interpreting the failure of the semagacestat trial. I made that point before (De Strooper, 2014). From the Doody et al. paper it was already clear that semagacestat had aberrant effects on γ-secretase: In blood measurements it actually increased the Aβ42 to Aβ40 ratio, which a classical inhibitor would not do. In my opinion, the drug failed mainly because of pharmacokinetic reasons, however. The dosage of semagacestat was such that side effects were maximized and therapeutic effects minimized. The result was that the steady-state levels of Aβ in the brain were not affected at all (De Strooper, 2014). This trial never tested the amyloid hypothesis in a serious way. On top of this criticism, the work of Okochi et al. now clarifies the unknown effects of semagacestat on alternate peptides that may have pathological effects on their own. 


    . A phase 3 trial of semagacestat for treatment of Alzheimer's disease. N Engl J Med. 2013 Jul 25;369(4):341-50. PubMed.

    . Lessons from a failed γ-secretase Alzheimer trial. Cell. 2014 Nov 6;159(4):721-6. PubMed.

    . Modulation of γ- and β-Secretases as Early Prevention Against Alzheimer's Disease. Biol Psychiatry. 2017 Aug 10; PubMed.

    . Restricted Location of PSEN2/γ-Secretase Determines Substrate Specificity and Generates an Intracellular Aβ Pool. Cell. 2016 Jun 30;166(1):193-208. Epub 2016 Jun 9 PubMed.

  2. This is a comprehensive study that further shows the complexity of the processing of APP and Aβ. The authors have used several different cell lines and a mouse model to show that the γ-secretase inhibitor (GSI) semagacestat may not act as previously believed.

    The obvious question is if the findings in the paper can be translated to humans. For example, Bateman and colleagues have shown that semagacestat lowers the production of CNS Aβ in a dose-dependent manner. We have shown that treatment with the GSIs semagacestat and avagacestat increases the levels of CSF Aβ1-15/16 while Aβ1-34 decreases. In another set of experiments we showed that cells treated with the GSI DAPT (at a high dosage) greatly increased Aβ1-15/16, while all longer Aβ species from Aβ1-17 up to Aβ1-42 disappeared (even though many of the peptides cannot be substrates for γ-secretase). However, we don’t know what happened with the intracellular Aβ.    

    In conclusion, there are still many questions left to fully understand the generation of Aβ and the involved enzymes.

  3. Based on the observations that semagacestat inhibited only secreted Aβ but not intracellular or Aβ byproducts, Tagami, et al. suggested that semagacestat is only a “pseudo” γ-secretase inhibitor (GSI). However, an obvious alternative interpretation of their data is that the processing of intracellular Aβ involves additional protease activit(ies) other than γ-secretase. L-684,458, labeled as a “true GSI” in this paper, and semagacestat represent transition state and non-transition state GSIs, respectively. The distinct pharmacological profiles have been described many years ago (Clarke et al., 2006). The fact that the Bmax of L-684,458 is twofold of that for non-transition state GSIs suggests that L-684,458 binds to additional aspartyl proteases in addition to γ-secretase. L-684,458 blocks broader protease activity than non-transition state GSIs. I disagree with the conclusion of this paper and believe semagacestat is a potent, selective GSI that blocks both APP and Notch processing in vitro and in vivo.

    I do agree that the semagacestat trial did not truly test the amyloid hypothesis, but for reasons different from that suggested by this paper. The pharmacological property of semagacestat, combined with poor brain penetration and dose-limiting toxicity, resulted in a lack of target engagement, i.e., there was no significant CSF Aβ reduction in patients treated with the drug.     

  4. This is a very interesting study lending additional support for the hypothesis that increased intracellular Aβ is a major culprit of neurodegeneration and cognitive decline, while extracellular pool of Aβ plays relatively minor role.

    One obvious conclusion of the study is that future cell-based assays in screening for Aβ-lowering drugs should include quantification of intracellular pool of AβA as their readout.

    It will be interesting to see if semagacestat causes intracellular accumulation of Aβ by affecting the function of intracellular Aβ-degrading proteases (i.e., insulin-degrading enzyme, IDE) through their direct inhibition or blocking their access to Aβ by impairing their trafficking.

  5. This is a very important paper, showing the unexpected result that semagacestat causes the accumulation of longer Aβ species and short peptide byproducts that are produced by the sequential carboxy-terminal trimming activity of γ-secretase.

    The paper highlights that we still have not fully understood the complex processing of C99 by γ-secretase at the mechanistic level. Now, with these findings by Tagami et al., it is getting even more complex. The data suggest that semagacestat is not inhibiting the enzyme but rather altering its processivity, leading to an accumulation of longer Aβ species such as Aβ46 and Aβ43. It is possible that C99 could perhaps not bind anymore if Aβ46 and Aβ43 would not be released from γ-secretase, leading to pseudoinhibition and reduction of Aβ and AICD as a secondary event. Likewise, byproducts that are not leaving the membrane potentially could also be inhibitory if they stay bound to the enzyme or alternatively if they flood the membrane and thus compete with C99 binding. It is possible that in particular the accumulation of intracellular long Aβ may add to the likely membrane-toxicity of uncleaved C99 that accumulates in the presence of semagacestat, although the amount of long Aβ, such as Aβ46, seems very minor as compared to that of C99.

    Interestingly, earlier studies also showed that the γ-secretase inhibitor DAPT causes the accumulation of longer Aβ45/46 inside cells, now indicating pseudoinhibition of this compound as well (Zhao et al., 2004; Qi-Takahara et al., 2005). However, it remains puzzling that in cell-free γ-secretase cleavage assays with CHAPSO-solubilized enzyme accumulation of such longer Aβ species in the presence of DAPT has not been observed. Phosphatidylcholine membranes are typically added in such assays that could take up accumulating longer Aβ peptides (Kakuda et al., 2006). However, DAPT clearly inhibits the formation of AICD in these assays as well as the carboxy-terminal processing to Aβ (Kakuda et al., 2006), and so does semagacestat (Chávez-Gutiérrez et al., 2012). 

    Why the short-peptide byproducts can accumulate in the membrane also remains mysterious. Tagami et al. now propose a new additional “pump-like” activity function of γ-secretase that releases the byproducts from the membrane, with which semagacestat interferes. However, intuitively, one would assume that the affinity of the 3–5 amino acid peptide byproducts will be so low that they should be released immediately into the hydrophilic space after they are generated, especially as we know from the structural studies that the active site of γ-secretase is in a solvent exposed cavity.

    Clearly, more studies are now needed to understand how inhibitors such as semagacestat cause the observed effects. 


    . The mechanism of γ-Secretase dysfunction in familial Alzheimer disease. EMBO J. 2012 May 16;31(10):2261-74. Epub 2012 Apr 13 PubMed.

    . Equimolar production of amyloid beta-protein and amyloid precursor protein intracellular domain from beta-carboxyl-terminal fragment by gamma-secretase. J Biol Chem. 2006 May 26;281(21):14776-86. PubMed.

    . Longer forms of amyloid beta protein: implications for the mechanism of intramembrane cleavage by gamma-secretase. J Neurosci. 2005 Jan 12;25(2):436-45. PubMed.

    . Identification of a new presenilin-dependent zeta-cleavage site within the transmembrane domain of amyloid precursor protein. J Biol Chem. 2004 Dec 3;279(49):50647-50. PubMed.

Make a Comment

To make a comment you must login or register.


Therapeutics Citations

  1. Semagacestat
  2. Avagacestat

News Citations

  1. γ-Secretase Gives Up Secrets About Its Cleavage
  2. Lodged in Late Endosomes, Presenilin 2 Churns Out Intraneuronal Aβ
  3. Lilly Halts IDENTITY Trials as Patients Worsen on Secretase Inhibitor
  4. Semagacestat Failure Analysis: Should γ-Secretase Remain a Target?
  5. sAPP Binds GABA Receptor, and More News on APP

Mutations Citations

  1. APP KM670/671NL (Swedish)
  2. PSEN1 I213T

Paper Citations

  1. . γ-secretase modulators and presenilin 1 mutants act differently on presenilin/γ-secretase function to cleave Aβ42 and Aβ43. Cell Rep. 2013 Jan 31;3(1):42-51. PubMed.
  2. . gamma-Secretase: successive tripeptide and tetrapeptide release from the transmembrane domain of beta-carboxyl terminal fragment. J Neurosci. 2009 Oct 14;29(41):13042-52. PubMed.
  3. . Modulators of γ-secretase activity can facilitate the toxic side-effects and pathogenesis of Alzheimer's disease. PLoS One. 2013;8(1):e50759. Epub 2013 Jan 7 PubMed.
  4. . Accumulation of murine amyloidbeta42 in a gene-dosage-dependent manner in PS1 'knock-in' mice. Eur J Neurosci. 1999 Jul;11(7):2577-81. PubMed.
  5. . Intra- or intercomplex binding to the gamma-secretase enzyme. A model to differentiate inhibitor classes. J Biol Chem. 2006 Oct 20;281(42):31279-89. PubMed.
  6. . Presenilins are not required for A beta 42 production in the early secretory pathway. Nat Neurosci. 2002 Sep;5(9):849-55. PubMed.
  7. . Neuropathological and biochemical assessments of an Alzheimer's disease patient treated with the γ-secretase inhibitor semagacestat. Am J Neurodegener Dis. 2014;3(3):115-33. Epub 2014 Dec 5 PubMed.
  8. . Lessons from a failed γ-secretase Alzheimer trial. Cell. 2014 Nov 6;159(4):721-6. PubMed.
  9. . Modulation of γ- and β-Secretases as Early Prevention Against Alzheimer's Disease. Biol Psychiatry. 2017 Aug 10; PubMed.
  10. . Alzheimer's-Causing Mutations Shift Aβ Length by Destabilizing γ-Secretase-Aβn Interactions. Cell. 2017 Jul 27;170(3):443-456.e14. PubMed.

Further Reading

No Available Further Reading

Primary Papers

  1. . Semagacestat Is a Pseudo-Inhibitor of γ-Secretase. Cell Rep. 2017 Oct 3;21(1):259-273. PubMed.