A massive transmembrane complex, γ-secretase churns out the infamous peptides that can start a person on the path in Alzheimer’s disease. Aβ is produced when the secretase shaves three residues at a time from APP fragments within cell membranes. What controls this process? A cryo-EM study published June 7 in Science elucidates the mechanism.

  • Cryo-EM resolves γ-secretase complexed with APP-C99, Aβ46, Aβ49, and Aβ43.
  • During processing, an α-helix unwinds by one turn, while a three-residue β-strand positions it for the next cut.
  • This mechanism aligns with the so-called “piston” model of γ-secretase processing.

Researchers led by Rui Zhou of Beijing’s Tsinghua University and Yigong Shi of Westlake University in Hangzhou resolved the structure of the secretase in complex with various Aβ peptides, including Aβ49, Aβ46, and Aβ43. The peptides lodged within a transmembrane channel of the enzyme, where each twisted into the same configuration: an α-helix connected via a short linker to a β-strand. For each successive cut, the substrate slid three residues, equivalent to one turn of the helix, deeper into the channel, while the C-terminus formed a hybrid β-strand with the enzyme that held the peptide in place for cleavage. In this manner, the secretase lops off three residues at a time.

“This new study is another tour de force in structure elucidation of the γ-secretase complex from [these investigators],” commented Michael Wolfe of the University of Kansas in Lawrence. “The snapshots of the protease bound to APP-derived [peptides] reveal critical details that, taken together, provide important insight into the mechanism of processive proteolysis.”

The study adds to prior work from Shi’s group that has brought γ-secretase into ever-clearer focus (Dec 2012 news; May 2015 news; Aug 2015 news).

Comprising the four subunits nicastrin, APH-1, PEN-2, and presenilin 1 or 2, the enzyme complex crisscrosses the cell membrane 20 times. Its catalytic heart beats within a channel formed by the sixth and seventh transmembrane domains of the presenilins. After β-secretase slices APP to leave a 99-amino-acid transmembrane fragment, aka β-CTF, γ-secretase goes to work. First using its endopeptidase activity, it clips C99 into either Aβ48 or Aβ49. Then, via processive, carboxypeptidase activity, it sequentially snips three to four residues from the C-terminus of those peptides. Depending on the endopeptidase cut, two production lines emerge: Aβ48 to 45 to 42, or Aβ49 to 46 to 43 to 40. Studies have demonstrated that familial AD mutations in the presenilins hobble both lines, leading to a glut of longer, more amyloidogenic peptides such as Aβ42 and Aβ43 and a dearth of Aβ38 and 40 (Apr 2019 news).

How does γ-secretase manage to precisely trim by three to four residues at a time? To find out, first author Xuefei Guo and colleagues resolved the structure of the enzyme in complex with peptides of the Aβ49 production line, including APP-C99, Aβ49, Aβ46, and Aβ43. To capture each substrate in the clutches of the enzyme, the researchers used trial and error to introduce cysteine residues in each peptide, and in the nicastrin subunit, that would allow them to cross-link the pair without distorting the substrate in the active site. Nicastrin helps capture and deliver substrates to the catalytic core of γ-secretase (Aug 2015 newsDec 2015 news). To hold the pair in an intermediate state, the researchers introduced a catalysis-quashing D385N mutation into PS1’s catalytic site.

First, the scientists resolved the C99 complex at 2.9 Å resolution. They spied residues 31-55 to be nestled within the catalytic transmembrane channel of PS1. The segment’s N-terminus had an α-helix that lay next to a three-residue linker sequence, connected to a five-residue β-strand at the C-terminus. Hydrogen bonds between the α-helix and PS1’s transmembrane strand held the substrate in place. Meanwhile, the C99 β-strand coupled with two β-strands from PS1 to position the scissile peptide bond, aka the bond about to be snipped, in the linker between residues 49 and 50. This would produce Aβ49 and C50-99, aka the APP intracellular domain, or AICD.

First Cut. A bit of APP-C99 (orange) lodges into the transmembrane catalytic channel of PS1 (blue, left). The substrate’s α-helix connect to a β-strand via a linker (middle). This β-strand forms a hybrid with PS1 (close-up, right). [Courtesy of Guo et al., Science, 2024.]

To see how the structure then changes to accommodate the next, carboxypeptidase cut, the scientists next resolved the Aβ49 complex. Lo and behold, a similar scene emerged. A segment of Aβ49 formed an α-helix connected to a C-terminal β-strand. Relative to the positioning of C99, the Aβ49 helix was buried by three amino acids, or one turn of the helix, deeper into the transmembrane cavity of PS1. It was held in place by hydrogen bonding with the same two residues from PS1 that had cradled C99’s helix. The linker and β-strand shifted by three residues as well, such that the three residues that had formed the linker of C99 now folded into a β-strand in Aβ49, and the five residues that had folded into a β-strand in C99 were released from the enzyme’s grip. Aβ49’s short β-strand hooked up with one β-strand from PS1, not two. This positioning poised the substrate for cleavage between V46 and I47, which would yield Aβ46.

Next on the Block. In the clutches of γ-secretase (left), Aβ49 (purple) forms an α-helix (right) and a three-residue β-strand that pairs with PS1 to position the peptide for a chop after residue 46 (red arrow). The helix slides one turn deeper into the catalytic cleft relative to the position of the C99 helix. [Courtesy of Guo et al., Science, 2024.]

You guessed it: A nearly identical structure emerged in the Aβ46 complex. An N-terminal α-helix, three amino acids deeper into the catalytic cleft, was held in place by the same two residues of PS1, while a new C-terminal three-residue β-strand formed a hybrid β-sheet with PS1. With the bond between T43 and V44 now splayed across the PS1 chopping block, the enzyme was primed to produce Aβ43.

Try as they might, the researchers were unable to resolve an Aβ43 peptide complexed with γ-secretase at atomic resolution. However, at 4.45Å resolution, they made out the characteristic α-helix and hybrid β-strand structure, suggesting Aβ43 was likely cleaved in the same way as its predecessors.

The findings support the so-called “piston” model of γ-secretase processing, whereby the whole substrate edges closer to the jaws of the PS1 catalytic site after each cleavage (Jan 2019 news). The size of each bite is dictated by the pitch of the α-helix, which is three amino acids long. The authors think the enzyme likely works this way for the other Aβ production line, leading from Aβ48 to Aβ42, as well as for other γ-secretase substrates, such as Notch-1.

One Turn at a Time. In the proposed model, each successively smaller APP fragment forms an α-helix paired with a short β-strand that holds the substrate in place for cleavage. Bolded numbers indicate β-strand residues for each structure. The tripeptide IAT (pink, amino acids 41-43) is shown for reference; it moves closer to the active site (red arrow) with each cut. [Courtesy of Guo et al., Science, 2024.]  

“The adjustment of the helical region of bound Aβ intermediates—into the hydrophobic presenilin cavity and toward the active site to set up tripeptide trimming—offers incontrovertible proof for the ‘piston’ model of processivity,” noted Wolfe. That said, he likened it to a ratchet rather than a piston, because the helical region only moves in one direction, not back and forth.

Of course, given its reliance on a series of static structures, cryo-EM cannot address exactly how this racheting takes place, Wolfe noted. “Conformational adjustments of the enzyme might involve fleeting dissociation of presenilin loop 1 from the substrate, and widening of the presenilin hydrophobic channel and active site,” he proposed.

Guo and colleagues’ reported structures come on the heels of another cryo-EM analysis of γ-secretase, led by Lucia Chavez-Gutierrez and Rouslan Efremov of KU Leuven in Belgium. These scientists elucidated γ-secretase in complex with the Aβ46 peptide. Their 27 May Nature Communications paper focused on how interactions between the B isoform of AphI and PS1 subunits might lead to allosteric regulation of proteolysis. With regard to how the enzyme latches onto Aβ, the two papers report somewhat different findings. For example, first author Ivica Odorčić and colleagues resolved an unstructured segment resting upon loop-1 of PS1, suggesting that “a close interaction with this loop reduces the flexibility of this substrate region and makes it detectable in cryo-EM,” Odorčić and Chávez Gutiérrez wrote to Alzforum. Guo’s study did not detect this interaction. “Its functional significance is unclear, but the presence of Alzheimer’s-disease-linked mutations in this loop support its involvement in efficient enzyme processivity,” they wrote (comment below).

A Familiar Fold. In complex with PS1 (pink/light blue), both Aβ46 (red) and APP-C83 (dark blue) form an α-helix connected to a β-strand. Hydrogen bonds connect PS1 to its catalytic prey. [Courtesy of Odorčić et al., Nature Communications, 2024.]

Important similarities between the papers were the β-sheet at the C-terminal end of the substrate that formed a hybrid with PS1, and hydrogen bonds between residues Tyr115, Ser169 and Trp165 in PS1 and the substrate α-helix (image at right). Like Guo and colleagues, Odorčić found the β-strand was longer in C83, before the endopeptidase cut, than in Aβ46 but, other than that, the structure was maintained. “Collectively, these data show that the substrate conformation is largely similar between initial and intermediate cuts, suggesting that GSEC shapes the substrate as the substrate rearranges during processive proteolysis at its C-terminus,” the authors wrote.

Odorčić et al. also reported that eliminating the helical hydrogen bonds substantially hobbled the processive capacity of the enzyme, resulting in longer Aβ peptides. “Disruption of these bonds, through mutations, likely explains the pathogenicity of early onset Alzheimer’s disease variants at these positions,” Odorčić and Chávez Gutiérrez wrote.—Jessica Shugart


  1. This new study is another tour de force in structure elucidation of the γ-secretase complex from the labs of Rui Zhou in Beijing and Yigong Shi in Hangzhou. The snapshots of the protease bound to APP-derived C99, Aβ49, Aβ46, and Aβ43 reveal critical details that, taken together, provide important insight into the mechanism of processive proteolysis. The adjustment of the helical region of bound Aβ intermediates—into the hydrophobic presenilin cavity and toward the active site to set up tripeptide trimming—offers incontrovertible proof for the “piston” model of processivity (Yang et al., 2019). The alternative “unwinding” model (Yang et al., 2019Szaruga et al., 2017), with a static helical region, is incompatible with the new set of cryoEM structures, which show movement of this region by roughly one turn of a helix with each processing step.

    This mechanism is likely operative for the Aβ48→Aβ45→Aβ42→Aβ38 pathway of C99 processing, as well as for the many other substrates of γ-secretase (Güner and Lichtenthaler, 2020), such as Notch receptors. I would suggest, however, calling this a “ratchet” model, rather than a “piston” model, as the helical region moves in only one direction during processive proteolysis, into the hydrophobic cavity of presenilin, and not back and forth as a piston.

    The study, of course, has its limitations in providing snapshots of bound Aβ intermediates. Most notably, it does not inform about how ratcheting takes place. Presenilin and the rest of the γ-secretase complex appear static from one Aβ-bound state to the next, but there must be conformational adjustments in the enzyme, to allow movement of the helical domain of Aβ intermediate and setting up of the next cleavage step. Conformational adjustments of the enzyme might involve fleeting dissociation from the substrate of presenilin loop 1 and widening of the presenilin hydrophobic channel and active site. Such dynamic processes are not captured by the new cryoEM structures.

    This same limitation makes it difficult to extrapolate the effects of FAD mutations in APP and presenilin on γ-secretase processing based on the new structures. In this regard, the suggestion in the discussion that FAD mutations would be expected to destabilize enzyme-substrate complexes does not consider dynamics. Indeed, molecular dynamics simulations suggest reduced conformational flexibility—with increased stabilization—of FAD-mutant enzyme-substrate complexes, and this idea is supported experimentally by fluorescence lifetime imaging microscopy in intact cells (Devkota et al., 2024).


    . Structural basis of Notch recognition by human γ-secretase. Nature. 2019 Jan;565(7738):192-197. Epub 2018 Dec 31 PubMed.

    . Alzheimer's-Causing Mutations Shift Aβ Length by Destabilizing γ-Secretase-Aβn Interactions. Cell. 2017 Jul 27;170(3):443-456.e14. PubMed. Correction.

    . The substrate repertoire of γ-secretase/presenilin. Semin Cell Dev Biol. 2020 Sep;105:27-42. Epub 2020 Jun 29 PubMed.

    . Familial Alzheimer mutations stabilize synaptotoxic γ-secretase-substrate complexes. Cell Rep. 2024 Feb 27;43(2):113761. Epub 2024 Feb 13 PubMed.

  2. We read with great interest the report by Xuefei Guo et al. We just published structures of γ-secretase in an apo state as well as in complex with the intermediate Aβ46 substrate. Comparing these two independent studies has been particularly exciting.

    The various structures reported by Guo et al. on the complexes between γ-secretase and both the initial and intermediate APP-derived substrates reveal a β-sheet structure between presenilin 1 (PSEN1) and both the initial and intermediate substrates (located C-terminal to the cleavage site) that is a common feature during the sequential proteolysis. This hybrid structure likely stabilizes the distinct enzyme-substrate (E-S) complexes prior to proteolysis.

    Regarding the recognition of Aβ peptides, it has been very interesting to look at the structures with different Aβs, and to contrast the two structures with the intermediate Aβ46 peptide.

    Both our and Guo et al.’s structures consistently show H-bonding intermolecular interactions involving the backbone of the substrate, Aβ, and PSEN1-Tyr115 (loop 1), Ser169, and Trp165. In addition, our data pointed to the PSEN1-Gly384 as another H-bonding interactor with the substrate. Our analysis further adds to the functional significance of the intermolecular H-bonding interactions by showing that removal of the H-bonding capability at the PSEN1 Tyr115 (loop 1), Ser169, and Trp165 positions results in impaired γ-secretase processivity, i.e., the generation of longer Aβ peptides from APPC99. Similarly, the G384A substitution promotes generation of longer Aβ peptides from APP by destabilizing enzyme-substrate (E-S) interactions (Szaruga et al., 2017). 

    Our study and Guo et al. highlight the importance of specific H-bonding interactions for the recognition and processing of Aβ. As discussed in both reports, the formation of these intermolecular enzyme-substrate H-bonds enables the recognition and stabilization of substrates presenting low sequence homology. Their disruption (through mutations), leading to enhanced production of longer Aβ peptides, likely explains the pathogenicity of early onset Alzheimer’s disease variants at these positions, as discussed by Guo et al. and ourselves.

    The comparison also highlights differences in the E-S complexes. In our structure with Aβ46, the N-terminal part of the substrate is unstructured and rests on the PSEN1 loop 1, indicating that a close interaction with this loop reduces the flexibility of this substrate region and makes it detectable in cryo-EM. This interaction is absent in all structures reported by Guo et al., and its functional significance is unclear, but the presence of Alzheimer’s-disease-linked mutations in this loop support its involvement in efficient enzyme processivity.

    Regarding the mechanisms of cleavage, the available structural and functional data seems to reach a consensus on the following: i) initial substrate cleaving and subsequent cuts are preceded by unwinding of the C-terminal part of the substrate helix and formation of a hybrid β-sheet that facilitates filling of key S’1-3 pockets (Bolduc et al., 2016); ii) in subsequent cuts, this is accompanied by a piston-like movement of the helical substrate towards the active site. The successive cuts thus result in the embedding of the polar ectodomain of the substrates (Lys28, Asn27, Asp23, etc. in APP) into the hydrophobic substrate channel, which limits the number of cuts (Koch et al., 2023). 

    That two independent scientific analyses highlight similar conclusions while also identifying differences is exciting and significant, as they offer valuable insights into the complexities and nuances of an enzyme that plays a major role in AD pathogenicity and represents a potential target for therapeutic intervention.

    Figure 5 in our paper illustrates this:

    Structural comparison of GSEC1B-Aβ46 with GSEC1A-APPC83 and experimental validation of potential hydrogen bonds between PSEN1 and APP. A) Structural alignment of GSEC1B-Aβ46 and GSEC1A-APPC83 (PDB: 6IYC; shown in gray) complexes. PSEN1 TMs are indicated with circled numbers. B) Close-up of extracellular side of the substrate and loop 1. The GSEC1A-APPC83 complex was stabilized by disulphide cross-link between V7C APPC83 (unresolved) and Q112C PSEN1. C) Closeup view on intracellular side of substrate binding site. D) Details of PSEN1-Aβ interactions in the trans-membrane region. Potential hydrogen bond interactions between the substrates and W165, S169 and G384 are indicated. E) Western blot analysis of solubilized membranes from Psen1-/-/Psen2-/- (dKO) mouse embryonic fibroblast cell lines rescued with WT or mutant PSEN1. NCTm and NCTi indicate mature glycosylated and immature NCT, respectively. Molecular weights of protein standards are indicated on the left. F) GSEC processivity of APPC99 in Psen1-/-Psen2-/- MEFs rescued with WT or mutated PSEN1. Data are presented as mean ± SD, n=6 for the WT and n=3 for the mutants. Multiple comparison ANOVA was used to determine statistical significance (P < 0.05); P(WT vs Y115A)<0.0001, P(WT vs Y115F) )<0.0001, P(WT vs W165F) )<0.0001, P(WT vs S169A) )=0.0001, P(Y115A vs Y115F)=0.0115. Source data are provided as a Source Data file.


    . The amyloid-beta forming tripeptide cleavage mechanism of γ-secretase. Elife. 2016 Aug 31;5 PubMed.

    . APP substrate ectodomain defines amyloid-β peptide length by restraining γ-secretase processivity and facilitating product release. EMBO J. 2023 Dec 1;42(23):e114372. Epub 2023 Oct 18 PubMed.

    . Alzheimer's-Causing Mutations Shift Aβ Length by Destabilizing γ-Secretase-Aβn Interactions. Cell. 2017 Jul 27;170(3):443-456.e14. PubMed. Correction.

  3. Guo and colleagues advance our understanding of the biochemical processes behind γ-secretase-mediated cleavage of APP. Using cryo-EM of γ-secretase bound to APP-C99, Aβ49, Aβ46, and Aβ43, we are presented with the greatest insight yet of the sequential generation of Aβ species.

    The work confirms the established tripeptide cleavage pathway pattern of Aβ49>46>43>40 at a molecular resolution (Matsumura et al., 2014; Takami et al., 2009). The stereology of the active site also greatly supports previous in vitro work using amino acid substitutions to unravel substrate loading into the active site (Fernandez et al., 2016; Lichtenthaler et al., 1999). Evidence suggests an unwinding of the α-helix within the active site by one turn, helping to explain the tripeptide cleavage phenomenon. The data strongly support the piston model of γ-secretase cleavage, rather than an unwinding model (Yang et al., 2019). 

    This detailed picture helps us understand the effect of familial AD mutations. In PSEN1, Guo et al. show that Y115, S169, and W165 are crucial residues that form hydrogen bonds with the substrate. Mutations around these three residues are some with the earliest ages at onset (e.g., Y115D AAO = 29, S169L AAO = 31, W165G AAO = 36, and L166P AAO = 15). Additionally, the PALP motif appears important for flexibility and progression to the next cleavage event. We have recently discussed the importance of this motif in reference to the novel P436S mutation (Arber et al., 2024). In APP, residues such as V717 are also crucial for progression to the subsequent cleavage site, supporting our previous finding of a roadblock-like effect in the Aβ49>46>43>40 pathway caused by APP V717I (Arber et al., 2020). It will be interesting to determine the relevance of pathway switching in fAD, as shown (Kakuda et al., 2021). 

    The complexity of Aβ generation is fascinating, and we are now able to visualize this biochemical process thanks to this elegant work.

    AAO as presented in Liu et al., 2022


    . The presenilin 1 mutation P436S causes familial Alzheimer's disease with elevated Aβ43 and atypical clinical manifestations. Alzheimers Dement. 2024 Jun 2; PubMed.

    . Familial Alzheimer's disease patient-derived neurons reveal distinct mutation-specific effects on amyloid beta. Mol Psychiatry. 2020 Nov;25(11):2919-2931. Epub 2019 Apr 12 PubMed.

    . Transmembrane Substrate Determinants for γ-Secretase Processing of APP CTFβ. Biochemistry. 2016 Oct 11;55(40):5675-5688. Epub 2016 Sep 30 PubMed.

    . Switched Aβ43 generation in familial Alzheimer's disease with presenilin 1 mutation. Transl Psychiatry. 2021 Nov 3;11(1):558. PubMed.

    . Mechanism of the cleavage specificity of Alzheimer's disease gamma-secretase identified by phenylalanine-scanning mutagenesis of the transmembrane domain of the amyloid precursor protein. Proc Natl Acad Sci U S A. 1999 Mar 16;96(6):3053-8. PubMed.

    . Identification of the Aβ37/42 peptide ratio in CSF as an improved Aβ biomarker for Alzheimer's disease. Alzheimers Dement. 2022 Mar 12; PubMed.

    . γ-Secretase associated with lipid rafts: multiple interactive pathways in the stepwise processing of β-carboxyl-terminal fragment. J Biol Chem. 2014 Feb 21;289(8):5109-21. Epub 2013 Dec 28 PubMed.

    . 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.

    . Structural basis of Notch recognition by human γ-secretase. Nature. 2019 Jan;565(7738):192-197. Epub 2018 Dec 31 PubMed.

  4. This is beautiful work. The question of whether the mechanism—the formation of a stabilizing hybrid β-sheet that allows unwinding of the preceding small part of the transmembrane helix to expose the cleavage sites—would also be seen for the subsequent stepwise trimming cleavages, had been around since the first publication by Zhou et al., 2019, showing γ-secretase in complex with APP C83. Now, these structural snapshots by Guo et al. nicely show that this is the case, and they allow us to marvel at how systematically γ-secretase does its work.

    Truly fascinating. It will be interesting to see if other substrates follow the same processing path as APP, with the same repetitive structural features. Based on cleavage site analyses, this is very likely, even if corresponding small peptide byproducts have not yet been identified for substrates other than APP.


    . Recognition of the amyloid precursor protein by human γ-secretase. Science. 2019 Feb 15;363(6428) Epub 2019 Jan 10 PubMed.

Make a Comment

To make a comment you must login or register.


News Citations

  1. First Crystal Structure of a Presenilin
  2. Gamma Secretase: Intramembrane Liaisons Revealed
  3. γ-Secretase Revealed in Atomic Glory
  4. Familial Alzheimer’s Mutations: Different Mechanisms, Same End Result
  5. Nicastrin Bounces Bulky Proteins from γ-Secretase
  6. CryoEM γ-Secretase Structures Nail APP, Notch Binding

Mutation Interactive Images Citations

  1. PSEN-1

Further Reading

No Available Further Reading

Primary Papers

  1. . Molecular mechanism of substrate recognition and cleavage by human γ-secretase. Science. 2024 Jun 7;384(6700):1091-1095. Epub 2024 Jun 6 PubMed.
  2. . Apo and Aβ46-bound γ-secretase structures provide insights into amyloid-β processing by the APH-1B isoform. Nat Commun. 2024 May 27;15(1):4479. PubMed.