Open the curtains to a clearer picture of who plays footsy with whom in the γ-secretase complex. On April 27 in the Proceedings of the National Academy of Sciences, researchers led by Yigong Shi at Tsinghua University in Beijing described the highest resolution structure of the presenilin-containing apparatus to date. The structure one-upped the group’s previous work, which had yielded a slightly lower-resolution image that could not distinguish between the transmembrane (TM) regions of the four-subunit complex. The new findings illuminate interactions forged between members of the horseshoe-shaped protease, and point to flexible regions of the conglomerate that may bend and tilt to accommodate substrates such as APP.

Four protein subunits interact within and outside of the plasma membrane to form γ-secretase. Presenilin 1 (PS1) is the catalytic core. Nicastrin is proposed to help capture enzyme substrates, while PEN2 and APH1 are thought to help form and stabilize the complex. A majority of familial AD (FAD) mutations occur in PS1, and most researchers believe they cause AD by reducing the processing efficiency of the enzyme, which leads to a predominance of amyloidogenic Aβ42. Loss of PS1 function may also cause neuronal damage by limiting the processing of other substrates, such as Notch1.

The transmembrane domains of γ-secretase form a horseshoe-like loop (open side is facing reader) in the cell membrane. Aph1, PS1, and Pen2 form the rounded "toe" of the shoe that dips into the background, and two "heels" that project into the foreground. Nicastrin interacts with both heels on the extracellular side and a single transmembrane domain dips into the horseshoe. Lysozyme, fused to PS1 to help identify it, dangles into the cytoplasm. [Image courtesy of Sun et al., PNAS 2015.]

Researchers have tried to resolve the complex’s structure for the past 15 years, but the intricate entanglement of the four proteins within the membrane has made it a tough nut to crack. Previous studies employed single-particle electron microscopy to generate a low-resolution image (see Lazarov et al., 2006). Together with other biochemical methods, this technique suggested that the complex could exist in different conformations depending upon substrate engagement, bound inhibitors, or FAD mutations (see Elad et al., 2014, and Li et al., 2014). A subsequent study used cryo-EM, which freezes the protein in its native conformation, to resolve the structure to 12A resolution (see Osenkowski et al., 2009). Last year, Shi’s lab used cryo-EM to go even deeper, reporting a structure at 4.5Ǻ (see Lu et al., 2014). However, while this resolved the extracellular regions of the protein well, the TM regions of the complex—where the real action takes place—were at lower resolution and the researchers could not determine which domain belonged to which subunit.

For the current study, first author Linfeng Sun and colleagues set their sights on that intramembrane landscape. To identify the TM domains belonging to PS1, the researchers expressed T4 lysozyme protein fused to the PS1 N-terminus, which juts into the cytoplasm. This would allow them to identify the first TM domain of PS1; then, given sufficient resolution, they planned to trace the rest of the protein as it snaked through the membrane. They used the mild detergent digitonin to purify the complex from HEK293 cells, and found that it processed the C99 fragment of APP in vitro, suggesting it was in an active conformation, even with T4 attached.

The researchers performed cryo-EM and made three-dimensional electron density maps to resolve the structure to 4.32Ǻ . Unlike in their previous study, the resolution was more uniform throughout the extracellular and TM domains. The basic structure agreed with what the researchers had seen before when they had used a different detergent: a horseshoe-shaped complex lying flat within the membrane. The horseshoe peeped out of the extracellular side of the membrane to bind nicastrin, an extracellular subunit. The researchers used the T4 lysozyme to identify the nine TM regions of PS1. They had already reported the crystal structure of nicastrin, which contains a single TM domain (see Oct 2014 news). Shi told Alzforum that the three transmembrane domains belonging to PEN2 were obvious due to their position within the complex and their small number. That left APH1 to account for the remaining seven TM domains.

Using tagging techniques and prior structural knowledge, researchers assigned each of the 20 TM domains in the 4.32 Ǻ γ-secretase structure to one of the four subunits in the complex. [Image courtesy of Sun et al., PNAS 2015.] 

Aph1’s TM domains 2 and 3 tilt at an angle to the membrane plane and form a pocket. This V-shaped pocket holds the C-terminus of PS1 (not shown). [Image courtesy of Sun et al., PNAS 2015.]

With all TM domains accounted for, the researchers could then paint a clearer picture of protein interactions. Nicastrin, an extracellular subunit save its single TM region, interacted with the extracellular portion of PEN2 on the thin end of the horseshoe, and with Aph1 on the thick end. Nicastrin’s lone TM region stacked closely against three of seven TM domains from APH1. The researchers found that while most of APH1’s TM domains were stacked roughly perpendicular to the membrane, two rogue TM domains were set askew, creating a V-shaped pocket that cradled the C-terminus of PS1.

None of PS1’s nine TM domains sat perpendicular to the membrane plane; all were tilted at various angles and loosely arranged. PS1 splits into N- and C-terminal fragments as a requisite to activation, and the catalytic domain is formed by the interface of these two fragments at TM6 and TM7. The researchers found that these two domains resided on the convex, rather than the concave, side of the horseshoe, suggesting that substrates may enter the complex laterally on the outside, rather than from the center of the horseshoe. Rounding out the horseshoe, PS1’s TM4 rubbed up against TM1 and TM3 of PEN2. This subunit has three, not two, transmembrane domains, the researchers discovered. Two of these, TM1 and TM2, only traversed half the span of the membrane from the intracellular side, whereas TM3 broke through to the extracellular side to interact with nicastrin.

The nine transmembrane domains of PS1 sat askew within the membrane. The catalytic aspartyl residues in TM6 and TM7 (red dots) sat on the convex side of the γ-secretase horseshoe.

In a comment to Alzforum, David Bolduc, Dennis Selkoe, and Michael Wolfe of Brigham and Women’s Hospital, Boston, wrote that the new structure revealed important new information about the secretase. “With this increased resolution and thus clearer assignment of each of the TMDs come two main surprises: the di-aspartyl active site of presenilin lies on the convex side of the horseshoe, not within the concave cleft as originally presumed; and Pen2 contains not two but three TMDs,” they wrote.

Lucia Chavez-Gutierrez and colleagues at KU Leuven in Belgium had previously reported that PS1 had multiple active conformations. How could Shi’s single structure represent this dynamism? Shi told Alzforum that one clear structure emerging from the cryo-EM study is a strong indicator that one predominant wild-type structure exists. However, he noted that some regions, such as the second and sixth TM domains of PS1, displayed weak electron density and fuzzy EM images. This indicates that these domains may assume more than one configuration, perhaps tilting or twisting to bind substrates.  

How might the structure change with an FAD mutation, or in the presence of a γ-secretase inhibitor or modulator? Shi’s lab is investigating these questions. In collaboration with Sjors Scheres at Cambridge University in England. Shi’s lab is also working to resolve the structure of γ-secretase to an atomic level using cryo-EM. Shi hypothesized that such a detailed structure would reveal important functional characteristics about PS1 with relevance to AD. Bolduc, Selkoe, and Wolfe agreed: “[An atomic-level structure] would provide essential clues to the mechanism(s) by which mutations in presenilin found in FAD patients affect amyloid precursor protein processing,” they wrote. “Such information should facilitate the development of modulators of γ-secretase as potential AD therapeutics, an area that has received less attention than it should.”—Jessica Shugart


  1. I read with great interest this report from Yigong Shi's group. It reveals unexpected features of the γ-secretase structure and it raises many questions about the structure-function relationships in the protease complex, as many other excellent studies have. Most remarkably, the intimate connection between PSEN C-terminal fragment and APH1 is very intriguing. Our previous studies points to APH1 as an allosteric subunit in the complex (Acx et al., 2013). That APH1 and PSEN form the protease core and share a large interface may provide the structural basis for the functional effect. I also find intriguing the poor electron densities observed for PSEN1 TM6 and TM2, as discussed by the authors, as this could be an indication of their participation in substrate gating/entry.

    Definitively, the structural model presented in this study is a valuable framework to project and understand previous experimental observations as well as to design further experimental work to decipher the mechanisms of γ-secretase and “assimilate” them into a full view of γ-secretase structure/function.

    Finally, I would like to say that the outstanding contribution of the Yigong Shi and colleagues to the field makes this a very exciting time for everyone sharing a fascination for γ-secretase!


    . Signature amyloid β profiles are produced by different γ-secretase complexes. J Biol Chem. 2014 Feb 14;289(7):4346-55. Epub 2013 Dec 13 PubMed.

    View all comments by Lucia Chavez-Gutierrez
  2. Structural details of the γ-secretase complex had remained elusive, until last year when Yigong Shi and colleagues solved the structure of human γ-secretase by cryo-EM (Lu et al., 2014). This structure revealed the overall architecture of the complex, with the ectodomain of nicastrin sitting on top of a horseshoe-shaped structure composed of the transmembrane domains (TMDs) of the catalytic component presenilin and those of the other γ-secretase subunits—Pen2, Aph1, and nicastrin, which has one TMD incorporated into the horseshoe. While the solution of this structure (at an overall resolution of 4.5 angstroms) provided a huge leap forward in our understanding of γ-secretase, the limited resolution of the TMDs (range of 5-7 angstroms) prevented accurate assignment of each TMD to each of the four γ-secretase components.

    In this latest advance, Sun et al.  were able to obtain a higher-resolution image of the TMDs of the complex, again using cryo-EM. The better overall resolution (4.32 angstroms)—along with the fusion of T4 lysozyme to the N-terminus of presenilin— led to what is very likely the proper assignment of each of the TMDs of all four γ-secretase components. The new assignment of the nine presenilin TMDs reveals a close match in fold with that of an archaeal presenilin homolog solved two years ago by the Shi group (Li et al., 2013). With this increased resolution and thus clearer assignment of each of the TMDs come two main surprises: the di-aspartyl active site of presenilin lies on the convex side of the horseshoe, not within the concave cleft as originally presumed; and Pen2 contains not two but three TMDs, with one of the original two dipping only halfway into the membrane region before doing a hairpin turn.  Thus, the total number of TMDs in the γ-secretase complex is increased from 19 to 20.

    Now that the overall architecture of the complex is known, the atomic details of the TMDs need to be resolved. This would provide essential clues to the mechanism(s) by which mutations in presenilin found in familial Alzheimer’s disease (FAD) patients affect amyloid precursor protein processing. Such information should facilitate the development of modulators of γ-secretase as potential AD therapeutics, an area that has received less attention than it should (De Strooper et al., 2014; Feb 2015 news).


    . Three-dimensional structure of human γ-secretase. Nature. 2014 Aug 14;512(7513):166-70. Epub 2014 Jun 29 PubMed.

    . Structure of a presenilin family intramembrane aspartate protease. Nature. 2013 Jan 3;493(7430):56-61. PubMed.

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

    View all comments by Dennis Selkoe

Make a Comment

To make a comment you must login or register.


News Citations

  1. Crystal Structure Suggests Nicastrin Binds γ-Secretase Substrates

Paper Citations

  1. . Electron microscopic structure of purified, active gamma-secretase reveals an aqueous intramembrane chamber and two pores. Proc Natl Acad Sci U S A. 2006 May 2;103(18):6889-94. PubMed.
  2. . Structural interactions between inhibitor and substrate docking sites give insight into mechanisms of human PS1 complexes. Structure. 2014 Jan 7;22(1):125-35. Epub 2013 Nov 7 PubMed.
  3. . Cryoelectron microscopy structure of purified gamma-secretase at 12 A resolution. J Mol Biol. 2009 Jan 16;385(2):642-52. PubMed.
  4. . Three-dimensional structure of human γ-secretase. Nature. 2014 Aug 14;512(7513):166-70. Epub 2014 Jun 29 PubMed.

Other Citations

  1. Elad et al., 2014

Further Reading


  1. . γ-Secretase: a horseshoe structure brings good luck. Cell. 2014 Jul 17;158(2):247-9. PubMed.

Primary Papers

  1. . Structural basis of human γ-secretase assembly. Proc Natl Acad Sci U S A. 2015 May 12;112(19):6003-8. Epub 2015 Apr 27 PubMed.