Analysis of 138 pathogenic mutations in presenilin-1on the in vitro production of Aβ42 and Aβ40peptides by γ-secretaseLinfeng Suna,1, Rui Zhoua,1, Guanghui Yanga,1, and Yigong Shi (施一公)a,2

aBeijing Advanced Innovation Center for Structural Biology, Tsinghua–Peking Joint Center for Life Sciences, School of Life Sciences, Tsinghua University,Beijing 100084, China

Contributed by Yigong Shi, November 10, 2016 (sent for review October 25, 2016; reviewed by Matthew Freeman, Gang Pei, Jie Shen, and Gang Yu)

A hallmark of Alzheimer’s disease (AD) is the aggregation ofβ-amyloid peptides (Aβ) into amyloid plaques in patient brain. Cleav-age of amyloid precursor protein (APP) by the intramembrane pro-tease γ-secretase produces Aβ of varying lengths, of which longerpeptides such as Aβ42 are thought to be more harmful. Increasedratios of longer Aβs over shorter ones, exemplified by the ratio ofAβ42 over Aβ40, may lead to formation of amyloid plaques andconsequent development of AD. In this study, we analyzed 138reported mutations in human presenilin-1 (PS1) by individually re-constituting the mutant PS1 proteins into anterior-pharynx–defectiveprotein 1 (APH-1)aL–containing γ-secretases and examining theirabilities to produce Aβ42 and Aβ40 in vitro. About 90% of thesemutations lead to reduced production of Aβ42 and Aβ40. Notably,10% of these mutations result in decreased Aβ42/Aβ40 ratios. Thereis no statistically significant correlation between the Aβ42/Aβ40 ratioproduced by a γ-secretase variant containing a specific PS1 mutationand the mean age at onset of patients from whom the mutationwas isolated.

Alzheimer’s disease | γ-secretase | Aβ peptides | cleavage activity |amyloid hypothesis

The first case of Alzheimer’s disease (AD) was reported morethan 100 y ago, and the brain of the deceased patient contained

characteristic senile plaques (1). These plaques were found to beamyloid in nature by electron microscopy (2), and analysis of aminoacid sequence revealed the amyloid to be derived from amyloidprecursor protein (APP) (3). APP is first cleaved by β-secretase toproduce a 99-residue transmembrane fragment C99, which thenundergoes additional cleavages by γ-secretase to generate a seriesof amyloidogenic β-amyloid peptides (Aβ) with 39–43 amino acids(4, 5). The slightly longer Aβ peptides, particularly Aβ42 and Aβ43,are thought to be more prone to aggregation than the shorter onessuch as Aβ40 (6–8). Indeed, the amyloid plaques contain mostlylonger Aβ peptides such as Aβ42 (9).Familial cases of AD were found to be genetically linked to

missense mutations in APP (10, 11). These observations, togetherwith prior knowledge on amyloid plaques, prompted proposition ofthe amyloid hypothesis, which regards the formation of β-amyloidplaques as the cause of AD through apoptosis induction of neu-ronal cells (12). Consistent with this hypothesis, Aβ42 was found toinduce cell death (13, 14). Human presenilin-1 (PS1) was clonedand found to be targeted by missense mutations in early-onset fa-milial AD (EOFAD) (15). Remarkably, these EOFAD-linked PS1mutants led to elevated molar ratios of Aβ42 over Aβ40 both in celllines and in the brains of transgenic animals (16), and the plasmalevels of Aβ42 and Aβ43 in FAD patients with PS1 or APP mu-tations were significantly increased compared with those of healthyindividuals (17). These observations led to refinement of the am-yloid hypothesis, which identifies increased ratios of longer Aβpeptides over shorter ones as a key early event in AD development(18). The increased molar ratio of Aβ42 over Aβ40 is widely usedas a surrogate indicator for the pathogenic effect of presenilin andAPP mutations.

Deficiency of PS1 in neuronal cultures caused a fivefold drop inAβ production (19). Two transmembrane aspartate residues in PS1were found to be essential for the endoproteolytic maturation ofPS1 and the cleavage of C99, linking PS1 to γ-secretase (20). Theγ-secretase activity was copurified with a PS1-containing complex(21); photoactivated inhibitors of γ-secretase specifically labeled theactive site of PS1, identifying PS1 as a component of γ-secretase(22). The other three components of γ-secretase—presenilin en-hancer protein 2 (PEN-2), anterior-pharynx–defective protein 1(APH-1), and Nicastrin (NCT)—were subsequently identified (23,24). Because formation of amyloid plaques is a hallmark of AD,γ-secretase is naturally linked to AD. Specifically, the increasedmolar ratio of Aβ42 over Aβ40 as a potential causal factor of AD isattributed to altered proteolytic activity of γ-secretase.Despite its popularity, the amyloid hypothesis is yet to be ex-

perimentally proven and is challenged by unexplained observations(25, 26). Because PS1 is required for Aβ generation, it was initiallysuggested that AD-derived mutations might reflect a gain offunction (19). However, FAD-derived mutations appeared to de-crease γ-secretase activity in vitro and in cells (27–31). Enzymaticinhibition of γ-secretase in the clinic failed to alleviate the symp-toms of AD patients (32). Thus, the overall protease activity ofγ-secretase is no longer considered a causal factor for AD. Thealtered ratio of Aβ peptides, most frequently referred to as theincreased ratios of Aβ42 or Aβ43 over Aβ40, is thought to induceformation of the β-amyloid plaque and consequently lead to AD.

Significance

Alzheimer’s disease (AD) is the most common form of dementia,but the cause of AD remains poorly understood. Using highly pu-rified recombinant γ-secretase, we examined the effect of 138AD-derived presenilin-1 (PS1) mutations on the production ofβ-amyloid peptides (Aβ42 and Aβ40). These 138 mutations covervirtually all AD-targeted amino acids in PS1. Our results reveal nosignificant correlation between the Aβ42/Aβ40 ratio produced by aγ-secretase variantwith a specific PS1mutation and themean age atonset of patients carrying this mutation. The comprehensive char-acterization of pathogenic PS1 mutations serves as a valuable re-source for the analysis of γ-secretase activities and AD pathogenesis.

Author contributions: L.S., R.Z., G.Y., and Y.S. designed research; L.S., R.Z., and G.Y. per-formed research; L.S., R.Z., G.Y., and Y.S. analyzed data; and L.S. and Y.S. wrote the paper.

Reviewers: M.F., Oxford University; G.P., Institute of Biochemistry and Cell Biology, ShanghaiInstitutes for Biological Sciences, Chinese Academy of Sciences; J.S., Brigham and Women’sHospital, Harvard Medical School; and G.Y., University of Texas SouthwesternMedical Center.

The authors declare no conflict of interest.

Freely available online through the PNAS open access option.

See Commentary on page 629.1L.S., R.Z., and G.Y. contributed equally to this work.2To whom correspondence should be addressed. Email: shi-lab@tsinghua.edu.cn.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1618657114/-/DCSupplemental.

E476–E485 | PNAS | Published online December 5, 2016 www.pnas.org/cgi/doi/10.1073/pnas.1618657114

Dow

nloa

ded

by g

uest

on

Apr

il 27

, 202

1

Unfortunately, the appearance and amount of β-amyloid plaquescorrelate poorly with the development and severity of AD, andnormal individuals may contain β-amyloid plaques in the brain (33,34). PS1 mutations have also been found in frontotemporal de-mentia, which exhibits the same functional defect but lacks amyloidpathology (35–37). Finally, if γ-secretase activity underlies ADgenesis, one would expect to find disease mutations in the otherthree components of γ-secretase which presumably modulate itsprotease activity. No AD-associated mutation has been reported inPEN-2, APH-1, or NCT.In contrast to the amyloid hypothesis, the presenilin hypothesis

posits that the loss of normal function by the mutant PS1 causesAD through dominant negative effect on the wild-type allele (38,39). This hypothesis provides a satisfactory explanation to thepuzzling observation that AD-derived mutations are exclusivelymissense in nature but not nonsense or frameshift. It also poten-tially explains why heterozygous mouse with one mutant PSEN1allele, but not PSEN1+/− mouse with one allele deleted, developsmemory impairment and neurodegeneration (40). At present,whether AD is caused by loss of function in presenilin (the pre-senilin hypothesis) or qualitative changes in Aβ production (theamyloid hypothesis) remains controversial (27, 38).Another important indicator of AD development is the mean

age at onset (AAO) for patients with a shared mutation in PS1,PS2, or APP. A more deleterious mutation is likely to be reflectedby a lower AAO. Analysis of small subsets of AD-derived muta-tions supported a strong correlation: the higher the Aβ42/Aβ40ratio, the lower the AAO (41, 42). In contrast, a different set ofPS1 mutations only suggested a weak correlation (43). Notably,each of these conclusions was drawn from a limited number ofmutations, lacking statistical significance.In this study, we report a systematic analysis of 138 AD-derived

PS1 mutations, which include all reported amino acids that havebeen targeted for mutations in AD patients by January 2016. Eachof the 138 PS1 mutations was recapitulated in an APH-1aL–containing γ-secretase variant, which was purified to homogeneityand analyzed for its ability to produce Aβ42 and Aβ40. Theseresults reveal no statistically significant correlation between thecombined production of Aβ42 and Aβ40 by a γ-secretase variantand the mean AAO for patients with the corresponding mutation,or between the Aβ42/Aβ40 ratio and the mean AAO.

ResultsPreliminary Analysis of 10 PS1 Mutations. To examine the relation-ship between Aβ42/Aβ40 ratio and the mean AAO, we performeda pilot experiment on 10 representative PS1 mutations, which affectsix residues on four transmembrane helices (TMs) and four aminoacids from the inter-TM loops. Each of these PS1 variants wasreconstituted into γ-secretase for overexpression, purification, andassessment of proteolytic activity using APP-C99 as the substrate(Fig. 1A). APH-1aL, which is widely expressed in both neuronaland glia-enriched regions in the brain (44), was used in all γ-sec-retase variants. To ensure detection sensitivity, we used a mono-clonal antibody-based, fluorescence-detecting AlphaLISA assay.The combined production of Aβ42 and Aβ40 was decreased for

7 of the 10 variants and increased for three: V272A, S365A, andL424V (Fig. 1B). The variants S365A and L424V also increasedproduction of Aβ40 compared with WT γ-secretase, whereas sixvariants, including S365A and L424V, increased Aβ42 production(Fig. 1C). Eight of these 10 variants exhibit increased ratios ofAβ42 over Aβ40 compared with WT γ-secretase (Fig. 1D). Amongthe other two variants, S365A displays the same Aβ42/Aβ40 ratioas WT γ-secretase, and A285V has a slightly lower ratio of 0.84 ±0.08 (Table S1).The combined production of Aβ42 and Aβ40 by these 10

γ-secretase variants was plotted against the mean AAOs thatcorrespond to the PS1 mutations, revealing no obvious correlation(Fig. 1E). For example, two variants with increased production are

placed on two extreme ends of the AAO range, with L424V andS365A corresponding to AAOs of 26 and 62.5 y, respectively (45,46). Next, the Aβ42/Aβ40 ratios were plotted against the meanAAOs, again revealing no obvious correlation (Fig. 1F). Themutation L424V yields an Aβ42/Aβ40 ratio of 2.12 ± 0.22,whereas the mutation F105I from patients with an AAO of 55 ycorresponds to a ratio of 9.13 ± 0.40 (47).

Experimental Design and Selection of PS1 Mutations. Our pre-liminary analysis suggested a lack of correlation between theAβ42/Aβ40 ratio for a specific γ-secretase variant and the meanAAO of the patients from whom the corresponding PS1 mu-tation was isolated. To reach a statistically significant conclu-sion, we must include a much larger set of AD-derived mutationsin PS1. By January 2016, there had been 228 reported mutationsin PS1 that were derived from AD patients (www.alzforum.org).These mutations affect 121 amino acids in PS1, among which 58were each mutated to two or more types of amino acids (Fig.S1A). To ensure a comprehensive, unbiased analysis of PS1, weincluded at least one mutation for each disease-targeted aminoacid. For 13 PS1 residues that are mutated to multiple types ofamino acids, we chose two or three mutations for investigation(Fig. S1B). Altogether, 138 distinct mutations were selected forin-depth analysis, covering all 121 AD-targeted residues in PS1.These 138 mutations include those that were subjected topreliminary analysis (Fig. 1).These 138 PS1 variants were individually cloned and coex-

pressed with the other three components of γ-secretase in mam-malian cells (48). The 138 resulting γ-secretase variants, eachcontaining a specific PS1 mutant and APH-1aL, were individuallypurified to homogeneity. All 138 variants were eluted from gelfiltration with a nearly identical elution volume, suggesting thatnone of the AD-associated PS1 mutations disrupts the folding oroverall structure of γ-secretase. Thus, these mutations likely affectthe functional aspects of PS1 and γ-secretase.Analysis by SDS/PAGE revealed five discrete bands for the

WT γ-secretase (Fig. S2, lane 1-9) and for the vast majority of thevariants (Figs. S2 and S3A, lanes 1-1 through 16-9). These fivebands correspond to NCT (120–150 kDa), the N-terminal fragmentof PS1 (PS1-NTF, 34 kDa), APH-1aL (29 kDa), the C-ter-minal fragment of PS1 (PS1-CTF, 19 kDa), and PEN-2 (12kDa). The apparent size of PEN-2 is larger than the predictedmolecular weight, in part due to presence of the FLAG affinitytag. The endoproteolytic processing of PS1 is incomplete for someγ-secretase variants exemplified by G209V, L235R, and E280G(Fig. S2, lanes 1-2, 1-5, and 1-6). Eight of these γ-secretase variantsnearly abrogated endoproteolysis, as evidenced by the intact PS1and missing NTF/CTF bands (Fig. S4A).

Generation of Aβ42 and Aβ40 by 138 γ-Secretase Variants. The 138γ-secretase variants were individually examined for their abilities tocleave APP-C99 into Aβ42 and Aβ40 using the AlphaLISA assay(Table S1). Compared with WT γ-secretase, only 10 variants in-creased production of Aβ40 (Fig. 2). The largest increase of 82%for Aβ40 production comes from the variant L226F, which re-places Leu226 by Phe in TM5 of PS1. All 10 variants with increasedAβ40 production, hereafter referred to as TOP10, yielded higheramounts of Aβ42 and Aβ40 (Fig. 3A). All 10 variants also en-hanced production of Aβ42; except L282V and I439V, the extentof increased production for Aβ42 is greater than that for Aβ40(Fig. 2). These eight variants exhibit increased ratios of Aβ42 overAβ40 (Fig. 3B).Compared with WT γ-secretase, 34 variants increased production

of Aβ42 (Fig. 2). All other 104 variants produced lower levels ofboth Aβ42 and Aβ40. Of these 104 variants, 67 exhibit a severelycompromised ability to produce Aβ40, which is defined by loss of atleast 95% activity compared with WT γ-secretase. By the samecriteria, only 14 variants exhibit a severely compromised ability to

Sun et al. PNAS | Published online December 5, 2016 | E477

BIOCH

EMISTR

YPN

ASPL

US

SEECO

MMEN

TARY

Dow

nloa

ded

by g

uest

on

Apr

il 27

, 202

1

Fig. 1. Preliminary analysis of 10 AD-derived PS1 mutations. (A) A schematic diagram of the major workflow in this study. The WT human γ-secretaseand 138 variants, each containing an AD-derived mutation in PS1, were individually overexpressed and purified. The peak fractions were incubated withAPP-C99 and examined for the production of Aβ42 and Aβ40 using the AlphaLISA assay. (B) Effect of 10 AD-derived PS1 mutations on the cleavageactivities of the corresponding γ-secretases. Shown here is the combined production of Aβ42 and Aβ40 peptides by the 10 γ-secretase variants. Theactivity of WT γ-secretase was normalized as 1. Each experiment was repeated three times, and the SD is shown. (C ) Effect of 10 AD-derived PS1mutations on the generation of Aβ42 and Aβ40. (D) All but two variants show increased molar ratios of Aβ42 over Aβ40 compared with WT γ-secretase.Only the variant A285V displays a slightly lower Aβ42/Aβ40 ratio, and S365A is nearly identical to WT γ-secretase. (E ) The combined production of Aβ42and Aβ40 by each of the 10 γ-secretase variants shows no obvious correlation with the mean AAO. (F ) The Aβ42/Aβ40 ratio exhibits no obvious cor-relation with the mean AAO.

E478 | www.pnas.org/cgi/doi/10.1073/pnas.1618657114 Sun et al.

Dow

nloa

ded

by g

uest

on

Apr

il 27

, 202

1

produce Aβ42, and they represent a subset of the 67 variants withcrippled Aβ40 generation (Fig. 2). These results demonstrate thatthe vast majority of AD-derived PS1 mutations negatively affect thecleavage activities of the resulting γ-secretase variants in vitro. Inaddition, a PS1 mutation is more likely to decrease the productionof Aβ40 than Aβ42 or to increase the generation of Aβ42 thanAβ40. Either scenario would result in an increased ratio of Aβ42over Aβ40.Our conclusion is corroborated by analysis of the total cleav-

age activities toward Aβ42 and Aβ40 (Fig. 3A). Because themolar quantity of Aβ40 generated by WT γ-secretase exceedsthat of Aβ42 by ∼10-fold, the combined amount of Aβ42 and

Aβ40 is dominated by Aβ40 production. Consequently, although34 variants increased Aβ42 production, only 13 variants exhibit ahigher level of combined Aβ42 and Aβ40 production comparedwith WT γ-secretase (Fig. 3A). In addition to TOP10, these 13variants include I143V, F177L, and V272A, each of which has aWT level of Aβ40 but markedly elevated Aβ42 production.Among these 13 variants, 6 have activities that are within 20%range of the WT value, and the most active variant, L226F, hasan activity that is ∼98% higher than WT γ-secretase (Fig. 3A). Insharp contrast, 101 variants each display a combined productionof Aβ42 and Aβ40 that is less than 50% of that by WT γ-secre-tase. Of these 101 variants, 57 exhibit a severely compromised

Fig. 2. Effect of 138 AD-derived PS1 mutations on the abilities of the corresponding γ-secretase variants to generate Aβ42 and Aβ40. The amounts of Aβ42 andAβ40 produced by WT γ-secretase are normalized. The relative amounts of Aβ42 and Aβ40 produced by these 138 γ-secretase variants are shown here. The vastmajority of these variants exhibit reduced production of both Aβ42 and Aβ40. In particular, production of Aβ42 as well as Aβ40 was abrogated to backgroundlevels for 11 γ-secretase variants, including S178P, S212Y, Q223R, S230I, I238M, K239N, T245P, L271V, T274R, T291P, and E318G. Only 10 γ-secretase variants exhibitincreased cleavage activities toward both Aβ42 and Aβ40 compared with WT γ-secretase.

Sun et al. PNAS | Published online December 5, 2016 | E479

BIOCH

EMISTR

YPN

ASPL

US

SEECO

MMEN

TARY

Dow

nloa

ded

by g

uest

on

Apr

il 27

, 202

1

ability, with about 42 variants abrogating generation of Aβ42 andAβ40 altogether. The Aβ42/Aβ40 ratios of these 42 γ-secretasevariants cannot be calculated because they only produce back-ground levels of Aβ42 and Aβ40.

Our conclusion is also confirmed by quantification of the Aβ42/Aβ40 molar ratios by the remaining 96 γ-secretase variants (Fig. 3B).Only 13 variants exhibit decreased Aβ42/Aβ40 ratios. Among the13 variants, 9 have Aβ42/Aβ40 ratios that are within 20% of the

Fig. 3. Effect of 138 AD-derived PS1 mutations on the abilities of the corresponding γ-secretase variants to modulate the Aβ42/Aβ40 ratios. (A) Effect of 138AD-derived PS1 mutations on the combined production of Aβ42 and Aβ40 by the corresponding γ-secretase variants. The vast majority of the mutationsnegatively affect the combined production of Aβ42 and Aβ40. Only 13 variants exhibit increased total cleavage activity compared with WT γ-secretase.(B) Effect of 138 AD-derived PS1 mutations on the abilities of the corresponding γ-secretase variants to modulate the Aβ42/Aβ40 ratios. The vast majority ofthese mutations lead to increased Aβ42/Aβ40 ratios. Only 13 variants exhibit decreased Aβ42/Aβ40 ratios compared with WT γ-secretase.

E480 | www.pnas.org/cgi/doi/10.1073/pnas.1618657114 Sun et al.

Dow

nloa

ded

by g

uest

on

Apr

il 27

, 202

1

normalized WT value, and only 4 variants (D333G, T354I, N405S,and A409T) exhibit ratios that are significantly less than the WTvalue. The lowest Aβ42/Aβ40 ratio of 0.51 ± 0.03 is associatedwith the PS1 mutation T354I, and all four corresponding muta-tions in PS1 affect residues in the CTF. In contrast to the mod-erate reduction of Aβ42/Aβ40 ratios, 83 variants exhibit higherratios, of which 64 are at least 100% higher and 31 are at leastfourfold higher (Fig. 3B). These 31 variants are hereafter referredto as RATIO31. The highest Aβ42/Aβ40 ratio of 171 ± 25 is as-sociated with the PS1 mutation G384A.

Aβ42/Aβ40 Ratios in Liposome- and Cell-Based Assays.Our proteolyticcleavage assay was performed on each of the 138 γ-secretasevariants in the presence of low amounts of phospholipids and0.5% zwitterionic detergent CHAPSO. Such mild conditions areexpected to maintain the structural integrity as well as biochemicalproperty for most membrane proteins and enzymes. Nevertheless,to rule out the possibility of potentially disruptive effects from thedetergents, we performed the cleavage assays in liposomes and incells. Liposomes were prepared using polar lipid extracts from pigbrain, and both WT γ-secretase and 11 variants were individuallyreconstituted into these liposomes in the presence of the substrateAPP-C99 (Fig. 4A). These liposomes, both before and afterγ-secretase reconstitution, remained intact as revealed by electronmicroscopy (Fig. S5).Following overnight incubation of these proteoliposomes at

37 °C, the cleavage products were analyzed by the AlphaLISA assay(Fig. 4A). Compared with the normalized value of WT γ-secretase,the overall results for the 11 variants are quite similar to thoseobtained using the detergent CHAPSO micelles (Fig. 4B). The ra-tios of Aβ42/Aβ40 are similar (within 30% difference of each other)between detergent micelle- and liposome-based assays for eightvariants, F105I, A246E, L248R, V272A, A285V, S365A, L381V,

and L424V. The other three variants exhibit some differences be-tween the two assay conditions, but the trend of the Aβ42/Aβ40ratios remains the same between the two assay conditions. For ex-ample, compared with WT γ-secretase, the variant I229F has ahigher Aβ42/Aβ40 ratio under both assay conditions, although thevalue measured in the liposome-based assay is ∼30% higher thanthat in detergent micelle (Fig. 4B).Next, we examined whether similar Aβ42/Aβ40 ratios would be

maintained in cell-based assays. Pilot experiments using the mouseneuronal N2a cells revealed significant background cleavage fromendogenous γ-secretase. To eliminate the background, we suc-cessfully applied the CRISPR/Cas9 technology to knock out thePSEN1 genes in N2a cells (Fig. 4C and Fig. S6). Each of the 11PS1 variants and the APP-C99 substrate were coexpressed in thePSEN1−/− N2a cells, and the production of Aβ42 and Aβ40was monitored by the AlphaLISA assay. Compared with WTγ-secretase, the cell-based results are generally comparable to thoseunder detergent micelles (Fig. 4D). The ratios of Aβ42/Aβ40 aresimilar (within 30% difference of each other) between detergentmicelle- and cell-based assays for six variants F105I, A246E,L248R, P264L, L381V, and L424V. The other five variants exhibitsome differences between the two assay conditions, but the trendremains the same for two variants. Compared with WT γ-secretase,the variants I229F and V272A each has a higher Aβ42/Aβ40 ratiounder both assay conditions, although in each case the valuemeasured in the cell-based assay is more than 30% higher than thatin detergent micelle (Fig. 4D). Of the remaining three variants,S365A has a WT-level Aβ42/Aβ40 ratio in detergent micelle and a35% higher ratio in the cell-based assay.These results demonstrate that the Aβ42/Aβ40 ratios derived

from liposome- and cell-based assays are qualitatively similar tothose obtained under detergent micelles. Notably, none of these threeconditions would faithfully recapitulate the cleavage environment in

Fig. 4. The Aβ42/Aβ40 ratios generated in liposome- and cell-based cleavage assays are similar to those obtained in detergent micelle-based assays. (A) Aschematic diagram of the liposome-based cleavage assay. The WT and mutant γ-secretases were individually reconstituted into the liposomes together withAPP-C99 and examined for their abilities to generate Aβ42 and Aβ40 using the AlphaLISA assay. (B) The Aβ42/Aβ40 ratios generated in the liposome-basedcleavage assays are similar to those obtained in detergent micelle-based assays. Results from 11 representative γ-secretase variants are shown here. (C) Aschematic diagram of the cell-based cleavage assay. The CRISPR/Cas9 technology was used to generate PS1−/− N2a cells. The pCAG vectors, each encoding WTor a distinct γ-secretase variant, were individually transfected into N2a cells for assessment of production of Aβ42 and Aβ40. (D) The Aβ42/Aβ40 ratiosgenerated in the cell-based cleavage assays are similar to those obtained in detergent micelle-based assays.

Sun et al. PNAS | Published online December 5, 2016 | E481

BIOCH

EMISTR

YPN

ASPL

US

SEECO

MMEN

TARY

Dow

nloa

ded

by g

uest

on

Apr

il 27

, 202

1

patient brain. Both liposome- and cell-based assays may sufferfrom poor reproducibility and uncertainties associated with lipidcomposition and/or cell growth status. For convenience and re-producibility, we had chosen the detergent micelle-based assay toassess WT γ-secretase and the 138 variants.

Lack of Correlation Between Aβ42/Aβ40 Ratios and AAOs.We plottedthe observed ratios of Aβ42 over Aβ40 for the 138 γ-secretasevariants against the mean AAOs for patients from whom these PS1mutations were derived (Fig. 5A). Strikingly, there is no obviouscorrelation between the Aβ42/Aβ40 ratios and the AAOs. A forcedfit of these data points into a straight line produced a correlationcoefficient of 0.038 and a P value of 0.06, which are statisticallyinsignificant (Fig. S7A). The PS1 mutation L166P corresponds to avery low AAO of 24 y (49), and the corresponding γ-secretasevariant exhibits a relatively low Aβ42/Aβ40 ratio of 2.71 ± 0.37. Incontrast, the mutation E273A was derived from patients with amean AAO of 63 y (50), and the corresponding γ-secretase varianthas an Aβ42/Aβ40 ratio of 2.55 ± 0.20, similar to that for L166P.The mutation G384A, corresponding to a mean AAO of 39 y (51),has the highest Aβ42/Aβ40 ratio of 171 ± 25 (Fig. 5A). Thecleavage activities of these 138 variants, as reflected by combinedproduction of Aβ42 and Aβ40, also display no correlation with themean AAOs (Fig. 5B).Because these 138 variants exhibit vastly different enzymatic

activities, we wondered whether select subsets of these variantsmight exhibit meaningful correlation between Aβ42/Aβ40 ratiosand mean AAOs. We plotted Aβ42/Aβ40 ratios and mean AAOsfor the TOP10 variants, which exhibit increased production forboth Aβ42 and Aβ40. The scattered distribution of this plot revealsno statistically significant correlation (Fig. S7B). A similar con-clusion was reached for the RATIO31 variants, which exhibit the31 highest Aβ42/Aβ40 ratios among all variants (Fig. S7C). Pre-vious studies suggested potential correlations between AAO and

the Aβ42/Aβ40 ratio on the basis of a small subset of FAD-derivedmutations (41–43). We examined the same mutations and obtainedsimilar results despite subtle shift of exact numbers (Fig. S8). Ob-viously, these correlations lack statistical significance.Our systematic analyses of nearly all reported missense muta-

tions from PS1 using the in vitro assay shows no statistically sig-nificant correlation between AAOs and Aβ42/Aβ40 ratios. Becausethese 138 mutations include all AD-targeted residues in PS1, thestatistical power embedded in our brute force approach cannot beunderstated. In fact, conclusions derived from a subset of the dataare often erroneous. For example, a subset of nine PS1 mutationsgive a perfect correlation (R2 = 0.99, P value < 0.001) for theconclusion that higher ratios of Aβ42/Aβ40 correspond to lowerAAOs, which is consistent with the amyloid hypothesis (Fig. 5C).Unfortunately, a subset of another nine PS1 mutations also yieldsan excellent correlation (R2 = 0.92, P value < 0.001) for the op-posite conclusion that higher ratios of Aβ42/Aβ40 correspond tohigher AAOs (Fig. 5D).

Effect of Random PS1 Mutations on Aβ42/Aβ40 Ratios. Why do mostFAD-derived PS1 mutations result in increased Aβ42/Aβ40 ratiosby the corresponding γ-secretase variants? Results of the in vitrocleavage assays strongly suggest that γ-secretase may have evolvedto optimize the Aβ42/Aβ40 ratio to a low value; namely, any PS1mutation is likely to reduce Aβ40 production more than Aβ42 or toincrease Aβ42 production more than Aβ40. If this assumptionwere true, random mutations of PS1 would be expected to mostlylead to increased Aβ42/Aβ40 ratios. To examine this assumption,we randomly chose 13 amino acids for missense mutation. These13 residues are located in eight TMs of PS1 (Fig. 6A); the vastmajority of these 13 mutations involve replacement of conservedamino acids such as L150V and L182V. The 13 correspond-ing γ-secretase variants were individually purified to homogeneity

Fig. 5. There is no statistically significant correlation between the Aβ42/Aβ40 ratios produced by γ-secretase variants and the mean AAOs. (A) A plot of theAβ42/Aβ40 ratio versus the mean AAO. Altogether, 90 data points are shown here. For 42 γ-secretase variants, the Aβ42/Aβ40 ratio cannot be computed dueto background levels of Aβ42 and/or Aβ40. For six PS1 mutations, there are no reported AAOs. (B) A plot of the total cleavage activity of γ-secretase variantversus the mean AAO. Altogether, 127 data points are shown here. For 11 PS1 mutations, there are no reported AAOs. (C) Higher ratios of Aβ42/Aβ40correlate with lower AAOs from a group of nine PS1 mutations. The correlation coefficient is 0.99, with a P value of less than 0.001. (D) Lower ratios of Aβ42/Aβ40 correlate with lower AAOs from a group of nine PS1 mutations. The correlation coefficient is 0.92, with a P value of less than 0.001.

E482 | www.pnas.org/cgi/doi/10.1073/pnas.1618657114 Sun et al.

Dow

nloa

ded

by g

uest

on

Apr

il 27

, 202

1

(Fig. S3B) and subjected to the detergent micelle-based, APP-C99cleavage assays.Strikingly, none of these 13 variants produced more Aβ40 or

Aβ42 than WT γ-secretase (Fig. 6B). Six variants (P88S, V97F,L182V, S230A, L443V, and Y446H) nearly abrogated productionof Aβ40 and diminished generation of Aβ42. Consequently, thesesix variants exhibit a severely compromised ability to generate theAβ peptides (Fig. 6C). The Aβ40 production by three variants(P88S, V97F, and Y446H) is at the background level and thusdisallows calculation of the Aβ42/Aβ40 ratios. In complete agree-ment with our assumption, the extent of reduction for Aβ40 isconsiderably more than that for Aβ42 for most of the 13 variants.Seven variants exhibit increased ratios of Aβ42/Aβ40 comparedwith WT γ-secretase, and three variants have a ratio that is similarto that by WT γ-secretase (Fig. 6D).The patterns of Aβ42/Aβ40 production by these 13 randomly

generated mutations in PS1 are reminiscent of those by 138 FAD-derived mutations. In both cases, the vast majority of the γ-sec-retase variants exhibit reduced production of Aβ42 and Aβ40 andincreased ratios of Aβ42/Aβ40. A sizable fraction of the variantsgenerate background levels of Aβ42 and/or Aβ40, disallowingcalculation of Aβ ratios.

DiscussionMutations in APP, PS1, and PS2 have been unequivocally linkedto the development of AD, and most AD-derived mutations inPS1 result in increased ratios of Aβ42/Aβ40. Although the amyloidhypothesis is yet to be proved, it is supported by these observationsand by the finding that oligomeric Aβ peptides induced synapseengulfment by microglia (52). A human monoclonal antibody,aducanumab, was shown to selectively target aggregated Aβ andreduce the amyloid plaque in patients with prodromal or mild AD.A dose-dependent slowing of clinical progression was observed in

the phase 1b trial and awaits further confirmation by the ongoingphase 3 trials (53).Notably, nearly 90% of FAD mutations occur in presenilin, but

no mutations have been identified in the other three subunits ofγ-secretase. Through knock-in studies, presenilin is shown to playan important role in synaptic plasticity (54–57). Such observationssupport the presenilin hypothesis, where presenilin mutations areconsidered to be closely associated with neurodegeneration byimpairing the γ-secretase–dependent and –independent PS func-tions. PS1 may function, either alone or in complex with otherproteins, as an ion channel to regulate calcium flux (58–61). Ex-amination of FAD-derived mutations in PS1 identified two mu-tational hotspots, each located in the center of a four-TM bundle(62). Whether this observation is linked to the function of PS1awaits further investigation.The mean AAO is an indicator of disease severity and pro-

gression, with a lower AAO suggestive of a more deleteriousmutation in PS1 and thus more debilitating disease. If increasedratios of longer Aβ peptides over shorter ones is a key early eventin AD development, then one might be able to find a statisticallysignificant correlation between such ratios and AAO. This study isaimed at investigating the effect of PS1 mutations on γ-secretaseactivity and examining a potential correlation between the Aβ42/Aβ40 ratio and disease progression as represented by AAO. Torule out bias from subsets of PS1 mutations, we collected a largenumber of reported PS1 mutations and individually generated thecorresponding γ-secretase variants. Except 13 mutations, all othersled to decreased production of Aβ42 and Aβ40 compared with WTγ-secretase. The vast majority of PS1 mutations lead to increasedratios of Aβ42/Aβ40. The Aβ42/Aβ40 ratios by these γ-secretasevariants, each containing a specific PS1 mutation and APH-1aL,show no statistically significant correlation with the mean AAOsfor the corresponding mutations.

Fig. 6. γ-secretase variants with engineered PS1 mutations exhibit similar biochemical properties as the variants with AD-derived mutations. (A) Location of the13 amino acids affected by engineered mutations. Except TM4, each of the other eight TMs contributes at least one residue for mutation. Shown here is a ribbonrepresentation of PS1 structure. The residues targeted for mutation are colored magenta, and the two catalytic Asp residues are shown in spheres. (B) All 13engineered mutations result in compromised protease activities of the corresponding γ-secretase variants for the generation of both Aβ42 and Aβ40. (C) All 13engineered mutations lead to decreased levels of protease activity as judged by the combined production of Aβ42 and Aβ40. (D) Seven mutations cause increasedratios of Aβ42/Aβ40 compared with WT. Only three mutations (I140V, I253V, and I414V) have no significant effect on the Aβ42/Aβ40 ratios.

Sun et al. PNAS | Published online December 5, 2016 | E483

BIOCH

EMISTR

YPN

ASPL

US

SEECO

MMEN

TARY

Dow

nloa

ded

by g

uest

on

Apr

il 27

, 202

1

In this study, APH-1aL was used to reconstitute γ-secretase, andconsequently, the general conclusions reported in our study arerestricted to APH-1aL–containing γ-secretase. To examine thevalidity of such conclusions under other settings, we generated10 γ-secretase variants, each reconstituted with APH-1b and aspecific PS1 mutation, and individually examined their abilities toproduce Aβ42 and Aβ40. The combined production of Aβ42 andAβ40 by these APH-1b–containing γ-secretase variants displays asimilar trend to that by the APH-1aL–containing γ-secretase (Fig.S9A). No significant correlation was found between the combinedproduction of Aβ42 and Aβ40 and the AAO (Fig. S9B). Exceptone PS1 mutation (F105I), the APH-1b–containing γ-secretasesgenerally exhibit higher Aβ42/Aβ40 ratios than APH-1aL–con-taining γ-secretases (Fig. S9C). The trend of Aβ42/Aβ40 ratiochanges is quite similar for these two classes of γ-secretases, sug-gesting general applicability of such preliminary conclusion. Nosignificant correlation between the Aβ42/Aβ40 ratios and AAOs isobserved (Fig. S9D).We acknowledge the caveat that our experimental conditions,

where homogeneous γ-secretase is used in an artificial in vitrosetting, are quite different from the complex circumstances in pa-tient brain, where PSEN1 is heterozygous and potential dominantnegative effect may be at work (39). Another complicating factor isthe in vivo existence of six distinct γ-secretase complexes formed bydifferent forms of presenilin (PS1 and PS2) and APH-1 (APH-1aL,APH-1aS, and APH-1b), which are localized to different regions ofthe brain (44, 63). Different Aβ profiles can be generated by thedistinct γ-secretase complexes (64). FAD mutations in PS1 mayshift its localization and increase the intracellular longer Aβ pep-tides (65).In addition, the cleavage of APP-C99 by γ-secretase involves

an endopeptidase activity, which generates Aβ48 or Aβ49, and asubsequent carboxypeptidase activity, which produces Aβ45/Aβ42from Aβ48 and Aβ46/Aβ43/Aβ40 from Aβ49. In our study, we onlyassessed the production of Aβ42 and Aβ40, which is presumablyaffected by both endopeptidase and carboxypeptidase activities.Our study does not differentiate between these two activities, butit has been shown that some PS1 mutations have differential ef-fects on these two activities. Some mutations had little effect onthe endopeptidase activity but exhibited altered carboxypeptidaseactivity (66). Five AD-derived PS1 mutations led to elevatedAβ42/Aβ40 ratios using synthetic, longer Aβ peptides, reflectingaltered carboxypeptidase activity of γ-secretase (67).Although our data appear to contradict the amyloid hypothesis,

the caveats listed above preclude a definitive conclusion. Notably,we cannot rule out the possibility that AD occurs through severaldifferent mechanisms and the amyloid hypothesis may constituteone of these mechanisms. Similarly, the presenilin hypothesis mayaccount for other mechanism(s). In fact, AD, characterized by apattern of cognitive and functional decline, is a very complexdisease, with Aβ accumulating years ahead of any detectablesymptom yet triggering a series of cellular and physiological re-sponses. Thus, it is possible that there is no simple or single causefor this complex disease. Regardless of these different scenarios,our results involving characterization of 138 PS1 mutations rep-

resent a valuable resource for future investigation of γ-secretaseand AD development.

Materials and MethodsProtein Preparation. Generation of all FAD-derived PS1mutants and expressionand purification of the resulting γ-secretase variants were described asreported (48). The purified γ-secretase was applied to Superose-6 column (GEHealthcare) in 0.1% digitonin and Hepes buffer. The peak fractions wereconcentrated for activity assays.

Detergent-Based Activity Assays. Purified 4 μg WT or mutant γ-secretase wasmixed with 4 μg APP-C99 substrate in 0.5% CHAPSO, 50 mM Hepes, pH 7.0,0.1% phosphatidylcholine, and 0.025% phosphatidylethanolamine and in-cubated in 37 °C for 16 h. Detection of Aβ production by the AlphaLISA assaywas performed as described (68).

Liposome-Based Activity Assay. Porcine brain polar lipid extract (Avanti) wasdried by nitrogen gas after being dissolved in a mixture of chloroform andmethanol (3:1 vol/vol) to 50 mg·mL−1 and then resuspended to 20 mg·mL−1 in25 mM Hepes, pH 7.4, 150 mM NaCl. The lipids were frozen in liquid nitrogenand thawed at room temperature gently for eight cycles followed by extrusionthrough 0.4 μm filter to yield liposomes. CHAPSO was added to a final con-centration at 1% (wt/vol) at 4 °C for 30 min. Purified γ-secretase and APP-C99were added to the lipids and incubated at 4 °C for 30 min. Detergents wereremoved by two-step incubation with 400 mg·mL−1 Bio-Beads SM2 (Bio-Rad)after which the proteoliposomes were frozen and thawed for five cycles. Theproteoliposomes were applied to centrifugation at 100,000 × g for 40 min at4 °C and washed twice by ice-cold lysis buffer to remove free γ-secretase andsubstrate. The proteoliposomes were resuspended in lysis buffer to 20 mg·mL−1.Integrity of the proteoliposomes was checked by negative staining electronmicroscopy. The proteoliposomes were incubated at 37 °C for 16 h followed bydetection of Aβ40 and Aβ42 using the AlphaLISA assay.

Generation of PSEN1 Knockout Cell Line. The adherent N2a cell line was pas-saged and maintained in DMEM (Life Technologies) supplemented with 10%(vol/vol) FBS and MEM NEAA (GIBCO) at 37 °C under 5% (vol/vol) CO2 and 75%relative humidity. SgRNAs targeting mouse PSEN1 gene were designed in silicoon a website (crispr.mit.edu), and several candidates were chosen. For genera-tion of the complete knockout cell line, the annealing product of two sgRNA-containing primers (forward: 5′-caccgGCCATGCGGTCCATTCGGGG-3′; reverse:5′-aaacCCCCGAATGGACCGCATGGCc-3′) was inserted into pSpCas9 vectordigested with BbsI. Then, 0.5 μg plasmids were transiently transfected usingneofect into ∼70% confluent cells in a 24-well plate. Twenty-four h aftertransfection, puromycin was added to a final concentration of 2 μg·mL−1 forselection. Finite serial dilution method was used to separate a single cell colonyin 96-well plate. The cells alive were harvested, and their genome was extractedusing tissue gDNA kit (Biomiga). The knockout efficiency was examined using aPCR-based method.

Cell-Based Activity Assay. Using 0.5 μg neofect with 15-min incubation,0.25 μg plasmid with a FAD-derived PS1 mutant and 0.25 μg plasmid con-taining the substrate APP-C99 were premixed and transfected into ∼70%confluent cells in a 24-well plate. Transfected cells were cultured for 60 h. Thenthe cell medium was harvested, and 2 μL was detected by the AlphaLISA assayas described above.

ACKNOWLEDGMENTS. This work was supported by funds from the Ministryof Science and Technology (2014ZX09507003006 to Y.S.) and the NationalNatural Science Foundation of China (31130002 and 31321062 to Y.S.). L.S. issupported by a postdoctoral fellowship of the Tsinghua–Peking Joint Centerfor Life Sciences.

1. Alzheimer A (1907) About a peculiar disease of the cerebral cortex. Centralblatt für

Nervenheilkunde Psychiatrie 30:177–179.2. Terry RD, Gonatas NK, Weiss M (1964) Ultrastructural studies in Alzheimer’s presenile

dementia. Am J Pathol 44:269–297.3. Glenner GG, Wong CW (1984) Alzheimer’s disease and Down’s syndrome: Sharing of a

unique cerebrovascular amyloid fibril protein. Biochem Biophys Res Commun 122(3):

1131–1135.4. Estus S, et al. (1992) Potentially amyloidogenic, carboxyl-terminal derivatives of the

amyloid protein precursor. Science 255(5045):726–728.5. Golde TE, Estus S, Younkin LH, Selkoe DJ, Younkin SG (1992) Processing of the amyloid

protein precursor to potentially amyloidogenic derivatives. Science 255(5045):728–730.6. Burdick D, et al. (1992) Assembly and aggregation properties of synthetic Alzheimer’s

A4/beta amyloid peptide analogs. J Biol Chem 267(1):546–554.

7. Jarrett JT, Berger EP, Lansbury PT, Jr (1993) The carboxy terminus of the beta amyloid

protein is critical for the seeding of amyloid formation: Implications for the patho-

genesis of Alzheimer’s disease. Biochemistry 32(18):4693–4697.8. Suzuki N, et al. (1994) An increased percentage of long amyloid beta protein secreted

by familial amyloid beta protein precursor (beta APP717) mutants. Science 264(5163):

1336–1340.9. Roher AE, et al. (1993) Structural alterations in the peptide backbone of beta-amyloid

core protein may account for its deposition and stability in Alzheimer’s disease. J Biol

Chem 268(5):3072–3083.10. Chartier-Harlin MC, et al. (1991) Early-onset Alzheimer’s disease caused by mutations

at codon 717 of the beta-amyloid precursor protein gene. Nature 353(6347):844–846.11. Mullan M, et al. (1992) A pathogenic mutation for probable Alzheimer’s disease in

the APP gene at the N-terminus of beta-amyloid. Nat Genet 1(5):345–347.

E484 | www.pnas.org/cgi/doi/10.1073/pnas.1618657114 Sun et al.

Dow

nloa

ded

by g

uest

on

Apr

il 27

, 202

1

12. Hardy JA, Higgins GA (1992) Alzheimer’s disease: The amyloid cascade hypothesis.Science 256(5054):184–185.

13. Pike CJ, Burdick D, Walencewicz AJ, Glabe CG, Cotman CW (1993) Neurodegenerationinduced by beta-amyloid peptides in vitro: The role of peptide assembly state. J Neurosci13(4):1676–1687.

14. Gschwind M, Huber G (1995) Apoptotic cell death induced by beta-amyloid 1-42peptide is cell type dependent. J Neurochem 65(1):292–300.

15. Sherrington R, et al. (1995) Cloning of a gene bearing missense mutations in early-onset familial Alzheimer’s disease. Nature 375(6534):754–760.

16. Borchelt DR, et al. (1996) Familial Alzheimer’s disease-linked presenilin 1 variantselevate Abeta1-42/1-40 ratio in vitro and in vivo. Neuron 17(5):1005–1013.

17. Scheuner D, et al. (1996) Secreted amyloid beta-protein similar to that in the senileplaques of Alzheimer’s disease is increased in vivo by the presenilin 1 and 2 and APPmutations linked to familial Alzheimer’s disease. Nat Med 2(8):864–870.

18. Selkoe DJ, Hardy J (2016) The amyloid hypothesis of Alzheimer’s disease at 25 years.EMBO Mol Med 8(6):595–608.

19. De Strooper B, et al. (1998) Deficiency of presenilin-1 inhibits the normal cleavage ofamyloid precursor protein. Nature 391(6665):387–390.

20. Wolfe MS, et al. (1999) Two transmembrane aspartates in presenilin-1 required forpresenilin endoproteolysis and gamma-secretase activity. Nature 398(6727):513–517.

21. Li YM, et al. (2000) Presenilin 1 is linked with gamma-secretase activity in the de-tergent solubilized state. Proc Natl Acad Sci USA 97(11):6138–6143.

22. Li YM, et al. (2000) Photoactivated gamma-secretase inhibitors directed to the activesite covalently label presenilin 1. Nature 405(6787):689–694.

23. Francis R, et al. (2002) aph-1 and pen-2 are required for Notch pathway signaling,gamma-secretase cleavage of betaAPP, and presenilin protein accumulation. Dev Cell3(1):85–97.

24. Yu G, et al. (2000) Nicastrin modulates presenilin-mediated notch/glp-1 signal trans-duction and betaAPP processing. Nature 407(6800):48–54.

25. Herrup K (2015) The case for rejecting the amyloid cascade hypothesis. Nat Neurosci18(6):794–799.

26. Pimplikar SW, Nixon RA, Robakis NK, Shen J, Tsai LH (2010) Amyloid-independentmechanisms in Alzheimer’s disease pathogenesis. J Neurosci 30(45):14946–14954.

27. Chávez-Gutiérrez L, et al. (2012) The mechanism of γ-Secretase dysfunction in familialAlzheimer disease. EMBO J 31(10):2261–2274.

28. Bentahir M, et al. (2006) Presenilin clinical mutations can affect gamma-secretaseactivity by different mechanisms. J Neurochem 96(3):732–742.

29. Shimojo M, Sahara N, Murayama M, Ichinose H, Takashima A (2007) Decreased Abetasecretion by cells expressing familial Alzheimer’s disease-linked mutant presenilin 1.Neurosci Res 57(3):446–453.

30. Cacquevel M, Aeschbach L, Houacine J, Fraering PC (2012) Alzheimer’s disease-linkedmutations in presenilin-1 result in a drastic loss of activity in purified γ-secretasecomplexes. PLoS One 7(4):e35133.

31. Heilig EA, Xia W, Shen J, Kelleher RJ, 3rd (2010) A presenilin-1 mutation identified infamilial Alzheimer disease with cotton wool plaques causes a nearly complete loss ofgamma-secretase activity. J Biol Chem 285(29):22350–22359.

32. Doody RS, et al.; Alzheimer’s Disease Cooperative Study Steering Committee; Sem-agacestat Study Group (2013) A phase 3 trial of semagacestat for treatment of Alz-heimer’s disease. N Engl J Med 369(4):341–350.

33. Nordberg A (2008) Amyloid plaque imaging in vivo: Current achievement and futureprospects. Eur J Nucl Med Mol Imaging 35(Suppl 1):S46–S50.

34. Villemagne VL, et al. (2008) The ART of loss: Abeta imaging in the evaluation ofAlzheimer’s disease and other dementias. Mol Neurobiol 38(1):1–15.

35. Raux G, et al. (2000) Dementia with prominent frontotemporal features associatedwith L113P presenilin 1 mutation. Neurology 55(10):1577–1578.

36. Tang-Wai D, et al. (2002) Familial frontotemporal dementia associated with a novelpresenilin-1 mutation. Dement Geriatr Cogn Disord 14(1):13–21.

37. Dermaut B, et al. (2004) A novel presenilin 1 mutation associated with Pick’s diseasebut not beta-amyloid plaques. Ann Neurol 55(5):617–626.

38. Shen J, Kelleher RJ, 3rd (2007) The presenilin hypothesis of Alzheimer’s disease: Evidencefor a loss-of-function pathogenic mechanism. Proc Natl Acad Sci USA 104(2):403–409.

39. Heilig EA, Gutti U, Tai T, Shen J, Kelleher RJ, 3rd (2013) Trans-dominant negativeeffects of pathogenic PSEN1 mutations on γ-secretase activity and Aβ production.J Neurosci 33(28):11606–11617.

40. Shen J, et al. (1997) Skeletal and CNS defects in Presenilin-1-deficient mice. Cell 89(4):629–639.

41. Duering M, Grimm MO, Grimm HS, Schröder J, Hartmann T (2005) Mean age of onsetin familial Alzheimer’s disease is determined by amyloid beta 42. Neurobiol Aging26(6):785–788.

42. Kumar-Singh S, et al. (2006) Mean age-of-onset of familial Alzheimer disease causedby presenilin mutations correlates with both increased Abeta42 and decreasedAbeta40. Hum Mutat 27(7):686–695.

43. Mehta ND, et al. (1998) Increased Abeta42(43) from cell lines expressing presenilin 1mutations. Ann Neurol 43(2):256–258.

44. Serneels L, et al. (2009) gamma-Secretase heterogeneity in the Aph1 subunit: Rele-vance for Alzheimer’s disease. Science 324(5927):639–642.

45. Guerreiro RJ, et al. (2010) Genetic screening of Alzheimer’s disease genes in Iberianand African samples yields novel mutations in presenilins and APP. Neurobiol Aging31(5):725–731.

46. Robles A, et al. (2009) Clinical picture of a patient with a novel PSEN1 mutation(L424V). Am J Alzheimers Dis Other Demen 24(1):40–45.

47. Raux G, et al. (2005) Molecular diagnosis of autosomal dominant early onset Alzheimer’sdisease: An update. J Med Genet 42(10):793–795.

48. Lu P, et al. (2014) Three-dimensional structure of human γ-secretase. Nature512(7513):166–170.

49. Moehlmann T, et al. (2002) Presenilin-1 mutations of leucine 166 equally affect thegeneration of the Notch and APP intracellular domains independent of their effecton Abeta 42 production. Proc Natl Acad Sci USA 99(12):8025–8030.

50. Kamimura K, et al. (1998) Familial Alzheimer’s disease genes in Japanese. J Neurol Sci160(1):76–81.

51. Cruts M, et al. (1995) Molecular genetic analysis of familial early-onset Alzheimer’sdisease linked to chromosome 14q24.3. Hum Mol Genet 4(12):2363–2371.

52. Hong S, Dissing-Olesen L, Stevens B (2016) New insights on the role of microglia insynaptic pruning in health and disease. Curr Opin Neurobiol 36:128–134.

53. Sevigny J, et al. (2016) The antibody aducanumab reduces Aβ plaques in Alzheimer’sdisease. Nature 537(7618):50–56.

54. Saura CA, et al. (2004) Loss of presenilin function causes impairments of memory andsynaptic plasticity followed by age-dependent neurodegeneration. Neuron 42(1):23–36.

55. Guo Q, et al. (1997) Alzheimer’s presenilin mutation sensitizes neural cells to apo-ptosis induced by trophic factor withdrawal and amyloid beta-peptide: Involvementof calcium and oxyradicals. J Neurosci 17(11):4212–4222.

56. Xia D, et al. (2015) Presenilin-1 knockin mice reveal loss-of-function mechanism forfamilial Alzheimer’s disease. Neuron 85(5):967–981.

57. Xia D, Kelleher RJ, 3rd, Shen J (2016) Loss of Aβ43 production caused by presenilin-1mutations in the knockin mouse brain. Neuron 90(2):417–422.

58. Yoo AS, et al. (2000) Presenilin-mediated modulation of capacitative calcium entry.Neuron 27(3):561–572.

59. Tu H, et al. (2006) Presenilins form ER Ca2+ leak channels, a function disrupted byfamilial Alzheimer’s disease-linked mutations. Cell 126(5):981–993.

60. Alattia JR, et al. (2013) Highly efficient production of the Alzheimer’s γ-secretaseintegral membrane protease complex by a multi-gene stable integration approach.Biotechnol Bioeng 110(7):1995–2005.

61. Chan SL, Mayne M, Holden CP, Geiger JD, Mattson MP (2000) Presenilin-1 mutationsincrease levels of ryanodine receptors and calcium release in PC12 cells and corticalneurons. J Biol Chem 275(24):18195–18200.

62. Bai XC, et al. (2015) An atomic structure of human γ-secretase. Nature 525(7568):212–217.

63. Lai MT, et al. (2003) Presenilin-1 and presenilin-2 exhibit distinct yet overlappinggamma-secretase activities. J Biol Chem 278(25):22475–22481.

64. Acx H, et al. (2014) Signature amyloid β profiles are produced by different γ-secretasecomplexes. J Biol Chem 289(7):4346–4355.

65. Sannerud R, et al. (2016) Restricted location of PSEN2/γ-secretase determines sub-strate specificity and generates an intracellular Aβ pool. Cell 166(1):193–208.

66. Quintero-Monzon O, et al. (2011) Dissociation between the processivity and totalactivity of γ-secretase: Implications for the mechanism of Alzheimer’s disease-causingpresenilin mutations. Biochemistry 50(42):9023–9035.

67. Fernandez MA, Klutkowski JA, Freret T, Wolfe MS (2014) Alzheimer presenilin-1mutations dramatically reduce trimming of long amyloid β-peptides (Aβ) by γ-secre-tase to increase 42-to-40-residue Aβ. J Biol Chem 289(45):31043–31052.

68. Sun L, et al. (2015) Structural basis of human γ-secretase assembly. Proc Natl Acad SciUSA 112(19):6003–6008.

Sun et al. PNAS | Published online December 5, 2016 | E485

BIOCH

EMISTR

YPN

ASPL

US

SEECO

MMEN

TARY

Dow

nloa

ded

by g

uest

on

Apr

il 27

, 202

1