Alzheimer’s disease-associated mutations increase amyloid precursor protein resistance to γ-secretase cleavage and the Aβ42/Aβ40 ratio

Mutations in the amyloid precursor protein (APP) gene and the aberrant cleavage of APP by γ-secretase are associated with Alzheimer’s disease (AD). Here we have developed a simple and sensitive cell-based assay to detect APP cleavage by γ-secretase. Unexpectedly, most familial AD (FAD)-linked APP mutations make APP partially resistant to γ-secretase. Mutations that alter residues N terminal to the γ-secretase cleavage site Aβ42 have subtle effects on cleavage efficiency and cleavage-site selectivity. In contrast, mutations that alter residues C terminal to the Aβ42 site reduce cleavage efficiency and dramatically shift cleavage-site specificity toward the aggregation-prone Aβ42. Moreover, mutations that remove positive charge at residue 53 greatly reduce the APP cleavage by γ-secretase. These results suggest a model of γ-secretase substrate recognition, in which the APP region C terminal to the Aβ42 site and the positively charged residue at position 53 are the primary determinants for substrate binding and cleavage-site selectivity. We further demonstrate that this model can be extended to γ-secretase processing of notch receptors, a family of highly conserved cell-surface signaling proteins.


Introduction
Alzheimer's disease (AD) is the most prevalent chronic neurodegenerative disease [1]. It is characterized by the formation of amyloid plaques in the brain that mainly consist of aggregated amyloid beta (Aβ) peptides. Aggregation-prone monomeric Aβ peptides are proteolytic cleavage products of amyloid precursor protein (APP), whose cleavage by β-secretase generates a membrane-bound fragment that contains the C-terminal 99 residues (C99). This fragment is further cleaved by γ-secretase into the intracellular AICD50 peptide and extracellular 37-42 amino-acid amyloid Aβ peptides. Aβ can oligomerize and form higher-order fibrils that give rise to Aβ plaques, whose primary component is the abnormally folded fibrillar form of Aβ [2,3]. It has also been hypothesized that the aberrant accumulation of the more insoluble and aggregation-prone Aβ42 (42 amino acid Aβ) over Aβ40 is an important trigger for the development and pathogenesis of AD [4]. In an alternative, non-amyloidogenic processing pathway, APP is sequentially cleaved by α secretase to generate a C-terminal 83-residue fragment, and then cleaved by γ-secretase into AICD50 and non-plaque-forming extracellular peptides [5]. Many familial AD (FAD)linked mutations have been found in the genes encoding γ-secretase and APP, which have led to the hypothesis that pathological processing of C99 and the subsequent formation of Aβ-containing plaques are causally related to AD pathogenesis [6]. Although the molecular basis of AD remains controversial, the amyloid hypothesis, extensively supported by genetics, cell biology, biochemistry and animal studies, has dominated the field of AD research [7][8][9].
γ-secretase is a multisubunit protease complex that cleaves several type Ι single-pass transmembrane proteins at residues within their transmembrane domains [10]. It is an integral membrane protein, comprising four parts: presenilin (PS), nicastrin, anterior pharynx-defective 1 and presenilin enhancer 2 [11,12]. PS is the catalytic subunit of γ-secretase [13], whereas the other three subunits are involved in assembly, maturation and stability of the complex [14,15]. The best-studied substrates of γ-secretase are APP [5,6] and Notch receptors [16], both of which are of great physiological and pathological importance.
Owing to the central role of γ-secretase in Aβ peptide generation and AD pathology, inhibitors and modulators of γ-secretase have been considered as AD therapeutics, yet little success has been achieved in clinical trials, possibly because of the complexity of substrate recognition and catalytic processing. Although mutations in the gene encoding the catalytic PS component of γ-secretase are strongly associated with early-onset FAD, many disease-associated mutations surprisingly cause a partial loss of function in the γ-secretase complex [17][18][19]. Although recently the near-atomic structure of γ-secretase has been solved [20][21][22], the relationship between γ-secretase and its FAD-linked APP substrates remains ambiguous.
In this study, we have established a variant of the Tango protein-protein interaction assay [23] as a simple and sensitive tool to investigate the cleavage of C99 by γ-secretase in cells. This assay measures the total amount of C99 cleaved, that is, the epsiloncleavage of C99 irrespective of the number of secondary cleavage events, and which we have therefore termed 'epsilon-cleavage' assay ( Figure 1a). We tested the effect of all FAD-linked C99 mutations and found that a significant portion of FAD-linked APP mutant proteins became partially resistant to γ-secretase cleavage. In addition, the APP mutations can be divided into two distinct groups, based on whether the mutations alter residues N-terminal (group 1) or C-terminal (group 2) to the Aβ42 cleavage site. Group 1 mutations have subtle effects on cleavage efficiency and Aβ40/Aβ42 specificity. In contrast, group 2 mutations cause partial resistance to γ-secretase cleavage and dramatically skewed cleavage specificity toward Aβ42 production. Unexpectedly, mutations that change the positively charged property at residue 53 nearly abolish APP cleavage by γ-secretase. On the basis of these data, we propose a substrate and charge-recognition model for γ-secretase and APP, which we have validated by Notch substrate cleavage. These results provide a structure-activity relationship profile of the FAD-linked APP mutations as well as molecular insights into substrate recognition and cleavage of APP and Notch receptors by γ-secretase.

Results
The epsilon-cleavage assay allows sensitive detection of Aβ cleavage in cells To investigate the relationship between γ-secretase and its C99 substrate, we developed the γ-secretase epsilon-cleavage assay, which was derived from the cell-based Tango assay [23] and was similar to the Gal4-UAS or Gal4-VP16 systems used previously for Aβ cleavage [24,25]. The Tango assay is a proteinprotein interaction assay, in which a membrane-bound interaction partner is fused to a synthetic tetracyclinecontrolled transcriptional activator (tTA, comprising a fusion of the tetracycline repressor-TetR with a tandem repeat of the C-terminal activation domain of herpes simplex virus-VP16 [26]) and a TEV protease cleavage site, whereas a soluble interaction partner is fused to TEV protease. Association between the two interaction partners leads to an efficient cleavage of the tTA moiety by TEV protease, which allows tTA to move to the nucleus to activate a luciferase reporter gene. We adopted the Tango assay to test the cleavage of APP by γ-secretase by fusing C99 with a reverse tetracycline-controlled transactivator (rTA), which can mediate tetracycline-independent gene activation upon APP cleavage (Figure 1a). We also introduced T4 lysozyme (T4L) between C99 and rTA to increase substrate stability, and a C-terminal FLAG tag for immunological detection. As a control, we generated the same construct lacking rTA. We then coexpressed these constructs together with the four subunits of human γ-secretase in HEK293-derivative cells with a stably integrated TA-responsive luciferase reporter (HTL cells) [23]. A more than 17 × increase in luciferase activity was detected in cells transfected with the C99-T4L-rTA-fused activator or C83-T4L-rTA-fused activator and γ-secretase (Figure 1b and Supplementary Figure S1a). Thus, the Tet-based assays appeared to be more sensitive than the GAL4-based systems [24,25]. γ-secretase expression was measured by Compared with a standard Tango assay to detect the interaction between a constitutively active rhodopsin (Rho(4M)) and a pre-activated arrestin (Arr(3A)) [27], the γ-secretase epsilon-cleavage assay showed a similar reporter activity for the C99 fusion construct when combined with the endogenous γ-secretase, but three times higher reporter activity for the C99 fusion construct or C83 fusion construct when co-transfected with γ-secretase (Figure 1b and Supplementary Figure S1a). In contrast, the C99 hybrid lacking rTA failed to elicit reporter gene activity above background (Figure 1b). We further validated the assay by testing the effect of seven different γ-secretase inhibitors [28][29][30][31][32], all of which reduced C99 cleavage to various degrees ( Figure 1c). These inhibitors did not affect the reporter activity in the Rho-Arr Tango assay (Supplementary Figure S3), indicating the inhibitor specificity. Inhibition was dose-dependent, as shown in Figure 1d for the two most potent inhibitors, LY-411575 and compound E, as expected for specific inhibition. The complete inhibition by these two inhibitors suggests that the luciferase activity can be all attributed to the γ-secretase cleavage activity toward the C99 substrate reporter.
To further demonstrate that the generated signal is indeed due to cleavage by γ-secretase, as well as to abolish background due to cleavage by endogenous γ-secretase, we introduced internal deletions into both chromosomal isoforms of PS genes, PS1 and PS2, in the HTL cell line using the CRISPR/Cas9 method [33,34] (see Materials and Methods). Successful genomic deletions were confirmed by immunoblotting with antibodies that recognize the mature PS1 C terminus and the PS2 C terminus (Figure 1e), as well as by quantitative reverse transcriptase polymerase chain reaction (qRT-PCR; Supplementary Figure S5a), PCR and sequencing of the reversibly transcribed PS1 and PS2 mRNAs (Supplementary Figure S5b and c). In the absence of co-transfected γ-secretase expression plasmids, reporter gene activity due to C99 or C83 cleavage was abolished in the PS1/PS2-deleted HTL cells, but could be completely restored by co-transfection with wild-type (WT) PS1 (Figure 1f and Supplementary Figure S1b). PS1 protein expression was shown in Supplementary Figure S2e. This indicates that the relatively high reporter gene signal in the absence of exogenous γ-secretase is indeed completely due to C99 cleavage by endogenous γ-secretase, rather than to cleavage by other proteases. Together, these results demonstrate that we have developed a sensitive cell-based assay for an easy and fast measurement of C99 cleavage by endogenous or transfected γ-secretase in cells.

Most AD pathogenic C99 mutations reduce C99 cleavage by γ-secretase
Mutations in the C99 fragment have been associated with clinical onset of AD; however, how these mutations affect their cleavage by γ-secretase has not been systematically examined. To address this question, we introduced all 28 FAD-linked APP single point mutations listed in the Alzforum database (http:// www.alzforum.org/mutations, see Supplementary  Table S2 for the corresponding C99 positions) into the C99 substrate reporter for the epsilon-cleavage assays. These mutated APP reporters were also introduced to HTL ΔPS1/PS2 cells, in which C99 is not cleaved, for expression-level validation. As shown in Figure 2a, the majority of the pathogenic APP mutations decreased C99 cleavage by γ-secretase, and C99 T43I was almost completely resistant to γ-secretase cleavage. Only 8 of the 28 mutant proteins, the majority of which resided N-terminal to the Aβ42 site (group 1), did not show a statistically significant decrease in reporter gene activity ( Figure 2a). Signal decrease was not due to low expression levels as shown by immunoblotting (Figure 2b). The observed decrease in γ-secretase cleavage toward the FAD-linked APP-mutated proteins suggests that these are loss-offunction mutations. The simplest interpretation of these data is that most of the mutant proteins have reduced (or lost) their ability to be efficiently recognized and thus cleaved by γ-secretase. The loss-offunction mutations of FAD-linked APP resemble the scenario of FAD-linked γ-secretase mutations, the majority of which are also loss-of-function mutations [17][18][19]. Although the β-amyloid hypothesis for AD pathogenesis has paramount supporting evidence, including cellular and animal toxicity of Aβ aggregates, the loss-of-function mutations in FAD-linked APP and γ-secretase remain difficult to be explained and could have important implications in drug discovery for AD. The recent failure of the semagacestat phase III clinical trial adds further complexity for the causal relationship of APP cleavage and γ-secretase activity with AD [35,36]. The loss-of-function mutations in FAD-linked APP and γ-secretase would suggest that non-selective γ-secretase inhibitors are unlikely to be effective treatments for AD.

Production of Aβ peptides
γ-secretase can cleave C99 at multiple sites, generating intermediate extracellular cleavage Recognition of APP and notch receptors by γ-secretase products of 43, 45, 46, 48, 49 and 51 amino acids that are further cleaved to the main final products Aβ40 and Aβ42 [37][38][39][40][41][42] (see Figure 3a). To gain insight into cleavage selectivity, we analyzed the effect of FAD-linked C99 mutants on production of Aβ40 and Aβ42, using a commercial AlphaLISA assay. This assay detects immunologically an epitope that is common to both Aβ40 and Aβ42 (VFFAE) as well as an Aβ42-selective epitope (GGVVIA; see Figure 3a) to determine the levels of Aβ40 and Aβ42, using purified peptides for calibration (Supplementary Figure S6). Wild-type C99 cleavage yielded almost 10 times higher levels of Aβ40 than Aβ42 (Figure 3b versus 3c, Figure 3d inlet), in agreement with the reported ratio [43]. Consistent with the results from the epsiloncleavage assays (Figure 2), most of the FAD-linked APP mutations reduced Aβ40 cleavage by γ-secretase (Figure 3b). The FAD-linked C99 proteins had similar expression levels (Figure 3e), indicating that differences in Aβ peptide levels are predominantly due to changes in C99 cleavage efficiency by γ-secretase. All mutations altering amino acids N-terminal to the Aβ42 cleavage site were associated with Aβ42 levels close to or below that of the WT APP. In striking contrast, all mutations affecting amino acids in the transmembrane helix C terminal to the Aβ42 cleavage site caused dramatically increased Aβ42/Aβ40 ratios (Figure 3d), mostly because of strong increases in Aβ42 levels together with significant decreases in Aβ40 levels (Figure 3b and c). The dramatic increase in Aβ42 cleavage of V44 and I45 mutations is consistent with the data reported previously [44,45]. We only found trace levels of Aβ in cell lysates, which were not significantly higher than those of the control cells (Supplementary Figure S7), consistent with the release of Aβ peptides into the extracellular space. Together, these data have important implications in substrate recognition and cleavage of APP by γ-secretase as discussed below. E22del  E22G  E22K  E22Q  D23N  L34V  A42T  T43A  T43I  V44A  V44M  I45F  I45V  I45T  V46F  V46G  V46I  V46L  T48P  L52P  K53N   0   25   50   75   100   125 Relative Activity    A2V  H6R  D7H  D7N  E11K  K16N  A21G  E22del  E22G  E22K  E22Q  D23N  L34V  A42T  T43A  T43I  V44A  V44M  I45F  I45V  I45T  V46F  V46G  V46I  V46L  T48P  L53P

Recognition of APP and notch receptors by γ-secretase 6
Determinants for γ-secretase substrate recognition and cleavage-site selectivity Our results demonstrate that most FAD-linked APP mutant proteins are partially resistant to cleavage by γ-secretase. On the basis of the locations and effects of mutations, the FAD-linked APP mutations can be categorized into two distinct groups. The first group of mutations changes residues N-terminal to the Aβ42 site. Except for mutations in residue D7, mutations in this group show relatively subtle effects on both C99 total cleavage and Aβ42/Aβ40 selectivity and do not exhibit a clear pattern to link these mutations to their processing by γ-secretase (Figures 2 and 3). The second group mutations reside in the C-terminal portion of the Aβ42 (residues 42-53), which appears to be a hotspot of FAD-linked mutations ( Figure 4). All FAD-linked mutations in this small region strongly increase the Aβ42/Aβ40 ratio (Figure 3d), with a decrease in the overall cleavage efficiency (Figure 2a) and Aβ40 cleavage (Figure 3b) that is accompanied by an increase in Aβ42 cleavage (Figure 3c).
Currently, there is little information regarding the mechanism of substrate recognition and cleavage-site selectivity of APP by γ-secretase. On the basis of the relation between the sites of FAD-linked APP mutations and γ-secretase cleavage (Figures 2a and 3a), we derive a model of substrate recognition of APP by γ-secretase (Figure 5a). In this model, we propose that the C-terminal region of the APP transmembrane domain (residues 42-53) is the primary substrate recognition/binding site of γ-secretase because most mutations in this region significantly reduce APP  Figure 4 Summary of γ-secretase cleavage efficiency and selectivity for FAD-linked C99 mutant proteins. Top: region of C99 affected by AD-associated mutations together with predicted secondary structure (predicted by PSIPRED Protein Sequence Analysis) [55]. The amino acids of the predicted C99 transmembrane helix are highlighted by a green box, and γ, ϵ and ζ γ-secretase cleavage sites are indicated by arrows. The detection antibody epitopes are indicated by red letters. Middle: amino acids altered by FAD-linked C99 mutations are color-coded based on their effect on total cleavage by γ-secretase and the relative cleavage into Aβ42 and Aβ40 peptides. Red letters indicate mutations in the epitope regions, which are not compared with WT in Aβ40 and β42 production and Aβ42/β40 ratio. Lower panel: color code and symbols for secondary structure elements.
Ting-Hai Xu et al.
cleavage by γ-secretase (group 2 in Figures 2a and 3a). In contrast, mutations in APP residues N-terminal to Aβ 42 have much less effect. Additional substrate recognition and binding may be contributed by residue K53 C-terminal to the APP transmembrane helix, and to a lesser extent by residues D7 from the N terminus of APP because APP mutations in both positions show decreased cleavage efficiency by γ-secretase (Figure 2a).
The above model of APP substrate recognition by γ-secretase helps to resolve a key puzzle regarding the cleavage-site selectivity of APP. It has been established that Aβ42 and Aβ40 are generated by sequential tri-peptide cleavage of Aβ48 (or Aβ51) and Aβ49, respectively, which are the initial ϵ-cleavage sites of γ-secretase [42]. However, it remained unclear why mutations C-terminal to the Aβ42 cleavage site affect the cleavage selectivity between Aβ42 and Aβ40 if the mutation site is already removed before the Aβ42/40 cleavage based on the sequential cleavage steps. Our model of substrate recognition indicates that binding of the APP segment C-terminal to Aβ42 determines the initial ϵ-cleavage sites of γ-secretase (Figure 5b). For WT APP, we reason that the default ϵ-cleavage site is preferential for Aβ49 over Aβ48 as the Aβ42/Aβ40 ratio is~1-10. Mutations that alter the C terminus of APP affect not only substrate-binding efficiency but also the initial cleavage-site selectivity, some of which tip the γ-secretase ϵ-cleavage sites toward Aβ48, which subsequently generates Aβ45 and Aβ42, thus resulting in a higher ratio of Aβ42/Aβ40. This reasoning is consistent with that mutations in the C-terminal APP, including T48P, L52P and K53N (Figure 3), are near the initial γ-secretase ϵ-cleavage sites at Aβ49 or Aβ48. The proposed model of the C-terminal APP amino acids (residues 42-53) as the major determinants for substrate binding and cleavage-site selectivity is consistent with the catalytic site of γ-secretase revealed by the cryo-electron microscopy (cryo-EM) structure [22] (Figure 6a), which is located at the cytoplasmic side that is predicted to be close to the C terminus of the APP transmembrane helix.
There are two predictions from the proposed model of substrate recognition of APP by γ-secretase. The first prediction is that the C-terminal portion of the APP transmembrane helix will be much closer to γ-secretase than the APP N-terminal portion; thus, mutations introducing large side-chain residues at the C-terminal portion of the APP transmembrane domain (TMD) will have more pronounced effects than the mutations in the N terminus of APP TMD. As shown in Figure 6b, large residue mutations at the APP TMD C-terminal portion greatly reduce its cleavage by γ-secretase, whereas corresponding mutations at the APP TMD N-terminal portion have little effect. The second prediction is that mutations in D7 and K53, other than their naturally occurring APP mutations, could also affect γ-secretase cleavage efficiency. This is indeed the case as shown in Figure 6c. D7 mutations (D7A, D7R, D7W and D7Y) show similar effect as the FAD-linked mutations (D7H and D7N). Strikingly,  WT  G29R  G29W  G29Y  A30R  A30W  A30Y  T43R  T43W  T43Y  V46R  V46W  V46Y  T48R  T48W  T48Y  L49W  L49Y   0   mutations at K53 that remove its positive charge nearly abolish APP cleavage by γ-secretase, whereas K53R mutation does not (which actually increases cleavage), indicating a requirement for a positively charged residue (K53 or R53) for cleavage of APP by γ-secretase (Figure 6c). The effect of K53 mutation is specific as the mutations in the following positively charged residues K54 and K55 have little effect (Figure 6c). The γsecretase is an aspartic protease with two negatively charged residues at its catalytic site. On the basis of the charge distribution of human γ-secretase (PDB code: 5A63 [22]) and the APP TM domain (PDB code: 2LP1 Model extension to Notch substrates. (a) Sequence alignment of a subset of known γ-secretase substrates (human APP and human Notch substrates) within the known α-like cleavage site and the 10 N-terminal residues of the intracellular domain (ICD). The column of arginine and lysine residues highlighted in the red box is the conserved positively charged TM junction residue corresponding to K53 in C99. The green bracketed region corresponds to the hydrophobic TM domain. The ϵ-cleavage sites and α-secretase cleavage site are indicated by black arrows and triangle, respectively. Blue: positively charged residues; red: negatively charged residues; green: hydrophobic residues; purple: polar uncharged residues. (b) Epsilon-cleavage assay cleavage efficiency of Notch1 proteins with mutations within the TM domain indicate that the C-terminal region of Notch 1 TMD is more important than its N-terminal TMD for cleavage. The corresponding amino acids of Notch1 are shown below the graph. Red letters indicate the mutated residues. (c) Replacement of the positively charged TM junction residue in the four human Notch-mini proteins with alanine or glutamic acid strongly reduced notch cleavage (error bars = s.e.m., n = 3, *Po 0.05; **P o0.01; ***Po0.001 (versus WT)).
Recognition of APP and notch receptors by γ-secretase [46]), we propose a charge-recognition model for γ-secretase substrate recognition, in which the positively charged cluster at K53 would form electrostatic interactions with the negatively charged catalytic pocket (Figure 6a).

A general model of γ-secretase substrate recognition
Besides APP, γ-secretase processes many other substrates, including the family of Notch receptors [47,48]. We further tested whether our model of APP recognition by γ-secretase can be extended to Notch receptors. First, we introduced human Notch substrates within the known α-like cleavage site and 10 N-terminal residues of the intracellular domain to our system ( Figure 7a) and focused on the TM helix of Notch 1. Similar to APP mutations, mutations introducing large residues at the C-terminal portion of the Notch 1 TM helix have much bigger effects than the corresponding mutations at the N-terminal portion (Figure 7b), suggesting that the C-terminal portion of Notch 1 TMD is the primary recognition site of γ-secretase. We then mutated the positively charged residues that follow the Notch TMD to alanine or a negatively charged glutamic acid. In all four Notch receptors, these mutations dramatically decreased γ-secretase cleavage efficiency (Figure 7c), in excellent agreement with the model of APP substrate recognition by γ-secretase. On the basis of sequence alignment of 69 γ-secretase substrates (Supplementary Figure S8), a positively charged residue is 100% conserved at the TM junction (three to four residues after the ϵ-cleavage sites), followed by a cluster of positively charged residues. In the case of N-cadherin, this cluster of positively charged residues is required for its cleavage by γ-secretase [49]. The absolute requirement of the positively charged residues at K53, together with the similarity of the TMD mutation profile between APP and Notch, indicates that the model of APP recognition and cleavage is generally applicable to other γ-secretase substrates, including Notch receptors.

Discussion
In this study, we have developed a simple cell-based γ-secretase epsilon-cleavage assay to determine the effects of the known FAD-linked APP mutations on C99 cleavage. Together with the measurements of the levels of the most critical cleavage products, Aβ40 and Aβ42, these experiments provide three important observations. First, most FAD-linked APP mutations cause partial resistance to cleavage by γ-secretase (Figure 2a), suggesting that these mutants are less efficiently recognized by γ-secretase. Second, FAD-linked APP mutations can be categorized into two distinct groups based on the locations and effects of mutations. Only mutations that affect residues C-terminal to the Aβ42 cleavage site (residues 42-53, which are a hotspot of FAD-linked mutations; Figure 4) markedly affect cleavage efficiency and strongly increase the Aβ42/Aβ40 ratio. These data suggest that the main γ-secretase recognition site is C-terminal to amino acid 42, and that mutations in this region determine the Aβ42/Aβ40 ratio by skewing the relative efficiencies of the ϵ-cleavage sites at residues 48 or 51 (precursors for Aβ42) and 49 (precursor for Aβ40; Figure 5b). These observations are consistent with previous results for a partial list of FAD-linked APP mutations [50][51][52][53]. Third, mutations that remove the positive charge from the invariant lysine or arginine residue at the TM junction (K53 for C99) greatly compromise the cleavage efficiency of APP by γ-secretase. Analysis of the charge distribution of γ-secretase reveals a negatively charged catalytic center that would require a positively charged residue near the initial ϵ-cleavage sites. This conclusion is supported by the conservation of K53 in many γ-secretase substrates (Supplementary Figure S8).
Together, the three key observations above have led us to propose a model of γ-secretase substrate recognition as illustrated in Figure 5a. In this model, the APP region C-terminal to the Aβ42 site serves as the primary site for γ-secretase recognition by being physically closer to γ-secretase than the APP N-terminal region. One possible mechanism for the effect of FAD-linked APP mutations on γ-secretase cleavage is that these APP mutations could cause tilting of the TMD helix, thereby altering the presentation of the substrate to γ-secretase and changing the initial ϵ-cleavage-site selection as proposed previously [53]. Additional determinants for APP recognition by γ-secretase reside in the flanking regions of the APP TM helix, particularly the positively charged residue K53. This model is consistent with all naturally occurring FAD-linked APP mutations (Figures 2  and 3), and we further validate this model extensively with additional mutations that introduce large side-chain residues in the APP TMD helix or mutations that remove the positive charge at residue 53 (Figure 6b and c). A recent study has proposed that the major γ-secretase substrate-binding site resides in the substrate TMD, whereas the ectodomain of substrate is largely dispensable for cleavage [54]. Our model is completely in agreement with this study and further localizes the γ-secretase substrate binding to the TMD region C-terminal to Aβ42 and the conserved positively charged residue at position 53. Importantly, the key features of this model of the APP substrate recognition by γ-secretase are highly conserved in Notch receptors, another major class of γ-secretase substrates (Figure 7). Given the extensive list of γ-secretase substrates, the results and the model presented here should have important implications for how diverse substrates are processed by γ-secretase.

Materials and Methods
Cell culture HTL cells were a gift from G Barnea and R Axel (Brown University and Columbia University). They are derived from HEK293 cells with a stably integrated luciferase reporter under the control of the bacterial tetO operator element. We replaced the tTA with a rTA for increased, tetracycline-independent gene activation [27]. The cells were routinely grown in Dulbecco's modified Eagle's medium (Invitrogen Life Technologies, Grand Island, NY, USA) supplemented with 10% (v/v) fetal bovine serum (Invitrogen Life Technologies) at 37°C under a humidified 5% CO 2 atmosphere.

PS1/PS2 deletion cell line
The HTL PS1/PS2-deleted cell line was established by the CRISPR/Cas9 method. In order to express two pairs of single guide RNAs, the expression cassette under the control of the human U6 promoter was repeated four times in the modified pX458 vector (Supplementary Figure S4a). Pairs of single guide RNAs targeting two different sequences within PS1 and PS2, respectively, were designed using the CRISPR Design Tool  γ-secretase epsilon-cleavage assay HTL cells were split at 50 000 per well in a 24-well plate. After 1 day of growth, the cells were transfected with 65 ng total DNA using X-tremeGENE 9 Reagent (Roche Diagnostics, Indianapolis, IN, USA) with a ratio of 1:3 (1 µg DNA:3 µl reagent) according to the standard protocol. For coexpression with γ-secretase, 20 ng substrate-encoding DNA, 5 ng phRG-tk Renilla normalization standard and 10 ng of each of the γsecretase subunit expression plasmids were transfected. For substrate only, 20 ng substrate, 5 ng phRG-tk Renilla and 40 ng pBSK mock plasmid were transfected with or without proper amount of inhibitors. For the control Tango assay, 10 ng Rho (4M)-TEV-site-rTA, 10 ng Arr(3A)-TEV, 5 ng phRG-tk Renilla and 40 ng pBSK mock plasmid were co-transfected. Cells were harvested and lysed the following day. Luminescence activities were measured using the Dual Luciferase Kit (Promega, Madison, WI, USA) according to the manufacturer's instructions. Renilla luciferase serves as transfection control. Relative activity is normalized activity using WT C99-T4L-rTA activity as 100.

RNA isolation and quantitative real-time PCR
Total RNA was extracted from cells with TriZol (Invitrogen Life Technologies), followed by chloroform extraction and isopropanol precipitation. The precipitate was washed with 0.5 ml 75% ethanol, pelleted, air-dried and dissolved in RNase-free water. The SuperScript First-Strand Synthesis Kit (Invitrogen Life Technologies) was used for first-strand complementary DNA synthesis according to the manufacturer's instructions. Quantitative real-time PCR amplification was carried out in 10 μl solution containing 5 μl Power SYBR Green Real-Time PCR Master Mix (Applied Biosystems, Foster City, CA, USA), 2.5 μl 1:5 diluted complementary DNA sample and 2.5 μl PCR primer mix (forward and reverse each 0.8 μM) using a Step One Plus Thermocycler (Applied Biosystems). Human GAPDH primers (Supplementary Table S1) were used as internal control. The value of threshold cycle (Ct) was generated at every cycle during a run. Fluorescent readings from real-time PCR reactions were quantitatively analyzed by determining the difference of Ct (ΔCt) between the PS and GAPDH genes. The gene expression of the PS1 or PS2 gene was determined as 2 − ΔCt . The relative gene expression was expressed as percentage of WT control.

AlphaLISA assay
To get a better profile of Aβ production, we used C99-Flag as substrates and the concentration of transfected DNA was optimized by preliminary tests and set to 320 ng per well with 0.96 µl X-tremeGENE 9 Reagent. After 1 day of growth, the assay was performed as a standard two-step protocol, in which 5 μl cultured supernatant (or cell lysate) were incubated with 5 μl AlphaLISA Aβ1-40/42 acceptor beads and biotinylated anti-Aβ antibody at 23°C for 1 h, followed by another 30-min incubation with 10 μl AlphaLISA Aβ1-40/42 donor beads in the dark at 23°C. Photon counts were determined in 384 plates using an Envision-Alpha Reader (PerkinElmer, Waltham, MA, USA).

Site-directed mutagenesis
All site-directed mutagenesis was carried out using the QuikChange method (Agilent). All constructs were confirmed using DNA sequencing.