Familial Alzheimer's disease mutations inhibit γ-secretase-mediated liberation of β-amyloid precursor protein carboxy-terminal fragment


Address correspondence and reprint requests to Mark Bothwell, University of Washington, Department of Physiology and Biophysics, Box 357290, Rm G424 Health Science Building, 1959 Pacific Way NE, Seattle, WA 98195, USA. E-mail: mab@u.washington.edu


Cleavage of the β-secretase processed β-amyloid precursor protein by γ-secretase leads to the extracellular release of Aβ42, the more amyloidogenic form of the β-amyloid peptide, which subsequently forms the amyloid-plaques diagnostic of Alzheimer's disease. Mutations in β-amyloid precursor protein (APP), presenilin-1 and presenilin-2 associated with familial Alzheimer's disease (FAD) increase release of Aβ42, suggesting that FAD may directly result from increased γ-secretase activity. Here, we show that familial Alzheimer's disease mutations clustered near the sites of γ-secretase cleavage actually decrease γ-secretase-mediated release of the intracellular fragment of APP (CTFγ). Concordantly, presenilin-1 mutations that result in Alzheimer's disease also decrease the release of CTFγ. Mutagenesis of the epsilon cleavage site in APP mimicked the effects of the FAD mutations, both decreasing CTFγ release and increasing Aβ42 production, suggesting that perturbation of this site may account for the observed decrement in γ-secretase-mediated proteolysis of APP. As CTFγ has been implicated in transcriptional activation, these data indicate that decreased signaling and transcriptional regulation resulting from FAD mutations in β-amyloid precursor protein and presenilin-1 may contribute to the pathology of Alzheimer's disease.

Abbreviations used

Alzheimer's disease


β-amyloid precursor protein


carboxy-terminal fragment


Dulbecco's modified Eagle's medium


enzyme-linked immunosorbent assay


familial Alzheimer's disease


Notch intracellular domain






Radio Immuno Precipitation Assay


wild type

Familial Alzheimer's disease (FAD) is an autosomal dominant disorder caused by missense mutations in β-amyloid precursor protein (APP), presenilin-1 (PS1) and presenilin-2 (PS2) (Selkoe 2001). APP is a type I transmembrane protein that is sequentially proteolytically processed by α-, β-, and γ-secretase. APP cleavage by α-secretase results in the release of the large extracellular domain and the generation of the C83 membrane-resident fragment. β-secretase cleavage of APP, mediated by the BACE atypical aspartyl protease (Vassar et al. 1999), also liberates the extracellular domain and generates the C99 membrane-resident fragment. Subsequently, C83 and C99 are processed by the recently characterized heterotetrameric γ-secretase complex consisting of Nicastrin, Aph-1, Pen-2, and PS1/2 (Francis et al. 2002; Edbauer et al. 2003; Kimberly et al. 2003; Takasugi et al. 2003). γ-Secretase cleavage of C83 and C99 produces the innocuous p3 peptide and the pathogenic Aβ peptide, respectively, and releases the intracellular fragment of APP from the membrane (referred to in this work as CTFγ). PS1 and PS2 form the catalytic core for the γ-secretase processing of APP (Schroeter et al. 2003; Cervantes et al. 2004). Generation of FAD pathology by mutations in either the APP substrate or the PS1/2 cleaving enzyme indicates that the etiology of the disorder resides in some alteration in the enzyme–substrate relationship.

Multiple alternative sites of cleavage of APP by γ-secretase generate a variety of Aβ peptides, of which Aβ40 and Aβ42 are the most studied. The FAD mutations in APP, PS1 and PS2 elevate the production of Aβ42, normally a minor Aβ peptide species, while having variable effects on production of Aβ40 (Suzuki et al. 1994; Borchelt et al. 1996; Duff et al. 1996; Scheuner et al. 1996; Citron et al. 1997; Murayama et al. 1999). Aβ42 is less soluble and has increased aggregative properties, relative to Aβ40, which are thought to promote Aβ plaque formation (Burdick et al. 1992; Xia et al. 1997). The consistency of this observation within cell culture, mouse models and human clinical investigations has led to the amyloid hypothesis – the primary etiological account of Alzheimer's disease (AD) pathology (Hardy and Selkoe 2002). A fundamental tenet of this hypothesis is that the mutations present in APP and PS1/PS2 alter γ-secretase activity to promote cleavage at the Aβ42 specific site. Consequently, the amyloid hypothesis ascribes a ‘gain of function’ to the FAD mutations.

Countervailing observations with Notch signaling, another γ-secretase substrate, suggest that the FAD mutations in the presenilins are ‘loss of function’ with respect to release of the Notch intracellular domain (NICD) (Baumeister et al. 1997; Song et al. 1999; Schroeter et al. 2003). Recently, γ-secretase mediated cleavage of APP was shown to initiate transcription in a manner that is reminiscent of Notch. APP forms a heterotrimeric complex with Fe65 and Tip60. γ-Secretase-mediated release of CTFγ permits nuclear accumulation of a transcriptionally active CTFγ/Fe65/TIP60 complex (Cao and Sudhof 2001; Baek et al. 2002). The perception that FAD mutations are ‘gain of function’ mutations for APP cleavage has led some to the reasonable speculation that enhanced release of a cytotoxic CTFγ fragment might contribute to the neuropathology of FAD (Passer et al. 2000; Kinoshita et al. 2002; Kim et al. 2004). However, several studies have suggested that FAD PS1 mutations might actually decrease release of CTFγ (Chen et al. 2002; Schroeter et al. 2003). In the present study, we have examined the influence of a wide variety of FAD mutations of both APP and PS1 on CTFγ production from APP in cultured cells. Because accurate quantification of the quickly degraded CTFγ product by direct biochemical means is problematic, in the present study we monitor CTFγ production employing an APP-Gal4VP16 fusion construct which yields γ-secretase-dependent transcription of a luciferase reporter plasmid. Our results demonstrate that a variety of FAD mutations in both APP and PS1 lead to a decrement in CTFγ release from the plasma membrane. We discuss models that may reconcile the observation that FAD mutations have both ‘gain of function’ and ‘loss of function’ attributes.

Experimental procedures

Plasmid generation and mutagenesis

The human APP695 cDNA was PCR amplified with 36 nucleotide primers containing a HindIII site in the 5′ primer and a SpeI site in the 3′ primer, which also deleted the stop codon. The 2-kb PCR product was purified, digested and subcloned into pCEFL (expression vector driven by EF1α promoter, gift of Dr Silvio, NIH). pSGVP (a gift from Dr Mark Ptashne, Sloan-Kettering Institute) contains the Gal4-VP16 coding sequence. This construct was used as a template to amplify the Gal4-VP16 coding sequence with a SpeI site engineered into the 5′ primer and an XbaI site built into the 3′ primer. The Gal4-VP16 coding sequence was subcloned into the EF1α-APP695 vector, creating an APP695-Gal4VP16 fusion construct (referred to as APPGV16). The normalization vector, EFnLACZ was generated by PCR amplifying the nuclear targeted NLS-LacZ (gift of R. Palmiter, University of Washington) with a 5′ primer containing a NheI site and 3′ primer containing a HindIII site. The purified PCR product was digested and subcloned into pcDNA3.1 (Invitrogen, Carlsbad, CA, USA). The CMV promoter was subsequently removed and the EF1α promoter was subcloned into the NruI and NheI sites resulting in EF1α-NLS-LacZ (referred to as EFnLacZ).

Mutagenesis of APPGV16 was performed using an adaptation of the standard QuikChange mutagenesis procedure (Stratagene, La Jolla, CA, USA). APPGV16 DNA (200 ng) was used in each reaction with mutagenesis primers into which silent restriction sites were employed to diagnose successful mutagenesis by restriction digest analysis. The 45 nucleotide mutagenesis primers were designed to overlap with one another by 20–25 nucleotides, leaving 20–25 nucleotides of unhybridized sequence per primer. An initial extension reaction was run for four cycles of PCR prior to the addition of the template APPGV16 DNA, yielding an approximately 65-nucleotide 5′-and 3′- mutagenesis primer. The standard QuikChange conditions were used for the mutagenesis reactions, with the exception that 5% dimethylsulfoxide (DMSO) was added to the mutagenesis reactions. All APPGV16 mutations were restriction digestion mapped to ensure the presence of the silent restriction site incorporated into the primers, and subsequently confirmed by sequencing.

CMV-Gal4VP16 was generated by using the Gal4VP16 fragment of APPGV16. The SpeI-XbaI Gal4VP16 coding sequence was subcloned into pcDNA3.1 (Invitrogen). The CTFγ-Gal4VP16 construct was generated by amplifying the carboxy-terminal fragment of APP between the β-amyloid cleavage site and carboxy-terminus using primers that included a BamHI site in the 5′-primer and a SpeI site in the 3′-primer. The APP PCR product was used to make a fusion with CMV-Gal4VP16 cut with BamHI and SpeI, resulting in APP-CTFγ being subcloned into the 5′-end of Gal4VP16 in frame with the coding sequence.

Cell culture and transfection conditions

Primate kidney Cos7 cells were grown in Dulbecco's modified Eagle's medium (DMEM) supplemented with 10% fetal bovine serum (HyClone, Logan, UT, USA) and 1 × penicillin and streptomycin (Invitrogen) at 37°C in 5% CO2. Immortalized mouse embryonic fibroblasts (Nyabi et al. 2003), both the PS1–/–PS2–/– and wild type, were grown under the same conditions (obtained from Dr B. de Strooper, Flanders Interuniversity Institute for Biotechnology, Belgium). Rat adrenal pheochromacytoma PC12 cells were grown in DMEM supplemented with 10% horse serum (HyClone), 5% fetal bovine serum (HyClone) and 1 × penicillin/streptomycin (Invitrogen) at 37°C in 5% CO2. Stock plates were grown to approximately 50–80% confluence prior to splitting cells for transfection.

The PS1 stable cell lines were generated by transfecting Cos7 and HT22 cells with the dual Zeo cassette PS1 constructs (a gift from Dr Selkoe, Harvard University) and placing the cells under Zeomycin (Invitrogen) selection, starting 48 h post-transfection. The cells were grown under selection for several passages over a 3-week period and then used for experiments. The resulting cell lines were tested for equivalence of PS1 expression levels by western blotting using the Chemicon PS1 loop region antibody.

All transactivation assays were carried out using cells seeded onto 24-well plates. The plates were grown to 80–90% confluence and transfected using Lipofectamine 2000 (Invitrogen). Cos7 cells were transfected using 1.5 μL of lipofectamine 2000 reagent per well. Control experiments were carried out to determine that this concentration of lipofectamine was optimal for this system in Cos7 cells. PC12 cells and immortalized MEFs were transfected using 3 μL of lipofectamine 2000 per well. The transfection mix was incubated on the cells for 2–4 h, prior to a replacement of the medium of fresh growth medium.

In all transactivation assays, approximately 1 μg DNA was used per well of the 24-well plate. APPGV16 concentrations are defined in each experiment in which titration assays are performed. If it is not specified, then 100 ng/well of APPGV16 DNA was used. In all experiments, the promoter concentration was held constant by the addition of either pcDNA3.1 (Invitrogen) to normalize CMV–promoter concentrations (PS1 experiments) or pCEFL to normalize the amount of EF1α promoter present in each condition. In experiments in which APPGV16 was titrated across the experiment, pCEFL was counter titrated to ensure that there would not be any promoter competition or variation as a result of promoter saturation across conditions. The Gal4-luciferase reporter used was pFRluc (Invitrogen), which was used at 400 ng/well. In experiments also over-expressing PS1 isoforms, 250 ng/well of the reporter construct was used and 150 ng/well of the PS1 construct was used (unless concentrations provided in the figure). Normalization was done using 50 ng/well of the EFnLACZ construct.

Transfections for western blot were performed in 6-well plates at approximately 90% confluence. APPGV16 DNA (1 μg) was used for each isoform (corresponding to approximately 250 ng/well of DNA in the 24-well plate assay). The cells were incubated with transfection mix for 4–5 h prior to replacing the media with fresh growth media. The cells were lysed 48 h post-transfection for examination by western blot.

Cleavage assay and inhibitors

Luciferase and βGal assays were performed by standard methods as previously reported (Matthews et al. 1994; Wiley et al. 1999). The cells were lysed on ice 36–48 h post-transfection and luciferase assays were performed and point to point normalized to βGal values. For the assays in which parallel protein concentration levels were assessed, 1 × proteasome inhibitor mix (Sigma, St Louis, MO, USA) was added. In comparative studies, the addition of proteasome inhibitor mix was found to have no effect on the luciferase assay results. All measurements were performed on a EG & G Berthold LB 96V luminometer.

The γ-secretase inhibitors used in these experiments were generously provided by Dr M. Wolfe (Harvard University). The WPE-II compound had been found to be 10-fold more effective at blocking β-amyloid secretion with an IC50 of 100 nm, while DAPT had an approximate IC50 of 1 μm (Wolfe, personal communication). These compounds were used at the concentrations described in the figures. For experiments in which explicit concentrations are not given, WPE-II was used at 1 μm and DAPT was used at 10 μm. The cells were treated with WPE-II and DAPT γ-secretase inhibitors starting approximately 24 h post-transfection and remained in the inhibitors for 16–20 h prior to lysis.

Western blots, immunoprecipitations and antibodies

Protein expression level comparisons were made from lysates taken from the 6-well plates transfected with just the APPGV16 isoforms. The cells were lysed using standard RIPA buffer, protein concentrations were quantitated, and 50 μg of protein was loaded from each condition. The blots were probed using rabbit anti-Gal4 DNA binding domain antibody (Zymed, San Francisco, CA, USA). Comparison of APPGV16 expression levels in the transactivation assays were performed by removing 50 μL of lysate from each sample and transferring it to 65°C 1 × phosphate-buffered saline (PBS)/1% sodium dodecyl sulfate (SDS). The samples were run on 15–4% gradient gels and probed using the rabbit anti-Gal4 DNA binding domain antibody (US Biologicals, Swampscott, MA, USA). The immunoprecipitation of the CTFγ-GV16 was performed in cell lysates prepared from Cos7 cells transiently transfected with wild-type or mutant APPGV16 in 10-cm plates. The cells were lysed in cold RIPA buffer with 1 × protease inhibitor cocktail (Sigma). The immunoprecipitation was performed using the anti-Gal4 antibody (US Biologicals). The immunoprecipitates were run on a standard 10% polyacrylamide gel and the blots were probed with the same anti-Gal4 antibody used in the immunoprecipitations.

Aβ40 and Aβ42 ELISA assays

β-amyloid sandwich ELISA assays were performed by coating 96-well Nunc MaxiSorp plates with the 6E10 capture antibody (Signet Laboratories, Dedham, MA, USA) generated against the amino-terminal region of the Aβ peptide. Standard curves were generated by applying titrated quantities of purified Aβ40 and Aβ42 peptides (rPeptide, GA) to the plates and rocking at 4°C overnight. Assay samples were generated by transiently transfecting Cos7 cells and HT22 cells with 500 ng/well of either APPGV16 wild-type, or the S3 site mutants (S3m1 and S3m3) in 24-well plates in a procedure that matched the protocol used for the transactivation assays. The transiently transfected cells were incubated in 200 μL of DMEM supplemented with 10% fetal bovine serum (FBS) for 48 h. The medium was removed and pooled from the Cos7 cells and HT22 cells and added to the 6E10-coated Nunc plate along with the peptide standards. After incubating the medium overnight, the plate was washed and incubated with either the Aβ40- or the Aβ42-specific biotinylated secondary antibodies and rocked at room temperature (20°C) for 2 h. Following washes, the plates were then incubated with Avidin-HRP (Vector Laboratories, Burlingame, CA, USA) for 1 h. The assay was developed by adding 100 μL of TMB (Promega, Madison, WI, USA) and incubating in the dark for 30 min at room temperature, and stopped by the addition of 1 N H2SO4 and the absorbance at A450 was measured.


APPGV16 screen detects γ-secretase activity

The highly unstable CTFγ fragment of APP is present in cells only at very low concentrations, which are difficult to reliably quantitate by biochemical means. Consequently, we employed the APPGV16 activator, γ-secretase-mediated cleavage of which initiates transcription of a reporter plasmid encoding luciferase downstream of a promoter that contains Gal4-binding sites. Initially we validated this assay by demonstrating that luciferase expression shows the same dependence on γ-secretase activity as found for release of CTFγ from native APP. Various concentrations of two well-characterized γ-secretase inhibitors, WPE-II-89 and DAPT, were used in pharmacological titration assays. Both compounds effectively inhibited APPGV16-dependent luciferase activity, achieving greater than 80% inhibition at 10 μm (Fig. 2a). WPE-II-89 was effective at lower concentrations than DAPT, consistent with results obtained assaying inhibition of Aβ release (personal communication, Dr M. Wolfe, Harvard University). Similar results were obtained in Cos7 cells (data not shown). Thus, most of the APPGV16-dependent luciferase activity observed results from cleavage of APPGV16 by γ-secretase. This conclusion was confirmed in experiments employing cells lacking PS1 and PS2, essential components of the γ-secretase protease complex (De Strooper et al. 1998; Herreman et al. 2000). PS1–/–PS2–/– immortalized mouse embryonic fibroblasts produced much less GAL4VP16-dependent luciferase activity than did wild-type MEFs (Fig. 2b), and activity was restored by transfection of PS1 (Fig. 2c). Thus, both pharmacological inhibition and genetic targeting of γ-secretase blocks APPGV16 induced Gal4-luciferase transactivation. This confirms that the APPGV16/Gal4-luciferase assay is a valid measure of γ-secretase activity.

Figure 2.

APPGV16 assay detects γ-secretase activity. The efficacy of the APPGV16 reporter system to detect γ-secretase activity was tested pharmacologically using known γ-secretase inhibitors and genetically using PS1/PS2 null PMEFs. (a) PC12 cells were transiently transfected by lipofectamine 2000 with the APPGV16 construct, pFRluc (Stratagene) and EFnLacZ. Dual titrations were performed using two known γ-secretase inhibitors (WPE-II-89 and DAPT) obtained from Dr M. Wolfe, Harvard University. (Ki of DAPT > WPE-II-89; determined using Aβ release assays, M. Wolfe, personal communication). The PC12 cells were treated with the γ-secretase inhibitors for 15 h, starting 24 h post-transfection. (b) PS1–/–PS2–/– and wild-type immortalized MEFs (obtained from Dr B. de Strooper, Flanders Interuniversity Institute for Biotechnology, Belgium) were transfected with various quantities of APPGV16 plasmid. Generation of luciferase activity in the wild-type and PS1–/–PS2–/– MEFs differed 8–10-fold, achieving statistical difference across the titration curve (two-way anovap < 0.001). (c) PS1–/–PS2–/– cells were transiently transfected with 100 ng/well of APPGV16 and increasing amounts of PS1 wild-type (wt) DNA. PS1–/–PS2–/– MEFs were harvested and assayed 24 h post-transfection.

Familial Alzheimer's disease mutations in APP decrease γ-secretase-mediated cleavage

The APP mutations associated with FAD generally lie in the extracellular region proximal to the α-secretase and β-secretase cleavage sites (herein referred to as class I) or, more commonly, within the transmembrane domain adjacent to the γ-secretase cleavage sites (referred to as class II). Multiple class II FAD mutations were introduced into APPGV16 to assess the effect on γ-secretase-mediated cleavage (Fig. 1, noted in purple). FAD mutant and wild-type APPGV16 isoforms were examined in pair-wise comparisons between wild-type and specific FAD mutant forms of APPGV16. These experiments employed titrated concentrations of the pair of APPGV16 activators. Titrated levels of transfected APPGV16 DNA were used to increase confidence that the information was consistent across a large range of expression levels. Further, assessing extent of cleavage over a range of levels of APPGV16 provides additional information about the nature of the perturbations in the cleavage process – differences in cleavage at saturating concentrations would be loosely comparable with alterations of enzymatic Vmax, whereas differences in the concentration required to reach plateau activity levels could be coarsely related to enzymatic Km. As γ-secretase docking and proteolysis of APP are separate processes (Annaert et al. 2001; Esler et al. 2002; Tian et al. 2002, 2003), it was initially anticipated that FAD mutations might exclusively effect one of these components. All tested class II β-APP mutations significantly decreased APPGV16-dependent luciferase activity (Figs 3a–f), while the single class I mutation tested (Swedish) did not differ significantly from wild type (Fig. 3g). The decrement in activity observed occurred across the concentration spectrum, validating the consistency of the effect. However, the decrease in activity within the plateau range of the titration suggests that FAD mutations primarily influence catalysis rather than affinity of association of APP with γ-secretase.

Figure 1.

Topological pattern of FAD mutations in APP695. The FAD mutations in APP cluster around the proteolytic processing sites. The extracellular FAD mutations are shown relative to the α-secretase and β-secretase cleavage sites. The majority of APP FAD mutations occur in the transmembrane domain, immediately proximal to the known γ-secretase cleavage sites. The mutations in black are relative to APP695. In instances in which individual mutations are associated with a historical name, that name is also provided. Listed in red are the positions of the FAD mutations relative to the full-length APP770 isoform. Because APP must be cleaved in at least three places within the transmembrane domain, the cleavage site corresponding to each is highlighted. The site highlighted in orange corresponds to the Aβ40 promoting cleavage site, while the yellow highlighting represents the pathogenic Aβ42 cleavage site. The S3/Notch-like cleavage site is represented in blue. The KM595/596 Swedish mutation, T639I, V640M, I641V, V642 (I and G), and L648P were used within the present study to examine the effects of these mutations upon γ-secretase-mediated liberation of the APP intracellular domain (commonly referred to as CTFγ). A schematic representation of the peptide product resulting from γ-secretase cleavage of either the α- or β-secretase products color coded within the diagram [the p3 alpha cleavage product (blue), and the Aβ product (orange)].

Figure 3.

Comparison of wild-type and FAD mutant APP695GV16. Direct comparisons were made of wild-type and individual FAD mutant versions of APPGV16. In each case, titrations of wild-type and FAD APPGV16 were used to assess potential differences in γ-secretase-mediated cleavage at multiple APPGV16 DNA concentrations. As the neuronal APP695 isoform was used to generate the constructs, the positional nomenclature is relative to the APP695 isoform. The FAD mutations compared with wild type are (a) T639I/Austrian mutation, (b) V640M/French mutation, (c) I641V/Florida mutation, (d) V642I/London mutation, (e) V642G/London, (f) L648P/Australian mutation, and (g) KM595/596NL/Swedish mutation. To ensure that the plateau effect observed was not as a result of report saturation, titration of the holo-APPGV16 was compared with activity resulting from the pre-cleaved CTFγ-GV16 expression vector (h). Because of high βGal values, all luciferase/βGal ratios are multiplied by 50 000 to bring numbers into a consistently non-fractional range. Statistical analysis employing the two-way anova demonstrated that all mutations, except Swedish, produced significantly lower levels of activity (p < 0.01). The variation in absolute numbers is likely because of variation in the assay time between 24 and 48 h (which impacts the ratio of induced luciferase to constitutive βGal expression levels), and inter-experimental variation in transfection efficiency.

Observed differences between mutant and wild-type APPGV16 are not likely to be because of unintended disparities in expression level as mutant and wild-type activities differ in the plateau region of the plasmid titrations where cleavage is apparently independent of APPGV16 concentration. To verify that this plateau of activity results from saturation of γ-secretase enzyme with APP substrate, rather than merely reaching an inherent maximal limit of Gal4-promoter activation, the titration experiment was repeated employing a plasmid encoding a pre-cleaved form of the activator. CTFγ-GV16 gave much greater activation across the titration than did the holo-APPGV16. This confirms that the saturation phenomenon is not because of maximal promoter activation and supports the conclusion that the observed plateau is as a result of enzymatic saturation (Fig. 3h).

Further, to eliminate the possibility that the plateau of transcriptional activity with increasing quantities of APPGV16 DNA does not merely result from a plateau of APPGV16 protein expression levels at elevated concentrations of the APPGV16 DNA, we employed western blot analysis to assess APPGV16 protein concentrations within the lysate used to perform the transactivation assay. For wild-type, Austrian and Florida mutant isoforms of APPGV16, the concentration of protein expressed increased roughly linearly with increasing amounts of plasmid transfected (Fig. 4a). No plateau in APPGV16 protein expression was observed over a range of plasmid quantities that yielded a plateau of transactivation. This observation is consistent with our conclusion that the transactivation plateau reflects generation of APPGV16 levels sufficient to saturate γ-secretase.

Figure 4.

APPGV16 wild-type and FAD mutant isoforms are consistently expressed and processed by α- and β-secretase. (a) APPGV16 protein levels were examined in the same lysates used for the transactivation assays in Fig. 3, with the Austrian and Florida mutants relative to wild-type. Protein levels increase to similar levels with both mutant and wild-type APPGV16. Protein levels appear to increase linearly across the range of DNA concentrations used. The lysates corresponding to 100 ng of transfected APPGV16 DNA show minimally detectable levels of APPGV16 protein. However, at this concentration of APPGV16 DNA transactivation levels begin to plateau in parallel transactivation assays (see Figs 3a and c).  (b) Similarity of protein expression levels resulting from equivalent amounts of transfected APPGV16 DNA was explored using the wt, T639I, V642G, V642I, and KM595/596NL isoforms. Protein (50 μg) was loaded into each lane and the blot was probed with rabbit anti-Gal4 antibody (Zymed). (c) Immunoprecipitation of wild-type and T639I APPGV16 protein from 10-cm plates of Cos7 cells transiently expressing either construct was performed in the presence or absence of the WPE-II-89 γ-secretase inhibitor. A band corresponding in molecular weight to CTFγ-GV16 was observed only with wild-type APPGV16 in the absence of WPE-II-89. The large band beneath the predicted CTFγ-GV16 is the light chain of the Gal4 antibody used in the immunoprecipitation. (d) The contribution of α-, β- and γ-secretase to the differential processing of wild-type and FAD mutant APPGV16 was assessed pharmacologically using inhibitors for each secretase. Cos7 cells were transfect with wild-type, T639I, V642G and V642I APPGV16 constructs for 24 h prior to the addition of the specific inhibitors for 15 h. TAPI, a widely used α-secretase inhibitor, was used at 50 μm; β-secretase inhibitor II (Calbiochem, San Diego, CA, USA) was used at 10 μm; WPE, the γ-secretase inhibitor, was used at 1 μm.

To ensure that the different isoforms of APPGV16 were expressing equivalently, several FAD mutant forms of APPGV16 were compared with wild type following transfection of a single quantity of plasmid (Fig. 4b). No significant difference in quantity of APPGV16 expressed between mutants and wild type was observed, with the possible exception of the Swedish mutation for which the protein levels appeared slightly higher. Consequently, the decreases among class II APP mutations in production of the transcriptionally active carboxy-terminal fragment of APPGV16 do not reflect differences in the amount of APPGV16 expressed, but rather appear to result from decreases in efficiency of γ-secretase-mediated processing of the mutant protein. Finally, for a single APP FAD mutant (T639I), we employed direct biochemical means to verify that differences in reporter activation reflect differences in APPGV16 cleavage. Immunoprecipitation of cell extracts revealed a band of the size expected for CTFγ-GV16, and this product was not present from cells exposed to a γ-secretase inhibitor (Fig. 4c). The quantity of CTFγ-GV16 produced from the T639I mutant was below the limits of detection. This confirms the inhibition of CTFγ production from the T639I mutation indicated by the transcriptional reporter system and generally supports the validity of the assay system.

Wild-type and FAD APPGV16 are normally proteolytically processed

In view of these surprising findings, we wished to confirm that APPGV16 behaves like native APP in the manner in which it is processed. γ-Secretase-mediated cleavage of APP requires prior cleavage by α- or β-secretase. We assessed whether cleavage of APPGV16 was subject to a similar requirement by examining the effect of inhibitors of these enzymes. For wild-type and mutant APPGV16, inhibition of either α- or β-secretase diminishes γ-secretase-mediated release of the carboxy-terminal fragment (Fig. 4d). Thus, APPGV16, like native APP, requires α- or β-secretase-mediated cleavage prior to γ-secretase-mediated cleavage. The different APPGV16 isoforms were inhibited equivalently by α- and β-secretase inhibitors. This indicates that the decreased γ-secretase-mediated transactivation by mutant forms of APPGV16 does not reflect diminished processing by the α- and β-secretases. The significant attenuation of wild-type and mutant APPGV16 cleavage caused by the γ-secretase inhibitor demonstrates the γ-secretase dependence of processing in each case. Further, the statistical difference maintained between wild-type and FAD mutant APPGV16 in the presence of α- and β-secretase inhibitors disappears in the presence of γ-secretase inhibitor.

FAD mutations in PS1 lead to decreases in APP cleavage by γ-secretase

The finding that FAD mutations of APP diminish CTFγ production led us to ask whether PS1 mutations associated with FAD have a similar effect. Wild-type APPGV16 plasmid was titrated into cells transiently over-expressing either wild-type or FAD mutant forms of PS1. All FAD mutations of PS1 tested resulted in a decrement in release of CTFγ from APPGV16 (Figs 5a–d). These results were confirmed using stable cell lines expressing the same PS1 constructs, for which even greater differences between wild-type and mutant PS1 were observed (Fig. 5e). To assess whether the effects of PS1 mutations might differ in neuronal and non-neuronal cell types, we established cultures of HT22 cells (a mouse hippocampal cell line), stably transfected with wild-type or mutant PS1. Neuronal cells expressing FAD mutant PS1 demonstrated the same reduction of APPGV16 cleavage observed in the Cos7 cell system (Fig. 5f). It is interesting to note that FAD mutations that cause the earliest onset of AD pathology (Czech et al. 2000) roughly correlate with the degree of impairment of APPGV16 cleavage in Cos7 cells stably transfected with mutant and wild-type PS1 (Fig. 5e). However, it must be noted that this correlation was not observed in the PS1 HT22 stably transformed cell lines.

Figure 5.

Comparison of cleavage of APPGV16 by wild-type and FAD mutant presenilin-1. γ-Secretase-mediated cleavage of APPGV16 was tested in both transiently and stably transfected Cos7 cells over-expressing either wild-type or FAD mutant isoforms of presenilin-1. 500 ng/well of presenilin-1 vector DNA was used in all transient transfection experiments (a–c) and assays were performed 48 h following transfection. Two-way anova analysis of the wild-type and FAD mutant PS1 transiently transfected cells was performed with the following results (L286V < wt, p < 0.08; M146L < wt, p < 0.0001; C410Y < wt, p < 0.001). (d) Similarly, the L166R mutant, previously characterized as deficient in APP-CTFγ production, was compared with wild-type using transient over-expression assays. (e) Stable presenilin-1 Cos7 cell transformants were generated using PS1 wt, A246E, L286V, M146L, and C410Y. Mixed populations of stably transfected clones were used to avoid clonal bias between the different presenilin-1 lines. The Cos7 presenilin-1 stable lines were all transfected with various quantities of APPGV16 (range 25 ng to 500 ng APPGV16) and harvested for assay 48 h post-transfection with APPGV16. Two-way anova analysis of the comparative titration shows that GV16-dependent luciferase expression generated by all the FAD mutant forms of presenilin-1 is significantly lower than that resulting from wild-type presenilin-1 (A246 < wt, p < 0.0001; L286V < wt, p < 0.01; M146L < wt, p < 0.0001; C410Y < wt, p < 0.0001). (f) HT22 cells, a mouse hippocampal neuroblastoma line, were used to generate a second set of cell lines stably expressing the PS1 constructs mentioned in (e). APPGV16 DNA (100 ng) was transiently transfected into the HT22 cells. Wild-type PS1 activity is significantly higher than FAD PS1 in all conditions (p < 0.01, student's t-test).

APP S3/ε-site mutagenesis decreases carboxy-terminal liberation

Our finding of reduced production of CTFγ in concert with FAD mutations that increase production of Aβ42 implies that the majority of CTFγ derives from an APP cleavage site that does not produce Aβ42. Recent studies have revealed that a major site of cleavage of APP by γ-secretase occurs at a position that would be predicted to produce a 49-residue Aβ-like peptide (Gu et al. 2001; Sastre et al. 2001; Yu et al. 2001; Weidemann et al. 2002). This site has been referred to as the epsilon site, or as the S3-like site by reference to the similar site of cleavage of Notch. One plausible model that could account for the invariable association of reduced CTFγ production with increased Aβ42 is that FAD mutations might inhibit cleavage at a normally preferred epsilon/S3-like site, increasing the amount of substrate available for cleavage at the site that generates Aβ42. If this were the case, one would expect that intentionally eliminating the epsilon/S3-like site would inhibit CTFγ production to approximately the same extent as FAD mutations. As noted by others, there is rough conservation of the whole tetrapeptide region immediately adjacent to the cytosolic face in APP, APLP1, APLP2 and Notch, all of which are cleaved by γ-secretase at this position (Gu et al. 2001). Consequently, we generated two mutations at this site: the first, LV645/6AA (S3m1), modified just two residues at the predominant epsilon site (relative to APP695 numbering scheme); the second (S3m3) replaced all four conserved residues between 645 and 8 (S3m3), with alanine residues. Titration experiments comparing transiently transfected wild-type and epsilon mutations in Cos-7 cells were performed. CTFγ production was markedly diminished by the epsilon site mutations (Fig. 6a). Interestingly, the effects of the two epsilon site mutants were remarkably similar, suggesting that mutation at the LV645/6 site was sufficient to repress epsilon cleavage activity. In order to assure that the difference was not imparted by differential expression levels, western blots were performed, with antibody against Gal4, to compare levels of expression of wild-type and epsilon mutant proteins. Wild-type and epsilon mutant forms of APPGV16 were found to be expressed at similar levels across the spectrum of transfected plasmid concentration (Fig. 6b). These results strongly suggest that the predominant site of cleavage responsible for CTFγ is the epsilon/S3-like site.

Figure 6.

APPGV16 S3-like/epsilon site mutants impair CTFγ release. Two distinct mutations were made in the APPGV16 construct at the S3-like/epsilon site. The first mutation, S3m1, altered the epsilon cleavage site alone, replacing them with alanines (ML647/8AA). The second mutation, S3m3, altered the epsilon site and the two n-terminal semiconserved residues (LVML645–8AAAA). Titrations of the wt, S3m1 and S3m3 were employed in transactivation assays in Cos7 cells (a). The wt activity was significantly higher than either the S3m1 or the S3m3 mutations (p < 0.0001; two-way anova). The protein expression levels were examined from individual distinct wells taken from the same experiment (b). There were no differences observed between the wt, S3m1 and S3m3 expression titrations that could account for the difference in activity observed.

S3/ε-site mutations increase Aβ42 production in HT22 and Cos7 cells

In order to assess the similarity of S3-site mutations upon amyloid production, beta-amyloid ELISA assays were performed for Aβ40 and Aβ42 in Cos7 and HT22 cells transiently transfected with wild-type or S3 mutant APPGV16. These assays were performed using the same transfection methodology employed with the transactivation assays described in the previous section. The medium from the cells was collected and pooled from triplicate sets of transfected cells for each condition for each cell type. The pooled medium was used to assay for either Aβ40 or Aβ42 using biotinylated antibodies specific for either the carboxy-terminus of Aβ40 or Aβ42. The results of these assays demonstrated that the mutations at the S3 site had little to no effect on Aβ40 production in either Cos7 or HT22 cells (Figs 7a and d). However, the S3 mutation did dramatically increase the amount of Aβ42 observed in both Cos7 and HT22 cells (Figs 7b and e). These data are plotted in terms of the ratio of Aβ42/Aβ40 (Figs 7c and f), in which over a threefold increase was noted in both S3-site mutations in both cell types tested. These data suggest that impairments in γ-secretase mediated proteolysis as a result of effects upon the S3 site can result in increased Aβ42 production. These observations are consistent with previous reports in which specific FAD mutations in PS1 lead to decreased CTFγ release and simultaneously result in elevated pathogenic amyloid production (Moehlmann et al. 2002).

Figure 7.

S3/ε-site mutations increase Aβ42 in Cos7 and HT22 cellsAβ40 and Aβ42 sandwich ELISA assays were performed on the medium derived from Cos7 and HT22 cells transiently transfected with either wt, S3m1 or s3m3 APPGV16. (a–c) Correspond to the Cos7 cells, while (d–f) relate to the HT22 cells. The top row shows Aβ40 levels for each cell type (a, d). The middle row depicts Aβ42 values (b, e) for the Cos7 and HT22 cells. The bottom row shows the values of the top two rows expressed as a ratio of Aβ42/Aβ40. The S3-site mutation increased the relative amount of Aβ42 at least threefold over wt APPGV16 within both the Cos7 and HT22 cells.


The primary finding of this work is that FAD mutations in APP and PS1 result in a decrement in the liberation of the intracellular CTFγ fragment. The comparative titrations performed with mutant and wild-type APPGV16 demonstrated that all the FAD mutations that topologically map to the transmembrane domain decrease APP cleavage by γ-secretase (Figs 3a–f) without influencing the level of expression of APPGV16 protein (Figs 4a and b). The one FAD mutation examined in this work that maps to the extracellular region of APP (Swedish), showed no decrease in liberation of CTFγ. This observation is not surprising as the Swedish mutation is believed to influence the extracellular β-secretase processing rather than the cleavage by γ-secretase (Forman et al. 1997). FAD mutations in PS1 caused decreases in release of the intracellular CTFγ fragment without exception (Figs 5a–f).

CTFγ production reached a plateau at high levels of APP, apparently because the cleavage capacity of γ-secretase became saturated. As cleavage of APPGV16 by γ-secretase requires prior cleavage by α- or β-secretase, it is formally possible that the plateau reflects the saturation of one of these two enzymes rather than γ-secretase. However, over-expression of TACE (an α-secretase) or BACE did not stimulate CTFγ production (data not shown), suggesting that it is γ-secretase that becomes saturated. Although the complex mechanism of APP cleavage by γ-secretase is unlikely to fit the simple assumptions of classic Michaelis–Menton analysis, the influence of FAD mutations on CTFγ production at apparently saturating levels of APPGV16 suggests that FAD mutations primarily diminish the catalytic efficiency of γ-secretase rather than diminishing its affinity for association with APP.

The methodology employed within this study should dynamically represent the rate of γ-secretase-mediated proteolysis of APPGV16 at steady-state conditions. The clear dependence of the output of this system upon γ-secretase activity (Figs 2a–c) shows, by the same standards as employed for biochemical assays, that it monitors γ-secretase mediated catalysis. Further, the addition of the Gal4VP16 tag to the carboxy-terminus of APP did not alter the normal proteolytic processing events – as α- or β-secretase-mediated proteolytic events are still required for activity (Fig. 4d). Consequently, the simplest conclusion is that the FAD mutations directly impair γ-secretase mediated catalysis essential for CTFγ release from the membrane.

While findings of reduced production of CTFγ might appear to conflict with the robust literature demonstrating elevated levels of β-amyloid associated with FAD mutations, multiple plausible mechanisms may contribute to this result. First, as CTFγ is generated both from the α-secretase product C83, which does not produce Aβ, and from the BACE product C99, which does produce Aβ, it is conceivable that CTFγ derives primarily from C83 and, indeed, it has been suggested that this is the case (Kametani 2004). However, our results suggest that CTFγ in our experimental system derives about equally from C83 and C99, because pharmacological inhibition by an α-secretase inhibitor or a β-secretase inhibitor each decreases CTFγ production by about 50%. A second set of possibilities derives from the fact that multiple Aβ-related peptides are generated by γ-secretase-mediated cleavage, and no study of the effect of FAD mutations of APP or PS1/2 has attempted to assay all of them. Some studies have indicated that enhanced production of Aβ42 in FAD may occur in concert with decreased production of Aβ40 (De Jonghe et al. 2001; Qi et al. 2003). As Aβ42 normally represents only a small fraction of total Aβ-related peptides, Aβ42 may increase in FAD while the sum of all Aβ-related peptides decreases, and CTFγ would decrease in proportion to the decrease of total Aβ-related peptides.

Completing the thorough book-keeping of cleavage products required to provide a clear understanding of this system has been complicated by the recent finding that the only CTFγ product that can be recovered from cells expressing APP represents cleavage at the epsilon/S3-like site – rather than from sites representing the C-terminus of Aβ40 or Aβ42. The relationship between epsilon cleavage and Aβ production is unclear, as noted by multiple groups (Weidemann et al. 2002; Cervantes et al. 2004; Hecimovic et al. 2004). While it is clear that both Aβ production and CTFγ formation are presenilin-dependent processes, the degree of coupling between γ-site cleavage and epsilon cleavage is unknown. One school of thought contends that γ-proteolysis, producing numerous species of Aβ (Wang et al. 1996), is largely independent of epsilon cleavage (Chen et al. 2002; Moehlmann et al. 2002; Schroeter et al. 2003; Hecimovic et al. 2004; Jankowsky et al. 2004). Alternatively, it has been suggested that epsilon cleavage occurs first, producing Aβ48/Aβ49, which are subsequently cleaved to produce Aβ40/Aβ42 (Sato et al. 2003; Funamoto et al. 2004).

The ambiguity as to the mechanism stems largely from the lack of evidence for what would appear to be necessary cleavage products supporting either perspective. The CTFγ cleavage products corresponding to proteolysis at the Aβ sites are C57 and C59. Yet, neither C57 nor C59 have been detected within in vitro or in vivo assays (Gu et al. 2001; Sastre et al. 2001; Yu et al. 2001; Weidemann et al. 2002). Generally, this supports an epsilon-first cleavage model. Conversely, an epsilon-first model suggests that lengthy Aβ species must exist, at least transiently. There are reports of longer Aβ species (Kuo et al. 2000; Roher et al. 2004), yet it remains unclear whether these species represent a consistent phenomenon because of the rarity of their detection.

As one of several plausible models to account for our observation of decreased CTFγ production associated with FAD mutations that increase Aβ42 production, we have suggested that FAD mutations of APP and PS1 may reduce cleavage at the epsilon site, thereby increasing the quantity of substrate available for cleavage at an alternative site that generates Aβ42. Our finding that intentionally eliminating the S3-like epsilon site decreases CTFγ production (Fig. 6) and increases Aβ42 production (Fig. 7), in a manner resembling the effect of FAD mutations, supports this model. Whichever model prevails, it is clear that Aβ42 and CTFγ production need not change equivalently in FAD. Although our work is the first demonstration of consistent decreases in CTFγ formation as a result of FAD mutations in both APP and PS1, other groups have examined the role of FAD mutations in CTFγ production. Numerous groups have observed that FAD mutations in PS1 lead to decrements in Notch processing (Baumeister et al. 1997; Song et al. 1999; Schroeter et al. 2003), which explains why FAD mutations fail to rescue sel-12 deficiency in Caenorhabditis elegans (Baumeister et al. 1997). However, reports upon the effects of PS1 FAD mutations upon CTFγ formation are contradictory. Several groups have observed general deficiencies in CTFγ formation sponsored by multiple FAD PS1 mutants (Chen et al. 2002; Schroeter et al. 2003). Others have reported less consistent results, showing impairment of only select FAD mutations (Moehlmann et al. 2002). Another study finds a 50% reduction in quantitative levels of CTFγ with PS2 mutations by quantitative sequencing, yet fail to identify this quantitative effect by western blotting following in vitro CTFγ synthesis (Sato et al. 2003). Previous work examining the effect of APP FAD mutations upon CTFγ production also are conflicting. Several studies have reported no observed differences in the level of CTFγ formation with APP FAD mutations (Bergman et al. 2003; Sato et al. 2003). One study to the contrary finds that two specific APP FAD mutations confer decreases in CTFγ formation, while others do not (Hecimovic et al. 2004). It is important to note that all reports show that Aβ42 levels increase with FAD mutations – even in the presence of demonstrable decreases in CTFγ levels.

The disagreement among these studies can probably be attributed to the various experimental systems employed, all of which are artificial in one or more aspects. Some of these studies employed reconstituted cell-free systems to assess APP processing, which cannot be assumed to faithfully reproduce all aspects of the in vivo environment, particularly as the interaction of the APP transmembrane domain with γ-secretase is likely to be influenced by the membrane and/or detergent composition (Fraering et al. 2004). Other studies examine processing from transfected constructs encoding C99 rather than APP, which may not traffic equivalently through endosomal, endoplasmic reticulum (ER) and Golgi compartments implicated in processing by γ-secretase. Several studies examine the impact of various FAD mutations on processing of the Swedish FAD mutation of APP, which is taken as the ‘normal’ control for these studies, a convention that has obvious caveats. Additionally, multiple groups have noted that a fraction of CTFγ remains membrane associated post-cleavage (Gu et al. 2001; Sastre et al. 2001; Weidemann et al. 2002). The various assay systems may differ with regard to the extent to which this pool of CTFγ is detected. Finally, apparent differences between our observations and reports of others may simply reflect our precise assay, which permits small differences to be reliably detected. In our studies, the least potent FAD mutations reduced CTFγ production by only 30–40%. Reliably detecting changes of this magnitude by western blot analysis is difficult.

It is interesting to consider whether the decreased liberation of APP-CTFγ associated with FAD mutations is of functional significance. A variety of signaling functions have been ascribed to the intracellular domain of APP, including regulation of c-jun N-terminal kinase (JNK) activity (Scheinfeld et al. 2002) and Go-mediated regulation of cAMP-dependent protein kinase (PKA) activity (Nishimoto et al. 1993). Whether γ-secretase mediated cleavage of APP might promote, or inhibit, these signaling pathways is unknown. Yet, it is tempting to speculate that APP signaling via intracellular factors would require a membrane resident fragment of APP (APPholo, C99 or C83), and hence would likely be regulated by γ-secretase-mediated proteolysis. Liberation of CTFγ from the membrane may also invoke intracellular signaling events, as a number of studies have indicated that APP cleavage by γ-secretase promotes nuclear trafficking by a trancriptionally active complex containing CTFγ. Decreased release of CTFγ associated with FAD mutations should inhibit this transcriptional regulation. Consistent with this notion, mice bearing brain-specific double PS1/PS2 knockouts have been observed to develop several cardinal pathological characteristics of AD (Saura et al. 2004), suggesting that diminished γ-secretase activity could contribute to AD pathogenesis. It will remain for future studies to assess whether perturbations in signaling functions of APP resulting from decreased CTFγ production contributes to aspects of Alzheimer's disease pathology.


The authors would like to acknowledge that the funding for this research was provided by the Alzheimer's Disease Association grant IIRG-00– 1935 and the NIH grant R01 AG211-27.