Amyloidogenic Processing but Not Amyloid Precursor Protein (APP) Intracellular C-terminal Domain Production Requires a Precisely Oriented APP Dimer Assembled by Transmembrane GXXXG Motifs*

The β-amyloid peptide (Aβ) is the major constituent of the amyloid core of senile plaques found in the brain of patients with Alzheimer disease. Aβ is produced by the sequential cleavage of the amyloid precursor protein (APP) by β- and γ-secretases. Cleavage of APP by γ-secretase also generates the APP intracellular C-terminal domain (AICD) peptide, which might be involved in regulation of gene transcription. APP contains three Gly-XXX-Gly (GXXXG) motifs in its juxtamembrane and transmembrane (TM) regions. Such motifs are known to promote dimerization via close apposition of TM sequences. We demonstrate that pairwise replacement of glycines by leucines or isoleucines, but not alanines, in a GXXXG motif led to a drastic reduction of Aβ40 and Aβ42 secretion. β-Cleavage of mutant APP was not inhibited, and reduction of Aβ secretion resulted from inhibition of γ-cleavage. It was anticipated that decreased γ-cleavage of mutant APP would result from inhibition of its dimerization. Surprisingly, mutations of the GXXXG motif actually enhanced dimerization of the APP C-terminal fragments, possibly via a different TM α-helical interface. Increased dimerization of the TM APP C-terminal domain did not affect AICD production.

The physical interaction between APP and the secretases or other partners is crucial for its processing, and yet it is very poorly understood. APP contains several GXXXG motifs at the junction between the juxtamembrane and transmembrane (TM) regions (19 -22). GXXXG motifs are known from the sequence of the glycophorin A (GpA) protein to mediate sequence-specific dimerization and very close apposition of TM helices (23). In glycophorin A, the sequence LIXXGVXXGVXXT mediates tight dimerization between TM helices (24) by direct glycine-glycine contacts (25). It has later been recognized that GXXXG motifs can mediate more generic oligomerization of TM domains (26). More importantly, it has been shown that glycine is compatible with ␣-helical secondary structure in lipid bilayers and that, because of its small size, this residue allows the close association of interacting helices (27,28). GXXXG motifs have been shown to play a role in the assembly, trafficking, and activity of several proteins of the ␥-secretase complex (29,30).
The presence of three GXXXG motifs in APP suggests that the glycine face of the APP TM helix may be involved in interactions with other proteins or with itself and offers a molecular basis for APP homo-and hetero-oligomerization. Strikingly, one genetic mutation that leads to early onset Alzheimer disease, the Flemish mutation, is represented by the APP A617G mutation, which creates a fourth in-register GXXXG motif preceding the TM domain of APP (31). Moreover, we have reported that the GXXXG motifs play a major role in fibrillization of A␤40 and A␤42 (21). The mechanisms by which homoor heterodimerization of APP (32) act on its processing are far from being understood.
Here we show that APP processing via the amyloidogenic pathway to both A␤40 and A␤42 depends on the presence of a small residue, either glycine or alanine, at the position of the GXXXG motifs. Pairwise replacement of glycine by leucine within these motifs in human APP695 leads to significantly less A␤ production. One APP mutant in particular, where glycines of the middle GXXXG motif (Gly-625 and Gly-629) were mutated to leucine, exhibited altered amyloidogenic processing with a drastic reduction of A␤ formation. Similar results were observed when glycines (Gly-625 and Gly-629) were mutated to isoleucines. The mutation of GXXXG motifs to LXXXL did not affect ␤-cleavage or the APP-BACE1 interaction but decreased the ␥-cleavage of APP without impairing the APP-PS1 interaction. Strikingly, the ␥-secretase-mediated release of AICD was not altered by the GXXXG to LXXXL mutation. Unexpectedly, the mutation of the middle GXXXG motif ( 625 GXXXG 629 ) to LXXXL, rather than weakening dimerization, enhanced the formation of homodimers of the APP C-terminal fragments. Because the leucine substitutions are not compatible with the close helix packing allowed by the GXXXG motifs, they must alter the helix orientation within the homodimer. Taken together our results show that the GXXXG motifs in the TM domain of APP are required for A␤, but not for AICD production, and that dimerization per se does not suffice to promote the amyloidogenic processing of APP. Both orientation and dimerization of the APP TM domain differently affect the cleavage of APP by ␥-secretase, a process critically involved in Alzheimer disease.

EXPERIMENTAL PROCEDURES
Antibodies and Reagents-The polyclonal anti-human A␤ used for immunoprecipitation was described previously (33). The human specific WO-2 and 6E10 antibodies were from The Genetics Co. (Schlieren, Switzerland) and Signet Laboratories (Dedham, MA), respectively. The polyclonal antibody directed against the APP C terminus (34) was kindly provided by N. Sergeant (INSERM U422, Lille, France). The anti-actin polyclonal antibody was from Sigma The polyclonal anti BACE1 and monoclonal MAB1563 antibodies directed against human PS1 were from Oncogene Research (Cambridge, MA) and Chemicon (Temecula, CA), respectively. The monoclonal anti-HA and polyclonal anti-Myc were purchased from Santa Cruz (Santa Cruz, CA) and Roche Applied Science, respectively. Secondary antibodies were from Amersham Biosciences. All cell culture media and antibiotics were from Invitrogen.
Plasmids-A cDNA encoding the full-length sequence of human APP695 was inserted into the SmaI/SalI restriction sites of pSVK3 vector (pSVK3-APP695). All the mutated APP695 and C99 HA-and Myc-tagged C99 constructs were derived from the pSVK3-APP695 parental vector. Mutated human APP695Gal4 constructs were derived from the pMst-APP (APPGal4) parental vector (15). cDNAs coding for the six double glycine mutations in the APP juxtamembrane and transmembrane domains were generated by overlap extension PCR with synthetic oligonucleotides (see the supplemental data). All constructs were verified by full sequencing.
Cell Cultures and Transfection-The culture of CHO cells expressing human APP and/or human PS1 has been previously described (35). For transient transfection, cells were seeded at the density of 3.10 5 cells/cm 2 24 h before transfection (2 g of pSVK3APP-695/well) with Lipofectamine TM according to the manufacturer's instruction (Invitrogen). Stable cell lines expressing human APP mutants were established by co-transfecting pSVK3-APP expression vector and PSV2Neo vector at a 10:1 ratio (2 g DNA/well). Whole populations expressing human APP695 were selected in the presence of 500 g/ml G418 and further subcloned.
APPGal4 Transactivation Assays-CHO cells were transfected with the following plasmids: pMst-APPGal4 expression vectors (0.4 g/well), pG5E1B-luc (0.4 g/well), pCMV5-Fe65 (0.2 g/well), phRG-TK (0.02 g/well). Cells were harvested 48 h post-transfection in 0.1 ml/well reporter lysis buffer, and the firefly and renilla luciferase activities were measured with the Dual Glo TM luciferase assay system (Promega). The firefly luciferase activity was standardized by the Renilla luciferase activity to control for transfection efficiency.
Immunoprecipitation and Co-immunoprecipitation-Immunoprecipitation of A␤ from the culture media with a polyclonal anti-A␤ antibody (30 l/ml) has been previously described (36). Cells were grown in 6-well dishes washed in ice-cold phosphate-buffered saline, solubilized in immunoprecipitation buffer (1 ml/well of 25 mM Tris, pH 7.6, 0.5% Triton X-100, 0.5% Nonidet P-40 containing a protease inhibitor mix). Cellular extracts were cleared by centrifugation (20,000 ϫ g, 5 min, 4°C). Immunoprecipitation of APP was carried out on the supernatants with 2.5 g/ml anti-human APP (WO-2) antibody. Co-immunoprecipitations were carried out in the same conditions with 2.5 l/ml of anti-BACE1, 2 l/ml anti-human PS1 (MAB1563), 2.5 g/ml anti-human APP (WO-2), 5 l/ml anti-HA, or 5 l/ml anti-Myc antibodies. Immunoprecipitates were analyzed by Western blotting as described below.
Surface Biotinylation of APP-Cell surface biotinylation was performed as previously described (38). CHO cells expressing human APP (ϳ5.10 6 cells) were washed with phosphate-buffered saline containing 2 mM Ca 2ϩ and 1.2 mM Mg 2ϩ and incubated with 0.8 ml of sulfo-NHS-biotin (Pierce) at 1.5 mg/ml in phosphate-buffered saline for 30 min at 4°C with mild shaking. Cells were then washed twice with cold phosphate-buffered saline containing 100 mM glycine, incubated with the same solution for 45 min at 4°C to quench the unbound biotin reagent, and further solubilized in lysis buffer (25 mM Tris-HCl, pH 6.8, 0.5% (v/v) Triton X-100, 0.5% (v/v) Nonidet P-40) supplemented with protease inhibitors (1 g/ml pepstatin, 10 g/ml leupeptin, 1 mM phenylmethylsulfonyl fluoride) for 1 h at 4°C with vigorous shaking. After centrifugation at 15,800 g at 4°C for 20 min, 150 l of supernatant fluid were incubated with an equal volume of streptavidin bead suspension (Pierce) for 1 h at room temperature. After centrifugation (15,800 ϫ g, 15 min, 4°C), supernatant fluids were collected for analysis of the nonbiotinylated intracellular fraction. Biotinylated cell surface proteins contained in the pellet were washed 4 times with 0.8 ml of lysis buffer and resuspended in 50 l of Western blotting sample buffer containing 50 mM dithiothreitol. Samples were analyzed by Western blotting.
Analytical Subcellular Fractionation-CHO cells were recovered in 0.25 M sucrose containing 1 mM EDTA, 3 mM imidazole buffered at pH 7.4, and complete protease inhibitors (Roche Applied Science). The cellular suspension was homogenized in a tight Dounce homogenizer. A low speed nuclear fraction was pelleted at 10 4 ϫ g for 10 min and washed 3 times by resuspension and sedimentation. Pooled postnuclear supernatants were further sedimented at 10 5 ϫ g for 60 min in a Ti50 rotor (Beckman Instruments). A high speed particular pellet was resuspended in 0.3 ml of homogenization buffer, mixed with 2.3 M sucrose to reach 1.28 g/ml in density, and layered at the bottom of a linear sucrose gradient (from 1.10 to 1.24 g/ml in density). After floatation by centrifugation at 2 ϫ 10 5 ϫ g for 22 h in a Sw40 rotor (Beckman), 12 fractions were collected and analyzed for protein content. Western blotting and quantifications were performed as described below. Results were represented as normalized histograms (39).
ELISA-Two-days after plating or transfection, cells were reincubated for 8 h in fresh culture medium. Culture media were recovered and incubated with protease inhibitors (1 g/ml pepstatin, 10 g/ml leupeptin, 1 mM phenylmethylsulfonyl fluoride). Samples were cleared by centrifugation (16,000 ϫ g, 5 min, 4°C). Immunoprecipitation buffer (1:10 v/v) was added to supernatants to a final concentration of 25 mM Tris, pH 7.6, 0.5% Triton X-100, 0.5% Nonidet P-40. A␤40 and A␤42 were quantified using fluorescent ELISA assays from BIOSOURCE (Camarillo, CA). Soluble ␣APP and soluble ␤APP were quantified in multiplex from 25 l of cellular medium using the electrochemiluminescence detection method of Meso Scale Discovery (Gaithersburg, MD).
Computational Searches-Low energy conformations of helix dimers were searched by rotating each helix through rotation angles 1 and 2 from 0°to 360°with sampling step sizes of 25-45°and with an interhelical separation of 9.0, 9.5, 10.0, and 10.5 Å. The 1 and 2 angles are equivalent to the ␣ and ␤ angles originally described by Adams et al. (40). Molecular dynamics simulations were run with simulated annealing at each rotational orientation of the dimer using the program X-PLOR along with the united atom topology and parameters sets, TOPH19 and PARAM19, respectively. Calculations were carried out for helices with both left-and right-handed geometries with initial crossing angles of 25°. Four different runs were carried out for each starting geometry using torsional angle dynamics. The rotation and crossing angles were allowed to vary during the simulations. To maintain an ␣-helical conformation, distance restraints were applied between O i and N iϩ4 atoms along the backbone. For the APP homodimers, computational searches were carried out on residues 620 -648. Position 624 (lysine) was neutral for the molecular dynamics simulations and charged for the final energy minimization.

GXXXG to LXXXL Mutation
Reduces the Amyloidogenic Processing of APP-To investigate the role of the GXXXG motifs in the processing of human APP695, we simultaneously replaced by site-directed mutagenesis the two Gly (G) residues of each GXXXG motif to either Ala (A) or Leu (L). The amino acid substitutions at positions 621, 625, 629, and 633 that correspond to the positions 25, 29, 33, and 37 of A␤ are indicated for each mutant (Fig. 1A). The mutated positions are located at the interface between the extracellular and TM regions (621/625, mutants 1 and 3) or in the predicted TM domain (625/629 and 629/633, mutants 2 and 3 and mutants 5 and 6). None of the mutations are directly located at a known APP cleavage.
CHO cells were transfected with the plasmids encoding APP695 and the different APP695 mutants (1 to 6, see Fig. 1). Cells were harvested 48 h after transfection, and the expression of full-length human APP was analyzed in cell lysates by Western blotting with the human specific WO-2 antibody. We detected the expression of the different mutants at the expected size (Fig. 1B). The release of soluble APP (s␣APP), which is generated by the ␣-secretase activity, was monitored in the corresponding extracellular culture medium. We detected production of the s␣APP by Western blotting of culture medium from transfected CHO cells with the same WO-2 antibody. All of the mutants produced comparable levels of s␣APP (Fig. 1B).
When A␤ production was analyzed by immunoprecipitation with polyclonal anti-human A␤ antibodies from the culture medium of transfected cells, we detected a significant decrease in A␤ production in mutants 4, 5, and 6, whereas when GXXXG motifs were replaced by alanine (mutants 1-3), there was no The amino acid substitution (Gly to Ala or Gly to Leu) generated for each mutant appears in bold underlined. The cleavage sites of ␣ (␣)-, ␤ (␤ and ␤Ј)-, and ␥ (␥ and ⑀)-secretase activities are indicated by arrows. The epitopes of the human-specific WO-2 antibody are also shown along with the C-terminal position recognized by the A␤40 and A␤42-specific antibodies for ELISA. B, the expression of cellular APP695 or APP mutants was analyzed 48 h after transfection by Western blotting revealed by the WO-2 antibody. The presence of APP is indicated by arrows. Actin was used as a protein loading control. Forty hours after transfection, CHO cells were conditioned in fresh culture medium for 8 h. The accumulation of s␣APP was analyzed by Western blotting revealed by the WO-2 antibody. A␤ was immunoprecipitated from the same culture medium and analyzed by Western blotting revealed by the WO-2 antibody. C, the A␤ to s␣APP ratio was calculated and represented as percentage of A␤/s␣APP production in non-mutated controls (APP695). Values are the means Ϯ S.E., n ϭ 4; *, p Ͻ 0.05; **, p Ͻ 0.01; ***, p Ͻ 0.001, compared with control. difference in A␤ production as compared with wild type APP (Fig. 1B). This decrease, especially for APP mut5, is also evident when the levels of extracellular A␤ were normalized to the levels of s␣APP and represented as the percentage of the A␤/s␣APP ratio measured in the culture medium of cells expressing wild type APP695 (Fig. 1C). Because the different APP mutations are located in the A␤ sequence, we verified that the WO-2 antibody was still able to recognize mutated A␤. In addition, the decrease in A␤ production was confirmed for APP mut5 with another antibody (6E10) directed against the N terminus of human A␤, and this decrease did not result from the accumulation of A␤ in intracellular compartments or in cell membranes (not shown). The very strong decrease observed, particularly in mut5 (G625L/G629L), is consistent with the idea that the balance between amyloidogenic and non-amyloidogenic processing of APP was affected when glycines from GXXXG motifs were replaced by leucines but not by alanines.

G625L/G629L Mutation Impairs Neither the Cellular Localization
Nor the ␤-Cleavage of APP-The consequences of the glycine to leucine mutations on APP amyloidogenic processing were analyzed in detail in CHO cell lines stably expressing APP695 or APP mut5 (G625L/G629L). A plausible explanation for these results was that the Gly to Leu mutation disrupted the interaction of APP with binding partners, leading to mislocalization of the protein. Consistently APP mutant5 exhibited an increase in the mature forms of full-length APP (Fig. 1B). This may indicate a more efficient processing through the Golgi apparatus or a failure in endocytosis (41). Analytical subcellular fractionation and cell surface biotinylation experiments were carried out to study the subcellular distribution of APP and APP mut5. Using analytical subcellular fractionation, we did not observe any significant difference in the cellular distribution of APP and APP mut5 (Fig. 2, A  and B). As shown in Fig. 2C, similar levels of APP and APP mut5 were detected at the cell surface after cell surface biotinylation. This indicated that the decrease in A␤ did not result from a cellular redistribution of APP induced by the glycine to leucine mutation.
We investigated a possible role of the GXXXG motifs in the ␤-cleavage of APP. The ␤-cleavage of APP is performed by the BACE1 protein (5). Because the TM domain of BACE1 contains two AXXXA motifs that may also promote the interaction of TM helices (42), we analyzed the effects of the G625L/G629L mutation on APP-BACE1 interaction. The association of APP and BACE1 in the cell membrane was analyzed by non-denaturating gel electrophoresis (Blue Native PAGE). The results shown in Fig. 3B strongly suggested that the formation of APP-BACE1 complexes was not impaired by the G625L/G629L mutation. This was confirmed by co-immunoprecipitation experiments (Fig.  3C). We further showed that soluble ␤APP, produced by the cleavage of APP by ␤-secretase, was not affected in APP mut5 (Fig. 3D). The quantification of s␣APP in the same sample allowed us to conclude that the s␤APP/s␣APP ratio was the same for APP695 and APP mut5. In addition, the cellular APP metabolite produced by ␤-secretase (␤CTF) did not decrease but, indeed, accumulated in APP mut5 (Fig. 3E). This indicated that the G625L/G629L mutation did not impair the ␤-cleavage of APP but seems to have blocked ␥-cleavage. Impact of G625L/G629L Mutation on ␥-Cleavage and A␤ Production-To investigate the role of the GXXXG motifs in the ␥-cleavage of APP, we measured A␤ production from cells expressing either APP or C99 bearing or not the G625L/G629L mutation (Fig. 4, C and E). The C99 protein corresponds to the ␤-cleaved C-terminal fragment of APP fused to the signal peptide sequence (Fig. 4A), and thus, A␤ is released from C99 by a single cleavage performed by ␥-secretase (43). We specifically measured A␤40 and A␤42 production by ELISA in CHO cell supernatants. Both A␤40 and A␤42 were present in the culture medium of cells expressing APP695 or C99 (Fig. 4, C and E). Importantly, A␤40 production was abolished, and the level of A␤42 was also strongly decreased in cells expressing APP mut5 or C99 mut5. The decrease in A␤40 and A␤42 in cells expressing APP mut5 was not compensated by an increase in shorter A␤ isoforms (supplemental Fig. 1). This was consistent with the overall decrease in A␤ measured by immunoprecipitation (Fig.  1C). Similar levels of C99 or C99 mut5 were detected in the cells. However, in cells expressing C99 mut5, a band corresponding to a possible C99 dimer was detected, whereas such a band was present only at very low levels in C99 cells (Fig. 4D).
This impairment of APP ␥-cleavage detected for both APP mut5 and C99 mut5 led us to investigate the effect of the GXXXG mutation on APP-PS1 interaction in co-immunoprecipitation experiments. PS1 is thought to harbor the catalytic core of the ␥-secretase complex (9), and it has been shown to interact with APP C-terminal fragments (44). Our data clearly show that the G625L/G629L mutation did not impair the inter-action between APP and full-length PS1 or PS1 fragments that were reported to represent the catalytically active form of PS1 (Fig. 5).
Taken together, the complete inhibition of A␤40 along with the strong decrease in A␤42 production observed for APP mut5 and C99 mut5 led us to conclude that G625L/G629L mutation inhibits specifically the ␥-cleavage that produces A␤40 and A␤42, but this was not due to a failure of APP-PS1 interaction.
G625L/G629L Mutation Does Not Change AICD Release-The cleavage of APP C-terminal fragments by the ␥-secretase complex also releases AICD, the APP intracellular C-terminal domain (18). Because the G625L/G629L mutation decreases A␤ production, we studied its impact on AICD release. This was first analyzed by using APPGal4 fusion proteins. This system allows measurement of the release of AICD in a quantitative and sensitive manner by a Gal4 transactivation assay (15,16). APP Gal4 and the corresponding G625L/G629L mutant (APPGal4 mut5) were transiently expressed in CHO cells (Fig.  6, A and B). We measured A␤40 production in the culture medium of APPGal4-expressing cells by ELISA. We observed a strong decrease in A␤ production in extracellular media from cells expressing APPGal4 mut5 (Fig. 6C). This was in line with the results obtained from cells expressing APP695. Very interestingly, transactivation assays showed that the mutation of the GXXXG motif had no detectable effect on AICD release (Fig. 6D).
The intracellular AICD levels were next monitored by Western blotting in cells expressing C99 and C99 mut5. AICD is a very labile peptide that is rapidly degraded by intracellular metalloproteases and particularly by insulin-degrading enzyme (45). The AICD levels were, therefore, measured in the presence of 100 M o-phenanthroline, a metalloprotease inhibitor (46). Results shown in Fig. 6E indicate that the levels of AICD are similar in cells expressing C99 and C99 mut5. We carried out the same experiments in cells expressing APP695 and APP mutant 5 and showed that AICD levels were again similar in cells expressing APP and APP mut5 (Fig. 6F).
In summary, we have found that the glycine to leucine mutation decreased A␤ formation (Figs. 1C, 4C, 4E, and 6C), whereas it had no effect on AICD release (Fig. 6, D, E, and F)    AICD and A␤ are believed to be produced by ␥-secretase activity. In addition, the interaction of APP with PS1 was not disrupted by the G625L/G629L mutation (Fig. 5). This led us to search for a possible role of the GXXXG motifs in homodimerization of APP C-terminal fragments, which could account for the observed effects on A␤ and AICD formation.
G625L/G629L Enhances the Formation of C99 Homodimers-One way for the GXXXG motifs to modulate A␤ production could be to promote close apposition of TM domains in APP or, especially, in C99. We asked whether amyloidogenic processing simply requires dimerization or whether the GXXXG motifs impose a specific interface of TM dimerization, which then promotes amyloidogenic processing. Because our G625L/ G629L mutation inhibited amyloidogenic processing, we tested its effect on C99 dimerization. Unexpectedly, results shown in Fig. 4D indicate that the G625L/G629L mutation is likely to trigger the formation of SDS-resistant C99 dimers.
We employed HA-and Myc-tagged C99 constructs to be able to test by co-immunoprecipitation experiments whether C99 or mutants of C99 do oligomerize. These proteins contain the signal peptide (residues 1-19) of APP fused to the tag (HA or Myc) followed by a linker corresponding to the 4 amino acids preceding the ␤-cleavage site and the C99 sequence (Fig. 7A).
When expressed in CHO cells, these constructs produce A␤40 and A␤42 (Fig. 7B), indicating that they are correctly processed by ␤and ␥-secretase. Here again, for the tagged versions of C99, the G625L/ G629L mutation completely blocked A␤40 production and very strongly decreased A␤42 (Fig. 7B). Western blotting on cell lysates revealed that the mutated HA-and Myc-C99 formed high levels of SDS-resistant dimers when compared with the corresponding nonmutated proteins (Fig. 7C), which is in accordance with our results with non-tagged C99 and C99 mut5. These dimers where recognized specifically by both anti-tag (HA and Myc) and WO-2 antibodies.
Cell lysates were immunoprecipitated with the anti-Myc antibody and were subsequently analyzed by Western blotting with the anti-HA antibody. C99 proteins were detected by Western blotting with anti-HA antibodies only in cell lysates co-expressing HA-C99 and Myc-C99 constructs and not in those expressing only HA-C99. This demonstrated the specificity of the antibody for the tag (Myc) used for immunoprecipitation. Bands were detected in Western blots at the expected molecular weight of tagged-C99 monomers. Their intensity was similar for C99 and C99 mut5 (containing the G625L/G629L mutation). Importantly, very high levels of C99 dimers were immunoprecipitated in cells co-expressing HA-and Myc-C99 mut5, whereas these dimers were present at low levels in cells expressing the nonmutated HA-and Myc-C99 proteins (Fig. 7D). The formation of C99 dimers did not result from the post-lysate aggregation of C99 proteins. When cell lysates separately expressing HA-and Myc-C99 were mixed and immunoprecipitated (Post-lysate mix lanes), no C99 dimer was detected (Fig. 7D). Altogether, our data show that the Gly-to-Leu mutation at the positions 625 and 629 promoted the formation of SDS-resistant C99 dimers and, thus, promoted dimerization of the C terminus domain of APP.
We further analyzed the dimerization state in relation to A␤ production for C99 proteins that carried a single Gly to Leu mutation in the 625 GXXXG 629 motif or for Gly to Ile mutants that were recently described (22). The different mutants are depicted in Fig. 8A. The Gly-to-Leu mutation at the position 625 had weak effects on the formation of SDS-resistant C99 dimers, whereas the G629L mutation lead to formation of similar levels of SDS-resistant C99 dimers as those observed for the G625L/G629L mutants (Fig. 8B). Strikingly, the decrease in A␤ secretion (A␤40 and A␤42) was inversely proportional to the levels of C99-SDS resistant dimers measured for these mutants. In addition, identical results were obtained for glycine to isoleucine mutants, among which was the very recently described G 629 I mutant (22) (Fig. 8, B and C). There was no significant difference observed in AICD levels produced by any of these mutants (Fig. 8B). These results indicate that the GXXXG motif may impose a specific dimer interface, which is required for amyloidogenic processing. Mutations that rotate this interface, even if they would enhance dimerization, would impair amyloidogenic processing and A␤ formation. We have used computational searches of low energy dimer structures to assess the likely dimerization interface imposed by the GXXXG motifs and the possible influence of mutation of the glycine residues on TM helix dimerization and helix orientation within a dimer. For the wild type APP sequence (residues 620 -648), we typically observed a low energy structure (Fig. 9A) where helix dimerization was mediated by the GXXXG motifs involving glycines 621, 625, 629, and 633. In contrast, for the mut5 sequence (G625L/G629L) and the G629I sequence, the presence of leucines at positions 625 and 629 or isoleucine at position 629 led to a rotation of the helices and a low energy dimer structure with the small residues Gly-634 and Ala-638 in the dimer interface (Figs. 9, B and C, respectively). Fig. 9D shows a plot of the interaction energies that stabilize the helix dimers in Figs. 9, A-C. The close proximity of the helices in the wild type APP dimer allows Ser-622 to form a stabilizing interhelical hydrogen bond. In the mut5 dimer, the most stabilizing interaction is an interhelical hydrogen bond formed by Asn-623. Thus, the dimeric interfaces are very likely to differ between the wild type APP TM domain the APP mut5 TM domain.

DISCUSSION
Our key finding is that the TM GXXXG motifs are required for A␤40 and A␤42 production from APP. The GXXXG motifs are not required for the generation of AICD, although both A␤ peptides and AICD are normally produced by a ␥-secretase-mediated process. Our APP mut5, where the middle GXXXG motif is mutated to LXXXL, is indistinguishable from wild type APP with respect to AICD generation but exhibits severely impaired generation of the A␤40 and A␤42 peptides. Because the leucine substitutions are not compatible with the close helix packing observed by the GXXXG motifs, our results demonstrate that helix dimerization and orientation differently affect processing of APP leading to A␤ or AICD production.
The GXXXG motif and "GXXXG-like" motifs (in which one or both glycine residues are substituted by other small residues, such as alanine or serine) were found to be essential for mediating close interactions between TM ␣-helices due to the very small size of glycine (47). Both alanines and glycines are small residues frequently found in the contact interfaces between TM helices (20), and AXXXA motifs have been reported to also promote interaction of TM ␣-helices (42). The APP TM domain and extracellular juxtamembrane region, which are thought to adopt ␣-helical conformations, contain three in-register GXXXG motifs. It is unclear whether these motifs contribute to the dimerization of APP, which contains a bulky extracellular domain that has been reported to mediate APP-APP interactions (48). However, after ectodomain shedding (by ␤-cleavage for instance), when the extracellular portion is small, the TM domains are likely to play a major role in dimerization of ␤CTF (C99). To investigate the role of GXXXG motifs in APP processing, we mutated the glycine residues of each GXXXG motif either to alanine or leucine and found that the middle motif is crucial for the amyloidogenic processing of APP. Impairment of the amyloidogenic processing of APP in the GXXXG to LXXXL mutants might, therefore, be due to the disruption of close TM helical packing when glycines were substituted by bulky leucine residues. Indeed, amyloidogenic processing was normal when glycines were mutated to alanines.
Three lines of evidence indicated that the GXXXG mutation might affect the ␥-cleavage of APP; (i) ␤-cleavage and APP-BACE1 interaction were not affected by GXXXG to LXXXL mutation (Fig. 3, B and C); (ii) higher levels of ␤CTFs were detected in APP mut5 in the context of unchanged ␤-cleavage (Fig. 3E); (iii) a similar decrease in A␤ was observed for APP mut5 and C99 mut5, in which A␤ is produced by the single ␥-cleavage (Fig. 4, C and E). Thus, the replacement of glycine by leucine in the middle GXXXG motif (G625L/G629L) of APP or C99 leads to impaired ␥-cleavage, which results in the inhibition of A␤40 and A␤42 production.
Because mutation of 625 GXXXG 629 to LXXXL resulted in decreased ␥-cleavage, it was conceivable that this mutation might impair the interaction of APP with the ␥-secretase. The multicomponent ␥-secretase complex contains at least four membrane proteins. The interaction of APP with PS1, the catalytic core of ␥-secretase complex, was not altered as far as association is concerned (Fig. 5). It is not clear whether the relative positioning of APP toward PS1 was modified by the leucine mutations, and this should be further investigated.
Very importantly, we also showed that in the context of decreased A␤ production, the intracellular levels of AICD were not affected. A␤ and AICD production require a functional presenilindependent ␥-secretase complex (18,49,50). However, A␤ is mainly produced by cleavage at the positions 40 and 42, whereas AICD results from the cleavage at position 49, distal to the ␥-cleavage site (51). Thus, the cleavage at position 40 or 42 is now referred as ␥-cleavage, whereas the cleavage at position 49 releasing AICD is referred to as ⑀-cleavage. The interrelation between ␥and ⑀-cleavage is a matter of debate. A previous study reported an equimolar production of A␤ and AICD by ␥-secretase, suggesting a direct relationship between ␥and ⑀-cleavage (52). On the other hand, loss of a component (TMP21) associated to the ␥-secretase complex results in an increase in A␤, dependent on ␥-cleavage of APP, without affecting the level of AICD, dependent on ⑀-cleavage (53). Moreover, it has been reported that PS1 mutations increase the production of A␤42 but inhibit cleavage at the ⑀ position (54). Although our data do not provide the experimental power to establish a mechanistic explanation for the differential processing of APP at the ␥and ⑀cleavages, they clearly establish that the G625L/G629L mutation of APP results in a decrease in total A␤ without affecting the level of AICD. These results are completely in line with a very recent study showing that the extracellular/luminal juxtamembrane region of APP is an important regulatory domain that differentially regulates ␥and ⑀-cleavage (55). This point should be further investigated to understand how the GXXXG motifs contribute to proper positioning of APP for ␥-cleavage without influencing ⑀-cleavage.
APP was reported to form homodimers/homo-oligomers (32,48). The isolated TM sequence of APP and other ␥-secretase substrates like Notch were shown to self-associate and form dimers in TOXCAT assays (56). A very recent study reported that the GXXXG motifs of APP are required for A␤42 production and that the same motif supports homodimerization of the APP TM domain in isolation in TOXCAT assays in bacteria (22). Mutation of Gly-629 to isoleucine impaired both dimerization in TOXCAT assays of a short sequence taken from the APP TM domain and the production of A␤42. In contrast, our results demonstrate that the G629I mutant promotes FIGURE 8. Gly to Leu and Gly to Ile mutations display similar effects on A␤ production, AICD release, and C99 oligomers formation. A, schematic representation of the TM and juxtamembrane domains of the C99 Gly to Leu or Gly to Ile mutants. B, the expression of C99 and C99 mutants was measured by Western blotting in cell lysates revealed by the WO-2 antibody (top). The presence of C99 is depicted by an arrow, and the asterisk indicates the presence of C99 SDS-resistant dimers. The presence of AICD and ␣CTFs was measured in the same cell lysates by Western blotting, revealed with the C-ter antibody (bottom). C, A␤1-40 and A␤1-42 production was monitored by ELISA in the culture medium of cells expressing the indicated Gly to Leu or Gly to Ile C99 mutants. Results are given as A␤ levels in pg/ml. Values are the means Ϯ S.E., n ϭ 4; ***, p Ͻ 0.001, compared with control. nd, non-detectable.
homodimerization of C99 (Fig. 8B) and provides the first experimental evidence that C99 homodimers are present in transfected cells. These observations do not fit to a model in which GXXXG mutations weaken the dimerization of TM helices, thus, facilitating the access of ␥-secretase to the cleavage sites that are buried in the membrane (22). A very recent study might explain why our results about dimerization of APP TM domain are not in line with those previously reported (22). We analyzed homodimerization of the ␤CTF of APP (C99) by co-immunoprecipitation, whereas dimerization of APP TM domain was previously studied in TOXCAT assays (22,56). Importantly, in the TOXCAT assays the dimerization of only the APP 29 -42 TM sequence (A␤ numbering) was analyzed (22), and this sequence lacks the first juxtamembrane GXXXG motif. As shown in Fig. 9, this motif contains a serine (serine 622, APP695 numbering) that forms the strongest interhelical contact in the calculated APP homodimer and is, thus, likely to be a major determinant of APP dimerization. More to the point, the first APP GXXXG motif was shown to be critically involved in A␤, but not in AICD production (55), with the single mutation of serine 622 to leucine drastically reducing A␤ production without affecting AICD release (55). These results agree with our observations and establish that the dimerization of the APP TM domain needs to be investigated in the context of the surrounding juxtamembrane region.
This raises the hypothesis that, although GXXXG motifs are required for amyloidogenic processing, enhancing dimerization of APP or C99 per se does not necessarily lead to an increase in the production of A␤ and, particularly, A␤42 (22,32), since we show that increased dimerization of C99 is associated with a very strong decrease in A␤40 and A␤42. Dimerization that occurs after ectodomain shedding has also been shown to decrease presenilindependent intramembrane cleavage (58). Molecular dynamics simulations of the TM domain of APP or of the G625L/G629L and G629I mutants show that a different dimer interface can be adopted by the wild type APP TM helix versus G625L/ G629L and G629I TM helices. All the Gly residues of the GXXXG motifs are predicted to be in the interface of the wild type TM dimer, whereas introduction of Leu or Ile residues will cause a rotation and the placement of other small residues in the interface (i.e. Gly-634 and Ala-638). This conformation, which allows for strong and SDS-resistant dimerization, has no effect on ⑀-cleavage but deeply affects A␤ production. This prediction provides a new general model to explain our observations and results reported by other groups (22).
Taken together, these data lead to a comprehensive model for the role of the GXXXG motifs in APP on the pathogenesis of Alzheimer disease. We propose that the sequence represented The glycines in the interface of wild type APP dimer allow Ser-622 to form a strong interhelical hydrogen bond. In the GG/LL and G/I mutants, the 634 GXXXA 638 sequence allows close approach of the helices. In these mutants, interhelical hydrogen bonding of Asn-623 provides the most stabilizing interaction.
by A␤, when present in the context of full-length APP, may adopt an ␣-helical structure. Once BACE has cleaved APP at the ␤ position, the ␤CTF will likely assume a dimer conformation, with the GXXXG motifs in the interface, as these motifs are classical mediators of TM dimerization (27). The ␤CTF dimers will then bind to the presenilin complex(es) and will be processed to form A␤ peptides and AICD. For these processing events, two distinct models have been proposed. In the first progressive cleavage occurs from the ⑀ to the ␥ sites (59), with expected equimolar production of A␤ and AICD, as previously described (52). In the second, the ␥and ⑀-secretase activities may be associated with different presenilin complexes, and PS1dependent unraveling of the ␣-helical TM sequence occurs around positions 40 and 42 to generate the ␥-cleavage. In this case one can imagine situations where A␤ production can increase or decrease relative to the level of AICD. Once A␤ is generated, the GXXXG motifs would promote a conformational change from ␣-helix to ␤-strand, as we previously reported (21). Amyloid fibrils associated with Alzheimer disease and a wide range of other neurodegenerative diseases have a cross-␤-sheet structure, where main chain hydrogen-bonding occurs between ␤-strands in the direction of the fibril axis. We have shown that the packing of Met-35 against Gly-33 in the C terminus of A␤40 and against Gly-37 in the C terminus of A␤42 (21) leads to the formation of strongly neurotoxic amyloid fibrils. Because a small fraction of these A␤ peptides may never leave the membrane lipid bilayer, they may also bind other GXXXG or AXXXA transmembrane proteins like APP or the ␥-secretase complex as has been suggested (19) and, thus, affect their function.
In summary, our present and previous data (21) indicate that the APP TM GXXXG motifs mediate close interactions between TM ␣-helices of APP and between ␤-sheets in fibrils formed by the A␤40 and A␤42 peptides and also that glycine facilitates the ␣-helix to ␤-sheet conversion, which all are key features required for amyloid deposition. As a result, a motif initially discovered to promote dimerization of glycophorin A via self-association of its TM domain may play a major role in A␤ production and pathogenic effects of APP processing in Alzheimer disease (19). In addition, the finding that AICD release by ⑀-cleavage can be dissociated from A␤ production is also of particular interest in the physiopathological function of APP. AICD is thought to act as a regulator of gene transcription (16,17,57), and our data indicate that the genetic program controlled by the amyloidogenic processing of APP might be distinguished from A␤ production, a key event in Alzheimer disease.