Induced Dimerization of the Amyloid Precursor Protein Leads to Decreased Amyloid-β Protein Production*

The amyloid precursor protein (APP) plays a central role in Alzheimer disease (AD) pathogenesis because sequential cleavages by β- and γ-secretase lead to the generation of the amyloid-β (Aβ) peptide, a key constituent in the amyloid plaques present in brains of AD individuals. In several studies APP has recently been shown to form homodimers, and this event appears to influence Aβ generation. However, these studies have relied on APP mutations within the Aβ sequence itself that may affect APP processing by interfering with secretase cleavages independent of dimerization. Therefore, the impact of APP dimerization on Aβ production remains unclear. To address this question, we compared the approach of constitutive cysteine-induced APP dimerization with a regulatable dimerization system that does not require the introduction of mutations within the Aβ sequence. To this end we generated an APP chimeric molecule by fusing a domain of the FK506-binding protein (FKBP) to the C terminus of APP. The addition of the synthetic membrane-permeant drug AP20187 induces rapid dimerization of the APP-FKBP chimera. Using this system we were able to induce up to 70% APP dimers. Our results showed that controlled homodimerization of APP-FKBP leads to a 50% reduction in total Aβ levels in transfected N2a cells. Similar results were obtained with the direct precursor of β-secretase cleavage, C99/SPA4CT-FKBP. Furthermore, there was no modulation of different Aβ peptide species after APP dimerization in this system. Taken together, our results suggest that APP dimerization can directly affect γ-secretase processing and that dimerization is not required for Aβ production.

plaques, one of the neuropathological hallmarks of Alzheimer disease (1,2). In the amyloidogenic pathway A␤ is released from the amyloid precursor protein (APP) (3) after sequential cleavages by ␤-secretase BACE1 (4 -6) and by the ␥-secretase complex (7,8). BACE1 cleavage releases the large ectodomain of APP while generating the membrane-anchored C-terminal APP fragment (CTF) of 99 amino acids (C99). Cleavage of ␤-CTF by ␥-secretase leads to the secretion of A␤ peptides of various lengths and the release of the APP intracellular domain (AICD) into the cytosol (9 -11). The ␥-secretase complex consists of at least four proteins: presenilin, nicastrin, Aph-I, and Pen-2 (12). Presenilin is thought to be the catalytic subunit of the enzyme complex (13), but how the intramembrane scission is carried out remains to be elucidated. Alternatively, APP can first be cleaved in the non-amyloidogenic pathway by ␣-secretase within the A␤ domain between Lys-16 and Leu-17 (14,15). This cleavage releases the APP ectodomain (APPs␣) while generating the membrane-bound C-terminal fragment (␣-CTF) of 83 amino acids (C83). The latter can be further processed by the ␥-secretase complex, resulting in the secretion of the small 3-kDa fragment p3 and the release of AICD.
APP, a type I transmembrane protein (16) of unclear function, may act as a cell surface receptor (3). APP and its two homologues, APLP1 and APLP2, can dimerize in a homotypic or heterotypic manner and, in so doing, promote intercellular adhesion (17). In vivo interaction of APP, APLP1, and APLP2 was demonstrated by cross-linking studies from brain homogenates (18). To date at least four domains have been reported to promote APP dimerization; that is, the E1 domain containing the N-terminal growth factor-like domain and copper binding domain (17), the E2 domain containing the carbohydrate domain in the APP ectodomain (19), the APP juxtamembrane region (20), and the transmembrane domain (21,22). In the latter domain the dimerization appears to be mediated by the GXXXG motif near the luminal face of the transmembrane region (21,23). In addition to promoting cell adhesion, APP dimerization has been proposed to increase susceptibility to cell death (20,24).
Interestingly, by introducing cysteine mutations into the APP juxtamembrane region, it was shown that stable dimers through formation of these disulfide linkages result in significantly enhanced A␤ production (25). This finding is consistent with the observation that stable A␤ dimers are found intracellularly in neurons and in vivo in brain (26). Taken together, these results have led to the idea that APP dimerization can positively regulate A␤ production. However, other laboratories have not been able to confirm some of these observations using slightly different approaches (23,27).
To further address the question of how dimerization of APP affects cleavage by ␣-, ␤-, and ␥-secretase, we chose to test this with a controlled dimerization system. Accordingly, we engineered a chimeric APP molecule by fusing a portion of the FK506-binding protein (FKBP) to the C terminus of APP such that the addition of the synthetic membrane-permeant bifunctional compound, AP20187, will induce dimerization of the APP-FKBP chimera in a controlled manner by binding to the FKBP domains. Using this system, efficient dimerization of APP up to 70% can be achieved in a time and concentration-dependent fashion. Our studies showed that controlled homodimerization of APP-FKBP leads to decreased total A␤ levels in transfected N2a cells. Homodimerization of the ␤-CTF/C99 fragment, the direct precursor of ␥-secretase cleavage, showed comparable results. In addition, induced dimerization of APP did not lead to a modulation of different A␤ peptides as it was reported for GXXXG mutants within the transmembrane domain of APP (21).

EXPERIMENTAL PROCEDURES
DNA Constructs-FKBP-tagged APP695 constructs were generated in a three-fragment ligation step. First, the 1054-bp EcoRI and XhoI 5Ј fragment was released from APP695 pUAST (17). Second, an APP fragment from the 3Ј end was generated by PCR from APP695 pUAST without a stop codon and appending an XbaI restriction site using the primer 5Ј-CGAGTTCCTACAACAGCAGCC-3Ј and an antisense primer, 5Ј-GCGGTGTCTAGAGTTCTGCATCTGCTC-3Ј, and digested with XhoI and XbaI. These two fragments were ligated into the vector pC4F1 (with one FKBP domain) or pC4F2 (with two FKBP domains) (Ariad Pharmaceuticals, Cambridge, MA) using the restriction sites EcoRI and XbaI. To generate FKBP-tagged SPA4CT constructs, a silent mutation was first introduced into SPA4CT DA pCEP4 (28) at position Glu-22 (A␤ numbering) to remove the internal EcoRI site by site-directed mutagenesis (QuikChange, Stratagene, La Jolla, CA). A PCR fragment of the entire SPA4CT DA open reading frame was subsequently generated, again without the stop codon and with an Xba1 site added to the 3Ј end and an EcoRI site added to the 5Јend, and ligated into vector pC4F1 (with one FKBP domain) and vector pC4F2 (with two FKBP domains). A control FKBP-tagged AICD construct starting at Val-50 (A␤ numbering) was generated with a methionine residue added as a start codon and ligated into pC4F1 (or pC4F2).
To generate an APP695 CT HA construct without the FKBP fusion, the 3Ј-untranslated region of APP was amplified from APP695 wt NT HA pcDNA3.1 neo (17) with sense primer 5Ј-GTTTATAACTCGAGACCCCCGCCACAGC-3Ј, containing an XhoI site, and antisense primer, 5Ј-CCCGCG-GCTCTAGATCTTAAAGCATATGTAAAG-3Ј, containing an XbaI site. The PCR product was digested with XhoI and XbaI restriction enzymes and subcloned into the same sites in the pcDNA3.1ϩ neo vector. Sense primer 5Ј-GTGGCCGAG-GAGGAGATTCAGGA-3Ј, starting at position 1481 in the APP695 open reading frame and antisense primer, 5Ј-GCT-GACCTCGAGTTATGCGTAGTCTGGTACGTCGTACG-GATAGTTCTGCATCTGCTC-3Ј, containing the 3Ј end of the APP695 open reading frame without a stop codon, the HA tag, and an XhoI restriction site, were used to amplify APP695 wt NT HA. The PCR product was digested with SacI and XhoI restriction enzymes. The EcoRI and SacI fragment from APP695 pUAST and the preceding SacI-XhoI 1736-bp PCR fragment were ligated into the EcoR1 and XhoI sites in the pcDNA3.1ϩ neo vector containing the 3Ј-untranslated region of APP. Subsequently, the L17C, K28C, K28S, and K28A mutations were introduced into the APP695 CT HA pcDNA3.1ϩ neo plasmid by site-directed mutagenesis (QuikChange).
Antibodies, Cell Lines, Transfections, and Cell Culture-Mouse Neuroblastoma N2a cells were maintained in Dulbecco's modified Eagle's medium supplemented with 10% fetal bovine serum and 1% penicillin/streptomycin. N2a cells were plated in 6-well plates and transfected with Lipofectamine 2000 according to the manufacturer's instructions. The dimerizing ligand, AP20187 (Ariad Pharmaceuticals), was added 7 h after transfection and continued from 1 to 17 h. Monoclonal antibodies included HA.11 to detect the HA epitope tag (Covance), 82E1 against the free N terminus of A␤ (IBL, Minneapolis MN), B436 recognizing the N-terminal region of A␤ (29), and 6E10 (Signet), WO2 (30), and Ab9 (31) against the amino acids 1-16 of human A␤. The polyclonal sAPP␤ antibody recognizes the ISEVKM sequence at the C terminus of wild type human sAPP␤ (Immuno-Biological Laboratories, Inc., Minneapolis, MN). Antibody 6E10 was also used for detection of sAPP␣, and ␤-secretase cleavage site-specific antibody 82E1 was also used for detection of the APP␤ C-terminal fragment. The ␤-tubulin antibody E7-a1 was obtained from Developmental Studies Hybridoma Bank.
SDS Electrophoresis and Immunoblotting-Cells were lysed for 15 min at 4°C in lysis buffer (50 mM Tris/HCl, pH 7.5, 150 mM NaCl, 5 mM EDTA, and 1% Nonidet P40) supplemented with protease inhibitors (Complete TM protease inhibitor mixture, Roche Applied Science). The supernatants were collected, and the protein concentration determined with a Micro BCA assay (Pierce). Equal amounts of protein samples were separated on 8 or 12% Bis-Tris SDS-PAGE or 12.5 or 15% Tris/glycine SDS-PAGE and then transferred to nitrocellulose membranes. To separate A␤38, A␤39, A␤40, and A␤42 species, media samples were immunoprecipitated with antibody B436 as above and separated on 10% Tris/Tricine gels containing 8 M urea (32). Synthetic A␤ peptide standards (AnaSpec) were separated in parallel with the experimental samples. After electrophoresis and transfer to nitrocellulose, the membranes were incubated with 82E1 monoclonal antibody which recognizes all fulllength A␤ peptides starting at position 1.
Blue Native Gel Electrophoresis-Blue native gel electrophoresis was performed according to a protocol modified from Schägger et al. (33). In brief, cells in one 10-cm cell culture dish were washed once and collected in phosphate-buffered saline at 4°C. The cell pellets were resuspended in 1 ml of homogenization buffer (250 mM sucrose in 20 mM HEPES, pH 7.4, with protease inhibitors) and then sheared by passing through a 27 ϫ gauge needle 10 times. Postnuclear supernatant was collected after a low speed spin at 1000 ϫ g for 15 min at 4°C. The membranes were pelleted after centrifugation at 100,000 ϫ g for 1 h at 4°C and washed once with 200 l of homogenization buffer, and the centrifugation was repeated once. The pellets containing the membranes were resuspended in 200 l of homogenization buffer. 100 g of protein were solubilized with Blue Native sample buffer (1.5 M amino caproic acid, 0.05 M Bis-Tris, 10% n-dodedecyl-␤-D-maltoside, and protease inhibitor at pH 7.5). The samples were incubated on ice for 30 min and then centrifuged for 10 min at 14,000 rpm at 4°C in a microcentrifuge. Blue Native loading buffer (5.0% Serva Coomassie Brilliant Blue G250 and 1.0 M aminocaproic acid) was added to the supernatant. The samples were separated on 4 -15% Tris-HCl gels (Criterion, Bio-Rad) overnight at 4°C with Coomassie Blue containing cathode buffer (10ϫ cathode buffer, pH 7.0, 0.5 M Tricine, 0.15 M Bis-Tris, 0.2% Coomassie Blue) and anode buffer (pH 7.0, 0.5 M Bis-Tris). The gel was transferred to a polyvinylidene difluoride membrane. The following molecular weight standards were used: thyroglobulin (669 kDa), apoferritin (443 kDa), catalase (240 kDa), aldolase (158 kDa), and bovine serum albumin (66 kDa), all from Sigma.
Immunoprecipitation of A␤-N2a cells in six-well plates were transiently transfected with Lipofectamine 2000 according to the manufacturer's instructions. Overnight conditioned media were collected and incubated overnight at 4°C with antibody B436 and antimouse IgG-agarose beads (American Qualex, San Clemente, CA). After 3 washes with buffer A (10 mM Tris-HCl, pH 7.5, 150 mM NaCl, 2 mM EDTA, 0.2% Nonidet P-40) and 1 wash with 10 mM Tris-HCl, pH 7.5, the beads were resuspended in 2ϫ Laemmli sample buffer plus dithiothreitol and separated by 12% Bis-Tris SDS-PAGE.
Mass Spectrometry Analysis-Matrix-assisted laser desorption/ionization time of flight (MALDI-TOF) mass spectrometry of A␤ peptides was performed on a 4800 MALDI-TOF-TOF (Applied Biosystems/MDS-Sciex, Foster City, CA). A␤ peptides were immunoprecipitated from conditioned medium with antibody B436 coupled to Seize TM beads (Pierce). The immunoprecipitates were eluted from the beads with 25% 0.1% trifluoroacetic acid, 75% acetonitrile, and the resulting samples were mixed 1:1 with ␣-cyano-4-hydroxycinnamic acid matrix in methanol:acetonitrile:water (36:56:8) (Agilent, Santa Clara, CA) and spotted on the MALDI target grid. Mass spectra were acquired from m/z 3,500 -5,000 Da in reflector positive mode at 10,000 shots/spectrum using single shot protection and delayed extraction times of 420 ns. The mass spectra are plotted as m/z versus relative intensity.

RESULTS
Cysteine Induced Dimerization of APP-Studies from Multhaup and co-workers (25) have shown that artificial dimeriza-tion of APP by introducing cysteine residues within the A␤ domain increased A␤ production. K28C mutation (using A␤ numbering) showed 6 -7-fold augmentation in A␤ secretion (25), whereas the L17C mutation (21) showed unchanged A␤40 levels but increased A␤42 release, suggesting that dimers of APP facilitate A␤ generation. However, a recent study reported that a K28S mutation showed a marked (90%) reduction in A␤ production (35), possibly due to impairment in ␥-secretase cleavage. As a result, we wish to first revisit these cysteine mutants as well as determine the effects of base substitutions at the lysine residue at position 28 (by A␤ numbering or position 624 by APP695 numbering). Accordingly, we examined APP FIGURE 2. Induced dimerization of C-terminal FKBP-tagged APP695 and SPA4CT analyzed by Blue Native gel electrophoresis. A, schematic representation of APP695 F1 and APP695 F2 constructs consisting of an APP695-FKBP fusion protein with a C-terminal HA tag. APP695 F1 contains one 12-kDa FKBP tag, and APP695 F2 contains two FKBP tags fused in a tandem repeat. AP20187 is an analogue of rapamycin, which binds to FKBP, thereby inducing dimerization of APP-FKBP fusion proteins. B, schematic representation of the constructs SPA4CT F1 and SPA4CT F2, containing one FKBP or two FKBP tags, respectively. C and D, dimerization was induced with AP20187 (100 nM) overnight in N2a cells transiently transfected with APP695 F1/APP695 F2 (C) or with SPA4CT F1/SPA4CT F2 (D). Note, formation of dimers and multimers in the APP chimeric proteins seen by Blue Native PAGE followed by Western blotting with anti-HA antibody.
processing and A␤ production in four different mutations: L17C, K28C, K28S, and K28A (Fig. 1A). All three mutations at A␤28 (K28A, K28C, and K28S) resulted in a significant reduction in A␤ secretion. For K28A and K28S, the A␤ levels were essentially undetectable. The magnitude of reduction in A␤ production in the K28C mutation was slightly variable, depending on the A␤ antibody used for immunoprecipitations (supplemental Fig. 1). In parallel, there was a strong reduction in sAPP␣ secretion in the L17C mutant, indicating that impairment of ␣-secretase cleavage might have led to increased ␤-secretase cleavage, as seen in the concomitant rise in the level of ␤-CTF. Although there was a mild reduction in sAPP␣ and sAPP␤ secretion in the K28C mutant, neither the K28A nor the K28S mutant showed any significant alterations in sAPP␤ secretion. In addition, there was a substantial change in ␣-secretase processing for mutants K28S and K28A. Nevertheless, our findings are consistent with those reported by Ren et al. (35) that substitutions of the basic lysine residue at position 28 by other hydrophilic (Cys or Ser) or hydrophobic (Ala) amino acids led to marked reduction in A␤ production, presumably by affecting ␥-secretase cleavage. However, in view of the K28C results it appears that dimerized APP does not consistently augment A␤ generation, especially if the K28C substitution led to other perturbations in APP processing.
Blue Native Gel Analysis of Induced and Constitutive Dimerization of APP695 and C99-In light of the above results, another approach to dimerize APP is needed to determine the impact of APP dimerization on A␤ production, especially one that minimizes secondary effects due to amino acid substitutions within the A␤ sequence. We used a regulated homodimerization system based on the human FKBP12 and a small bifunctional ligand, AP20187 (36). We engineered a chimeric APP construct containing the full-length human APP cDNA fused to one or two FKBP domains and a HA epitope tag at the C terminus ( Fig.  2A). AP20187, an analogue of rapamycin, is a membrane-permeable compound that induces rapid dimerization of the FKBP or, in this case, APP695-FKBP. Either one FKBP domain (APP F1) or two FKBP domains (APP F2) were fused to the C terminus of APP to enable APP dimerization. To analyze the impact of APP dimerization on ␥-secretase processing directly, we also fused one or two FKBP domains to a C99 or ␤-CTF construct, SPA4CT, as above, and these are termed SPA4CT F1 and F2, respectively (Fig. 2B). SPA4CT mimics the C99 ␤ C-terminal fragment generated after BACE1 cleavage and is a direct substrate of ␥-secretase cleavage (28).
To examine dimerization of these APP-FKBP chimeric constructs, APP695 F1 and APP695 F2 were transiently transfected into mouse N2a cells and treated overnight with AP20187. Blue Native PAGE showed that the majority of APP molecules were in the dimer form (about 70%), whereas the untreated control sample showed very little APP dimers (Fig. 2C). For comparison, both APP cysteine mutants (APP L17C and K28C) showed comparable levels of APP dimer formation using Blue Native PAGE (Fig. 3A). Dimerization of APP F2 was more efficient than APP F1 due to the additional FKBP domain with almost no APP monomer present (Fig. 2C). In addition, APP multimers, likely tetramers, were also detected. Of note, in the Blue Native gel system, the APP monomer migrated with a mobility of ϳ300 kDa (Fig. 2C), whereas the APP dimer had a molecular mass of twice the monomer size (ϳ600 kDa). The apparent molecular mass of APP monomers and dimers in this gel system is in accordance with a recent publication by Sato et al. (37) and likely does not reflect their true molecular masses. This is consistent with the observation that migration of the two APP cysteine mutants was essentially the same as the APP-FKBP chimeric proteins using Blue Native PAGE (Fig. 3A). Furthermore, in non-reducing and SDS-denaturing gels, the cysteine mutants migrated at the predicted molecular masses and showed comparable ratios of monomers and dimers as seen in the Blue Native gels (supplemental Fig. 1).
In the SPA4CT F1 construct the addition of AP20187 also led to about 70% dimerization, similar to what was observed for APP F1 (Fig. 2D). Similarly, dimerization of SPA4CT F2 showed strong formation of dimers and higher order mul- timers with almost complete absence of the monomeric species (Fig. 2D).
To determine the dose response of APP-FKBP chimeric protein to AP20187, APP F1-transfected cells were incubated overnight with different concentrations of AP20187 (50 nM, 100 nM, 500 nM, and 1 M). After Blue Native gel electrophoresis, essentially the same level of dimerization (ϳ70%) was seen in all four concentrations of AP20187 after overnight treatment (Fig. 3B). To determine the time course of dimerization, APP F1-transfected cells were incubated with either 500 nM or 1 M AP20187 and analyzed from 10 to 60 min after treatment. Both concentrations led to rapid dimerization such that abundant dimers were present after only 10 min of treatment in accordance with previously published studies using this system ( Fig. 3C for 1 M, data not shown for 500 nM). Therefore, dimerization of the APP-FKBP chimera with AP20187 was rapid and efficient. The level of induced dimerization was also comparable with the APP cysteine mutants but has the advantage of not altering any A␤ amino acid residues.
APP695 F1 and APP695 F2 Are Functional ␥-Secretase Substrates-To confirm that the addition of the FKBP sequences to the C terminus of APP did not perturb A␤ generation, APP695 F1 and APP695 F2 were transiently transfected into N2a cells, and the cells were analyzed for APP processing (Fig. 4). Abundant APP CTFs and AICD were detected in cells transfected with either construct. The latter species were confirmed by mobility to AICD F1 or AICD F2 control constructs, and production of this fragment was abolished after treatment with the ␥-secretase inhibitor DAPT. The stability of AICD is likely due to the presence of the FKBP tag because a number of reports have shown stabilization of AICD after fusion with various protein tags (9, 38 -40). Furthermore, high levels of A␤ were released from cells transfected with the APP F1 and APP F2 constructs, indicating no overt perturbation of ␥-secretase cleavage. DAPT treatment inhibited A␤ production as well as AICD production from cells transfected with APP F1 and APP F2 (Fig. 4).

Dimerization of APP695 F1 Leads to Decreased A␤
Production-We next assessed the effects of induced dimerization on APP proteolytic processing. However, we first wish to ascertain that AP20187 does not by itself alter APP processing or A␤ generation. Accordingly, N2a cells stably transfected with wild type full-length APP695 (without the FKBP tag) were treated with increasing doses of AP20187. Indeed, at concentrations up to 1 M, we did not detect any appreciable changes in full-length APP or in the levels of secreted sAPP␣, sAPP␤, A␤, indicating that this compound does not perturb APP metabolism (Fig. 5). Next, N2a cells were transiently transfected with the APP F1 construct and then treated overnight with 50 or 100 nM AP20187. After this treatment, steady state levels of full-length APP were consistently higher than control cells not treated with AP20187 (Fig. 6A). Levels of A␤ were essentially unchanged, but when corrected for the increased APP levels, there was a significant 50% reduction in A␤ secretion. Interestingly, AICD and ␤-CTF levels were slightly increased, whereas ␣-CTFs, sAPP␣, and sAPP␤ levels showed a mild reduction or no change. Similar results were obtained with the APP F2 construct (supplemental Fig. 2). In addition, similar changes were seen in intracellular A␤ levels after AP20187 treatment, suggesting that APP-FKBP-induced dimerization did not result in a selective retention of A␤ in intracellular compartments (supplemental Fig. 4).
Because it has been suggested that dimerization of APP modulates the production of shorter A␤ species (21), we proceeded to determine whether dimerization of APP695 induced by AP20187 resulted in the generation of different A␤ species. Therefore, media of N2a cells transfected with the APP695 F1 construct and treated with 50 and 100 nM AP20187 were analyzed on Tricine-urea gels to separate the  2). AICD, which co-migrated with a control AICD F1 (or AICD F2) construct, can be detected from both APP full-length (FL) constructs, and generation of this fragment can be blocked by DAPT. Similarly, A␤ is released into the medium in a DAPT-dependent manner from both APP F1 and APP F2 constructs. Full-length APP and CTFs in the cell lysates were immunoblotted with an anti-HA antibody, and ␤-tubulin was detected with antibody E7. A␤ was immunoprecipitated from conditioned medium with A␤ antibody B436 and immunoblotted with ␤-site-specific 82E1 antibody. FIGURE 5. AP20187 does not affect APP695 wt processing. N2a cells stably expressing wild type full-length (FL) APP695 (without FKBP tag) were treated with increasing concentrations of AP20187 overnight. Cell lysates were analyzed via immunoblotting with antibody 6E10 for APP and antibody E7 for ␤-tubulin. A␤ was immunoprecipitated with antibody B436 and immunoblotted with antibody 82E1. sAPP␣ and sAPP␤ in the media were detected by immunoblotting with 6E10 and sAPP␤-specific antibodies, respectively. various A␤ species (Fig. 7A). By this approach the profile of A␤-(38 -42) peptides from the APP695 F1 construct (and the APP695 F2 construct, data not shown) was essentially unchanged, with A␤40 being the predominant species as expected. Quantification of the different A␤ species from three independent experiments showed no significant change in their ratios. Finally, A␤ from cells transiently transfected with APP F1 or APP F2 were analyzed by MALDI-TOF after treatment with 500 nM AP20187 (Fig. 7, B  and C). The results showed that there were no substantial changes in A␤ species from either APP F1 or APP F2 cells with or without AP20187 treatment. Taken together, our results showed that induced dimerization of APP reduced A␤ levels without modulation of individual A␤ species.
Dimerization of C99 Leads to Decreased A␤ Production-The above results suggested that dimerization of APP lowers A␤ generation, possibly at the level of ␥-secretase cleavage. To confirm this interpretation, we next analyzed the processing of SPA4CT, the ␤-CTF fragment generated after BACE1 cleavage. Accordingly, we assessed A␤ and AICD levels in cells transfected with the C99/SPA4CT F1 or C99/SPA4CT F2 constructs, both of which can be efficiently dimerized (see Fig. 2D). Treatment with AP20187 led to an increase in C99/␤-CTF fragment and AICD levels (Fig. 8A, supplemental Fig. 3), similar to the observations with full-length APP (see Fig. 6, supplemental Fig.  2). When normalized to steady state levels of ␤-CTF, levels of secreted total A␤ were markedly diminished (70 -80%) in cells treated with 100 nM AP20187. The medium was also analyzed by ELISA to quantify the relative levels of A␤40 and A␤42 production (Fig. 8,  D and E). Indeed, both species were reduced by ϳ70%, with slightly more reduction seen for the A␤40 peptides. In sum, studies with both full-length APP and N-terminaltruncated APP (SPA4CT) constructs showed similar levels of reduction in A␤ secretion after induction of dimerization.

DISCUSSION
It has been reported that APP contains a number of domains that mediate homo-and heterodimerization, the latter interacting with APLP1 and APLP2, and in support of the functionality of these domains, small amounts of APP are recovered as dimers in vitro and in vivo (17,18,21,25). To date, a number of diverse physiological roles have been attributed to this dimerization event. In addition, based primarily on artificial cysteine mutations designed to induce disulfide bond formation, it has been suggested that APP dimerization accelerates A␤ production (25). However, subsequent studies have not consistently confirmed these initial observations (23,27). In this study we chose to use a different approach to induce APP dimerization to explore the relationship between dimerization and A␤ generation without having to introduce cysteine mutations by creating APP-FKBP chimeric proteins. The addition of the bivalent compound AP20187 led to rapid and efficient formation of APP dimers in transfected cells. Fusion of one FKBP domain resulted in ϳ70% dimer formation, whereas fusion of two FKBP domains resulted in dimers and multimers, with almost a complete absence of APP monomers. However, when APP dimers or multimers were formed, there was a consistent decrease in A␤ secretion that ranged from 50 to 80% depending on whether one or two FKBP domains were inserted. Therefore, we conclude that APP dimerization does not directly promote A␤ generation and in fact negatively impacts A␤ production.
Previous studies analyzing the putative role of APP dimerization all used cysteine substitutions within the A␤ domain to create artificial disulfide bonds between adjacent APP molecules. This approach leads to efficient and constitutive dimerization of APP. However, inconsistent results have been reported regarding the effects on A␤ generation. Given the interests in the mechanisms of A␤ production, we decided to test the role of APP dimerization on APP processing by engineering APP-FKBP chimeric constructs that can be efficiently dimerized with the bivalent compound, AP20187. The advantage of this approach is that no mutations are introduced into the APP sequence, as the latter mutations may perturb processing by ␣-, ␤-, or ␥-secretase by interfering with one of the enzymatic activities. For both of the published cysteine mutants (L17C and K28C), this is a critical point because, for example, the mutation L17C in APP is directly located at the ␣-secretase cleavage site and has the potential of interfering with ␣-secretase activity (21). Indeed, this is precisely the outcome we observed, as there was a significant reduction in ␣-secretase cleavage and a corresponding increase in ␤-secretase cleavage as measured by the levels of sAPP␣ and sAPP␤, respectively (Fig. 1). In fact, this alteration alone may explain the increase in A␤ release without potentially having to take into account any effects of actual APP dimerization. A second advantage of the FKBP-AP20187 system is the time-and concentration-dependent regulation of dimerization as compared with the constitutive dimerization of APP induced by cysteine mutations (Fig. 3). In our studies, although a tag of 122 amino acids was appended to the C terminus of APP, there were no overt perturbations in APP processing that we could observe. Both mass spectrometry and Tricine-urea gel electrophoresis showed the normal predicted profile of A␤ peptides without any alterations before or after dimerization from both APP F1 and APP F2 constructs. These findings are consistent with the apparent normal processing of APP GFP fusions used by a number of laboratories (38,(41)(42)(43)(44)(45). However, with this approach, our results unequivocally showed that forced dimerization of APP lowered rather than increased A␤ generation. Similar results were obtained with both full-length APP and C99 (SPA4CT) constructs, leading us to hypothesize that this reduction is due to an effect on ␥-secretase cleavage and not on other aspects of APP processing, such as ␣or ␤-secretase cleavage.
In our studies, although A␤ generation was significantly reduced, we did observe an increase in the levels of AICD after the addition of the dimerization compound. This was not expected because both A␤ and AICD are derived by ␥-secretase activity (9 -11), and it has been reported that the generation of A␤ and AICD are equimolar, at least under basal conditions (46). A␤ generation appears to initially start at the ⑀-cleavage near the cytoplasmic face that first releases AICD from the membrane tether, before -cleavage, and finally, ␥-secretase cleavages occur (47,48). In this model these processive cleavages can be perturbed while keeping ⑀-cleavage relatively intact. In fact, a recent publication showed that mutations within the GXXXG motif in the APP transmembrane domain led to a reduction in A␤40 and A␤42 without altering AICD levels (23). In this latter case, however, the drop in A␤40 and A␤42 levels was accompanied by an increase in A␤34 species (21), indicating that mutations at the GXXXG motif were specific for the generation of A␤ peptides, presumably due to alterations to ␥-secretase cleavage. For this reason we specifically examined the profile of A␤ peptides in the APP-FKBP chimera by both Western blotting and MALDI-TOF (Fig. 7), and we did FIGURE 7. No modulation of different A␤ species after induction of APP dimerization. A, N2a cells transfected with APP695 F1 were treated overnight with 50 or 100 nM AP20187. Full-length (FL) APP in cell lysates was detected with an anti-HA antibody. The conditioned media were immunoprecipitated with A␤ antibody B436 and separated on a 10% Tricine-urea gel. No significant changes in the relative levels of the major A␤ species were observed. Synthetic A␤ 38 , A␤ 40 , and A␤ 42 peptides were run as control size markers. B and C, MALDI analysis of A␤ peptides secreted from APP695 F1 (B)-or APP695 F2 (C)-transfected N2a cells after treatment with 500 nM AP20187 as compared with ethanol treated controls.
not detect any alterations in the profile of A␤ peptides generated from the APP-FKBP chimeric molecules before or after dimerization. Another possibility is that the stability of AICD fusion peptides is different from untagged molecules. Normally, AICD is rapidly turned over, but it is likely that the stability of AICD-FKBP fusion peptides is enhanced, a characteristic shown by other APP C-terminal fusion tags (9,35,38). Given this, it is possible that dimerized AICD-FKBP species are even more stable than monomeric AICD-FKBP peptides. Finally, there are other examples wherein A␤ and AICD production are not linked. For example, a number of mutations within the APP transmembrane region were shown to have distinct effects on the two cleavage events (23,49,50). And, TMP21, a member of the p24 cargo protein family, appeared to selectively inhibit APP processing at the ␥-site without affecting ⑀-cleavage (51).
Our results are consistent with the report by Ren et al. (35) showing that the SPA4CT-GVP mutant K28S showed a 90% reduction of total A␤ production, whereas AICD levels were basically unaffected. Because both the K28A and K28C mutations also led to a marked reduction in A␤ secretion (Fig. 1), our results are in agreement with the observations that these mutations likely perturbed ␥-secretase cleavage. In particular, although the K28C resulted in high levels of APP dimers, this itself was insufficient to augment A␤ generation and release.
On the other hand, the A␤ reduction we saw in induced dimerization of APP-FKBP chimeras may be supra-physiologic, as the degree of APP dimerization that takes place normally in cultured cells appears to be far less than what was achieved by AP20187-induced dimerization. Furthermore, dimerization at the APP C terminus may not be equivalent to dimerization within other APP domains. However, it remains to be determined what percentage of APP is dimerized in vivo, as this has not been examined in detail. It has also been suggested that APP dimerization increases the susceptibility of cells to a variety of injury, including A␤ (24). Therefore, it is possible that even though APP dimerization may not directly impact A␤ production, this process may still play an important role in synaptic loss or neuronal death in Alzheimer disease. Last, these observations are all in accord with recent publications showing that ␥-secretase is likely a monomeric rather than a dimeric complex, as was originally proposed (37,(52)(53)(54). Given these findings, ␥-secretase is likely to function in the monomeric state and cleaves one rather than two substrate APP molecules within the complex.