Presenilin 1 and Presenilin 2 Target γ-Secretase Complexes to Distinct Cellular Compartments*

γ-Secretase complexes achieve the production of amyloid peptides playing a key role in Alzheimer disease. These proteases have many substrates involved in important physiological functions. They are composed of two constant subunits, nicastrin and PEN2, and two variable ones, presenilin (PS1 or PS2) and APH1 (APH1aL, APH1aS, or APH1b). Whether the composition of a given γ-secretase complex determines a specific cellular targeting remains unsolved. Here we combined a bidirectional inducible promoter and 2A peptide technology to generate constructs for the temporary, stoichiometric co-expression of six different combinations of the four γ-secretase subunits including EGFP-tagged nicastrin. These plasmids allow for the formation of functional γ-secretase complexes displaying specific activities and maturations. We show that PS1-containing γ-secretase complexes were targeted to the plasma membrane, whereas PS2-containing ones were addressed to the trans-Golgi network, to recycling endosomes, and, depending on the APH1-variant, to late endocytic compartments. Overall, these novel constructs unravel a presenilin-dependent subcellular targeting of γ-secretase complexes. These tools should prove useful to determine whether the cellular distribution of γ-secretase complexes contributes to substrate selectivity and to delineate regulations of their trafficking.

␥-Secretase complexes achieve the production of amyloid peptides playing a key role in Alzheimer disease. These proteases have many substrates involved in important physiological functions. They are composed of two constant subunits, nicastrin and PEN2, and two variable ones, presenilin (PS1 or PS2) and APH1 (APH1aL, APH1aS, or APH1b). Whether the composition of a given ␥-secretase complex determines a specific cellular targeting remains unsolved. Here we combined a bidirectional inducible promoter and 2A peptide technology to generate constructs for the temporary, stoichiometric co-expression of six different combinations of the four ␥-secretase subunits including EGFP-tagged nicastrin. These plasmids allow for the formation of functional ␥-secretase complexes displaying specific activities and maturations. We show that PS1containing ␥-secretase complexes were targeted to the plasma membrane, whereas PS2-containing ones were addressed to the trans-Golgi network, to recycling endosomes, and, depending on the APH1-variant, to late endocytic compartments. Overall, these novel constructs unravel a presenilin-dependent subcellular targeting of ␥-secretase complexes. These tools should prove useful to determine whether the cellular distribution of ␥-secretase complexes contributes to substrate selectivity and to delineate regulations of their trafficking.
␥-Secretase complexes were mainly characterized through their association with Alzheimer disease (AD) 2 as they achieve the proteolytic cleavages of the amyloid precursor protein (APP) leading to the release of amyloid peptides (A␤s), which are considered to play a central role in the neurodegenerative process (1). The inhibition of ␥-secretase activity was therefore regarded as a potential therapy for AD. However, clinical trials of ␥-secretase inhibitors were halted because of increased occurrences of adverse effects in treated patients, including skin cancers and even a worsening of cognitive function (2,3). These undesired consequences are probably due to the broad spectrum of ␥-secretase inhibitors that are blocking the processing of not only APP, but also many other ␥-secretase substrates. Indeed, numerous transmembrane proteins (up to 90 to date), including Notch, have been identified as ␥-secretase substrates and are implicated in important signaling functions regulating cell fate, adhesion, migration, neurite outgrowth, or synaptogenesis (for reviews see Refs. 4 and 5). Research efforts are currently under way to develop alternative strategies to specifically inhibit/modulate ␥-secretase activity toward APP and spare other substrates. A better understanding of the subcellular trafficking of ␥-secretase complexes and their substrates may provide useful clues toward achieving this goal.
Importantly, as mentioned above, two homologues of PS (PS1 and PS2) and APH1 (APH1a and APH1b) were identified in humans, and their different combinations with other subunits can generate at least four distinct ␥-secretase complexes. Moreover, alternatively spliced forms of PS (39) and APH1a (APH1aL and APH1aS) (40) were identified, increasing further the theoretical diversity of ␥-secretase complexes. The levels of expression of the variable subunits PS and APH1 can differ between tissues and/or cell types, but ␥-secretase complexes of distinct composition can co-exist within the cell (41,42). The influence of ␥-secretase subunit variability on ␥-secretase subcellular trafficking has not been fully explored, but it is worthy of attention because the targeting of a given ␥-secretase complex to a specific cellular compartment may underlie the substrate specificity of this complex and thus yield useful information for designing more specific inhibitors of APP processing.
Recently, we used bigenic expression vectors to co-express the four ␥-secretase subunits, including PS1 and different variants of APH1, and used bimolecular fluorescence complementation (BiFC) to visualize newly assembled ␥-secretase complexes (36). We observed that PS1-containing ␥-secretase complexes were mainly targeted to the plasma membrane and that those including APH1b were more retained/retrieved in the ER than those including APH1a.
In the present work, we generated new tools for the inducible, temporary, and stoichiometric expression of six different combinations of the four ␥-secretase subunits. We show that these constructs allow the formation and maturation of functional ␥-secretase complexes and their visualization in post-ER compartments. Remarkably, we observe distinct subcellular targeting of ␥-secretase complexes depending on their presenilin subunits. PS1-containing ␥-secretase complexes are targeted mainly to the plasma membrane, whereas PS2-containing ones are targeted to different types of endosomes, depending on their APH1 subunit. These results unravel the importance of ␥-secretase composition for its cellular targeting. Our novel constructs should prove very useful to further explore the subcellular trafficking of ␥-secretase complexes and to investigate the link between ␥-secretase cellular compartmentalization and substrate specificity.

Experimental Procedures
Plasmid Constructs-All ␥-secretase subunit cDNAs used here were of human origin. All of the constructs described here were verified by sequencing. The plasmid backbone for all of the constructs was pTRE3G-BI (Clontech), which harbors a doxycycline-inducible bidirectional promoter flanked by two multicloning sites. A BstBI site was introduced into pTRE3G-BI by cloning the following annealed primers into the BamHI and NotI sites: 5Ј-GATCCATCTTCGAAGATGC-3Ј and 5Ј-GGCCGCATCTTCGAAGATG-3Ј. Transgene expression from pTRE3G-BI requires the co-transfection of pCMV-Tet3G (Clontech) for the expression of a transactivator protein that activates transcription from pTRE3G-BI promoter when bound to doxycycline.
APH1aL sequence was PCR-amplified using pEGFP-C1-APH1 as a template and the following primer set: 5Ј-CGCTC-
Cell Culture, Transfections, and Induction Protocols for Biochemical Analyses-COS-7 cells (ECACC) were plated in 100-mm dishes at a density of 3 ϫ 10 4 cells/cm 2 in Dulbecco's modified Eagle's medium (Gibco) supplemented with 10% FBS (Sigma). Eighteen hours after plating, cells were co-transfected with 2.4 g of pCMV-Tet3G-and 9.6 g of pTRE3G-BI-based constructs (see above and Fig. 1) per 100-mm dish, using jetPRIME transfection reagent according to the manufacturer's instructions (Polyplus transfection). Twenty-two hours after transfection, transgene expression was induced transiently for 5 h with 1 g/ml doxycycline in culture medium and then stopped with three washes in 1ϫ PBS before placing the cells back in culture medium. Nineteen hours after induction, cells were processed to prepare membrane-enriched fractions or lysed for co-immunoprecipitation experiments (see below).
Preparation of Membrane-enriched Fractions and in Vitro ␥-Secretase Assay-Membrane-enriched fractions were prepared mainly as described previously (48) and used for an in vitro ␥-secretase assay (49). The production of ␥-secretase metabolites was analyzed by Western blotting using 16.5% Tris-Tricine gels and a combination of mouse monoclonal anti-FLAG M2 (F1804, Sigma) and anti-A␤(1-12) 2H3 (50) antibodies (kind gift of Dr. D. Schenk, Elan Biosciences, South San Francisco, CA) to visualize APP intracellular domain (AICD) and A␤, respectively. The peak heights of signal intensities from AICD and A␤ bands were quantified with ImageJ software using the central third of the lane width for analysis according to Gassmann et al. (51).
Preparation and Separation of Native ␥-Secretase Complexes-Membrane-enriched fraction pellets from COS-7 cells (see above) were resuspended in 60 l of 1ϫ NativePAGE sample buffer (Invitrogen), 1% digitonin, 1ϫ protease inhibitor mixture (Sigma), followed by a 25-min incubation on ice. Insoluble material was pelleted (100,000 ϫ g, 20 min at 4°C), and the supernatants corresponding to solubilized proteins were collected. Thirteen micrograms of solubilized proteins were separated at 6°C using the NativePAGE Novex BisTris 3-12% polyacrylamide gradient gel system from Invitrogen according to the manufacturer's recommendations, which is based on the blue native PAGE technique developed by Schägger and von Jagow (53). NativeMark unstained protein standard (Invitrogen) was used for molecular weight estimation. Following blue native PAGE, proteins were transferred onto a PVDF membrane in NuPAGE transfer buffer (Invitrogen). The membrane was then incubated in 8% acetic acid for 15 min, rinsed in deionized water, and air-dried. The membrane was rewet in methanol prior to immunodetection with anti-GFP antibodies (described above).
ELISA Quantification of A␤ Peptides-The secreted medium collected from transfected HEK293 APP WT /BACE1 stable cells (see transfection and induction protocols above) was diluted 1:40 for A␤40 analysis and 1:3 for A␤42 analysis using specific ELISA kits from Invitrogen.
Fluorescence Microscopy-COS-7 cells were plated on 18-mm acid-washed coverslips in 12-well plates for immunocytochemistry experiments or on 25-mm acid-washed coverslips in 6-well plates for live imaging at a density of 7.5 ϫ 10 3 cells/cm 2 . Cells were co-transfected 16 h after plating with 0.2 g of pCMV-Tet3G and 0.8 g of pTRE3G-BI constructs (see "Plasmid Constructs") per well (for 12-well plates or twice these amounts for 6-well plates) using jetPRIME transfection reagent (Polyplus transfection). Twenty-two hours after transfection, transgene expression was induced transiently for 2 h with 1 g/ml doxycycline.
Primary Neuronal Cultures and Transfections-The following experimental procedures were in accordance with the European Communities Council Directive of 24 November 1986 (86/609/EEC) and local French legislation. Hippocampal neurons were prepared from Sprague-Dawley rat embryos (Rattus norvegicus, RjHan:SD strain, mixed male and female day 18 embryos, Janvier Labs) according to Kaech and Banker (26), plated at a density of 12,500 cells/cm 2 on poly-D-lysine-coated (0.1 mg/ml, 70,000 -150,000 g/mol polymers; Sigma) glass coverslips (Menzel Gläser) equipped with paraffin dots, and cultured in Neurobasal medium supplemented with B27 (both from Gibco) and 1-␤-D-arabinofuranosylcytosine (5 M; Sigma), facing a glial feeding layer grown on 60-mm dishes. Glial cells were prepared from the cortex of Sprague-Dawley rat embryos (embryonic day 18) 2 weeks before the neuronal culture according to Zeitelhofer et al. (31). Cultured neurons were transfected at 6 days in vitro (DIV) using Lipofectamine 2000 (5 l; Invitrogen), 0.6 g of pCMV-Tet3G, and 2.4 g of pTRE3G-BI constructs diluted in a final volume of 200 l of Opti-MEM medium (Gibco) for each 60-mm dish. Transgene expression was induced at 16 DIV with doxycycline (500 ng/ml) for 5 h. Neurons were fixed at 17 DIV, and coverslips were mounted using the same procedure as for COS-7 cells (see above).
Statistical Analysis-Statistical analyses were performed with Prism version 4 (GraphPad Software) using repeated measure one-way analysis of variance followed by Newman-Keuls multiple comparison tests.

Results
Generation of New Tools for the Inducible, Temporary, and Stoichiometric Co-expression of Four ␥-Secretase Subunits and the Visualization of Assembled ␥-Secretase Complexes-Our previous work with BiFC constructs (36) showed us that a continuous expression of ␥-secretase subunits leads to the accumulation of unassembled or preassembled ␥-secretase subunits within the ER that can obscure the visualization of mature ␥-secretase complexes in post-ER compartments. To avoid this drawback, we chose to express given sets of ␥-secretase subunits using a commercial vector containing a bidirectional and doxycycline-inducible promoter (Fig. 1A), allowing a temporary and equimolar co-expression of two transgenes. By stopping the induction, we should allow ␥-secretase complex maturation and trafficking away from the ER to their cellular destination(s) before their examination.
We also aimed at favoring a stoichiometric expression of the four ␥-secretase subunits; therefore, we could not rely on the co-transfection of two vectors with bidirectional promoters. To achieve this goal, we took advantage of the 2A peptide-linked multicistronic vector technology, which allows the expression of multiple proteins from a single open reading frame (43). The "self-cleaving" 2A peptide sequence was initially characterized ␥-Secretase Subcellular Targeting Relies on Presenilins in foot and mouth picornavirus (55), and since then, many other 2A peptide sequences have been identified in other viruses and some parasites. The 2A peptide "cleavage" actually occurs through a ribosomal skip mechanism at the end of 2A peptide sequences (56). These sequences can be placed between genes within a single vector to allow stoichiometric production of discrete protein products. Because the "cleavage" occurs at the end of the 2A sequence, most of the peptide (Ͻ20 amino acids) will consequently remain attached to the protein placed upstream of the 2A peptide. This was an important aspect to consider for choosing the position at which to clone ␥-secretase subunit sequences around a 2A peptide sequence. We chose to subclone the C-terminal EGFP-tagged nicastrin (NCT-EGFP) and PEN2 sequences separated by the P2A peptide (see "Exper-imental Procedures") on one side of the bidirectional promoter ( Fig. 1A) because PEN2 is destabilized by C-terminal tags that interfere with interactions with other ␥-secretase components (44), whereas nicastrin was successfully tagged at its C terminus without affecting interactions with other ␥-secretase components (11,(57)(58)(59). We subcloned the different combinations of APH1 (APH1aL, APH1aS, or APH1b) and presenilin (PS1 or PS2) sequences separated by the T2A sequence on the other side of the bidirectional promoter (Fig. 1A), because the C terminus of PS1 is required for ␥-secretase assembly and activity (20,60) and could be perturbed by a tag, whereas APH1 was C-terminally tagged without noticeable effect on APH1 interaction with other ␥-secretase subunits or ␥-secretase assembly (7, 61). Overall, the combination of the bidirectional promoter A, schematics of the plasmid constructs generated to express six different combinations of the four ␥-secretase subunits varying by their presenilin (1 or 2) and APH1 (aL, aS, or b) subunits. The pTRE3G-BI promoter (gray bidirectional arrow) allows for simultaneous, equivalent, and inducible expression of the two transgenes cloned beside it. The 2A peptide sequences (T2A and P2A, white boxes) cloned between two genes, allow for equivalent production of two discrete proteins through a "cleavage" event within the 2A peptide sequence (43). Nicastrin was C-terminally tagged with EGFP separated by a 10-amino acid flexible linker. These constructs are referred to as follows (from top to bottom): 4␥-PS1/APH1aL, 4␥-PS1/APH1aS, 4␥-PS1/APH1b, 4␥-PS2/APH1aL, 4␥-PS2/APH1aS, and 4␥-PS2/APH1b. B, schematics of the control plasmid constructs generated to co-express only three ␥-secretase subunits together. These constructs are referred to as follows (the missing subunit is indicated in parentheses): 3␥-(ϪAPH1), 3␥-(ϪPS1), and 3␥-(ϪPEN2). C, schematics of the ␥-secretase subunits synthesized from 4␥-PS/APH1 constructs indicating the positions of 2A and EGFP tags. The proline residue remaining after the 2A-peptide "cleavage," at the N terminus of the ␥-secretase subunits placed downstream of 2A sequences, is indicated (P).

␥-Secretase Subcellular Targeting Relies on Presenilins
and 2A peptide technology should achieve the synthesis of discrete ␥-secretase subunits, including T2A-tagged APH1 and EGFP-P2A-tagged nicastrin (Fig. 1C). For better clarity, we will refer hereafter to these constructs as "4␥-PS/APH1," because they allow the co-expression of the four ␥-secretase subunits (4␥) and to indicate the six specific combinations of the variable subunits PS and APH1, namely 4␥-PS1/APH1aL, 4␥-PS1/ APH1aS, 4␥-PS1/APH1b, 4␥-PS2/APH1aL, 4␥-PS2/APH1aS, or 4␥-PS2/APH1b. We decided to tag nicastrin with EGFP as a reporter of ␥-secretase localization, because it is a component common to all ␥-secretase complexes. Our previous experiments using BiFC constructs showed us that no signal from fluorescently tagged ␥-secretase subunits can be detected in post-ER compartments without the co-transfection of the four ␥-secretase subunits and the formation of new ␥-secretase complexes (36). We therefore assumed that with the above-described new constructs, the fluorescence derived from EGFP-tagged nicastrin in post-ER compartments would genuinely reflect ␥-secretase localization and not only nicastrin localization. However, to confirm this postulate, we generated control constructs in which only three ␥-secretase subunits will be co-expressed (Fig.  1B) that we will refer to as "3␥" followed by the missing subunit: 3␥-(ϪAPH1), 3␥-(ϪPS), or 3␥-(ϪPEN2). Overall, our new vectors are the first constructs permitting inducible and temporary stoichiometric expressions of all ␥-secretase subunit combinations and should allow the visualization of specific ␥-secretase complexes in the cell to assess the putative differences in their subcellular localization and trafficking.
The Co-induced Temporary Expression of the Four ␥-Secretase Subunits Is Required and Sufficient for the Maturation of Newly Formed ␥-Secretase Complexes and Their Detection Outside the ER-To validate our strategy to visualize mature ␥-secretase complexes in the cell only through the detection of NCT-EGFP, we first compared the expression of ␥-secretase subunits in COS-7 cells transfected with 4␥-PS1/APH1aL or its control constructs ( Fig. 2A). In 4␥-PS1/APH1aL-transfected cells, we observed an increased intensity of the bands corresponding to PS1-NTF or PEN2 as compared with empty vectortransfected cells and the appearance of new bands for NCT-EGFP-P2A and APH1-T2A at their expected molecular weights, slightly over endogenous subunits, showing that the 2A peptide "cleavages" occurred efficiently and that our 4␥-PS1/APH1aL construct achieved the expression of discrete ␥-secretase subunits. Using an anti-GFP antibody to detect transgenic NCT-EGFP, we observed a band doublet only for the cells transfected with 4␥-PS1/APH1aL, whereas cells transfected with all control "3␥" constructs presented only the lower band with a higher intensity. The upper band of the doublet appeared thicker and more intense than the lower one and probably corresponded to mature glycosylated NCT-EGFP. Indeed, nicastrin complex glycosylation requires its exit from the ER and trafficking through the Golgi apparatus, and ␥-secretase complex assembly is necessary for ER exit (22). In 4␥-PS1/APH1aL-transfected cells, full-length PS1 (PS1 FL) was still detectable (Fig. 2A); however, a clear increase of endoproteolyzed PS1 N-terminal fragment (PS1 NTF), as compared with empty vector-transfected cells, was also observed. This , or 3␥-(ϪPEN2) and the transactivator-protein expression plasmid. Transgene expression was induced with doxycycline for 5 h the next day, and cells were lysed 20 h after induction and analyzed by Western blot using specific antibodies for each ␥-secretase subunit or an anti-GFP antibody to discriminate NCT-EGFP from endogenous NCT. B, COS-7 cells were co-transfected with the constructs indicated on the left and the transactivator-protein expression plasmid; transgene expression was induced with doxycycline for 2 h the next day, and cells were fixed 22 h after induction, processed for immunocytochemistry to label the ER marker calregulin, and analyzed using wide field epifluorescence microscopy. Scale bar, 10 m.
We next analyzed NCT-EGFP fluorescence in COS-7 cells transfected with 4␥-PS1/APH1aL or its control "3␥" constructs, for which transgene expression was transiently induced 22 h before fixation and microscope observation to allow newly synthesized ␥-secretase complex maturation and trafficking to their cellular destinations. Cells transfected with "3␥" constructs displayed an EGFP fluorescence restricted to the ER, particularly intense around the nucleus, overlapping with the ER marker calregulin (Fig. 2B). Conversely, in 4␥-PS1/ APH1aL-transfected cells, NCT-EGFP fluorescence did not overlap with calregulin staining. In particular, the nuclear envelope, which is contiguous to the ER and clearly visible in control cells, was barely detectable. In these cells, NCT-EGFP fluorescence was mainly detected at the plasma membrane, highlighting the cell edges. These results, together with the observation of a combined NCT-EGFP glycosylation and PS1 endoproteolysis only in 4␥-PS1/APH1aL-transfected cells ( Fig. 2A), demonstrate that the NCT-EGFP detected at the plasma membrane was genuinely associated with mature ␥-secretase complexes and truly reflected the localization of newly assembled ␥-secretase complexes (i.e. containing PS1 and APH1aL).
␥-Secretase Subunit Compositions Drive Distinct Levels of Maturation into Functional ␥-Secretase Complexes-We next analyzed the expression of ␥-secretase subunits in membraneenriched fractions prepared from COS-7 cells transfected with the different 4␥-PS/APH1 constructs (Fig. 3A). We detected a band doublet for NCT-EGFP, presenting a more intense mature upper band except for the cells transfected with 4␥-PS1/APH1b, for which the immature band appeared more prominent. This observation reflects a slower maturation of PS1-and APH1b-containing ␥-secretase complexes and could indicate a higher degree of ER retention/retrieval of these ␥-secretase complexes. This difference was clearly linked to the nature of the PS because the NCT-EGFP mature band was similarly prevalent in all 4␥-PS2/APH1-transfected cells.
Full-length PS1 or PS2 was still detectable 20 h after the end of induction, in all 4␥-PS/APH1-transfected cells, concomitant with a clear increase in PS1 NTF or PS2 CTF immunoreactivities (as compared with empty vector control or non-induced cells (Fig. 3A, lanes 1 and 2)), indicating the assembly and maturation of new ␥-secretase complexes in 4␥-PS/APH1-transfected cells. However, despite comparable levels of expression of PEN2 and APH1-T2A between all of the 4␥-PS/APH1-transfected cells, obvious differences were observed in the PS1 NTF/  4 -8), and the transactivator protein expression plasmid; transgene expression was induced or not with doxycycline (Dox.) for 5 h the next day, and cells were processed to recover a membrane-enriched fraction 1 day after induction and analyzed by Western blot using specific antibodies for each ␥-secretase subunit or an anti-GFP antibody to detect NCT-EGFP and an anti-2A peptide antibody to detect APH1-T2A. B, membrane-enriched fractions analyzed in A were used for a ␥-secretase in vitro activity assay using an APP-C100-FLAG substrate. The production of AICD-FLAG and A␤ was analyzed by Western blot using a combination of anti-FLAG and anti-A␤ antibodies. In lane 9, the ␥-secretase inhibitor N-[N-(3,5-difluorophenacetyl)-L-alanyl]-S-phenylglycine t-butyl ester was added to the reaction using the membrane fraction corresponding to lane 3 (4␥-PS1/ APH1aL). C and D, quantification of AICD-FLAG (C) and A␤ (D) production in ␥-secretase in vitro assays. A.U., arbitrary units. Results are presented as mean Ϯ S.E. (error bars) (n ϭ 4). Statistical analysis using repeated measures analysis of variance revealed that means for AICD-FLAG and A␤ were significantly different (p Ͻ 0.01); Newman-Keuls multiple-comparison tests revealed significant differences between cells transfected with empty vector control and the ones transfected with PS1-containing vectors (*, p Ͻ 0.05; **, p Ͻ 0.01). The full results of Newman-Keuls pairwise comparisons are presented in Table 1.

␥-Secretase Subcellular Targeting Relies on Presenilins
PS1 FL and PS2 CTF/PS2 FL ratios, indicating the following. First, PS endoproteolysis was more efficient for PS2-containing ␥-secretase complexes than for those harboring PS1; second, the extent of PS1 endoproteolysis varied according to the nature of the APH1 isoform, with APH1b leading to a lower extent of maturation. This slower PS1 endoproteolysis of PS1/ APH1b-containing ␥-secretase complexes agreed well with their lower degree of nicastrin glycosylation.
To assess the functionality of newly formed ␥-secretase complexes, we used the membrane fractions characterized above to perform ␥-secretase in vitro activity assays (49), and we analyzed the production of AICD and A␤ (Fig. 3B). We did not detect any transfected subunit expression in non-induced cells. However, we chose to use empty vector-transfected cells as a control, because they went through the exact same treatments. AICD production was significantly increased in COS-7 cells transfected with PS1-containing constructs ( This significant increase of ␥-secretase in vitro activity over endogenous activity observed in COS-7 cells transfected with PS1-containing constructs demonstrates that the GFP-P2A tag attached to transgenic nicastrin or the T2A tag attached to transgenic APH1s did not alter ␥-secretase activity of the new ␥-secretase complexes made from 4␥-PS1/APH1 constructs. These results are in agreement with the original results showing that the co-expression of the four core components of ␥-secretase is sufficient to increase ␥-secretase in vitro activity in mammalian cells (7,8,62). The absence of significant increases in AICD and A␤ productions in the cells transfected with 4␥-PS2/ APH1 constructs is in accordance with the reported lower in vitro activity toward APP-based substrates of PS2-containing ␥-secretase complexes as compared with PS1-containing ones (59,63,64).

Assembly and Distinct Maturation of the Four ␥-Secretase Subunits into High Molecular Weight ␥-Secretase Complexes-
The concomitant glycosylation of NCT-EGFP and increased endoproteolysis of presenilins observed only when the four ␥-secretase subunits were co-expressed, as well as the increased ␥-secretase activity measured with the transfection of 4␥-PS1/ APH1 constructs, suggested the formation of new ␥-secretase complexes but did not prove directly the interaction of newly synthesized subunits into a complex. To confirm this hypothesis, we performed co-immunoprecipitation experiments, using an anti-GFP antibody to immunoprecipitate NCT-EGFP (Fig.  4, A and B). APH1-T2A, PEN2, and full-length and endoproteolyzed fragments of PS1 (Fig. 4A) or PS2 (Fig. 4B) were coimmunoprecipitated with both mature and immature NCT-EGFP. These observations demonstrate the interaction and assembly of transfected ␥-secretase subunits into genuine de novo ␥-secretase complexes. This also confirmed the presence of both mature ␥-secretase complexes (containing fully glycosylated NCT-EGFP and endoproteolyzed PS) and immature complexes (made of immature NCT-EGFP and full-length PS) in transiently induced cells, the latter ones being more abundant in 4␥-PS1/APH1-transfected cells.
Next, to determine whether ␥-secretase subunits expressed from 4␥-PS/APH1 constructs assembled into high molecular weight complexes and, thus, behave as their endogenous counterparts, we solubilized native ␥-secretase complexes from membrane-enriched fractions in 1% digitonin buffer and performed blue native PAGE followed by Western blotting using an anti-GFP antibody to identify the protein complexes where NCT-EGFP was incorporated (Fig. 4C). A major immunoreactive band was observed above the 480 kDa marker (high molecular mass band) for all cells transfected with the different 4␥-PS/APH1 constructs and appeared to be the only band for PS2-containing ␥-secretase complexes. Considering the additional mass brought by EGFP and 2A peptide tags (ϳ30 kDa), this major band agrees well with the expected mass described for endogenous ␥-secretase complexes (ϳ440 kDa) by several groups using this technique and 1% digitonin solubilization (42,65). Additional bands were detected for PS1-containing ␥-secretase complexes. The major one appeared just below the 480 kDa marker (lower molecular mass band) and was as intense as the high molecular mass band for APH1b-containing complexes but slightly less intense for APH1aL-or APH1aSand PS1-containing ␥-secretases. The presence and the intensity of this lower molecular mass band clearly correlate with the lower levels of PS endoproteolysis and NCT-EGFP maturation specifically observed with PS1-containing complexes (Fig. 3A). Importantly, we did not detect any signal in the vicinity or below the 146 kDa marker, showing that 24 h after the end of the induction, all of the NCT-EGFP that was synthesized was either incorporated into higher molecular weight complexes or degraded. This observation further strengthens our interpretation that mature and immature ␥-secretase complexes co-exist at this time point and that even the immature NCT-EGFP is probably incorporated in partially assembled or immature ␥-secretase complexes.
Distinct ␥-Secretase Subunit Combinations Differently Affect A␤ Peptides Secretion-In ␥-secretase in vitro assays, ␥-secretase complexes and a recombinant APP substrate are brought together in a reconstituted membrane environment. Therefore, a specific ␥-secretase complex with a given PS/APH1 combination could have processed an APP substrate that it does not physiologically encounter within the cell because of distinct cellular targeting of this particular complex and APP ␤-CTF.
To determine whether ␥-secretase complexes assembled from our 4␥-PS/APH1 constructs can differently contribute to the processing of ␤-CTFs in a more integrated system, we analyzed A␤40 and A␤42 cellular secretions from HEK293 cells stably expressing APP WT and BACE1 (Fig. 5). Analysis of transgenic ␥-secretase subunits expression in HEK293 APP WT /BACE1 stable cells (Fig. 5A) confirmed the maturation differences described for COS-7 cells (i.e. a lower degree of glycosylation of

␥-Secretase Subcellular Targeting Relies on Presenilins
NCT-EGFP and reduced PS1 endoproteolysis in 4␥-PS1/ APH1b-transfected cells and, more broadly, a lower level of PS endoproteolysis in cells transfected with PS1versus PS2-containing constructs). Overall, our data indicate that the maturation/endoproteolysis differences observed in 4␥-PS/APH1transfected cells were not associated with cell-specific effects but genuinely linked to the nature of the heteromeric complexes.
A␤40 secretion was significantly increased in HEK293 APP WT /BACE1 cells transfected with any 4␥-PS1/APH1 construct (Fig. 5B), whereas it appeared similar to control in all 4␥-PS2/APH1-transfected cells. These results support our in vitro observations in COS-7 cells of a lower activity of PS2containing ␥-secretase complexes as compared with PS1-containing complexes.
Thus, despite an endogenous set of ␥-secretase activities in HEK293 APP WT /BACE1 stable cells, we were able to delineate striking APH1 component-related differences in the A␤42/ A␤40 ratios. In particular, cells transfected with APH1b-containing constructs presented significantly higher A␤42/A␤40 ratios than those transfected with APH1aL-containing con- . ␥-Secretase subunit interaction and assembly into high molecular weight complexes. A and B, COS-7 cells were co-transfected with empty vector, or 4␥-PS1/APH1 constructs (A), or 4␥-PS2/APH1 constructs (B) and the transactivator-protein expression plasmid; transgene expression was induced with doxycycline for 5 h the next day, and cells were lysed in 1% CHAPSO buffer 1 day after induction and processed for immunoprecipitation using an anti-GFP antibody (GFP IP). The immunoprecipitation of NCT-EGFP and the co-immunoprecipitation of the ␥-secretase components were analyzed by Western blot using specific antibodies for each ␥-secretase subunit or an anti-GFP antibody to detect NCT-EGFP and an anti-2A peptide antibody to detect APH1-T2A. *, immunoglobulin heavy chain band. C, COS-7 cells were co-transfected and induced as in A and B and were processed to recover a membrane-enriched fraction 1 day after induction, followed by solubilization in a 1% digitonin buffer. Solubilized proteins were separated by blue native PAGE followed by Western blot using an anti-GFP antibody to detect native protein complexes where NCT-EGFP was incorporated. HMW, high molecular weight. structs ( Fig. 5D and Table 1). These observations are in full accordance with those of Acx et al. (59), who showed, in a ␥-secretase null background, that APH1b-containing ␥-secretase complexes favored the production of longer A␤ peptides as compared with their APH1aL-containing counterparts. These results therefore further identify our 4␥-PS/APH1 constructs as appropriate tools to yield functional ␥-secretase complexes harboring their proper enzymatic activities and, thus, to examine their cellular trafficking.
We also tested our 4␥-PS/APH1 constructs in HEK293 cells stably expressing m⌬ENotch (66). Western blot analysis did not reveal alteration in the processing of this substrate NICD (Notch intracellular domain)/m⌬ENotch ratio (supplemental Fig. S1). However, we cannot exclude the possibility that subtle changes may not be detected due to the lower sensitivity of this approach (as compared with the ELISA measurements of A␤ peptide secretion) and to the fact that these cells display endogenous ␥-secretase.
PS1-and PS2-containing ␥-Secretase Complexes Are Targeted to Distinct Subcellular Compartments-Are ␥-secretase complexes of different compositions targeted to distinct cellular compartments? To address this question, we first transfected COS-7 cells with the various 4␥-PS/APH1 constructs and transiently induced ␥-secretase expression 1 day before microscope observation. In all 4␥-PS1/APH1-transfected cells, NCT-EGFP fluorescence was mainly detected at the plasma membrane, thereby delineating the contours of the cells (Fig.  6A). Whereas some cells presented only this cell surface fluorescence (such as the one presented in Fig. 2B), other cells still displayed some fluorescence probably associated with the ER, were co-transfected with empty vector or with the indicated 4␥-PS/APH1 constructs and the transactivator protein expression plasmid; 7 h after transfection, transgene expression was induced or not with doxycycline (Dox.) for 16 h; 8 h after induction, cells were placed for 16 h in secretion medium, which was collected prior to cell lysis. A, cell lysates were analyzed by Western blot using specific antibodies for each ␥-secretase subunit or an anti-GFP antibody to detect NCT-EGFP and an anti-2A peptide antibody to detect APH1-T2A. B-D, secretion media were analyzed by ELISA to measure secreted A␤40 (B) and secreted A␤42 (C), and the A␤42/A␤40 ratio was then calculated (D). Results are presented as mean Ϯ S.E. (error bars) (n ϭ 3). Statistical analysis using repeated measures analysis of variance revealed that means for secreted A␤40, secreted A␤42, and the A␤42/A␤40 ratio were significantly different (p ϭ 0.0015, 0.0002, and 0.0069, respectively), Newman-Keuls multiple comparison tests reveal significant differences between control cells transfected with empty vector and those transfected with 4␥-PS1/APH1 constructs where indicated (*, p Ͻ 0.05; **, p Ͻ 0.01). The full results of Newman-Keuls pairwise comparisons are presented in Table 1.  Fig. 3 (C and D) and Fig. 5 (B-D) For clarity, empty vector-transfected cells are designated as "v", and 4␥-PS/APH1transfected cells are identified with presenilin number and APH1 variant letters (e.g. 1aL represents 4␥-PS1/APH1aL). NS, not significant (p Ͼ 0.05).

␥-Secretase Subcellular Targeting Relies on Presenilins
in continuity with the nuclear envelope, especially in 4␥-PS1/ APH1b-transfected cells. Strikingly, in 4␥-PS2/APH1-expressing cells, NCT-EGFP fluorescence was concentrated in intracellular organelles, presenting mostly a vesicular aspect but also sometimes a tubular appearance (Fig. 6A). Two main patterns of distribution were observed for all 4␥-PS2/APH1-transfected cells: 1) several large vesicular structures near one side of the nucleus and many other smaller vesicles scattered throughout the cell (Fig. 6A, bottom left and middle panels) or 2) one large and bright structure near the nucleus and numerous small vesicles scattered throughout the cell (Fig. 6A, bottom right). Occasionally, faint fluorescence was also detected at the cell surface,

␥-Secretase Subcellular Targeting Relies on Presenilins
suggesting that PS2-containing ␥-secretase complexes may transit through the plasma membrane before reaching the intracellular structures where they accumulate. In addition, no ER-associated NCT-EGFP fluorescence was visible around the cell nucleus in the cells transfected with PS2-containing constructs. When observed by time lapse video microscopy, these intracellular structures appeared to display different dynamics, depending on their pattern of distribution (described above; supplemental Movie 1), with some cells presenting more tubules able to bend, split, and turn within the cell (supplemental Movie 2). These strikingly distinct repartitions of NCT-EGFP fluorescence in 4␥-PS1/APH1-and 4␥-PS2/APH1-transfected cells confirmed that the observed fluorescence indeed came from NCT-EGFP incorporated into newly assembled ␥-secretase complexes and did not correspond to free, unassembled NCT-EGFP. According to the likely contribution of ␥-secretase in Alzheimer disease, we assessed the relevance of our observation by examining the distribution of PS1-and PS2-containing ␥-secretase complexes in primary cultured hippocampal neurons. A homogeneous cell surface-associated fluorescence was observed in the soma, axons, and dendrites of neurons transfected with all 4␥-PS1/APH1 constructs (Fig. 6, B-D, respectively). In addition, these neurons presented NCT-EGFP fluorescence around the nucleus (Fig. 6B), reminiscent of the ER-associated fluorescence observed in COS-7 cells. In neurons transfected with PS2-containing constructs, NCT-EGFP fluorescence was concentrated in intracellular vesicular structures throughout the soma, axons, and dendrites (Fig. 6, B-D). Therefore, the PS-related distinct distributions observed in COS-7 cells fully stand in cultured neurons, indicating that they could not be accounted for by cell-specific effects and appear relevant with regard to AD brain pathology.

Discussion
In the present study, we generate new constructs to transiently induce the stoichiometric co-expression of the four core components of ␥-secretase complexes with specific combinations of their variable subunits presenilin and APH1 and to visualize their cellular trafficking through the EGFP-tagged nicastrin incorporated in newly formed complexes. Interestingly, the biochemical characterization of these tools reveals different composition-dependent rates of maturation of the complexes into functional ␥-secretase. Strikingly, PS1-containing complexes are mainly targeted to the plasma membrane, whereas PS2-containing complexes are addressed to the TGN and a subset of endosomal compartments and are particularly enriched in recycling endosomes. Additionally, cellular distribution of PS2-containing complexes differs according to the APH1 component. Thus, PS2/APH1aS-or PS2/APH1b-containing complexes are also observed in late endosomes and lysosomes in which PS2/APH1aL-containing complexes were never detected.
Differential Maturation Rates of ␥-Secretase Complexes Depending on Their Composition-One main advantage of our constructs is that they allow us to transiently synthesize ␥-secretase subunits through doxycycline induction in a few hours, thereby allowing observation of their expression at a given time after the end of induction. We mostly analyzed ␥-secretase subunit expression between 20 and 22 h after the end of induction to permit their assembly, maturation, and trafficking outside the ER where they are synthesized. Using these settings, and despite the optimized design of our constructs for the stoichiometric expression of the four ␥-secretase subunits, we still detect full-length PS1 and PS2, showing that a fraction of transfected presenilins was not endoproteolyzed. Surprisingly, this fraction is more important for PS1-containing ␥-secretase complexes than for those containing PS2. This observation suggests a less efficient endoproteolysis of PS1 versus PS2 or a slower assembly of PS1-containing ␥-secretase complexes. This latter explanation is corroborated by the detection of smaller intermediate complexes in 4␥-PS1/APH1transfected cells in addition to the main high molecular mass complexes (Ͼ480 kDa) that are the only ones observed in 4␥-PS2/APH1-transfected cells. This interpretation is further strengthened by the observation of a residual ER-associated NCT-EGFP fluorescence in the cells transfected with PS1-containing constructs that may correspond to partially assembled complexes and that is absent in the cells transfected with PS2containing constructs.
The levels of full-length PS1 are even higher in cells transfected with 4␥-PS1/APH1b that also presented a lower degree of NCT-EGFP glycosylation. This result suggests a higher degree of ER retention/retrieval of PS1-and APH1b-containing ␥-secretase complexes, and accordingly, 4␥-PS1/APH1b-transfected cells present a more intense ER-associated NCT-EGFP fluorescence. These results corroborate our previous findings,

␥-Secretase Subcellular Targeting Relies on Presenilins
obtained with different constructs driving a continuous expression of BiFC-tagged ␥-secretase subunits, showing a higher ER retention/retrieval of ␥-secretase complexes containing PS1 and APH1b than those containing PS1 and APH1aL or APH1aS (36). Interestingly, this lower level of nicastrin glycosylation of PS1/APH1b-containing ␥-secretase complexes is also observed, but is not specifically discussed, by Acx et al. (59), who rescued PS/APH1s knock-out mouse embryonic fibroblasts with stable expression of specific combinations of human PS1 or PS2 and human APH1aL or APH1b. It remains to be determined whether PS1-and APH1b-containing ␥-secretase complexes serve a particular function within the ER or whether additional signaling/stimuli can trigger/facilitate the ER exit of these complexes.
Differential Subcellular Targeting of ␥-Secretase Complexes Depending on Their Presenilin Subunit-We demonstrate for the first time a striking difference in the subcellular targeting of PS1-containing ␥-secretase complexes that are enriched in the plasma membrane and PS2-containing complexes that are addressed to the TGN and a subset of endosomes. This difference has obvious implications for the substrate specificity of PS1-or PS2-containing ␥-secretase complexes. Indeed, the were co-transfected with the indicated 4␥-PS2/APH1 constructs and the transactivator protein expression plasmid; transgene expression was induced with doxycycline for 2 h the next day, and cells were fixed 1 day after induction and processed for immunocytochemistry using antibodies against the trans-Golgi network marker TGN46 (A), the recycling endosome marker transferrin receptor (TfR (B)), or the late endosome marker CD63 (C). Alternatively, cells were labeled with Lysotracker Red prior to fixation (D). Fluorescent signals were acquired using laser confocal microscopy. Insets, magnifications of the boxed areas. Scale bars, 5 m.

␥-Secretase Subcellular Targeting Relies on Presenilins
regulated intramembrane proteolysis of ␥-secretase substrates requires a rate-limiting cleavage resulting in the shedding of their ectodomain (4). The cellular site for this precleavage and the events immediately following it (such as the endocytosis of the cleaved substrate) probably influence the susceptibility of the substrate to further processing by PS1-or PS2-containing ␥-secretase complexes. Similarly, the location of the ␥-secretase cleavage, at the plasma membrane or in endosomal compartments, may influence the signaling efficiency of the generated intracellular domain. For example, the nuclear translocation and signaling of AICD require endocytosis (68,69).
The detection of PS1-containing ␥-secretase complexes at the plasma membrane is in agreement with our previous results obtained with BiFC-tagged ␥-secretase subunits (36) as well as the results obtained by Kaether et al. (29) with HEK293 cells stably expressing EGFP-tagged PS1. It also agrees well with cell surface biotinylation data showing the presence of PS1 at the plasma membrane in different cell systems (28, 29, 70 -74).
The subcellular localization of PS2 is much more poorly documented. We describe here the targeting of PS2-containing ␥-secretase complexes to the TGN and a subset of endosomes with a particular enrichment in transferrin receptor-positive recycling endosomes, whatever the nature of their APH1 subunit. The presence of active PS2-containing ␥-secretase complexes in the TGN is corroborated by the comparable increased production of A␤42 in cells co-expressing mutant PS2 and an APP-C100 substrate tagged or not with a TGN-targeting sequence (75). Our live cell imaging experiments also reveal the presence of PS2-containing ␥-secretase complexes in tubular structures that are characteristics of the recycling endosomes and are often lost during paraformaldehyde fixation. Supporting this localization, the hydrophilic loop of presenilins was shown to interact with Rab11, a GTPase associated with recycling endosomes, that is efficiently co-immunoprecipitated with PS2 in COS cells (76,77). Interestingly, transferrin recycling is delayed by the pharmacological inhibition of ␥-secretase or in PS1-and PS2-deficient embryonic stem cells, suggesting that presenilins may regulate recycling endosome function (70). In addition, we detected some PS2-containing ␥-secretase complexes in organelles labeled with late endosome/lysosome markers when they also contain APH1aS or APH1b but not APH1aL, showing that the nature of the APH1 subunit can modulate the endocytic itineraries of PS2-containing ␥-secretase complexes. In agreement with an increased abundance of PS2-containing ␥-secretase complexes in intracellular compartments as compared with PS1-containing ones, higher intracellular accumulation of A␤42 was observed in cells expressing familial AD mutant PS2 rather than mutant PS1 (78).
The localizations of PS1-and PS2-containing ␥-secretase complexes revealed by our 4␥-PS/APH1 constructs seem in apparent contradiction with the results we obtained on APP processing, considering current knowledge on the subcellular sites for A␤ production. Indeed, many studies show that A␤ is generated mainly in the TGN and endosomes, where the ␤-secretase BACE1 is also found (79,80). We observed an increased secretion of A␤ peptides only with the expression of PS1-containing ␥-secretases (Fig. 5) that we found to be present at the plasma membrane (Fig. 6) and no significant variation of this secretion with the expression of PS2-containing ␥-secretases that are actually targeted to the TGN and endosomes (Fig.  7). We cannot exclude the possibility that a portion of the total A␤ peptides generated were missed in our experiments, because we measured secreted A␤ but did not assess intracellular A␤. However, during these experiments, a substantial part of the content of the TGN and endosomes should have been released in the extracellular medium through vesicular recycling. Furthermore, in vitro ␥-secretase assays (Fig. 3), in which access to the substrate is not restricted by subcellular localization, confirm previous findings by other groups of the lower efficiency of PS2-containing ␥-secretase complexes to process APP ␤-CTFs (59,63,64). This characteristic also probably contributes to the lower A␤ peptide secretion obtained with these complexes. Therefore, our results argue for the production of A␤ peptides at the plasma membrane by PS1-containing ␥-secretase complexes, in agreement with reports from other groups (28,32,81)who notably showed that the inhibition of endocytosis did not prevent ␥-secretase processing of ␤-CTFs and even increased A␤ secretion. Our results highlight an unexpected degree of complexity in the understanding of the ␥-secretase processing of a particular substrate; not only the subcellular site(s) where this substrate is produced (i.e. where ectodomain shedding occurs) and the cellular localization of a specific ␥-secretase complex should be considered, but also the relative affinity/efficiency of a specific ␥-secretase complex to process this particular substrate. It will be interesting to further explore the physiological relevance of the sequential processing of APP by ␤and ␥-secretases along the endocytic/recycling routes and whether APP metabolites can inform the cell of the intensity of these vesicular turnovers.
New Relevant Tools to Decipher the Details of ␥-Secretase Complex Trafficking-Our new constructs are designed for an inducible, temporary, and stoichiometric expression of specific combinations of the four ␥-secretase subunits. We demonstrate here that these constructs allow for the formation of functional ␥-secretase complexes displaying the major characteristics associated with their specific presenilin and APH1 content that were described recently (59) and their visualization in distinct cellular compartments. These tools will therefore be very useful to address the many remaining questions related to the trafficking of the different ␥-secretase complexes and their substrates in cell lines and in primary neuronal cultures. Our constructs can be easily modified or implemented to study specific aspects of ␥-secretase trafficking and to determine the molecular determinants of their differential targeting evidenced here. For example, it will be particularly interesting to study whether some mutations associated with familial forms of AD on PS1 or PS2 can affect the trafficking of the ␥-secretase complexes in which these mutated subunits are incorporated. Our constructs are also well suited for live imaging of ␥-secretase trafficking and will be useful to determine whether this trafficking can be altered by specific stimuli, notably in primary neuronal cultures. For instance, the activation of ␤ 2 -adrenergic receptors or ␦-opioid receptors triggers the targeting of ␥-secretase to late endosomes and lysosomes (82).
Finally, because multiple reports have demonstrated that neuronal activity modulates A␤ production and synaptic release (83)(84)(85), our tools will be helpful to determine the importance of ␥-secretase trafficking in these processes in future studies.
Author Contributions-X. M. designed, performed, and analyzed experiments. X. M. and F. C. discussed data and wrote the manuscript.