Fe65 Stimulates Proteolytic Liberation of the β-Amyloid Precursor Protein Intracellular Domain*

The β-amyloid precursor protein (APP)-binding protein Fe65 is involved in APP nuclear signaling and several steps in APP proteolytic processing. In this study, we show that Fe65 stimulates γ-secretase-mediated liberation of the APP intracellular domain (AICD). The mechanism of Fe65-mediated stimulation of AICD formation appears to be through enhanced production of the carboxyl-terminal fragment substrates of γ-secretase and direct stimulation of processing by γ-secretase. The stimulatory capacity of Fe65 is isoform-dependent, as the non-neuronal and a2 isoforms promote APP processing more effectively than the exon 9 inclusive neuronal form of Fe65. Intriguingly, Fe65 stimulation of AICD production appears to be inversely related to pathogenic β-amyloid production as the Fe65 isoforms profoundly stimulate AICD production and simultaneously decrease Aβ42 production. Despite the capacity of Fe65 to stimulate γ-secretase-mediated APP proteolysis, it does not rescue the loss of proteolytic function associated with the presenilin-1 familial Alzheimer disease mutations. These data suggest that Fe65 regulation of APP proteolysis may be integrally associated with its nuclear signaling function, as all antecedent proteolytic steps prior to release of Fe65 from the membrane are fostered by the APP-Fe65 interaction.

The ␤-amyloid precursor protein (APP)-binding protein Fe65 is involved in APP nuclear signaling and several steps in APP proteolytic processing. In this study, we show that Fe65 stimulates ␥-secretase-mediated liberation of the APP intracellular domain (AICD). The mechanism of Fe65-mediated stimulation of AICD formation appears to be through enhanced production of the carboxyl-terminal fragment substrates of ␥-secretase and direct stimulation of processing by ␥-secretase. The stimulatory capacity of Fe65 is isoform-dependent, as the non-neuronal and a2 isoforms promote APP processing more effectively than the exon 9 inclusive neuronal form of Fe65. Intriguingly, Fe65 stimulation of AICD production appears to be inversely related to pathogenic ␤-amyloid production as the Fe65 isoforms profoundly stimulate AICD production and simultaneously decrease A␤42 production. Despite the capacity of Fe65 to stimulate ␥-secretase-mediated APP proteolysis, it does not rescue the loss of proteolytic function associated with the presenilin-1 familial Alzheimer disease mutations. These data suggest that Fe65 regulation of APP proteolysis may be integrally associated with its nuclear signaling function, as all antecedent proteolytic steps prior to release of Fe65 from the membrane are fostered by the APP-Fe65 interaction.
The ␤-amyloid precursor protein (APP) 2 -binding protein Fe65 associates with the carboxyl-terminal region of APP, a protein integrally involved in Alzheimer disease (AD) pathogenesis (1). Several groups have demonstrated that Fe65 plays a key nuclear signaling function, in association with the histone acetyltransferase Tip60, following ␥-secretase-mediated cleavage of APP (2)(3)(4). Although the exact mechanism of APP/Fe65 nuclear signaling is not resolved (5,6), it remains clear that Fe65 is essential in APP-mediated transcriptional regulation. Trun-cation and mutagenic studies have identified the tyrosine residues in the 682 YENPTY 687 motif (numbered relative to APP695) within the intracellular domain of APP as the Fe65binding site (7,8). Three core domains within Fe65, two PTB domains and one WW domain, mediate the physical associations between APP and other Fe65 interacting factors (1). The carboxyl-terminal PTB domain (or PTB2) is the Fe65 domain that associates with the APP YENPTY motif (1,(7)(8)(9) in a phosphotyrosine-independent manner.
Previous work has demonstrated that Fe65 is involved in the regulation of APP processing. The intracellular interaction between APP and the carboxyl-terminal portion of the low density lipoprotein-related protein 1 (LRP) is mediated by Fe65 via its two PTB domains (10). The LRP interaction with APP, mediated by Fe65, may coordinate numerous aspects of APP proteolytic processing, including APP carboxyl-terminal fragment generation and ␤-amyloid secretion (11). APP is processed in a sequential fashion, in which either ␣or ␤-secretase proteolytic cleavage is a prerequisite for ␥-secretase-mediated proteolysis. Consequently, Fe65 regulation of APP proteolytic processing could play a fundamental role in numerous aspects of APP function. Previous work implicating Fe65 in modulation of ␤-amyloid secretion (10 -12) suggests that Fe65 may play a role in proteolytic events associated with AD pathogenesis.
Consistent with the connection between Fe65 and AD etiology, epidemiological studies have suggested that a specific isoform of Fe65 may be protective against late onset sporadic AD. A careful examination of Fe65 isoforms expressed in AD patients and unafflicted controls isolated an allele of the Fe65 gene in which a CTA element in intron 13 was deleted adjacent to the splice donor site (13). The deletion results in differential splicing of the nascent Fe65 transcript, leading to an alternative coding sequence at the carboxyl-terminal portion of the Fe65 protein (14). The alternative allele is referred to as the a2 allele and is reported to confer resistance to the late onset form of AD (13)(14)(15). Splicing events also generate a second form of variability in Fe65 transcripts. In neuronal cell types, the six nucleotide exon 9 is incorporated into Fe65 transcripts, resulting in the insertion of Arg-Glu into the first PTB domain (16). The role of the a2 and neuronal (exon 9 inclusive) Fe65 isoforms in regulating APP proteolytic processing is explored in this work, which demonstrates that all Fe65 isoforms promote ␥-secretase-mediated AICD release, but with the neuronal form of Fe65 being the least efficacious.

EXPERIMENTAL PROCEDURES
Cell Culture and Transfection Conditions-COS7 cells were grown in Dulbecco's modified Eagle's medium supplemented with 10% fetal bovine serum (Hyclone) and 1ϫ penicillin/streptomycin (Invitrogen) at 37°C in 10% CO 2 . Stock plates of COS7 cells were grown to 50 -80% confluence prior to splitting cells for transient transfection experiments. All transactivation assays were done using cells seeded onto 24-well plates. The plates were grown to ϳ90% confluence prior to transfection. All transfections employed Lipofectamine 2000 (Invitrogen), using 1.5 l of Lipofectamine 2000 per well and transfection medium supplemented with 10% fetal bovine serum lacking antibiotics. NIH 3T3 cells, HEK293 cells, and JEG3 cells were grown in 10% fetal bovine serum/Dulbecco's modified Eagle's medium and were transfected using the same methods as employed for COS7 cells. The transfection mix was incubated on the cells for 2-4 h prior to replacement of the transfection medium with fresh growth medium described above.
In all transactivation assays, 1.5-2 g of DNA was used per well. APPGV16 concentrations are defined in each experiment in which titration assays are done. If it is not otherwise specified, then 250 ng/well of APPGV16 DNA was used. In all transactivation assays in which Fe65 values are not stated, 1000 ng per well of Fe65 DNA were employed for an APP:Fe65 vector ratio of 1:4. In all experiments with Fe65, the vector preparations used for the transfection were measured and double checked by comparative gel electrophoresis. To ensure that none of the observed effects were because of undetectable variations in vector preparation quality, all experiments were replicated with different preparations of Fe65 vectors. In all experiments, the promoter concentration was held constant by the addition of either pcDNA3.1 (Invitrogen) to normalize cytomegalovirus promoter concentrations or pCEFL to normalize the amount of EF1␣ promoter present in each condition. In experiments in which APPGV16 was titrated across the experiment, pCEFL was counter-titrated to ensure that there would not be any promoter competition or variation because of promoter saturation across conditions. The pFRluc (Invitrogen) Gal4-luciferase reporter was transfected at 400 ng/well to measure the Gal4VP16-mediated transactivation levels. In experiments also overexpressing PS1 isoforms, 150 ng/well of the PS1 construct was used. Normalization was done using 50 ng/well of the EF2-␤Gal construct previously described (17). For experiments unaccompanied by transactivation assays, cells were grown in 6-well plates, and 1000 ng/well of the APPderived constructs and 4000 ng/well of the Fe65 vectors were used to maintain the 1:4 APP:Fe65 vector ratio. The 6-well plates were transfected using 7.5 l/well of Lipofectamine 2000. The transfection times and other conditions were identical. All experiments were assayed 24 -48 h following the transfection.
Plasmid Generation and Mutagenesis-All of the APP vectors use PCR amplification of the human APP695 cDNA fused in-frame to Gal4-VP16 driven by the elongation factor 1␣ promoter (EF1␣). The original APP-Gal4VP16 (APPGV16) vector generation has already been described (17), and all of the subsequent vectors used the same strategy. 5Ј primers were generated that contain a HindIII restriction site and a start codon in front of the coding portion of the cDNA representing the 5Ј end of C99, C83, or C50. The 3Ј primer used for all of the APP constructs was identical to that used to make APPGV16, in which a 5Ј XbaI site resides in front of the complementary sequence at the 3Ј end of APP695 with the stop codon deleted. Following PCR amplification of C99, C83, and C50, the PCR product was isolated, digested with HindIII and XbaI, and subcloned into the HindIII and XbaI sites of the EF1␣-APPGV16 vector from which the HindIII-XbaI APP coding cassette had been removed. This resulted in C99, C83, and C50 vectors driven by the identical portion of the EF1␣ promoter cloned in-frame with Gal4VP16.
The APPGV16 682 YENPTY 687 point mutants were generated by the method employed previously (17). Briefly, mutagenic primers were designed with changes in the coding sequence to alter either Tyr 682 or Tyr 687 to alanine and insert a silent restriction site using the standard QuikChange site-directed mutagenesis approach (Stratagene). The resulting APPGV16 Y682A and Y687A vectors were mapped and sequenced to verify the presence of the mutation. The Myc epitope-tagged Fe65 vectors were generously provided by Dr. Qubai Hu and Dr. George Martin. The generation of the pcDNA3.1-Fe65myc vectors has been reported previously (14,18), whereby the human neuronal, non-neuronal, and non-neuronal a2 sequence is cloned into pcDNA3.1/Myc-His(ϩ)B at the BamHI and NotI sites (Invitrogen). Untagged versions of these vectors were also generated by leaving the stop codon intact following the subcloning of the cDNA. The amino-terminally truncated Fe65 M 260 Fe65 M327L vector was also provided by the laboratory of Dr. George Martin, encoding the human neuronal form of Fe65 cDNA beginning at Met 260 and containing an M327L mutation cloned into pcDNA3.1. The set of Fe65 truncation/ deletion mutations was obtained from the laboratory of Dr. T. Sudhof and has been described previously (2). The BACE expression vector was obtained from Curagen Corp., in which the human BACE cDNA was PCR-amplified and subcloned into pcDNA3.1. The human presenilin 1 wild-type and FAD mutant expression vectors were described previously (17).
Cleavage Assay, Protein Preparation, and Inhibitor-Luciferase and ␤-galactosidase assays were performed by standard methods as reported previously (16). The APPGV16 protein is cleaved by ␥-secretase, liberating the APP AICD-Gal4VP16 moiety to transactivate the pFRluc Gal4-luciferase construct as described previously (17). The cells were lysed on ice 24 -48 h post-transfection, and luciferase assays were performed and point to point normalized to ␤-galactosidase values driven by the EF2-␤Gal vector. For the assays in which parallel protein concentration levels were assessed, see below for details on protein preparation. All measurements were performed on an EG&G Berthold LB 96V luminometer. The ␥-secretase inhibitor DAPT (Sigma) was used at 10 M, unless otherwise stated, and the cells were treated with inhibitor ϳ24 h post-transfection.
Western Blots, A␤ ELISA, and Antibodies-Western blots examining protein levels in parallel with transactivation assays used 30 l per well (90 l total) of lysates, placed into 180 l of 1ϫ RIPA with 1ϫ proteosome inhibitor mixture (Sigma) or a separate well from the same transfection experiment lysed in 1ϫ RIPA supplemented with 1ϫ proteosome inhibitor mixture. The lysates were rocked at 4°C for 10 min and spun down and prepared for loading. The samples were run on NOVEX 4 -12% NuPAGE gradient gels and probed using the rabbit anti-Gal4 DNA binding domain antibody (Calbiochem), the rabbit APP carboxyl-terminal antibody (Cell Signaling), mouse c-Myc antibody (BD Biosciences), and the FE518 Fe65 antibody (14). Western blots employing 6-well plates were transfected and assayed 48 h following transfection. The cells were lysed in 1ϫ RIPA with 1ϫ proteosome inhibitor mixture (Sigma), and prepared as described above. The A␤40 and A␤42 sandwich ELISA measurements were performed using the media drawn off of the transiently transfected cells, 48 h post-transfection. The ELISA procedure was performed as reported previously (17).

RESULTS
Fe65 Overexpression Stimulates Liberation of the APP Intracellular Domain-The pivotal role of Fe65 in mediating the interaction between APP and LRP (10, 11) led us to examine its role in the modulation of ␥-secretase-mediated proteolysis of APP. It is difficult to quantify the total extent of cleavage of APP by direct biochemical means, because the amino-terminal product of the cleavage is a heterogeneous mix of multiple A␤-related peptides, and the labile carboxyl-terminal product (AICD) does not accumulate to appreciable levels. Consequently, for many experiments, we employed a system in which cleavage of a carboxyl-terminally modified APP drives transcription of a luciferase reporter plasmid, as we have described previously (17). Preliminary experiments indicated that Fe65 promoted ␥-secretase-mediated proteolysis of APP. Consequently, this brought up the question of whether different splice variant products of Fe65 would differ in ability to promote proteolytic liberation of AICD. To assess the role of the neuronal specific exon 9/E9 Arg-Glu insertion into PTB1 domain, we employed Myc-tagged versions of the neuronal and non-neuronal forms of human Fe65 (described under "Experimental Procedures"). Because a second splice variation occurs that alters the carboxyl terminus of Fe65, via incorporation of an alternative exon 14, a Myc-tagged version of the non-neuronal expression vector incorporating the a2 allele was also implemented in this work. Throughout these experiments, a threeway comparison is made between human Fe65 neuronal (E9), Fe65 non-neuronal, and Fe65 non-neuronal a2 to compare the relative efficacy of the PTB1 insertion (neuronal versus nonneuronal) and the a2 allele (non-neuronal versus non-neuronal a2). These vectors were overexpressed in COS7 cells in concert with a titrated range of APPGV16 concentrations which, once cleaved, activates the Gal4-luciferase reporter. Luciferase expression is normalized to constitutively expressed EF2-␤Gal. This methodology has been described previously (17).
Consistent with previous experiments, we observed that increasing levels of APPGV16 resulted in a plateau in the transactivation of the Gal4-luciferase reporter. All three isoforms of Fe65 substantially increased the proteolytic liberation of AICD-GV16 from the membrane at statistically significant levels (Fig.  1A). However, the three Fe65 isoforms demonstrated significantly different stimulatory capacity, as the neuronal form was considerably less efficacious in promoting cleavage than either the non-neuronal or a2 isoform (Fig. 1A, upper two curves represent the non-neuronal and a2 isoforms). Although in some experiments the a2 allele promoted slightly greater AICD-GV16 liberation than the non-neuronal, this difference was not consistently observed across experiments.
To verify that Fe65 stimulation of APP proteolytic liberation was not because of variation in expression following transfection, a portion of the lysates generated for the transactivation assay was employed for immunoblot analysis. Consistently, APPGV16 expression increased across the vector titration. To obviate small differences that might appear within individual blots, the lysates were run and probed with two separate antibodies to interrogate APPGV16 and Fe65 expression levels (Fig.  1B). We did not observe any difference in APPGV16 expression levels within the different Fe65 isoform transfection conditions. Conversely, we were surprised to see that increasing levels of APPGV16 resulted in a consistent decrement in Fe65 protein COS7 cells were transiently transfected with titrated levels of the APPGV16 expression vector with or without vectors encoding the three Fe65 isoforms. The concentrations used in the titration were 0, 25, 100, and 500 ng of expression vector per transfected well. The Fe65 expression vectors were co-transfected at 1000 ng/well. A, the Gal4-luciferase data resulting from transactivation of the pFRluc reporter is plotted for all four conditions. All statistical analysis employed the GraphPad Prism software. Statistical analysis was performed using a two-way ANOVA to compare differences between conditions in which Fe65 is present and absent across the APPGV16 titration. APPGV16 expression alone compared with co-expression with Fe65 neuronal resulted in statistically significant difference at p Ͻ 0.0001. The 2-ANOVA analysis performed between Fe65 neuronal and Fe65 non-neuronal resulted in a significant difference at p Ͻ 0.0001. The two-way ANOVA between the Fe65 nonneuronal and a2 co-transfected titration curves demonstrated slight statistical significance at p Ͻ 0.05. However, this difference is attributable to a divergence exclusively at the 100 ng of APPGV16 concentration. Student's t test analysis shows a p ϭ 0.01 value for the 100-ng conditions. However, no other point in the titration is statistically different between Fe65 non-neuronal and a2 isoforms by point by point t test analysis. B, Western blots were performed using the lysates from the transactivation assay. The top two blots were probed for APPGV16 using APP carboxyl-terminal antibody and the Gal4 antibody. The bottom two blots were probed for Fe65. The first Fe65 blot used the Myc antibody targeting the Myc tag epitope on the carboxyl terminus of Fe65, whereas the second blot employed the FE518 Fe65 antibody. levels for all isoforms (Fig. 1B, bottom two blots). Although no strict quantitation was performed, it appears that APPGV16 may result in a greater decrement of the Fe65 neuronal form across the titration.
Fe65-mediated Stimulation Is ␥-Secretase-mediated-The assay system employed in this work detects release of the AICD-GV16 moiety from the membrane. In previous work, we have shown this to be ␥-secretase-dependent. However, the intracellular domain of APP contains a known caspase cleavage site (19 -22). To ensure that Fe65 was stimulating proteolytic release of AICD-GV16 through the ␥-secretase processing pathway, COS7 cells were transiently transfected with APPGV16 and the Fe65 isoforms in the presence or absence of the ␥-secretase inhibitor DAPT. Consistently, all Fe65 isoforms promote statistically significant increases in AICD-GV16 production, and the non-neuronal and a2 isoforms are more efficacious than the neuronal form (Fig. 2). The blockade of transactivation observed in the presence of DAPT confirms that Fe65 stimulation of AICD-GV16 is dependent on ␥-secretase, and, consequently, is not likely to be mediated by caspase cleavage. This result was confirmed by comparing effects of transfected Fe65 isoforms on cleavage of AICD-GV16 in wild-type primary mouse embryo fibroblasts and mouse embryo fibroblasts derived from PS1 Ϫ/Ϫ /PS2 Ϫ/Ϫ mice. Fe65 isoforms promoted APPGV16mediated transactivation exclusively in the wild-type cells (data not shown).

Fe65 Stimulation of APP Proteolysis Requires Direct Protein
Interaction-Given the demonstrated role of Fe65 in transcriptional regulation (2, 3, 6), we sought to determine whether Fe65 stimulation of ␥-secretase-mediated proteolysis was because of direct physical association or through an indirect mechanism. Fe65 association with APP is mediated by the interaction of the phosphotyrosine binding domain (PTB2) with the tyrosine residues within the YENPTY motif (7,9,23). Consequently, APP-Gal4VP16 mutants were made that contain alanine substitutions at either of the two tyrosine residues within the YENPTY motif. The resulting mutations APPGV16 Y682A and Y687A were transfected into COS7 cells either alone or in the presence of the Fe65 isoforms. The Fe65 isoforms stimulated cleavage of the wild-type APPGV16, yet all three isoforms failed to stimulate cleavage of APPGV16 Y682A or APPGV16 Y687A (Fig. 3). These data confirm that direct association between APP and Fe65 is necessary for Fe65 to promote ␥-secretase-mediated AICD release. The observation that both tyrosine mutations completely blocked transactivation was surprising, as each tyrosine residue has been demonstrated to contribute to the interaction with Fe65. Blockage of Fe65 stimulation of APP proteolysis by mutations to either Tyr residue in the YENPTY motif suggests that both residues are essential for the functionally effective association of Fe65. Comparing untreated samples demonstrated statistically significant differences between APPGV16 and Fe65 neuronal (p Ͻ 0.0001). The difference between Fe65 neuronal and Fe65 non-neuronal was also statistically significant (p Ͻ 0.01). However, there was no significant difference between Fe65 non-neuronal and the a2 isoform. The ␥-secretase dependence of Fe65 stimulation is observed in the consistent levels of repression with DAPT treatment. In all conditions, DAPT repressed between 80 and 90% of the Gal4-luciferase signal; the level of repression was slightly higher with co-expression of the Fe65 isoforms (APPGV16 alone, 80%; Fe65 co-expression, 88 -89%).

FIGURE 3. Fe65 stimulation of APPGV16 proteolysis is dependent upon direct association.
Site-directed mutagenesis of APPGV16 at the tyrosine residues within the YENPTY motif resulted in two alternative APPGV16 constructs, APPGV16 Y682A and APPGV16 Y687A. Wild-type APPGV16, APPGV16 Y682A, and APPGV16 Y687A were transiently transfected into COS7 cells in the presence or absence of the three Fe65 isoforms. All transfection conditions were assayed 48 h post-transfection. Gal4-luciferase values were normalized to constitutive ␤-galactosidase (␤Gal) expression. Fold stimulation with all three Fe65 isoforms was consistent with previous observations for wild-type APPGV16, ranging from 4-to 10-fold for the three isoforms. Student's t test analysis demonstrated that APPGV16 and Fe65 isoforms maintained statistically significant differences at p Ͻ 0.0002 or less. However, the APPGV16 Y682A and Y687A mutants failed to demonstrate statistically significant levels of difference even at the p Ͻ 0.01. With APPGV16 Y682A and Y687A the fold stimulation observed with any of the Fe65 isoforms never reached 2-fold.

Fe65 Stimulates AICD
Fe65 Stimulation of APP Proteolysis Is Dependent on PTB1-Based upon the above observation that direct physical interaction between APP and Fe65 is critical for Fe65 enhancement of ␥-secretase-mediated proteolysis, we sought to identify the critical Fe65 domains. Fe65 has three functional domains, an amino-terminal WW domain (8) and two carboxyl-terminal PTB domains (9). Consequently, it is possible that Fe65 stimulation of proteolysis occurs by assembling a heterotrimeric or heterotetrameric complex with APP. The Fe65 deletion mutants have been described previously (2). The domain structure of Fe65 and the amino acid positions flanking the three functional domains are diagrammed in Fig. 4A. The region deleted within each of the PTB mutants is diagrammed above the protein, as shown in Fig. 4A, and the position of the WW domain mutation is indicated with an asterisk.
Mapping the critical domains of Fe65 to stimulate APP proteolysis was accomplished by co-transfecting APPGV16 with the Fe65 variants in COS7 cells. The results were normalized as fold stimulation. Consistent with previous observations, fulllength Fe65 stimulated an approximate 8-fold increase in proteolysis. Deletion of any of the three core domains of Fe65 decreased its capacity to stimulate APP proteolysis at statistically significant levels (Fig. 4B). However, deletion of the PTB1 domain resulted in the greatest decrease. Indeed, this mutant appears to inhibit, not stimulate, cleavage of APPGV16. The difference between APPGV16 alone and the PTB1 deletion is statistically significant. Consequently, it appears that not only is PTB1 critical to Fe65 function, but lack of it results in a dominant-negative form of the protein with respect to enhancing APP proteolysis. This effect probably reflects interference with the activity of Fe65 natively expressed in COS7 cells. As the two PTB domains of Fe65 are involved in the previously observed linker functions of Fe65 (2, 10), Fe65 stimulation of APP proteolysis may require the assembly of multiprotein complexes. Intriguingly, the deletion of the amino-terminal portion of Fe65 (128 -711 variant) resulted in a dramatic and statistically significant increase in the proteolytic release of AICD-GV16 (Fig.  4B). Further truncation of the amino terminus of Fe65 to the beginning of the WW domain reduced the stimulatory effect to a level comparable with full-length Fe65. This prompted us to speculate that a region of Fe65 that can stimulate proteolytic events is normally masked by the amino-terminal portion of the protein, an event that potentially might be regulated in vivo. Similar effects were observed by Sudhof and co-workers (2) for signaling by APP-Gal4. Additional truncation of Fe65 past the WW domain (residues 287-711) resulted in a significant reduction in Fe65 stimulatory efficacy. However, it is interesting to note that truncation of Fe65 down to just the PTB1 domain (residues 287-531) reduces but does not entirely eliminate its ability to stimulate processing of APPGV16.
To ensure that the observed effects were not because of simple differences in level of protein expression, fractions of the lysates used in the transactivation assay were subjected to immunoblot analysis with antibodies against APP and Fe65 (Fig. 4B). The equivalency of plasmid concentrations in Fe65 vector preparations was verified by agarose gel electrophoresis prior to initiating the transfection. Modest differences in protein levels achieved were not sufficient to account for the observed differences in functional activity. The PTB2 mutant protein did appear to be present at a reduced level in the experiment shown. However, this mutant yielded similarly diminished activity in other experiments in which its level of expression was equivalent to wild-type Fe65. It is unclear whether the The ⌬PTB1 and ⌬PTB2 representations diagram the region of each domain deleted in those mutants. B, Gal4-luciferase data were normalized to fold stimulation using the average value of APPGV16 as the referent for each individual value within all conditions. Co-expression with all forms of Fe65 resulted in statistically significant differences from APPGV16 alone (p Ͻ 0.0001), using Student's t tests to perform the analysis. However, overexpression of ⌬PTB1 repressed AICD-GV16 production at significant levels. Mutations in either the WW domain or PTB2 resulted in significant differences from full-length Fe65 (p Ͻ 0.003 and p Ͻ 0.0008, respectively). apparently reduced expression of the PTB2 mutant in the experiment shown reflects poor efficiency of expression of the plasmid vector or differences in stability or post-translational processing of expressed protein. It should be noted that at longer exposures many of the Fe65 variants did have multiple bands not encountered with the full-length Fe65. Consequently, it is plausible that cleavage products associated with the variants may account for some of the observed differences. However, this is unlikely as the primary protein product of the Fe65 expression vectors was vastly more abundant than the cleavage products observed. Equivalency of Full-length Fe65 and Fe65 Derived from an Alternative Start Site-The Fe65 knock-out mouse generated by Martin and co-workers (24) was determined to be a hypomorphic mutant as a shorter Fe65 protein product was still expressed. The targeting strategy employed a deletion of exon 2 of the murine Fe65 gene that contains the start codon. Another start site that is in-frame and outside of the exon 2 targeted region occurred at position Met 260 . Consequently, we sought to test the functional efficacy of the protein resulting from the alternative start site relative to the full-length Fe65 protein. The truncation product associated with the alternative start site maps to the beginning portion of the WW domain. However, it is unclear whether this partial truncation is sufficient to destroy the interaction capacity of the WW domain as Met 260 resides after the amino-terminal tryptophan, but two tryptophan residues remain within the WW domain carboxyl terminus to the alternative start site. Interestingly, we observe no difference between effects of full-length Fe65 or M 260 Fe65 M327L on APPGV16 proteolysis (Fig. 5). This result is consistent with our antecedent domain mapping studies, as there was little difference between full-length Fe65 and the truncated 242-711 Fe65 (the truncation mutation that preserved the WW domain).
Isoform Difference Is Not Dependent on Protein Concentration-APPGV16 may result in a decrement in expression of all three Fe65 isoforms, in which the neuronal form of Fe65 is impacted most profoundly (Fig. 1). To assess whether this observed decrement in protein levels is responsible for the differential efficacy of the Fe65 isoforms in stimulating APP proteolysis, APPGV16 expression vector levels were held constant, and the expression vector for each isoform of Fe65 was titrated from 0 to a 10-fold excess relative to APPGV16. Fe65 protein concentrations, as assessed by immunoblot analysis, increased almost linearly with increasing quantity of transfected plasmid. However, the stimulation of APPGV16-dependent luciferase transcription (measured from the same experiment as the immunoblot analysis) plateaus near a 1:3 ratio of expression vectors (100 ng/well of APPGV16 DNA was used) (Fig. 6). The linear increase in protein levels of neuronal, non-neuronal, and a2 Fe65 variants across the titration appeared equivalent despite the difference in transactivation observed. This indicates that the differences in the ability of the Fe65 isoform to promote proteolytic cleavage of APP are not because of protein concentration effects.
Fe65 Stimulation of APP Proteolysis Is Consistent across Heterologous Cell Types-To assess whether Fe65 stimulation of APP proteolysis is a general phenomenon, titration experiments were performed in NIH 3T3, HEK293, and JEG3 cells.
The titration experiments were performed in which either Fe65 was incrementally introduced into cells expressing a fixed amount of APPGV16, or APPGV16 was titrated into cells expressing a fixed amount of each of the Fe65 isoforms. These experiments demonstrated that titrating in Fe65 consistently increased APPGV16 proteolysis in each cell type (supplemental Fig. S1, A, C, and E). In these experiments 100 ng of the APPGV16 expression vector was employed, corresponding to early plateau values of reporter activity in previous APPGV16 titration experiments (Fig. 1). Consistent with the observation in COS7 cells, the Fe65 neuronal form increased the level of transactivation the least, ϳ2-3-fold in all three cell types. Concordantly, the non-neuronal and the a2 isoform increased AICD release to a much greater extent, up to 8-fold in NIH 3T3 cells. The overall extent of stimulation did vary between cell types, with NIH 3T3 cells and HEK293 cells showing the highest levels of stimulation. Interestingly, the a2 isoform resulted in marginally higher levels of AICD release than the non-neuronal isoform in all three cell types. In the experiments employing APPGV16 titration with fixed levels of Fe65 vector employed in each transfection, the same pattern emerged with an attenuated difference between the non-neuronal and a2 isoforms (supplemental Fig. S1, B, D, and F). Again, there was a greater difference between isoforms observed in the NIH 3T3 and HEK293 cells than in JEG3 cells, suggesting some potential cell type specificity to the potency of the Fe65 isoforms. Yet, in The M 260 Fe65 M327L is a truncation mutation mapping to an alternative start site resident within Fe65. The additional M327L mutation was added as overexpression of the truncated version of Fe65 resulted in protein species initiating at the Met 327 site. To avoid the generation of biologically unobserved Fe65 protein species, Met 327 was mutated to examine exclusively the Fe65 variant starting at Met 260 . The alternative start site abolishes the first portion of the WW domain but preserves the Trp residue mutated in the previous experiment (Fig. 4). Both forms of Fe65 elicited statistically significant differences in proteolysis compared with APPGV16 alone at p Ͻ 0.0002 or less by Student's t test. However, the two co-expressed forms of Fe65 demonstrated no statistical difference in stimulating AICD production. Induced Gal4-luciferase values were normalized to constitutive ␤-galactosidase expression.
both sets of experiments it was observed that Fe65 promotes the proteolytic release of the AICD fragment, suggesting that the role of Fe65 in stimulating APP proteolysis is a cell type independent phenomenon.
Fe65 Stimulates Precursor Formation and ␥-Secretase-mediated APP Proteolysis-Previous work has suggested that Fe65 may positively regulate both APPs␣ secretion as well as A␤ production (12,25). Furthermore, LRP association with APP, mediated by Fe65, regulates proteolytic processing and AICD generation (10,11). This suggests that Fe65 could stimulate ␥-secretase-mediated liberation of AICD-GV16 by promoting processing of APP by ␣and/or ␤-secretase, thereby increasing the concentration of substrate for ␥-secretase-mediated cleavage. Additionally, it has been shown that Fe65 can both stabilize and target the ␥-secretase cleaved AICD to the nucleus (26,27). Consequently, we sought to explore whether Fe65 stimulation occurs by increasing ␥-secretase substrate production, directly stimulating ␥-secretase, or through stabilization of the released AICD-GV16 moiety. To accomplish this, we generated expression vectors coding for ␣or ␤-secretase APPGV16 cleavage products, C99-GV16 and C83-GV16. The C50-GV16 ␥-secretase cleavage product was also generated to test the Fe65 stabi-lization effects. These constructs were transfected into COS7 cells either alone or in the presence of the Fe65 isoforms.
The results indicate that Fe65 may play a role at each level of regulation of AICD generation. The most robust stimulation of AICD-GV16 production was observed with the full-length APPGV16 expression vector. Here, the neuronal form produced an approximate 4-fold increase in proteolysis, whereas the non-neuronal and a2 forms stimulated an approximate 7-8-fold increase (Fig. 7A). The augmented stimulation achieved by the non-neuronal and a2 forms of Fe65 was significantly higher than the neuronal form. Fe65-mediated stimulation of AICD-GV16 mobilization was also observed with the immediate substrates for ␥-secretase, as the three Fe65 isoforms produced a 3-4-fold increase in proteolysis with both C99-GV16 and C83-GV16. However, the Fe65 isoform dependence in proteolytic stimulation is dramatically attenuated with C99-GV16 and C83-GV16 (Fig. 7A). Although the action of the three isoforms is statistically significantly different with the APP holoprotein, there is no statistical difference between any of the Fe65 isoforms with C99-GV16 or C83-GV16. This further supports the notion that the observed differences in Fe65 stimulation with the holoprotein are not because of small differences in Fe65 protein abundance, as there were identical amounts of Fe65 expression vector co-transfected with the holo-APPGV16, C99-GV16, and C83-GV16. Additionally, there was a small increase in C50-GV16 transactivation with all three Fe65 isoforms, consistent with its previously reported role in AICD stabilization. However, this is only slightly above the nonspecific stimulation observed with the GV16 expression vector. Consequently, it appears that Fe65 may play a role in regulating both production of the immediate substrates of ␥-secretase, by promoting ␣-secretase or ␤-secretase-mediated cleavage of APP, and by directly stimulating ␥-secretase-mediated cleavage of those substrates.
To verify that the results observed do not reflect differences in efficiency of expression of different constructs, protein levels were assessed by Western blotting. The Gal4 antibody was employed to visualize the APPGV16 species, and the Myc antibody was used to elucidate Fe65 expression levels represented within the transactivation experiment (Fig. 7B). Immunoblots with anti-Gal4 demonstrated equivalent levels of APPGV16, C99-GV16, and C83-GV16 across conditions. The observed levels of C50-GV16 were less consistent, likely because of the instability of the ␥-cleaved C50 product. Fe65 expression levels also appeared roughly equivalent within the APPGV16 and C99-GV16 conditions, with perhaps slightly less of the Fe65 neuronal form. Interestingly, overexpression of C83-GV16 appeared to diminish the level of expression of Fe65. However, the remaining level of expression of Fe65 is apparently still sufficient to maximally promote C83-GV16 cleavage, as revealed by titrating the amount of Fe65 plasmid transfected (Fig. 6). It is unclear whether the apparent decrement of Fe65 expression resulting from expression of C50-GV16 and GV16 is meaningful. This difference may simply reflect variability in the immunoblot procedure, as the C50-GV16 and GV16 samples were run on a discrete gel from the APPGV16, C99-GV16 and C83-GV16 samples. Intriguingly, in the holo-APPGV16 and C99-GV16 transfected cells, a smaller Fe65 protein fragment  . Fe65 titration demonstrates isoform difference is independent of protein concentration. Fe65 isoforms were titrated into COS7 cells cotransfected with 100 ng of APPGV16 expression vector per well. Empty pcDNA3.1 cloning vector was counter-titrated with Fe65 to maintain constant cytomegalovirus promoter levels. A, all isoforms of Fe65 stimulated APPGV16 proteolysis significantly (p Ͻ 0.0001, two-way analysis of variance (ANOVA)) with saturating levels of Fe65 transfection observed prior to 1000 ng/well. Near maximal stimulation was achieved by 300 ng of Fe65 expression vector DNA (3:1 vector ratio with APPGV16). Increasing Fe65 expression vector concentration above a 3:1 ratio had little to no effect upon Gal4-luciferase levels. There was no difference in absolute values between 300 and 1000 ng per well within each Fe65 isoform. This is reflected in the lack of statistical difference between the 300 and 1000 ng per well conditions within any of the Fe65 isoforms. However, the Fe65 neuronal form was statistically significantly weaker than both the non-neuronal and the a2 isoforms across the titration (p Ͻ 0.0001, two-way ANOVA). B, Western blots were performed using the Gal4 antibody against the APPGV16 construct and the Myc antibody against the Fe65 Myc epitope employing 1 well of cells from the identical experiment. Across conditions, APPGV16 concentrations were equivalent. Conversely, Fe65 protein concentrations increased roughly linearly across the Fe65 expression vector titrations, with little to no differences in protein levels observed between the distinct Fe65 isoforms.
was generated, which was most robustly observed in the nonneuronal and a2 isoforms of Fe65 (denoted by the asterisk in Fig. 7B, bottom blot). The size of the observed protein matches the p65Fe65 endoproteolytic protein product reported previously (18). Interestingly, this proteolytic cleavage maps to the amino-terminal portion of Fe65. Based upon our observation that the amino-terminal portion of Fe65 is repressive to its stimulatory function (Fig. 4), this may suggest an isoform-specific mechanism for Fe65-promoted AICD-GV16 liberation. Fe65 Promotes Formation of the Immediate Substrates of ␥-Secretase-The difference between Fe65 stimulation of APPGV16 holoprotein cleavage, and that of C99-GV16 or C83-GV16, suggests that a large portion of Fe65 stimulation occurs at the level of ␣and ␤-secretase action. In order to verify this conclusion by more direct means, COS7 cells were transfected with APPGV16 alone or in association with the Fe65 isoforms. The ␥-secretase activity was blocked by application of DAPT, to allow accumulation of the products of APPGV16 by ␣and ␤-secretase. The production of C83-GV16 and C99-GV16 was assessed by immunoblot analysis. C83-GV16 and C99-GV16 were expressed in parallel as molecular weight references. Concordant with previous observations, the build up of C83-GV16 was greatest with overexpression of Fe65 non-neuronal and a2 isoforms (Fig. 8A). Little C99-GV16 was detected. This does not necessarily indicate the absence of ␤-secretase-mediated cleavage, however, because we observed that C99-GV16 was proteolytically processed into C83-GV16. This indicates that ␤-secretase and ␣-secretase may cleave sequentially. Again, a smaller Fe65 protein fragment was generated predominantly from the non-neuronal and a2 forms. Longer exposures did demonstrate small amounts of the endoproteolytic product present in the Fe65 neuronal condition as well. This suggests the interesting possibility that isoform-dependent differences may be associated with differential endoproteolytic processing of Fe65 isoforms. However, the Fe65 neuronal isoform appears to be the most sensitive to destabilization by APP. Consequently, there may be less of the p65Fe65 endoproteolytic product associated with the neuronal form simply because of lower levels of neuronal Fe65 protein. The examination of lysates for intracellular APPs did reveal a band consistent in size with the soluble ectodomain of APP that was increased in cells co-transfected with Fe65 (Fig. 8B).
Fe65 Stimulates Release of the APP Intracellular Domain Independent of Amyloid Production-Because Fe65 appears to promote processing of APP at the level of cleavage by ␣and ␤-secretase and promotes the cleavage of the products of those reactions by ␥-secretase, we sought to explore the role of Fe65 in the regulation of A␤40 and A␤42 production. COS7 cells were transfected with APPGV16 in the presence or absence of the Fe65 isoforms. Because of the lack of C99-GV16 build up with DAPT (Fig. 8), we assumed that endogenous ␤-secretase activity is low in COS7 cells. Consequently, BACE was co-transfected to facilitate an examination of the role of Fe65 in amyloid generation. As observed in the previous experiments, Fe65 markedly promoted AICD-GV16-mediated transactivation in an isoform-dependent manner (Fig. 9B). However, in contrast, using the media from the same cells analyzed in the transactivation assay, Fe65 induced only a small statistically insignificant FIGURE 7. Fe65 stimulation of AICD-GV16 liberation using APPGV16 and derivatives representing the ␣-, ␤-, and ␥-secretase cleavage products. The holo-APP form of APPGV16 and species corresponding to the ␤-secretase cleaved form (C99-GV16), the ␣-secretase cleaved form (C83-GV16), and the form resulting from ␥-secretase cleavage at the ⑀-site (C50-GV16) were transiently transfected into COS7 cells either alone or co-transfected with the three Fe65 isoforms. The GV16 expression vector was also transfected alone or in association with the Fe65 isoforms as a negative control for nonspecific effects of Fe65. A, normalized Gal4-luciferase transactivation values are plotted as fold stimulation using the average value of each APPGV16 species alone as a referent for all samples within that condition. All three Fe65 isoforms produced the greatest effect upon cleavage of the holo-APP species of APPGV16. The Fe65 neuronal form induced an approximate 4-fold stimulation, whereas the non-neuronal and a2 isoforms stimulated an approximate 7-and 8-fold increase, respectively. All Fe65 species induced statistically significant differences from APPGV16 alone at p Ͻ 0.005 or less. Additionally, the Fe65 neuronal form induced significantly less cleavage than either the nonneuronal or a2 isoforms (p Ͻ 0.02). The Fe65 non-neuronal and a2 isoforms were not statistically different from each other at the 95% confidence interval. The Fe65 fold stimulation was attenuated with the C99-GV16 and C83-GV16 species, ranging from 3-to 4-fold stimulation for all Fe65 isoforms. All isoforms of Fe65 induced statistically significant differences from either C99-GV16 or C83-GV16 expression alone (p Ͻ 0.003), with the exception of the Fe65 neuronal form with C83-GV16 in which the sample had elevated variance. The fold stimulation of the Fe65 isoforms with C50-GV16 was diminished, ϳ2-2.5-fold. This value was slightly higher than nonspecific effects observed with Fe65 co-expressed with GV16, which achieved 1.4 -1.9-fold stimulation. B, Western blots were run using a fraction of the lysates prepared for the transactivation assay. The samples were broken into two groups as follows: APPGV16, C99-GV16, and C83-GV16 were group one; C50-GV16 and GV16 conditions were group two. These sample sets were run on two gels for each antibody. The samples were probed with the Gal4 antibody to identify APPGV16, C99-GV16, C83-GV116, and C50-GV16. Conversely, the Myc antibody was used to examine expression levels of the Fe65 isoforms in the different transfection conditions (bottom blot). The asterisk denotes the presence of a small molecular weight species of Fe65, which was predominantly observed in association with the non-neuronal and a2 isoforms. The APPGV16, C99-GV16, and C83-GV16 protein species were equivalently expressed, whereas only small amounts of C50-GV16 were detected. The Fe65 neuronal (N), non-neuronal (n), and a2 (a) isoforms all had confirmed expression in the co-transfections with each APPGV16 species with some variability in detected levels.

Fe65 Stimulates AICD
increase in A␤40 levels but a consistent decrease in A␤42 levels (Fig. 9A, right axes). The decrement in A␤42 levels is statistically significant with the non-neuronal and a2 isoforms of Fe65. The lack of statistical significance with the neuronal isoform of Fe65 results from increased variance within the test set, although the mean A␤42 level was still lower with Fe65.
Immunoblot analysis with anti-Gal4 revealed that the level of expression of APPGV16 was consistent across the transfection conditions (Fig. 9C). Consequently, the Fe65-mediated reduction in A␤42 production is not because of variation in APPGV16 expression. Expression levels of Fe65 also were similar across conditions. In accordance with previous observations, there is marginally less of the neuronal form of Fe65 observed at the protein level.
Fe65 Influences Processing of APP by Wild-type and FAD Mutant PS1 Equivalently-The conclusion that Fe65 directly promoted processing of APP by ␥-secretase led to speculation that Fe65 might restore the impaired function of ␥-secretase previously observed with FAD mutant PS1 (17). To test this hypothesis, COS7 cells were transfected with APPGV16, in the presence or absence of neuronal Fe65, in conjunction with either wild-type or FAD mutant PS1. Two concentrations of APPGV16 plasmid were used to achieve levels of expression of APPGV16 that were known to be functionally subsaturating or saturating for the cellular processing capacity (17). The ratio of plasmid for APPGV16 and Fe65 was maintained at 1:4 across conditions. Consistent with previous observations, overexpression of FAD mutant PS1 resulted in statistically significant decreases in AICD-GV16 production (Fig. 10). Co-expression of Fe65 resulted in an increase in AICD-GV16 production in cells transfected with either FAD mutant or wild-type PS1 (Fig.  10). However, the fold increase with Fe65 was equivalent in wild-type and FAD mutant instances. Importantly, AICD-GV16 liberation maintained statistically significant differences between FAD mutant and wild-type PS1 transfected cells, irrespective of Fe65 co-expression. Consequently, it is unlikely that Fe65 will rescue deficiencies in AICD production resulting from mutant forms of ␥-secretase associated with FAD.

DISCUSSION
Fe65 is implicated in regulating numerous aspects of APP proteolytic processing and function, from A␤ secretion to nuclear signaling. In this study, we extend the previously ascribed roles for Fe65 in proteolytic regulation of APP to include the promotion of liberation of the AICD. A primary focus of this work is an exploration of the differences in efficacy of different Fe65 isoforms that result from allelic variation of the gene and from alternative splicing. In neuronal systems alternative splicing incorporates the six-nucleotide exon 9 encoding Arg-Glu into the nascent transcript (16). The Arg-Glu insertion occurs within the middle region of the PTB1 domain of Fe65. The functional significance of the difference between the neuronal E9 inclusive and the non-neuronal forms of Fe65 has remained enigmatic. Brains of AD patients examined post-mortem revealed that the neuronal form of Fe65 was down-regulated in degenerated regions of the brain, whereas both neuronal and non-neuronal isoforms were up-regulated in unaffected areas of the brain (28). This suggests that Fe65 may be protective against the onset of AD pathogenesis.
An allelic variant of Fe65, known as the a2 isoform, was identified as a polymorphism that conferred resistance to late onset dementia of Alzheimer-type (13). This conclusion has been disputed by some groups (29 -31) and confirmed by another working with a more age-advanced population (15). The polymorphism falls within the splice donor site of intron 13 and leads to splicing to an alternative splice acceptor site, resulting in a complete change in the coding sequence for the penultimate carboxyl-terminal portion of the Fe65 protein (14).
This study compared the function of the neuronal, non-neuronal, and the non-neuronal a2 forms of Fe65. The consistent and compelling observation is that the non-neuronal and nonneuronal a2 isoforms have substantially greater ability to pro-  with DAPT for an additional 24 h. One set of cells was transfected with either C83-GV16 or C99-GV16 as molecular weight markers, which were also treated with DAPT for the stipulated period. A, cells were lysed and immunoblot analysis was performed to assess APPGV16 expression, levels of accumulation of products corresponding to CTF␣ and CTF␤, and to verify consistency of Fe65 expression. The APP carboxyl-terminal antibody was used to probe holo-AP-PGV16 levels, as that also demonstrated the consistency of endogenous APP levels in the sample lysates. The blot probed with the Gal4 antibody demonstrates the levels of C99-GV16 and C83-GV16. Increased C83-GV16 accumulation is observed in the conditions co-transfected with Fe65 non-neuronal and a2 isoforms. Interestingly, C99-GV16 appears to be proteolytically processed, resulting in a molecular weight species that is indistinguishable from C83-GV16. Anti-Myc antibody was used to probe Fe65 expression levels. Similar levels of expression different Fe65 isoforms were observed, with a slight decrement in expression of neuronal Fe65. Additionally, a lower molecular weight species was observed in the non-neuronal and a2 isoforms of Fe65 (unlabeled arrow, bottom blot). B, blots were probed with the 22C11 antibody to detect uncleaved APP and intracellular APPs. A band consistent with the size of APPs was detected at increase levels in cells co-transfected with Fe65 (denoted in the figure by an asterisk).
mote APP processing than the neuronal form, whereas the activity of the non-neuronal and non-neuronal a2 isoforms is similar. The differences observed experimentally do not result from differences in the nature or quality of plasmid DNA employed, because the plasmids do not differ significantly except for the coding region inserted (14), and the results were obtained with numerous different plasmid preparations, each of which was carefully characterized with regard to purity and concentration. The majority of the experiments discussed here were performed in COS7 cells; however, similar results were obtained with NIH 3T3, HEK293, and JEG3 cells (supplemental Fig. S1). The use of neuronal cell types was explored, however, as previously noted endogenous expression of Fe65 is greatly elevated in neuronal systems (28); consequently, endogenous Fe65 masked the effects of transfected Fe65. FIGURE 9. Fe65 differentially stimulates AICD and A␤ production. COS7 cells were transiently transfected with APPGV16 alone or co-transfected with each of the Fe65 isoforms. All cells were co-transfected with the BACE expression vector to drive ␤-amyloid production. Eighteen h following transfection, the media were changed, and 300 l of fresh media was added to each well. The cultures were maintained for an additional 36 h, following which time the cells the media were drawn off each sample, and 100 l were added to two wells within a plate pre-prepared with the 6E10 amyloid capture antibody. One well was used for the A␤40 and the other for the A␤42 amyloid assay. The remainder of the cells were then washed, lysed, and analyzed by both the transactivation assay and Western blot. A, ␤-amyloid sandwich ELISA was performed as described under "Experimental Procedures." Detectable levels of A␤40 and A␤42 were found in all samples, plotted relative to the A␤40 values (left axis) or A␤42 values (right axis). The values were normalized to the linear regression generated from the A␤40 and A␤42 standards run in association with the assay. A␤40 levels were slightly elevated with the addition of Fe65. However, the differences in A␤40 levels between APPGV16 alone and co-expression of Fe65 were not statistically significant. In contrast, A␤42 levels declined with co-expression of Fe65. The difference in A␤42 levels between APPGV16 alone and co-expression of Fe65 non-neuronal (p Ͻ 0.006) and a2 (p Ͻ 0.03) were statistically significant. B, transactivation of the Gal4-luciferase reporter by AICD-GV16 was marked potentiated by all three Fe65 isoforms. The differences between APPGV16 alone and coexpression of Fe65 was statistically significant for all isoforms (p Ͻ 0.0001). Furthermore, the non-neuronal and a2 isoforms were both significantly more potent than Fe65 neuronal (p Ͻ 0.003). C, immunoblots were probed for APPGV16 with the Gal4 antibody and Fe65 using the Myc antibody. The expression levels of APPGV16 were consistent across conditions in which Fe65 neuronal (N), non-neuronal (n), and a2 (a) were co-transfected. The lower blot demonstrates Fe65 expression. The last lane is a mock-transfected control. . Fe65 stimulates AICD-GV16 production in wild-type and FAD mutant PS1. APPGV16 was transiently transfected into COS7 cells in the presence or absence of neuronal Fe65. CMV-PS1 expression vectors containing the wild-type or FAD mutant (M146L and C410Y) human cDNA sequences were co-transfected. Two concentrations of APPGV16 were used within each PS1 set. The low and high concentrations of APPGV16 were 50 and 250 ng of expression vector per well, respectively. The ratio of APPGV16:Fe65 was maintained at a 1:4 ratio across conditions. Levels of AICD-GV16 production were elevated in wild-type relative to FAD mutant PS1 within all conditions examined. PS1 wild-type (wt) was significantly different from M146L and C410Y in low and high concentrations of APPGV16, and in the presence and absence of Fe65 (p Ͻ 0.005; condition to condition analysis with Student's t test). Fe65 induced increases in AICD-GV16 production with both PS1 wild-type and FAD mutant transfected cells. However, the fold stimulation with Fe65 was indistinguishable between wild-type and FAD mutant conditions (low APPGV16 concentration, 4.1-fold (wild type), 3.6-fold (M146L), 3.6-fold (C410Y); high APPGV16 concentration, 3.0-fold (wild type), 3.4-fold (M146L), and 3.0-fold (C410Y)).

Fe65 Stimulates AICD
Previous work with the APPGV16 reporter system has demonstrated that APPGV16-mediated transcriptional activation is dependent upon ␣-, ␤-, and ␥-secretase proteolytic processing of APP (17). Consistent with this conclusion, this study demonstrates that Fe65-mediated stimulation of this reporter system requires ␥-secretase action, as DAPT blocks stimulation of transactivation by Fe65 (Fig. 2). Consequently, it is unlikely that the Fe65 induction of reporter activity is because of nonspecific transcriptional effects. Furthermore, as DAPT blocks the Fe65 augmentation of APPGV16 reporter activity, it is unlikely that other proteolytic events associated with the APP carboxyl terminus, such as caspase cleavage (20,32), are responsible for the Fe65-mediated effects observed.
The stimulation of AICD liberation by Fe65 apparently reflects both increased rate of ␥-secretase-mediated cleavage of its immediate substrates (C83 and C99) as well as increased rate of production of those substrates via cleavage of APP by ␣and ␤-secretases. The latter effect apparently predominates, because Fe65 promotes liberation of AICD from APPGV16 more strongly than from C83-GV16 or C99-GV16 (Fig. 7). The ability of Fe65 to promote formation of such ␥-secretase substrates was also demonstrated directly, as coordinate expression of APPGV16 and the three Fe65 isoforms in the presence of the ␥-secretase inhibitor DAPT shows elevated levels of C83-GV16 accumulation with the non-neuronal and a2 isoforms (Fig. 8A). We observed that all three Fe65 isoforms promoted liberation of AICD from both C83-GV16 and C99-GV16. These observations are consistent with previous studies showing that other members of the Fe65 gene family directly stimulate ␥-secretase activity (33). Therefore, Fe65 may promote a complex series of proteolytic events in which both steps prior to Fe65 liberation from the membrane are fostered by its association with APP.
Various studies have concluded that Fe65 either inhibits or promotes secretion of APP s␣ (10,12,18). The results of our analysis of AICD production indicate that Fe65 promotes cleavage of APP by ␣-secretase, a conclusion that implies that Fe65 should increase rather than decrease production of APP s␣ . We did not attempt to assess the effects of Fe65 overexpression on secretion of APP s␣ . However, we did observe that Fe65 overexpression increased the quantity of a band consistent in size with APP s␣ in cellular fractions (Fig. 8B), suggesting that Fe65 may stimulate ␣-secretase-mediated cleavage of APP within the Golgi or within endocytic compartments, both sites in which APP processing has been described (34 -38). Indeed, variation among cell types in the predominant locus of APP ␣-cleavage (plasma membrane, endosome, or trans-Golgi) might account for the disagreement in published studies concerning whether Fe65 promotes or inhibits secretion of APP s␣ .
Mutations of either tyrosine residue within the YENPTY motif of the Fe65-binding site of APP abolished stimulation of APP proteolysis (Fig. 3) by all three Fe65 isoforms. Although mutagenic scanning has identified the amino-terminal tyrosine as critical to the association of Fe65 with APP (7), the NPTY forms the core of the PTB domain recognition sequence (1,8,39). The YENPTY motif is highly conserved, and hence perturbations would be expected to disrupt associations between binding partners. Mutation of the critical APP-binding site abrogates the Fe65-mediated stimulation, suggesting that direct association is important. Although the association of Fe65 with APP is apparently important for Fe65-mediated effects on APP processing, the affinity of that interaction does not appear to be a major determinant of the effectiveness of the interaction, because the Fe65 a2 isoform is known to associate more weakly with APP than the non-a2 alleles of Fe65 (14); yet the a2 allele stimulates APP proteolysis with equivalent or greater efficacy to the non-neuronal and neuronal isoforms.
Our domain mapping studies demonstrate that all three core domains of Fe65 play a function in stimulating APP proteolysis (Fig. 4). Deletion of PTB1 has the most profound effect upon APP processing, as deletion of PTB1 converts Fe65 from an activator to an inhibitor of APP processing, possibly by acting as a dominant-negative inhibitor of the action of endogenous Fe65. The PTB1 domain of Fe65 mediates association with several binding partners, including LRP (10), Tip60 (2), Tau (40), and CP2/LSF/LBP1 (41), suggesting that the association of one or more of these proteins with Fe65 is necessary for stimulation of APP processing. Interestingly, the neuronal form of Fe65 inserts the residues Arg-Glu into PTB1. This suggests that the differing effects of non-neuronal and neuronal Fe65 on APP processing reflect differing association with one of the aforementioned binding partners. LRP is likely to be the differentially associating binding partner, as association of LRP with the APP-Fe65 complex is known to promote APP processing (10,11).
Deletion of the PTB2 domain of Fe65, which mediates association of Fe65 with APP (1,7,9), results in only a partial loss of Fe65 function. This surprising result is consistent with other studies demonstrating that both Fe65 and Fe65 lacking PTB2 promote APP s␣ secretion (10). This suggests the possibility that PTB1 may contribute weakly to the association of Fe65 with the YENPTY motif of APP.
Selective mutations to the WW domain that blocks its functionality (2), or deletion of the amino-terminal portion of Fe65 containing the WW domain, also attenuate the Fe65 stimulation of APP proteolysis. Surprisingly, we repeatedly observed small residual levels of stimulation with the Fe65 truncation mutants in which only the PTB1 domain remained. Although the mechanism of this effect requires further investigation, the effect underscores the critical role of PTB1 in Fe65 function.
Truncation of Fe65 amino-terminal to the recently identified acidic residue cluster (ARC) domain (18) enhances Fe65 stimulation of APP proteolysis (Fig. 4). Intriguingly, further truncation of the amino-terminal region past the ARC domain but prior to the WW domain, reduces Fe65 stimulatory capacity to levels similar to the full-length protein. This suggests that the most amino-terminal portion of Fe65 masks the augmentative role of the ARC domain. Consistent with this notion, it was recently noted that alternative start sites for Fe65 drive the formation of an amino-terminally truncated version of Fe65 in the partial Fe65 knock-out mouse (24). This start site maps to residue 260, leaving the majority of the WW domain intact, including two tryptophan residues. Overexpression of Fe65 from the alternative start site demonstrates indistinguishable levels of APPGV16 proteolysis from the full-length Fe65 protein (Fig. 5), consistent with the deletion studies. NOVEMBER 16, 2007 • VOLUME 282 • NUMBER 46

JOURNAL OF BIOLOGICAL CHEMISTRY 33323
There have been numerous reports of Fe65 modulation of ␤-amyloid secretion. However, both positive (12) and negative effects (42) have been reported. In this study, we find that all three isoforms decreased secretion of pathogenic A␤42 in concert with increased production of AICD-GV16 (Fig. 9, A and B). It is curious to note that none of the Fe65 isoforms significantly affect A␤40 levels. However, this result is reminiscent of the findings of our previous work, in which we showed that mutation of the APP S3 cleavage site resulted in decreased liberation of AICD-GV16 accompanied by increased production of A␤42, whereas A␤40 levels were unaffected (17). As Fe65 increases AICD production, we speculated that overexpression of Fe65 might partially rescue the loss of function observed with PS1 mutants. Although overexpression of Fe65 in this study did increase AICD-GV16 liberation via FAD mutant PS1, the Fe65mediated fold stimulation of APPGV16 processing was the same with both wild-type and FAD mutant PS1 (Fig. 10). Consequently, Fe65 is unlikely to be a useful means to restore the loss of function associated with the PS1 FAD mutants.
The significance of our finding that Fe65 isoforms differ in their ability to promote APP cleavage is due in part to the characterized role of Fe65 as a nuclear signaling moiety. It was originally suggested that transcriptional regulation was mediated by nuclear accumulation of a tripartite complex of AICD-Fe65-Tip60 (2, 3). Recent work has drawn into doubt the necessity of either the AICD or Tip60 for functional nuclear signaling, suggesting that Fe65 alone is sufficient for transcriptional initiation (6). The observation that hold up, prior to ␥-secretase-mediated cleavage, functions to sequester Fe65 from the nucleus is widely recognized (2,6,43). Consequently, the liberation of Fe65 from the cytosolic face of the membrane may be necessary for Fe65 nuclear signaling, whether it functions in a complex with AICD or independently. Although there is little doubt that Fe65 plays a transcriptional regulatory role, as deletion of Fe65 in mice negatively impacts both neurodevelopment and memory formation (24,44), which genes are actively regulated remains a hotly contended issue (3,(45)(46)(47)(48)(49).
The role of Fe65 in stimulating APP processing at multiple cleavage steps suggests a tight coupling between the nuclear signaling function of Fe65 and its role in proteolytic regulation of APP. The biological significance of the finding that the Fe65 isoform specific to neuronal systems promotes APP processing less effectively than other Fe65 isoforms is unclear. Possibly the lower activity of the neuronal form counter-balances the much greater level of Fe65 protein expressed in neurons (16). Our results indicate that Fe65 stimulates cleavage of APP by ␣-secretase and decreases pathogenic amyloid production. Because stimulation of ␣-secretase-mediated cleavage of APP and A␤42 suppression are both speculative venues for AD therapeutics, the mechanism involved in Fe65 regulation of APP proteolysis may provide insights into novel therapeutic approaches to AD.