PAT1a Modulates Intracellular Transport and Processing of Amyloid Precursor Protein (APP), APLP1, and APLP2*

Understanding the intracellular transport of the β-amyloid precursor protein (APP) is a major key to elucidate the regulation of APP processing and thus β-amyloid peptide generation in Alzheimer disease pathogenesis. APP and its two paralogues, APLP1 and APLP2 (APLPs), are processed in a very similar manner by the same protease activities. A putative candidate involved in APP transport is protein interacting with APP tail 1 (PAT1), which was reported to interact with the APP intracellular domain. We show that PAT1a, which is 99.0% identical to PAT1, binds to APP, APLP1, and APLP2 in vivo and describe their co-localization in trans-Golgi network vesicles or endosomes in primary neurons. We further demonstrate a direct interaction of PAT1a with the basolateral sorting signal of APP/APLPs. Moreover, we provide evidence for a direct role of PAT1a in APP/APLP transport as overexpression or RNA interference-mediated knockdown of PAT1a modulates APP/APLPs levels at the cell surface. Finally, we show that PAT1a promotes APP/APLPs processing, resulting in increased secretion ofβ-amyloid peptide. Taken together, our data establish PAT1a as a functional link between APP/APLPs transport and their processing.

For APP, a number of intracellular interaction partners, such as Fe65 (19), Fe65L1 (20), or X11␣ and X11␤ (21), are known to affect its processing. All of these proteins interact with the NPTY motif in the intracellular domain of APP, APLP1, and APLP2 via their phosphotyrosine binding (PTB) domain (22). Protein interacting with APP tail 1 (PAT1) binds to the basolateral sorting sequence (BaSS) of APP and is associated with microtubules. Further, an influence on APP cleavage at the cell surface has been proposed (23). Therefore, a kinesin light chain (KLC)-like function has been proposed (23). However, the structural and primary sequence homology between PAT1 and KLC is low. Additionally, PAT1 was shown not to interact with kinesin heavy chain (23), and its association to microtubules is elevated in the presence of ATP (23), indicating that PAT1 has no KLC-like function. It has also been reported that PAT1 affects the half-life time of the APP intracellular domain (AICD), suggesting a regulatory function of PAT1 in the putative nuclear signaling of the AICD (24). In this study, AICD fragments with a length of 57 (C57) and 59 (C59) amino acids were shown to be affected by PAT1. However, the majority of AICD released to the cytoplasm results from ⑀-cleavage, and it is not clear whether C57 and C59 are released to the cytosol in vivo (15,(25)(26)(27)(28)(29). Additionally, Ara67, which is 99.6% identical to PAT1, was reported as an androgen receptor-interacting protein, interrupting androgen receptor cytoplasmic-nuclear import and modulating androgen receptor signaling. These studies suggest that PAT1 might be a regulatory factor that controls the subcellular localization of multiple proteins with divergent functions.
Here we show that PAT1a, which is 99.0% identical to PAT1, exists in a common complex with APP, APLP1, and APLP2 in vivo and co-localizes with APP/APLPs bearing intracellular vesicles. Moreover, we demonstrate that PAT1a regulates the levels of APP family members at the cell surface. Finally, we provide evidence that PAT1a increases the proteolytic conversion of APP/APLPs and promotes the generation of A␤, suggesting an important regulatory role of PAT1a in the transport of APP/APLPs with possible implications for Alzheimer disease pathogenesis.

EXPERIMENTAL PROCEDURES
Cloning of Human PAT1a-The PAT1a cDNA was amplified from a human brain cDNA library (Clontech) using open reading frame-flanking primers (sense, 5Ј-gaa gga aga tgg cgg ccg tgg-3Ј; antisense, 5Ј-cct cag cag ctc ggt ccc tcg aca ttc tg-3Ј) and cloned into the pCRII-TOPO vector (Invitrogen). For recombinant expression of PAT1a, the encoding cDNA was subcloned into pcDNA3.1 (Invitrogen). Alternatively, a hemagglutinin (HA) tag was fused to the 3Ј-end of the PAT1a cDNA by PCR and cloned into pcDNA3.1. For this purpose, the following oligonucleotides were used: sense, 5Ј-aat tgg tac cgc cgc cac cat ggc ggc cgt gga act aga gtg-3Ј, and antisense, 5Ј-gta att gcg gcc gct caa gcg tct tag ggg acg tcg tat ggg tag cag cac ggt ccc tcg aca ttc tg-3Ј.
Cell Fractionation and Carbonate Extraction-COS-7 or SH-SY5Y cells were co-transfected with PAT1a-HA and APP, APLP1, or APLP2 cDNAs. 28 h after transfection, cells were lysed in homogenization buffer (HOM) (250 mM sucrose, 5 mM HEPES, pH 7.4, 1 mM MgCl 2 , 10 mM KCl, 1 mM phenylmethylsulfonyl fluoride) by passing 20 times through a 27-gauge needle and centrifuged at 100 ϫ g for 5 min at 4°C. The supernatant was added to a 9-fold higher volume of carbonate buffer (0.1 M Na 2 CO 3 , pH 11) and incubated for 15 min on ice. The postnuclear homogenate was then layered on a 1/3 volume of sucrose/carbonate buffer (carbonate buffer containing 0.25 M sucrose) and centrifuged at 100,000 ϫ g for 45 min. Soluble proteins in the supernatant and membrane fractions in the pellet were analyzed by SDS-PAGE followed by Western blotting. The amounts of PAT1a associated with APP/APLPs in the membrane fraction or present in the fraction containing soluble proteins were quantified using Advanced Image Data Analyzer (AIDA) software, version 3.20.116 (Raytest Isotopen-Messgeräte GmbH).
Immunocytochemical Assays-For immunocytochemical analysis, primary neurons were fixed with 4% paraformaldehyde/phosphate-buffered saline for 10 min and permeabilized with 0.1% Nonidet P-40 in phosphate-buffered saline for 10 min. Staining of cells was performed using standard procedures (33,34). The fluorescent signals were visualized with an Axiovert 200 M Inverted Microscope (Zeiss) supplied with a CCD Camera (Hamamatsu), and images were acquired and enhanced using the MetaMorph Imaging System (Universal Imaging Co.).
Cell Culture and Transfection-COS-7, HEK293, and SH-SY5Y cells were cultured under standard conditions. Primary neurons were prepared as described (35) with the exception that mouse cortical neurons of embryonic day 14 were used. Neurons were grown on poly-L-lysine-coated 15-mm coverslips (Marienfeld) in serum-free neurobasal medium (Invitrogen) with B-27 supplement (Invitrogen), 25 M glutamate, and 0.5 mM glutamin. Cells were transfected with plasmid DNA using Lipofectamine Plus (Invitrogen) or Lipofectamine 2000 (Invitrogen) according to the manufacturer's instructions and collected for analysis 20 -72 h after transfection.
Silencing of PAT1a Expression-For the down-regulation of PAT1a, a short hairpin (sh) construct (shPAT1a) was cloned into pSilencer1.0-U6 (Ambion) using the sequence previously described by Zhang et al. (36). A scrambled RNAi sequence directed against PAT1a (scrPAT1a) was cloned accordingly and served as a control. For cassette cloning of the shPAT1a con-struct, the following two oligonucleotides were used: forward, 5Ј-cga agg cag aac agt taa ttt tca aga gaa att aac tgt tct gcc ttc ttt ttt g-3Ј, and reverse, 5Ј-aat tca aaa aag aag gca gaa cag tta att tct ctt gaa aat taa ctg ttc tgc ctt cgg gcc-3Ј. The non-functional scrPAT1a construct was generated by the same strategy, using the following oligonucleotides: forward, 5Ј-cga agg tag atc agc taa ttt tca aga gaa att aac tgt tct gcc ttc ttt ttt g-3Ј, and reverse, 5Ј-aat tca aaa aag aag gca gaa cag tta att tct ctt gaa aat tag ctg atc tac ctt cgg gcc-3Ј. The identity of the construct was confirmed by sequencing. SH-SY5Y cells were transfected with pSilencer sh-PAT1a and src-PAT1a using Lipofectamine 2000 (Invitrogen) according to the manufacturer's instructions and collected for analysis 48 -72 h after transfection.
Co-Immunoprecipitation Assay-Adult mouse brains were homogenized in ice-cold co-immunoprecipitation buffer (10 mM HEPES, pH 7.2, 143 mM KCl, 5 mM MgCl 2 , 1 mM EGTA, 1% Triton X-100, protease inhibitor mix (Roche Applied Science). Before immunoprecipitation, either anti-PAT1a antibody or preimmune serum as a control was incubated with protein A-Sepharose beads. Alternatively, antibodies were covalently coupled to an agarose gel (ProFundTM co-immunoprecipitation kit (Pierce)). For IPs, brain lysates of young adult mice (C57Bl/6) containing 400 g of protein were incubated with beads loaded with anti-PAT1/PAT1a antibody or preimmune serum for 4 h at 4°C. After extensive washing with co-immunoprecipitation buffer, immunocaptured protein samples were collected in sample buffer containing ␤-mercaptoethanol and analyzed by SDS-PAGE and Western blotting. The antigen/ antibody-complexes were visualized using ECL detection (Amersham Biosciences).
Analyses of APP/APLPs Processing-SH-SY5Y cells were transfected as described previously. For the analysis of secreted sAPP/APLPs and A␤, fresh growth medium was added to the cells after transfection (2.5 ml for each 6-cm dish and 1 ml for each well of a 6-well plate), conditioned for 30 h, and subsequently assayed for A␤ (ELISA) or analyzed by Western blotting. For the analysis of the fulllength APP/APLPs and CTFs, cells were lysed 30 h after transfection for 30 min in ice-cold lysis buffer (50 mM Tris/HCl, pH 7.5, 150 mM NaCl, 1% Triton X-100, 5 mM EDTA, and Complete TM protease inhibitor mixture (Roche Applied Science)) and centrifuged at 10,000 ϫ g. The supernatants were denatured in sample buffer and analyzed by SDS-PAGE on 8% Tris-glycine or 15% Tris-Tricine gels and Western blotting.
Quantification of A␤ Using Sandwich ELISA Assay-Secreted A␤ levels were measured by a Sandwich ELISA assay (hAmyloid ELISA High Sensitive, The Genetics Company, Zürich, Switzerland) as described (37). Mouse monoclonal antibodies specific to A␤40 (G2-10) or A␤42 (G2-13) were used as capture antibodies, and an N-terminal A␤ antibody (W0 -2) was used for detection. A␤ levels in the samples were determined by comparison with the signal from media spiked with known quantities of A␤40 and A␤42.
Surface Biotinylation-48 h after transfection with control plasmid, PAT1a cDNA, or shPAT1a, SH-SY5Y cells were washed with cold phosphate-buffered saline containing 0.2 mM CaCl 2 and 2 mM MgCl 2 followed by incubation with sulfo-NHS-LC-biotin (0.5 mg/ml, Pierce) for 1 h at 4°C. The reaction was stopped by extensive washing with phosphate-buffered saline and quenched with 50 mM NH 4 Cl, 1% bovine serum albumin in 50 mM Tris, pH 8.0. Cell lysates were prepared and precipitated with streptavidin-Sepharose (Amersham Biosciences). The precipitated biotinylated proteins were subjected to Western blot analysis. For detection of PAT1a or APP/APLPs, the indicated antibodies were used and visualized using horseradish peroxidaseconjugated secondary antibodies and enhanced chemiluminescence technique (Amersham Biosciences).

RESULTS
Polymorphism of the PAT1/APPBP2 Gene-As a prerequisite for our studies of the PAT1 protein, we amplified the coding region of PAT1 cDNA from a human brain cDNA library. Surprisingly, we consistently found seven nucleotide exchanges in eight independent clones, resulting in six amino acid substitutions (F296L, K325R, A337A, S340C, L354V, P369R, N544K) ( Fig. 1). Data base analyses of the PAT1 sequence (cDNA, AF017782; aa, AAC83973.1) revealed that besides PAT1, a sequence (APPBP2 cDNA, NM_006380; aa, NP_006371.2) completely identical with the sequence we amplified from the

PAT1a-dependent APP/APLPs Transport and Processing
brain cDNA is annotated in the data base. Furthermore, we found that the genomic sequence, which is located on chromosome 17, is also identical to our brain cDNA PCR product and APPBP2 (LocusID: 10513). The Ensembl website gives no indication for a duplication of this gene in the human genome. We analyzed 53 expressed sequence tags covering the region of PAT1 cDNA containing the nucleotide exchanges and found that all tested expressed sequence tag sequences were identical with our brain cDNA PCR products and APPBP2, but none was equal to PAT1. These data indicate that the PAT1 sequence used in previous analyses (23,24,36) carries six amino acid exchanges. The PAT1 amino acid sequence is thus 99% identical to APPBP2. We conclude that the PAT1 sequence represents a rare variant of APPBP2 rather than a sequencing error as independently of the PAT1 sequence, six of seven nucleotide exchanges in the PAT1 sequence are annotated as single nucleotide polymorphisms (SNPs) in the genomic sequence. In addition, Ara67 has been described to be 99.6% identical with PAT1 (36), suggesting that Ara67 is a further variant of PAT1/AP-PBP2. Thus, PAT1, Ara67, and APPBP2 likely represent different alleles of one single gene, of which APPBP2 is the most common form. For simplification, we propose to rename APPBP2 to PAT1a.
Co-localization and Interaction of PAT1a with APP, APLP1, and APLP2 in Neurons-To test whether PAT1a interacts with APP and possibly also with APLP1 and APLP2 in vivo, we performed co-fractionation, co-immunoprecipitation, and co-localization experiments. In a cell fractionation assay, we found that about 50% of PAT1a co-fractionated with membranes containing APP/APLPs. Moreover, PAT1a can be released into the cytosolic fraction by carbonate extraction, showing that it is not a transmembrane protein nor coupled covalently to the membrane (data not shown). Together these data suggest that PAT1a is associated with membrane compartments containing APP, APLP1, and APLP2. To perform co-immunoprecipitation analysis, we generated a polyclonal anti-PAT1a antibody directed against an epitope in the C terminus of human PAT1a (amino acids 542-572). This region contains only one of the amino acid substitutions (N544K) when compared with PAT1/ Ara67. Since we have not tested whether the antibody recognizes PAT1 and Ara67, we cannot exclude the possibility of cross-reactivity. However, we show that this anti-PAT1a antibody can be used for Western analyses and immunoprecipitation studies of recombinant HA-tagged PAT1a and endogenous PAT1a (possibly also PAT1 and Ara67) ( Fig. 2 and see also Fig. 5). Co-immunoprecipitation experiments with mouse brain extracts using the anti-PAT1a antibody revealed that APP, APLP1, and APLP2 can be co-immunoprecipitated with PAT1a, suggesting that all three APP family members are interacting with PAT1a in vivo (Fig. 2). Co-immunoprecipitations with PAT1a preimmune serum served as control (Fig. 2). Noteworthily, the mature forms of APP, APLP1, and APLP2 were selectively co-immunoprecipitated with PAT1a, indicating specific interaction in post-Golgi compartments.
The specificity of the anti-APP, anti-APLP1, and anti-APLP2 antibodies was demonstrated in Western analyses of brain extracts from wild-type mice when compared with APP-, APLP1-, and APLP2-knock-out mice (supplemental Fig. 1). We next investigated the cellular localization of PAT1a in mixed cortical neurons. Since we so far could not establish conditions that allow immunocytochemical analyses of endogenous PAT1a, we transfected primary neurons differentiated for 7 days in vitro with a cDNA encoding HAtagged PAT1a. We observed a diffuse cytosolic staining as well as punctuate vesicular PAT1a immunoreactivity, suggesting that PAT1a is partly associated with intracellular vesicles (Fig. 3). We also detected a strong PAT1a immunoreactivity in the somatodendritic compartment and in MAP2negative neurites, indicating that PAT1a is localized to both dendrites and axons (data not shown). In non-neuronal cells, a FIGURE 2. Co-immunoprecipitation (IP) of PAT1a with APP, APLP1, and APLP2 from mouse brain extracts. A, immunoprecipitation of recombinant PAT1a with the anti-PAT1a antibody in SH-SY5Y cells. Cell lysates of transiently transfected SH-SY5Y cells expressing HA-tagged PAT1a were incubated with anti-PAT1a antibody or preimmune serum (P.I.) as a control. The immunoprecipitates were detected in Western blot analysis with the anti-HA antibody. 20% of the cell homogenate (Input) was loaded as a control. In the lanes loaded with the immunoprecipitates besides PAT1a, the heavy chain of the IgG was detected (marked by an asterisk). B, in co-immunoprecipitation experiments from mouse brain extract, PAT1a was immunoprecipitated with the anti-PAT1a antibody. Samples lacking the PAT1a antibody (Ϫ) and samples with the corresponding preimmune serum served as controls. 1% of the brain homogenates (Input) were loaded as control. Western blot detection of APP, APLP1, and APLP2 was performed with the antibodies 22C11, CT11, and D2-II, respectively. PAT1a was detected with the anti-PAT1a antibody. DECEMBER 29, 2006 • VOLUME 281 • NUMBER 52 mixed nuclear and cytosolic localization of PAT1 was reported (24), which could be confirmed in our studies using neuronal (SH-SY5Y, N2a, primary cortical neurons) and non-neuronal cells (COS-7 and HEK293) (supplemental Fig. 2). Finally, we investigated whether PAT1a co-localizes with APP, APLP1, or APLP2. For this purpose, primary neurons were transfected with cDNAs encoding HA-tagged PAT1a and Myc-tagged APP, APLP1, or APLP2. Immunocytochemical analysis revealed co-localization of PAT1a with APP, APLP1, or APLP2 in vesicular structures of distal neurites (Fig. 3). Remarkably, not all APP/APLP-positive vesicles were co-stained with PAT1a. This suggests that PAT1a is associated with a specific subset of APP/APLP-containing vesicles in the early and late secretory pathway or in endosomes. Taken together, these data show that PAT1a is associated with vesicular membranes containing APP/APLPs, suggesting a function of PAT1a in APP/ APLPs trafficking.

PAT1a-dependent APP/APLPs Transport and Processing
PAT1a Binds Directly to the BaSS of APP, APLP1, and with the Highest Affinity to APLP2-To investigate whether the association of PAT1a with APP-, APLP1-, and APLP2-containing vesicles is mediated via a direct binding of PAT1a to the intracellular domain of APP/APLPs, we performed GST pulldown experiments with in vitro-translated PAT1a and GST fused to the intracellular domain of APP (AICD), APLP1 (ALICD1), and APLP2 (ALICD2), respectively. We found that PAT1a was specifically retained on beads loaded with GST-AICD but did not bind to beads loaded with GST alone (Fig. 4). However, the binding capacity of PAT1a to AICD was low (only 2-fold above background), which is consistent with previous studies (23). The APLP1 and APLP2 intracellular domains also bound PAT1a. Interestingly, ALICD2 showed stronger binding to PAT1a than ALICD1 or AICD (10-fold above background) (Fig. 4). Kinesin light chain 1 (KLC1), which has a similar structure as PAT1a, did not bind to GST-AICD, -ALICD1, or -ALICD2 fusion proteins under the same conditions, demonstrating the specificity of our GST pulldown analysis (30). Deletion mapping experiments, based on two-hybrid analysis and GST pulldown studies of APP and PAT1, have previously shown that PAT1 binds to the BaSS of APP (23). To investigate the role of the BaSS in PAT1a binding to APLP1 and APLP2, we generated a mutant GST-ALICD1 and GST-ALICD2 lacking the BaSS (GST-ALICD1⌬BaSS and GST-ALICD2⌬BaSS). Neither GST-ALICD1⌬BaSS nor GST-ALICD-2⌬BaSS were interacting with PAT1a, demonstrating a direct binding of PAT1a to the BaSS of APP, APLP1, and APLP2.
PAT1a Facilitates the Trafficking of APP, APLP1, and APLP2 to the Cell Surface-To test whether PAT1a plays a role in the regulation of subcellular transport of APP/APLPs, we examined the influence of PAT1a expression on APP/APLP trafficking to the cell surface. For this purpose, we generated a short hairpin construct to mediate a knockdown of PAT1a (shPAT1a). A non-functional RNAi construct (scrPAT1a) served as a control.
Endogenous PAT1a levels were determined with our anti-PAT1a antibody. PAT1a was detected in Western blot analyses of cell lysates from SH-SY5Y cells overexpressing HA-tagged PAT1a and in cell lysates of non-transfected SH-SY5Y at 67 kDa, corresponding to the calculated molecular mass (66.9 kDa) (Fig. 5A). Under PAGE conditions optimized for separation of proteins with a molecular mass between 60 and 80 kDa, two forms of PAT1a could be differentiated (Fig. 5B). The difference in the apparent molecular weight might be explained by different phosphorylation states of PAT1a or other post-translational modifications. An anti-HA antibody recognized proteins with the same molecular weight in PAT1a-HA expressing cells but not in control cells transfected with empty vector. The overexpression rate of PAT1a was about 3-5-fold higher than endogenous PAT1a protein levels. Additionally, both antibodies, the anti-PAT1a and the anti-HA antibody, recognized lower molecular weight bands after longer exposure times, which may represent C-terminal cleavage products of PAT1a (data not shown). PAT1a RNAi treatment of COS-7 and SH-SY5Y cells reduced PAT1a levels to 30 -50%, whereas ␤-tubulin protein levels were not affected ( Fig. 5B and data not  shown).
To evaluate whether PAT1a influences surface trafficking of APP/APLPs, we compared the amounts of surface APP, APLP1, and APLP2 in the presence of increased or reduced expression levels of PAT1a by cell surface biotinylation studies. Cells overexpressing PAT1a showed increased levels of surface APP and APLP2, whereas the PAT1a knockdown caused a reduction of APP and APLP2 at the plasma membrane. Altered PAT1a expression affected the APLP1 surface expression in a similar way, but in contrast to the influence on APP and APLP2, not significantly. However, manipulating PAT1a expression had no effect on total levels of APP, APLP1, and APLP2, suggesting a PAT1a-dependent relocalization of APP/APLPs to the cell surface.
PAT1a Affects Processing of APP, APLP1, and APLP2-Assuming that PAT1a regulates the anterograde transport of APP/APLPs to the cell surface, it should consequently affect the proteolytic conversion of APP/APLPs. Besides the Golgi apparatus, the plasma membrane is one of the two major sites where ␣-secretase cleavage of APP takes place (9,38). In contrast, BACE 1 is most active at mildly acidic pH (11), making endosomes the most likely location for BACE cleavage of APP. Moreover, BACE 1 was reported to be cycling between the cell surface and endosomes (39). It was also shown that internalization of cell surface APP via endocytosis leads to elevated A␤ generation and release into the medium (40). To test the impact of elevated or reduced levels of PAT1a on processing of APP/ APLPs, we used SH-SY5Y and COS-7 cells expressing APP, APLP1, or APLP2 fused to a tandem z-domain (2z), which was used as it facilitates the detection of all APP/APLP CTFs (8,15). The cells were additionally co-transfected with HA-tagged PAT1a, the PAT1a RNAi construct, or a non-functional RNAi construct (srcPAT1a) as a control. Cells coexpressing PAT1a showed elevated levels of sAPP, sAPLP1, and sAPLP2, as well as increased amounts of the corresponding CTFs. On the other hand, the reduction of PAT1a expression led to decreased secretion of sAPP/sAPLPs and lower CTF levels (Fig. 5). Remarkably, PAT1a affected ␣-, ␤-, and ␥-CTF generation to a very similar extent (Fig. 6), which argues in favor of a function in anterograde transport of APP/APLPs.
Influence of PAT1a on the Generation of A␤40/42-Since PAT1a influenced APP/APLPs CTF generation, we assumed that PAT1a should also affect A␤ production. To test this assumption, we collected conditioned medium of human neuroblastoma cells (SH-SY5Y) expressing APP, APLP1, or APLP2 fused to a 2z-domain and different levels of PAT1a (as a consequence of transfection with PAT1a cDNA, scrPAT1a, or shPAT1a) 48 h after transfection. A␤40 and A␤42 levels were determined using ELISA measurements. Consistent with the observed effect of PAT1a on ␤-CTF and ⑀-CTFs generation, we found that cells overexpressing PAT1a generated elevated levels of A␤, whereas cells with reduced levels of PAT1a produced less A␤. No difference in the ratio of A␤40 to A␤42 was observed. These data demonstrate that altered PAT1a expression affects A␤ production, suggesting a regulatory role in the pathogenesis of Alzheimer disease.

DISCUSSION
In this study, we show that PAT1a binds to all APP family members in vivo and associates with APP-, APLP1-, and APLP2-containing vesicles in neurons, likely through direct interaction with their ICDs. We also demonstrate that PAT1a regulates the cell surface levels of APP/APLPs. Moreover, we show that PAT1a promotes the cleavage of APP/APLPs by ␣-, ␤-, and ␥-secretases, resulting in elevated secretion of sAPP/ sAPLPs, increased production of ␣-, ␤-, and ⑀-cleaved CTFs, and elevated levels of A␤.
Our data base and PCR analyses of brain cDNA revealed that the sequence differences in PAT1 and Ara67 are due to SNPs in the PAT1a/APPBP2 gene, indicating a high variance in this gene. Detailed investigations will be necessary to determine the precise frequency of SNPs and whether these nucleotide exchanges have an impact on the binding affinity of PAT1a to APP/APLPs, which could affect APP processing.
PAT1a was reported to be a nucleocytoplasmic protein, but so far, only cytosolic functions of PAT1a/Ara67 have been proposed (23,24,36). However, we also observed nuclear localization of PAT1a in non-neuronal cell lines, 7 but we did not find any specific PAT1a immunoreactivity in the nuclei of primary neurons, although PAT1a was expressed at high levels. Therefore, a nuclear function of PAT1a in neurons appears unlikely. However, we found that in primary neurons, PAT1a co-fractionates and co-localizes with membrane compartments containing APP/APLPs. Although we did not determine the specific character of these vesicles, their morphology and localization in distal neurites suggests that they are vesicular components of the secretory or endocytotic pathway. Thus, under consideration of our studies showing an influence of PAT1a on the amount of APP/APLPs at the cell surface, we propose that PAT1a either facilitates the transport of APP/ APLPs to the cell surface or inhibits their endocytosis rate. Consistently, we found that PAT1a binds directly to the juxtamembrane region of the intracellular domain of APP/APLPs. The BaSS has previously been shown to be important for transport of APP to the cell surface, APP internalization, and basolateral transport of APP and APLP2 in MDCK cells (41)(42)(43).
All APP family members are predominantly cleaved in the late secretory pathway, including the plasma membrane and endosomes (44). We found that PAT1a affects APP/APLPs processing in a way that simultaneously increases or decreases ␣-, ␤-, and ␥-cleavage depending on increased or decreased expression of PAT1a, respectively. Since all processing steps including ⑀-CTF and A␤ production were similarly affected, it might be possible that PAT1a regulates ␣-, ␤-, and ␥-secretase activities. However, this is unlikely as activation of the involved proteases relies on different molecular mechanisms. Interestingly, several adaptor proteins are known to influence APP transport and processing. Fe65 overexpression was shown to  Thus, PAT1a appears as a single band or as a double band, respectively. Probing of the same extract with a HA antibody revealed an identical signal. After loading of higher amounts of protein extracts (60 g), the anti-PAT1a antibody also detected endogenous PAT1a in non-transfected cells with the same apparent molecular weight. The overexpression rate of PAT1a was about 3-5-fold higher than endogenous PAT1a protein levels. B, SH-SY5Y cells were treated with transfection reagent only, with shPAT1a, or with a non-functional RNAi construct (scrPAT1a). 48 h after transfection, cell lysates were analyzed on a 8% Tris-glycine gel and subjected to Western analysis using anti-PAT1a and anti-␤-tubulin antibodies. C, biotinylation assay 48 h after transfection with PAT1a cDNA, for control pCDNA3.1 or shPAT1a, and for control scrPAT1a. Cells were incubated at 4°C for 1 h with a biotinylation reagent. Cell extracts were prepared, and biotinylated surface proteins were immunoprecipitated. Western analysis was performed with anti-APP, anti-APLP1, or anti-APLP2 antibodies. D, for quantification, densitometric measurements of three independent experiments were performed. The data were normalized to the amount of total APP, APLP1, and APLP2. The mean Ϯ S.E. is indicated. Statistically significant differences in comparison with control cells transfected with pCDNA3.1 (served as control for PAT1a overexpression) or the scrPAT1a construct (served as control for PAT1a knockdown) (Mann-Whitney U-test) with a p value Ͻ0.05 are marked by an asterisk. FIGURE 6. PAT1a affects APP/APLP processing. The influence of PAT1a on APP/APLP processing was analyzed in SH-SY5Y cells co-transfected with APP-, APLP1-, or APLP2-2z plasmids and PAT1a cDNA (black column), pCNA3.1 (empty vector, gray column), srcPAT1a (non-functional RNAi construct, gray column), or shPAT1a (RNAi, black column). A, after 48 h, medium containing sAPP, sAPLP1, or sAPLP2, and the cell extracts containing full length (fl) and CTFs of APP, APLP1, and APLP2 were subjected to Western analyses. B and C, the amounts of sAPP/APLPs (B) and ␣-/␣-like CTFs (C) generated in cells with different PAT1a expression levels were quantified. The relative amounts of CTFs and secreted fragments in comparison with srcPAT1a or pCDNA3.1-transfected cells (indicated by pCDNA3.1 (served as control for PAT1a overexpression) or the scrPAT1a construct (served as control for PAT1a knockdown) were set to 1) are shown. APP/APLPs full-length protein levels varied in the range of less than Ϯ 5%. Error bars show S.E. (n Ն 3). D, 48 h after co-transfection of SH-SY5Y cells with APP-2z and PAT1a cDNA, pCDNA3.1, scrPAT1a, or shPAT1a, the conditioned medium was subjected to ELISA analysis to determine A␤40 and A␤42 levels. Error bars show standard deviations of the mean (n Ն 3). Statistical significant differences (Mann-Whitney U-test) with a p value Ͻ0.05 and Ͻ0.01 are marked by a single asterisk or two asterisks, respectively. E, determination of the ratio of A␤40/A␤42. DECEMBER 29, 2006 • VOLUME 281 • NUMBER 52 enhance APP secretion (19,45). In contrast, X11␣ overexpression blocks APP maturation and secretion (21,46), whereas the knockdown of X11␣ increases APP ␣and ␤-CTFs but decreases A␤ levels (47).

PAT1a-dependent APP/APLPs Transport and Processing
As described in this work, the observed concomitant increase or decrease in all processing steps strongly suggests that PAT1a is a limiting factor to deliver APP/APLPs to compartments with secretase activity. In light of the distinct subcellular localization of the different secretase activities, namely the plasma membrane for ␣-secretase (9, 38) and endocytic compartments for ␤/␥-secretases (11, 39, 48 -50), a function of PAT1a in APP/ APLPs transport to the cell surface appears most likely. This view is further corroborated by the regulatory effect of PAT1a on cell surface levels of APP/APLPs. Although controversial, inhibition of endocytosis using a dynamin mutant was reported to retain elevated levels of APP at the cell surface, leading to increased ␣-cleavage. Surprisingly, ␤-secretase cleavage and A␤ production were enhanced as well (51,52). Therefore, we cannot exclude that PAT1a may also reduce the endocytosis rate of APP/APLPs.
Further, the manipulation of PAT1a expression had different effects on processing and cell surface levels of APP/APLPs. For example, APLP1 processing was clearly affected by altered PAT1a expression, whereas the cell surface levels of APLP1 were not significantly changed. This indicates that PAT1a might affect the intracellular transport of APP/APLPs by increasing the proportion of APP/APLPs in compartments with active ␣and ␤-secretases, which does not necessarily lead to a pronounced accumulation of the full-length APP/APLPs at the cell surface.
In conclusion, we hypothesize that PAT1a is controlling the intracellular localization of APP/APLPs, and by these means, also the proteolytic cleavage of all APP family members.