Ectodomain phosphorylation of beta-amyloid precursor protein at two distinct cellular locations.

The β-amyloid precursor protein (βAPP) is a transmembrane protein that is exclusively phosphorylated on serine residues within its ectodomain. To identify the cellular site of βAPP phosphorylation, we took advantage of an antibody that specifically detects the free C terminus of β-secretase-cleaved βAPP containing the Swedish missense mutation (APPssw-β). This antibody previously established the cellular location of the β-secretase cleavage of Swedish βAPP as a post-Golgi secretory compartment (Haass, C., Lemere, C., Capell, A., Citron, M., Seubert, P., Schenk, D., Lannfelt, L., and Selkoe, D. J. (1995) Nature Med. 1, 1291-1296). We have now localized the selective ectodomain phosphorylation of βAPP to the same compartment. Moreover, the phosphorylation sites of βAPP were identified at Ser198 and Ser206 of βAPP695 by tryptic peptide mapping, mass spectrometry, and site-directed mutagenesis. Intracellular phosphorylation of βAPP was inhibited by Brefeldin A and by incubating cells at 20°C, thus excluding phosphorylation in the endoplasmic reticulum or trans-Golgi network. Ectodomain phosphorylation within a post-Golgi compartment occurred not only with mutant Swedish βAPP, but also with wild type βAPP. In addition to phosphorylation within a post-Golgi compartment, βAPP was also found to undergo phosphorylation at the cell surface by an ectoprotein kinase. Therefore, this study revealed two distinct cellular locations for βAPP phosphorylation.

spanning ␤-amyloid precursor protein (␤APP; Ref. 3). ␤APP can be proteolytically processed within two general pathways: an amyloidogenic and a nonamyloidogenic processing route (summarized by Haass and Selkoe (4)). Within the latter pathway, ␤APP is constitutively cleaved by a protease referred to as ␣-secretase. This cleavage occurs near the middle of the A␤ region, thus inhibiting A␤ formation (5,6) and resulting in the secretion of APP s wt -␣ (for terminology, see Fig. 2A) into the media of cultured cells (7). In the amyloidogenic pathway, ␤APP is first cleaved by ␤-secretase at the N terminus of the A␤ domain and subsequently by ␥-secretase at its C terminus, resulting in the constitutive secretion of A␤ (8 -11).
One cellular mechanism for the generation of A␤ involves reinternalization of full-length ␤APP from cell surface to endosomes (12), in which the ␤-secretase cleavage can occur (13). During recycling of endosomes to the cell surface (13), the resulting 12-kDa C-terminal fragment is cleaved by ␥-secretase to release A␤. Missense mutations, found in a few families with familial autosomal dominant AD, frame the A␤ domain (reviewed by Mullan and Crawford (14)). All familial autosomal dominant AD-linked mutations found in the ␤APP gene have now been shown to influence directly A␤ generation. A mutation just before the N terminus of the A␤ region at the ␤-secretase cleavage site (the "Swedish" mutation; Ref. 15) results in a 3-6-fold increased production of A␤ (16 -18). Missense mutations close to the ␣-secretase site also cause an increased production of A␤, but the increase is paralleled by alternative N-terminal cleavages of A␤ (19). Mutations at the C terminus of the A␤ domain (just after the ␥-secretase site) result in the generation of longer A␤ peptides ending at amino acid 42 instead of amino acid 40 (20). The former peptides have been shown to aggregate more rapidly (21), presumably leading to an accelerated amyloid plaque formation.
Recently, we (1) and others (22) showed that the increased production of A␤ from ␤APP molecules bearing the Swedish mutation is due to a cellular mechanism distinct from that principally involved in A␤ generation from wild type ␤APP. ␤-Secretase cleavage of ␤APP appears to generally occur within the endocytic pathway (13). During reinternalization, only small amounts of full-length, uncleaved ␤APP molecules are available, because substantial quantities of ␤APP have already been cleaved by ␣-secretase. However, in the case of Swedish mutant ␤APP, we found that ␤-secretase cleavage occurs at an earlier time point in ␤APP trafficking, namely within the secretory pathway on the way to the cell surface, predominantly in a post-Golgi compartment, most likely secretory vesicles (1). Therefore, ␤-secretase cleavage of Swedish mutant ␤APP, in contrast to the principal ␤-secretase cleavage of wild type ␤APP, occurs in competition with ␣-secretase cleavage in the secretory pathway.
␤APP matures by undergoing NЈ-and OЈ-glycosylation, sulfation, and phosphorylation during transport from the endoplasmic reticulum to the cell surface (7,8,23). Protein phosphorylation is known to be involved in the regulation of cellular processes such as differentiation, metabolism, and signal transduction (for review, see Ref. 24). Besides many intracellular phosphoproteins, some phosphorylated secretory proteins have been described, e.g. fibronectin (25), fibrinogen (26), prolactin (27), chromogranin B, secretogranin II (28), and L-29, a soluble lectin (29). Although the cellular locus for phosphorylation of most of these secretory proteins is not identified, it has been shown that chromogranin B and secretogranin II are phosphorylated in the secretory pathway within the transcisternae of the Golgi (28).
In addition to numerous intracellular protein kinases, ectoprotein kinases acting at the surface of intact cells have been characterized (30 -32). These enzymes use extracellular ATP as cosubstrate to phosphorylate endogenous cell surface proteins as well as soluble proteins and have been implicated in a number of biological phenomena, including cell growth inhibition (33), long-term potentiation in neurons and synaptogenesis (34,35), and parasite-host interactions (36,37). Ubiquitously occurring casein kinase-like ectoprotein kinases can be released from the cell surface upon interaction with extracellular protein substrates (38,39), thus allowing them to act at a distance to their cellular origin.
In this study, we have determined the subcellular locations of the phosphorylation of ␤APP. We used an antibody (192sw (40); see Fig. 2A) specifically detecting APP s sw -␤, the derivative we found to be generated in high quantities within a well defined post-Golgi secretory compartment (1). Through biochemical and cell biological experiments we demonstrate that intracellular phosphorylation of Swedish ␤APP as well as wild type ␤APP occurs within this compartment, i.e. after the trans-Golgi, most likely within secretory vesicles. Ectodomain phosphorylation was mapped to Ser 198 and Ser 206 of ␤APP695, which represent potential phosphorylation sites for casein kinase (CK)-2 and CK-1, respectively. Further, we show that ␤APP can be phosphorylated by an ectoprotein kinase activity on the cell surface. Therefore, our data demonstrate that ␤APP undergoes ectodomain phosphorylation at two distinct cellular locations.

MATERIALS AND METHODS
Cell Culture, Metabolic Labeling, and Drug Treatment-Kidney 293 cells were stably transfected with the wt ␤APP 695 cDNA (9,41) or with the ␤APP 695 cDNA containing the Swedish double mutation (18). Chinese hamster ovary cells stably transfected with the amyloid precursorlike protein 2 (APLP2) cDNA have been described previously (42). Metabolic labeling and treatment of cells with 10 g/ml of Brefeldin A (BFA; solubilized in ethanol) was carried out as described earlier (23,43,44). Ethanol was added in identical concentrations to the corresponding control cells.
Incubation of Cells at 20°C-Cells were incubated at 20°C as described (1,45). Briefly, cells were metabolically labeled with 150 Ci of [ 35 S]methionine or 1.5 mCi of [ 32 P]orthophosphate for 3 h in methionine-free or sodium phosphate-free Dulbecco's minimal essential medium buffered with 10 mM HEPES. Tissue culture dishes were sealed with Parafilm and incubated in a water bath at 20°C or 37°C. The temperature was controlled carefully throughout the experiment.
Pulse-chase Experiments-Pulse-chase experiments were carried out as described (43). Briefly, cells stably transfected with the Swedish ␤APP mutation were pulse-labeled with [ 35 S]methionine for 5 min in methionine and serum-free media. Cells were than chased for the indicated time points in media containing excess amounts of methionine and 10% fetal calf serum. Cell lysates were immunoprecipitated with antibody C7 (to detect full-length ␤APP) and antibody 192sw (to detect intracellular APP s sw -␤). Media were immunoprecipitated with antibody 192sw (to detect secreted APP s sw -␤). Phosphoamino Acid Analysis-Phosphoamino acid analysis was carried out by two-dimensional high voltage electrophoresis (47). Radiolabeled proteins electrotransferred onto polyvinylidene difluoride-membrane were hydrolyzed in 6 M HCl for 90 min at 110°C. Subsequently, supernatants were dried in a SpeedVac concentrator, and pellets were dissolved in 5 l of pH 1.9 buffer (7.8% acetic acid, 2.5% formic acid) and spotted onto cellulose-TLC plates together with unlabeled phosphoamino acids (Ser(P), Thr(P), and Tyr(P); 1 g each). High voltage electrophoresis was carried out for 20 min (pH 1.9 buffer) at 1.5 kV and for 16 min (pH 3.5 buffer; 5% acetic acid, 0.5% pyridine) at 1.3 kV, respectively. Radioactive phosphoamino acids were identified by autoradiography and comparison with ninhydrin-stained standards.
Matrix-assisted Laser Desorption/Ionization-Mass Spectrometry-Approximately 10 g of unlabeled ␤APP together with a trace of 32 Plabeled ␤APP were digested with trypsin, and the resulting peptides were separated on 10 -20% Tris/Tricine gels as described. Radiolabeled peptide bands were cut out from the gel, extracted twice for 10 min with 100 l of 0.1% aqueous trifluoroacetic acid followed by 100 l of 60% acetonitrile. The combined supernatants were subjected to a 5-mm micro precolumn (LC Packings) packed with Poros R2 (Perseptive Biosystems). The peptides were eluted in 10 l with a step gradient of 80% acetonitrile, 0.1% trifluoroacetic acid. Molecular masses (isotopic average) of the eluted peptides were determined by a Vision 2000 (Finnigan) mass spectrometer equipped with a nitrogen laser and operated in reflection mode at an accelerating voltage of 5000 V. 1 l of the peptide solution was crystallized in matrices consisting of 1% 2,4-dihydroxybenzoic acid in 0.1% aqueous trifluoroacetic acid. All peptide spectra were externally calibrated by using the monoisotopic masses of sodium (M r 23.0) and fullerene C70 (M r 840.0). Peptides were identified by computer-assisted analysis using the Swiss-Prot sequence data bank and the special program package HUSAR (developed at the Department of Molecular Biophysics, German Cancer Research Center, Heidelberg).
Phosphorylation was started by the addition of 0.5-1.5 M [␥-32 P]ATP and allowed to proceed for 0 -30 min at 37°C. Reactions were terminated by removing cell supernatants followed immediately by two washes of the cells with ice-cold phosphorylation buffer containing 2 mM unlabeled ATP. Subsequently, cells were lysed in presence of 2 mM ATP for 7 min on ice. Cell lysates (prepared as described by Haass et al. (43)) were centrifuged for 10 min at 14,000 ϫ g, and cellular ␤APP was isolated by immunoprecipitation as described above and separated by SDS-PAGE. Radiolabeled proteins were detected by autoradiography of dried gels. Cell viability during phosphorylation assays was evaluated by several criteria (49).
In Vitro Mutagenesis-The ␤APP cDNA construct containing a stop codon at the ␣-secretase cleavage site was described previously (23). The C-terminal deletion construct of ␤APP was described by Haass et al. (44). A cDNA construct containing a stop codon at the ␤-secretase site of wt ␤APP was generated as described (23) using the following annealed oligonucleotides: GATCTCTGAAGTGAAGATGTAGGCAG (stop ␤-wt sense) and AATTCTGCCTACATCTTCACTTCAGA (stop ␤-wt antisense).
The cDNA construct containing a stop codon at the ␤-secretase site of Swedish ␤APP was generated as described (23) using the following annealed oligonucleotides: GATCTCTGAAGTGAATCTGTAGGCAG (stop ␤-sw sense) and AATTCTGCCTACAGATTCACTTCAGA (stop ␤-sw antisense).
The corresponding cDNAs were stably transfected into kidney 293 cells as described (1,51), and single cell clones were isolated using cloning cylinders (51).
All mutations were confirmed by sequencing both DNA strands.

RESULTS
Ectodomain Phosphorylation of APLP2-In order to obtain a general validation of ectodomain phosphorylation of ␤APP (23), we examined the phosphorylation of the highly related APLP2. APLP2, APLP1, and ␤APP are members of a conserved gene family of homologous proteins (52)(53)(54)(55). APLP2 is particularly similar to ␤APP because it shares some of its characteristic biochemical properties and also matures through the constitutive secretory pathway, where its ectodomain is secreted into culture media (42,55,56). To analyze the potential phosphorylation of APLP2, Chinese hamster ovary cells stably transfected with the APLP2 cDNA were metabolically labeled with [ 35 S]methionine or [ 32 P]orthophosphate. Conditioned media were precipitated with antibody D2-1 raised against fulllength mouse APLP2 (42). As shown in Fig. 1, immunoprecipitation of conditioned media from [ 35 S]methionine labeled cells resulted in the detection of the two major APLP2 species. In close agreement with the data reported (42), we observed a high molecular weight species that corresponds to the chondroitin sulfate glycosaminoglycan-modified form of APLP2 and a lower molecular weight species representing the unmodified form of APLP2. Both forms of APLP2 were also observed after labeling with [ 32 P]orthophosphate (Fig. 1). These results show that APLP2, similar to ␤APP, is phosphorylated within its ectodomain, indicating that ectodomain phosphorylation of secreted derivatives of proteins belonging to the APP gene family is a general phenomenon.
Stability of Ectodomain Phosphorylation of ␤APP-To assess the stability of ␤APP phosphorylation, we examined protein phosphatase activity in AD brain extracts as well as in the conditioned media of cultured cells (57,58). No ␤APP dephos-phorylating activity was detected in any of the AD brain extracts, both those from temporal and occipital regions, indicating a relative resistance of ␤APP to protein phosphatase activity. In addition, APP s did not undergo dephosphorylation in conditioned media (data not shown). These experiments demonstrate that ectodomain phosphorylation of ␤APP is relatively resistant to protein phosphatase activities, suggesting a long lasting biological function of phosphorylated APP s molecules.
Intracellular Phosphorylation of Swedish Mutant ␤APP-Little is known about the cell biology of ectodomain phosphorylation. In order to determine the subcellular locus for ␤APP phosphorylation, we used an antibody (192sw; Fig. 2A) that specifically recognizes APP s sw -␤, which we previously detected in high quantities within the lysates of kidney 293 cells stably transfected with the Swedish ␤APP cDNA (1). To determine if ectodomain phosphorylation also occurs on intracellular APP s sw -␤, we radiolabeled kidney 293 cells expressing Swedish mutant ␤APP with [ 32 P]orthophosphate. Upon immunoprecipitation of cell lysates and media we detected phosphorylated intracellular and secreted APP s sw -␤ as well as phosphorylated intracellular full-length ␤APP (Fig. 2B). This result indicates the occurrence of intracellular phosphorylation of the ␤APPectodomain. To prove that APP s sw -␤ was indeed produced de novo and not taken up by fluid phase endocytosis, we pulselabeled kidney 293 cells stably transfected with the Swedish cDNA. The cells were then chased in the presence of excess unlabeled methionine. Aliquots of the cell lysates were immunoprecipitated either with antibody C7 (to detect maturation of full-length ␤APP) or with antibody 192sw (to detect intracellular APP s sw -␤). In addition, conditioned media were immunoprecipitated with antibody 192sw to detect secreted APP s sw -␤. As shown in Fig. 3, full-length ␤APP is processed within 45 min from immature NЈ-glycosylated form to mature NЈ-and OЈglycosylated form. Shortly after, the amount of full-length ␤APP declines due to the secretion of APP s . Consistent with our previous results (1), the highest level of intracellular APP s sw -␤ was detected after 45 min. After this time point the levels of intracellular APP s sw -␤ declined, and an increase of secreted APP s sw -␤ in the media was observed (Fig. 3). The precursor product relationship clearly indicates that intracellular APP s sw -␤ is produced de novo and not due to a fluid phase mediated uptake of secreted species.
Mapping of Phosphorylation Sites within ␤APP-To determine which amino acids were phosphorylated in Swedish mutant ␤APP, we performed phosphoamino acid analysis of intracellular as well as secreted APP s sw -␤. Both species are phosphorylated exclusively on serine residues (Fig. 4). This result is in line with recent studies showing that wt ␤APP is constitutively phosphorylated solely on serine residues (23). It also confirms that phosphorylation of intracellular APP s sw -␤ is an amino acid phosphorylation, not an incorporation of phosphate into sugar moieties of ␤APP.
In order to identify the site(s) of ␤APP phosphorylation, we performed tryptic peptide mapping of in vivo phosphorylated ␤APP molecules. Kidney 293 cells stably transfected with wild type ␤APP695 or cDNA constructs deleting large portions of the N-terminal half (AX construct (23) (Fig. 5A) or the Cterminal half (XB construct (23)) were labeled with [ 32 P]orthophosphate or [ 35 S]methionine. Secreted forms of the respective ␤APP molecules were immunoprecipitated with antibody 1736. In agreement with data published earlier (23), we found that phosphorylation occurs exclusively within the Nterminal portion of ␤APP, since no phosphate incorporation occurred in cells expressing the N-terminal deletion construct (Fig. 5B). Phosphorylated full-length ␤APP as well as the phosphorylated C-terminal deleted ␤APP (XB) were digested with trypsin, and the digestion products were separated on a 10 -20% Tris/Tricine gel. A single phosphorylated peptide of approximately 4.8 kDa was detected for both full-length ␤APP and XB constructs (Fig. 5C). Computer analysis of the potentially generated tryptic peptides revealed that the radiolabeled peptide could represent solely the amino acid sequence from 181-224 of ␤APP 695 . To prove this in more detail, the radiolabeled ϳ4.8-kDa peptide was eluted from Tris/Tricine gel and subjected to matrix-assisted laser desorption/ionization-mass spectrometry (see "Materials and Methods"). Three monoisotopic masses of 2286.5, 3673.5, and 4877.3 (Ϯ 10) were detected in the eluate. The masses of 2286.5 and 3673.5 could not be matched to tryptic peptides of ␤APP and presumably represent peptides of autocatalytically cleaved trypsin, migrating close to the phosphorylated ␤APP tryptic peptide. In contrast, the mass of 4877.3 matches that of the sequence of amino acids 181-224 of ␤APP695 in a double phosphorylated form (4714.7 ϩ 160 Da). Since the amino acid sequence of this peptide contains four serine residues, we searched for putative phosphor acceptor sites by computer-assisted analysis. Serine residues 198 and 206 were identified within an acidic sequence of this peptide, representing potential phosphorylation sites for CK-2 and CK-1, respectively (Fig. 5D). These serines were therefore mutagenized to alanines, and the corresponding cDNA constructs were stably transfected into kidney 293 cells. Single cell clones were metabolically labeled with [ 32 P]orthophosphate or [ 35 S]methionine, and secreted ␤APP s was immunoprecipitated from conditioned medium with antibody B5. Phosphate incorporation was quantified by phosphor imaging. As shown in Fig.  5E, phosphorylation of ␤APP containing the S198A mutation was reduced by about 80%, while that of the S206A mutation was reduced by about 15%. Similar data were obtained after immunoprecipitation of full-length ␤APP from cell lysates (data not shown). Taken together, these data might therefore indicate that both serines represent in vivo phosphorylation sites (see "Discussion" for details).
Phosphorylation of ␤APP Occurs within Golgi-derived Vesicles-Ectodomain phosphorylation of ␤APP was found on all types of secreted APP s molecules, regardless of whether Swedish or wt ␤APP was cleaved at either the ␣or the ␤-secretase site. To produce APP s molecules with defined C termini corresponding to ␣or ␤-secretase-cleaved APP s wt/sw , we stably transfected kidney 293 cells with cDNA constructs containing stop codons at sites corresponding to these scissions. These transfectants were then metabolically labeled with [ 35 S]methionine or [ 32 P]orthophosphate, and their conditioned media were precipitated with antibody B5, which detects all secreted APP s species. As shown in Fig. 6, APP s wt -␣, APP s wt -␤, and APP s sw -␤ were each secreted as phosphorylated species. Thus, membrane insertion of ␤APP is not necessary for its phosphorylation, and APP s can be phosphorylated regardless of which secretase activity cleaved the precursor, indicating a general cellular mechanism for the ectodomain phosphorylation of mutant and wt ␤APP.
To determine whether phosphorylation of ␤APP occurs in the same compartment as the ␤-secretase cleavage of Swedish ␤APP (1), we investigated the effect of BFA on phosphorylation of Swedish ␤APP. BFA is known to cause a collapse of the Golgi network, resulting in a block of forward transport at the cis-Golgi compartment (61). Kidney 293 cells stably transfected with Swedish ␤APP were metabolically labeled with either [ 35 S]methionine or [ 32 P]orthophosphate in the absence or presence of BFA. Cell lysates were precipitated with antibody C7 (to detect full-length ␤APP) or antibody 192sw (to detect intracellular APP s sw -␤), and conditioned media were precipitated with antibody 192sw (to detect secreted APP s sw -␤). As reported previously, BFA treatment not only inhibited the maturation of full-length ␤APP but also completely inhibited the generation of intracellular APP s sw -␤ and its secretion ( Fig. 7A; Refs. 1 and 44)). Treatment with BFA also resulted in an inhibition of ␤APP ectodomain phosphorylation (Fig. 7B), clearly showing that phosphorylation does not occur within the endoplasmic reticulum or the early Golgi. The trace amounts of phosphorylated species detected after BFA treatment are due to ␤APP molecules that escaped the BFA block at the beginning of the experiment.
To determine whether ectodomain phosphorylation of ␤APP occurs within the trans-Golgi network, kidney 293 cells expressing Swedish ␤APP were incubated at 20°C. Under such conditions, membrane proteins accumulate within the trans-Golgi network (45). As reported previously (1) incubation at 20°C resulted in the accumulation of full-length NЈ-and OЈglycosylated ␤APP within cell lysates; no APP s sw -␤ was detected in cell lysates or conditioned media ( Fig. 8A; Ref. 1). As shown above, after labeling with [ 32 P]orthophosphate at 37°C, mature phosphorylated ␤APP was precipitated from cell lysates and phosphorylated APP s sw -␤ from both lysates and media (Fig.  8B). In contrast, incubation of cells at 20°C completely inhib- FIG. 5. Identification of the phosphorylation sites of ␤APP within its ectodomain. A, schematic of wild type ␤APP (WT) and the AX and XB constructs, missing large portions of the N-terminal and the C-terminal half of the ␤APP ectodomain, respectively. The A␤-domain is represented by a striped bar, and vertical lines represent cellular membranes. The numbers above denote amino acid residues with the restriction sites used to generate the constructs indicated (23). B, kidney 293 cells stably transfected with wild type (WT) ␤APP695, AX, or XB were labeled with [ 35 S]methionine ( 35 S) and [ 32 P]orthophosphate ( 32 P), respectively, and conditioned media were immunoprecipitated with antibody 1736. Radiolabeled proteins were visualized by autoradiography after separation by SDS-PAGE. C, phosphopeptide map of radiolabeled, secreted forms of wild type (WT) ␤APP or XB. After SDS-PAGE, proteins were transferred to nitrocellulose membrane and digested with trypsin as described under "Materials and Methods." The resulting tryptic peptides were separated on a 10 -20% Tris/Tricine gel and analyzed by autoradiography. The position of the phosphorylated 4.8-kDa peptide is marked by an arrowhead. D, amino acid sequence of the phosphorylated tryptic peptide (amino acids 181-224), which was identified by mass spectrometry and computer-assisted analysis (see "Materials and Methods"). The serine residues representing potential phosphorylation sites of CK-1 (Ser 206 ) and CK-2 (Ser 198 ) are shown in boldface letters. E, quantification of in vivo phosphorylation of wild type ␤APP (WT) and ␤APP carrying serine to alanine mutations at positions 198 (S198A) and 206 (S206A). Kidney 293 cells stably expressing wild type or mutated forms of ␤APP (S198A, S206A) were labeled with [ 35 S]methionine or [ 32 P]orthophosphate for 2 h. Quantification of protein expression and phosphate incorporation in the different forms of ␤APP were carried out by phosphor imaging. Bars represent means Ϯ S.E. of three independent experiments. ited phosphorylation of full-length ␤APP (Fig. 8B), although large amounts of full-length ␤APP were present as shown by labeling with [ 35 S]methionine (Fig. 8A). Taken together, these data strongly suggest that the intracellular ectodomain phosphorylation of Swedish ␤APP occurs within a post-Golgi compartment, most likely secretory vesicles, and not in the trans-Golgi network itself.
Analogous experiments were then carried out to determine the cellular locus of the ectodomain phosphorylation of wt ␤APP. When kidney 293 cells expressing wt ␤APP were labeled at 37°C with [ 35 S]methionine, antibody C7 precipitated the expected doublet of full-length ␤APP from cell lysates representing the immature and mature forms of the precursor (Fig.  8C). Precipitation with antibody 1736, which specifically identifies APP s wt -␣ and does not cross-react with full-length ␤APP or APP s wt -␤, results in the detection of intracellular APP s wt -␣ from cell lysates as well as secreted APP s wt -␣ from conditioned media (Fig. 8C). The detection of intracellular APP s wt -␣ is in good agreement with data published previously (51,59,60), indicating ␣-secretase cleavage within the secretory pathway. When cells were incubated at 20°C, an accumulation of mature ␤APP was observed; however, the generation of intracellular APP s wt -␣ and consequently its secretion was completely inhibited (Fig. 8C). When cells were metabolically labeled with [ 32 P]orthophosphate at 37°C, we detected mature phosphorylated full-length ␤APP, and precipitation of cell lysates with antibody 1736, specific for APP s -␣, resulted in the detection of intracellular phosphorylated APP s wt -␣ (Fig. 8D). However, incubating the cells at 20°C completely inhibited phosphorylation of wild type ␤APP; no phosphorylated full-length ␤APP or intracellular and secreted APP s wt -␣ was detected (Fig. 8D). Taken together, these data show that intracellular ectodomain phosphorylation of wild type as well as Swedish ␤APP occurs within a post-Golgi compartment, most likely within secretory vesicles, suggesting that this compartment represents a general subcellular site of ectodomain phosphorylation of ␤APP.
Ectodomain Phosphorylation Can Occur on the Cell Surface-Because mature full-length ␤APP is also present at the cell surface, we examined whether membrane-bound ␤APP can be a substrate for ectoprotein kinases. Intact kidney cells, transfected with wild type ␤APP cDNA, were incubated in the presence of 1 M [␥-32 P]ATP in the cell supernatant, allowing specific detection of ectoprotein kinase activity (30). Full-length ␤APP was then precipitated from cell lysates and APP s wt -␣ from cell supernatants. As shown in Fig. 9A (wt) cell surface-bound full-length ␤APP was phosphorylated by ectoprotein kinase activity. Moreover, phosphorylated APP s wt -␣ was recovered The relatively higher amounts of APP s wt -␣ in supernatants from transfectants expressing ⌬C-␤APP as compared with that from transfectants expressing wt-␤APP is due to a higher rate of ␣-secretase cleavage, which is in close agreement with previous results (13,44). C, phosphorylation of soluble APP s -␣ by ectoprotein kinase on the surface of kidney 293 cells. APP s -␣ was collected from supernatants of kidney 293 cells stably transfected with a cDNA construct containing a stop codon corresponding to the ␣-secretase site (compare Fig. 4) for 1 h. The supernatant was taken off and split into two halves. One half was incubated with untransfected kidney 293 cells (ϩ Cells), and the other half was incubated in a Petri dish without cells (Ϫ Cells). Both dishes were incubated for 15 min at 37°C in the presence of 1 M [␥-32 P]ATP. APP s -␣ was immunoprecipitated with antibody 1736 and separated by SDS-PAGE. D, two-dimensional phosphoamino acid analysis of cell surface ␤APP, showing that ␤APP is exclusively phosphorylated on serine residues by ectoprotein kinase. from cell supernatants (Fig. 9A, Media). Similar experiments with kidney 293 cells expressing Swedish ␤APP showed that cell surface ␤APP sw is also phosphorylated by ectoprotein kinase activity (data not shown). Cell surface phosphorylation was also investigated with cells expressing a C-terminal truncated form of ␤APP, which inserts in cell membranes but does not undergo reinternalization (13,44). As with full-length ␤APP (Fig. 9A), the C-terminal truncated form of ␤APP was also phosphorylated (Fig. 9B, ⌬C), indicating that reinternalization of ␤APP is not necessary for its phosphorylation. Again, phosphorylated APP s wt -␣ was recovered from cell supernatants (Fig. 9B, Media). To prove whether phosphorylated APP s wt -␣ does exclusively derive from its phosphorylated precursor or if soluble APP s wt -␣ can be phosphorylated after proteolytic cleavage, cell-free supernatant containing APP s wt -␣ was incubated with [␥-32 P]ATP either in the absence or in the presence of untransfected intact kidney 293 cells. As shown in Fig. 9C, APP s wt -␣ was phosphorylated only in the presence of intact cells, indicating that soluble APP s wt -␣ could serve as a substrate for ectoprotein kinase. Thus, neither membrane insertion nor reinternalization is necessary for ␤APP phosphorylation. However, ␤APP was not phosphorylated in the absence of cells (Fig.  9C, Cells), showing that ectoprotein kinase activity is not cosecreted with ␤APP s -species. As revealed by two-dimensional phosphoamino acid analysis, phosphorylation of ␤APP by ectoprotein kinase occurs exclusively on serine residues (Fig. 9D). The results clearly demonstrate that cell surface-bound ␤APP and its soluble derivatives can be phosphorylated by membrane-associated ectoprotein kinase on the surface of intact cells.

DISCUSSION
In summary, our data show that full-length ␤APP and its ␣and ␤-secretase-cleaved derivatives can be phosphorylated at two different subcellular locations. In both cases, ␤APP is exclusively phosphorylated on its ectodomain but not in the cytoplasmic tail. Ectodomain phosphorylation of ␤APP has been demonstrated previously (23) and was further supported by the data presented here; ␤APP can be phosphorylated on the cell surface by incubating cells with [␥-32 P]ATP, and secreted APP S derived from recombinant cDNA constructs with stop codons inserted at the ␣and ␤-secretase site of mutant and wild type ␤APP still result in the secretion of phosphorylated APP S . Moreover, APP S incubated with living cells is phosphorylated by a cell surface ectoprotein kinase. Therefore, evidence from multiple experiments proves exclusive ectodomain phosphorylation of ␤APP. Intracellular APP s and full-length ␤APP molecules are phosphorylated within a post-Golgi compartment, most likely secretory vesicles. This is the cellular compartment to which we have localized the ␤-secretase activity cleaving Swedish ␤APP (1). Therefore, phosphorylation of APP s occurs during or immediately before or after the secretory cleavages of ␤APP.
The in vivo phosphorylation sites of ␤APP were identified as serine residues 198 and 206 by phosphopeptide mapping, sitedirected mutagenesis, and mass spectrometry. Moreover, in vivo secreted APP s was detected exclusively in double phosphorylated form. Ser 198 is followed by acidic amino acid residues and therefore represents a putative phosphorylation site for CK-2 (63), while Ser 206 is preceded by an acidic domain and represents a CK-1 phosphorylation site (64). However, individual mutations of Ser 198 and Ser 206 differently affected the phosphate incorporation. The S198A mutation resulted in a reduction of phosphorylation of about 80%, while the S206A mutation reduced phosphorylation by about 15%. This might be explained by sequential phosphorylation events, in which the first phosphorylation at Ser 198 facilitates the subsequent phos-phorylation at Ser 206 by acidifying this domain. A similar process has been described involving protein kinases A and CK-1 (65,66).
Interestingly, in addition to the intracellular phosphorylation, our data also demonstrate a second cellular site for phosphorylation of membrane-bound ␤APP: an ectoprotein kinase activity at the cell surface. In contrast to the intracellular phosphorylation of ␤APP, which appears to be a constitutive event (23), phosphorylation by ectoprotein kinases could represent a regulated mechanism. Because ATP is known to be released into the extracellular environment by a variety of cellular stimuli (for review see Refs. 67 and 68), the availability of this cosubstrate for ectoprotein kinases could represent a biological regulation mechanism for phosphorylation of cell surface ␤APP. Since ␣-secretase activity is present within cell lysates (51,59,60), as well as on the cell surface (12,62), full-length surface ␤APP will contribute to the pool of phosphorylated APP s molecules in conditioned media. In addition, secreted derivatives of ␤APP (APP s wt and APP s sw ) released by ␣or ␤-secretase into the cell supernatant also serve as substrates for ectoprotein kinase. Our study demonstrates for the first time the unusual phenomenon that ␤APP and its principal secreted derivatives can undergo selective ectodomain phosphorylation at two distinct subcellular locations. It will now be important to determine whether both mechanisms result in the phosphorylation of identical amino acid residues or if ␤APP is phosphorylated by different protein kinases on two or more sites within the same molecule. The functional consequences of this complex regulation of ␤APP ectodomain phosphorylation are unknown so far. However, one might speculate that extracellular function(s) of ␤APP, e.g. the modulation of neuronal excitability by APP s (69), could be regulated by selective ectodomain phosphorylation.