Alzheimer’s Disease (cid:98) A4 Protein Release and Amyloid Precursor Protein Sorting Are Regulated by Alternative Splicing*

We show here that alternative splicing influences the polarized secretion of amyloid precursor protein (APP) as well as the release of its proteolytic 3–4-kDa frag- ments (cid:98) A4 and p3. In Madin-Darby canine kidney II cells stably transfected with various APP isoforms and APP mutants, APPsec was consistently secreted basolaterally. In contrast, Madin-Darby canine kidney II cells transfected with L-APP677, which occurs naturally by alternative splicing of exon 15, secreted this isoform both apically and basolaterally, while maintaining the basolateral sorting of endogenous APPsec. This suggests that the alternative splicing of APP exon 15 modulates the polarized sorting of secretory APP. The same alter- native splicing event also decreased the production of (cid:98) A4 relative to p3. This is the first example of alternative splicing regulating polarized trafficking of a secretory protein. The Alzheimer (cid:98) A4-amyloid precursor protein (APP) 1 (1) is a ubiquitously expressed transmembrane glycoprotein (2, 3). It occurs in eight different isoforms that are generated by alternative splicing of the APP exons 7,

We show here that alternative splicing influences the polarized secretion of amyloid precursor protein (APP) as well as the release of its proteolytic 3-4-kDa fragments ␤A4 and p3. In Madin-Darby canine kidney II cells stably transfected with various APP isoforms and APP mutants, APPsec was consistently secreted basolaterally. In contrast, Madin-Darby canine kidney II cells transfected with L-APP677, which occurs naturally by alternative splicing of exon 15, secreted this isoform both apically and basolaterally, while maintaining the basolateral sorting of endogenous APPsec. This suggests that the alternative splicing of APP exon 15 modulates the polarized sorting of secretory APP. The same alternative splicing event also decreased the production of ␤A4 relative to p3. This is the first example of alternative splicing regulating polarized trafficking of a secretory protein.
The Alzheimer ␤A4-amyloid precursor protein (APP) 1 (1) is a ubiquitously expressed transmembrane glycoprotein (2,3). It occurs in eight different isoforms that are generated by alternative splicing of the APP exons 7, 8, and 15 (4). APP lacking the residues encoded by exon 15 is denoted L-APP (5). L-APP isoforms have been found to be the major isoform in several different tissues (4). While exon 15 is part of the divergent region of the APP family, a similar splice pattern exists for another member of the APP family, the APLP2 gene (6,7). In both proteins, alternative splicing may result in the addition of a chondroitin sulfate side chain (8 -10). All APP isoforms including L-APP are converted to secretory forms by as yet unidentified proteases termed APP secretases. Secretase activities are involved in release of APPsec as well as in the production of a 4-kDa (␤A4) and 3-kDa (p3) peptides (for a recent review, see Refs. 11 and 12). ␤A4 is the principal component of the amyloid plaques found in Alzheimer's disease brains. Point mutations in the APP gene increasing the amount or the length of ␤A4 released segregate with familial disease and lead to Alzheimer's disease at a certain age of the affected person, demonstrating an association between ␤A4 production and Alzheimer's disease pathogenesis.
Here, we have analyzed the influence of splicing on APP sorting and ␤A4 release. For this purpose we used MDCK cells, which are a widely used model for investigations of polarized secretion of proteins. Two cellular domains can be distinguished: the basolateral side, which in the kidney would be exposed to the basal membrane and is engaged in cell-matrix interactions, and the apical side, which would be exposed to the kidney tubular lumen. Sorting of transmembrane proteins, including APP (13)(14)(15) to the basolateral domain generally appears to be dominantly determined by short cytoplasmic signals. However, identification of the signals required for basolateral targeting of soluble proteins or for apical sorting in general, with the exception of glycosylphosphatidylinositolanchored proteins, has been a long and unsuccessful process (for a recent review, see Refs. 16 and 17).
The APP cytoplasmic signal including the QYTSI motif has been shown to result in efficient sorting of transmembrane APP to the basolateral side (14). Removal of this signal causes the transmembrane APP to become distributed equally on both sides and concomitantly increases the release of ␤A4 to the apical rather than the basolateral side, to which it is usually restricted. However, this mutation does not affect the polarized secretion of APPsec to the basolateral domain, suggesting that an additional determinant of basolateral sorting is present in the luminal domain and that the majority of APPsec is generated intracellularly (14,15,18).
In this study, we investigated the polarized secretion of an alternatively spliced form of APP, L-APP. Surprisingly, we found that L-APPsec is sorted differently than is APPsec. Furthermore, we show that this alternative splicing also results in a decrease in the generation of ␤A4 relative to p3.

MATERIALS AND METHODS
DNA-Human APP cDNAs APP695, APP751, APP770, L-APP677, L-APP677A, L-APPex16A STOP, APP695⌬CHO, APP695␥, and SPA4CT (3) were cloned into the vector pHD (19) for MDCK studies. Human APP cDNAs APP695, APP751, and L-APP677 were also cloned into the vector pCEP4 (Invitrogen) for COS cell studies. L-APP677A and L-APPex16A STOP , containing a Ser to Ala, respectively, Ser to Leu substitution at position 563 (L-APP677 numbering), was generated by polymerase chain reaction-mediated mutagenesis. APPex16A STOP , contains an artificial stop codon at position 565 (L-APP677 numbering). The sequence of the polymerase chain reaction-amplified regions was * This work was supported by SFB 317, the Bundesministerium fü r Bildung und Forschung of Germany through Grant 030666A, Fonds der Chemischen Industrie, the Forschungsschwerpunkt BadenWü rttenberg, and the National Health Medical Research Council of Australia. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
§ Both authors were funded by the Deutsche Forschungsgemeinschaft (Graduiertenkolleg Neurobiologie Universitä t Heidelberg).
¶ To whom correspondence should be addressed: Zentrum fü r Molekulare Biologie University of Heidelberg, D-69120 Heidelberg, INF 282. Tel.: 49-6221-546847; Fax: 49-6221-545891; E-mail: iy9@ix.urz. uni-heidelberg.de. 1 The abbreviations used are: APP, amyloid precursor protein (for convenience, all APP isoforms containing the residues encoded by exon 15 will be referred to as APP, those devoid of these residues as L-APP); MDCK, Madin-Darby canine kidney; APLP2, amyloid precursor like protein; GAG, glycosaminoglycan. confirmed by sequencing both DNA strands of the subcloned polymerase chain reaction product. APP695␥ was generated by EcoR I and ClaI digestion of pBluescript ␤A4 (20, 21) and subcloning of the resulting fragment into pBluescript APP695. This cDNA thus ends with ␤A4 residue 42. APP␥ encoded by this construct has been shown not to integrate into the cell membrane and to be readily secreted (21,22); L-APPex16A STOP ends in APP-exon 16, does not integrate into the cell membrane, and will be readily secreted. APP695⌬CHO was constructed by fusion of the first PvuII site to the EcoRI site inside of the coding region of APP695. 2 Cell Culture and Transfections-All cell culture media were obtained from Sigma. MDCK II cells were obtained from Kai Simons (EMBL, Heidelberg, Germany) and cultivated as described (23). The pHD vector containing APP cDNA was cotransfected with the selectable vector pSV2neo at a ratio of 1:20 into MDCK II cells using Lipofectin (Life Technologies, Inc.). Stable transfectants were selected in 300 g/ml gentamycin (Life Technologies, Inc.) and subcloned. A minimum of three independent clones expressing the transfected protein in roughly equal amounts to the endogenous APP were selected for further analysis, except for L-APPex16A STOP (two clones). COS 7 cells were cultivated in MEM/F-12 medium containing 10% fetal calf serum. The pCEP4 vector containing APP cDNA was transfected into COS 7 cells using Lipofectin (Life Technologies, Inc.) according to the manufacturer's instructions. Stable transfectants were selected in 300 g/ml Hygromycin (Boehringer Mannheim).
Polarity Assay-Only experiments in which endogenous (Kunitz-type protease inhibitor containing) APP was sorted to the basolateral chamber with an efficiency Ն90% were selected for further analysis. In addition, all established MDCK II cell clones were analyzed for polarity using the [ 35 S]methionine uptake assay (24). Uptake was Յ1% apical.
Sorting Assay-Stably transfected MDCK cells were plated on 24-mm collagen transwell chambers (Costar) with a pore size of 0.4 mm and analyzed after 4 -6 days. Sodium butyrate was not used to enhance expression. Cells were used for up to 12 passages (total including transfection). Sorting of endogenous Kunitz-type protease inhibitor containing APP was used as an internal control to assay the polarity of the cells.
Immunoprecipitation and Western Blot Detection-APP from overnight-conditioned medium was precipitated using the anti-FdAPP polyclonal antibody (1:1000) as described previously (26). The entire samples were separated by electrophoresis using 7.5% polyacrylamide gels according to (27). The gels were then blotted onto nitrocellulose filters and blocked with 10% skim milk powder (Merck) in Tris-buffered saline, pH 8. The filter was incubated first with a monoclonal anti APP antibody (22C11, 1:10,000) for 2 h and then with a secondary horserad-ish peroxidase-coupled anti-mouse antibody (Amersham Corp; 1:5000) for 1 h. Detection was done by chemiluminescence (Amersham ECL system). Band intensity was quantified using a computerized elscript 400 densitometer (Hirschmann, Germany) according to the manufacturer's instructions.
Pulse-Chase-Stably transfected COS 7 cells were grown to near confluence, washed three times with MEM lacking methionine, and incubated for 30 min. in MEM without methionine. Cells were pulsed for 10 min in the presence of 600 Ci/ml of [ 35 S]methionine (Amersham). After the pulse, the labeling medium was removed by washing the cells with MEM containing 150 mg/l L-methionine. Total conditioned medium was removed at various time points and replaced with equal amounts of MEM (containing 150 mg/liter L-methionine).

Sorting of APP and L-APP in Polarized MDCK II Cells-
Recent evidence suggested that APP contains two independent basolateral sorting signals. One is located in the cytoplasmic domain and controls the sorting of transmembrane APP; the second one, used for basolateral sorting of APPsec, is thought to reside in the luminal domain of APP (14,15,18) and does not influence the distribution of transmembrane APP. In order to identify the luminal sorting signal, we expressed different APP constructs ( Fig. 1) in stably transfected MDCK II cells and analyzed the amount of secreted APP in both the apical and basolateral chambers. As expected, APP695, APP751 (data not shown), APP695␥, and endogenous APP containing the Kunitztype protease inhibitor exon (APP751 and/or APP770) were sorted basolaterally (mean value 93-95% Ϯ 5 basolateral) ( Fig.  2 and Fig. 4). In contrast, L-APP677 tended to be secreted apically (65% apical), although the absolute values were variable between individual experiments (absolute variation Ϯ 25 apical, standard deviation Ϯ 15.3) (Fig. 3 and Fig. 4). This variation did not depend on the expression level or on the individual clone analyzed (data not shown). The observation that L-APP is secreted differently than is APP suggests that L-APP contains a different sorting signal, probably generated by the alternative splicing of exon 15  dominant luminal apical sorting signal. To establish if the basolateral sorting signal resides on exon 15 itself, a large deletion clone (APP695⌬CHO) was introduced into the MDCK cells. This deletion extends from exon 12 to the beginning of ␤A4 (exon 16), so that the construct lacks exon 15 as well as the neighboring regions, which could influence a newly generated apical signal in L-APP. If exon 15 contains the dominant basolateral sorting signal for APPsec, the mutant protein, like L-APP, should be sorted apically. However APP695⌬CHOsec was stringently sorted basolaterally (mean value 95% basolateral) ( Fig. 4 and Fig. 5). It is concluded from this that the basolateral sorting signal, if any, resides in the first half of the luminal domain of APP, or ␤A4 and exon 15, is unlikely to contain the dominant luminal basolateral sorting signal.
The possibility that the fusion of exon 14 to exon 16 generates a dominant apical signal was analyzed using a point mutation (L-APP677A). It has been postulated that in some proteins, carbohydrates could function as an apical sorting signal (16,17). This could be the case for L-APP, because a chondroitin sulfate glucosaminoglycan (GAG) may be added to at least some of the L-APP molecules (9,10,29). This addition can be inhibited by mutation of the serine residue 563 (APP677 numbering) at the beginning of exon 16 (9). If the GAG or any other modification at this site functioned as an apical sorting signal, the mutant protein should revert to basolateral sorting. Expression of L-APP677A, in contrast to L-APP677, did not generate high molecular weight APP, demonstrating that the L-APP677A does not contain GAG (Fig. 6). However, expression of L-APP677A in MDCK II cells resulted in apical sorting of L-APP677Asec equal to or only slightly less than that of L-APPsec, indicating that the GAG has little or no enhancing effect on apical sorting of L-APP (Fig. 4). Also, a soluble L-APP mutant (L-APPex16A STOP ), which is not a substrate to APP secretases (␣, ␤, ␥, ␦), nor does it contain their cleavage sites, was secreted apically as well as basolaterally, similar to the other L-APP constructs (Figs. 3B and 4).
␤A4 and p3 Generation-Upon observing that L-APP is sorted differently than APP, we decided to investigate whether it also differs with regard to ␤A4/p3 production. We therefore analyzed the polarized secretion of ␤A4 and p3 for APP695 and L-APP677 as well as for the truncated construct SPA4CT lacking the complete luminal domain of APP except for the extramembranous part of ␤A4. In all cases, we found the secretion of the peptides ␤A4 and p3 to be restricted to the basolateral chamber and thus parallels the secretion of APP and confirms earlier reports of ␤A4 and p3 secretion for non L-APP isoforms of APP (data not shown; Refs. 13-15). Since ␤A4 production in this MDCK II strain is low, we conducted further investigations using African green monkey kidney epithelial cells (COS 7), stably transfected with L-APP677, APP695, or APP751.
The observation, that L-APP is sorted differently than is APP, raised the question whether L-APP is also processed differently by APP secretases. Among the secreted low molecular weight fragments, p3 is generated by the action of ␣-secretase while ␤A4 is generated by ␤-secretase. ␤A4 is not degraded into p3 but both peptides seem to be formed via parallel mechanisms from different APP pools or degradative intermediates (30). If APP and L-APP isoforms differ with respect to ␤A4 or p3 generation, APP695 and APP751 should be processed similarly, while L-APP677 should differ. COS 7 cells were thus stably transfected with L-APP677, APP695, and APP751, metabolically labeled with [ 35 S]methionine, and the p3 and ␤A4 peptides immunoprecipitated with specific antibodies (Fig. 7, A  and B).
In order to verify that our polyclonal antibody 692 is equally sensitive to ␤A4 and p3, we used the monoclonal antibody G210, which is specific for ␤A4 sequences ending with amino acid 40 (␤A4 numbering). Such an antibody should detect ␤A4, and p3 with equal efficiency. We found that the ␤A4/p3 ratio did not differ from the results obtained with antibody 692 (data not shown).
Whereas the ␤A4/p3 ratio was 0.21 for APP695 and APP751, the ratio decreased to 0.14 for L-APP (p Ͻ 10 Ϫ7 ; Student's t test). These ratios were highly reproducible in different experiments and were independent of the absolute levels of (L-)APP, ␤A4 and p3 being produced (Fig. 7B). Also, clonal variation seemed not to influence the ␤A4/p3 ratios, since stably expressing cells were used from three different transfections for APP695 and L-APP677 (two for APP751) and found to secrete ␤A4/p3 in the same ratio for each construct independent of the transfection. In addition to p3 and ␤A4, we observed one or two bands migrating between the major ␤A4 band and p3 (reviewed in Ref. 12), which we refer to as p3 related peptides. These peptides presumably contain longer p3 species that were cleaved more N-terminally from the major ␣-secretase site at Lys 16 /Leu 17 , since they were not detected by monoclonal antibodies specific for ␤A4 1-8 (data not shown). The intensity of these two bands was correlated with p3 rather than with ␤A4, as both p3 and p3 related peptides were increased with L-APP expressing cells compared with APP-expressing cells. However, the upper band could not always be observed.
Pulse-chase experiments showed overall parallel kinetics of the transfected secretory proteins and the 3-4 kDa peptides (Fig. 8, A and B). DISCUSSION APP is synthesized in cells as a transmembrane precursor protein that is cleaved to generate soluble APP (APPsec), which is subsequently secreted. This secretion has been shown to be polarized in MDCK cells and to depend on unknown luminal signals. In contrast, the sorting of transmembrane APP and other transmembrane proteins has been found to depend on short signals on the cytoplasmic tail of these proteins. The mutant protein APP695␥ unlike APP and L-APP will not integrate into cellular membranes and lacks cytoplasmic sequences responsible for the polarized sorting of transmembrane APP. However, APP695␥ was secreted basolaterally. It therefore shows the influence of APP-luminal sequences on polarized sorting of soluble APP. Similar results have been obtained for APP695␣ and APP695␤, mutants that end at ␣ and ␤-secretase sites, respectively (14,15). These observations make it very likely that APP contains at least one luminal sorting signal for soluble APP.
In this study, we show that the sorting of soluble APP is regulated by alternative splicing. The very same splicing event also influences the release of ␤A4 and p3 and thus might be of concern in Alzheimer's disease pathology. Expression of L-APP resulted in increased apical secretion of the soluble derivative L-APPsec as compared with expression of APP in MDCK II cells (Figs. 3A and 4). L-APP is distinguished from APP by the absence of the residues encoded by the alternatively spliced exon 15. It has recently been reported that the absence of exon 15 residues results in the generation of a GAG attachment signal (9,10,29). Glycosylation has been speculated to be a possible sorting signal (16,17).
We thus analyzed the polarized sorting of L-APP677A. This mutant lacks the serine residue in the GAG attachment site and hence is not modified by GAG. The sorting of L-APP677A was similar to the sorting of L-APP677 (Fig. 4) and it is concluded that the reported carbohydrate addition specific to L-APP does not play a major role in APP sorting in MDCK II cells. The observation that L-APP is secreted differently as compared with APP must therefore depend on a different mechanism generated by the alternative splicing of exon 15. Several mech- Overnight conditioned media from plastic grown MDCK II cells was analyzed using anti-FdAPP for immunoprecipitation and antibody 22C11 for Western blot detection. In contrast to L-APP677, L-APP677A does not generate diffuse high molecular weight bands resulting from GAG attachment. Only a fraction of L-APP carries the GAG modification in MDCK II cells. Lanes 1, 4, and 7 show L-APP677A expressing clones; lanes 10 and 11 show L-APP677 expressing clones; lanes 2, 3, 5, 6, 8, and 9 show non-expressing clones. Note the appearance of high molecular weight immunoreactivity with L-APP, which is not observed with L-APP677A transfected clones. However, L-APP high molecular weight immunoreactivity was always less than that of unmodified L-APP and therefore clearly detectable only with clones expressing high amounts of L-APP.
anisms might be postulated. First, exon 15 itself might contain a dominant luminal basolateral sorting signal or an essential part of this signal.
The basolateral secretion, observed with the deletion clone APP⌬CHO, provides strong evidence that exon 15 itself does not contain the basolateral sorting signal for APPsec. If it were, the deletion of exon 15 as well as the neighboring regions would remove the dominant basolateral sorting signal for APPsec and would cause the mutant protein, like L-APP, to be sorted randomly. However, APP⌬CHOsec was stringently sorted basolaterally (Figs. 4 and 5).
Thus exon 15 alone cannot be responsible for basolateral sorting. Furthermore the results obtained with APP⌬CHO suggest, that if basolateral sorting for soluble proteins requires the presence of a specific signal, such a signal must be present on the first half of the luminal domain of APP or that more than one basolateral signal coexists on APPsec. APP⌬CHOsec does contain in addition to the first N-terminal half of the APP luminal domain also the luminal domain of ␤A4. We cannot exclude that the latter domain influences sorting of the transmembrane APP/APP⌬CHO. However, it seems unlikely that they are important for the sorting of APPsec since APP␣ as well as APP␤ (which is devoid of the ␤A4 domain) have been reported to be secreted basolaterally (14,15).
A second possibility is that in L-APP a more distant signal is inactivated due to a change in conformation. Since neither the GAG modification of L-APP is necessary for apical sorting nor exon 15 itself contains a basolateral signal, it might be postulated that the effect observed in L-APP would be conferred by a more complex structural phenomenon. A structural difference between L-APP and APP is further supported by the finding that L-APP expressed in E. coli, without carbohydrate modification migrates at an aberrantly high molecular weight in SDS gel electrophoresis, as compared with APP. 3  Such a structure has been described as the luminal sorting signal of chromogranin B, a protein normally sorted to secretory granules in the regulated secretory pathway (32). A structural domain important for sorting could be either disrupted or created by alternate splicing out exon 15, i.e. removal of exon 15 could either interfere with the binding of L-APP to proteins involved in basolateral sorting or generate a new signal that leads to apical sorting. In principle, a third explanation for differences in polarized sorting could be an unequal distribution of APP secretases. However, the results obtained here and by others (14,15) show that soluble APP mutants (APP␤, APP␣, APP␥, and L-APPex16A STOP ), that are not substrates for APP secretase are still sorted according to the presence or absence of exon 15. This conclusion is further supported by Haass et al. (14), which showed that transmembrane APP, which has been targeted by mutation of the cytoplasmic signal QYTSI to 50% to the apical membrane, still was released stringently basolaterally after cleavage by the APP secretases. A possible accumulation of intracellular L-APPsec after secretase cleavage in a compartment incapable of sorting seems unlikely, since L-APPex16A STOP , which does not undergo secretase cleavage, was also secreted similarly to L-APP. In addition, L-APPex16A STOP contains none of the binding sites identified for APP secretases.
Since APP is a transmembrane molecule with secretory derivatives, a differential basolateral versus apical localization of APP ␣-secretase could be used as a mechanism for polarized sorting. However, this mechanism seems unlikely to play a role in the sorting differences observed for the truncated constructs L-APPex16A STOP and APP␥, which are not substrates for APPsecretases, but are sorted according to the presence or absence of exon 15-encoded residues. Since the APP secretases are not known, we cannot completely rule out the possibility of differences at the level of secretases, but in light of the results with the above soluble mutants, this possibility seems less likely.
The conditions for polarized sorting of soluble proteins are not known. It has been speculated that either basolateral or apical sorting does not require a specialized sorting signal or that ubiquitous signals are being used for a default pathway (16,17). If basolateral sorting was used as a default pathway, the results obtained with APPsec, L-APPsec, APP695␥, and APP⌬CHO could be explained by the generation of a new (dominant) apical sorting signal in L-APP due to the fusion of exon 14 to exon 16. If apical sorting was used as the default pathway, a basolateral signal would be disrupted or interfered in L-APP, resulting in constitutive apical sorting. Sorting of L-APPsec is more variable than sorting of APPsec. This could reflect true variability or may indicate that L-APPsec sorting is additionally modulated by factors that do not influence the sorting of APP.
While alternate splicing of exon 15 can lead to alterations in polarized sorting, the actual identity of a putative luminal (apical) sorting signal remains unclear.
Amyloid Production-The difference found in polarized sorting of APP and L-APP raised the possibility that ␤A4 production or sorting is also affected by alternative splicing. ␤A4 was sorted basolaterally by APP as well as by L-APP expressing cells. While transmembrane APP was not studied here, this is interesting as it shows that the transmembrane products of the APP-secretases ␣ and ␤ (A4CT and p3CT) are sorted according to the cytoplasmic signal and independently of the L-APPsec sorting signal.
While ␤A4 was sorted basolaterally when produced from APP as well as from L-APP, a decrease of ␤A4 relative to p3 was observed in L-APP expressing cells, which was not due to differences in the overall secretory kinetics of either ␤A4, p3, APPsec, or L-APPsec in the transfected cells.
This difference in p3 and ␤A4 generation might be of importance in the generation of plaques, since N-terminally shortened ␤A4 peptides (including p3) should contribute differently to the aggregation process of ␤A4 to amyloid. Altered production of ␤A4 could be either directly or indirectly linked to the altered polarized sorting of L-APP. This could be due to a difference in subcellular compartmentalization of APP and L-APP. Alternatively, a difference in primary or secondary structure in L-APP could also inhibit the ␤-secretase. The relative increase of p3-related peptides possibly suggests that these peptides are derived by a mechanism related to ␣-secretase and not ␤-secretase.
In summary, the results presented here show that L-APP is both sorted and processed differently than APP. The difference in polarized sorting shown here, combined with the different tissue distributions (4), suggests that L-APP differs in function from APP and that the GAG attachment may be involved in this function.
Very recently, a downstream exon has been identified in the nonhomologous region of APLP2 that is alternatively spliced to generate a chondroitin sulfate attachment site (8). However, this modification does not influence the polarized secretion in MDCK cells (33), which is in line with our findings, that this GAG modification alone does not influence polarized secretion of APP. The general homology of APP and APLP2 splicing does seem to make it plausible that L-APP and L-APLP2 might be sorted similarly, but this is not the case. At the protein level, however, the sequences at and around exon 15 of APP show no homology to APLP2. Thus a prediction for the sorting of L-APP and L-APLP2 based on the homology might be difficult.
Several lines of evidence suggest that the amount of ␤A4 produced is an important factor in the development of Alzheimer's disease (31). In light of the differences in ␤A4 generation between APP and L-APP shown in this report, one could speculate that L-APP might contribute differently than APP to the development of amyloid plaques or even have have a protective effect. It is interesting that the brain is the only organ with established ␤A4 amyloid deposition in Alzheimer's disease and that neurons are the only cell type expressing high amounts of APP but only low amounts of L-APP(4).