Enhanced production and oligomerization of the 42-residue amyloid beta-protein by Chinese hamster ovary cells stably expressing mutant presenilins.

Mutations in the presenilin 1 (PS1) and presenilin 2 (PS2) genes cause the most common and aggressive form of early onset familial Alzheimer's disease. To elucidate their pathogenic mechanism, wild-type (wt) or mutant (M146L, C410Y) PS1 and wt or mutant (M239V) PS2 genes were stably transfected into Chinese hamster ovary cells that overexpress the β-amyloid precursor protein (APP). The identity of the 43-45-kDa PS1 holoproteins was confirmed by N-terminal radiosequencing. PS1 was rapidly processed (t1/2 = 40 min) in the endoplasmic reticulum into stable fragments. Wild-type and mutant PS2 holoproteins exhibited similar half lives (1.5 h); however, their endoproteolytic fragments showed both mutation-specific and cell type-specific differences. Mutant PS1 or PS2 consistently induced a 1.4-2.5-fold increase (p < 0.001) in the relative production of the highly amyloidogenic 42-residue form of amyloid β-protein (Aβ42) as determined by quantitative immunoprecipitation and by enzyme-linked immunosorbent assay. In mutant PS1 and PS2 cell lines with high increases in Aβ42/Aβtotal ratios, spontaneous formation of low molecular weight oligomers of Aβ42 was observed in media, suggesting enhanced Aβ aggregation from the elevation of Aβ42. We conclude that mutant PS1 and PS2 proteins enhance the proteolysis of β-amyloid precursor protein by the γ-secretase cleaving at Aβ residue 42, thereby promoting amyloidogenesis.

genes that strongly predispose individuals to the premature development of AD have been identified to date. First, missense mutations in the APP gene (1)(2)(3)(4)(5), which encodes the precursor of A␤, increase the production of A␤ peptides, particularly A␤ 42 , in vitro and in vivo (6 -10). Second, inheritance of the ⑀4 polymorphism of the apolipoprotein E gene increases the number and density of A␤ deposits in the brain (11)(12)(13)(14)(15). The third and fourth genes to be linked to AD, presenilin 1 (PS1) and presenilin 2 (PS2), cause the most common form of early onset familial AD (16 -18). These genes encode highly homologous proteins predicted to span the membrane 7-8 times (19). Missense mutations in PS1 and PS2, more than 30 of which have already been identified (20), result in markedly accelerated clinical and neuropathological features of AD.
A clue to the mechanism of the presenilins has come from the recent report of selective elevations in A␤ 42 levels in plasma and skin fibroblast media of subjects harboring PS1 or PS2 mutations (10). Because primary fibroblasts expressing different PS1 or PS2 mutations show very low A␤ secretion that cannot be easily studied mechanistically, we examined stably transfected Chinese hamster ovary (CHO) cell lines in which the sole variable is the introduction of normal or mutant presenilin genes, and any other host-derived factors are eliminated (21). Moreover, the formation of A␤ oligomers can be detected in the conditioned media of CHO cells (22). We stably introduced different PS1 and PS2 mutant genes into CHO cells, and characterized the expressed proteins by radiosequencing, pulse-chase experiments and pharmacological treatment. In contrast to PS1-expressing cells, cells expressing mutant PS2 showed presenilin endoproteolytic patterns that differed from wt PS2 cells in both mutation-specific and cell type-specific ways. Using two methods of quantitation, we found that a direct effect of the familial AD-linked presenilin mutations is to increase selectively and significantly the cellular production of the highly amyloidogenic A␤ 42 peptide. In mutant PS1 and PS2 cell lines with high increases of A␤ 42 secretion, spontaneous A␤ oligomer formation was observed, demonstrating that heightened production of A␤ 42 by cells results in enhanced A␤ aggregation.
Metabolic Labeling and Immunoprecipitation-In pulse-chase experiments, cells were first incubated in methionine-free, fetal bovine serum-free media for 45 min before pulse-labeling with 100 Ci/ml [ 35 S]methionine for 5 to 20 min. Cells were then changed to regular Dulbecco's modified Eagle's medium and chased for 0.5-5 h. For some experiments, brefeldin A (10 g/ml) was present during the chase period. Cells were lysed in a buffer containing 50 mM Tris, pH 7.6, 150 mM NaCl, 2 mM EDTA, 1% Nonidet P-40, 12 mM CHAPS (Pierce), and a protease inhibitor mixture (5 g/ml leupeptin, 5 g/ml aprotinin, 2 g/ml pepstatin A, and 0.25 mM phenylmethylsulfonyl fluoride; Sigma). Immunoprecipitation was performed as described (27). All quantitation of A␤ immunoprecipitates were performed with a PhosphorImager 400A, using ImageQuant software (Molecular Dynamics).

RESULTS
Stable Co-expression of Human APP and PS1 or PS2 Genes in CHO Cells-Wild-type and mutant (M146L, C410Y) PS1 genes were stably transfected into CHO cells that stably overexpress APP (cell line 7W) (see "Materials and Methods"). Wt PS1/APP transfectants (PS70) were pulse-labeled for 5 min and chased for 0 -5 h with or without BFA. In the absence of BFA, we found that the half-life of APP remained unchanged (ϳ30 min), compared with that in the 7W parental cell line (29). The expected NЈ-and OЈ-glycosylation of APP was observed, and the characteristic 10-and 12-kDa C-terminal fragments of APP appeared when cells were chased for more than 0.5 h (Fig.  1). In the presence of BFA, APP did not undergo normal maturation by glycosylation (Fig. 1), in agreement with published data (30). BFA also blocked the proteolytic processing of fulllength APP into C-terminal fragments (Fig. 1). In the same cell lysates, the PS1 holoprotein was identified as a 43-45-kDa doublet, and this holoprotein had a half-life of ϳ40 min. BFA treatment did not change the electrophoretic mobility of the PS1 proteins or alter the rate of PS1 turnover ( Fig. 1), suggesting that full-length PS1 is primarily processed within the endoplasmic reticulum (ER). Consistent with this conclusion, we observed no [ 3 H]glucosamine or [ 35 S]sulfate incorporation into PS1 proteins, in contrast to the modification of APP by these groups in the same cells (data not shown). The identity of the 43-and 45-kDa PS1 holoproteins was confirmed by radiolabeled sequencing with [ 35 S]methionine. We found methionines at positions 1 and 16 in both bands, clearly demonstrating that both proteins begin with the N-terminal methionine predicted from the PS1 cDNA sequence (16). These data, together with the consistent precipitation of both bands by an extreme Cterminal PS1 antibody (4627) (Fig. 1), suggest that the two polypeptides are SDS-stable conformers of full-length PS1.
Stable 7W CHO cells expressing human PS2 proteins were also established (see "Materials and Methods"), and the metabolism of PS2 was examined by pulse-labeling for 20 min and chasing for 0.5-5 h, followed by immunoprecipitation with the N-terminal PS2-specific antibody, 2972 (Fig. 2). PS2 transfectants overexpressed full-length PS2 proteins as a broad doublet at 46 -55 kDa, as well as the characteristic higher molecular weight PS2 aggregates (31). Both wt and mutant (M239V) full-length PS2 holoproteins had a half-life of ϳ1.5 h. We also searched for endoproteolytic products of PS2, which have not been previously reported. We observed the appearance of a 35-kDa N-terminal fragment in both wt and mutant cells after chasing for 0.5 h. Once formed, the 35-kDa fragment was very stable, and there was no degradation after 5 h in both cell lines. In addition, an 18-kDa band was precipitated by the N-terminal antibody in the wt PS2 transfectants but not in the M239V mutant PS2 cells (Fig. 2). This 18-kDa band was absent when the antibody was preabsorbed with its peptide immunogen. The 18-kDa fragment was immediately detected after 20-min pulse labeling of wt PS2 cells and then rapidly disappeared during the chase period. To confirm its identity, we performed radiosequencing of the 18-kDa fragment with [ 35 S]methionine and found methionines at positions 1 and 5, entirely consistent with the N-terminal amino acid sequence predicted from the PS2 cDNA. Similar pulse-chase experiments were conducted to assess any alteration of APP maturation in the PS2 stable cell lines, and we detected no change in the half-life and posttranslational modification of APP molecules in wt and mutant PS2 transfectants. We also failed to observe any incorporation of [ 3 H]glucosamine or [ 35 S]sulfate into PS2 proteins, compared with the expected modification of APP by these groups in the same cells (data not shown). PS70 cells stably expressing wt PS1 and APP were pulse-labeled for 5 min and chased for 0 -5 h with or without BFA (10 g/ml). Cell lysates were co-immunoprecipitated with 4627 (to PS1) and C7 (to APP). The half-life of wt PS1 holoprotein was ϳ40 min, and BFA did not block the turnover of PS1 proteins. Overexpression of PS1 did not change the half-life and the post-translational modification of APP. Maturation of full-length APP and proteolytic formation of its 10-and 12-kDa C-terminal fragments were only detected in the absence of BFA. Note that the lower portion of the gel was overexposed to show the PS1 holoprotein, due to its relatively low abundance compared with APP.
FIG. 2. Pulse-chase of PS2 holoprotein and its derivatives in wt and mutant PS2 transfectants. Wild-type and mutant (M239V) PS2 cells were pulse-labeled for 20 min and chased for 0 -5 h. The holoprotein (top arrowhead) and high molecular weight PS2 aggregates were quickly turned over, with a half-life of ϳ1.5 h. A stable 35-kDa Nterminal fragment (middle arrowhead) appeared after 0.5 h. The 18-kDa N-terminal fragment (lower arrowhead), which was only observed in the wt PS2 cells, appeared immediately after pulse labeling and was degraded at a similar rate as the holoproteins.

Differential Endoproteolysis of Mutant PS2 Proteins in CHO
and 293 Cells-The above results (Fig. 2) raised the possibility that the processing of mutant PS2 proteins differs from that of wt PS2. This phenomenon was not observed for the PS1 mutants examined here. As reported previously (32), PS1 holoproteins underwent endoproteolysis to form stable 27-28-kDa Nterminal and 17-18-kDa C-terminal fragments, and the fragment pattern did not differ between wt and mutant PS1 proteins expressed in the CHO cells (data not shown). In the case of the PS2 stable cell lines, we detected both the 35-kDa and 18-kDa N-terminal fragments in wt PS2 transfectants, but not in untransfected CHO cells, as determined under steadystate conditions by combined immunoprecipitation-Western blotting of unlabeled cell lysates with antibody 2972 (Fig. 3). The 35-kDa fragment was the major endoproteolytic product, whereas the 18-kDa fragment occurred at low abundance. The latter result was consistent with that of the pulse-chase experiments, which indicated a quick turnover of the 18-kDa fragment versus considerable stability of the 35-kDa fragment (Fig.  2). In CHO cells expressing the M239V (Italian) or N141I (Volga-German) mutations, only the 35-kDa fragment was detected ( Fig. 3), suggesting a different proteolytic processing mechanism for wt and mutant PS2 proteins. To confirm that this difference was not confined to CHO cells, we examined the endoproteolysis of PS2 in human kidney 293 cells stably expressing wt or each of the mutant isoforms (Fig. 3). Here, we observed the 35-kDa fragment in both untransfected and PS2transfected cells, suggesting a greater level of endogenous PS2 expression in 293 than CHO cells. No 18-kDa fragment was detected in any of the 293 cell lines. Moreover, the Italian PS2 mutation selectively induced a significant reduction of the 35-kDa fragment in both cell types, but this was associated with the formation of an additional 30-kDa N-terminal fragment only in the 293 cells (Fig. 3). Taken together, these results demonstrate a striking variability in the proteolytic processing of PS2 holoproteins as a function both of different mutations (i.e. the Volga-German versus Italian mutation) and of different cell types (e.g. CHO versus 293 cells).
Comparison of the Levels of Total A␤ and A␤ 42 in Wild-type and Mutant PS1-or PS2-transfected CHO Cells-To determine whether mutant PS proteins alter APP processing, we examined numerous CHO cell lines stably overexpressing wt and mutant PS genes. A total of 13 clones expressing wt PS1 (clones PS70, PS106, and PS111), mutant PS1 (ML45, ML60, ML86, CY6, CY10, and CY11), wt PS2 (PS2-1 and PS2-2), or mutant PS2 (MV31 and MV42) were used for quantitative analysis of APP processing and A␤ production. The various clones of wt or mutant PS1 or PS2 cell lines expressed substantially different levels of PS protein (Fig. 4, A and B). In the PS70 and PS106 wt lines and the ML45, ML60, CY6, and CY10 mutant lines, the PS doublet at 43-45 kDa and the characteristic PS oligomers at ϳ100 -140 kDa (31) were readily seen (Fig. 4A). Cell lines PS111, ML86, and CY11 expressed the same set of proteins but at lower levels (data not shown). The difference between the clones with the lowest and the highest PS1 expression levels was about 10-fold. Two wt PS2 clones and two M239V mutant PS2 clones expressed high levels of PS2 holoproteins at ϳ46 -55 kDa and higher molecular weight oligomers at ϳ110 -140 kDa that were similar to those of PS1 (Fig. 4B). The higher molecular weight bands that were consistently detected in all immunoprecipitations have been shown to represent aggregates of PS proteins in transfected cells (31).
We quantitated APP expression levels in all of these PS1 and PS2 stable transfectants. Compared with the parental 7W cells that were untransfected with PS genes, the biosynthesis and the steady state levels of full-length APP did not change significantly in any of these 13 clones (data not shown). The levels of A␤ total and A␤ 42 were then compared by immunoprecipitation of conditioned media with antibodies that are highly specific for each derivative. Medium from the same culture dish was aliquoted and precipitated with antibodies R1282 (for A␤ total ) or 21F12 (for A␤ peptides ending at residue 42). Phosphorimaging of the A␤ total signal showed no significant alteration in cells expressing mutant PS1 or PS2 (data not shown). However, when the same media were precipitated by 21F12, we observed an increase in the amounts of the A␤ 42 and p3 42 gel bands in the mutant cells (Fig. 4C). The ratio of A␤ 42 to total A␤ was calculated, and the fold increase in this ratio was established by normalizing the mean ratios obtained in all wt PS1 (n ϭ 40) or wt PS2 (n ϭ 10) determinations to 1.0 ( Fig. 5A and Table I). A 1.4 -2.5-fold increase in mean A␤ 42 /A␤ total ratios was obtained in the mutant PS1 and PS2 clones (p Ͻ 0.001, except in MV31, p Ͻ 0.005; two-tailed Student's t test). Cell lines ML60, MV31, and MV42 secreted more A␤ 42 (Figs. 4C and  5A and Table I) than the other lines. The fold increase in A␤ 42 /A␤ total ratio in the latter three lines was in the same range as that of a cell line expressing the APP V717F mutation (Fig. 4C), which is known to induce substantial overproduction of A␤ 42 (8).
To confirm these results, the concentrations of A␤ 42 and A␤ total in the conditioned media of PS1 and PS2 transfectants were measured by sensitive and specific sandwich ELISAs. A scattergraph of all ELISA results is shown in Fig. 5B. When the mean ratios obtained in all wt PS1 (n ϭ 24) or wt PS2 (n ϭ 11) determinations were normalized to 1.0, the relative A␤ 42 / A␤ total ratios were increased 1.3-2.2-fold in the mutant PS1 and PS2 lines. The differences between the wt and mutant lines were highly statistically significant (p Ͻ 0.001, except in CY10, p Ͻ 0.02), and correlated well with the quantitation by immunoprecipitation in the same cell lines.
Oligomerization of Secreted A␤ in PS1 and PS2 Mutant Cell Lines-In the course of the immunoprecipitation studies, we observed the occurrence of low molecular weight oligomers of A␤ in the conditioned media of the APP V717F cell line, as reported previously (22). The V717F line produced more A␤ 42 than the APP wild-type 7W cell line, as expected (8), and it showed 42-specific oligomeric A␤ bands by immunoprecipitation with 21F12 and SDS-PAGE (Fig. 4C, lane V717F). Inter-estingly, the three PS1 or PS2 mutant clones with the highest level of A␤ 42 production, ML60, MV31, and MV42 (Fig. 5, A and  B), showed low molecular weight oligomeric species indistinguishable from those of APP V717F (Fig. 4C, lanes ML60, MV31,  and MV42). Because the presence of two additional hydrophobic residues at the C terminus of A␤ 42 is known to increase its fibrillogenic potential in vitro and lead to its seeding the aggregation of the A␤ 40 peptide (33), the appearance of A␤ 42 containing oligomers in the medium further supports the substantially heightened production of A␤ 42 by cell lines expressing mutant PS1 or PS2.

DISCUSSION
The PS1-and PS2-linked cases of AD are distinguishable from common sporadic AD cases principally by their earlier clinical onset and more severe neuropathology (34 -36). Therefore, elucidating the genotype-to-phenotype relationship of the presenilin mutations should shed light on factors that are important in the pathogenesis of AD in general. Here, we show that expression of a mutant presenilin gene in cultured cells results in a selective and statistically significant increase in the secretion of the highly amyloidogenic A␤ 42 peptide. For most clones, the increase in the A␤ 42 /A␤ total ratio obtained by immunoprecipitation was closely similar to that obtained by ELISA (Table I). Thus, the results were internally consistent. Our results are in agreement with the selective elevation of A␤ 42 levels in plasma and skin fibroblast media obtained from living PS1 and PS2 patients (10). One of the PS1 mutations examined in the latter studies is at the same codon (146) as one we tested. Moreover, the degree of increase we observed, 1.3-2.5-fold, is the same as that obtained in vivo (10) and in the same range as that resulting from the APP 717 mutation (8).
We observed no obvious relationship between the level of PS1 expression and the degree of increase in A␤ 42 /A␤ total ratio. For example, the clone with the highest increase (M146L-60) had lower PS1 protein levels than clone C410Y-6, which showed a relatively small but still significant (p Ͻ 0.001) increase. This result strongly suggests that mutant presenilin proteins confer a dominant negative gain of function, which can be seen even at low levels of expression and in the presence of the endogenous wt presenilins.
The fact that expressing solely a mutant presenilin cDNA in a cultured peripheral cell leads directly to a selective increase in A␤ 42 similar to that observed in vivo (i.e. in plasma and primary fibroblasts) indicates that factors unique to the brain or to the AD state are not required for this A␤ overproduction.  Excessive A␤ 42 production thus appears to be an initial consequence of mutant presenilin expression, a conclusion that is supported by the findings that carriers of PS1 mutations can show elevated A␤ 42 levels in plasma presymptomatically, while the majority of symptomatic subjects with sporadic AD show no increase in A␤ 42 levels despite advanced clinical disease (10).
Our results are consistent with the recent demonstration of a significant 2-fold increase in the density of A␤ 42 plaques in the brains of subjects bearing a PS1 mutation, compared with their density in severe sporadic AD cases (34). Furthermore, A␤ 42 has been shown to be the initial constituent of plaques in AD and Down's syndrome, preceding the development of the other cytopathological features of AD by many years or decades (37)(38)(39).
Of particular interest in this study was the spontaneous appearance of A␤ 42 oligomers in the conditioned media of PS1 or PS2 mutant cells with high (Ͼ2-fold) elevation in A␤ 42 secretion. Peptides ending at A␤ 42 appear to be a major constituent of the diffuse plaques seen initially in AD and Down's syndrome (34, 36, 40 -44), and they have been proposed to serve as a nidus for the aggregation of the more abundant A␤ 40 peptides (33). In contrast to wt APP expressing cells, cells expressing V717F mutated APP consistently showed A␤ 42 oligomer formation in the media, and these cells are known to undergo a selective increase in A␤ 42 secretion. Similar rises in A␤ 42 secretion caused by mutant PS1 or PS2 also led to A␤ 42 oligomer formation. The appearance of A␤ 42 oligomers correlated with the relative increase in A␤ 42 levels; in cells expressing the same PS1 mutation but showing A␤ 42 elevations of less than 2-fold, no A␤ 42 oligomer was detected under the same in vitro condition. Based on these findings, we postulate that PS1 and PS2 mutations, expressed throughout the lifetime of the host, gradually lead to sufficient elevation of extracellular A␤ 42 levels in the brain to induce A␤ 42 oligomerization, with subsequent diffuse plaque formation and later accrual of A␤ 40 peptides. The sequential appearance of first A␤ 42 and then A␤ 40 immunoreactive plaques in Down's patients of increasing age (38,39) are consistent with this hypothesis.
The mechanism of the selective increase in A␤ 42 production caused by mutant presenilins remains to be elucidated. It is unlikely that PS proteins generally affect APP synthesis or metabolism, as the posttranslational modification and turnover of total APP proteins were not changed by introduction of wt or mutant PS genes. Rather, we speculate that APP proteins may interact directly with PS1 or PS2 proteins in a presenilinrich compartment of the cell (e.g. the ER or early Golgi) and that APP interacts with mutant presenilin in a manner that leads to its increased exposure to the ␥-secretase cleaving specifically after residue 42 of the A␤ region. Evidence for the existence of 40-and 42-specific ␥-secretases in mammalian cells has recently been presented (26). Wild-type PS might, for example, form a complex with APP (or the 12-kDa C-terminal fragment arising from ␤-secretase cleavage), and this complex might prevent access of APP to a 42-specific ␥-secretase in the ER; mutations in PS could prevent the proper formation of such complexes.
Since PS1 is known to undergo constitutive endoproteolysis, and very little intact holoprotein is detectable in cells or tissues expressing this gene (32), 2 we searched for evidence of a change in endoproteolytic patterns in cells bearing mutant presenilins. We did not observe a significant change in fragment formation in the two PS1 missense mutations we examined, despite a consistent and significant increase of A␤ 42 /A␤ total in all clones expressing these mutations. However, the mutant form of PS1 that has a deletion of exon 9 undergoes no endoproteolysis (32), and yet this mutation also causes a selective increase in A␤ 42 secretion in transfected cells (45). 3 Therefore, interference with the normal endoproteolysis of PS1 appears not be an obligatory step in producing the mutant phenotype. In the case of our analyses of PS2 mutations, we searched for PS2 endoproteolytic products and identified a major 35-kDa N-terminal fragment. Its size is larger than the 28-kDa derivative, which is the major N-terminal fragment of PS1, indicating that the size of PS2 endoproteolysis is substantially C-terminal to that of PS1, which occurs at and around residue 298 (46). The proteolytic fragment patterns varied when two distinct mutations were expressed in the same cell type (293) and also varied when the same mutation (M239V) was expressed in two different cell types (CHO and 293). Despite these complex differences in PS2 endoproteolysis, the PS2 M239V mutant cell line still showed increased A␤ 42 secretion in both CHO and 293 cells (this study). 3 Our findings, therefore, point to the need for caution in interpreting changes in presenilin processing as a result of PS mutations. Any mutation should be analyzed in at least two cell types, and multiple mutations should be examined before any conclusion about the pathogenic role of altered endoproteolysis is reached.
In summary, our results strongly support the hypothesis that familial AD-linked mutations in PS1, PS2, and APP all cause AD by increasing the cellular production of A␤ 42 , thereby accelerating the polymerization of this and other A␤ peptides and promoting cerebral accumulation of A␤ as an essential early event in AD pathogenesis. These findings provide further impetus for current efforts to identify compounds which inhibit the production or the aggregation of A␤ as therapeutic agents for AD.