The Alternatively Spliced Kunitz Protease Inhibitor Domain Alters Amyloid β Protein Precursor Processing and Amyloid β Protein Production in Cultured Cells

The insoluble amyloid deposited extracellularly in the brains of patients with Alzheimer's disease (AD) is composed of amyloid β protein, a ∼4-kDa secreted protein that is derived from a set of large proteins collectively referred to as the amyloid β protein precursor (βAPP). During normal processing the βAPP is cleaved by β secretase, producing a large NH2-terminal secreted derivative (sAPPβ) and a COOH-terminal fragment beginning at Aβ1, which is subsequently cleaved by γ secretase releasing secreted Aβ. Most secreted Aβ is Aβ1-40, but ∼10% of secreted Aβ is Aβ1-42. Alternative βAPP cleavage by α secretase produces a slightly longer NH2-terminal secreted derivative (sAPPα) and a COOH-terminal fragment beginning at Aβ17, which is subsequently cleaved by γ secretase releasing a ∼3-kDa secreted form of Aβ (P3). Several of the βAPP isoforms that are produced by alternative splicing contain a 56-amino acid Kunitz protease inhibitor (KPI) domain known to inhibit proteases such as trypsin and chymotrypsin. To determine whether the KPI domain influences the proteolytic cleavages that generate Aβ, we compared Aβ production in transfected cells expressing human KPI-containing βAPP751 or KPI-free βAPP695. We focused on Aβs ending at Aβ42 because these forms appear to be most relevant to AD. Using specific sandwich enzyme-linked immunosorbent assays, we analyzed full-length Aβ1-42 and total Aβ ending at Aβ42 (Aβ1-42 + P3(42)). In addition, we analyzed the large secreted derivatives produced by α secretase (sAPPα) and β secretase (sAPPβ). In mouse teratocarcinoma (P19) cells expressing βAPP695 or βAPP751, expression of the KPI-containing βAPP751 resulted in the secretion of a lower percentage of P3(42) and sAPPα and a correspondingly higher percentage of Aβ1-42 and sAPPβ. Similar results were obtained in human embryonic kidney (293) cells. These results indicate that expression of the KPI domain reduces α secretase cleavage so that less P3 and relatively more full-length Aβ are produced. Thus, in human brain and in animal models of AD, the amount of KPI-containing βAPP produced may be an important factor influencing Aβ deposition.

In the brains of patients with Alzheimer's disease (AD), 1 one characteristic pathological feature is the deposition of amyloid in senile plaques. In many patients, amyloid is also deposited in the walls of cerebral and meningeal blood vessels. AD amyloid is composed of amyloid ␤ protein (A␤), a ϳ4-kDa secreted protein released from an ϳ120-kDa amyloid ␤ protein precursor (␤APP) through cleavage by proteases referred to as secretases. Soluble, secreted A␤ is normally present in human cerebrospinal fluid (1,2) and plasma (3). It is present in high concentration in medium conditioned by mixed human fetal brain cells (2) or a variety of transfected cells expressing ␤APP (1,4), and it can be detected at lower concentration in the medium of cultured cells expressing only endogenous ␤APP (5,6). Recent studies have established that the ␤APP is normally cleaved by ␤ secretase to produce a large NH 2 -terminal secreted derivative (sAPP␤) (7) and a cell-associated COOH-terminal derivative beginning at A␤1 (8). Subsequent cleavage of this COOH-terminal fragment by ␥ secretase releases 4-kDa A␤ for secretion. Alternative cleavage of the ␤APP by ␣ secretase produces a slightly longer NH 2 -terminal secreted derivative (sAPP␣) (9) and a COOH-terminal fragment beginning at A␤17 (9), which is cleaved by ␥ secretase releasing a truncated ϳ3-kDa form of A␤ (P3) for secretion. None of the three secretases (␣, ␤, or ␥) has been isolated or cloned, so the specific protease(s) responsible for each secretase activity are currently unknown.
␤APP751 and ␤APP770, which contain a 56-amino acid Kunitz protease inhibitor (KPI) domain encoded by exon 7, are expressed both in the central nervous system and in peripheral tissues, whereas the KPI-free ␤APP695 is expressed almost exclusively in brain, where it is more abundant than the KPIcontaining isoforms. The secreted form of the KPI-containing ␤APP751 is identical to protease nexin II, a plasma serine protease inhibitor (21,22), and it has been shown to inhibit proteases such as trypsin (23)(24)(25), chymotrypsin (22,24,25), and human coagulation factor XIa (22,23).
To determine whether the KPI domain influences the proteolytic cleavages that generate A␤, we compared A␤ production and sAPP secretion in transfected cells expressing human KPI-containing ␤APP751 or KPI-free ␤APP695. A␤ production was examined by sandwich ELISAs developed previously in our laboratory (26,27) and by immunoprecipitation with A␤-specific antibodies. The secretion of sAPP was evaluated by immunoprecipitating ␣ or ␤ secretase-derived sAPP (sAPP␣ or sAPP␤) with specific antibodies.

EXPERIMENTAL PROCEDURES
Cell Culture-Mouse teratocarcinoma P19 cells and human embryonic kidney 293 cells expressing ␤APP695 and ␤APP751 under the control of a cytomegalovirus enhancer and chick ␤-actin promoter were prepared by transfection with modified pCXN2 vectors (28) constructed as described by Fukuchi et al. 2 P19 cells were maintained in Dulbecco's modified Eagle's medium supplemented with 10% heat-inactivated fetal bovine serum and penicillin/streptomycin in the presence of G418. To select for high expression and clonal copy number, the P19 clones designated a were maintained in 0.8 mg/ml G418, and the P19 clones expressing at a lower level (b clones) were maintained in 0.4 mg/ml G418. P19 cells were split every other day, and the medium was changed every day. After transfection with Lipofectin (Life Technologies, Inc.), polyclonal 293 cells stably expressing ␤APP695 or ␤APP751 were selected with G418. The stable polyclonal 293 lines were maintained in Dulbecco's modified Eagle's medium supplemented with 10% heat-inactivated fetal bovine serum and penicillin/streptomycin in the presence of 0.5 mg/ml G418. For time course and synthesis rate measurements, P19 cells were seeded into six-well plates at 2-4 ϫ 10 5 /well 1 day prior to the experiment, and 293 cells were seeded into six-well plates at 0.2-6 ϫ 10 6 /well 1-3 days prior to experimentation. In some experiments, 293 cells were plated into 10-cm plates, and A␤ levels were determined at 48 -96 h postplating.
Radiolabeling Experiments-Subconfluent cells on 10-cm plates were labeled for 5 or 8 h with Tran 35 S-label (250 Ci/ml) in 5 ml of minimum essential medium deficient for methionine/cysteine. Radiolabeled A␤ released into the medium was immunoprecipitated with A␤specific antibodies: R1280 (rabbit polyclonal anti-A␤1-40 provided by D. J. Selkoe), 4G8, BA-27, or BC-05 conjugated to tresyl-activated agarose beads. Following separation by 10 -20% Tris-Tricine SDS-polyacrylamide gel electrophoresis, the immunoprecipitated A␤ was visualized and quantitated by PhosphorImaging (1,30). For measurement of the ␤APP synthesis rate, cells seeded at 2-4 ϫ 10 5 /well in six-well plates on the day prior to experimentation were labeled for 20 min with Tran 35 S-label (250 Ci/ml) in 0.5 ml of minimum essential medium deficient for methionine/cysteine. Using antibody to the ␤APP COOH terminus, the radiolabeled ␤APP was immunoprecipitated from lysates, separated on 10% Tris-Tricine gels, and quantitated by PhosphorImaging as described above. To assess sAPP␣ and sAPP␤, cells seeded at 2-4 ϫ 10 5 /well in six-well plates on the day prior to experimentation were labeled for 2.5-10 h as described above. BAN-50 (anti-A␤1-16) was used to immunoprecipitate sAPP␣, and Ab53 (provided by B. Greenberg), which is specific for the COOH terminus of sAPP derived from ␤ secretase cleavage, was used to immunoprecipitate sAPP␤. The immunoprecipitated sAPPs were processed and quantitated as described above.

A␤ Accumulation in the Medium of Transfected P19 Cells
Expressing ␤APP695 or ␤APP751-To investigate the effect of the KPI domain on the processing events that determine ex-tracellular A␤ concentration, we compared the A␤ that accumulates in the medium of transfected mouse teratocarcinoma (P19) cells expressing ␤APP695 or ␤APP751. Using sandwich ELISAs developed in our previous work (26,27), we followed A␤ accumulation over a 10-h time course by analyzing A␤1-40 (BAN-50/BA-27 ELISA) and A␤1-42(43) (BAN-50/BC-05 ELISA) at 0, 2.5, 5, 7.5, and 10 h. At each time point, we also analyzed A␤ using a BC-05/4G8 ELISA. In this ELISA, A␤s ending at A␤42(43) are captured with BC-05, which is specific for A␤s with this COOH terminus, and they are detected with 4G8, which is specific for the A␤17-24 epitope. Thus, this ELISA detects not only full-length A␤1-42(43) but also NH 2terminally truncated A␤s ending at 42(43) (e.g. A␤17-42(43)) (27). To evaluate ␤APP synthesis in each transfected cell line, we used PhosphorImaging to quantitate the newly synthesized ␤APP radiolabeled during a 20-min pulse of Tran 35 S-label.
In the P19 clones analyzed in our initial experimentation, ␤APP synthesis ( Fig. 1A) was 6-fold higher in the P19-␤APP695 clone than in the P19-␤APP751 clone. The results of the A␤ measurements made in our initial set of five independent experiments are shown in Fig. 1 and Table I. In line with the increased ␤APP expression in the P19-␤APP695 cells, more A␤ accumulated in medium conditioned by the P19-␤APP695 cells than in medium conditioned by the P19-␤APP751 cells. As expected from our previous studies, both the P19-␤APP695 and P19-␤APP751 lines showed (i) increasing accumulation of A␤1-40 and A␤1-42 over the 10-h time period, with A␤1-40 Previously published reports have shown that full-length A␤ and P3 (␣ secretase-derived) are the major forms of A␤ produced by cultured cells (4,5,31). Thus, the increased signal seen with the BC-05/4G8 ELISA compared with the BAN-50/ BC-05 ELISA for A␤1-42(43) is presumably due primarily to P3 ending at A␤42(43), P3 (42). The concentration of this putative P3 (42), calculated by subtracting the BAN-50/BC-05 signal from the BC-05/4G8 signal, is shown in Table I. In medium conditioned by P19-␤APP695 cells for 10 h, 55 Ϯ 4% of A␤ ending at A␤42 was P3 (42), whereas in P19-␤APP751 cells, the percentage of P3 (42) was substantially lower at 26 Ϯ 8% (p Ͻ 0.03 by nonparametric rank sum Mann-Whitney test). Essentially identical results were obtained in two experiments analyzing a second pair of P19 clones in which ␤APP synthesis in the P19-␤APP695 line was 0.9 times that in the P19-␤APP751 line (Table I). In the second P19-␤APP695 clone, 63 Ϯ 28% of A␤ ending at A␤42 was P3 (42), whereas in the second P19-␤APP751 clone, the percentage of P3 (42) was substantially lower at 17 Ϯ 25%. To pursue these observations, we analyzed P3(42) twice more in the b clones and five more times in the a clones. When the data from all 14 measurements were pooled, 70 Ϯ 5% of A␤ ending at A␤42 was P3 (42) in cells expressing ␤APP695, whereas the percentage of P3(42) was only 35 Ϯ 5% in cells expressing ␤APP751 (Table I, p Ͻ 0.001 by nonparametric rank sum Mann-Whitney test). This effect was significant for both the b clones (75 Ϯ 12 for ␤APP695 versus 28 Ϯ 12 for ␤APP751, p Ͻ 0.04 by rank sum Mann-Whitney test) and the a clones (68 Ϯ 5 for ␤APP695 versus 38 Ϯ 6 for ␤APP75, p Ͻ 0.01 by rank sum Mann-Whitney t test).
To confirm that P3(42) accounts for the difference between BAN-50/BC-05 and BC-05/4G8 assays, we labeled cells from the initial P19-␤APP695 clone for 8 h with Tran 35 S-label, used BC-05 to immunoprecipitate the A␤ ending at A␤42 which was secreted into the medium, separated 4-kDa A␤ and 3-kDa P3 by Tris-Tricine SDS-polyacrylamide gel electrophoresis, and quantitated the relative amounts of A␤ and P3 by PhosphorImaging. This clone was chosen for quantitative analysis of the percentage of P3(42) by immunoprecipitation because it produced far more A␤ than the other clones. For comparison, we analyzed the A␤ ending at A␤40 which is produced by this clone by immunoprecipitating with BA-27. As expected from previous reports (4, 5, 31) and the results of our sandwich ELISAs, P19-␤APP695 cells secreted substantial amounts of P3 ( Fig.  2A). Quantitation showed that P3(42) comprised 68% of the total A␤ immunoprecipitated by BC-05, a result in excellent agreement with the percentages (68 Ϯ 5%) obtained in the 10 experiments where sandwich ELISA data were used to determine the percentage of P3(42) (Fig. 2B). P3(40) comprised 55% of the total A␤ immunoprecipitated by BA-27. Thus, the percentage of P3 was similar for the major secreted A␤ ending at A␤40 and the minor A␤ ending at A␤42(43). We could not analyze A␤ ending at A␤40 with a BA-27/4G8 ELISA assay The value for P3(42)/total A␤ ending at A␤42 obtained by immunoprecipitation was determined by calculating the ratio of 3-kDa A␤/(3-kDa A␤ ϩ 4-kDa A␤) using the pixel volumes obtained by PhosphorImaging (panel A). Panel C, subconfluent P19-␤APP695 and P19-␤APP751 cells (a clones) were plated in a 10-cm dish and labeled with Tran 35 S-label for 5 h. A␤ released into the conditioned medium was immunoprecipitated with polyclonal antibody R1280. After separation of the immunoprecipitated A␤ on a 10 -20% Tris-Tricine gel, the gel was fixed, dried, and the radiolabeled 4-kDa and 3-kDa A␤ (P3) were visualized and quantitated by PhosphorImaging. equivalent to the BC-05/4G8 assay for A␤ ending at A␤42 because the sensitivity of the BA-27/4G8 ELISA is very low. We were unable to detect the radiolabeled full-length A␤(42) and P3(42) secreted by our P19-␤APP751 clones after BC-05 immunoprecipitation, presumably because the efficiency of BC-05 immunoprecipitation was insufficient to detect the smaller amount of A␤ secreted by the P19-␤APP751 clones. Thus, to compare full-length A␤ and P3 in P19-␤APP695 and P19-␤APP751 a clones, we immunoprecipitated radiolabeled A␤ with R1280, an efficient polyclonal rabbit antiserum that recognizes both 4-kDa A␤ and 3-kDa P3 (Fig. 2C), although it is not selective for forms ending at A␤40 or A␤42. After R1280 immunoprecipitation, the percent P3 (P3/(A␤ ϩ P3) ϫ 100%) in P19-␤APP695 and P19-␤APP751 cells was 57 and 26%, respectively. Thus, immunoprecipitation with this antibody (i) showed a reduced percent P3 in P19-␤APP751 cells as expected from our sandwich ELISA data and (ii) gave P3 percentages for the P19 a clones which were in reasonable agreement with the percentages obtained by sandwich ELISA.
To assess the generality of the phenomenon observed in the P19 cell lines, we analyzed the relative abundance of A␤1-42 and P3 (42) in conditioned media of polyclonal transfected human embryonic kidney (293) cells expressing ␤APP695 or ␤APP751. In 293-␤APP695 cells, 87 Ϯ 1% of A␤ ending at A␤42 was P3 (42), whereas in the 293-␤APP751 cells, the percentage of P3(42) was significantly (p Ͻ 0.0001) lower at 74 Ϯ 2% (Table  II). Thus, 293 cells make relatively more P3(42) than P19 cells, but in both cell types the percentage of P3(42) is reduced significantly in cells expressing ␤APP751 compared with those expressing ␤APP695.
sAPP␣ and sAPP␤ in the Medium of Transfected P19 Cells Expressing ␤APP695 or ␤APP751-There is good evidence that P3 is derived from COOH-terminal ␤APP derivatives produced by ␣ secretase (4,5,31,32). Thus, a reduction of ␣ secretase activity could account for the reduced percentage of P3 observed in transfected P19 cells expressing ␤APP751. If ␣ secretase activity is reduced, then the secretion of ␣ secretasederived sAPP (sAPP␣) should be decreased. To determine whether sAPP␣ is decreased in P19-␤APP751 cells, we radiolabeled P19 cells expressing ␤APP751 or ␤APP695 for 2.5-10 h with Tran 35 S-label, used antibodies specific for sAPP␣ or sAPP␤ to immunoprecipitate these derivatives from the conditioned medium, separated the immunoprecipitated, radiolabeled sAPP by SDS-polyacrylamide gel electrophoresis, and quantitated the sAPP by PhosphorImaging. As in our previous experiments, ␤APP synthesis in each transfected cell line was analyzed in sister cultures by quantitating the newly synthesized ␤APP radiolabeled during a 20-min pulse of Tran 35 Slabel. Cells from two pairs of P19-␤APP695 and P19-␤APP751 clones were examined. Analysis of the synthesis rates of these lines showed overlapping levels of ␤APP expression as illustrated in Fig. 3A.
We used BAN-50 (anti-A␤1-16) and Ab53 (specific for the COOH terminus of sAPP␤) to immunoprecipitate specifically sAPP␣ and sAPP␤, respectively. Both sAPP␣ and sAPP␤ were detected readily in medium conditioned by P19-␤APP695 or P19-␤APP751 cells (Fig. 3B). To determine whether the relative amounts of sAPP␣ and sAPP␤ are different in cells ex-pressing ␤APP695 or ␤APP751, we compared sAPP␣/(sAPP␣ ϩ sAPP␤) and sAPP␤/(sAPP␣ ϩ sAPP␤) in three experiments (Table III). The relative amounts of sAPP␣ and sAPP␤ were highly reproducible in lines expressing the relevant ␤APPs at widely varying levels (Fig. 3A). Overall, the relative amount of sAPP␣ was significantly (p Ͻ 0.001 by both nonparametric rank sum Mann-Whitney test and unpaired t test) lower for P19 -751 cells than for P19 -695 cells with means of 63 Ϯ 1% and 83 Ϯ 2%, respectively.

DISCUSSION
Our results show, for the first time, that ␤APP751 is processed differently than ␤APP695. Specifically, our comparisons of transfected cells expressing ␤APP695 or ␤APP751 at various levels indicate that, independent of their level of expression, cells expressing ␤APP751 make relatively less sAPP␣ and relatively less P3(42) than cells expressing ␤APP695. Thus, it appears that the expression of the KPI domain diminishes cleavage by ␣ secretase. This diminution of ␣ secretase cleavage could be due to inhibition of ␣ secretase by protease nexin II released extracellularly and/or into an intracellular compartment, to a conformational change in the ␤APP751 holoprotein which renders it relatively less susceptible to cleavage by ␣ secretase or to some other mechanism such as altered trafficking which reduces the exposure of the ␤APP751 holoprotein to ␣ secretase. Our preliminary studies on P19 cells suggest that inhibition of ␣ secretase by extracellular protease nexin II may not be the major mechanism by which ␤APP751 exerts its effect, since medium conditioned by P19-␤APP751 cells had little effect on the production of P3(42) by P19-␤APP695 cells. Additional studies are clearly needed, however, to identify the mechanisms involved.
We wish to emphasize that the rather subtle differences identified were best demonstrated by carefully analyzing cells expressing ␤APP695 or ␤APP751 in parallel for the relative  3. Quantitation of the sAPP␣ and sAPP␤ secretase produced by P19-␤APP695 and P19-␤APP751 cells. Panel A, ␤APP synthesis rates in P19-␤APP695 and P19-␤APP751 cells. Two different pairs of ␤APP695 and ␤APP751 clones designated a and b were assessed. For clone a of both lines, cells were plated at two different densities: 2 ϫ 10 5 /well (a1) and 1 ϫ 10 5 /well (a2); for clone b, cells were plated at 2 ϫ 10 5 /well. The arrowheads indicate the positions of fulllength ␤APP695 (on the left) and ␤APP751 (on the right). The relative amounts of sAPP␣ and sAPP␤ measured in this experiment are quantitated in Table III, Experiment 1. Panel B, sAPP immunoprecipitated from the media of P19-␤APP695 and P19-␤APP751 cells after a 10-h labeling period. sAPP␣ was precipitated by monoclonal antibody BAN-50 (anti-A␤1-16), and sAPP␤ was precipitated by polyclonal antibody Ab53, which is specific for the COOH terminus of sAPP derived from ␤ secretase cleavage. The relative amounts of sAPP␣ and sAPP␤ produced by the b clones measured in this experiment are quantitated in Table III (Tables I and II) showed a much higher percentage of P3 (42) in the medium of 293 cells, an observation in good agreement with the previous results of Seubert et al. (7), who showed that more than 90% of the sAPP produced by 293-␤APP695 cells is sAPP␣. Thus, the relative amount of ␣ secretase activity is influenced by cell type; but in both of the cell types that were analyzed relatively less P3(42) and sAPP␣ were produced from ␤APP751 than from ␤APP695. Studies of A␤ secretion by primary cultures indicate that there are substantial differences in the relative amount of 4-kDa A␤ and P3 which is secreted by neuronal and nonneuronal cells. Neurons, which express ␤APP695, produce substantially higher levels of P3 (5); whereas astrocytes (5) and skin fibroblasts (33), which express predominantly ␤APP751/ 770, produce mostly full-length A␤. The results obtained in the present study suggest that the isoform expressed by these cell types may participate in determining the relative amount of P3 and full-length A␤ which is produced, although intrinsic differences in ␣ secretase activity obviously could also contribute to the differences observed.
Our findings also have important implications concerning the amyloid deposition process in aging, AD, and in experimental models of AD. There is evidence that the cerebral APP751/ APP695 mRNA ratio increases with age (34,35) as might be expected as neurons, which have relatively high expression of ␤APP695, are lost during normal aging. In addition there is evidence that the neuronal APP751/APP695 mRNA ratio is correlated positively with plaque density in brains of AD and control patients (36). Finally, the available evidence suggests that efforts to model A␤ deposition in transgenic mice overexpressing ␤APP are more effective when KPI-containing forms are expressed. Recently Games et al. (37) reported Alzheimerlike pathology in mice expressing human ␤APP with the ␤APP V717F mutation linked to familial AD. The ␤APP V717F transgene in these mice was driven by platelet-derived growth factor promoter, and it contained portions of APP introns 6 -8, allowing alternative splicing of exons 7 and 8 by the host. With this construct, this group achieved marked overexpression of human ␤APP in the transgenic mouse brain, and they observed extracellular amyloid deposition, dystrophic neuritic components, and gliosis that resembled AD. In the brains of these transgenic mice, the predominant human mRNA expressed was APP751. In a series of studies on transgenic mice expressing human ␤APP751 at lower levels than those in the transgenic mice of Games et al., Cordell and her colleagues have reported an age-related, subtle increase in A␤ deposition and tau staining (38 -41) as well as age-dependent deficits in spa-tial learning in a water maze task and in spontaneous alternation in a Y maze (42). In their discussion of these animals with subtle histopathological changes (42), the authors indicate that transgenic mice expressing ␤APP695 do not develop similar histopathological (38,39) or behavioral (29) changes. Based on these suggestive findings in transgenic mice and the results that we report here, we speculate that the level of expression of the KPI domain may influence the development of AD by altering ␣ secretase cleavage in a way that changes the specific species of A␤ which are secreted.  Fig. 3A. b p Ͻ 0.001 with respect to ␤APP751 clones by nonparametric Mann-Whitney test.