Processing of beta-amyloid precursor protein by cathepsin D.

The events leading to the formation of β-amyloid (βA4) from its precursor (βAPP) involve proteolytic cleavages that produce the amino and carboxyl termini of βA4. The enzyme activities responsible for these cleavages have been termed β- and γ-secretase, respectively, although these protease(s) have not been identified. Since βA4 is known to possess heterogeneity at both the amino and carboxyl termini, β- and γ-secretases may actually be a collection of proteolytic activities or perhaps a single proteolytic enzyme with broad amino acid specificity. We investigated the role of cathepsin D in the processing of βAPP since this enzyme has been widely proposed as a γ-secretase candidate. Treatment of a synthetic peptide that spans the γ-secretase site of βAPP with human cathepsin D resulted in the cleavage of this substrate at Ala42-Thr43. A sensitive liquid chromatography/mass spectrometry technique was also developed to further investigate the ability of cathepsin D to process longer recombinant βAPP substrates (156 and 100 amino acids of βAPP carboxyl terminus) in vitro. The precise cathepsin D cleavage sites within these recombinant βAPP substrates were identified using this technique. Both recombinant substrates were cleaved at the following sites: Leu49-Val50, Asp68-Ala69, Phe93-Phe94. No cleavages were observed at putative γ-secretase sites: Val40-Ile41 or Ala42-Thr43, suggesting that cathepsin D is not γ-secretase as defined by these βA4 termini. Under conditions where the βAPP156 substrate was first denatured prior to cathepsin D digestion, two additional cleavage sites near the amino terminus of βA4, Glu−3-Val−2 and Glu3-Phe4, were observed, indicating that cathepsin D cleavage of βAPP is influenced by the structural integrity of the substrate. Taken together, these results indicate that in vitro, cathepsin D is unlikely to function as γ-secretase; however, the ability of this enzyme to efficiently cleave βAPP substrates at nonamyloidogenic sites within the molecule may reflect a role in βAPP catabolism.

The events leading to the formation of ␤-amyloid (␤A4) from its precursor (␤APP) involve proteolytic cleavages that produce the amino and carboxyl termini of ␤A4. The enzyme activities responsible for these cleavages have been termed ␤and ␥-secretase, respectively, although these protease(s) have not been identified. Since ␤A4 is known to possess heterogeneity at both the amino and carboxyl termini, ␤and ␥-secretases may actually be a collection of proteolytic activities or perhaps a single proteolytic enzyme with broad amino acid specificity.
We investigated the role of cathepsin D in the processing of ␤APP since this enzyme has been widely proposed as a ␥-secretase candidate. Treatment of a synthetic peptide that spans the ␥-secretase site of ␤APP with human cathepsin D resulted in the cleavage of this substrate at Ala 42 -Thr 43 . A sensitive liquid chromatography/mass spectrometry technique was also developed to further investigate the ability of cathepsin D to process longer recombinant ␤APP substrates (156 and 100 amino acids of ␤APP carboxyl terminus) in vitro. The precise cathepsin D cleavage sites within these recombinant ␤APP substrates were identified using this technique. Both recombinant substrates were cleaved at the following sites: Leu 49 -Val 50 , Asp 68 -Ala 69 , Phe 93 -Phe 94 . No cleavages were observed at putative ␥-secretase sites: Val 40 -Ile 41 or Ala 42 -Thr 43 , suggesting that cathepsin D is not ␥-secretase as defined by these ␤A4 termini. Under conditions where the ␤APP156 substrate was first denatured prior to cathepsin D digestion, two additional cleavage sites near the amino terminus of ␤A4, Glu ؊3 -Val ؊2 and Glu 3 -Phe 4 , were observed, indicating that cathepsin D cleavage of ␤APP is influenced by the structural integrity of the substrate. Taken together, these results indicate that in vitro, cathepsin D is unlikely to function as ␥-secretase; however, the ability of this enzyme to efficiently cleave ␤APP substrates at nonamyloidogenic sites within the molecule may reflect a role in ␤APP catabolism.
A major component of the extracellular deposits found in Alzheimer's disease (AD) 1 brain tissue is a 39 -43-residue pep-tide called ␤-amyloid (␤A4) (1,2). The initial discovery of ␤A4 as a constituent in AD tissue prompted a number of investigations into the role of ␤-amyloid in this disease. The physical properties of ␤A4 combined with the genetic evidence linking mutations around the ␤A4 domain of the ␤-amyloid precursor protein (␤APP) to early onset forms of the disease (reviewed in Refs. 3 and 4) provide compelling evidence that this peptide is involved in the etiology of Alzheimer's disease. Unraveling the cellular mechanism of ␤APP processing and of ␤A4 production is, therefore, important in understanding the pathogenesis of AD.
␤A4 is formed by the proteolytic cleavage of a 695-770-amino acid integral membrane protein, the ␤APP at two distinct sites near the carboxyl terminus of this molecule. The proteolytic cleavage that generates the amino terminus of ␤A4 occurs approximately 28 residues extralumenal to the transmembrane domain of ␤APP, while that which generates the carboxyl terminus of ␤A4 takes place within the transmembrane domain. The protease activities that mediate these cleavage events are called ␤and ␥-secretase, respectively. Cleavage at both the ␤and ␥-secretase sites produces a soluble form of ␤A4 that is 40 amino acids long and ends at Val 40 (5); however, longer and potentially more pathogenic forms of ␤A4 that end at Ala 42 -Thr 43 , have also been identified (2,6,7). Whether these various forms of ␤A4 are produced by a single ␥-secretase is not known. A third major cleavage that is not involved in the formation of ␤A4 occurs within the ␤A4 domain of ␤APP and is generated by a protease activity called ␣-secretase.
A number of recent studies have revealed important clues regarding the proteolytic processing of the ␤APP molecule. From these studies, two processing pathways have been proposed. The first is a constitutive secretion pathway that generates soluble forms of ␤APP, which are ultimately released by cells expressing the precursor protein (8). This pathway involves the primary proteolytic cleavage of ␤APP at the ␣-secretase site located within the ␤A4 domain (9, 10), releasing the large, soluble extracellular portion of ␤APP while simultaneously generating a 10-kDa membrane-associated ␤APP carboxyl-terminal fragment (11). This cleavage event precludes the formation of ␤A4. A minor cleavage event that also serves to liberate a soluble, albeit shorter extracellular domain of ␤APP, occurs at the ␤-secretase site (12). It is not yet clear whether the protease responsible for this cleavage is a true component of the secretory pathway, or whether it is involved in the processing of ␤APP via a degradative pathway.
The second ␤APP processing pathway is one that leads to the formation of amyloidogenic carboxyl-terminal fragments of ␤APP (13-16) that appear to be the direct precursors to ␤A4 itself (17). Factors that may regulate the processing of ␤APP through this amyloid forming pathway include ␤APP muta-tions and the expression of aberrant levels and/or isoforms of ␤APP (18). This amyloidogenic pathway was initially thought to involve the lysosomal-endosomal system based on the observation that lysosomal inhibitors stabilized amyloidogenic carboxyl-terminal fragments (15, 16, 19 -22). The observation that full-length surface ␤APP can be reinternalized (15) via a coated-pit mechanism (23,24) added further support for the role of the lysosomal-endosomal system. However, the observation that lysosomal inhibitors have little or no effect on ␤A4 levels suggests that the subcellular site of ␤A4 production may alternatively involve nonlysosomal compartments such as the trans-Golgi or early endosome (17,21,25,26).
Using a variety of peptide or protein substrates that span the ␣-secretase cleavage site, a number of protease activities have been implicated as ␣-secretase (27), but the true enzyme(s) has yet to be definitively identified. The search for ␣-secretase is confounded by the possibility that multiple proteases may be involved in cleaving this site (28).
The ␤and ␥-secretases responsible for processing ␤APP through the amyloidogenic pathway also remain a mystery despite numerous attempts to identify them. Using a variety of synthetic peptide and recombinant ␤APP substrates, a number of proteases have been implicated as candidates for ␤-secretase based on their ability to cleave at or near the Met 0 -Asp 1 bond of ␤A4. These proteases include multicatalytic proteases, serine proteases, metalloproteases, and aspartic acid proteases (as reviewed in Refs. 27 and 29). Proteases implicated as ␥-secretases include multicatalytic protease, prolyl endopeptidase, and cathepsin D (27,29). Of these enzymes, cathepsin D is of particular interest, since many of the properties associated with this protease favor its role in the amyloidogenic processing of ␤APP.
The features of cathepsin D relevant to ␤A4 generation include the following observations. Cathepsin D is an abundant aspartic protease in brain tissue (30) and is located in and is active within acidic lysosomal and endosomal compartments (31). This protease is also associated with amyloid deposits in AD tissue (32,33), is up-regulated in AD neurons (34), and shows an age-related change in activity in human and rodent brain tissue (35,36). Furthermore, cathepsin D exists in both a soluble and a membrane-associated form (37,38), the latter of which favors its role in processing membrane-associated proteins such as ␤APP. Cathepsin D also displays an amino acid specificity consistent with the amino acid sequences in the vicinity of putative ␤and/or ␥-secretase sites (39). Evidence against a role of cathepsin D in the formation of ␤A4 includes the fact that the levels of cathepsin D activity in AD tissue are not significantly different from those of controls (40) and that the overexpression of this enzyme in cell culture does not enhance the level of ␤A4 secretion (41).
To determine the ability of cathepsin D to function as either a ␤or ␥-secretase, in vitro assays were developed to study the proteolytic activity of cathepsin D on synthetic peptide substrates that span either the ␤or ␥-secretase sites (27,29,42,43). Although these studies showed that cathepsin D is able to cleave these substrates at sites consistent with either a ␤or a ␥-secretase, better evidence implicating cathepsin D as a secretase enzyme was obtained by studying the in vitro activity of cathepsin D on ␤APP itself. The production of amyloidogenic carboxyl-terminal fragments of ␤APP starting from a ␤APP695 substrate was a demonstration of cathepsin D's ability to cleave ␤APP near the ␤-secretase site and, hence, an illustration of a possible ␤-secretase activity for this enzyme (44). Likewise, cathepsin D was able to further process an amyloidogenic carboxyl-terminal ␤APP substrate (corresponding to a ␤-secretase product), suggesting a ␥-secretase activity for the enzyme (42).
However, in these studies, since the precise cathepsin D cleavage sites were not characterized at the amino acid sequence level and since no ␤A4 was detected, the ability of cathepsin D to cleave ␤APP substrates at relevant sites was not fully established.
In this study, we further investigated the potential role of cathepsin D in ␤APP processing by developing sensitive in vitro assays for evaluating the proteolytic activity of this enzyme on various ␤APP substrates. Improvements in the methods used to detect and to characterize hydrophobic cleavage products has allowed us to identify precise ␤APP cleavage sites within the vicinity of the ␤A4 domain. These cleavage sites provide important clues regarding the function of cathepsin D in ␤APP processing.

MATERIALS AND METHODS
Enzyme, Synthetic Substrate, and Antibodies-Bovine cathepsin D (EC 3.4.23.5), 15 units/mg of protein, was from Sigma. Human liver cathepsin D (8 units/mg of protein) was either from Calbiochem or purified according to published procedures (45). Enzyme purity was assessed by standard SDS-polyacrylamide gel electrophoresis and reversed phase methods. In addition, control assays performed in the presence of the aspartic protease inhibitor, pepstatin, showed complete inhibition of all protease activity, confirming the lack of contaminating protease activities. One cathepsin D unit produces an increase in A 280 of 1.0/min at pH 3.0, 37°C measured as trichloroacetic acid-soluble products using hemoglobin as a substrate. The synthetic peptide: Ac-Orn-GGVVIATVI-Orn-NH 2 was synthesized by standard solid phase methods on an Applied Biosystems Biosynthesizer (Applied Biosystems, Foster City, CA) and purified to single peak homogeneity by RP-HPLC. This peptide spans the ␤A4 ␥-secretase site (Gly 37 -Ile 45 ) but is very hydrophobic since it comprises a portion of the transmembrane domain of ␤APP. To enhance the solubility of this peptide in the low pH range, ornithine residues were included at the amino-and carboxyl-terminal ends. The polyclonal antiserum, BC-1, directed to the cytoplasmic domain of ␤APP (705-730 of ␤APP751 sequence) was described previously (17). Anti-FLAG M2 affinity column was from Eastman Kodak Co. Brij35 and Brij30 were purchased from Sigma.
Expression and Purification of FLAG-␤APP156-bFGF Substrate-In order to construct the DNA encoding ␤APP156 tagged with FLAG at the amino terminus and a bFGF epitope at the carboxyl terminus, a FLAG-␤APP751 was first made by inserting a double-stranded synthetic oligonucleotide encoding the signal sequence of ␤APP, a FLAG epitope, and the first three amino acids of ␤APP into a KpnI-SmaI-digested pGEM vector containing ␤APP751. This insertion added the FLAG sequence immediately 5Ј to the ␤APP751 sequence. Next, the sequence encoding a bFGF epitope was added to the 3Ј end of the ␤APP sequence. This was accomplished by first introducing a MluI restriction site at the 3Ј end of ␤APP751 by in vitro mutagenesis. A double-stranded synthetic oligonucleotide encoding the bFGF tag was then ligated into this MluI site. This FLAG-␤APP751-bFGF vector was then used to construct a baculovirus FLAG-␤APP156-bFGF expression vector.
The FLAG-␤APP751-bFGF plasmid DNA was digested with KpnI and BsmI to remove a 1745-base pair sequence encoding the aminoterminal residues of ␤APP751 from Pro 4 to Trp 602 . A 23-base pair double-stranded oligonucleotide was ligated into the KpnI-BsmI sites to create a fusion between ␤APP751 Pro 4 -Leu 596 . This FLAG-␤APP156-bFGF construct was then digested with SmaI-XmnI and the sequence encoding FLAG-␤APP156-bFGF was blunt-end ligated into filled-in NheI-digested baculovirus expression vector (Invitrogen, San Diego, CA). The chimeric baculovirus expression vector was used to express the ␤APP derivative in SF9 insect cells.
A 1-liter culture of infected SF9 cells was harvested and lysed in PBS containing 1% (v/v) Triton X-100 (reduced) and 5 mM EDTA. The cell suspension was adjusted to 0.1% Triton X-100 by dilution with PBS and centrifuged at 27,000 ϫ g, for 20 min. The resulting supernatant was applied to an anti-bFGF affinity column and eluted with 0.1 M sodium citrate, pH 3.0 containing 0.1% Triton X-100. Fractions containing FLAG-␤APP156-bFGF were immediately neutralized with Tris, adjusted to 0.1 M CaCl 2 , and applied to an anti-FLAG M1 affinity column. FLAG-␤APP156-bFGF was eluted with PBS containing 0.1% Triton X-100 and 5 mM EDTA.
Expression and Purification of ␤APP100-FLAG Substrate-The ␤APP100-FLAG used in this study was expressed and purified as de-scribed by Mackay et al. 2 Digestion of Ac-Orn-GGVVIATVI-Orn-NH 2 with Cathepsin D-A 6-g sample of Ac-Orn-GGVVIATVI-Orn-NH 2 was digested with 1 g (0.008 units) of human cathepsin D in 80 l of 50 mM sodium citrate, pH 3.5, for 2 h at 37°C. The enzyme to substrate molar ratio was 1:200. Following digestion, the sample was immediately analyzed by RP-HPLC on a Hewlett Packard HP 1050Q automated LC system equipped with a Vydac C18 (0.46 ϫ 15 cm) column equilibrated in 0.1% (v/v) trifluoroacetic acid. Elution was with a linear acetonitrile gradient (1%/min) at room temperature. Absorbance was monitored at 215 nm.
Digestion of FLAG-␤APP156-bFGF with Cathepsin D-A 3-ml sample of purified FLAG-␤APP156-bFGF (90 g) was diluted to 5 ml with 50 mM sodium citrate, pH 3.5. A 55-g sample (0.83 unit) of bovine cathepsin D was added to the substrate to yield a final enzyme to substrate molar ratio of 1:5. The mixture was allowed to digest at 37°C for variable time periods. A 2-h digestion period was routinely used after it was found that the major product peaks were present in greatest abundance after this time period. After digestion, a 500-l sample was immediately subjected to RP-HPLC separation on a Vydac C4 column (0.46 ϫ 5 cm) equilibrated in 20 mM ammonium acetate, pH 6.8. Elution was performed at a constant flow rate of 1.0 ml/min with a linear acetonitrile gradient (3.75%/min) at room temperature. Absorbance was monitored at 215 nm. Urea-denatured ␤APP156 was prepared by resuspending 100 g of ␤APP156 in 100 l of 8.0 M urea. A 50-l sample was then added to 450 l of 50 mM sodium citrate, pH 3.5, and 3.0 g (0.05 unit) of bovine cathepsin D were added. Digestion was allowed to proceed for 2 h at 37°C. Following digestion, a 200-l aliquot was injected onto a Vydac C18 column (0.46 ϫ 15 cm) equilibrated in 0.1% trifluoroacetic acid. The sample was eluted with an acetonitrile gradient (1.3%/min). Absorbance was monitored at 215 nm.
Digestion of ␤APP100-FLAG with Cathepsin D-A 30-l sample of purified ␤APP100 (24 g) was diluted to 75 l with 50 mM sodium citrate, pH 3.5, and digested with a 2.4-g sample (0.02 unit) of human cathepsin D. The enzyme to substrate molar ratio was 1:30. The mixture was allowed to digest at 37°C for variable time periods. As with the ␤APP156 substrate described above, a 2-h digestion period was routinely used after it was found that the major product peaks were present in greatest abundance after this time period. After digestion, a 10-l sample was subjected to capillary RP-HPLC separation using an HP1050 LC system adapted for capillary HPLC separations as described in Guzzetta et al. 3 A constant LC flow rate setting of 0.2 ml/min provided a column flow rate of 8 l/min. The capillary column used in these experiments was a 25 cm ϫ 0.32-mm inside diameter, 5-m Vydac C18 column (300 Å) (Microtech Scientific, Sunnyvale, CA). The column was equilibrated with 0.1% trifluoroacetic acid and eluted with acetonitrile at room temperature. Absorbance was monitored at 215 nm using a Kratos/ABI 757 UV detector (Applied Biosystems, Foster City, CA) equipped with a capillary UV flow cell (LC Packings, Inc., San Francisco, CA).
Western Blot Analysis of ␤APP156 after Digestion with Cathepsin D-Peak fractions collected off the RP-HPLC were dried down, redissolved in Laemmli sample buffer, and separated on a 16.5% Tris-HCl-Tricine polyacrylamide gel. Following electrophoresis, the gel was transferred to a polyvinylene difluoride membrane, blocked with blocking buffer (4% (w/v) fish gelatin, 2% (w/v) nonfat dry milk in PBS) for 1 h, then transferred to blocking buffer containing a 1:300 dilution of BC-1 polyclonal antisera (17). The filter was then developed with an anti-rabbit Vectastain ABC Elite Kit (Vector Laboratories, Burlingame, CA) and reacted with the horseradish peroxidase substrate, 3,3Ј-diaminobenzidine (Vector Laboratories).
Amino Acid Sequence and Mass Spectral Analysis-Peak fractions from the cathepsin D digestion of ␤APP156 were collected off a RP-HPLC column (Vydac C4, 0.46 ϫ 15 cm), dried, and subjected to Edman degradation using an Applied Biosystems 477A protein sequencer inline with a 120A phenylthiohydantoin analyzer (Applied Biosystems, Foster City, CA). Alternatively, to reduce sample loss and to enhance sensitivity, capillary RP-HPLC fractions were collected directly onto Porton filters (Beckmen Instruments) and subjected to sequence analysis using an ABI 470A gas-phase sequencer in-line with an Hewlett Packard HP1090 HPLC. This latter technique allowed the direct sequencing of as little as 2-5 pmol of sample.
HPLC-mass spectral (LC/MS) analysis was performed using an SSQ 7000 Finnigan MAT mass spectrometer equipped with a Finnigan electrospray source. The spray voltage was set at 4500 volts.

RESULTS
Digestion of Synthetic Peptide with Cathepsin D-To determine whether cathepsin D is capable of cleaving at the ␥-secretase site, a synthetic peptide spanning residues 37-45 of ␤A4 (Ac-Orn-GGVVIATVI-Orn-NH 2 ) was subjected to digestion with human liver cathepsin D. The digestion products were separated by RP-HPLC, and the resulting chromatogram is shown in Fig. 1. The full-length peptide substrate eluted with a retention time of 14.7 min. This peak was well separated from two product peptides that eluted with retention times of 8.2 and 11.4 min. Mass spectral analysis of these two products identified the 8.2-min peak as TVI-Orn-NH 2 (MH ϩ mass ϭ 445.3) and the 11.4 min peak as Ac-Orn-GGVVIA (MH ϩ mass ϭ 671.4). Thus, the intact peptide substrate was cleaved by cathepsin D predominantly (Ͼ95% of total digestion) between Ala 42 and Thr 43 . The peaks eluting with retention times of 2, 3.5, and 12.5 min are background peaks (perhaps buffer components) not related to the peptide substrate or any of the products formed during the digestion. In addition to the Ala 42 -Thr 43 cleavage, the mass spectral analysis of the digestion revealed a minor (Ͻ5%) product peptide formed by a cleavage between Thr 43 and Val 44 (data not shown). Identical results were obtained using bovine cathepsin D as well, thus cathepsin D is capable of cleaving this synthetic peptide predominantly carboxyl-terminal to Ala 42 with a minor cleavage after Thr 43 .
Digestion of FLAG-␤APP156-bFGF with Cathepsin D-Although the synthetic peptide Ac-Orn-GGVVIATVI-Orn-NH 2 can be used to assay cathepsin D cleavage at a ␥-secretase-like site, caution should be displayed when interpreting this cleavage activity as true ␥-secretase activity, since a synthetic peptide might not display all the primary, secondary, and tertiary structural elements of a natural protein substrate. A more appropriate substrate for ␥-secretase is full-length ␤APP which presumably contains many of the structural requisites for protease recognition. However, since full-length ␤APP is a large, complex protein consisting of multiple domains and glycosylation sites, the use of this large protein as a secretase substrate is not straightforward. Therefore, we decided to simplify this protein by omitting much of the extracellular domain. The final ␤APP substrate used in these experiments (FLAG-␤APP156-bFGF) includes the last 156 amino acids of ␤APP751 (␤APP156) and contains the entire ␤A4 domain as well as the transmembrane and intracellular carboxyl-terminal domains. To facilitate the purification and characterization of this protein and its proteolytic products, an 8-amino acid FLAG epitope and a 10-amino acid bFGF epitope were engineered at the amino and carboxyl termini, respectively.
Since the major product peaks were the most abundant after a 2-h digestion, FLAG-␤APP156-bFGF was digested with bovine cathepsin D for this amount of time, and the resulting digest was separated by RP-HPLC (Fig. 2). The large amount of enzyme used here was determined to generate efficient cleavage of the substrate within a short time period where the aggregation and precipitation of the recombinant substrate was not a problem. Carboxyl-terminal fragments of ␤APP, including ␤APP156, are well known to aggregate (9,13,18). When the digestion was carried out at lower enzyme to substrate ratios for extended time periods, a considerable amount of substrate was found to precipitate and resulted in much lower cleavage efficiency. A series of product peaks were generated and the seven primary product peaks were collected for Western blot analysis using BC-1 anti-sera. The resulting Western blot (Fig. 3) identified an ϳ5-kDa carboxyl-terminal fragment of FLAG-␤APP156-bFGF present in product peaks 6 and 7 that contains the cytosolic BC-1 epitope. This fragment is of interest since, based on the known amino acid sequence of

␤-Amyloid Precursor Processing
FLAG-␤APP156-bFGF, the cleavage of this substrate at the ␥-secretase site predicts a BC-1 reactive carboxyl-terminal fragment of this approximate size. Additional Western blot analysis of these fractions using antibodies directed toward other regions of ␤APP confirmed that this ϳ5-kDa fragment is a carboxyl-terminal fragment of ␤APP formed by a proteolytic cleavage at or near the predicted ␥-secretase site. In addition, digestion with human cathepsin D gave identical results (data not shown). All seven major HPLC peaks from this digest were collected separately over several runs to isolate enough material for amino acid sequence analysis. The primary sequence obtained for each of the seven fractions is shown in Table I, and a diagram of the resulting cleavage sites is shown in Fig. 4. FLAG-␤APP156-bFGF was cleaved by cathepsin D at four primary sites. The amino acid sequence obtained for the ϳ5-kDa fragment containing the BC-1 epitope present in fraction 7 shows that this fragment was produced by a cleavage between Leu 49 and Val 50 of ␤APP156. This site is within the putative transmembrane domain of ␤APP, but 7-9 amino acid residues downstream of the carboxyl terminus of the ␤A4 sequence. Fraction 1 consisted of a FLAG-␤APP156-bFGF fragment possessing a carboxyl-terminal of ␤APP fragment formed by cleavage at Phe 93 -Phe 94 . The amino acid sequence of this fragment was also observed in fractions 3, 4, and 6. Fraction 2 contained a FLAG-␤APP156-bFGF fragment generated by a cleavage at Asp 68 -Ala 69 within the cytoplasmic domain of ␤APP, downstream of the putative transmembrane domain. Double amino acid sequences were detected in fractions 4, 5, and 6. Common among these fractions was the amino-terminal FLAG sequence of FLAG-␤APP156-bFGF, indicating that these fractions contained, in part, amino-terminal fragments of the substrate. Fraction 5 also contained a unique secondary sequence generated by a cleavage between Asp Ϫ44 and Ser Ϫ43 of the substrate. Full-length (intact) ␤A4 was not detected by this method of analysis.
In order to determine whether any structural features of FLAG-␤APP156-bFGF might dictate cathepsin D cleavage of this protein, a second experiment was performed in which the substrate was first denatured in 8 M urea prior to digestion with cathepsin D. RP-HPLC of the digest (Fig. 5) and amino acid sequence analysis of the resulting products (Table II) identified three of the four primary cleavage sites described above. In addition, two other cleavage sites were identified. The first was 3 amino acid residues amino-terminal to the ␤A4 domain (Glu Ϫ3 -Val Ϫ2 ), and the second was just inside the ␤A4 domain (Glu 3 -Phe 4 ). Since these cleavages were not detected in the previous digestion of nondenatured ␤APP156, the physical state/structure of the substrate appears to influence the susceptibility of ␤APP156 to cathepsin D cleavage near the amino terminus of the ␤A4 domain. Digestion of ␤APP100-FLAG with Cathepsin D-A more appropriate protein substrate for determining the role of cathepsin D as a ␥-secretase is the carboxyl-terminal 100 residues of ␤APP (␤APP100) since a ␥-secretase cleavage of this carboxylterminal fragment of ␤APP should directly generate a ␤A4-like peptide independent of a ␤-secretase cleavage (17). The ␤APP100-FLAG substrate used in this study contains the last 100 amino acid residues of ␤APP751, starting at Met 0 at the amino terminus of ␤A4 and continuing to the carboxyl terminus of ␤APP751. Previous cell culture results showed that the expression of a similar fragment in COS cells generates large amounts of soluble ␤A4, indicating that this ␤APP fragment contains all the requisite information for a ␥-secretase cleavage (18,25). In addition, the final ␤APP100-FLAG substrate contains an 8-amino acid FLAG epitope at the carboxyl terminus to facilitate the purification of this recombinant protein.
Pure ␤APP100-FLAG was digested with human liver cathepsin D at an enzyme to substrate molar ratio of 1:30 for 2 h, at 37°C. As with the FLAG-␤APP156-bFGF substrate, these conditions prevented extensive aggregation and precipitation of the substrate prior to digestion. The 2-h time point was again chosen for further analysis, since the major products were in the most abundance at this time. The digest was separated by RP-HPLC on a capillary LC system. The resulting chromatogram is shown in Fig. 6. The improved resolution and sensitivity of this system allowed for the detection and collection of 15 major product peaks which were individually collected for both amino acid sequence and electrospray MS analysis. The results are summarized in Table III. Reliable amino-terminal sequences were obtained for 12 of the 15 fractions collected. While most fractions showed one primary sequence, fractions eluting late in the gradient possessed multiple sequences; the two most abundant sequences are shown. To fully characterize the fragments present in each fraction, electrospray MS analysis was also performed, resulting in MH ϩ masses for 10 of the 15 fractions collected. Single mass values were observed for fractions 1-11 (no MS data was obtained for fractions 1, 4, 6, 12, and 15), while fractions 13 and 14 contained multiple fragments as detected by MS analysis. Source collision-induced dissociation for many of these fractions helped to definitively identify the fragments present in each fraction. The amino acid sequences obtained by MS analysis were in good agreement with those obtained by direct sequencing and taken together, provided a complete identification of the ␤APP100 residues comprising each fragment (Table  III). These fragments can be categorized according to their carboxyl-terminal residues. By doing so, four groups of fragments were observed, ending with either Phe 19 , Asp 68 , Phe 93 , or Lys 107 . These related fragments were thus formed by proteolytic cleavages at three primary sites within ␤APP100-FLAG (the Lys 107 -containing fragments originated from the preexisting carboxyl terminus of ␤APP100-FLAG). The heterogeneity observed at the amino-terminal end of these carboxylterminally related fragments indicates that minor cleavages adjacent to or near the primary these cleavage sites occurred. Together, these cleavages identify specific regions of ␤APP that are particularly susceptible to cathepsin D cleavage. Based on the amino-terminal residue of the fragment present in fraction 8, one additional primary proteolytic cleavage site was identified at Leu 49 -Val 50 , located 7-9 amino acid residues downstream of the ␤A4 domain.
The major cleavage sites found within ␤APP100-FLAG were compared to those identified for FLAG-␤APP156-bFGF and were in good agreement (Fig. 4), despite the fact that, in contrast to the Ac-Orn-GGVVIATVI-Orn-NH 2 peptide substrate, neither the Ala 42 -Thr 43 nor the Thr 43 -Val 44 sites were cleaved in the recombinant protein substrates. Three major cleavage sites identified in ␤APP100-FLAG (Leu 49 -Val 50 , Asp 68 -Ala 69 , and Phe 93 -Phe 94 ) were present in FLAG-␤APP156-bFGF. However, the cleavage site observed at Phe 19 -Phe 20 within the ␤A4 domain of ␤APP100-FLAG was not detected for FLAG-␤APP156-bFGF. It is possible that, in FLAG-␤APP156-bFGF, this site is either inaccessible to cathepsin D, due to a conformational difference between the two substrates, or is exposed and cleaved, but the resulting products were not present in great enough abundance for detection and analysis.
The majority of the cathepsin D cleavage sites identified include a hydrophobic amino acid residue at the P-1 position. This is consistent with the known specificity of cathepsin D for such amino acids (39). In addition, cleavages frequently occurred on the carboxyl side of acidic residues. This is not surprising since, under the acidic conditions used in these assays, acidic side chains may be protonated and uncharged, rendering them better targets for cathepsin D recognition. Although other potential cathepsin D cleavage sites exist in both FLAG-␤APP156-bFGF and ␤APP100-FLAG substrates, cleavages at these sites were not observed, again suggesting that the structure of the native substrate molecule is important in dictating preferred cleavage sites of ␤APP.

DISCUSSION
The ␥-secretase cleavage site located at the carboxyl terminus of the ␤A4 domain is situated within the transmembrane domain of the ␤APP molecule. Because of the high degree of hydrophobicity associated with synthetic peptides that span this region, the design of reliable in vitro assays has been FIG. 5. Urea-denatured FLAG-␤APP156-bFGF digested with cathepsin D. A 50-g sample of FLAG-␤APP156-bFGF was denatured in 8 M urea and diluted 1:10 in 500 l of 50 mM sodium citrate, pH 3.5. Bovine cathepsin D was added, and the digestion was allowed to proceed for 2 h at 37°C. After this period, a 200-l aliquot of the digest was separated by RP-HPLC as described under "Materials and Methods." Curve A, digestion of urea-denatured FLAG-␤APP156-bFGF with cathepsin D at an enzyme to substrate molar ratio of 1:40; curve B, urea-denatured FLAG-␤APP156-bFGF alone; curve C, cathepsin D alone.  6. Digestion ␤APP100-FLAG with cathepsin D. A 24-g sample of ␤APP100-FLAG was digested with bovine cathepsin D in 75 l of approximately 50 mM sodium citrate, pH 3.5, for 2 h at 37°C. Capillary RP-HPLC was used to separate the resulting ␤APP100-FLAG products. The 15 major cleavage products (1-15) generated from this digestion were collected for further amino acid sequence and mass spectral analysis as described under "Materials and Methods." difficult and has required alternative methods for handling hydrophobic substrates (42,43). Despite these challenges, studies have identified protease activities that can cleave peptide substrates spanning the ␥-secretase site of ␤APP. Recently, an endogenous aspartic acid protease activity was observed in human brain homogenates that cleaved a synthetic peptide spanning residues 711-716 of ␤APP770 at Ala 42 and at Thr 43 . This activity closely resembled that of purified cathepsin D (43). A similar observation was reported using an immobilized synthetic peptide spanning a larger region around the ␥-secretase site (42). The ability of cathepsin D to form potentially amyloidogenic carboxyl-terminal fragments from a fulllength ␤APP695 precursor protein (44) and to further proteolytically cleave amyloidogenic carboxyl-terminal ␤APP fragments (42) provides evidence that this enzyme plays an important role in ␤APP processing.

VMLKKKQYTSI--GVV
Based on the known specificity of cathepsin D toward hydrophobic amino acids (39), the fact that cathepsin D is an abundant aspartic protease in human brain tissue (30) and that membrane associated forms of cathepsin D are known to exist (37,38), this enzyme is ideally suited for the proteolytic attack of membrane-bound proteins such as ␤APP. Hence, it is a strong candidate for ␥-secretase. We decided to investigate in greater detail the role of cathepsin D as a potential ␥-secretase by using both synthetic peptides and native protein substrates that span the ␥-secretase site of ␤APP. Improved methods for the handling and analysis of hydrophobic proteins and peptides lead to the identification of precise cathepsin D cleavage sites within the ␤APP molecule.
We first used a synthetic peptide (Ac-Orn-GGVVIATVI-Orn-NH 2 ) to determine if cathepsin D can cleave a sequence of amino acid residues that spans the putative ␥-secretase site. The primary cathepsin D cleavage sites within this peptide were found to be carboxyl-terminal to Ala 42 (95% of total digestion) and Thr 43 (5% of total digestion) while cleavage at Val 40 was not observed. Thus, consistent with the earlier studies described above (42,43), cathepsin D is capable of cleaving a synthetic peptide substrate at biologically relevant sites corresponding to the longer (1-42 and 1-43) and potentially more pathogenic forms of ␤A4 (3). Digestion with either human or bovine cathepsin D gave identical results, indicating that the cleavage specificity of this enzyme is conserved across species. Caution must be displayed, however, when interpreting these results to mean that cathepsin D is ␥-secretase since the Ac-Orn-GGVVIATVI-Orn-NH 2 peptide used in these studies contains flanking ornithine residues that could influence cathepsin D cleavage. Furthermore, as with other synthetic peptides, a short sequence of amino acids may lack many of the structural features of the natural ␤APP substrate that might be important in directing protease cleavage. This concern was addressed by using recombinant ␤APP proteins as substrates for cathepsin D.
Two recombinant ␤APP substrates were used in this study. The longer of the two forms, FLAG-␤APP156-bFGF, consists of the last 156 amino acids of ␤APP and contains a portion of the extracellular domain in addition to the entire ␤A4 sequence, the transmembrane domain and the cytoplasmic carboxyl-terminal domain. The second ␤APP substrate, ␤APP100-FLAG is shorter and only consists of the carboxyl-terminal 100 amino acids of ␤APP, starting at Met 0 of the ␤A4 domain. Since the latter substrate has a preformed ␤A4 amino terminus, an authentic ␥-secretase cleavage at the carboxyl terminus of the ␤A4 domain would directly generate ␤A4 peptide species identical to those identified in vitro and in vivo, i.e. terminating at positions 39 -44. Previous results from this laboratory showed that ␤A4 is efficiently produced by cells expressing a similar amino-terminally truncated ␤APP protein indicating that ␤APP100-FLAG possesses all the requisite cleavage sites for ␤A4 production (18).
Unlike the Ac-Orn-GGVVIATVI-Orn-NH 2 peptide substrate described above, neither FLAG-␤APP156-bFGF nor ␤APP100-FLAG was cleaved by cathepsin D directly at putative ␥-secretase sites. Thus, in cleaving ␤APP protein substrates, cathepsin D does not display a ␥-secretase activity. It is plausible that the flanking ornithine residues present in the synthetic peptide, but not in the recombinant substrates, impart an amino acid sequence that is readily recognized by cathepsin D. Alternatively, structural features of the longer ␤APP substrates prevent cathepsin D from recognizing or binding to the putative ␥-secretase site. Additional studies are required to prove either concept.
Although cathepsin D failed to show authentic ␥-secretase activity on recombinant ␤APP substrates (as defined by cleavage at positions 39 -44), both FLAG-␤APP156-bFGF and ␤APP100-FLAG were readily cleaved by this enzyme at other sites along the sequence including the Phe 19 -Phe 20 site of ␤APP100-FLAG located within the ␤A4 domain. These major cleavage sites combined with adjacent minor cleavage sites, localized specific regions of ␤APP that were particularly susceptible to cathepsin D cleavage. The cleavage of ␤APP at these sites suggests that cathepsin D might alternatively be involved in the catabolism of ␤APP or ␤A4 (17,18). The Leu 49 -Val 50 cleavage site located within the putative transmembrane domain, and the Asp 68 -Ala 69 and the Phe 93 -Phe 94 cleavage sites located within the cytoplasmic domain of ␤APP were found to be preferred cathepsin D cleavage sites in both recombinant protein substrates.
The similarity in cleavage patterns between FLAG-␤APP156-bFGF and ␤APP100-FLAG indicates that cathepsin D cleavage on the carboxyl-terminal side of the ␤A4 domain is not influenced by the presence of amino acid residues upstream of the ␤A4 domain. Of the three cathepsin D cleavage sites common to both FLAG-␤APP156-bFGF and ␤APP100-FLAG, the Leu 49 -Val 50 is of particular interest since this site is located 7 amino acid residues downstream of the carboxyl terminus of ␤A4. It is conceivable that cathepsin D might play a role in the generation of ␤A4 by initially processing ␤APP at this downstream site (contingent on prior removal/exposure of this ␤APP site from within the membrane), followed by a secondary processing event(s) to liberate the final collection of carboxyl termini of ␤A4 ending at positions 39 -44. Such a secondary processing event might involve a carboxypeptidase as discussed previously (46,47). Whether ␤A4 (1-40) or (1-42) is ultimately produced might then be regulated by the extent of carboxypeptidase processing (46,47).
The inability of cathepsin D to cleave ␤APP at other potential cathepsin D (hydrophobic) sites suggests that the structural configuration of the substrate influences and/or defines susceptibility to proteolytic cleavage. Evidence for this was obtained by digesting urea-denatured FLAG-␤APP156-bFGF with cathepsin D. In addition to three of the cleavage sites common to both ␤APP156 and ␤APP100 identified above, urea denatured ␤APP156 was cleaved at two additional sites near the amino terminus of the ␤A4 domain, Glu Ϫ3 -Val Ϫ2 and Glu 3 -Phe 4 . The Glu Ϫ3 -Val Ϫ2 cleavage is 3 amino acids amino-terminal to the Asp 1 of ␤A4, while the Glu 3 -Phe 4 cleavage is 2 amino acids carboxyl-terminal to Asp 1 . Cleavages at these sites suggest that ␤APP is folded in such a way that the amino terminus of ␤A4, under native conditions, is sterically/structurally unavailable to cathepsin D cleavage. Upon denaturation, this portion of ␤APP may become more susceptible to proteolytic cleavage.
The Glu 3 -Phe 4 amino-terminal cleavage is consistent with reports that describe forms of ␤A4 beginning with Phe 4 observed to be produced by cultured cells (48,49). In addition, the Glu Ϫ3 -Val Ϫ2 cleavage was previously identified using a ␤A4 peptide substrate that spans residues Ile Ϫ5 -Asp 7 (N-dansyl-␤APP-(591-601)-amide) (44). This cleavage is consistent with a minor form of ␤A4 produced by cultured cells, the aminoterminal of which begins with Val Ϫ2 (21). Thus, the Glu 3 -Phe 4 and the Glu Ϫ3 -Val Ϫ2 cleavages suggest that cathepsin D may function as a ␤-secretase. Actual in vivo ␤-secretase activity by cathepsin D would depend on enzyme:substrate accessibility and on favorable intracellular conditions for cathepsin D activity. The recent finding that cathepsin D is capable of appropriately cleaving peptide substrates containing the Swedish KM-NL mutation is in support of the potential role of this enzyme as a ␤-secretase (50), although it remains to be seen whether ␤-secretase cleavage involves a single or multiple proteolytic activities.
␤APP contains a type 1 transmembrane domain and as such is an integral membrane protein. Since the ␥-secretase site is located within the transmembrane domain, it must normally be situated within a phospholipid bilayer and hence, protected from proteolytic cleavage. In order for ␥-secretase to cleave at this site, it must first become accessible to the enzyme. A number of mechanisms have been proposed to explain how a ␥-secretase cleavage can occur in ␤APP when this site is buried within the transmembrane domain. These mechanisms include membrane damage or degradation, a shorter transmembrane domain, allowing exposure to cleavage (51), or a less rigid and more permeable membrane (52). All mechanisms potentially explain how the ␥-secretase site may become accessible to proteolytic cleavage. For mechanisms not involving membrane damage, what role, if any, the intact membrane has on directing proteolytic cleavage once the ␥-secretase site becomes exposed is unknown.
In summary, this study has found that, although cathepsin D possess many of the features expected of a ␥-secretase, this enzyme does not show a precise ␥-secretase activity on recombinant ␤APP substrates in vitro. Thus, it is unlikely that cathepsin D is a ␥-secretase candidate. Our conclusion is supported by results with cathepsin D "knock-out" mice whose cultured neurons efficiently produce ␤A4. 4 The ability of cathepsin D to readily cleave ␤APP at other "nonamyloidogenic" sites, suggests that it might alternatively play a role in ␤APP and/or ␤A4 catabolism. The in vitro assays developed in this report will help in the characterization of other known and novel brain proteases that may be involved in the amyloidogenic processing of ␤APP. The identification of the authentic ␤APP processing enzyme(s) will pave the way toward the development of therapeutic inhibitors to prevent the formation of ␤A4.