Beta-secretase processing of the beta-amyloid precursor protein in transgenic mice is efficient in neurons but inefficient in astrocytes.

Alzheimer's disease is characterized by the extracellular deposition of β-amyloid peptide (Aβ) in cerebral plaques and evidence is accumulating that amyloid is neurotoxic. Aβ is derived from the β-amyloid precursor protein (APP). Proteolytic processing of APP by the enzyme, β-secretase, produces the N terminus of Aβ, and releases a secreted ectodomain of APP (β-s-APP). To develop animal models for measuring β-secretase activity in specific brain cells in vivo, we have targeted the expression of the full-length human APP to either neurons or astrocytes in transgenic mice using the neuron- specific enolase (NSE) promoter or a modified glial fibrillary acidic protein (GFAP) gene, respectively. The APP cDNAs expressed were mutated (KM to NL at 670/671) to encode amino acid substitutions that enhance amyloidogenic processing in vitro. Western analyses revealed abundant production of β-s-APP in the brains of NSE-APP mice and enzyme-linked immunosorbent assay analyses showed production of Aβ in fetal primary mixed brain cultures and brain homogenates from these transgenic animals. Because the NSE promoter drives expression primarily in neurons, this provides in vivo evidence that the β-secretase cleavage necessary for generation of β-s-APP and Aβ is efficiently performed in neurons. In contrast, only little β-s-APP was detected in brain homogenates of GFAP-APP mice, indicating that astrocytes show very little β-secretase activity in vivo. This provides strong in vivo evidence that the major source of Aβ in brain is from neurons and not from astrocytes.

The extracellular deposition of ␤-amyloid peptide (A␤) 1 in senile plaques is an early and invariant feature of Alzheimer's disease (AD). This 39 -43-amino acid peptide is the major component of plaques and is proteolytically processed from the ␤-amyloid precursor protein (APP) (1,2). APP is expressed in all tissues, and the relative amount of A␤ processed from APP varies in different cell types in culture (3)(4)(5). The cellular source of A␤ deposited into plaques in the brain is unknown. Mutations in APP are responsible for some forms of familial AD, supporting the hypothesis that APP and A␤ are central to the disease process (6). Missense mutations immediately Nterminal to the A␤ region of APP 2 lead to a 5-10-fold enhancement of A␤ produced from APP in vitro, strongly supporting the role of A␤ in the development of AD in this family (7,8). Other families, with mutations at the 717 position of APP, have been shown to produce increased amounts of the more amyloidogenic 42-amino acid form of A␤ from APP (9). These findings suggest that factors governing the metabolic processing of APP play a direct pathogenic role in Alzheimer's disease.
The majority of APP is cleaved in the middle of the A␤ region, releasing a secreted ectodomain containing the first 16 amino acids of A␤ (␣-s-APP). This processing, mediated by an unidentified enzymatic activity termed "␣-secretase" precludes A␤ formation (10). In an alternative pathway, cleavage between Met 671 and Asp 672 by a likewise unidentified enzyme named "␤-secretase" produces the N-terminal end of A␤ and releases a secreted ectodomain composed of a truncated form ending with Met 671 (␤-s-APP) (4). Further processing of the C-terminal end of A␤ leads to the release of A␤ (3,5,11,12). This alternative processing pathway has been identified in vitro in a variety of cell types; it appears to be more prominent in primary fetal mixed brain cultures (4) than in cells of peripheral origin. While previous studies have allowed enormous progress in elucidating the processing of APP in vitro, an in vivo model for ␤-secretase processing of APP would allow for characterization of these pathways in a living animal.
To obtain in vivo models for studying ␤-secretase processing of APP in the brain, we have generated transgenic mice expressing the 751-amino acid form of human APP in either neurons or astrocytes. A mutated form of APP was expressed and a specific antibody to ␤-s-APP of that form used to assay the products of ␤-secretase in brain extracts. We report that ␤-secretase processing of APP was highly effective when this molecule was expressed in neurons and relatively ineffective when it was expressed in astrocytes of transgenic mice. In addition to helping identify the main cellular source of A␤ in the brain, these and related models should prove valuable for the screening for drugs that inhibit ␤-secretase activity.

MATERIALS AND METHODS
Generation of Vectors-The rat NSE (13) and murine GFAP (14,37) regulatory sequences have been described previously. A 4.2-kb BglII * This work was supported by National Institutes of Health Grant AG11385 (to L. Mucke) and by Athena Neurosciences, Inc. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
¶ To whom correspondence should be addressed: Athena Neurosciences, 800 Gateway Blvd., South San Francisco, CA 94080. Tel.: 415-877-0900; Fax: 415-877-8370; E-mail: amyloid!lisam@ uunet.uu.net. 1 The abbreviations used are: A␤, ␤-amyloid peptide; APP, ␤-amyloid precursor protein; AD, Alzheimer's disease; ␤-s-APP, secreted ectodomain of APP produced by ␤-secretase; ␣-s-APP, secreted ectodomain of APP produced by ␣-secretase; NSE, neural specific enolase; GFAP, glial fibrillary acidic protein; PCR, polymerase chain reaction; kb, kilobase pair(s); PEI, polyethyleneimine; ELISA, enzyme-linked immunosorbent assay; KLH, keyhole limpet hemocyanin. fragment containing the promoter, exon 1, intron 1, and a portion of exon 2 was obtained from S. Forss-Petter (15). The 4.1-kb fragment extending from the upstream BglII site to the initiating NSE ATG was fused to the initiating methionine of the 751 form of the APP cDNA (16,17) using PCR mutagenesis and standard subcloning techniques. Two PCR reactions containing either oligonucleotides 260 (GCG GGA GAT CTT TGC TCG GCA CGG CCC) and 1072 (CCC AGC CAT CAT GCT GCC CGG GTT GGC) or oligonucleotides carman 2 (ACC TGC CAC TAT ACT GGA ATA) and 1073 (GCC AAC CCG GGC AGC ATG ATG GCT GGG ATC TC) generated fragments that were linked together in a third PCR reaction using oligonucleotides 260 and carman 2. The product of this reaction was cleaved with KpnI to provide the 1-kb NSE/APP fusion DNA fragment. The 3.1-kb BglII to KpnI fragment of the NSE promoter was ligated to the 5Ј end of the NSE/APP fusion fragment and the 3Ј end of the NSE/APP fusion fragment was ligated to the KpnI to SpeI APP cDNA fragment. The PCR reactions were carried out using standard Cetus/Perkin-Elmer reagents with 25 rounds of 96°C for 90 s, 62°C for 30 s, and 72°C for 90 s, followed by 10 min at 72°C. Thus splicing sequences 5Ј to the APP cDNA sequences are provided by intron 1 of the NSE gene in all NSE constructs. Swedish (KM to NL at positions 670/671 of the 770-amino acid APP form) mutations were introduced into APP by PCR mutagenesis and subcloned into APP using BglII and EcoRI (Swedish). All PCR-generated regions were verified by sequencing. In NSE-APP constructs, the SpeI site in the APP cDNA was fused either with SV40 early polyadenylation signals provided by the BclI to EcoRI fragment of SV40 or with the XhoI to PstI fragment of pL1 containing SV40 16S/19S late splice sequences (18) plus SV40 early polyadenylation signals. The BamHI and EcoRI sites in the SV40 DNA were destroyed. The PvuII to EcoRI fragment of pBR322 providing plasmid replication sequences was flanked with NotI sites and placed with the PvuII site abutted to the 5Ј end of the NSE promoter.
For the generation of the GFAP-APP fusion gene, a cDNA encoding full-length human 751 form APP carrying the Swedish mutations (introduced as described above) was excised with NruI and SpeI, ligated with NotI linkers, and cloned via NotI into exon 1 of the modified murine GFAP gene as described (14).
Derivation of Transgenic Mice-The APP-expressing vectors were used to make transgenic animals from B6ϫDBA (NSE) or B6ϫSJL (GFAP) mice using standard microinjection techniques. Founder transgenic animals, in which multiple copies of unrearranged vector DNA was inserted, were identified by Southern analysis of tail-derived DNA. The Chelsea 58 NSE line was subsequently crossed for three generations onto the B6ϫSJL background to rule out potential effects of strain differences. The strain background had no effect on experimental results.
Primary Culture-Primary cortical cultures were made from embryonic day 17 embryos by the method of Koh and Choi (19) with the following modifications. Falcon six-well plates were coated with polyethyleneimine (PEI, Sigma). A 5% PEI solution was diluted 1:100 in 150 mM sodium borate, pH 8.3-8.5, and incubated on the plates at room temperature overnight. The PEI was removed and plates washed two times in sterile phosphate-buffered saline. Plating medium was placed on the plates for at least 1 h prior to plating the cells. The tissue was trypsinized in 0.05% trypsin, 0.53 mM EDTA for 20 min at 37°C, 2 ϫ 10 6 cells were plated per well, and cells were incubated in a 5% CO 2 incubator. Cells were not further treated except to replace plating medium twice a week.
Western Analysis-For analysis of total APP levels, brains were homogenized in Nonidet P-40 buffer (1% Nonidet P-40, 50 mM Tris, pH 7.5, 10 mM EDTA, and a mixture of protease inhibitors containing 5-10 g/ml leupeptin, 2-4 g/ml pepstatin A, 5-10 g/ml aprotinin, and 1-2 mM phenylmethylsulfonyl fluoride). The homogenates were ultracentrifuged in a Beckman TL-100 rotor at 55,000 rpm (100,000 ϫ g) for 10 min and the supernatants subjected to Western analysis using either human specific anti-5 or C-terminal reactive anti-6 antibodies at 0.4 g/ml (also called anti-BX5 or anti-BX6) (20). A monoclonal version of the anti-5 antibody, 8e5 antibody, was used for some Western analyses (21). For detection of ␤-s-APP, brains were homogenized in 50 mM Tris plus 10 mM EDTA in the mixture of protease inhibitors described above with ( Fig. 6) or without (Figs. 4 and 5) 150 mM NaCl, and homogenates were ultracentrifuged in a Beckman TL-100 rotor at 55,000 rpm (100,000 ϫ g) for 10 min. Supernatants (Sup 1) were reserved for analysis; the pellets were solubilized with the above buffer containing 1% Nonidet P-40, respun as above, and the resulting supernatants (Sup 2) were also analyzed. Equivalent amounts of supernatant protein were analyzed by Western blotting using 2 g/ml of the Swedish192 antibody, which is specific for the Swedish mutant form of ␤-s-APP (22).
ELISA Assays-Brain homogenates were prepared for ELISA as described (21,23). ELISAs specific for either human (12,23) or rodent A␤ (24) were performed as described. A polyclonal rabbit antibody specific for rat and mouse A␤ was generated against the rat A␤ 1-28. Immunogen was prepared as follows. 10 mg of KLH (keyhole limpet hemocyanin, Pierce) was reconstituted and buffer exchanged into 0.3 M phosphate buffer, pH 7.8, and diluted in the same buffer to a concentration of 2 mg/ml. 5 mg of rat A␤ 1-28 peptide was dissolved in 5 ml of deionized water and added to the KLH. 50 mg of disuccinimidyl suberate (Pierce) was dissolved in 1 ml of dimethyl sulfoxide (Sigma). 300 l of the disuccinimidyl suberate stock was added to the KLH/peptide solution and rocked on a nutator for 3 h at room temperature. The conjugate was dialyzed extensively against phosphate-buffered saline. Coupling was determined by a gel band shift by SDS-polyacrylamide gel electrophoresis on a 6% Novex gel. Rabbits were immunized by standard protocol (Josman Laboratories) with the immunogen. Antibody was affinity-purified using Rat 1-28 peptide coupled to actigel resin.

RESULTS AND DISCUSSION
The purpose of this study was to examine the metabolism of human APP in transgenic mice in two major cell types of the adult central nervous system, neurons and astrocytes. Regulatory sequences from the rat NSE gene and from the murine GFAP gene were used to target expression of human APP to neurons or astrocytes, respectively (Fig. 1). Since the APP 670/671 KM to NL "Swedish" mutation has been shown in tissue culture systems to increase the products of ␤-secretase activity, A␤ (7,8) and ␤-s-APP (25), 3 we anticipated that expression of APP containing this mutation would enhance our ability to measure ␤-secretase activity in vivo. Furthermore, inclusion of this mutation allowed us to specifically detect ␤-secretase processing of the transgene-derived human APP. Therefore, the transgenic lines in this study were generated using fusion genes encoding this mutant form of APP. The NSE-APP and GFAP-APP transgenic mice generated allowed comparison of ␤-secretase activities in neurons versus astrocytes.
The expression of APP in the brains of these animals was determined by Western analysis of brain homogenates (Fig. 2). 3  Mutated or wild type human cDNAs encoding full-length APP 751 were placed downstream of regulatory sequences from the rat neuron-specific enolase (NSE) gene or ligated into exon 1 of a modified murine glial fibrillary acidic protein (GFAP) gene (see "Materials and Methods" for details). The diagram depicts the main components of the fusion genes used to generate transgenic mice (not drawn to scale). Construct "a" contains splice sequences both 5Ј and 3Ј to the 14PP cDNA, whereas construct "b" contains only a 5Ј splice sequence.
Western blots were probed with the human specific APP antibody, anti-5 (also called anti-BX5 (20), to measure the amount of transgenic expression ( Fig. 2A), or with an antibody, anti-6, made to the C-terminal region of APP (also called anti-BX6 (20), which reacts with both endogenous mouse and human APP (Fig. 2B). APP-specific bands run at approximately 110 kDa. Similar to results obtained with other NSE-and GFAPdriven transgenes (13,14,37), transgene expression levels varied across different transgenic lines (Table I), but were similar in mice derived from the same transgenic line (data not shown or see below). Brains from several NSE-APP and GFAP-APP lines showed a robust signal of human APP by Western blot analysis (Fig. 2, Table I). However, consistent with results obtained previously (13,26,27), total (human plus murine) APP levels in brains of transgenic mice were only marginally increased over endogenous murine APP levels found in brains of non-transgenic controls (Fig. 2). RNase protection analyses showed that the highest NSE expressing line expressed approximately the same APP mRNA levels as that in the NSE-hAPP751 line described previously (26) or approximately 20% that of endogenous mouse APP (data not shown). The anti-5 antibody occasionally will react with a lower molecular weight protein in brain homogenates that is unrelated to the APP transgene and is also present in the non-transgenic control (Fig. 2).
In general, mice transgenic for NSE-driven constructs that did or did not contain a hybrid 16S/19S SV40 late gene splice 3Ј to the APP coding sequence (Fig. 1) showed similar levels of APP expression (Table I). This is in contrast to results obtained with fusion genes containing the SV40 small t splice in a similar position, as deletion of this splice in the 3Ј position resulted in improved expression levels when a 5Ј splice sequence is present (28 -30). This discrepancy could relate to differences in the SV40 splice sequences used and/or to effects exerted by other elements of the distinct transgenes expressed.
Diverse neural injuries result in reactive astrocytosis, which is characterized by a variety of changes in astroglial morphology and molecular profile (31). Brain injury has also been shown to up-regulate the astroglial expression of APP (32-34), a phenomenon of potential relevance to AD, which is accompanied by a prominent reactive astrocytosis (35). Notably, because of the injury responsiveness of the GFAP gene, gliosisprovoking maneuvers such as focal penetrating brain lesions effectively up-modulate the astroglial expression of the GFAPdriven transgenes (36,37). Therefore transgene expression was also tested 2 days after the placement of focal cerebral penetrating lesions. As expected, lesioning resulted in a substantial increase in the cerebral expression of human APP in mice transgenic for GFAP-APP (Table I, Fig. 5).
While the copious amounts of A␤ in brain homogenates of Alzheimer's diseased patients can be readily measured (38), the detection of A␤ present in normal brain tissue is technically difficult. Because it has also been difficult to measure A␤ in brain homogenates of transgenic mice with these moderate levels of human APP expression, an in vitro approach was first used to obtain evidence for the production of human A␤ peptide by neurons of NSE-APP transgenic mice. Primary cortical cultures were generated from heterozygous fetal transgenic mice from line Hillary 14 (Table I)  As demonstrated previously (13) the anti-5 antibody also reacts with a lower molecular weight protein, which is present at similar levels in brain homogenates from transgenic and non-transgenic mice (A). The C-terminal anti-6 antibody shows more bands than the human-specific anti-5 antibody, since it cross-reacts with all mouse isoforms. Lines Byers and Tilly do not contain the Swedish mutation and are included to show low (Byers 46) or medium (Tilly 169) expression levels.

TABLE I APP expression in various transgenic lines
Relative expression levels of human APP in brains of heterozygous transgenic mice from different lines were assessed semiquantitatively by Western blot analysis and are indicated on the right. In cases where the transgene expression level was modulated by lesioning, the expression level obtained after lesioning is indicated by an asterisk. "ϩ splice" refers to the presence of the 3Ј splice sequences in addition to the 5Ј splice sequences present in all NSE vectors. "Ϫ" indicates the absence of detectable levels of APP expression. human A␤ in transgenic cultures were higher than the levels of endogenous mouse A␤ (Fig. 3). This provides in vitro evidence that neurons of NSE-APP transgenic mice can produce abundant amounts of human A␤. In order to assess ␤-secretase activity in vivo, brain homogenates were assayed by Western analysis for the presence of ␤-s-APP. This approach was chosen for the following reasons. ␤-s-APP is a direct product of ␤-secretase activity, and its production parallels the production of A␤ in vitro under conditions that are expected to either directly modulate the activity of ␤-secretase or to modulate the accessibility of APP to ␤-secretase. The incorporation of the Swedish mutation into APP (7, 8) 3 enhances both ␤-s-APP and A␤ production. Treatment of 293 cells with bafilomycin reduces both ␤-s-APP and A␤ production without any effect on ␣-secretase products (22). Taken together, these results indicate that levels of ␤-s-APP serve as a good marker for ␤-secretase activity. Previously, we described an antibody specific for ␤-s-APP that does not cross-react with ␣-s-APP or uncleaved APP (4). Because this antibody does not recognize the Swedish mutant form of ␤-s-APP, we have likewise generated a similar antibody specific for the Swedish mutant form of ␤-s-APP, Swedish192 (22). This antibody does not cross-react with the wild type version of ␤-s-APP, ␣-s-APP, nor with full-length APP produced by cells in culture (22). 3 For the present study, we determined that Swedish192 also does not cross-react with these molecules in brain homogenates of transgenic animals (data not shown). Brain homogenates, solubilized in either low salt or 2% Nonidet P-40, from mice of different transgenic lines were compared for ␤-s-APP content by Western blot analysis using the Swedish192 antibody (Fig.  4). ␤-s-APP was clearly detectable in 6/6 different NSE-APP expressor lines tested (Hillary lines 14 and 103, Chelsea lines 32, 58, 62, and 68). ␤-s-APP levels were roughly proportional to the levels of total APP produced in these lines (data not shown).
There is still a debate concerning which cell type is responsible for the generation of A␤ in brain (39). In vitro data indicate greater A␤ production in astroglial cultures than in neuronal cultures (3). We wished to address this question in an in vivo system.
Our demonstration of the abundant presence of ␤-s-APP in NSE-APP transgenic animal brains demonstrates that neurons are capable of vigorous ␤-secretase activity and are a potential source of A␤ in the brain. Astrocytes are also an abundant cell type in brain, and we next asked if they too were capable of generating ␤-secretase products of APP in the GFAP-APP transgenic animals. Notably, GFAP-APP mice produce substantially less ␤-sAPP than NSE-APP mice (Fig. 5). While basal levels of human APP expression in the GFAP APP line FIG. 5. Low levels of ␤-secretase processing of human APP in GFAP-APP transgenic animals. Western analysis was done on high speed supernatants of brain homogenates (solubilized in 1% Nonidet P-40 for the human-specific APP antibody anti-5 or in low salt buffer for ␤-s-APP-specific Swedish192 antibody as described under "Materials and Methods") from transgenic (ϩ) mice of NSE-APP line Hillary 14 and GFAP-APP line 835-18 and from non-transgenic controls (Ϫ). Brains were obtained from either lesioned animals (ϩ) or unmanipulated mice (Ϫ). Western blots were immunostained with either the human APP-specific antibody anti-5 (left panel) or with the Swedish mutant ␤-s-APP-specific antibody, Swedish 192 (right panel). The arrow indicates the location of the APP-specific bands. Molecular mass markers indicated are in kilodaltons. The weak ␤-s-APP-specific band in GFAP-APP transgenic samples is seen just above the two nonspecific cross-reacting bands that are also present in non-transgenic control samples. 835-18 were slightly lower than those of the NSE-APP line Hillary 14, injury-induced astroglial activation resulted in an up-modulation of human APP expression in the GFAP-driven line to levels in excess of those seen in the NSE-APP transgenic mice (Fig. 5). Strikingly, compared with the NSE-APP line Hillary 14, relatively little ␤-s-APP was produced in GFAP-APP line 835-18 in either lesioned or unlesioned brains. Western analyses on Nonidet P-40 soluble extracts (Sup 2) with Swedish192 antibody again showed robust amounts of ␤-s-APP in NSE-APP animals and very little ␤-s-APP in GFAP-APP mice (data not shown), indicating that ␤-s-APP inside cells is likewise reduced in the GFAP-APP transgenic animals. The same results were obtained when NSE-APP line Chelsea 58 and GFAP-APP line 835-16 were compared and when NSE-APP line Chelsea 58 was crossed onto the same background mouse strain as the GFAP-APP lines (Fig. 6). While lesioning of either GFAP-APP or the NSE-APP transgenic mice had no effect on the ratio of ␤-s-APP to total APP, this ratio was substantially lower in GFAP-APP than in NSE-APP transgenic mice ( Fig. 5 and data not shown). Although it is difficult to measure A␤ in brain homogenates of transgenic mice with these moderate levels of human APP expression, we were able to measure detectable levels (5-fold over background fluorescence) of A␤ in brain homogenates from NSE-APP but not GFAP-APP mice (Fig. 6). This supports our Western data and indicates that in vivo, ␤-S-APP measurements accurately reflect ␤-secretase activity and A␤ production.
In conclusion, the demonstration of abundant ␤-s-APP generation and the production of A␤ in the central nervous system of NSE-APP transgenic mice indicates that neurons are capable of vigorous ␤-secretase activity and are a potentially important source of A␤ in the brain. While astrocytes are also an abundant cell type in the brain and respond to neural injuries with increased APP expression, based on the relatively low ␤-s-APP and undetectable A␤ production in lesioned and unlesioned GFAP-APP transgenic mice, ␤-secretase cleavage and A␤ production appears to be substantially less effective in astrocytes than in neurons. As illustrated here, the established APP transgenic mice can serve as convenient models to assess ␤-secretase activity in different central nervous system cell types in vivo. These mice should also prove valuable for the preclinical testing of ␤-secretase inhibitors, compounds that may be useful in the treatment of AD.