The Familial Dementia BRI2 Gene Binds the Alzheimer Gene Amyloid-β Precursor Protein and Inhibits Amyloid-β Production*

Alzheimer disease (AD), the most common senile dementia, is characterized by amyloid plaques, vascular amyloid, neurofibrillary tangles, and progressive neurodegeneration. Amyloid is mainly composed by amyloid-β (Aβ) peptides, which are derive from processing of the β-amyloid precursor protein (APP), better named amyloid-β precursor protein (AβPP), by secretases. The AβPP intracellular domain (AID), which is released together with Aβ, has signaling function, since it modulates apoptosis and transcription. Despite its biological and pathological importance, the mechanisms regulating AβPP processing are poorly understood. As cleavage of other γ-secretase substrates is regulated by membrane bound proteins, we have postulated the existence of integral membrane proteins that bind AβPP and regulate its processing. Here, we show that BRI2, a type II membrane protein, interacts with AβPP. Interestingly, 17 amino acids corresponding to the NH2-terminal portion of Aβ are necessary for this interaction. Moreover, BRI2 expression regulates AβPP processing resulting in reduced Aβ and AID levels. Altogether, these findings characterize the BRI2-AβPP interaction as a regulatory mechanism of AβPP processing that inhibits Aβ production. Notably, BRI2 mutations cause familial British (FBD) and Danish dementias (FDD) that are clinically and pathologically similar to AD. Finding that BRI2 pathogenic mutations alter the regulatory function of BRI2 on AβPP processing would define dysregulation of AβPP cleavage as a pathogenic mechanism common to AD, FDD, and FBD.

extracellularly (sAPP␤) or into the lumen of intracellular compartments, the COOH-terminal fragment of 99 amino acids (C99) remains membrane bound. In a second, intramembranous proteolytic event, C99 is cleaved, with somewhat lax site specificity, by the ␥-secretase. Two peptides are released in a 1:1 stoichiometric ratio: the amyloidogenic A␤ peptide, consisting of 2 major species of 40 and 42 amino acids (A␤40 and A␤42, respectively), and an intracellular product named AID or AICD, which is very short-lived and has been identified only recently (7)(8)(9). In an alternative, nonamyloidogenic proteolytic pathway, A␤PP is first processed by ␣-secretase in the A␤ sequence leading to the production of the sAPP␣ ectodomain and the membrane-bound COOH-terminal fragment of 83 amino acids (C83). C83 is also cleaved by the ␥-secretase into the P3 and AID peptides. While A␤ is implicated in the pathogenesis of Alzheimer disease, AID mediates most of the A␤PP signaling functions. A pathogenic role for A␤PP processing in AD has been ascertained by the finding that mutations in presenilins (10 -13), key components of the ␥-secretase, and A␤PP (14) cause autosomal dominant familial forms of AD. Thus, because of its biological and pathological importance, understanding how A␤PP cleavage is regulated is of primary significance.
Membrane-bound proteins prompt Notch cleavage by secretases and the release of a transcriptionally active intracellular fragment (15). Considering the remarkable similitude between A␤PP and Notch signaling, we have hypothesized that A␤PP processing is similarly regulated. We report herein that the type II membrane protein BRI2 (16), also known as E25 (17) and Itm2 (18), fulfills this description.

EXPERIMENTAL PROCEDURES
Split-ubiquitin Yeast Two-hybrid Screening-The split-ubiquitin system provides an attractive alternative to analyze interactions between integral membrane proteins (19). The split-ubiquitin system and human brain libraries were purchased from Dualsystems Biotech (Zurich, Switzerland). The screenings were performed according to the manufacturers protocol. Briefly, human A␤PP (amino acids 1-695), human A␤PP (amino acids 1-664; A␤PPNcas), or human APLP2 were cloned into pTMV4, pAMBV4, and pAMBV4 bait vectors, respectively, to obtain APP family bait proteins fused to the COOH-terminal half of ubiquitin (Cub), followed by a reporter fragment (LexA, a DNA-binding protein, fused to VP16, a transcriptional activation). Human brain libraries express proteins fused at the NH 2 -terminal half of mutated ubiquitin (NubG). For each library we screened ϳ5 ϫ 10 6 transformants. Clones coding for proteins that can interact with A␤PP/APLP2-Cub will promote the NubG-Cub interaction followed by recruitment of ubiquitin-specific protease(s), cleavage of the A␤PP/APLP2-Cub bait, release of the LexA-VP16 transcription factor, and the transcriptional activation of the two reporter genes (lacZ and HIS3). Library plasmids were recovered from HIS3-and lacZ-positive yeast transformants, identified by nucleotide sequencing, and cloned into pcDNA3.1 with an NH 2 -terminal FLAG tag and directly tested its ability to interact with A␤PP by immunoprecipitation as described below. Screening for co-activator of both reporter genes resulted in the identification of known A␤PP/APLP2-binding proteins, such as Fe65 (20).
Immunoprecipitation and Western Blot-Unless otherwise noted, all immunoprecipitation procedures were performed at 4°C. The transfected cells were lysed in Buffer A (20 mM Hepes/NaOH, pH 7.4, 1 mM EDTA, 1 mM dithiothreitol, 150 mM NaCl, 0.5% (w/v) Triton X-100) containing 10% (v/v) glycerol for 30 min, and the lysates were cleared at 20,000 ϫ g for 10 min. For FLAG immunoprecipitation, the cleared lysates were mixed with 20 l of FLAG-M2 beads (Sigma) for 2 h and washed three times with Buffer A. The precipitants were boiled in 60 l of 2 ϫ SDS sample buffer and subjected to Western blot. For other immunoprecipitation, the cleared lysates were incubated with antibodies for 1 h and mixed with 20 l of protein A/G beads (Pierce), washed in Buffer A, and processed as above. Human brains from normal individual were homogenized in Buffer A containing 10% (v/v) glycerol using a Dounce homogenizer. The proteins were extracted overnight with the protein concentration at 5 mg/ml. Extracted proteins were cleared at 20,000 ϫ g for 1 h. The supernatants were incubated with the indicated antibodies and protein A/G beads blocked with PBS containing 1%(w/v) bovine serum albumin. Precipitants were washed and processed as described above.
Metabolic Labeling-HEK293APP cells transfected with pcDNA3 or BRI2 were incubated in DMEM without methionine and cysteine (Invitrogen) supplemented with penicillin, streptomycin, and 10% fetal bovine serum, for 2 h. After, cells were labeled 30 min by adding to the culture media 35 S-labeled methionine and cysteine (ICN). The labeled cells were washed extensively and chased in DMEM supplemented with penicillin, streptomycin, and 10% fetal bovine serum for the indicated periods of time. After the chase, cells were lysed and immunoprecipitated with ␣A␤PPct as described above. The media of labeled cells were cleared at 20,000 ϫ g for 10 min and immunoprecipitated with the indicated antibodies.
Luciferase Assays-The assays were performed as described (25), except that the A␤PP-Gal4 fusion (26) was used as a Gal4 source. Luciferase activity was normalized by the activity of ␤-galactosidase co-transfected to monitor the transfection efficiency.
Enzyme-linked Immunosorbent Assay (ELISA)-HEK293APP cells were transiently transfected with pcDNA3 or BRI2. 24 h after the transfection, the cells were conditioned for 24 h, and A␤40 and A␤42 in the media were measured using human A␤ ELISA kits (KMI Diagnostics), according to the manufacturer's protocol. The transfected cells were lysed and cleared as above, and the amount of extracted protein was used to normalize the amount of A␤ detected by ELISA.
Protein Determination-Protein concentrations were determined by Bradford protein assay (Bio-Rad) and bovine serum albumin as a standard.

RESULTS AND DISCUSSION
To test whether membrane-tethered proteins might regulate A␤PP processing, we have used the split-ubiquitin system to identify interactions between membrane proteins. Screening of a human brain cDNA library for proteins that interact with A␤PP family proteins resulted in the identification of BRI2 (17) and BRI3 (27), members of a gene family of Type II membrane proteins containing a Brichos domain (28). Although the function of BRI proteins is unknown, BRI2 mutations are found in patients with FBD (16) and FDD (29). Of note, neuro-pathological findings in FBD and FDD include amyloid and/or preamyloid parenchyma plaques, congophilic amyloid angiopathy, neurofibrillary tangles, and neurodegeneration, similar to AD. Hence, because mutations in BRI2 cause AD-like familial dementia we have further studied the physiological relevance of this BRI2-A␤PP interaction.
To assess the BRI2-A␤PP interaction in mammalian cells, HeLa cells were co-transfected with BRI2 and A␤PP constructs (Fig. 1a). Immunoprecipitation of cell lysates with an ␣FLAG antibody showed that BRI2 interacted with full-length A␤PP (Fig. 1, b and d), C99 (Fig. 1, b and d), and A␤PPNcas, which present a deletion of most of the intracellular region of A␤PP ( Fig. 1c) but not C83 (Fig. 1, b and d). A␤PP runs as a doublet. The lower A␤PP band represents nonglycosylated, immature A␤PP; the upper form is instead composed of mature, glycosylated A␤PP. Of note, only the mature, glycosylated forms of A␤PP and A␤PPNcas interacted with BRI2 ( Fig. 1, b-d). It should also be noted that BRI2 overexpression dramatically increases the levels of C99 (Fig. 1, b and d). The significance of this finding will be discussed below.
Deletion of most of the BRI2 ecto-domain did not abolish the binding to APP (BRI2 1-131 , Fig. 1d). The reverse immunoprecipitation with an ␣A␤PP antibody revealed that A␤PP immunoprecipitates BRI2 (Fig. 1e). Additionally, a proteolytic ϳ17-kDa BRI2 NH 2 -terminal fragment detected in transfected HeLa cells (BRI2nt, which is similar in size to BRI2 1-131 ) was FIG. 1. BRI2 is a ligand of A␤PP. a, schematic representation of the BRI2 and A␤PP constructs used. The constructs are numbered 1 (amino acids 1-266, full length) and 2 (amino acids 1-131). b-d, Western blots (WB) of anti-FLAG immunoprecipitates (IP ␣FLAG) and total lysates (TL) from HeLa cells expressing the indicated proteins show the specificity of BRI2/A␤PP association and map the interaction sites. pc indicates the empty vector (pcDNA3.1); the numbers 1 and 2 indicate the BRI2 constructs shown in a; * indicates nonreduced anti-FLAG antibody; m, denotes mature, glycosylated forms of A␤PP, while i indicates the immature, nonglycosylated A␤PP. For Western blots, ␣A␤PP represents the monoclonal antibody 22C11, while ␣A␤PPct is a rabbit polyclonal raised against the COOH terminus of A␤PP. e, the lysates of Hela cells transfected with both A␤PP and BRI2 were precipitated with either a rabbit polyclonal control (RP) or ␣A␤PPct. Immunoprecipitants and total lysates were blotted with either the ␣A␤PP monoclonal antibody 22C11 or ␣FLAG. BRI2, as well as a ϳ17-kDa BRI2 NH 2 -terminal fragment (BRI2nt), were precipitated by ␣A␤PPct together with A␤PP. also precipitated with A␤PP (Fig. 1e). The specificity of these interactions was supported by the evidence that BRI2 did not bind to ApoER2, another type I integral membrane protein (Fig. 1c). These findings attest that while the intracytoplasmic tail of A␤PP and most of the A␤PP and BRI2 ectodomain are not important for BRI2/A␤PP interaction, a 17-amino acid region in the ectodomain of A␤PP, juxtaposed to the transmembrane region and containing the NH 2 -terminal A␤ sequence, is essential for this binding. These data strongly suggest that BRI2 and A␤PP do not interact in trans (i.e. as receptor/ligand expressed on distinct membranes) but, rather, form a molecular complex in cellular membranes.
A␤PP and BRI2 are both expressed in mature neural tissues. We therefore sought to determine whether A␤PP also interacts with BRI2 in the adult human brain. First, we tested four anti-BRI2 antibodies to determine whether they could immunoprecipitate human BRI2. For these tests, HeLa cells were transfected with FLAG-BRI2 and immunoprecipitated with the four BRI2 antibodies and controls. As shown in Fig. 2a, only the EN3 anti-BRI2 antibody was able to precipitate BRI2. Next, we made homogenates of human brains and performed immunoprecipitation with either the ␣A␤PPct antibody or EN3. As shown in Fig. 2b, C99 (and larger COOH-terminal A␤PP fragments) was precipitated with both anti-A␤PP as well as EN3 antibodies, while C99 was not precipitated with a rabbit polyclonal IgG. Of interest, also in this case C83 did not precipitate with BRI2, although it was precipitated by ␣A␤PPct. Moreover, full-length A␤PP was also precipitated by EN3, albeit at low levels. Altogether, these experiments indicate that endogenous A␤PP and BRI2 associate in the adult human brain. In addition, they show that BRI2 preferentially interacts with C99 and larger A␤PP COOH-terminal fragments but not with C83.
As shown in Fig. 1, b and d, expression of BRI2 constructs invariantly results in increased levels of C99. It is likely that this dramatic increase in C99 levels is dependent on an effect of BRI2 on APP processing. To directly test for this, we expressed FLAG-tagged BRI2 in HeLa, HEK293, and N2a cells together with A␤PP-Gal4, a luciferase reporter under the control of a Gal4 promoter and ␤-galactosidase. A␤PP-Gal4 is a fusion of the yeast transcription factor Gal4 to the cytoplasmic domain of A␤PP. ␥-Cleavage of A␤PP-Gal4 will release AID-Gal4 from the membrane to the nucleus with consequent activation of luciferase transcription (26). As shown in Fig. 3a, BRI2 reduces luciferase activity in all three cell lines, suggesting an inhibition of AID formation. Instead, transfection of mouse ␤-secretase (BACE) resulted in increased AID release, as expected (Fig. 3b). The BRI2 1-131 mutant, which still interacts with A␤PP and produces increased C99 levels (Fig. 1d), also inhibits AID release (Fig. 3c). Last, mixing experiments show that, for BRI2 to repress AID-Gal4 release, it must be co-expressed with A␤PP-Gal4 in the same cell. In fact, mixing cells expressing BRI2 with cells expressing A␤PP-Gal4 does not affect release of AID (Fig. 3d). This further suggests that BRI2 and A␤PP interact in cis rather than in trans.
To further validate this system, we have measured A␤ in the conditioned media of HEK 293 transfected with BRI2 and found that BRI2 significantly diminished A␤40 and A␤42 levels (Fig. 3b). Again, BACE transfection increased A␤40 and A␤42 secretion (Fig. 3f).
Inhibition of AID and A␤ production by BRI2 suggests that BRI2 expression reduces cleavage of A␤PP by the ␥-secretase. However, it is also possible that BRI2 could modulate the ␤and ␣-cleavage of A␤PP. As discussed above, cleavage of A␤PP by either ␤or ␣-secretase releases sAPP␤ and sAPP␣ in the supernatant, respectively. While increased amounts of either sAPP␣ or sAPP␤ indicate increased ␣or ␤-cleavage, reduction of either sAPP␣ or sAPP␤ reflect decreased ␣or ␤-cleavage. Thus, to determine whether BRI2 affects either ␤or ␣-secretase, we measured the amounts of sAPP␣ and sAPP␤. In these same experiments, we also measured intracellular levels of C99 and C83. HEK293-APP cells were transfected with FLAG-BRI2 or a vector control. Transfected cells were pulse-labeled with [ 35 S]methionine-cysteine for 30 min, then chased for 0, 1, 2, and 4 h at 37°C (Fig. 3c). The cell lysates were immunoprecipitated with ␣A␤PP-ct antibody at each time point (Fig. 3c). To measure secreted A␤PP (sAPP␣ and sAPP␤), supernatants were collected from cells labeled for 4 h and precipitated with the anti-A␤PP antibodies P21 (which precipitates both sAPP␣ and sAPP␤) or 6E10 (which only precipitates sAPP␣) (Fig. 3d). BRI2 transfection resulted in decreased amounts of C83 (Fig.  3c) and sAPP␣ (Fig. 3d). Conversely, the levels of C99 (Fig. 3c) and sAPP␤ (Fig. 3d) were augmented. Notably, BRI2 was coimmunoprecipitated by the ␣A␤PP-ct antibody at all time points. Thus, BRI2 expression reduces cleavage of A␤PP by ␣-secretase while increasing its processing by ␤-secretase. The concomitant inhibition of ␥-secretase and increase in ␤-cleavage of A␤PP explains the dramatic increase in C99 levels.
A␤PP is a member of a family of proteins that includes APLP1 and APLP2. APLP1 and APLP2 are also ␥-secretase substrates (22) and among the numerous ␥-secretase substrates are those that bear more sequence similarity to A␤PP. Thus, to test whether BRI2 generally affects ␥-secretase or specifically inhibits ␥-cleavage of A␤PP, we transfected BRI2 with either APLP1 and APLP2. Western blot using anti-APLP1 or anti-APLP2 COOH-terminal antibodies indicates that BRI2 expression does not promote accumulation of COOH-terminal fragments of APLP1 (data not shown) and APLP2 (Fig. 3i). This result supports the notion that BRI2 specifically blocks the ␥-activity on A␤PP but not on other ␥-substrates.
Altogether, our studies suggest that BRI2 and A␤PP form a multimolecular complex in cell membranes. While the stoichiometry of A␤PP and BRI2 in such complexes has to be investigated and whether BRI2 and A␤PP are found in a structure comprising other proteins is unknown, our data suggest that BRI2 functions as an endogenous regulator of A␤PP processing. More specifically, we found that BRI2 expression decreases both ␣and ␥-cleavage of A␤PP while increasing its ␤-processing. Although the detailed molecular mechanisms responsible for these functions must be directly addressed, the finding that BRI2 interacts with a region of A␤PP comprising the ␣and ␥-cleavage sites insinuates that BRI2 physically masks the two target sequences from the secretases. Increased ␤-activity might be consequential to the reduced ␣-processing.
Recently, mutations in BRI2 have been found in FBD (16) and FDD (29) patients. Both wild type and mutant BRI2 are processed by furin (30) with this processing resulting in the FIG. 3. BRI2 regulates A␤PP processing by secretases. a-c, BRI2 reduces the A␤PP-Gal4-driven luciferase activity in HeLa, N2a, and HEK293 cells. Cells were co-transfected with A␤PP-Gal4 together with pcDNA3.1 (pc) or FLAG-BRI2, BRI2 1-131 , or BACE. Data are expressed as percentage of the luciferase activity measured in cells transfected with the empty vector. BRI2, BACE, and pcDNA3.1-transfected cells express similar levels of A␤PP-Gal4 (data not shown). Error bars represent S.D. for three independent experiments. d, cells were transfected with either A␤PP-Gal4, pcDNA3.1 (pc), or BRI2. Cells transfected with A␤PP-Gal4 were than mixed at the indicated ratio with either BRI2 or pcDNA3.1 transfected cells. Samples were analyzed for luciferase activity as described above 24 h after transfection and mixing. e and f, BRI2 inhibits production of A␤40/A␤42. HEK293 cells stably expressing A␤PP (HEK293A␤PP) were transfected with an empty vector (pcDNA3.1 (pc)), BACE, or FLAG-BRI2, and A␤40 and A␤42 secreted in the media were measured by ELISA. The A␤ amount was normalized by the protein content of the lysates of the transfected cells. Error bars represent S.D. for three independent experiments. g, pulse-chase experiment, representative of four independent experiments, of transfected HeLa cells. HEK293A␤PP cells were transfected with an empty vector (Vector) or FLAG-BRI2. The lysates of metabolically labeled cells were precipitated with ␣A␤PPct. The numbers above each lane indicates the hours the cells were chased (c). BRI2 expression decreases C83 production while dramatically increasing the generation of C99. h, the conditioned media of similarly transfected HEK293A␤PP cells were harvested after 4 h labeling and were precipitated with either p21 or 6E10. While the amounts of sAPP␣ were decreased by BRI2, the total sAPP (sAPP␣ϩsAPP␤) did not show a significant change indicating an augmentation of sAPP␤ production and a shift of A␤PP processing from ␣to ␤-secretase. Vec., vector. i, cells were transfected with APLP2 together with either pcDNA3.1 or BRI2. 24 h after transfection, cells were analyzed by Western blot for BRI2 and APLP2 peptides. Vec., vector. secretion of a COOH-terminal peptide. Furin cleavage of wild type BRI2 releases a 23-aa-long peptide. In FBD patients, a point mutation at the stop codon of BRI2 results in a readtrough the 3Ј-untranslated region and the synthesis of a BRI2 molecule containing 11 extra amino acids at the COOH terminus. Furin cleavage of this mutated BRI2 generates a longer peptide, the ABri peptide, which is deposited as amyloid fibrils. In the Danish kindred, the presence of a 10-nucleotide duplication one codon before the normal stop codon produces a frameshift in the BRI2 sequence generating a larger-thannormal precursor protein, of which the amyloid subunit, also released by furin processing, comprises the last 34 COOHterminal amino acids. The deposition of ABri and ADan amyloids is considered the pathogenic cause of these chromosome 13 dementias, reinforcing the dogma that distinct amyloidogenic peptides can begin similar neurodegenerative processes. According to this view, amyloidogenesis, and not the particular amyloidogenic peptide, is key in initiating neurodegeneration. However, the finding that BRI2 regulates A␤PP processing is intriguing and prompts us to speculate that altered A␤PP processing is also a pathogenic factor in FBD and FDD. Intriguingly, in FDD patients A␤ co-deposits with ADan in vascular amyloid lesions (31). Further studies will address whether this hypothesis is correct.