Brain Endothelial Cells Produce Amyloid β from Amyloid Precursor Protein 770 and Preferentially Secrete the O-Glycosylated Form*

Deposition of amyloid β (Aβ) in the brain is closely associated with Alzheimer disease (AD). Aβ is generated from amyloid precursor protein (APP) by the actions of β- and γ-secretases. In addition to Aβ deposition in the brain parenchyma, deposition of Aβ in cerebral vessel walls, termed cerebral amyloid angiopathy, is observed in more than 80% of AD individuals. The mechanism for how Aβ accumulates in blood vessels remains largely unknown. In the present study, we show that brain endothelial cells expressed APP770, a differently spliced APP mRNA isoform from neuronal APP695, and produced Aβ40 and Aβ42. Furthermore, we found that the endothelial APP770 had sialylated core 1 type O-glycans. Interestingly, Ο-glycosylated APP770 was preferentially processed by both α- and β-cleavage and secreted into the media, suggesting that O-glycosylation and APP processing involved related pathways. By immunostaining human brain sections with an anti-APP770 antibody, we found that APP770 was expressed in vascular endothelial cells. Because we were able to detect O-glycosylated sAPP770β in human cerebrospinal fluid, this unique soluble APP770β has the potential to serve as a marker for cortical dementias such as AD and vascular dementia.

Alzheimer disease (AD) 3 is characterized by intracellular accumulation of neurofibrillary tangles and extracellular deposits of amyloid ␤ (A␤) peptides in the brain (1,2). A␤ is generated from amyloid precursor protein (APP) by the se-quential actions of ␤-secretase, BACE1 (␤-site APP-cleaving enzyme 1) (3)(4)(5)(6), and ␥-secretase (7). Because most earlyonset familial AD patients have gene mutations that influence the processing or aggregation of A␤, and the neurites associated with A␤ plaques are often damaged (2), the process of A␤ deposition in the brain seems to be strongly associated with AD pathogenesis. Even though APP has been proposed to have a receptor-like function and binds to multiple extracellular matrix proteins such as heparin and collagen (8,9), understanding the biological functions of APP remains an important scientific and intellectual challenge (10). Two paralogs of APP are known in mammals and are designated APPlike proteins 1 and 2. Interestingly, triple knock-out mice lacking all three APP family members die shortly after birth (11), suggesting redundant functions of the APP family proteins.
APP has three alternatively spliced isoforms: APP695, APP751, and APP770 (12,13). Compared with APP695, APP751 contains an additional Kunitz-type protease inhibitor (KPI) domain, whereas APP770 contains a KPI domain plus an OX2 domain. APP695 is most abundantly expressed in neurons, whereas APP751 and APP770 show more ubiquitous expression patterns (14). The secreted form of APP containing a KPI domain, also known as the protease nexin 2, potentially inhibits certain serine proteases, most notably several prothrombotic enzymes (15); very limited information is available concerning the functions of the OX2 domain. In the present study, we show that brain microvascular endothelial cells (BMECs) express significant levels of APP770, from which A␤40 and A␤42 are produced, and that the endothelial APP770 has multiple ⌷-glycosylation chains, which potentially play important roles in APP processing.

EXPERIMENTAL PROCEDURES
Materials-The sources of the materials used in this study were as follows: tissue culture media and reagents including DMEM from Invitrogen; recombinant peptide N-glycosidase (PNGase) from New England BioLabs; O-glycosidase from Roche Applied Science; Arthrobacter ureafaciens sialidase from Nacalai Tesque; protein A-Sepharose Fast Flow from GE Healthcare; protein molecular weight standards from Bio-Rad; peanut agglutinin (PNA; biotinylated Arachis hypogaea) lectin and a series of lectin-conjugated agaroses from Seikagaku Corp.; A␤40 from Peptide Institute; BCA protein assay reagents and sulfo-NHS-LC-biotin from Thermo Fisher Scientific; all other chemicals were from Sigma or Wako Chemicals. An anti-APP (C15) rabbit polyclonal antibody against endogenous membrane-bound (intact) APP was a generous gift from Dr. Kei Maruyama (Saitama Medical University, Saitama, Japan). The commercially available antibodies used were as follows: mouse monoclonal anti-APP 22C11 (Chemicon); anti-human sAPP␣ (6E10; Signet Laboratories); anti-GAPDH (Chemicon); anti-KPI (Chemicon); anti-OX2 (Chemicon); anti-human sAPP␤ (IBL Co.); goat polyclonal anti-platelet endothelial cell adhesion molecule (Santa Cruz Biotechnology). The clinical study was approved by the Ethical Committee of RIKEN and Fukushima Medical University (No. 613). All animal experiments were performed in compliance with the institutional guidelines for animal experiments of RIKEN.
Expression Plasmids and Cell Culture-Human APP770 FLAG-pcDNA was constructed by inserting a human APP770 sequence generated by PCR (primers 1 and 2) at the SalI and HindIII sites of pcDNA, and a FLAG domain fragment was generated by annealing primers 3 and 4 at the HindIII and XbaI sites of pcDNA. A ViraPower Adenoviral Expression system (Invitrogen) was used to produce recombinant adenoviruses carrying human APP-FLAG, according to the manufacturer's protocol. A series of APP770-pcDNA3.1 mutants was generated using a QuikChange site-directed mutagenesis kit (Stratagene). To generate the APP OX2All mutant, in which all of the possible O-glycosylation sites in the OX2 domain were deleted, APP S346A,S348A was constructed first, followed by creation of the APP OX2All mutant using APP S346A,S348A as a template and the primers APP SOX2A forward and reverse. To generate the APP all mutant, in which all the reported O-glycosylation sites (16) in the whole APP770 region were deleted, we used the APP OX2all mutant as a template and the primers APP T366A,T367A and APP T651A to change Thr 366 , Thr 367 , and Thr 651 to Ala. The primer information is shown in Table 1, A and B. Human BMECs (Applied Cell Biology Research Institute) were cultured in CS-C complete medium with or without FBS and used within four passages. Human umbilical vein endothelial cells (TaKaRa) were cultured in EMB-2 (TaKaRa) containing 2% FBS and EGM-2 TM SingleQuots (TaKaRa) and used within four passages. Primary liver sinusoidal endothelial cells were prepared from mouse livers using CD146 MicroBeads (Miltenyi Biotech) according to the manufacturer's protocol (17). COS-7 cells were cultured in DMEM containing 10% FBS. Cortical neurons were isolated from C57BL/ 6CrSlc mice as described previously (18).
Patients and Samples-A patient (60 years of age) who died from a nonketotic hyperosmolar coma was recruited for this study. Cerebrospinal fluid (CSF) samples were collected from patients with AD.
PCR Analysis of APP Transcripts-Total RNA was isolated from cells using the Sepasol reagent (Nacalai Tesque, Inc.), and 5 g of the isolated RNA was reverse-transcribed with random hexamers using a SuperScript III RT kit (Invitrogen) according to the manufacturer's protocol. The obtained cDNA samples were subjected to PCR analyses with primers A and D for APP695, APP751, and APP770, primers B and D for APP751 and APP770, and primers C and D for APP770. PCR was performed for 28 cycles (95°C for 40 s, 56°C for 40 s, and 72°C for 90 s). The primer information is shown in Table 1C.
Immunohistochemistry-Brain tissue was fixed in phosphate-buffered 15% formalin solution and embedded in paraffin. A pair of serial sections (5-m thick) was stained with hematoxylin and eosin or incubated with an anti-OX2 antibody (1:100) overnight at 37°C followed by biotinylated antirabbit IgG (1:200). Bound antigens were visualized with the avidin and biotin-peroxidase complex method (ABC kit, Vector Laboratories).
A␤ Quantification-BMECs were infected with adenovirus preparations for APP770-FLAG overexpression and cultured in Opti-MEM for 8 h. The levels of A␤40 and A␤42 in the media were determined using a human amyloid ␤(1-40) or (1-42) assay kit (IBL Co.) for BMECs and a human/rat ␤-amyloid(40) or (42) ELISA Kit (Wako Chemicals) for mouse neurons.
Glycosidase Digestion-Cell lysates, APP precipitates from cell lysates, culture media, and heparin-precipitated samples from culture media were incubated in the presence or absence of Arthrobacter ureafaciens sialidase (4 milliunits) and/or O-glycosidase (2 milliunits) for 18 h.
Cell Surface Biotinylation-BMECs grown in 10-cm culture plates were labeled with Sulfo-NHS-LC-Biotin for 30 min at 4°C. After washing the plates three times with 0.1 M glycine in PBS (pH 8.0) and once with PBS, cell lysates were prepared. NHS-SS-biotin was used for internalization assays. After two washes with ice-cold PBS, the cells were cultured in prewarmed CS-C media for 0 or 1 h at 37°C. After washing once with ice-cold CS-C media and twice with PBS/10% FBS, the remaining cell surface biotin was removed by incubating the cells in buffer containing the nonpermeant reducing agent GSH (50 mM GSH, 75 mM NaCl, 0.3% NaOH, and 10% FBS) twice for 20 min each at 4°C (19). After the reaction was quenched by 5 mg/ml of iodoacetamide in PBS containing 1% BSA, cell lysates were prepared. Biotinylated cell surface proteins were pulled down from the lysates with streptavidin-Sepharose (GE Healthcare) for further analysis.
Benzyl N-Acetyl Galactosamine (GalNAc) Treatment-Subconfluent BMECs grown on 10-cm tissue culture plates were incubated in the presence of benzyl GalNAc (2 mM) for 18 h, before cell lysates and culture media samples were prepared for further analysis.

RESULTS
Brain Endothelial Cells Express APP770-The human brain expresses three alternatively spliced APP mRNA isoforms: APP695, APP751, and APP770 (12, 13) (Fig. 1A), whereas neurons exhibit restricted expression of APP695 (14,20,21). This suggests that cell type-specific APP splicing events occur in the brain. In the present study, we focused on brain endothelial cells and established primary human BMECs to characterize the brain endothelial APP. Western blot analysis using an anti-APP C-terminal antibody (C15) showed that BMECs expressed equivalent or rather higher levels of APP compared with primary neurons (Fig. 1B). Interestingly, the endothelial APP exhibited two discrete bands, based on gel mobility after SDS-PAGE. Because the anti-KPI and anti-OX2 antibodies also detected these signals, we concluded that endothelial cells expressed APP770. Neither the anti-KPI nor the anti-OX2 antibodies detected neuronal APP, thereby confirming previous reports that neurons solely express APP695  (14,20). When we analyzed BMEC lysates with the anti-KPI or anti-OX2 antibodies, we observed an additional band between the two APP bands. This additional band is most likely the processed form of APP, sAPP, which lacks the C-terminal sequence of APP. Using the OX2 antibody, we could detect APP770 in endothelial cells prepared from mouse brains (supplemental Fig. S1). Furthermore, human umbilical vein endothelial cells also expressed the two forms of APP, although their expression levels were significantly lower than those in BMECs (Fig. 1B), indicating that APP770 is broadly expressed in endothelial cells.

Analysis of Brain Endothelial APP
To analyze the APP transcripts, we isolated RNA from endothelial cells and primary neurons. Reverse-transcribed cDNA samples were then analyzed by PCR using a series of oligonucleotide primers to detect APP695, APP751, and APP770. The results confirmed the cell type-specific APP expression, as neurons expressed APP695, whereas endothelial cells expressed APP770 and not APP695 (Fig. 1C).
Endothelial APP770 Has Sialylated Core 1 Type O-Glycans-Because APP potentially has two N-and multiple O-glycans (16,22,23), we expected that the high molecular weight APP770 (APP-H) would be heavily glycosylated. We first performed glycosidase treatment, in which sialidase-treated and -untreated APP were incubated with PNGase. We found that the low molecular weight APP770 (APP-L) was resistant to sialidase treatment, indicating that APP-L contains no sialic acid ( Fig. 2A). After PNGase treatment, both APP-H and APP-L moved slightly faster on SDS-PAGE gels, indicating that both forms contain N-glycan chains, although their molecular weight difference cannot be explained by N-glycans alone. Furthermore, the mobility difference between APP-H and de-N-glycosylated APP-H on SDS-PAGE gels was similar to that after sialidase treatment. This suggests that few sialic acids, if any, were present in the N-glycans of APP-H. Indeed, it seems that both APP-H and APP-L were sensitive to endoglycosidase H, which cleaves high mannose and some hybrid types of N-glycans (supplemental Fig. S2). After sialidase treatment, the mobility of APP-H shifted markedly and became closer to that of APP-L (Fig. 2, A and B). Similarly, after treatment with O-glycosidase, which only cleaves unsubstituted core 1 and 3 type O-glycans, the mobility of APP-H perfectly matched that of APP-L (Fig. 2B). We also found that sialidase-treated APP-H reacted with PNA lectin that specifically detects core 1 type O-glycans (Fig. 2C). Furthermore, a series of lectin pulldown assays showed that APP-H and APP-L have affinities for Sambucus sieboldiana agglutininand concanavalin A-lectin, respectively (Fig. 2D). In addition to the above results, judging from the fact that Sambucus sieboldiana agglutinin lectin recognizes Sia␣2,6Gal/GalNAc, we concluded that APP-H has ␣2,6-sialylated core 1 type O-glycans.
When we overexpressed APP695, APP751, or APP770 in COS cells, mobility differences between the two APP forms were always clearly observed in the case of APP770 (Fig. 3A). These preliminary data suggested the presence of O-glycans in the OX2 domain. Therefore, we generated a series of APP770 mutants in which each Ser/Thr residue within the OX2 domain was individually changed to Ala (Fig. 3B). We then overexpressed wild-type APP770 and its mutants in COS cells. Even though the O-glycosylation machinery is not well developed in COS cells, we detected a lower level of APP-H (Fig. 3C). Mutations at Ser 346 , Ser 348 , and Thr 352 had no effect on the mobility of the APP mutants on SDS-PAGE gels, whereas the APP T353A mutant, in which Thr 353 was changed to Ala, migrated faster than wild-type APP770 (Fig. 3, B and  C). These results suggest that an O-glycan is attached to Thr 353 . The findings that both APP T353A and APP OX2all , in which all of the Ser/Thr residues within the OX2 domain were mutated to Ala, still exhibited two discrete bands suggested the presence of additional O-glycosylation sites in domains other than the OX2 domain in APP770. Perdivara et al. (16) recently showed that APP695 is modified with O-glycans at Thr 291 , Thr 292 , and Thr 576 (Thr 366 , Thr 367 , and Thr 651 for APP770). Therefore, we next generated the APP770 mutant APP all , in which Thr 366 , Thr 367 , Thr 651 , and Thr 652 were additionally changed to Ala from APP OX2all . APP all mostly showed a single band (Fig. 3C), indicating that we had mutated all of the O-glycosylation sites of APP770 in the APP all mutant. O-Glycosylated APP770 Is Preferentially Secreted to Media-Next, we analyzed the APP metabolites in endothelial cells. We performed Western blot analyses of BMEC lysates and culture media with the anti-APP 22C11 antibody, which recognizes the N-terminal APP region and therefore detects sAPP. We observed three bands in the cell lysates using the anti-22C11 antibody. In the media, we detected a single sAPP band that migrated with the middle band observed in the cell lysates (Fig. 4A), suggesting that sAPP is solely derived from APP-H. Indeed, sAPP was sensitive to both sialidase and Oglycosidase treatments (Fig. 4B). Furthermore, after sialidase treatment, most of the sAPP was precipitated with PNA lectin (Fig. 4C), indicating that sAPP mostly contains sialylated core 1 type O-glycan chains. It should be noted that we failed to detect sAPP without O-glycans. Considering that APP cleavage at the ␣-site seems to occur at the cell surface, whereas cleavage at the ␤-site occurs during the endocytotic pathway (24), it is possible that APP without O-glycosylation is unable to move to the cell surface to encounter either ␣or ␤-secretase. However, cell surface biotinylation experiments showed that both APP-H and APP-L reached the cell surface (Fig.  4D). We next studied how cell surface APP is metabolized in the cell. To analyze APP internalization, we next labeled the surface of BMECs with a disulfide cleavable biotinylation reagent (19). Internalization was then induced at 37°C for 1 h. Interestingly, we found that only APP-H was endocytosed (Fig. 4E). This suggests that cell surface APP-H uses a specific intracellular trafficking pathway to encounter both ␣and ␤-secretase. Furthermore, benzyl GalNAc, an inhibitor of Oglycan chain elongation, failed to inhibit sAPP secretion into the media (Fig. 4F).
A␤ is Produced from Brain Endothelial Cells-Since there is limited information regarding the expression levels and activities of the endothelial ␣-secretase and ␤-secretase enzymes for the processing of APP770, we characterized the soluble sAPP770 secreted from BMECs. Using specific antibodies against sAPP␣ and sAPP␤, we successfully detected both sAPP770␣ and sAPP770␤ (Fig. 5A). Since the amyloidogenic ␤-secretase pathway is present in endothelial cells, it is rea-  Total proteins, or internalized biotinylated proteins that were precipitated with streptavidin-Sepharose, were analyzed by Western blotting with the anti-APP C15 antibody. F, BMECs were cultured in the presence of benzyl GalNAc, and then intact APP and sAPP were analyzed by Western blotting with the anti-APP C15 and 22C11 antibodies, respectively. sonable to consider that these cells also have ␥-secretase activity to produce A␤ peptides. Even though endogenous A␤ was not detectable, we detected both A␤40 and A␤42 in the culture media of BMECs overexpressing APP770, and the ratio of endothelial A␤42/A␤40 was similar to that in neurons (Fig. 5B).
APP770 is Expressed in Cerebral Vessels and sAPP770␤ Is Secreted into the CSF-To clarify whether APP770 is indeed expressed in cerebral vessels, we first analyzed cerebral cortex sections with an anti-OX2 antibody to determine the localization of APP770. The luminal regions of the venous and venular endothelial cells, but not smooth muscle cells, were stained with the anti-OX2 antibody (Fig. 5C). No immunohistochemical signals were observed in the vessels of the arachnoid. Next, we investigated whether the CSF contains sAPP770␤, the N-terminal ␤-secretase cleavage product of APP770. sAPP was pulled-down from CSF using heparinagarose and then immunostained with anti-APP 22C11, anti-OX2 and anti-sAPP␤ antibodies. We detected two bands with the anti-APP 22C11 antibody, of which only the upper band was detected with the anti-OX2 antibody (Fig.  5D), indicating that the upper band is derived from APP770. Since both the upper and lower bands were de-tected with the anti-sAPP␤ antibody, both forms contain ␤-secretase cleavage products.

DISCUSSION
The dementia of AD is closely associated with accumulation of A␤ in the brain parenchyma and in the walls of blood vessels in the brain (25)(26)(27). A␤ deposition in the cerebral vasculature contributes to cerebral amyloid angiopathy, which shows a prevalence of Ͼ80% among AD individuals and 10 -40% in elderly people without AD (28). Microbleeds in the brain occur in most cerebral amyloid angiopathy cases (25). Although important roles for cerebral vascular smooth muscle cells in vascular A␤ clearance were recently highlighted (29), the regions where vascular A␤ is produced for deposition remain to be determined.
We have shown for the first time that brain endothelial cells express APP770. First, we unambiguously detected APP770 expression in human BMECs. Using an anti-APP770 antibody that recognizes the OX2 domain for immunohistochemical staining of the cerebral cortex, we found that APP770 was expressed in venous and venular endothelial cells. We also showed that A␤40 and A␤42 were produced in BMECs. Therefore, our study points out the possibility that endothelial A␤ peptides could be the source of the A␤ deposits in cerebral vessel walls. Even though our immunohistochemical study is somewhat different from the previous observations in which A␤ deposits are frequently found in cortical arteries, the major distribution of APP770 in the brain might be different from where A␤ is mainly produced from APP770 to cause vascular A␤ deposits.
Perdivara et al. (16) recently identified the core 1 type Oglycans attached to residues Thr291, Thr292 and Thr576 of APP695. Here, we have shown that APP770 has an additional O-glycan chain at residue Thr353 within the OX2 domain. Furthermore, a series of site-directed mutations within the KPI domain had no effect on the mobility of APP770 (Fig. S3), indicating the absence of O-glycans in the KPI domain. A large ectodomain of APP consists of several subdomains, such as the E1 (30), E2 (31), and KPI domains (32). The web application POODLE-S (Prediction of Order and Disorder by Machine Learning) predicts that the other regions, in which all of the O-glycans are located, are intrinsically unstructured (33) (supplemental Fig. S4). Therefore, it is unlikely that the addition of any O-glycans to APP770 would result in conformational changes that might affect the ␣or ␤-secretase cleavage. Because we found that O-glycosylated APP is selectively internalized, how O-glycosylation of APP would modulate its intracellular trafficking to the same intracellular compartment as ␣and ␤-secretase remains an interesting and unanswered question. Because APP695 was also shown to contain O-glycans, whether our finding applies to neuronal APP also remains to be resolved. We found that BMECs expressed higher amounts of O-glycosylated APP770 than did COS cells, suggesting that endothelial cells possess a highly developed O-glycosylation machinery. Interestingly, endothelial O-glycan deficiency causes blood/lymphatic misconnections (34), indicating a fundamental role for O-glycans in vascular development. In this study, we were able to discriminate CSF sAPP770␤ from other types of sAPP␤. As BACE1 seems to be a stressresponse protein (35), one of our ongoing projects is elucidation of the critical factors required to increase sAPP770␤ secretion. We are now establishing a sandwich ELISA system to quantify sAPP770␤ in CSF or plasma. Because the CSF sAPP770␤ appears to be mainly derived from brain endothelial cells, sAPP770␤ would be a potential biomarker for diagnosing cerebrovascular dementia or AD.