In Silico Screening for Secreted Proteins.

Among the currently available cDNA collections, the cDNA collections from RIKEN are excellent resources because they include >90% of the predicted mouse gene products. Furthermore, >70% of the cDNAs from these collections are full-length (4), and can therefore be used to express full-length proteins for functional screening. We used the following criteria for computational screening of >60,000 cDNAs to identify secreted proteins: (i) the presence of a signal sequence at the N terminus; (ii) an absence of transmembrane domains; (iii) an ORF coding for at least 300 amino acid residues; and (iv) functionally unknown. These criteria are based on the fact that most known ECM genes encode proteins that are >300 aa in length, including an N-terminal secretion signal sequence. A total of 181 clones met the selection criteria.

To verify that the candidate proteins were secreted, we established a high-throughput expression system in which the full-length proteins encoded by candidate cDNAs were expressed in mammalian cells with C-terminal GFP tags, and their secretion into conditioned media was examined by Western blot analyses using anti-GFP mAb. Among the 181 clones, 146 clones were successfully expressed, and 93 of these proteins were secreted into the media (data not shown).

In Vitro Functional Screening for Candidate ECM Proteins.

The secreted proteins were subjected to in vitro high-throughput functional screening assays. The abilities of the candidate proteins to form pericellular deposits were examined by using de novo expression of GFP-tagged proteins in 293T cells or incubation of differentiated myoblasts or mouse embryonic fibroblasts (MEFs) in conditioned media containing the secreted proteins [Fig. 2A and supporting information (SI) Fig. S1]. To screen for interactions with known ECM molecules, conditioned medium containing the candidate proteins was placed in 96-well plates preadsorbed with 19 different ECM molecules, including various types of collagens, laminins, and GAGs as well as fibronectin. Bound proteins were quantified by using the fluorescent signals derived from the GFP tags (Fig. 2B and Fig. S2). To screen for GAG modifications, mobility shifts of the secreted proteins after treatment with enzymes that cleaved heparan sulfate and chondroitin sulfate chains were examined by Western blot analysis. The types of GAGs and their covalent attachments to the core proteins were further analyzed by using separate treatments with heparan sulfate- or chondroitin sulfate-cleaving enzymes, followed by Western blot analysis with anti-GFP mAb and various anti-stub mAbs that specifically recognized neoepitopes generated by the enzymes (Fig. 2C and Fig. S3). To screen for the promotion of cell–substrate adhesion, the secreted proteins were captured on 96-well plates via anti-GFP antibody and used as substrates in cell adhesion assays by using MG63 osteosarcoma, HT1080 fibrosarcoma, and HeLa cells (Fig. 2D and Fig. S4). These in vitro functional assays identified 24 proteins as putative ECM proteins of which 19 localized to ECMs, 14 bound to known ECM proteins, 4 contained GAG chains, and 11 promoted cell–substrate adhesion (summarized in Fig. 3).

Open in a separate window

Fig. 2.

Representative results obtained in the screening assays. (A) Pericellular deposits were examined by using the de novo expression of GFP-tagged candidate proteins in 293T cells and incubation of differentiated C2C12 myoblasts or MEFs in conditioned media containing proteins secreted from transfected 293F cells. (Scale bars, 10 μm.) (B) The ECM molecules used in the solid-phase binding assays were as follows: 1, collagen I; 2, collagen II; 3, collagen III; 4, collagen IV; 5, collagen V; 6, collagen VI; 7, gelatin; 8, fibronectin; 9, laminin-111; 10, laminin-211/221; 11, laminin-511/521; 12, heparin; 13, heparan sulfate; 14, chondroitin sulfate A; 15, dermatan sulfate; 16, chondroitin sulfate C; 17, chondroitin sulfate D; 18, chondroitin sulfate E; 19, hyaluronic acid; and 20, BSA (negative control). The amount of bound GFP-fusion protein was determined by measuring the intensity of GFP fluorescence. (C) Attachment of GAG chains was examined by incubating GFP-fusion proteins in conditioned media from transfected 293F (ECM290) or COS (ECM898) cells with heparinase and heparitinase (h), chondroitinase ABC (c), or buffer alone (n), followed by Western blot analyses using anti-GFP mAb to detect mobility shifts of the candidate proteins. After incubation with GAG-degrading enzymes, the smeared bands (asterisks) disappeared (or diminished) with a concomitant appearance (or increase) of the faster migrating bands (arrows). A putative proteolytic fragment of the ECM290-GFP fusion protein is indicated by an arrowhead. (D) The cell adhesion-promoting activities were examined by incubating MG63, HT1080, and HeLa cells in 96-well plates containing immobilized GFP-tagged candidate proteins. Cells adhering to the substrates were visualized by staining with Diff-Quick. (Scale bars, 100 μm.) (E) Immunohistochemical localizations of the candidate proteins. Sagittal sections of E16.5 mouse embryos were stained with affinity-purified antibodies. The following representative localization data are shown: ECM201 (1), periosteum of the humerus; ECM322 (2), ligaments and associated perichondrium surrounding the spinal cartilage; ECM314 (3), rib cartilage; ECM306 (4), intervertebral disks in the spinal cord; ECM661 (5), lung epithelium; ECM392 (6), lip epithelium; ECM270 (7), choroid plexus epithelium and associated blood vessels; ECM290 (8), hair follicle. ca, cartilage; pc, perichondrium; po, periosteum; tl, tendon or ligament; id, intervertebral disk; hu, humerus. (Scale bars, 100 μm.)

Open in a separate window

Fig. 3.

A summary of the screening results. Accession numbers, genomic loci in mice, numbers of amino acids, PSG and GvH scores, and the domain structures of the putative ECM proteins encoded by the clones isolated with the in silico and in vitro screening assays are indicated. Proteins that were given new names in this study are underlined (see also Table S1). Results obtained from in vitro screening and in vivo IHC screening are also indicated on the right. Labeling in yellow and gray denotes positive and negative results, respectively. Labeling in white (blank) denotes “not determined.” Proteins localized at atypical ECM structures are blue. Positive staining of the interstitial matrices of cartilage (ca), tendon/ligament (tl), perichondrium (pc), and periosteum (po) are indicated. *, The predicted protein structures based on the arrangement of the Pfam domains. FNIII, fibronectin type III domain; LG, laminin G domain; VWA, von Willebrand factor type A domain; VWC, von Willebrand factor type C domain; KZ, KAZAL-type serine protease inhibitor domain; Tg, thyroglobulin type 1 repeat; EF, EF-hand calcium-binding domain; TSP1, thrombospondin type 1 domain; LRR, leucine-rich repeat; IGFBP, insulin-like growth factor binding protein; Ig, Ig domain; LRRNT, leucine-rich repeat N-terminal domain; OLF, olfactomedin-like domain. RGD motifs are denoted with red stars. **, The results have been published (5, 6).

To assess the reliability and specificity of our strategy, we performed pilot experiments in which 26 cDNAs encoding known ECM proteins were selected from RIKEN′s cDNA collections and subjected to the in vitro assays. Each of these cDNAs conformed to the criteria for the computational screening and the encoded proteins were secreted after transfection. Twenty-three of these proteins were positive in at least one of the in vitro assays (Table S2), verifying the utility of our strategy.

The in vitro assays revealed a marked degree of functional diversity among the selected proteins. Among the 19 ECM-assembling proteins, 18 were immobilized in the ECM of either 293T cells or differentiated myoblasts, whereas only three were capable of assembling into both ECMs (Fig. 3 and Fig. S1). A number of the proteins were found to bind with heparin (e.g., ECM482/eratin), although some of them displayed more specific binding to dermatan sulfate (e.g., ECM331/coffeecrisp) or chondroitin sulfate E (e.g., ECM392/ependolin) (Fig. 2B). Some of the proteins exhibited affinities for collagens with different binding spectra (e.g., ECM482/eratin). Among the 11 cell-adhesive proteins, seven showed some degree of cell-type specificity; for example, ECM482/eratin mediated the adhesion of MG63 and HT1080 cells, but not HeLa cells, whereas ECM742/cradin mediated the adhesion of HeLa and HT1080 cells, but not MG63 cells (Fig. 2D and Fig. S4). Five of the cell-adhesive proteins contained an RGD motif (Fig. 3), and at least two of them [ECM742/cradin and ECM661/MAEG (5)] promoted cell adhesion in an RGD-dependent manner (Fig. S4). These differences are indicative of the functional diversity of the selected proteins and support the use of the complementary in vitro screening assays.

Immunohistochemical Localization of the Candidate Proteins.

To confirm that the selected proteins were indeed ECM proteins, we immunohistochemically examined their tissue localizations. Among the 24 candidate proteins, at least 16 were confirmed to be ECM proteins based on their localizations in ECM structures (Fig. 2E and Fig. S5). These proteins displayed clear tissue-specific expression patterns; three were localized in the periostea, three in the perichondria, four in tendons and/or ligaments, seven in cartilage, and seven at the BM zones of various cell types (Fig. 3 and Fig. S5). Antibodies against six other proteins gave restricted staining patterns that were not indicative of ECM proteins (Fig. S6).

More than half of the proteins thus identified were encoded by the genes that are currently only represented by mammalian ESTs in public databases. Many of the proteins contained consensus domains frequently found in ECM proteins. Fibronectin type III repeats were found in four proteins; von Willebrand factor type A and type C domains, MAM domains, and multiple small leucine-rich repeats were found in two proteins, respectively (Fig. 3). One protein (ECM290/nectican) showed similarity with agrin in that both proteins shared a domain structure characterized by the presence of three laminin G-like domains interspersed with EGF-like domains. It should be noted that the frequency of such ECM domains in the proteins that were selected with the in silico screening but failed in the in vitro screening assays was significantly lower than the frequency in the proteins that passed the in vitro assays (Table S3). Moreover, three proteins without any known domains (ECM392/ependolin, ECM314/emprin, and ECM517/RAINB2) were also identified as ECM proteins, underscoring the advantages of our function-based and localization-based screening strategies compared with screening based solely on the primary protein structure.

Screening for ECM Deposition.

293T cells were plated in 96-well glass-bottomed plates precoated with fibronectin (10 μg/ml) and transfected with the GFP fusion plasmids. After incubation for 3–4 days, the cells were fixed with 2% formaldehyde and labeled with anti-GFP mAb (Santa Cruz Biotechnology) under nonpermeable conditions followed by rhodamine-conjugated secondary antibodies. Alternatively, C2C12 myoblasts plated in 96-well glass-bottomed plates precoated with 0.1% gelatin were cultured in DMEM supplemented with 1% horse serum and insulin (0.4 unit/ml) to induce differentiation. At 4–5 days after the induction of differentiation, the medium was replaced with conditioned medium from transfected 293F cells (Invitrogen), and the cells were cultured for 1 day. Confluent monolayer cultures of MEFs were fed with a 1:1 mixture of complete medium and conditioned medium from transfected 293F cells and cultured for 3 days. Samples were visually screened for the deposition of GFP-fusion proteins in ECM-like structures by using an LSM5 PASCAL confocal microscope (Carl Zeiss).

Screening for ECM Molecule-Binding Activity.

Black 96-well plates were coated with ECM proteins or GAGs at 4°C overnight. The GAG chains were conjugated to dipalmitoyl phosphatidylethanolamine (24) to ensure their adsorption to the plastic surface. The plates were blocked with 3% BSA and then incubated with conditioned media from transfected 293T or 293F cells overnight at 4°C. After washing the plates, the amount of bound GFP-fusion protein was determined by measuring the intensity of GFP fluorescence with a Fluoroskan Ascent fluorometer (Thermo Scientific). The mean background fluorescence intensity (n = 2) after subtraction of the intensity of sGFP (the negative control) was usually <0.001; therefore, we set the threshold value for positivity in the assays to 0.005. At least two independent experiments were performed and representative data are shown. When the mean background intensity was >0.001, the threshold was set to a value that was 5 times higher than the mean value.

Screening for Cell Adhesion-Promoting Activity.

To immobilize the GFP-fusion proteins, conditioned media containing the GFP-fusion proteins were added to anti-GFP polyclonal antibody-coated 96-well plates and incubated at 4°C overnight. The plates were blocked with 1% BSA, and seeded with MG63, HT1080, or HeLa cells suspended in DMEM containing 0.1% BSA at a density of 3 × 10³ cells per well. After incubation for 40 min, the cells were washed twice with PBS, fixed with 3.7% formaldehyde, and stained with Diff-Quik (Sysmex). Under these conditions, no background activity was detected irrespective of the examined cell types.

Screening for GAG Modifications.

Recombinant proteins in conditioned media from transfected 293F or COS cells were treated with a mixture of heparitinase (0.005 unit/ml), heparinase (0.005 unit/ml), and chondroitinase ABC (0.25 unit/ml) in 50 mM sodium acetate (pH 7.0), 4 mM CaCl₂, and 0.2 mM Pefabloc (Roche) or with buffer alone at 37°C for 2 h. The samples were separated by SDS/PAGE under reducing conditions, and subjected to Western blot analysis with anti-GFP mAb. Proteins that migrated faster after treatment with the GAG-cleaving enzymes were treated separately with a mixture of heparinase and heparitinase or chondroitinase ABC to determine which types of GAG chains were covalently attached to the proteins.

We thank T. Emoto, H. Fujiwara, and Y. Hayashi for antigen production; N. Sugimoto, C. N. Weber, T. Hasegawa, M. Kondo, and Y. Iwase for digital image construction; A. Ueda, M. Mutoh, and A. Morimoto for webpage construction; K.M. Yamada and M. Okuhara for helpful comments; T. Nishida and K. Seki for providing tissue samples; H. Nakamura and S. Futaki for web server maintenance; and M. Saito for critical discussions. This study was supported in part by New Energy and Industrial Technology Development Organization of Japan Research Contract 06001294-0 and by a grant from the RIKEN Genome Exploration Research Project.