Molecular Basis of the Ligand Binding Specificity of αvβ8 Integrin*

αvβ8 is an integrin that recognizes an Arg-Gly-Asp (RGD) motif and interacts with fibronectin, vitronectin, and latent TGF-β1. We comprehensively determined the binding activity of the αvβ8 integrin toward 25 secreted proteins having an RGD motif. The αvβ8 integrin strongly bound to latent TGF-β1 but showed marginal activity for other RGD-containing proteins, including fibronectin and vitronectin. Site-directed mutagenesis of latent TGF-β1 demonstrated that the high affinity binding of αvβ8 integrin to latent TGF-β1 was defined by Leu-218 immediately following the RGD motif within the latency-associated peptide of TGF-β1. Consistent with the critical role of Leu-218 in latent TGF-β1 recognition by αvβ8 integrin, a 9-mer synthetic peptide containing an RGDL sequence strongly inhibited interactions of latent TGF-β1 with αvβ8 integrin, whereas a 9-mer peptide with an RGDA sequence was ∼60-fold less inhibitory. Because αvβ3 integrin did not exhibit strong binding to latent TGF-β1 or distinguish between RGDL- and RGDA-containing peptides, we explored the mechanism by which the integrin β8 subunit defines the high affinity binding of latent TGF-β1 by αvβ8 integrin. Production of a series of swap mutants of integrin β8 and β3 subunits indicated that the high affinity binding of αvβ8 integrin with latent TGF-β1 was ensured by interactions between the Leu-218 residue and the β8 I-like domain, with the former serving as an auxiliary recognition residue defining the restricted ligand specificity of αvβ8 integrin toward latent TGF-β1. In support of this conclusion, high affinity binding toward the αvβ8 integrin was conferred on fibronectin by substitution of its RGDS motif with an RGDL sequence.

␣v␤8 is an integrin that recognizes an Arg-Gly-Asp (RGD) motif and interacts with fibronectin, vitronectin, and latent TGF-␤1. We comprehensively determined the binding activity of the ␣v␤8 integrin toward 25 secreted proteins having an RGD motif. The ␣v␤8 integrin strongly bound to latent TGF-␤1 but showed marginal activity for other RGD-containing proteins, including fibronectin and vitronectin. Site-directed mutagenesis of latent TGF-␤1 demonstrated that the high affinity binding of ␣v␤8 integrin to latent TGF-␤1 was defined by Leu-218 immediately following the RGD motif within the latency-associated peptide of TGF-␤1. Consistent with the critical role of Leu-218 in latent TGF-␤1 recognition by ␣v␤8 integrin, a 9-mer synthetic peptide containing an RGDL sequence strongly inhibited interactions of latent TGF-␤1 with ␣v␤8 integrin, whereas a 9-mer peptide with an RGDA sequence was ϳ60-fold less inhibitory. Because ␣v␤3 integrin did not exhibit strong binding to latent TGF-␤1 or distinguish between RGDL-and RGDA-containing peptides, we explored the mechanism by which the integrin ␤8 subunit defines the high affinity binding of latent TGF-␤1 by ␣v␤8 integrin. Production of a series of swap mutants of integrin ␤8 and ␤3 subunits indicated that the high affinity binding of ␣v␤8 integrin with latent TGF-␤1 was ensured by interactions between the Leu-218 residue and the ␤8 I-like domain, with the former serving as an auxiliary recognition residue defining the restricted ligand specificity of ␣v␤8 integrin toward latent TGF-␤1. In support of this conclusion, high affinity binding toward the ␣v␤8 integrin was conferred on fibronectin by substitution of its RGDS motif with an RGDL sequence.
Integrins are a family of adhesion receptors that bind to a variety of extracellular ligands, typically cell adhesion proteins in the extracellular matrix (ECM). 2 Integrins play mandatory roles in embryonic development and the maintenance of tissue architecture by providing essential links between cells and the ECM (1). Integrins are composed of two non-covalently asso-ciated subunits, termed ␣ and ␤. In mammals, 18 ␣ and 8 ␤ subunits have been identified, and combinations of these subunits give rise to at least 24 distinct integrin heterodimers, among which 18 isoforms function as ECM receptors. Based on their ligand binding specificities, ECM-binding integrins are classified into three major groups as follows: laminin-, collagen-, and Arg-Gly-Asp (RGD)-binding integrins (1,2), of which the RGD-binding integrins have been most extensively investigated. The RGD-binding integrins include ␣5␤1, ␣8␤1, ␣IIb␤3, and ␣v-containing integrins, which interact with a variety of ECM ligands containing RGD motifs with distinct binding specificities.
The integrin ␣v subunit was originally identified as a receptor for vitronectin (3). The ␣v-containing integrins are widely expressed on many cell types, including neural crest cells, glial cells, muscle cells, osteoclasts, epithelial cells, and vascular endothelial cells during embryonic development (4 -8) and during angiogenesis in response to tumors (9). Mouse embryos containing a null mutation in the ␣v gene exhibit placental defects and intracerebral hemorrhage, indicative of its important role in placentation and vasculogenesis (10,11). To date, the ␣v subunit has been shown to combine with five ␤ subunits, ␤1, ␤3, ␤5, ␤6, and ␤8. Among the five ␣v-containing integrins, ␣v␤8 is the major ␣v integrin and has a critical role in placentation and vasculogenesis, because the phenotypes of mice lacking integrin ␤8 expression largely overlap with those lacking the ␣v subunit (12).
The amino acid sequence of the ␤8 subunit is highly conserved among vertebrates but is divergent from other integrin ␤ subunits (13), suggesting the ␣v␤8 integrin may have unique functions among the ␣v-containing integrins. The exogenous expression of integrin ␤8 inhibited cell growth, spreading, and focal contact formation (14,15), in contrast to the exogenous expression of other ␣v-associated integrin ␤ subunits. The ␤8 subunit is found in the brain, kidneys, airways, and placenta and is localized at brain vessels, synapses, glial cells, and dendritic spines, implying a specific function in the brain (13,16). To date, the ␣v␤8 integrin has been shown to bind vitronectin (14), fibronectin (17), and a latency-associated peptide of TGF-␤1/3 (18,19), among which TGF-␤1 is the most characterized ligand for ␣v␤8 integrin. The ␣v␤8 integrin binds to the latent form of TGF-␤1 (designated latent TGF-␤1) and activates it by releasing the mature TGF-␤1 from the latency-associated peptide (18). However, the binding capability of the ␣v␤8 integrin to other RGD-containing proteins has not been comprehensively analyzed at the molecular level, suggesting unknown proteins containing the RGD motif might also serve as ligands that specifically interact with the ␣v␤8 integrin.
Although ligand recognition by RGD-binding integrins is primarily determined by the RGD motif in the ligands, several studies have demonstrated that residues outside the RGD motif also define binding specificities and affinities toward individual RGD-binding integrins (2,20). For example, ␣5␤1 integrin specifically binds to fibronectin through the bipartite recognition of an RGD motif in the 10th type III repeat, together with the PHSRN sequence and several basic residues within the 9th type III repeat, the latter serving as a "synergy site" (22,23). ␣8␤1 integrin selectively binds to nephronectin via a bipartite interaction with the RGD motif and LFEIFEIER sequence, the latter located at the C-terminal ϳ10 amino acids from the RGD motif (24). The high affinity binding of ␣v␤6 integrin to its ligands, foot-and-mouth disease virus and latent TGF-␤1, requires the RGD motif and an LXX(L/I) sequence, of which the latter forms an ␣-helix to align the two conserved hydrophobic residues along the length of the helix (25,26). Thus, the ligand-binding specificity of RGD-binding integrins might be defined by the bipartite recognition site comprising an RGD motif and residues flanking the RGD motif or those in neighboring domains that come into close proximity with the RGD motif in an intact ligand protein. However, the ligand-binding specificity of the ␣v␤8 integrin toward a broad range of RGD-containing proteins as well as the molecular mechanisms defining its ligand specificity remain poorly understood.
In this study, we comprehensively investigated the binding activities of the ␣v␤8 integrin toward 25 RGD-containing proteins selected by in silico screening for putative secreted proteins containing a conserved RGD motif. Our results showed that the ␣v␤8 integrin has a restricted binding specificity toward latent TGF-␤1, which is primarily defined by the Leu-218 residue located immediately after the RGD motif. The mechanism by which the ␣v␤8 integrin recognizes the Leu-218 residue was investigated by constructing a series of swap mutants between integrin ␤8 and ␤3 subunits.

Experimental Procedures
Cells, Antibodies, and Reagents-FreeStyle TM 293-F cells were obtained from Life Technologies, Inc., and cultured in FreeStyle 293-F expression medium. Human plasma fibronectin was purified from outdated human plasma by gelatin affinity chromatography as described previously (27). HRP-conjugated mAbs against FLAG and penta-His tags were purchased from Sigma and Qiagen (Valencia, CA), respectively. An anti-"Velcro" (ACID/BASE coiled-coil) antibody was raised in rabbits by immunization with coiled-coil ACID and BASE peptides, as described previously (28), and biotinylated by an EZ-link NHS-Sulfo-LC-biotin kit (Pierce) to detect recombinant integrins. HRP-conjugated streptavidin was purchased from Pierce. Synthetic peptides were purchased from Thermo Fisher Scientific (Dreieich, Germany) and dissolved in 100% DMSO.
In Silico Screening of RGD Motif-containing Proteins-The Protein Information Resource (PIR) Perfect Peptide Match program was used to screen proteins registered in the UniProt Knowledgebase (UniprotKB) (29,30). Non-redundant proteins possessing at least one RGD motif were selected and then further screened for their ability to be secreted into the extracellular space, based on their annotation in the Uniprot database or the presence or absence of signal peptides and transmembrane regions, respectively, which were predicted by using PSORT II (31) and SOSUI (32). Conservation of RGD motifs in vertebrates was assessed using the Ensembl genome database (33).
A cDNA encoding the extracellular region of human integrin ␤8 was amplified by reverse transcription-PCR using total RNA extracted from WiDr human colon carcinoma cells as a template. The PCR-amplified cDNA was digested with BamHI/ PmeI and inserted into the corresponding restriction sites of the pEF expression vector in-frame to the sequence encoding the "BASE" peptide and a His 6 tag at the 3Ј end of the integrin ␤8 cDNA as described previously (35). Another pEF vector encoding the extracellular region of integrin ␤8 lacking a C-terminal His 6 tag was also constructed by overlap extension PCR for expression of recombinant ␣v␤8 integrin lacking the His 6 tag (designated as ␣v␤8(⌬His)). The expression vectors for the extracellular regions of integrin ␣v and ␤3 were described previously (28,35). cDNAs encoding a series of swap mutants of the extracellular region of integrin ␤8 and ␤3 were amplified by PCR using cDNAs encoding the integrin ␤8 and ␤3 as a template, respectively. The primer sequences for PCR are available upon request. After verification by DNA sequencing, PCR-amplified cDNA fragments were digested with BamHI and NheI and inserted into the corresponding restriction sites of pEFintegrin ␤8-BASE-His 6 or pEF-integrin ␤3-BASE-His 6 and verified by DNA sequencing, respectively.
Site-directed Mutagenesis-Site-directed mutagenesis of latent TGF-␤1 and a truncated form of fibronectin (FNIII7-10) was accomplished by overlap extension PCR with KOD polymerase using pSecTag-TGF-␤1 and pSecTag-FNIII7-10 as templates, respectively. The primer sequences for the site-directed mutagenesis are available upon request. After verification by DNA sequencing, the PCR products containing the mutations were subcloned into the pSecTag2B vector.
Expression and Purification of Recombinant Proteins and Integrins-GST-fused EGF-like protein 6 (EGFL6), GST-fused FRAS-1-related extracellular matrix protein 1 (FREM1), laminin-␣5 (LAMA5, as laminin-511), and nephronectin were purified as described previously (24, 36 -38). For purification of other RGD-containing proteins, FreeStyle 293-F cells (Life Technologies, Inc.) were transiently transfected with individual expression vectors according to the manufacturer's instructions. The conditioned media were collected at 72 h after transfection and centrifuged to remove cells and debris, followed by addition of Pefabloc SC (Roche Diagnostics, Basel, Switzerland; 0.4 mM), imidazole (10 mM), and sodium azide (0.02%). The conditioned media were incubated with nickel-nitrilotriacetic acid (Ni-NTA)-agarose beads (Qiagen), followed by washing with TBS. Bound proteins were eluted with TBS containing 200 mM imidazole. The eluted fractions except for latent TGF-␤1, vitronectin, thrombospondin-1, and PCSK6 were applied to columns of anti-FLAG M2-agarose beads (Sigma), and the bound proteins were eluted with 100 g/ml FLAG peptide (Sigma). Recombinant latent TGF-␤1, vitronectin, thrombospondin-1, and PCSK6 were purified by one-step Ni-NTA affinity chromatography. The purified proteins were dialyzed against TBS and quantified by a protein assay kit (Bio-Rad) using BSA as a standard. Recombinant integrins were expressed in FreeStyle 293-F cells by cotransfection with expression vectors encoding integrin ␣ and ␤ subunits and purified as described above by two-step affinity chromatography, except for ␣v␤8(⌬His), which was purified by one-step chromatography using anti-FLAG-M2-agarose (28,35).

SDS-PAGE and Western
Blotting-SDS-PAGE was carried out according to Laemmli (39) using 8, 12, or 5-20% gradient gels. Separated proteins were visualized by Coomassie Brilliant Blue staining or transferred onto PVDF membranes (Millipore, Billerica, MA) for immunoblotting. The membranes were treated with TBS containing 5% skim milk and 0.05% Tween 20 for detection of FLAG tags or anti-His blocking reagent (Qiagen) containing 0.05% Tween 20 for detection of His 6 tags. The membranes were then probed with HRP-conjugated antibodies against FLAG or penta-His tags, followed by visualization with the ECL Western blotting substrate (GE Healthcare).
Integrin Binding Assay-Integrin binding assays were performed as described previously (40). Briefly, microtiter plates (Nunc-Immuno TM MicroWell TM 96-well plates; Thermo Fisher Scientific) were coated with various RGD-containing proteins (10 nM) overnight at 4°C and then blocked with TBS containing 10 mg/ml BSA. The plates were incubated with integrins in the presence of 1 mM MnCl 2 with or without 10 mM EDTA. In the inhibition assays, integrins were incubated on the plates in the presence of synthetic peptides at various concentrations to evaluate their inhibitory activities. The plates were washed with TBS containing 1 mM MnCl 2 , 0.1% BSA, and 0.02% Tween 20 with or without 10 mM EDTA, followed by quantification of bound integrins by an ELISA using a biotinylated rabbit anti-Velcro antibody and HRP-conjugated streptavidin. Integrin binding assays were also performed in a reverse manner. Briefly, microtiter plates were coated with 10 nM ␣v␤8(⌬His) integrin and then incubated with various RGDcontaining proteins having a His 6 tag. Bound proteins were quantified by ELISA using an HRP-conjugated anti-penta-His antibody. The results represent the means of triplicate determinations. Apparent dissociation constants were calculated by saturation binding assays as described previously (40).

Results
In Silico Screening of RGD-containing Proteins as Candidates for ␣v␤8 Integrin Ligands-Among the currently available databases for protein sequences, UniProtKB is a suitable resource for the screening of putative ␣v␤8 integrin ligands because it contains ϳ91,800 human protein sequences, some of which are manually annotated with information extracted from the literature. We performed protein sequence-based screening for proteins containing an RGD motif(s) using the Protein Information Resource Perfect Peptide Match program and extracted 5,083 proteins containing at least one RGD sequence. Given that this program often assigns different ID numbers for alternatively spliced variants and/or fragments, we reassigned the same ID numbers for such variants, thus yielding a non-redundant protein list comprising 1,909 proteins possessing at least one RGD motif. Because integrins are cell-surface receptors that recognize extracellular proteins, we next screened for putative secreted proteins based on the annotation of UniProtKB or the presence of a signal peptide and the absence of transmembrane region(s), respectively, as predicted by PSORTII and SOSUI. This yielded 190 putative secreted proteins with an RGD motif. Because integrin ␤8 is evolutionarily conserved among vertebrates, we assumed that the ligands for the ␣v␤8 integrin should have an RGD motif conserved among vertebrates. Screening of 190 putative secreted proteins among vertebrate orthologs that harbored the conserved RGD motif yielded 29 candidates for the ␣v␤8 integrin ligands (Table 1).
Recombinant Expression and Purification of the Candidates for ␣v␤8 Integrin Ligands-We expressed and purified the 29 RGD-containing proteins listed in Table 1. Fibronectin, EGFL6, FREM1, laminin-␣5 (LAMA5, as laminin-511), and nephronectin were purified as described previously (24, 27, 36 -38). Latent TGF-␤1 was expressed as a recombinant protein using a mammalian expression system with a His 6 tag at the N terminus and purified from conditioned medium by Ni-NTA affinity chromatography. Other proteins were expressed with N-terminal FLAG and C-terminal His 6 tags to ensure the secretion of fulllength proteins. Secreted recombinant proteins were purified from the conditioned medium by Ni-NTA affinity chromatography. When protein purity was not sufficient, they were further purified by anti-FLAG mAb affinity chromatography. The authenticity of the purified proteins was verified by SDS-PAGE under reducing conditions, followed by Coomassie Brilliant Blue staining and immunoblotting against the FLAG and His 6 tags. Purified latent TGF-␤1 gave three bands migrating at 50 kDa (unprocessed latent TGF-␤1), 37 kDa (the latency-associated peptide), and 10 kDa (mature TGF-␤1) (Fig. 1A), confirming the authenticity of the purified protein (41). Except for thrombospondin-1, proprotein convertase subtilisin/kexintype 6 (PCSK6), semaphorin-3C (SEMA3C), extracellular matrix protein 2 (ECM2), lysine-tRNA ligase (KAR2), and PCSK5, the other proteins were also purified at their predicted molecular mass regions (Fig. 1B). Thrombospondin-1 and PCSK6 were obtained as proteolytically processed forms com-prising N-and C-terminal fragments by Ni-NTA affinity chromatography. It was difficult to obtain sufficient amounts of recombinant SEMA3C, ECM2, KAR2, and PCSK5 for subsequent integrin binding assays because of their low levels of expression and/or proteolytic degradation.
␣v␤8 Integrin Preferentially Binds to Latent TGF-␤1-A total of 25 RGD-containing proteins (hereafter designated as "RGD proteins") were purified and subjected to integrin binding assays using recombinant ␣v␤8 integrin expressed and purified as a disulfide-linked heterodimer of the extracellular domains of ␣v and ␤8 chains. The purified ␣v␤8 integrin gave a single band migrating at ϳ250 kDa upon SDS-PAGE under non-reducing conditions and was resolved into four bands, i.e. ϳ150 kDa (unprocessed ␣v chain), ϳ120 kDa (␣v heavy chain), ϳ100 kDa (␤8 chain), and ϳ30 kDa (␣v light chain), under reducing conditions (Fig. 1C). The integrin binding assays were performed in the presence of 1 mM Mn 2ϩ to fully activate ␣v␤8 integrin. Among the 25 RGD proteins examined, EDIL3, fibrillin-1, fibrillin-2, fibronectin, osteopontin, latent TGF-␤1, and vitronectin bound to the ␣v␤8 integrin with different binding affinities, but other RGD proteins did not show any significant binding to the ␣v␤8 integrin ( Fig. 2A). Because the coating efficiency varies among the 25 RGD proteins, we also performed reverse binding assays in which microtiter plates were coated with ␣v␤8 integrin without a His 6 tag and then incubated with a panel of RGD proteins added in the solution phase. Bound RGD proteins were detected with an HRP-conjugated antipenta-His antibody. EGFL6, fibronectin, FREM1, laminin-␣5, and nephronectin without a His 6 tag were not included in this assay. Among the 20 RGD proteins tested, only latent TGF-␤1   bound strongly to the ␣v␤8 integrin in a divalent cation-dependent manner (Fig. 2B). ANGPTL7 and IBSP also gave positive signals in the assay, but the signals still persisted in the presence of 10 mM EDTA, suggesting that they did not represent authentic interactions of the ␣v␤8 integrin with its ligand through the divalent cation in the ␤8 subunit. The results obtained in two separate assays of the reverse format indicated that the ␣v␤8 integrin preferentially bound to latent TGF-␤1 with an affinity far exceeding those of fibronectin and vitronectin, the known ligands for the ␣v␤8 integrin. Saturation binding assays revealed that ␣v␤8 integrin bound to latent TGF-␤1 with an apparent dissociation constant of 2.3 Ϯ 0.2 nM, which was approximately 1 or 2 orders of magnitude lower than that of fibrillin-1, fibronectin, and vitronectin (Fig. 2C). Substitution of the RGD motif of latent TGF-␤1 with an inactive RGE sequence completely abrogated the ability of latent TGF-␤1 to bind to the ␣v␤8 integrin, confirming the RGD-dependent interaction of latent TGF-␤1 with ␣v␤8 integrin. The low binding affinities of the ␣v␤8 integrin toward vitronectin, fibronectin, and fibrillin-1 were not caused by inactivation of their RGD ligand activity because they retained the ability to bind to ␣v␤3 integrin, an integrin that binds to these RGD proteins (Fig. 2D) (42)(43)(44). It should be noted that the ␣v␤3 integrin showed only marginal binding activity to latent TGF-␤1. Its activity was nullified by substitution of the RGD motif with an inactive RGE sequence (Fig. 2D). These results indicated that the ligand specificity and binding affinity of ␣v␤8 integrin toward latent TGF-␤1 differ significantly from the ␣v␤3 integrin.
Leu-218 Residue Immediately Following the RGD Motif Is Required for High Affinity Binding to ␣v␤8 Integrin-To explore the molecular basis of the restricted ligand specificity of ␣v␤8 integrin, we focused on the LXXI sequence immediately following the RGD motif, because the LXXI sequence was required for the high affinity binding of latent TGF-␤1 to ␣v␤6 integrin in concert with the RGD motif (25,26). We constructed latent TGF-␤1 mutants, in which the Leu-218 and/or Ile-221 residues of the LXXI sequence were substituted with Ala, and we assessed their ability to bind to the ␣v␤8 integrin (Fig. 3). Although the I221A substitution did not affect the binding of latent TGF-␤1 to ␣v␤8 integrin, the L218A substitution almost completely abrogated the binding, similar to an RGD 3 RGE substitution. L218A/I221A double substitution also inactivated binding to ␣v␤8 integrin. These results indicated that both the RGD motif and the Leu-218 residue were strictly required for the high affinity binding of latent TGF-␤1 to ␣v␤8 integrin. To confirm the importance of Leu-218 immediately following the RGD motif in latent TGF-␤1 binding by the ␣v␤8 integrin, we examined whether synthetic peptides modeled after the RGDLcontaining sequence in latent TGF-␤1 could inhibit the binding of latent TGF-␤1 to ␣v␤8 integrin. We synthesized a 9-mer peptide containing the RGDL sequence (RRGDLATIH, designated as RGDL) and its mutant forms with RGDL 3 RGDA, RGDL 3 RGEL, and RGDL 3 RGEA substitutions (Fig. 4A), and we examined their inhibitory effects on the binding of the ␣v␤8 integrin to latent TGF-␤1. The RGDL peptide strongly inhibited the binding of ␣v␤8 integrin to latent TGF-␤1 with an IC 50 of ϳ0.3 M (Fig. 4B and Table 2). Substitution of the Leu residue with Ala resulted in an ϳ60-fold decrease in the potency of the peptide to inhibit the ␣v␤8 integrin-latent TGF-␤1 interaction, although the RGEL peptides, in which the Asp residue of the RGD motif was substituted with Glu, resulted in an ϳ500-fold decrease. The RGEA peptide showed  little inhibitory effect at the highest peptide concentration used. These results indicated that although the RGD motif is the primary determinant, the Leu residue immediately after the RGD motif is critically required for the binding of the ␣v␤8 integrin to latent TGF-␤1. We also examined the potency of the peptides to inhibit the binding of ␣v␤3 integrin to fibronectin (Fig.  4C). Both the RGDL and RGDA peptides were equally inhibitory with an IC 50 of 0.3-0.4 M, regardless of the presence or absence of the Leu residue following the RGD motif, although RGEL and RGEA peptides were only weakly inhibitory. These results indicated that the Leu residue immediately after the RGD motif is not involved in ligand recognition by the ␣v␤3 integrin, although it is indispensable for latent TGF-␤1 recognition by the ␣v␤8 integrin.
To further corroborate the critical role of Leu-218 immediately after the RGD motif in latent TGF-␤1 recognition by ␣v␤8 integrin, we examined whether the high affinity binding of latent TGF-␤1 toward the ␣v␤8 integrin could be conferred on fibronectin by substituting the RGDL sequence for the RGDS cell-adhesive motif in the 10th FNIII domain. We produced a truncated form of fibronectin consisting of the 7th to 10th type III domains (FNIII7-10) and corresponding mutants with substitution of its RGDS sequence for RGDL or RGEL, and we assessed their abilities to bind to the ␣v␤8 integrin (Fig. 5A). Although control FNIII7-10 bound to the ␣v␤8 integrin with only moderate affinity, the RGDL mutant bound strongly to the ␣v␤8 integrin with an apparent K d of 2.2 nM, which was comparable with that of latent TGF-␤1 (Fig. 5B). Substitution with the RGEL sequence completely abrogated the ability of FNIII7-10 to bind to ␣v␤8 integrin, underscoring the prerequisite role of the RGD motif in ligand recognition by ␣v␤8 integrin. These results provide further support for a critical role of the Leu residue immediately after the RGD motif in high affinity binding of the ␣v␤8 integrin to its ligands.
␤8 I-like Domain Defines the Binding Specificity of ␣v␤8 Integrin to Latent TGF-␤1-Given that the ␣v␤8 and ␣v␤3 integrins share the same ␣v subunit, the ␤8 subunit should be responsible for latent TGF-␤1 recognition by the ␣v␤8 integrin. To explore the region within the ␤8 subunit that defines the ligand-binding specificity of ␣v␤8 integrin, we focused on the ␤ I-like and hybrid domains of the ␤8 subunit because accumulating evidence suggests that these domains are directly involved in ligand recognition by ␣v␤3 and ␣IIb␤3 integrins (45)(46)(47). Given that ␣v␤3 integrin lacks the ability to bind to latent TGF-␤1, we produced a mutant of the ␣v␤3 integrin, designated ␣v␤3-8BI/HYB, in which the ␤3 I-like and hybrid domains were swapped with those of the ␤8 subunit (Fig. 6A), and we assessed its binding activities toward latent TGF-␤1, fibronectin, vitronectin, and fibrillin-1. Although wild-type ␣v␤3 integrin showed weak binding to latent TGF-␤1, the ␣v␤3-8BI/HYB mutant exhibited strong binding, recapitulating the high affinity binding of ␣v␤8 integrin to latent TGF-␤1 (Fig. 6B). In contrast, the binding activities toward fibronectin, vitronectin, and fibrillin-1 were abrogated in ␣v␤3-8BI/HYB, demonstrating that ligand-binding specificity of the ␣v␤8 integrin was conferred to ␣v␤3 integrin by swapping the ␤ I-like and hybrid domains. Saturation binding assays revealed that the binding affinity of the ␣v␤3-8BI/HYB mutant toward latent TGF-␤1 was similar to that of wild-type ␣v␤8 integrin, yielding an apparent dissociation constant of 3.3 Ϯ 0.5 nM (Fig.  6C and Table 3). These results indicated that the ligand specificity and binding affinity of ␣v␤8 integrin to latent TGF-␤1 is defined by the ␤8 I-like and hybrid domains.
To identify further the region responsible for latent TGF-␤1 binding by ␣v␤8 integrin, we constructed another swap mutant of ␣v␤3 integrin, ␣v␤3-8〉⌱, in which only the ␤3 I-like domain was swapped with the ␤8 I-like domain (Fig. 6A). The binding specificity toward latent TGF-␤1 was retained by the ␣v␤3-8BI mutant, although its binding activity was lower than for the ␣v␤8 integrin and ␣v␤3-8BI/HYB mutants (Fig. 6B). The  apparent dissociation constant of the ␣v␤3-8BI mutant for latent TGF-␤1 was 18 Ϯ 1 nM (Table 3), demonstrating that the binding affinity toward latent TGF-␤1 was ϳ6-fold lower than those of ␣v␤8 integrin and ␣v␤3-8BI/HYB. These results indicated that the ␤8 I-like domain primarily defines the ligand specificity of the ␣v␤8 integrin, although the ␤8 hybrid domain potentiates the binding affinity toward latent TGF-␤1. Consistent with these results, ␣v␤8 -3BI, a mutant of the ␣v␤8 integrin whose ␤ I-like domain was swapped with the ␤3 I-like domain, lost the ability to bind to latent TGF-␤1 but bound avidly to fibronectin, vitronectin, and fibrillin-1 (Fig. 6B), recapitulating the ligand specificity of wild-type ␣v␤3 integrin. The binding of ␣v␤3-8BI to latent TGF-␤1 was strongly inhibited by the RGDL peptide with an IC 50 of ϳ0.07 M, although substitution of the Leu residue with Ala resulted in an ϳ60-fold decrease in the inhibitory potency of the peptide (Fig. 6D and Table 2). These results indicated that the ␤ I-like domains primarily determine the binding specificity of ␣v-containing integrins  Table 3. D, inhibition of ␣v␤3-8BI binding to latent TGF-␤1 by synthetic peptides. The assays were performed as described in the Fig. 4 legend. The results represent the means of three independent determinations. Closed circles, RGDL (9-mer containing both RGD motif and Leu residue); open circles, RGDA (9-mer with the Leu 3 Ala substitution); closed triangles, RGEL (9-mer with RGD 3 RGE substitution); open triangles, RGEA (9-mer with RGDL 3 RGEA double substitution). and that the ␤8 I-like domain recognizes the Leu residue following the RGD motif and is necessary for the high affinity binding of ␣v␤8 integrin to latent TGF-␤1.

Role of the Disulfide-linked Loop in Latent TGF-␤1 Recognition by ␣v
Integrins-To identify the integrin ␤8 subunit region involved in latent TGF-␤1 binding further, we focused on a small disulfide-linked loop consisting of 6 -8 amino acid residues that resides on the top of the ␤ I-like domain, designated as the "disulfide-linked loop (DLL)" (Fig. 7A). The DLL was reported to determine the ligand-binding specificities of ␣v␤1 and ␣v␤3 integrins (48). To address the role of the ␤8 subunit DLL (␤8-DLL) in latent TGF-␤1 recognition by the ␣v␤8 integrin, we produced a swap mutant of the ␣v␤3 integrin termed ␣v␤3-8DLL, in which the ␤3-DLL was swapped with the corresponding residues of the ␤8 subunit. The ␣v␤3-8DLL mutant exhibited binding to latent TGF-␤1 while retaining the ability of the ␣v␤3 integrin to bind to fibronectin, vitronectin, and fibrillin-1 (Fig. 7B). A saturation binding assay demonstrated that the binding affinity of ␣v␤3-8DLL to latent TGF-␤1 was significantly lower than that of wild-type ␣v␤8 integrin, although the affinity toward fibronectin and vitronec-tin was only slightly affected (Fig. 7C). These results indicated that ␤8-DLL confers latent TGF-␤1 binding activity to the ␣v␤3 integrin without compromising the ability of the ␣v␤3 integrin to bind to its cognate ligands. We also produced another swap mutant, ␣v␤8 -3DLL, in which the ␤8-DLL was swapped with the ␤3-DLL. Unexpectedly, the overall binding specificity of the ␣v␤8 integrin remained unaltered after DLL swapping; ␣v␤8 -3DLL selectively bound to latent TGF-␤1 without acquiring the ability to bind to fibronectin, vitronectin, and fibrillin-1 (Fig. 7B). Saturation binding assays indicated only a small decrease in the binding affinity toward latent TGF-␤1 with ␣v␤8 -3DLL (Fig. 7C). Consistent with these results, the binding affinities of ␣v␤8 -3DLL toward fibronectin and vitronectin were also unchanged after DLL swapping. These results indicated that DLL is not the primary determinant for ligand specificity of the ␣v␤8 integrin, although it conferred latent TGF-␤1 binding activity to the ␣v␤3 integrin upon DLL swapping. Therefore, it is puzzling why ␤8-DLL, which is dispensable for defining the ligand specificity of ␣v␤8 integrin, confers latent TGF-␤1 binding activity to ␣v␤3 integrin with-  Table 3.
out compromising its ability to bind to fibronectin, vitronectin, and fibrillin-1.
␤8-DLL Is Not Involved in the Specific Recognition of Leu-218 Immediately after the RGD Motif-To address the apparent discrepancy of the role of ␤8-DLL in defining the ligand specificity of the ␣v␤8 integrin, we examined whether ␤8-DLL is involved in the recognition of Leu-218 required for high affinity binding of latent TGF-␤1 to ␣v␤8 integrin. We examined the inhibitory effects of the RGDL and RGDA peptides on the binding of latent TGF-␤1 to ␣v␤8 -3DLL and ␣v␤3-8DLL, both of which were capable of binding to latent TGF-␤1 (Fig. 8). The binding of ␣v␤8 -3DLL to latent TGF-␤1 was strongly inhibited by the RGDL peptide with an IC 50 of 0.06 M, although substitution of the Leu residue with Ala resulted in an ϳ80-fold decrease in the inhibitory potency of the peptide (Fig. 8A and Table 2). This indicated that Leu-218 is recognized by ␣v␤8 -3DLL to sustain its specific binding to latent TGF-␤1. In contrast, RGDL and RGDA peptides equally inhibited the binding of ␣v␤3-8DLL to latent TGF-␤1 with an IC 50 of 0.17 and 0.13 M, respectively, irrespective of the presence or absence of a Leu residue immediately after the RGD motif ( Fig. 8B and Table 2). These results demonstrated that ␤8-DLL in ␣v␤3-8DLL does not recognize the Leu residue immediately after the RGD motif, consistent with the conclusion that ␤8-DLL is not involved in the Leu-218dependent high affinity binding of ␣v␤8 integrin to latent TGF-␤1. The RGEL peptide did not inhibit the interaction of ␣v␤8 -3DLL and ␣v␤3-8DLL mutants with latent TGF-␤1, highlighting the critical importance of the RGD motif in latent TGF-␤1 recognition by the ␣v␤8 integrin. Taken together, these results indicate that ␤8-DLL is dispensable for the specific binding of the ␣v␤8 integrin to latent TGF-␤1, although the ␤8 I-like domain is necessary and sufficient for recognition of the Leu-218 residue by the ␣v␤8 integrin.

Discussion
This study demonstrated that ␣v␤8 integrin binds strongly and preferentially to latent TGF-␤1 with an affinity of ϳ100fold higher than for other RGD proteins, including fibronectin, vitronectin, and fibrillin-1. The high affinity interaction of the ␣v␤8 integrin with latent TGF-␤1 is determined by the Leu-218 residue immediately following the RGD motif within the latency-associated peptide of latent TGF-␤1. Substitution of the Leu-218 residue with Ala resulted in a dramatic reduction of the latent TGF-␤1 binding affinity of ␣v␤8 integrin, even though the RGD motif remained unperturbed. Accumulating evidence indicates that the binding affinities of RGD-containing ligands toward integrins are potentiated by sequences residing outside the RGD motif. The occurrence of such an auxiliary binding sequence was originally proposed in the central cellbinding domain of fibronectin, where a set of residues within the 9th type III repeat (designated synergy site) potentiates the ␣5␤1 integrin-mediated cell adhesive activity of the RGD motif within the 10th type III repeat, although electron microscopic analyses failed to confirm a direct interaction of the 9th type III module harboring the synergy site with ␣5␤1 integrin (49). Nephronectin also contains an auxiliary sequence LFEIFEIER required for the high affinity binding of nephronectin to the ␣8␤1 integrin, which functions in concert with an RGD motif (24). DiCara et al. (25) reported that the interaction of latent TGF-␤1 with ␣v␤6 integrin is determined by an LXXI sequence immediately C-terminal to the RGD motif, in which two hydrophobic residues Leu-218 and Ile-221 are required for the high affinity binding of latent TGF-␤1 to the ␣v␤6 integrin. Our results show that Leu-218 is critically required for latent TGF-␤1 recognition by ␣v␤8 integrin, but Ile-221 is dispensable for recognition, because the substitution of Ile-221 with Ala did not affect the high affinity binding of latent TGF-␤1 to ␣v␤8 integrin. The substitution of Ile-221 with Pro had no impact on the latent TGF-␤1 recognition by ␣v␤8 integrin either (data not shown). In support of this conclusion, a 9-mer peptide containing an RGDL sequence strongly inhibited the interaction of latent TGF-␤1 with ␣v␤8 integrin, whereas a 9-mer peptide with an RGDA sequence had an ϳ60-fold lower inhibitory effect on the interaction. The critical role of Leu-218 in latent TGF-␤1 recognition by ␣v␤8 integrin was further corroborated by site-directed mutagenesis of fibronectin, in which the high affinity binding toward ␣v␤8 integrin was conferred on fibronectin by substitution of its RGDS motif with an RGDL sequence. The Leu-218 residue immediately following the RGD motif is conserved in latent TGF-␤1 among vertebrates, under- scoring the importance of Leu-218 as an auxiliary recognition residue defining the high affinity interaction of latent TGF-␤1 with ␣v␤8 integrin.
The high affinity binding of ␣v␤8 integrin with latent TGF-␤1 was conferred upon ␣v␤3 integrin (which normally exhibits marginal latent TGF-␤1 binding) by swapping the I-like domain of the ␤3 subunit with the ␤8 subunit. This demonstrated the ␤ I-like domain primarily defines the high affinity interaction of ␣v␤8 integrin with latent TGF-␤1. Recently, Dong et al. (26) reported the crystal structure of ␣v␤6 integrin complexed with an 11-mer peptide HGRGDLGRLKK derived from latent TGF-␤3. They demonstrated that the ␤6 I-like domain interacted with the LGRLK sequence immediately following the RGD motif, which forms an amphipathic ␣-helix and confers high affinity binding of ␣v␤6 integrin to latent TGF-␤3. The binding to latent TGF-␤1 by ␣v␤3-8BI, in which the I-like domain of ␤3 subunit was swapped with the ␤8 subunit, was strongly inhibited by the RGDL peptide but less so by the RGDA peptide, indicating the ␤ I-like domain of the ␤8 subunit might harbor the region(s) responsible for Leu-218 recognition. To explore this possibility further, we predicted the structures of the ␤ I-like domain of ␣v␤8 and ␣v␤3 integrins complexed with an RGDL-containing peptide using Swiss-Model with the crystal structure of ␣v␤6 integrin complexed with the latent TGF-␤3 peptide (26) and that of the ␣v␤3 integrin (50) as templates (Fig. 9). The ␤8 I-like domain was predicted to assume a structure of open conformation that allows the side chain of the Leu residue to fit into the ␤8 I-like domain, thus enabling the high affinity binding of latent TGF-␤1 with ␣v␤8 integrin. In contrast, the ␤3 I-like domain was predicted to assume a closed structure, failing to accommodate the side chain of the Leu residue, and therefore resulting in a low affinity binding to latent TGF-␤1.
The DLL of ␤ I-like domain has been shown to modulate the ligand specificity of ␣v-containing integrins (48). Although cells expressing ␣v␤1 integrin did not bind to the substrate coated with von Willebrand factor and fibrinogen, which are high affinity ligands for the ␣v␤3 integrin, cells expressing the mutant ␣v␤1 integrin, where ␤1-DLL was swapped with ␤3-DLL, bound to these ligands, demonstrating that the ligandbinding specificity of ␣v␤3 integrin can be conferred upon ␣v␤1 integrin by swapping the DLL. In contrast to this report, our results showed that ␤3-DLL had little involvement in the recognition of fibronectin, vitronectin, and fibrillin-1 by ␣v␤3 integrin, because ligand-binding specificities of ␣v␤3 integrin toward these proteins remained unchanged after DLL swapping. Furthermore, the ability of ␣v␤8 integrin to bind to latent TGF-␤1 was retained after swapping ␤8-DLL with ␤3-DLL, indicating that ␤8-DLL is not the primary determinant of the specific binding of the ␣v␤8 integrin to latent TGF-␤1. Because Leu-218 is still recognized by ␣v␤8 -3DLL, which retains specific binding to latent TGF-␤1, it seems likely that the Leu-218 binding pocket is maintained in the ␤8 I-like domain even after replacement of ␤8-DLL with ␤3-DLL. However, why ␤8-DLL confers latent TGF-␤1 binding activity to ␣v␤3 integrin, despite ␤8-DLL being dispensable for the recognition of latent TGF-␤1 by ␣v␤8 integrin remains to be elucidated. The replacement of ␤3-DLL with ␤8-DLL might alter the conformation of the ␤3 I-like domain so that the resulting ␤ I-like domain adopts a structure reminiscent of the open conformation that is competent for latent TGF-␤1 binding, whereas the FIGURE 9. Predicted structures of the head region of ␣v␤8 integrin with an 11-mer peptide containing RGD motif and Leu residue. A, ribbon models of ␤ I-like domain of ␣v␤8 integrin (left) and ␣v␤3 integrin (right) with 11-mer peptide HGRGDLGRLKK derived from latent TGF-␤3 were created using the crystal structure of ␣v␤6 integrin with this peptide (Protein Data Bank code 4UM9) as the template. The models were predicted with the Swiss-Model and fine-tuned by energy minimization with UCSF Chimera. ␤ I-like domains of ␤8 and ␤3 are shown in pink and cyan, respectively. 11-mer peptides are colored in violet red, and the Leu residue immediately following the RGD motif is shown with side chain in yellow. B, molecular surfaces of the ␤ I-like domain of integrin ␤8 and ␤3 subunits with the 11-mer peptide containing RGD motif and Leu residue were generated with the Chimera. The ␤8 I-like domain is predicted to assume a structure of open conformation that allows the side chain of the Leu residue to fit into the ␤8 I-like domain, although the ␤3 I-like domain is predicted to assume a closed structure, failing to accommodate the side chain of the Leu residue.
normal ␤3 I-like domain assumes a closely packed structure that constrains accommodation of Leu-218.
Several lines of evidence indicate that the ␤ hybrid domain acts as a mechanical device that regulates the affinity state of integrins by rearrangement at the interface between the ␤ ⌱-like domain and the ␤-propeller domain of the ␣ subunit, which together form the ligand-binding site of integrins. Thus, binding of ␣5␤1 integrin to its ligand causes a dramatic change in the position of the ␤ hybrid domain relative to the ␤ ⌱-like domain to induce an open conformation of the integrin headpiece (49). Xiao et al. (46) demonstrated that the hybrid domain of the ␤3 subunit extends laterally away from the ligand-binding site to stabilize the open headpiece conformation. Consistent with the role of the ␤ hybrid domain in the affinity state modulation of integrins, the limiting swing-out of the ␤3 hybrid domain prevents ␣IIb␤3 integrin from binding to its high affinity ligand fibrinogen, indicating that the swing-out of the ␤3 hybrid domain is required for activation of the ␣IIb␤3 integrin (47). Our results showed that ␣v␤3-8BI/HYB and ␣v␤8 integrin had a similar binding affinity to latent TGF-␤1, although ␣v␤3-8BI had a lower latent TGF-␤1 binding affinity. These results are consistent with previous studies and support the consensus that the ␤ hybrid domain regulates, in collaboration with the ␤ I-like domain, the conformational change of the integrin headpiece from a low to a high affinity state for ligand binding.
There is compelling evidence that the ␣v␤8 integrin plays a dominant role in activating latent TGF-␤1 in the developing brain. Mice deficient in the expression of the integrin ␤8 subunit exhibit variable embryonic lethality because of vasculogenesis failure and severe brain hemorrhage (12), as is the case with mice deficient for integrin ␣v subunit expression (11). Notably, when mice with a TGF-␤1 gene knock-in mutation that causes an RGE substitution of the RGD motif are crossed with TGF-␤3-deficient mice they die as a result of severe brain hemorrhage (51), recapitulating the phenotype of integrin ␤8-deficent mice. Given the similarities in phenotypes between ␣v␤8 integrin-deficient mice and TGF-␤1(RGE)/TGF-␤3 double mutant mice, ␣v␤8 integrin has been proposed to act as an "angiogenic switch" in the brain through TGF-␤ activation (52,53). Consistent with the role of ␣v␤8 integrin in TGF-␤1 activation, Yamazaki et al. (54) reported that ␤8 integrin was specifically expressed by Schwann cells and involved in latent TGF-␤ activation in the bone marrow, thereby regulating hematopoietic stem cell hibernation. Furthermore, the activation of latent TGF-␤1 by ␣v␤8 integrin may occur via mechanisms distinct from those for TGF-␤1 activation by ␣v␤6 integrin, which requires the ␤6 subunit cytoplasmic domain and a functional actin cytoskeleton. However, TGF-␤ activation by ␣v␤8 integrin is considered independent of the ␤8 subunit cytoplasmic domain and actin cytoskeleton (18,21). Indeed, the amino acid sequence of the cytoplasmic domain of integrin ␤8 subunit differs significantly from those of other ␤ subunits (13), making it unlikely that the ␤8 cytoplasmic domain interacts with the actin cytoskeleton and generates traction force for cell spreading and migration as well as the conformational activation of latent TGF-␤s (15). Given its restricted binding specificity and prominent expression in astrocytes surrounding cerebral blood vessels (52), ␣v␤8 integrin may function as an anchor specialized for latent TGF-␤1, thereby ensuring the localized action of active-TGF-␤1 at the neurovascular unit where astrocytes cross-talk with endothelial cells to facilitate brain vascular development.
In conclusion, we have shown that the ␣v␤8 integrin binds strongly and preferentially to latent TGF-␤1. Its high affinity binding is primarily defined by the Leu-218 residue located immediately after the RGD motif within the latency-associated peptide of latent TGF-␤1. Although it remains to be determined how the ␤8 I-like domain recognizes the Leu-218 residue at the atomic level by x-ray crystallography of the ␣v␤8 integrin complexed with a latent TGF-␤1, our comprehensive study on the binding activities of ␣v␤8 integrin and its mutant proteins toward a wide range of RGD proteins has identified molecular mechanisms involved in the specific recognition of latent TGF-␤1 by the ␣v␤8 integrin that might help our understanding of its physiological and pathological roles.
Author Contributions-A. O. designed and performed the experiments, analyzed the results shown in all figures, and wrote the paper. K. S. conceived and coordinated the study, analyzed the results, and wrote the paper. Y. S. coordinated the study, performed and analyzed the experiments shown in Fig. 1, and wrote the paper. T. I. performed and analyzed the experiments shown in Fig. 1. T. U. constructed vectors for expression of integrins and performed preliminary experiments exploring their ligand specificity. J. T. provided expression vectors for the integrin ␣v and ␤3 subunits and contributed to interpretation of the results. All authors reviewed the results and approved the final version of the manuscript.