αvβ5 Integrin Sustains Growth of Human Pre-B Cells through an RGD-independent Interaction with a Basic Domain of the CD23 Protein*

CD23 is a type II transmembrane glycoprotein synthesized by hematopoietic cells that has biological activity in both membrane-bound and freely soluble forms, acting via a number of receptors, including integrins. We demonstrate here that soluble CD23 (sCD23) sustains growth of human B cell precursors via an RGD-independent interaction with the αvβ5 integrin. The integrin recognizes a tripeptide motif in a small disulfide-bonded loop at the N terminus of the lectin head region of CD23, centered around Arg172, Lys173, and Cys174 (RKC). This RKC motif is present in all forms of sCD23 with cytokine-like activity, and cytokine activity is independent of the lectin head, an “inverse RGD” motif, and the CD21 and IgE binding sites. RKC-containing peptides derived from this region of CD23 bind αvβ5 and are biologically active. The binding and activity of these peptides is unaffected by inclusion of a short peptide containing the classic RGD sequence recognized by integrins, and, in far-Western analyses, RKC-containing peptides bind to the β subunit of the αvβ5 integrin. The interaction between αvβ5 and sCD23 indicates that integrins deliver to cells important signals initiated by soluble ligands without the requirement for interactions with RGD motifs in their common ligands. This mode of integrin signaling may not be restricted to αvβ5.

CD23 is a type II transmembrane glycoprotein synthesized by hematopoietic cells that has biological activity in both membrane-bound and freely soluble forms, acting via a number of receptors, including integrins. We demonstrate here that soluble CD23 (sCD23) sustains growth of human B cell precursors via an RGD-independent interaction with the ␣v␤5 integrin. The integrin recognizes a tripeptide motif in a small disulfidebonded loop at the N terminus of the lectin head region of CD23, centered around Arg 172 , Lys 173 , and Cys 174 (RKC). This RKC motif is present in all forms of sCD23 with cytokine-like activity, and cytokine activity is independent of the lectin head, an "inverse RGD" motif, and the CD21 and IgE binding sites. RKCcontaining peptides derived from this region of CD23 bind ␣v␤5 and are biologically active. The binding and activity of these peptides is unaffected by inclusion of a short peptide containing the classic RGD sequence recognized by integrins, and, in far-Western analyses, RKC-containing peptides bind to the ␤ subunit of the ␣v␤5 integrin. The interaction between ␣v␤5 and sCD23 indicates that integrins deliver to cells important signals initiated by soluble ligands without the requirement for interactions with RGD motifs in their common ligands. This mode of integrin signaling may not be restricted to ␣v␤5.
The structures of the derCD23 protein, a naturally occurring sCD23 fragment generated by action of the derp1 protease from Dermatophagoides pteronyssinus, and of a 25-kDa sCD23 (residues 150 -321), have recently been solved by heteronuclear nuclear magnetic resonance spectroscopy (14) and x-ray crystallography (15), respectively. Although there are pronounced differences between the structures derived by the two methods, both show a generally consistent overall structure for the C-type lectin head domain comprising eight ␤ sheets and two ␣ helices (14,15). The NMR structure reveals a striking distribution of acidic and basic residues on opposites faces of the lectin head domain and demonstrates unequivocally that the interaction surfaces for IgE and CD21 binding are spatially distinct (14); the data also support unequivocally earlier studies that suggested that these binding sites and the structures responsible for expression of cytokine-like activities are distinct (16). The crystal structure confirms that CD23 contains only a single calcium binding site in the lectin head and that conformational changes in the CD23 structure accompany calcium coordination in this site (15). The binding site for integrins was not identified in either study, and, as reported here, this resides at the N-terminal region of the C-type lectin domain.
The ␣v␤5 integrin is a member of the vitronectin receptor (VnR) family. VnRs are heterodimers of the ␣v integrin chain in association with any one of five ␤ subunits (17) and have well documented roles in cell attachment, spreading, and migration, rescue from apoptosis, and angiogenesis (18 -21). VnRs bind to the extracellular matrix proteins vitronectin (Vn) and fibronectin (Fn) by recognition of an Arg-Lys-Asp (RGD) motif in target proteins (22); both the ␣ and ␤ integrin subunits contribute to ligand binding (23,24). Integrins are known to bind to proteins via other peptide motifs (22), usually resulting in enhancement of the outcome of occupation of the RGD site. In particular, ␣v␤5 binds to a basic domain on the HIV Tat protein (25), although no functional consequence for this interaction has been established. However, the greater affinity of ␣v␤5 for the Tat basic domain compared with an equivalent Vn domain (25) suggests that ligands other than Vn could interact with the ␣v␤5 basic domain binding site and could have distinct signaling functions.
This report demonstrates that the ␣v␤5 integrin interacts with sCD23 using a site distinct from the RGD binding pocket and that recognition of a basic motif on CD23 that is distinct from its interaction sites for IgE, CD21, or itself, enhances survival of human pre-B cells.
Cell Culture, siRNA Procedures, and Flow Cytometry-The SMS-SB cell line was derived from a female patient presenting with acute lymphoblastic leukemia (26); Nalm-6 and Blin-1 cell lines were from laboratory stocks. Cell lines were maintained in RPMI 1640 medium supplemented with 10% (v/v) heat-inactivated fetal calf serum, 2 mM fresh glutamine, and penicillin and streptomycin, at 37°C in a 5% CO 2 in air in a humidified atmosphere. Human telomerized fibroblasts were cultured as described previously (27). Cytokines, obtained from R&D Laboratories, were used at 5-10 ng/ml, but had no effect over a wide dose-response range. SMS-SB cells were also propagated in protein-free hybridoma medium-II (PFHM, Invitrogen), at 2.5-5 ϫ 10 5 cells/ml ("normal cell density"). In stimulation experiments, SMS-SB cells were cultured at 2500 cells/100-l culture (low cell density (LCD)) a seeding density at which the cells are prone to apoptosis (10). NALM-6 and Blin-1 cells were washed extensively in PFHM prior to culture at 2500 cells/ 100-l culture. Cultures were propagated in the presence or absence of cytokines, mAbs or peptides, at 37°C for 72 h followed by addition of 0.3 Ci/well tritiated thymidine ([ 3 H]TdR) for 18 h prior to harvest; incorporation was determined by liquid scintillation spectrometry.
Integrin ␣v knockdown was achieved using siRNA targeting the sequence, 5Ј-AGCAACTTTATTATAGATTTA-3Ј (Hs_ITGAV_1_HP siRNA), whereas the control non-silencing siRNA (cat. no. 1022076) targeted the sequence, 5Ј-AATTCT-CCGAACGTGTCACGT-3Ј. 89 l of 2 M siRNA was diluted in 2.9 ml of Dulbecco's modified Eagle's medium (serum-free) then 44.5 l of HiPerFect (Qiagen) was added prior to 10-min incubation at room temperature. This solution was added to a 10-cm plate that had been seeded 2 h previously with 6 ϫ 10 5 cells in 12 ml of Dulbecco's modified Eagle's medium containing 20% (v/v) fetal calf serum. The medium was replaced 48 h later, and further control and silencing siRNA molecules were added as above. Cells were harvested after a further 24 h and analyzed for binding of CD23, peptides, and anti-␣v integrin mAbs.
For flow cytometry, 5 ϫ 10 5 cells were stained with either FITC-conjugated or unlabeled primary mAb for 30 -60 min; unlabeled primary antibody was visualized using a secondary FITC-conjugated anti-mouse IgG or, in the case of biotinylated anti-␣v␤5, using streptavidin-Quantum Red. Cells were analyzed on a FACScan flow cytometer, using CellQuest software.
Affinity Isolation and Analysis of Cellular Proteins-10 8 SMS-SB cells were harvested, washed twice in ice-cold phosphate-buffered saline, suspended in 1.5 ml of ice-cold OGP extraction buffer (1% (w/v) OGP in 50 mM HEPES/KOH, pH 7.4, 5 mM CaCl 2 , 140 mM NaCl, 1 mM phenylmethylsulfonyl fluoride, 1 mM aprotinin, 1 mM leupeptin), and lysed with 40 strokes of a chilled glass homogenizer. The homogenate was centrifuged at 1,000 ϫ g for 10 min at 4°C, and the resulting supernatant further was centrifuged at 35,000 ϫ g for 45 min at 4°C. Cellular extracts were added to bovine serum albumin-Affi-Gel pre-equilibrated with OGP extraction buffer and incubated at 4°C for 6 h. The matrix was pelleted, and unbound proteins were recovered, added to pre-equilibrated sCD23-Affi-Gel, and incubated at 4°C overnight. The sCD23-Affi-Gel was pelleted, and the unbound fraction was retained. Both matrices were exhaustively washed, and specifically bound material eluted by boiling in sample buffer and subjected to SDS-PAGE under reducing conditions on a 10% (w/v) acrylamide gel. Eluates were transferred to nitrocellulose membranes and probed with anti-VnR-component antibodies, followed by an HRP-labeled secondary antibody and ECL. For far-Western analysis, purified integrin proteins were subjected to SDS-PAGE under non-reducing conditions and transferred to nitrocellulose membranes. The blots were probed with biotinylated CD23-derived peptides, in the presence or absence of competing peptides, and

CD23-␣v␤5 Integrin Interaction Is RGD-independent
binding was visualized using HRP-conjugated streptavidin and ECL. Areas of the peaks visualized by ECL/autoradiography were determined using ImageJ software. Lanes containing ␤5 bands were scanned, and identical peak widths were selected for calculation of areas under the selected ␤5 peaks.
Biotinylated peptides were captured on 96-well streptavidin-coated enzyme-linked immunosorbent assay trays and blocked with 1% (w/v) casein. Purified ␣v␤5 integrin (0.2 g) was added to each well, and binding was quantitated by use of P1F6 mAb and HRP-anti-mouse IgG followed by tetramethylbenzidine as substrate. Binding of biotinylated peptides to cells was visualized with fluorochrome-conjugated streptavidin and flow cytometry; mean fluorescence intensity data were derived using the CellQuest program. In cell stimulation experiments, peptides were used in the 10 nM to 6 M concentration range, and appropriate solvent vehicle controls (e.g. acetonitrile and Me 2 SO) were always performed. Biotinylated and non-biotinylated peptides provoked identical responses in functional assays.
Interactions between CD23-derived peptides and ␣v␤5 integrin were assessed at 25°C using a BIAcore 2000 instrument (Biacore AB). The ␣v␤5 integrin was immobilized at ϳ4000 response units on a C1 chip, and an underivatized reference cell was employed as a control surface. CD23-derived peptides in HBS-EP buffer (0.01 M HEPES-KOH, pH 7.4, 0.15 M NaCl, 3 mM EDTA, 0.005% (v/v) Surfactant P20) were injected over the test and control chip surfaces at a flow rate of 40 l⅐min Ϫ1 with a 3-min association phase followed by a 6-min dissociation phase. In all experiments, peptides bound exclusively to the test cell, there was no nonspecific binding to the control reference cell, and regeneration of a stable baseline (using 0.2 M glycine-HCl, pH 2.5, where necessary) was readily achieved. Data were standardized according to standard double referencing data subtraction methods (28), using control and blank injections, prior to kinetic analysis using BIAevaluation 3.0.2 software.

RESULTS
The ␣v␤5 Integrin Is a CD23-binding Protein-Recombinant sCD23 rescues cultures of an acute lymphoblastic leukemia- Apoptosis was determined by staining harvested cells with propidium iodide and Hoechst 33342 for 1 min prior to analysis on a Coulter Elite flow cytometer as previously described (10). Normal cell density cultures (2.5-5 ϫ 10 5 cells/ml) served as a control. B, extracts of SMS-SB cells in OGP buffer were passed over a bovine serum albumin-Affi-Gel column, and the flow-through was collected and subsequently incubated on CD23-Affi-Gel; aliquots of flow-through and eluted fractions from both columns were electrophoresed under reducing conditions. Proteins were transferred to nitrocellulose and probed with anti-␣v (panel i) or anti-␤5 (panel ii) anti-peptide antibodies, and binding was visualized with protein A-HRP and ECL. C, SMS-SB cells were stained (solid lines) with mAbs specific for ␣v␤5, CD47, or ␣v␤3 and visualized with the appropriate secondary reagent; isotype control staining is shown in the shaded area. derived human pre-B cell-like cell line, SMS-SB (26), from low cell density-induced apoptosis in a dose-dependent manner (10) (Fig. 1A, panel i); sCD23 is the only cytokine tested that rescues SMS-SB cells from apoptosis ( Fig. 1A, panel ii). SMS-SB cells do not express CD23 itself or any of its known receptors (10) (Fig. 1C, and data not shown), and this suggested that the interaction of sCD23 with acute lymphoblastic leukemia-derived pre-B cells was mediated via a hitherto undefined receptor. CD23-Affi-Gel affinity chromatography of lysates of [ 35 S]methionine-labeled SMS-SB cells enriched two proteins of ϳ120 and 80 kDa, molecular masses that are consistent with those of mature ␣v and ␤5 integrin proteins, respectively (data not shown). The interpretation that these two species were indeed ␣v and ␤5 was confirmed by performing CD23-Affi-Gel affinity chromatography on lysates of unlabeled cell extracts followed by Western blotting. Both ␣v and ␤5 proteins were detected in the eluate from the CD23 matrix but not in the eluate from the bovine serum albumin column (Fig. 1B). The polyclonal anti-␣v antiserum used in this experiment (Fig. 1B, panel i) binds to a C-terminal epitope on ␣v that is located on the 25-kDa fragment generated during biosynthetic maturation of ␣v (29). No ␤3 protein was detected in the eluate from the CD23-Affi-Gel column or in whole cell lysates (data not shown). SMS-SB cells stain with both the P1F6 mAb that recognizes the assembled ␣v␤5 heterodimer (30), and anti-CD47 VnR-associated protein mAbs, but do not stain with the ␣v␤3specific LM609 mAb (30) (Fig. 1C). Reverser transcription-PCR analysis of SMS-SB RNA yielded amplicons of the expected sizes for CD47, CD51/␣v, ␤1 and ␤5, but no correctly sized PCR products were detected for ␤3, ␤6, or ␤8 coding sequences (data not shown).
The demonstration that SMS-SB cells express ␣v␤5 and that it interacts with CD23 suggests that this integrin is linked to pre-B cell growth and survival. To test this proposal, SMS-SB cells were incubated with ␣v␤5-specific monoclonal antibodies (mAb) or with Vn and Fn, the normal ligands for ␣v␤5. Incubation with the ␣v-specific mAb AMF7 stimulated proliferation of SMS-SB cells in response to increasing concentrations of mAb, whereas an isotype-matched control mAb was without effect ( Fig. 2A). Importantly, a second pre-B cell line, NALM-6, also showed enhanced proliferation when stimulated by the AMF7 mAb (Fig. 2B). The 15F11 mAb that recognizes fully assembled ␣v␤5 heterodimers but does not impede ␣v␤5 binding to Vn (31), stimulated SMS-SB proliferation (Fig. 2C). However, the P1F6 mAb, which recognizes a distinct epitope on assembled ␣v␤5 heterodimers and does inhibit ␣v␤5 binding to Vn (30), failed to stimulate proliferation (Fig. 2C). These data are reminiscent of those of Hermann and colleagues (13) who demonstrated that an anti-␣v␤3 reagent directed to the RGD binding site could not impede CD23-driven production of cyto- kine release by monocytic cells. Moreover, finding that neither Vn nor Fn stimulated proliferation of SMS-SB cells (Fig. 2D) supports the contention that the binding sites for RGD-containing ligands and sCD23 are distinct. These data show that ␣v␤5 regulates a pro-survival response in pre-B cell lines and that this response seems to be both ligand-selective and potentially independent of the integrin binding site for RGD. The lack of an RGD sequence in sCD23 suggests that the motif on CD23 recognized by the integrin is not an RGD-related sequence (see Fig. 3B; there is an inverse RGD sequence of DGR but this is, of course, stereochemically different to RGD). One precedent for such recognition is the non-RGD-dependent interaction of ␣v␤5 with a basic motif of the HIV Tat protein (25).
␣v␤5 Binds a Basic Motif in CD23-The sequence of 25-kDa human CD23 (Met 151 to Ser 321 ) is shown in Fig. 3A together with a schematic representation of CD23 showing the major functional regions of the protein (Fig. 3B). A library of 83 biotinylated overlapping nonapeptides was constructed based on the 25-kDa sCD23 sequence (Fig. 3A) and probed with purified ␣v␤5 to identify binding sites for the integrin. Purified ␣v␤5 binds specifically and consistently to a series of four adjacent peptides (peptides #9 -#12) encompassing residues Lys 166 -Gly 180 (data not shown), and the sole feature common to these four peptides is a tripeptide of sequence Arg-Lys-Cys (RKC) beginning at Arg 172 in the CD23 sequence. The structural data for sCD23 (14,15) demonstrate that the RKC motif resides in a small disulfide-bonded loop at the N-terminal end of the C-type lectin domain. The interaction between the four CD23-derived RKC-containing peptides and ␣v␤5 was assessed by surface plasmon resonance (SPR) analysis. This analysis demonstrates that the four CD23-derived, RKCcontaining peptides bind to immobilized ␣v␤5 integrin in a rank order of #11 Ͼ #12 Ͼ #9 Ͼ #10, whereas peptide #58, which does not have an RKC motif, failed to bind to the integrin (Fig. 3C). A longer RKC-containing, CD23-derived pentadecapeptide (the "long peptide" or LP), spanning all 15 amino acids encompassed by peptides #9 to #12, displayed greater binding to ␣v␤5 than the RKC nonapeptides (Fig. 3D). The slow dissociation of the LP suggests that, although the RKC motif may be essential for the CD23-␣v inte-  SEPTEMBER 14, 2007 • VOLUME 282 • NUMBER 37 grin interaction, the complete binding site may extend well beyond the small common RKC tripeptide motif. Fig. 3E illustrates a representative sensorgram for the binding of derCD23 to immobilized ␣v␤5 integrin; the affinity of this interaction (6.04 ϫ 10 Ϫ6 M) is of the same order of magnitude as those for binding of derCD23 to CD21 (8.7 ϫ 10 Ϫ7 M) and to the Fc region of human IgE (1.3 ϫ 10 Ϫ6 M), respectively (14). Fig. 3F states the affinities of each of peptides #9 to #12 and long peptide for immobilized ␣v␤5 integrin. LP had the highest affinity, consistent with the sensorgrams shown in Fig.  3 (C and D), whereas the remaining peptides had affinities for the integrin similar to that of derCD23. There was no consistent correlation between affinity of a peptide for ␣v␤5 and its biological activity. A similar trend of data was observed when purified ␣v␤3 integrin was immobilized on the sensor surface (data not shown). The SPR data demonstrate that CD23-derived, RKC-containing peptides bind to ␣v␤5 integrin. Fig. 3G demonstrates that the binding of CD23 to SMS-SB cells was greatly attenuated by inclusion of excess peptide #11 but not by a variant of peptide #11 where the arginine and lysine have been replaced by alanines (peptide #11AA). Excess LP also inhibited binding of CD23 to the cells (data not shown). Fig. 3H shows the reciprocal experiment, where excess CD23 reduced the binding of peptide #11 to background levels. These data indicate that CD23 and peptides derived from it compete for binding to the same sites on SMS-SB cells.

CD23-␣v␤5 Integrin Interaction Is RGD-independent
CD23-derived, RKC-containing Peptides Bind to Cells and Are Biologically Active-The data of Fig. 3 indicate that CD23-derived RKCcontaining peptides bind to ␣v␤5 in vitro. The binding of biotinylated CD23-derived peptides to SMS-SB cells was detected by flow cytometry, and all four RKC-containing peptides bound to SMS-SB cells (Fig. 4, A and B), and to other ␣v␤5 ϩ pre-B cell lines such as NALM-6 (data not shown). CD23-derived peptides immediately adjacent to peptides #9 and #12, namely #8 and #13, or peptide #15, that had similar net charge to the peptides that bound to ␣v␤5, or a randomly chosen peptide, #57, all failed to bind to ␣v␤5 or to cells (Fig. 4B). Peptides #61-63, which have an RKL tripeptide sequence, or peptides #78 -80

CD23-␣v␤5 Integrin Interaction Is RGD-independent
that have the inverse RGD (DGR) sequence, also did not bind to ␣v␤5 or to cells (data not shown). Binding of biotinylated peptide #9 to SMS-SB cells was significantly inhibited by the presence of excess non-biotinylated peptides #10 and #12, but not peptide #8 (Fig. 4C, panels i and ii): a similar result was obtained for NALM-6 cells (data not shown). The LP inhibited binding of the ␣v␤5 heterodimer-specific mAb P1F6 to Blin1 pre-B cells (Fig. 4D), indicating that binding of this peptide to the integrin on the cell surface perturbs the conformation of the integrin such that the integrity of the epitope recognized by the antibody is compromised. Finally, the contribution of ␣v␤5 to binding of peptides and CD23 was investigated by using an RNA interference-based approach. It did not prove possible to deliver siRNA constructs to SMS-SB or other pre-B cell lines to achieve a reduction in integrin RNA or protein expression in these cells (data not shown). As an alternative, siRNA specific for the ␣v subunit (siRNA␣v) was introduced into telomerized human fibroblasts, and the effects on expression of ␣v integrins and binding of CD23 and CD23-derived peptides to the cells was determined. The ␣v subunit was targeted because successful attenuation of ␣v expression will reduce expression of all ␣v-containing integrins due to the requirement for ␣ and ␤ chains to assemble in the endoplasmic reticulum prior to transport to the cell surface. Moreover, because the ␤5 subunit pairs only with ␣v (17), a successful knockdown of ␣v expression will also diminish ␣v␤5 cell surface levels. Fig. 4E shows that exposure of telomerized fibroblasts to siRNA␣v, but not the control siRNA, significantly reduced staining by the AMF7 and P1F6 reagents (Fig. 4E, panels i and ii, respectively), showing both total ␣v and ␣v␤5 expression at the cell surface were reduced; the background level of staining with the isotype control IgG1 was unaffected by either siRNA (Fig. 4E, panel iii). Similarly, binding of biotinylated CD23 (Fig. 4E, panel iv) or peptide #9 (Fig. 4E, panel v) to siRNA␣v-treated cells was reduced by a substantial amount, whereas the binding of a negative control peptide (#41, Fig. 4E, panel vi) was unaffected. These data indicate that loss of ␣v integrin expression in general, and ␣v␤5 in particular, correlates with a dramatic reduction of binding of CD23 and RKC-containing CD23-derived peptides to the cells, confirming that ␣v integrins are the target for CD23 binding. Because the fibroblasts also express ␤1 and ␤2 integrin subunits, the expression of which will be unaffected by siRNA␣v, it is clear that none of these integrins, and in particular the ␤1 integrin subunit, can substitute for ␣v in binding CD23. These data are entirely consistent with the SPR data of Fig. 3, which show that CD23 binds the ␣v␤5 integrin.
The biological activities of CD23-derived peptides were assessed in standard low cell density proliferation assays. Peptides #9 and #12 stimulated thymidine incorporation by SMS-SB cells, demonstrating that CD23-derived peptides that bind ␣v␤5 via their RKC motif can mimic the effect of CD23 (Fig. 5A), but peptides #10 and #11, which bound to SMS-SB cells, did not stimulate proliferation, raising the possibility that these peptides might act as inhibitors of the stimulatory activity of peptides #9 and #12 (Fig. 5A). Accordingly, SMS-SB cells were incubated with an excess of peptides #10 or #11 and then challenged with either peptide #9 or #12. For both peptide #9 and #12, peptides #10 and #11 suppressed proliferation, whereas the irrelevant peptide #57 did not (Fig. 5B, panels i  and ii). These data indicate that peptides #10 and #11 can indeed inhibit the action of stimulatory peptides #9 and #12. However, in striking contrast, a tetrapeptide of sequence RGDS failed to inhibit either binding of CD23-derived RKCcontaining peptides to SMS-SB cells or stimulation of proliferation of the cells by peptides #9 and #12 (Fig. 5C, panels  i and ii). These data demonstrate that the integrin does not use the RGD binding site to capture sCD23.
sCD23 Cytokine Activity Is Independent of the IgE and CD21 Binding Sites-The hypothesis that ␣v␤5 integrin recognizes the dibasic peptide region in CD23 was tested by substituting Arg 172 and Lys 173 in peptides #9 and #12 with alanine (peptides #9AA and #12AA, respectively). Peptide #9AA shows a severely impaired ability to sustain growth of SMS-SB cells compared with wild-type peptide #9, being comparable to that achieved with peptide #58 (Fig. 6A, panel i). Biotinylated peptide #9AA displayed greatly reduced binding to SMS-SB cells (Fig. 6A,  panel ii), and SPR analysis demonstrated clearly that essentially all peptide #9AA binding to immobilized ␣v␤5 integrin was lost when the two basic residues of peptide #9 were substituted with alanine (Fig. 6A, panel iii). Similar data were obtained for peptide #12AA, which induced SMS-SB proliferation at levels identical to those seen with control peptide #58 (Fig. 6B), and which also shows reduced binding to SMS-SB cells (data not shown). Peptide #11 showed the highest level of binding to immobilized ␣v␤5 (see Fig. 3D) and to cells (see Fig. 4B), and the peptide #11AA double-alanine mutant failed to completely bind to immobilized ␣v␤5 as determined by SPR analysis (Fig. 6C). Finally, a CD23-GST fusion protein comprising residues 48 -248 of CD23 (sCD23 48 -248 ), which contains the RKC motif but lacks the inverse RGD motif, the CD21 and IgE interaction surfaces, and all four protein residues that participate in coordination of calcium binding in the lectin head (15), sustained growth of SMS-SB cells in a dose-dependent manner (Fig. 6D). The approximate EC 50 for the sCD23 48 -248 construct, in the 5-10 nM range, is broadly similar to that observed for 25-kDa sCD23 (10) (Fig. 1A, panel i). The LP, comprising the RKC motif flanked by six residues on either side, also promoted growth of SMS-SB cells (Fig. 6E) supporting the conclusion that  SEPTEMBER 14, 2007 • VOLUME 282 • NUMBER 37 the RKC motif is necessary and sufficient for expression of sCD23 cytokine activity in SMS-SB cells. The data are consistent with the interpretation that the cytokine activity of CD23 mediated via the ␣v␤5 integrin in pre-B cells does not require any contribution from the inverse-RGD motif or from the IgE and CD21 binding sites.

CD23-␣v␤5 Integrin Interaction Is RGD-independent
RKC-containing Peptides Derived from CD23 Bind Integrin ␤ Chains-The data of Fig. 6 indicate that the RKC-containing CD23-derived peptides do not bind to the RGD binding site of the ␣v integrins, and this is fully supported by the x-ray crystallographic data for ␣v␤3 in association with an RGD-containing cyclic pentapeptide that suggest that the lysine is too large to be accommodated in the RGD binding site (24). As a first step in seeking the binding site for RKC-containing peptides in the ␣v␤5 integrins, a far-Western approach was adopted in which purified ␣v␤5 and ␣v␤3 integrins were electrophoresed under non-reducing conditions and the separated subunits were transferred to nitrocellulose and probed with biotinylated RKC-containing peptides. Fig. 7A (lanes 1 and 2) illustrates that peptide #9 bound strongly to the ␤5 and ␤3 chain but only very weakly, if at all, to the ␣v subunit; treatment of identical blots with an irrelevant biotinylated peptide (#76), gave no signal (Fig. 7A,  lanes 3 and 4). Fig. 7B (panel i) demonstrates that binding of peptide #9 to the ␤5 subunit is not impeded by the presence of excess, negative control peptide #58 (lane 2), where no reduction of peak areas of the ␤5 bands was evident (Fig. 7B, panel ii); inclusion of excess peptide #10 (lane 3) inhibited binding of biotinylated peptide #9 to the separated ␤5 subunit by ϳ90% (Fig. 7B, panel ii). These data are completely consistent with those of Fig. 5 showing that peptide #10 inhibited peptide #9-driven proliferation of SMS-SB cells. The data of Fig. 7 indicate that CD23-derived, RKC-containing peptides bind to a structure present on the ␤ chains of the ␣v integrins.

DISCUSSION
This study identifies the ␣v␤5 integrin as a CD23 receptor, defines the recognition site for the integrin on the CD23 protein, and demonstrates a new role for CD23 in regulation of human B cell precursor growth. The data demonstrate that survival signaling via the integrin is ligand-selective and is mediated via a structure on the integrin that is distinct from the RGD binding site and recognizes a basic domain on CD23; this second site appears to reside on the ␤ subunit. Importantly, given the protein-free conditions of the proliferation assays and the sizes of the ligands used, the data demonstrate that integrins send important signals to cells without the need for ligation of the RGD binding site. This property may not be limited to ␣v␤5. The data illustrate further the complex roles of CD23 in regulating human B cell function and show clearly that the pleiotropic functions of CD23 are controlled by discrete structural motifs on the protein.
The RKC motif recognized by the ␣v␤5 integrin is located at positions 172-174 of the human CD23 protein and is preserved in all forms of sCD23 that possess cytokine activity (1). Fig. 8A places the RKC motif in the context of its general position in the CD23 protein and demonstrates that the RKC-centered integrin binding site is completely distinct from those used to cap-

CD23-␣v␤5 Integrin Interaction Is RGD-independent
ture either IgE or CD21 (14). It is also separate from the sequence in the stalk region reported to bind to major histocompatibility complex class II molecules (33). The finer detail provided by the NMR structure of derCD23 (14) and crystal structure of sCD23 (15) locates the RKC in a small disulfidebonded loop near the N-terminal end of the lectin domain at a solvent-exposed tight turn between the ␤0 and ␤1 strands, with the side chain of Cys 174 participating in formation of a disulfide bond with Cys 163 (Figs. 3B and 8). The data of Fig. 6, showing loss of binding and activity of peptide #9 when alanine groups replace the basic arginine and lysine residues, are consistent with a role for these solvent-available basic side chains on the corner of the loop in binding to the integrin. The failure of peptide #11AA to competitively inhibit binding of sCD23 to SMS-SB cells further supports this interpretation. Comparing 20 NMR structures for derCD23 with respect to the RKC motif (Fig. 8C), it is clear that Cys 174 occupies relatively few positions, presumably by virtue of disulfide bond formation with Cys 163 , and Lys 173 is also found in relatively few distinct conformations. However, Arg 172 , exposed on the corner of the tight bend, samples a wide range of orientations. The crystal structure of sCD23 and the NMR structure of derCD23 do show striking differences, and in the specific case of the RKC motif, both Arg 172 and Lys 173 are clearly exposed to the solvent in the derCD23 structure (Fig. 8D, gray lines), whereas the sCD23 crystal structure shows Lys 173 occupying a less solvent-available position close to the ␣1 helix and the arginine in a different but still exposed position (Fig. 8D, blue lines, and Fig. 8B, panels ii and iv). Finally, although binding of calcium causes notable conformational changes in the structure of sCD23 as a whole (15), there is relatively little effect on the orientations and relative positions of the RKC residues in the calcium-free and bound states (Fig. 8E). The NMR structure can accommodate a direct binding model for the ␣v␤5-CD23 interaction, based on having both Arg 172 and Lys 173 available for binding, but the crystal structure does not preclude an induced-fit model in which Arg 172 provides an initial point of contact that allows the two protein structures to alter such that a larger interaction surface becomes available. A comparison of the human and murine sequences in the context of the ␣v␤5 binding site defined in this report offers a partial explanation for the apparent lack of cytokine-like activity in murine CD23 mediated via the ␣v integrins. The human CD23 sequence between residues 169 and 181 is NFQRKCYYFGKGT while the equivalent murine sequence is HFQQKCYYFKGS; importantly, the RKC motif that may be the primary determinant of binding and activity is altered to QKC in the murine CD23 sequence. In contrast to the large, mobile, and basic arginine side chain, glutamine may lack both sufficient length and flexibility to enter a binding cleft and/or may lack the charged termini required to bind securely to the target integrin in a conformation capable of signaling to the cell interior. Although Arg 172 alone is unlikely to be the sole determinant of cytokine activity, it is clearly a key residue for binding and, by extension, biological activity.
The data of Fig. 2 demonstrate that different ␣v␤5 ligands (CD23, Vn, and Fn) and mAbs directed to different epitopes of ␣v␤5 itself elicit characteristic responses in pre-B cell lines by acting via distinct, RGD-dependent and RGD-independent, binding sites on ␣v␤5. There are precedents for ligand-selective signaling via ␣v␤3 in monocytes where CD23 promotes proinflammatory cytokine synthesis, while Vn drives cell spreading but no cytokine production (13). The importance of adhesion interactions between pre-B cells or acute lymphoblastic leukemia cells with stromal cells for survival is widely appreciated (e.g. the VLA-4-VCAM-1 interaction (34 -36)), and, in the bone marrow environment where several ␣v ligands are present (i.e. Fn and Vn (37) and CD23 (38)), the availability of ligand-selective responses, mediated via two distinct binding sites on the same integrin, may be important. If engagement of the RGD binding site was linked to cell survival, then Vn and Fn in the bone marrow would sustain growth of precursor cells in a nonselective manner, resulting in failure to eliminate redundant or malignant precursors. However, the ␣v␤5 integrin may allow B cell precursors to adhere to the stromal matrix via Vn in an interaction that is neutral with respect to cell survival. Signals for inhibition of apoptosis and promotion of cell growth would then be delivered via the second, non-RGD binding site on the ␣v␤5 integrin. Whether the ␣v␤5 integrin delivers a growthsustaining signal to cell types other than lymphoid precursors remains to be established. . RKC-containing peptides bind to integrin ␤ chains. A, purified ␣v␤3 and ␣v␤5 integrins (0. 5 g) were electrophoresed under non-reducing conditions, and separated subunits were transferred to nitrocellulose filters and probed with 4.5 g/ml biotinylated peptides #9 (lanes 1 and 2) or #76 (lanes 3 and 4). Peptide binding was visualized with streptavidin-HRP and ECL. Blots of separated ␣v␤5 integrin chains, generated as described above, were probed with biotinylated peptide #9 (lane 1) in the presence of a 2-fold molar excess of peptide #58 (lane 2) or #10 (lane 3) (B, panel i). Peak area analysis was performed using ImageJ software (B, panel ii).
In our system, RGD-containing peptides do not sustain cell growth themselves, and fail to inhibit either peptide #9 or #12 binding to cells or their ability to sustain cell growth (Fig. 5), and far-Western data (Fig. 7) suggest that RKC-containing peptides bind separated ␤3 and ␤5 chains. The x-ray crystallographic model of ␣v␤3 integrin in association with an RGD-containing cyclic pentapeptide fully supports the interpretation that ␣v␤5 binds CD23 using a structure distinct from the RGD binding site. The RGD-peptide arginine is secured by a bidentate salt link with Asp 218 and a second such link with Asp 150 from the ␣v chain (24), whereas the aspartate side chain makes contacts with Tyr 122 , Arg 214 , and Asn 215 from the ␤3 chain, plus a Mn 2ϩ ion (24). The peptide glycine residue resides at the interface between the ␣v and ␤3 domains making multiple hydrophobic interactions, including a dominant contact with the carbonyl oxygen of Arg 216 of ␣v (24). The long side chain of the RKC peptide lysine would collide with this carbonyl moiety and so preclude stable insertion of the RKC motif into the RGD binding site.
Integrins recognize sequences other than RGD (22). Thus, the ␣2␤1 integrin binds a basic tetrapeptide sequence (RKKH) in the snake venom metalloprotease, jararhagin, a potent inhibitor of platelet binding to collagen (39), and the I-domain of the integrin is known to be the binding site. However, it is becoming increasingly apparent that small groups of basic residues, such as described in this report, are often targets for integrin binding, and in many cases recognition of such motifs by integrins is linked to, or cooperates with, binding to the RGD binding site in facilitating adhesion reactions. For example, both binding of the angiogenic factor CCN3 to ␣v␤5 and CCN3-induced migration of fibroblasts is inhibited by RGD-containing peptides even though the protein does not possess an RGD, or related, sequence (40). The C-terminal domain of the closely related CCN1 protein has two distinct regions, each containing a pair of adjacent lysine residues, that bind the ␣M␤2 integrin and support monocyte adhesion (41). CCN3 has equivalent sequences in its C-terminal domain, although whether these are bound by ␣v␤5 remains to be formally proven. The binding of the non-RGD-containing cobra venom cardiotoxin A5 to ␣v␤3 and its ability to inhibit osteoclastogenesis is inhibited by RGD-containing peptides (42). Interestingly, the related A6 toxin binds only weakly to ␣v␤3 and is not cardiotoxic, and this is explained by replacement of Lys 28 and Lys 29 FIGURE 8. Structural contexts of the CD23 RKC motif. Structural data files for the NMR structure of derCD23 (1T8C (14)) and the crystal structures of calcium-free and calcium bound sCD23 (2H2R and 2H2T, respectively (15)) were acquired from the Brookhaven protein data base, and images were generated using the PyMOL software suite (W. L. DeLano (2002) PyMOL, DeLano Scientific, San Carlos, CA). A, the position of the IgE (blue), CD21 (red), and ␣v␤5 (yellow) binding sites on the 1T8C NMR derCD23 structure (14) as a surface model. B, compares the derCD23 and 2H2R Ca 2ϩ -free sCD23 structures (15), with Arg 172 (red) and Lys 173 (blue) highlighted and the positions of the extended side chains shown in panels ii and iv. C, compares as a ribbon diagram twenty individual NMR structures close to the RKC motif (14); the individual orientations of Arg 172 , Lys 173 , and Cys 174 are illustrated as sticks. D, RKC motif in detail, superimposing the derCD23 (1T8C) structure as gray sticks with the Ca 2ϩ -free sCD23 structure (2H2R) as blue sticks. E, orientation of the RKC motif in sCD23 structures with calcium either bound (yellow sticks) or free (blue sticks) (15). SEPTEMBER 14, 2007 • VOLUME 282 • NUMBER 37 of the A5 molecule with valine and alanine, respectively (42). The ␣v␤5 integrin recognizes the HIV Tat protein via a non-RGD motif that is basic in character (25). Studies of binding of ␣IIb␤3 to peptide fragments of fibrinogen demonstrated that the integrin recognizes a cluster of basic residues located between residues 370 and 381 of the ␥C domain that is devoid of an RGD sequence (43). Intriguingly, the structure of the ␥C domain (32) reveals two pairs of basic residues on exposed surfaces of the protein that are crucial for the interaction with the integrin. In particular, substitution of either Lys 380 or Lys 381 of ␥C peptides with alanine reduces peptide binding to ␣IIb␤3, and a double substitution completely abolishes binding (43). The data from other systems are completely consistent with our data showing loss of binding and activity of CD23-derived peptides when Arg 172 and Lys 173 are substituted with alanine.

CD23-␣v␤5 Integrin Interaction Is RGD-independent
A range of integrin-mediated responses may be driven by recognition of basic domains on other extracellular proteins, and such responses may be independent of occupation of the RGD binding site or may show cross-talk with RGD-specific binding events. Indeed, the cellular response initiated following ligation of the basic domain recognition site on ␣v or other integrins may also be dependent on the integrin heterodimer being engaged. Definition of the precise basic domain binding structure on the ␤ chain of the ␣v integrins will help address these issues.