Spontaneous Formation of L-Isoaspartate and Gain of Function in Fibronectin*

Isoaspartate formation in extracellular matrix proteins, by aspartate isomerization or asparagine deamidation, is generally viewed as a degradation reaction occurring in vivo during tissue aging. For instance, non-enzymatic isoaspartate formation at RGD-integrin binding sites causes loss of cell adhesion sites, which in turn can be enzymatically “repaired” to RGD by protein-l-isoAsp-O-methyltransferase. We show here that isoaspartate formation is also a mechanism for extracellular matrix activation. In particular, we show that deamidation of Asn263 at the Asn-Gly-Arg (NGR) site in fibronectin N-terminal region generates an αvβ3-integrin binding site containing the l-isoDGR sequence, which is enzymatically “deactivated” to DGR by protein-l-isoAsp-O-methyltransferase. Furthermore, rapid NGR-to-isoDGR sequence transition in fibronectin fragments generates αvβ3 antagonists (named “isonectins”) that competitively bind RGD binding sites and inhibit endothelial cell adhesion, proliferation, and tumor growth. Time-dependent generation of isoDGR may represent a sort of molecular clock for activating latent integrin binding sites in proteins.

Isoaspartate formation in extracellular matrix proteins, by aspartate isomerization or asparagine deamidation, is generally viewed as a degradation reaction occurring in vivo during tissue aging. For instance, non-enzymatic isoaspartate formation at RGD-integrin binding sites causes loss of cell adhesion sites, which in turn can be enzymatically "repaired" to RGD by protein-L-isoAsp-O-methyltransferase. We show here that isoaspartate formation is also a mechanism for extracellular matrix activation. In particular, we show that deamidation of Asn 263 at the Asn-Gly-Arg (NGR) site in fibronectin N-terminal region generates an ␣ v ␤ 3 -integrin binding site containing the L-isoDGR sequence, which is enzymatically "deactivated" to DGR by protein-L-isoAsp-O-methyltransferase. Furthermore, rapid NGRto-isoDGR sequence transition in fibronectin fragments generates ␣ v ␤ 3 antagonists (named "isonectins") that competitively bind RGD binding sites and inhibit endothelial cell adhesion, proliferation, and tumor growth. Time-dependent generation of isoDGR may represent a sort of molecular clock for activating latent integrin binding sites in proteins.
Fibronectins are adhesive proteins that mediate a variety of cellular interactions with extracellular matrix and play important roles in hemostasis, thrombosis, inflammation, wound repair, angiogenesis, and embryogenesis (1,2). About 20 isoforms of human fibronectin can be generated as a result of alternative splicing of the primary transcript (1,3). Fibronectins are large glycoproteins (ϳ450 kDa) composed of two nearly identical disulfide-bonded subunits present in most body fluids and extracellular matrix of many tissues. Each subunit consists of three types of repeating homologous modules termed FN-I, FN-II, and FN-III repeats. Alternatively spliced modules, called EDA, EDB, and IIICS, can also be present (1,3). Single modules or groups of modules may contain binding sites for different molecules, including sulfated glycosaminoglycans, DNA, gelatin, heparin, and fibrin (1,3,4). Furthermore, fibronectins contain binding sites for about half of the known cell surface inte-grin receptors (5,6). In particular, the FN-III 10 repeat contains an RGD site that can bind ␣ 3 ␤ 1 , ␣ 5 ␤ 1 , ␣ v ␤ 1 , ␣ v ␤ 3 , ␣ v ␤ 5 , ␣ v ␤ 6 , ␣ 8 ␤ 1 , and ␣II b ␤ 3 integrins, while the FN-III 9 repeat contains the so-called "synergy site" PHSRN that cooperates with RGD in the binding of ␣ 5 ␤ 1 and ␣II b ␤ 3 (1,7).
Primary and tertiary structure analysis of human fibronectin showed that this protein contains two GNGRG loops, located in FN-I 5 and FN-I 7 modules, that are conserved in bovine, murine, rat, amphibian, and fish (8). Two additional NGR sites, less conserved, are also present in human FN-II 1 and FN-III 9 (see Fig. 1). Recent experimental work showed that peptides containing the NGR motif can inhibit ␣ 5 ␤ 1 -and ␣ v ␤ 1 -mediated cell adhesion to fibronectin (9).
These notions prompted us to investigate the functional role of NGR in fibronectin. We observed that the NGR sequence of FN-I 5 (residues 263-265) promotes endothelial cell adhesion via an unusual mechanism based on non-enzymatic deamidation of Asn 263 to L-isoAsp, generating isoDGR, a new cell adhesion motif. Furthermore, we show that ␣ v ␤ 3 integrin is an important receptor of isoDGR and that the deamidated FN-I 5 module (named "isonectin-1") regulates endothelial cell adhesion and inhibits tumor growth. Finally, we show that this motif is regulated in a negative manner by protein-L-isoAsp-O-methyltransferase (PIMT). 2 Based on these findings we propose a new "activation/deactivation" model for isoAsp "formation/removal" in fibronectin.
Preparation and Characterization of Recombinant FN-I 4 -5 and FN-I 4 -5 SGS-The cDNA coding for human fibronectin fourthfifth type I repeats (FN-I 4 -5 ; residues 184 -273 of fibronectin) was prepared by reverse transcriptase PCR on MSR-3-mel cells total RNA (13) using the following primers: 5Ј-CTGGATCCGAGAA-GTGTTTTGATCATGCTGCTGGG (forward) and 5Ј-TATAT-TAAGCTTTCAGTGCCTCTCACACTTCC (reverse). A control fragment with NGR replaced with SGS (FN-I 4 -5 SGS), was generated by PCR on FN-I 4 -5 plasmid using the above forward primer and the following reverse primer, 5Ј-TATATTAAGCTTTCAG-TGCCTCTCACACTTCCACTCTCCACTGCCGCTG. Amplified fragments were cloned into a pRSET-A plasmid (Invitrogen), expressed in BL21(DE3)pLysS Escherichia coli cells as soluble proteins (with a His tag at the N terminus), and purified from cell extracts by metal-chelate affinity chromatography.
Preparation and Characterization of Synthetic Peptides-Various peptides (biotinylated and non-biotinylated) were prepared by chemical synthesis using an Applied Biosystem model 433A peptide synthesizer. Amino acids in L-configuration were used except when indicated. A synthetic peptide corresponding to the fifth type I repeat of human fibronectin (FN-I 5 ), residues 230 -274 of mature protein (Swiss-Prot accession number P02751), was also synthesized with an additional N-terminal acetyl-glycine and with Glu 250 in place of Asp 250 . To promote disulfide formation between Cys 258 -Cys 270 and Cys 231 -Cys 260 in FN-I 5 (see Fig. 1), Cys 258 and Cys 270 were protected with S-trityl groups (removable by trifluoroacetic acid after sidechain deprotection and cleavage from the resin), whereas Cys 231 and Cys 260 were protected with S-acetamidomethyl groups (removable by treatment with iodine). After RP-HPLC purification of peptide, the Cys 258 -Cys 270 disulfide bridge was formed by incubating 20 mol of peptide in 300 ml of 8% Me 2 SO at pH 7 (room temperature, overnight). The peptide was then purified by RP-HPLC and lyophilized. To form the second Cys 231 -Cys 260 disulfide bridge, 13.5 mol of peptide was dissolved in 67 ml of 80% v/v acetic acid and mixed with 0.135 ml of 1 N hydrochloric acid. Then, 21 mg of iodine, dissolved in 2 ml of methanol, was added to the peptide solution under stirring. After 90 min, 1 mmol of ascorbic acid was added to quench iodine. The peptide was then purified by RP-HPLC. All peptides were dissolved in sterile water and stored in aliquots at Ϫ20°C. Peptide purity was analyzed by RP-HPLC. Free sulfhydryl groups in all peptide preparations were Ͻ0.1% as checked by titration with Ellman's reagent (Pierce). Peptide identity was checked by MALDI-TOF or electrospray ionization mass spectrometry. The molecular mass of peptides used throughout this work was similar to the expected value. Electrospray ionization mass spectrometry analysis of FN-I 5 showed the presence of an additional component of ϩ144 Da, corresponding to FN-I 5 with unremoved acetamidomethyl groups.
Accelerated Aging (Heat Treatment) of Fibronectin Fragments and Peptides-Fibronectin fragments, peptides, and peptide-TNF conjugates were diluted in 0.1 M ammonium bicarbonate buffer, pH 8.5, incubated for 16 h at 37°C, and stored at Ϫ20°C until analysis. These products are hereinafter referred to as "heat treated." Isoaspartate (isoAsp) Quantification-isoAsp content in proteins and peptides was quantified using the IsoQuant isoaspartate detection kit (Promega). The isoAsp content in fibronectin (freshly isolated from human plasma) and FN-30 kDa fragment was 0.048 and 0.026 pmol/pmol of protein, respectively.
Cell Adhesion Assay and PIMT Treatment-Untreated and heat-treated fibronectin fragments, peptides, and peptide-TNF conjugates were diluted at the desired concentration with 150 mM sodium chloride, 50 mM sodium phosphate, pH 7.3, and added to 96-well polyvinyl chloride microtiter plates (Falcon; BD Biosciences). After overnight incubation at 4°C, the plates were washed, seeded with EA.hy926 cells in DMEM containing 0.1% BSA (40,000 cells/well), and left to incubate for 2-3 h at 37°C, 5% CO 2 . Adherent cells were fixed and stained with crystal violet as described (11). The effect of PIMT on the proadhesive properties of different fragments was investigated as follows. Microtiter plates were coated with various products as described above and washed with 0.9% sodium chloride. Each well was then filled with 45 l of a solution containing 0.02 mM S-adenosyl-L-methionine in 150 mM sodium chloride, 50 mM sodium phosphate buffer, pH 6.8, and 5 l of PIMT solution (from the IsoQuant isoaspartate detection kit; Promega) (final volume 50 l/well) and incubated at 37°C for 16 h. After incu-bation the plates were washed with 0.9% sodium chloride, and cell adhesion assay was performed as described above.
Binding of Peptides and Proteins to Integrins-Human ␣ v ␤ 3 , ␣ 5 ␤ 1 , and ␣ 1 ␤ 1 integrin solutions, 0.5-2 g/ml in phosphatebuffered saline with Ca 2ϩ and Mg 2ϩ (DPBS; Cambrex), were added to 96-well polyvinylchloride microtiter plates (50 l/well) and left to incubate overnight at 4°C. All subsequent steps were carried out at room temperature. The plates were washed with DPBS and further incubated with DPBS containing 3% BSA (200 l/well, 1 h). The plates were washed and filled with CNGRC-TNF or TNF solutions (5 g/ml, 50 l/well in 3% BSA-DPBS) and left to incubate for 2 h. After washing with DPBS, each well was incubated with purified rabbit anti-murine TNF IgGs in 3% BSA-DPBS containing 1% normal goat serum (10 g/ml, 50 l/well, 1 h) followed by a goat anti-rabbit peroxidase conjugate in the same buffer (50 l/well, 1 h). Bound peroxidase was detected by adding o-phenylendiamine chromogenic substrate. Binding of biotinylated peptides (CNGRC-TNF 1-11 , CARAC-TNF 1-11 , and FN-I 5 ) to purified integrins was studied using streptavidin peroxidase conjugate com- plexes. Complexes were prepared by mixing various quantities (0.5-1 g) of heat-treated biotinylated peptides in DPBS containing 3% BSA with 0.03 units of streptavidin peroxidase (binding capacity 1 g of biotin/unit of streptavidin peroxidase) (final volume 15 l). Complexes were diluted in 3% BSA-DPBS (1:500), added to microtiter plates coated with integrins as described above, and incubated for 2 h at room temperature. After washing with DPBS, bound peroxidase was detected by chromogenic reaction as described above. Each assay was carried out in triplicate.
Inhibition of Endothelial Cell Proliferation-HMEC-1 cells (8 ϫ 10 3 ) in complete medium were seeded in 96-well plate (100 l/well). After 2 h of incubation at 37°C, 5% CO 2 , 100 l of peptide solution in complete medium (100 M) was added to each well. Cells were further incubated for 72 h at 37°C, 5% CO 2 . The number of viable cells present within each well was then determined by 3-(4,5-dimethylthiazol-2yl)-2,5-diphenyltetrazolium bromide assay using a calibration curve generated by plating different amounts of cells.
In Vivo Studies-Studies on animal models were approved by the Ethical Committee of the San Raffaele H Scientific Institute and performed according to the prescribed guidelines. C57BL/6N mice (Charles River Laboratories, Calco, Italy) weighing 16 -18 g were challenged with subcutaneous injection in the left flank of 7 ϫ 10 4 RMA living cells; 4 days later, mice were treated daily with 200 g of heat-treated FN-I 5 (100 l) in 0.9% sodium chloride (intraperitoneal). Tumor growth was monitored by measuring tumors with calipers as previously described (18). Animals were sacrificed before tumors reached 1.0 -1.3 cm in diameter. Tumor sizes are shown as mean Ϯ S.E. (five animals/group).

Accelerated Aging of Fibronectin
Fragments Generates NGR-dependent Adhesion Sites-The adhesion of EA.hy926 cells to natural, recombinant, and synthetic fibronectin fragments containing the NGR motif was studied. First, the following proteolytic fragments of fibronectin were studied: (a) FN-70 kDa, containing the FN-I 1-9 and FN-II 1-2 repeats; (b) FN-30 kDa, containing the FN-I 1-5 repeats; and (c) FN-45 kDa, containing the FN-I 6 -9 and FN-II 1-2 repeats (see Fig. 1 for a schematic representation). All fragments, after adsorption to microtiter plates, induced cell adhesion and spreading ( Fig. 2A), suggesting the presence of pro-adhesive sites in these regions.
To investigate the role of the NGR motif in this phenomenon we analyzed the pro-adhesive properties of FN-I 4 -5 and FN-I 4 -5 SGS fragments, the latter corresponding to a mutant with SGS in place of NGR (residues 263-265). Recombinant FN-I 4 -5 , but not the SGS mutant, promoted cell adhesion (Fig.  2B). When we incubated these fragments in 0.1 M ammonium bicarbonate buffer, pH 8.5, for 16 h at 37°C (from now on this treatment will be called "heat treatment"), we observed an increase of cell adhesion to FN-I 4 -5 , but not to FN-I 4 -5 SGS (Fig. 2B). This suggests that a cell adhesion site, somehow related to the NGR sequence, was generated by accelerated aging. Similar results were obtained with synthetic FN-I 5 (Fig. 2C).
To assess whether the NGR motif was sufficient for mediating this phenomenon, we investigated the pro-adhesive properties of short peptides containing the NGR motif before and after heat treatment. Because the NGR tripeptide is unlikely to fold and to bind to microtiter plates, we introduced two flanking cysteines. The rationale for using flanking cysteines was based on the fact that the predicted conformation of CNGRC peptide, by molecular dynamic simulation, is similar to that of the GNGRG loop of FN-I 5 repeat (8). Furthermore, we fused this peptide to TNF (CNGRC-TNF) or to the first eleven residues of TNF (CNGRC-TNF 1-11 , devoid of TNF activity) to enable adsorption to microtiter plates. As expected, heat treatment of CNGRC-TNF increased cell adhesion (Fig. 2D). No adhesion was observed to TNF alone either before or after heat treatment (data not shown). Similarly, heat treatment of CNGRC-TNF 1-11 , but not of CARAC-TNF 1-11 (a control peptide), increased cell adhesion (Fig. 2D, right).
These results support the hypothesis that the NGR motif is sufficient for promoting cell adhesion after heat treatment. Interestingly, a CDGRC-TNF conjugate, prepared by recombinant DNA technology, was completely inactive (Fig. 2E), suggesting that Asn is a crucial residue for the enhanced activity. In conclusion, these results suggest that structural changes related to the Asn residue of NGR motif in FN-I 5 lead to generation of a pro-adhesive site.
Deamidation of the NGR Motif Is Associated with Increased Cell Adhesion-It is well known that the Asn residues, particularly when followed by Gly, can undergo non-enzymatic deamidation via succinimide intermediate at physiological pH (19 -23). This reaction leads to formation of Asp and isoAsp, predominantly in L-configuration (24). Accordingly, heat treatment of FN-I 5 , CNGRC-TNF, and CNGRC peptides increased their molecular mass by ϳ1 Da as measured by mass spectrometry analysis (data not shown). Furthermore, Ͼ0.5 mol isoAsp/ mol of heat-treated CNGRC-TNF subunit was detected using the IsoQuant kit.
To assess whether the enhanced adhesive properties of FN-I 5 after heat treatment depended on NGR deamidation, we incubated the heat-treated FN-I 5 and CNGRC-TNF with PIMT, an enzyme that converts L-isoAsp and D-Asp residues to L-Asp (25)(26)(27)(28)(29). PIMT almost completely inhibited the pro-adhesive activity of heat-treated FN-I 5 and CNGRC-TNF (Fig. 3A). Moreover, this enzymatic treatment partially inhibited the proadhesive properties of natural FN-30 kDa (Fig. 3B) and FN-45 kDa fragments (data not shown), suggesting that Asn deamidation can occur also in natural fibronectin fragments. To assess the specificity of this reaction we evaluated the effect of PIMT on retronectin, an FN fragment that is known to promote cell adhesion via RGD (see Fig. 1). As expected, in this case no inhibition was observed after enzymatic treatment (Fig. 3B). Conversely, in this case we observed a modest but significant increase. This was not totally unexpected as also RGD can undergo isomerization of Asp residues, with formation of isoAsp and loss of function (20,24,30,31). Thus, in this case PIMT is expected to "repair" RGD and increase cell adhesion.
In conclusion, these results suggest that NGR deamidation, in contrast to RGD isomerization, is associated with a "gain of function" in cell adhesion assays.
Mechanism of NGR Deamidation-The mechanism and the kinetics of peptide deamidation in DMEM, pH 7.53, and in 0.1 M ammonium bicarbonate, pH 8.5, were then investigated. To this aim the CNGRCGVRY peptide (called NGR-2C) was synthesized and analyzed by reverse-phase HPLC before and after incubation at 37°C. Residues GVRY were added to the C terminus of CNGRC to enable detection and column adsorption. In addition, two peptides called CDGRCGVRY (DGR-2C) and CisoDGRCGVRY (isoDGR-2C) corresponding to the same sequence of NGR-2C except for the presence of L-Asp and L-isoAsp in place of L-Asn, respectively, were prepared and analyzed by HPLC. The half-life of NGR-2C at pH 7.53 and pH 8.5, estimated from the height of the main chromatographic peak (Fig. 4A, peak 1), was ϳ4 and 2 h, respectively (Fig. 4A, inset). In contrast, DGR-2C and isoDGR-2C under the same conditions were considerably more stable than NGR-2C (not shown).
Further studies on NGR-2C stability in 150 mM sodium chloride, 25 mM Hepes, pH 7.4, and in 150 mM sodium chloride, 50 mM sodium phosphate, pH 7.3, showed that 1 and 5% of peptide, respectively, was degraded after 1 day at 4°C. These values increased to 27 and 86%, respectively, after 1 day at 37°C. Of note, the peptide was stable for more than 1 week at 4 and 37°C when stored in water (pH 5.6). Thus, the degradation reaction strongly depends on buffer composition. To identify the degradation products corresponding to peaks 2 and 3, formed in ϳ1:3 ratio, we next analyzed heattreated NGR-2C before and after spiking with NGR-2C or DGR-2C or isoDGR-2C. The results showed that peaks 2 and 3 correspond to DGR-2C and isoDGR-2C, respectively (Fig.  4B), suggesting that both degradation products are deamidated forms.
To verify this hypothesis and to assess whether the mechanism of deamidation occurs via succinimide intermediate we monitored the NGR-2C deamidation reaction by MALDI-TOF. This analytical method showed, as expected, a progres-sive decrease during incubation of the molecular species corresponding to NGR-2C (expected monoisotopic mass, MH ϩ 1025.43) and an increase of species characterized by ϩ1 Da (Fig.  4C). Noteworthy, a peak of Ϫ17 Da was also observed during the first 80 min of treatment, likely corresponding to the succinimide intermediate, which disappeared at later time points (Fig. 4C). These results strongly support the hypothesis that deamidation of Asn in the NGR motif occurs via loss of ammonia (Ϫ17 Da) followed by hydrolysis of the succinimide ring (ϩ 18 Da) with a global gain of ϳ1 Da, as has been demonstrated for other Asn-containing peptides (19 -23). ␣ v ␤ 3 Is a Receptor for the L-isoDGR Motif-In an attempt to identify the receptors of deamidated fibronectin fragments we analyzed the binding of heat-treated FN-I 5 , as well as that of CNGRC-TNF and various peptides, to purified ␣ v ␤ 3 , ␣ 5 ␤ 1 , and ␣ 1 ␤ 1 integrins. To this aim, direct and competitive ELISAs with integrins adsorbed on microtiter plates were performed. Direct ELISA showed binding of all NGR-containing molecules (heat treated) to ␣ v ␤ 3 but little, or not at all, to the other integrins (Fig. 5, A and B). No binding was observed with heat-treated TNF or with a control CARAC-TNF 1-11 peptide (Fig. 5, A and  B), suggesting that NGR was critical. Of note, we observed binding of CNGRC-TNF to ␣ v ␤ 3 even before heat treatment, although to a lower extent (data not shown), possibly due to deamidation occurring during preparation and/or assay incubation. Accordingly, pretreatment with PIMT decreased the binding of both "heat-treated" and "untreated" CNGRC-TNF to ␣ v ␤ 3 (data not shown).
To verify the importance of the NGR loop of FN-I 5 for integrin binding we performed competitive binding experiments with recombinant FN-I 4 -5 and FN-I 4 -5 SGS, the latter lacking NGR. As expected, only the fragment with NGR competed the binding of FN-I 5 to ␣ v ␤ 3 (Fig. 5C, left). Binding competition was observed also with a peptide corresponding to the entire 256 -271 loop of FN-I 5 (see Fig. 1C), but not with a control peptide with a scrambled sequence at the GNGRG site (Fig. 5, right). Similarly, the binding of CNGRC-TNF to ␣ v ␤ 3 was efficiently competed with heat-treated FN-I 5 (EC 50 , 0.4 M) (Fig. 6A).
To identify the NGR deamidation product responsible for integrin binding, other competitive ELISAs were performed using synthetic peptides with Asn replaced with Asp or isoAsp or Ser. The results showed that isoDGR-2C can compete 600-fold more efficiently than DGR-2C (EC 50 , 0.1 versus 60 M), whereas little or no competition occurred with the control peptide SGR-2C (EC 50 , Ͼ1000 M) (Fig.  6A). Furthermore, low efficiency was observed with D-isoDGR-2C, a peptide with D-isoAsp in place of L-isoAsp, indicating that the binding was stereospecific. Considering that the isoDGR-2C is stable under these assay conditions (see above), these results suggest that L-isoDGR is an ␣ v ␤ 3 binding motif.
Kinetics of Integrin Binding Site Formation in Fibronectin Fragments-To investigate the kinetics of integrin binding site formation in fibronectin fragments, we tested, by ELISA, the capability of FN-I 5 to compete the binding of CNGRC-TNF to ␣ v ␤ 3 after incubation for different times at 37°C in DMEM. Interestingly, when we plotted the inhibitory concentrations (IC 50 ) versus incubation time we observed that maximal competitive binding activity was reached after 24 -48 h of incubation (half-life 3.4 h) (Fig. 6B). This value corresponds very well to the time course of deamidation reaction of NGR-2C described above and further supports the concept that the integrin binding properties of heat-treated FN-I 5 were related to deamidation of the NGR site. Because the structural basis of heat-treated FN-I 5 function in cell adhesion is related to the presence of isoAsp we have named this product "isonectin-1" and we propose to call "isonectins" all bioactive fragments containing the deamidated FN-I 5 module.
L-isoDGR Is a Competitive Antagonist of RGD Ligands of ␣ v ␤ 3 and ␣ 5 ␤ 1 Integrins-To assess whether L-isoDGR binds the RGD binding site of ␣ v ␤ 3 -integrin we next performed competitive binding studies with various doses of ACDCRGDCFC-TNF, a known high affinity ligand of ␣ v ␤ 3 integrin (15), and various doses of isoDGR-2C and RGD-2C (CRGDCGVRY). The binding curves and Schild plot of binding data showed that both peptides efficiently antagonize, in a competitive manner, the binding of ACDCRGDCFC-TNF to ␣ v ␤ 3 -integrin (Fig. 7). This strongly suggests that the isoDGR motif binds within, or close to, the RGD binding site of ␣ v ␤ 3 with an affinity similar to that of the RGD motif (K d 0.57 and 0.41 M, respectively).
Because peptides containing CRGDC can also bind the ␣ 5 ␤ 1 integrin (32), we next performed competitive binding experiments with this integrin. As shown in Fig. 7C, the binding of ACDCRGDCFC-TNF to ␣ 5 ␤ 1 was weaker than that to ␣ v ␤ 3 . The binding was competed by RGD-2C and isoDGR-2C with similar potency (Fig. 7C), suggesting that the isoDGR motif can also bind the RGD binding site of ␣ 5 ␤ 1 . However, the IC 50 was 4 -5-fold higher than that required for ␣ v ␤ 3 , suggesting that the affinity for this integrin was lower.
Isonectin-1 Inhibits Tumor Growth in Vivo-Because ␣ v ␤ 3mediated endothelial cell adhesion is critical for endothelial cell survival and proliferation in tumors, we tested the hypothesis that isonectin-1 and isoDGR-2C peptide could affect tumor FIGURE 6. Inhibition of CNGRC-TNF binding to ␣ v ␤ 3 by heat-treated FN-I 5 and by synthetic isoDGR-2C. A, biotinylated CNGRC-TNF (4 g/ml) was mixed with different amounts of the indicated peptides, added to ␣ v ␤ 3coated plates, and left to incubate for 2 h at room temperature. Biotinylated CNGRC-TNF binding was detected with streptavidin peroxidase. Mean Ϯ S.E. (n ϭ 2). The concentration of each peptide able to compete 50% of binding (IC 50 ) is also shown. B, kinetics of FN-I 5 deamidation. FN-I 5 was diluted with DMEM and incubated for various times at 37°C. The capability of each product to inhibit the binding of deamidated CNGRC-TNF to ␣ v ␤ 3 was then tested by competitive ELISA. The concentrations able to inhibit 50% of CNGRC-TNF binding (IC 50 ) obtained with each product were plotted versus the incubation times. Bars represent the confidence intervals. FIGURE 7. L-isoDGR is a competitive antagonist of RGD ligands of ␣ v ␤ 3 and ␣ 5 ␤ 1 integrins. A, competitive binding of ACDCRGDCFC-TNF to ␣ v ␤ 3 with CisoDGRCGVRY (isoDGR-2C) and CRGDCGVRY (RGD-2C). Microtiter plates were coated with ␣ v ␤ 3 and incubated with mixtures of ACDCRGDCFC-TNF (agonist) and peptides (antagonists) at the indicated concentrations. ACDCRGDCFC-TNF binding was detected with anti-TNF antibodies as described under "Experimental Procedures." B, Schild plot of binding data obtained using the Prism program (GraphPad Software, Inc., San Diego, CA). C, competitive binding of ACDCRGDCFC-TNF to ␣ v ␤ 3 -or ␣ 5 ␤ 1 -coated plates with the indicated peptides. In this case a single concentration of ACDCRGD-CFC-TNF (0.1 g/ml, left panel; 3 g/ml, right panel) and various concentrations of competitor were used.

DISCUSSION
It is well known that proteins may contain isoAsp residues due to post-translational Asn deamidation or Asp isomerization reactions (21,31,(37)(38)(39). These reactions can occur in vivo, e.g. in extracellular matrix proteins with slow turnover (30,31,37), and in vitro during protein isolation and storage (22,40,41). In general these phenomena are associated with protein "loss of function" and, for this reason, are generally viewed as deleterious events associated with protein aging. In this work, for the first time we have shown that Asn deamidation of the NGR site of the fifth fibronectin type I repeat (FN-I 5 ) is associated with a gain of function, because deamidated fragments containing this module (named isonectins) are able to affect endothelial cell adhesion and proliferation in different assays.
This view is supported by the observation that (a) accelerated aging of synthetic FN-I 5 or recombinant FN-I 4 -5 , but not of FN-I 4 -5 -SGS (a mutant with the NGR sequence replaced with SGS), increased their cell adhesion properties; (b) accelerated aging was associated with Asn deamidation, as shown by mass spectrometry and isoAsp analysis of products; (c) treatment of aged fragments and peptides with PIMT, an enzyme that converts L-isoAsp and D-Asp residues to L-Asp (25,27), completely inhibited their pro-adhesive activity.
We have obtained evidence to suggest that Asn deamidation at NGR sites occurs via succinimide intermediate, which upon hydrolytic cleavage leads to formation of Asp and isoAsp in a 1:3 ratio (see Fig. 9 for a schematic representation). Racemization and hydrolysis of the succinimide intermediate leading to the formation of D-Asp and D-isoAsp is also possible, but this typically occurs with a much lower efficiency (19,20,24). Thus, deamidated products with L-configuration are likely quantitatively more relevant. Deamidation reactions can take a few hours, days, or years to occur depending on neighboring amino acid sequence, temperature, buffer composition, and ionic strength (19,21). For instance the presence of a Gly residue after Asn has a dramatic destabilizing effect (21,22,24). Accordingly, we have found that the kinetics of NGR deamidation in synthetic fibronectin fragments and peptides are surprisingly rapid (half-life 3-4 h), likely due to favorable conformation.
Three-dimensional structure analysis of FN-I 5 showed that the GNGRG motif forms an exposed loop, likely accessible to water and to receptors (see Fig. 1B). Of note, molecular dynamic simulation of an NGR peptide with flanking cysteines , isoDGR-2C, and SGR-2C. Vitronectin (3 g/ml) was adsorbed to microtiter plate, and cell adhesion assay was performed as described under "Experimental Procedures," using cell culture medium containing the soluble competitors. Mean Ϯ S.E. (n ϭ 3). B, inhibition of HMEC-1 cell proliferation by soluble isoDGR-2C, RGD-2C, and SGR-2C peptide. The assay was performed as described under "Experimental Procedures" using cell culture medium containing the soluble competitors. Mean Ϯ S.E. (n ϭ 6). C, anti-tumor effect of repeated administration of isonectin-1 (200 g, intraperitoneal) to RMA tumor-bearing mice or isoDGR-2C or RGD-2C or SGR-2C peptide (100 g, intraperitoneal) to B16F1 tumor-bearing mice as indicated. Animals (five/ group) were treated at the indicated times (arrows). **, p Ͻ0.0005; *, p Ͻ0.05; statistical analysis by two-tailed t test.
(CNGRC), used throughout this study as a molecular surrogate of the fibronectin NGR motif, showed that its structure is superimposable to that of the FN loop (8).
The question arises as to whether the receptor binding site in deamidated fibronectin is DGR or isoDGR. The following observations suggest that isoDGR is the biologically relevant motif. First, cell adhesion to deamidated FN-I 5 (named isonectin-1) was competed by peptides containing isoDGR but not the DGR motif. Noteworthy, no competition was observed with D-isoDGR-containing peptides, pointing to stereospecific interactions. Second, enzymatic conversion of L-isoAsp into L-Asp residues by PIMT completely inhibited the pro-adhesive activity of deamidated fibronectin fragments. This view is further supported by the observation that the CDGRC-TNF conjugate was completely inactive in direct cell adhesion assays. Thus, L-isoDGR is the bioactive motif.
Studies aimed at identifying the cellular binding sites showed that ␣ v ␤ 3 integrin efficiently binds deamidated FN-I 5 (isonectin-1) as well as peptides containing L-isoDGR, suggesting that this integrin is an important L-isoDGR receptor. A weaker binding was observed also with ␣ 5 ␤ 1 . Noteworthy, D-isoDGR was 60-fold less efficient in ␣ v ␤ 3 binding. Furthermore, L-DGR was 600-fold less efficient in ␣ v ␤ 3 recognition, supporting the concept that L-isoDGR, and not L-DGR, is the bioactive motif. Furthermore, studies aimed at characterizing the binding site on ␣ v ␤ 3 and ␣ 5 ␤ 1 showed that L-isoDGR acts as a competitive inhibitor of the RGD binding site of these integrins.
Can Asn deamidation also occur in natural fibronectins? We have found that natural FN-30 kDa fragment as well as intact fibronectin freshly isolated from human plasma contain 0.026 -0.048 pmol of isoAsp/pmol of protein. Furthermore, treatment of FN-30 kDa with PIMT partially inhibited its pro-adhesive properties. This suggests that formation of isoAsp can occur also in natural fibronectins in sufficient quantity to affect cell adhesion. We cannot exclude, however, that the kinetics of Asn deamidation in fibronectin might be slower than that measured with peptides or with FN-I 5 , considering the strong influence that the Asn molecular microenvironment might have on deamidation kinetics. One interesting possibility that deserves to be investigated is that conformational changes occurring after deposition in tissues or local changes in the microenvironment composition or the presence of specific proteases might affect the kinetics of this reaction. Considering the low turnover of fibronectin after deposition in tissues (42) and the abundance of this protein in plasma and in tissues, it is very likely that a significant amount of deamidated fibronectin is formed also in vivo. Another question is whether generation/removal of isoDGR in fibronectin could play a role in normal or pathological conditions. We observed that isonectin-1 can inhibit the adhesion of endothelial cells to vitronectin, a ligand of ␣ v ␤ 3 ; furthermore, peptides containing the isoDGR motif inhibited the proliferation of microvascular endothelial cells in vitro. We have also observed that daily administration of isonectin-1 to RMA lymphoma-bearing mice significantly inhibits tumor growth in vivo. These findings suggest that deamidated fragments may play a role in endothelial cell adhesion and tumor biology. Several investigators implicated ␣ v ␤ 3 as a receptor for various proteolytic fragments of ECM proteins that can act as anti-angiogenic factors (43)(44)(45). An interesting possibility is that deamidated fibronectin fragments contribute, together with other ECM protein fragments, to regulate angiogenesis in normal and pathological conditions. How can cells control isoDGR generation in fibronectin? Asn deamidation is a thermodynamically spontaneous reaction independent from enzymatic regulation. Whether specific deamidases further accelerating this process exist or not is presently unknown. However, the finding that enzymatic removal of isoAsp by PIMT inhibits isonectin-1 pro-adhesive activity suggests a potential enzymatic mechanism for negative regulation of this site. Formation of ␤-linked isopeptide bond (isoAsp) in proteins and subsequent conversion in ␣-linked peptide bond (Asp) by PIMT are generally regarded as a sort of "damage repair" mechanism of aged proteins (31). It has also been hypothesized that in some proteins these damage repair reactions may have some useful function, e.g. as a sort of molecular clock for protein degradation or intracellular localization (21,31,46). Based on our findings, a new "activation-deactivation" model may also be envisaged for isoAsp formation-removal occurring at certain sites in extracellular proteins (selected by evolution), such as fibronectin. Thus, whereas PIMT may be considered a sort of "repairing" enzyme for Asp residues undergoing isomerization, e.g. to rescue RGD in aged fibronectin and collagen (30,31), it may also be viewed as an enzyme that "destroys" the function of isoDGR, pointing to a new function for this enzyme. Of note, increased amounts of extracellular PIMT have been observed in injured tissues and wound healing (47,48).
Finally, considering that ␣ v ␤ 3 integrin is a good marker of angiogenic vessels, exogenous fibronectin fragments or short peptides containing the L-isoDGR motif may be exploited, in principle, as ligand for targeted delivery of drugs, cytokines, toxins, apoptotic peptides, radionuclides, viral particles, genes, or imaging compounds to angiogenic vessels in tumors or in other angiogenesis-related diseases.
In conclusion, spontaneous conversion of NGR to isoDGR in fibronectin fragments represents a novel mechanism for gener- FIGURE 9. Schematic representation of the NGR deamidation reaction. Asparagine deamidation occurs via hydrolysis of the succinimide intermediate, leading to formation of DGR or isoDGR, with changes in charges and peptide bond length.
ating ␣ v ␤ 3 ligands that may regulate endothelial cell functions and tumor growth. Generation of isoDGR sites in proteins by NGR deamidation (or DGR isomerization) may represent a novel mechanism for regulating their function.