Mechanism of Perturbation of Integrin-Mediated Cell-Matrix Interactions by Reactive Carbonyl Compounds and Its Implication for Pathogenesis of Diabetic Nephropathy

Purified α_vβ₃ and α₁β₁ integrins were purchased from Chemicon (Temecula, CA); α_v- and β₁-integrin monoclonal antibodies were purchased from Santa Cruz (Santa Cruz, CA); fibronectin, pyridoxamine hydrochloride, glyoxal, GLA, and MGO were obtained from Sigma-Aldrich (St. Louis, MO); d-glucose, mouse type IV collagen, and laminin-1 isolated from EHS tumor were purchased from Invitrogen (Carlsbad, CA); vitronectin was obtained from TaKaRa Biomedicals (Shiga, Japan).

Human umbilical vein endothelial cells were obtained from BioWhittaker (Charlotte, NC). Cells were grown in EGM-2 MV medium (BioWhittaker) and used between passages 4 and 7. Conditionally immortalized mouse mesangial cells were provided by Dr. Ambra Pozzi (Vanderbilt University Medical Center, Nashville, TN) (30). Conditionally immortalized human podocytes were derived as previously described (31). Mesangial cells were propagated in Dulbecco’s modified Eagle’s medium supplemented with 10% fetal bovine serum, 100 units/ml penicillin, 100 μg/ml streptomycin, and 100 units/ml interferon-γ at 33°C; podocytes were propagated in RPMI medium with the same supplements but without interferon-γ. For differentiation, cells were cultured in the above medium without interferon-γ at 37°C. Low passages (<20) of the immortalized cells were used to ensure the preservation of the specific properties characteristic for differentiated state (30,31).

Recombinant human α3NC1 and RGD-α3NC1 domains of type IV collagen with the NH₂-terminal FLAG sequence were stably expressed in HEK 293 cells and purified from conditioned medium by affinity chromatography on anti-FLAG agarose (Sigma) as previously described (29).

Proteins were immobilized on 96-well plates (Nunc, Rochester, NY) in Tris-buffered saline at 4°C overnight. Plates were washed twice with 200 mmol/l sodium phosphate buffer, pH 7.5, and incubated in the same buffer with or without different additives in the wet chamber to prevent changes in sample volume due to evaporation; 0.02% sodium azide was added to suppress bacterial growth. All incubations were carried out in the dark at 37°C. No protein desorption was detected after different treatments as determined by alkaline phosphatase competition assay (32) and enzyme-linked immunosorbent assay (ELISA) using anti-FLAG antibody (data not shown).

Physiologically relevant Amadori-BSA was prepared with diabetic concentration of glucose (30 mmol/l) using a previously described procedure (33).

Cell adhesion was quantified using crystal violet staining as previously described (29). Binding of integrins to different matrix proteins was determined in a solid-phase binding assay and quantified using corresponding integrin antibodies and ELISA as previously described (29).

Modification of protein lysine residues to carboxymethyl lysine (CML) were measured by ELISA using polyclonal anti-CML antibody R618 as described previously (34).

Human recombinant RGD-α3NC1 (30 μg) was incubated in 0.2 mol/l sodium phosphate buffer, pH 7.5, with or without 2 mmol/l MGO for 20 h at 37°C. After the removal of unreacted MGO, protein was reduced, alkylated with iodoacetamide, and digested with chymotrypsin (1:100 enzyme-to-protein ratio, wt/wt) in 0.1 mol/l NH₄HCO₃ buffer, pH 8, for 4 h at room temperature. Resulting peptide mixtures were snap-frozen in liquid nitrogen and stored at −70°C until mass-spectrometry analysis.

The liquid chromatography–tandem mass spectrometry (liquid chromatography-MS/MS) analysis was performed on a ThermoFinnigan LTQ linear ion trap mass spectrometer using previously described protocol (35). Protein was identified using the SEQUEST 2.0 algorithm (36) and the SEQUEST Browser software. A list of peptides was created and each peptide was run through P-Mod software to check for possible chemical modifications (37).

Data were expressed as mean ± SD, and statistical analysis was performed using Student’s t test for unpaired samples or ANOVA followed by post hoc Student-Newman-Keuls comparisons. Differences were considered statistically significant if P values were less than 0.05.

We have recently reported that endothelial cell adhesion to COOH-terminal fragment of human α3 chain of type IV collagen is mediated through both non-RGD regions of α3NC1 domain and adjacent RGD motif of collagenous domain using purified recombinant proteins with preserved native structure (29). In this paper, we used the same model to elucidate the mechanism by which diabetic conditions may affect cell-matrix interactions, particularly in renal glomerulus. The use of both RGD-α3NC1, a fragment of the human α3(IV) chain, and EHS collagen, purified α1(IV) and α2(IV) chains, allowed us to more accurately model specific protein components of native glomerular matrices.

In choosing the experimental conditions for MGO treatment, we were guided by the finding that the incubation of human albumin with 0.5 mmol/l MGO for 24 h resulted in modification of ∼1 Arg residue per molecule (38). Modification by MGO strongly inhibited endothelial cell adhesion to RGD-α3NC1 with maximum effect achieved at 2 mmol/l MGO after 24 h (Fig. 1). Similar experimental conditions were then applied to determine the effect of MGO on glomerular cell adhesion. Kidney glomerulus is characterized by specific distribution of collagen IV α-chains (39). In particular, the α3 chain is localized specifically to the GBM, which supports podocytes and endothelial cells, whereas α1 and α2 chains form the mesangial matrix supporting mesangial cells (39). Mesangial cells also interact with GBM in paramesangial regions (40). Along with collagen IV, laminin-1 is another major constituent of mesangial matrix (41). In the GBM, laminin-1 is present only during kidney development and is replaced by laminin-11 in the mature kidney (42). Because the purified laminin-11 is not available commercially, we modeled podocyte adhesion using laminin-1. Both α1 chain of laminin-1 and α5 chain of laminin-11 interact with integrins via RGD site (43,44). As shown in Fig. 2, MGO-modification of RGD-α3NC1 domain, full-length collagen IV, or laminin-1 resulted in inhibition of both mesangial cell and podocyte adhesion to these matrix proteins.

To investigate how other factors of diabetic milieu may affect cell-matrix interactions, we addressed the hypothesis that a direct modification of ECM proteins by glucose, particularly of lysine residues, may inhibit cell adhesion (5). Unlike MGO, which is a highly active electrofile that can react with both arginine and lysine protein side chains (20), glucose reacts preferentially with ε-amino group of lysine or α-amino group of NH₂-terminal amino acid forming advanced glycation end products via lysyl-Amadori intermediate (45). Incubation of RGD-α3NC1 with high (up to 1 mol/l) concentrations of glucose for 10 days at 37°C resulted in a dramatic increase in CML, a glucose-derived modification of lysine residues (45), as was determined using ELISA (Fig. 3A, inset). However, this modification did not affect endothelial cell adhesion (Fig. 3A). Similarly, modification of full-length collagen IV in the presence of glucose, as indicated by the extensive conversion of lysine residues to CML (Fig. 3B, inset), had no effect on cell adhesion (Fig. 3B).

We then investigated another possible mechanism of glucose-induced inhibition of cell adhesion, i.e., through formation of lysyl-Amadori intermediate on circulating proteins followed by cross-linking with arginine residues of matrix proteins. To this end, we prepared a physiologically relevant albumin-Amadori intermediate that mimics closely the degree of modification of serum albumin in diabetes (33). Preincubation of this glycated albumin with RGD-α3NC1 for up to 20 days did not affect endothelial cell adhesion to RGD-α3NC1 (Table 1). Similarly, endothelial cell adhesion to full-length collagen IV was not affected after preincubation with BSA and very high (500 mmol/l) concentration of glucose (data not shown).

To investigate the mechanism of inhibition of cell-matrix interaction by MGO, we used recombinant proteins α3NC1 or RGD-α3NC1, which represent either NC1 domain of α3 chain of human collagen IV or a larger fragment of α3 chain also including the adjacent collagenous region containing RGD sequence. MGO strongly inhibited endothelial cell adhesion to RGD-α3NC1 in a concentration-dependent manner (Fig. 4A). Modification with MGO inhibited cell adhesion to RGD-α3NC1 to the level of α3NC1 but had no effect on cell adhesion to α3NC1 itself (Fig. 4A). These results suggest that the arginine residue of RGD motif within RGD-α3NC1 is a target for the inhibition of cell adhesion by MGO.

In the competition experiments using free amino acids, only arginine protected from MGO-induced inhibition of cell adhesion to RGD-α3NC1, whereas lysine had no effect (Fig. 4B). This finding further indicates that the inhibition of cell adhesion to RGD-α3NC1 by MGO resulted from modification of an arginine residue within RGD sequence. Inability of the excess of free lysine to compete out the MGO effect suggests that modification of lysine residues within RGD-α3NC1 protein is not involved in inhibition of cell adhesion by MGO, a notion consistent with the results of the glucose experiments (Fig. 3).

Modification of an arginine residue within RGD sequence was directly demonstrated by mass-spectrometric analysis of MGO-modified RGD-α3NC1. Several modified proteolytic peptides containing RGD motif were detected in MGO-treated sample, whereas no modified peptides were found in untreated control (Table 2). The analysis of the MS/MS data using P-Mod algorithm (37), as shown for one of the peptides in Fig. 5, assigned the modifications to the arginine residue with the molecular masses of modifications corresponding to two known MGO derivatives of arginine, 5-hydro-5-methyl-4-imidazolone and 4-carboxy-4,6-dimethyl-5,6-dihydroxy-1,4,5,6-tetrahydropyrimidine (20,46). The same two MGO-derived modifications were also found on several other arginine residues of RGD-α3NC1 protein: Arg⁷⁸, Arg¹⁰¹, and Arg²⁰² (data not shown) [see (29) for amino acid positions]. However, because all of these residues are located within α3NC1 domain, their modification by MGO had no effect on cell adhesion (Fig. 4A). No modifications were found on the rest of arginines: Arg⁶, Arg¹², Arg⁴³, Arg⁵⁸, Arg¹⁰⁹, Arg¹⁷¹, Arg¹⁸¹, and Arg²⁰⁵, probably because of steric hindrance in the context of natively folded RGD-α3NC1 protein (data not shown). Proteolytic peptides containing Arg²³¹ have not been detected (data not shown).

Similar to results obtained with RGD-α3NC1 protein, free arginine but not lysine protected cell adhesion to full-length collagen IV upon modification by MGO (data not shown). Modification by MGO also strongly inhibited cell adhesion to fibronectin and vitronectin (data not shown), in agreement with a known critical role of RGD motifs in mediating cell adhesion to these ECM proteins (47,48). These data are consistent with MGO-induced inhibition of cell adhesion in major ECM proteins via modification of specific arginine residues involved in integrin-mediated cell-matrix interactions.

Integrin α_vβ₃ is a member of an extended family of integrins, cell surface receptors that mediate cell-matrix interactions. As with a number of other integrins, α_vβ₃ interacts with matrix proteins predominantly via specific binding to RGD sequence (27) and, as shown in our previous study, is a major endothelial cell receptor for RGD-α3NC1 (29). In the solid-phase binding assay, α_vβ₃ demonstrated strong binding to RGD-α3NC1, whereas virtually no binding to α3NC1 domain was detected (Fig. 6A, inset). Thus, any effect on α_vβ₃ binding would be due to modification of an arginine residue within RGD sequence of RGD-α3NC1 protein. Integrin binding to MGO-treated RGD-α3NC1 was strongly inhibited with half-maximal inhibitory concentration decreasing dramatically (from 160 to 5 μmol/l) as incubation time increased from 24 h to 20 days (Fig. 6A). In fact, after a 20-day incubation, a significant effect was already apparent at 1 μmol/l MGO (Fig. 6A), an MGO concentration close to that found in diabetic plasma (9,11). Similar to MGO, two other reactive carbonyl compounds that can be generated during autoxidation of glucose and glycated protein intermediates, glyoxal and GLA, also inhibited α_vβ₃ binding, albeit less efficiently (Fig. 6A, inset).

The interaction of collagen IV with the various cell types is mediated mainly via binding of the integrins α₁β₁ and α₂β₁ to the specific site in the triple-helical part of the molecule (49). Moreover, it has been demonstrated that Arg⁴⁶¹ within the α2 chain is essential for binding of α₁β₁ integrin (50). Modification of EHS collagen IV by MGO inhibited binding of purified α₁β₁ integrin to collagen IV (Fig. 6B). Like with RGD-α3NC1, the effective MGO concentration significantly decreased as incubation time increased.

Pyridoxamine, a drug that has shown promise in treatment of diabetic nephropathy (51) and is capable of scavenging reactive carbonyl compounds (34,52), prevented MGO-induced inhibition of cell adhesion to collagen IV and α_vβ₃ binding (Fig. 7A and B). This protection depended on pyridoxamine concentration and was already significant at equimolar concentrations of MGO and pyridoxamine (Fig. 7B).

Our results suggest that the weakening of cell-matrix interactions under diabetic conditions may occur because of modification of specific arginine residues within integrin-binding sites of matrix proteins. These arginine modifications are derived from reactive carbonyl compounds, most importantly MGO, that are elevated in diabetes (8–12). Inhibition of cell adhesion to major ECM proteins (collagen IV, laminin, fibronectin, and vitronectin) was also consistent with this mechanism. Furthermore, the mechanism was addressed in a context of cell-ECM interactions in renal glomerulus as demonstrated by a significant decrease in adhesion of both mesangial cells and podocytes to several ECM proteins modified by MGO.

Arginine residues are critical structural determinants in many binding sites of matrix proteins, which mediate ECM interactions with multiple members of integrin superfamily (27,28). For example, studies of collagen IV proteolytic fragments demonstrated that cell adhesion is mediated by several binding sites within triple helical and NC1 domains of the molecule (53,54), with the major cellular receptors being α₁β₁ and α₂β₁ integrins (55). Subsequently, Asp⁴⁶¹ in the α1(IV) chain and an Arg⁴⁶¹ in the α2(IV) chain have been identified as critical residues for α₁β₁ binding to the triple helical domain (50). Yet another integrin binding site containing critical arginine residue, GFOGER (in which O is hydroxyproline), was found in collagen I (56). Interestingly, triple-helical GFOGER peptide potently inhibited α₂β₁ integrin–mediated cell adhesion to collagen IV (56). Moreover, the same sequence exists in α1 chain of collagen IV (residues 385–390), where it may serve as a binding site for α₂β₁ integrin.

Like in collagen IV, specific arginine residues involved in mediation of cell adhesion have also been found in laminin, another major constituent of renal basement membranes and mesangial matrix (3). In addition to a cryptic RGDN motif of the A chain (43) and the YIGSR region in the B1 chain (57) of laminin-1, two RGD sites in α5 chain of laminin-10 and -11 are also involved in cell adhesion (44). Other glomerular matrix proteins fibronectin, vitronectin, and entactin appear to mediate cell adhesion predominantly via RGD motif (47,48,58). Thus, in native extracellular matrices, a set of several critical arginine residues within structurally diverse integrin binding sites would be available for targeting by reactive carbonyl species. It also implies that cell-matrix interactions in different tissues may be affected by nonenzymatic reactions via the same mechanism. Diminished adhesion of osteoblasts to MGO-modified collagen I has been reported (59).

In our experiments, glucose modification of immobilized RGD-α3NC1 domain or full-length collagen IV did not affect cell adhesion to these proteins (Fig. 3). Although glucose does not react with arginine residues directly, it may contribute as a source of reactive carbonyl species formed during glycoxidative reactions, which may, in turn, modify arginine. Interestingly, incubation of collagen I with ribose, a sugar prone to rapid oxidative degradation, caused inhibition of osteoblast adhesion to this matrix protein (60). However, in vivo glucose autoxidation may play a secondary role in inhibition of cell-matrix interactions because glyoxal is the only α-dicarbonyl compound formed in this reaction (13). Because the concentration of glyoxal in plasma of diabetic patients and model diabetic animals is three- to sixfold lower than that of MGO (9,61) and glyoxal is significantly less active at inhibiting integrin binding (Fig. 6), MGO-producing cellular processes such as degradation of glucose-derived triose phosphates may play a more important role in perturbation of cell-matrix interactions in diabetes.

Glucose may also modify arginine residues by formation of Lys-Arg cross-links via protein-Amadori intermediate (62,63). However, this contribution may be limited by the low reactivity of glucose-Amadori intermediate compared with that of reactive carbonyl compounds and by steric constrains imposed by the matrix network structure on the formation of cross-links. In our study, albumin-Amadori intermediate, which could potentially form cross-links with arginine residues of matrix proteins, had no effect on cell adhesion (Table 1).

Glucose may also affect cell-matrix interactions by regulating the expression of cellular integrins (64). However, in our experiments, these effects were not a factor because the cells were not directly treated with either glucose or reactive carbonyl compounds.

The weakening of cell-matrix interactions due to modification of specific arginine residues in matrix proteins by reactive carbonyl species provides a mechanism by which the “carbonyl stress” of diabetes may bring about effacement of glomerular epithelium and podocyte shedding characteristic of diabetic nephropathy (2). Interestingly, podocytes found in urine of rats with experimental diabetes were viable and could attach to collagen-coated plates (65). This suggests a primary role of inhibition of cell-matrix interactions in diabetes compared with cell death, at least for podocytes. Consequently, loss of cell contact with ECM can trigger apoptotic death (66). Similarly, weakening of cell-ECM interactions could cause the loss of endothelial cells in glomerular capillaries linked to impaired blood flow, development of renal ischemia, and progression of glomerular sclerosis (67). Integrin-mediated interactions with ECM may also play a key role in the regulation of cell growth (68). Within this context, modification of matrix components by reactive dicarbonyl compounds under diabetic conditions may facilitate migration and proliferation of mesangial cells, contributing to the expansion of mesangial matrix and the increase in glomerular volume.

The steady-state levels of MGO increase significantly in plasma of diabetic patients compared with normal subjects ranging from 0.4 to 2 μmol/l (8,9). In our in vitro study, the lowest effective MGO concentration was within this physiological range, i.e., 1 μmol/l in solid-phase integrin binding experiments after a 20-day exposure. It is important to note that in diabetes, proteins of glomerular matrix would be exposed to the elevated levels of reactive carbonyl compound for years before the appearance of the initial clinical signs of nephropathy. A long time needed for development of diabetic nephropathy, very slow turnover of matrix proteins, the fact that the concentration of circulating MGO correlates positively with duration of diabetes (8), and the presence of MGO- and 3-DG–derived arginine modifications in renal glomeruli of diabetic rats and in glomerular mesangial matrix of diabetic patients (23,25,69) support the notion that our proposed mechanism may be operable under diabetic conditions in vivo.

Our results also suggest that scavenging of reactive carbonyl compounds may protect integrin-mediated cell-matrix interactions, thus delaying or preventing the deterioration of kidney filtration function in diabetes. Interestingly, pyridoxamine, which protected cell adhesion and integrin binding in our experiments (Fig. 7), has also shown promising results in treatment of diabetic nephropathy. In animal studies, pyridoxamine inhibited diabetes-induced albuminuria and an increase in glomerular volume (70). In clinical trials, it significantly decreased the rate of serum creatinine accumulation in patients with overt nephropathy (71). The pyridoxamine protection of cell-matrix interactions in vivo is possible because its concentration in plasma of pyridoxamine-treated diabetic animals can be as high as 100 μmol/l (70), a significant molar excess over circulating MGO and other carbonyl compounds. Consequently, in diabetic animal models, pyridoxamine treatment decreased plasma MGO levels and specific MGO-derived protein modification (52,61). The reported therapeutic effects of pyridoxamine in diabetic patients and in animal models of diabetes further support our hypothesis that inhibition of cell adhesion by MGO and other dicarbonyl compounds may contribute to the pathogenesis of diabetic nephropathy.

B.G.H. has received support from National Institute of Diabetes and Digestive and Kidney Disease Research Grant DK-065138. P.A.V. has received support from National Institute of Diabetes and Digestive and Kidney Disease Research Grant DK-066415.

We thank Dr. Ambra Pozzi for a gift of mouse glomerular mesangial cells and Anjana Dey for excellent technical assistance. We also thank Vanderbilt University for institutional support of the Proteomics Laboratory in the Mass Spectrometry Research Center through the Academic Venture Capital Fund.