Increased Dicarbonyl Metabolism in Endothelial Cells in Hyperglycemia Induces Anoikis and Impairs Angiogenesis by RGD and GFOGER Motif Modification

All chemicals were obtained from Sigma-Aldrich (Poole, Dorset, U.K.), unless otherwise stated.

Human dermal microvascular endothelial HMEC-1 cells were cultured as described previously (27). Primary human aortic endothelial cells (HAECs; Cambrex Bioscience) were cultured in endothelial growth medium with 2% fetal bovine serum, 10 ng/ml endothelial growth factor, 1 mg/ml hydrocortisone, 50 mg/ml gentamicin, and 50 μg/ml amphotericin-B and studied during passages 4–6. Levels of methylglyoxal, glyoxal, and 3-deoxyglucosone were determined in HMEC-1 and HAECs and culture medium by derivatization with 1,2-diaminobenzene and quantitation by liquid chromatography with tandem mass spectrometric detection (LC-MS/MS). Internal standardization was achieved by use of stable isotope-substituted standards and calibration by analysis of authentic standards. Briefly, 50 μl culture medium or 1 million cells were deproteinized by addition of 20 μl 10% trichloroacetic acid. Internal standards, 5 pmol each of [¹³C₂]methylglyoxal, [¹³C₂]glyoxal, and [¹³C₆]3-deoxyglucosone (prepared in-house), were added, and the dicarbonyls were derivatized by incubation with 100 μmol/l 1,2-diaminobenzene, 100 μmol/l diethylenetriaminepenta-acetic acid, and 1.2 mmol/l sodium dithionite stabilizer for 4 h in the dark. The samples were then analyzed by LC-MS/MS. The column was a Waters ODS Symmetry (100 × 2.1 mm) and guard column (10 × 2.1 mm). The mobile phase was 0.1% trifluoroacetic acid in water with a linear gradient of 0–50% methanol over 0–20 min and isocratic 50% methanol thereafter; the flow rate was 0.2 ml/min. The capillary voltage was 3.5 kV, the cone voltage 50 V, the interscan delay time 100 ms, the source and desolvation gas temperatures 120 and 350°C, respectively, and the cone gas and desolvation gas flows were 150 and 550 l/h, respectively. Mass transitions (parent ion>fragment ion; collision energy), retention time, limit of detection, and recoveries were as follows: glyoxal 130.9 > 77.1 Da, 28 eV, 18.7 min, 0.21 pmol, and 99%; methylglyoxal 144.8 > 77.2 Da, 28 eV, 20.8 min, 0.25 pmol, and 94%; and 3-deoxyglucose 235.0 > 199.2 Da, 16 eV, 15.5 min, 0.13 pmol, and 76%.

Wells of 96-well polystyrene plates (Corning) were coated with glycated and unglycated collagen (human placental, type IV, and pepsin-extracted triple helical domain [28]) by incubation with 70 μg/ml collagen (50 μl) for 24 h at 37°C. Wells were washed with PBS (3 × 100 μl). Surface coating was assayed with 3-(4-carboxybenzoyl)quinoline-2-carboxaldehyde reagent (Invitrogen, Paisley, Scotland), releasing the collagen coat by washing with PBS containing 0.1% Triton X-100 (3 × 100 μl). Total protein recovery was 94 ± 4%. HMEC-1 cells were plated on collagen substrates and incubated for 1 h at 37°C. Nonadherent cells were removed, and adherent cells were quantified by staining with 3-(4,5-dimethylthiazole-2-yl)-2,5-diphenyltetrazolium bromide and lysed with dimethylsulfoxide, and the blue formazan product was detected by absorbance at 560 nm (29). HMEC-1 cells grown in normal conditions (5 mmol/l glucose) and hyperglycemia (20 mmol/l) were plated in wells coated with collagen preincubated with medium of HMEC-1 cultures in normal and high glucose concentrations (5 and 20 mmol/l, respectively) for 7 days. Normal and high glucose concentration medium–conditioned collagen and collagen conditioned by medium from HMEC-1 cells incubated with the cell permeable glyoxalase I inhibitor S-p-bromobenzylglutathione cyclopentyl diester (BrBzGSHCp₂) (1 μmol/l, a nontoxic concentration) for 48 h were prepared by incubating type IV collagen (6 mg/ml) for 24 h with the medium and then dialyzed. HMEC-1 cells were also plated in wells coated with collagen glycated minimally with glucose (AGE_min-collagen). AGE_min-collagen was prepared by incubating type IV collagen (6 mg/ml) with 50 mmol/l glucose in 100 mmol/l sodium phosphate buffer, pH 7.4, at 37°C, for 21 days and then dialyzed. HMEC-1 cells were also plated in wells coated with collagen glycated by 500 μmol/l methylglyoxal with incubation for 6 h (MG_min-collagen) and for 12 and 24 h (MG-collagen), and control collagen was incubated similarly with aminoguanidine (500 μmol/l) to scavenge methylglyoxal and prevent glycation or without glycating agent. The adhesion assay was also performed with HMEC-1 cells preincubated for 30 min with anti–α₁ and anti–α₂β₁ monoclonal antibodies (30 μg/ml; Chemicon). Endotoxin levels in collagen preparations were measured by limulus assay (QCL-1000; BioWhittaker) and were <18 pg/ml when deployed.

Collagen was modified by methylglyoxal. Type IV collagen (6 mg/ml) was incubated with 500 μmol/l methylglyoxal in 100 mmol/l sodium phosphate buffer, pH 7.4, at 37°C for 6, 12, and 24 h and dialyzed as described previously (19). The resultant collagen glycated by methylglyoxal was analyzed for glycation adducts by exhaustive enzymatic hydrolysis and LC-MS/MS (20). Total methylglyoxal-derived adducts were quantified by preparing similar glycated collagen derivatives with radiolabeled [2-¹⁴C]methylglyoxal (19).

The locations of MG-H1 residues in MG-collagen were identified by peptide mapping with thermolysin (Bacillus thermoproteolyticus rokko, type X lyophilized powder) and electrospray cationic LC-MS, with corroboration peptide mapping with endopeptidase Glu-C (Staphylococcus aureus V8, type XVII-B lyophilized powder) (19). LC-MS peptide responses were normalized to the COOH-terminal peptide response (which is not glycated). Loss of peptide by methylglyoxal modification was deduced by comparison of normalized peptide responses in methylglyoxal-modified collagen with that of control collagen in triplicate (19).

Aortas were removed from streptozotocin (STZ)-induced diabetic and normal control rats, collected and described in a previous study (30). Briefly, male Sprague-Dawley rats, 250 g, were kept in two per cage at 21°C, 50–80% humidity with daily 14-h light cycle and had free access to food and water. Diabetes was induced by intravenous injection with 55 mg/kg STZ. Body weight and moderate hyperglycemia were stabilized by subcutaneous injection of 2 units Ultralente insulin every 2 days. Thiamine and benfotiamine (70 mg · kg⁻¹ · day⁻¹) were given orally in the chow over 24 weeks. Aortas were cleaned from external debris, rinsed in ice-cold PBS, minced, and homogenized in 0.5 ml PBS. Samples were centrifuged (6,000g, 4°C, and 20 min), and supernatant was removed. The pellets were washed twice with PBS, stirred in 2 ml chloroform:methanol (2:1) at 4°C for 24 h, washed twice with water, blotted dry, and weighed. Aliquots of aortal collagen were hydrolyzed by exhaustive enzymatic hydrolysis for AGE analysis by LC-MS/MS or by limited proteolysis with both Glu-C and thermolysin for peptide mapping analysis by matrix-assisted laser desorption ionization mass spectrometry (Bruker Daltonics Reflex IV mass spectrometer).

The crystal structure of the α₂I domain from α₂β₁ integrin bound to triple helical GFOGER collagen peptide (Protein Data Bank identification number 1dzi) at 2.1Å (9) was obtained from the Research Collaboratory for Structural Bioinformatics database. Arg-12 in the crystal structure corresponds to Arg-390 in the α₁ chain of human type IV collagen and was replaced with the hydroimidazolone MG-H1 residue. The native and modified protein structures were equilibrated using the Generalized Born solvation model (31) and then subjected to the Dreiding force field (32) until the total energy of the system reached equilibration. Energy minimizations of modifications and molecular graphics were performed using DS Viewer Pro 5.0 (Accelrys, San Diego, CA).

HMEC-1 and HAECs were incubated on methylglyoxal-modified and unmodified control collagen for 24 h. The nonadherent cells were then removed and fixed in 1% paraformaldehyde and stained using the TdT-mediated dUTP-biotin nick-end labeling (TUNEL) method (Apoptag kit; Chemicon), and the number of apoptotic cells assessed by cell counting on a confocal microscope (Bio-Rad). Cells were counterstained with propidium iodide for cell cycle analysis, and percentage with sub-G₁ DNA content was taken as a measure of apoptosis. Nonadherent HMEC-1 and HAECs were double stained with Hoescht 33342 and propidium iodide. Necrotic cells were stained red with propidium iodide and were assessed by cell counting on a confocal microscope (33).

Matrigel (ECM from the Engelbreth-Holm-Swarm murine tumor; Sigma) modified minimally by methylglyoxal (MG_min-ECM) was prepared by incubation of ECM gel (6 mg/ml) with 500 μmol/l methylglyoxal for 24 h at 37°C and dialyzed. Control ECM was incubated without MG. Modification of ECM by methylglyoxal was also prepared in the presence of aminoguanidine (500 μmol/l). ECM preparations were plated onto a 96-well plate and allowed to gel at 37°C for 12 h. Microvascular endothelial HMEC-1 cells were then seeded onto each well and incubated for 24 h at 37°C. Endothelial cell tube formation was quantified by image analysis of 100× phase contrast images of eight random fields using NIH Image software. The effect of model hyperglycemia and glyoxalase I inhibition was studied by preincubation of HMEC-1 cells with 20 mmol/l glucose and 1 μmol/l BrBzGSHCp₂ for 48 h.

Data are means ± SD. Significance of difference between mean and medians was assessed by Student’s t test and the Mann-Whitney U test, as appropriate.

Human microvascular endothelial HMEC-1 cells were incubated in medium with 5 mmol/l glucose (normoglycemia) and 20 mmol/l glucose (model hyperglycemia). The concentration of methylglyoxal in the culture medium was fourfold higher than glyoxal. The concentration of both dicarbonyls in the culture medium increased two- to threefold in hyperglycemia and fourfold in the presence of BrBzGSHCp₂ (Fig. 2A). The cellular concentration of dicarbonyls was increased less markedly than that of the culture medium in hyperglycemia and in the presence of BrBzGSHCp₂ (Fig. 2B).

Coating of microplates with control human type IV collagen led to 1.52 ± 0.25 μg protein/well (48% of total protein) and was not significantly different for the glycated collagen derivatives. Hyperglycemic culture medium–conditioned collagen decreased the adhesion of HMEC-1 cells incubated in normoglycemia by 37%, with respect to unconditioned control collagen. This decrease in adhesion was prevented by aminoguanidine (Fig. 2C). To control for changes in integrin expression in hyperglycemia (34), we also studied the collagen adhesion properties of HMEC-1 cells cultured in hyperglycemia. Hyperglycemic culture medium–conditioned collagen decreased the adhesion of HMEC-1 cells incubated in hyperglycemia by 55%, with respect to unconditioned control collagen. This was partially prevented by addition of aminoguanidine during the conditioning of the collagen with the medium and totally inhibited with addition to aminoguanidine during the conditioning and cell adhesion assay (Fig. 2D). Adhesion of endothelial cells to AGE_min-collagen was also decreased by 49%, with respect to unconditioned control collagen. Conditioning of collagen with culture medium from endothelial cells incubated in normoglycemia with the BrBzGSHCp₂ also decreased HMEC-1 cell adhesion, which was also prevented by aminoguanidine (Fig. 2E). Cell adhesion to collagen was blocked by monoclonal antibodies to α₁ and α₂β₁ integrins. In all cases, collagen conditioned with hyperglycemic culture medium, AGE_min-collagen, and MG_min-collagen, the major and common modification of collagen was MG-H1 residues (Table 1; Figs. 1A and 3A and B). Similar effects were also found with cultures of primary HAECs (data not shown).

To model the effect of MG-H1 residue formation in type IV collagen in hyperglycemia, we modified collagen by methylglyoxal in vitro. Modification of human type IV collagen with methylglyoxal in vitro for 6 h formed 3.6 molar equivalents of MG-H1 residues (Table 1) and decreased endothelial cell binding by 58%, which was inhibited by aminoguanidine (Fig. 2G). Increasing extent of modification of collagen led to increased MG-H1 residues and further decrease in endothelial cell adhesion with the greatest marginal decrease in adhesion for the initial 3.6 molar equivalents of MG-H1 residues (Fig. 2H).

Human type IV triple helical collagen was glycated to low extents with methylglyoxal under physiological conditions in vitro. AGE residue contents were (in mol/mol protein; means ± SD, n = 3; at 6 h) MG-H1 3.60 ± 0.17, Nε-(1-carboxyethyl)lysine 0.029 ± 0.007, argpyrimidine 0.052 ± 0.019, and methylglyoxal-derived lysine dimer 0.0015 ± 0.0001; and (at 24 h) MG-H1 10.51 ± 0.61, Nε-(1-carboxyethyl)lysine 0.221 ± 0.011, argpyrimidine 0.224 ± 0.039, and methylglyoxal-derived lysine dimer 0.0098 ± 0.0014. Total methylglyoxal-derived adducts, determined by incubation with [2-¹⁴C]methylglyoxal, were 3.96 ± 0.46 and 9.19 ± 0.50 mol/mol protein for the 6- and 24-h incubations, indicating that all methylglyoxal-derived adducts were detected in the LC-MS/MS assay. The main methylglyoxal-derived glycation adduct in methylglyoxal-modified MG-collagen was MG-H1 residues, representing >95% of total methylglyoxal-derived glycation adducts (Fig. 3A and B; Table 1).

The sites in human type IV collagen susceptible to glycation by methylglyoxal were identified by mass spectrometric peptide mapping with limited proteolysis by thermolysin and corroborated by similar studies with endopeptidase Glu-C. This revealed glycation hotspots for formation of MG-H1 residues at Arg-390 (α₁) (58 mol %), Arg-889 (α₂) (57 mol %), and Arg-1452 (α₂) (55 mol %). The modified peptide of Arg-390 (α₁), R_MG-H1GE, was detected in MG-collagen but not in the control collagen. These modified arginine residues are within α1 chain GFOGER and α2 chain RGD integrin binding sites (Fig. 3C and D; Table 2). Other unlocated MG-H1 residues were also present.

We then sought evidence that collagen of the vascular basement membrane may be modified by methylglyoxal-forming MG-H1 residues at functional sites in vivo. We analyzed aortal collagen of rats, mainly collagen types I and III with some type IV collagen. MG-H1 residues were detected as major AGEs in aortal collagen of control and STZ-induced diabetic rats. Aortal collagen MG-H1 residue content increased approximately twofold in diabetic rats and was decreased by high-dose therapy with thiamine and benfotiamine. Aortal collagen MG-H1 residue contents were (mmol/mol Arg; mean ± SD or median, minimum–maximum) normal control 0.55 ± 0.30 (n = 13); diabetic control 1.26, 0.38–4.23 (n = 11, P < 0.001 with respect to normal control); diabetic + 70 mg/kg thiamine 0.65 ± 0.26 (n = 7, P < 0.05, with respect to diabetic control); and diabetic + 70 mg/kg thiamine benfotiamine 0.49 ± 0.22 (n = 8, P < 0.01, with respect to diabetic control). Limited proteolytic digests of rat aortal collagen by matrix-assisted laser desorption ionization mass spectrometry showed the unmodified peptide, ARGDDG (590.3 Da), in control and STZ-induced diabetic rats and detectable MG-H1-modified RGD peptide AR_MG-H1GDDG (644.5 Da) in some control rats (three of six) and all diabetic rats (six of six). These are tryptic peptides of rat type I collagen. Similar modification of type IV collagen is expected.

The structural basis for impairment of integrin-collagen interaction on modification of arginine residues in GFOGER sites by methylglyoxal was examined by molecular modeling. The interaction of α₂β₁ integrin with collagen GFOGER peptide triple helix (Fig. 1B) involves two of the three strands (middle and trailing strands) interacting with the integrin α₂-I domain (Fig. 1C). The middle strand GFOGER motif interacts with the α₂-I domain where the Glu-389 carboxylate group coordinates the Mg²⁺ ion; the Mg²⁺-carboxylate bond distance was 2.3Å. Other ligands of the Mg²⁺ ion are Ser-153, Ser-155, Thr-221 of the α₂-I domain and two water molecules (not shown). The Thr-221 side chain OH group makes an H-bond with the neighboring Mg²⁺-coordinated carboxylate group of Glu-389. The arginine residue of the trailing strand GFOGER, equivalent to Arg-390, is hydrogen-bonded to His-258 in loop 3 (2.09Å) of the α₂-I domain and reaches into an acidic pocket within the α₂-I domain to establish weak ionic interactions with Glu-256 (3.16Å) (9). Modeled substitution of Arg-390 for MG-H1 revealed significant distortions in the protein-protein interaction (Fig. 1D). The MG-H1 residue bends out of the acidic pocket, and the hydrogen bond to His-258 is lost. The collagen Glu-389 has lost hydrogen bonding to Thr-221 and coordination to Mg²⁺, with predicted rotation and separation of Mg²⁺-carboxylate groups to a bond length of 3.79Å. Similar structural changes in human type IV collagen likely account for decreased integrin binding and endothelial detachment.

Endothelial detachment from basement membrane collagen may lead to apoptosis (Fig. 1E). Nonadherent HMEC-1 and HAECs exposed to MG-collagen showed significant increase in apoptotic cells versus cells incubated with control collagen, as measured by TUNEL staining and by cell-cycle analysis. Apoptotic HMEC-1 cells increased from 6.1 ± 0.4 to 17.9 ± 1.1% and apoptotic HAECs increased from 2.2 ± 0.1 to 6.9 ± 0.2% by TUNEL staining (Fig. 1F and G). There was no significant endothelial cell necrosis and no significant increase in apoptosis and necrosis in endothelial cells attached to collagen.

Increased glycation by methylglyoxal of basement membrane protein components may impair endothelial cell–ECM interactions mediating angiogenesis (7). We investigated the effect of minimal glycation of murine tumor ECM on the formation of tube-like structures of endothelial cells in vitro, a model of angiogenesis (35). ECM modified minimally by methylglyoxal inhibited the formation of tube-like structures by endothelial cells in vitro. This effect was prevented by aminoguanidine (Fig. 4A–C). Tube formation was also inhibited by RGD peptide and monoclonal antibodies to α₁ and α₂β₁ integrins. Endothelial cells incubated in hyperglycemia and with BrBzGSHCp₂ also showed impaired formation of tube-like structures (Fig. 4D).

This study implicates increased formation of MG-H1 at critical functional sites in type IV collagen as modifications leading to endothelial cell anoikis and decreased angiogenesis in hyperglycemia associated with diabetes. Loss of microvascular endothelial cells with development of acellular capillaries in diabetes has been linked to retinopathy and generalized microangiopathy (36). Increased apoptosis of microvascular endothelial cells in hyperglycemic culture in vitro and experimental and clinical diabetes in vivo suggested activation of programmed cell death (37). Herein, we show for the first time that increased formation of methylglyoxal in hyperglycemia and glycation of collagen is implicated in endothelial cell apoptosis, detachment, and impairment of angiogenesis.

In hyperglycemia, increased uptake of glucose by predominantly GLUT1 transport into endothelial cells leads to cytosolic hyperglycemia, biochemical dysfunction (2), and increased formation of methylglyoxal (38). This contributes to increased methylglyoxal concentration in plasma in hyperglycemia associated with experimental and clinical diabetes (16,30). The increases of methylglyoxal were less marked inside cells than in extracellular medium and plasma, an effect attributed to the presence of cellular glyoxalase I activity (15). Cellular methylglyoxal concentrations was increased in normoglycemia by inhibition of glyoxalase I with BrBzGSHCp₂ (39). Glyoxalase I is the major enzymatic activity that catalyzes the metabolism of glyoxal and methylglyoxal (15). Increased formation and plasma concentrations of methylglyoxal in diabetes leads to increased formation of AGEs, increasing hydroimidazolone MG-H1 residues in plasma protein, hemoglobin, and cytosolic proteins of glomeruli, retina, and peripheral nerve (19–21). Formation of methylglyoxal occurs during glycation of proteins by glucose via fragmentation reactions (40), and hence increased MG-H1 residues were also found in AGE_min-collagen. Analysis of protein glycation, oxidation, and nitration adducts in collagen showed that glycation adducts MG-H1 and FL were major types of protein damage quantitatively in collagen glycated in vitro. We also found increased glycation of aortal collagen by methylglyoxal and glucose in experimental diabetes.

An important functional effect of arginine modification in ECM collagen was impairment of endothelial cell attachment. Binding of endothelial cells to type IV collagen is mediated by α₁ and α₂β₁ integrins (9,26), confirmed in this study by inhibition of endothelial cell binding to type IV collagen with anti-integrin antibodies. Glycation of human type IV collagen by methylglyoxal formed by biochemical dysfunction of endothelial cells in model hyperglycemia in vitro led to loss of integrin binding and cell detachment. Glyoxal and 3-deoxyglucosone are other dicarbonyls increased in hyperglycemia and diabetes, but hydroimidazolone residues derived from these dicarbonyls in collagen were not increased significantly in collagen glycated in vitro (Table 1). This may be because the concentration of glyoxal was relatively low in endothelial cell cultures and 3-deoxyglucosone is much less reactive than methylglyoxal.

Peptide mapping of human type IV collagen indicated that methylglyoxal modification of collagen occurred at arginine residues in GFOGER and RGD sites in vitro and in vivo. The dissociation constant of the α₂β₁ integrin–GFOGER complex is ∼1 nmol/l (41). Replacement of Arg to Lys in a GFOGER peptide decreased the peptide binding to α₂β₁ integrin by twofold (10), whereas chemical modification of the Arg by p-azidophenylglyoxal decreased binding by 100-fold (26). Our studies indicate that similar modification of collagen by methylglyoxal occurs in vivo. Increased formation of MG-H1 residues at these sites in diabetes may decrease the binding of endothelial cells to integrins with induction of anoikis and impairment of angiogenesis.

Ligation of α₁β₁ and α₂β₁ integrins to unmodified RGD and GFOGER sites of triple helical collagen provide survival signals for vascular microvascular and aortal endothelial cells. Modification of these residues in type IV collagen and the highest marginal decrease in cell binding for the initial 3.6 molar equivalents of MG-H1 formation reflects hotspot glycation with concomitant functional impairment. Unligated integrins activate apoptosis in endothelial cells via regulation of integrin-linked kinase (42). Interactions with the globular NC1 domain (α_Vβ₃ and α_Vβ₅) are also involved. Pedchenko et al. (43) found that methylglyoxal modification of critical arginine residues in the α₃ noncollagenous domain also impaired endothelial cell adhesion. MG-H1 formation was implicated in detachment of human umbilical vein endothelial cells, mesangial cells, and podocytes from ECM proteins. A further study has linked glycation of fibronectin by methylglyoxal to apoptosis of retinal pericytes (44). Administration of exogenous methylglyoxal to rats induced damage to the vasculature similar to that found in diabetes (45).

ECM modification by methylglyoxal also inhibited angiogenesis. Inhibition of angiogenesis occurs in the early stages of vascular disease in diabetes. This may contribute to tissue ischemia, impairment of wound healing, and coronary collateral vessel development (8). Emerging therapeutic interventions to prevent vascular complications of diabetes, high-dose therapy with thiamine, benfotiamine, and pyridoxamine, decreased protein glycation by methylglyoxal (30,46). Prevention of endothelial cell anoikis and impaired angiogenesis may contribute to their therapeutic effects.

This research has received support from the Wellcome Trust (U.K.) and the Juvenile Diabetes Research Foundation International (New York).