Glycation of type I collagen selectively targets the same helical domain lysine sites as lysyl oxidase–mediated cross-linking

Nonenzymatic glycation of collagen has long been associated with the progressive secondary complications of diabetes. How exactly such random glycations result in impaired tissues is still poorly understood. Because of the slow turnover rate of most fibrillar collagens, they are more susceptible to accumulate time-dependent glycations and subsequent advanced glycation end-products. The latter are believed to include cross-links that stiffen host tissues. However, diabetic animal models have also displayed weakened tendons with reduced stiffness. Strikingly, not a single experimentally identified specific molecular site of glycation in a collagen has been reported. Here, using targeted MS, we have identified partial fructosyl-hydroxylysine glycations at each of the helical domain cross-linking sites of type I collagen that are elevated in tissues from a diabetic mouse model. Glycation was not found at any other collagen lysine residues. Type I collagen in mouse tendons is cross-linked intermolecularly by acid-labile aldimine bonds formed by the addition of telopeptide lysine aldehydes to hydroxylysine residues at positions α1(I)Lys87, α1(I)Lys930, α2(I)Lys87, and α2(I)Lys933 of the triple helix. Our data reveal that site-specific glycations of these specific lysines may significantly impair normal lysyl oxidase–controlled cross-linking in diabetic tendons. We propose that such N-linked glycations can hinder the normal cross-linking process, thus altering the content and/or placement of mature cross-links with the potential to modify tissue material properties.

Among the known and suspected cross-links in fibrillar collagens, perhaps the least understood but most speculative pathologically are advanced glycation end-products (AGEs). 2 Collagens are well recognized targets for such addition products from glucose and other carbonyl-containing reactants, because they are typically very long-lived proteins with halflives ranging from 1 to 2 years in bone and 10 years in skin (1) to a 100 years or more in cartilage (2) and tendon (3,4). It is evident that nonenzymatic AGE cross-links increase over time and are associated with tissue stiffening (5,6). Remarkably, after decades of research and thousands of publications, no specific sites of AGE cross-linking have yet been experimentally identified in a collagen. There is evidence, however, that fibrillar collagen glycation is not random but likely to occur at favored sites (7).
AGEs derive spontaneously with time by further reactions of initial lysine side-chain glycation adducts. These potentially could include new intra-and intermolecular cross-links (8). Unlike glycosylation (an enzymatically regulated intracellular process during collagen synthesis), which covalently attaches sugar molecules to the hydroxyl moiety of certain hydroxylysines (Hyl), glycation is a nonenzymatic and opportunistic chemical addition of the carbonyl group of free reducing sugars (primarily glucose) to the ⑀-amine of a Lys side chain. Glycation is also the initial step of the classical Maillard reaction pathway, which begins with a relatively unstable Schiff base adduct of the Lys residue that Amadori rearranges to the more stable ketoimine, fructosyl-Lys. With time these can dehydrate, condense, fragment, and cross-link, thereby forming a complex array of AGEs. The most prominent end-product seems to be glucosepane, a fructosyl-Lys to arginine addition product and potential cross-link, formed by dehydration and carbonyl migration along the sugar's carbon chain. The fructosyl-Lys and an Arg side chain need to be Յ7 Å apart for such a crosslink to form (9). Several potential sites of glucosepane formation in collagen type I fibrils have been predicted theoretically by molecular modeling, but direct proof by tissue analysis is lacking (10,11).
Hyperglycemia accelerates the age-related decrease in connective tissue quality. One of the biochemical hallmarks of diabetes is tissue stiffening thought to be caused by AGE crosslinking of collagen (12). An accompanying decrease in collagen solubility is consistent with this concept (13). However, reports on the altered material properties of diabetic tendons are inconsistent. For example, conflicting animal model studies have reported both increases (14,15) and decreases (16 -18) in tendon stiffness. Despite decades of research, the pathogenesis of the secondary musculoskeletal effects in diabetes is still poorly understood.
In contrast to stochastic AGE cross-link formation, lysyl oxidase-mediated collagen cross-linking is a biologically con-  cro ARTICLE trolled and site-specific process that appears to plateau at a relatively young tissue age (19,20). For example, in type I collagen of mouse tail tendon, the predominant lysyl-oxidase mediated cross-links are acid-labile aldimine bonds. These form in growing fibrils by the addition of telopeptide Lys aldehydes to Hyl at residues 87 or 930/933 in ␣-chains of adjacent 4D-staggered molecules. In the current study, we examined type I collagen from TallyHo mice, a naturally occurring model of obesity and type 2 diabetes, for glycation sites by peptide MS. The results showed that the same sites targeted by telopeptide aldehydes in lysyl-oxidase driven collagen cross-linking were also the predominant glycation sites.

Diabetic and control mice
Evidence of increased glycation in TallyHo mouse proteins was associated with elevated HbA1c levels in all 5 diabetic mice ( Table 1). All diabetic mice were tested for HbA1c levels Ͼ7% and blood glucose levels Ͼ250 mg/dl. Inclusion criteria for control mice (C57BL6) included Ͻ6% HbA1c and Ͻ200 mg/dl blood glucose. Diabetic mice (36.8 Ϯ 3.2 g; n ϭ 5) were heavier and had notably fattier tissue upon dissection than control mice (33.4 Ϯ 1.5 g; n ϭ 5).

Reduced collagen extractability
Differences in type I collagen extractability in 3% acetic acid from lyophilized tissue samples supported an effect on intermolecular cross-linking in diabetic tendon collagens. Densitometry of stained collagen chains (␣, ␤, and ␥) from SDS-PAGE with sample loads normalized to the dry weight of tendon revealed that total collagen was less extractable from TallyHo mouse tail tendons (30%; n ϭ 4) compared with control ( Fig. 1, left panel). This decrease in extractability was observed equally across all collagen chains (␣, ␤, and ␥). Interestingly, the decrease in extractability was less prominent after pepsin digestion (19%; n ϭ 4) (Fig. 1, right panel). Also notable from the SDS-PAGE band profiles is the lack of any obvious new bands that might arise from new inter-helical stable crosslinks introduced as AGEs.
Screening the telopeptide cross-linking Lys residues from type I collagen showed no evidence of glycation from diabetic or control mouse tendons, only the peptides with unmodified Lys residues. In addition to nonglycated peptides from the three Lys-containing telopeptide domains, six other peptides containing helical Lys sites were also recovered and found to be nonglycated from diabetic mouse tendon. These unmodified sites included Lys at the Xaa position and Lys/Hyl at the Yaa Table 1 Diabetic (TallyHo) and control (C57BL6) mice weights, HbA1c levels, and pyridinoline content TallyHo mice are a naturally occurring obesity-derived model of type 2 diabetes generated from selective mouse breeding. Inclusion criteria for the diabetic mouse included HbA1c level Ͼ7% and blood glucose levels Ͼ250 mg/dl. Concentration of pyridinoline cross-linking residues in diabetic and control mouse tail tendon expressed as mol/mol of collagen. HP, hydroxylysyl pyridinoline.

Mouse tail tendon
Mass

Figure 1. SDS-PAGE reveals a decrease in collagen extractability in diabetic mouse tail tendon.
Left panel, acid labile aldimine cross-links are broken with mild acetic acid treatment, allowing native type I collagen monomers to be extracted from the tissue. TallyHo mouse tendon was less acidextractable than control mouse tendon collagen (ϳ30%; n ϭ 4). Right panel, pepsin digestion showed a similar but slightly reduced effect (ϳ19%; n ϭ 4), between control and diabetic tissues. These findings support the presence of AGE cross-links. Sample loads were normalized to the dry weight of tendon prior to loading. Densitometry was performed to quantify stained collagen chains (␣, ␤, and ␥) on SDS-PAGE. The disparity between acid and pepsin extraction also supports a possible alteration to acid labile cross-links in Tal-lyHo mouse.

Nonenzymatic collagen glycation sites
position of the Gly-Xaa-Yaa repeat domains of the ␣1(I) and ␣2(I) chains. Although Lys/Hyl glycation might be theoretically random in nature, the results clearly show restriction to the helical domain basic motifs that characterize the sites of lysyl oxidasegenerated aldehyde interaction. Their percentage occupancy (%) and the sites of glycated Lys residues in type I collagen from diabetic and control mouse tendon are summarized in Fig. 6.

Glycated lysine residues are excluded from normal collagen cross-linking
The fructosyl-Hyl glycations described here and lysyloxidase-controlled collagen cross-linking are mutually exclusive processes. Telopeptide Lys aldehydes and reducing sugar carbonyls interact with the ⑀-amine of helical domain Lys or Hyl residues. Not surprisingly, only nonglycated aldimine cross-links were identified in borohydride-reduced diabetic tendon. We were also unable to identify a glycated form of the ␣2(I)Lys 87 -to-C-telopeptide aldimine dimeric crosslinked peptide, which might hypothetically have been 30% gly-cated at ␣2(I)Lys 90 based on the precursor linear peptide results (Fig. 7). The absence of such a glycated cross-linked peptide suggests that the described glycations are able to both chemically and sterically hinder physiological collagen cross-linking involving telopeptide Lys aldehydes (Fig. 8). In the case of ␣2(I)Lys 90 , glycation cannot occur after a telopeptide has crosslinked to the neighboring ␣2(I)Lys 87 , presumably because of steric hindrance.

Pyridinoline cross-link analysis
As an independent downstream measure of the effect of nonenzymatic glycations on cross-linking Lys residues, pyridinoline cross-links were quantified after acid hydrolysis of mouse tail tendon. Pyridinoline cross-links are only minor cross-links in mouse tail tendon collagens; however, their fluorescence does provide a convenient quantitative index of cross-link content (26). The results revealed a small decrease in hydroxylysyl pyridinoline (HP) cross-links in diabetic tissue (0.036 mol/mol of collagen; n ϭ 5) compared with control (0.041 mol/mol of collagen; n ϭ 5). This 13% lower HP content in diabetic versus control tendon inversely correlates closely with the increased levels of glycation (ϳ10 -20%) at the cross-linking residues ( Table 1).

Discussion
Diabetes has become a global pandemic. By recent estimates, over 8% of adults worldwide are affected (27). There is abundant clinical evidence of the adverse effects of hyperglycemia on connective tissues including restricted joint mobility, increased risk of injury, and impaired healing (28 -30). An accumulation of AGE cross-links is thought to be an underlying factor in affected tissues (14). In fact, tissue stiffness, mechanical fragil-

Nonenzymatic collagen glycation sites
ity, increased failure loads, and a higher collagen denaturation temperature have all been associated with elevated intermolecular AGE cross-links (17,31). However, no direct experimental evidence for a single primary glycation adduct or molecular site of AGE cross-linking resulting from glucose addition in vivo has yet been reported for any collagen type.
Using peptide-targeted tandem MS, the current results reveal the primary sites of glycation by glucose in type I collagen. Unexpectedly, the only significant glycation sites that could be found coincided with those that undergo cross-linking by the lysyl-oxidase cross-linking pathway. Although the literature has focused on Lys residue glycations (32), the present study identified only fructosyl-Hyl glycations, because the glycated sites are 100% Hyl in tendon type I collagen. However, in analyses of type I collagen from other tissues and from mouse models of lysine underhydroxylation (22), both lysine and hydroxylysine peptide isoforms were glycated without preference in proportion to their inherent synthetic ratio at each site (data not shown). In addition to a unique MS/MS profile described in the results, the structure of fructosyl-Hyl was also supported by a 2-Da mass increase of the glycated peptides when tissue was collagenase-digested and analyzed after borohydride reduction. This is consistent with the conversion of the keto-imine fructosyl-Hyl to the secondary amine product (data not shown).
Our findings imply that glycations have the potential not only to go on to form AGE cross-links in proteins but also to chemically hinder normal lysyl oxidase-mediated collagen cross-linking. Specifically, glycation of helical domain crosslinking Hyl residues (Lys 87 and Lys 930/933 ) will prevent the modified amino-group from cross-linking to a telopeptide-generated aldehyde. In other words, both collagen telopeptide aldehydes (normal) and glucose carbonyls (abnormal) target the same helical domain cross-linking lysine ⑀-amine side chains and are therefore mutually exclusive (Figs. 8 and 9). Steric hindrance is also indicated, because ␣2(I)Lys 90 glycation apparently physically blocked the neighboring ␣2(I)Lys 87 from becoming cross-linked (assuming that in the mouse, ␣2(I)Lys 90 is not an alternative binding partner for a telopeptide aldehyde). The absence of a glycated aldimine cross-linked peptide from this site in the collagenase digests supports this interpretation. Diabetic rat tail tendons were previously shown to have significantly less of the reducible collagen cross-link, hydroxylysino ⌬ norleucine, compared with age-matched control tendon (33). This is consistent with the concept that elevated glycation levels in diabetic collagen can inhibit lysyl-oxidase derived cross-link formation. The lower pyridinoline content of hyperglycemic mouse tendon (Table 1) is also consistent with decreased lysyl oxidase cross-linking.
Glycations have been shown to increase the mean lateral distance between collagen molecules in tendon fibrils (34), which could distress the supramolecular packing of the fibril structure and so indirectly reduce the cross-linking potential. Also, if glycation was most concentrated on collagen molecules at fibril surfaces, as is likely from size-exclusion considerations, this could prevent newly made collagen molecules from cross-linking to existing fibrils during tissue growth and remodeling. Aldimine cross-links and free Lys aldehydes are likely to be in a continual state of equilibrium on the surface of collagen fibrils, even in mature tendons. Indeed, it has been demonstrated in tendon construct cultures that inhibiting lysyl oxidase activity after 1 month in culture resulted in mechanical failure of the entire fabric (35).
The current study also adds support to a proposed association between glycation and tissue stiffening. An observed decrease in collagen extractability from diabetic tendon is consistent with the notion that glycation goes on to introduce intermolecular AGE cross-links (13,33). The decrease in collagen extractability implies an increase in acid-stable intermolecular cross-links. The smaller effect on collagen solubility upon pepsin digestion compared with acid extraction supports the addition of inter-triple-helical cross-links in diabetic mouse tendon (although no evidence was noted for such bonds in the form of new bands in Fig. 1). Accumulated AGE cross-links have been associated with both increased (14,15) and decreased (16 -18) tendon stiffening. However, there is also other evidence to indicate that AGEs do not affect individual collagen fibril stiffness (36) but may, in fact, significantly dimin-  ). B, the Lys 90 site from diabetic mouse tendon is significantly more glycated than control (ϳ30% fructosyl-Hyl; 433.2 4ϩ , 577.2 3ϩ , 864.8 2ϩ ). C, MS/MS reveals that the sugar attachment is located on residue Lys 90 . The blue hexagon indicates fructosyl-moiety; K* indicates Hyl.

Nonenzymatic collagen glycation sites
ish overall tendon viscoelasticity (37). Nevertheless, it is clear that factors that impact collagen synthesis, turnover, or enzymatic cross-linking can also impact tissue mechanical properties. A reduced number of lysyl oxidase-mediated cross-links coupled with an increase in AGE cross-links, dependent on where the latter were placed, could have the net effect of both reducing the strength and increasing the stiffness of tendon.
Neighboring charged amino acids seem the most likely drivers of glycation at Lys 87/90 and Lys 930/933 . For example, Lys side chains with decreased pK a values are proposed to react prefer-   (47). It should be noted that in mouse, ␣2(I)Lys 90 is glycated rather than ␣2(I)Lys 87 . In rat and human, there is no Lys at position 90. Glycation of helical domain cross-linking residues appeared to be selective, because all other peptides we recovered by LC-MS that contained Lys and/or Arg residues from the triple helical and telopeptide regions of both ␣-chains proved to be nonglycated. J indicates pyroglutamic acid. Figure 7. Glycated collagen hydroxylysines are sterically or chemically prohibited from participating in collagen cross-linking. Identification from tendon of an aldimine cross-linked peptide between collagen (␣2(I)Lys 87 and an ␣1(I) C-telopeptide) is illustrated. A and B, MS profile from borohydride-reduced control and diabetic tendon reveals only a nonglycated reduced aldimine structure (712.7 5ϩ , 890.6 4ϩ , and 1186.9 3ϩ ). C, MS/MS fragmentation spectrum of the parent ion (712.7 5ϩ ) supports the structure of the reduced aldimine cross-link.

Nonenzymatic collagen glycation sites
entially with glucose (38). Nearby negative carboxyls on Glu or Asp are predicted to lower the pK a of Lys residues (38) and mechanistically help catalyze the subsequent Amadori rearrangement by stabilizing the Schiff base intermediate. Nearby basic residues (Lys, Arg, and His) also promote Lys residue glycation (39). As candidate glycation sites, the helical domain cross-linking Lys residues are ideally located adjacent to Lys, Arg, and His residues. This consensus sequence is evident at every helical domain cross-linking site (mouse sequences shown): ␣1(I)Lys 87 (GMKGHR); ␣1(I)Lys 930 (GDRGIKGHR); ␣2(I)Lys 87 (GFKGVKGHS); and ␣2(I)Lys 933 (GHRGLPGLK). Clearly, the same local factors that have evolved to favor telopeptide aldehyde additions also favor glucose carbonyl additions.
Simulations of atomistic models have recently been used to predict theoretical sites of AGE cross-linking in collagen (10,11,40). In these studies, both ␣1(I)Lys 87 (10) and ␣2(I)Lys 87 (11) came up as prominent predicted sites of glycation. Our results show that ␣2(I)Lys 90 was favorably glycated over crosslinking residue Lys 87 ; however, this Lys (GFKGVKGHS) appears to be unique to the mouse sequence and in other higher mammals, including rat and humans, is replaced by Arg (GFKGIRGHN). We have evidence that supports ␣2(I)Lys 87 being the residue glycated in higher mammals other than mouse (data not shown).
Protein glycation is a rapid and, initially, potentially reversible reaction (42), with the degree of glycation being proportional to the level of glucose in the blood. The product from glucose (fructosyl-Lys) can break down to carboxymethyl-Lys, which has been predicted to cause tissue damage through its ability to chelate transitional metals (43). The latter AGE can also produce reactive oxygenated species (43). Unlike the initial glycation product, the predicted AGE cross-links are not reversible (8). In future studies, it will be important to identify specific sites of the major AGEs, such as glucosepane predicted to bridge Lys and Arg side chains, and determine whether any are intermolecular and therefore potentially detrimental to tissue properties. It will also be important to understand at the level of the fibril and suprafibrillar architecture whether glycation is unevenly distributed, for example concentrated on the surface more than the interior of fibrils.

Diabetic mice
All animal work was performed with institutional approval by the University of Colorado Anschutz Medical Center Institutional Animal Care and Use Committee. Eight-week-old TallyHo mice (n ϭ 5) (TALLYHO/JngJ; stock no. 005314) and control mice (n ϭ 5) (C57BL6/J; stock no. 000664) were purchased from The Jackson Laboratory (Bar Harbor, ME). Tal-lyHo mice are a naturally occurring polygenic model of obesity and type 2 diabetes derived from the selective breeding of hyperglycemia in mice over multiple generations. Only male mice were used in this study (the disease is not penetrant in females). The animals were kept under standard conditions until euthanasia at 26 weeks. The animals were maintained on a standard chow diet (catalogue no. 2920X; Envigo, Madison, WI), sterilized by irradiation, and given free access to food and water. Body mass was measured weekly. Blood samples were drawn to measure blood glucose and HbA1c levels at 17 weeks of age and at euthanasia. Inclusion criteria for the study included HbA1c levels (Ͼ7%) and blood glucose levels (Ͼ250 mg/dl). The study was not blinded (TallyHo mice are white, and the C57BL/6J mice are black).

Collagen extraction
Tail tendon collagens were characterized in TallyHo (test) and C57BL6 (control) mice. Intact type I collagen was solubilized from the tissues by acid extraction in 3% acetic acid for 24 h at 4°C. Total collagen was extracted by limited pepsin digestion. Tissues were also digested with bacterial collagenase with and without borohydride reduction, and total collagenase digests were resolved into peptide fractions by C8 reversephase HPLC (44). Acid-extracted and pepsin-extracted colla- Figure 8. Lysyl oxidase-derived collagen cross-linking altered by hydroxylysine glycations. Type I collagen cross-linking in tendon uses Lys/ Hyl aldehyde precursors in the telopeptide domains. In growing fibrils, the aldehydes target specific Lys or Hyl residues at helical domain sites (Lys 87 and Lys 930 /Lys 933 ) in neighboring collagen molecules. Similarly, glucose can covalently bind to Lys and Hyl side chains through the same ⑀-amino group, and if the same residues were the preferred target, they could be mutually exclusive processes.

Nonenzymatic collagen glycation sites
gen ␣-chains were resolved by SDS-PAGE and stained with Coomassie Blue R-250. Collagen extractability was determined based on band intensities using National Institutes of Health ImageJ software as previously described (45).

Collagen cross-linking analysis
The collagen pyridinoline cross-link content was determined by fluorescence monitoring with reverse-phase HPLC. Pyridinoline cross-links were analyzed in mouse tail tendon by HPLC after acid hydrolysis in 6 M HCl for 24 h at 108°C. Dried samples were dissolved in 1% (v/v) n-heptafluorobutyric acid for quantitation of HP by reverse-phase HPLC and fluorescence monitoring as previously described (46).

Mass spectrometry
Glycations (fructosyl-Lys and fructosy-Hyl) and glycosylations (glucosylgalactosyl-Hyl and galactosyl-Hyl) were quantified at specific sites in collagen ␣-chains as previously described (47). Collagen ␣-chains were cut from SDS-PAGE gels and subjected to in-gel trypsin digestion. Tendons were also digested with bacterial collagenase, with and without borohydride reduction, and resolved by C8 reverse-phase HPLC prior to analysis by MS (44,48). Electrospray MS was carried out on the trypsin-and collagenase-digested peptides using an LTQ XL linear quadrapole ion-trap mass spectrometer equipped with in-line Accela 1250 LC and automated sample injection (ThermoFisher Scientific). Proteome Discoverer software (ThermoFisher Scientific) was used for peptide identification. Tryptic peptides were also identified manually by calculating the possible MS/MS ions and matching them to the actual MS/MS spectrum using Thermo Xcalibur software. Differences in post-translational modifications were determined manually by averaging the full scan MS over several LC-MS minutes to include all the post-translational variations of a given peptide. Protein sequences used for MS analysis were obtained from the Ensembl genome database.  . Model of altered collagen cross-link formation in a glycated tendon. A, in the fibril, collagen molecules are spatially staggered in an arrangement that promotes ordered intermolecular cross-linking. B, in normal mouse tendon, type I collagen initially forms intermolecular aldimine cross-links between helical domain Hyl and telopeptide Lys aldehydes. In diabetic collagen, partially glycated helical domain cross-linking Hyl are sterically and/or chemically hindered from participating in normal collagen cross-linking chemistry. The net effect could be a tendon with compromised material properties resulting from fewer physiologically stable cross-links yet subsequent tissue stiffening from potential inappropriately placed AGE cross-links in the fibrillar matrix architecture.