Sites of non-enzymatic crosslinks pentosidine and glucosepane in collagen I of human cortical bone

Abstract

Structural alterations of bone extracellular matrix (ECM) may affect bone quality, which, along with bone mass, affects the ability of bone to resist fracture. In particular, an increase in advanced glycation end-product (AGE) crosslinking of type I collagen has been implicated in contributing to increased bone fragility in diabetes and ageing. However, the location of these crosslinks and their relative abundance is unknown, thus hindering our understanding of how AGE crosslinks impact bone quality. We analyzed physiological crosslinks, glucosepane (GN) and pentosidine (PE), in collagen I from cortical bone of 12 male and 12 female cadaveric femurs using liquid chromatography-tandem mass spectrometry. The analysis identified 11 distinct crosslinks at 8 different sites within the collagen I triple helix. Crosslinks at two sites formed only within the same collagen chain, while the remaining crosslinks formed within the same triple helical molecule and/or between different molecules of the neighboring microfibrils. Most of the GN and PE crosslinks were located to the D-periodic overlap zone of the microfibril. The relative crosslink levels varied significantly at different sites: 0.005% to 2.1% for PE and 0.06% to 24.9 % for GN. Female donors had fewer crosslink sites compared to male donors. While total PE and total fAGE levels did not correlate with bone mechanical properties in the present sample size, the level of GN at one site positively correlated with ultimate stress, while the level of PE at another site negatively correlated with post-yield toughness. However, these correlations were weak, and several other correlations between site-specific GN and PE levels and these mechanical properties were not consistent between the sexes. ln conclusion, this study identified the native locations of GN and PE within collagen I of bone, an innovation essential for addressing the question of how the specific structural features of bone ECM affect bone quality.

1. Introduction

One of the factors thought to affect bone quality in ageing, diabetes, osteoporosis and other diseases of bone is non-enzymatic post-translational crosslinking modifications of bone extracellular matrix (ECM). These crosslinks derive from glucose and glucose oxidation products and constitute a subclass of advanced glycation end-products (AGEs). Even though multiple AGE crosslinks have been found in human tissues such as crosslines, vesperlysines, glyoxal-lysine dimer (GOLD), methylglyoxal lysine dimer (MOLD), and 3-deoxyglucosone-derived imidazolium (DOGDIC) [1], glucosepane (GN) and pentosidine (PE) are highly relevant to disease since GN is the most abundant AGE crosslink found in human and animal tissues [2] and since PE is commonly used as a surrogate marker of glycation and glycoxidation in vivo [3]. Both PE and GN were initially discovered in model proteins incubated with glucose, followed by their identification in animal and human tissues [2, 4, 5]. The levels of these crosslinks have been shown to significantly increase in diabetes, ageing and other diseases [2, 6–8].

Unlike enzymatic crosslinks, accumulation of AGE crosslinks is thought to have detrimental effects on the fracture resistance of bone [9]. Clinical studies have revealed associations between the elevated serum and/or urinary PE levels and the increased risk of bone fracture [10–13]. Serum GN levels have been shown to correlate with progression of osteoarthritis in an animal model [14]. Specific mechanisms by which AGE crosslinks cause bone fragility are poorly understood but their accumulation may impede the deformation capacity of collagen fibrils and/or cause structural changes to type I collagen that favor the formation of damage [15]. Investigations of the crosslink effect on bone mechanics are impeded by the unknown locations of AGE crosslinks within bone ECM.

As examples from studies of collagenous soft tissues, elevated GN levels contributed to reduced density of collagen packing [16]; GN- and PE-crosslinked collagen fibrils formed thicker bundles with reduced stress relaxation rates and resisted remodeling by macrophages [17]; and in vitro glycation reduced the sliding between neighboring collagen fibrils and decreased stress relaxation of rat tail tendons loaded in tension at incremental levels of applied strain [18]. Glycation disrupted the longitudinal ordering of the molecules in collagen fibrils and caused the fibrils to have a more negative charge [19]. In silico modeling predicted destabilizing effect of GN on collagen I structure near the integrin and proteoglycan binding sites, which may disrupt key protein-protein interactions involving collagen [20]. Models of GN-modified collagen also predicted increased hydration, which may result in fibril swelling [21], and increase in the tensile and the lateral Young’s moduli at low strains, thus potentially contributing to fibrillar stiffening [22]. High levels of AGEs, particularly GN, caused decrease in energy dissipation via friction between sliding tropocollagen molecules, resulting in a more abrupt fibrillar failure [23]. Moreover, the structure of AGE crosslinks, along with their levels, was an important contributor to changes in mechanical behavior of collagen fibrils in a computational model [24].

Because the hard mineral exists within the soft collagen fibrils and also between the mineralized collagen fibrils in bone, the apparent stiffening and viscoelastic dampening effect of AGE crosslinks on bone fragility is not straightforward. The effects of crosslinking on collagen structure and mechanical properties would likely depend on specific crosslink microenvironment and on crosslink abundance at the specific collagen sites. However, information about physiological crosslink sites within collagen I triple helix and collagen fibril is limited. Physiological post-translational modification (PTM) sites in collagen I have been investigated so far only for enzymatic PTMs, i.e. hydroxyproline, hydroxylysine, galactosyl-hydroxylysine and glucosylgalactosyl-hydroxylysine [25, 26]. For GN, some collagen sites have been predicted in silico using proximity and energy minimization criteria [27, 28]; however, we are unaware of any studies experimentally demonstrating the presence of GN in bone matrix. As a proof of principle, GN sites have been experimentally identified in bovine ribonuclease A, a model protein modified by glucose in vitro [29], but this study has little physiological relevancy. Site-specific PTM databases are overwhelmingly focused on enzymatic PTMs and the information about the sites of non-enzymatic PTMs, including AGEs, is lacking [30]. To address this knowledge gap, we employed liquid chromatography mass spectrometry (LC-MS) to produce a quantitative map of GN and PE crosslinks in collagen I from male and female bone donors. The map revealed sites of the intra- and intermolecular crosslinks, differences and similarities between male and female donors, and crosslink locations within collagen I fibril with assistance from molecular modeling.

2. Materials and Methods

2.1. Bone samples.

Male and female donors were between 58 and 96 years of age with median (interquartile range) of 80.0 (66.8, 91.3) and between 35 and 94 with median (interquartile range) of 73.3 (58.0, 84.0), respectively. The donors had no history of diabetes. Upon receiving the fresh-frozen femurs from donor services (NDRI, Philadelphia, PA and MTF Biologics, Edison, NJ), they were stored at −80°C. Using a low speed, water-irrigated, diamond-embedded, circular cutting saw (South Bay Technology Inc, Model 660, San Clemente, CA, USA) strips of cortical bone were extracted from the medial quadrant of each femur mid-shaft at room temperature. The periosteal and endosteal surfaces of the cortical strips were milled using a custom fixture to machine a uniform, rectangular beam specimen (nominal dimensions: L=50.1 ± 2.13 mm × B= 4.04 ± 0.15 mm × W= 5.03 ± 0.23 mm). Two 2-mm thick cross-sections were cut from the distal end of the beam for PE measurements by high-performance liquid chromatography (HPLC) and for mass spectrometry (MS) experiments.

2.2. Total pentosidine and fAGE measurements.

These commonly reported bulk AGE markers were used in correlation analyses for comparison with the site-specific crosslinks. As previously described [31], a bone section per donor (~30 mg) was demineralized in 20% ethylenediamine tetraacetic acid (EDTA) at 4 °C for 5 weeks. The samples were then hydrolyzed in 6 N HCl at 110 °C for 22 h. After hydrolysis, the hydrochloric acid was evaporated using a SpeedVac Concentrator System (ThermoFisher) with a cold trap. Next, each hydrolysate was resuspended in HPLC-grade water and centrifuged at 15,000g and 4°C for 20 min. Supernatant aliquots corresponding to ~1 mg of dry bone was used for each assay.

The pentosidine (PE) crosslinks were measured using a HPLC system (Agilent 1260 Infinity, Agilent Technologies, Santa Clara, CA) equipped with a Spherisorb 4.6 × 150 mm column (Waters Co, Milford, MA). The HPLC conditions were as follows: buffer A was 0.22% heptafluorobutyric acid (HFBA), buffer B was 100% acetonitrile, and the gradient program was 0.00–4.50 min, 13% B; 4.75–10.00 min, 17% B; 10.25–16.00 min 25% B; 16.75–21.00 min, 100% B; and 21.25–26.00 min, 13% B. The PEN was quantified using fluorescent detector at λ_ex/em = 328 nm/378 nm and a PEN standard (Cayman Chemical, Ann Arbor, MI). Pyridoxine (Sigma, St. Louis, MO), used as an internal standard, was quantified at λ_ex/em = 295 nm/420 nm within the same HPLC run. The PEN values were calculated based on the standard curve and normalized to collagen levels measured in the same samples.

A known fraction of each hydrolysate and quinine sulphate standards were dissolved in 0.1 M H₂SO₄ and analyzed as previously described [32]. Briefly, 100 μL of standard and sample solutions was pipetted in duplicate into a 96-well plate. Fluorescence was measured at λ_ex/λ_em = 370/440 nm using BioTek Synergy H1 microplate reader. The levels of fAGEs were calculated based on the standard curve and normalized to collagen levels measured in the same samples.

Hydroxyproline (Hyp) levels were measured by a colorimetric assay as described in Stoilov I et al. [33]. Briefly, the assay was performed in 96-well plates following the addition of the oxidizer and Ehrlich's solution to Hyp standards and the samples. The color was developed upon the incubation of the plates at 65 °C for 20 min and consequent rapid cooling. The measurements were taken at 550 nm using Synergy H1 microplate reader (BioTek Instruments, Winooski, VT). Collagen levels were calculated based on 290 Hyp residues per triple helical collagen molecule [34]. This colorimetric assay has been previously validated using HPLC [35].

2.3. Extraction of bone proteins.

Extraction of proteins from cortical bone samples was carried out as previously described [34], with minor changes. Briefly, the other bone section (~40-mg) per donor was pulverized under liquid nitrogen using a SPEX 6770 freezer/mill (SPEX SamplePrep LLC, Metuchen, NJ). The powdered samples (30 mg) were mixed with 0.5 mL of the 10 mM Tris buffer, pH 7.5, containing 6 M guanidinium hydrochloride (GdnHCl), 50 mM EDTA and Halt’s protease inhibitor cocktail. Extraction was carried out for 72 h at 4°C under continuous rotation. Supernatant was collected following centrifugation at 13,000g for 15 min at 4°C and subjected to three cycles of washing with 10 mM Tris buffer using 3 KDa Amicon Ultra-4 centrifugal filtration unit to achieve complete buffer exchange. The total protein concentration in the extracts was determined using Pierce^™ BCA protein assay kit (ThermoFisher Scientific, Waltham, MA). This concentration was used to guide the sample preparation for liquid chromatography-tandem mass spectrometry (LC-MS²) experiments.

2.4. Incubation with glucose.

This in vitro experiment was conducted to assess the feasibility of site-specific identification of AGE crosslinks at natural abundances in vivo. An aliquot of extracted bone protein was incubated with 100 mM glucose in 70 mM sodium phosphate buffer, pH 7.5 at 37°C for 60 days; the control incubation did not contain glucose. Because oxidation is not required for glucosepane formation, 1 mM diethylenetriaminepentaacetic acid (DTPA) was added to the incubation buffer to inhibit production of oxidative AGEs, e.g. carboxymethyllysine.

2.5. Mass spectrometry (LC-MS²).

The site-specific modifications in bone collagen I were analyzed as described previously [36], with some changes. Samples of bone protein extracts were diluted with 100 mM ammonium bicarbonate and reduced with dithiothreitol. Cysteine residues were carbamidomethylated with iodoacetamide, and two separate aliquots from each sample were digested with either sequencing-grade trypsin (Promega) or ArgC protease (Genovis) for 8 h at 37°C. Following digestion, the resulting two separate peptide preparations were desalted, reconstituted with 0.1% formic acid, and loaded onto a capillary column packed with 20 cm of C18 reverse phase material (Jupiter, 3 μm beads, 300Å, Phenomenex). Peptides were loaded onto this capillary, reverse phase, analytical column (360 μm O.D. × 100 μm I.D.) using a Dionex Ultimate 3000 nano-LC and autosampler. The mobile phase solvents consisted of 0.1% formic acid, 99.9% water (solvent A) and 0.1% formic acid, 99.9% acetonitrile (solvent B). Peptides were gradient-eluted at a flow rate of 350 nL/min using a 90-minute gradient. The gradient consisted of the following steps: 2%−35% B from 1–73 min; 35%−95% B from 73–78 min; constant 95% B from 78–79 min; 95% B back to 2% B from 79–80 min; constant 2% B from 80–90 min (column re-equilibration).

A Q Exactive Plus mass spectrometer (Thermo Scientific), equipped with a nanoelectrospray ionization source, was used to mass analyze the eluting peptides using a data-dependent method. The instrument method consisted of MS1 using an MS AGC target value of 3×10⁶, followed by up to 20 MS² scans of the most abundant ions detected in the preceding MS scan. A maximum MS/MS ion time of 100 ms was used with a MS2 AGC target of 1×10⁵. Higher-energy collisional dissociation was set to 27 NCE, and peptide match, isotope exclusion, and dynamic exclusion (10s) were enabled. The original mass spectra are publicly available [37–39].

Crosslinked peptides were identified in pLink3 [40] against a 189 protein database derived from previous Sequest searches [36]. Glucosepane and pentosidine chemical composition and linker masses were added in pConfig for the search. A multi-process workflow was performed with tryptic digestion allowing for four missed cleavages, peptide mass range 600 to 6000 Da, peptide length 6 to 60, MS1 tolerance of 10 ppm, and MS2 tolerance of 20 ppm. Hydroxyproline and oxidized methionine were selected as variable modifications. Results were filtered using a separated false discovery rate (FDR) set to 10% at the spectra (rCSMs) level. Opening of FDR to 10% identified additional candidates from the bone matrix proteome sub-set, which were validated manually to ensure accurate assignments. No E-values were calculated, and no quantitation step was included in the workflow. Annotated spectra were visualized in pLabel.

For data analyses, a set of GN and PE-crosslinked peptide pairs was constructed based on previous in silico predictions [27, 28]. The searches were configured to include non-enzymatic lysine-arginine residue crosslink modifications glucosepane (GN, ΔM=108.0211) and pentosidine (PE, ΔM=57.9844) (see Fig. 1 for structures). As previously described [36], the additional searches accounted for enzymatic PTMs commonly present in bone ECM, i.e. hydroxylated lysine and proline residues (ΔM=15.9949), galactosyl (ΔM=178.0477) and glucosylgalactosyl (ΔM=340.1006) hydroxylysine residues, as well as for modifications derived from sample processing, i.e. carbamidomethylation (ΔM=57.0215) of cysteine and oxidation (ΔM=15.9949) of methionine. The LC-MS² data were manually analyzed for the presence of the predicted precursor ions (mass tolerance of 4 ppm) and corresponding MS² product ions (mass tolerance of 20 ppm); all the identified crosslinks were validated by manual spectral inspection in every donor.

Figure 1. Structures of glucosepane and pentosidine crosslinks.

The contributing structures of lysine and arginine sidechains are indicated by the shaded areas. The carbon atoms in the 7-membered and 6-membered ring structures, respectively, are derived from a glucose molecule. The modification monoisotopic mass differences are shown below the structure names.

Accurate mass measurements of the tryptic and ArgC peptides were used to generate extracted ion chromatograms (XICs). A window of 10 ppm around the theoretical monoisotopic m/z values of the observed precursor ions was utilized for making XICs of the unmodified and crosslinked peptide forms. Using QualBrowser (Xcalibur 4.0), the integrated area under the curve (AUC) for each XIC peak was determined. The relative abundance of each crosslinked peptide pair was then calculated as a percentage of the summed AUC obtained for all the identified crosslinked and unmodified forms of these peptides.

2.6. Molecular modeling.

Backbone coordinates for the 30-mer collagen triple helix (1CGD) [41] were computationally mutated to create two 3D models of human triple helical collagen I spanning α1 chain residues 248–278 and 733–763 (see Fig. 2 for residue numbering). The glucosepane (GN) lysine-arginine crosslink structure was constructed using the program Avogadro [42]. Covalent bonds and electrostatic interactions were energy minimized using the universal force field (UFF), a broadly applicable classical force field, which contains parameters for most atoms in the periodic table [43]. The gradient descent method, an iterative optimization algorithm, was utilized to find the local energy minimum of the structure. The individual collagen helices were manually positioned using PyMOL [44] to achieve a configuration that allowed for formation and visualization of the GN crosslink while avoiding steric clashes. Placement was guided by ensuring that the helices maintained a distance between 5–10 Å to enable interhelix crosslink formation while avoiding steric hindrance between side chains and backbone atoms. The placement of the helices was refined iteratively, based on visual inspection, to achieve a balance between structural integrity and crosslink viability. Electrostatic surface potentials were rendered using the APBS (Adaptive Poisson–Boltzmann Solver) PyMOL plugin [45]. Partial charges and atomic radii were calculated with the Schrodinger, Inc. Protein Preparation Workflow plugin [46]. Ray traced images were generated using PyMOL. Surface area and volume calculations were conducted with ChimeraX [47].

Figure 2. Location of the glucosepane (GN) and pentosidine (PE) non-enzymatic crosslink sites in type I collagen from human male bone.

Residue numbering for α1 (I) chain is shown above the sequence and numbering for α2 (I) chain is shown below the sequence. Non-triple helical N-terminal and C-terminal telopeptides are indicated by black solid line boxes. Amino acid position numbering is given for full length collagen α1(I) or α2(I) chain sequences (UniProtKB – P02452, CO1A1_HUMAN and P08123, CO1A2_HUMAN, respectively). The triple helical sequence regions that were not detected by LC-MS² are highlighted in gray; the overall sequence coverage of collagen I triple-helical region was 95.7% and 90.3% for α1 and α2 chains, respectively. The guide sequence numbering is shown in green font above (α1(I) chains) or below (α2(I) chain) the sequences. The glucosepane- and/or pentosidine-crosslinked lysine (K) and arginine (R) residues are shown in red font with the crosslinks at the specific sites identified by numbers as follows: #1: α1a K265 – α1a R268; #2: α1b R268 – α2 K654; #3: α2 K380 – α1a R469; #4: α2 R573 – α2 K1064; #5: α1b K742 – α1a K745; #6: α2 R671 – α1a K759; #7: α2 K1014 – α1a R1105; and #8: α2 K1064 – α2 R1067. The crosslink color coding is as follows: purple - PE and GN; blue - PE; orange - GN. See Tables S1–S13 for MS² identification of the crosslinked peptide pairs at specific collagen I sites and Figure S3 for site-specific quantitation of the crosslink levels. Note 1: crosslinks #2 and #4 can only exist as interhelix crosslinks (see Fig. 7); they are shown here within a single triple helix for clarity. Note 2: mass spectrometric analysis cannot distinguish between the two α1 chains within the triple helix, therefore, when necessary, we utilized the published in silico modeling data to determine the exact crosslink positions (see manuscript text for details).

Density functional theory (DFT) calculations were performed using Gaussian 16 [48]. To compute the electrostatic potentials (ESPs). Geometry optimization was carried out at the B3LYP/6–31G* theory level, with Merz–Kollman (MK) population analysis employed with the ESP specific keyword IOP(96/33=2) to derive properly fit atomic charges. Upon convergence, a subsequent frequency analysis was conducted to confirm that the optimized structure had no imaginary frequencies, indicating a true minimum on the potential energy surface. Following the successful geometry and frequency convergence, cube files for both the electron density and ESPs were generated using Gaussian's “cubegen” utility with coarse refinement, which provided spatially resolved representations of the electron density and ESPs. Visualization and analysis of the ESP charges were performed using ChimeraX [47], where the ESP data were colorimetircally mapped onto the electron density volumes.

2.7. Mechanical testing.

The mechanical testing of human cortical bone was performed in our recently published study [49], and the selected mechanical parameters were also used here for correlation analysis. Therefore, for experimental details about the preparation of cadaveric bone and testing of mechanical specimens see Berestesky et al. [49]. The bone samples described above came from a section of each single-edge notched beam (SENB) that was loaded-to-failure to determine fracture toughness properties. The section was well outside the region that undergoes crack propagation. The tensile specimens were also extracted from the medial quadrant, next to the strip of bone that was machined into a SENB.

2.8. Statistical analyses.

Statistical analyses were performed using GraphPad Prism 10.1.2 analytical tools. The analysis included an unpaired two-tailed t-test or a nonparametric Mann-Whitney test (based on assumptions about the normality of data distribution), as well as Pearson correlation analysis and Grubbs’ outlier test. To determine if the relationship between a mechanical property of bone and an AGE crosslink or fAGE depended on sex, we used linear regression models in a backward stepwise manner removing the interaction term first and then the sex term if not significant. Thus, the linear regression lines could be different between male and female donors or be one line through the combined female and male data. Outliers were detected in a small number of tests but were not removed from the analyses. The differences were considered significant if the p-value was less than 0.05; corrections for multiple comparisons were made using the Bonferroni method.

3. Results

3.1. Site-specific analysis of glucosepane crosslink in vitro and in vivo.

Identification of specific sites of non-enzymatic AGE crosslinking in vivo is experimentally challenging because the presence of numerous Lys (K) and Arg (R) residues within the α1 and α2 chains of collagen I (Fig. 2) makes it difficult to predict mass-to-charge ratios (m/z) of the two crosslinked proteolytic peptides without prior knowledge about their proximity within the collagen structure. To circumvent this difficulty, we were guided by the results of the published in silico modeling studies [27, 28]. Previously, several GN crosslink sites were identified in a model protein, bovine RNase A, incubated in vitro with a high concentration of glucose [29]. Therefore, as an initial step, we searched for GN in bone matrix isolated from the mineralized bone and incubated in vitro with 100 mM glucose for 60 days. The analysis of the MS² spectra revealed the presence of GN in the α2 chain of collagen I linking K1064 and R1067 sites (Figure S1 and Table S14). Subsequent analyses of the bone matrix isolated from the male donor demonstrated that GN was present in vivo at the same site (Figure S2). These findings showed the feasibility of site-specific identification of AGE crosslinks at natural abundances in vivo.

3.2. Analysis of major AGE crosslinks in collagen I of human bone.

In the male samples, we identified 11 crosslinks (6 GN and 5 PE) at 8 sites within the collagen I triple-helical region. Thus, at 3 sites (#1, #3 and #8) both GN and PE crosslinks were identified at the same site (purple connecters in Fig. 2). Two crosslink sites were intrachain, i.e. located within a single α1 or α2 chain (Fig. 2 and Fig. 4A). Six crosslink sites were interchain, i.e. connected two different collagen I chains; and two of those were interhelix, i.e. connected two different triple helical molecules (Fig. 2 and Fig. 4 B and C). In the female samples, we identified 6 crosslinks (3 GN and 3 PE) at 5 distinct sites within the collagen I triple-helical region. All 5 sites were interchain with two of them being interhelix; and no intrachain sites were detected in female samples (Fig. 3). For both male and female samples, the corresponding unmodified peptides have also been identified (Tables S1–S13 and S15–S20). The levels of these peptides were used for normalization.

Figure 4. Levels of pentosidine (PN) and glucosepane (GN) at the crosslink sites found in both male and female donors.

The site IDs, crosslink sequence positions and p-values (M vs. F) are shown at the top of each panel. M, male; F, female. The asterisks (*) next to the data points indicate a formal same-donor outlier; the outliers have not been removed from the data analyses unless specifically stated.

Figure 3. Location of the glucosepane (GN) and pentosidine (PE) non-enzymatic crosslink sites in type I collagen from human female bone.

Residue numbering for α1 (I) chain is shown above the sequence and numbering for α2 (I) chain is shown below the sequence. Non-triple helical N-terminal and C-terminal telopeptides are indicated by black solid line boxes. Amino acid position numbering is given for full length collagen α1(I) or α2(I) chain sequences (UniProtKB – P02452, CO1A1_HUMAN and P08123, CO1A2_HUMAN, respectively). The triple helical sequence regions that were not detected by LC-MS² are highlighted in gray; the overall sequence coverage of collagen I triple-helical region was 95.1% and 90.3% for α1 and α2 chains, respectively. The guide sequence numbering is shown in green font above (α1(I) chains) or below (α2(I) chain) the sequences. The glucosepane- and/or pentosidine-crosslinked lysine (K) and arginine (R) residues are shown in red font with the crosslinks at the specific sites identified by numbers as follows: #2: α1b R268 – α2 K654; #3: α2 K380 – α1a R469; #4: α2 R573 – α2 K1064; #5: α1b K742 – α1a K745; and #7: α2 K1014 – α1a R1105. The crosslink color coding is as follows: purple - PE and GN; blue - PE; orange - GN. See Tables S15–S20 for MS² identification of the crosslinked peptide pairs at specific collagen I sites and Figure S4 for site-specific quantitation of the crosslink levels. Note 1: crosslinks #2 and #4 can only exist as interhelix crosslinks (see Fig. 7); they are shown here within a single triple helix for clarity. Note 2: mass spectrometric analysis cannot distinguish between the two α1 chains within the triple helix, therefore, when necessary, we utilized the published in silico modeling data to determine the exact crosslink positions(see manuscript text for details).

Quantitative analysis revealed that the normalized relative abundances of the identified crosslinked peptides varied depending on the site, ranging from 0.006% to 1.4% for PE and from 0.06% to 24.9% for GN in the male samples (Fig. S3). In the female samples, the ranges were from 0.36% to 0.49% for PE and from 4.0% to 22.0% for GN (Fig. S4). These data are consistent with the reported higher levels of GN vs. PE in vivo [50]. At the same time, the identified crosslinked peptides and their levels demonstrated high donor-to-donor reproducibility with only a single crosslinked peptide pair, corresponding to PE at Site #3 (α2 K380-α1a R469), not detected (only unmodified peptides detected) in one male donor (Fig. S3).

3.3. Quantitative AGE crosslink map of bone collagen I from male and female donors.

This map is complimentary to a previously published map of non-crosslinking AGE modifications within the triple-helical region of human bone collagen I [36]. We mapped only two AGE crosslink species, namely GN and PE, because GN is abundant in ECM proteins [2] in vivo and PE is a standard AGE marker [2–4].

In the male map, seven out of the eight crosslink sites coincided with the previously reported non-crosslinking AGE modification clusters (Fig. 2 and [36]). The major differences between male and female maps were the absence of crosslink sites #1 (GN and PE), #6 (GN) and #8 (GN and PE) in the female map (Fig. 3). Otherwise, the remaining crosslinks were located to the same sequence positions as in the male map (Figs. 2 and 3). Interestingly, male collagen I showed positive correlations between total PE levels and crosslink levels at specific sites (sites #1, #2, #6 and #8), while no such correlations were found in female donors (Table S21). The PE levels at site #2 were significantly lower in female vs. male donors, while at site #4 the GN levels were elevated in female vs. male donors (Fig. 4).

3.4. Relationship between crosslink levels and bone mechanics.

For the sample size of the present study, the post-yield toughness from tensile testing and the final J-integral from fracture toughness significantly decreased with donor age (Fig. 5A and 5B). However, both correlations were not significant when accounting for multiple comparisons. Ultimate tensile stress and crack initiation toughness (K_init) did not correlate with age (Table 1). Sex was not a significant covariate in these correlations with age (Table S22).

Figure 5. Multivariate regression analyses for selected correlations (see Table 1).

Correlations of bone mechanical properties (post-yield toughness, ultimate stress and J-integral) vs. the donor’s age and the levels of crosslinks at specific collagen I sites, and correlations between bound water and the donor’s age and total fAGEs, were analyzed by stepwise linear regression as described under Materials and Methods. For those correlations independent of sex, regression lines represent data from the combined male and female donors (panels A, B, D and I). The p-value in the red box (panel G) indicates significant correlation following Bonferroni correction for multiple comparisons.

Table 1. Pearson’s correlation coefficients (p-values) between characteristics of cortical bone and different collagen crosslink sites.

Characteristics included mechanical properties (ultimate stress, post-yield toughness, J-integral and crack initiation toughness (K_init)), volume fraction of bound water (BW), and age for 12 male and 12 female donors, i.e. n=24. The analyses were performed only for the crosslink sites present in both male and female donors. Shown are the original p-values not corrected for multiple comparisons. For selected correlations (shaded boxes), multivariate regression analysis data are shown in Fig. 5.

Property	Age	Ultimate Stress	PY Toughness	J-Integral	Kinit	BW
Age	1	−0.304	−0.555	−0.405	−0.228	−0.534
		(0.1481)	(0.0049)a	(0.0495)a	(0.2830)	(0.0071)a
PE at Site #2	+0.391	−0.120	−0.442	−0.178	0.183	−0.020
	(0.0590)	(0.5773)	(0.0307)a	(0.4058)	(0.3922)	(0.9266)
PE at Site #3	−0.122	+0.115	+0.053	−0.115	+0.225	−0.020
	(0.5706)	(0.5925)	(0.8050)	(0.5935)	(0.2899)	(0.9270)
PE at Site #5	−0.246	See Fig. 5Ha	−0.004	See Fig.5Ea	0.135	+0.347
	(0.2457)		(0.9840)		(0.5290)	(0.0965)
GN at Site #3	−0.283	−0.073	See Fig.5Ga	+0.223	+0.390	+0.423
	(0.1798)	(0.7362)		(0.2949)	(0.0597)	(0.0394)
GN at Site #4	−0.190	+0.548	+0.403	+0.121	+0.253	+0.170
	(0.3732)	(0.0056)b	(0.0510)b	(0.5725)	(0.2323)	(0.4262)
GN at Site #7	−0.035	−0.022	−0.287	−0.228	+0.158	−0.166
	(0.8712)	(0.9170)	(0.1734)	(0.2847)	(0.4616)	(0.4387)
Total fAGEs	+0.067	+0.047	−0.306	−0.299	−0.006	See Fig. 5Fa
	(0.7617)	(0.8305)	(0.1561)	(0.1653)	(0.9789)
Total PE	+0.287	+0.107	−0.315	+0.063	+0.365	+0.154
	(0.1745)	(0.6196)	(0.1338)	(0.7701)	(0.0791)	(0.4727)

^aSee Figure 5 for multivariate regressions.

^bAn outlier influences the strength of the correlation. See Fig. S6.

While total PE and fAGE did not significantly correlate with the mechanical properties of human cortical bone (Table 1), as might be expected from previous studies involving the mechanical testing of cadaveric bone samples [15, 51–53], PE levels at site #2 negatively correlated with post-yield toughness in the combined male and female data (Fig. 5D) in which sex was not a significant covariate (Table S22). PE at site #5 did not however significantly corelate with the energy to propagate a crack to failure (final J-integral) in male donors (Fig. 5E). Unexpectedly, PE levels at this site positively correlated with this mechanical parameter (Fig. 5E) in female donors, but this correlation was rather weak and did not remain significant upon accounting for multiple comparisons. GN levels at site #3 negatively correlated with post-yield toughness in male donors, while they positively correlated with post-yield toughness in female donors (Fig. 5G). The levels of PE at site #5 negatively correlated with ultimate tensile stress in male donors but not in female donors (Fig. 5H). Lastly, there was an outlier in the levels of the GN crosslink at site #4 (marked by asterisk in Fig. 4) that influenced its positive correlation with post-yield toughness (Fig. S6A and S6B) and ultimate tensile stress (Fig. S6C and S6D). Notably, these correlations were not significant at the conservative level of significance required by Bonferroni correction for multiple comparisons.

Bound water levels in the matrix of human cortical bone have also been found to decrease with age [53]. This was verified in the present study (Table 1), though the correlation was driven by the female donors (Fig. 5C). There was a significant interaction between fAGEs and sex such that bound water significantly decreased with an increase in fAGEs in female bone (slope = −0.080, p=0.0258) but not in male bone (slope = +0.026, p=0.1073). With respect to site-specific crosslinks, only GN at Site #3 correlated with the volume fraction of bound water (Table 1). It had weak positive correlation with bound water, so the abundance of this crosslink site in collagen I was not an explanation for the age-related decrease in bound water. In summary, due to insufficient statistical power, almost all correlations were rendered insignificant upon correction for multiple comparisons with a notable exception of the levels of GN at site #3, which showed significant sex-dependent correlations vs. post-yield toughness, even after a conservative Bonferroni correction for multiple comparisons (Fig. 5G).

3.5. Molecular modeling of crosslink sites in collagen I fibril of human bone.

The identified GN crosslinks were modeled in the context of the human collagen I triple helical structure at the specific crosslink sites (Fig. 6). Three different GN crosslink configurations were modeled: the intrachain GN (sites #1 and #8), which links Lys and Arg residues within the same polypeptide chain (Fig. 6A); the interchain GN (sites # 3, #5, #6 and #7) connects two chains within the same triple helix (Fig. 6B); and the interhelix GN (sites #2 through #7) connects two different collagen triple helices (Fig. 6C). The modeling confirmed that the triple helix can accommodate structural requirements for the crosslink formation at the identified collagen I sites. The main structural consequence of GN crosslink formation was the increase in steric bulk (Fig. 6D) and the decrease in overall electropositivity at the crosslink site due to replacement of electropositive Lys and Arg side chains with hydrophobic seven-membered ring containing two electronegative hydroxyl groups (Fig. S5). PE was more electropositive compared to GN with a relatively smaller increase in the steric bulk (Fig. S5). Gibbs free energy (ΔG) was negative for both crosslinks, consistent with spontaneous non-enzymatic mechanism of formation (Fig. S5J).

Figure 6. Models of glucosepane (GN) crosslink at different positions within collagen I and GN molecular volume.

(A) The intrachain GN crosslink between the residues of the collagen α1a chain (e.g. site #1 in Fig. 2); (B) the interchain GN crosslink between α1a and α1b chains of the same triple helix (e.g. site #5 in Fig. 2); and (C) the interhelix crosslink between α1b and α2 chains of the different triple helices (e.g. site #2 in Fig. 2); (D) Molecular volume for Lys + Arg and GN (see Fig.S5 for comparisons between GN and PE). The modeling was performed using the programs PyMOL and Avogadro as described in Materials and Methods. The chain color coding is as follows: α1a (light green), α1b (magenta), α2 (cyan).

4. Discussion

We have generated a map of pentosidine (PE) and glucosepane (GN), the two most prominent AGE crosslinks in collagen I from human bone. This map can be used to guide molecular dynamics simulations and computational models that investigate how these non-enzymatic modifications affect bone quality. The map identifies, for the first time, the sites and the levels of these AGE crosslinks in collagen I that form in vivo within one of the major structural proteins. To the best of our knowledge, this is the only in vivo AGE crosslink map of any human ECM protein and is complimentary to previously published map of non-crosslink AGE modifications in bone collagen I [36]. The map demonstrates high donor-to-donor reproducibility of the crosslink positions within the primary sequences of collagen triple helical region in both males and females.

Fig. 7 illustrates the locations of the identified crosslinks within collagen I fibril. Triple helical collagen molecules (tropocollagens) assemble in a staggered manner into fibrils where one D-period repeat (~67 nm) contains the complete tropocollagen sequence from the elements of five tropocollagen segments. These D-periodic, 5-meric subunits of a collagen fibril are called microfibrils that form a quasi-hexagonal assembly pattern within the fibril [54–58]. Due to interdigitation of the neighboring microfibrils [56], crosslinks can potentially form between the microfibrils at six crosslink sites (dashed blue lines in Fig. 7). Thus, depending on the site, the crosslinks can exist within collagen fibril in the following configurations: intrachain (i.e. within the same collagen α-chain), interchain (i.e. between two different α-chains of the same triple helix), intramicrofibrillar (i.e. between two triple helices of the same microfibril), and intermicrofibrillar (i.e. between two neighboring microfibrils). At two sites (#1 and #8), the crosslinks exist only in the intrachain configuration because they were identified within a single tryptic peptide by mass spectrometry (Fig. 7, horizontal solid green lines). At four sites (#3, #5, #6, and #7), crosslinks can exist in the interchain (Fig. 7, solid red lines) and/or the intermicrofibrillar (Fig. 7, dashed blue lines) configurations. At the remaining two sites (#2 and #4), crosslinks can exist in the intramicrofibrillar (Fig. 7, solid red lines) and/or the intermicrofibrillar (Fig. 7, dashed blue lines) configurations. Interestingly, the crosslinks are almost exclusively located to the overlap zone of the fibril with minimal presence in the gap zone (Fig. 7), possibly due to the gap zone being more structurally flexible and a region where mineral accumulates [56], which may create a less favorable environment for crosslink formation. In the overlap zone with its tighter molecular packing [56], even low stoichiometry crosslinks are more likely to perturb fibril structure. On the other hand, the overlap zone should provide sufficient flexibility, possibly via interdigitation of microfibrils and right-handed supertwist [56], to ensure spatial proximity between Lys and Arg side chains necessary for crosslink formation, especially at sites #2 and #4.

Figure 7. Location of the glucosepane (GN) and/or pentosidine (PE) non-enzymatic crosslink sites in type I collagen fibril from human male bone.

Triple helical collagen molecules (tropocollagens) assemble in a staggered manner into fibrils where one D-period repeat (~67 nm) contains the complete tropocollagen sequence from the elements of five tropocollagen segments (TS). One of the segments does not transverse the entire D-period creating an overlap and a gap zones. Microfibril segment (MS) is one D-period-long segment of a microfibril (MF), which is a D-periodic 5-meric subunit of a collagen fibril (F). The expansion shows a two-dimensional representation of a microfibril segment (MS1) with two neighboring microfibril segments (MS2 and MS3). Amino acid sequence of the segments is given for full length collagen α1(I) or α2(I) chains (UniProtKB – P02452, CO1A1_HUMAN and P08123, CO1A2_HUMAN, respectively). The glucosepane- and/or pentosidine-crosslinked lysine (K) and arginine (R) residues are shown in red font with the crosslinks at specific sites depicted as solid red (intramicrofibrillar crosslinks) and dashed blue (intermicrofibrillar crosslinks) lines. The intrachain crosslinks at sites #1 and #8 which cannot form intermicrofibrillar connections are shown in green. The lysine triple helical residues participating in lysyl oxidase-catalyzed intermolecular C-terminal and N-terminal crosslinks are indicated within the MS3 sequence by the letters a and b, respectively. See Fig. 2 for more details on crosslink locations and identities. Note that the figure shows the identified crosslink locations and does not intend to represent differential levels of crosslinks at different sites. Note that mass spectrometric analysis cannot distinguish between the two α1 chains within the triple helix, therefore, when necessary, we utilized the published in silico modeling data to determine the exact crosslink positions (see manuscript text for details). F, fibril; MF, microfibril; TS1-TS5, tropocollagen monomers of a 5-meric microfibril. Crosslink site ID numbers, corresponding to those in Figs. 2 and 3, are shown at the bottom of the microfibril segment MS3.

In both male and female donors, most of the crosslinks were located to previously identified AGE clusters within the collagen triple helix [36]. This is consistent with the existence of specific triple-helical regions which are particularly susceptible to AGE damage. Like in our previously published map of non-crosslinking AGEs, there were fewer sites of AGE crosslinks found in female donors than in male donors (Fig. 3). Additionally, at many crosslink sites, we found significant correlations between crosslink levels and the levels of total PE, a known surrogate marker of AGE accumulation in ECM [59]; this correlation was present in male but not in female donors (Table S21). Overall, these results are consistent with previous observations for non-crosslinking AGEs [36] and may be a consequence of generally lower levels of oxidative stress in pre-menopausal women [60, 61], which would be less favorable for glycoxidative pathways producing AGEs. Even though this protection is lost after the menopause [61], the pre-menopausal protection may still significantly impact the overall rates of AGE formation, since AGEs can accumulate for several decades in the long-lived ECM collagens [62]. It is important to note that there was a tendency towards younger age in female donors (35–94 years) compared to male donors (58–96 years), which could have contributed to the observed sex-related differences in the AGE crosslink levels.

Many AGE crosslink sites on the map exhibited relatively low modification levels (<1% relative abundance) as was expected for physiological AGE levels in vivo [50, 63]. The GN levels were generally higher than the PE levels rising to ~40–60% relative abundance in some individual patients (see, for example, site #7 in Figs. 4, S3 and S4). This is consistent with the reports of GN being the most abundant AGE crosslink in human ECM [2].

Knowledge of the AGE crosslink positions within the collagen I sequence allows for evaluation of their impact on collagen structure and function. The analysis of the male AGE crosslink map suggests potential interference of GN and PE at site #1 and possibly GN at site #7 with N- and C-terminal intermolecular enzymatic collagen I crosslinking (Fig. 7 and [64]), which is consistent with earlier reports [65–67]. Conversely, at site #1 both GN and PE were present at the lowest levels compared to all the other sites in males and were absent in females (Figs. S3 and S4), which may be due to interference from lysyl oxidase-catalyzed C-terminal intermolecular crosslinking at K²⁶⁵ [64]. The PE site #5 overlaps with keratin sulfate proteoglycan binding site ([64] and Fig. 2), and sites #1, #5 and #7 overlap with the binding sites for HSP47, a molecular chaperone that plays a key role in the biosynthesis and secretion of collagen I [64]. This may be particularly significant for the GN site #7 (Fig. 2), where modification levels were relatively high (Figs. 4, S3 and S4).

Modeling the effects of GN crosslinks on collagen structure showed a significant increase in steric bulk on the exterior surfaces of the collagen triple helices (Fig. 6). This structural modification is likely to disrupt both native intra- and interhelical interactions by introducing spatial constraints that influence the packing of the adjacent molecules. Additionally, the crosslinks are likely to alter the electrostatic topology of the collagen surface. Such changes can disrupt the native hydrogen bonding network, which are critical for maintaining the structural as well as the functional integrity of collagen in its native state. Hypothetically, site-specific GN and PE levels may be more sensitive than bulk measures of PE and fAGE to the prediction of the mechanical properties of human cortical bone, which would suggest differential influence of individual crosslink sites on bone mechanics. In this regard, significant correlations between crosslink levels at site #3 and post-yield toughness (Fig. 5G) may be of potential interest. However, the overall weak correlations and the sex-related differences in the direction of some correlations point to a complex relationship between crosslink locations within collagen structure and bone mechanical properties, which cannot be clearly discerned in this study due to a number of factors discussed below in the Limitations section. Given a multitude of factors affecting bone mechanics, e.g. cortical porosity, degree of mineralization and its heterogeneity, primary orientation of collagen fibrils, presence of denatured collagen and microdamage, number of intact cement lines, and levels of glycosaminoglycans, contribution of AGE crosslinks to increase in bone fragility remains uncertain.

5. Limitations

There are several notable limitations to our study including relatively small sample size (n=12) due to financial and temporal challenges typical for site-specific PTM analysis by MS. Therefore, the studies lacked statistical power to clearly resolve potential relationship between the crosslink levels at specific collagen sites and bone mechanical properties, including potential sex differences. In addition, the male and female groups were not perfectly age-balanced, with female group being somewhat younger, which made sex comparisons potentially confounded by age.

Efficiency and representativeness of protein extraction are common uncertainties in this type of studies, especially when the substantial pelleted material remains upon the extraction. Therefore, it is possible that additional crosslink sites exist or that the AGE site abundances are altered within the collagen fraction that remained insoluble in this study. It is also possible that significant GN and PE crosslink levels exist in non-collagenous bone ECM proteins, which may contribute to bone mechanical properties but were not investigated in this study.

This paper is focused specifically on AGE crosslinks. Even though the sites of major non-crosslink AGEs in bone collagen I have been mapped [36], more studies are needed to elucidate the effects of both crosslink and non-crosslink AGEs on bone mechanics with respect to other known determinants of the cortical bone fragility (e.g. porosity, primary directions of collagen fibrils, and denatured or uncoiled collagen). Finally, the use of cadaveric tissue for which health information is limited, along with the limitations on the group size, hindered our ability to perform more comprehensive covariate regression analyses.

6. Conclusions

We identified 6 positions of major physiological AGE crosslinks, 3 GN sites and 3 PE sites, within the collagen I triple helix of human bone that existed in all female and male donors; the additional two sites were present only in male donors. To the best of our knowledge, this is the first experimental evidence of the GN presence in the bone matrix. The map, in combination with molecular modeling, revealed the crosslink sites to primarily form in the overlap zone of collagen I fibril. The triple helical region of collagen I contained overall higher levels of GN compared to PE crosslinks but the levels varied significantly between the sites and the individual patients. Unlike the levels of total AGEs (PE and fAGEs), crosslink levels at specific collagen sites showed sex-related and sex-independent correlations with bone mechanical properties. However, these correlations were weak, primarily due to low statistical power, imprecise group matching and other limitations, and should not be overinterpreted. Thus, additional studies are needed to elucidate a complex role of these AGEs in bone fragility with respect to other determinants of bone toughness, fracture toughness, and strength.