Is direct oxidative damage of bone collagen associated with poor human cortical bone tissue fracture resistance?

Abstract

The limitations of clinical X-ray-based tools and fracture risk assessment tools, such as FRAX, particularly for disease groups, including type 2 diabetes (T2D) and CKD, suggest that important determinants of bone fracture resistance are not assessed. One such determinant is the bone collagen network, whose nativity and connectivity contribute to cortical bone fracture resistance, but whose degradation mechanisms remain unclear. Oxidative damage to the collagen network, driven by oxidative stress, has been proposed as a contributing factor. Two pathways are recognized: an indirect pathway involving glycoxidation and advanced glycation end-product formation, examined in a precursor study, and a direct pathway, examined herein, involving protein fragmentation and amino acid side chain modifications, such as carbonylation. To investigate the direct pathway, bulk collagen carbonyl content (measured using a fluorescence-based assay and normalized to collagen content) and intact collagen α-chain content (the opposite of α-chain fragmentation and measured via α-chain band intensities using gel electrophoresis) were quantified ex vivo using cortical bone specimens from 80 human donors with and without T2D and/or CKD. Relationships with fracture toughness and other bone quality measures from a precursor study were assessed using Spearman correlations and multiple linear regression. Bulk collagen carbonyl content was not directly associated with collagen network connectivity or cortical bone fracture toughness. In contrast, intact collagen α-chain content emerged as an independent explanatory variable for stable crack-growth fracture toughness measures (adj-R² = 46.3%, p = .028) and was negatively associated with bulk collagen carbonyl content (r = −0.37, p < .001). These findings suggest that collagen α-chain integrity may provide insight into variation in cortical bone fracture resistance, whereas bulk collagen carbonyl content did not support the hypothesized direct oxidative damage pathway in this donor cohort.

Introduction

Clinical X-ray based technologies that measure BMD are widely used to assess bone fracture risk.¹ These tools, however, have been shown to underestimate bone fracture risk in certain individuals and disease groups.^2,3 For example, almost 50% of all non-vertebral fractures occurring in individuals aged 55 and over are associated with reports of normal BMD.⁴ In addition, individuals with type 2 diabetes (T2D) typically report normal to high BMD but suffer higher incidences of fracture compared to individuals without diabetes.^3,5 The World Health Organization’s fracture risk assessment tool (FRAX), which combines BMD measurements with other clinical risk factors, such as age, sex, fracture history, and body mass, among others, has also been shown to underestimate fracture risk in individuals with diabetes.⁶ This suggest that some factors associated with increased bone fragility and contributing to the fracture risk of an individual are invisible to X-ray based tools and not considered by current screening tools, such as FRAX. One such factor is the state of the organic phase of bone, which has been shown to be vital to the fracture resistance of bone at the tissue level.⁷ Of particular interest to our group is type I collagen, which constitutes about 90% of the organic phase of cortical bone. The type I collagen molecule has a triple helix structure, consisting of three α chains (two α-1 chains and one α-2 chain) bound together by hydrogen bonding.⁸ Type I collagen has been shown to contribute to the inelastic and viscoelastic behaviors of bone that enable it to better resist fracture.⁹ In fact, its specific importance to the fracture resistance of bone has been shown in various studies.

Zioupos et al. showed that the collagen network connectivity degrades with age, and this was correlated with decreased cortical bone work-to-fracture, a measure of its fracture resistance.¹⁰ Collagen network connectivity measured using a method known as hydrothermal isometric tension (HIT) testing gives a measure of how well the collagen molecules are tied together in a network within the bone tissue.¹¹ Generally, this connectivity is a function of crosslinking and polypeptide chain lengths.^12,13 Wang et al.¹⁴ also showed that the strength, stiffness, and toughness of the collagen network decrease with age. In addition, they found that their measure of denatured collagen content increased with age, and this correlated strongly with reduced fracture resistance.¹⁴ Furthermore, a more recent study by our group and collaborators,¹⁵ which encompassed a more general and heterogenous donor population, also showed a moderate to strong positive correlation between the collagen network connectivity and the fracture resistance of cortical bone. It was shown that the changes in collagen network connectivity helped explain approximately 49% of the variation observed in the fracture resistance of the cortical bone samples.¹⁵ These studies highlight the important contribution of bone collagen to the overall fracture resistance of cortical bone tissue and that the collagen network nativity and connectivity degrade with age and in bone damaging diseases. However, the underlying mechanisms of collagen network nativity and connectivity degradation leading to bone fragility remain unclear.

Oxidative stress is regarded as a fundamental feature of aging and many acute and chronic diseases, including osteoporosis, osteoarthritis, cancers, hypertension, T2D, and CKD.¹⁶ Most of these diseases are thought to be associated with poor bone quality, with T2D and CKD of particular interest as patients have increased bone fracture risk compared to controls. However, current clinical tools do a poor job of fracture risk prediction for individuals with these diseases.^6,17,18 Oxidative stress occurs when there is an abundance of ROS over antioxidants that scavenge these ROS.¹⁹ Interestingly, oxidative stress is known to lead to deleterious changes to cells, tissues, and macromolecules, including proteins, and these changes are termed oxidative damage.¹⁹ For long lived proteins, such as collagens, there are 2 general pathways of oxidative damage: indirect and direct.

The indirect pathway involves the oxidation of other macromolecules (lipids, carbohydrates, and sugars) by ROS, which in turn react with proteins, causing oxidative damage. For instance, various sugar molecules can be oxidized by ROS forming highly reactive substrates that react with proteins. This process is known as glycoxidation and can result in the formation of advanced glycation end-products (AGEs). Advanced glycation end-products have been one of the key foci of attention of bone quality research for approximately 2 decades. In a precursor study to this current report, AGEs were examined.²⁰ Cortical bone specimens taken from the femurs of 80 different donors with and without a history of T2D and/or CKD were analyzed for various AGE contents, collagen network-based properties and fracture resistance measures. We found 3 AGE adducts [carboxy-methyl-lysine (CML), carboxy-ethyl-lysine (CEL), and 5-hydromethyl-imidazolone (MG-H1)] and pentosidine to be significantly higher in T2D/CKD group compared to controls. However, this distinction between groups was not observed for our collagen network connectivity and fracture toughness measures with significant data overlap between the groups. These results suggest a continuous variation rather than categorical distribution for bone quality and fracture resistance irrespective of disease status (specifically T2D and CKD for the precursor study). Other recent studies involving T2D and control groups have also found this continuous variation for various mechanical properties of bone with no significant differences between the 2 groups.^21,22

In the direct pathway, the ROS directly attack the backbone and/or side groups of proteins. This can result in protein fragmentation and/or conformational changes.²³ This pathway predominately leads to the formation of carbonyl containing compounds (eg, aldehydes). Consequently, carbonylation, a highly stable and established biomarker, is widely used to measure the extent of protein oxidative damage by the direct pathway.²⁴ Given that oxidative damage by the direct pathway can cause polypeptide chain fragmentation, this is a plausible mechanism that might partially contribute to the degradation of the collagen network connectivity previously associated with reduced fracture resistance. Collagen network connectivity is partially a function of polypeptide chain length.¹² This mechanism has not yet been studied.

Whereas our precursor study focused on AGEs as markers of indirect oxidative damage, the present study evaluated 2 previously unexamined measures associated with direct oxidative damage to cortical bone collagen: bulk collagen carbonyl content, as an in situ biomarker of direct oxidative damage, and intact collagen α-chain content, as an in situ biomarker of collagen α-chain fragmentation. The objective was not to reproduce the findings of the precursor study, but rather to determine whether markers of direct oxidative damage provide additional insight into the continuous variation observed in collagen network properties and cortical bone fracture resistance regardless of T2D/CKD status. Accordingly, T2D/CKD status was treated as secondary, and relationships between direct oxidative damage, collagen network properties, and cortical bone fracture toughness were examined across all donors. To achieve this objective, bulk collagen carbonyl content and intact collagen α-chain content were measured in the same 80 cortical bone specimens from the precursor study, and their associations with fracture toughness measures, collagen network quality-related properties, and key determinants of bone tissue quality, including porosity, were investigated. We hypothesized that bone collagen carbonyl content would correlate negatively with collagen α-chain integrity and collagen network connectivity, and that these relationships would contribute to statistical models explaining the continuous variation observed in human cortical bone fracture toughness measures.

Materials and methods

Specimen acquisition

This study was approved by the University of Waterloo’s Office of Research Ethics Committee (ORE #40888). Eighty human femoral cadavers were acquired from 80 donors (43 males and 37 females) from various tissue banks across North America (Mount Sinai Allograft Technologies, Lake Superior Centre for Regenerative Medicine, National Disease Research Interchange, United Tissue Network, Innoved Institute, and Science Care). The ages of the donors were between 19 and 103 yr (mean ± SD = 68.9 ± 17.5 yr). Thirty-four of the donors (19 males and 15 females) had a history of T2D with or without CKD (mean ± SD = 70.1 ± 13.4 yr). The rest of the 46 donors (24 males and 22 females) had no history of T2D and/or CKD (mean ± SD = 68.0 ± 20.0 yr). These are referred to as the control group. Donors included in this study did not have a history of heavy smoking and/or drinking, cancer, liver disease, COVID-19, hysterectomy/ovariectomy, radiation/radiotherapy or chemotherapy treatment, surgeries, such as hip or knee arthroplasty, taking selective serotonin reuptake inhibitors, bisphosphonates or other antiresorptive medications, and glucocorticoids as well as sex hormone imbalances. Femurs were stored in a −80 °C freezer until specimen preparation.

Overview of precursor study experiments

In our precursor study,²⁰ femurs from these same donors were used to prepare cortical bone single edged notched bend (SENB) specimens and used in 3-point bend fracture tests as well as various characterization tests to measure the collagen network properties, AGE contents, tissue porosity and mineralization profiles. These experiments are described in detail in the precursor paper²⁰ with a brief overview provided below.

Using a combination of a pathology bone band saw with an irrigated diamond blade (IMEB), a mini-CNC vertical mill (Sherline), a low-speed metallurgical saw with a wafering diamond blade (Isomet, Buehler), beams of dimensions 50.5 ± 0.55 mm × 5.0 ± 0.05 mm × 2.48 ± 0.07 mm were obtained from the distal lateral diaphysis of each femur (Figure 1). Razor sharpened starter notches of length 2.55 ± 0.1 mm and side grooves of depth 0.49 ± 0.1 mm were created at the mid-point of these beams to create SENB specimens. These SENB specimens underwent quasi-static, displacement-controlled three-point bending fracture tests using a micromechanical test system (Psylotech). A digital camera (Point Grey, 5 MP, 2/3″ detector) attached to an optical microscope (BXFM, Olympus Corporation) mounted over the test system allowed for images of the region around the growing crack to be acquired during the test. The images were then used to compute crack extensions during the fracture test using an approach based on digital image correlation. From these, alongside the load-line deflection curves, J-integral resistance (J-R) curves were generated from which three single value fracture toughness measures were computed for each sample. These were: (1) the crack initiation fracture toughness (J_IC) denoting the onset of crack growth, (2) crack instability fracture toughness (J-int) denoting approximately the point where crack growth transitions to unstable fracture, and (3) the normalized work to fracture (Wfx_n) capturing the total energetic cost during the stable fracture process per unit area of crack extension.

Figure 1.

Illustration of a single edged notched bend (SENB) specimens prepared from the lateral distal diaphysis of a human cadaveric femur and showing the side groove on one side along with specimen dimensions.

After the fracture test, each SENB specimens was divided into various portions for further characterization tests, including: (1) micro-CT, (2) liquid chromatography–mass spectrometry (LC-MS), (3) HIT testing, and (4) differential scanning calorimetry (DSC) (Figure 2). It is important to note, LC-MS, HIT, and DSC tests were conducted on demineralized portions from the SENB specimen. The demineralization was achieved by soaking specimens in 0.5 M ethylenediaminetetraacetic acid for 6 wk at room temperature (RT).

Figure 2.

Schematic documenting the division of a single edged notched bend (SENB) specimen after fracture testing prior to the various characterization tests in this study and the precursor study.²⁰ For the precursor study, the characterization tests included micro-CT to assess cortical porosity, apparent volumetric BMD (avBMD), tissue mineral density (TMD), and mineralization heterogeneity, hydrothermal isometric tension (HIT) test to measure the collagen network stability and connectivity, differential scanning calorimetry (DSC) to assess the collagen network’s nativity, stability, heterogeneity and denaturation energy, and liquid chromatography-mass spectrometry to measure the collagen network’s lysyl oxidase (LOX) mediated crosslink contents (LNL, HLNL, DHLNL, DPD, and PYD) and advanced glycation end-product (AGE) contents (CML, CEL, MG-H1, and PEN). In the current study, fluorescence-based carbonyl microplate assays were conducted to measure carbonyl contents, a biomarker for direct oxidative damage to proteins, as well as sodium dodecyl sulfate polyacrylamide gel electrophoresis (SDS PAGE) to assess collagen chain fragmentation by quantifying the intact α chains.

Micro-CT was conducted using an inCiTe 3D X-ray microscopy system (KA Imaging) with an isotropic voxel size of 7 μm with the following settings: tube voltage of 80 kVp, tube current of 166 μA, integration time of 1500 ms, 1000 projections, an averaging frame of 5, 1-mm aluminum filter, and a micro-CT hydroxyapatite phantom (QRM-D32, PTW) to allow for density calibration. After reconstruction, cortical porosity, apparent volumetric BMD (avBMD), tissue mineral density (TMD), extent of mineralization (M_peak), and mineralization heterogeneity (M_FWHM) were computed.

Liquid chromatography–mass spectrometry was carried out using a LaChrom Elite L-2000U system (Hitachi) and a maXis 3G Ultra-High Resolution QqTOF-MS by Bruker Daltonics, equipped with an electrospray ionization (ESI) source in positive ion mode. The LC-MS system was used to measure the immature lysyl oxidase (LOX) crosslinks: lysinonorleucine (LNL), hydroxylysinonorleucine (HLNL), and dihydroxylysinornorleucine (DHLNL), as well as the matured LOX crosslinks: pyridinoline (PYD) and deoxypyridinoline (DPD). Additionally, four different AGEs: pentosidine and three AGE adducts; CML, CEL, 5-hydro-5-methyl-4-imidazolon-2-yl-ornithine 1 (MG-H1) were also measured.

Hydrothermal isometric tension was performed using a custom-built system that allows 6 samples to be tested simultaneously. Here, the samples were held at isometric constraint and heated in a water bath from RT to 90 °C at a rate of 1.4 °C/min. From the resulting isometric stress vs temperature curves, the maximum rate of change in the stress, referred to as maximum slope (Max.Slope), and denaturation temperature (T_d) were computed. DSC was conducted using a TA Instruments DSC Q-2000 system (TA instruments). The samples used in this test were enclosed in hermetically sealed aluminum pans and heated from RT to 95 °C at a rate of 1.4 °C/min to mimic the HIT heating rate. The heat flow vs temperature curves obtained from these test were used to compute the temperature at onset of denaturation (T_onset), temperature at which the maximum heat flow was recorded (T_peak), the full width of the endotherm at half of the maximum heat flow (FWHM), which gives a measure of the collagen network disorder (heterogeneity), and the enthalpy of denaturation (ΔH; normalized to both the wet and dry mass of the samples), which provides a measure of the bone collagen nativity.

For this report, demineralized portions were taken from the same SENB specimens and used in 2 additional assays: a fluorescence-based microplate carbonyl content assay and SDS-PAGE (Figure 2).

Bulk carbonyl content quantification by fluorescence-based microplate assay

To assess the extent of direct oxidative damage to the collagen network of each cortical bone specimen, bulk carbonyl content in the collagen was measured using a fluorescence-based microplate assay. This assay was based on work done by Mohanty and co-workers.²⁵ The fluorescence probe, fluorescein-5-thiosemicarbazide (FTC), binds 1:1 to the carbonyls (aldehydes and ketones) on the bone collagen and the associated fluorescence from the bound FTC is used to measure the bulk carbonyl content. Between 15 and 20 mg of the demineralized bone collagen from each specimen was isolated, cut into small pieces, and underwent trypsin digestion for 72 h at 37 °C. The trypsin digest involved using trypsin from porcine pancreas dissolved in Hank’s buffer to form a 1 mg/mL solution. After the trypsin digestion, two aliquots of the supernatants of the digested specimens were collected. The first aliquot was used in a colorimetric hydroxyproline assay to measure the amount of hydroxyproline present in each demineralized portion. The hydroxyproline assay was carried out by following the method of Hofman and colleagues.²⁶ Briefly, 25 μL of hydrochloric acid was mixed with a 25 μL aliquot of the trypsin-digested supernatant and heated at 110 °C for 4 h. The resulting hydrolysate was neutralized with NaOH and reacted with 40 μL of 0.05 N chloramine-T and 40 μL of Ehrlich’s reagent (15% w/v p-dimethylaminobenzaldehyde in a 2:1 v/v isopropanol: perchloric acid solution). Triplicate 80 μL aliquots of each reaction product were then pipetted into wells of a transparent 96-well microplate (Corning). The absorbance at 560 nm was measured using a SpectraMax Plus 384 microplate reader (Molecular Devices) and used to compute the hydroxyproline content for each specimen based on a standard curve generated from absorbance readings of hydroxyproline standards of known concentrations.

The second aliquot of 50 μL was used in the fluorescence-based microplate assay. This was mixed with 50 μL of 0.2 mM FTC dissolved in 1:9 (v/v) dimethyl sulfoxide and 4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid (HEPES) buffer and incubated in the dark overnight. After incubation, the collagen peptides were precipitated into pellets using 30% trichloroacetic acid. Three acetone washes were performed to remove unbound FTC from the precipitated pellets. The pellets were left to air dry in a fume hood for 1.5 h to evaporate remaining acetone. The air-dried pellets were then solubilized with 6 M guanidine hydrochloride and re-suspended in sodium phosphate monobasic buffer. Triplicates of each solution were pipetted into a black opaque 96 well microplate (Greiner Bio-One) and fluorescence was measured at an excitation of 480 nm and emission of 530 nm using a BioTek Synergy H4 Hybrid multi-mode microplate reader (Agilent). The bulk collagen carbonyl content of each specimen was determined using a standard curve prepared from FTC standard stock solutions of known concentrations. The bulk collagen carbonyl content measured for each specimen was then normalized to the hydroxyproline content measured as described above.

Sodium dodecyl sulfate polyacrylamide gel electrophoresis

Sodium dodecyl sulfate polyacrylamide gel electrophoresis was conducted to measure the amount of intact α chains in each sample’s collagen network. This was done by measuring the relative intensity of the bands associated with the intact α-I and α-II chains of the collagen molecules. In nominally healthy collagen networks, the number of intact α chains is presumably consistent for the same amount of bone collagen. Therefore, the loss of band intensity was considered a measure of α-chain fragmentation in the collagen network. That is, a relative decrease in α-chain band intensities for the same amount of collagen loaded in the gel could only plausibly be explained by fragmentation of the α chains. We refer to this herein as intact collagen α-chain content.

Demineralized portions from each specimen roughly 20-25 mg were isolated, divided into small portions, lyophilized, and then digested using pepsin. The pepsin digestion was carried out for 6 d at RT using pepsin from porcine gastric mucosa dissolved in 0.5 M acetic acid (HOAc) and a 1:10 (w/w) pepsin to dry mass of bone specimen ratio. The supernatant from each pepsin digested specimens were collected and neutralized with sodium hydroxide. Like the fluorescence-based carbonyl assay, two aliquots were taken from the neutralized supernatants. The first aliquot underwent the colorimetric hydroxyproline assay following the same procedure as described in the Section “Bulk carbonyl content quantification by fluorescence-based microplate assay” to quantify the amount of hydroxyproline present in the pepsin digested supernatant. This was used to measure the amount of collagen present (hydroxyproline forms about 14% of the collagen molecule). The second aliquot was used in SDS-PAGE. This 25 μL aliquot was mixed 10:3 (v/v) with a buffer solution of 4x Laemmli buffer (Bio-Rad) and 2-mercapto-ethanol (Millipore Sigma) in a 5:1 (v/v) ratio. The resulting solution was heated at 95 °C for 5 min. After cooling, a volume from this solution was taken such that it contained 30 μg of collagen based on the hydroxyproline assay carried out on the same supernatant. Using a consistent mass of collagen ensured each sample if healthy (native) would contain the same amount of intact α chains. These were loaded into pre-cast 7.5% polyacrylamide gels with 10 wells (Mini-protean TGX, Bio-Rad). Each run included an unstained protein standard (Precision Plus, Bio-Rad), a purified rat tail collagen standard (RTC; Millipore Sigma), and the pepsin-HOAc solution used for digestion as a molecular weight standard, a positive control and a negative control, respectively. The tests were run at 200 V for 50 min. After the tests, the gels were stained with Coomassie Blue (Bio-Rad) for 2 h and then destained for 36 h using a 10% (v/v) methanol plus 10% (v/v) HOAc aqueous destaining solution. The stained gels were then imaged using a gel imaging system (ChemiDoc MP, Bio-Rad). ImageJ was used to measure the relative intensity of the intact α-I and α-II chain bands. The α chains have molecular weights between 100 and 150 kDa, and the intensity of the Coomassie Blue stained band is indicative of the number of intact α chains present within the constant mass loaded into the lane.²⁷ These were then normalized to the intensity of the α-chain bands for the RTC control from each corresponding gel. Note that the same RTC control solution was run on every gel (Figure 3).

Figure 3.

Representative sodium dodecyl sulfate polyacrylamide gel electrophoresis (SDS-PAGE) gel showing polypeptide bands detected by Coomassie blue staining. The numbers on top of the gel denote the well numbers, with the legend on the right side of the figure indicating the type of sample in each well. Well 5 contained the protein standard and the number directly on top of each band in well 5 represents the molecular weight of the band in kDa. The box highlights the 2 α-chain bands for each sample used to quantify the measure of collagen chain fragmentation.

Statistical analysis

One tail student t-tests were conducted to test for differences in the bulk collagen carbonyl content and intact α-chain content between the T2D/CKD group and the control group. This was done to assess whether the idea of continuous variation irrespective of T2D/CKD status held regarding measures related with direct oxidative damage of the collagen network. To ensure differences detected between the groups could not be attributed to age differences because donors were not age matched, analyses of covariance (ANCOVA) were also performed.

Spearman correlations and multiple linear regression (MLR) were conducted to investigate relationships between the bulk collagen carbonyl contents, the intact collagen α-chain content, and the previously measured collagen network properties as well as cortical bone fracture toughness measures from the precursor study. Spearman correlations were used because some of the variables did not follow a normal distribution. This was tested using the Kolmogorov–Smirnov test.

For the MLR models, the fracture toughness measures were the dependent/response variables. The bulk collagen carbonyl contents and the intact collagen α-chain contents together with the other measures from the precursor study were all considered as independent/explanatory variables. Age and sex were not included in the multivariable models as the objective of the analysis was to evaluate covariance among tissue-level material properties rather than clinical predictors of fracture risk. Age was a significant covariate with many of the other explanatory variables. Emphasis was placed on assessing associations between material properties. Therefore, sex was not included in the analyses.

Using Max.Slope and porosity as base explanatory variables, a forward selection stepwise regression approach based on each model’s adjusted R² was adopted to add additional explanatory variables. The final MLR models were generated such that either bulk collagen carbonyl content or the intact collagen α-chain contents were included as a final explanatory variable. If two explanatory variables were co-variates (r > 0.55 from Spearman’s correlation), they were not used in the same model to prevent multi-collinearity. Variance inflation factor (VIF) values for final explanatory variables in each model were computed to further ensure multi-collinearity was avoided (see Supplementary Material found in Table S1). Two-way interaction terms between relevant explanatory variables and the bulk collagen carbonyl contents or the intact collagen α-chain contents were also considered in generating the MLR models. Linearity and normality were also checked to ensure the models met the assumptions for MLR models (see Supplementary Material Section 1.2 as well as Figures S1 and S2).

T-tests, ANCOVA and Spearman’s correlations were conducted using the MATLAB Statistics Toolbox (The Mathworks Inc.), whereas the MLR were performed using RStudio software (Posit). Statistical significance was defined as p-value ≤.05.

Results

As reported in the precursor study,²⁰ two specimens were excluded as outliers from the various statistical analyses because of their unusually high porosity (>30%). From the one tail t-tests, no differences between the two groups (Control vs T2D ± CKD) were detected for both bulk collagen carbonyl content (p = .639) and the intact collagen α-chain content (p = .342) (Figure 4). Analyses of covariance confirmed that the lack of detectable differences between the groups were not attributable to age.

Figure 4.

Box plots with overlayed data points for (A) bulk carbonyl content normalized to hydroxyproline content and (B) normalized α-chain band intensity from Coomassie Blue staining, comparing the T2D/CKD and control groups. Statistical differences were not detected between the two groups for both measures. This suggests that disease status (T2D and/or CKD) does not determine direct oxidative damage to bone collagen nor collagen chain fragmentation.

Table 1 details the Spearman correlation coefficients for relationships between the bulk collagen carbonyl content as well as the intact collagen α-chain content and the fracture toughness measures, various AGEs, and the collagen network-based properties. The bulk carbonyl content did not correlate with any of the fracture toughness measures but did correlate positively with the previously measured AGE adducts (0.28 < r < 0.32, p < .01). Interestingly, the bulk carbonyl content had a moderate negative (r = −0.37, p < .001) correlation with the intact collagen α-chain content.

Table 1.

Spearman’s correlation coefficient table between carbonyl contents, intact α-chain contents, and the other characterization measures and age.

Characterization measures		Carbonyl content	Intact α chains content
Fracture toughness measures	J IC	ns	0.21+
	J-int	ns	ns
	Wfx n	ns	ns
AGEs	CML	0.32**	−0.56***
	CEL	0.29**	−0.56***
	MG-H1	0.31**	−0.59***
	Pentosidine	0.21+	−0.46***
Immature LOX crosslinks	LNL	ns	0.45***
	HLNL	ns	0.54***
	DHLNL	ns	ns
Mature LOX crosslinks	DPD	ns	ns
	PYD	ns	ns
Collagen quality-based properties	Max.Slope	ns	0.20+
	T d	ns	0.19+
	T onset	ns	ns
	T peak	0.39***	ns
	FWHM	ns	−0.23*
	ΔH (wet mass)	ns	0.36**
	ΔH (dry mass)	ns	0.37***
	Intact α chains	−0.37***	-
Bone mineral-based properties	Porosity	ns	0.19+
	avBMD	ns	−0.24*
	TMD	ns	ns
	M peak	ns	ns
	Mineral heterogeneity	ns	ns

The superscript indicates: ^*p-value <.05, ^**p-value <.01, ^***p-value <.001, ⁺p-value <.1, ns represents not significant.

The intact collagen α-chain content did not correlate with any of the fracture toughness measures. However, its correlation with JIC fell just outside the statistical significance criterion having p = .06. In addition, the intact α-chain content had a relatively strong negative correlation with all of the AGEs measured in the precursor study (−0.46 < r < −0.59, p < .001). Surprisingly, a significant correlation was not detected between the intact α chain content and the collagen network connectivity (Max.Slope) as well as thermal stability (T_d) with the p-values for both falling just outside statistical significance at p = .077 and p = .098, respectively. However, the intact α-chain contents correlated with other collagen network properties, specifically its denaturation enthalpies (ΔH), its heterogeneity (FWHM) and its immature crosslink contents (LNL and HLNL) (Table 1). Figure 5 shows scatterplots for the bulk carbonyl content vs J-int (Figure 5A) and intact α chain content (Figure 5C) as well as the intact α chains content vs J-int (Figure 5B) and Max.Slope (Figure 5D). Other relevant scatter plots for intact α chains and carbonyls are presented in the Supplementary Material (Figures S3 and S4).

Figure 5.

Scatter plots for (A) normalized α-chain band intensity vs J-int (B) bulk carbonyl content vs J-int (C) normalized α-chain band intensity vs bulk carbonyl content and (D) normalized α-chain band intensity vs Max.Slope. No statistically significant relationships were detected between normalized α-chain band intensity as well as bulk carbonyl contents and J-int, a measure of stable crack growth fracture toughness. However, there was a moderate negative relationship detected between normalized α chain band intensity and bulk carbonyl contents.

Table 2 provides an overview of the MLR models that explain the variation in fracture toughness measures with the intact α chain contents and bulk carbonyl content included as explanatory variables. The bulk carbonyl content was not detected as an explanatory variable for any of the fracture toughness measures whereas intact α chain contents were detected as an independent explanatory variable for J-int and Wfxn but not JIC. For both J-int and Wfxn, a combination of Max.Slope, an interaction term between Max.Slope and porosity as well as the intact α chain contents explained 46.3% and 41.0% of their variation, respectively. Interestingly, when the intact α chain content was replaced with an interaction term between it and Max.Slope, a roughly equal adj-R² was obtained, that is, an adj-R² of 46.3% vs 46.6% for J-int and adj-R² of 41.0% vs 41.3% for Wfx_n. As in the precursor study, Max.Slope remained the most significant positive contributor to the models of fracture toughness measures indicated by its strong positive standardized beta-coefficients (𝛽) values.

Table 2.

Summary of multiple linear regression (MLR) models showing the intact α-chain content and other variables as explanatory variables along with their standardized beta co-efficient () that explained the various fracture toughness measures.

Fracture toughness measures	Explanatory variables				Adj-R  2
J IC  model 1	Max.Slope , p<.001)	Max.Slope* Porosity ()			29.1
J IC  model 2	Max.Slope 4)	avBMD )	M peak )	ΔH (wet mass) , p=.048)	31.5
J-int model 1	Max.Slope , p<.001)	Max.Slope* Porosity (  p=.001)			43.4
J-int model 2	Max.Slope , p<.001)	Max.Slope* Porosity (  p<.001)	Intact α chains (  p=.028)		46.3
J-int model 2	Max.Slope , p<.001)	Max.Slope* Porosity (  p<.001)	Max.Slope* Intact α chains (  p=.023)		46.6
Wfx n  model 1	Max.Slope , p<.001)	Max.Slope* Porosity (  p=.002)			38.7
Wfx n  model 2	Max.Slope , p<.001)	Max.Slope* Porosity (  p=.001)	Intact α chains (  p=.054)		41.0
Wfx n  model 3	Max.Slope , p<.001)	Max.Slope* Porosity (  p=.001)	Max.Slope* Intact α chains (  p=.042)		41.3

Discussion

This study was conducted as an extension to our precursor study²⁰ to gain insight into the possible association between direct oxidative damage to the collagen network, the collagen network’s α-chain fragmentation, the collagen network connectivity as well as the fracture resistance of human cortical bone. In the precursor study, various AGE adducts (CML, CEL, and MG-H1) were found to be correlated strongly and negatively with immature LOX crosslink content, especially HLNL. In turn, an interaction term between HLNL and our measure of the collagen network connectivity (Max.Slope) was a significant explanatory variable in the MLR models for two of the fracture toughness measures, J-int and Wfx_n (adj-R² = 47.6% and 40.9%, respectively). As such, it seems that AGE adducts, a form of indirect oxidative damage, may disrupt immature LOX crosslink formation during bone remodeling, which in turns degrades the collagen network connectivity over time, therefore negatively impacting the contribution of the collagen network to the overall fracture resistance of the cortical bone tissue.

In this study, bulk collagen carbonyl content, a measure of direct oxidative damage in the collagen network was measured ex vivo in human femoral cortical bone specimens from the same 80 donors from the precursor study.²⁰ To the best of our knowledge, this is the first study to measure bulk carbonyl content in human cortical bone collagen ex vivo. In addition, the intact α-chain content, the opposite of collagen α-chain fragmentation, was also measured. We hypothesized that bulk collagen carbonyl content would correlate negatively with the intact α-chain content. In addition, we hypothesized that both measures would be useful explanatory variables in models of the human cortical bone’s fracture resistance irrespective of T2D and/or CKD status.

Unlike the AGE adducts and pentosidine measured in the precursor study, differences in bulk collagen carbonyl content and intact collagen α-chain content between the T2D/CKD group and the control group were not detected. It is important to note that donors in the control group had other morbidities beyond T2D and CKD, such as osteoporosis, osteoarthritis and hypertension, which are also associated with elevated oxidative stress.¹⁶ The oxidative stress associated with these diseases may drive the formation of carbonyl containing compounds in bone collagen. Therefore, the lack of a detectable difference might be explained by the presence of these other morbidities in the control group, and the associated elevated oxidative stress possibly giving rise to carbonyls in the collagen network in amounts similar to those in the T2D/CKD group. It is difficult to control for this and hence a reason to examine factors contributing to continuous variations in the collagen network and fracture resistance.

A moderate negative correlation was observed between bulk collagen carbonyl content and intact collagen α-chain content measures (r = −0.37 p < .001; Table 1). Oxidative damage by the direct pathway is known to lead to fragmentation of polypeptide chains predominantly resulting in the formation of carbonyl containing compounds.²⁸ One consequence of direct oxidative damage to cortical bone collagen may be fragmentation of its α chains.

Interestingly, both the bulk collagen carbonyl content and the intact collagen α-chain content were correlated with the AGE content measured in the precursor study (Table 1). This presumably occurs because both pathways of oxidative damage have a common driving force, oxidative stress, resulting in multiple collinearities. The relationships between the intact α-chain content and AGEs are made more interesting by the correlations between the intact α-chain content and the collagen network’s denaturation energy (ΔH), its disorder (FWHM) as well as the immature crosslinking contents (LNL and HLNL; Table 1). Advanced glycation end-product contents measured in the precursor study were also correlated with the same collagen network-based properties.²⁰ These multiple collinearities draw attention to a plausible common driving force, possibly oxidative stress.

Collagen denaturation occurs when the hydrogen bonds that bind the three α chains together into the triple helix are broken and the α chains unravel and possibly separate.²⁹ This can occur due to thermal energy or mechanical work.³⁰ Both require relatively large amounts of energy.^31,32 Consequently, if fragmentation of the intact α chains occurs, one might expect that the energetic cost of denaturation would be lower. This may explain the relationship between the intact α chain content and the denaturation energy, ∆H, measured in this study (Table 1). In addition, there may be greater disorder in the collagen network, possibly explaining the relationship between the intact α chain content and the FWHM measured using DSC (Table 1). The FWHM is thought to capture heterogeneity in the semi-crystalline structures of the collagen fibrils because packing and crosslinking control the thermal stabilities of the molecules within the collagen fibrils.^33–35

It is challenging to interpret the relationship between the intact α chains and the immature crosslinks. In the precursor study, AGE adducts were shown to clearly and negatively correlate with the collagen immature crosslink contents, especially HLNL. It was proposed that the AGE adducts disrupt or block the formation of immature crosslinks either by competing with LOX for the lysine and hydroxylysine sites on the collagen at which the immature crosslinks form and/or that they may form at these sites because the action of LOX is suppressed by oxidative stress. From this, it is possible that the relationship between the intact α chains and the immature crosslinks is merely correlative. Alternatively, this relationship may be a consequence of oxidative stress.

On the other hand, in a study where ribose pretreatment was used to counter the deleterious effect of gamma irradiation necessary for the sterilization of bone allografts, it was found that increased crosslinking in the collagen network due to the ribose incubation protected the collagen network from α-chain scission driven by gamma irradiation.¹³ Gamma radiation causes protein polypeptide chain fragmentation because it drives water radiolysis, which generates free radicals including ROS that attack and fragment the protein polypeptide chains.³⁶ One might extrapolate that the increased crosslinking provided greater stability to the collagen network which in turn protected the α chains from scission. Therefore, it is possible that the reduced stability of the collagen network due to the disruption of the immature LOX crosslinks due to AGE adduct formation may make the collagen molecules more susceptible to fragmentation by ROS. In this case, the relationship between the intact α chains and the immature crosslinks as well as the AGE adducts found in this study would be causative. However, there is little evidence in the literature to support this hypothetical mechanism. Hence, this is an area for further exploration. Consequently, it is posited that the relationship between the intact α chains and immature LOX crosslinks is a consequence of elevated oxidative stress driving both the disruption of immature crosslinks formation as well as collagen chain fragmentation.

An important outcome of this study is that the primary hypothesis was not supported. Although direct oxidative damage was hypothesized to contribute to reduced fracture resistance through degradation of collagen network connectivity, bulk collagen carbonyl content did not demonstrate statistically detectable associations with fracture toughness or collagen network connectivity. Consequently, the present data do not support bulk collagen carbonyl content as a direct explanatory variable of cortical bone fracture resistance within this donor cohort. While more specific measures of oxidative protein damage may yield different findings, such possibilities remain speculative and are not supported by the current dataset.

A possible explanation for this measure’s lack of contribution to the MLR models may be its lack of specificity and lower sensitivity. The fluorescence-based microplate assay measures all ketones and aldehydes in a solubilized specimen. As such, it is possible that the carbonyl assay quantified derivatives that were not products of collagen fragmentation. That is, products from other post-translational modifications to the collagen network. For example, oxidative deamination of lysine and hydroxylysine by the LOX enzyme results in allysine and hydroxyallysine, which are highly reactive aldehydes.³⁷ These could also be captured in the bulk carbonyl content measure if LOX were effective in the oxidative deamination of the lysines and hydroxylysines, but the following stages of crosslink formation were interrupted. In future studies, it will be beneficial to use methods that provide greater specificity.²³

It is worth noting that the correlation between the intact collagen α-chain content and the collagen network connectivity (Max.Slope) as well as the J_IC fell just outside significance (p = .077 and p = .06, respectively). While inclusion of intact α-chain content increased the adjusted R² values of the fracture toughness models by only approximately three percentage points, the variable remained statistically significant after accounting for collagen network connectivity and porosity. Therefore, the contribution of intact α-chain content should be interpreted as modest. Nevertheless, its independent contribution suggests that α-chain integrity captures information not fully represented by collagen network connectivity measures alone.

Despite its modest contribution to the R² and its small effect size (β ~ .18-.20), the intact collagen α chain content was found to be a statistically significant independent explanatory variable for the MLR models developed for J-int and Wfx_n (Table 2). Note that J-int and Wfx_n are measures of stable crack growth fracture toughness of the cortical bone tissue. This result, and the positive correlation between the intact α-chain content and the measures of ∆H is consistent with the contribution of intact collagen molecules towards sustaining stable crack growth in the human cortical bone tissue.³⁰ Interestingly, when the intact α chain content was replaced with an interaction term between it and Max.Slope in the MLR models, the standardized β and adj-R² values remained approximately the same as when the intact α-chain content was used as an independent explanatory variable (Table 2). This implies that the intact α chain content impacts the contribution of the collagen network connectivity, Max.Slope, which has a strong standardized β of ~.56-.67, to cortical bone fracture resistance. The MLR models for J-int and Wfx_n that included intact α chain content produced similar adjusted R² values as the MLR models from our precursor study, where HLNL (one of the immature LOX crosslinks) and the lysine-based AGE adducts were detected as explanatory variables (both independently and in interactions with Max.Slope; see Supplementary Material Table S2). Furthermore, the intact collagen α-chain content had a similar standardized β as the lysine-based AGE adducts (β ~ .17-.20). This again highlights multiple collinearities presumably driven by a common driving force. Large changes in R² and large standardized β values were not expected in this study because molecular-scale measures alone are insufficient to explain all variation in tissue-level fracture toughness. Cortical bone fracture resistance emerges from interactions across multiple hierarchical length scales, including molecular modifications of collagen, collagen fibril organization, collagen–mineral interactions, osteonal architecture, cortical porosity, and crack-deflection mechanisms.

Like all studies, this study had some limitations. First, the donors in the T2D/CKD and control groups were not age nor BMI matched. This occurred because supply became restricted, as we collected the cadaveric femurs during the COVID-19 pandemic. This issue was handled by using ANCOVA as described in the Materials and methods and Results sections. This approach was taken as the main aim of the study was to investigate the impact of oxidative damage of cortical bone tissue fracture resistance and not simply test for differences between the two categorical disease groups. We were unable to match BMI as this information was not readily available for all the donors from the various tissue banks. In addition, T2D duration and HbA1c, a measure of blood glucose levels over the past 2-3 mo, were not available for all donors. Furthermore, homogeneity within the specimens used for the various characterization tests was assumed. This was a reasonable assumption as tissue from a 50 mm × 5 mm × 2.5 mm cortical bone specimen from each donor was used in all the experiments.

In conclusion, bulk collagen carbonyl content was not associated with collagen network connectivity nor cortical bone fracture toughness in this donor cohort, indicating that the hypothesized direct oxidative damage pathway was not supported by the present data. In contrast, intact collagen α-chain content emerged as a modest but statistically significant explanatory variable for stable crack-growth fracture toughness measures, was negatively associated with bulk carbonyl content, and influenced the contribution of collagen network connectivity to these models. Therefore, direct oxidative damage may contribute to collagen α-chain fragmentation, which may in turn affect collagen network integrity and its role in cortical bone fracture resistance. Combined with findings from our precursor study, we propose that oxidative stress may influence cortical bone fracture resistance through both direct and indirect pathways that alter collagen network quality, as summarized in the hypothetical framework presented in Figure 6. Controlled longitudinal studies using sensitive and specific measures of oxidative damage are needed to establish the proposed mechanistic links.

Figure 6.

Hypothetical framework for future investigation of the proposed mechanistic pathway by which oxidative stress may lead to oxidative damage via direct and indirect pathways within the collagen network resulting in degradation of the collagen network connectivity. This could result in reduced cortical bone fracture resistance because collagen connectivity is an important contributor to cortical bone’s ability to sustain stable crack growth.

Author contributions

Daniel Y. Dapaah (Conceptualization, Methodology, Investigation, Formal analysis, Writing—original draft), Shoutaro Arakawa (Methodology, Investigation, Writing—original draft), Gwennyth A. Carroll (Methodology, Investigation), Yiran Wang (Formal analysis), Stewart McLachlin (Methodology, Supervision, Funding acquisition), Mitsuru Saito (Methodology, Supervision, Writing—review & editing), and Thomas L. Willett (Conceptualization, Methodology, Supervision, Funding acquisition, Writing—review and editing)

Funding

This work was supported by Project Grant funding (PJT-159617) from the Canadian Institutes of Health Research (T.L.W.), infrastructure funding from the Canadian Foundation for Innovation (T.L.W.), the Ontario Research Fund (T.L.W.), and Natural Sciences and Engineering Research Council (NSERC) of Canada Research Tools and Instruments (S.M.) as well as scholarship funding from the NSERC CREATE Training Program in Global Biomedical Technology Research and Innovation and the Government of Ontario (D.Y.D.).

Conflicts of interest

The authors declare that they do not have any conflicts of interest.

Data availability

Data is available upon reasonable request to the corresponding author.

Ethics approval

This study was approved by the University of Waterloo’s Office of Research Ethics Committee (ORE #40888).