Altered tissue composition, microarchitecture, and mechanical performance in cancellous bone from men with type 2 diabetes mellitus

Abstract

People with type 2 diabetes mellitus (T2DM) have normal to high bone mineral densities, but, counterintuitively, have greater fracture risks than people without T2DM, even after accounting for potential confounders like BMI and falls. Therefore, T2DM may alter aspects of bone quality, including material properties or microarchitecture, that increase fragility independently of bone mass. Our objective was to elucidate the factors that influence fragility in T2DM by comparing the material properties, microarchitecture, and mechanical performance of cancellous bone in a clinical population of men with and without T2DM. Cancellous specimens from the femoral neck were collected during total hip arthroplasty (T2DM: n=31, age=65±8 years, HbA1c=7.1 ± 0.9%; non-DM: n=34, age=62±9 years, HbA1c=5.5±0.4%). The T2DM specimens had greater concentrations of the advanced glycation endproduct pentosidine (+36%, p<0.05) and sugars bound to the collagen matrix (+42%, p<0.05) than the non-DM specimens. The T2DM specimens trended toward a greater bone volume fraction (BV/TV) (+24%, NS, p=0.13) and had a greater mineral content (+7%, p<0.05) than the non-DM specimens. Regression modeling of the mechanical outcomes revealed competing effects of T2DM on bone mechanical behavior. The trend of higher BV/TV values and the greater mineral content observed in the T2DM specimens increased strength, while the greater values of pentosidine in the T2DM group decreased post-yield strain and toughness. The long-term medical management and presence of osteoarthritis in these patients may influence these outcomes. Nevertheless, our data indicate a beneficial effect of T2DM on cancellous microarchitecture, but a deleterious effect of T2DM on the collagen matrix. These data suggest that high concentrations of AGEs can increase fragility by reducing the ability of bone to absorb energy prior to failure, especially for the subset of T2DM patients with low BV/TV.

INTRODUCTION

People with type 2 diabetes mellitus (T2DM) have normal to high bone mineral densities (BMDs), but counterintuitively, also have greater fracture risks than people without diabetes.^(1,2) Furthermore, epidemiologic studies that account for hip BMD, age, body mass index (BMI), fracture history, neuropathy, retinopathy, and falls consistently underpredict fractures in men and women with T2DM.^(1,3,4) Because these factors do not fully account for the increased fracture risk observed in patients with T2DM, additional factors, including bone quality,⁽⁵⁾ may also contribute to the pathophysiology of T2DM-realted fractures. Aspects of bone quality that may contribute to decreased fracture resistance in T2DM include bone microarchitecture and tissue material properties, which can arise from changes in bone remodeling. Hyperglycemia and hyperinsulinemia, two conditions that typically characterize T2DM, alter bone remodeling and therefore are hypothesized to alter microarchitecture and tissue material properties.

Hyperglycemia may alter bone tissue material properties through disruption of bone remodeling via osteoblasts and osteoclasts and through formation of advanced glycation endproducts. In T2DM, bone remodeling can be altered by suppression of bone formation as well as bone resorption, as evidenced by in vitro studies,^(6–8) reports of decreases in the bone formation markers^(9–12) and bone resorption markers,^(10,11,13) and histomorphometric analyses.⁽¹²⁾ Moreover, serum glucose correlates with a decrease in bone turnover markers,⁽¹⁴⁾ and excess oral glucose intake is associated with decreased bone formation in otherwise healthy individuals.⁽¹⁵⁾ In contrast to hyperglycemia, insulin is an anabolic agent, and hyperinsulinemia may help explain the greater BMD observed in people with T2DM.⁽¹⁶⁾ Insulin signaling helps regulate osteoblastic proliferation and supports osteoclastogenesis.⁽¹⁷⁾ Additionally, insulin levels post-oral-glucose-tolerance-test are positively correlated with BMD at the hip and spine in men and women with and without T2DM.⁽¹⁸⁾ Nevertheless, the extent to which bone remodeling and material properties are affected by the competing effects of hyperglycemia and hyperinsulinemia is not yet known.

In addition to altered remodeling, excess glucose can lead to changes in bone tissue material properties through the accumulation of advanced glycation endproducts (AGEs). AGEs are the reaction products of reducing sugars with free amino groups in proteins and result in a diverse array of structures including crosslinking and non-crosslinking products. Crosslinking AGEs have been implicated in embrittling the collagen matrix in rodent models of T2DM⁽¹⁹⁾ and in in vitro ribosylation/glycosylation studies of human and bovine bone tissue.^(20,21) Non-crosslinking AGEs, like carboxymethyllysine (CML), can also be deleterious to bone tissue through interactions with the receptor for AGEs (RAGE), which induces oxidative stress and inflammation.⁽²²⁾ In a study of CML-modified collagen compared to unmodified collagen, apoptosis of bone lining cells was increased in vivo and in osteoblast cell cultures.⁽²³⁾ Together, AGE accumulation and the downstream effects of AGE-RAGE interactions may compound the changes in tissue material properties with T2DM and worsen mechanical performance.

Cross-sectional studies also indicate that bone microarchitecture may be altered in people with T2DM compared to people without T2DM, though the precise mechanisms and relationships to hyperglycemia and hyperinsulinemia are unknown. People with T2DM had similar or greater trabecular volumetric BMD at the femoral neck⁽²⁴⁾ and tibia versus non-diabetic controls,⁽²⁵⁾ and high-resolution peripheral quantitative computed tomographic assessment of tibial and radial trabecular bone revealed greater trabecular bone volume and plate volume fraction in women with T2DM versus women without T2DM.⁽²⁶⁾ Observations of cortical bone microarchitecture are mixed; cortical porosity of tissue from people with T2DM compared to people without T2DM has been reported to decrease at the proximal femoral shaft,⁽¹⁴⁾ to increase at the distal radius,⁽²⁷⁾ and to not significantly differ in the distal radius and distal tibia.^(11,28) The divergent findings between trabecular and cortical microarchitecture highlight the need for compartment-specific analyses in T2DM bone tissue in humans.

Although microarchitectural changes have been documented in people with T2DM, the functional consequences of these changes are largely unknown. Two studies have compared mechanical performance of tissue from men and women with and without T2DM, and the differences were found in cortical tissue. One study reported decreased resistance to in vivo impact indention in cortical bone at the tibial diaphysis in patients with T2DM versus those without T2DM,⁽¹¹⁾ and the other found decreased resistance to cyclic reference point indentation in cortical proximal femur tissue.⁽²⁹⁾ With the exception of these two studies, the majority of evidence that T2DM deleteriously affects bone mechanical performance is from rodent models. In rodent models, strength^(19,30,31) and work-to-failure^(19,31,32) were consistently decreased in rodents with T2DM compared to non-DM controls⁽³³⁾ and were often accompanied by concomitant changes in bone quality including altered microarchitecture,^(30–32,34) BMD,^(31,32,34) osteoid surface and thickness,⁽³²⁾ and AGE content.⁽¹⁹⁾

Similarly, the availability of data on tissue material properties in bone from humans with T2DM is limited. There are reports of greater concentrations of tissue AGEs⁽³⁵⁾ and greater mean calcium content in bone from people with T2DM compared to non-DM controls;⁽³⁶⁾ however, rodent models of T2DM provide the bulk of evidence that T2DM alters tissue material properties. Multiple rodent models of T2DM report increased mineral content (assessed by mineral:matrix ratio) in Zucker diabetic Sprague-Dawley (ZDSD) rats^(37,38) and KK-Ay mice compared to controls.⁽³⁹⁾ The same studies report no differences in concentrations of mature enzymatic crosslinks (pyridinoline and deoxypyridinoline) or in the AGE pentosidine;^(37–39) however, the collagen maturity (ratio of mature to immature enzymatic crosslinks) of the KK-Ay mice was increased compared to controls.⁽³⁹⁾ In contrast, a study of WBN/Kob rats that found a decrease in concentration of enzymatic crosslinks with a simultaneous increase in pentosidine concentration,⁽¹⁹⁾ which suggests phenotype-specific changes in tissue material properties with T2DM.

In summary, although bone from people with T2DM may possess altered tissue material and microarchitecture properties, the effects of such changes on bone tissue mechanical properties have not been assessed, and the mechanism by which T2DM affects bone fracture resistance is unknown. Moreover, most fragility fractures occur at cancellous sites, and the mechanical properties of cancellous bone from patients with T2DM have not yet been fully characterized.

Therefore, the objectives of this study were to relate material properties and microarchitecture to the mechanical performance of cancellous bone from the femoral neck in men with and without T2DM. We focused on cancellous bone because it preferentially accumulates AGEs relative to cortical bone⁽⁴⁰⁾ and on men because they have poor health outcomes associated with fragility fractures. The trends of greater fracture risk with T2DM are parallel in men and in women, although absolute hip fracture risk is higher in women than in men.⁽¹⁾ Furthermore, men are undertreated for fragility fractures and tend to experience worse health outcomes following fractures. Specifically, a smaller proportion of men at high risk of fracture are treated than women at high risk of fracture,⁽⁴¹⁾ and older men are twice as likely to die after a hip fracture compared to age-matched women.⁽⁴²⁾ We hypothesized that cancellous bone specimens from men with T2DM would have increased AGE concentrations, have greater bone volume fraction, and would be more brittle relative to non-DM controls.

METHODS

Study Cohort

Men undergoing total hip arthroplasty for osteoarthritis were sequentially recruited at a single, metropolitan training hospital (Hospital for Special Surgery, New York, NY) where participating surgeons obtained informed consent. Prior to arthroplasty, HbA1c was measured in all participants. All procedures were approved by the institutional review boards of the Hospital for Special Surgery and Cornell University.

Initial power analyses to detect 20% differences in mean mechanical properties at 80% power and 5% level of significance indicated 17 specimens/group were needed. A total of 75 men were recruited and allocated to two groups based on T2DM diagnosis at the time of surgery: men with a type 2 diabetes mellitus diagnosis (T2DM, n = 35) and men without a T2DM diagnosis (non-DM, n = 40). Four T2DM subjects and six non-DM subjects were excluded based on the following criteria: diagnosis of type 1 diabetes mellitus; a prior fragility fracture; any disease of bone such as osteogenesis imperfecta, fibrous dysplasia, or malignancy; renal or hepatic disorder involving the bone such as hyperparathyroidism or vitamin D deficiency; a history of avascular necrosis of the hip; treatment with medications that affect bone metabolism such as thiazolidinediones, teriparatide, glucocorticoids, bisphosphonates, or anticonvulsants; or pathological evidence of bone metastasis. In total, 31 T2DM subjects and 34 non-DM subjects were included in the analyses for this study.

Specimen Retrieval and Preparation

Femoral head and neck tissue retrieved from total hip arthroplasty was wrapped in saline-soaked gauze and stored at −20 ˚C prior to specimen preparation. The cancellous specimens were aligned along the principal trabecular axis which was determined from a radiograph. The specimens were then cut to 10 mm in length using a high-speed precision saw and a custom jig that ensured parallel ends. The tissue removed from the ends of the original core (~3 mm total length removed per end) was retained for compositional analyses. Figure 1 shows the allocation of tissue for all analyses. For 20 specimens (T2DM: n = 11, non-DM: n = 9), a 10-mm-long, uniform core could not be obtained because the proximal femur specimen was too small after retrieval and gross sectioning for pathological analysis. These 20 specimens, therefore, did not undergo mechanical testing or microCT analysis; however, cancellous tissue along the principal trabecular axis was still excised for compositional analyses.

Figure 1:

Anatomical location, size of cancellous specimen excised, and allocation of tissue for each characterization technique.

In all specimens, the tissue allocated for compositional testing was prepared as follows: Bone marrow from the excess tissue was removed with a dental water pick, defatted using three 15-minute soaks in isopropyl ether, and rinsed for 15 minutes in DI water. The tissue was then homogenized and allocated into two groups for compositional analysis: 1) FTIR and 2) HPLC and fluorescence spectroscopy.

Micro-computed Tomography

Micro-computed tomography was used to assess trabecular microarchitecture. Specimens were scanned in saline at 55 kVp and a resolution of 10 μm (μCT 35, Scanco Medical, Brüttisellen, Switzerland). Bone volume fraction (BV/TV), bone surface to volume ratio (BS/BV), connective density (Conn.D), trabecular number (Tb.N), trabecular thickness (Tb.Th), trabecular separation (Tb.Sp), and degree of anisotropy (DA) was evaluated (μCT 40, Scanco Medical, Brüttisellen, Switzerland).

Mechanical Testing

Monotonic compression testing was performed to determine the apparent-level mechanical properties of the specimens. Specimens were thawed, press-fit into custom, shallow-welled, brass endcaps, and secured with cyanoacrylate glue. The glue was allowed to cure for at least 10 minutes while maintaining specimen hydration. The load regimen consisted of 10 preconditioning cycles from 0% to 0.1% strain at a rate of 0.5% s⁻¹, followed by a single compressive load to 3% strain at a rate of 0.5% s⁻¹ before unloading to zero load (858 Mini Bionix, MTS, Eden Prairie, MN). Load was measured with a load cell (SSM-1000, Transducer Techniques, Temecula, CA), and strain was measured with an external extensometer (634.12 Axial Extensometer, MTS, Eden Prairie, MN) attached to the specimen endcaps.

Stress-strain curves were used to evaluate apparent-level Young’s modulus, yield strain, yield, stress, ultimate strain, ultimate stress, post-yield strain, post-yield toughness, toughness, and residual strain. Yield strain and stress were determined using the 0.2% strain offset method; ultimate strain and stress were determined from the maximum load; and post-yield strain was calculated as difference between ultimate strain and yield strain. Two different measures of toughness were determined: 1) toughness, which was defined as the toughness from zero strain to ultimate strain and 2) post-yield toughness, which was defined as the toughness from yield strain to ultimate strain. In five non-DM specimens, failure was not observed before the end of the test at 3% strain; therefore, ultimate strain, ultimate stress, and post-yield toughness data were not available for those specimens.

Fourier Transform Infrared Spectroscopy

FTIR spectroscopy was used to evaluate the compositional properties of cancellous bone tissue. The demarrowed and defatted tissue was lyophilized in a vacuum drier then powdered using a cryomill (6770, SPEX SamplePrep, Metuchen, NJ). Two mg of dried, powdered tissue was added to 200 mg of dried KBr, then pressed into a pellet using a 13-mm-diameter die. FTIR spectra were collected at a spectral resolution of 4 cm⁻¹ over the spectral range of 800 to 2000 cm⁻¹ using an FTIR spectrometer (Spotlight 400, Perkin-Elmer Instruments, Waltham, MA).

Using custom Matlab code (2014a, The MathWorks, Inc., Natick, MA), the spectra were baseline corrected (as detailed in ⁽⁴³⁾) and the following parameters were calculated: 1) the mineral:matrix ratio (area ratio of the phosphate ν₁-ν₃ peak [916–1180 cm⁻¹] to amide I peak [1596–1712 cm⁻¹]), which characterizes tissue mineral content;⁽⁴⁴⁾ 2) the collagen maturity (intensity ratio of 1660 cm⁻¹ to 1690 cm⁻¹), which is related to the ratio of pyridinoline to divalent crosslinks;⁽⁴⁴⁾ 3) the mineral crystallinity (intensity ratio of 1030 cm⁻¹ to 1020 cm⁻¹), which is related to crystal size and stoichiometric perfection;⁽⁴⁴⁾ the carbonate:phosphate ratio (area ratio of the carbonate ν₂ peak [852–890 cm⁻¹] to phosphate ν₁-ν₃ peak [916–1180 cm⁻¹]), which characterizes carbonate substitution into hydroxyapatite;⁽⁴⁴⁾ and the acid phosphate content (intensity ratio of 1127 cm⁻¹ to 1096 cm⁻¹), which characterizes acid phosphate substitution into stoichiometric hydroxyapatite.⁽⁴⁵⁾

Additionally, the sugar:matrix ratio, was examined to assess sugars bound to the collagen matrix. The sugar:matrix ratio (area ratio of ν CO and ν CC peaks [900–1100 cm⁻¹]⁽⁴⁶⁾ to amide I peak [1596–1712 cm⁻¹]) is related to the amount of sugars (including glucose and ribose) attached to the matrix.⁽⁴⁷⁾ The sugar:matrix ratio was evaluated following decalcification (see Supplemental S1) because the prominent ν₁ phosphate overlaps the sugar peaks. For a subset of the specimens, there was not enough tissue after demineralization to detect an infrared signal (T2DM: n = 5, non-DM: n = 5).

High-Performance Liquid Chromatography

The concentrations of pentosidine and two enzymatic crosslinks, pyridinoline (Pyd) and deoxypyridinoline (Dpd), were determined using HPLC and expressed normalized to collagen concentration (e.g., mol/mol collagen). The total content of mature enzymatic crosslinks (sum of Pyd and Dpd) was calculated, as well as the ratio of non-enzymatic crosslinks (pentosidine) to mature enzymatic crosslinks, which indicates the ratio of physiologically disadvantageous crosslinks to advantageous crosslinks (similarly described in reference ⁽⁴⁸⁾).

Tissue allocated for HPLC and fluorescence spectroscopy analyses was lyophilized then hydrolyzed in 6 N HCl at 110 °C for 18 h. An aliquot of the hydrolysate was reserved for fluorescence spectroscopy, and the remainder was dried in a vacuum centrifuge (Speed Vac SC110A, Savant, Farmingdale, NY). The dried hydrolysate was re-suspended in an internal standard solution containing 10 nM pyridoxine and 2.4 μM homoarginine (30 μl internal standard per 1 mg dried bone) and filtered with a 0.45 μm membrane. The re-suspended samples were diluted 1:5 with 10% acetonitrile (v/v) and 0.5% heptafluorobutyric acid (v/v).

Pentosidine, Pyd, and Dpd were separated on a C-18 column (Gemini-NX C-18, Phenomenex, Torrance, CA) using two isocratic steps⁽⁴⁹⁾ and a programmable HPLC system (Model 126, Beckman Coulter, Inc., Fullerton, CA) attached to a fluorescence detector (Model FP1520, Jasco, Easton, MD). The calibration standard containing pentosidine (Case Western Reserve University, Cleveland, OH), Pyd (8004, Quidel, Athens, OH), and Dpd (8004, Quidel, Athens, OH) was created through serial dilution.

The concentrations of pentosidine, Pyd, and Dpd were normalized by collagen concentration determined by hydroxyproline concentration (assuming 300 mol hydroxyproline per mol collagen). To measure hydroxyproline, amino acid analysis was performed on an aliquot of the further diluted sample from the crosslink analysis.⁽⁵⁰⁾ Briefly, the crosslink sample used for separation was diluted 1:50 with 6 μM homoarginine (amino acid internal standard) in 0.1 M borate buffer (pH 11.4). For enhanced detection, the sample solution was derivatized using 6 mM 9-fluorenylmethyl chloroformate for 40 minutes and extracted three times with pentane. After derivatization, 25% (v/v) acetonitrile in 0.25 M boric acid (pH 5.5) was added. A calibration standard of purified hydroxyproline (Sigma-Aldrich, St. Louis, MO) and 6 μM homoarginine in 0.1 M borate buffer (pH 11.4) was created through serial dilution. The separation of amino acids was run using the injection sequence described by Bank et al.⁽⁵⁰⁾ Data analysis for crosslink and hydroxyproline determination was performed using 32 Karat Workstation software (v. 5.0, Beckman Coulter, Inc., Brea, CA) and Matlab (2014a, The MathWorks, Inc., Natick, MA).

Fluorescence Spectrometry

Total fluorescent AGEs (fAGEs) were measured using fluorescence spectrometry and normalized to collagen concentration. The endogenous fluorescence of cancellous bone tissue was compared to the fluorescence of a quinine standard (149504, Sigma Aldrich, St. Louis, MO). The aliquot of hydrolysate, reserved from that used for the HPLC separations, was diluted with DI water to a concentration of 3 μg bone/ml solution. The fluorescence of diluted bone hydrolysate and the quinine standards were measured in a 96-well plate using a multi-mode microplate reader (Synergy H1, BioTek, Winooski, VT) at an excitation of 360 nm and an emission of 460 nm.

A colorimetric assay of hydroxyproline was used to determine collagen content to normalize the bulk fluorescence. For the bone tissue samples, the hydrolysate measured for bulk fluorescence was diluted with deionized water to 0.3 μg bone/ml. To initiate the reaction, chloramine-T was added to the hydroxyproline standards and the diluted sample hydrolysates. The standards and samples were incubated for 20 minutes at room temperature, then 3.15 M perchloric acid was added to stop the reaction. After sitting 5 minutes at room temperature, p-dimethylaminobenzaldehyde was added. The standards and samples were incubated at 60 °C in a water bath for 20 minutes, then cooled in cold water in darkness to room temperature. The absorbance of the specimens and standards was measured at a wavelength of 570 nm in a 96-well plate a multi-mode microplate reader. Total fAGEs are reported in units of ng quinine fluorescence/mg collagen.

Statistical Analyses

Between-group differences in the participant characteristics, compositional, microarchitectural, and mechanical outcomes were determined by Student’s t-tests, Welch’s t-tests, Mann-Whitney U tests, or Kolmogorov-Smirnov tests, as appropriate, after testing for normality and homogeneity of variances. In the T2DM group only, dichotomous variables (yes/no) for insulin use, metformin use, and other anti-DM medication use (Table 1) were created, and ANOVAs were calculated for the compositional, microarchitectural, and mechanical outcomes with medication use as the grouping variable. Univariate linear regressions were performed with all mechanical, compositional, and microarchitectural outcomes as the dependent variable and pre-operative HbA1c-- a serum metric routinely assessed clinically--as the independent variable to determine if recent glycemic control over the three months prior to surgery is related to characteristics of bone tissue that are only measurable with bone biopsies. A significance level of p = 0.05 was used for all analyses.

Table 1:

Group characteristics of the full cohort and the mechanical testing cohort. Values are shown as means ± standard deviation, unless otherwise noted. Coexisting conditions were determined from medical records at the time of total hip arthroplasty. p values assessed by Student’s t-test, Welch’s t-test, Mann-Whitney U test, Kolmogorov-Smirnov test, or Pearson chi-square test, as appropriate. Abbreviations: non-DM: non-diabetic group, T2DM: type 2 diabetic group; THA: total hip arthroplasty; TKA: total knee arthroplasty.

	Full Cohort			Mechanical Testing Cohort
Characteristic	non-DM	T2DM	p	non-DM	T2DM	p
	(n = 34)	(n = 31)		(n = 25)	(n = 20)
Demographic
Age (years)	61.6 ± 8.6	64.8 ± 8.1	0.131	60.6 ± 8.0	64.8 ± 7.1	0.073
Past surgical THA or TKA, n (%)	10 (29.4)	6 (19.4)	0.347	9 (36.0)	4 (20.0)	0.239
Biochemical
Pre-operative HbA1c (%)	5.53 ± 0.39	7.07 ± 0.89	< 0.05	5.52 ± 0.40	7.21 ± 1.01	< 0.05
Anthropometry
Height (cm)	178.9 ± 6.9	177.2 ± 8.0	0.369	180.0 ± 6.9	176.4 ± 8.1	0.126
Weight (kg)	93.7 ± 21.1	99.6 ± 17.7	0.222	97.7 ± 21.9	97.2 ± 18.6	0.945
Body mass index (kg/m2)	29.8 ± 6.3	32.3 ± 5.3	0.084	30.9 ± 6.7	32.2 ± 6.0	0.508
Coexisting conditions
Hypertension, n (%)	15 (44.1)	20 (64.5)	0.099	11 (44.0)	14 (70.0)	0.081
Hyperlipidemia, n (%)	15 (44.1)	18 (58.1)	0.261	11 (44.0)	14 (70.0)	0.081
Chronic kidney disease, n (%)	1 (2.9)	2 (6.5)	0.501	1 (4.0)	2 (10.0)	0.427
Coronary artery disease, n (%)	1 (2.9)	6 (19.4)	< 0.05	1 (4.0)	2 (10.0)	0.427
Current medications/supplements
Insulin, n (%)	0 (0.0)	7 (22.6)	< 0.05	0 (0.0)	4 (20.0)	< 0.05
Biguanides (Metformin), n (%)	0 (0.0)	23 (74.2)	< 0.05	0 (0.0)	13 (65.0)	< 0.05
Other antidiabetic medications1, n (%)	0 (0.0)	14 (45.2)	< 0.05	0 (0.0)	7 (35.0)	< 0.05
Vitamin D, n (%)	3 (8.8)	8 (25.8)	0.068	2 (8.0)	8 (40.0)	< 0.05

¹Antidiabetic medications include sulfonylureas (SUs): glyburide, glipizide, glimepiride; glinides: nateglinide; glucagon-like peptide 1 (GLP-1) agonists: dulaglutide, liraglutide, exenatide; and depeptidyl peptidase 4 (DPP-4) inhibitor: sitagliptin

A principal component analysis (PCA) was used to determine whether variation in the large number of interrelated measured compositional variables could be explained in terms of a smaller number of independent variables (factors). Because many of the primary compositional variables are interrelated (e.g., pentosidine and fAGEs; crystallinity and carbonate:phosphate ratio), we expected that PCA would yield a smaller number of independent compositional factors for use in regression modeling. Specifically, PCA was performed on mineral:matrix ratio, collagen maturity, crystallinity, carbonate:phosphate ratio, acid phosphate content, Pyd concentration, Dpd concentration, pentosidine concentration, and total fAGEs to detect underlying relationships in the compositional variables. Details of the rationale for creation of factors are summarized in Supplemental Section S4.

Regression modeling was used to determine which compositional and microarchitectural factors were the most important explanatory variables of mechanical performance and how a single explanatory variable can influence a mechanical outcome when holding all other influential factors (like BV/TV) constant. Four mechanical properties were modeled: 1) Young’s modulus, 2) ultimate stress, 3) post-yield strain, and 4) post-yield toughness. These properties respectively describe the bulk tissue stiffness, strength, ductility, and toughness, and together offer a comprehensive evaluation of the monotonic compressive properties of bulk cancellous tissue. First, stepwise selection regressions using the Akaike information criterion with the small sample size correction (AICc)^(51,52) were used to determine which compositional (AGEs, enzymatic crosslinks, mineral properties) and microarchitectural parameters were the most important determinants of mechanical performance. Linear^(53,54) and power law^(55,56) relationships were assessed to determine the influence of BV/TV on mechanical properties (see Supplemental Section S2 and Supplemental Table 1), and a linear BV/TV term was elected for use in these models. For highly collinear variables (e.g., Tb.Th, Tb.N, Tb.Sp), only one variable was included in the stepwise model. Patient age was included as a fixed effect, and the factor that emerged from the principal component analysis of the compositional outcomes was included as a potential model effect in place of the variables that constituted the factor. The interaction of Group with all potential compositional and microarchitecture predictors was also included. To ascertain whether or not T2DM-realted complications (e.g., CVD, CKD, Table 1) and influenced the measured mechanical properties, dichotomous variables for these complications were also included as potential effects. While participants with diagnosed vitamin D deficiency were excluded, a dichotomous variable for whether or not the participant was taking vitamin D supplements (Table 1) was included in the models to ascertain whether vitamin D supplementation influenced the mechanical properties. To maintain power for the regression analyses (T2DM: n = 20, non-DM: n = 25), the sugar:matrix ratio was not included as a potential effect. A goodness of fit check was performed for each model (see Supplemental Section S5).

A visual representation that shows the direction and magnitude of each predictor in the post-yield toughness model was generated. The grand mean of all predictor variables was calculated for the non-DM and T2DM groups, and these values were entered into the model for post-yield toughness to ascertain the predicted average post-yield toughness. Next, the predicted post-yield toughness was calculated for an increase of one standard deviation to the grand mean of each predictor variable, while holding all other predictors at their respective grand means (Figure 4A). Group was set to T2DM, and age was set to 63 years (grand mean). The values of the material parameters were fixed at the grand mean + 1 SD, and the effects of varying BV/TV through high (grand mean + 1 SD), medium (grand mean), and low (grand mean – 0.5 SD) values were examined. (The low BV/TV value was set to grand mean – 0.5 SD instead of grand mean – 1 SD because the latter resulted in a negative value of post-yield toughness, which was not physically meaningful.) Figure 4B-D shows the model-predicted post-yield toughnesses of these three scenarios.

Figure 4:

The regression model for post-yield toughness (Table 3) was used to calculate predicted post-yield toughness under different circumstances. (A) The predicted post-yield toughness of a specimen with grand mean values of BV/TV, MMF, pentosidine, and total fAGEs is shown with the top gray box and the vertical black dashed line. The predicted post-yield toughness of each predictor parameter was varied one at a time from the grand mean to the grand mean plus one standard deviation to show the magnitude and direction of change and is indicated with a gray box. For (B-D), MMF, pentosidine, and total fAGEs were evaluated at the mean plus one standard deviation and BV/TV was evaluated at (B) the grand mean of BV/TV plus one standard deviation, which illustrates a low risk scenario; (C) the grand mean of BV/TV, which illustrates a medium risk scenario; and (D) the grand mean of BV/TV minus one half of one standard deviation, which illustrates a high risk scenario. The bottom black box in Panels B-D indicates the predicted post-yield toughness for that scenario, and it is also the cumulative effects of all predictor parameters. The horizontal gray lines indicate the 95% CI.

RESULTS

Group Characteristics

In the full cohort, the T2DM group had a higher pre-operative HbA1c than the non-DM group (+28%, p < 0.05), thus confirming hyperglycemia in the T2DM group and euglycemia in the non-DM group (Table 1). Age, past surgical history, height, weight, and BMI did not differ across groups, and the T2DM group possessed a higher prevalence of coronary artery disease (p < 0.05) with a trend toward a higher prevalence of hypertension (p = 0.099) compared to the non-DM group (Table 1). Twenty-three percent of the T2DM group used insulin (n = 7/31), 74% used metformin (n = 25/31), and 45% used other antidiabetic medications (n = 13/31) (Table 1). None of the compositional, microarchitectural, or mechanical properties varied within the T2DM group as a result of insulin, metformin, or other antidiabetic medications use, as determined by ANOVA.

The mechanical testing cohort differed slightly from the full cohort. First, although the mean ages of the T2DM and non-DM groups were not statistically different in the full cohort, the T2DM group was borderline older than the non-DM group in the mechanical testing cohort (+7%, p = 0.073) (Table 1). Accordingly, mechanical testing data were adjusted for age (see Statistical Analyses). Second, in the mechanical testing cohort, the T2DM group had a higher prevalence vitamin D supplementation than the non-DM group (p < 0.05) (Table 1).

Advanced Glycation Endproducts

The concentration of the AGE pentosidine was 36% greater and the sugar:matrix ratio was 42% greater in the T2DM group compared to the non-DM group (both p < 0.05) (Figure 2A). Concentration of total fAGEs did not differ across groups (Figure 2A). Pentosidine, the sugar:matrix ratio, and total fAGEs were not correlated among each other.

Figure 2:

Compositional properties of the full cohort organized by (A) measures of advanced glycation endproducts, (B) measures of enzymatic crosslinks, and (C) measures of mineral properties. * p < 0.05 and ^# p < 0.10 assessed by Student’s t-test, Welch’s t-test, Mann-Whitney U test, or Kolmogorov-Smirnov test. Abbreviations: non-DM: non-diabetic group; T2DM: type 2 diabetic group; Total fAGEs: total fluorescent advanced glycation endproducts; Pyd: pyridinoline; Dpd: deoxypyridinoline; XST = crystallinity; C:P = carbonate:phosphate ratio; AP = acid phosphate content.

Enzymatic Crosslinks

The concentration of the enzymatic crosslink Pyd trended towards being lower in the T2DM group than the non-DM group (−10%, p = 0.071) (Figure 2B), while the concentration of Dpd did not differ across groups (Figure 2B). The total content of mature/trivalent enzymatic crosslinks (sum of Pyd and Dpd) measured with HPLC and collagen maturity (XLR, Figure 2B) measured with FTIR did not differ across groups.

Ratio of Non-enzymatic to Enzymatic Crosslinks

The ratio of non-enzymatic crosslinks to mature crosslinks (or disadvantageous crosslinks to advantageous crosslinks⁽⁴⁸⁾), calculated as pentosidine/(Pyd+Dpd), was 61% greater in the T2DM group compared to the non-DM group (T2DM: 9.94 ± 6.42, non-DM: 6.17 ± 2.42, p < 0.05).

Mineral Properties

The mineral:matrix ratio was 7% greater in the T2DM group versus the non-DM group (p < 0.05) (Figure 2C). The crystallinity, carbonate:phosphate ratio, and acid phosphate content did not differ across groups (Figure 2C).

Microarchitecture

In T2DM compared to non-DM specimens, Tb.Sp was lower (−13%, p < 0.05), Tb.N trended toward a greater value (+20%, p = 0.059), and Conn.D trended toward greater value (+96%, p = 0.061) (Table 2). All other microarchitectural parameters (BV/TV, BS/BV, Tb.Th, and DA) did not differ across groups (Table 2). Neither weight nor BMI were correlated with BV/TV suggesting that body mass did not increase bone volume.

Table 2:

MicroCT assessed microarchitecture of the mechanical cohort specimens. Values are shown as means ± standard deviation. p values assessed by Student’s t-test, Welch’s t-test, Mann-Whitney U test, or Kolmogorov-Smirnov test. Abbreviations: non-DM: non-diabetic group, T2DM: type 2 diabetic group; BV/TV: bone volume fraction; BS/BV: specific bone surface; Conn.D: connectivity density; Tb.N: trabecular number; Tb.Th: trabecular thickness; Tb.Sp: trabecular separation; DA: degree of anisotropy.

Microarchitecture Parameter	non-DM	T2DM	T2DM vs. non-DM	p
	(n = 25)	(n = 20)	(% difference)
BV/TV (%)	15.81 ± 7.30	19.63 ± 8.69	24	0.125
BS/BV (mm2/mm3)	19.01 ± 3.79	18.22 ± 4.24	−4	0.520
Conn.D (1/mm3)	10.52 ± 9.19	20.65 ± 21.52	96	0.061
Tb.N (1/mm)	1.27 ± 0.29	1.52 ± 0.49	20	0.059
Tb.Th (mm)	0.15 ± 0.03	0.16 ± 0.07	7	0.534
Tb.Sp (mm)	0.80 ± 0.15	0.71 ± 0.12	−13	< 0.05
DA	1.77 ± 0.24	1.84 ± 0.23	4	0.394

Mechanical Properties

Compression testing revealed that Young’s modulus was 86% greater (p < 0.05) (Supplemental Figure S1A), yield stress was 91% greater (p < 0.05) (Supplemental Figure S1C), and ultimate stress was 90% greater (p < 0.05) (Supplemental Figure S1F) in the T2DM versus non-DM group. Yield strain, ultimate strain, post-yield strain, post-yield toughness, toughness, and residual strain did not differ across groups (Supplemental Figure S1B, S1E, S1G, S1H, S1I, S1K).

The mechanical property outcomes of Young’s modulus, yield stress, ultimate stress, and both measures of toughness were linearly normalized by each specimen’s respective BV/TV (mechanical property/BV/TV, see Supplemental Section S2 and Supplemental Table 1). Normalized Young’s modulus (Figure 3A), normalized yield stress, and normalized ultimate stress (Figure 3B) remained 51%, 55%, 45% greater in the T2DM versus non-DM group, respectively (all p < 0.05). Normalized post-yield toughness (Figure 3C) and normalized toughness did not differ across groups.

Figure 3:

(A) Young’s modulus normalized by BV/TV, (B) Ultimate stress normalized by BV/TV, and (C) Post-yield toughness normalized by BV/TV. Horizontal lines indicate group means. * p < 0.05 by Student’s t-test. Abbreviations: non-DM: non-diabetic group, T2DM: type 2 diabetic group; BV/TV: bone volume fraction. A version of this figure that also shows the relationship of mineral content with the mechanical properties normalized by BV/TV can be seen in Supplemental Figure S3.

Mineral Maturity Factor

PCA was used to determine whether variation in the interrelated compositional variables could be explained in terms of a smaller number of independent variables (factors). Using PCA, we identified a factor (factor loadings > |0.8|, eigenvalue = 2.68) that we termed the mineral maturity factor (MMF) because it consisted of FTIR crystallinity, carbonate:phosphate ratio, and acid phosphate content, all variables that characterize the composition and size of bone mineral crystals. The original variables were standardized and coded such that higher MMF values indicate a more mature mineral crystal, with larger/more perfect crystals and lower carbonate and acid phosphate substitution. The MMF captured 89% of the variance of the three included parameters and was included in the subsequent regression analyses.

The MMF was not statistically different across groups (mean ± SD, T2DM: MMF = −0.101 ± 0.878, non-DM: MMF = 0.129 ± 1.073, p = 0.384).

Regression Analyses with Pre-operative HbA1c

Correlations between pre-operative HbA1c and compositional parameters revealed that pre-operative HbA1c was weakly and positively correlated the mineral:matrix ratio (R² = 0.182, p < 0.05). All other compositional parameters were not correlated with pre-operative HbA1c.

Correlations between pre-operative HbA1c and microarchitectural parameters revealed that pre-operative HbA1c weakly and positively correlated with Tb.N (R² = 0.077, p < 0.05) and weakly and negatively correlated with Tb.Sp (R² = 0.091, p < 0.05). No other microarchitectural parameters were correlated with pre-operative HbA1c.

Correlations between pre-operative HbA1c and mechanical properties revealed that Young’s modulus was weakly and positively correlated with HbA1c (R² = 0.088, p < 0.05). Pre-operative HbA1c was not correlated with any other mechanical outcomes.

Predicted Regression Analyses for Mechanical Properties

A combination of microarchitectural and compositional parameters explained between 12% and 83% of the observed variation in Young’s modulus, ultimate stress, post-yield strain, and post-yield toughness (Table 3). Age, CVD status, CKD status, and vitamin D supplementation were included as fixed effects for each model to control for variations in patient age and the possible influence of known complications (Table 1). Neither age, CVD status, CKD status, or vitamin D supplementation were significant explanatory variables in any of the models of mechanical properties.

Table 3:

Final regression models of selected mechanical properties organized by parameter type (fixed effect, microarchitecture, or composition). Values are shown as parameter coefficients with standard error in parentheses. Regression coefficients were determined from forward stepwise regressions using the Akaike information criterion (AICc). Abbreviations: BV/TV: bone volume fraction; Tb.Th: trabecular thickness; Total fAGEs: total fluorescent AGEs.

Parameter	Young’s Modulus (MPa)	Ultimate Stress (MPa)	Post-yield Strain (%)	Post-yield Toughness (kPa)
Intercept	−1444.4 (676.1)	−12.44 (7.45)	1.446 (0.485)	−3.86 (38.60)
Group	−58.2 (36.4)	---	---	−8.79 (4.68)
Fixed Effect
Age (years)	3.4 (4.1)	0.05 (0.06)	−0.005 (0.008)	−0.08 (0.58)
Microarchitecture
BV/TV (%)	43.8 (66.6)	0.51 (0.06)	---	5.57 (0.73)
TbTh (mm)	2731.6 (1634.5)	---	---	---
TbTh*Group	−3039.4 (980.9)	---	---	---
Composition
Mineral:Matrix (unitless)	120.5 (100.6)	1.15 (1.20)	---	---
Mineral Maturity Factor (unitless)	---	---	−0.164 (0.063)	−10.56 (5.62)
Pentosidine (mmol/mol collagen)	---	---	−0.038 (0.032)	−6.10 (2.72)
Total fAGEs (nq quinine/mg collagen)	---	---	---	−0.10 (0.09)
Total fAGEs*Group	---	---	---	0.17 (0.09)
Adjusted R2	0.826	0.694	0.116	0.739

Eighty-three percent of the variation in Young’s modulus was explained by BV/TV, Tb.Th, mineral:matrix ratio, Group, and the interaction of Group with Tb.Th; 70% of the variance in ultimate stress was explained by BV/TV and mineral:matrix ratio; 12% of the variance in post-yield strain was explained by the mineral maturity factor and pentosidine; and 74% of the variance in post-yield toughness was explained by BV/TV, the mineral maturity factor, pentosidine, total fAGEs, Group, and the interaction of Group with total fAGEs (Table 3). The model-predicted values of each mechanical property for both groups fell within the 95% CI of the experimentally measured mechanical property values (see Supplemental Section S5 and Supplemental Table S2).

Hypothetical Risk Scenarios Determined by Predicted Post-yield Toughness

Regression analysis of individual predictors for post-yield toughness indicated that BV/TV had a large and positive effect, as expected, while MMF, pentosidine, and total fAGEs had more modest, negative effects. Specifically, a one standard deviation increase in BV/TV from the grand mean increased the predicted post-yield toughness by 68%, while a one standard deviation increase in MMF, pentosidine, or total fAGEs decreased predicted post-yield toughness by 14%, 16%, or 19%, respectively (Figure 4A). As anticipated, the directions of these changes corresponded to the signs of the predictor coefficients (Table 3).

Three risk scenarios were examined for post-yield toughness based on the regression model coefficients (Table 3) and Figure 4A. The compositional predictors of MMF, pentosidine, and total fAGEs were held at their respective grand means plus one standard deviation, and group was set to T2DM for each scenario. For the low-risk scenario, a high BV/TV (grand mean +1 SD) was considered (Figure 4B); for the medium-risk scenario, an average BV/TV was considered (grand mean) (Figure 4C); and for the high-risk scenario, a moderately low BV/TV (grand mean – 0.5 SD) was considered (Figure 4D). The predicted post-yield toughness with all parameters set to their respective grand means was 66 kPa. For the low-, medium-, and high-risk scenarios, the predicted post-yield toughnesses were 79 kPa, 34 kPa, and 11 kPa, respectively.

DISCUSSION

In this work, we describe the compositional, microarchitectural, and mechanical properties of cancellous bone at the femoral neck in men with and without T2DM. To further our understanding of how changes in composition and microarchitecture with T2DM affect mechanical performance, we generated linear regression models for key mechanical properties. As hypothesized, the concentration of AGEs was greater in cancellous tissue in the T2DM group. Contrary to our hypothesis, post-yield properties were not different across groups at the femoral neck; however, our statistical models showed that the altered tissue composition in the T2DM group decreases post-yield stain and post-yield toughness after accounting for the large influence of BV/TV.

Compositional characterization revealed a greater concentration of pentosidine and a greater sugar:matrix ratio in the T2DM specimens compared to the non-DM specimens. These results confirm increased accumulation of the AGE pentosidine and sugars bound to the collagen matrix in cancellous tissue from men with T2DM and are consistent with other reports of bone tissue pentosidine in men and women with and without T2DM.⁽³⁵⁾ Total fAGE concentration in this study was not different between groups, consistent with another report of tissue fAGEs in proximal femoral cancellous bone in men and women with and without T2DM.⁽²⁹⁾ In vitro glycosylation/ribosylation studies of human cancellous bone offer the ability to isolate the effects of AGE accumulation. For age-matched groups, fAGE concentrations reported by Tang et al.⁽²¹⁾ for in vivo controls and in vitro glycated femoral cancellous bone (62 years) are within the range of 150–200 ng quinine/mg collagen reported here.

The apparent discrepancy between a highly specific measure of AGEs (i.e., pentosidine), a global measure of sugars attached to the collagen matrix (i.e., sugar:matrix ratio), and a non-specific measure of AGEs that fluoresce at a specific wavelength (i.e., total fAGEs) underscores the current limitations of measuring AGE accumulation. For example, pentosidine is just one of hundreds of AGEs, so despite its ability to precisely characterize pentosidine concentration, its overall contribution to total AGE accumulation is unknown. Conversely, measurement of total fAGEs may capture a wider range of AGEs, but only fluorescent AGEs are measured, and all non-fluorescent AGEs like CML are undetected. Finally, the ratio of non-enzymatic crosslinks to mature enzymatic crosslinks was 61% greater in the T2DM specimens compared to the non-DM specimens. Together, these findings provide clear evidence of detrimental crosslink accumulation in T2DM tissue and have translational relevance because crosslink accumulation stiffens the collagen matrix, which could in turn make bone in people with T2DM more brittle and less resistant to fracture.

In addition to a greater accumulation of AGEs in the T2DM specimens compared to the non-DM specimens, we observed additional evidence of altered collagen and mineral properties. The concentration of the trivalent enzymatic crosslink Pyd measured with HPLC trended towards being lower in the T2DM specimens compared to the non-DM specimens. This finding, though not statistically significant, may indicate a less mature collagen matrix and is consistent with a study of WBN/Kob rats that found reductions in enzymatic crosslinks with T2DM.⁽¹⁹⁾ Reductions in mature crosslinks in tissue from patients with T2DM may lead to decreased strength as evidenced by the association of mature pyridinoline crosslinks and tissue strength.^(57,58)

The mineral:matrix ratio, which characterizes tissue mineral content, was 7% greater in specimens in the T2DM group than the non-DM group. Our finding of increased mineral content with T2DM is consistent with findings of greater mean calcium content in the proximal femur of men and women with T2DM compared to non-diabetic controls.⁽³⁶⁾ A more mineralized tissue is consistent with decreased ductility,⁽⁵⁹⁾ thus the greater mineral content observed in the tissue from the T2DM group likely contributes to the increased fracture risk in people with T2DM. The observed greater mineral content may indicate an increased tissue age (i.e., time since formation),⁽⁶⁰⁾ of the T2DM versus non-DM tissue. Alternatively, an increase in tissue mineral content may be due to an increase in mineralization nucleation sites.⁽⁶¹⁾ FTIR-assessed mineral properties and the mineral maturity factor (MMF), which combined mineral crystallinity with carbonate substitution and acid phosphate content, were not different across groups. This suggests that the overall maturity of the mineral in the T2DM group is similar to that of the non-DM group.

The simultaneous increase in mineral content without an increase in mineral maturity conveys complex effects of T2DM on mineralization, and it is likely that multiple processes responsible for the observed properties are occurring concurrently (reviewed recently in ^(2,62)). One potential process is that hyperglycemia and AGE accumulation, which have been implicated in decreased osteoblastic function and proliferation in vitro,^(6–8,23) may disrupt remodeling and could account for the increased bulk tissue mineral content. A second possible process is that AGE accumulation decreases mineralization rates and alters normal mineralization processes,⁽⁴⁸⁾ as evidenced by decreases in mineral apposition rates, bone formation rates, and the number of bone formation sites in AGE-modified rat tissue.⁽⁶³⁾ A third possibility is that the formation of AGEs inherently alters the charge profile of the collagen fibers,^(64,65) thereby disrupting cell-matrix interactions that could cause downstream effects for collagen and mineral maturation.^(61,66) The current study was not designed a priori to study these biochemical pathways; however, it is feasible that a simultaneous reduction in remodeling rates and disruption of mineralization and collagen maturation with T2DM via these proposed processes would culminate in the tissue properties observed here.

In the microarchitectural analyses, the T2DM specimens had smaller separation distances between trabecula than the non-DM specimens. Additionally, the T2DM specimens had a borderline greater number of trabecula (+20%, p = 0.059) and borderline greater connective density (+96%, p = 0.061) than the non-DM specimens. The BV/TV of the T2DM specimens was not different than that of the non-DM specimens (+24%, p = 0.125). Even though some of the differences in microarchitecture outcomes did not meet the p < 0.05 threshold for statistical significance, overall, the T2DM specimens had more numerous, more connected, and closer trabeculae, which are known to contribute to a greater BV/TV. Therefore, the 24% greater BV/TV observed in the T2DM group versus non-DM group, though not statistically significant, is consistent with these other measured microarchitectural parameters. Furthermore, the 24% difference in BV/TV is similar to significant differences reported in larger studies of trabecular microarchitecture in patients with T2DM⁽⁶⁷⁾ and is consistent with prior observations of greater trabecular hip vBMD in patients with T2DM compared to those without.⁽²⁴⁾ Interestingly, cancellous microarchitecture may vary between the hip, spine, and distal sites. Analyses of DXA images for trabecular bone score (TBS) indicate that the TBS of the lumbar spine was lower in women with T2DM compared to those without T2DM, despite higher BMD at this site,^(68,69) thereby indicating a poor-quality microarchitecture.⁽⁷⁰⁾ Peripheral QCT assessment of the microarchitecture of the distal tibiae and radii of participants in the MrOS cohort showed that men with T2DM had a lower strength to body weight ratio than men without T2DM at these sites.⁽⁷¹⁾ In contrast, a recent analysis from the MrOS cohort did not find a substantial increase in vertebral fractures in men with T2DM compared to men without T2DM, even after accounting for the greater BMD in the T2DM group.⁽⁷²⁾ Together, these studies highlight how microarchitecture can vary from site to site and with T2DM status, and suggest that the effects of T2DM on skeletal tissue are related to the proportion of cancellous and cortical bone.

The Young’s modulus and ultimate stress of specimens from the T2DM group were greater than those from the non-DM group, indicating that the T2DM specimens were stiffer and stronger. The post-yield properties, including post-yield strain and post-yield toughness, however, did not differ across groups. Because the apparent stiffness, strength, and toughness of bulk cancellous tissue are all strongly influenced by the quantity of bone in the tested specimen (i.e., BV/TV),^(56,73,74) we normalized these apparent properties by each specimen’s BV/TV. Contrary to our hypothesis, the stiffness and strength remained greater in the T2DM specimens after normalization, though to a lesser magnitude (Figure 3). The post-yield properties remained similar across groups after normalization by BV/TV. Thus, the decreased resistance to fracture at the whole-bone level observed clinically in T2DM patients was not mirrored in the monotonic compressive behavior of cancellous explants.

Although modulus and ultimate stress remained higher in the T2DM group vs. the non-DM group after normalization by BV/TV (Figure 3), other significant group-wise differences contributed to the observed differences in mechanical properties. In particular, 1) multiple concomitant group-wise differences in compositional properties (e.g. mineral:matrix, pentosidine) had opposing effects on the measured mechanical properties (Table 3, Figure 2), and 2) significant interactions between the observed compositional changes (Table 3, Figure S2) affected the observed group-wise differences in mechanical properties. The regression modeling enabled us to account for these group-wise differences and interactions while also controlling for BV/TV, and to isolate the effect of each bone quality factor on the mechanical properties. The model results demonstrate that the greater mineral content of the T2DM specimens compared to the non-DM specimens is largely responsible for the greater stiffness and strength of the specimens in the T2DM group observed after normalization by BV/TV. The specimens with higher mineral content (as measured by the FTIR mineral:matrix ratio) had consistently greater moduli, ultimate stresses, and post-yield toughnesses before and after accounting for BV/TV (see Supplemental Figure S3). The importance of bone volume fraction and the mineral content on the bulk cancellous bone behavior is further exemplified by the regression models (Table 3), which show the strong contribution of these two parameters on Young’s modulus and ultimate stress. These results demonstrate that tissue mineral content is consistently important for elastic behavior (which has been previously documented⁽⁵⁹⁾).

In the regression model of post-yield strain, which modeled the tissue’s ductility or ability to deform plastically prior to failure, measures of bone quality (i.e., the mineral maturity factor [MMF] and pentosidine) were significant explanatory variables (Table 3). Similarly, the predictive factors for post-yield toughness, a quantity that characterizes the tissue’s ability to absorb energy prior to failure and depends on the product of strength and ductility, were the combination of the predictive factors observed in strength (BV/TV) and in post-yield strain (MMF, pentosidine). The MMF and AGE outcomes had distinctly opposite effects as BV/TV and the mineral:matrix ratio. For example, greater values of BV/TV increased predicted post-yield toughness, while greater values of AGEs, like pentosidine, decreased predicted post-yield toughness (Table 3). Our regression models of post-yield properties clearly demonstrate the important role of tissue composition after controlling for the greater bone volume fraction of cancellous bone in patients with T2DM. One limitation of our regression models, however, is the moderately large confidence intervals for each parameter that arise from the limited sample size (n = 40–45) and the inherent variability in human data. Nevertheless, the model-predicted values of each mechanical property for both groups fell within the 95% CI of the experimentally measured mechanical property values (Table S2), confirming that the models resulting from the forward stepwise regression method were physically meaningful.

The regression modeling elucidates the relative contributions of microarchitecture and tissue material properties with T2DM that occur concurrent with changes in BV/TV. Here, BV/TV was such a potent modulator of the mechanical properties that the between-group differences in BV/TV—although not statistically significant (+24% T2DM vs. Non-DM, NS, p=0.13)—were sufficiently large that they dominated the experimentally observed mechanical properties and essentially masked the effects of changes in other bone properties. By performing regression modeling, we revealed compositional and microarchitectural changes in T2DM tissue that can be observed after controlling for BV/TV and accounting for patient age, CVD status, CKD status, and vitamin D supplementation. For example, post-yield toughness was not different by group, but was strongly influenced by BV/TV (Table 3). Using stepwise regression models to isolate the effect of BV/TV, we determined that the mineral maturity factor, pentosidine, and total fAGEs affected post-yield toughness, and these effects were more modest and opposite in sign to that of BV/TV (Table 3). Of the mineral maturity factor, pentosidine, and total fAGEs, only pentosidine was significantly different by group (Figure 2A); however, the presence of the mineral maturity factor and total fAGEs in the predicted models indicates that these two predictors capture aspects of post-yield toughness that are not necessarily as heavily influenced by group. Although patient age, CVD status, CKD status, and vitamin D supplementation may affect bone mechanical performance, the magnitude of influence for these parameters was not appreciable in relation to the other predictor variables here.

The different scenarios hypothesized in Figure 4B-D for low, medium, and high risks further exemplify the influence of tissue AGEs and mineral maturity, while also showing the influence of BV/TV. For example, a one-standard-deviation increase from the mean in mineral maturity, pentosidine, or total fAGEs decreases the predicted post-yield toughness compared to the predicted post-yield toughness for an average specimen (Figure 4A). The scenarios proposed in Figure 4B, 4C, and 4D demonstrate how deleterious changes in tissue composition can affect bone toughness, especially in the absence of the compensatory effect of BV/TV. For example, in the low-risk scenario, the high BV/TV increased the predicted post-yield toughness more than the deleterious compositional changes decreased the predicted post-yield toughness (Figure 4B). In the medium-risk scenario, BV/TV partially compensated for the adverse influences of the compositional changes, resulting in a cumulative decrease in predicted post-yield toughness (Figure 4C). Finally, in the high-risk scenario where BV/TV was moderately lower than the grand mean, BV/TV no longer compensated for the compositional changes, resulting in a drastic reduction in predicted post-yield toughness (Figure 4D). While the regression models presented here are only representative of this study cohort, our models demonstrate how skeletal fragility risk associated with changes in microarchitecture or tissue composition may be stratified. Furthermore, while tissue properties such as mineral maturity, pentosidine, or total fAGEs are likely to be systemic, BV/TV varies throughout the skeleton, suggesting that low BV/TV sites may be particularly susceptible to fracture in T2DM patients with high values of MMF, pentosidine, and total fAGEs.

The variable Group modulated trabecular thickness in the model of modulus and total fAGEs in the model of post-yield toughness (Table 3). Neither trabecular thickness nor total fAGEs alone were statistically different between groups (Figure 2); nevertheless, the significant effect of Group in these two models indicates that there are important characteristics relating to trabecular thickness and total fAGEs that were not fully captured by the models.

Tying together the differences in material properties in T2DM versus non-DM specimens and the regression model results, it is evident that T2DM has both beneficial and adverse effects on the mechanical performance of bone. First, T2DM patients had a greater bone volume fraction at the femoral neck, though not statistically significant, which is consistent with clinical assessments of BMD at the femoral neck^(1,75) and hip.⁽⁷⁶⁾ Furthermore, the greater quantity of bone had a large and favorable influence on the monotonic apparent stiffness, strength, and toughness. Second, the greater mineral content in the T2DM specimens compared to the non-DM specimens resulted in greater apparent-level strength; however, greater strength typically comes at the expense of ductility and results in more brittle tissue. While we did not detect a group-wise difference in post-yield strain or post-yield toughness, the tradeoff between strength and toughness is well documented.^(59,77) It is possible that the large effect of BV/TV overshadowed any smaller effects of T2DM on post-yield properties and inhibited our ability to detect such differences in group comparisons.

One key benefit of our regression models is that they allowed us to estimate the potential effect of material properties separate from the compensatory effects of increased BV/TV with T2DM, which cannot be fully accomplished with simple group-wise comparisons. For example, our data show greater concentrations of AGEs (e.g., pentosidine) in cancellous tissue from patients with T2DM than those without T2DM, and after accounting for differences in BV/TV, our models revealed the adverse effect of pentosidine accumulation on the post-yield properties. This indicates that if a patient with T2DM has very large accumulation of tissue AGEs, like pentosidine, but does not have a concurrent increase in BV/TV, the adverse effects of T2DM can outweigh any potential compensating effects of increased BV/TV with T2DM (Figure 4D). Because post-yield toughness is indicative of the amount of energy that can be absorbed before failure, our model demonstrates how poor bone quality without an advantageous increase in bone mass can lead to decreased fracture resistance in patients with T2DM.

Although the results of the mechanical testing (which showed that bulk cancellous tissue from the T2DM group was stronger than that from non-DM controls) appear in conflict with clinical observations of increased fracture risk in patients with T2DM, they highlight the complexity of identifying the key factors that contribute to fracture behavior across multiple levels of structural hierarchy. Although no evidence exists directly demonstrating decreased whole-bone strength in T2DM patients, several meta-analyses found greater fracture risk in T2DM patients vs. non-DM controls even after controlling for numerous potential confounders such as BMI, peripheral neuropathy, and falls.^(1,3) These findings suggest that the bones of T2DM patients may have impaired structural or material properties, although other non-bone factors not captured in the epidemiologic studies may also contribute to increased fracture risk. Indeed, recent reviews address how cellular and molecular processes, micro- and macro-architectural abnormalities, and medications all contribute to bone fragility in T2DM;^(2,78) thus a one-size-fits-all approach to understanding and treating bone fragility in patients with T2DM is likely not possible. In this study, we identified material properties as potentially important modulators of bone strength (mineral:matrix) or fragility (AGEs) in patients with T2DM, and our findings are consistent with previous works that found whole-bone mechanical failure is sensitive to tissue material properties. Specifically, poor tissue material properties reduced whole-bone strength by 39–57% in the absence of changes in BMD or geometry,⁽⁷⁹⁾ and this is particularly salient because a 20% reduction in whole-bone strength is estimated to increase fracture risk by 50–70% or more.⁽⁸⁰⁾ Thus, although management of bone fragility in patients with T2DM is still undergoing refinement, our results are broadly consistent with recent guidelines that suggest intervention with osteoporosis medication following identification of patients at the highest risk of fracture as determined by a history of prior fragility fracture, BMD, or other clinical risk factors.⁽⁷⁸⁾

Our study focused on the cancellous tissue properties at the millimeter-scale, and our results suggest that patients with T2DM have a greater quantity of bone that is more highly mineralized. While these changes will ultimately increase cancellous bone strength at the millimeter-scale, a more highly mineralized bone will be inherently more brittle and less capable of plastic deformation and thus potentially less fracture resistant.^(59,81) In addition, our models of post-yield strain and post-yield toughness demonstrate an adverse effect of AGE accumulation on energy absorption once the large influence of BV/TV is controlled for. Finally, it is important to note that both cancellous and cortical bone contribute to structural performance at the whole-bone level, and cortical bone is more critical for load bearing in the hip.⁽⁸²⁾ In the first study relating bone tissue AGES to mechanical properties from tissue from patients with T2DM, Karim et. al. reported a trend toward greater AGEs in the cortex of the femoral neck (+19%, p=0.09) and greater creep indentation distance (+18%) and indentation distance increase (+20%) in cortical bone from patients with T2DM vs. non-DM controls; however, no relationships between bone AGEs and the indentation variables were observed.⁽²⁹⁾ These results highlight the complexity of assessing the relationships between AGEs and bone fragility across levels of structural hierarchy and suggest an important area for future investigation. Because growing evidence indicates that T2DM compromises cortical tissue integrity,^(11,25,28) it is possible that adverse changes due to T2DM in the cortical compartment may outweigh any beneficial changes in the cancellous compartment, and this may help to explain fragility observed clinically at the hip.

Our study design has several important limitations and advantages. The main shortcoming of this study was the limited availability of participant information related to T2DM. Even though the pre-operative HbA1c and diagnosis of T2DM were recorded at the time of total hip arthroplasty, the participants’ long-term HbA1c history or duration of T2DM were not available. Therefore, our information on the participants’ glycemic control was only on the order of months, which is much shorter than the years required for skeletal changes due to T2DM to fully manifest. Furthermore, the severity and duration of T2DM are known to greatly affect fracture risk,^(28,83) and therefore may also affect the degree of compositional changes due to T2DM. Thus, the single time point for HbA1c level limits our ability to relate clinical measures of glycemic control to the tissue properties that may govern fracture behavior and likely influenced our ability to detect stronger correlations between HbA1c and tissue material properties. Though the long-term severity of T2DM of the men enrolled in this study is unknown, the low prevalence of T2DM-related complications and the single time point HbA1c indicate that the men in the T2DM group had relatively well-controlled T2DM prior to surgery, thus the findings of this study may not be generalizable to patients with a wide range of glycemic control. Moreover, large cohort studies and meta-analyses demonstrate that people with T2DM who take insulin have a greater fracture risk than people with T2DM who do not take insulin.^(1,71) We did not observe a significant effect of T2DM treatment type (insulin, metformin, other, or none) on the compositional, microarchitectural, or mechanical property outcomes. Similar to the absence of long-term glycemic control measures, our ability to account for other T2DM-related complications that may affect bone, such as systemic inflammation,⁽⁸⁴⁾ was limited by the lack of detailed information on these comorbidities collected prior to surgery. We were able to account for the presence of CVD and CKD in our models; however, our study was not designed or powered to assess the effects of CVD or CKD. We also could not assess whether bone remodeling was altered in the T2DM group via serum biomarkers or dynamic bone labeling. Lastly, although we collected and characterized femoral neck tissue at the farthest distance possible from the joint surface, all subjects in our study had osteoarthritis. Differential effects of osteoarthritis on trabecular tissue in the femoral heads of people with and without T2DM were not reported in a similar cohort of patients undergoing total hip arthroplasty for osteoarthritis; however, it is possible that osteoarthritis may have altered the tissue properties in our specimens.⁽⁸⁵⁾ The medical management of patients with T2DM, as well as osteoarthritis, may affect the results and may also help explain some of the apparent discrepancies between our findings and the findings in T2DM animal models. Finally, our study did not include women; therefore, the extent to which results are generalizable to women are unknown. While the surgical setting of our specimen collection imposed several limitations, one major benefit of our study is that the explants characterized here are from the femoral neck, a highly clinically relevant fracture site with a large volume of cancellous bone. Moreover, tissue from the femoral neck allows for tissue characterization at the bulk level and for compartment-specific analysis of cancellous and cortical bone. Finally, our study is among the first to characterize changes in bone quality and mechanical properties in cancellous bone from a clinical population of men with T2DM.

CONCLUSION

In this study, we quantified the accumulation of AGEs, characterized the mineral and matrix composition, assessed the microarchitecture, and mechanically assessed cancellous bone from men with and without T2DM. Our findings showed clear evidence of AGE accumulation and altered tissue material properties in a clinical population with T2DM. Additionally, we used statistical models to predict and compare the mechanical performance of hypothetical low, medium, and high risk scenarios based on the observed changes in tissue composition and microarchitecture with T2DM. Our regression models showed, as expected, that bone volume fraction was by far the greatest determinant of compressive mechanical properties. Furthermore, the T2DM group had a greater tissue mineral content than the non-DM group which increased strength. After accounting for the greater bone volume fraction in the T2DM group, our analyses revealed the deleterious effects of T2DM such as AGE accumulation and altered mineral maturity. Using regression modeling, we illustrated that poor bone quality independent of increased bone volume fraction may drastically reduce the ability of cancellous bone to absorb energy prior to failure in patients with T2DM. Our findings echo epidemiological evidence that the quantity of bone alone does not explain fracture risk, and our results provide a foundation for future investigations of changes in bone quality with T2DM. Although our methods are not available in a typical diabetes clinic, our findings are clinically relevant because they demonstrate that different populations of T2DM patients may have distinctly different bone fragilities as a result of varying tissue composition and microarchitecture.

Supplementary Material

Supplemental Figure S1:

Mechanical properties from monotonic compression testing of cancellous specimens by group. Abbreviations: non-DM: non-diabetic group; T2DM: type 2 diabetic group.

Supplemental Figure S2:

Methods overview.

Supplemental Figure S3:

(A) Young’s modulus normalized by BV/TV, (B) Ultimate stress normalized by BV/TV, and (C) Post-yield toughness normalized by BV/TV. Horizontal lines indicate group means. Mineral:matrix data (Figure 2C) were used to divide the 65 FTIR samples (both DM and non-DM groups) into tertiles of lower (white squares), middle (light grey squares), and upper (dark grey squares) (see Supplemental Section S3). Abbreviations: non-DM: non-diabetic group, T2DM: type 2 diabetic group; BV/TV: bone volume fraction.