Examining the Relationships between Bone Tissue Composition, Compositional Heterogeneity and Fragility Fracture: A Matched Case Controlled FTIRI Study

Abstract

Fourier transform infrared imaging (FTIRI) provides information on spatial distribution of the chemical composition of thin tissue specimens at ~7um spatial resolution. This study of 120 age-and BMD-matched patients was designed to investigate the association of FTIRI variables, measured in iliac crest biopsies, with fragility fractures at any site. An earlier study of 54 different women found hip BMD to be a significant explanatory variable of fracture risk for cortical bone, but not for cancellous bone. In the current study, where age and BMD were controlled through matching, no such association is seen, validating the pairing scheme. Our first study of un-matched iliac crest biopsies found increases in collagen maturity (cancellous and cortical bone) and mineral crystal size (cortical bone only) to be a significant explanatory variable of fracture when combined with other covariates. The ratio for collagen maturity has been correlated to the amount of enzymatic collagen cross-links. To assess the impact of other FTIRI variables (acid phosphate substitution, carbonate-to-phosphate ratio, and the pixel-distribution (heterogeneity) of all relevant FTIRI variables), we examined biopsies from a matched case-controlled study, in which 60 women with fractures, were each paired with an age-and BMD-matched female control. With the matched data set of 120 women, conditional logistic regression analysis revealed that significant explanatory variables of fracture were decreased carbonate-to-phosphate ratio in both cancellous (Odds Ratio =0.580, 95% confidence interval (CI)=[0.37, 0.909], p= 0.0176) and cortical bone (Odds Ratio= 0.519, 95% CI=[0.325, 0.829], p= 0.0061), and increased heterogeneity (broadened pixel distribution) of collagen maturity for cancellous bone (Odds Ratio = 1.549, 95% CI= [1.002, 2.396], p=0.0491). The observation that collagen maturity was no longer linked to fracture in age-and BMD-matched samples suggests that age-dependent variation in collagen maturity may be a more important contributory factor to fragility fractures than previously thought.

Introduction¹

The strength of a bone is determined by both its quantity and its quality. Quality, here, refers to properties beyond bone mineral density (BMD) that contribute to bone strength ^(1,2), described in terms of chemical composition, architecture, geometry and morphology. Fourier transform infrared imaging (FTIRI) provides a “chemical photograph” of the composition of thin sections of bone, at a spatial resolution of ~6 microns. This data portrays the material properties of biopsied bones in terms of relative amounts of mineral, collagen and composition of these components. FTIRI was previously used to provide information on the composition of cancellous and cortical bones in states of health and disease ^(3–13). The current study was designed to investigate the association of FTIRI variables, measured in iliac crest biopsies, with fragility fractures in any other skeletal site.

FTIRI previously identified crystallinity and collagen maturity, as two bone quality variables associated with fracture, in biopsies from 54 otherwise healthy women, with and without fractures, when combined with other covariates ⁽¹²⁾. In that study, hip BMD was significantly associated with fracture when cortical but not cancellous bone FTIRI data was considered. These two predictors were independent of three other explanatory variables: age, history of estrogen treatment, and mineral-to-matrix ratio ⁽¹²⁾. Carbonate-to-phosphate ratio was also associated with fragility fracture ⁽¹⁴⁾. This last variable, however, was not included in the first analysis of FTIR variables and fragility fracture ⁽¹²⁾ due to instrument limitations. The extent of acid phosphate substitution, inversely related to mineral maturity ⁽¹⁵⁾, is indicative of new bone formation, but has yet to be reported in human fracture cases. Additionally, some studies have suggested that the mean values of FTIR variables and their spatial distribution (heterogeneity) change with disease status. Here, we define heterogeneity as the line-width at half-maximum of the FTIRI pixel distributions. In small populations, alterations in heterogeneity were noted in osteopenic patients ⁽⁶⁾ and in patients treated with bisphosphonates, ^(5,13) compared to small, naive populations of age-and sex-matched controls. Specifically, in femoral neck sections from patients with hip fractures, heterogeneity, as described above, of mineral-to-matrix ratio and carbonate-to-phosphate ratio were decreased relative to patients without fractures, while the heterogeneity of crystallinity was increased in hip fracture cases⁽⁶⁾. In iliac crest biopsies of peri-menopausal women treated for 3 years with alendronate, heterogeneity of mineral-to-matrix ratio, crystallinity, and collagen maturity were decreased relative to that in women who received a placebo ⁽¹³⁾ Postmenopausal women taking bisphosphonates, who had low-energy intertrochanteric and subtrochanteric femoral fractures, similarly had reduced heterogeneity of crystallinity and collagen maturity relative to bisphosphonate-naive patients with similar types of-fractures ⁽⁵⁾.

The purpose of the current study was to determine which bone quality variables assessed by FTIRI were associated with fragility fracture, with the long-term goal of designing studies to determine cause and effect of these associations. Fragility fracture here was defined as a non-traumatic fracture due to a fall from a standing height or less, excluding digits, face or skull. To include all of the pertinent variables in a multivariate analysis, we examined biopsies of 60 women with fractures and 60 age-and BMD-matched female patients without fractures. This enabled us to analyze 10 bone quality variables while pairing for age and BMD. These variables (mineral-to-matrix ratio, carbonate-to-phosphate ratio, collagen maturity, crystallinity, and acid phosphate substitution, and the heterogeneity of each of the above variables) included 7 new parameters not previously related to fracture risk: carbonate-to-phosphate ratio ⁽⁶⁾, acid phosphate substitution ⁽¹⁵⁾ and the heterogeneity of each of the 5 parameters. Heterogeneity, for these purposes, was defined as above. Our hypothesis was that loss of heterogeneity in each of the 5 cortical and 5 cancellous FTIRI variables, in addition to the previously reported changes in the mean values of some FTIRI variables, is associated with fragility fractures. This paper now demonstrates that decreased carbonate-to-phosphate ratio in cancellous and cortical bone is seen in iliac crest biopsies from women with fragility fractures compared to age-and BMD-matched controls. Opposite to our hypothesis, greater heterogeneity of the collagen maturity was found in fracture cases compared to controls; no other significant differences in heterogeneity were noted. While this information, which requires analyses of bone biopsies, will be difficult to include in clinical screening tools such as FRAX(16), it does provide insight into potential mechanisms of fragility fractures and will be of use for evaluation of pharmaceutical agents where biopsies are available.

Materials and Methods

Biopsies

Iliac crest biopsies used for this study were provided by Dr. Recker from his Bone Quality Study, with consent of the “Bone Quality Analysis Study Team”. Biopsies were obtained from 120 women, all had consented to participate. The study was performed under IRB 07–14738 from Creighton University with additional approval of the IRB of the Hospital for Special Surgery (#93019). Biopsies from women with fractures were paired with non-fracture cases by matching the non-fracture subjects to within 5 years of the age, and within 10% of the hip BMD of each fracture case. The analysis had been powered, based on other variables in Dr Recker’s Bone Quality Study; specifically on finding a 25% difference in histomorphometric activation frequency, trabecular bone connectivity density, and nano-indentation hardness, with an α=0.05, β= 0.80; the analyses in the current manuscript were considered secondary. All biopsies were obtained at least 6 months after the fracture had occurred, and within 5 years of the index fracture. Index fractures occurred as follows: wrist (n=21), ankle (n=15), humerus (n=7), patella (n=4), shoulder (n=3) and other locations with 1 or 2 fractures (n=10). None of the women reported being treated with any of the anti-resorptive agents (bisphosphonate, calcitonin, estrogen, etc.) or with PTH. BMD was measured using a Hologic Delphi densitometer at the time of enrollment for fracture cases. Control subjects were measured at variable times after their respective fracturing case was identified. Each control was identified and enrolled within a few weeks or months following identification based on archived BMD. The archived BMD, however, was not used for the matching of controls since that BMD might have been performed as long as several years prior to enrollment. Consequently, new BMD measurements were acquired for controls as well as fracture cases. The enrollment and exclusion criteria are summarized in Supplemental Table 1. The number of prior fractures a person suffered was self-reported, but confirmed through medical records review.

FTIRI Data Acquisition and Processing

All FTIRI data were obtained from biopsy sections cut to 1–2um thickness and scanned at 6.25 um × 6.25 um/pixel resolution, unless otherwise noted, on an infrared Imaging system (Perkin-Elmer Spotlight 300, Perkin Elmer, Waltham, MA). Two sections cut as alternate serial sections from each biopsy, were analyzed with five separate areas of intact cortical and five of intact trabecular bone (sometimes consisting of several trabeculae, hence referred to as cancellous bone), each at least 300×500 um² in area, acquired from each section. The cortical bone regions were always 500 um in the longitudinal dimension but the 300um dimension was stretched to accommodate the width of the entire cortex, as needed. The cancellous bone regions corresponded to intact trabeculae. The case-control status was not known until the entire analyses were completed. The average bone area (pixels) scanned was comparable between members of a pair, and ranged for the entire population of cortical and cancellous bone from 4250 to 11350 pixels, mean ± SD 7179.7 ±2501.2, no-fracture; 6694.17±1355.99, fracture, p=0.59.

After base-line correction and subtraction of the PMMA contribution (ISYS 5.0 Image Analysis Software, Malvern Instruments, Columbia, MD), we pooled spectra from either cortical or cancellous regions of bone and analyzed second-derivative spectra to confirm previously selected band peak positions. Since second-derivative band intensity is inversely proportional to the square of the original band half-width, this produces an enhancement of sharp lines in complex spectra (17). These band positions were subsequently used to calculate the following variables: mineral-to-matrix peak area ratio (900cm⁻¹–1200cm⁻¹/1550cm⁻¹–1800 cm⁻¹), representing the relative mineralization of the collagen matrix ⁽¹⁸⁾; carbonate-to-phosphate peak area ratio (850cm⁻¹–890 cm⁻¹/900cm⁻¹–1200 cm⁻¹), representing the extent of carbonate substitution into phosphate and hydroxyl positions in the hydroxyapatite lattice⁽⁶⁾; crystallinity (1030cm⁻¹/1020 cm⁻¹) intensity ratio, reflecting hydroxyapatite crystal size and perfection, as determined by X-ray diffraction line broadening⁽¹⁸⁾; acid phosphate substitution (1128 cm ⁻¹/1096cm⁻¹ intensity ratio), representing the substitution of HPO4 ions into the hydroxyapatite lattice, characteristic of younger mineral ⁽¹⁵⁾; and collagen maturity (1660cm⁻¹/1690cm⁻¹ intensity ratio ), the ratio of non-reducible (mature) to reducible (immature) enzymatic cross-links⁽¹⁰⁾. The variables were averaged, and expressed as mean ± SD for the cortical and cancellous regions of each biopsy. The same software was used to calculate the pixel distribution for each image, fit the distribution with a Gaussian function and provide the line width at half maximum, recorded here as the heterogeneity.

Statistics

All comparisons of FTIRI data were based on the average of multiple images at 6.25 um resolution. This average represents the values that are in both the organic and inorganic components of the tissue sections, and excludes the contributions of the PMMA and barium fluoride window which have been removed by subtraction and assigned a zero value; thus different areas can be compared as they represent in tissue ratios only. In a univariate analysis, mean values for all of the fracture and non-fractured cases for each variable were tabulated (age, number of fractures, hip BMD, and FTIR parameters and their heterogeneities) and compared using a Wilcoxon signed-rank test. In multivariate analysis on paired data, conditional logistic regression models were constructed and odds ratios with 95% confidence intervals were estimated to determine independent predictors for fracture. The outcome variables were fracture (1) or no fracture (0). For cortical bone, the predictor variables were the five FTIR cortical variables and each of their heterogeneities. For cancellous bone, the predictor variables were the five cancellous FTIRI variables and their heterogeneities. A backwards selection procedure was applied to the conditional logistic regression to build the most parsimonious model. Significance level was set at 0.05. Both univariate and multivariate analyses were performed using SAS version 9.3 (SAS Institute, Cary, NC).

An investigational analysis of the relation between number of fractures and each of the FTIRI variables in the fracture cases, was conducted using the nonparametric Spearman’s correlation with GraphPad Prism, version 3.03 (Carlsbad, CA). Similar analyses were done on age vs. mean FTIRI data for all cases and on carbonate-to-phosphate ratio vs. the following histomorphometric bone formation markers (unpublished data from Dr Recker’s study): osteoid thickness, mineral apposition rate, osteoblast surface, mineralizing surface and bone formation rate, in all cases. To compare the FTIRI data from an entire section to data from individual sub-regions, a paired t-test was used (GraphPad version 3.0).

Results

The data in this study came from post-menopausal women, age 49–79, with fragility fractures, matched to women without fractures based on age and BMD. Their demographics are summarized in Table 1. The matched design was successful in creating groups with similar mean age and BMD (Table 1). The reproducibility of our FTIRI analysis design was demonstrated by comparing low resolution (25um × 25um pixel resolution) scans of the entire biopsy with five segments of the entire cortex taken at 6.25um×6.25um pixel resolution (supplemental Figure 1). The total number of pixels for each of these analyses (whole section or sub-regions) was ~5600.

Table 1.

Study Demographics

Parameter a	Controls	Fracture	Difference (No fracture – Fracture)	p-value b
Number of subjects (n)	60	60	0
Age at Biopsy (yrs.) [95% Confidence Limits]	61.82±7.34 [59.92 to 63.72]	62.19±7.44 [62.260.27–64.11]	−0.3726 [−3.0446 to 2.2993]	0.512
Hip BMD (gm/cm2) [95% Confidence Limits]	0.8266±0.0823 [0.8051 to 0.848]	0.8263±0.087 [0.8035 to 0.8492]	0.00026 [−0.0308 to 0.0313]	0.414
Average number of fractures	0.0820±0.276c	2.5±1.4	2.42	<0.0001

^aAll values unless otherwise noted are mean ± SD.

^bComparisons based on Wilcoxon signed-rank test.

^cFive women reported traumatic fractures and were included in the No Fracture (control) group.

Typical FTIR images for one fracture case along with her paired control (identified after the code was broken) ages 56, 58, BMD = 0.89 g/cm², Figure 1) show the magnitude and spatial distribution for each variable and their respective pixel heterogeneities. Each image represents the calculated value of each FTIR variable at each pixel thereby illustrating the spatial distribution of the indicated variable. The pixel histogram adjacent to each image indicates the mean and Gaussian fit to the distribution of the variable in the corresponding image. These histograms were used for calculating the heterogeneity of each image. All images for a given parameter (e.g., mineral-to-matrix ratio), are presented on a common scale for the purpose of comparison. As expected for both biopsies, fracture (Figure 1B, D) and paired non-fracture (Figure 1A, C), the mineral-to-matrix ratio is highest in the older bone (center of the cortex and center of the trabeculae). The histogram of the carbonate-to-phosphate ratio appears sharper in the bone from the fracture case (Figure 1B, D). The image of collagen maturity appeared to have larger distribution of areas with high values in the fracture case (Figure 1F). Crystallinity appears increased in the fracture case (Figure 1F) while acid phosphate substitution had a less even distribution in the fracture case (data not shown). The distributions in the pixel histograms were not perfectly Gaussian.

Figure 1.

Typical FTIRI images from two paired biopsies from women aged 56 and 58, respectively; BMD =0.89 for both). Images from the no-fracture case (rows A, C and E) and from the fracture case (rows B, D and F) are shown with their respective pixel histograms to the right of each the image. The mean and SD are superimposed on the pixel histogram. Rows A, B, E and F show cancellous bone, rows C and D cortical bone. The color scale for all images for an indicated variable (mineral-to-matrix (Min/Mat), carbonate-to-phosphate (CO3/PO4), collagen maturity (XLR) and crystallinity (XST) are shown below the images for that group. All pixel histograms for the indicated variable are on the same scale, noted below the lowermost histogram. The bar represents 100 um.

Univariate analyses showed no significant differences between the mean chemical composition in images of biopsies from women who fractured and their non-fracture controls (Figure 2A, Table 2) with the exception of cortical carbonate-to-phosphate ratio which was decreased by 3.9%, p=0.0015 in the fracture cases. Mean heterogeneities were also not different in cortical or cancellous regions of the biopsy when fracture and non-fracture cases were compared (Figure 2B, Table 3).

Figure 2.

Box-whisker plots showing distribution in cortical and cancellous bone of individual average FTIRI values for each fractured and. non-fractured case. A) FTIRI variables B) heterogeneity for each variable. The data points represent the mean value for each case for cortical and cancellous bone. The box shows the 25th to 75th percentile of the data. The “whiskers” or error bars extend from the edge of the box to the 5th and 95th percentile. The mean and standard deviation are shown by the error bar to the right of each box. Note the Y-axis scales have been adjusted to allow visualization of individual data points.

Table 2.

Univariate Analysis: FTIRI Variables ^a in Iliac crest biopsies from Fracture and Non-Fracture Cases

Variable	Non-Fracture	Fracture	Difference (Non-fracture – fracture)	p-value b
Cortical bone
Mineral-to-matrix ratio [95% Confidence Limits]	4.285 ±0.259 [4.218 to 4.352]	4.275±0.316 [4.193 to 4.356]	0.0105 [−0.0941 to0.1151]	0.785
Carbonate-to-phosphate ratio [95% Confidence Limits]	0.00846±0.00044 [0.00835to 0.00858]	0.00813±.00063 [0.00796–0.00829]	0.000336 [0.000138 to 0.000533]	0.0015
Crystallinity [95% Confidence Limits]	1.214±0.028 [1.207to 1.222]	1.216±0.027 [1.209to 1.223]	−0.00111 [−0.0111 to 0.00884]	0.741
Acid phosphate substitution [95% Confidence Limits]	0.352±0.0264 [0.345 to 0.358]	0.350±0.0241 [0.344 to 0.357]	0.00167 [−0.00749 to 0.0108]	0. 598
Collagen maturity [95% Confidence Limits]	3.760±0.279 [3.688 to 3.832]	3.813±0.271 [3.743 to 3.883]	−0.0532 [−0.1525 to 0.0461]	0.200
Cancellous bone
Mineral-to-matrix ratio [95% Confidence Limits]	4.337 ±0.494 [4.269 to 4.406]	4.347±0.294 [4.2712 to 4.4234]	−0.010 [−0.1111to0.0915]	0.884
Carbonate-to-phosphate ratio [95% Confidence Limits]	0.00828±.0.000490 [0.00815 to 0.00840]	0.00808±0.00054 [0.00794 to −0.00823]	0.00019400 [0.0000064 to 10.00038100]	0.147
Crystallinity [95% Confidence Limits]	1.194±0.0294 [1.186 to 1.203]	1.197±0.0273 [1.190 to 1.204]	−0.003 [−0.0137 to 0.0068]	0.561
Acid phosphate substitution [95% Confidence Limits]	0.3926±0.0363 [0.3833 to 0.402]	0.386±0.037 [0.376 to 0.395]	0.007 [−0.00616 to 0.0202]	0.360
Collagen maturity [95% Confidence Limits]	3.683±0.275 [3.612 to 3.7542]	3.749±0.281 [3.676 to 3.821]	−0.066 [−0.1662 to0.035]	0.089

^aAll values are dimensionless ratios. Pooled differences are shown.

^bP-value based on Wilcoxon signed-rank test. Significant differences are highlighted in bold.

Table 3.

Univariate Analysis: Heterogeneity ^a of FTIRI Variables

Variable	Non-Fracture	Fracture	Difference b	p-value c
Cortical Bone
Mineral-to-matrix ratio [95% Confidence Limits]	1.537±0.251 [1.473 to 1.602]	1.551±0.265 [1.483 to 1.619]	−0.0136 [−0.1068 to 0.0797]	0.599
Carbonate-to-phosphate ratio [95% Confidence Limits]	0.00261±0.00104 [0.00234 to 0.00288]	0.00250±0.00052 [0.00237 to 0.00264]	0.000107 [−0.00019 to 0.000403]	0.686
Crystallinity [95% Confidence Limits]	0.139±0.0467 [0.127 to 0.151]	0.145±0.079 [0.124 to 0.165]	−0.00586 [−0.0293 to 0.0176]	0.861
Acid phosphate substitution [95% Confidence Limits]	0.103±0.094 [0.079 to 0.127]	0.093±0.026 [0.086 to 0.099]	0.010 [−0.0149 to 0.0349]	0.911
Collagen maturity 95% Confidence Limits	0.455±0.078 [ 0.453 to 0.503]	0.472±0.117 [0.442 to 0.502]	0.0058 [−0.033 to 0.0446]	0. 809
Cancellous Bone
Mineral-to-matrix ratio [95% Confidence Limits]	1.584±0.285 [1.510 to1.658]	1.646±0.332 [1.561 to 1.732]	−0.062 [−0.1743–0.0497]	0.414
Carbonate-to-phosphate ratio [95% Confidence Limits]	0.00252±0.00034 [0.00243 to 0.00261]	0.00300±0.0027 [0.0037 to 0.00268]	−0.000480 [−0.00117to0.000208]	0.1142
Crystallinity [95% Confidence Limits]	0.1468±0.0461 [0.1349 to 0.159]	0.149±0.064 [0.132 to 0.165]	−0.0020 [−0.0221 to 0.0181]	0.439
Acid phosphate substitution [95% Confidence Limits]	0.105±00.0237 [0.0994–0.112]	0.108±0.026 [0.104 to0.115]	−0.003 [−0.0114 to 0.00642]	0.693
Collagen maturity [95% Confidence Limits]	0.455±0.0778 [0.435 to 0.475]	0.30.479±0.113 [0.450 to 0.508]	−0.024 [−0.0591 to 0.0108]	0.373

^aHeterogeneity calculated from the full-width at half maximum of pixel histograms for each of the indicated variables and are dimensionless ratios.

^bDifference between pooled non-fracture and fractured cases.

^cp-value based on Wilcoxon signed-rank test.

In the multivariate analysis for cortical (Table 4) and cancellous (Table 5) bone, conditional logistic regression showed cortical and cancellous carbonate-to-phosphate ratio was significantly associated with fragility fractures. In detail, after selection, the model indicated reduced carbonate-to-phosphate ratio in cortical bone (Odds Ratio= 0.519, 95% CI=[0.325, 0.829], p= 0.0061) and in cancellous bone (Odds Ratio =0.580, 95% CI=[0.37, 0.909], p= 0.0176) associated with fractures, i.e. for cortical bone, one-unit increase in carbonate-to-phosphate ratio is associated with 48.1% decrease in odds of fracture; for cancellous bone, one-unit increase in carbonate-to-phosphate ratio is associated with 42% decrease in odds of fracture. In the cancellous multivariate analyses (Table 5), increased heterogeneity of collagen maturity (Odds Ratio = 1.549, 95% CI= [1.002, 2.396], p=0.0491) was also associated with fracture after selection, i.e., i.e. one-unit increase in heterogeneity of collagen maturity was associated with 54.9% increase in odds of fracture.

Table 4.

Conditional Logistic Regression Analysis: Odds Ratio (OR) for Cortical Bone^a

Full Model
Parameter	OR	95% Confidence Limits	p-value
Mineral-to-matrix ratio	1.101	0.625 to 1.939	0.7385
Mineral-to-matrix ratio heterogeneity	1.197	0.789 to 1.815	0.3973
CO3-to-PO4 ratio	0.433	0.231 to 0.811	0.0089
CO3-to-PO4 ratio heterogeneity	0.861	0.535 to 1.386	0.5377
HPO4	1.042	0.637 to 1.706	0.8693
HPO4 heterogeneity	1.044	0.658 to 1.654	0.8557
Collagen maturity	1.42	0.759 to 2.657	0.2728
Collagen maturity heterogeneity	0.799	0.440 to 1.453	0.4624
Crystallinity	0.901	0.506 to 1.604	0.722
Crystallinity heterogeneity	1.346	0.761 to 2.38	0.3077
Model After Selection
Parameter	OR	95% Confidence Limits	p-value
CO3-to-PO4 ratio	0.519	0.325 to 0.829	0.0061

^aPaired analysis with cases and controls matched by BMD and age. CO3-to-PO4= carbonate to phosphate ratio; HPO4 = acid phosphate substitution; Odds Ratio (OR) represents the change in the odds of fracture for one-unit increase in a particular predictor; 95% confidence interval of OR is shown.

Table 5.

Conditional Logistic Regression Analysis: Odds Ratio (OR) for Cancellous Bone ^a

Full model
Parameter	OR	95% Confidence Limits	p-value
Mineral-to-matrix ratio	1.212	0.721 TO 2.039	0.4686
Mineral-to-matrix ratio heterogeneity	1.131	0.654 TO1.955	0.6596
CO3-to-PO4 ratio	0.647	0.395 TO 1.062	0.0851
CO3-to-PO4 ratio heterogeneity	6.559	0.417 TO 103.133	0.1809
Collagen maturity	0.966	0.582 TO 1.602	0.8923
Collagen maturity heterogeneity	1.792	0.996 TO 3.223	0.0515
Crystallinity	1.557	0.847 TO 2.864	0.1543
Crystallinity heterogeneity	1.075	0.706 TO 1.638	0.7365
HPO4	1.029	0.577 TO 1.835	0.9236
HPO4 heterogeneity	0.774	0.456 TO 1.314	0.3428
Model after selection
Parameter	OR	95% Confidence Limits	p-value
CO3-to-PO4 ratio	0.580	0.37 TO 0.909	0.0176
Collagen maturity heterogeneity	1.549	1.002 TO 2.396	0.0491

^aPaired analysis with cases and controls matched by BMD and age. HPO4 = acid phosphate substitution; CO3-to-PO4 – carbonate/phosphate ratio; ; Odds Ratio (OR) represents the change in the odds of fracture for one-unit increase in a particular predictor. 95% confidence interval of OR is shown

The probability of fragility fracture increases after the first fracture ⁽¹⁹⁾ and fracture risk increases with age ⁽²⁰⁾, therefore, we did two investigative analyses using Spearman’s correlation of: (i) each FTIRI variable for only the fracture cases against the number of fractures reported (Supplemental Table 2) and (ii) for all of the cases, independent of fracture status, each variable vs. age (Supplemental Table 3). In terms of numbers of fractures, cortical carbonate-to-phosphate ratio decreased with increasing number of reported fractures (Spearman’s r = −0.2275, p = 0.0125); cancellous collagen cross-link ratio tended to increase with increasing number of fractures (Spearman’s r = 0.1379; p = 0.1331) and cancellous carbonate-to-phosphate heterogeneity tended to increase with increasing number of fractures (Spearman’s r = 0.1461; p = 0.1114). Cancellous mineral-to-matrix heterogeneity and of carbonate-to-phosphate ratio heterogeneity tended to increase with the number of fractures (Spearman’s r = 0.1538; p = 0.0934, Spearman’s r = 0.1461, p = 0.1114, respectively). There were no other changes that correlated with fracture number. While there were few significant age-dependent differences (Supplemental Table 3), there are some that should be noted: cortical mineral-to-matrix ratio increased with age (Spearman r=0.2228; p= 0.0140), cancellous acid phosphate substitution decreased with age (Spearman’s r = −0.1897; p = 0.0380); and cancellous mineral-to-matrix heterogeneity increased with age (Spearman’s r = 0.2302; p=0.0114). There was no correlation of carbonate-to-phosphate ratio with any of the histomorphometric bone turnover markers examined.

Discussion

Factors Associated with Fragility Fractures

This study was designed to investigate the association of FTIRI variables, measured in iliac crest biopsies, with fragility fractures at other sites. In our earlier study ⁽¹²⁾, using multivariate analysis, we found hip BMD to be a significant explanatory variable of fragility fracture when cortical FTIRI data was considered, but not cancellous data. In our current study, where age and BMD were controlled through matching, no such association is seen, validating the pairing scheme used.

Our first un-matched iliac crest biopsy study ⁽¹²⁾, associated increases in collagen maturity (cancellous and cortical bone) and mineral crystal size (cortical bone only), with fracture cases when combined with other covariates. A small study (n=10) of hip fracture cases, where femoral neck biopsies were obtained within 48 hours of fracture, matched by age to necropsy controls and examined by quadrants, showed decreased cancellous and cortical mineral-to-matrix ratio, increased carbonate-to-phosphate ratio, increased collagen maturity and increased crystallinity in the fracture cases ^{(6 ).} In contrast to the hip fracture study ⁽⁶⁾, our current, age-and BMD-matched study indicates an association between decreased carbonate-to-phosphate ratio and fracture risk in both cortical and cancellous bone. However, compared to previous findings, using biopsies paired by age and BMD and a statistical model, we no longer detect a significant association of fracture risk with an elevated collagen cross-link ratio (collagen maturity), elevated crystallinity or decreased mineral-to-matrix ratio. One possible explanation for these differences in our findings is that the paired iliac crest biopsies were not obtained immediately following fracture and changes in iliac crest bone quality occurred during that interval. Additionally, the observed compositional changes in the current study, while significant, were small (<4% in univariate analysis of cortical bone, and no higher than 22 % in individual pairs of cancellous bone). This suggests that age-dependent variation in composition may be a more important factor than previously thought. It also is probable that due to the greater sample size of the current, compared to previous reports, there was an increased likelihood of including patients in whom other unidentified factors contributed to fragility fracture risk. Factors, other than BMD and bone composition, contributing to fragility, include: defective cortical thickness and/or bone shape ⁽²¹⁾, increased cortical porosity⁽²²⁾, defective trabecular microanatomy⁽²³⁾, muscle weakness ^(24), impaired balance or vision ^{( 25)}, or other as yet, unidentified factors. The previous study ⁽¹²⁾ also included premenopausal women; that too, could have influenced earlier findings.

This study has additional limitations. First; biopsies were primarily taken from white, middle-aged Caucasian women from the Midwest; thus, the study may not be representative of all women. None the less, the large sample size of the groups reported here improves the generalizability of our results in post-menopausal women with and without fractures relative to those of preceding studies. Next: men were not included, yet they too suffer osteoporosis and are subject to fragility fractures ⁽²⁶⁾. Thirdly: biopsies were obtained within 5 years of fracture, but not less than 6 months after the fracture; thus we are assuming minimal changes in this tissue over this time period. In earlier studies ^(6,12), some biopsies were obtained within a few days of fracture, and this may have affected the findings. Looking ahead, due to reports of conflicting variations in heterogeneities in fracture cases, it will be necessary to relate changes in heterogeneity to biological mechanisms including turnover and determine whether the changes are associated with proliferation of micro-cracks. As bone is a hierarchical tissue, we must also determine whether changes in tissue composition are seen at all levels – macro, micro, and nano. Lastly; we used iliac crest biopsies rather than those obtained at or near the fracture site. Bone quality parameters obtained at the iliac crest does not necessarily reflect bone quality in the spine, rib, radius, wrist, etc. Bone remodeling (turnover) and tissue age and quality are site-specific ^(27,28), as is micro-damage accumulation ⁽²⁹⁾. We feel justified in the use of such biopsies. First, because they provide a means for obtaining tissue safely from patients with bone disease that can be used for diagnosis ^(30,31). Second, analyses of iliac crest biopsies are routinely used to assess bone quality, based on histomorphometry⁽³²⁾, computed-tomography^(33,34) and spectroscopic imaging⁽³⁵⁾, assuming these biopsies reflect general changes in bone tissue quality. In addition, micro-crack sizes and distributions have been assessed ⁽³⁶⁾ from such biopsies. Analyses of iliac crest biopsies in some studies have even correlated findings with mechanical testing of the same or comparable materials ^(e.g.,37,38). Furthermore, our earlier cadaver studies⁽⁵⁾ demonstrated that the mean values of a variety of FTIRI parameters, including those measured in the current study, agreed from site-to-site when normalized to iliac crest biopsies, although there was variation in heterogeneity.

Collagen maturity

Collagen cross-link ratio, the ratio of mature to immature collagen cross-links, also referred to as collagen maturity, was the most consistent explanatory variable of fracture in our previous study⁽¹²⁾. In the present study it was not a significant factor, suggesting that this ratio is more closely associated with increasing age than with fragility fractures, PER SE. Age is an independent risk factor for fragility fractures ^(39,40,41), with risk increasing exponentially with age after age 70 ^(39,41). Other studies have also reported increases in collagen maturity with fragility fracture ^{(32,42,43,44,45)}, however none of these studies were cross-matched for patient age or BMD. Enzymatic collagen cross-links, those detected by FTIRI⁽¹⁰⁾, are known to increase with age^{(46,47,48,49,50)}, at least until 50 in the vitreous of the eye⁽⁴⁸⁾, 25 years of age in human bone⁽⁴⁹⁾, and until about 60% from the edge of the osteon (maturing tissue age⁽⁵⁰⁾ ) in baboons. This explains, in part, how collagen maturity can be elevated in fragility fracture cases but no longer an explanatory variable, once corrected for age. An additional explanation may be the length of time between the index fracture and biopsy, particularly where there was no correlation between age and collagen maturity in the fracture data from the current study.

Crystallinity

The crystal size and perfection, or crystallinity, as measured by FTIRI, is linearly related to crystal size and perfection in the c-axis-(002) plane of hydroxyapatite determined by X-ray diffraction analysis ⁽⁵¹⁾. Crystallinity was reported by our group and by others to be elevated in cortical and cancellous bone of fragility fracture cases ^{(12,52,53,54,55)}. Increased crystal size is also related to increased bone brittleness in young mice ⁽⁵⁶⁾. Larger and more perfect crystals are found in older bone, left after remodeling ^{(12,50, 53,57,58,59,60)}. The finding that there was no increase in crystallinity with fracture may be related to the paired nature of the study and the time at which the biopsies were obtained. Raman analysis of the femoral cortex in humans and rodents showed a linear increase in crystallinity with increasing animal age ^{(59,60,61,62)}. Crystallinity is inversely related to carbonate-to-phosphate ratio ⁽⁶⁾, i.e., lower carbonate-to-phosphate ratio should be associated with larger crystals, and carbonate-to-phosphate ratio was reduced in the paired fracture cases; the absence of an increase in crystallinity was thus unexpected. This may reflect the sensitivity of the crystallinity measurements, especially since carbonate-to-phosphate ratio and crystallinity are inversely correlated when cancellous and cortical values from the current study were compared (Spearman’s r= −0.2257, p = 0.0098).

Carbonate-to-Phosphate Ratio

Fracture cases, in both cortical (based on univariate and multivariate analyses) and cancellous (based on multivariate analysis) regions of bone, had slightly albeit significantly lower carbonate-to-phosphate ratios than did paired non-fracture cases. Carbonate-to-phosphate ratio also decreased with increasing fracture number. Changes in bone carbonate-to-phosphate ratio are difficult to interpret, because carbonate substitution rises as the tissue develops, and falls as the tissue remodels. In bone, carbonate ions can substitute for both phosphate and hydroxide ions when the hydroxyapatite crystal lattice is formed ⁽⁶³⁾. The extent of such substitutions, increases during early development ^(64,65). This phenomenon is likely due to carbonate incorporating at a rate faster than that of new mineral deposition. Thus, regardless of developmental age, the highest values of carbonate-to-phosphate ratio are found in the oldest bone (at the center of the trabeculae and cortex). In contrast, when hydroxyapatite mineral crystals perfect, or dissolve and reform during remodeling, carbonate substitution decreases; as a result, smaller changes occur in the mean carbonate-to-phosphate ratio values as a function of chronologic aging following maturity ⁽⁶⁶⁾. In the osteons of humans ⁽⁶⁷⁾ and male baboons ⁽⁵⁰⁾, the average carbonate-to-phosphate ratio increases with chronologic age. This agrees with findings using Raman spectroscopy to study cortical bone from rats ⁽⁶⁰⁾ and humans ^(59,61). Within an individual osteon; the oldest bone (the lamellae furthest away from the Haversian canal) may have a carbonate-to-phosphate ratio that is lower than the closest adjacent lamellae, due to remodeling ⁽⁵⁰⁾. Carbonate-to-phosphate ratio-is low, relative to controls, in micro damaged areas of bone ⁽⁶⁸⁾ and in biopsies from some women treated with bisphosphonates ^{(13, 69)}, not all ⁽⁵⁾. Low carbonate-to-phosphate ratios were attributed to a relative increase in the amount of mineral (phosphate) and/or the loss of younger, more substituted mineral, at the expense of older more-perfect mineral (68). Our decreased carbonate-to-phosphate ratio did not correlate with any histomorphometric bone turnover markers. In contrast, using a less robust correlation (Pearson’s), studying renal osteodystrophy patients, carbonate-to-phosphate ratio was reported to inversely correlate with quantitative histomorphometric osteoblast and osteoclast surface fractions, markers of bone remodeling activity ⁽⁹⁾. Our results, more closely resemble another study of patients with chronic kidney disease ⁽⁷⁰⁾ wherein carbonate-to-phosphate ratio was not correlated with any histomorphometric bone turnover markers, using Spearman’s correlation. Taken together, the inconsistent correlations suggest that this ratio may not predict whether remodeling is increased or decreased, but rather they indicate that the mineral content is abnormal, reflecting an area typical of older or micro-damaged bone.

With active remodeling, one would also expect an increased mineral-to-matrix ratio, increased crystallinity, and decreased acid phosphate content; any changes in these values, as noted above, were not significant in either univariate or multivariate models, probably due to the paired nature of this study. Mineral-to-matrix ratio values did rise with age in our correlation analysis. Acid phosphate substitution in cancellous bone only, decreased with age in these our correlations, likely due to age-related increased remodeling. The observation in the present study that carbonate-to-phosphate ratios are lower, despite correction for age and BMD, speaks to the mechanism of fragility, suggesting that the fracture risk is associated with abnormal, perhaps damaged, bone.

Importance of Heterogeneity

It was our contention that greater heterogeneity in bone composition is associated with a more fracture resistant material ^(5,7,30). This contention was based on the observations that regions of heterogeneous material properties act as intrinsic toughening mechanisms in bone ^(71,72) and that micro-cracks stop proliferating at cement lines or other barriers ⁽⁷³⁾. For example, a reduction in heterogeneity of Raman vibrational spectroscopic variables (similar to FTIRI variables) was reported in the femoral cortex in aging patients ⁽⁵⁹⁾. This was in contrast to reports that mineralization homogeneity, as measured by BMDD, was increased in trabecular regions of non-fracture cases rather than in fracture cases ⁽⁷⁴⁾, and that mineralization heterogeneity increased with age in both cortical and cancellous bone ^(74,75,76). Our data agreed in part with a recent study comparing femoral neck cortical BMDD in aging vs. osteoporotic samples. No differences in homogeneity were found among these samples ⁽⁷⁶⁾. The observation in the current study of 120 patients with an increase in heterogeneity of cancellous collagen maturity in fracture cases was initially surprising. The data, however, paralleled both BMDD data on cancellous calcium distributions ^{(65, 73,74)} and our own correlations of cortical collagen cross-link heterogeneity with number of fractures. The increase in heterogeneity is small, possibly reflecting the presence of a broader distribution of collagen enzymatic cross-links. The clinical significance of small changes in heterogeneity is difficult to assess, as they may reflect turnover changes, sensitivity of the methodology and the location from which the tissue was obtained. The FTIRI images illustrate differences in distributions, which currently are best quantified based on pixel distribution, however future advances in image analyses may provide better ways to describe these differences. It is interesting to note that the number of fractures trended toward a rise with cancellous carbonate-to-phosphate and mineral-to-matrix heterogeneity, suggesting that better metrics for documenting heterogeneity may reveal significant differences. The lack of differences in heterogeneity with fracture status in the current study is likely due to the paired nature of the study, especially as heterogeneity was reported to be reduced as a function of age ⁽⁵⁴⁾. BMD accounts at best for 60% of the predicted fragility fractures in women ^(78,79). Other factors ^(22–25), including uncertainty in BMD measurements, variations in bone quality parameters, etc., all likely contribute to these fractures. The specific bone quality measures most closely associated with fragility fractures can only be addressed in studies in which mechanical properties, BMD and FTIRI properties are assessed in the same cohort. We anticipate that when all the arms of Dr Recker’s Bone Quality study are complete, this question can be addressed and hypotheses developed to test how these variables affect fragility fractures.

Conclusions

This study, containing the largest database of iliac crest biopsies to date, has confirmed that FTIRI analysis of these age-and BMD-paired biopsies, does detect significant compositional differences between patients with or without fractures. Specifically, we report that carbonate-to-phosphate ratio was decreased in the iliac crest biopsies in both cancellous and cortical bones of those patients with fragility fractures at other skeletal sites. This data will be critical in examining changes in bone composition in osteopenic and osteoporotic women treated with pharmacologic agents ⁽⁸⁰⁾. Our hypothesis that heterogeneity of fracture cases would be decreased was not supported. Correlations of heterogeneity with age and number of fractures suggest the need for further study of bone heterogeneity, at all levels.