Mechanical Properties of Adult Vertebral Cancellous Bone: Correlation With Collagen Intermolecular Cross-Links


  • X. Banse M.D.,

    Corresponding author
    1. Orthopedic Research Laboratory, Université Catholique de Louvain, Brussels, Belgium
    • Université Catholique de Louvain, Orthopedic Research Laboratory, Tour Pasteur 5388, Avenue E. Mounier 53, B-1200 Bruxelles, Belgium
    Search for more papers by this author
  • T. J. Sims,

    1. Collagen Research Group, Division of Molecular and Cellular Biology, University of Bristol, Langford, Bristol, United Kingdom
    Search for more papers by this author
  • A. J. Bailey

    1. Collagen Research Group, Division of Molecular and Cellular Biology, University of Bristol, Langford, Bristol, United Kingdom
    Search for more papers by this author

  • The authors have no conflict of interest.


Although the mechanical strength of cancellous bone is well known to depend on its apparent density, little is known about the influence of other structural or biochemical parameters. This study specifically investigates the cross-linking of the collagen in human vertebral bone samples and its potential influence on their mechanical behavior. Multiple cylindrical samples were cored vertically in the vertebral bodies of nine subjects (aged 44–88 years). Three spinal levels (T9, T12 or L1, and L4) and three sample sites within a vertebral body (anterior, posterior, and lateral) were used, for a total of 68 samples. The density was measured with peripheral quantitative computed tomography (pQCT) and all cylinders were mechanically tested in compression. After mechanical testing, they were unmounted and used for biochemical analysis. The amount of collagen (wt/wt of bone) and its content in reduced immature cross-links, that is, hydroxylysinonorleucine (HLNL, mol/mol of collagen) and dihydroxylysinornorleucine (DHLNL), as well as stable mature cross-links, that is, hydroxylysyl-pyridinoline (HP), lysyl-pyridinoline (LP), and pyrrole cross-link were determined for each cylinder. None of the biochemical parameters correlated to the density. On multiple linear regression, the prediction of the mechanical properties was improved by combining density data with direct collagen cross-link assessment. The HP/LP ratio appeared as a significant predictor to the strength (r = 0.40; p = 0.001) and stiffness (r = 0.47; p < 0.001) samples with a high HP/LP ratio being stronger and stiffer. Additionally, the ultimate strain correlated to the HP or LP concentration (r = 0.38 or 0.49; p < 0.01). Different subjects had different HP/LP ratios and different HP or LP concentrations in their vertebral bone samples, and the location of origin within a subject had no influence on the concentration. These observations suggest that the nature of the organic matrix in adult vertebral bone is variable and that these variations influence its mechanical competence.


Considerable variability in the mechanical competence of vertebral cancellous bone has been reported, with modulus values ranging widely from 100 to 700 Mpa.(1,2) The amount of bone per unit of volume (BV/TV) or apparent density certainly is the major parameter influencing the stiffness and the strength of the overall structure, but a substantial part of the variance of the mechanical properties (typically 20–30%) remains unaccountable.(3–6)

Part of this unexplained variance is caused by different trabecular architecture. Several studies have investigated additional parameters, describing the organization of the trabecular network as, for example, thickness, spacing, and connectivity of the trabeculae.(7,8) Hence, despite extensive efforts, a complete knowledge of cancellous bone architecture and density does not allow a total prediction of its properties.(9)

Part of the unexplained variance in bone strength may be linked to differences in the nature of bone tissue itself. Bone tissue can be considered simplistically as a two-phase system of a collagen scaffold interspersed with the mineral hydroxyapatite. The collagen fibers are composed of parallel aligned, end-overlapped and quarter-staggered molecules and this precise organization allows both stabilization of the fibers by intermolecular cross-linking and nucleation of mineralization within the gap region of this molecular alignment.(10) In general terms, the collagen scaffold provides tensile strength and the mineral rigidity in compression.

The fibrillar type I collagen accounts for >90% of the organic matrix. The initial head-to-tail cross-linking of the molecules occurs enzymatically through lysyl and hydroxylysyl aldehydes in the nonhelical telopeptide regions, which react with a lysine or hydroxylysine residue within the triple helical domain of an adjacent molecule. As turnover decreases, these cross-links, primarily keto-imines, react with further lysyl or hydroxylysyl aldehydes to give the mature trivalent cross-links hydroxylysyl- pyridinoline (HP) and lysyl-pyridinoline (LP) and pyrroles.(10,11) The relative amounts of each of these cross-links depends on the level of hydroxylation of both the triple helical and the telopeptide domains of the molecule, each believed to be controlled by different lysyl hydroxylases.(12) The pyrrole and pyridinoline appear to be located at different ends of the molecule,(13) supporting the concept of a direct influence of the type of cross-links on the fibrillar structure of the collagen.

A reduction of intermolecular cross-linking is well known to result in fragile bones, for example, in osteolathyrism. However, little attempt has been made to include the nature of the cross-links of the extracellular collagen in the prediction of mechanical properties.(14) A study of partial lathyrism in young bone revealed that a 10% decrease in the immature cross-links produced a 15% decrease in mechanical strength.(15) Other animal models showed that alterations of post-translational modifications of the collagen led to a reduction in mechanical properties.(16–19)

In the case of human cancellous bone, few studies have combined investigations on the nature of the extracellular matrix with mechanical testing of samples. Bailey et al.(20) found no change in the cross-link profile of iliac crest cancellous bone samples with age, and the relationship with strength was most influenced by the total collagen content and the density. In contrast, changes in the cancellous bone of femoral heads have been reported in osteoporosis(21) and osteoarthritis(22) and the bone has been reported to be weaker.(23)

Despite the obvious relationship of collagen cross-links and bone strength there are no clear correlations to date. However, taking benefit of recent improvements of the methods used in cancellous bone testing,(24) the aim of this study was to show a correlation of bone mechanical properties with the intermolecular cross-linking of the collagen of adult human vertebral cancellous bone.


Sample collection

Sampling and density measurements have been described previously in detail.(25) Briefly, specimens came from 9 autopsy subjects. There were 3 women and 6 men aged 44–88 years and with no relevant bone disease. Three vertebral bodies were obtained in each subject at different vertebral levels: one thoracic (T9), one thoracolumbar (T12 or L1), and one lumbar (L4). The discs were removed carefully and the absence of prevalent fracture was checked by contact X-ray of the specimen. Eight cylindrical samples (8.2 mm in diameter) were cored vertically from each subject, two in the T9 vertebral body, three in the T12-L1, and three in the L4. Location within the vertebral body (sample site) was either anterior, posterior, or lateral (Fig. 1). There was no cortical shell in the cores. In one T12 specimen, the lateral sample was not obtained and in 1 subject the L4 vertebral body was discarded for degenerative disc lesions, leaving 68 samples for the study. These cores were not of uniform height (median, 25 mm; range, 19–29 mm). The bone marrow was not removed and samples were kept frozen until mechanical testing.

Figure FIG. 1.

Cylindrical samples (diameter, 8.2 mm) were cored vertically in the vertebral bodies from 9 subjects. The origin of a sample was defined by the name of the individual subject and two location variables: the vertebral level in the spine and the sample site of coring. Usually, two samples (one anterior and one posterior) were cored in the thoracic level (T9), and three samples (one anterior, one posterior, and one lateral) in the thoracolumbar (T12 or L1) and lumbar (L4) levels for a total of eight samples in a single subject. They were scanned with pQCT to measure the apparent density. After the mechanical testing in compression, each cylinder was dismounted, end plates were removed, and the collagen cross-link content of the central part of the core was assayed.

Bone density determination

Density of each sample (g/cm3) was measured with a peripheral quantitative computed tomography (pQCT) Research SA+ (Stratec, Pfrorzheim, Germany). The accuracy of the technique has been validated previously.(25,26)

Mechanical testing

All samples were tested mechanically using the “end-cap” technique as described by Keaveny.(5,24) Samples were fixed on brass end-caps and tested in compression up to a 2% strain at room temperature on a computer-controlled screw-driven testing machine (Zwick model Z50/TH3A; ZwickUlm, Germany). The forces were measured with a 2-kN load cell and deformations were recorded with a Multisens extensometer (Zwick). No preconditioning cycles were performed. The following parameters were determined: (i) The stiffness or elastic modulus (MPa) was calculated by linear regression of the slope of the stress-strain curve between 0.001 and 0.2% strain; (ii) The strength (MPa) was the stress at the first maximum on the stress-strain curve (called ultimate point); (iii) The strain to failure (%) was the strain (a normalized estimate of the deformation) at the ultimate point. This parameter indicates the capacity of the sample to support deformation and, hence, to absorb energy.(27)

Biochemical analysis: collagen properties

After mechanical testing, the samples were detached from the brass support, the glue being dissolved with acetone. At 2.5 mm from both its upper and lower extremities, they were cut using a cutting grinding band saw (Exakt, Norderstedt, Germany). The central part was retained for biochemical analysis (Fig. 1). As the end plates were removed (with a 2-mm margin of safety), the sample consisted exclusively in cancellous bone with no contamination by cortical bone, cartilage, or calcified cartilage. Samples were washed thoroughly under a jet of deionized water to remove bone marrow and blood cells, defatted in a 1:1 (vol/vol) chloroform-methanol solution, rinsed with methanol and then deionized water, and finally freeze-dried.

Collagen concentration

The amount of collagen was determined by hydroxyproline assay of an aliquot of the acid hydrolysate (6 M of HCl, 110°C, 16 h) using a continuous-flow Autoanalyzer (Burkard Scientific, Uxbridge, UK) based on the method of Grant.(28) The accuracy of the method was checked occasionally by determination of hydroxyproline on the amino acid analyzer. The collagen content was calculated on the dry weight of the bone assuming 14% hydroxyproline in the type I collagen. The resulting data then were used to calculate the cross-link values as mole per mole of collagen.

Intermolecular cross-links

The intermediate cross-links of the newly synthesized collagen were stabilized by borohydride reduction before acid hydrolysis, and both reduced and mature cross-links were determined after acid hydrolysis on a modified amino acid analyzer as previously described in detail.(29,30)

Briefly, the freeze-dried bone samples were homogenized, suspended in PBS, and reduced with potassium borohydride. The samples were washed, freeze-dried, reweighed, and hydrolyzed in 6 M of HCl for 24 h at 110°C in screw-topped glass hydrolysis tubes (Medline Scientific, Oxon, UK). The excess acid was removed by freeze drying, and the residue was dissolved in 0.5 ml of distilled water.

The hydrolysates were separated initially on a CF1 (Whatman, Kent, UK) cellulose column to remove the non-cross-linking amino acids. The samples were assayed for the borohydride reduced form of immature cross-links, that is, hydroxylysinonorleucine (HLNL) and dihydroxylysinornorleucine (DHLNL) together with the stable mature cross-links HP and LP using a modified gradient on an amino acid analyzer (Alpha Plus; Pharmacia, Loughborough, UK), as previously described. The location of the cross-links on the analyzer had been confirmed previously with samples of authentic cross-links prepared in the laboratory. Quantification of the cross-links was achieved using ninhydrin color reaction and their known leucine equivalents. The ratio of reducible cross-links (DHLNL/HLNL ratio) and the ratio of mature cross-links (HP/LP ratio) were computed.

Pyrrolic cross-link

Bone matrix (100 mg) was dispensed into polypropylene tubes (Sarstedt, Leicester, UK) and initially decalcified by suspension (20 mg/ml) in 0.5 M of tetrasodium EDTA, pH 7.5, for 3 days at 4°C. The EDTA was decanted after centrifugation (10.000 rpm) and the decalcified matrix was resuspended in distilled water, shaken, and subsequently centrifuged (10.000 rpm). After repeating this wash step, 1200 μl of 0.1 M of 3-[N-tris(hydroxymethyl)methylamino]-2-hydroxy-porpanesulphonic acid (TAPSO, Sigma-Aldrich, Poole, UK), pH 8.2, was dispensed into each tube and the sample was heat-denatured at 100°C for 30 minutes and then allowed to cool to 37°C. Next, samples were treated with 1000 U of trypsin (N-tosyl-l-phenylalanine chloromethyl ketone [TPCK]-treated) dispensed from a trypsin stock (5000 U/ml of TAPSO buffer) and left to digest by shaking gently for 18 h at 37 C. The pyrrole content of the digest was assayed by reaction with p-dimethylaminobenzaldehyde (DAB) using a microtiter plate format (Micro-test III flexible plate; Becton Dickinson, San Jose, CA, USA).(20) Quantification was facilitated by the inclusion of a 1-methylpyrrole (Aldrich, Dorset, UK) standard (range, 1–10 mmol/liter) prepared from a stock solution (50 mM in ethanol). A 50-μl aliquot was removed for total collagen content by hydroxyproline assay described previously. A further 250 μl of each sample was combined with 50 μl of DAB (11.4% wt/vol in 60% perchloric acid). Within 15 minutes of adding the DAB, all samples were filtered using a 0.2-μm syringe filter (13 mm, HPLC technology) and 180 μl of the sample was dispensed into a 96-well microtiter plate and read at 570 nm using a microtiter plate reader (Labsystems Multiskan MCC 340; Helsinki, Finland). The pyrrole content of the bone digests was calculated based on the extinction coefficient of 25,000 for 1-methylpyrrole and expressed as moles of pyrrole per mole of collagen based on the hydroxyproline assay.(31)

Statistical analysis

The statistical analysis was performed with SPSS, Inc. 10.0 (SPSS, Inc., Chicago, IL, USA). Linear regression was used to explore the relationship between the density and the other parameters. The origin of a given bone cylinder can be described by the name of the individual subject, the level in the spine, and the sample site in the vertebral body. The proportion of variance (η2, %) that could be explained by each of these three independent variables was estimated using a mixed model. The name of the subject was introduced as a random factor and the location variables were presented as fixed factors, meaning that the differences between levels and areas were considered as equal in each subject, the usual assumption in ANOVA.

Multiple linear regressions were performed with either the stiffness, strength, or strain to failure as dependent variables. These regression models always included the density as first predictor of the mechanics (this was mandatory because of the major effect of the density on cancellous bone mechanics) plus one of the biochemical parameters. This analysis was first done on the whole set of data and then on the nine mean values for a single subject.


Bone density

The mean apparent density of the samples was 0.17 g/cm3 (Table 1), with values ranging from 0.09 to 0.34 g/cm3. The density varied significantly among patients but the analysis revealed systematic differences within the vertebral body (Table 2). For example, the posterior samples were 20% more dense than the anterior or lateral.(25)

Table Table 1.. Biomechanical and Biochemical Parameters of the Bone Samples
original image
Table Table 2.. Between- and Within-Subject Variation in the Parameters
original image

Mechanical properties

Five samples were lost, leaving 63 samples with successful mechanical testing (Table 1). The stiffness ranged from 127 to 725 MPa (mean, 352 MPa). The strength values ranged from 0.60 to 6.17 MPa (mean, 2.37 MPa) and the strain to failure ranged from 0.72 to 2.01% (mean, 1.19%). The stiffness and strength of the cylinders were highly dependent on their apparent density (r = 0.91 and 0.90) and the strain to failure was moderately dependent (r = 0.41).

Intermolecular cross-links

Both reduced intermediate cross-links HLNL and DHLNL were identified, the latter being approximately double the concentration of the former. The major cross-link in these adult bone samples was the mature HP, the LP being about one-half that of the HP value. The total pyridinolines amounted to ∼0.2–0.3 mol cross-link/mole of collagen, and the pyrrole accounted for ∼0.1 mol/mole of collagen (Table 1). These figures are consistent with other published values in the literature for bone collagen cross-links.

None of the biochemical parameters correlated with the density. About 70% of the variance of the collagen content depended on the individual subject (Table 2). The most subject-specific aspect of the collagen cross-link profile was the HP/LP ratio (Fig. 2) as well as the concentration of LP in the bone (with 82% and 81% of the variance associated with the subject). The sample site within the vertebral body (anterior, posterior, or lateral) had no systematic influence on the biochemical parameters. Only a small part of the variance of the HP and LP concentrations (10%) was dependent on the level the samples were taken from, thoracic, thoracolumbar, or lumbar.

Figure FIG. 2.

A box plot showing the subject-to-subject variability in the amount of pyridinoline cross-links is shown. Different individual subjects (or patients) had different bone collagen properties; some had much more HP or LP than others and the ratio between these two important cross-links also varied between subjects. See text and Table 2 for statistical analysis.

Biochemical parameters and mechanics

The regression analysis indicated that after controlling the major effect of density, a substantial part of the “unaccounted variance” was explained by the biochemical properties of the collagen (Table 3). The HP/LP ratio was found to be a significant predictor of the stiffness (r = 0.47; p < 0.001) and strength (r = 0.40; p = 0.001). The concentration of pyridinoline cross-links (HP and LP) in the collagen also were correlated with the ultimate strain (r = 0.39 and r = 0.40, respectively; p < 0.01). Neither of the intermediate cross-links was determined by borohydride reduction (DHLNL or HLNL), nor were the pyrrole cross-links correlated with any of the three mechanical parameters.

Table Table 3.. Influence of the Collagen Properties on the Mechanical Parameters
original image

Because the cross-link profile was subject specific, we applied the regression procedure to the mean values of each subject (n = 9). The HP/LP ratio was still significantly predictive of the stiffness (r = 0.73; p < 0.05; Fig. 3) and the strength was (r = 0.65; p < 0.05). For the prediction of the ultimate strain, the concentration of LP remained a valuable predictor (r = 0.66; p < 0.05).

Figure FIG. 3.

After accounting for the density, samples from a given subject were stiffer than those from another subject. This unaccounted stiffness can be estimated as the residue of the regression between the stiffness and the density (vertical axis and residual stiffness, MPa). Each subject had a characteristic HP/LP ratio. The mean HP/LP ratio of an individual subject (horizontal axis) turned out to be a significant predictor (p < 0.05) of the residual stiffness. The vertical bar is ±SEM of the residual stiffness.


The data presented here suggest that the type of collagen cross-linking contributes significantly to the prediction of the biomechanical properties of cancellous bone from the vertebral bodies.

Cross-links and mechanics

Like others,(5,6) we observed that ∼20–30% of the variance in stiffness and strength of the vertebral cancellous bone was not accounted for by the density. Our results indicate that a significant part of this unaccounted variance was related to differences in matrix composition of the trabeculae. The HP/LP ratio was a good predictor of the stiffness and strength of the bone cylinders. Samples with a high HP/LP ratio possessed a better bone tissue quality than samples with a low ratio. Because this biochemical parameter varied significantly across subjects (Table 2), those subjects with a high HP/LP ratio had a relatively stiffer and stronger bone (Fig. 3).

The variability of the HP/LP ratio between subjects is caused by the different degree of hydroxylation of the specific lysine residues involved in cross-linking (Fig. 4). There are now strong indications that this hydroxylation is controlled by different lysyl hydroxylases acting on specific amino acid sequence domains of either the triple helix or the nonhelical telopeptides of the collagen molecule.(12) Variations in the degree of hydroxylation of these specific lysine residues lead to different cross-link profiles in different tissues, for example, the absence of pyridinoline and pyrrole in skin in contrast to their major presence in bone collagen and the absence of the pyrrole in cartilage collagen. The present observation that different subjects had different HP/LP ratios indicates that the activity of the triple helix hydroxylase(s) varies from one subject to another. Whether this is controlled genetically or depends on external conditions currently is unknown.

Figure FIG. 4.

The ratio between HP/LP depends on the hydroxylation of specific lysine residues on the triple helix. This hydroxylation is controlled by the activity of a lysyl hydroxylase (different from that which hydroxylates the telopeptide residues). Hyl, hydroxylysine; Lys, lysine.

The noncollagenous proteins are known to contribute significantly to the mineralization process.(32) Previously, we have shown that the lysyl hydroxylation in type I collagen of the tendon was reduced on mineralization of the tendon, suggesting that the change might be associated with the mineralization process.(33) To speculate further, one could consider the hypothesis that the type of cross-link present in the collagenous matrix also may influence the quality and extent of mineralization (i.e., the size and shape of the crystals), which in turn could affect the mechanical elastic properties of the bone.

The absence of any correlation of the modulus or compressive strength with cross-links in human iliac crest bone(20) is unclear and may be caused by the absence of specific loading direction of the fibers, which recently has been shown to be important,(34) or to the use of different protocols for mechanical testing. We previously reported a relationship of strength with the pyrrole cross-link in avian bone, but this may be related to the fact that the concentration of the pyrrole was double that of the pyridinoline in these animal.(15) In this study, analyses were confined to human trabecular bone and the pyrrole was less abundant than pyridinoline. Additionally, the avian cortical bone was tested in three-point bending, which leads to a tensile mode of failure.

Our data also indicate that samples with high pyridinoline content (HP and LP) had a relatively higher strain to failure. The failure of vertebral trabecular bone ultimately results from tensile fracture of horizontal trabeculae and from bending and buckling of the vertical elements.(35) Presumably, pyridinoline increases the resistance of the matrix to tensile and shear strength of the trabeculae and consequently preserves them as mechanically competent during the progressive deformation of the cylinder. This is consistent with the recent observation of a relationship between the energy to fracture and collagen denaturation in baboon cortical bone.(36)

Subject to subject variations and cross-link assay

Urinary pyridinolines have become the most commonly used marker of bone resorption for clinical research.(11) Generally, urinary LP values are considered to be more specific than HP for bone collagen, because of its higher abundance in bone than other collagenous tissues. Both free or peptide-bound LP can be quantified in urine, the amount released daily from bone depending on the amount of bone collagen degraded by the osteoclasts.

On direct analysis of the vertebral bone collagen, the concentration of LP varied considerably among subjects (Fig. 2). For example, MG had 0.03 ± 0.005 (mean ± SD) mol of LP per mole of collagen and LF had 0.11 ± 0.020. If both of these subjects degrade the same quantity of vertebral bone (g), the first will release much less LP than the second. Consequently, the urine cross-links certainly are sufficiently sensitive to follow the evolution of a given patient but maybe not to compare the metabolic activity of different patients.

Limitations and strengths of the study

The main limitation of this study is the relatively small number of subjects used to collect the samples (n = 9). We increased the number of samples by coring and analyzing eight samples in each spine. Consequently, the validity of the regression analysis on the overall number of cylinders (n = 63) is reduced by the repeated nature of the data. Conclusions from such regressions have to be limited to the relationship between the nature of a given cylinder (density and biochemical composition of the matrix) and its own mechanical behavior. After reduction of the data to a single mean value for a given subject (using n = 9), the correlation also was significant but the regression with two predictors on such a small number of values is questionable.

However, our study was strong in that we carried out all the measurements on human vertebral cancellous bone. The use of human material from an anatomical site that has the highest risk of fracture with aging certainly is important if we ultimately are to draw clinically relevant conclusions. This is much more difficult with animal models. The biochemical and mechanical analyses were performed on the same samples. Our sampling strategy was rigorous and proved its sensitivity to the within-subject variations in the structure or density.(25) The end-cap technique was used for measurement of properties of vertebral cancellous bone because it was designed specially to eliminate end artifacts thus allowing a precise measurement of the stiffness.(24)

In this study, we improved the prediction of the biomechanical properties of vertebral cancellous bone by combining pQCT (density) data and direct collagen cross-link assessment. Because both density measurements and pyridinoline levels (HP/LP in urine) are accessible in clinical practice, our observations open an interesting way of progress in the prediction of fracture risk.


This work was supported by the National Funds for Scientific Research (FNRS, Belgium).