2.1.

Bone Samples

Five human tibiae from three individuals (one male and two females) of ages 58, 72, and 88 were obtained from the Vesalius Clinical Training Centre at the University of Bristol, with appropriate ethical approval and material transfer agreements. The tibiae were examined by an orthopedic surgeon and showed no signs of bone disease (deaths were due to soft-tissue cancers and stroke).

Additional long bones were collected from other species (metacarpal bones from Thoroughbred racehorses, the metacarpus of a 1-year-old red deer stag, and the femur of a skeletally mature Welsh Mule domestic sheep) were also studied in order to strengthen the biological significance of our hypothesis. All the bones were frozen after being excised and stored in a freezer (193 K). Soft tissue and periostea were removed with scalpels before spectroscopic examination.

2.2.

Raman Data Collection

The Raman spectrometer used to collect the data was custom built by Cobalt Light Systems Ltd. (Oxfordshire, United Kingdom). It utilized an 830-nm near-infrared laser, which delivered 300 mW to the sample (1 mm spot size). The collection optics imaged the scattered light into the fiber-optic bundle and then into a spectrograph with a CCD detector (Andor iDus 420 BR-DD, Belfast, Northern Ireland) at its output. To aid overall collection efficiency, the low-loss Optran WF fiber bundle used a round configuration of 33 fibers to collect the filtered Raman signal; the spectrograph end of the fiber bundle was configured as a linear array to optimally fill both the spectrograph input slit and the available vertical extent of the CCD. The detector had a spectral resolution of $\sim 8{\mathrm{cm}}^{-1}$, and its software removed spurious signals that were due to cosmic-ray events.

The spectrum-collection process was automated by loading the bones on a motorized translation stage that was controlled by the instrument’s software. The software enabled automated movement of the sample stage and acquisition of Raman spectra. The flat medial surface of each tibia was then scanned along the full length of the shaft; a 30-s spectrum ($30\times 1\mathrm{s}$ accumulations) was taken at each step (every 1 mm). The cranial surfaces of the horse, sheep, and deer bones were scanned in the same way. The order of the measurements was randomized to negate the influence of instrumental and environmental drifts. The optical setup of the instrument and the scattering/absorption properties of near-infrared light meant all the Raman signal that was collected originated from within $\sim 1$ to 2 mm of the scanned surface.

In some of the previous mechanical and ultrasonic studies discussed above, the variation of elastic moduli and fracture stresses along the cephalocaudal axis were ascribed to the anisotropic manner in which the osteons are orientated in the bone material.^10,11 Unlike some studies previously reported in the literature,^23,24 the Raman spectrograph in the present study was largely insensitive to orientation/polarization effects (it had no polarizer, a large spot size, utilized optical fibers, and had low numerical aperture²⁵). The instrument’s insensitivity to polarization was confirmed by probing highly aligned tendon samples at different orientations and also by comparing different Raman band ratios from bone samples (i.e., the variation in the phosphate ${\nu }_1/\mathrm{amideI}$ ratio and the variation in phosphate ${\nu }_4/\mathrm{amideIII}$ ratio).²³

2.3.

Raman Data Processing

The variable sensitivity of the Cobalt instrument (i.e., of the filters and detector) across the spectral range was corrected using an HgAr calibration lamp. The broad fluorescence backgrounds of the bone spectra were removed using a polynomial fitting routine (script written in-house on MATLAB®, The Mathworks Inc., version 2007b).²⁶

The mineral-to-collagen ratios were determined by comparing band heights [phosphate $\nu 1$ band ($\sim 960{\mathrm{cm}}^{-1}$) for mineral and proline/hydroxyproline bands (average height across the band envelope $\sim 830$ to $900{\mathrm{cm}}^{-1}$) for collagen] and were normalized to 100% (divided by the mean) for the compositional trend plots. The phosphate-to-carbonate ratio was calculated by dividing the phosphate $\nu 1$ band intensity by the carbonate band ($\sim 1070{\mathrm{cm}}^{-1}$) intensity (illustrated in Fig. 2 of Ref. 21).

The trend plots (Figs. 3, 4, and 6) were smoothed with the use of rolling averages (each data point is the average of a spectrum and two neighbors on either side) and averaged using an interpolation function (OriginLab, OriginPro 8.5).

2.4.

Analysis of Bulk Composition

The data in Fig. 2(b) was collected with a Renishaw inVia Raman microscope ($\times 5$ Leica objective) with an 830-nm laser that produced 100 mW at sample (Renishaw, Gloucestershire, United Kingdom). One of the human tibiae (from a 58-year-old male) was sectioned after the other data had been collected, and it was divided in half along the coronal plane so as to expose the full thickness of the cortex for further spectroscopic examination. Ten spectra were collected from the cortex at four anatomical positions (distal end, proximal end, and two near the mid-shaft); the spectra were collected in a line across the full cortex (from the periosteal surface to the endosteal surface). The data were processed in the same way as above.

2.5.

Biochemical Analysis

3.1.

Raman Spectroscopy

Typical Raman spectra of human tibial bone are shown in Fig. 1; the spectra are very similar, but inspection of the mineral bands (phosphate and carbonate bands) shows that the bone from the mid-shaft (halfway along the bone) has a higher ratio of mineral-to-collagen than bone from a more proximal region (spectrum was taken from point 80% along the total length of the tibia, measured from distal end). Figure 2(a) shows that a similar difference can be seen at the distal end of the tibia; this time the values are shown for five human tibiae. The average mineral-to-collagen ratio for the mid-shaft tibial bone is $\sim 20\%$ higher ($p<0.005$) than that found $\sim 7\mathrm{cm}$ from the knee (15% of the total length from the distal end). Figure 2(b) shows that this difference is not a surface effect and that the $\sim 20\%$ difference in mineral-to-collagen ratio holds across the thickness of the cortex ($p<0.005$).

Fig. 1

Raman spectra of human tibial bone. The bone halfway along the tibia (red dotted spectrum) has a greater mineral-to-collagen ratio than that from a more proximal part of the cortex [the black spectrum is taken 80% along the length of the bone (from distal end)].

Fig. 2

(a) The histogram shows the average of the five tibiae. The ratio of mineral to collagen is smaller near the end of the tibia (15% from distal end) than at mid-shaft (50% from distal end). (b) The bulk bone composition across the full thickness of the tibial cortex. Ten spectra were collected from the cortex at four anatomical positions (distal end, proximal end, and two near the mid-shaft); the spectra were collected in a line from the periosteal surface to the endosteal surface of a sectioned human tibia.

When the mineral-to-collagen ratio is measured at hundreds of points along the cephalocaudal axis, a macroscopic-scale structure is revealed; the mineral-to-collagen ratio along the mid-section of the diaphysis is relatively stable (from $\sim 25$ to $\sim 75\%$ of the total tibia length the ratio varies by $\sim 5\%$),²⁹ but in the extremities, the mineral-to-collagen ratios decrease by 10% and keep decreasing (Fig. 3). Figure 4 shows similar macroscopic scale mineralization patterns in the distal cortices of equine metacarpal bones, the distal cortex of an ovine femur, and along the whole length of a cervine metacarpal bone.

Fig. 3

Mineral-to-collagen ratio along the length of the human tibia; the bone near the mid-shaft has a larger mineral-to-collagen ratio than bone nearer the metaphysis (mean of five tibiae).

Fig. 4

Mineral-to-collagen ratio along the length of various long bones. (a) The distal end (65% of the length, proximal end not shown) of four equine metacarpal-bone cortices (mean). (b) The distal end (75% of the length, proximal end not shown) of a cervine metacarpal-bone cortex. (c) The whole cortex of an ovine femur.

The mineral-to-collagen ratio of the tibia (measured with Raman spectroscopy) was compared with structural and architectural properties previously published by Capozza et al.²² The Capozza data were normalized to 100 (%) and plotted against similarly treated Raman data (Fig. 5). The comparison revealed a simple relationship between the mineral-to-collagen ratio and the thickness of the cortex, with the thinner areas being the ones with lowest ratio of mineral-to-collagen [Fig. 5(a)]. The analysis also showed more complicated relationships [Fig. 5(b)] between the mineral-to-collagen ratio and the bone mineral content, the circularity of the bone, and the cross-sectional moment of inertia. Below mineral-to-collagen ratio of 100 (i.e., below average), the structural properties adopt either one of two distinct values; these data points all come from the ends of the bones. Above mineral-to-collagen ratio of 100 (i.e., above average), the structural properties adopt wider range of values (these data points come from the mid-shaft).

Fig. 5

Mineral-to-collagen ratio versus structural/architectural properties²² of the human tibia; (a) as the cortex of the human tibia increases, the mineral-to-collagen ratio of bone material also increases. (b) The bone mineral content (BMC), cross-sectional moment of inertia (CSMI), and circularity have similar (more complicated) relationships with the Raman ratio. They each have two extreme values when the mineral-to-collagen is less than 100 (corresponding to the two ends of the tibia), and a range of values when the mineral-to-collagen is greater than 100 (corresponding to mid-shaft). All structural data from Capozza et al.²²

The variation in the phosphate-to-carbonate ratio along the length of the tibia is shown in Fig. 6. The pattern has similarities with the mineral-to-collagen ratio profile in Fig. 3 in that it is reasonably uniform along the mid-shaft ($<3\%$) but decreases (by 6%) toward the ends of the bone. There are, however, differences between the phosphate-to-carbonate pattern and the mineral-to-collagen pattern; by comparing Fig. 6 with Fig. 3, it can be seen that the carbonate substitution does not vary as much as the mineralization (note figures are plotted on the same scale); also the decrease of the phosphate-to-carbonate ratio at the proximal end of the bone begins nearer the midpoint of the bone and decreases more slowly.

Fig. 6

The composition of the bone apatite is not uniform along the length of the human tibia (average of five tibiae); the ratio of phosphate to carbonate is larger near the mid-shaft than toward the metaphyses.

3.2.

Biochemical Analysis

The results of the biochemical analyses are shown in Fig. 7(a). The mineral content of the bone samples, calculated by the weight change following extraction with EDTA, was $68.4\%\pm 6.0$ ($\text{mean}\pm \mathrm{SD}$) and levels showed regional variation that agreed with the Raman data. The water content ranged from 2.6 to 18.8% ($7.5\%\pm 4.1$) and was lower in the mid-diaphysis than at the diaphysis ends. Collagen content showed a reverse pattern of that seen in the mineral content along the diaphysis ($17.4\%\pm 3.9$). Those regions of bone with the highest mineral content had the lowest sulfated GAG content in the organic component.

Fig. 7

(a) The biochemistry results show the percentage of mineral, water, and collagen for different samples of bone taken at different points along the length. The fraction of glycosaminoglycan (μg per mg of dry demineralized bone) is also shown. (b) These results are summarized in this ternary diagram, which shows the relative proportions, by weight, of the main constituents of human tibia bone; mid-shaft specimens are represented as circles and specimens from the proximal and distal ends of the bone are represented as triangles. All the bones in nature would lie near the shaded area.

Figure 7(b) shows a version of the ternary diagrams similar to those in Refs. 1 and 6; the shaded gray area roughly corresponds to the area where all the bones found in nature would lie. In this ternary diagram, the bone is from different regions of the human tibia rather than functionally adapted bones from different species (that are plotted in the Refs. 1 and 6). The tibia samples spread across the shaded area with the mid-shaft samples clustering toward the highly mineralized corner and the proximal/distal samples clustering toward the antler end where less mineralized bones are found.^1,6

4.1.

Raman Spectroscopy of Whole Bones

Raman spectroscopic techniques have been used to create spatial maps of chemical changes in bones in a number of different studies; these include studies of undamaged, strained, and failed regions of bone, developing murine skulls, and across individual osteons.^30–32 Raman spectroscopy has also been used to investigate composition variation within the cortices of individual bones; a study looking at the mechanical property variation between cranial and caudal quadrants of a murine femur found that the composition did not differ between the two but that the orientation of the collagen fibrils did.³³ Other studies looked at the mechanical property and Raman spectral differences between newly formed bone near, and older bone further from, the periosteum³⁴ and at the differences between cubes of metaphyseal bone and diaphyseal bone.³⁵ Typically, these experiments mapped much smaller areas and length scales than those we report here.

In Sec. 3, it was stated that the composition of the bone along the mid-section of the diaphysis is relatively stable (from $\sim 25$ to $\sim 75\%$ of the total tibia length the ratio varies by $\sim 5\%$); our group has recently published a paper that quantified this mid-shaft compositional variation; we showed that across 10 cm of diaphyseal (mid-shaft) cortical bone, the phosphate-to-amide I ratio can vary by as much as 8%, and the phosphate-to-carbonate ratio by as much as 5%.²⁹ Both this study and the Raman tomographic imaging of large, intact sections of a canine limb bone, which have been reported by another group,³⁶ differ from the present study in that they did not look for variations at the macroscopic scale along (the cephalocaudal axis of) the entire bone.

4.2.

Regional Variation of Mineralization in Long Bones

Long bones grow in length from the growth plates (physeal plates) at each end; bone material that is furthest from a growth plate has (if we neglect remodeling for the present) taken longer to mature and should have reached a later stage of mineralization. These differences in the age of the bone material have been used to explain findings from an x-ray tomographic study of mineralization gradients in the femora of 6- to 9-week-old rats¹⁸ and a Raman study that showed differences between metaphyseal bone and diaphyseal bone from the femora of 2- to 12-week-old mice.³⁵ Distance from the growth plate is not as relevant to most of the data presented here as all the bones were obtained from skeletally mature individuals [the exception being the red deer stag in Fig. 4(b), which did show the same pattern but was only 1 year old].

If more remodeling were to occur near the ends of long bones, the bone material there would, on average, be younger than the mid-diaphysis bone (and the associated resorption cavities would also contribute to lowering the density). There have been studies that show that this is the case in dogs.^37–39 In Marotti’s study, skeletons of stray dogs were collected and various long bones (metacarpal bone, metatarsal bone, femur, ulna, rib, etc.) were sectioned and examined histologically. He found that there were more secondary osteons per unit area near the ends of the bones than nearer the mid-shaft.³⁸ In younger animals, this phenomenon has been ascribed to the conversion of the endosteal surface of the cortex to trabecular bone in growing bones.⁴⁰ The Raman and biochemical data in the current study were taken across the cortex (rather than just the endosteal surface), but the site-specific remodeling described by Marotti may still be relevant.

It has long been proposed that the density patterns along long bones are a structural feature of developmental origin, with the density increasing as an adaptation to mechanical loading.^14,15 These density patterns could also be due to site-specific remodeling or they could be related to the site-specific adaptation of the mineral content of bone.

4.3.

Compositional Variation and Material Properties

Much of the variation in mechanical properties along the length of bones (particularly the variation in tensile strength) is due to variation in collagen fibril orientation.^10,11,33,41 However, the difference in the mineral-to-collagen ratio could also have a functional significance, for instance, a decrease in the mineral-to-collagen ratio of 10% could be expected to lead to a 7.5 to 10% decrease in the Young’s modulus.³ This may be beneficial to the function of the tibia if in regions where loads were transferred from one bone to the next, the cartilage was protected from destructive forces. Cortical bone material and the cancellous bone material are more pliant (less stiff) and more able to absorb energy (tough) when they have lower mineral-to-collagen ratios. Excepting the complications introduced by microstructure⁴² and remodeling, this has already been shown for cancellous bone material (the trabeculae do have a lower mineral volume fraction,^42,43 a smaller mineral-to-collagen ratio as measured with Raman spectroscopy,⁴⁴ and the material is 20 to 30% less stiff than cortical bone^43,45).

Les et al.,⁸ investigating the distribution of material properties in equine metacarpal bones, suggested that the variations were adaptive, concluding that they served to enhance and modulate sagittal bending; however, they were unable to address whether the cephalocaudal variation was the result of adaptation to applied loads or was genetically determined, independent of exercise regimes.⁸

4.4.

Raman Spectroscopy and Biochemical Analysis

It was shown some time ago that the mineral content of bone (as measured with ashing) is correlated with the phosphate-to-amide I ratio (as measured with FTIR).⁴⁶ It has also been shown that Raman mineral-to-collagen ratios behave as expected for bones, which are known to have greatly differing mineral contents.³ The present study, in which the mineral content was measured by biochemical means, also showed regional variation in composition which was in agreement with the Raman data. Figure 7(b) shows that a change in the Raman mineral-to-collagen ratio actually represents a change in the relative proportion, by weight, of mineral, protein, and water.

4.5.

Regional Phosphate-to-Carbonate Variations

Figure 6 shows that there is a subtle difference between the composition of the bone apatite at the ends of the tibia and that in the shaft. As discussed above, there is evidence to suggest that increased carbonate inclusion in the apatite crystals reduce crystallinity and affect the mechanical properties of the bone.^19,47 Studies of aging in mice²¹ and osteoporosis in humans²⁰ have shown increased phosphate-to-carbonate ratio to be accompanied by decreased mineral-to-collagen ratio and altered bone mechanical properties, specifically increased capacity to deform (mice), and fewer low-trauma fractures (osteoporosis in humans), i.e., tougher/less stiff bone. Whether the mechanical properties of bone material near the ends of long bones is tuned by composition or whether the carbonate inclusions are more likely to occur in the less mineralized bone cannot be addressed by the present study. That a broad flat Raman band associated with the organic phase of bone is known to overlap with the carbonate band (from $\sim 1000$ to $\sim 1100{\mathrm{cm}}^{-1}$) further complicates the analysis.

4.6.

Limitations and Future Directions

As described above, variations along the length of long bones have been recorded using other analytical techniques (e.g., x-ray, ultrasound, gravimetric, collagen orientation, etc.). Comparisons of those techniques with the Raman data presented here provide scope for interesting future work.

Figures 3 and 4 show that the transition from more highly mineralized (i.e., stiffer) bone to the less mineralized bone (i.e., more pliant) is gradual, meaning there are no sharp interfaces between materials of very different Young’s modulus and, thus, no large stress discontinuities. Future investigations will look at compositional trends in different long bones and look for differences related to gender or age. It would be interesting to know if a disruption in the trend (perhaps medically induced) would affect the bone’s performance and lead to increased or decreased risk of fractures in the shaft.⁴⁸ If the variation in composition along the cephalocaudal axis was altered by disease and stiffer mid-shaft bone was formed near the end of the bone, it could lead to destructive forces being transmitted to the cartilage/joint each time the bone was loaded.^49,50 One could speculate that the loss of bone, or the adaptation to new loading conditions (e.g., muscle wastage), could alter the mineralization patterns in long bones. In this event, any regions that developed more highly mineralized materials would suffer from reduced toughness and greater susceptibility to microcracks and/or fracture. As spatially offset Raman spectroscopy is developed as a noninvasive tool for monitoring bone transcutaneously, it will be possible to explore these hypotheses in vivo.^25,36,51

The results of the biochemical analysis in Fig. 7(b) hint that the compositional adaptation of bone material within a long bone fits the same paradigm as that of specialized functionally adapted bones from different creatures (i.e., bulla, antler, etc.).^1,2,7 If the composition of bone is an independent design parameter of bone matrix, then understanding how regions of bone (that are only separated by a few millimeters) can have such large variations in composition will be of interest. The cues (be they genetic or local) for osteoblast matrix synthesis may present novel therapeutic targets for controlling bone composition and, thus, bone’s material properties.