Infrared Microscopic Imaging of Bone: Spatial Distribution of CO32−



This article describes a novel technology for quantitative determination of the spatial distribution of CO32− substitution in bone mineral using infrared (IR) imaging at ∼6 μm spatial resolution. This novel technology consists of an IR array detector of 64 × 64 elements mapped to a 400 μm × 400 μm spot at the focal plane of an IR microscope. During each scan, a complete IR spectrum is acquired from each element in the array. The variation of any IR parameter across the array may be mapped. In the current study, a linear relationship was observed between the band area or the peak height ratio of the CO32− v3 contour at 1415 cm−1 to the PO43− v1,v3 contour in a series of synthetic carbonated apatites. The correlation coefficient between the spectroscopically and analytically determined ratios (R2 = 0.989) attests to the practical utility of this IR area ratio for determination of bone CO32− levels. The relationship forms the basis for the determination of CO32− in tissue sections using IR imaging. In four images of trabecular bone the average CO32− levels were 5.95 wt% (2298 data points), 6.67% (2040 data points), 6.66% (1176 data points), and 6.73% (2256 data points) with an overall average of 6.38 ± 0.14% (7770 data points). The highest levels of CO32− were found at the edge of the trabeculae and immediately adjacent to the Haversian canal. Examination of parameters derived from the phosphate v1,v3 contour of the synthetic apatites revealed that the crystallinity/perfection of the hydroxyapatite (HA) crystals was diminished as CO32− levels increased. The methodology described will permit evaluation of the spatial distribution of CO32− levels in diseased and normal mineralized tissues.


THE MINERAL phase in bone consists of poorly crystalline, carbonate-containing hydroxyapatite (HA). Structurally, replacement of PO43− in the HA lattice by CO32− leads to a change in lattice dimensions and increased disorder.(1–7) The level of CO32− in bone mineral varies with the age of the individual and on average probably exists at ∼6 wt%(8) although a range of values has been suggested.(9, 10) The level may depend on anatomical location (e.g., trabecular vs. cortical).(11) Levels different from the range encountered in normal bone have been reported for various pathological conditions.(12–16) The relationship between CO32− content and the structure of the mineral phase was investigated as a basis for understanding age-related alterations.(17) Carbonate substitution affects the mineral solubility(5, 8, 9) and the presence of CO32− is thought to play a significant role in bone resorption.(18) Studies with point-by-point infrared (IR) microspectroscopy have shown that qualitative changes in the carbonate/phosphate ratio can provide insight into the effects of matrix modification and age in transgenic animals and human bone.(19–21) This work describes the development of a method to determine quantitatively the spatial distribution of CO32− in mineralized tissues with IR imaging and will enhance our understanding of the significance of carbonate in normal and diseased mineralized tissues.

IR spectroscopic investigations of homogenized bones and teeth(22–24) have shown that CO32− ions generally substitute for the two anionic sites in the HA lattice for PO43− ions (type B carbonate, the major substitution site in bone and dentin) or for OH ions (type A carbonate, which occurs at a significant level in dental enamel). In addition to these insertion sites, a labile carbonate species was identified and is thought to represent surface carbonate.(7) The vibrational mode widely used for identification of these species is the out-of-plane deformation, the v2 mode, at 879 cm−1 in the free ion, which undergoes substitution site-dependent spectral shifts on insertion into the HA lattice. The spatial distribution of the percent of each type of carbonate substitution in bone has been reported elsewhere.(7, 8, 22–24) Polarized IR studies of the v2 mode also have provided information about the orientation of the CO32− planes in the HA of mineralized turkey tendon.(25)

A limitation of the application of traditional IR spectroscopy to the study of tissues is the necessity for sample homogenization before spectral examination. This process renders it impossible to monitor spatial variations in the tissue constituents. Yet, spatial variations in these properties are evidently a major determinant of biological function. The application of IR microscopy techniques (either using point-by-point methods or array detector-based microscopic imaging) serves to overcome the limitations of sample homogenization. In previous publications(26, 27) our laboratories have established the feasibility of acquiring IR microscopic images from normal and abnormal states of mineralized tissue and cartilage. Several IR spectroscopic parameters have been developed for molecular characterization of the mineral and protein components. Although some of these parameters that measure the mineral content, mineral crystallinity, and collagen cross-linking can be applied in both point-by-point Fourier transform infrared (FTIR) microspectroscopy and IR imaging,(28, 29) the aforementioned carbonate analysis is not applicable because the v2 mode of CO32− lies outside the operating range of the current generation of mercury-cadmium-telluride array detector elements, which have a low frequency cut-off of ∼900 cm−1. The only other CO32− mode that is currently feasible for IR imaging is the v3 mode, which for B type substituted CO32−, consists of a spectral doublet with components at ∼1419 cm−1 and 1450 cm−1. This region of the spectrum overlaps vibrations from both the protein components of the tissue and the embedding material (polymethylmethacrylate [PMMA]) most frequently used for sectioning the tissue, thus rendering quantitative determination of CO32− somewhat awkward. Nevertheless, the current study shows the feasibility of using spectral subtraction techniques to overcome spectroscopic interference in this spectral region and therefore to use the v3 mode for quantitative imaging of CO32− in bone.


Preparation of carbonated HA

Type B carbonate apatites were prepared following the procedure of Penel et al.(30) Briefly, a phosphate solution [0.216 M (NH4)2 HPO4 in 30 ml of H2O and 5 ml of NH3 · H2O with varying levels of NaHCO3] was added very slowly over a 3-h period to a constantly stirred calcium-containing solution [0.25 M Ca(NO3)2 · 4H2O dissolved in 30 ml of H2O and 10 ml of NH3 · H2O] at 80°C. The precipitates were allowed to mature at 80°C for an additional 2 h. The precipitates were filtered, washed briefly with distilled water, and dried at 60°C overnight. All reagents were analytical grade. Reaction conditions were kept constant except for different levels of sodium bicarbonate.

FTIR and X-ray diffraction analysis

Carbonate apatite crystals were analyzed by FTIR as potassium bromide (KBr) pellets (200:1, wt/wt) on a Mattson RS-1 spectrometer (Mattson, Madison, WI, USA) with 512 scans coadded at 4 cm−1 resolution. Interferograms were apodized with a triangular function and Fourier-transformed with one level of zero filling. The sample compartment of the spectrometer was purged constantly with dry air generated from a Whatman gas generator (Whatman, Haverhill, MA, USA).

X-ray spectra were collected on a Siemens D-5000 Powder Diffractometer with automatic sample feeder (Siemens, Iselin, NJ, USA) using Ni-filtered Cu-Kα (1.545 A) radiation. Ground samples were scanned from 24° to 37° (2θ) at 0.05° intervals. Data were accumulated for 4 h to achieve good signal/noise ratios.

Two-dimensional IR analyses

A detailed description of the two-dimensional (2D) IR correlation spectroscopy techniques developed by Noda(31) as applied to the study of HA crystallinity has been published elsewhere.(32) The spectra are normalized to the integrated peak areas to compensate for 2D features arising simply from differences in sample concentrations rather than from alterations in spectral contours.

FTIR imaging

Iliac crest biopsy specimens were acquired as part of routine diagnoses and provided under an Institutional Review Board (IRB)-approved protocol by the Pathology Department of the Hospital for Special Surgery. Samples had been fixed in ethanol, embedded in PMMA, and were cut into ∼5-μm-thick sections before placement between two BaF2 windows on the instrument stage. Spectra were acquired with a BioRad Sting-Ray system (BioRad, Cambridge, MA, USA). In this device, a 64 × 64 element mercury-cadmium-telluride (MCT) focal plane array detector, mapped to a 400 μm × 400 μm spot at the focal plane of an IR microscope, is coupled to a step-scan interferometer. Routinely, data from one scan of 1024 steps with 81 coadditions at each step were collected at 8 cm−1 spectral resolution with one level of zero filling.

Carbonate analysis

The CO32− contents of the synthesized apatites were obtained by standard carbon-hydrogen-nitrogen elemental analyses. The content of CO32− in the synthetic HA samples was taken as 5× the analytical carbon content.

Data analysis

The coherent crystal lattice dimension in the c-axis direction (c-axis particle size and perfection) of synthetic apatites was calculated from the Scherrer formula:

equation image

where Bhkl is the half-width of the hkl diffraction peak, λ is the wavelength of the radiation, K is a constant assumed here to be 0.9, and θhkl is the value in degrees (2θ) of the diffraction maxima for the hkl reflection.

IR imaging data were compiled from the 4096 single spectra, which were baseline-corrected in the region of interest before any spectral manipulations. For the quantitative determination of CO32−, the contributions of embedding material (PMMA) and organic matrix (collagen) to the original spectra were subtracted based on spectra of the pure compounds. Thus, the intensity of the C = O stretching mode of PMMA at ∼1720 cm−1 and the amide I frequency at ∼1650 cm−1 (for collagen) were used as standards for spectral subtraction because there were no mineral contributions at these frequencies in the tissue spectra. For each spectrum in the IR imaging data sets, these correction factors were applied globally to the entire spectrum, thereby minimizing contributions from nonmineral components of the tissue to the CO32− v3 region.


Determination of CO32− in synthetic HA

A series of FTIR spectra (750-1500 cm−1) of synthetic HA with increasing amounts of CO32− is shown in Fig. 1. As the levels of CO32− increase, several spectral changes are evident. First, the v2 band near 875 cm−1, known(18–20) to be comprised of at least three underlying subbands corresponding to A-type CO32− at 878 cm−1, B-type CO32− at 871 cm−1, and relatively minor nonspecific or surface CO32− at 866 cm−1 gains in intensity relative to the PO43− v1,v3 contour between 950 and 1200 cm−1. Second, a similar relative intensity increase is observed in the CO3= v3 spectral region. The latter is comprised of a single band near 1419 cm−1 in the isolated ion (D3h symmetry), but develops a broad high-frequency shoulder near 1450 cm−1 on insertion into the HA lattice. The observed ratio of the 1419-1450 bands is in good accord with the IR data of Rey et al.(7) for B type CO32− substitution. This point is discussed in the following paragraph. Finally, increasing levels of CO32− are accompanied by broadening of the spectral features within the phosphate v1,v3 contour, which are indicative of decreased mineral crystallinity/perfection.

Figure FIG. 1..

A series of mid-IR spectra of HA synthesized with (bottom to top) increasing levels of CO32− (wt% in the apatitic phase) as follows: a, 1.05; b, 3.33; c, 4.35; d, 7.80, e, 8.78. The spectroscopic features of interest in the current work are labeled.

The quantitative relationship between IR spectral parameters and the analytically determined carbonate levels in these synthetic samples is shown in Fig. 2. The most useful IR parameter for eventual imaging applications is the band area ratio of the CO32− v3 contour to the PO43− v1,v3 contour. It is noted that comparisons of parameters arising from single bands (peak heights, areas, etc.) are subject to uncertainties arising from thickness variations between samples. Ratios of areas or peak heights therefore are used to control for sample-to-sample variations in the weights of the HA phase used in the analysis. In addition, the use of parameters involving ratios in IR microscopy compensates for sample-to-sample thickness variations in microtomed sections. The drawback of the current measurement is the inherent assumption that the molar integrated area of the phosphate v1,v3 contour does not change as the mineral crystallinity/perfection is altered. This conjecture cannot be tested easily so that the aforementioned intensity ratio must be considered as an approximation to the relative carbonate concentration. However, the high value of the correlation coefficient between the spectroscopically and analytically determined ratios in Fig. 2 (R2 = 0.989) attests to the practical utility of the current intensity ratio for IR determination of bone carbonate levels.

Figure FIG. 2..

The linear relationship between the analytically determined weight percent CO32− and the intensity ratio of CO32− v3 mode to the PO43− v1,v3 mode. The regression equation for this determination is Y = 6.85 × 10−3 + 0.0349X, where Y = I1415/1030 and X = weight percent of carbonate determined analytically. The correlation coefficient is R2 = 0.989.

Effect of CO32− on HA crystallinity/maturity

As is evident from Figs. 1 and 2, increasing levels of carbonate cause substantial alterations in the PO43− v1,v3 contour. Shape changes in this band are caused by alterations in the positions and widths of the underlying subbands and reflect alterations in mineral crystallinity. Several standard data reduction protocols including curve-fitting, derivative spectroscopy, and Fourier self-deconvolution have been previously used to analyze quantitatively changes in this contour when sample crystallinity/perfection is altered. As discussed elsewhere,(32) each of these methods offers advantages and disadvantages for the analysis of complex contours. Recently, we have shown that the resolution enhancement techniques of 2D IR spectroscopy can be applied to this contour to examine the subtle changes induced by alterations in crystallinity/perfection.(32) Figure 3 shows the 2D synchronous plot generated from the set of samples with increasing levels of CO32− in the spectral range of 900-1500 cm−1. The two components of the CO32− v3 contour are correlated negatively with the 1030-cm−1 component of the phosphate v1,v3 contour. The latter band is known to increase in intensity as mineral crystallinity/perfection is enhanced. In previous studies,(33) we showed that the 1045-cm−1 band reflects the amount of B-type CO32−. In the current work, a small peak in the 2D synchronous plot is visible at this position, although the bulk of the change in the contour arises from variation of the 1030-cm−1 component.

Figure FIG. 3..

2D IR synchronous plot for the series of synthetic samples with increasing levels of CO32− ion. The solid lines are positive correlation contours. The dashed lines are negative correlation contours.

CO32−-induced changes in HA crystallinity for this set of synthetic samples also were evaluated with X-ray diffraction. A series of X-ray powder patterns (2θ from 24° to 37°) of HA with increasing levels of CO32− is shown in Fig. 4. There is substantial carbonate-induced line broadening in all the reflections, but because of limitations with sample volume, calculations of changes in the a-axis dimension were not conclusive. The calculated crystal particle size and perfection derived from the Scherrer equation is plotted for the 002 reflection in Fig. 5 as a function of the level of CO32− substitution. Similar trends were noted for the 300 reflection.

Figure FIG. 4..

X-ray powder diffraction patterns for a series of samples of HA synthesized with (bottom to top) increasing levels of CO32− (wt% in the apatitic phase) as follows: a, 1.05; b, 3.33; c, 4.35; d, 7.80, e, 8.78.

Figure FIG. 5..

Calculated c-axis particle size/perfection as a function of CO32− concentration determined from the Scherrer equation applied to the X-ray powder pattern line widths for the 002 reflection.

Applications to IR microscopy

The primary difficulty with the use of the CO32− v3 mode for IR imaging is the potential interference from overlapped vibrational modes of both the protein component of the tissue and the PMMA embedding material. To compensate for contributions from these constituents to the CO32− v3 mode, a double subtraction protocol was used. The success of this procedure is shown in Fig. 6. The original baseline-corrected spectra from sites across trabecular bone (i.e., varying from low to high levels of mineral) selected from IR imaging data are shown in Fig. 6A. The same set of spectra following the scaled subtraction process to eliminate interference from PMMA is displayed in Fig. 6B. As evident in Fig. 6B, the subtraction process occasionally results in a small residual derivative-like feature, arising from slight spectral shifts in the PMMA C = O stretching band. The contribution of protein to the 1419 cm−1 band was estimated from the ratio of the amide I intensity to the 1419 feature in collagen spectra, which was used as a standard for subtraction. The resultant set of spectra is shown in Fig. 6C. The double subtraction process evidently is reasonable, because it results in CO32− v3 contours that closely resemble those in the series of synthetic HA derivatives (compare Fig. 6C with Fig. 1). To illustrate the quantitative determination of relative CO32− levels in mineralized tissue as determined from IR microscopic imaging, the results of analyses of four trabecular regions are reported. In addition, a series of 36 spectra were selected from IR imaging data across a region from one of the trabecular bone samples. The spatial variation in the carbonate/phosphate ratio across trabecular bone as monitored by the intensity ratio of the CO32− v3/PO43− v1,v3 bands following the double subtraction process is shown in Fig. 7A. The highest relative levels of carbonate occur at the edges, with the lowest relative levels evident at the center of the tissue. The variation in mineral crystallinity/perfection as monitored from the intensity ratio (I 1030/I 1020) is plotted in Fig. 7B and shows the trends in the direction opposite to those seen in the carbonate/phosphate levels. The I 1030/I 1020 index is highest at the center of the bone and decreases toward the edges.

Figure FIG. 6..

(A) A series of IR spectra selected from imaging data across trabecular bone showing contributions from PMMA at 1720 cm−1 and from the protein amide I mode near 1660 cm−1. (B) The same series of IR spectra as in panel A in which the contributions from PMMA at 1720 cm−1 have been removed mathematically while the protein amide I mode near 1660 cm−1 are still present. (C) The same series of IR spectra as in panel B from which the protein amide I mode near 1660 cm−1 have been removed mathematically.

Figure FIG. 7..

(A) Variation in the intensity ratio of CO32− v3 mode to the PO43− v1,v3 mode across a trabecular bone sample from a human iliac crest biopsy specimen. The spectra from which these data were obtained were selected from the set of IR imaging data shown in Fig. 6 (B) Variation in the intensity ratio of two peaks at 1030 cm−1 and 1020 cm−1 within the PO43− v1,v3 contour across the trabecular bone sample shown in panel A.

Finally, the variation in relative CO32− levels in one of 400 μm × 400 μm sections of trabecular bone as determined from an IR imaging data set is depicted in Fig. 8A. The area ratio of the CO32− v3/PO43− v1,v3 bands is converted to weight percent CO32− in the mineral phase according the regression equation generated from the data in Fig. 2, color coded, and plotted as an image. In four images of trabecular bone the average CO32− levels were 5.95 wt % (2298 data points), 6.67% (2040 data points), 6.66% (1176 data points), and 6.73% (2256 data points). The average value of CO32− from these images is ∼6.38 ± 0.14 wt% (7770 data points from the four trabecular bones). In one image of cortical bone from human iliac crest biopsy specimens the CO32− levels were 6.40% (3792 data points). The highest levels of CO32− were found at the edge of the trabeculae. The highest relative levels of CO32−, approaching 10-11%, occur at in a very small number (<1%) of pixels near the edges of the bone. Such high levels of carbonate were seen only in this trabecular bone sample. In the other samples studied, the highest weight percent CO32− observed was 8% or 9%. A histogram depicting the overall CO32− distribution from the four trabecular bone samples are shown in Fig. 8B.

Figure FIG. 8..

(A) Variation in the intensity ratio of CO32− v3 mode to the PO43− v1,v3 mode throughout a trabecular bone sample from a human iliac crest biopsy specimen, acquired from IR microscopic imaging. The color-coding reflects the actual intensity ratios measured from the IR imaging data set. The right hand scale is that derived from the regression equation from the data in Fig. 2. The number of data points used in the analysis is less than 4096 (the number of detectors in the array) because the sample does not fill the field of view. (B) The distribution of CO32− values summed over the four trabecular bone samples studied.


The current work describes the first quantitative determination of the spatial variation of CO32− in mineralized thin tissue sections using IR imaging. Such a characterization may lead to improved understanding of the functional role of the ion in normal and diseased states. It can be used to correlate variations in microarchitecture with carbonate level or to examine the relative carbonate content adjacent to active osteoclasts or quiescent osteoblasts. It also can be used in tooth sections to distinguish enamel and dentin. The values of the intensity ratio across trabecular bone reveal an average value of 6.38 ± 0.14 wt% CO32− in the apatitic phase. As reported values of bone CO32− range from 4 to 7.4%, the average value determined here lies well within the currently accepted range.

The effect of CO32− substitution on HA crystallinity as reported previously(5, 9, 34) also is evident from the 2D IR data shown in Fig. 3. The 2D IR synchronous plots show a strong negative correlation between the 1030-cm−1 component of the PO43− v1,v3 contour and both components of the CO32− v3 mode. In the current instance the correlation is strengthened by the X-ray diffraction data from the synthetic apatites (Fig. 4). These results are in accord with our previous investigations that show the 1030-cm−1 component to be enhanced as the HA crystallinity/perfection is increased. Increased levels of carbonate thus are indicative of an HA phase with reduced crystallinity/perfection.

The determination of CO32− in homogenized bone by IR is well documented. Some 35-40 years ago, Emerson and Fischer(35) and Baxter et al.(36) observed a doubling of the v3 mode. They attributed the additional spectral feature near 1450 cm−1 to symmetry lowering of the ion in the tissue. The use of the low-frequency component of v3 to determine CO32− in apatites was shown by Featherstone et al.(37) They determined that the absorbance ratio of this feature to the phosphate v4 band at 575 cm−1 allows an estimate of carbonate levels to better than ±10% in the biologically relevant range of 1-12 wt%. The phosphate v4 band is precluded from the current study because of detector limitations but the v1,v3 mode used instead offers the advantage of being the strongest band in the IR spectra of mineralized tissues.

An important issue in the current experiments is the substitution pattern of CO32−. Rey et al.(7, 22, 23) have made a thorough study of the effect of the substitution type on the CO32− v3 spectral region. Pure type A CO32− in synthetic samples exhibited a well-defined doublet in synthetic samples with components at 1463 cm−1 and 1435 cm−1, a pattern not observed in the current bone samples. It is therefore concluded that there is no significant type A substitution in our samples. Type B CO32− also presents a well-defined doublet in synthetic samples with components observed by Rey et al.(7) at 1456 and 1422 cm−1 with a peak height ratio (calculated from Fig. 3B of Rey et al.(7)) of 1.21. In our synthetic samples, we have observed the same doublet with a linear decrease in the component intensity ratio from 1.27 to 1.15 as the CO32− level increased from 0 to 10%. The variation in the ratio may be related to a differential broadening in the widths of the two components as the CO32− level increases. It is evident from a comparison of Rey's data(23) with ours that the current bone samples exhibit negligible interference from type A substitution.

Extensions of the current experiment appear feasible. For example, with appropriate control samples, it should be possible to image spatial variations in both A- and B-type CO32− in dental enamel. In addition, the variation of CO32− between normal and abnormal states can be determined in a straightforward fashion. In addition, relationships between the spatial distribution of CO32− and cellular activity will be accessible.


This work was supported by a Public Health Service Grant (AR 41325) to A.L.B. and R.M. IR images were collected in the IR imaging core at the Hospital for Special Surgery, supported by AR 46121.