Alterations of Mass Density and 3D Osteocyte Lacunar Properties in Bisphosphonate-Related Osteonecrotic Human Jaw Bone, a Synchrotron µCT Study

Osteonecrosis of the jaw, in association with bisphosphonates (BRONJ) used for treating osteoporosis or cancer, is a severe and most often irreversible side effect whose underlying pathophysiological mechanisms remain largely unknown. Osteocytes are involved in bone remodeling and mineralization where they orchestrate the delicate equilibrium between osteoclast and osteoblast activity and through the active process called osteocytic osteolysis. Here, we hypothesized that (i) changes of the mineralized tissue matrix play a substantial role in the pathogenesis of BRONJ, and (ii) the osteocyte lacunar morphology is altered in BRONJ. Synchrotron µCT with phase contrast is an appropriate tool for assessing both the 3D morphology of the osteocyte lacunae and the bone matrix mass density. Here, we used this technique to investigate the mass density distribution and 3D osteocyte lacunar properties at the sub-micrometer scale in human bone samples from the jaw, femur and tibia. First, we compared healthy human jaw bone to human tibia and femur in order to assess the specific differences and address potential explanations of why the jaw bone is exclusively targeted by the necrosis as a side effect of BP treatment. Second, we investigated the differences between BRONJ and control jaw bone samples to detect potential differences which could aid an improved understanding of the course of BRONJ. We found that the apparent mass density of jaw bone was significantly smaller compared to that of tibia, consistent with a higher bone turnover in the jaw bone. The variance of the lacunar volume distribution was significantly different depending on the anatomical site. The comparison between BRONJ and control jaw specimens revealed no significant increase in mineralization after BP. We found a significant decrease in osteocyte-lacunar density in the BRONJ group compared to the control jaw. Interestingly, the osteocyte-lacunar volume distribution was not altered after BP treatment.


Introduction
Under healthy conditions bone undergoes continuous remodeling to adapt to spatially and temporarily variable demands, through a delicate equilibrium between resorption and formation, which is performed by osteoclast and osteoblast cells, respectively.
Bisphosphonates (BP), which are commonly prescribed in the treatment of osteoporosis and bone metastasis, have been shown to reduce significantly the risk of fracture [1,2]. The action of BP relies on the reduction of bone resorption by inhibiting osteoclast activity. However, a severe and most often irreversible adverse effect of high-dosage BP treatment is the potential occurrence of osteonecrosis of the jaw [3][4][5][6][7]. Although multiple hypotheses have been formulated recently, the underlying pathophysiological mechanisms of bisphosphonate-related osteonecrosis of the jaw (BRONJ) are still not completely understood [8][9][10][11].
Bone remodeling results in a heterogeneous distribution of mineralized tissue units, with variable degrees of mineralization. This heterogeneity can be assessed from the bone mineralization density distribution by using techniques such as quantitative backscattered electron imaging (qBEI) [12][13][14], microradiography [15], or synchrotron radiation micro-CT [14,16]. The degree of mineralization of bone is a quality factor that influences the mechanical properties of bone [17].
Osteoclast and osteoblast activity is thought to be orchestrated by osteocytes, which are the most abundant type of bone cell and form a well-distributed network within the mineralized matrix [18]. These cells reside in cavities called lacunae measuring several hundreds of mm 3 in volume, and are interconnected through cell dendrites extending in tiny canals called canaliculi. The canalicular diameter of human bone has been reported to be within the range between 200 and 900 nm [19].
It was recently shown in a murine lactating model that not only osteoclasts are able to resorb bone matrix, but also that osteocytes remodel their peri-lacunar and peri-canalicular matrix [35]. Alterations in lacunar size have also been observed in response to changes of the mechanical environment, for example enlarged lacunae were reported in mice following space flight [36], or after glucocorticoid treatment [37]. In ovariectomized rats, both lacunar size and density were found to be altered in newly-formed bone after antiresorptive and anabolic pharmaceutical treatment [38].
So far, investigation of the three-dimensional (3D) structure of the LCN has been limited by the imaging techniques available [21,39]. Synchrotron radiation micro-computed tomography (SR mCT) enables 3D imaging of bone tissue at the cellular length scale and has been shown to be an appropriate tool for investigating 3D lacunar morphology [40][41][42][43]. At the sub-micron resolution, SR mCT enables 3D imaging of the LCN with a large field of view [44,45]. Synchrotron X-ray nano-CT with phase contrast, which provides sensitivity to the mass density variations that is several orders of magnitude higher than conventional attenuation contrast SRmCT, was recently used to investigate the bone LCN and the 3D collagen orientation at the nanometer length-scale [46,47].
However, very limited data is currently available at the submicron length scale for human jaw bone, both in terms of the distribution of osteocyte lacunae and in terms of mass density distribution [48].   In the present study, we investigated the differences in lacunar morphology and peri-lacunar tissue properties at the submicrometer length scale in human jaw bone tissue samples obtained from both healthy subjects and patients suffering from BRONJ. Additionally, cortical bone samples collected from the femur and tibia of donors not treated with BP were analyzed and compared to BRONJ and healthy control jaw bone samples. All investigations were based on SR mCT using phase contrast images measured with a 350 nm isotropic pixel size. Coupled with phase retrieval, this simultaneously provides information about the 3D distribution of the osteocyte lacunae and the local mass density of bone [46]. In order to quantify the mass density distribution, which is closely related to the bone mineral density distribution (BMDD), we used indices similar to those introduced for BMDD [13]. We hypothesized that (i) the bone turnover of the healthy human jaw bone, assessed by analyzing the mass density distribution, would be increased in comparison to the other anatomical sites and (ii) that extracellular matrix density and lacunar volumes of samples originating from patients suffering from BRONJ would be altered in order to compensate for the mineral homeostasis disturbed by the inhibited osteoclast activity. We therefore reported and compared the osteocyte lacunae volume distribution and the spatial arrangement of lacunae, as well as descriptors for the mass density distribution of the perilacunar tissue.

Ethics Statement
Ethical approval for the jaw samples was granted by the Ä rztekammer Bremen (Studien-Nr. 310). All donors signed an informed consent form. The present study also involved the use of cadaver specimens.
Ethical approval for the femur samples was granted by the Ethical Commission of the Medical University of Vienna, see [49]. Ethical approval for the tibia was granted by the University of California, San Francisco Committee on Human Research, see [50].

Specimen Preparation
Nineteen human jaw bone sections (blocks of about 1-3 mm 3 in size) were extracted from 12 female and 7 male donors, of whom 8 female and 2 male donors were suffering from BRONJ. The healthy control samples were from debris obtained during tooth removal. The BRONJ samples were obtained from surgeries necessary for the treatment of the necrosis. Furthermore, 7 cadaver specimens originating from the human femoral midshaft and 3 cadaver specimens originating from the human tibia midshaft of other donors were included in the present study. Detailed information on gender, donor age, anatomical origin, and BP treatment can be found in Table 1.
The femora were collected and prepared as described previously [49]. After removal, the jaw bone sections were embedded in Tissue-TekH O.C.T. TM (Sakura Fintec Europe B.V., Alphen aan den Rijn, Netherlands) solution and stored frozen at 220uC until further processing. Following thawing, the specimens were drilled down to a diameter of about 0.5 mm using a high precision lathe [51]. For the BRONJ samples, this sub-volume was selected from a region in which no necrotic tissue had been observed. The cut bone samples were fixed in 70% ethanol for transport. Tibia bone cores were machined with a coring tool and precision circular saw to a length of 4 mm and diameter of 4 mm, similar to the sample preparation described in [50]. Then, to fit the field of view of the imaging setup, the tibia samples were cut from cortical bone (diameter: 500 mm, height: 1 mm) using a high precision drilling machine. About 12 hours before imaging, the samples were placed inside the measurement hutch in order to allow adjustment to humidity and temperature.

Synchrotron Radiation Phase Contrast mCT
The SR mCT data were obtained at ESRF (European Synchrotron Radiation Facility, Grenoble, France) at beamline ID22NI. The X-ray beam was focused using Kirkpatrick-Baez reflective optics [46]. The scans used in the present study were performed for each sample by collecting 1201 projections, each with 0.2 s exposure time, over a total range of 360u. The energy was set to 16.874 keV, and the sample-detector distance was 282 mm, resulting in a (350 nm) 3 isotropic voxel size in the reconstructed image. Due to the coherence of the synchrotron source, the intensity of the recorded radiograph includes phase contrast [52,53]. Reconstruction was performed using Paganin's method [54], coupled to the conventional filtered back projection algorithm.
In the Paganin method, the phase is retrieved by simply assuming a linear relationship between the absorption index (b) and the refractive index decrement (d). For cortical bone (ICRU-44), the d/b (delta/beta) ratio at the given energy was set to 199 based on the XOP software [55]. The high ratio of delta/beta demonstrates the higher sensitivity for imaging the phase (d) compared to imaging the attenuation (b). The reconstructed 3D image made of 2048 3 voxels corresponds to a map of the refractive indices stored in units of 2p/l, with l being the wavelength of the X-ray beam (here l = 0.0735 nm). This map is linearly related to mass density [46] which was shown to be associated with the degree of mineralization [56]. The spatial resolution with these settings allowed an easy distinction of osteocyte lacunae and larger pores from the mineralized tissue matrix (Fig. 1A-C). However, the canalicular network could be not resolved.

Image Segmentation
In order to segment osteocyte lacunae inside the bone tissue volume (BV), the histogram of the whole 3D image was computed ( Fig. 2A). A threshold was determined for each image using the multi-class Otsu's method in the open-source software ITK (Kitware) [57] to separate mineralized tissue from non-mineralized pores. The resulting binary image was then labeled using a 3D connected component (CC) analysis method [41]. Objects smaller than 50 mm 3 or larger than 1000 mm 3 were considered not to be lacunae and were excluded from further analysis (Fig. 3).

Extraction of Quantitative Parameters
The 3D image of each sample was virtually divided along the sample length into 3 equal-sized volumes of interest (VOI). The lacunae connected to the border, which could be truncated, were removed to avoid including bias in the analysis. Bone volume (BV) was considered as the entire mineralized tissue excluding osteocyte lacunae and other pores. The lacunae were segmented as described above and their volumes were computed for each VOI. The median (Lc.V med ) and the variance (Lc.V var ) were extracted from the histogram of the lacunar volumes. Lacunar porosity was derived as the ratio of the total volume of all lacunae to the bone volume (Lc.TV/BV) and the lacun MDD r ae density was defined as the number of lacunae per bone volume (N.Lc/ BV). Furthermore, for each VOI, the distance within which 50% of the mineralized bone tissue is located with respect to the closest lacunar surface (Lc.Dist 50 ) was computed as the median of the Euclidean distance transform of the bone tissue [43].
Additionally, we used the reconstructed complex refractive index distribution, which is linearly related to the mass density, to compute the apparent mass density distribution (MDD r ) of each sample (Fig. 2B). The MDD r was calculated within the BV domain and was normalized by its area under the curve. Since the reconstructed complex refractive index might be biased due to the constant delta over beta ratio used in the Paganin phase retrieval [58], absolute values of mass density could not be retrieved and hereinafter the superscript r denotes relative values for all mass density parameters.
However, since the different bone samples can be considered comparable in terms of size and composition, this allows quantitative comparison of the relative difference in mass density between the bone samples.
Following the well-established approach for the description of BMDD by Roschger et al. [13], five parameters were extracted from the MDD r , i.e. MDD r Mean (the mean relative mass density within the evaluated distribution), MDD r Peak (the most frequent relative mass density value), MDD r Low and MDD r High (the 0.5 th and 99.5 th percentiles), and MDD r FWHM (the full width at half  maxima of the distribution). The threshold of P = 0.005 was arbitrarily chosen and is a compromise between maintaining good sensitivity for low and high values in the MDD r and minimizing potential artifacts using the partial volume effect for MDD r Low : Eventually, the vascular porosity was estimated after coarsening the segmented volumes by a factor of five and cleaning the volumes from objects smaller than 1600 voxel (8575 mm 3 ). The canal volume (Ca.V) was quantified using voxel counting and canal surface (Ca.S) was determined from the number of voxel located within one voxel Euclidean distance to the pore boundary. The following parameters were quantified: ratio of canal volume to bone volume (Ca.V/BV), ratio of canal surface to bone volume (Ca.S/BV) and ratio of canal surface to canal volume (Ca.S/ Ca.V).
All post-processing was done using MATLAB 2012a (The MathWorks Inc., Natick, MA, USA).

Statistical Testing
All statistical analyses were performed using the statistics toolbox in MATLAB. The normality of the distributions of each investigated parameter was determined by the Jarque-Bera test [59]. Differences with respect to anatomical sites, healthy and BRONJ groups were assessed by analyses of variance (ANOVA), followed by post hoc multiple comparison Bonferroni tests. The sample size did not allow a robust analysis of the effects of age and gender. All statistical results were considered significant for p, 0.05.

Results
Cross sectional mCT images and the corresponding segmented lacunar areas of a jaw bone sample are shown in Fig. 1. A 3D volume rendering of the osteocyte lacunae with subsections of one jaw and a femur sample image are shown in Fig. 3. These representative images exhibit distinct differences in the distribution and alignment of the lacunae in the different anatomical sites. The relative mass density histogram of the jaw specimen shown in Fig. 1 is shown in Fig. 2. The normalized histograms of the three adjacent sub-volumes of the same specimen in Fig. 4 illustrate the local intra-sample variability of MDD r : The average properties for the three different anatomical sites and the BRONJ samples are summarized in Table 2 and Table 3.

Differences between Anatomical Sites
The average lacunar volume ranged from 194 mm 3 in the tibia samples to 277 mm 3 in the jaw bone. Both median values and variance (Figs. 5A-B) were significantly higher in the jaw bone compared to the other anatomical sites. Distinct distributions of lacunar volume between samples from the jaw bone and those obtained from the peripheral skeletal sites are also illustrated in Fig. 6, which shows comparable, almost normal distributions for tibia and femur samples, but a remarkable asymmetry towards high volumes in the jaw bone samples. The average distance, in which 50% of the tissue matrix with respect to the closest lacunae is located (Lc.Dist 50 ), was highest in the femur, followed by the jaw bones and lowest in the tibiae (Fig. 5E). However, the standard deviation in jaw bone was considerably higher than in the other skeletal sites. For all investigated sites the average lacunar density (N.Lc/BV) was found to be larger than or equal to 20000 mm 23 (Fig. 5D). Variations between the anatomical sites reflected those observed for Lc.Dist 50 , i.e. the lacunar density was highest in the tibia and lowest in the femur. The lacunar density in the jaw bones was between those of the other two sites.
Mean, low, and high values of the relative mass density distribution were significantly higher in the tibia compared to jaw and femur (Figs. 5H-J). In contrast, the heterogeneity of MDD r within the evaluated sub-volumes, as expressed by MDD r FWHM , was higher in the jaw bone than in tibia samples.
The average and standard deviation of Ca.V/BV, Ca.S/BV and Ca.S/Ca.V of all sections were found to be (765) %, (0.00460.003) mm 21 , and (0.0660.03) mm 21 , respectively. All values are summarized in Table 3. ANOVA revealed no significant differences between the different anatomical sites or between jaw sections from healthy donors and BRONJ.

Differences between BRONJ and Control Jaw Bones
Significantly lower lacunar densities (F = 5.1, p,.028) were observed in the BRONJ sample group ( Table 2). The lower lacunar density was associated with higher Lc.Dist 50 values (F = 6.7, p,.012). It should be noted that the values observed in the BRONJ samples are still within the range observed at other skeletal sites (see Table 2). All other evaluated parameters were statistically not significantly different when comparing BRONJ samples and healthy jaw bone controls. Most strikingly, the   lacunar volume distributions of BRONJ and control jaw bone samples were almost identical (Fig. 7).

Discussion
The present study aimed to determine the potential impact of alterations of the osteocyte lacunar network and mass density of the extracellular matrix, in the event of osteonecrosis after BP treatment. To address this question, we used synchrotron radiation phase micro-tomography with a 350 nm voxel size. We analyzed human cortical bone specimens extracted from the mandibular jaw of 10 patients suffering from BRONJ and of 9 healthy persons for control. In addition, we investigated samples from anatomical sites in which BP treatment does not usually induce necrosis, such as the tibia and femur.
The imaging technique used allows the investigation of relatively large sample volumes in 3D, without the necessity for demineralization or any other tissue preparation steps (except for ethanol fixation and drying), and combines a large field of view, a very high spatial resolution and a high signal-to-noise ratio with a good sensitivity to mass density fluctuations [52,60]. Although the absolute mass density could not be derived, the derived relative  mass density distributions (MDD r ) enabled a quantitative comparison between the evaluated samples.
Our results suggest that the osteocyte lacunar number, volume and distribution, as well as the mass density in the extracellular matrix, are closely linked to the anatomical site. While there are several studies reporting the human lacunar density based on 2D imaging modalities [61,62] there are only a few reporting lacunar density in 3D, and these are from attenuation contrast tomography [43] at a lower resolution (1.47 mm voxel size) [40,42]. The lacunar density values we found in the femoral samples are consistent with those we reported recently [43] and with those reported by Carter et al. [40] for the same anatomical region in women across their lifespan. In another study, significant variations were observed between anterior-posterior and mediallateral regions with differences of up to 30% between the regions in single individuals, but no significant impact of age on N.Lc/BV [40]. Moreover, they found a significant decrease in the lacunar volume with respect to donor age (R 2 = .46), whereas the lacunae from the younger group were ,30% larger (age: ,50 years) than those of the older group (age: .50 years). In contrast, no impact of age on lacunar size was reported by others [27].
We observed pronounced differences between jaw bones in comparison to the two peripheral skeletal bone sites. This may be explained by the different origin of the cells and the different remodeling rates in those tissues, or the different mechanical environment. While osteocytes in the peripheral skeleton derive from the trunk lateral plate mesoderm, in the mandible they derive from the paraxial mesoderm [63,64]. The remodeling rate in jaw bone is believed to be higher than in femur and tibia [11,65], which leads to lower mineralization and mass density [12] in the jaw. The latter has been confirmed only partly in our study by means of lower distribution values of MDD r (mean, low and high) and a higher intra-specimen variability (in terms of FWHM) in the jaw bones compared to tibia (Fig. 5F-J). However, we observe a trend towards smaller values for jaw compared to femur (mean, low and high), which is in line with the lower remodeling rate in the femur reported earlier [66].
The majority of the osteocyte lacunar volumes were in the range between 100 and 400 mm 3 , which is in agreement with values reported in other studies [40,42,43]. However, a small fraction of very large osteocyte lacunae with volumes between 400 and 1000 mm 3 was found in the jaw bone samples (Fig. 6). Most strikingly, this characteristic asymmetric distribution of osteocyte lacunar volume was almost identical in healthy jaw bone samples and those treated with BP (Fig. 7). The only significant effect of the BP treatment in jaw bone appeared to be the reduction of lacunar density, which was also evident in the increased average distance between the extracellular matrix and the closest osteocyte lacuna. On average, the samples treated with BP had a lacunar density 14.8% lower than the healthy jaw bone controls, but still 15% higher than the femur samples. Mass density (mean, low, and high values) was not significantly higher in the jaws treated with BP in comparison to the healthy controls, although these values were marginally higher and the heterogeneity (FWHM) was slightly lower than those in the control group. Therefore, it must be concluded that no remarkable changes of mass density occur in the extracellular matrix after BP treatment. For comparison, the mass density distribution values of the jaw bones treated with BP do not reach the values observed in tibia and femur samples. However, with the current analysis we cannot exclude potential local changes, for example directly at the pore tissue boundary. It is well accepted that osteocytes actively remodel their direct extracellular environment [35] and it has been shown by fluorescence-labeled risedronate injected intravenously in a murine model that BP is indeed deposited around the osteocyte lacunae [67]. Interestingly, the results of that study have suggested that the deposition rate was not uniform in all osteocytes, but decreased along with the distance to the next vascular channel. Therefore, a dedicated regional analysis of the lacunar-matrix interface merits further investigation. Overall, except for the reduction in osteocyte lacunar number, the changes of the lacunar network and the mass density of the extracellular matrix appeared marginal. In particular, the BP-induced parameter alterations in the jaw stayed within the variations observed between different anatomical sites. Consequently, the observed parameter variations are not likely to be the primary causes for the development of BRONJ in the jaw.
In this study, we have not investigated the lacunar occupation rate and viability of osteocytes, as the increased number of abandoned osteocyte lacunae is a well-documented sign of BP treatment in the jaw [64,68]. Moreover, the filling of empty lacunae with mineral has been reported in several studies [69][70][71], which is in line with the reduced lacunar density we observed in the jaw bones treated with BP. It is known that BPs have a high affinity to hydroxyapatite [63]. Therefore, we hypothesize that higher perfusion and turnover rates initially lead to a higher deposition of BP in the human jaw compared to other human sites, which is in line with the findings from previous studies in animals [11]. At higher doses BP becomes cytotoxic [72], which presumably promotes the gradual depletion of osteocytes and allows the abandoned pores to be filled with more BP-loaded mineral. Finally, the acidification of this tissue, caused for example by an inflammation, which is known to occur in BRONJ, can result in an excessive dissolution of BP-loaded mineral and a release of toxic doses of BP. Among the multifold biological factors promoting the development of necrosis in the jaw, the fraction of large lacunae (.400 mm 3 ) found in jaw bone tissue but not in the tissues from the other skeletal sites may be a structural factor, since it allows the deposition of larger amounts of highly concentrated BP-loaded mineral in such abandoned lacunae, which when washed out results in higher and thus more toxic local BP doses.
We did not observe significant differences in vessel porosities between the different groups, which supports the idea that the observed differences in lacunar properties and mass density distributions between the different groups are not linked to the potential differences in vessel porosities in the investigated specimens. The reported pore-volume ratios are within the range previously shown for femoral cortical bone [43]. However, the quantification of vessel porosities was limited in our study, due to the size of the field of view, which was small in comparison to the average distance between individual vessels. Therefore, the field of view cannot be considered to be a representative volume with respect to the vessel network.
This study has several limitations. One drawback of the present study is that the comparison between sites and even within the jaw bone is hampered by our limited ability to control the exact anatomical location and orientation of the harvested samples and the underlying diagnosis. Moreover, the samples from femur, tibia, and jaw bones could not be collected from the same donors, and the duration and type of BP treatment was not uniform within the BRONJ group. Potential influencing factors, such as intraspecimen variability, age and gender, as well as the BP treatment conditions, could have biased our analysis. Additionally, in contrast to previous investigations in which we have demonstrated the feasibility to extract absolute mass density values from the phase contrast images [46,47], we report relative values in this study. This is due to the fact that (i) the large distance between sample and detector in the current configuration violated the nearfield condition, which is an essential prerequisite for the Paganin-phase retrieval used [58] and (ii) we used a constant ratio of delta over beta for the Paganin retrieval, even though it could have been different between samples with different degrees of mineralization [73].
Additionally, since we observed a bimodal distribution of the apparent mass densities (Fig. 4), and the samples were comparable small in terms of being representative for the ratio of interstitial to osteonal tissue, MDD r Peak may not be an appropriate parameter to quantify MDD r : Nevertheless, the subtle ultra-structural alterations observed in the jaws treated with BP underline the need for further sophisticated investigations of large tissue volume with sub-micron resolution, high sensitivity to local changes in mineral density and chemical composition of the tissue. Such changes are not likely to be depicted by conventional X-ray methods. Phase contrast tomography with voxel sizes ranging down to about 50 nm is now available at SR sources and may provide new hints towards the ultra-structural mechanisms leading to the pathogenesis of BRONJ.