In osteoarthritis (OA), changes occur in both cartilage and subchondral bone. The subchondral bone plate facilitates normal cross-talk between articular cartilage and trabecular subchondral bone, and adaptive changes in the plate due to OA may therefore disturb cross-talk homeostasis. To investigate these changes over time, we examined the cartilage–subchondral bone interface using a combined approach of histologic analysis and in vivo microfocal computed tomography.
Sixteen-week-old male C57BL/6 mice (n = 32) received intraarticular injections of collagenase in 1 joint to induce instability-related OA and received saline injections in the contralateral knee joint (control joint). At 2, 4, 6, 10, and 14 weeks after injection, changes in the tibial subchondral bone plate and subchondral trabeculae were analyzed.
Two weeks after injection, collagenase-injected joints had significantly more cartilage damage and osteophytosis than did control joints. Osteoclast activity directly underneath the subchondral bone plate was significantly elevated in collagenase-injected joints compared to control joints (mean ± SEM osteoclast surface/bone surface 11.07 ± 0.79% versus 7.60 ± 0.81%), causing the plate to become thinner and creating a large increase in subchondral bone plate porosity (mean ± SEM cumulative porosity volume 0.05 ± 0.04 × 10−3 mm3 in control joints versus 2.52 ± 0.69 × 10−3 mm3 in collagenase-injected joints). Four weeks after injection, the previously formed perforations disappeared, coinciding with a significant rise in osteoblast activity in the subchondral trabecular bone in collagenase-injected joints compared to control joints (mean ± SEM bone formation rate/bone surface 0.62 ± 0.13 μm3/μm2 per day versus 0.30 ± 0.03 μm3/μm2 per day).
The current study is the first to provide quantitative longitudinal data on the dynamic changes in the subchondral bone plate after OA induction. The development of plate perforations may enhance mutual interaction between subchondral trabeculae, bone marrow cells, and the articular cartilage in OA.
In osteoarthritis (OA), changes in bone are thought to accompany cartilage deterioration, although it remains unclear which process is responsible for the initial homeostatic disturbance. Located directly underneath the articular cartilage, the subchondral bone plate provides structural support and acts as a portal for biochemical interactions between the cartilage layer and bone marrow and/or subchondral trabecular bone (1–5). Therefore, any changes in its structure may well have an effect on the overlying cartilage as well as on the underlying subchondral bone. In OA, subchondral bone is hypomineralized and of inferior quality as a consequence of abnormally high turnover (6–10). The exact cause of this increased bone turnover is as yet unknown, but the involvement of changes in the overlying articular cartilage is proposed as the principal cause (11–13). In contrast, other studies have shown that the OA process starts with an increase in subchondral bone turnover, which in turn, initiates cartilage damage (12–15). Whichever theory is true, extensive interaction between cartilage and subchondral bone is likely to occur. In 2003, Burr and Radin proposed that microcracks in the subchondral bone plate or calcified cartilage may initiate targeted remodeling of the subchondral bone, accounting for the increased turnover and reduced material density of the subchondral plate (16). In addition, an increase in subchondral bone plate porosity has been described in OA patients as well as in animal models of OA (17, 18), but it is unclear if this increase in porosity also leads to increased interaction between cartilage and subchondral bone.
To study changes in the subchondral bone, the use of microfocal computed tomography (micro-CT) imaging has become an invaluable tool (19, 20). In contrast to histologic assessment, micro-CT enables 3-dimensional (3-D) monitoring and quantification of changes in dense tissues such as bone. Over the last few years, the use of in vivo micro-CT has become the new standard (21). This technique allows longitudinal followup within 1 animal and enables better discrimination of cause and consequence. Nevertheless, relatively few studies have looked at in vivo changes in the subchondral bone plate (22, 23). Therefore, in this study, we examined changes in the subchondral bone plate with the development of OA, using a combined approach of histologic analysis and in vivo micro-CT.
MATERIALS AND METHODS
Thirty-two 16-week-old male C57BL/6 mice (Harlan) were used. They were maintained on a 12-hour light/12-hour dark cycle, housed in individually ventilated cages with 4 littermates per cage, and fed ad libitum. Body weight of all animals was measured at weekly intervals throughout the experiment. For induction of OA, we used the collagenase OA model as described previously (24–26). On day 0 of the experiment, mice were anesthetized using a 5% isoflurane/N2O/O2 mixture, and intraarticular injections of either 1 unit of purified bacterial type VII collagenase (Sigma) dissolved in 6 μl saline (right knee joints) or 6 μl saline only (left knee joints; hereinafter called control left knee joints) were given using a syringe with a 33-gauge needle. The animals received pain medication subcutaneously while anesthetized (buprenorphine hydrochloride; 0.01 mg/kg body weight). All injections were repeated on day 1.
Animals were divided into 3 cross-sectional (cross) groups, termed cross/0–2, cross/0–4, and cross/0–14, with numbers referring to followup time (in weeks), and 1 longitudinal (lt) group (lt/0–14), with 8 mice per group. Sixteen mice (cross/0–2, n = 4; cross/0–4, n = 4; and cross/0–14, n = 8) received an intraperitoneal calcein injection (0.01 mg/gm body weight; Sigma) 8 days and 1 day before they were killed. After the mice were killed, their hind legs were excised and fixed in 4% buffered formalin. All procedures were performed in accordance with applicable federal guidelines and institutional policies.
To monitor bone changes, we used a SkyScan 1076 in vivo x-ray microtomograph. Prior to scanning, each animal was anesthetized, and a small synthetic foam block was mounted between its hind legs to clear the scan area of surrounding soft tissues (Figure 1A). The mouse was placed supine in a polystyrene bed. During scanning, the x-ray beam was directed at the proximal tibia, scanning an area 17 mm in length and 35 mm in width, while leaving the rest of the animal shielded from radiation. The following settings were used: voxel size 9 μm, voltage 40 keV, current 250 μA, exposure time 2,356 msec, frame averaging 2, beam filtration filter 1 mm aluminum. Data were recorded every 1.5-degree rotation step through 180 degrees. Left and right legs were scanned in the same scan session. The total time per scan was 21 minutes; the hind legs were irradiated for a total of 11 of these minutes due to the use of a shutter that shielded the beam between recordings.
Mice in groups cross/0–2, cross/0–4, and cross/0–14 were scanned at time zero (right before OA induction) and immediately after they were killed 2 weeks (cross/0–2), 4 weeks (cross/0–4), and 14 weeks (cross/0–14) after OA induction. Mice in group lt/0–14 were scanned at time zero and at weeks 2, 4, 6, 10, and 14 after OA induction (Figure 1B). After mice were killed, all excised hind limbs were scanned again using an extended ex vivo scanning protocol with improved signal-to-noise ratio, where the rotation angle was lowered to 0.8 degrees, and frame averaging was increased to 3. All other settings remained unchanged. Reconstruction was performed using SkyScan NRecon software, version 1.4.3. Resulting grayscale cross sections were composed of isotropic voxels measuring exactly 8.88 μm in all dimensions.
Radiation dose per scan was measured with a Solidose 400 dosimeter equipped with a DCT16-77 ionization chamber probe (RTI Electronics). The received radiation dose was 0.08 Gy per minute using the in vivo scan settings. This dose rate needs to be corrected for tissue depth since part of the radiation dose will be absorbed by the skin and overlying tissues. SkyScan recommends a tissue correction factor of 0.656 for the mouse hind leg, which means that the hind legs of each animal received 11 minutes × 0.08 × 0.656 = 0.58 Gy per scan.
Quantification of bone morphometric parameters.
Each knee joint data set, consisting of reconstructed cross sections, was aligned to the same spatial orientation. Scans were segmented with a local threshold algorithm (27). Using 3-D data analysis software (CTAnalyzer; SkyScan), the proximal tibia was selected for further analysis (Figure 1A). The cortex (including the subchondral bone plate) and trabeculae of the proximal tibia were separated using software developed in house (available upon request from the corresponding author). By applying subsequent steps of image dilation and erosion on the binary data sets, the software extracts different anatomic compartments of the proximal tibia: subchondral bone plate, subchondral and metaphyseal trabeculae, bone marrow, and perforations inside the subchondral bone plate. Perforations were defined as tunnels through the subchondral bone plate, enabling contact between the articular cartilage and the subchondral bone marrow. The different compartments were analyzed with the freely available software package 3D-Calculator (http://www.erasmusmc.nl/47460/386156/Downloads) using a binary selection feature.
To measure subchondral bone plate thickness (in μm), we selected medial and lateral subsections of the weight-bearing region of the subchondral bone plate measuring 0.50 mm mediolateral in width and 1.15 mm ventrodorsal in length (Figure 1A). For reasons of clarity, medial and lateral thickness values were averaged after measuring. Total perforation volume inside the subchondral bone plate was measured in subsections expanded to 1.00 mm mediolateral width, and medial and lateral volumes of each cross section were summed.
For subchondral trabeculae (Figure 1A), the following 3-D morphometric parameters were calculated: trabecular bone volume fraction (BV/TV), representing the ratio of trabecular bone volume (BV; in mm3) to endocortical tissue volume (TV; in mm3); trabecular thickness (TbTh; in μm); trabecular separation (TbSp; in μm); and connectivity density (in mm−3), indicating the number of trabecular connections per unit volume (28). Medial and lateral values were averaged. In order to evaluate whether bone changes also occurred at distal sites in the mouse hind leg due to the collagenase injection, BV/TV of the calcaneus (scanned ex vivo) was measured in mice killed 4 and 14 weeks after saline or collagenase injection.
To visualize the subchondral bone changes that took place within 1 animal, scans derived from 1 animal at time zero and at 4 and 14 weeks were spatially matched by rotation and translation using the registration software M.I.R.I.T. (29). This software uses an optimization criterion based on maximizing mutual information. The resulting rotated data sets were used to create overlaps between different time points after injection, and differences were highlighted by color coding. All 3-D images were made using the software program ANT (SkyScan).
Left and right hind limbs of groups cross/0–2 (n = 4), cross/0–4 (n = 4), and lt/0–14 (n = 8) were decalcified in 10% EDTA/5% paraformaldehyde for 14 days and embedded in paraffin. Hind limbs of animals that received calcein injections remained undecalcified and were embedded in methylmethacrylate.
Coronal (frontal) histologic sections (6-μm thick) were taken through the joint at 100-μm intervals. Paraffin sections were stained with Safranin O–fast green. Cartilage damage was scored at the medial and lateral tibial plateau in 3 histologic sections per joint using the semiquantitative grading and staging system devised by the Osteoarthritis Research Society International Working Group (30). Medial and lateral damage values were summed and subsequently averaged for the 3 sections, yielding a maximum obtainable cartilage damage score of 48. Loss of proteoglycans (stained by Safranin O) was scored using a 0–3-point scale, where 0 represents no loss of proteoglycans and 3 indicates complete loss of staining for proteoglycans in more than half of the cartilage layer. Osteophyte presence was scored according to the 4-point scoring system described by Kamekura et al (31), where 0 = no osteophytes, 1 = formation of cartilage-like tissues, 2 = increase in cartilaginous matrix, and 3 = endochondral ossification. Finally, osteoclasts were visualized using histochemical staining for tartrate-resistant acid phosphatase activity as described previously (32), using Gill's hematoxylin as counterstain. Osteoclast quantification included osteoclast surface/bone surface (OcS/BS) and osteoclast number/bone surface (OcN/BS) using Bioquant Osteo software, version 7.20.
Methylmethacrylate-embedded sections were photographed in brightfield and fluorescent light to visualize the outline of the bone and calcein labels, respectively. These photomicrographs were merged in Paint Shop Pro software, version 7.02 (Jasc Software) (i.e., the brightfield photomicrograph was made 70% transparent and added as layer to the fluorescent photomicrograph). The whole surface area of epiphyseal trabecular bone was analyzed, and single and double labels were manually traced using Bioquant Osteo software, version 7.20. Mineralized surface/bone surface (MS/BS; in %), mineral apposition rate (MAR; in μm/day), and bone formation rate/bone surface (BFR/BS; in μm3/μm2 per day) were then calculated.
Results were statistically analyzed with GraphPad Prism software, version 5.02. For the cartilage damage, osteophytosis, osteoblast, and osteoclast data, a repeated-measures two-way analysis of variance (ANOVA) was applied with treatment (intraarticular injection with saline or collagenase) and time as factors and Bonferroni posttests. For the micro-CT–derived morphometric data, a two-way ANOVA was used with treatment (intraarticular injection with saline or collagenase) and time as factors and Bonferroni posttests. When studying effects in time within a single group or treatment, a one-way ANOVA with Tukey's posttest was chosen. In all cases, P values less than 0.05 were considered significant. In all graphs, data are presented as the mean ± SEM.
Effect of the operation and scanning.
All animals were closely monitored up to 12 hours after the operation, corresponding to the therapeutic effective time- span of the applied analgesic, and since cage activity was normal, no further analgesia was applied. Although some weight loss was measured in the first week after surgery (2–6% of the initial body weight at time zero), this change was not significant, and all animals had normal weight gain thereafter. From 10 weeks after injection onward, the average body weights of both the cross/0–14 group (scanned 2 times, including once after mice were killed) and the lt/0–14 group (scanned 6 times, including once after mice were killed) were significantly higher than the body weights measured at time zero (data not shown). The bone volume fraction in the proximal tibia of control left knee joints did not differ significantly between cross/0–14 (mean ± SEM 12.6 ± 0.71%) and lt/0–14 (mean ± SEM 10.17 ± 1.04%) at 14 weeks, indicating that repeated scanning did not have a measurable deleterious effect on normal bone metabolism.
To assess whether the animals would preferentially unload one of their hind legs due to the operation, we analyzed the amount of trabecular bone in the calcaneus of collagenase-injected and contralateral control legs. In collagenase-injected legs, the mean ± SEM BV/TV was 104.9 ± 6.3%, 98.6 ± 5.2%, and 96.0 ± 3.8% of that in contralateral control legs and the mean ± SEM TbTh was 95.7 ± 2.7%, 99.5 ± 1.9%, and 96.9 ± 4.3% of that in contralateral control legs at 2, 4, and 14 weeks, respectively, after collagenase injection. No significant differences were found between collagenase-injected and contralateral control joints at any time point.
OA induction and subchondral bone changes.
Two weeks after OA induction, cartilage damage was significantly higher in collagenase-injected knee joints than in contralateral control joints (Figure 2A). Mineralized osteophytes were also observed after 2 weeks in collagenase-injected joints (Figure 2A). Reduction of Safranin O staining, indicative of proteoglycan loss, was significant at 4 weeks after OA induction, but not at 2 or 14 weeks (data not shown).
In the subchondral volumes of interest of collagenase-injected knee joints, bone loss was observed. Compared to contralateral control knee joints, the thickness of the subchondral bone plate in the collagenase-injected knee joints decreased from 2 weeks after injection onward (Figure 2B), indicating increased osteoclastic activity. This initial thickness decrease was temporary, since at 10 weeks after injection, the bone plate returned to its initial thickness, although the average plate thickness measured in the contralateral control joints (which also increased in time) could not be equaled.
When we quantified the presence of osteoclasts located directly underneath the subchondral bone plate (Figure 2C), we found similar time-dependent dynamics in OcS/BS and OcN/BS in saline- and collagenase-injected knees. A significant increase in OcS/BS was detected in collagenase-injected knee joints only during the first period after injection (at 2 weeks) (Figure 2C). The OcN/BS was elevated as well, although this difference did not reach significance (P = 0.09) (Figure 2C).
In the trabecular bone underlying the subchondral bone plate, similar temporal bone loss patterns were found, as assessed by BV/TV (P < 0.05 between curves) (Figure 3A) and TbTh (P < 0.01 between curves) (data not shown), whereas TbSp was significantly elevated (P = 0.001 between curves) (Figure 3A) in collagenase-injected knee joints compared to contralateral control joints. As previously found for the subchondral bone plate, the initial decrease in BV/TV was followed by an increase, indicating a rise in osteoblastic activity. This was confirmed by the calcein labeling, which showed increased osteoblast activity in the subchondral trabecular bone of collagenase-injected knee joints at 4 weeks (Figure 3B). We observed significant increases in MS/BS and BFR/BS (Figure 3C) as well as in MAR (data not shown) in collagenase-injected knee joints compared to contralateral control joints. At 14 weeks, these parameters also increased significantly in control knee joints (Figure 3C), indicating normal growth and providing an explanation for the temporal increase in subchondral bone plate thickness in controls (see Figure 2B).
The changes in subchondral bone (i.e., subchondral bone plate and subchondral trabeculae) within the same anatomic location in 1 animal are shown in Figure 4A. To further visualize the bone remodeling process, binarized grayscale images derived from time zero, 2 weeks, and 14 weeks were overlapped, and the differences in the amounts of bone between the time points were visualized (Figure 4B). It became clear that at 2 weeks, primarily bone loss was observed at both the medial and lateral sides of the tibial plateau, whereas at 14 weeks, bone gain was more prominent.
Since we observed that subchondral bone plate thinning was one of the main outcomes of the experiment, we scrutinized the micro-CT data of the subchondral bone plate. This analysis demonstrated that at multiple locations, the subchondral bone plate had become so thin that perforations had started to appear, forming connecting tunnels between the subchondral trabecular bone/bone marrow and the articular cartilage (Figures 5A and B). To exclude the possibility that the perforations were due to higher noise levels in the in vivo scans, we checked whether the same perforations could also be found in scans made with the ex vivo scan protocol (which had a higher signal-to-noise ratio), and this proved to be the case (Figure 5A, right). In addition, we identified the perforations by histologic analysis (Figure 5C). Subchondral bone plate porosity was observed in both saline-injected and collagenase-injected knee joints, but the number and size of the perforations were larger in the collagenase-injected knee joints (Figure 6A). The perforation volume of the tibial plateau differed significantly between collagenase-injected and contralateral control joints (0.05 ± 0.04 × 10−3 mm3 in control joints versus 2.52 ± 0.69 × 10−3 mm3 in collagenase-injected joints) despite some temporal variation between animals (Figure 6B).
This study is the first to show quantitative longitudinal data on the formation of perforations in the subchondral bone plate in an instability-induced model of OA appearing within 2 weeks after induction. Increased osteoclast activity acting locally underneath the subchondral bone plate is the most likely cause of the formation of these perforations in the initial phase after inducing instability, followed by repair by increased osteoblast activity. These data show the high temporal dynamics in subchondral bone plate metabolism in OA. Finally, we demonstrated that in vivo micro-CT is sensitive enough to quantify the anatomic changes following altered osteoblast/osteoclast activities, without disturbing normal bone metabolism. Besides generating statistically powerful data, this technique also helps to reduce the number of experimental animals needed.
The dynamic changes in OA subchondral bone that we found in the present study match previously reported data. For instance, Hayami et al also found subchondral bone loss to occur in the rat knee within 2 weeks after surgical destabilization, followed by significant increases in subchondral bone volume relative to sham operation up to 10 weeks after surgery (33). Importantly, the authors observed increased vascular invasion into the calcified cartilage after 1 week, which coincided with the onset of cartilage surface fibrillation. Two in vivo micro-CT studies also found early bone loss to occur in the rat knee following destabilization (22) or after intraarticular injection of monosodium iodoacetate (23). Similar findings were observed in dogs (18), guinea pigs (34), rabbits (35), and cats (36).
Subchondral bone loss also occurs in OA patients, especially in progressive OA. Bettica et al found that levels of urinary markers of type I collagen degradation were significantly higher in patients with progressive knee OA, similar to levels in patients with osteoporosis (10), and Buckland-Wright et al concluded that risedronate was able to correct the subchondral trabecular bone loss found in progressive knee OA (37). Thus, although subchondral sclerosis is the final disease outcome, these data indicate that bone loss generally occurs in OA, which may explain why net bone loss is observed at early stages of the disease process. To confirm this in human patients is difficult, since clinical data on changes in the subchondral bone plate are generally derived from patients with established OA. This is a pity, since our current data justify in-depth analysis of the subchondral bone plate in case of (mild) joint symptoms. This may identify early stages of OA, which then may be modifiable or treatable.
Our findings that subchondral bone plate perforations are present in normal tibial plateaus and increased after inducing knee joint instability fit well with previous data. The first reports of holes penetrating the subchondral bone plate date from the 1980s, when Duncan et al (38) described numerous perforations through the subchondral plate in nonarthritic tibial plateaus, some of which extended to the bone marrow (38). Similar findings were described by Lyons et al (3) and Clark and Huber (39). The latter study estimated the prevalence of these perforations in the human tibial plateau to be <10/cm2. More recently, imaging studies showed that small molecules such as gadolinium (2) or sodium fluorescein (40), initially present in the bone marrow, are able to pass the subchondral bone plate and reach the articular cartilage. These studies showed that next to a mechanical interaction, biochemical interactions between bone and cartilage are also likely to occur. Finally, Hwang et al found an increasing density of enlarged subchondral canals penetrating the calcified cartilage in samples with partially eroded cartilage and linked this finding to increased hydraulic conductance of osteochondral tissue samples with progression of OA (41). Hwang et al also postulated that this increased hydraulic conductance of the subchondral bone plate may cause the articular cartilage to lose its high interstitial fluid pressurization (important to withstand joint loads) because of an abnormally high fluid exudation toward the subchondral bone.
From our current findings and data from literature discussed above, we would like to propose a concept of the consecutive events that take place in bone and cartilage in instability-induced OA in the murine knee joint. In this concept, we hypothesize that the increased perforation of the subchondral bone plate found in our current study leads to an increase in the mutual interaction between subchondral bone and cartilage.
At time zero, directly after inducing instability, mechanical stress is induced on the articular cartilage and underlying subchondral bone, causing an increase in osteoclast activity just below the articular cartilage. This causes a decrease in subchondral bone plate thickness in the first 2 weeks. Enlarged plate perforations are formed within the subchondral bone plate, increasing its permeability and causing a higher than normal fluid exudation toward the subchondral bone, leading to a net loss of fluid from the cartilage, which subsequently becomes damaged.
Meanwhile, in an attempt at repair, the damaged cartilage produces anabolic (growth) factors but also catabolic factors such as matrix metalloproteinases (42, 43) that reach the subchondral bone via the plate perforations, while factors derived from the subchondral bone may also travel toward the articular cartilage (12, 15). This will start a cascade of events in which the fine balance of bone buildup and breakdown becomes disturbed. In addition, osteocytes inside the subchondral bone plate, known to respond to fluid flow, may play a modulating role as well (44). Then, 4 weeks after the induction of instability, the functional coupling that exists between osteoclasts and osteoblasts causes a rise in osteoblast activity in the subchondral trabecular bone, decreasing the amount of perforations back to control levels and increasing the subchondral bone plate thickness. At this stage, osteoclast activity is suppressed. With more time progressing, osteoblast and osteoclast activity return to control levels, but the bone turnover is still in a disordered phase, and at the end of the followup period (at 14 weeks), the subchondral bone plate remains thinner compared to the contralateral controls.
A limitation of this study is the absence of a group of healthy animals. Therefore, we cannot rule out the possibility that some of the parameters measured in the contralateral control joints may be different compared to intact (i.e., nonoperated) joints (e.g., due to altered cage activity). However, recurrent monitoring of the mice did not indicate any changes in daily cage activities until the end of the followup period (14 weeks), and we therefore have no reason to believe that this might be the case. Also, we did not note progression of cartilage damage, which is commonly observed in humans. The cause of this is not yet clear, but it may be related to the OA model we used and/or it may be explained by the length of the followup period, which may still be too short. The fact that subchondral bone osteosclerosis was not observed in our experiment might be explained by the relatively early phase of the OA disease process; despite the presence of fully mineralized osteophytes, the amount of cartilage damage at 14 weeks was still mild, and full-thickness erosion of the articular cartilage layer was only sporadically observed. Finally, although we acknowledge the fact that anterior cruciate ligament transection or destabilization of the medial meniscus is currently more commonly used as a standard model for instability-induced OA, we believe that the collagenase model represents instability-induced OA as well. Since its establishment by Van der Kraan et al in 1990 (24), this model has been described in several reports and in multiple species (45–47). Our own reported data have demonstrated that the laxity characteristics of murine knee joints are indeed markedly increased after injection of bacterial collagenase (48).
In conclusion, we used a longitudinal approach in which subchondral bone changes in instability-induced OA were quantified using in vivo micro-CT combined with histologic analysis. We found that, as a result of increased osteoclast activity, subchondral bone loss occurred within 2 weeks after destabilization, resulting in increased subchondral bone plate perforation. These perforations directly connect 2 of the most affected tissues in OA, articular cartilage and subchondral bone, providing the possibility of increased cross-talk between cartilage and bone and contributing to disease progression. Finally, our findings may open up inspection of subchondral plate perforation as an early indicator of OA, and it will be a challenge to investigate this concept in a clinical study.
All authors were involved in drafting the article or revising it critically for important intellectual content, and all authors approved the final version to be published. Dr. van Leeuwen had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Study conception and design. Botter, van Osch, Waarsing, Weinans, van Leeuwen.
Acquisition of data. Botter, Clockaerts, van Leeuwen.
Analysis and interpretation of data. Botter, van Osch, Clockaerts, Weinans, van Leeuwen.
We kindly acknowledge Yvonne Sniekers for help with operating the in vivo micro-CT scanner, Nicole Kops for assistance with histology, and the personnel of the Experimental Animal Facility of the Erasmus Medical Center for taking care of the animals during the experiment.