Similarities and discrepancies in subchondral bone structure in two differently induced canine models of osteoarthritis



In osteoarthritis (OA), cartilage degradation is accompanied by subchondral bone changes. The pathogenesis and physiology of bone changes in OA are still unclear. The changes in subchondral bone architecture and cartilage damage were compared in differently induced experimental models of OA. Experimental OA was induced bilaterally by anterior cruciate ligament transection (ACLT) or by cartilage trauma (Groove model); bilateral sham surgery served as control. Lysylpyridinoline (LP, bone resorption) and C-telopeptide of type II collagen (CTX-II, cartilage breakdown) were measured over time. At 20 weeks after surgery, the subchondral cortical plate and trabecular bone of the tibia were analyzed by micro–computed tomography (µCT) and cartilage degeneration was analyzed histologically and biochemically. In both models, cartilage degeneration and cortical subchondral plate thinning were present. CTX-II levels were elevated over time in both models. Subchondral trabecular bone changes were observed only in the ACLT model, not in the Groove model. Correspondingly, LP levels were elevated over time in the ACLT model and not in the Groove model. Interestingly, the trabecular bone changes in the ACLT model were extended to the metaphyseal area. The early decrease in plate thickness, present in both models, as was cartilage damage, suggests that plate thinning is a phenomenon that is intrinsic to the process of OA independent of the cause/induction of OA. On the other hand, trabecular changes in subchondral and metaphyseal bone are not part of a common pathway of OA development and may be induced biomechanically in the destabilized and less loaded ACLT joint. © 2010 American Society for Bone and Mineral Research


Osteoarthritis (OA) often presents with pain and is accompanied by stiffness, crepitus, and swelling of the joint. Structural changes underlying these clinical features are damage of the articular cartilage, changes in the subchondral bone structure, osteophyte formation at the joint margins, and synovial inflammation. The etiology of OA is complex and includes various genetic, biochemical, and mechanical factors. In fact, the pathogenesis of OA is poorly understood.1, 2

Radiographs of osteoarthritic joints show an increased density of the subchondral bone, defined as subchondral sclerosis. Changes in architectural structure are suggested to be responsible for this increase in density. In established OA, studies of bone structure have shown that the subchondral cortical plate has thickened,3 and the volume of the trabecular bone has increased.4, 5 It is suggested that the increased stiffness of sclerotic bone in OA might play a role in the progression of cartilage degeneration.6 Whether these subchondral bone changes observed in established OA precede, occur simultaneously with, or are the result of cartilage degeneration is still subject of discussion.3, 7 Studies in human end-stage hip OA showed both cartilage deterioration without trabecular bone changes4 and bone changes without local cartilage changes.5 Unfortunately, in human studies (in vivo or ex vivo), mostly established (severe) OA is studied, whereas longitudinal data on subchondral bone changes from onset until full blown clinical OA are lacking. Early OA is difficult to detect clinically. Sensitivity of existing noninvasive (in vivo) analytic methods of subchondral bone changes is low, and small changes are difficult to quantify.8

Several studies in the rabbit and canine anterior cruciate ligament transaction (ACLT) model of OA9 showed a decrease of trabecular bone volume10–13 and a thinning of the subchondral plate13–15 early in the disease process. These changes are in contrast to those found in human end-stage OA.4, 5, 16 However, changes of subchondral bone in experimental guinea pig models at a more advanced stage were found comparable with those in human end-stage OA.17, 18 This suggests that bone remodeling in preclinical animal models of OA is biphasic: an early decrease in trabecular bone volume followed by a phase in which the subchondral bone becomes denser and stiffens. Recently, the early overall decrease in bone volume also was found in early human OA,19 supporting the biphasic response theory.

In this study we focused on the early phase of OA because early diagnosis and treatment are the challenge for the future. There are many different models of early OA with different causes. For example, in the canine Groove model, OA is induced by surgically damaging the cartilage of the femoral condyles, leading to a progressive cartilage degeneration, also in the surgically untouched tibial plateau.20–22 In the canine ACLT model, joint instability is the trigger for features of OA.9 It might well be that despite a final common outcome regarding cartilage damage with plate thickening and increased subchondral trabecular bone volume, the initial process with respect to bone changes may differ among various subpopulations (and models) of OA depending on the cause (or induction).

Most recently, Sniekers and colleagues published a pilot study suggesting that early in the disease process, subchondral trabecular bone changes occur in the ACLT model but not in the Groove model, whereas cartilage damage is comparable.13 As such, trabecular bone changes would be unrelated to cartilage pathology. However, this study was limited in number of animals and used a unilateral variant of both models with potential altered loading in control joints owing to the presence of OA in the contralateral joint. Therefore, the suggested uncoupling of cartilage damage and subchondral trabecular bone changes prompted us to study the structural changes of both bone and cartilage early in the process of OA in more detail in the bilateral versions of the two different models of OA.23, 24 Biomarkers were used to study cartilage degradation and bone loss over time, and as endpoint outcomes, micro–computed tomography (µCT) of bone and biochemical and histologic evaluation of cartilage were performed.

Materials and Methods


Skeletally mature Beagle dogs (n = 18 females, mean age 1.9 ± 0.1 years, weighing 11.8 ± 0.3 kg) were obtained from the animal laboratory of Utrecht University, The Netherlands. They were housed in groups of two dogs in indoor-outdoor pens and were let out on a large patio in groups for at least 2 hours a day. The feeding consisted of a standard diet and water ad libitum. The dogs were divided randomly in three groups of six animals each. The study was approved by the Utrecht University Medical Ethical Committee.

Surgical procedures and postoperative treatment

Under general anesthesia, surgery was performed through a 2- to 2.5-cm medial incision next to the ligamentum patellae in both knees of the 18 animals.

In 6 animals, OA was induced bilaterally according to the ACLT model9, 20, 24; the anterior cruciate ligament was transected using a pair of blunt curved scissors, making sure that no other structures were damaged. A successful procedure was established by a positive anterior drawer sign.

In 6 other animals, OA was induced bilaterally according to the Groove model.23 The cartilage of the lateral and medial femoral condyles of both knees was damaged using a Kirschner wire (1.5 mm diameter), which was bent at a 90-degree angle at 0.5 mm from the tip. This ensures that the depth of the grooves was restricted to 0.5 mm. In utmost flexion, approximately 10 longitudinal and diagonal grooves were made on the weight-bearing parts of the femoral condyles without damaging the subchondral bone. There was no absolute visual control over the procedure, but macroscopic evaluation at the end of the experiment showed a similar pattern in all affected knees.

In the remaining 6 animals, sham surgery was performed. Identical procedures were followed, visualizing the femoral condyles and the anterior cruciate ligament, except for the actual grooving of the cartilage or transection of the anterior cruciate ligament.

ACLT, groove, and sham dogs, as well as left and right joints, were operated on alternately to prevent surgical differences between the different groups and joints. Bleeding and soft tissue damage were prevented as much as possible. After surgery, synovium, fasciae, and skin each were sutured. The animals received analgesics (buprenorphine 0.01 mg/kg, im injections twice daily) and antibiotics (amoxicillin 400 mg/kg, oral administration twice daily) during the first 3 days after surgery. Starting 2 days after surgery, the dogs were allowed out on the patio again on a daily basis.

Longitudinal markers of bone and cartilage breakdown

Urine samples were collected twice before and every 2 to 3 weeks during the development of OA. As representative of bone resorption, lysylpyridinoline (LP) was measured in these samples. LP is a type I collagen cross-link residue considered specific for bone breakdown.25 The mature collagen cross-link LP was determined in acid hydrolysates of the urine samples by HPLC, as described previously.26 In addition, urine samples were assayed for cross-linked C-telopeptide II (CTX-II), a marker considered representative of cartilage degradation.27 CTX-II was measured by ELISA (Cartilaps, Nordic Bioscience Diagnostics, Herlev, Denmark) according to manufacturer's instructions. Both LP and CTX-II were corrected for urine dilution by urine creatinine level (Cayman Chemicals Co., Ann Arbor, MI, USA).

Endpoint parameters

Twenty weeks after surgery, the dogs were euthanized by an intravenous injection of Euthasate (sodium pentobarbital). Within 4 hours, the hind legs were amputated. From each animal alternately, the left (n = 3) or right (n = 3) knee joint was evaluated (the other knee joints were used for alternative purposes). Because in the Groove model surgical grooving could have led to subchondral bone changes not intrinsic to the OA process, in all further analyses, only the surgically untouched tibial plateaus were evaluated.

After opening the knee joint, digital high-resolution photographs were taken from the tibial plateau for blinded macroscopical scoring of cartilage damage. Subsequently, cartilage samples were taken from predefined locations of the weight-bearing area of the tibial plateaus as described previously.20 All samples were weighed (accuracy 0.1 mg) and stored for further analysis.

µCT analysis

The proximal part of the tibias was scanned in a µCT scanner (Skyscan 1076, Skyscan, Antwerpen, Belgium) with a voxel size of 18 µm. The reconstructed data set was segmented with a local thresholding algorithm.28

In both the medial and the lateral parts of each tibial scan, a cylinder with a diameter of 4.0 mm and a height of 3.5 mm (medial) or 3.1 mm (lateral) was selected. By use of anatomic landmarks, the cylinders were located in the middle of the weight-bearing area (Fig. 1A). They contained trabecular bone and subchondral plate (Fig. 1B).

Figure 1.

(A) The three locations at which evaluations were performed, indicated by squares, two subchondral in the epiphysis and one in the metaphysis. (B) A 3D representative picture of a cylinder of diameter 4 mm × height 3.1 to 3.5 mm consisting of subchondral plate and underlying trabecular bone as evaluated by µCT (see also Fig. 3).

The trabecular bone and subchondral plate were separated automatically using in-house software (Erasmus MC, Rotterdam, The Netherlands; for representatives, see Fig. 3). For the trabecular bone, bone volume fraction, which describes the ratio of bone volume over tissue volume (BV/TV); 3D trabecular thickness (Tb.Th)29; structure model index (SMI), a quantification of how rodlike or platelike the bone structure is30; and connectivity density (CD), describing the number of connections per volume,31 were calculated. For the subchondral plate, the 3D plate thickness (Pl.Th)29 was calculated. For these bone parameters, the data from the lateral and medial epiphyseal cylinders of each animal were averaged and used for statistical evaluation.

To analyze whether the trabecular changes were specific for the subchondral area, an additional region farther away from the joint space was analyzed. In advanced stages of OA, changes in the underlying bone, as shown by imaging techniques (X-ray/MRI/scintigraphy), are restricted to the subchondral/periarticular area and do not extend trough the metaphyseal area. If changes occur in the quickly adaptive trabecular bone of the metaphysis, they are more likely to be the result of a change in mechanical load of the whole proximal tibia. Bone changes in the metaphyseal area, distant from the cartilage, were considered not intrinsically related to cartilage degeneration because they are not likely to be influenced directly by changes in mechanical characteristics of cartilage or by chemical factors released from cartilage and synovium during the process of joint degeneration. Therefore, a cylinder (width 5.5 mm, height 3.5 mm) was selected in the metaphyseal area of the tibia, distal from the growth plate remains. This cylinder contained only trabecular bone (Fig. 1A), of which the same trabecular bone characteristics as described earlier were calculated.

Cartilage analysis

Macroscopic cartilage degeneration was evaluated on photographs by two observers unaware of the source of the photographs. Severity of cartilage degeneration of the tibia was graded from 0 to 4: 0 = smooth surface; 1 = roughened; 2 = slightly fibrillated; 3 = fibrillated, and 4 = damaged.32 Scores of the two observers were averaged (maximum of four) and used for statistical analysis.

For histology, four samples from predefined locations of the tibial plateau (two lateral and two medial) were fixed in 4% phosphate-buffered formalin containing 2% sucrose (pH 7.0). Cartilage degeneration was evaluated in safranin-O/fast-green iron hematoxylin-stained sections by light microscopy according to the slightly modified33 criteria of Mankin.34 Specimens were graded in random order by two observers unaware of the source of the cartilage. The average score (a maximum of 11) of the medial and lateral tibial compartments of each animal was used for statistical evaluation.

For biochemical analysis, proteoglycan (PG) content, one of the main components of cartilage, was determined from six samples taken from predefined locations. Details of biochemical analysis, using alcian blue staining, have been described previously by Mastbergen and colleagues.22, 32 In addition, damage of the cartilage collagen (type II) was assessed from four samples from predefined locations by selective proteolysis using α-chymotrypsin, which cleaves off only damaged, denatured collagen and leaves the intact triple-helical collagen behind. The soluble fraction (denatured collagen) was separated quantitatively from the insoluble fraction (intact collagen). Hydroxyproline levels were determined colorimetrically in both fractions after acid hydrolysis. The percentage of denatured collagen was calculated as (hydroxyproline in supernatant/total hydroxyproline) × 100%.10

Results for PG content and denatured collagen were averaged for all the samples from the lateral and medial compartments.

Calculations and statistics

Absolute mean values ± SDs of six sham, six groove, and six ACLT animals are presented. To analyze differences between OA and sham and between the two OA models, the unpaired nonparametric Mann-Whitney U test was used.


Longitudinal evaluation of bone breakdown

Urine level of lysylpyridinoline (LP) as a representative of bone resorption was elevated from week 6 until week 20 (the end of the experiment) in the ACLT model. In the Groove model, no such increase in bone resorption was observed. At all time points from week 6 on, the levels were statistically significantly higher in the ACLT model than in the sham surgery group (Fig. 2A; a: p < .05) as well as compared with the groove model group (all p < .05). Also, the area under the curve of the whole 20 weeks was statistically significantly higher in the ACLT group than in both other groups (p < .05 and p < .01 for sham and groove groups, respectively).

Figure 2.

Levels of lysylpyridinoline (LP) (A) and CTX-II (B) in urine normalized for urinary creatinine during the course of OA development for the three groups of animals. Mean values for each time point are given. a and b indicate statistically significant differences (p < .05) from the sham group for the ACLT and groove models, respectively.

Subchondral bone

Subchondral plate: The subchondral plate thickness was reduced significantly in the groove model and in the ACLT model compared with the sham group (both statistically significant; Fig. 3). Although the decrease in thickness tended to be larger in the ACLT model, this was not statistically significantly different from the groove model.

Figure 3.

(A) µCT analysis of the tibial subchondral plate. Mean ± SD of the plate thickness (Pl.Th) is given; a indicates p < .05; ns indicates not statistically significant. (B) Representative single-slide images of the original reconstructed CT scans and of segmented images with separation of the subchondral plate from the underlying trabecular bone, as used for 3D calculation of plate thickness for the three groups (sham, groove, and ACLT, respectively).

Subchondral trabecular bone: Subchondral trabecular bone in the ACLT model showed statistically significantly less bone volume fraction (BV/TV; Fig. 4A, left panel) and less trabecular bone thickness (Tb.Th; Fig. 4B, left panel). This also was reflected in the higher SMI (Fig. 4C, left panel) and higher CD (Fig. 4D, left panel), which indicated a more rodlike structure and generation of more pores in the original structure, resulting in more connections per volume. Interestingly, in the groove model, these differences compared with sham animals were seen only as a tendency, not statistically significantly different from the sham group.

Figure 4.

µCT analysis of tibial trabecular bone. Mean ± SD of (A) bone volume/total volume (BV/TV), (B) trabecular thickness (Tb.Th), (C) structure model index (SMI), and (D) connectivity density (CD) are given; a indicates p < .05; b indicates p < .01; ns indicates not statistically significant. Left panels represent the subchondral area, and right panels represent the metaphyseal area.

Metaphyseal trabecular bone

Because of the significant difference in subchondral trabecular bone characteristics between both models, trabecular bone changes in the metaphyseal part of the tibia were studied as well. Potential changes in this part of the bone are considered not to be intrinsically related to the cartilage degenerative process. Although absolute values of healthy bone at the two locations (subchondral epiphysis and metaphysis) were different due to differences in actual structure of the trabecular bone, it appeared that in the ACLT model bone volume fraction and trabecular thickness were lower than in the sham group, whereas the SMI was higher, similar to that found in subchondral trabecular bone (Fig. 4AC, right panel). In the groove model, metaphyseal trabecular bone structure was not statistically different from that of the sham group, as observed for the subchondral trabecular bone. Again, a slight tendency to a change in the direction of the ACLT model was observed. CD in the metaphyseal bone in both models was not different from that in the sham group (Fig. 4D, right panel).

Longitudinal evaluation of cartilage breakdown

Urinary CTX-II levels as a representative of cartilage breakdown were elevated from week 10 until week 20 in the ACLT model and from week 1 in the groove model. At all time points from week 10 on (except for one), the levels were statistically significantly higher for the ACLT and groove models than for the sham surgery group (Fig. 2B). In addition, the area under the curve was statistically significantly higher in the ACLT and groove models than in the sham group (p < .05 and p < .01, respectively). No difference was observed between both OA models except for an earlier (at 1 and 3 weeks) increase in CTX-II levels in the groove model compared with the ACLT model, which was anticipated based on the way cartilage damage is induced in both models (direct cartilage damage and joint instability, respectively).

Cartilage matrix integrity

Cartilage matrix integrity was measured by several parameters—a macroscopic score, a histologic score, and biochemical analysis of proteoglycan content and collagen damage.

Macroscopic evaluation of cartilage of the tibial plateau, located above the subchondral bone areas, evaluated by µCT, showed a statistically significant increase in cartilage damage compared with the sham group in both the groove and ACLT models (both p < .05; Fig. 5, top left panel). Histologic evaluation of cartilage in both models confirmed the macroscopic results, as represented by the increased modified Mankin grade (Fig. 5, top right panel; both p < .05 compared with the sham group). Representative photographs of the cartilage in all three groups have been added (Fig. 5B), showing the characteristic features of OA, including loss of safranin-O staining, fibrillation of the articular surface, and chondrocyte clustering. Loss of safranin-O staining was corroborated by the decrease in PG content (Fig. 5, bottom left panel) as determined biochemically. This loss of PGs was accompanied by an increase in the amount of denatured collagen (Fig. 5, bottom right panel) in the groove and ACLT models when compared with the sham group. All these changes appeared very similar in both models, although histologic cartilage damage was slightly more pronounced in the groove model than in the ACLT model (p < .05).

Figure 5.

Cartilage integrity parameters. (A) Mean ± SD of macroscopic (top left) and histologic (top right) cartilage damage (Mankin grade). (B) Representative micrographs of cartilage for each of the three groups showing surface damage, cell clustering, and loss of safranin-O staining for the groove and ACLT models. (C) Biochemical [cartilage matrix GAG content (bottom left) and collagen damage (bottom right)] analysis of cartilage obtained from the tibial surface. a at the top of a bar indicates a statistically significant (p < .05) difference from the sham group. The differences between the groove and ACLT groups have been indicated as well, a indicating p < .05 (ns = not statistically significant).


This study demonstrates that thinning of the subchondral plate coincides with degenerative changes in articular cartilage independent of the model used and as such is considered an intrinsic part of the OA process. This is in contrast to the subchondral trabecular bone changes that were clear in the ACLT model but hardly present in the groove model and that coincided with similar changes in metaphyseal bone. As such, subchondral trabecular bone changes might not be an intrinsic part of the cartilage degenerative process and are differently regulated in different models (owing to different causes) of OA. It is suggested that mechanical unloading of the bone is causative in early subchondral trabecular bone changes rather than the degenerative process itself.35

Subchondral plate thickness was significantly reduced in both OA models, as were cartilage integrity parameters. The similarity in cartilage damage for both models was corroborated by the similarity in elevation of the cartilage collagen marker CTX-II during the course of development of the cartilage degeneration. The early reduction in plate thickness is in concurrence with a previously published pilot study that demonstrated that subchondral plate thinning at 3 and 10 weeks of OA development even may precede cartilage damage.13 Also, others have demonstrated that plate thinning is an early feature in the OA process.15, 36 Unfortunately, in our study group, size and variation in bone and cartilage parameters were too small to perform meaningful correlations between values of plate thickness and cartilage degeneration for each group.

Although speculative, there might be a direct role for biochemical factors between cartilage and bone in the process of thinning of the underlying subchondral plate.37 The tidemark, which was thought previously to be a strict barrier between cartilage and bone, allows perfusion of chemo- and cytokines released by the diseased chondrocytes and inflamed synovial tissue.38 In line with this assumption is that if mechanics instead of chemical components would be involved, a compensatory increase in bone tissue is anticipated owing to the loss of the mechanical function and with that shock absorbance of the cartilage. This would increase local stresses on the subchondral bone and result in an increase in bone tissue volume/thickness. Whereas the opposite, loss of bone tissue, is observed, a role for chemo- and cytokines rather than mechanics in early plate thinning is suggested.

Thinning of the plate was hardly represented by an increase in urine LP levels. In the groove model, only a very slight, not statistically significant increase in urine LP levels over time was observed. In contrast, LP levels were significantly increased in the ACLT model. This increase is expected to depend on the significant changes in subchondral trabecular bone that were present only in the ACLT model and, although a tendency was observed, not clearly in the groove model, whereas cartilage damage actually was most pronounced in the groove model. The trabecular changes in the ACLT model were not restricted to the subchondral trabecular area but extended throughout the bone. It is hypothesized, as has been reported previously, that ACL transection results in clearly diminished loading of the whole paw (braking, stance, and propelling force)39, 40 as a consequence of joint instability. Although not measured objectively, in this study a clear decrease in overall activity in the ACLT group owing to bilateral hind limb joint instability, in addition to more lameness in both hind limbs, was observed, probably compensated for by increased loading of the front limbs. Note that dogs can easily transfer their load to their front limbs.41 This changed loading pattern was not observed in the sham group or the groove group, corroborating the observation in the unilateral versions of both models.39–41 The significant increase in bone resorption over time as evaluated by urine LP levels is supportive in this respect. Importantly, the difference in bone resorption marker between both OA models with comparable cartilage damage should be taken into account when evaluating such biomarkers prospectively in clinical studies. It could well be that depending on the original cause of OA, bone resorption (turnover) levels are increased or not.

In contrast to the changes in trabecular bone volume, thickness, and SMI, the CD of the metaphyseal bone did not show the same changes as in the subchondral area. It is known that CD can either increase or decrease as a result of bone volume loss.42 With loss of bone volume, trabeculae may disappear, leading to a decrease in CD. However, bone loss also can increase CD by an increased porosity of the trabecular structures. Therefore, changes in this parameter are less conclusive.

Several studies showed that subchondral trabecular volume decreased early in the OA process, followed by an increase later in the process. All these models are based on joint instability (ACLT or collagenase-induced ligament weakening) or spontaneous OA of unknown origin. As such, the initial decrease in trabecular bone volume in these models also could be caused by unloading in the early phase of OA owing to changed biomechanical conditions of the joint, pain, or joint stiffness. Unfortunately, other primarily cartilage-damage-induced OA models have not been evaluated for bone characteristics in such a detailed way to be supportive of this concept in this respect.

This study underscores that cartilage damage in these models is not secondary to more stiffened bone because the decrease in bone volume is likely to result in more softened and fragile bone. These early changes in mechanical characteristics of the subchondral plate and trabecular bone even may have a natural protective function because the cartilage may be spared secondarily from excessive forces. On the other hand, the fragility of the bone also can be causative in the process of cartilage degeneration.

As such, the significance of the decrease in subchondral plate thickness and trabecular bone volume is still unknown. Do these early changes normalize over time, or are they the necessary trigger resulting in subchondral sclerosis? Although a short longitudinal pilot study has been performed,13 it is clear that longer follow-up studies for both models are needed. Do late-stage plate thickening and increase in trabecular volume also develop in the groove model, or is it a consequence of the early decrease in trabecular volume parameters seen only in the ACLT model? At present, we chose to compare the two models early in the disease process, but the results from this comparison urge a comparison later in the disease process as well.

Despite common degenerative features of clinical end-stage OA (ie, cartilage damage, subchondral sclerosis, and joint inflammation), the pathway from a healthy joint to a destroyed joint can be diverse. In this study, different causes of joint degeneration are represented—joint instability (ACLT model) and primary cartilage damage (groove model)—both common causes for OA in the human clinic. Gaining knowledge on the differences and similarities of pathogenic events in the development of OA with different etiologies can bring us one step closer to a more patient-specific approach in the early treatment of OA. A patient with a chondral defect may not benefit from pain medication (maintaining proper joint loading) or bisphosphonate therapy (inhibiting bone resorption) in contrast to a patient with an insufficient cruciate ligament in its early phase, preventing the bone changes early in the process by maintaining proper joint loading and/or arresting bone turnover.

In conclusion, the occurrence of early trabecular bone changes depends on the cause/induction of joint degeneration and is not an indispensable factor in cartilage degeneration. There could be an important role for unloading because the natural mechanics of the joint are disrupted in the ACLT model and not in the groove model. The early decrease in plate thickness, concomitant with the cartilage damage, present in both models, suggests that plate thinning is a process intrinsic to cartilage degeneration.


None of the study sponsors had a role in the study design; in the collection, analysis, and interpretation of the data; or in the decision to submit the manuscript for publication. All the authors state that they have no conflicts of interest.


This study was supported by Pfizer, Inc., the Dutch Arthritis Association, and the Anna Fund.