Cyclo-Oxygenase 2 Function Is Essential for Bone Fracture Healing


  • Ann Marie Simon,

    1. Department of Orthopaedics, University of Medicine and Dentistry of New Jersey, New Jersey Medical School, Newark, New Jersey, USA
    2. Department of Biomedical Engineering, Rutgers University, Piscataway, New Jersey, USA
    Search for more papers by this author
  • Michaele Beth Manigrasso,

    1. Department of Orthopaedics, University of Medicine and Dentistry of New Jersey, New Jersey Medical School, Newark, New Jersey, USA
    2. Department of Biomedical Engineering, Rutgers University, Piscataway, New Jersey, USA
    Search for more papers by this author
  • J. Patrick O'Connor Ph.D.

    Corresponding author
    1. Department of Orthopaedics, University of Medicine and Dentistry of New Jersey, New Jersey Medical School, Newark, New Jersey, USA
    2. Department of Microbiology and Molecular Genetics, UMDNJ-New Jersey Medical School, Newark, New Jersey, USA
    • Medical Sciences Building, Room G580, Department of Orthopaedics, UMDNJ-New Jersey Medical School, 185 South Orange Avenue, Newark, NJ 07103, USA
    Search for more papers by this author

  • The authors have no conflict of interest


Despite the molecular and histological similarities between fetal bone development and fracture healing, inflammation is an early phase of fracture healing that does not occur during development. Cyclo-oxygenase 2 (COX-2) is induced at inflammation sites and produces proinflammatory prostaglandins. To determine if COX-2 functions in fracture healing, rats were treated with COX-2-selective nonsteroidal anti-inflammatory drugs (NSAIDs) to stop COX-2-dependent prostaglandin production. Radiographic, histological, and mechanical testing determined that fracture healing failed in rats treated with COX-2-selective NSAIDs (celecoxib and rofecoxib). Normal fracture healing also failed in mice homozygous for a null mutation in the COX-2 gene. This shows that COX-2 activity is necessary for normal fracture healing and confirms that the effects of COX-2-selective NSAIDs on fracture healing is caused by inhibition of COX-2 activity and not from a drug side effect. Histological observations suggest that COX-2 is required for normal endochondral ossification during fracture healing. Because mice lacking Cox2 form normal skeletons, our observations indicate that fetal bone development and fracture healing are different and that COX-2 function is specifically essential for fracture healing.


FRACTURE HEALING is the culmination of a highly orchestrated series of physiological and cellular pathways to restore the function of broken bones. Osteogenesis during fracture healing occurs by intramembraneous and endochondral ossification that resembles fetal skeletogenesis.(1–3) However, the localized tissue hypoxia, the fracture hematoma, and subsequent inflammation at the fracture site, as well as the frank remodeling of the fracture callus at the later stages of healing, are unique physiological and cellular responses to bone fractures that have no known corresponding counterpart during fetal development of the skeleton.

It has been hypothesized that the early physiological responses to a bone fracture, namely, hypoxia and inflammation, induce gene expression pathways and promote cell proliferation and migration into the fracture site to promote healing.(4,5) Production or release of specific growth factors, cytokines, and local hormones at the fracture by these physiological processes would create the appropriate microenvironment to (1) stimulate periosteal osteoblast proliferation and intramembraneous ossification at the fracture site, (2) stimulate cell proliferation and migration into the fracture site, and (3) stimulate chondrocyte differentiation in the soft callus with subsequent endochondral ossification. Remodeling of the fracture callus by osteoclastic resorption and subsequent osteogenesis converts the fracture callus woven bone into cortical bone and thereby restores the shape and mechanical integrity of the fractured bone.

One class of factors that could mediate certain events of fracture healing is the prostaglandins. The effects of prostaglandins on bone metabolism are complex because prostaglandins can stimulate bone formation as well as bone resorption.(6) Prostaglandins are synthesized by osteoblasts, and different cell stimuli can alter the amount and possibly the spectrum of prostaglandins produced by osteoblasts.(7–9) Therefore, signal transduction, mechanical perturbations, or other physiological signals can affect bone metabolism through alteration of prostaglandin production.

Prostaglandin synthesis begins with the release of arachidonic acid from membrane phospholipids by phospholipase activity. Subsequently, arachidonic acid is converted into prostaglandin H2 (PGH2) by cyclo-oxygenase via two independent catalytic steps.(10) Synthase enzymes then convert PGH2 into the specific prostaglandins produced by that cell such as PGD2, PGE2, PGF, prostacyclin, and thromboxane. Thus, cyclo-oxygenase activity is essential for normal prostaglandin production and cyclo-oxygenase is believed to be the rate-limiting enzyme in the prostaglandin synthetic pathway.

There are two known forms of cyclo-oxygenase (cyclo-oxygenase 1 [COX-1] and COX-2), which are encoded by two genes.(11,12) COX-1 is expressed constitutively by many tissues and provides a homeostatic level of prostaglandins for the body and specific organs such as the stomach.(13) In contrast, COX-2 is expressed inductively in vitro by a diverse array of cell stimuli such as exposure to lipopolysaccharide,(14,15) certain cytokines and growth factors, (9,12) or mechanical stress.(8,16) COX-2 expression can be stimulated in vivo by wounding and inflammation.(17–19)

Inhibiting the cyclo-oxygenase activity of COX-1 and COX-2 can reduce prostaglandin synthesis by preventing the conversion of arachidonic acid into PGG2, the precursor of PGH2. This is commonly done to reduce inflammation and pain with aspirin and nonsteroidal anti-inflammatory drugs (NSAIDs) such as indomethacin. Most NSAIDs inhibit the cyclo-oxygenase activity of COX-1 and COX-2 with near equal potency, which often leads to detrimental gastrointestinal or kidney side effects.(20,21) Use of COX-2-selective NSAIDs has become very popular because these drugs, such as celecoxib (Celebrex; Pharmacia Corp., Peapack, NJ, USA) and rofecoxib (Vioxx; Merck and Co., West Point, PA, USA), preferentially inhibit the cyclo-oxygenase activity of COX-2 with selectivity relative to COX-1 of ∼8-fold for celecoxib and 35-fold for rofecoxib.(22)

Very little is known about prostaglandin production during fracture healing. Prostaglandin levels in and around the healing callus of rabbit tibia that had been severed by osteotomy showed that PGE and PGF levels were elevated between 1 and 14 days and 7 and 14 days postosteotomy, respectively.(23) No survey of the temporal pattern or variety of prostaglandins produced during fracture healing has been reported for other rodents or man. Nonspecific NSAIDs have been shown to delay but not stop fracture healing in experimental animal models.(24–26) In addition, nonspecific NSAIDs have been shown to reduce the incidence and severity of heterotopic bone formation in humans after certain fractures or orthopedic surgical procedures.(27,28) These observations suggest that prostaglandins are necessary for bone formation but given the limitations of nonspecific NSAID use, it is unknown whether prostaglandins produced by COX-1, COX-2, or both enzymes are essential for fracture healing.

Because prostaglandins do affect bone metabolism and formation and because inflammation is an early physiological response to bone fracture, we hypothesized that prostaglandin synthesis by COX-2 would be essential for normal bone fracture healing. Our experimental approach was to assess fracture healing using a standard rat closed femur fracture model in which COX-2 function was inhibited in vivo with the COX-2-selective NSAIDs celecoxib and rofecoxib. The results were striking in that femur fracture healing in rats treated with celecoxib or rofecoxib was impaired dramatically. These observations were confirmed by examining fracture healing in Cox1 and Cox2 null mice. Histological observations indicated that the defect in fracture healing caused by the COX-2-selective NSAIDs or by lack of Cox2 occurred in the endochondral ossification pathway.


Animals, drug dosage, and administration

A total of 253 male Sprague-Dawley rats (584 ± 62 g) were fed a standard diet and kept caged separately in a constant temperature and humidity environment. All rats were 6-9 months old at the beginning of the experiment. Drugs were administered daily by gavage beginning 2 days before fracture. Animals were selected randomly for each treatment group. The rats were gavaged with aqueous suspensions of indomethacin (1 mg/kg), celecoxib (4 mg/kg), or rofecoxib (3 mg/kg). No drug (control) rats were not initially gavaged, but later, rats were gavaged with water and no difference was noted between the no-drug rats that had been gavaged and those that had not. No statistically significant differences were found in animal weight changes during the experiments. The disposition of the rats used in this study is detailed in Table 1.

Table Table 1.. Disposition of Rats Used in This Study
original image

Retired breeder female Cox1−/− (B6; 129P2-Ptgs1tm1) and Cox2−/− (B6; 129P2-Ptgs2tm1) mice were obtained from Taconic Farms (Germantown, NY, USA). Closed femur fracture production in the mice was done using a method similar to that described in the following section. All animal procedures were approved and conducted in accordance with the New Jersey Medical School Institutional Animal Care and Use Committee.

Closed femur fracture production

The rats were anesthetized by intraperitoneal injection of ketamine (40 mg/kg) and xylazine (5 mg/kg). Under aseptic conditions, a medial parapatellar incision (0.4-0.5 cm) was made in the right hindlimb and the patella was dislocated laterally. The medullary canal was entered through the intercondylar notch and reamed with an 18G needle. A 1.1-mm stainless steel pin (Small Parts, Inc., Miami Lakes, FL, USA) was then inserted into the canal and secured in the proximal part of the greater trochanter by tamping. Then, the distal portion of the pin was cut flush with the femoral condyles and the patella dislocation was reduced. The soft tissue and skin were closed with 4-0 Vicryl sutures. After closing, the diaphysis of the pinned femur was fractured by means of a three-point bending device as described by Bonnarens and Einhorn.(29)


Radiographs were made postfracture to confirm the position and quality of each fracture and at death to determine the degree of healing. In addition, several rats were selected randomly to produce serial radiographs (at least two rats per treatment group). Radiographs were made of these rats weekly under anesthesia until the experimental endpoint (8 weeks). Radiographs of mice also were made under anesthesia. All radiographs were made using a 43805N Faxitron (Hewlett-Packard, McMinnville, OR, USA) and Eastman Kodak Co. MinR-2000 mammography film (Eastman Kodak Co., New Haven, CT, USA).

Mechanical testing

Animals within each treatment group were killed at 4, 6, and 8 weeks postfracture by CO2 asphyxiation. Animals with oblique, comminuted, or infected fractures were not used for mechanical testing. Both femora were removed and cleaned of all soft tissue, leaving the fracture callus undisturbed, and then immediately processed for mechanical testing. The samples were wrapped in saline-soaked gauze to prevent dehydration between steps. Measurements of the femora were taken using digital calipers to determine femur length and external callus dimensions. The intramedullary pin was removed from the fractured femur and a 1-mm-diameter stainless steel pin (∼0.8 cm in length) was inserted at the proximal and distal end perpendicular to the long axis of the bone to prevent slipping in the potting material. The intact femur also was pinned as described previously. The femoral ends were potted in 1-in hexnuts using a low-melt temperature metal (Wood's metal; Alfa Aesar, Ward Mill, MA, USA). Once potted, the gauge length (L) of each femur was measured. Torsional testing was conducted using a servohydraulics testing machine (MTS Systems Corp., Eden Prairie, MN, USA) with a 20-Nm reaction torque cell (Interface, Scottsdale, AZ, USA). The testing was carried out to failure at a rate of 2°/s and a data-recording rate of 20 Hz. Both the fractured and the intact femora were tested in internal rotation in proper anatomic orientation. The peak torque and angle at failure were calculated from the load-deformation curves. Internal fracture callus dimensions were measured after mechanical testing. From the callus dimensions, the polar moment of inertia (J) was calculated based on a hollow ellipse model.(30,31) The equations used to derive torsional rigidity, shear stress, shear modulus, and J were as follows(32): (i) torsional rigidity, (Tmax · L)/φ, where Tmax is the peak torque value in Newton-millimeters, L is the gauge length in millimeters, and φ is the angle at failure in radians; (ii) shear stress, (Tmax · Rmax)/J, where Rmax is the largest radial dimension of the fracture callus in millimeters (ao) and J is the polar moment of inertia; (iii) shear modulus (G), (Tmax · L)/Jφ; and (iv) polar moment of inertia (J), [φ (ab3 + a3b − (at)(bt)3 − (at)3(bt)]/4, where a is [ai + [ (ao −ai)/2], b is [bi + [(bobi)/2], t is the average bone thickness at the site of failure and is calculated as [ (aoai) + (bobi)]/2, where ao is the callus maximum outside radius, ai is the maximum interior radius, bo is the least outside radius, and bi is the least interior radius in millimeters. Only torsional testing data for which the fractured and control femur tested without incident were used.


Rats were killed at 2, 3, 4, 6, and 8 weeks postfracture by CO2 asphyxiation. Both femora were resected and the stainless steel pin was removed from the medullary canal. The harvested femora were fixed in 10% buffered formalin and embedded in polymethylmethacrylate following standard histological techniques for calcified tissue. The samples were sectioned sagittally through the fracture callus using an Isomet diamond saw (Buehler, Ltd., Lake Bluff, IL, USA), mounted on plexiglas slides, and polished to a thickness of 100 μm. The slides were then stained with van Gieson's picrofuchsin and Stevenel's blue to identify new bone growth and cartilage formation.(33) At least three fracture callus specimens were examined at each time point for each treatment group (range, 3-8; average of 5). Mice femora were fixed, decalcified, paraffin embedded, sectioned, and stained with Masson's trichrome stain. The samples were viewed and photomicrographs were taken using an Olympus BH2-RFCA microscope or an Olympus SZ40 microscope (Olympus Optical Co., Ltd., Shinjuku-ku, Tokyo, Japan).


Treatment with COX-2-selective NSAIDs leads to fracture nonunions and incomplete unions

Femur fracture healing was followed by serial radiographic analysis of rats treated with celecoxib, rofecoxib, indomethacin, or gavaged daily with water (no drugs group). Radiographs were made immediately after fracture production and then every week till the endpoint of the experiments (8 weeks postfracture). Representative results are shown in Fig. 1.

Figure FIG. 1..

Radiographic analysis of fracture healing in NSAID-treated rats. High-resolution radiographs were made immediately postfracture and then every week until the endpoint of the experiment (8 weeks) using a Hewlett-Packard Faxitron. Shown are radiographs of the fractured right femurs from the same rats taken at 2 (top), 4 (middle), and 8 weeks (bottom) postfracture (dorsal-ventral view). (A-D) Radiographs from a no-drug rat, an indomethacin-treated rat, a celecoxib-treated rat, and a rofecoxib-treated rat, respectively. Note that the fracture is still clearly evident in the 8-week postfracture radiographs of the celecoxib- and rofecoxib-treated rats.

We found that femur fracture healing proceeded normally in the no-drug rats as expected. At 1 week postfracture, formation of the hard callus could be detected radiographically but was more evident at 2 weeks (Fig. 1A). By 4 weeks postfracture, calcification of the soft callus region was clearly evident indicating that endochondral ossification had occurred. Additionally, by 4 weeks postfracture, the new bone formed during fracture repair had almost bridged the fracture gap. Bridging of the fracture and remodeling of the fracture callus were evident at 6 weeks postfracture (not shown). Continued remodeling of the callus as well as remodeling of the original femoral cortical bone at the fracture site is clearly evident by 8 weeks postfracture. These radiographic observations are typical of normal fracture healing.

Indomethacin treatment appeared to delay but not prevent fracture healing consistent with previous reports.(24–26) By 2 weeks postfracture, an X-ray dense hard callus is clearly evident in the indomethacin-treated rats (Fig. 1B). However, bridging of the fracture gap did not appear to occur until 5-6 weeks postfracture as compared with ∼4-5 weeks postfracture in the untreated rats (compare Fig. 1A with Fig. 1B). Bridging and remodeling were evident in the 8-week postfracture radiographs of the indomethacin-treated rats.

Celecoxib or rofecoxib treatment did not prevent formation of an X-ray dense callus as can be seen in the 2-week and 4-week postfracture radiographs (Figs. 1C and 1D). However, the original fracture was still plainly evident in the celecoxib- (Fig. 1C) and rofecoxib (Fig. 1D)-treated rats even after 8 weeks. No rofecoxib-treated rat was observed to have a normally bridged callus by radiography. However, nonunions, incomplete unions, and unions of the fractured femurs were observed by radiography in the celecoxib-treated rats. The incomplete unions were typified by the radiograph seen in Fig. 1C in which one cortex of the fracture callus was bridged but in which the original cortical bone ends of the fractured femur had not joined and the fracture was still clearly evident.

In addition to the serial radiographs made for certain rats, all animals in this study were examined radiographically immediately postfracture and when killed. A random, blinded sample of the 8-week postfracture radiographs was independently examined by six observers and scored as a union (1 point), incomplete union (0.5 points), or nonunion (0 points). The no-drug (n = 8), indomethacin-treated (n = 12), celecoxib-treated (n = 13), and rofecoxib-treated (n = 14) fracture radiographs had average scores of 0.75 ± 0.05 (SEM), 0.58 ± 0.05, 0.54 ± 0.04, and 0.38 ± 0.04, respectively. Despite the known difficulties associated with judging fracture healing from radiographs,(34–36) the data were significant between all treatment groups (p < 0.001; Kruskal-Wallis analysis of variance [ANOVA] on ranks) and between the no-drug and celecoxib or rofecoxib treatment groups but not the indomethacin treatment group at p < 0.05 (post hoc Dunn's Test; SigmaStat software, SPSS, Inc., Chicago, IL, USA). The radiographic comparisons were consistent with a delayed healing effect by indomethacin and an inhibitory effect by the COX-2-selective NSAIDs.

The mechanical properties of the healing femur fracture callus are diminished by NSAID treatment

In conjunction with the radiographic analysis, torsional mechanical testing of fractured femurs was performed also. The fractured femur and contralateral control femur from rats at 4, 6, and 8 weeks postfracture were tested to failure in torsion for each treatment group (no drugs, indomethacin, celecoxib, and rofecoxib). The data from these tests are summarized in Fig. 2 and Table 2. Peak torque is the maximum twisting force generated during torsional testing of the femur. Torsional rigidity is a measure of a structure's resistance to torque. Thus, a bone with high torsional rigidity would fail after only a few degrees of rotation, but soft tissue would not reach its peak torque until after a large angular deflection. Maximum shear stress is a measure of the ultimate shearing force withstood by the femur before failure and is a function of the applied torque and polar moment of inertia, which is dictated by callus geometry. Shear modulus measures the elastic resistance to deformation by a shearing stress for a given material and is constant for a given material.

Table Table 2.. Torsional Mechanical Testing Data
original image
Figure FIG. 2..

COX-2-selective NSAIDs alter the mechanical properties of fractured femurs. Mechanical testing data were obtained or derived as described in the Experimental Procedures section. The data for each fractured femur were normalized as a percentage of the value obtained from that animal's contralateral, unfractured femur, except for shear modulus. Shown are the mean normalized values at each time point and for each treatment group for (A) peak torque, (B) torsional rigidity, (C) shear stress, and (D) the mean shear modulus. The error bars represent SEs of the mean. Pairwise t-tests were made between the no-drug and experimental treatments within a time point. Significant differences (p < 0.05) are noted with asterisks. The p values that approached significance are indicated with letters (a = 0.052, b = 0.080, and c = 0.065).

We found in the no-drug rats that the normalized peak torque (92%) and torsional rigidity (81%) of the fractured femur was restored by 8 weeks postfracture as compared with the contralateral control femurs from each animal (Fig. 2). However, the shear modulus (0.96 GPa) and normalized shear stress (44%) of the fractured femurs at 8 weeks postfracture were still less than the contralateral control femurs (Fig. 2; Table 2). This is the expected result because during fracture healing, the ultimate mechanical integrity of the fractured bone, that is, peak torque, is maintained at a high level by increasing bone diameter via the fracture callus. Because the mechanical properties of the initially soft tissue within the callus and later the newly formed bone are much weaker than the mechanical properties of mature cortical bone; shear stress and shear modulus were, as expected, less than the contralateral control femurs. As the newly formed bone within the fracture callus matures by remodeling, the mechanical properties of the fractured bone increase. This is evident in our results as shear stress and shear modulus increase with time (Fig. 2). The high normalized torsional rigidity found for the fractured femurs in the no-drug rats at 6 (82%) and 8 (81%) weeks postfracture indicates that the fracture had been bridged by new bone as would be expected.

We also observed that all of the 6-week and 8-week postfracture no-drug femurs and all the contralateral unfractured femurs from all the treatment groups failed as predicted middiaphyseal spiral fractures during the torsional mechanical testing.

Indomethacin treatment reduced the mechanical properties of the healing femur fractures at earlier time points (Fig. 2). However, by 8 weeks postfracture, the normalized peak torque, torsional rigidity, and shear stress values obtained from the indomethacin-treated rats were not significantly different from the no-drug rats. In contrast, at 6 weeks postfracture, the normalized peak torque, torsional rigidity, and shear stress values (42, 31, and 14%, respectively) obtained from the indomethacin-treated rats were less than the no-drug rats at 6 weeks postfracture (71, 82, and 33%, respectively). Pointedly, the significantly low torsional rigidity of the fractured femurs from the indomethacin-treated rats at 6 weeks postfracture indicates that the fracture had not been bridged by bone. Of the eight fractured femurs tested at 8 weeks postfracture, six failed as unions, one failed as a incomplete union, and one failed as a nonunion (Table 3). These observations indicate that the nonselective NSAID, indomethacin, delays but does not prevent fracture healing, which is consistent with previous studies and shows the validity of our assay methods.(24–26)

Table Table 3.. Frequency of Unions, Incomplete Unions, and Nonunions in the Mechanically Tested Fractured Femurs at 8 weeks Postfracture
original image

Rofecoxib treatment had a drastic effect on the mechanical properties of the fractured femurs. At 8 weeks postfracture, for all values measured or derived, the mechanical properties of the fractured femurs from the rofecoxib-treated rats were significantly less than the no-drug rat fractured femurs (Fig. 2, Table 2). At 8 weeks postfracture, the fractured femurs of the rofecoxib-treated rats had only obtained 45, 29, and 16% of peak torque, torsional rigidity, or maximum shear stress of the contralateral unfractured femurs, respectively. The low torsional rigidity and shear modulus (0.35 GPa) values obtained from the fractured femurs of the rofecoxib-treated rats are consistent with healing failure and the formation of nonunions. In addition, whereas all of the contralateral control femurs from the rofecoxib-treated rats failed as middiaphyseal spiral fractures, four of the five fractured femurs at 8 weeks postfracture failed as nonunions and the other failed as an incomplete union (Table 3). These observations show that, at the dose and treatment regime used, the COX-2-selective NSAID rofecoxib stops fracture healing.

Unlike rofecoxib treatment, no significant differences were found in the mechanical properties of the healing fractured femurs from the celecoxib-treated rats as compared with no-drug rats. Despite the overall similarities in the mechanical values obtained between no-drug rats and celecoxib-treated rats, 3 of six fractured femurs from the celecoxib-treated rats, at 8 weeks postfracture failed as nonunions during the mechanical testing procedure and the other 3 failed as incomplete-unions (Table 3). The relatively low normalized torsional rigidity (45%) and shear stress (23%) found for fractured femurs from the celecoxib treated rats at 6 weeks postfracture indicates that the fracture site had not been bridged with bone (Fig. 2). Although not statistically different from the no drug rat fractured femurs, the data obtained from the celecoxib treated rat fractured femurs parallels closely the patterns obtained from the femurs of the indomethacin treated rats. Together these observations suggest that, at the celecoxib dose and treatment regime used, fracture healing is delayed and to a lesser extent than that found for the rofecoxib treatment regime, inhibited.

A κ2 analysis was performed on visual inspection data obtained from the 8-week postfracture femurs after mechanical testing (Table 3). The fractured femurs were considered to have failed as (a) unions if a spiral fracture developed through the diaphysis of the femur, (b) nonunions if the femur failed completely along the original fracture site, and (c) incomplete unions if some new bone bridging of the fracture site was evident but that the femur still failed primarily along the original fracture site. The data from the no drug, celecoxib, and rofecoxib treatment groups were compared with that from the indomethacin treatment group. Our analysis indicates that no statistical difference exists between the no-drug and the indomethacin treatment groups but that the celecoxib and rofecoxib treatment groups are significantly different from the indomethacin treatment group. These observations indicate that inhibition of COX-2 dramatically inhibits fracture healing.

No significant differences in the mechanical properties of the contralateral femurs were found between treatment groups. This indicates that the experimental treatment regimes did not alter the intrinsic properties of the rat bone by enhanced bone resorption or deposition, at least for the relatively short time frame examined.

COX-2-selective NSAID treatment alters cartilage formation during fracture healing

To assess the early stages of endochondral ossification in the fracture callus, we undertook a histological analysis of fracture healing in the drug-treated rats. At 2 weeks postfracture, gross abnormalities were present in the histology of the healing femur fractures of the indomethacin-, celecoxib-, or rofecoxib-treated rats as compared with an untreated control rat (Fig. 3). In all four experimental groups, significant mineralized callus was evident at the fracture site as expected from our radiographic data. Endochondral ossification appeared to be proceeding normally in the no-drug rat specimens. In contrast, the histological specimens from the indomethacin and COX-2-selective NSAID-treated rats had abnormally formed cartilage elements within the callus. The positional extent of new bone formed in the callus of the NSAID-treated rats also appeared to be abnormal in that it did not fully extend along the cortical bone to the fracture site. In the no-drug rats, new bone in the callus extends to the very ends of the cortical bone fracture site. This is not so in the NSAID-treated rats where this region of the callus generally is occupied by cartilage.

Figure FIG. 3..

COX-2-selective NSAIDs disrupt endochondral ossification during fracture healing. Shown are fracture calluses from no-drug, indomethacin, celecoxib-treated, and rofecoxib-treated rats at 2, 3, and 4 weeks postfracture as indicated. The specimens were embedded in polymethylmethacrylate, sectioned, and stained with van Gieson's picrofuchsin and Stevenel's blue so that bone is red, calcified cartilage is orange to red, cartilage is deep blue to purple, and fibrous tissue and muscle is pale blue. Each section is oriented with the cortical bone on the bottom, fracture callus on top, and fracture site in the middle. (D, G, and J) Note the abnormal cartilage morphology in the calluses of the NSAID-treated rats and the lack of cartilage in the (H and I) celecoxib- and (K and L) rofecoxib-treated rats at 3 weeks and 4 weeks postfracture.

NSAID treatment grossly altered fracture callus morphology at 3 weeks and 4 weeks postfracture (Fig. 3). Fracture healing proceeded normally in the no-drug rats with near bridging of the callus apparent by 4 weeks postfracture in concurrence with our radiographic data (Fig. 1). However, healing was delayed in the NSAID-treated rats. The fracture calluses of the indomethacin-treated rats were not bridged at 4 weeks but still appeared to be undergoing endochondral ossification based on the presence of cartilage within the soft callus at 3 weeks and 4 weeks postfracture. In contrast, little or no cartilage was evident in the fracture calluses of the celecoxib- or rofecoxib-treated rats at 3 weeks and 4 weeks postfracture, indicating that endochondral ossification had ceased. Instead, the fracture gap at 3 weeks and 4 weeks postfracture in the celecoxib- and rofecoxib-treated rats was filled with fibroblast-like cells that were lightly stained with Stevenel's Blue (Figs. 3 and 4). Additionally, massive resorption of the woven bone in the callus of the NSAID-treated rats appeared to leave a shell-like callus on the ends of the fractured bone.

Figure FIG. 4..

Abnormal bone resorption during fracture healing in COX-2-selective NSAID-treated animals. Shown are fracture calluses from rofecoxib-treated rats at (A and B) 3 weeks and (C) 4 weeks postfracture. The orientation of panels A and B are the same with external callus on top and fracture site to the immediate left of the panel. (C) The external callus is to the left and the fracture site is at the immediate bottom of the panel. The specimens were embedded in polymethylmethacrylate, sectioned, and stained with van Gieson's picrofuchsin and Stevenel's blue so that bone is red, calcified cartilage is orange to red, cartilage is deep blue to purple, and other cell types are shades of blue. Original photographic magnification is indicated. CB, cortical bone; WB, woven bone; CC, calcified cartilage; Ca, cartilage; F, fracture site; ab, air bubble; M, area of magnification shown in panel B; and Oc, osteoclasts. The NSAID-treated rats often developed areas of high bone resorption at the cortical bone, fracture site, and external callus junction (M) as seen in panel A. The air bubble (ab) seen in panel A is an artifact of the polymethylmethacrylate embedding. At higher magnification, osteoclasts (Oc) can be seen lining the cortical bone surface of area M in the 3-week fracture callus as denoted by the arrows. Shown in panel C is an identical area of a 4-week postfracture callus as shown in panel B. The extent and area of bone resorption appears to be greater at 4 weeks postfracture and also often encompassed all surfaces of the cortical bone at the fracture site. Similar bone resorption patterns were seen in celecoxib-treated rats and to a lesser extent in the indomethacin-treated rats.

Celecoxib and rofecoxib treatment often caused a massive bone resorption event at the distal ends of the fracture bones, leaving what appear to be indentations into the callus (Figs. 3 and 4). The magnitude of this bone resorption phase is indicated by the large number of osteoclasts that were found on the femur periosteal surface at the distal ends of the fractured bone near the apparent indentation (Fig. 4). This was a reproducible observation. Although none of the no-drug (seven rats) or indomethacin- treated rats (seven rats) showed more than one area of osteoclastic resorption, 5 of 10 celecoxib- and 8 of 11 rofecoxib-treated rats (seven rats) had multiple areas of osteoclastic resorption at the distal ends of the fractured bone at 3 weeks and 4 weeks postfracture.

At 6 weeks and 8 weeks postfracture, the fractured femurs of the no-drug rats appeared to be healing normally with active remodeling of the cortical bone ends and fracture callus (not shown). Fractured femurs from indomethacin-treated rats also appeared to be healing at 6 weeks and 8 weeks postfracture with evident bridging and active remodeling (not shown). In contrast, no further healing was evident in the celecoxib- or rofecoxib-treated rat fractured femurs (not shown). The callus in the COX-2-selective NSAID-treated rats was smaller at 6 weeks and 8 weeks postfracture but the fracture gap was still clearly evident and often filled with fibrous tissue. These observations are consistent with our radiographic and torsional mechanical testing data showing that celecoxib and rofecoxib can inhibit fracture healing.

Complications associated with use of COX-2-selective NSAIDs

Pin slippage was a severe complication with as many as 30% (23 of 76) of the rofecoxib-treated rats being affected (Fig. 5). Pin slippage is dislodgement of the intramedullary stainless steel rod used to stabilize the fracture and permit the rat to weight-bear on the fractured femur. Once fracture stability was lost from pin slippage, the rat was killed because the rat could no longer weight-bear on the femur and because the callus would be reinjured and thus alter healing. Sixteen of the 23 rofecoxib-treated rats with pin slippage were killed before the experimental endpoint and without further analysis. The etiology of the pin slippage is unknown. Because animals with these complications were excluded, our final data set is skewed in favor of rats that had healed. Other complications included anesthesia death during weekly radiographs and infections that often excluded specimens from further analysis. It was found using a κ2 analysis to compare each experimental group value to the no-drug rat value that the pin slippage rate was significant for all NSAID treatment groups (p < 0.0001) and that the infection rate for the rofecoxib-treated rats also was significant (p < 0.0001).

Figure FIG. 5..

Experimental complications associated with NSAID treatment during fracture healing. Complications that necessitated the premature killing or resulted in the premature death of a rat during the course of these experiments were compiled and used to determine the effects of NSAID treatment on anesthetic death, infection, and pin slippage. Treatment groups and the total number of rats in each treatment group are indicated in the figure. Experimental treatment group values were compared with the no-drug values using a κ2 analysis. Significant differences are noted with an asterisk. As can be seen, pin slippage was by far the most common complication and was significantly different for each treatment group relative to the no-drug rats with p values of <1E−4, 1E−7, and 1E−24 for the indomethacin-treated, celecoxib-treated, and rofecoxib-treated rats, respectively. The rofecoxib-treated rats were found also to have a statistically significant higher infection rate as compared with the no-drug rats (p < 0.0001). Death from anesthesia was not different between groups.

Normal fracture healing fails in Cox2 null mice

The observations made using the COX-2-selective NSAID-treated rats do not distinguish between a specific effect on COX-2 and a nonspecific effect of the NSAIDs on fracture healing. Therefore, to address specifically whether fracture healing requires Cox1 or Cox2 gene function, femur fracture healing was assessed in Cox1 and Cox2 knockout mice.(37,38) Closed femur fractures were produced in three female Cox1−/− and three female Cox2−/− mice. The animals were examined radiographically immediately postfracture and then at 7, 10, 14, 21, 28, and 42 days postfracture. Fracture healing appeared to proceed normally in the Cox1−/− mice relative to our previous observations in outbred and inbred strains of mice (Manigrasso and O'Connor, unpublished data, 2001). In contrast, only a slight periosteal hard callus was detected in any of the Cox2−/− mice. As can be seen in Fig. 6, the apparent difference in fracture callus size was most obvious at 2 weeks postfracture when the Cox1−/− mice had formed a large fracture callus but little or no callus was evident in the Cox2−/− mice.

Figure FIG. 6..

Cox2 but not Cox1 is essential for normal bone fracture healing. The right femora of Cox1−/− and Cox2−/− mice were fractured and examined radiographically and histologically at 2 weeks postfracture. (A and D) Radiographs of fractured femurs from a Cox1−/− and a Cox2−/− mouse, respectively. Note the lack of mineralized tissue (X-ray dense) in the fracture callus region of the Cox2−/− mouse. (B and C) Sagittal section through the fractured femur of a Cox1−/− stained with Masson's trichrome stain (cell nuclei, purple; muscle and cytoplasm, red; collagen and bone, blue). (E and F) Sagittal section through the fractured femur of a Cox2−/− stained with Masson's trichrome stain. Note the presence of chondrocytes within the Cox2−/− callus but the lack of endochondral ossification relative to the Cox1−/− mouse fracture callus. Original photographic magnification is indicated. The Cox2−/− mouse fracture callus specimens are shown at higher magnification because the callus was smaller. B, Bone; C, chondrocytes and cartilage; E, area of endochondral ossification; M, area of intramembraneous bone formation; F, fracture site.

One mouse of each genotype was killed at 2 weeks postfracture and the fractured femur was examined histologically (Fig. 6). New bone and differentiating chondrocytes were abundant within the Cox1−/− callus, indicating that COX-1 activity is not essential for fracture healing. In contrast, there was a plainly evident lack of new bone formation in the Cox2−/− fracture callus with only some apparent intramembraneous bone formation occurring at the edges of the callus. Chondrocytes were observed throughout the Cox2−/− soft fracture callus. However, the Cox2−/− chondrocytes failed to form a mineralized matrix as evident by the radiolucency and histological appearance of the soft callus. The amount of endochondral ossification appeared greatly reduced in the Cox2−/− specimen. Although vascularization of the Cox2−/− fracture callus was observed, this parameter was not quantified but appeared less than that seen in the Cox1−/− mouse fracture callus. These observations clearly indicate that normal fracture healing and endochondral ossification are stopped or dramatically reduced in the Cox2− /− mice and confirm our observations made in the COX-2-selective NSAID-treated rats.


Using a standard rat closed femur fracture model, we have shown that COX-2-selective NSAID treatment can stop normal fracture healing and induce the formation of incomplete unions and nonunions. These observations are in distinct contrast to those observations we and others have obtained by treating rats with nonselective NSAIDs such as indomethacin when it was observed that fracture healing was delayed but not prevented.(24,39) Our observations suggest that COX-2 has an essential function during normal fracture healing and that COX-2-selective NSAID inhibition of prostaglandin synthesis stops normal fracture healing.

Consistent with the effects of COX-2-selective NSAIDs on rat femur fracture healing, mice homozygous for a targeted mutation in Cox2 but not Cox1 also showed inhibited fracture healing (Fig. 6). This excludes the possibility that any additional inhibitory activity against other cellular processes or proteins by celecoxib or rofecoxib is responsible primarily for negatively affecting fracture healing.(40–42)

The amount of rofecoxib used to treat the rats (200 mg/70 kg) in this study was approximately four times the nominal, maximum human daily dose of 50 mg that is used to manage acute pain. In contrast, the celecoxib dose used to treat rats in this study was in the recommended dose range for humans. The rats received 280 mg/70 kg body weight of celecoxib once per day, whereas the recommended human maximum daily dose of celecoxib is 200 mg twice a day. Additionally, the rats in this study received daily doses of each drug till the endpoint of the experiment, which is unlike common clinical scenarios when COX-2-selective NSAIDs are used to manage acutely pain, inflammation, and swelling after a fracture. In contrast, arthritis patients, in particular, do use COX-2-selective NSAIDs on a daily basis for extended periods. Based simply on animal body weight and drug dose and without accounting for pharmacokinetic variables, the indomethacin dose used (1 mg/kg) would be predicted to inhibit most COX-1 activity but only partially inhibit COX-2 activity; the celecoxib dose used should inhibit most if not all COX-2 activity and possibly inhibit some COX-1 activity; and the rofecoxib dose used should inhibit completely COX-2 activity but not affect COX-1.(43) The estimated plasma half-lives for celecoxib and rofecoxib in male rats after a single drug dose are ∼4 h and 5 h, respectively.(44,45) However, in humans the estimated elimination half-life for celecoxib is 11 h and the plasma half-life for rofecoxib is ∼10-17 h, depending on drug dose.(46,47) Unfortunately, these data were unknown to us when this study was initiated and consequently our rat drug dosing regime may actually be an underrepresentation of the COX-2 inhibition level over a 24-h cycle as that for humans receiving similar drug doses. Additionally, celecoxib and rofecoxib may have different inhibitory concentrations for rat versus human COX-2. Experiments are in progress to address directly the levels of COX-1 and COX-2 inhibition in the fracture callus of the drug-treated rats. Despite the pharmacokinetic variations between rats and humans receiving COX-2-selective NSAIDs, our data clearly indicate that these drugs have a dramatic negative effect on fracture healing in mammals, and, thus, caution in the use of these COX-2-selective NSAIDs in humans with bone fractures or after certain orthopedic surgical procedures may be warranted.

The fractured femurs of the celecoxib-treated rats had increased mechanical properties and healed as incomplete unions, rather than nonunions, in ∼50% of the rats tested (Tables 1 and 2). Had the experimental endpoint been extended, some of the celecoxib-treated rats may have gone on to heal their femur fractures. However, the delay in any such healing among the celecoxib-treated rats would have been significantly longer than in no-drug or indomethacin-treated rats.

After mechanically testing the celecoxib- and rofecoxib-treated fractured femurs, we observed that the fracture callus had a shell-like morphology. The periphery of the fracture callus was bone that sometimes partly bridged the fracture gap and thus formed an incomplete union in the celecoxib-treated rats. However, little or no bone was present between the peripheral bone of the callus and the original femoral cortical bone ends. Often, this space appeared to be filled with fatty marrow. Also strikingly apparent was the lack of new bone or primary bone healing at the cortical bone ends of the fracture site.

The indomethacin dose used in this study (1 mg/kg) was shown previously to delay fracture healing in rats and was used principally as a positive control.(26) Increasing indomethacin doses to levels that would completely inhibit COX-2 (and COX-1 activity) causes a steep increase in rat mortality from gastrointestinal bleeding.(25,48) Consequently, it would be difficult to compare directly the effects of COX-2-selective and traditional NSAIDs on fracture healing based solely on COX-2 inhibition levels. An alternative approach would be to measure prostaglandin levels within and around the fracture callus during healing in control and drug-treated rats and to then correlate those observations with different healing parameters. Currently, we are pursuing these experiments.

The abnormal osteoclastic response observed at the fracture site in the NSAID-treated rats is counter to experimental observations in which prostaglandins stimulate bone resorption and COX-2 function promotes osteoclast formation.(49,50) Lack of prostaglandins in the fracture callus could induce osteoclastic activity through an undescribed mechanism or perhaps through an indirect mechanism such as increased mechanical instability at the fracture site. An additional possibility is that the amount and/or repertoire of prostaglandins produced at the fracture site in the NSAID-treated rats is competent for inducing osteoclastic activity but insufficient for osteogenesis to proceed normally. We favor this last possibility because the relative short half-life of celecoxib and rofecoxib in rats should produce daily periods in which COX-2- dependent prostaglandin synthesis could occur and because a similar osteoclastic response was not observed in the Cox2−/− mouse fracture callus (Fig. 6). Pin slippage leading to fracture destabilization was the major morbidity complicating these experiments (Fig. 5). Although the etiology of this pin slippage is unknown, inhibition of COX-2 activity dramatically increased the incidence. One possibility is that an imbalance in osteoclast and osteoblast activity leads to pin destabilization in a mechanism similar to the large number of osteoclasts observed at the fracture site in the NSAID-treated rats. An additional possibility is that the early structural integrity of the NSAID-treated calluses is poor, which leads to simple mechanical dislodgement of the pin. This possibility correlates with our findings that inhibition of COX-2 activity delays and inhibits fracture healing.

Chondrocyte differentiation and persistence in the fracture callus appears to be altered by COX-2-selective NSAID treatment and by lack of Cox2 (Fig. 6). Cartilage that is evident in the 2-week fracture calluses of COX-2-selective NSAID-treated rats either disappears or is reduced dramatically in the 3-week and 4-week fracture calluses (Fig. 3). In control rats, the fracture has bridged or is almost bridged by 4 weeks postfracture because the cartilage present at early times had undergone normal endochondral ossification and is no longer evident. In contrast, cartilage within the indomethacin-treated fracture callus does persist at 4 weeks postfracture, indicating a reduced rate of endochondral ossification. These observations suggest that the chondrocytes within the COX-2-selective NSAID-treated rats deteriorated without forming a cartilage matrix in which endochondral ossification could occur. This phenomenon then leads to the development of fracture nonunions and incomplete unions. Similar histological observations were seen in the Cox2−/− mouse fracture in which chondrocytes present in the callus failed to form a mineralized matrix (Fig. 6). In support of this hypothesis, teratocarcinoma chondrocytes developed from Cox2−/− embryonic stem cells were found to be deteriorating, hypotrophic, and undergoing apoptosis.(51) Additionally, the Cox2−/− mouse fracture callus was much smaller than that from the Cox1−/− mouse, suggesting that mesenchymal cell proliferation and migration into the Cox2−/− mouse fracture site is reduced or that the mesenchymal cells proportionately differentiate into fewer chondrocytes. Thus, it would appear that COX-2 function is essential for normal progression of endochondral ossification during fracture healing.

There are several steps during endochondral ossification at which COX-2 could exert an essential regulatory function. Endochondral ossification is a complicated process that begins when chondrocytes mature to produce a cartilage matrix. Eventually, the chondrocytes become hypertrophic and undergo apoptosis as the cartilage matrix matures and becomes calcified. Osteoclasts partially resorb the calcified cartilage with concurrent angiogenesis at the site of endochondral ossification. Osteoblasts proliferate and differentiate on the calcified cartilage and begin forming new bone. Remodeling of the new bone subsequently occurs to increase the mechanical properties of the bone. Prostaglandins produced by COX-2 could be used to enhance osteoblast proliferation and differentiation.(6) Prostaglandins also may be necessary to promote terminal differentiation of chondrocytes and formation of the cartilage matrix. Such an effect on chondrocytes could occur through a direct effect on differentiation or indirectly by preventing premature apoptosis because inhibition of COX-2 function by COX-2-selective NSAIDs has been shown to induce apoptosis in cancer cell lines.(42) The potential effects of prostaglandins on osteoblasts and chondrocytes during endochondral ossification are not mutually exclusive. An additional possibility is that COX-2- dependent signaling occurs between osteoblasts and chondrocytes to initiate and maintain endochondral ossification. Prostaglandins also are known to promote osteoclast activity, which may be essential for the normal endochondral ossification process. However, our histological observations indicate an exuberant osteoclast response at the fracture site in the NSAID-treated rats; thus, we do not favor this potential mechanism. Angiogenesis also is inhibited by COX-2-selective NSAIDs, at least in certain experimental models,(40) and vascularization of the Cox2−/− mouse fracture callus appeared to be reduced. Thus, an additional possibility is that lack of angiogenesis precludes proper delivery of osteoclasts and osteoblasts to the cartilage matrix interface for continued endochondral ossification.

Our observations suggest that COX-2-selective NSAIDs may be effective in reducing or preventing pathological heterotopic ossification. One particular genetic disease for which COX-2-selective NSAIDs may be efficacious is fibrodysplasia ossificans progressiva (FOP).(52) Children afflicted with FOP develop debilitating heterotopic bone through an endochondral ossification pathway that appears to initiate at sites of inflammation.(53) Thus, the COX-2-selective NSAIDs may be useful in reducing inflammation and stopping endochondral ossification at presumptive heterotopic ossification sites in children with FOP.

Our data indicate that COX-2 function is essential for fracture healing. In contrast, adult mice homozygous for targeted mutation of Cox2 appear to have normally formed skeletons. Together, these observations show that fetal osteogenesis and fracture healing, although similar in many ways, are different and probably are initiated and maintained through different molecular mechanisms. Therefore, it would appear that Cox2 is the first gene to be identified that is specifically essential for fracture healing but not fetal osteogenesis. Targeted mutation of other mouse genes involved in prostaglandin synthesis and signaling should enable further analysis of the prostaglandin pathway(s) involved in fracture healing and skeletal biology in general.


We thank Chris Sabatino for help in gavaging rats and performing mechanical tests. We also thank Dr. Sylvia Christakos, Dr. Bob Harten, Dr. Russ Parsons, Dr. Mitch Reiter, and Dr. Heather Beam for helpful discussions and critical reading of this study. We thank Dr. Van Thompson (UMDNJ-New Jersey Dental School) for use of the HP-Faxitron. This work was supported by grants from the Foundation for University of Medicine and Dentistry of New Jersey (UMDNJ), the New Jersey Medical School (NJMS) Dean's Biomedical Research Fund, the New Jersey Chapter of the Arthritis Foundation, and the Orthopaedic Research and Education Foundation (to J.P.O.C.) and by generous contributions from the Department of Orthopaedics, Dr. Fred Behrens, Chair.