ARTICLE IN PRESS

Journal of Biomechanics 41 (2008) 1222–1228 www.elsevier.com/locate/jbiomech www.JBiomech.com

An improved murine femur fracture device for bone healing studies Joseph E. Marturanoa, Benjamin C. Clevelanda, Melissa A. Byrnea, Shannon L. O’Connellb,1, John J. Wixtedb,1, Kristen L. Billiara,b, a

Department of Biomedical Engineering, Worcester Polytechnic Institute, 100 Institute Road, Worcester, MA 01609, USA Department of Surgery, University of Massachusetts Medical School, 55 Lake Avenue North, Worcester, MA 01609, USA

b

Accepted 24 January 2008

Abstract Murine models are commonly used to investigate bone healing and test new treatments before human trials. Our objective was to design an improved murine femur fracture device and determine optimal mass and velocity settings for maximal likelihood of transverse fracture. Fracture reproducibility was maximized using an adjustable kinetic energy level, a novel mouse positioning system and an electromagnet striker release assembly. Sixty wild-type mice of 8–12-week-old male and female with a weight of 26.476.1 g were subjected to an experimental postmortem fracture in the left and right femur (n ¼ 120) using variable kinetic energy inputs. A best-fit prediction equation for transverse fracture was developed using multivariate linear regression. Transverse fracture was shown to correlate most highly with kinetic energy with a maximum likelihood at mv2 ¼ 292 where m is mass (g) and v is velocity (m/s). Model validation with a group of 134 anesthetized C57BL/6 mice resulted in a favorable transverse fracture rate of 85.8%. Simple modifications to existing fracture devices can improve accuracy and reproducibility. The results may assist researchers studying the effects of genetic modifications and novel treatments on boney healing in murine femur fracture models. Maintaining kinetic energy parameters within suggested ranges may also aid in ensuring accuracy and reproducibility. r 2008 Elsevier Ltd. All rights reserved. Keywords: Fracture healing; Biomechanics; Transgenic mouse; Bone

1. Introduction Debilitating skeletal bone fractures occur over five million times each year in the United States alone (Holstein et al., 2007b). The increasing life expectancy of the population creates an ever-increasing risk of fracture. To study bone healing in a controlled manner, pre-clinical small animal models have been developed and utilized extensively. The original fracture models from the 1970s and 1980s used the rat (Bonnarens and Einhorn, 1984; Jackson et al., 1970), and its popularity has extended to present-day studies (Schmidmaier et al., 2004). More Corresponding author at: Department of Biomedical Engineering, Worcester Polytechnic Institute, 100 Institute Road, Worcester, MA 01609, USA. Tel.: +1 508 831 5384; fax: +1 508 831 5541. E-mail address: [email protected] (K.L. Billiar). 1 Also at: Department of Orthopedic Surgery, University of Massachusetts Memorial Hospital, 55 Lake Avenue North, Worcester, MA 01655, USA.

0021-9290/$ - see front matter r 2008 Elsevier Ltd. All rights reserved. doi:10.1016/j.jbiomech.2008.01.029

recently, mouse models have sparked great interest due to their well mapped genome and ability to delete or mutate specific genes (Baldik et al., 2005; Cheung et al., 2003; Hiltunen et al., 1993), which affords the opportunity to observe the effect of an individual gene in the healing process (Manigrasso and O’Connor, 2004). Multiple bones have been studied in these models, with the most prevalent being the femur and tibia. Although the mouse tibia is easier to access than the femur, the tibia is not an ideal model for fracture studies because of its curved major axis that complicates mechanical testing and the minimal local soft tissue surrounding the bone (Holstein et al., 2007a; Manigrasso and O’Connor, 2004). Furthermore, the proximity of the fibula to the tibia may change healing rate if it accidentally fractures, which can occur at rates up to 30% (Thompson et al., 2002). For these reasons, the transgenic mouse femur is an exciting fracture model for modern orthopedic research. The most widely utilized small animal fracture device is the Bonnarens and Einhorn (1984) rat tibia fracture device.

ARTICLE IN PRESS J.E. Marturano et al. / Journal of Biomechanics 41 (2008) 1222–1228

Its simple gravity-driven three-point bending design is easy to construct, operate, and maintain. However, the device as originally described has several shortcomings for use with transgenic mice. First, while it is relatively simple to position the tibia of a rat centered between the two anvils, positioning the femur of a mouse is much more difficult due to the intra-torso location and small size. Another disadvantage of the device is the use of the reset spring, which experiences fatigue and may not create a reproducible fracture over time. Perhaps the most significant device limitation has been the absence of a systematic study to determine the ideal parameters for creating a transverse fracture in the mouse femur using a gravity-driven fracture device (Carmouche et al., 2005; Taguchi et al., 2005; Thompson et al., 2002). Taken together, these issues may result in undesirable fractures, which can be particularly detrimental when using transgenic mice that are typically difficult to obtain. The goal of this work is to develop a device and method to improve quality and reproducibility of experimental fractures in a murine femur model. Using the results of this study, researchers will be able to more efficiently study the effects of genetic modifications and novel treatments on boney healing in murine femur fracture models. 2. Materials and methods 2.1. Theoretical considerations Based on physics fundamentals, fracture mechanics, and bone biomechanics, the following four device parameters were considered: impact mass, impact velocity, ‘‘gap’’ between two fracture anvils in threepoint bending and ‘‘depth’’ that the striker is permitted to traverse past the top skin surface of the mouse femur. The impact mass and velocity are related in the equation of kinetic energy, E K ¼ 12 mv2 . The effect of kinetic energy load on cortical bone has been directly related to fracture type in previous research (McGee et al., 2004), and has long been known to have an important relationship with the release of strain energy and propagation of stress waves as developed by Mott (1948). By rearranging and solving the integrand in Mott’s strain energy relationship, the following equation is obtained: sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð2kÞE k E 2 V¼ (1) ra2 s2 where V is the speed of stress waves, k is a geometric material constant, Ek is the kinetic energy, E is the elastic modulus of the material, r is the density of the material, a is the crack length and s is the amplitude of stress waves. Recent investigations have demonstrated a critical stress wave velocity level for materials above which comminuted or oblique fracture is likely to occur due to insufficient material shifting time (Anderson, 2005). As kinetic energy is the only controllable variable in Eq. (1), strict attention to the total kinetic energy is essential to minimize probability of comminuted or oblique fracture. Two important geometric considerations are the ‘‘gap’’ between threepoint bending anvils and the ‘‘depth’’ of striker penetration. The gap distance is a primary determinant of the kinetic energy necessary for fracture. Increasing anvil separation distance decreases the energy required to fracture the bone. In addition, depth is a second determinant of necessary kinetic energy level and has previously been considered as an important fracture parameter (Jackson et al., 1970). Very low values of depth would require high kinetic energies, and maintaining a consistent

1223

depth is critical to accurately correlate kinetic energy to fracture type and avoid pin bending.

2.2. Device construction The development of two novel systems, a springless electromagnetic drop assembly and a mouse positioning system, satisfied the requirements of the four design parameters. 2.2.1. Electromagnet drop assembly The device in this study combines the traditionally separate impact mass and shaft and uses the shaft as the impact mass. The shaft travels on a dual linear bearing system and an electromagnet is situated above the shaft that connects to a circuit with a momentary switch to release the striker when fracture is desired (see Fig. 1). The mass of the shaft is adjusted using slotted weights attached to the upper threaded portion of the shaft. Impact velocity is set using a thumbwheel on a threaded portion of the shaft to different vertical heights above a stopblock; this thumbwheel acts as the stopping mechanism. The impact velocity is calculated assuming free-fall using V ¼ (2gh)1/2 where V is the impact velocity, g is the gravity constant, and h is the release height. This system provides an adjustable means for a single user to apply a specific and reproducible kinetic energy load into the mouse femur without the use of a reset spring. 2.2.2. Mouse positioning system A stainless steel half-pipe supporting the mouse torso swivels on an arc track with 1201 of allowable rotation (see Fig. 2). Two semi-elliptical sections are excised from the top surface of each anvil to provide tactile feedback to the user during femur positioning. Prior to fracture, a 30-gauge stainless steel pin is inserted in retrograde fashion through the distal femur which serves as a post-fracture fixation device (Carmouche et al., 2005; Holstein et al., 2007a, b; Manigrasso and O’Connor, 2004; Mullis et al., 2006; Taguchi et al., 2005). The pin is kept at a length such that approximately 5 mm is protruding from the knee. Since the threedimensional position of the pin is directly correlated with the position of the femur, the pin provides visual feedback to the user. Thus femur positioning is adjusted using a torso half-pipe, anvil notches and an intramedullary canal pin.

2.3. Fracture parameters The specific values for device parameters used in this study are based on previous findings. The impact mass and velocity ranges are generated from the kinetic energy level of Taguchi et al. (2005), and the impact mass to mouse weight ratio of van Griensven et al. (2002). The anvil gap is based on the gap-tibia length ratio from a recent tibia fracture study (Thompson et al., 2002) and an average femur length of 14.8 mm in a 30-g ICR outbred mouse (Manigrasso and O’Connor, 2004). The depth was chosen partially from Jackson et al. (1970) and from a pilot study (n ¼ 20 femurs, weight ¼ 25.372.74 g), which showed that the selected depth permitted the full range of break types was easy to reproduce, and still used the thumbwheel as the stopping mechanism based on audible conformation (data not shown). The ranges and values used are provided in Table 1.

2.4. Experimental design Sixty wild-type mice of various strains with an age range between 8 and 12 weeks were subjected to an experimental postmortem fracture in the left and right femur (n ¼ 120) using the fracture parameters from Table 1. Culled wild-type animals were sacrificed pre-fracture via CO2 inhalation followed by cervical dislocation in accordance with IACUC guidelines. Although the animals were sacrificed prior to the procedure, for the purpose of this study postmortem surgery was carried out in accordance with IACUC-approved fracture protocols.

ARTICLE IN PRESS 1224

J.E. Marturano et al. / Journal of Biomechanics 41 (2008) 1222–1228

Fig. 1. (Left) Computer scale model of murine fracture device with an adjustable mass shaft and stopping thumbwheel held by electromagnet above adjustable height base; striker impacts mouse femur held in place by mouse positioning system. (Right) Photograph of built prototype with electromagnet foot pedal switch (‘a’); dimensions are 3700  1200  1200 .

Fig. 2. (Left) Computer scale model of mouse positioning system; mouse torso rests in half-pipe and swivels on a 1201 arc track depending on desired fracture leg, femur is positioned perpendicular to striker using anvil notches and entire system is vertically adjustable using a four-post bolt system to adjust velocity at impact. (Right) Photograph of built mouse positioning system.

Table 1 Fracture parameters used in testing of final prototype; gap and depth were fixed for simplicity and mass and velocity varied to test for different levels of kinetic energy

Min R-2000 mammography film (Eastman Kodak Co., Rochester, NY, USA). The fracture types on the film were verified by an orthopedic surgeon (author Wixted) who was blind to the experimental grouping.

Fracture parameter

Range tested

2.5. Data analysis

Impact mass Impact velocity Anvil gap distance Striker penetration depth

350–650 g 0.98–1.31 m/s 6 mm Level with top anvil surface

The mouse weight, impact mass and impact velocity for a given fracture were recorded and associated with the output break-type. The break-type data were subjected to a best-fit parameter analysis in SAS statistical software (v9.1, SAS Institute Inc., Cary, NC, USA) to develop a mathematical prediction relationship. A total of six different multiple regression models were used with prediction variables including mouse weight, impact mass, impact velocity and kinetic energy. The equation with the most significantly correlated variance in an F-test and the least number of insignificant variables from a t-test with the coefficients against zero was chosen to predict break type with a statistical confidence level of 95%.

To simulate a live fracture test, fracture occurred within 10 min after sacrifice to minimize the stiffening effect of rigor mortis. Within 30 min of post-fracture, fracture type was verified using dorsal–ventral radiography; plain radiographs were obtained using a Faxitron MX-20 cabinet X-ray system (Faxitron X-ray LLC, Wheeling, IL, USA) and Kodak

ARTICLE IN PRESS J.E. Marturano et al. / Journal of Biomechanics 41 (2008) 1222–1228

2.6. Model validation To verify accuracy of the prediction model, an additional study was conducted with 134 live C57BL/6 inbred mice. This strain of mice was selected for its popularity in animal research, and because its average cortical femur cross-sectional area is included within one standard deviation of the mean value of 29 common inbred mouse strains (Wergedal et al., 2005). A single combination of impact mass and velocity was chosen using the results of the prediction model to fracture the right femur of anesthetized mice. The energy value selected was within the model’s 95th percentile for transverse fracture confidence, and all other fracture testing parameters were identical to the protocol used to develop the prediction model. Deep anesthesia was induced using an intraperitoneal injection of Ketamine at 100 mg/kg body weight and Xylazine at 10 mg/kg body weight. Resultant fracture types were evaluated using the radiography technique previously described.

1225

The weight of the mice used to derive the prediction model ranged from 18 to 41 g with a mean of 26.476.1 g. A histogram of mouse weight is shown in Fig. 3. An example of three femur fractures—one transverse, one oblique and one comminuted fracture—is provided in Fig. 4. The distribution of resulting fracture types from the energy ranges in Table 1 is given in Table 2. The six different multiple regression models of break type are provided in Table 3. The equation most significantly correlated to the data from the F-test and with the least number of insignificant parameters from the t-test is model (6), which corresponds to the best-fit transverse fracture prediction model. The values of C and b in model (6) are provided in the following: BT ¼ 0:292 þ 0:001mv2 ,

(2)

3. Results We found that the device performs in a reproducible manner, is simple to operate with a single user, and is capable of rapid adjustment. The shaft does not visually reverberate during impact of the thumbwheel and stopblock. Further, the cylindrical linear bearings eliminate rotation except when significant manual torque is applied.

where BT is the output break type (unitless), m the impact mass (g) and v the impact velocity (m/s); constant C, the first term on the right-hand side of the equation, is unitless and coefficient b has units of s2/m2g. Break type (BT) values are 1 (no fracture), 0 (transverse fracture) and 1 (comminuted/oblique fracture). Thus, setting BT ¼ 0 provides ideal impact mass and velocity combinations to achieve transverse fracture. The relationship in Eq. (2) is represented graphically in Fig. 5; the outer boundary lines represent a predicted 80% or 95% confidence of achieving a transverse fracture using the impact mass and velocity specified. The accuracy of this model equation was verified with the results of subsequent fracture study (Table 4). Table 2 Summary table of fracture types observed using range of energy inputs from Table 1; diaphyseal fractures defined as occurring within the middle one-third of the cortical bone segment; fracture type and pin angulation assessed by an orthopaedic surgeon blind to experimental grouping (author Wixted) using the described radiography technique

Fig. 3. Frequency histogram of mouse weights as measured post-sacrifice; the average weight is 26.476.1 g.

Fracture type

Diaphyseal

Non-diaphyseal

Total

Pin angle4101

Null Transverse Comminuted Oblique

– 65 10 29

– 2 1 6

7 67 11 35

0 0 4 11

Total

111 (93%)

9 (7%)

120

15 (13%)

Fig. 4. Radiograph examples of a transverse fracture (a), an oblique fracture (b), and a comminuted fracture (c); radiopaque horizontal object through femur is a 30-gauge stainless steel intramedullary canal pin. Fracture type determined by an orthopedic surgeon who was blind to experimental grouping.

ARTICLE IN PRESS J.E. Marturano et al. / Journal of Biomechanics 41 (2008) 1222–1228

1226

Table 3 Transverse fracture prediction models with significance of prediction equation as a whole to data and non-significant parameters if any No.

Mice used

Model equation

Model P-value

Non-significant parameters

1 2 3 4 5 6

All All All All All All

BT ¼ C+wW+gM+dV BT ¼ C+wW+gM+lV2 BT ¼ C+wW+dV+eM2 BT ¼ C+wW+bMV2 BT ¼ C+gM+dV BT ¼ C+bMV2

o0.001 o0.001 o0.001 o0.001 o0.001 o0.001

W, M W, M W, M2 W M None

Model P-value represents significance of predictive ability of independent variables on break type using an F-test. Mouse weight was not found to be a significant parameter in any model. Key: BT ¼ break type, C ¼ constant, W ¼ mouse weight, M ¼ impact mass, V ¼ impact velocity, (b, g, l, w, d) ¼ coefficients.

Fig. 5. Prediction contour plot of transverse fracture using all mice tested (n ¼ 120); shaded area represents 80% prediction confidence and darkest center represents 95% confidence; plot indicates appropriate impact mass and velocity levels to use with mice in the weight range of this study (26.476.1 g). Table 4 Testing of prediction model using consistent impact mass and velocity in anesthetized mice; fracture methods identical to the technique used to develop prediction model Mouse characteristics

Fracture type

Strain

Null

Weight Number Model

C57BL/6 inbred 25.9971.76 g 134 R. femur, live

Fracture input parameters Mass 350 g Velocity 0.99 m/s Model 95th percentile

Results 11 (8.2%)

Transverse Comminuted/oblique Total

115 (85.8%) 8 (6.0%) 134 (100%)

Additional information Diaphyseal Pin angulation o101 Survival 7-days postfracture

114 (85.1%) 126 (94%) 134 (100%)

4. Discussion The specific objective of this study was to adapt the popular Bonnarens and Einhorn (1984) gravity-driven fracture device for use with the mouse model and to improve on its functionality in three areas: femur positioning, impact velocity consistency, and development of the ideal energy inputs for transverse fracture. Our

results demonstrate that the device and method developed for this study produce reproducible transverse fractures of the mouse femur when the mass and velocity are selected appropriately. Femur positioning was improved by reducing the amount of positioning required by the user. The consistency of impact velocity was resolved by removing the return spring and combining the shaft and impact mass into a single unit. Finally, optimal kinetic energy inputs were determined using the systematic impact mass and velocity experiment in conjunction with mathematical modeling and verified with a subsequent validation study. The first parameter studied was the weight of mice used in the prediction model study. The weight distribution in Fig. 3 is significantly wider than most previous fracture studies (Cheung et al., 2003; Hiltunen et al., 1993; Holstein et al., 2007a, b; Manigrasso and O’Connor, 2004; Mullis et al., 2006; Thompson et al., 2002; van Griensven et al., 2002); this broad distribution was utilized to determine if mouse weight was a significant variable in outcome fracture. The resulting fracture distribution recorded in Table 2 was considered satisfactory to generate a prediction model since there was at least a 5% occurrence rate of each fracture type, over 50% of breaks were the prediction model’s target transverse fractures, and less than 10% of fractures did not occur strictly inside the target middle onethird diaphyseal region. It is interesting to note the low intramedullary canal pin bending rate compared with previous findings that also used our value of striker penetration depth (Table 1), often resulting in consistently deformed pins (Jackson et al., 1970). While we suspect that this discrepancy is due to a combination of pin material and geometric differences, we submit that further investigation is needed to fully discern the relationship between striker depth and pin angulation. The results of the prediction relationship using the variables in Eq. (2) describe the influence of kinetic energy in output fracture type in a murine femur model. The specific relationship of 0.292 ¼ mv2 where m (g) and v (m/s) were found to produce a transverse fracture with the greatest confidence. Mouse weight was not found to be a significant determinant of break type in any of the prediction models studied, as shown in Table 3, and does not need to be considered as an influential factor provided

ARTICLE IN PRESS J.E. Marturano et al. / Journal of Biomechanics 41 (2008) 1222–1228

that mouse weights are within the range (17–41 g) of this study. The reliance on kinetic energy in the prediction model, rather than other combinations of mouse weight, impact mass and impact velocity, correlates well with fracture mechanics theory by observing that level of kinetic energy is the major predicting factor in output fracture type (Eq. (1)). The accuracy of the prediction model was evaluated using a secondary study of 134 live C57BL/6 inbred mice with results presented in Table 4. Using the parameters suggested by the model resulted in a favorable transverse rate of 85.8%, which while not the predicted 95% rate, is a promising improvement over the 75.2% rate recently published using a modified Bonnarens and Einhorn fracture device (Manigrasso and O’Connor, 2004). The observed 14.9% rate of fractures occurring outside of the middle one-third of the diaphysis was troubling, but was attributed primarily to user positioning error rather than inaccuracy of applied energy. We consider the success rate of the validation study to be encouraging and testament to the accuracy of the developed prediction model and reproducibility of the device. There are several limitations to our study. First, the measurement of velocity was calculated assuming that the shaft was descending in frictionless free-fall through the two linear bearings, and thus the actual velocity at impact is slightly less than reported. Second, since the prediction model mice were sacrificed pre-fracture, the stiffness and strength of the femur may have been different than a living anesthetized animal. However, sacrifice should have a minor effect on fracture outcome, since muscle stiffness while anesthetized is similar to that 10 min postmortem. A study using Sprague-Dawley rats has shown that rigor mortis takes at least 2973 min to reach 10% maximal stiffness in red gastrocnemius muscle at 37 1C (Kobayashi et al., 2004). Further, our validation study demonstrates the applicability of the model to live mice. The third limitation was that the prediction results are only valid for the ranges of the parameters tested. Using parameters outside the ranges cited may result in inaccuracies of predicted break type; clearly very large or small animals would require a different level of kinetic energy. Lastly, the prediction model does not address inherent differences that are present in differing strains (Wergedal et al., 2005), since only culled post-sacrifice wild-type mice were used without regard to mouse strain. We consider the design as an improvement over previous devices used for generating fractures in a murine model. Limitations in previous devices when used with the mouse were minimized using the recommended kinetic energy levels and proposed modifications. Implementing the electromagnet/linear bearing assembly for consistent single-user operation, coupled with the mouse positioning system, enable reproducible and precise kinetic energy applications to the diaphysis of the femur. In addition to the novel design, the statistical model equation to predict transverse fracture and the resulting contour plot may be

1227

useful in reducing the chance of poor experimental fracture as demonstrated by the success rate of our prediction model validity study.

Conflict of interest statement We, the authors hereby declare that individually and as a collective we have not acquired any personal, professional or financial conflicts of interest related to the work. To the best of our knowledge the work was completed with the highest level of objectivity and interpretations were free from bias. The results and conclusions of this work were reached knowing that their outcome would not have any personal, professional or financial effects on the authors or the institutions at which they are employed. In addition, we agree that each of the authors has contributed to the following three areas: (1) the conception and design of the study, or acquisition of data, or analysis and interpretation of data, (2) drafting the article or revising it critically for important intellectual content, (3) final approval of the version to be submitted.

References Anderson, T, 2005. Fracture Mechanics: Fundamentals and Applications, third ed. CRC Press, Boca Raton, FL. Baldik, Y., Diwan, A.D., Appleyard, R.C., Fang, Z.M., Wang, Y., Murrell, G.A.C., 2005. Deletion of iNOS gene impairs mouse fracture healing. Bone 37, 26–32. Bonnarens, F, Einhorn, T, 1984. Production of a standard closed fracture in laboratory animal bone. Journal of Orthopaedic Research 2, 97–101. Carmouche, J, et al., 2005. Lead exposure inhibits fracture healing and is associated with increased chondrogenesis, delay in cartilage mineralization, and a decrease in osteoprogenitor frequency. Environmental Health Perspectives 113, 749–755. Cheung, K, Kaluarachi, K, Andrew, G, Lu, W, Chan, D, Cheah, K, 2003. An externally fixed femoral fracture model for mice. Journal of Orthopaedic Research 21, 685–690. Hiltunen, A, Vuorio, E, Aro, H, 1993. A standardized experimental fracture in the mouse tibia. Journal of Orthopaedic Research 11, 305–312. Holstein, J, Menger, M, Culemann, U, Meier, C, Pohlemann, T, 2007a. Development of a locking femur nail for mice. Journal of Biomechanics 40, 215–219. Holstein, J, et al., 2007b. Erythropoietin (EPO)—EPO-receptor signaling improves early endochondral ossification and mechanical strength in fracture healing. Life Sciences 80, 893–900. Jackson, R, Reed, C, Israel, J, Abou-Keer, F, Garside, H, 1970. Production of a standard experimental fracture. The Canadian Journal of Surgery 13, 415–420. Kobayashi, M, Takemori, S, Yamaguchi, M, 2004. Differential rigor development in red and white muscle revealed by simultaneous measurement of tension and stiffness. Forensic Science International 140, 79–84. Manigrasso, M, O’Connor, J.P., 2004. Characterization of a closed femur fracture model in mice. Journal of Orthopaedic Trauma 18, 687–694. McGee, A, Qureshi, A, Porter, K, 2004. Review of the biomechanics and patterns of limb fractures. Trauma 6, 29–40. Mott, N, 1948. Fracture of metals: theoretical considerations. Engineering 165, 16–18.

ARTICLE IN PRESS 1228

J.E. Marturano et al. / Journal of Biomechanics 41 (2008) 1222–1228

Mullis, B, Copland, S.T., Weinhold, P.S., Miclau, T, Lester, G, Bos, G, 2006. Effect of COX-2 inhibitors and non-steroidal anti-inflammatory drugs on a mouse fracture model. Injury 37, 827–837. Taguchi, K, Ogawa, R, Migita, M, Hanawa, H, Ito, H, Orimo, H, 2005. The role of bone marrow-derived cells in bone fracture repair in a green fluorescent protein chimeric mouse model. Biochemical and Biophysical Research Communications 331, 31–36. Thompson, Z, Miclau, T, Hu, D, Helms, J, 2002. A model for intramembranous ossification during fracture healing. Journal of Orthopaedic Research 20, 1091–1098.

Schmidmaier, G., Wildemann, B., Melis, B., Krummrey, G., Einhorn, T.A., Haas, N.P., Raschke, M., 2004. Development and characterization of a standard closed tibial fracture model in the rat. European Journal of Trauma 30, 35–42. van Griensven, M, et al., 2002. Polymicrobial sepsis induces organ changes due to granulocyte adhesion in a murine two hit model of trauma. Experimental and Toxicologic Pathology 54, 203–209. Wergedal, J, Sheng, M, Ackert-Bicknell, C, Beamer, W, Baylink, D, 2005. Genetic variation in femur extrinsic strength in 29 different inbred strains of mice is dependent on variations in femur cross-sectional geometry and bone density. Bone 36, 111–122.