The authors state that they have no conflicts of interest.
Article first published online: 18 JUN 2007
Copyright © 2007 ASBMR
Journal of Bone and Mineral Research
Volume 22, Issue 10, pages 1548–1556, October 2007
How to Cite
Uthgenannt, B. A., Kramer, M. H., Hwu, J. A., Wopenka, B. and Silva, M. J. (2007), Skeletal Self-Repair : Stress Fracture Healing by Rapid Formation and Densification of Woven Bone. J Bone Miner Res, 22: 1548–1556. doi: 10.1359/jbmr.0070614
Published online on June 18, 2007
- Issue published online: 4 DEC 2009
- Article first published online: 18 JUN 2007
- Manuscript Accepted: 15 JUN 2007
- Manuscript Revised: 25 MAY 2007
- Manuscript Received: 8 FEB 2007
- stress fracture;
- woven bone;
- fatigue loading;
- rat ulna;
- bone damage
Stress fractures of varying severity were created using a rat model of skeletal fatigue loading. Periosteal woven bone formed in proportion to the level of bone damage, resulting in the rapid recovery of whole bone strength independent of stress fracture severity.
Introduction: A hard periosteal callus is a hallmark of stress fracture healing. The factors that regulate the formation of this woven bone callus are poorly understood. Our objective was to produce stress fractures of varying severity and to assess the woven bone response and recovery of bone strength.
Materials and Methods: We used the forelimb compression model to create stress fractures of varying severity in 192 adult rats. Forelimbs were loaded in fatigue until the displacement reached 30%, 45%, 65%, or 85% of fracture. The osteogenic responses of loaded and contralateral control ulnas were assessed 7 and 14 days after loading using pQCT, μCT, mechanical testing, histomorphometry, and Raman spectroscopy.
Results: Loading stimulated the formation of periosteal woven bone that was maximal near the ulnar midshaft and transitioned to lamellar bone away from the midshaft. Woven bone area increased in a dose-response manner with increasing fatigue displacement. Whole bone strength was partially recovered at 7 days and fully recovered at 14 days, regardless of initial stress fracture severity. The density of the woven bone increased by 80% from 7 to 14 days, caused in part by a 30% increase in the mineral:collagen ratio of the woven bone tissue.
Conclusions: Functional healing of a stress fracture, as evidenced by recovery of whole bone strength, occurred within 2 wk, regardless of stress fracture severity. Partial recovery of strength in the first week was attributed to the rapid formation of a collar of woven bone that was localized to the site of bone damage and whose size depended on the level of initial damage. Complete recovery of strength in the second week was caused by woven bone densification. For the first time, we showed that woven bone formation occurs as a dose-dependent response after damaging mechanical loading of bone.
The mammalian skeleton has a remarkable capacity for self-repair of injuries ranging from complete fractures to partial “stress” fractures. Complete fractures typically heal through a process of endochondral bone formation, whereby rapid formation of granulation tissue gives way to a fibrocartilaginous “soft callus” that is gradually replaced by a woven bone “hard callus.” The process by which a stress fracture heals is less well understood.
Stress fractures are common in athletes and military recruits. Distance runners in the United States suffer >400,000 stress fractures annually, and incidence rates in U.S. military recruits are as high as 4–7%.(1) These injuries have a mechanical etiology involving repetitive skeletal loading and bone fatigue.(2) Bone, like all solid materials, is susceptible to fatigue, a process by which repetitive loading at subfracture stress levels leads to the formation and propagation of cracks and progressive loss of stiffness and strength.(3,4) Fatigue-induced microdamage can trigger localized bone remodeling that will repair the damage.(5,6) Current theories suggest that if the rate of damage accrual outpaces the rate of repair, microcracks will lengthen and coalesce to form a stress fracture.(2,7)
Less attention has been given to the other tissue response that occurs in a healing stress fracture—the direct formation of a periosteal callus of woven bone.(8,9) Periosteal woven bone formation increases bone cross-sectional area and thereby should reduce mechanical strains engendered by skeletal loading, which in theory will diminish or prevent the accrual of further fatigue damage.(7) Another consequence of woven bone formation is that the acute loss of strength attributed to the stress fracture can be reversed, effectively repairing the structure.(10)
In addition to its presence at the site of healing stress fractures, woven bone forms during skeletal development, fracture healing, and after high-strain mechanical loading. Woven bone is characterized by a lack of organization(11) and is believed to form under conditions where a rapid rate of matrix deposition is needed.(12) When a slower rate of deposition suffices, bone formation occurs as ordered lamellar bone. Studies of bone adaptation to nondamaging, dynamic loading indicate that lamellar bone formation is stimulated in a mechanical strain-dependent manner (i.e., the rate of formation increases with increasing strain magnitude).(13,14) In contrast, the limited available evidence suggests that woven bone formation after loading occurs as an “all or none” response when a threshold value of strain is exceeded.(14) It is unknown if the same “all or none” behavior characterizes woven bone formation at the site of a healing stress fracture.
Experimental approaches have been described for creating stress fractures under controlled loading conditions. By applying cyclic compression to the rat forelimb (fatigue loading), an ulnar stress fracture can be created that triggers the two main tissue-level responses seen in humans—intracortical resorption and periosteal woven bone formation.(10,15) Moreover, by monitoring the progressive increase in applied displacement that occurs during fatigue loading, the level of ulnar damage can be controlled.(16,17) Our objectives were to use the rat forelimb loading model to produce ulnar stress fractures of varying severity and to assess the woven bone response and the recovery of bone strength. Based on a recent study wherein skeletal fluoride uptake increased with stress fracture severity,(18) we hypothesized that the magnitude of woven bone formation increases in proportion to stress fracture severity, leading to the rapid recovery of bone strength regardless of the level of damage.
MATERIALS AND METHODS
Study design and in vivo loading
A total of 192 adult male rats (age, 4.5–5.5 mo; Fischer 344; Harlan) were randomly assigned in equal number to one of four loading groups (30%, 45%, 65%, or 85% of fracture) and one of two survival time-points (7 or 14 days). Under anesthesia (1–3% isofluorane), the right forelimb of each rat was loaded in axial compression through the carpus and olecranon using a materials testing machine (Model 1331/8500R; Instron, Norwood, MA, USA). The loading method is one we described in detail recently.(17) Briefly, a preload of 0.3 N was applied followed by a single bout of fatigue loading (2-Hz haversine) with a peak compressive force adjusted for body weight (0.055 N/g; mean: 20.6 N). Loading was terminated when the increase in actuator displacement reached 30%, 45%, 65%, or 85% of the average displacement at complete fracture (2 mm). The average numbers of loading cycles for the four loading groups were 1193 ± 1593, 2229 ± 3380, 2415 ± 3748, and 2340 ± 2878, respectively (p = 0.20). In our previous study,(17) we related the increase in actuator displacement to bone damage in ulnas from rats that were killed immediately after loading. Ulnar stiffness and strength decreased and crack length increased progressively with increasing actuator displacement. Thus, in our loading model, the actuator displacement provides a means to modulate the level of time-zero ulnar damage.
After loading, rats were dosed with an analgesic (0.05 mg/kg buprenorphine hydrochloride, IM) and returned to their cages. They were allowed unrestricted activities and ad libitum access to water and chow until death by CO2 asphyxiation 7 or 14 days after loading. Ulnas designated for histology were dissected immediately postmortem and fixed in formalin; ulnas for mechanical testing were stored at −20°C until use. This study was approved by our Animal Studies Committee.
Of the 192 rats we started with, 4 were excluded because of problems related to in vivo loading (one sustained a forelimb fracture, one did not reach the displacement limit in the allotted 3-h time, one died of an overdose of anesthesia, and in one case, the Instron machine malfunctioned during loading). Another nine rats were excluded because of technical errors (two ulna were fractured during dissection, seven were lost or mislabeled during histological processing). Remaining were ulnas from 179 rats.
Before mechanical testing or histology, we assessed the distribution of ulnar bone formation along the bone length using pQCT (XCT Research; Norland/Stratec) on a subset of samples (n = 9–12 per group). Eight transverse pQCT slices (0.07-mm in-plane resolution; 0.5-mm thickness) were taken at 2-mm intervals along the length of each ulna, centered near the mid-diaphysis (Fig. 1A). We determined bone area and BMC at each section using a simple threshold (500 mg/cm3) and the manufacturer's CALCBD routine. We computed relative differences in cross-sectional bone area and BMC between loaded (right) and paired control (left) ulnas.
Bone-forming surfaces were labeled in vivo by double fluorochrome injection. Rats (n = 10–13 per group) received a first injection of calcein green (20 mg/kg; Sigma) and a second injection of alizarin complexone (30 mg/kg; Sigma). Rats killed on day 7 were injected on day 0 (immediately after loading) and day 5, whereas rats killed on day 14 were injected on day 7 and day 12. Ulnas were dissected postmortem, fixed in 10% formalin for 48 h, and dehydrated in ethanol followed by xylene. They were infiltrated and embedded in methylmethacrylate (Fisher). Sections (100 μm thickness) were cut on a saw microtome (Leica SP1600) at locations D1, D4, and D7 (Fig. 1A) and mounted on glass slides. Sections were visualized with an inverted microscope at ×4 objective (Olympus). Analysis of the fluorescent labels and the amount of woven bone was performed using ImageJ (NIH). The length of periosteal bone surface labeled with woven bone (Wo.B.LS), double lamellar label (dLS), single lamellar label (sLS), or not labeled (nLS) was determined and expressed as a fraction of bone surface (BS; Fig. 2). In addition, woven bone area (Wo.B.Ar) and original cortical bone area (B.Ar) were measured. All loaded ulnas (n = 10–13 per group) were analyzed; eight contralateral ulnas (n = 2–3 per loading group) were analyzed and pooled for control data.
Before mechanical testing, ulnas from a subset of rats (n = 4–7 per group) were embedded in 1.5% agarose gel, positioned within a 16-mm acrylic specimen tube, and scanned by μCT at 16-μm isotropic resolution (μCT 40; Scanco Medical). Scans were collected across the central 14 mm and analyzed by a single operator (MK) using the manufacturer's software. Periosteal woven bone was readily distinguished from the preexisting cortical bone based on the relative differences in their CT attenuation. Woven bone area and apparent mineral density (calibrated to the manufacturer's hydroxyapatite [HA] mineral phantom) were determined at 2-mm increments.
Ulnas were loaded to fracture in three-point bending ex vivo to determine whole bone mechanical properties. Ulnas were thawed to room temperature and maintained in a moist state using saline. They were placed on supports spaced 15 mm apart and loaded transversely at their midpoint with a single ramp waveform at 0.5 mm/s to fracture. This bending mode produces tension laterally and compression medially and corresponds to the one that occurs during forelimb compression.(19) Force-displacement data were collected and analyzed to obtain ultimate force, stiffness, postyield displacement, and energy to fracture.
After mechanical testing, the fractured ends of a subset of loaded and control ulnas (n = 3–4 from the 65% and 85% displacement groups at 7 and 14 days) were sectioned using a diamond wafering blade (Buehler Isomet) to create an even surface. We analyzed samples from the two highest displacement groups because they had the greatest amount of woven bone. The cut surface was analyzed using a fiber optically coupled Raman microprobe (HoloLab Series 5000 Raman Microscope; Kaiser Optical Systems) with a frequency-doubled 532-nm Nd-YAG laser excitation source. The laser was focused on the surface at a spatial resolution of ∼1 μm and a power <500 μW. The spectral range of 100–4000 Δcm−1 was simultaneously detected with a CCD array detector with 2048 channels and a spectral resolution of 2.5 cm−1. Spectral acquisition time was 32 × 4 s per spectrum. Raman spectra were taken at 12 preselected and photographically documented positions along a medial-lateral traverse of the cut surface of each loaded ulna, resulting in six woven bone spectra and six cortical bone spectra. Similarly, Raman spectra were taken at six positions along each nonloaded control ulna. Each spectrum was deconvolved in the ranges of 700–1200 and 2750–3150 Δcm−1 to calculate the peak positions, widths, and areas characteristic of mineral and collagen vibrations.
pQCT analysis indicated that loading caused increases in BMC and bone area that were localized to the central region of the diaphysis, with a symmetrical response centered 1 mm distal to the mid-diaphysis (Fig. 1B). Cross-sections 6 mm proximal or 8 mm distal to the mid-diaphysis showed little or no change, whereas cross-sections 0 and 2 mm distal to the mid-diaphysis showed maximal increases. Based on these findings, we focused our histological analysis on sections 1, 4, and 7 mm distal to the mid-diaphysis.
Periosteal woven bone was observed on 92 of 93 ulnas examined histologically, in amounts that depended strongly on fatigue displacement and anatomical location. Woven bone near the mid-diaphysis increased with increasing levels of fatigue displacement (Fig. 3), indicating clearly that woven bone formation occurred as a dose-response to the level of bone damage. Notably, there was negligible new bone labeled after day 7, and quantitative analysis indicated that woven bone area did not change from 7 to 14 days (p > 0.05). Total bone area (woven plus original cortical bone) showed the same relative differences between groups as the woven bone area, because the original cortical area was not different between groups (data not shown). Thus, total bone area also did not change from 7 to 14 days. In addition, consistent with the pQCT results, woven bone area depended on longitudinal location, with progressively less woven bone at sections further from the mid-diaphysis (p < 0.05; Fig. 4A). Moreover, there was a consistent distribution of woven bone within the cross-sections, with more woven bone medially than laterally as we have noted previously.(10) The longitudinal and transverse distribution of woven bone corresponded to the pattern of cracking, which was consistent with previous descriptions.(15–17) Cracks were observed on histological sections from 61 of 93 ulnas, with 60 ulnas exhibiting cracks at D1, 6 at D4, and 0 at D7. Invariably, these cracks were located on the medial half of the cross-section (Fig. 3A).
Consistent with the displacement- and location-dependent differences in woven bone area, the fraction of lamellar and woven bone labeled surfaces changed significantly with fatigue displacement and longitudinal location (Fig. 4B). For ulnas in the lowest displacement group, new woven bone covered 35–50% of the bone surface 1 mm distal to the mid-diaphysis, whereas lamellar bone formation occurred over 40–50% of the surface. In contrast, ulnas from the three highest damage groups had 80–90% of the periosteal surface covered with woven bone and 8–18% with new lamellar bone. Unlike the progressive displacement-dependent increase in woven bone area, the fraction of woven bone-labeled surface reached a plateau at 45% fatigue displacement. Thus, once the fraction of the periosteal surface covered with woven bone approached 90%, further increases in woven bone area between groups were achieved by the formation of a thicker layer of woven bone. Regarding the effect of longitudinal location, bone formation shifted from woven to lamellar with increasing distance from the mid-diaphysis.
We performed bending tests to determine the recovery of ulnar strength and stiffness after the creation of a stress fracture. We had previously assessed changes in ulnar mechanical properties on day 0 and reported a progressive loss of strength and stiffness with increasing fatigue displacement(17); these time-zero data were combined with new data from ulnas collected 7 and 14 days after fatigue loading. Together, the data revealed a partial recovery of strength by day 7 and a complete recovery by day 14 (Fig. 5). Notably, in contrast to the displacement-dependent loss of strength at day 0, there were no differences in the strength of loaded ulnas between displacement groups on day 14. Moreover, the initial loss of stiffness (which ranged from 15% to 90%) was almost completely recovered by day 14, and postyield displacement and energy to fracture were recovered such that they were either not different or improved versus controls on day 14 (Table 1).
The large increases in mechanical properties between 7 and 14 days were in apparent contradiction to the lack of increase in woven bone area during this time. To determine the material basis for the recovery of strength and stiffness after day 7, we measured woven bone mineral density using μCT. Woven bone density increased dramatically from 7 to 14 days (average 80%, p < 0.001; Table 2), but did not vary greatly with the level of fatigue displacement. In contrast, and consistent with the histomorphometric findings, woven bone area did not change with time (p = 0.11) but did depend strongly on fatigue displacement (p < 0.001).
Finally, we used laser Raman spectroscopy to assess whether the increases in apparent mineral density of woven bone between days 7 and 14 were caused in part by increased mineralization of the solid bone tissue. Analysis of Raman spectra for woven bone at day 7 revealed a mineral:collagen ratio (960/2940 and 960/1003 Δcm−1 peak areas) ∼40% less than for control cortical bone (p < 0.05; Fig. 6; Table 3). Mineralization of woven bone increased from 7 to 14 days (p < 0.05), although 14-day woven bone was still ∼20% less mineralized than control bone (p < 0.05). Other notable differences between woven bone and control spectra included greater intersite heterogeneity within the same bone, increased width of the mineral peaks, and a downshift of ∼3 Δcm−1 in the position of the 2940 Δcm−1 collagen peak.
Periosteal hard callus is the hallmark of a healing stress fracture,(8,9) yet the factors that regulate the formation of this woven bone callus are poorly understood. We created stress fractures of varying severity using a rat model of fatigue loading to test the hypothesis that the magnitude of woven bone formation increases in proportion to stress fracture severity. Our findings support this hypothesis and show clearly that the magnitude of the woven bone response was scaled to the level of applied fatigue displacement and thus to the level of initial bone damage. To our knowledge, this is the first evidence that woven bone formation occurs in a dose-response manner to a mechanical stimulus. This result is consistent with our recent study that found increased skeletal uptake of fluoride in proportion to stress fracture severity using the same loading model,(18) although in that study, we could not discriminate bone formation from damage and/or vascular responses. Our conclusion that the woven bone response is scaled to the severity of the stress fracture indicates that woven bone formation is a well-regulated response to skeletal injury. This extends previous studies that have described a role for woven bone in the adaptive response to elevated mechanical strain.(14,20)
It is well documented that lamellar bone formation occurs in a dose-response manner after application of cyclic mechanical strain of increasing magnitude.(13,14) Few investigators have focused on quantification of a woven bone response, although results from one study indicated that woven bone formation occurred as an “all or none” response once a threshold value of strain was exceeded.(14) There are several differences between our study and the previous study(14) that might explain our different conclusions. First, we used the forelimb loading model rather than the tibial four-point bending model; the “all or none” response may be unique to the tibial bending model. Second, we applied a single bout of fatigue loading that produced measurable bone damage, whereas the earlier study applied multiple bouts (36 cycles/day; 2 wk) that did not result in detectable damage. Third, the earlier conclusion was based on a measure of woven bone surface rather than woven bone area. We observed that woven bone surface increased from the 30% to the 45% displacement groups, but did not increase further because it had already reached ∼90% (Fig. 4B). In contrast, woven bone area increased progressively across all four displacement groups, suggesting that this measure may be a more sensitive indicator of a dose-response. Taken together, the woven bone surface and area data from our study indicate that once woven bone formation was activated on nearly the entire periosteal surface, the area of woven bone could increase further by the deposition of a thicker collar of woven bone. However, regardless of the thickness of the woven bone layer, the space that is occupied by woven bone is essentially established by day 7, as evidenced by the diffuse uptake of the fluorochrome labels given on days 5 and 7, with negligible new bone labeled at day 12.
We also hypothesized that the woven bone response would lead to the rapid recovery of bone strength regardless of the level of damage. Our findings support this hypothesis and show that the damage-dependent woven bone response resulted in nearly complete functional healing within ∼2 wk for all damage levels. The recovery of whole bone strength such that loaded ulnas from the four displacement groups were not different from control (and were not different from each other) at 14 days is particularly impressive, because the initial reductions in bone strength in loaded ulnas in these groups ranged from 10% to 60%. The recovery of other mechanical properties (stiffness, fracture energy, postyield displacement) was less uniform across the displacement groups, with evidence that the higher displacement groups recovered at a slower rate than the lower displacement groups. Additional studies are needed to confirm this finding and determine whether there is a functional significance of having different rates of recovery of these different properties.
The pattern of woven bone formation in the rat ulnar fatigue model corresponds closely with the pattern of bone damage (Fig. 7). Detailed descriptions of the damage/crack distribution has been reported previously by us(17) and others.(15,16) Briefly, we observe cracks only on the medial half of the cross-section, in a region near the midpoint of the ulna. The appearance of these cracks is consistent with a shear fracture, where the fracture plane is ∼45° from the direction of loading. The association between the spatial pattern of bone damage and bone formation suggests that the cracks in some way are the stimulus for the periosteal woven bone response. In further support of this notion, we observed that the amount of bone formation increased with increasing fatigue displacement, which we have previously established is directly related to the severity of cracking both within the cross-section and along the longitudinal direction.(17) Thus, we propose that the stress fracture itself is a primary stimulus that dictates the location and extent of the woven bone response.
Our study provides novel data on the relative contributions of bone size and mineralization to the recovery of bone strength after a stress fracture and highlights the time-dependent interplay between size, mineralization, and strength. We attribute the partial recovery of strength in the first 7 days to the rapid formation of a collar of woven bone, whose size depends on the level of initial damage. However, because bone area did not increase significantly from 7 to 14 days, the increases in ulnar strength that occur in the second week cannot be attributed to changes in bone size. Moreover, we do not attribute the recovery of mechanical properties to intracortical remodeling and repair of bone damage, based on the fact that intracortical osteoclasts take ∼10 days to appear in this model.(5,10,21) Rather, we attribute the increases in bone strength in the second week to woven bone densification. The apparent mineral density of the woven bone increased by 80% from 7 to 14 days, caused in part by an increase of 30% in the mineral:collagen ratio of the bone tissue. By day 14, the woven bone tissue was still relatively hypomineralized compared with cortical bone, but other spectroscopic features of the mineral and collagen of 14-day woven bone were equivalent or only modestly different from mature cortical bone. The rapid maturation of woven bone tissue is comparable to the rapid chemical and mechanical maturation reported after lamellar bone deposition.(22)
There are several limitations to our study. First, the creation of a stress fracture by a single bout of loading on the order of 1 h does not replicate the time-course of the development of a clinical stress fracture. Most stress fractures likely develop over a period of days or weeks,(23) and the prevailing view is that they are not caused solely by fatigue damage but by an interplay between fatigue damage, damage-initiated bone remodeling, and localized loss of bone stiffness.(2,7) Although our model did not allow for remodeling activity before the creation of the stress fracture, it is unclear whether a prior remodeling state would affect the periosteal response, which was the focus of our study. A second limitation is that we cannot separate the effects of dynamic strain from those of bone damage (the actual stress fracture). Inherent in a fatigue test is that, as the displacement and damage increase, the dynamic strain at the site of damage also increases. Thus, the dose-response increase in woven bone area we observed may be caused by greater dynamic strains and/or more severe stress fractures in the higher displacement groups. Nonetheless, the use of dynamic loading represents a physiologically relevant scenario in which dynamic strain and bone damage are present together at the site of a stress fracture. Regardless of the physical stimulus that triggers the woven bone response, the key discovery remains that woven bone formation in association with a healing stress fracture is not an “all or none” response but is modulated by mechanical factors.
One clinical implication of our findings is that the size of the periosteal callus may be an indicator of stress fracture severity. Although this may seem intuitive, our study provides quantitative data to support the concept. A second implication is that the severity of the stress fracture may not be indicative of the time needed for functional healing. This seems to contradict available guidelines for clinical treatment, which indicate that the severity of bone damage (as inferred by the size and intensity of scintigraphic uptake on a bone scan) dictates the prescribed rest period.(24) One reason for the discrepancy between clinical recommendations and our results is that the functional healing we measured may precede the relief of pain that is the basis for clinical management. Alternatively, when a bone sustains microdamage but not a true stress fracture, it may heal only by the slower process of intracortical remodeling. Paradoxically, only when bone damage progresses to the point of a stress fracture may the woven bone “repair” process become activated and lead to rapid functional healing.
In summary, our study showed the self-repair process by which whole bones regain their structural integrity within 2 wk after sustaining stress fractures of varying severity. In the first week, woven bone forms at the site of the stress fracture in proportion to the level of initial damage. The new woven bone is hypomineralized compared with normal cortical bone but nonetheless leads to a partial recovery of strength and stiffness. During the second week, the area of woven bone does not increase further, but the woven bone density nearly doubles as the voids begin to fill and the solid phase of the woven bone becomes more mineralized. Thus, we attribute the partial recovery of whole bone strength in the first week after a stress fracture to the rapid formation of a collar of woven bone that is localized to the site of bone damage and whose size depends on the level of initial damage; full recovery of strength in the second week is caused by the densification of this woven bone.
This study was funded by a grant from the NIH/NIAMS (R01 AR050211).
- 12001 Incidence and prevalence of stress fractures in military and athletic populations. BurrDB, MilgromC (eds.) Musculoskeletal Fatigue and Stress Fractures. CRC Press, Boca Raton, FL, USA, 1–14.
- 22001 Bone fatigue and remodeling in the development of stress fractures. BurrDB, MilgromC (eds.) Musculoskeletal Fatigue and Stress Fractures. CRC Press, Boca Raton, FL, USA, 161–182.
- 72001 The role of bone remodeling in preventing or promoting stress fractures. BurrDB, MilgromC (eds.) Musculoskeletal Fatigue and Stress Fractures. CRC Press, Boca Raton, FL, USA, 183–201.
- 81963 Histogenesis of stress fractures. J Bone Joint Surg Am 45: 1542., , , ,
- 92001 The histological appearance of stress fractures. BurrDB, MilgromC (eds.) Musculoskeletal Fatigue and Stress Fractures. CRC Press, Boca Raton, FL, USA, 151–159., ,
- 112004 Orthopaedic Pathology, 4th ed. Mosby, Edinburgh, UK.
- 242001 Early diagnosis and clinical treatment of stress fractures. BurrDB, MilgromC (eds.) Musculoskeletal Fatigue and Stress Fractures. CRC Press, Boca Raton, FL, USA, 295–303.,