Strain Gradients Correlate with Sites of Exercise-Induced Bone-Forming Surfaces in the Adult Skeleton


  • Stefan Judex,

    1. McCaig Centre for Joint Injury and Arthritis Research, Departments of Mechanical Engineering and Surgery, University of Calgary, Calgary, Alberta, Canada
    Search for more papers by this author
  • Ted S. Gross,

    1. Department of Orthopaedic Surgery, University of Cincinnati, Cincinnati, Ohio, U.S.A.
    Search for more papers by this author
  • Ronald F. Zernicke

    Corresponding author
    1. McCaig Centre for Joint Injury and Arthritis Research, Departments of Mechanical Engineering and Surgery, University of Calgary, Calgary, Alberta, Canada
    • Department of Surgery, McCaig Centre for Joint Injury and Arthritis Research, University of Calgary, 3330 Hospital Drive NW, Calgary, AB, T2N 4N1 Canada
    Search for more papers by this author

  • Portions of this research have been presented at the annual meetings of the Canadian Society of Biomechanics (1996) and the American Society for Bone and Mineral Research (1996).


Physical activity is capable of increasing adult bone mass. The specific osteogenic component of the mechanical stimulus is, however, unknown. Using an exogenous loading model, it was recently reported that circumferential gradients of longitudinal normal strain are strongly associated with the specific sites of periosteal bone formation. Here, we used high-speed running to test this proposed relation in an exercise model of bone adaptation. The strain environment generated during running in a mid-diaphyseal tarsometatarsal section was determined from triple-rosette strain gages in six adult roosters (>1 year). A second group of roosters was run at a high speed (1500 loading cycles/day) on a treadmill for 3 weeks. Periosteal surfaces were activated in five out of eight animals. Mechanical parameters as well as periosteal activation (as measured by incorporated fluorescent labels) were quantified site-specifically in 12 30° sectors subdividing a mid-diaphyseal section. The amount of periosteal mineralizing surface per sector correlated strongly (R2 = 0.63) with the induced peak circumferential strain gradients. Conversely, peak strain magnitude and peak strain rate were only weakly associated with the sites of periosteal activation. The unique feature of this study is that a specific mechanical stimulus (peak circumferential strain gradients) was successfully correlated with specific sites of periosteal bone activation induced in a noninvasive bone adaptation model. The knowledge of this mechanical parameter may help to design exercise regimens that are able to deposit bone at sites where increased structural strength is most needed.


THE DYNAMIC NATURE OF BONE is reflected in its ability to generate morphological changes in response to altered mechanical loading. Physical exercise can elevate strain levels in weight-bearing bones,1–3 and thus it has been suggested that exercise regimens may serve to mitigate osteoporosis by enhancing bone mass in the mature skeleton. The specific osteogenic effects of different exercise regimes (e.g., running, swimming, weightlifting) on the adult skeleton, however, have been equivocal. While cross-sectional studies suggest that long-term exercise may elevate substantially the mass of weight-bearing bones in humans,4,5 short-term longitudinal studies demonstrate only negligible or modest increases.6–8 This may reflect the limited sensitivity of some measuring techniques to detect focal bone adaptation as well as the lack of knowledge of the specific mechanical parameter(s) responsible to initiate new bone formation.

At the tissue level, several mechanical parameters have been proposed to initiate bone formation including, but not limited to, strain magnitude,9 strain frequency,10 strain history related parameters,11 strain rate,12 strain energy density,13 and circumferential strain gradients.14 Most of the relations between mechanical parameters and bone formation (at least those that were observed experimentally) were derived for a given cross-section (i.e., the larger the specific aspect of the mechanical stimulus, the larger the response of the bone). Although it has been proposed that relations determined for a cross-section may also be valid for any given volume of bone, this has not been confirmed. To test this experimentally, the distribution of the mechanical signal could be associated with the distribution of the response across a bone section. If such a site-specific relation exists between a particular mechanical parameter and changes in bone morphology, then the knowledge of this specific osteogenic component of the mechanical environment may provide information about a mechanism by which bone cells perceive their mechanical environment. Also, if a mechanical parameter consistently predicts the specific sites of bone formation in different models, then prophylactic exercise programs may be designed that deposit bone at sites where additional structural strength is required to reduce the risk of fractures.

Recently, gradients of longitudinal normal strain (in particular, peak circumferential strain gradients) have been found to be correlated highly with specific sites of bone formation induced by controlled exogenous loading.14 Strain gradients reflect differential deformations across a volume of tissue in a given direction. Physiologically, strain gradients are relevant as they generate pressure differentials within bone and, thereby, may contribute to fluid flow in bone.15 Here, we examined whether circumferential strain gradients of longitudinal normal strain initiated new bone formation in an exercise model of bone adaptation. Circumferential strain gradients induced by increased gait speed provide a focal mechanical stimulus,2 because their magnitudes are nonuniform in distribution at a given cross-section in the rooster tarsometatarsus (TMT) and increase significantly in a nonuniform fashion with increased locomotor speed. TMT mid-diaphyseal bone formation in response to increased physical activity is also focal in nature because the deposition of tissue is limited to specific sites within a mid-diaphyseal section.16,17 Consequently, we hypothesized that sites of bone-forming surfaces stimulated by high speed running would be significantly correlated with maximal circumferential strain gradients.


Experimental procedures

Twenty-two, skeletally mature, White Leghorn roosters were divided into three groups. The first group of six 1-year-old animals previously acclimatized to treadmill running were strain gaged, and the mid-diaphyseal distribution of selected mechanical parameters was determined. A second group of eight, 60- to 80-week-old animals ran (1.66 m/s) on the treadmill for 3 weeks, 6 days/week, 9 minutes/day (∼1500 step-cycles/day). The speed of 1.66 m/s reflected a balance between a gait speed that significantly elevated circumferential strain gradient magnitudes2 and a speed at which roosters were capable of running continuously for 9 minutes. Prior to starting the exercise regimen, animals were acclimatized to treadmill running for 2 weeks. Calcein (Sigma Chemical Co., St. Louis, MO, U.S.A.) was administered (20 mg/kg, intravenously) on days 7 and 21 of the exercise regimen to reveal sites of periosteal activation. Animals were killed 2 days after the second calcein injection with a 1 ml intravenous injection of Euthanyl (MTC Pharmaceuticals, Cambridge, ON, Canada). We then correlated the specific sites of mineralization observed in this group of roosters to the distribution of mechanical parameters determined from the group of strain gaged roosters. The third group of eight, 60–80 week roosters served as (sedentary) controls and received calcein injections but no treadmill exercise. All roosters were group-housed in pens (3 × 4 m) and received the same standard diet and water ad libitum.

Strain gage procedures

In the first group of six roosters, triple rosette strain gages (FRA-1-11-1L, Tokyo Sokki Kenkyujo Co., Japan) were attached to the medial, anterior, and lateral aspects of the left TMT mid-diaphysis using cyanoacrylate adhesive (Krazy Glue, Borden Co., Willowdale, ON, Canada). After a 24-h recovery, each animal ran on a treadmill at 1.66 m/s while data were sampled (200 Hz) for 10-s periods and filtered at 1 kHz. Postmortem, the orientation of the rosette strain gages relative to the longitudinal bone axis was obtained from microradiographs (Cabinet X-Ray System, Faxitron X-Ray Corp., Buffalo Grove, IL, U.S.A.). Using a diamond wafer saw, mid-diaphyseal transverse TMT sections (300 μm) were cut from the left legs, and microradiographs of these sections were scanned on a flatbed scanner. The locations of the strain gages were obtained directly from the cross-sectional microradiographs. Custom software calculated values of longitudinal normal strain and longitudinal shear strain at each gage site for each sampled time point.

Determination of the distribution of mechanical parameters

The distribution of longitudinal normal strain acting upon a mid-diaphyseal cross-section was quantified using linear beam theory.18 This method models the TMT mid-diaphysis as a prismatic beam subjected to axial loading and bending moments; it is assumed that points in a plane passing through the unloaded bone remain in this plane during loading. The relative accuracy of beam theory to determine bone strain distributions during functional loading has been verified experimentally18 and by a finite element model.19 The orientation of the neutral axis during running was expressed with respect to an axis parallel to the anterior cortex (Fig. 1A). All calculated angles were averaged across three consecutive gait cycles in each animal.

Figure FIG. 1.

(A) Mid-diaphyseal TMT cross-section subdivided into 12 equal-angle sectors with the centroid as an origin. Mechanical parameters were calculated for the periosteal center of each sector. γ is the angle that was used to express the orientation of the neutral axis. M, medial; A, anterior; L, lateral. (B) For each 30° sector, circumferential strain gradients of longitudinal normal strain were computed as the difference in longitudinal normal strain magnitude (Δε) between the borders of each sector, normalized to the linear distance (d) between these points.

Pixel-counting routines (PV-Wave, Visual Numerics, Houston, TX, U.S.A.) were used to divide the binary images of the cross-sections into 12 30°-angle sectors with the centroid as an origin (Fig. 1A). Note that all strain gradients defined in this manuscript are strain gradients of longitudinal normal strain. Circumferential surface strain gradients were calculated in each 30°-sector as the absolute difference between longitudinal normal strain on the periosteum of the borders of each sector normalized to the linear distance between these points (Fig. 1B). Radial strain gradients were determined, in analogy to circumferential strain gradients, between the midpoints of the periosteal and endocortical surface in each sector.

The distribution of longitudinal strain gradients acting on periosteal surfaces of the mid-diaphysis was estimated between two transverse sections, 3 mm above and below the gage section. In each sector, the magnitude of the longitudinal surface strain gradient was determined by longitudinal changes in strain magnitude and periosteal surface geometry. Changes in strain magnitude were estimated by extrapolating longitudinal normal strain distributions from the strain gaged section via linear beam theory. Briefly, the forces and moments acting on the gage section were computed and increased (decreased) by 4% for the sections 3 mm above (below) the gage section. The scaled forces and moments were then used to calculate the strain distribution in these sections. Longitudinal changes in mid-diaphyseal periosteal surface geometry were determined from a different set of six age-matched roosters from the same cohort as the original strain gaged roosters since the mid-diaphyses from strain gaged roosters were not intact any more. TMTs from those roosters were embedded in polymethyl-methacrylate (PMMA) and sectioned at the mid-diaphysis (three sections per bone at 3 mm intervals). Radiographs of the sections were taken and scanned. Images were then aligned according to groves in the PMMA blocks, and geometrical differences in periosteal midpoints of each sector were calculated between the proximal and distal section.

To contrast strain gradients with other mechanical parameters, we also determined peak longitudinal normal strain and peak strain rate at the periosteal center of each sector. Peak values of all mechanical parameters were averaged across three consecutive step cycles for each animal. We determined the peak magnitude of each mechanical parameter that occurred in each sector during the gait cycle as we assumed that this would be the single most important factor for the potential activation of bone cells.

Finite element modeling (shear strain distributions)

The distribution of longitudinal shear strain across a mid-diaphyseal section was determined by modeling the TMT as a prismatic cantilever subjected to end shearing forces and a torsional moment. Finite element meshes were developed in PATRAN (MacNeal Schwendler Corp., Costa Mesa, CA, U.S.A.) for each strain gaged TMT consisting of 2460–2760 isoparametric 20-node brick elements arranged in 20 longitudinal and 3 radial layers. ABAQUS (Hibbit, Karlson & Sorenson Inc., Newark, CA, U.S.A.) was used to determine initially the shear strain distribution for an arbitrary torsional moment and end shearing forces. Subsequently, the predicted and the recorded strain values were compared at all three gage sites and the torsional moment and end shearing forces were scaled.20 Longitudinal shear strain acting on a mid-diaphyseal section was expressed as the vector sum of εxz and εyz at the periosteal center of each of the 12 30°-sectors (see Fig. 7 for orientation of coordinate system).

Figure FIG. 7.

Distribution of peak longitudinal periosteal shear strain (mean ± SE) contrasted against the distribution of mineralizing surface (mean ± SE) demonstrating a lack of correlation (R2 = 0.06). A sample finite element mesh of a middiaphyseal section is presented in the inset.


Cross-sections (300 μm) were cut from left and right mid-diaphyseal TMTs of the exercised group as well as the control group, and consequently, the location of these sections coincided with the location of strain gage attachments in the first group of animals. Sections were hand-ground to 100 μm, and microradiographs were taken. Microradiographs were enlarged and scanned on a stereomicroscope (Stemi 2000-C, Carl Zeiss Jena GmbH, Jena, Germany) with a video camera attached to it. Edge detection programs were used to binarize the image, and pixel counting routines calculated bone cross-sectional area and second moments of inertia about the bending axis imposed during running (as determined from the group of strain gaged animals) as well as about the axis transverse to the bending axis.

Sections were then ground to 60–70 μm and examined under an epifluorescent microscope at magnifications of up to 800 times. Photographs of each quarter of the section were taken, scanned on a flatbed scanner, and assembled using image software (Photoshop, Adobe Systems Inc., San Jose, CA, U.S.A.). Using the method described above, sections were subdivided into 12 sectors. The beginning and end of each periosteal fluorescent label was determined interactively on the monitor. Custom software (PV-Wave) quantified the fluorescently labeled periosteal surface relative to the total periosteal surface within each sector. Symmetrical gait of the bipedal birds was assumed, and thus, histomorphometrical data of the left and right TMT were averaged for each animal.


Linear regression analysis determined the R2 values for the correlations between the distribution of each mechanical parameter and the distribution of mean percent labeled bone surface per sector. A nonparametric Mann-Whitney U test was used to test for significant differences in mineralizing surface between the groups of exercised and sedentary roosters and to test for differences in areal properties between roosters that responded to the exercise regimen and roosters that lacked a response. A significance level of 0.05 was employed for all tests. All values reported in the Results are the means ± SE for the respective animal subset.


Mechanical environment

High-speed running generated an extremely nonuniform strain environment in the middiaphyseal TMT. Peak compressive strains of −1850 ± 80 με were engendered in the anterior cortex. Peak tensile strains of 1220 ± 200 με were located in the posterior cortex. The neutral axis spanned the lateral and posteromedial cortices and varied little during stance phase (Fig. 2B). During the time period of stance phase at which longitudinal strains of 1000 με or more were generated, the total range of rotation was only 10 ± 2°. Additionally, the orientation of the neutral axis was similar for each rooster at the point in time at which peak strains were induced in the cortex: 14 ± 5° (see Fig. 1A for definition of the angle). This information is important because the distributions of circumferential and radial strain gradients are primarily dependent on the orientation and not on the location of the neutral axis.

Figure FIG. 2.

(A) Sample of longitudinal normal strain magnitude over time recorded from the anterior gage in a rooster running at a speed of 1.66 m/s. Tensile strains are (by definition) positive and compressive strains are negative. Strain magnitudes were well below 500 με during swing phase and reached their maximum during midstance. (B) Distribution of longitudinal normal strain (με) across the mid-diaphyseal tarsometatarsal section in a rooster running at a speed of 1.66 m/s. The distribution in this figure was determined at the point in time at which maximal strains were induced in the cortex (midstance). The neutral axis is dashed with strain isopleths plotted parallel to it. Large compressive strains (negative sign) were generated on the anterior cortex, and large tensile strains were generated on the posterior cortex (see Fig. 1A for orientation).

The magnitudes of peak circumferential strain gradients were highly nonuniform in distribution, with the smallest circumferential strain gradients located on the anterior and posterior cortices, and with the largest circumferential strain gradients located on the medial and lateral cortices near the neutral axis (Fig. 3). The magnitudes of peak radial strain gradients were in the same range as peak circumferential strain gradients, and sites of maximal radial strain gradients coincided with the sites of minimal circumferential strain gradients and vice versa. Due to the nearly prismatic shape of the TMT mid-diaphysis, maximal longitudinal strain gradients were an order of a magnitude lower than maximal radial and circumferential strain gradients.

Figure FIG. 3.

Distribution of circumferential strain gradients of longitudinal normal corresponding to the strain distribution presented in Fig. 2B. Circumferential strain gradients generated during stance phase were extremely nonuniform in distribution with the largest strain gradients on the medial and lateral cortices. A, anterior; M, medial.

Maximal periosteal shear strains generated in the cortex averaged 1580 ± 190 με. The largest shear strain magnitudes were observed at the anterior aspect of the cortex (sectors 3–5) and coincided roughly with the sites of largest compressive strains. Peak shear strains generated in the remaining sectors were relatively uniform in distribution (Fig. 7). Similarly to longitudinal normal strain, shear strains in all sectors reached their peak values during midstance. By superposition, peak shear strains at any point of the cortex were the result of a combination of applied shearing forces and a torsional moment. When applied separately, the combined end shearing forces generated shear strain magnitudes that were only between 12 and 29% (depending on the sector) of the shear strain magnitude generated by the torsional moment.

Activation of periosteal surfaces and areal measurements

Focal periosteal activation was observed in five out of eight exercised roosters undergoing the exercise regimen (Fig. 4). Partially double-labeled osteonal and endocortical surfaces verified that both calcein injections had worked in all roosters. However, even under high magnifications (800×), a periosteal double label could be detected only in one exercised rooster. In this animal, the double-labeled surface accounted for less than 35% of the total labeled periosteal surface. Periosteal surfaces of seven out of eight control sedentary roosters were completely quiescent, and one rooster demonstrated minimal periosteal mineralization of the anterior cortex (less than 1.5% of the total surface). The ratio of mineralizing surface (MS) to bone surface (BS) was significantly increased in the group of exercised roosters compared with the control group and, consequently, mineralizing periosteal surfaces observed in the exercised roosters could be attributed to the exercise regimen. In contrast, scattered labels were observed on endocortical surfaces in control roosters. Because the goal of this study was to relate specific mechanical stimuli to sites where a previously quiescent bone surface was activated, we limited our analysis to periosteal surfaces.

Figure FIG. 4.

Scanned photograph of a mid-diaphyseal TMT cross-section in which the high-speed running regimen caused activation of medial and anterolateral periosteal surfaces (bright calcein label).

Middiaphyseal cross-sectional area and cross-sectional moment of inertia about the locomotion-induced bending axis were not significantly different between the roosters that responded to the exercise regimen and the nonresponders (area, 24.1 ± 0.7 mm2 vs. 22.5 ± 1.1 mm2; Ibending, 92.1 ± 3.6 mm4 vs. 95.8 ± 17.6 mm4). Across all eight roosters, the ratio between the second moment of inertia about the bending axis and the moment about the axis transverse to it was 0.54 ± 0.01.

Correlation between mechanical parameters and periosteal activation

The distribution of peak circumferential strain gradients correlated significantly with the distribution of mineralizing surfaces across a mid-diaphyseal section (R2 = 0.63, p < 0.002) (Fig. 5A, B). The correlation between peak circumferential strain gradients and activated surfaces was similar in the left (R2 = 0.59) and right (R2 = 0.57) TMTs. Due to the inverse relation between radial and circumferential strain gradients, the distribution of peak radial strain gradients produced a significant but negative correlation (R2 = 0.72). Peak longitudinal strain gradients were not correlated (R2 = 0.02). Combining strain gradients in the three directions via multiple regression analysis generated a R2 value of 0.72.

Figure FIG. 5.

(A) Distribution of peak circumferential strain gradients and percentage of labeled surface per sector superimposed on a middiaphyseal TMT cross-section (mean ± SE). Sectors with the largest gradients correlated highly (R2 = 0.63) with the sectors that exhibited the largest regions of bone forming surfaces. (B) Correlative plot between peak circumferential strain gradients and labeled surface per section. Each data point represents the mean circumferential strain gradient and the mean value of activated periosteal surface per sector.

Because bone surfaces were activated near locations of least, rather than greatest, strains the specific sites of mineralization were not positively correlated with peak absolute longitudinal normal strain (R2 = 0.24, negative correlation). When the sign was incorporated into the correlation between strain magnitude and periosteal bone activation, the R2 value decreased to 0.06 (Fig. 6). Because the distribution of peak strain rate was highly associated with the distribution of peak strain magnitude, strain rate also failed to produce a significant positive correlation with the osteogenic response (R2 = 0.28, negative correlation). Similarly, the distribution of peak longitudinal periosteal shear strain (Fig. 7) could not account for the distribution of activated surfaces observed (R2 = 0.06).

Figure FIG. 6.

Relation between the means of peak longitudinal normal strain and activated periosteal surface per sector. The R2 value produced by this correlation was 0.06. When absolute strain magnitude was used, the R2 value increased to 0.24 but the nature of the correlation remained negative.


The unique feature of this study is that a specific mechanical stimulus was successfully correlated with specific sites of periosteal bone activation induced in a noninvasive bone adaptation model. We determined the distribution of locomotion-induced mechanical parameters acting on the TMT mid-diaphysis in one group of roosters and hypothesized that bone would form in another group of exercised roosters at sites subjected to the largest circumferential strain gradients. The brief treadmill running regimen (9 minutes/day) activated previously quiescent periosteal surfaces, and specific sites of mineralization correlated significantly (R2 = 0.63) with peak circumferential strain gradients.

Despite this success, the study has several limitations. The design of the study did not allow us to associate the distribution of the mechanical stimulus with the specific sites of periosteal activation in each individual animal. The long-term utility of our observations, however, is with the potential for designing exercise regimens that will optimally enhance bone mass in a group of patients. We found a strong positive linear correlation between circumferential strain gradients and sites of periosteal activation due to the coincidence of bimodal distributions for these two parameters. A nonlinear relation between strain gradients and periosteal activation would alter the future determination of what magnitude stimulus is required to initiate bone formation. The mean values for mineralizing surfaces were influenced by the three roosters with absent periosteal labels; however, correlation coefficients were not affected by the nonresponders (i.e., correlations performed without these three roosters produced the same R2 values). The spatial relation between peak circumferential strain gradients and sites of periosteal activation does not necessarily imply causation, and other parameters such as strain magnitude or strain rate may have influenced the magnitude of the observed response. Finally, this was a short-term experiment to investigate the ability of strain gradients to predict sites of exercise-induced periosteal activation. From these data, we cannot say whether circumferential strain gradients associated with a longer term exercise program will produce persisting, structurally relevant, alterations in bone mass and/or morphology.

In our study, peak normal strain magnitude was poorly linked to specific sites of periosteal activation in the adult rooster skeleton. These results conflict with a recent report in which peak magnitude strains were related to sites of increased bone formation in the rat ulna.21 One difference between the studies lies with the active forming and resorbing surfaces present in the young growing rat, as compared with the primarily quiescent adult skeleton present in the roosters in our study. In their study, bone formation was increased most over normal formation in locations of maximal normal strains (approximately 4000 με). It is interesting to note, however, that the areas of greatest bone formation in the control rat ulnas corresponded closely with the location of the neutral axis during normal locomotion (i.e., large circumferential strain gradients, minimal strain magnitude).

Structurally, it seems counterintuitive that increased mechanical loading should produce bone apposition in regions where mechanical integrity is least challenged. Nevertheless, it has been observed that the cross-sectional moment of inertia of long bones in a variety of animals is the least about the bending axis present during locomotion, suggesting that bone sacrifices strength for predictable loading conditions.22,23 Our data support this observation, since the cross-sectional moment of inertia in the rooster TMT mid-diaphysis about the bending axis is approximately 50% of the moment about an axis transverse to the bending axis. While the addition of bone mass along the neutral axis does little to reduce peak strains during activity, it would fortify an already stable strain environment (in terms of strain distribution). Alternatively, our results may be approached from a physiological, rather than structural optimization, perspective. Strain gradients (volumetric) are related to fluid flow in bone,15 and may, therefore, induce a number of osteogenically relevant processes such as fluid shear stresses24 or streaming potentials.25 Here, we assessed gradients of longitudinal normal strain and not volumetric strain gradients; however, their distributions are similar as the magnitude of a volumetric strain gradient in a given direction is dominated by the magnitude of the corresponding gradient of longitudinal normal strain.

Recent experimental and analytical data reveal the sensitivity of bone cell populations to fluid flow–induced shear stresses (shear strains).26–29 Consistent with previous theoretical30 and experimental studies,31,32 we observed a lack of correlation between the distribution of bone matrix shear strain and sites of periosteal activation. This may imply that direct stimulation of bone cells by matrix shear strain is unlikely to be part of the pathway by which bone cells respond to mechanical stimuli. Additionally, because matrix shear strains only have a negligible influence on fluid flow in bone33 and strain gradients are proportional to fluid flow,15 the poor correlation between shear strain and sites of periosteal activation is consistent with our hypothesis that strain gradients stimulate the activation of bone cell populations by influencing fluid flow.

Unlike studies in which exogenous loads generated strain environments substantially different from the normal strain milieu,14,34,35 strain distributions engendered in our model were similar to those induced during normal (walking) locomotion.2 The maximal circumferential strain gradients induced during running where larger than those observed during slow walking, but were located at the same sites.2 While our study indicates that knowledge of the bone's complete strain history may not be necessary to predict sites of bone formation induced by increased exercise, it is possible that the roosters' daily activity levels affected the responses we observed. Quantification of the 24-h strain history for group-housed roosters indicates that large strain magnitudes induced during daily activity are rare compared with those induced by treadmill running.36 However, activity levels among animals may vary and may account for the absence of a periosteal response in three of eight roosters within the exercise group. Even with the null response in three roosters, it is significant that every rooster showing periosteal activation had this response at sites of large circumferential strain gradients.

If circumferential strain gradients affect fluid flow–related cell activation mechanisms, the question arises as to why maximal radial and longitudinal strain gradients were unrelated to sites of periosteal activation. Longitudinal strain gradients in this study were an order of magnitude smaller than circumferential or radial strain gradients, perhaps mitigating the larger fluid permeability coefficient in this direction.37 Similarly, dielectric and electrokinetic considerations reflecting bone's microstructure37,38 suggest that cortical bone is less permeable radially which would impede fluid flow in that direction. Alternatively, fluid flow in a radial direction may have little influence on the activation of periosteal surfaces but may instead influence a different physiological process such as intracortical remodeling. We were unable to test this hypothesis in the present study because of the timing of the fluorescent labels.

Some researchers have reported a minimal or negligible effect of exercise on the adult skeleton.6,39,40 In these studies, bone quantity was averaged across a region of bone, as measured, for instance, by dual X-ray absorptiometry (DXA). While these data give important information about how a larger region of bone responds to a prescribed exercise program, focal significant increases in bone mass within this region may not have been detected. In our study, exercise activated only 23 ± 7% of the total periosteal surface in those roosters that responded to the exercise regimen. This small value also emphasizes that it is critical to have a means of controlling the sites of new bone deposition to obtain a structural benefit from the exercise regimen.

If bone continues to form and consolidate at the sites of activated periosteal surface, circumferential strain gradients may provide a means of optimizing exercise regimens to reduce fracture risk associated with bone loss pathologies. As we confirmed a previous study14 that circumferential strain gradients are highly associated with sites of bone forming surfaces induced in adult skeletons, simple exercises with weights could be designed in which the induced neutral axis (large circumferential strain gradients) is aligned with those sites most likely to enhance the strength of the tissue. If, indeed, there is a causal relationship between circumferential strain gradients and activation of periosteal surface, we also hypothesize that exercise should have larger effects on periosteal surface modeling when bones are subjected to bending moments (large strain gradients) than to torsional or axial loading (small strain gradients). To test those hypotheses in a clinical setting, the mechanical environment could be analyzed noninvasively by simple stress analysis, and site-specific changes in bone morphology could be assessed by techniques such as high-resolution quantitative computer tomography.

In summary, this study related specific sites of periosteal activation to an exercise-induced strain parameter. Each of the animals that demonstrated periosteal activation in response to the brief running regimen deposited bone at cortical sites experiencing small strain magnitudes but large circumferential strain gradients. The strong association between circumferential strain gradients and the specific regions of bone modeling suggests that it may be feasible to design exercise regimens that initiate bone formation at sites most in need of enhanced structural integrity.


This research was supported, in part, by the Natural Sciences and Engineering Research Council of Canada and the Alberta Heritage Foundation for Medical Research. The authors would like to thank Dr. Adrian Wilson for an introduction to PATRAN and ABAQUS.