Skeletal responsiveness to increased mechanical loading is known to be maturity-dependent. The pre- and early pubertal skeleton is able to markedly alter its mass, size, and shape in response to increased loading, thereby increasing its strength to a greater extent than the mature skeleton.1–3 For example, girls who started playing racket sports before or at menarche showed two- to fourfold larger side-to-side differences in bone mineral content (BMC) than girls who started more than 15 years after menarche.4 The failure of exercise interventions to increase areal bone mineral density (aBMD) in postmenarcheal girls5–7 also suggests that there is a decreased responsiveness of bone to loading after menarche. In a cross-sectional study using magnetic resonance imaging (MRI) to quantify bone structural adaptations to exercise during growth, we have previously reported that repetitive loading during prepuberty was accompanied by greater periosteal expansion, whereas loading after menarche lead to greater endosteal (medullary) contraction.8 This maturity- and surface-specific response to loading has a marked influence on the capacity of the skeleton to increase bone strength. Bone's enlargement through newly added bone on the periosteal surface increases bone strength to a greater extent than bone added to the endosteal surface because the bending strength of a unit area of bone is proportional to the fourth power of its distance from the neutral or long axis of the bone.9
It has been suggested that the responsiveness of bone to increased mechanical loading depends on the rate of bone apposition induced by growth.10 According to this hypothesis, exercise-induced periosteal expansion should be facilitated before puberty when rapid bone apposition occurs on the periosteal surface. Conversely, increased loading during puberty should promote endocortical apposition, particularly in girls, because estrogens act on the endocortical surface by inhibiting bone resorption during rapid growth and promoting bone formation after menarche.11 This hypothesis is supported by the findings from unilateral loading in animal studies that showed that male rats had a sixfold greater increase in bone area in response to loading than female rats.12 However, when these benefits were expressed relative to growth rate, the gender differences disappeared because male rats had a 2.5-fold greater increase in body weight and periosteal new bone formation than female rats.12 Whether exercise-induced changes in BMC and cortical bone geometry depend on growth-induced changes in these variables has never been tested longitudinally in humans. Tennis players represent an ideal model to test this hypothesis because growth-induced changes in BMC and bone geometry can be assessed in the nonplaying arm, and exercise-induced skeletal benefits can be assessed by comparing the playing and nonplaying arms.
The objectives of this study were (1) to compare the magnitude of growth- and exercise-induced changes in BMC and cortical bone geometry over 12 months between pre/peri- and postmenarcheal girls and (2) to identify the determinants of exercise-induced changes in BMC and cortical bone geometry over 12 months. We hypothesized that exercise-induced changes in BMC and cortical bone geometry would be (1) greater in pre/peri- than postmenarcheal girls, (2) small compared with growth-induced changes, and (3) proportional to growth rate and current training volume. We also hypothesized that the predominant response to continued loading would be periosteal expansion in prepuberty and endocortical apposition in late puberty.
Materials and Methods
Forty-five competitive female tennis players aged 10 to 17 years were recruited from tennis clubs in Melbourne, Australia. Players were included if they had been playing competitive tennis for a minimum of 2 years and were currently playing at least 2 hours per week (Table 1). Thirty-seven girls (82%) used a double-handed backhand. Girls were excluded if they had sustained a previous fracture of the humerus or had any medical disorder or treatment known to affect bone metabolism. The Ethics Committee of Deakin University approved the study, and written consent was obtained from all participants and their parents.
Table 1. Anthropometric Characteristics, Body Composition, and Training History at Baseline and Follow-up in Pre/Peri- and Postmenarcheal Tennis Players
Pre/peri (n = 13)
Post (n = 32)
Absolute change (95% confidence interval)
Absolute change (95% confidence interval)
Note: Data are given as mean ± SD. Assignment into pre/peri or post group was determined according to menarcheal status at baseline.
Changes in BMC and cortical bone geometry induced by both growth and exercise were assessed by analyzing (1) the annual changes in these variables in the nonplaying arm and (2) the side-to-side differences in the annual changes between the playing and nonplaying arm, respectively. The study design is depicted in Fig. 1. Loading-induced changes in bone and muscle variables were calculated as the absolute difference between the percent increase in BMC in the playing arm over 12 months minus the percent increase in BMC in the nonplaying arm over 12 months. For example, for a given training volume, if player A showed a 15% increase in BMC in the playing arm over 12 months versus only a 10% increase in the nonplaying arm (5-point difference), whereas player B exhibited a 12% increase in BMC in the playing arm and a 10% increase in the nonplaying arm (2-point difference), player B was less responsive to training than player A.
Because the duration of the follow-up ranged from 9.9 to 15.6 months among the players (mean 12.9 months), annual changes (absolute and relative) for each individual were calculated by normalizing the changes between baseline and follow-up to the duration of the follow-up (in months):
As reported previously,8 pubertal status was self-assessed with parental guidance using the standard five-scale Tanner stages for breast development and pubic hair.13 Menarcheal status was assessed through a questionnaire including medical history. Girls who indicated at baseline that they had not had their first menstrual cycle were contacted annually to determine their menarcheal status. Players were assigned to the pre/peri group if they were premenarcheal at baseline regardless of their menarcheal status at follow-up. The remaining players were those who were already postmenarcheal at baseline and therefore were assigned to the post group.
Bone geometry and mass
Bone dimensions at the humerus were determined with a 1.5-T MRI whole-body unit (Signa Excite GE Medical Systems, Milwaukee, WI, USA; software Version 12.0) with a commercial four-channel torso coil. For logistical reasons, the MRI scans were performed at two different locations, but the same MRI device was used at both sites. For all scans, T1-weighted spin-echo sequences were performed to obtain cross-sectional images along the entire length of the right and left humeri. Repetition time of 600 ms and echo time of 14 ms, matrix size 512 × 512, pixel size 390 µm2, and 5-mm slices (with 5-mm gaps between slices) were used at the first site. Repetition time of 3900 ms and echo time of 10 ms, matrix size 512 × 512, pixel size 430 µm2, and 4-mm slices (with 8-mm gaps between slices) were obtained at the second site. A similar proportion of pre/peri and post girls was scanned on both machines (39% pre/peri and 50% post at site 1), and there were no significant differences in muscle CSA or any of the bone geometric parameters between the girls that were measured on the different machines. Therefore, the data obtained on the two machines were pooled.
Humeral length was measured using calipers, as described previously.14 Two regions of interest (ROIs) were selected along the length of the humerus, each representing 10% of the length of the bone. Starting from the proximal end, the midhumerus was defined as the region extending from 40% to 50% of humeral length, whereas the distal humerus was defined as the 60% to 70% region. Parameters were averaged over the two ROIs, therefore representing 20% of the overall length of the humeral shaft. All axial images were analyzed using the OSIRIS imaging software program, Version 3.2 (Digital Imaging Unit, Centre of Medical Informatics, University Hospital of Geneva, Geneva, Switzerland, 1995). Periosteal (total) area (ToA, mm2) was the external size of the bone (ie, periosteal border), and cortical area (CoA, mm2) was periosteal minus the medullary area (MedA, mm2).8, 14 The average periosteal, cortical, and medullary areas were determined by summing the cross-sectional areas of all slices in the two ROIs (middle and distal humerus) divided by the total number of slices in the two ROIs. The short-term precision of image analysis was assessed by the root-mean-square coefficient of variation.15 The root-mean-square coefficient of variation was 1.1% for cortical bone area and less than 1% for total bone area and medullary area.
Bone mineral content (BMC) values of the playing and nonplaying humeri were measured using dual-energy X-ray absorptiometry (DXA; Prodigy, GE Medical Systems Lunar, Madison, WI, USA, pediatric software), as described previously.14 The root-mean-square coefficient of variation for BMC was less than 1%.
Training history was assessed by questionnaire. The participants, with the assistance of their parents, were asked to record the age at which they started playing regular tennis (at least 2 hours per week), the number of years of practice, and their training volume (hours per week) since they started playing, as well as injuries resulting from tennis.
Background variables between pre/peri and post were compared using independent t tests. The annual changes in background variables between pre/peri and post were assessed using a repeated-measures analysis of variance (ANOVA).
Effects of past training
Side-to-side comparisons in bone and muscle parameters at baseline were assessed using paired t tests. Relative side-to-side differences (percentage) were tested using a one-sample t test against zero.
Effects of growth
Growth effects over 12 months in the nonplaying arm were compared between pre/peri and post using repeated-measures ANOVA (baseline versus follow-up).
Effects of continued training
The effects of continued exercise training over the 12 months were first analyzed by comparing the side-to-side differences in muscle and bone parameters between baseline and follow-up using a repeated-measures ANOVA. The effects of continued loading were investigated further by comparing the annual changes in bone and muscle parameters between the playing and nonplaying arms. The effects of exercise were compared between pre/peri and post using repeated-measures ANOVA (annual changes in playing versus nonplaying arms). Analysis of covariance (ANCOVA) was used to adjust annual changes in bone and muscle parameters in the playing arm for the corresponding parameters in the nonplaying arm and changes in training volume. Exercise-induced skeletal benefits were calculated as the difference in the annual changes in bone and muscle parameters between both arms. Linear regression analysis was used to investigate potential predictors of exercise-induced skeletal benefits, such as training volume and baseline asymmetries. Growth rate (increase in height) rather than growth-induced changes in bone parameters was included as a predictor because growth-induced changes in muscle and bone parameters were used to calculate exercise-induced skeletal benefits (ie, the side-to-side differences in the annual changes in muscle and bone parameters; Fig. 1). Therefore, growth-induced changes in muscle and bone parameters would be correlated automatically with exercise-induced skeletal benefits by design, which would be a statistical bias, as explained previously.16, 17 All data were reported as mean ± SD unless otherwise stated. All statistical procedures were performed with SPSS for Windows, Version 17.0.1 (SPSS, Inc., Chicago, IL, USA, 2008).
The background characteristics of the pre/peri and post players for anthropometry, body composition, BMC, and training volume are given in Table 1. At baseline, 13 players were premenarcheal (pre/peri: gynecologic age −0.9 years, range −2.2 to −0.1 years), and 32 players were postmenarcheal (post: gynecologic age +1.8 years postmenarche, range 0.1 to 5.5 years). Among the pre/peri group (n = 13), 10 players experienced menarche during the 12-month follow-up, and 3 attained menarche after the follow-up period. Repartition of the players into each of the five Tanner stages at baseline is reported in Table 1. The mean age of menarche was not significantly different between pre/peri (13.2 ± 1.2 years) and post girls (12.7 ± 1.3 years). All players started playing before menarche (mean age 6.9 ± 1.9 years). Pre/peri girls started playing at a similar age and had a comparable training volume as post girls but a shorter training history (p < .05). Annual growth rates, that is, increase in height and humeral length as well as gains in body weight and lean tissue mass, were greater in pre/peri than in post girls (p < .001; Table 1). Pre/peri girls also gained approximately four times more bone mineral mass at the whole body over 12 months than post girls (410 ± 180 g versus 110 ± 100 g, p < .001; Table 1).
Growth-induced changes in BMC and cortical bone geometry in the nonplaying arm are shown in Fig. 2. In pre/peri girls, the nonplaying arm experienced significant increases in BMC (+20%, p < .001) that were associated with gains in ToA (+10%, p < .001) and CoA (+16%, p < .001), whereas MedA did not change significantly over the 12-month follow-up. Similar significant changes were observed in the post group, but the magnitude of the changes was three- to fourfold less than in the pre/peri group (BMC +5%, ToA +3%, CoA +5%, p < .05 to p < .001), and MedA decreased significantly by 1% (p < .05). Muscle CSA increased over 12 months in the pre/peri group (+8%, p < .05) but decreased in the post group (−4%, p < .05). In all girls combined, growth rate (increase in height) was strongly correlated with the annual changes in BMC, ToA, MedA, CoA, and MCSA in the nonplaying arm (r = 0.41 to 0.70, p < .01 to .001). Positive but weaker correlations also were found for the increase in body weight (data not shown). After adjustment for growth rate, differences between the pre/peri and post groups regarding the annual increases in bone variables and MCSA in the nonplaying arm disappeared.
Effects of past training on BMC, bone geometry, and muscle area
Baseline side-to-side differences in BMC, cortical bone geometry, and MCSA in the pre/peri and post groups are given in Table 2. Baseline BMC was 16% to 18% greater in the playing arm compared with nonplaying arm in both the pre/peri and post groups, and this was associated with a 9% to 17% side-to-side difference in ToA and CoA (all p < .001). There was no effect of past exercise on MedA in the pre/peri group, but in the post group, medullary area was 7% smaller in the playing arm compared with the nonplaying arm (p < .05). MCSA was 8% to 9% greater in the playing arm compared with the nonplaying arm in both the pre/peri and post groups (p < .001). Bone asymmetries were not greater in the post group than in the pre/peri group despite a longer history of training. In all girls, baseline side-to-side differences in BMC, CoA, and ToA were positively correlated with the side-to-side differences in MCSA (r = 0.28 to 0.38, p < .05). Baseline asymmetry in BMC also was correlated with baseline training volume (r = 0.30, p < .05), and baseline asymmetry in ToA was correlated with the number of years playing tennis (at least 2 hours per week; r = 0.47, p < .01).
Table 2. Relative Side-to-Side Differences (%) in Bone Mineral Content (BMC), Total Bone Area, Cortical Area, Medullary Area, and Muscle Area at Baseline and 12 Months in Pre/Peri- and Postmenarcheal Tennis Players
Exercise-induced changes on cortical bone surfaces over 12 months
Over 12 months, the relative side-to-side differences in favor of the playing arm increased further for BMC (p = .09), ToA, CoA, and MCSA in the pre/peri group (p < .05 to .01; Table 2). In the post group, the relative side-to-side differences in BMC and ToA continued to increase (p < .05), but there was no significant effect of continued training on CoA (+0.8%, p = .15; Table 2). The relative side-to-side differences in MedA did not change over 12 months in either the pre/peri or post players.
The annual absolute and percentage gains in BMC, ToA, CoA, and MCSA between the playing and nonplaying arms are illustrated in Fig. 2. In the pre/peri group, the playing arm showed greater annual increases than the nonplaying arm for BMC, and this was associated with greater gains in both ToA and CoA. In the post group, the differences in the annual increases between arms were significant for BMC, ToA, and CoA (Fig. 2).
When the results were expressed as the annual accrual (grams or square millimeters per year) over the 12 months, the pre/peri players accrued 3.1 g of extra BMC in the playing humerus compared with the nonplaying humerus (+11.8 g versus +8.7 g, respectively, p = .06), which corresponds to a 36% gain. This gain in BMC translated into a 34% greater increase in ToA and a 39% greater increase in CoA in the playing arm (p < .05). In the post group, the playing humerus accrued 2.1 g of extra BMC compared with the nonplaying humerus (+4.5 g versus +2.4 g, p < .01, +88%), which translated into a 39% greater increase in ToA and a 28% greater increase in CoA in the playing arm (p < .05). When comparing the annual gains in the playing arm between the pre/peri and post groups after accounting for the corresponding changes in the nonplaying arm, the increases in ToA and CoA were greater in the pre/peri group than in the post group (p < .05 to .001; Fig. 2). Adjustment for changes in training volume over the year and baseline bone asymmetries did not change the results. The increase in MCSA in the playing arm, which was greater in the pre/peri group than in the post group after adjusting for the corresponding increase in the nonplaying arm, was no longer different between groups after adjusting for the decrease in training volume over 12 months.
Determinants of the exercise-induced changes in BMC and cortical bone geometry
In the whole sample, the best predictors of the exercise-induced skeletal benefits in BMC, CoA, and MCSA were baseline side-to-side differences in bone and muscle variables: the smaller the preexisting side-to-side differences, the greater the skeletal benefits over 12 months. The strongest correlations were found for BMC (r = −0.48, p < .001), followed by CoA (r = −0.39, p < .01) and MCSA (r = −0.34, p < .05; Fig. 3). Growth rate (increase in height over 12 months) was a significant determinant of the magnitude of the exercise-induced skeletal benefits in ToA and CoA (r = 0.50 and 0.46, respectively, p < .01). So was the increase in whole-body BMC over 12 months (r = 0.33 and 0.34, respectively, p < .05). Exercise-induced changes in MCSA were predictive of the exercise-induced skeletal benefits in BMC in the pre/peri group only (r = 0.57, p = .04). In the whole sample, the change in training volume over 12 months was predictive of the exercise-induced changes in MCSA (r = 0.37, p < .05) but not the gains in bone parameters. Playing history (number of years playing tennis) was negatively correlated with the exercise-induced gains in ToA and CoA over 12 months (r = −0.34 and −0.44, respectively, p < .05).
The findings from this prospective study in currently active young female tennis players showed that continued training over 12 months induced further significant exercise-induced gains in both the mass and geometry of cortical bone in pre/peri- and postmenarcheal players. In line with our hypothesis, we found (1) that the exercise-induced benefits over 12 months were greater in the pre/perimenarcheal girls and (2) that the continued gains in cortical area with training in the pre/perimenarcheal players was due to periosteal apposition and not changes at the endocortical surface. While these results support the notion that the skeletal responsiveness to loading is greatest during the pre- and perimenarcheal years, our findings that postmenarcheal players experienced small but significant exercise-induced skeletal gains with continued training suggests that exercise during this period still may play an significant role in optimizing peak bone mass.
It is widely recognized that growing bone has the capacity to adapt to increased loading, but whether there is an optimal period during growth for exercise to enhance the mass, structure, and strength of bone is not clear. Although this study included only two prepubertal players, the magnitude of the baseline bone asymmetries in these players and more generally the pre/perimenarcheal group suggest that skeletal adaptations to loading started before puberty. Added to the fact that the skeletal responses to continued training over 12 months were greater in the pre/peri- than in the postmenarcheal players, our results are consistent with previous cross-sectional studies of racquet sport athletes and exercise intervention trials that indicated that the pre- and early pubertal years (Tanner stages I to III) may represent the optimal time to enhance bone mass.2, 4, 8, 18 Importantly, our study provides further insight into the structural basis underlying the exercise-induced changes in bone during growth. It has been suggested that exercise preferentially results in changes in bone structure at the surface(s) already undergoing rapid bone formation due to normal growth.10, 19 In line with this hypothesis, we found that the 39% greater annual accrual in cortical bone area in the playing compared with nonplaying arm in the pre/perimenarcheal players over 12 months was due to periosteal apposition and not changes at the endocortical surface. Several previous cross-sectional studies of racquet sport athletes also have reported that the side-to-side differences in cortical area found in girls who trained during the pre- and early-pubertal years or commenced training prior to menarche was due largely to bone formation on the periosteal surface.2, 8 Exercise-related gains in bone size have been shown to be associated with marked improvements in bone strength2, 8 because bone's resistance to bending or torsion is exponentially related to its diameter.9, 20
To our knowledge, there are few prospective studies that have investigated the surface-specific responses of cortical bone to continued loading late in puberty or postmenarche. Intervention trials conducted in postmenarcheal girls have shown that exercise results in either modest gains in bone mass or has no significant osteogenic effect.5–7, 21 It should be noted that bone mineral accrual slows down after menarche but remains positive.22 Rising estrogen levels in females inhibits periosteal apposition and may lower the bone (re)modeling threshold and sensitize bone next to marrow to mechanical loading, leading to endocortical apposition.23, 24 Consistent with this, the cross-sectional data of the postmenarcheal players at baseline showed that the greater cortical area in the playing compared with nonplaying arm was due to endocortical contraction coupled with periosteal apposition. However, we found no evidence of any further endocortical apposition with continued training over 12 months. This could be related to the reduction in training volume in the postmenarcheal players over the follow-up period or the short duration of the follow-up. Alternatively, the role of estrogens and their interaction with loading on cortical bone geometry may change during the years following menarche. We speculate that the greatest changes at the endocortical surface in response to loading in girls may occur predominantly throughout puberty and early postmenarche, with no further exercise-induced endocortical apposition occurring from around 2 years postmenarche. This hypothesis is supported by evidence of exercise-induced periosteal expansion with no endosteal apposition in female tennis players who started playing at least 1 year after menarche (mean starting age 26.4 ± 8.0 years).2 We did observe a small but significant exercise-induced gain in total bone area and BMC in the postmenarcheal players with continued training, suggesting that the large bending forces imparted to the dominant humerus during tennis typically induce tension and compression along the surfaces of the cortical shell and stimulate periosteal bone apposition.
One of the key determinants of the exercise-induced skeletal benefits in our study, independent of menarcheal status, was the preexisting bone asymmetries due to past training. The smaller the side-to-side differences at baseline, the greater the skeletal gains with continued training. For instance, some pre/perimenarcheal girls, despite showing an accelerated rate of growth, displayed little additional skeletal benefits from 12 months of unilateral loading (Fig. 3). In contrast, some postmenarcheal girls who exhibited relatively small asymmetries at baseline (eg, less than 10% difference in BMC between humerii) experienced further skeletal benefits by pursuing training over 12 months (Fig. 3). This point illustrates the importance of investigating past history of physical activity when conducting an exercise intervention in a research setting.16, 25, 26 These findings are consistent with the basic training principles of initial values and diminished return. According to these principles, the greatest changes in bone in response to loading generally occur in those with the lowest initial values, and following any initial adaptations to a given level of loading, any further gains in bone are likely to be slow and of a small magnitude.26 Our results also may suggest that there is a limit as to how much bone one can gain from exercise. This limit is likely to be specific to each individual because cases of extreme skeletal adaptation to mechanical loading have been reported. For example, a 22-year-old male baseball pitcher was found to have a 63% greater bone mass and cortical thickness in his dominant humerus, almost doubling the capacity of the bone to withstand torsional forces.27
Muscle forces produced by muscle contractions have been proposed to dominate the skeleton's postnatal structural adaptation to loading.28, 29 We found that the exercise-induced changes in muscle CSA predicted 32% of the variance in the exercise-induced skeletal benefits in BMC in the pre- and perimenarcheal girls. Such a relationship was not found in postmenarcheal girls, possibly owing to the fact that high estrogen levels may affect bone tissue more than muscle tissue, thereby modifying the muscle-bone relationship.30 Assuming that muscle CSA provides a surrogate estimate of peak muscle forces, these results suggest that factors other than muscle forces contributed to the skeletal gains with continued training. These findings are reinforced by our previous work in pre-, peri-, and postpubertal female tennis players, where we found that asymmetry in muscle area accounted for only 12% to 16% of the variance of the asymmetry in humeral bone mass, bone size, and bending strength.30 Training volume over the 12 months was not a predictor of the gains in bone parameters; however, training volume represents only a rough surrogate of exposure to loading. A quantification of the average number of tennis strokes over several training sessions might have provided a more accurate assessment of the impacts typically imparted to the playing arm of tennis players. It is also possible that a small portion of the bone asymmetries found in tennis players can be attributed to activities associated with daily living rather than regular tennis playing because female adolescents who do not engage in sport-related unilateral loading typically show 0% to 5% side-to-side differences in bone mineral mass and density.4, 31 Another limitation of this study is the absence of data on calcium intake, which can modulate the osteogenic response to exercise.
In summary, this prospective study demonstrates that continued training over 12 months in active pre/peri- and postmenarcheal female tennis players leads to significant exercise-induced gains in bone mass and geometry. The benefits in cortical bone geometry were greater in the pre/perimenarcheal girls, suggesting that the pre/perimenarcheal years may represent the optimal window of opportunity to enhance bone structural properties. However, since small skeletal gains were also observed in the postmenarcheal players, it would appear that continued exercise training after menarche still may play an important role in optimizing peak bone mass. In conclusion, we recommend that girls should be encouraged to be active before the onset of puberty to optimize bone structure, and to maintain participation in regular physical activity throughout puberty to maximize those skeletal benefits.
All the authors state that they have no conflicts of interest.
We thank the late Pr Charles Turner for his contribution to the conception and design of this study. We are very grateful to Jeni Black (Musculoskeletal Research Centre, La Trobe University, Heidelberg, Australia) and Brendan Henderson and Briony Hill (Centre for Physical Activity and Nutrition Research, Deakin University, Burwood, Australia) for their efforts in managing the project and analyzing the data. We thank radiographers Chris Evans and David Di Domenico for their technical assistance. We also thank the players and their parents for their time and commitment, as well as the staff from Box Hill Tennis School. This study was funded by the National Health and Medical Research Council (NHMRC). RMD was supported by an NHMRC Career Development Award (ID 425849). SLB was supported by an NHMRC Career Development Award (ID 229320).