Journal of Applied Physiology

Selected Contribution: Variation and heritability for the adaptational response to exercise in genetically heterogeneous rats

Michael Lee Troxell, Steven Loyal Britton, Lauren Gerard Koch


Adaptational response to aerobic exercise was artificially selected for across one generation in a founder population of 20 female and 20 male genetically heterogeneous rats (N:NIH). Selection for low and high response was based on the change in treadmill running capacity, assessed by meters (m) run to exhaustion before and after 24 days of modest treadmill running. The training response of the founder population averaged +222 m, with wide variation from a negative gain (−) of −110 m to a positive gain (+) of +430 m. Six pairs of the lowest (+13 m) and highest (+327 m) responders were mated. Mean response to training of the low-line (+242 m) offspring did not differ from the founder. The high-selected line gained 383 m from training, +161 m above the founder population. Narrow sense heritability estimated from regression of offspring on midparent values for response to training was 0.43 (P< 0.007). One generation of selection resulted in a 58% divide between the low and high lines. Selectively bred models of both intrinsic (untrained) and adaptation response can be useful in resolving the genetic basis of variation in aerobic capacity.

  • treadmill
  • training
  • artificial selection
  • breeding
  • N:NIH rats

energy transfer via aerobic metabolism defines a large part of biology for essentially all multicellular organisms (1, 22, 26). Aerobic capacity is assuredly a complex trait, as determined by the multifactorial interplay between genetic and environmental factors. Recent evidence in humans suggests two genetic substrates contribute to the natural variation that exists for an organism's current phenotype for aerobic endurance capacity. First, there is a complement of genes that determines variation of aerobic capacity in the untrained (intrinsic) state (4, 16). In addition to intrinsic capacity, another array of genes apparently dictates the adaptational responses to exercise (3, 5). Furthermore, there is considerable heterogeneity in responsiveness to training where some individuals experience little or no gain (3). The recent observations that modest exercise is associated with a 58% decline in the risk of acquiring Type 2 diabetes (17) and that exercise capacity is a predictor of mortality (21) exemplify the central roles of adaptational responses to exercise. Although evidence from twin and heritability studies suggests a substantial genetic component to aerobic capacity, none of the genes that contribute to variation in this phenotype has been identified (4, 24).

Given such complexity, the creation of animal models in which both genetic and environmental variations approach minimums can be of substantial value (6, 15). Artificial selection at the extremes of a complex trait produces ideal genetic models for three primary reasons. First, the low- and high-line traits can often be made to differ substantially, which increases the signal measurements. Second, if the coefficient of inbreeding is kept low, the major population complement of contrasting alleles causative of the trait difference will be concentrated in the divergent lines. Third, selection across many generations interprets into inadvertent selection for lack of sensitivity to subtle differences in environment. This can be of large benefit because inbred strains that differ markedly for a trait, that did not originate from selection, often demonstrate wide trait variation in response to similar environments (8).

Selection is possible if sufficient additive genetic variance exists in a population for that trait (11). On the basis of R. A. Fisher's 1930 Theorem of Natural Selection (12), traits peripherally associated with evolutionary fitness, such as morphology and complex physiology, demonstrate more additive genetic variance because of less pressure from natural selection (20). In 1996, our laboratory (18) started the successful development of rat genetic models for intrinsic (i.e., untrained) low- and high-aerobic treadmill running capacity by artificial divergent selective breeding.

The purpose of this study was to test the hypothesis that significant trait variation and heritability exist in a population of N:NIH rats for a measure of the adaptational response to exercise via modest aerobic treadmill endurance training. As a secondary product, we also obtained information about the relationship between initial intrinsic capacity and the magnitude of the response to training. Our results demonstrate that 1) wide variations in the response to training exist in an outbred population from which artificial selection can proceed, 2) there is a significant genetic component to aerobic training that carries an estimated narrow sense heritability (h2) of 0.43 across one generation, and 3) from initial evidence, a low intrinsic aerobic capacity might have linkage with a high response to training.


The overall plan was to test a founder population of rats for adaptational response to 8 wk of treadmill running and mate six breeding pairs that demonstrated the lowest and highest training response. After this, we could phenotype the first-generation offspring to estimate heritability and the response to divergent selection.

Founder population.

The starting population was 23 female and 24 male genetically heterogeneous rats (N:NIH stock) obtained from a colony maintained at the National Institutes of Health (13). We requested that each rat in the founder population be of different parentage so that selection was not among brothers and sisters, which broadens the genetic variance (14). As a result, the rats ranged in age from 8–16 wk old on arrival at our institution. Animals were divided by sex and housed two per cage. Each rat was provided food (Ralston Purina 5001) and water ad libitum. Animal housing and testing rooms had an average ambient temperature of 71°F and were on a 12:12-h light-dark cycle with the light cycle coinciding with daytime. All procedures were carried out with approval by our Institutional Animal Care and Use Committee and were conducted in accordance with the “Guiding Principles in the Care and Use of Animals” as approved by the American Physiological Society. The pretraining estimate of capacity for the founder rats was initiated after a 3-wk accommodation when the rats were 11–19 wk old. Rats were tested and trained between 4:00 PM and 8:00 PM, with the specific time of day for any rat randomized. The entire protocol to measure the phenotype for response to exercise training from pretesting period through training to posttesting period required 11 wk.

Estimation of endurance running capacity.

The first week of the protocol consisted of introducing each rat to treadmill (model Exer-4, Columbus Instruments, Columbus, OH) exercise for gradually increasing durations each day so that each could attain 5 min at a speed of 10 m/min on a 15° slope. This amount of exposure to treadmill running is likely below that required to produce a significant change in aerobic capacity (2, 10).

The first 2 days of introduction to treadmill running consisted of simply placing the rat on the belt that was moving at a velocity of 7 m/min (15° slope) and picking the rat up and moving it forward if it started to slide off the back of the belt. During introductiondays 3–5, the belt speed was gradually increased up to 10 m/min; failure to run caused the rats to slide off of the moving belt and onto a 15 × 15-cm electric shock grid that delivered a 1.2-mA current at 3 Hz. The rats were left on the grid for ∼1.5 s and then moved forward onto the moving belt. This process was repeated until the rats learned to run to avoid the mild shock. The ability to achieve this minimal level of running at least once constituted the threshold performance necessary for inclusion in endurance running capacity evaluation the following week.

During the second week, each rat was evaluated for maximal endurance running capacity on 3 days (either Monday-Wednesday-Friday or Tuesday-Thursday-Saturday) with the use of a previously described ramped-speed protocol (18). Each daily endurance trial was performed at a constant slope of 15° with the starting velocity at 10 m/min. Treadmill velocity was increased by 1 m/min every 2 min, and each rat was run until exhausted. Exhaustion was operationally defined as the third time a rat could no longer keep pace with the speed of the treadmill and remained on the shock grid (despite gentle pushing) for 2 s rather than run. At the moment of exhaustion, the current to the grid was stopped, and the rat was removed from the treadmill and weighed.

For each of the three trials, the total distance run (m) was used as the estimate of aerobic endurance capacity. The single best daily run of three trials for each rat was considered the trial most closely associated with the heritable component of endurance running capacity. Estimates of endurance capacity reported here as pretraining and posttraining values are based on the single best-day criterion.

Training protocol.

The goal was to apply an absolute training regime over an 8-wk period that would produce a training response that most of the founder population rats could complete. Day 1 training was set at a rate of 10 m/min for 20 min (distance of 200 m), and the result from each day of training was used to estimate the parameters for the next day of training. If more than 90% of the rats completed the training, the parameters for the next day were increased slightly. If less than 90% of the rats completed the training, the parameters for the next day were either not changed or scaled down slightly in an attempt to allow 90% or more of the rats to complete the training. The most critical factor was to have a population of rats that had received exactly the same amount of training. Table1 shows the target time, rate, and distance for each of the 24 days of training for both the founder andgeneration 1 populations. To manage the workload, the rats were trained in two groups. One group trained Tuesday-Thursday-Saturday, and the other group trained Monday-Wednesday-Friday. After the first 4 wk of training, both the days of the week on which the groups trained and the trainers of the groups were changed to reduce operator bias.

View this table:
Table 1.

Parameters for training protocol

Selection process.

Selection for low and high response was based on the change in treadmill running capacity assessed by meters run to exhaustion pretraining and posttraining to 24 days of modest treadmill running. The six lowest and six highest responding rats of each sex were selected from the founder population and randomly mated to establish the low- and high-selected lines. The resultant offspring from each of the 12 families were weaned 28 days after birth, and animals were divided by sex and housed two per cage. All offspring (35 females and 21 males) were tested for response to training in the same manner as the founder population with two minor alterations: 1) pretraining estimate of capacity was started uniformly when rats were 10 wk of age and 2) rats were trained with 12% more distance over the 8 wk (Table 1). Divergent selection, as employed here, served as a control for environmental shifts across time.

Analysis of data.

Narrow sense heritability (h2) was estimated from the slope of the linear regression of the training response of each offspring on the mean value of both parents (midparent value) (11). The average response to selection for the low and high lines was taken as the difference for the trait between the mean of the founder population and the mean of first-generation progeny from the selected parents. The average unadjusted response to selection was estimated from the regression of change (delta) in distance run to exhaustion expressed in meters for each rat (dependent variable) on generation (independent variable) using SPSS software (19). One-way ANOVA followed by Tukey's test was used to evaluate differences between the founder population and the low and high lines. If the data failed the assumption of normality, the Kruskal-Wallis one-way ANOVA for ranks was applied; if significance was found, the Dunn method was employed for pair-wise comparisons. An F test was applied to test whether the variance was equal between females and males.

Possible association between intrinsic aerobic capacity (pretraining values) and the response to training (delta) was assessed from the regression of posttraining (dependent variable) on pretraining values (independent variable). A slope of the linear regression line (post- vs. pretraining values) equal to 1 was interpreted as rats that respond with equivalent magnitude to training independent of their intrinsic capacity. The slope of the linear regression line (post- vs. pretraining values) of <1 indicated that rats with low intrinsic capacity respond more to the exercise training than rats with high intrinsic capacity. Post- vs. preslope values of >1 were interpreted as high-intrinsic capacity rats that respond more to training than low-intrinsic capacity rats. Slopes were considered significant if the 95% confidence interval (CI95) was not inclusive of a slope value of 1.

Except for descriptive data on the founder population (first paragraph of results), all data comparisons are from rats that completed the total cumulative training distance. The 5% level of confidence was arbitrarily used for assigning significance, and data are presented as means ± SE.


Forty (20 female and 20 males) of the original 47 founder rats completed the 8-wk absolute training protocol. Twenty-three of the 40 rats completed 100% of target distance (8,260 m), whereas the other 17 rats reached 85% of cumulative target distance and averaged 7,001 m of training. The entire founder population (n = 40) averaged a training response of 182 m with a wide variation that ranged between −110 and +430 m gained. This response did not correlate with age (r = 0.0655; P = 0.69) or pretraining body weight (r = 0.083; P = 0.61) of the rats. Parents were selected from the extremes of the phenotypic distribution and represented 30% of the entire founder population at each end of the extreme. Figure1 shows the response to training in these 40 founder rats. On average, the change in running capacity for female rats was 193 ± 28 m and the male response was 171 ± 33 m, with no significant difference in the variance of response between the sexes (P = 0.23). The six females with the lowest response to training averaged a gain of +41 ± 24 m, whereas the male response was lower (−14 ± 28 m) but not significantly different from females (P = 0.17). The six females and six males selected as high-line breeders demonstrated an increase in running capacity that averaged 321 ± 25 and 333 ± 22 m, respectively.

Fig. 1.

Change in capacity with training (m) for each rat in the founder population. The 6 lowest responding pairs, averaging 13 ± 29 m gain in capacity, were bred to propagate the low line. The 6 highest responding pairs gained 327 ± 15 m in response to training and were mated to start the high line.

For phenotype comparison between founder and first-generation populations, only data from rats (females and males combined) that fully completed the total target training distance (23/40 founders, 28/40 generation 1 low line, and 28/42 generation 1 high line) are presented. That is, only data from rats that had exactly the same amount of training applied as an environmental intervention were considered.

Figure 2 displays the founder andgeneration 1 values for response to training in the form of frequency histograms. The low-line offspring training response averaged 242 m (range = −131 to 650 m) and was not significantly different from the average of the founder population (222 m). The high-line offspring gained 383 m (range = 44–684 m), which was significantly different from the founder population response (222 m) by 63% (P < 0.003). Table2 provides a summary of the distance and time run to exhaustion and body weights of rats for pretraining, posttraining, and the delta produced by training. In addition, there were no differences in age (145.6 vs. 146.8 days old at posttest) or litter size (8.0 vs. 8.9 pups) between the first-generation offspring of the low and high lines.

Fig. 2.

Frequency histograms for change in capacity with training for the founder (A) and generation 1(B) populations. After 1 generation of selection, the average response to training for the low line (242 m) was significantly different from the response in the high line (382 m).

View this table:
Table 2.

Summary of founder and generation 1 offspring

The response to training for each family was regressed on the corresponding midparent values to estimate h2 (Fig.3). Eleven families (5 high line and 6 low line) are represented; one of the high-line families failed to produce offspring. The slope of the linear regression line was 0.43 (P < 0.007), estimating that ∼43% of the response to training in the offspring was determined genetically. Figure4 shows the response to artificial divergent selection for gain in capacity due to training across one generation. The regression of response to training as a function of generation (i.e., founder and generation 1) was not significantly different from zero for the low line. The slope for the high line was significant (P < 0.001), providing an initial estimate that selection for high response to training will increase by 161 m each generation.

Fig. 3.

Regression of the offspring on midparent values was used to estimate narrow sense heritability. Approximately 43% of the variation in the offspring is associated with heritable factors. AnF test revealed no difference in variance between the female and male parents for response to training.

Fig. 4.

Response to 1 generation of artificial divergent selection for adaptational response to training. The low line showed no significant response to selection, but the high line gained 161 m in capacity over the founder population. Values are means ± SE.

Figure 5 displays the relationship between the initial pretraining capacity and the adaptational response to training, as measured by the posttraining estimate of capacity. In each panel, slopes of regression lines are compared with a line of identity (slope = 1) derived from theoretical data, where posttraining values exactly match pretraining values (i.e., no response to training). For the founder population (Fig. 5 A), the linear regression line had a slope (0.47) that was significantly less than 1 (CI95 = 0.127 to 0.804). Thus, on average, the rats with lower pretraining capacity had larger responses to training than rats with higher initial pretraining capacity. The half portion of the population with the lower initial pretraining capacity increased 282 m, whereas the half portion of the population with higher pretraining capacity increased only 160 m with training.Generation 1 of the low line also produced a pre- vs. posttraining regression with a slope (0.17) of <1 (CI95 −0.253 to 0.592) (Fig. 5 B). The bottom half portion of the population with lower pretraining capacity increased 332 m, whereas the top half portion of the rats with higher pretraining capacity increased 153 m with training. In contrast to the founder and first-generation low-line rats, the offspring in generation 1 of the high line produced an association not different from random between pre- and postvalues (Fig.5 C). That is, the regression between pre- and posttraining had a slope (0.80) not different from 1 (CI95 = 0.101 to 1.491).

Fig. 5.

Regression of posttraining on pretraining values for founder (A) and generation 1 low (B) and high (C) lines. Slopes were <1 for both founder andgeneration 1 low-line rats, indicating that low intrinsic capacity (pretraining) was associated with a greater response to training. For the generation 1 high-line results, the slope was not different from 1, suggesting no association between pretraining level of capacity and response to training. As a reference, a line of identity was derived from data in which the posttraining values are identical to pretraining (i.e., no adaptational response to training).


Traits such as the response to training are referred to as complex to emphasize their determination by multiple genetic and environmental interactions. Multiple systems such as cardiovascular, metabolic, sensorimotor, and neuromuscular as well as aspects of social, behavioral, and present local environmental factors are presumed to combine to determine the response to training. As a starting point, we created a simple model as a guide to understanding variation in physical capacity (current phenotype), as applied in this case to aerobic endurance (Fig. 6). The model has four additive components comprising genetic and environmental contributions to intrinsic capacity (i.e., untrained), capacity accrued from the response to training, and an error term. Symbolically, PC = GI + EI + GT + ET + RSe, where PC = an individual's (ith) current measured phenotype, GI = genes causative of intrinsic capacity, EI = environment influencing intrinsic capacity, GT = genes causative of adaptational response to training, ET = environment influencing response to training, and RSe = random and systematic error of measure of PC. Until more explicit information is available, each of the four components that sum has been assigned equal fractional representation that follows a Gaussian distribution (18). An individual that approaches a sum of zero from the additive components is defined as having less than viable capacity and does not contribute to the population. The one-hundredth percentile represents the theoretical optimal combination for alleles and environment. Each individual (ith) will have a unique set of these four components, and various combinations can result in a similar current phenotype. The goal in artificial selection is to concentrate G components while minimizing the E components by maintaining a standardized environment.

Fig. 6.

Four-component additive model of current phenotype for aerobic endurance capacity. Each individual (ith) derives randomly each of its endurance components (◊) from a Gaussian distribution. GI, genes causative of intrinsic capacity; EI, environment influencing intrinsic capacity; GT, genes causative of response to training; ET, environment influencing response to training. For simplicity, random and systematic error (RSE) for the measure of PC (individual's current measured phenotype) was taken as zero.

Complex traits not closely associated with evolutionary fitness can demonstrate a response to artificial selection (20). The most critical factors for a selection response are sufficient additive genetic variance and a measure of the trait with low random and systematic error. Additive genetic variance refers to that associated with the average effects of substituting one allele for another and is presumably greater for traits peripheral to evolutionary fitness because of less selective pressure-causing fixation (12). In devising a test for estimating the response to moderate endurance training that would likely capture additive genetic variance, we had four criteria in mind: 1) simple to perform, 2) objectively interpretable, 3) gradated on a continuous numerical scale, and 4) demonstrating a wide range between the low and high values of the phenotype (6). Results from the founder population thus predicted that generation 1would display the attained significant response to selection and accompanying h2 (0.43).

Training protocols are most often formulated as some function of initial capacity adjusted periodically to account for training-induced alterations in capacity (28). The general success of this kind of relative training is in accord with the model in Fig. 6 and also accounts for the concept of a threshold for a training effect. We presume that the genes causative of response to training follow a hierarchical expression pattern as a function of the training protocol; that is, training response genes follow a low to high threshold based on training intensity, frequency, and duration. Low-trained subjects would have access to a larger number of currently “not induced” low-threshold training response genes and thus obtain gain with a lesser training program. Subjects who already acquired some training effects will achieve an equivalent gain with a higher training protocol because low-threshold genes have already been activated.

The present study applied a genetically based approach to evaluate the effects of training. Physiologically, relative training appears to be appropriate for attaining large training responses in populations (28). For these studies, we applied the same absolute training stimulus to each rat to test for response. If we had applied relative training, the gain would have been a function of both intrinsic capacity and a differential activation of training genes unique for each individual. Thus it seemed appropriate to apply absolute training uniformly and allow the allelic variation for training to determine the response for each individual. Uniform training for a population means the amount of training applied is a function of the capacity deemed acceptable for subjects with the lowest training capacity. Because we also have a capacity threshold for inclusion in the study, we are, in turn, selecting for compliance with the training protocol. Thus first-generation offspring were able to train ∼12% greater in total distance than the founder population. Our idea was to increase the training distance to no higher than 10,000 m and then maintain that value for all subsequent generations. The relatively large health benefits of modest exercise training underscore the importance of understanding the allelic variation for extremes of this trait (17).

Models incorporating higher threshold training genes could be derived from a founder population with high intrinsic capacity such as the HCR (high capacity runners) line that is being developed by artificial selective breeding (18). Presumably, members of the HCR line have retained genetic variation for response to training with the exception of a substrate lost via the inbreeding subsequent to selection in a finite population (∼1% per generation). Although of large interest, information about higher threshold training alleles would presumably apply to a smaller elite population.

Training responses can demonstrate an inverse association with intrinsic capacity, and this phenomenon has been termed the principle of initial values (23, 25). Data in Fig. 5, Aand B, show a slope significantly less than one for post- vs. pretraining values for both the founder and generation 1low line and are thus in accord with this principle. Although it is premature to draw concrete conclusions, at least two factors could account for this association. First, the low-line rats might self-train themselves across time by having more spontaneous movement in their home cages. Second, some of the genes causative of intrinsic capacity (i.e., pretraining capacity) might be physically linked to genes determining the response to training. Genome sequence data have revealed that genes of related function, and those connected in the same metabolic pathway, can occur as tandem arrays (7). Thus it is possible that evolution favored linkage patterns that created the potential for the simultaneous transmission of blocks of genes that encode for both low intrinsic capacity and high response. Despite this possibility, the generality of the principle of initial value may be overestimated because regression is often used to quantitate the relationship between pretraining capacity (independent variable) and the change in capacity posttraining (dependent variable). Because the change with training, often estimated as post- minus prevalues, is a function of pretraining, a significant negative relationship is derived axiomatically. That is, a regression requires that the variables be independent. Entry of random numbers for pre- and posttraining values when plotted as pre- vs. postvalues minus prevalues results in a negative slope.

The distance run (m) at exhaustion was used as the estimate of aerobic endurance capacity and thus breeding value. The single best trial of three was used because it was deemed the best indicator of capacity determined by intrinsic genetic composition (9). This idea of estimating the genetic component from the one best day of performance, rather than the average for all trials, for example, has two origins. First, the environment can have an infinite negative influence on capacity by reducing the distance run to zero. Factors such as subtle differences in housing or daily handling could cause a genetically superior rat to perform below its normal on a given day. Second, the environment can have only a finite positive influence on any test of capacity for a given genotype.

Two studies from Bouchard and colleagues that used individuals from the HERITAGE Family Study provide fundamental information about the response to exercise in humans. In the first study, Bouchard et al. (3) tested the hypothesis that individual differences in the response of maximal oxygen uptake to a standardized exercise training protocol aggregate within families. Twenty weeks of training on a cycle ergometer produced on average an increase in maximal oxygen uptake of ∼400 ml/min. Similar to our present finding in rats, Bouchard et al. found substantial variation in the response to training that ranged from essentially zero response to a little over 1,000 ml O2/min gain from training. Such variance can be partitioned into that originating from genetic and environmental factors. ANOVA revealed that there was 2.5 times more variance between families compared with within families. The maximal heritability for trainability was estimated at 47% for the change in maximal oxygen consumption adjusted for age and sex. Thus it appears that the response to aerobic training includes a substantial genetic component. In a second study, Bouchard et al. (5) performed a genomic scan to identify chromosomal regions linked to maximal oxygen consumption under two conditions: 1) individuals in the untrained, sedentary condition and 2) the same individuals after 20 wk of endurance training on a bicycle ergometer. For the sedentary condition, markers on 4q, 8q, 11p, and 14q were linked with maximal oxygen consumption, whereas markers on 1p, 2p, 4q, 6p, and 11p were linked with maximal oxygen consumption in response to training. A recent genome scan in rat genetic models identified quantitative trait loci on rat chromosomes 3 and 16 for aerobic capacity in untrained animals (27).

In summary, our data demonstrate wide variation for response to modest treadmill training, and ∼43% of this variation is determined by genetic factors (i.e., h2 = 0.43). Divergent artificial selection produced a 58% divide for response to training between the low and high lines in the first generation, with all of the difference due to gain in training for the high line. The positive gain for training of 161 m for the high line, with no significant change for the low line, suggests natural selection may have imposed constraints on selection for the low response to training. Given the complexity of aerobic capacity, the development of animal models that allow us to study genetic and environmental uniformity will be of great value in resolving this phenotype at the gene and molecular levels of organization.


We acknowledge the expert technical support of Krista Pettee, Mary Nelson, and Jonathan Shields and thank Marianne Miller Jasper for preparation of the manuscript.


  • This work was supported by National Heart, Lung, and Blood Institute Grant HL-64270.

  • Address for reprint requests and other correspondence: L. G. Koch, Physiology and Molecular Medicine, Medical College of Ohio, 3035 Arlington Ave., Toledo, OH 43614-5804 (E-mail: lkoch{at}

  • The costs of publication of this article were defrayed in part by the payment of page charges. The article must therefore be hereby marked “advertisement” in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

  • First published November 1, 2002;10.1152/japplphysiol.00851.2002


View Abstract