Genetic architecture of voluntary exercise in an advanced intercross line of mice

Abstract

Exercise is essential for health, yet the amount, duration, and intensity that individuals engage in are strikingly variable, even under prescription. Our focus was to identify the locations and effects of quantitative trait loci (QTL) controlling genetic predisposition for exercise-related traits, utilizing a large advanced intercross line (AIL) of mice. This AIL (G₄) population originated from a reciprocal cross between mice with genetic propensity for increased voluntary exercise [high-runner (HR) line, selectively bred for increased wheel running] and the inbred strain C57BL/6J. After adjusting for family structure, we detected 32 significant and 13 suggestive QTL representing both daily running traits (distance, duration, average speed, and maximum speed) and the mean of these traits on days 5 and 6 (the selection criteria for HR) of a 6-day test conducted at 8 wk of age, with many colocalizing to similar genomic regions. Additionally, seven significant and five suggestive QTL were observed for the slope and intercept of a linear regression across all 6 days of running, some representing a combination of the daily traits. We also observed two significant and two suggestive QTL for body mass before exercise. These results, from a well-defined animal model, reinforce a genetic basis for the predisposition to engage in voluntary exercise, dissect this predisposition into daily segments across a continuous time period, and present unique QTL that may provide insight into the initiation, continuation, and temporal pattern of voluntary activity in mammals.

according to dickinson and colleagues (17), “Locomotion, movement through the environment, is the behavior that most dictates the morphology and physiology of animals.” From an evolutionary perspective, sustained long-distance running may be a derived capacity of the genus Homo, originating approximately 2 million years ago, and appears to have been vital in shaping modern human physiological and anatomic architecture (see, e.g., Refs. 3, 9). Movement is also intimately associated with the ecology of animals and is vital for procuring food, finding mates, predator avoidance, and dispersal (see, e.g., Ref. 32). From a human health perspective, substantial evidence indicates that physical inactivity is an important risk factor for a number of chronic diseases, chief of which may be obesity and cancer (Refs. 30, 67; but see Ref. 69).

Despite the documented importance of exercise to health-related quality of life (2, 22, 47, 62), there remains considerable variation in human activity levels, even within a given society, sex, and age cohort, with many people remaining inactive or not exercising enough to realize the rewards (see, e.g., Ref. 19; see also Ref. 67). Consequently, emerging studies are now beginning to elucidate the genetic architecture underlying the predisposition for voluntary exercise, in order to better understand the nature of this important interindividual variability.

It has been well established in both human beings and mice that the predisposition to engage in voluntary activity is heritable (see, e.g., Refs. 21, 41, 63), but the locations of specific genetic markers associated with this predisposition are just beginning to be elucidated in humans (e.g., Refs. 8, 16, 57) and mice (e.g., Refs. 34, 42, 46, 72). Like studies will continue to improve our understanding of the biological factors controlling individual variation in voluntary physical activity levels and, in conjunction with data reviewed by Bray et al. (4), may aid clinicians in designing more effective physical activity-based therapies with targeted dosages and intensities (see Refs. 10, 40, 54).

The focus of the present study was to identify the locations and magnitudes of quantitative trait loci (QTL) controlling exercise-related traits, utilizing a large, moderately advanced intercross line (AIL) of mice. Through random intercrossing over multiple generations, the production of AILs can provide a more accurate approach to mapping loci by accumulating recombination events and providing increased mapping resolution (14). This G₄ population originated from a reciprocal cross between mice with genetic propensity for increased voluntary exercise [high-runner (HR) line] and the inbred strain C57BL/6J (B6). The HR line originated from a long-term replicated artificial selection experiment for high voluntary wheel-running behavior on days 5 and 6 of a 6-day wheel exposure (reviewed in Refs. 24, 64). By generation 16, and continuing through generation 50 and beyond, the HR lines (4 replicates) had diverged from the control lines (C lines, 4 replicates) with an approximate 2.5- to 3.0-fold increase in total revolutions per day. This was caused primarily by HR mice running faster rather than for more minutes each day, but the relative importance of the two components differs between the sexes (males show a significant increase in amount of time spent running) and among the four replicate HR lines (see Ref. 63 and Fig. 4 in Ref. 52). These lines of mice have been the focus of numerous studies characterizing the morphological, physiological, and behavioral traits that have evolved in concert with high levels of voluntary activity (reviewed in Refs. 24, 64).

Fig. 4.

G₄ QTL maps of average running speed (total revolutions/time spent running) on each of 6 days of wheel access, the mean from days 5 and 6, and running trajectories across the 6-day test. Slopes were not calculated for individuals missing 1 or more days of wheel-running data. Red traces are the simple mapping output, and black traces are the GRAIP permutation output. Gray shaded regions are either suggestive (P ≤ 0.1) or significant (P ≤ 0.05) at a genomewide level in the GRAIP results. Dotted line represents the permuted 95% LOD threshold.

In the present study, we genotyped >800 G₄ mice representing reciprocal crosses between HR and B6 with a genomewide single nucleotide polymorphism (SNP) panel. Our primary goal was to map QTL related to running traits on days 5 and 6 of the 6-day exposure to wheels, as this was the criterion for which the HR mice were selectively bred. However, it has long been recognized that wheel running activity varies temporally (60). Still, despite these long-standing observations, little is known with regard to the mechanistic underpinnings of the initiation, continuation, trajectory, or day-to-day variation in wheel running in rodents (39, 56). Thus a secondary goal of the present study was to map running traits associated with the initiation, continuation, and temporal pattern of voluntary wheel running behavior across the 6 days of wheel access.

MATERIALS AND METHODS

G₄ creation and phenotyping.

Full details of the creation and phenotyping of the G₄ population have been provided elsewhere (36), and only the pertinent features are presented here. Progenitor HR and B6 mice (n = 44, 22 males and 22 females per line) mice underwent a reciprocal breeding protocol to produce a F₁ generation. In subsequent generations (F₂, F₃, G₄), the two reciprocal cross-line populations (HR♀ × B6♂ and B6♀ × HR♂) were not mixed and always comprised 32 mating pairs each. From these mating pairs, no fewer than 16 unique families were represented in each reciprocal cross population. In each generation, interfamilial matings were assigned with a Latin square design to avoid inbreeding and increase the effective population size. After the F₃ generation, a large G₄ population was produced through extra parities for extensive phenotypic and genotypic data collection. Extra parities were generated by allowing the same sets of parents to produce multiple litters.

G₄ individuals (n = 815) at 8 wk of age were weighed (±0.1 g) and then exposed to running wheels (model 80850, Lafayette Instruments, Lafayette, IN; circumference = 1.1 m) for 6 days. Voluntary wheel running was recorded electronically in 1-min intervals for 23–24 h of each of the 6 days of wheel access. After the sixth day of wheel access, mice were weighed and killed via decapitation, and tissues were harvested. Throughout phenotyping, mice were provided a repeatable synthetic control diet (Research Diet D10001; 21 kcal% protein, 68 kcal% carbohydrate, 13 kcal% fat) and water ad libitum. All procedures were approved by and are in accordance with guidelines set forth by the Institutional Animal Care and Use Committee at the University of North Carolina at Chapel Hill.

From the wheel-running recordings, the following daily traits were calculated: distance (total revolutions), time spent running (cumulative 1-min intervals in which at least 1 revolution was recorded), average speed (total revolutions/time spent running), and maximum speed (highest number of revolutions in any 1-min interval within a 24-h period). In addition to daily traits, we calculated mean values of distance, time, average speed, and maximum speed on days 5 and 6 of the 6-day test. These traits are of particular interest because the mean number of total revolutions on days 5 and 6 was the criterion for which the HR line was selectively bred (63). Furthermore, using least-squares linear regressions, we estimated the slope and intercept for distance, time, average speed, and maximum speed across the 6 days of wheel exposure. If an individual did not have trait values for all 6 days of wheel exposure, then the corresponding slope and intercept values were omitted from analyses.

Descriptive statistics.

Descriptive statistics for body mass (before wheel access) and voluntary wheel-running traits (described above) are presented in Table 1 (for a comparison to the parental strains, see Ref. 45). Partial phenotypic correlations were performed in SAS (version 9.1, SAS Institute, Cary, NC) for body mass and mean running distance, time, average speed, and maximum speed on days 5 and 6 of the 6-day exposure to running wheels (Table 2). Correlations were adjusted for factors with known phenotypic effects (see Ref. 36), parent of origin [whether a G₄ individual was descended from a progenitor (F₀) cross of HR♀ × B6♂ or B6♀ × HR♂; coded as 1 or 0 respectively], sex, and wheel freeness (the number of wheel revolutions following acceleration to a given velocity). P values from partial correlations were adjusted for multiple comparisons with the false discovery rate procedure (12), controlling the overall Type I error rate at 5%. For simplicity, and because mean distance (on days 5 and 6) was the selection criterion for the HR line, we chose to present only the partial correlations for a subset of the 37 traits.

Table 1.

Descriptive statistics for phenotypic traits measured in G₄ population

Trait	n	Mean	SD	Range
Body mass	800	26.03	4.67	16.30–39.30
Revolutions
Day 1	753	8,525	3,104	287–23,739
Day 2	754	8,824	2,843	379–16,976
Day 3	784	8,996	3,174	848–22,161
Day 4	694	9,259	3,097	537–22,158
Day 5	797	10,278	3,121	2,828–22,053
Day 6	769	11,000	3,621	2,287–24,068
(Days 5 + 6)/2	767	10,663	3,251	2,600–23,061
Slope (days 1–6)	618	458	657	−1,747 to 2,663
Intercept (days 1–6)	618	8,014	3,038	−2,632 to 17,693
Time
Day 1	753	685	186	92–1,164
Day 2	754	573	146	89–963
Day 3	784	539	143	103–994
Day 4	694	523	130	68–874
Day 5	797	545	120	174–922
Day 6	769	560	132	210–991
(Days 5 + 6)/2	767	554	121	210–937
Slope (days 1–6)	618	−24	29	−134 to 129
Intercept (days 1–6)	618	668	173	−79 to 1,087
Average speed
Day 1	753	12.21	2.41	3.12–22.62
Day 2	754	15.23	2.65	4.26–26.90
Day 3	784	16.46	3.00	8.24–28.09
Day 4	694	17.34	3.01	5.10–28.14
Day 5	797	18.69	3.25	9.30–30.72
Day 6	769	19.42	3.48	10.49–32.01
(Days 5 + 6)/2	767	19.10	3.27	11.45–31.20
Slope (days 1–6)	618	1.37	0.61	−0.49 to 3.62
Intercept (days 1–6)	618	11.69	2.35	3.04–22.54
Maximum speed
Day 1	753	25.72	3.56	11.33–39.50
Day 2	754	28.15	3.93	14.67–43.00
Day 3	784	29.74	4.27	17.67–47.92
Day 4	694	30.86	4.21	17.33–49.92
Day 5	797	32.44	4.71	22.42–52.00
Day 6	769	33.24	4.87	20.50–51.67
(Days 5 + 6)/2	767	32.86	4.62	21.46–51.84
Slope (days 1–6)	618	1.45	0.87	−1.09 to 4.79
Intercept (days 1–6)	618	24.82	3.70	10.43–40.85

The following traits were measured for a 6-day exposure to running wheels: body mass (g) before exposure to running wheels, running distance (revolutions/day), time spent running (i.e., cumulative 1-min intervals in which at least 1 revolution was recorded), average speed (total revolutions/time spent running), and maximum speed (highest number of revolutions in any 1-min interval within a 24-h period). (Days 5 + 6)/2 is the mean of days 5 and 6 of a 6-day exposure to running wheels, the criterion for which the high-runner (HR) strain was selectively bred (63). Slope and intercept values were calculated from a linear regression across the 6-day test; slopes and intercepts were not calculated for individuals missing 1 or more days of wheel-running data.

Table 2.

Pearson partial correlations for mean voluntary running traits from days 5 and 6 of 6-day exposure to running wheels

Trait	Distance	Time	Average Speed	Maximum Speed
Body mass	0.034	0.045	0.011	0.066
Distance		0.796*	0.753*	0.643*
Time			0.222*	0.164*
Average speed				0.877*

Pearson partial correlations (r; controlling for sex, parent of origin, and wheel freeness) are shown for a subset of the 37 phenotypic traits presented in Table 1.

^*P < 0.05 after correction for multiple comparisons with the false discovery rate procedure (12).

Genotyping and linkage map.

A total of 815 G₄ mice were genotyped for 764 SNPs. SNPs were selected on the basis of their relatively even spacing across the genome and their predicted informativeness based on initial genotyping of representative individuals from the F₀ parental strains (n = 12, HR; n = 1, B6) using the Mouse Diversity array (71). Specifically, we used 362,000 SNPs present in the training array to identify SNPs with identical homozygous genotypes in the HR samples and informative with respect to B6. Genotyping in the G₄ was performed with the Sequenom (San Diego, CA) MassARRAY system as previously described (23). After genotyping, we confirmed that markers were fully informative by comparing F₀ HR mice (n = 32) to a subset of F₀ B6 (n = 8). We excluded any SNPs where common alleles were shared between HR and B6 F₀ parental strains. All fully informative SNPs were checked for errors in approximate Mendelian inheritance and segregation distortion. Additionally, taking into account pedigree structure and higher levels of recombination relative to a F₂ population, we estimated genotyping errors with the error detection function in Merlin (1) and dropped individual calls that were deemed extremely unlikely. The final set of SNPs (n = 530, with an average spacing of 4.7 Mb) used for QTL analyses is provided in Supplemental Table S1.¹ A genetic map was calculated with the R/qtl package (6) for the R environment (v.2.8.1) (51), treating the G₄ population as an F₂ (Supplemental Fig. S1).

QTL analyses.

In total, we evaluated 37 quantitative traits (listed in Table 1) for location and magnitude of underlying QTL. To appropriately account for family structure (nonindependence of individuals) in the G₄ population, we employed the Genome Reshuffling for Advanced Intercross Permutation (GRAIP) procedure (49). GRAIP uses a permutation scheme to create “randomized” populations that respect family structure. Parental (F₃) genotypes were first estimated with Merlin (1). GRAIP-randomized populations were then created. Each population was created by permuting the identities of the parents respecting sex. From each set of simulated F₃ progenitors, a simulated G₄ population was then created by simulating inheritance and recombination. These simulated populations respect the family structure of the original population, but any association between genotype and phenotype has been removed. Since family structure affects the association between genotype and phenotype, locus-specific and genomewide empirical P values can be estimated by using the distribution of P values for these permuted maps.

To generate permuted P values, QTL analyses were performed for the original population and the GRAIP-permuted populations (n = 50,000) with R/qtl. Within R/qtl, the multiple imputation method (55) was employed to handle missing data, drawing 16 times from possible genotypes at each missing locus. Appropriate statistical models had previously been defined (36) and included parent-of-origin type, sex, and wheel freeness. Parity (order of litters from individual F₃ dams) was not included in the model as there was no statistically significant effect on any wheel-running trait. When analyzing body mass, wheel freeness was excluded from the model and parity was added as an additional covariate. The X chromosome was analyzed in two ways. Because R/qtl is currently designed for F₂ populations, and requires the identity of the parental grandmother (coded as 0, 1) to most appropriately cope with the X chromosome, we analyzed the X chromosome by treating it as an autosome and utilizing the same additive covariates as described above. For comparison, we treated the X as a sex chromosome, allowing R/qtl to convert the X chromosome data to an internal standard by using the provided sex identifiers and by inferring the direction of the cross.

After R/qtl mapping of the original and permuted populations, we computed locus-specific P values as previously described (49, 50). In brief, utilizing the output from R/qtl, locus-specific P values were calculated for each marker of the original data set by utilizing the value for that specific marker in each of the permuted genome maps at each locus as a null distribution. We compared the null distribution for each marker with the value for the original G₄ mapping data in order to generate locus-specific P values at marker positions. P values were then interpolated at regular physical points on the genome, based on the known physical position of markers, and placed on a scaffold at regularly spaced sets of physical positions. Finally, we computed genomewide, adjusted P values by finding the minimum possible P values [or highest −log P, logarithm of odds (LOD)] from each GRAIP-permuted map by generating locus-specific P values for each permuted map as described above and extracting the lowest locus-specific P value from each permuted map. From this set of best locus-specific P values, we then generated an ordered list. It should be noted that genomewide GRAIP-adjusted significance thresholds were generated by utilizing 50,000 permutations. Therefore, for the GRAIP output, a minimum possible P value with 50,000 permutations is 0.00002 (1/50,000), so the maximum −log P = 4.7. Loci that met or exceeded 95th and 90th percentiles of this ordered list were deemed significant and suggestive, respectively. These percentiles are equivalent to an empirical genomewide P = 0.05 and P = 0.10, respectively. Confidence intervals (90–95%) of QTL locations were approximated by 1-LOD-drop support intervals in megabases (relative to the GRAIP-permuted LOD score) (5, 43, 44). The percent variation explained by each significant and suggestive QTL was extracted by standard linear regression by fitting the imputed QTL marker genotypes, and the additive QTL effects were expressed in phenotypic standard deviation units and as a percentage of the residual variance. In additional analyses, to test for possible covariate interactions with a QTL (i.e., the effect of the QTL varying with the covariate), we included QTL × sex and QTL × parent of origin factors in the model in a stepwise fashion. Significant interactions were identified when LOD_Full − LOD_Additive = LOD ≥ 3.0 (55).

The production of AILs provides an effective approach to map loci, but because of the complex breeding history the assumption of independence among individuals has been conclusively shown to be false, and several additional methods currently exist to account for family structure (33, 48, 68). Our multigenerational breeding protocol expanded the final generation by producing multiple litters from the same set of crosses. The 30 unique families were represented by 57 breeding pairs (for complete details on breeding history see Ref. 36). Each breeding pair contributed an approximately equal number of litters (mean 2, range 1–3) to the G₄ generation with a mean size of 7.5 (range 2–13). Although each individual in our testing population (G₄) was not derived from a unique pair in the breeding population (F₃), as assumed in Darvasi and Soller's (14) simulations, we maximized the number of crosses while minimizing the number of offspring resulting from each cross in an attempt to reduce the effects of family. Given the relatively short number of intercrosses and the generally well-balanced mating design used in this study, it is possible that the GRAIP-adjusted LOD scores are overly conservative for our population, and we thus in some cases present and discuss the naive or unadjusted LOD scores from the simple mapping output (i.e., Supplemental Table S2).

RESULTS

Descriptive statistics and partial phenotypic correlations are presented in Tables 1 and 2, respectively. All traits were either approximately normally distributed or slightly skewed and reasonably symmetric. In the G₄ population, mean wheel-running traits on days 5 and 6 of the 6-day test were significantly correlated with one another, while no running trait was significantly correlated with body mass after controlling for sex, parent of origin, and wheel freeness (Table 2).

Results for all QTL analyses are presented in Table 3, Figs. 1–5, and Supplemental Table S2. In total, 41 significant (P ≤ 0.05, LOD ≥ 3.9) and 20 suggestive (P ≤ 0.1, LOD ≥ 3.5) QTL were observed for the voluntary wheel traits and body mass after controlling for potential family structure with the GRAIP procedure. Additionally, we provide QTL detected and respective statistics for body mass and voluntary wheel-running traits from unadjusted output that were significant at the genomewide level (P ≤ 0.05, LOD ≥ 3.9), but did not remain significant or suggestive (P ≤ 0.1, LOD ≥ 3.5) after the GRAIP procedure (Supplemental Table S2). Regardless of the analysis method (see materials and methods), we did not observe any significant or suggestive QTL on the X chromosome.

Table 3.

QTL detected and respective statistics for body mass and voluntary wheel-running traits

Trait	Nearest Marker	MMU	Peak Position, Mb	Naive LOD	GRAIP LOD	CI, Mb	% Var	Additive ± SE	Dominance ± SE
Body mass	JAX00263199	1	115.6	6.9	3.5	95–141	1.1	0.6 ± 0.2	0.4 ± 0.3
	JAX00127022	5	10.6	9.4	4.7*	−16	1.8	−0.6 ± 0.2	−0.4 ± 0.3
	JAX00139789	6	36.3	10.6	4.7*	25–40	1.0	0.6 ± 0.2	−0.1 ± 0.3
	JAX00415862	16	24.3	7.3	3.5	11–28	1.5	−0.7 ± 0.2	0.4 ± 0.3
Distance
Day 1	JAX00240652	1	3.5	6.2	4.1*	−12	2.2	−669.3 ± 164.1†	−141.5 ± 226.4
	JAX00008045	1	112.7	5.9	3.9*	107–139	1.5	−492.7 ± 153.6	208.5 ± 227.8
	JAX00608826	6	46.8	6.4	4.4*	38–52	3.6	−761.2 ± 146.7†	−234.1 ± 225.0
	JAX00155508	7	108.9	7.4	4.7*	97–116	4.4	870.1 ± 148.2†	−258.8 ± 225.9
Day 2	JAX00009649	1	134.3	4.6	3.7	111–139	2.3	−463.3 ± 138.4	402.3 ± 212.0
	JAX00581735	5	50.0	4.5	3.5	48–66	2.6	375.3 ± 147.7	794.6 ± 205.1
	JAX00139789	6	36.3	7.3	4.7*	−60	2.8	−625.2 ± 135.8†	−12.8 ± 206.6
	JAX00155961	7	114.9	4.5	3.5	97–119	3.9	545.7 ± 141.6	286.8 ± 212.3
Day 3	JAX00582506	5	52.9	4.6	4.7*	51–59	2.9	325.6 ± 165.1	1,008.8 ± 224.7†
Day 4	JAX00155961	7	114.9	5.7	4.7*	101–130	3.2	648.2 ± 158.1†	537.5 ± 238.5
Day 5	JAX00155961	7	114.9	4.1	4.2*	98–129	2.6	595.8 ± 149.2	469.1 ± 226.6
Day 6	JAX00155508	7	108.9	4.0	3.7	100–120	2.1	638.3 ± 170.1	341.5 ± 264.4
(Days 5 + 6)/2	JAX00155508	7	108.9	4.2	4.2*	99–124	2.3	607.4 ± 152.9	273.0 ± 237.4
Slope (days 1–6)	JAX00025338	11	24.0	5.3	3.8	20–38	3.8	172.4 ± 38.2†	14.0 ± 52.8
Intercept (days 1–6)	JAX00240652	1	3.5	5.3	3.6	−23	3.7	−846.7 ± 174.2†	−78.6 ± 241.7
	JAX00008766	1	122.5	6.4	4.7*	110–136	3.9	−688.6 ± 166.3†	518.1 ± 249.1
	JAX00139789	6	36.3	7.3	4.4*	19–55	3.7	−774.9 ± 160.6†	−82.1 ± 242.0
	JAX00155743	7	112.1	5.0	3.5	102–118	3.0	664.0 ± 162.3†	−386.7 ± 243.7
Time
Day 1	JAX00009797	1	136.3	11.6	4.7*	92–151	5.1	−49.4 ± 9.0†	30.7 ± 13.6
	JAX00645408	7	82.6	6.7	4.1*	75–86	3.8	31.7 ± 9.4	−59.8 ± 13.3†
Day 2	JAX00253602	1	66.2	5.8	3.9*	30–77	2.5	−30.2 ± 8.0	10.9 ± 11.0
	JAX00582506	5	52.9	5.9	4.7*	49–58	3.1	28.4 ± 7.6	35.3 ± 10.5
	JAX00139228	6	28.8	6.8	4.7*	22–42	2.3	−29.7 ± 7.0†	−1.5 ± 10.6
	JAX00155508	7	108.9	5.7	4.0*	99–119	2.6	31.5 ± 7.0†	−0.2 ± 10.8
	JAX00350930	13	15.7	4.9	3.6	−22	2.2	−28.6 ± 7.3	7.3 ± 10.8
Day 3	JAX00154099	7	90.0	6.4	4.7*	75–117	3.1	27.1 ± 6.8†	−30.4 ± 10.4
Day 4	JAX00156517	7	122.4	7.7	4.7*	91–132	4.8	39.9 ± 6.8†	−1.12 ± 10.0
Day 5	JAX00155508	7	108.9	6.7	4.7*	92–131	3.5	29.5 ± 5.6†	2.3 ± 8.5
	JAX00478815	19	46.4	5.6	4.3*	41–49	2.2	−21.8 ± 6.6	16.8 ± 8.7
Day 6	JAX00155508	7	108.9	8.1	4.7*	93–127	4.1	34.5 ± 6.2†	7.0 ± 10.0
(Days 5 + 6)/2	JAX00155508	7	108.9	8.5	4.7*	91–129	4.3	32.3 ± 5.6†	5.4 ± 8.7
Slope (days 1-6)	JAX00008766	1	122.5	9.4	4.7*	58–141	6.6	9.6 ± 1.6†	−3.5 ± 2.3
	JAX00139228	6	28.8	5.9	3.8	22–48	3.3	6.7 ± 1.6†	3.3 ± 2.3
	JAX00026075	11	33.9	5.5	3.6	22–37	3.5	6.4 ± 1.6	5.0 ± 2.3
Intercept (days 1–6)	JAX00009649	1	134.3	9.9	4.7*	85–142	5.5	−46.9 ± 9.1†	32.1 ± 13.8
	JAX00139789	6	36.3	8.1	4.7*	−45	3.3	−41.7 ± 9.2†	−5.2 ± 13.8
	JAX00041702	13	10.5	6.3	4.6*	−23	2.6	−38.1 ± 9.8	3.4 ± 14.1
Average speed
Day 2	JAX00436582	17	33.2	5.4	4.7*	27–47	3.0	0.6 ± 0.1†	0.4 ± 0.2
Day 3	JAX00496243	2	91.8	4.4	3.6	81–106	2.3	−0.3 ± 0.2	0.7 ± 0.2
	JAX00441944	17	52.3	4.5	3.8	29–69	2.0	0.6 ± 0.2†	−0.01 ± 0.22
Day 5	JAX00385288	14	79.9	3.8	3.8	68–92	2.0	0.5 ± 0.2	0.5 ± 0.2
Day 6	JAX00097778	2	99.0	5.1	4.2*	80–103	3.4	−0.8 ± 0.2	0.4 ± 0.3
	JAX00385288	14	79.9	4.1	3.7	68–92	2.3	0.6 ± 0.2	0.6 ± 0.3
(Days 5 + 6)/2	JAX00097778	2	99.0	4.3	3.8	81–103	3.0	−0.7 ± 0.2	0.4 ± 0.2
	JAX00385288	14	79.9	4.2	3.9*	69–92	2.3	0.6 ± 0.2	0.5 ± 0.2
Intercept (days 1–6)	JAX00037863	12	76.6	4.8	3.9*	73–81	3.4	0.1 ± 0.1	0.9 ± 0.2†
Maximum speed
Day 2	JAX00496243	2	91.8	4.1	3.7	85–96	2.8	−0.4 ± 0.2	1.0 ± 0.3
Day 3	JAX00096585	2	82.8	5.3	4.2*	80–105	2.8	−0.7 ± 0.2	0.7 ± 0.3
	JAX00024300	11	9.9	4.7	3.9*	−13	2.8	1.0 ± 0.2†	0.3 ± 0.3
	JAX00311223	11	53.2	4.4	3.6	46–68	2.7	1.0 ± 0.2†	0.1 ± 0.3
Day 4	JAX00311223	11	53.2	5.1	4.7*	48–61	3.5	0.9 ± 0.2†	0.6 ± 0.3
Day 5	JAX00498192	2	102.8	3.9	3.5	81–115	2.6	−0.7 ± 0.3	0.9 ± 0.3
Day 6	JAX00496243	2	91.8	5.5	4.2*	78–114	4.3	−1.0 ± 0.3	1.1 ± 0.4
	JAX00311223	11	53.2	5.2	4.4*	46–62	2.5	1.0 ± 0.2†	0.4 ± 0.4
(Days 5 + 6)/2	JAX00496243	2	91.8	4.8	4.1*	80–115	3.8	−0.8 ± 0.3	1.1 ± 0.4
	JAX00024300	11	9.9	4.1	4.1*	7–14	1.8	0.9 ± 0.3	0.1 ± 0.3
	JAX00311223	11	53.2	4.6	4.2*	45–61	2.1	0.9 ± 0.2	−0.4 ± 0.3

The following traits were measured for a 6-day exposure to running wheels: body mass (g) before exposure to running wheels, running distance (revolutions/day), time spent running (i.e., cumulative 1-min intervals in which at least 1 revolution was recorded), average speed (total revolutions/time spent running), and maximum speed (highest number of revolutions in any 1-min interval within a 24-h period). (Days 5 + 6)/2 is the mean of days 5 and 6 of a 6-day exposure to running wheels, the criterion for which the HR strain was selectively bred (63).Slope and intercept values were calculated across the 6-day test; slopes and intercepts were not calculated for individuals missing 1 or more days of wheel-running data.

^*Genome Reshuffling for Advanced Intercross Permutation (GRAIP) logarithm of odds (LOD) exceeding the 95% (P ≤ 0.05, LOD ≥3.9) permutation threshold; other QTL exceeded the 90% (P ≤ 0.1, LOD ≥3.5) threshold. Confidence intervals (CIs) for QTL positions were obtained with a 1.0-LOD drop in Mb (relative to the GRAIP-permuted LOD score). %Var is the percentage of phenotypic variance accounted for by the QTL effect. For additive and dominance effects, positive values indicate increasing effect of the HR allele or increasing effect of the heterozygote, respectively.

^†Additive and dominance effects statistically significant at P < 0.05.

Fig. 1.

G₄ quantitative trait locus (QTL) maps of body mass before running wheel exposure. Red traces are the simple mapping output, and black traces are the Genome Reshuffling for Advanced Intercross Permutation (GRAIP) permutation output. Genomewide GRAIP-adjusted significance thresholds were generated utilizing 50,000 permutations. Therefore, for the GRAIP output, a minimum possible P value with 50,000 permutations is 0.00002 (1/50,000), so the maximum −log P = 4.7. Gray shaded regions are either suggestive (P ≤ 0.1) or significant (P ≤ 0.05) at a genomewide level in the GRAIP results. Black and gray lines represent the permuted 95% and 90% logarithm of odds (LOD) thresholds, respectively.

Fig. 5.

G₄ QTL maps of maximum running speed (highest number of revolutions in any 1-min interval within a 24-h period) on each of 6 days of wheel access, the mean from days 5 and 6, and running trajectories across the 6-day test. Slopes were not calculated for individuals missing 1 or more days of wheel-running data. Red traces are the simple mapping output, and black traces are the GRAIP permutation output. Gray shaded regions are either suggestive (P ≤ 0.1) or significant (P ≤ 0.05) at a genomewide level in the GRAIP results. Dotted line represents the permuted 95% LOD threshold.

After adjusting for the family structure in the G₄ population utilizing the GRAIP procedure, two significant and two suggestive QTL were detected for body mass on MMU5, MMU6, MMU1, and MMU16, respectively. Figure 1 depicts both the unadjusted and the GRAIP-adjusted permuted output.

In total, GRAIP-adjusted output revealed 11 significant and 7 suggestive QTL across the 9 different running distance (revolutions/day) traits. These QTL represented both daily running distances, the mean on days 5 and 6, and the slope and intercept across all 6 days of running (Fig. 2). Running distance QTL individually accounted for 1.5–4.4% of the total phenotypic variation. QTL on MMU7 were either significant or suggestive for running distance on all days (except day 3) and the mean on days 5 and 6. On day 3, a peak on MMU7 was not significant or suggestive in the GRAIP-permuted output (LOD = 3.0), but the unadjusted mapping output revealed a LOD score of 3.5 (Fig. 2). Analyses of total revolutions across all 6 days revealed results (unadjusted LOD = 6.4 at 108.9 Mb on MMU7) similar to those for the daily measures.

Fig. 2.

G₄ QTL maps of running distance (revolutions/day) on each of 6 days of wheel access, the mean from days 5 and 6, and running trajectories across the 6-day test. Slopes were not calculated for individuals missing 1 or more days of wheel-running data. Red traces are the simple mapping output, and black traces are the GRAIP permutation output. Gray shaded regions are either suggestive (P ≤ 0.1) or significant (P ≤ 0.05) at a genomewide level in the GRAIP results. Dotted line represents the permuted 95% LOD threshold.

Although QTL on MMU7 exhibited a strong and consistent day-to-day pattern, additional significant and suggestive QTL were found to be unique only to days 1–3. On days 1–3, QTL were found on MMU1 (n = 3), MMU5 (n = 2), and MMU6 (n = 2) (Table 3 and Fig. 2). Thus it appears the genetic architecture for running distance can change across time, with some QTL remaining constant while others appear only during the initial exposure to wheels. With regard to slope of wheel running distance across all 6 days, a suggestive QTL was discovered on MMU11. QTL were also discovered for the intercept of the linear regression on MMU1, MMU6, and MMU7, and the locations were close to those observed for running distance on day 1.

For time spent running (i.e., cumulative 1-min intervals in which at least 1 revolution was recorded), 16 significant and 3 suggestive QTL were discovered, many of them appearing to colocalize with those observed for running distance. QTL represented daily duration values, the mean on days 5 and 6, and the slope and intercept across all 6 days (Fig. 3). QTL individually accounted for 2.2–6.6% of the total phenotypic variation for time spent running. As observed for running distance, QTL on MMU7 (significant or suggestive) exhibited a consistent pattern for running duration on all days and the mean on days 5 and 6.

Fig. 3.

G₄ QTL maps of time spent running (i.e., cumulative 1-min intervals in which at least 1 revolution was recorded) on each of 6 days of wheel access, the mean from days 5 and 6, and running trajectories across the 6-day test. Slopes were not calculated for individuals missing 1 or more days of wheel-running data. Red traces are the simple mapping output, and black traces are the GRAIP permutation output. Gray shaded regions are either suggestive (P ≤ 0.1) or significant (P ≤ 0.05) at a genomewide level in the GRAIP results. Dotted line represents the permuted 95% LOD threshold.

Running time QTL were also observed that were inconsistent across the entire wheel access period. As observed for running distance, significant and suggestive QTL were discovered on days 1–3 that were not observed on days 4–6 or for the mean running duration on days 5 and 6. On days 1–3 significant QTL were found on MMU1 (n = 2), MMU5 (n = 1), MMU6 (n = 1), and MMU13 (n = 1, suggestive) (Table 3 and Fig. 3). Additionally, analysis of day 5 running duration revealed a significant QTL on MMU19 that was not observed on any other day. Significant QTL were discovered on MMU1 (slope and intercept), MMU6 (slope and intercept), MMU11 (slope only), and MMU13 (intercept only) and found in regions similar to those observed for daily traits.

Average running speed (total revolutions/time spent running) analyses revealed four significant and five suggestive QTL found on MMU2, MMU12, MMU17, and MMU14. QTL represented daily running average speed and the mean average speed on days 5 and 6 (Fig. 4). QTL individually accounted for 2.3–3.4% of the total phenotypic variation for average running speed. Daily average running speed QTL (Fig. 4) represented less of a temporal pattern compared with running distance (Fig. 2) or time spent running (Fig. 3), with no QTL observed on the same chromosome for more than two consecutive days. No QTL were detected for slope, but one significant QTL, on MMU12, was discovered for the intercept of the linear regression across all 6 days, but it did not appear to colocalize with any QTL observed for average speed on individual days (Table 3).

Analyses of maximum running speed (highest number of revolutions in any 1-min interval within a 24-h period) revealed eight significant and three suggestive QTL across MMU2 and MMU11. QTL represented daily maximum running speed and the mean average speed on days 5 and 6 (Fig. 5). QTL individually accounted for 1.8–4.3% of the total phenotypic variation for maximum running speed. Although not significant in the GRAIP-permuted output, peaks on day 4 (MMU2) and day 5 (MMU11) each had unadjusted LOD scores of 3.4 (Fig. 5). Considering the former, daily QTL were reasonably consistent across all days with the exception of day 1, where no significant or suggestive signals were observed. Contrary to what was observed for running distance and duration, no QTL was unique to the initial wheel exposure or any single day. No QTL were detected for the slope or intercept when examining trajectory of maximum running speed across all 6 days.

Most QTL had increasing effects resulting from the HR allele, but these effects were often day dependent (Table 3). For example, for running distance, increasing effects of the B6 allele were often observed for the initial days of wheel exposure (especially day 1), while for the final days increasing effects were noted for the HR allele. Average additive QTL effects were frequently significant and exhibited similar temporal patterns. Average dominance effects were large for most running traits examined. And, notably, in three cases we found significant dominance effects in the absence of significant additive effects: running distance on day 3, running time on day 1 (MMU7), and the intercept of average running speed.

Separate analyses (of the QTL presented in Table 3) investigated QTL × sex and QTL × parent of origin factors in a stepwise fashion and revealed statistical evidence for parent-of-origin-specific QTL in three cases. Here we present unadjusted LOD scores from these analyses, as we have already demonstrated significance after accounting for family structure. First, body mass QTL on MMU6 showed a significant QTL × parent of origin interaction (LOD_Full − LOD_Additive = 7.6). Separate analyses of the parent-of-origin types revealed unadjusted LOD scores of 0.2 for individuals descended from a progenitor cross (F₀) of HR♀ × B6♂ and 17.8 for individuals descended from B6♀ × HR♂. Second, we observed a significant QTL × parent of origin interaction (LOD_Full − LOD_Additive = 3.4) for distance QTL on MMU1 (112.7 Mb). Separate analyses revealed a LOD of 6.9 for mice descended from HR♀ × B6♂ compared with a LOD of 1.2 for from the reciprocal cross. Finally, a significant interaction (LOD_Full − LOD_Additive = 3.2) was observed for the slope of time spent running (MMU11) (LOD = 0.6, HR♀ × B6♂; LOD = 8.7, B6♀ × HR♂).

DISCUSSION

To date, QTL associated with mouse wheel-running activity have been mapped in either second-generation intercross or backcross populations (e.g., Refs. 42, 46). Although many methods exist to map individual QTL (13), the AIL approach employed here enables finer mapping of many QTL with a single population. By generating higher levels of recombination throughout the genome, the entire genome is lengthened in terms of centimorgan distance (compared to a F₂), providing increased mapping resolution in the AIL and reductions in the confidence intervals of map locations (14). In this study, the production of a G₄ population resulted in an approximate threefold expansion (averaged across all chromosomes) of the genetic map relative to a new standard map for the laboratory mouse (see the Revised Shifman map lengths in Table 1 of Ref. 11) (comparisons are depicted in Supplemental Fig. S1). This map expansion is, as expected, less than what was observed for more advanced intercrosses (see, e.g., Ref. 48). However, use of this intermediate stage of the AIL permitted quicker access into the genetic architecture of voluntary exercise, and we have maintained the AIL (now at G₉) for potential follow-up fine mapping targeted at the genomic regions identified here.

We observed the strongest signals for wheel running distance and duration. Our results revealed a generally consistent pattern (as evidenced by overlapping confidence intervals) for running distance and duration across all 6 days, with QTL primarily found in a region on MMU7, with significant additive effects resulting from the HR allele. These pleiotropic effects are reasonable given that running distance is a product of the amount of time spent running and the speed at which an individual runs. And, given the high correlation between running distance and running time, further analysis of the mean distance on days 5 and 6 was conducted with running time as an additional covariate. As expected, this analysis resulted in a reduction of the LOD score of the QTL on MMU7 (naive LOD; without time as a covariate = 4.2, with time as a covariate = 1.5). We did not observe any significant or suggestive QTL on MMU7 for average running speed or maximum running speed. This pattern is different from that previously observed by Lightfoot et al. (42) and Nehrenberg et al. (46), where significant or suggestive QTL for running speed were found to colocalize with regions for running distance.

Lightfoot et al. (42) identified four QTL that were deemed to be significant. These QTL represented running duration (DUR13.1), speed (SPD9.1 and SPD13.1), and distance (DIST13.1), with the QTL for running speed (SPD9.1) accounting for the largest percentage of phenotypic variance (11.3). These major QTL do not directly overlap with the QTL identified here, but direct comparisons to Lightfoot et al. (42) are difficult because they examined running values across all 21 days of wheel access, while we primarily examined daily values and mean values on days 5 and 6 of wheel access. Moreover, Lightfoot et al. (42) generated their F₂ mapping population from different mouse strains (C57L/J and C3H/HeJ) than those utilized here. A forthcoming common set of mice (the Collaborative Cross), derived from a diverse set of eight founder strains and designed for the analysis of complex traits, should, in our opinion, partially mitigate the need for comparisons of isolated mapping populations (65). However, we do feel that the creation of intercross and backcross populations involving phenotype-specific strains (such as HR) will remain important.

Nehrenberg et al. (46) found little evidence of significant QTL for running time in general. Contrary to the present investigation, Nehrenberg et al. (46) employed a backcross design and an alternate replicate HR line (4 currently exist). The HR line utilized in Nehrenberg et al. (46) is fixed for a Mendelian recessive allele (26) that causes an approximate 50% reduction in hindlimb muscle mass and has been mapped a to a 2.6335-Mb region between 67.453 and 70.0865 Mb on MMU11 (29). In addition to alterations in muscle mass, this replicate line exhibits a number of phenotypic differences compared with the HR line utilized here, most importantly increases in running speed (Ref. 28 and references therein). However, the QTL previously detected by Nehrenberg et al. (46) for running distance and speed and the QTL observed here for distance and duration were both found in reasonably close approximation to the tyrosinase (tyr) gene (∼94.6 Mb) on MMU7. This is particularly intriguing given evidence that tyrosinase can serve as a precursor for dopamine, a neurotransmitter previously demonstrated to be involved in voluntary movement and predatory aggression (53). The other prominent QTL identified by Nehrenberg et al. (46) on MMU6 (for maximum running speed) does not directly overlap with those identified here.

Although individual days generally shared some common QTL, the initial exposure (days 1 and 2) to wheels and the trajectory of running traits across the entire access period revealed some novel findings. During the initial exposure to running wheels (days 1 and 2), we have demonstrated that unique genomic regions are least partially responsible for running distance and duration as revealed by significant and suggestive QTL on MMU1, MMU5, and MMU6. In most of these cases, the B6 allele had significant additive effects, with the notable exception of the QTL detected on MMU7, where the HR allele always had an additive effect (and in most cases a significant one). These temporal differences in additivity may be illustrative of variation in anxiety- or fear-related behavioral differences (e.g., as might be measured by open-field behavior) between HR and B6 mice.

Regions on MMU1 have previously been implicated in both home-cage activity (34) and open-field behavior (27). Kas et al. (34) utilized a chromosome substitution strain to identify a 312-kb QTL interval at 80 Mb on MMU1 containing a single gene (A830043J08Rik) associated with home-cage activity. Gene expression profiling further identified a gene (Epha4) outside of the QTL interval as a strong candidate downstream involved in motor activity via the neuronal circuitry controlling movement. Distinct from home-cage activity, but still located on MMU1, loci for open-field behavior have been mapped in close proximity to 145 Mb (70), 175 Mb (31), 100 Mb (15), and 190 Mb (58, 59). These regions have been shown to harbor genes involved in anxiety-like behavior in rodents, and human homologs have been associated with panic disorder (38). Thus, on the basis of our present findings, we preliminarily conclude that fear, or lack thereof, of a novel object (e.g., a running wheel), or more general anxiety resulting from novel solitary housing conditions, may contribute to wheel running during initial exposure to wheels. Additionally, given the results of Kas et al. (34), regions on MMU1 may play a role in the initial “learning” (broadly involving neural circuitry) process involved with wheel running. Follow-up investigations will be needed to elucidate a clearer picture of the regions on MMU1 identified here and their putative role in wheel-running behavior. It is worth noting that variation in the regulation of sex hormones may also be playing an important role during the initiation and continuation of wheel running (see Ref. 39); however, we did not quantify estrogen/testosterone levels in the present study, and this may have diminished our power of QTL detection.

Our efforts, along with those of Nehrenberg et al. (46), have now led to the identification of multiple QTL underlying activity-related phenotypes in the context of an artificial selection experiment for increased voluntary wheel running. Although these QTL individually and collectively only explain a small fraction of the phenotypic variance in activity measures, they potentially represent genomic regions that have been (or currently are) under positive selection. We acknowledge the difficulties in relating the importance of the present results (and those of Ref. 46) to the phenotypic divergence in wheel running seen between HR and control mice (e.g., see Fig. 1 in Ref. 37). First, we have utilized B6 in the creation of the G₄ as opposed to the control lines derived from the Hsd:ICR strain [Harlan-Sprague-Dawley (HSD), Indianapolis, IN]. Second, we cannot rule out genetic drift as we are only examining one of the four replicate HR lines. However, given that nearly all of the allelic effects from mean running traits on days 5 and 6 associated the HR allele with increased running with partial replication [compared with Nehrenberg et al. (46)], we feel this provides reasonably strong evidence that at least some of the identified genomic regions have been influential during the evolution of voluntary wheel running in the context of this artificial selection experiment. Many adaptive changes in exercise physiology, as well as motivational aspects of voluntary running, have been observed in HR mice compared with their ICR controls (see Refs. 24, 53, 64). Presently, we do not know which component (motivation or ability) most accounts for variation in wheel running traits or QTL identified in this mapping population. However, follow-up investigations are profiling gene expression in brain and muscle tissue in a selection of G₄ mice in the hopes of providing some insight into these two aspects of voluntary exercise, which may or not be mutually exclusive.

Average dominance effects of QTL were in most cases large and appear to be playing an important role in the regulation of voluntary wheel running. These findings support those of previous investigations examining wheel running in F₁ populations. Dohm et al. (18) observed net dominance in the direction of high wheel running in an F₁ population resulting from wild-captured house mice and ICR (the base population of HR) mice. Additionally, Nehrenberg et al. (45) observed significant heterotic inheritance of wheel running behavior in F₁ individuals from crosses of HR and C57BL/6J mice (identical to the strains utilized here). And, to our knowledge, the most comprehensive examination of heterotic inheritance of wheel running in mice was conducted by Bruell (7) and involved 4,000 mice from 13 inbred strains and 31 hybrid groups, with heterosis observed for a significant number of the hybrids.

In addition to what might initiate wheel running, we also attempted to identify genomic regions controlling temporal variation (or the trajectory) in wheel running. We mapped the slope and intercept of a linear regression for running distance, duration, average speed, and maximum speed across all 6 days of the testing period (for hypothetical examples, see Fig. 4 in Ref. 25). Here, we report the first ever, to our knowledge, QTL associated with the trajectory of running across multiple days of wheel exposure. As expected, the intercept QTL were found in regions similar to the QTL peaks identified on the initial day of exposure. However, the QTL observed for the slope of the exercise-related traits often did not coincide with locations of the individual day QTL. For example, we identified a peak on MMU11 for the slope of wheel running distance but did not observe a peak on MMU11 for running distance on any of the individual days. Therefore, it is possible that the global trajectory of exercise behavior on longer timescales is at least partially controlled by different genomic regions than the behavior on individual days. Although further studies are needed, these regions may prove especially important given the importance of physical activity in the maintenance of weight regulation.

Previously in this G₄ population, we reported significant effects of sex and parent of origin, and in some cases interactions between these two effects, on voluntary wheel traits and body composition (36). Formerly, we hypothesized that the mechanistic regulation of these observed parent-of-origin effects may be genetic (i.e., X-linked or mtDNA variations), epigenetic (i.e., genomic imprinting), or environmental (i.e., in utero environment or maternal care) phenomena. Given the lack of observed QTL on the X chromosome, we can preliminarily rule out direct genetic effects as an explanation for the observed parent-of-origin effects on voluntary wheel-running traits. With regard to genomic imprinting, we observed QTL × parent of origin interactions for only a small number of QTL. However, we only examined potential interactions for the QTL that were initially significant by utilizing additive models (Table 3). Future studies will be needed to more thoroughly understand QTL × parent of origin interactions across the entire genome, whether these potentially significant effects lie within known imprinting regions, and the explanatory power of the parent-of-origin specific QTL to the percent phenotypic variance.

Results of the present investigation are an important step in continuing efforts to elucidate the genetic architecture of voluntary exercise levels. The large number of QTL discovered here (and by others) suggests that many genomic elements contribute to the predisposition for voluntary exercise, but the identities and nature of the underlying genetic variation are not yet well understood. However, as studies involving all aspects of activity (wheel running, home cage, open field, etc.) in rodents are beginning to emerge and converge, the intricacies of such a complex behavior as voluntary exercise are beginning to become clearer. And, while translation from mouse to human is uncertain, given the parallels detailed in Ref. 20 we are optimistic that investigations into the genetic architecture of voluntary wheel running in rodents will have positive consequences for our understanding of the variation in exercise behavior in human populations.

GRANTS

This work was supported by National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK) Grant DK-076050 to D. Pomp and in part by National Institute of General Medical Sciences National Centers of Systems Biology Program Grant GM-076468 to F. P. M. de Villena. S. A. Kelly was supported through a National Institute of Mental Health-funded (5T32MH075854-04) Interdisciplinary Obesity Training (IDOT) program. Phenotypes were collected with the Animal Metabolism Phenotyping core facility within UNC's Clinical Nutrition Research Center (funded by NIDDK Grant DK-056350).

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the author(s).