Exercise and diet affect quantitative trait loci for body weight and composition traits in an advanced intercross population of mice

Abstract

Driven by the recent obesity epidemic, interest in understanding the complex genetic and environmental basis of body weight and composition is great. We investigated this by searching for quantitative trait loci (QTLs) affecting a number of weight and adiposity traits in a G₁₀ advanced intercross population produced from crosses of mice in inbred strain C57BL/6J with those in a strain selected for high voluntary wheel running. The mice in this population were fed either a high-fat or a control diet throughout the study and also measured for four exercise traits prior to death, allowing us to test for pre- and postexercise QTLs as well as QTL-by-diet and QTL-by-exercise interactions. Our genome scan uncovered a number of QTLs, of which 40% replicated QTLs previously found for similar traits in an earlier (G₄) generation. For those replicated QTLs, the confidence intervals were reduced from an average of 19 Mb in the G₄ to 8 Mb in the G₁₀. Four QTLs on chromosomes 3, 8, 13, and 18 were especially prominent in affecting the percentage of fat in the mice. About of all QTLs showed interactions with diet, exercise, or both, their genotypic effects on the traits showing a variety of patterns depending on the diet or level of exercise. It was concluded that the indirect effects of these QTLs provide an underlying genetic basis for the considerable variability in weight or fat loss typically found among individuals on the same diet and/or exercise regimen.

it is well established that body weight, weight gain, and obesity have a complex genetic and environmental basis (13, 46). Among the many environmental factors influencing these sorts of traits, diet and exercise probably have been the most studied. With regard to diet, increased caloric (particularly high fat) consumption appears to be the major factor that accounts for an increase in the average body weight in human populations during the last three decades (59). Other factors clearly are involved, however, because some individuals who consume a high-fat diet gain little weight (51). With regard to exercise, it has long been known that physical activity of various kinds tends to decrease the probability of obesity and associated health problems such as various cancers (3, 6, 8, 53). Yet this trend is not always clear, and there typically is considerable variability in the extent of weight loss even among individuals who routinely engage in exercise (25, 30, 60).

Although these and other environmental factors affecting weight and/or weight gain have received the most attention because of the present obesity epidemic, much is known about the genetic basis of these traits as well. The bulk of the genetic variation in weight and obesity in most populations appears to be quantitative (29). Consistent with this, by 2005 the number of quantitative trait loci (QTLs) found to affect obesity and associated traits exceeded 200 in humans and 400 in mice (49). And by the end of 2010, single nucleotide polymorphism (SNP) variants with replicable effects on body weight were identified in 17 different regions throughout the human genome (28).

Currently there is much interest in identifying QTLs for body weight and related traits that significantly interact with diet or exercise and thus at least partially determine the extent to which individuals will lose or gain weight depending on their diet (23, 45, 54) or level of exercise (22, 48). Mice have been particularly useful models for these searches, and in fact several studies have uncovered these sorts of interactions. For example, both Gordon et al. (24) and Cheverud et al. (13) made use of intercross populations of mice fed either a high- or a low-fat diet and found that roughly half of the QTLs affecting various obesity-related traits also showed interactions with the dietary environment. Studies scanning the genome for body weight or obesity QTL × exercise interactions appear to be quite rare, although one example is an investigation by Leamy et al. (37), who discovered a number of such interactions in mice that indicated QTLs affecting several physical activity traits behave variably depending on the phenotypic values for body weight.

For this study, we made use of a G₁₀ advanced intercross population produced from crosses of mice in inbred strain C57BL/6J with those in a strain selected for high voluntary wheel running (HR) to search for QTLs affecting a number of exercise, weight, and adiposity traits. The mice in this population were fed either a high-fat or a control diet, and weight and body composition were measured both before and after exercise, so this afforded an excellent opportunity for us to test for QTL by diet and QTL by exercise interactions as well. Kelly et al. (31, 33) previously used mice in the G₄ generation from this cross and found a number of QTLs for these same traits. It therefore also was of interest to compare our results with theirs to assess the extent of QTL replication. Because the G₁₀ population accumulated additional recombinants since the G₄ generation, we expected to find increased mapping resolution with reduced QTL support intervals (17), thus permitting reasonable inferences about potential candidate genes underlying these QTLs.

MATERIALS AND METHODS

Population and phenotypic traits.

The advanced intercross population used in this study was originally produced from reciprocal crosses of mice from the C57BL/6J (B6) inbred strain with those from the HR strain (33). The HR strain of mice was derived from one of four lines subjected to long-term artificial selection for high voluntary wheel running on days 5 and 6 after wheel exposure (58). This line was derived from the outbred, genetically variable Hsd:ICR (HSD) strain (Harlan Sprague-Dawley, Indianapolis, IA) of Mus domesticus and had experienced 44 generations of artificial selection at the time of the original crosses with B6 mice (58).

Mice in the F₁ and three subsequent generations (F₂, G₃, G₄) were derived from the two reciprocal mating types (B6 males × HR females and B6 females × HR males) and were always kept separate. In the next three generations, interfamilial matings were done according to a Latin square protocol designed to reduce inbreeding to the extent possible (33). Following the G₄ generation, only one of the two reciprocal types (B6 females × HR males) was carried through the G₁₀ and subsequently utilized here. Interfamilial matings for generations G₅–G₉ were made as before. By the ninth generation, mice from the B6 female × HR male mating type were crossed to produce the G₁₀ population, with each mating pair being allowed to produce two litters. At ∼4 wk of age, all G₁₀ mice were randomly allocated into either a group fed a high-fat diet or a group fed a control diet (see Table 1). Altogether, 473 G₁₀ mice were generated and available for the analysis, including 234 males (116 in the control and 118 in the high-fat diet group) and 239 females (118 in the control and 121 in the high-fat diet group).

Table 1.

Description of the control (D10001) and high-fat diets (D0902903)

	Control Diet		High-fat Diet
	g%	kcal%	g%	kcal%
Protein	20	21	25	21
Carbohydrate	66	68	41	34
Fat	5	12	24	45
Kcal/g	3.90		4.77
Ingredients	g	kcal	g	kcal
Casein	200	800	200	800
dl-Methionine	3	12	3	12
Corn starch	150	600	50	200
Maltodextrin	0	0	100	400
Sucrose	500	2000	173.8	695
Cellulose	50	0	50	0
Corn oil	50	450	50	450
Lard	0	0	145	1,305
Mineral mix S10001	35	0	35	0
Vitamin mix V10001	10	40	10	40
Choline bitartrate	2	0	2	0
Total	1,000	3,902	818.8	3,802

At an average of 8 wk (53–59 days) of age, these mice were weighed (0.1 g), and their fat tissue and lean tissue were measured (0.01 g) with a quantitative magnetic resonance imaging system (Echo Medical Systems, Houston, TX). All fat and lean tissue weights were divided by the total weight of the mice and then multiplied by 100 to convert them into percentages of fat tissue and lean tissue.

After these (pre-exercise) measurements were made (59–65 days of age), the G₁₀ mice were individually exposed to running wheels. Four separate voluntary exercise traits were generated from electronic recordings taken in 1 min intervals during each of 6 consecutive days of wheel access as previously described for the G₄ population (31). The timings of matings were such that mice were measured in one of 13 different cohorts that were tested sequentially. The traits measured included 1) distance run (total daily wheel revolutions), 2) time spent running (cumulative number of intervals run where at least one revolution occurred), 3) average speed (total revolutions/time), and 4) maximum speed (highest number of revolutions in any interval during a 24 h period). Although these four traits were measured on each of the 6 days, we chose to use only their average on days 5 and 6 since this was the criterion used for selection in the HR line (58).

Immediately after the exercise period (postexercise) the G₁₀ mice again were weighed and their percentages of fat and lean tissue calculated. Beyond pre- and postexercise measures, this allowed the calculation of the percentage change in body weight and composition in response to the 6 days of voluntary wheel running. For weight, this change was calculated as

$$\frac{\text{postexercise weight}-\text{pre-exercise weight}}{\text{pre-exercise weight}}\times 100$$

Changes in the percentage of fat and lean tissue were calculated in a similar fashion. In addition, food was weighed (0.1 g) before and after wheel access, and these values were subtracted and then expressed as a fraction of total weight to provide a measure of food intake during the exercise period. After these measurements were completed, all mice were killed, tail clips were taken for DNA analysis, and the carcasses were stored at −30°C. All rearing and measurement procedures were approved by the Institutional Animal Care and Use Committee at the University of North Carolina at Chapel Hill.

Altogether, therefore, 14 traits were measured that were subjected to analysis. They included 10 weight and body composition traits (pre- and postexercise body weights, fat and lean mass percentages, the percent change in body weight, % fat, and % lean tissue resulting from exercise, and food intake during the exercise period) and four exercise traits (distance, time, average speed, and maximum speed). All 14 body composition and exercise traits were available in all 473 mice.

Genotypic data.

DNA was prepared by standard methods from tail biopsies, and genotyping was outsourced for analysis using the Mouse Universal Genotyping Array (MUGA) (15), a 7,851 SNP array built on the Illumina Infinium platform. Markers on the MUGA are distributed genome-wide with an average spacing of 325 kb and standard deviation of 191 kb. Based on genotyping of representative individuals from the F₀ parental strain (n = 13, HR; n = 2, B6), we selected a total of 2,058 fully informative SNPs for use in the QTL mapping for all traits. Taking into account the high levels of recombination expected in the G₁₀ population, we checked all fully informative SNPs for significant segregation distortion, and genotyping errors were estimated with the error detection function in Merlin (1). Individual calls that were deemed extremely unlikely were dropped from the analysis. A list of these SNP markers with their locations (in Mb) is given in Supplemental Table S1.¹

The frequencies of the HR/HR and B6/B6 genotypes at each of the SNPs on each chromosome are illustrated in Supplemental Fig. A1, and their overall averages across all SNPS are listed in Supplemental Table A1. Supplemental Fig. A1 shows that the heterozygote (HR/B6) frequencies across most chromosomes consistently track around the expected level of 50%, suggesting that the breeding scheme was successful in minimizing inbreeding. The drift in homozygote frequencies is much more pronounced (as might be expected with the smaller sample sizes involved), with averages for the individual chromosomes as low as 11% and as high as 41% (Supplemental Table A1).

Preliminary statistical analysis.

Before conducting the QTL analyses, we first ran univariate and multivariate analyses of covariance with the MIXED procedure in SAS to test for potential effects of several variables on each of the 20 traits. For all traits, this model included sex, diet, and parity as fixed classification variables, litter size and age as covariates, and cohort as a random classification variable. All variables except parity showed multivariate significance. For the four exercise traits, we also included wheel freeness in the model as a covariate to adjust for the number of wheel revolutions following acceleration to a given velocity (33).

We calculated residuals from the analyses described above and used them to examine the distributions of each trait. Eight of the 14 traits exhibited normality (P > 0.05 in Kruskal-Wallis tests) in their distributions, whereas the other six mostly showed mild skewness. No attempt was made to transform these six variables, however, because we wanted to compare our results with those from the G₄ mice (31, 33) where transformations were not done. In addition, the permutation procedure used to establish threshold likelihood of odds (LOD) scores for the QTLs (see below) does not require that the distributions of these traits be normal. We also used the residuals to calculate correlations for each of the three sets of traits. Probabilities generated from these correlations were evaluated with the false-discovery-rate procedure (5).

QTL analysis.

We searched for QTLs throughout the mouse genome for each of the 14 traits using the QTLRel package implemented in R (9, 10). This program adjusted for the relatedness among individuals generated in our advanced intercross (G₁₀) mice by the calculation of condensed identity coefficients (38) from the pedigree data provided. We used the Haley-Knott interval mapping (26) option in this package that resulted in the imputation of index values between any adjacent markers separated by >1 cM. Altogether, an additional 965 genotypic index values were included with the 2,058 SNPs, for a total of 3,023 markers that were used in the scans. The model used for each of the traits included additive and dominance genetic effects as well as the fixed and random variables as outlined above in the calculation of residuals. Running this model in QTLRel produced likelihood ratio values at each of the markers on all chromosomes that were converted into LOD scores.

To establish threshold levels of significance for the LOD scores generated for each trait, we used the traditional permutation method of Churchill and Doerge (14). In this method, the phenotypic data were shuffled 1,000 times, and the analysis was run on each of these samples and the highest LOD score recorded. The 95th percentile value among these 1,000 LOD scores then was used as the 5% experiment-wise significance threshold. Similarly, the 95th percentile values for the highest LOD score on each of the 20 chromosomes were used as the 5% chromosome-wise thresholds.

Putative QTLs were considered to be present at the sites of all LOD scores on each chromosome that reached the chromosome-wise threshold levels of significance. Where two or more peaks occurred on the same chromosome with LOD scores exceeding the threshold value for that chromosome, these were taken to represent multiple QTLs if the peaks were separated by a drop of at least 1.5 LOD units. Confidence intervals for each QTL were defined by 1.5 LOD drops on each side of the peak position, this being more appropriate than a 1.0 LOD drop when markers are ∼1 cM apart (40).

Additive (a) and dominance genotypic values (d) at each QTL site were estimated in QTLRel by partial regressions and tested for significance (P < 0.05) via individual t-tests. The additive genotypic value is defined as one-half the difference between the values for the two homozygotes whereas the dominance genotypic value is defined as the difference between the midhomozygous and the heterozygous values (19). The a values may be positive or negative in sign, with positive values indicating that the HR alleles increase the mean of the trait, whereas negative values indicate that the HR allele decreases the mean of the trait. A d value of 0 (or one that is not significantly different from 0) indicates that the phenotypic value of the heterozygote is intermediate between the two homozygotes and thus that there is no dominance, whereas a d value equal to +a (or −a) indicates that the phenotypic value of the heterozygote is equal to that of one of the homozygotes and that there is complete dominance. When the value of the heterozygote is outside of the range of both homozygotes (d > +a or d < −a), this indicates overdominance. QTLRel also estimated the percentage of the total phenotypic variation for the traits explained by each QTL.

Once all QTLs were discovered for each of the 14 traits, we tested them for sex-specific effects. This was accomplished by subtracting the likelihood values generated in models run with and without sex by QTL interactions and evaluating these differences for significance at the conventional 5% level. We used the same procedure to test for diet-specific QTL effects as well. Significant QTL interactions were interpreted as meaning that the genotypes affected the trait means differentially depending on the particular sex or diet. We also tested the seven weight/body composition postexercise traits for QTL-by-exercise interactions and QTL-by-exercise-by-diet interactions. For this purpose, we used each of the four exercise traits: distance, time, average speed, and maximum speed. Significant interactions using these continuous exercise variables were taken to indicate differential genotypic effects on the associations of the postexercise traits with the exercise traits within (QTL × diet × exercise) or across (QTL × exercise) the dietary environments. Where these occurred, we calculated partial regressions (adjusting for sex, diet, etc.) of the postexercise traits on the exercise traits for each of the genotypes to discover the patterns of differences.

QTL replication.

Once the analysis of the G₁₀ mice was completed, it was of interest to compare our QTL results with those previously achieved in the G₄ population (31, 33). However, the G₁₀ population was derived from only one of the reciprocal mating types (B6 females × HR males), whereas both mating types were used to map QTLs in the G₄ generation (31, 33). Furthermore, Kelly et al. (32) found significant parent-of-origin effects for many of the body composition and exercise traits. To appropriately assess QTL replication, therefore, we conducted QTL analyses as before (31, 33) for each of the 14 traits in the G₄ population, but only for the mice produced from crosses of B6 females × HR males. We then tested for replication at each of the sites of these newly identified G₄ QTLs. To accomplish this, we followed the same procedure as already described above for the G₁₀ analyses but tested for QTLs only for the SNPs at the sites closest to the G₄ QTLs. Those sites with a LOD score of 1.30 (P < 0.05) or higher were taken to indicate QTL replication (18, 61).

RESULTS

Basic statistics.

Basic statistics for all (unadjusted) 14 traits in male and female G₁₀ mice fed the control and the high-fat diet are given in Table 2. All values are shown for the separate sexes because the majority of traits, especially weight and percent fat, showed significant sex differences in preliminary ANOVAs. For example, pre- and postexercise weight is greater in males than in females, but the reverse is true for pre- and postexercise fat percentages. All four exercise traits also showed significant sexual dimorphism, with female values being greater than male values in all instances.

Table 2.

Means and SD for the body composition, exercise, and skeletal traits in male and female mice fed the control diet and the high-fat diet

	Males-Control (n = 116)		Males-High Fat (n = 118)		Females-Control (n = 118)		Females-High Fat (n = 121)
	Mean	SD	Mean	SD	Mean	SD	Mean	SD
Wt pre-exercise, g	28.62	2.90	31.35	3.20	20.85	2.45	22.70	3.36
% Fat pre-exercise	12.74	3.76	15.83	5.06	14.58	4.27	19.00	6.02
% Lean pre-exercise	80.18	4.04	77.25	4.92	79.96	4.64	76.62	5.66
Wt postexercise, g	27.18	2.34	29.50	2.50	20.24	1.83	21.45	2.10
% Fat postexercise	7.86	2.15	10.64	3.62	9.77	2.58	12.37	3.83
% Lean postexercise	83.28	2.92	80.50	3.98	83.23	3.01	81.05	4.08
% Change in Wt	−4.76	4.70	−5.59	5.18	−2.48	6.06	−4.74	7.06
% Change in % fat	−34.82	20.85	−31.00	17.09	−30.63	17.57	−32.66	16.68
% Change in % lean	4.04	4.74	4.40	4.81	4.36	5.99	6.15	6.80
Food intake/Wt	0.79	0.17	0.68	0.17	0.95	0.27	0.83	0.24
Distance, revolutions	7,801.46	3,746.98	7,931.62	3,681.99	10,886.67	3,621.89	11,430.68	3,678.26
Time, cumulative intervals run	438.17	122.51	442.18	129.21	551.81	104.47	581.14	116.75
Average speed, revolution/time	17.17	4.49	17.28	4.08	19.42	4.41	19.41	4.10
Maximum speed	31.15	6.18	31.95	5.23	34.58	6.22	34.36	5.71

Wt, weight.

Among all mice, lean tissue averaged ∼80%, with fat tissue consistently <20%. Across both sexes and dietary environments, it can be seen that the mean weight and percent fat values decreased with exercise (compare pre- and postexercise means), although the magnitude of this decrease is much greater for fat (−31 to −35%) than for weight (−2.5 to −5.6%). The percentage of lean tissue, on the other hand, consistently increased with exercise, its magnitude varying from 4.0 to 6.2%. Student t-tests using the residuals calculated from adjusting for the various fixed and random variables showed that these percent changes with exercise in fact were statistically different from 0 (P < 0.0001) for all three traits. Significant diet effects in the ANOVAs were found for nine of the 10 weight and body composition traits, but not for any of the four exercise traits. Pre- and postexercise weight and percent fat show consistently higher means for mice in the high-fat compared with the control diet, whereas the reverse is true for the percent of lean tissue (Table 2). Food intake during the exercise period was less in mice fed the high-fat diet.

Pairwise correlations among the 10 weight and body composition traits for the G₁₀ mice are given in Table 3. These values range from −0.93 to +0.84, and 41 (of the 45 total) reach statistical significance. In general, the pre- and postexercise weight traits are positively correlated with the percent of fat tissue but negatively correlated with the percent of lean tissue. However, pre-exercise weight is negatively correlated with the percent change in the percent fat and positively correlated with the percent change in lean tissue. Postexercise weight shows no association with the percent change in either fat or lean tissue. Not surprisingly, the percentages of fat and lean tissue are negatively correlated. Also, food intake (per body weight) shows negative correlations with pre- and postexercise weight and percent fat but positive correlations with the pre- and postexercise percent of lean tissue.

Table 3.

Pairwise correlations of body weight and composition traits in the G₁₀ mice before and after 6 days of voluntary exercise

	Pre-exercise		Postexercise
	% Fat	% Lean	% Wt	% Fat	% Lean	% Change in % Wt	% Change in % fat	% Change in % lean	Food Intake/Wt
Wt pre-exercise	0.55†	−0.59†	0.84†	0.36†	−0.36†	−0.60†	−0.26†	0.44†	0.49†
% Fat pre-exercise		−0.93†	0.27†	0.68†	−0.60†	−0.61†	−0.42†	0.66†	−0.61†
% Lean pre-exercise			−0.28†	−0.64†	0.59†	0.67†	0.40†	−0.75†	−0.60†
Wt postexercise				0.36†	−0.37†	−0.10*	0.07	0.05	−0.25†
% Fat postexercise					−0.93†	−0.14†	0.32†	0.04	−0.31†
% Lean postexercise						0.12†	−0.32†	0.08	0.29†
% Change in Wt							0.59†	−0.73†	0.55†
% Change in % fat								−0.73†	0.40†
% Change in % lean									−0.51†

All values are adjusted for sex, diet, cohort, litter size, and age differences.

^*P < 0.05;

^†P < 0.01.

Table 4 shows correlations among the four exercise traits and between the exercise and postexercise weight/composition traits. All six correlations among the four exercise traits are moderate to high in magnitude and statistically significant. All postexercise traits except the percent change in weight show significant correlations with one or more of the exercise traits. However, these associations are generally quite low, the maximum correlation (postexercise percent lean tissue with time) being +0.31. In general, therefore, the postexercise traits have minimal associations with the exercise traits.

Table 4.

Correlations within the four exercise traits and between the exercise and postexercise weight and composition traits for mice in the G₁₀ population

	Distance	Time	Average Speed	Maximum Speed
Distance		0.83†	0.84†	0.67†
Time			0.42†	0.30†
Average speed				0.84†
Wt postexercise	0.02	−0.11	0.14†	0.17†
% Fat postexercise	−0.15†	−0.27†	0.00	0.16†
% Lean postexercise	0.16†	0.31†	−0.04	−0.18†
% Change in Wt	0.00	−0.10	0.07	0.08
% Change in % fat	−0.05	−0.19†	0.08	0.15†
% Change in % lean	0.02	0.15†	−0.08	−0.09
Food intake/Wt	0.18†	0.08	−0.02	0.13†

All traits are adjusted for potential effects of several covariates.

^*P < 0.05;

^†P < 0.01.

Weight and body composition QTLs.

Table 5 gives the basic statistics for all QTLs discovered affecting the 10 weight and body composition traits in the G₁₀ mice. Altogether, 71 separate QTLs were identified, with 15 reaching the experiment-wise level of significance and the remainder significant at the chromosome-wise threshold levels. The number of QTLs varies among the traits, being highest for pre- and postexercise percentages of fat (11 and 10), and lowest for the percent change with exercise in fat tissue (2). QTLs are found on 16 of the 20 chromosomes (all except chromosomes 4, 12, 15, and X), with chromosomes 3 (10 occurrences) 7 (8 occurrences), and 8 (9 occurrences) being most highly represented. QTLs on chromosome 3 in particular affect seven of the 10 traits and show three separate peaks at 74.6 Mb, 118.8–121.2 Mb, and 140–143.3 Mb. Confidence intervals for the 71 QTLs average 11.8 Mb with a standard deviation of 6.61 Mb, although those for the 15 QTLs reaching experiment-wise significance are somewhat lower and considerably less variable (mean = 10.4 Mb, standard deviation = 3.83 Mb).

Table 5.

QTL statistics for the weight and body composition and exercise traits in the G₁₀ mice

Trait	Ch	Position, Mb	Confidence Interval, Mb	LOD	a	d	%	Interactions
Pre-exercise (8 wk of age)
Wt	3	140.0	131.7–148.1	3.48	−0.845‡	−0.389	1.13
	5	144.9	142.1–147.6	3.52	0.337	0.905‡	1.01
	7	19.9	17.0–25.7	3.52	−0.613‡	−0.809‡	1.41
	8	89.8	80.1–91.5	5.22†	1.409‡	1.304‡	2.36
	9	112.3	111.5–116.3	3.24	−0.567‡	−0.806‡	0.96
	13	59.0	40.1–74.4	3.07	−0.920‡	−0.971‡	1.30
	18	69.6	65.4–81.7	3.37	1.096‡	−0.087	1.46	D
	19	59.5	56.6–60.7	2.90	0.410	−0.790*	1.17
% Fat	3	74.6	68.1–81.3	3.65	−1.617‡	−0.956	3.16
	3	143.3	137.0–144.3	5.71†	−1.916‡	−0.427	6.20	D
	5	116.6	110.6–125.8	2.90	−1.300‡	−0.868	2.57
	7	31.2	17.3–47.8	3.68	−1.578‡	−0.351	4.01
	8	81.5	77.5–90.6	9.62†	3.246‡	1.706‡	11.40	D
	10	92.3	87.1–113.8	2.64	−1.309‡	1.424‡	1.86
	11	34.1	30.9–36.3	3.78	−1.268‡	0.736	3.78
	11	107.3	92.8–108.0	3.40	1.369‡	0.431	3.03
	13	59.0	56.9–70.7	4.37†	−1.942‡	−1.522‡	4.71	D
	18	52.9	49.3–57.4	3.36	0.435	−1.823‡	4.85
	18	69.1	65.4–70.2	5.09†	2.214‡	−0.636	4.85
% Lean	3	143.3	131.7–145.4	4.19	1.612‡	0.551	5.09	D
	7	25.2	18.5–47.7	3.07	1.324‡	0.861	4.20
	7	94.8	91.3–104.2	3.14	1.122‡	−0.642	2.61
	8	87.4	80.2–90.6	9.29†	−3.109‡	−1.809‡	11.45
	11	105.1	100.9–108.0	4.79†	−1.684‡	−0.328	4.82
	13	59.0	56.9–70.7	5.34†	2.125‡	1.751‡	6.36
	13	79.8	74.9–84.5	3.42	1.489‡	0.813	3.35
	18	69.6	65.4–72.1	4.20	−2.188‡	0.611	4.81
Postexercise
Wt	3	121.2	117.6–128.3	2.97	0.100	0.690‡	0.61	E3,E4
	5	137.6	127.8–142.1	2.97	0.880‡	−0.426	1.42
	7	128.0	124.8–136.4	3.19	0.760‡	0.317	1.02	DE1,DE3,DE4
	8	95.5	88.6–97.3	3.36	0.740‡	0.175	0.99	DE3
	16	74.9	73.5–83.4	3.18	−0.323	1.055‡	0.83	E2
	16	98.0	93.5–98.0	2.96	0.502‡	−0.458*	0.68
% Fat	1	146.8	142.1–157.9	3.88	0.910‡	0.193	3.38
	2	69.9	63.7–70.4	4.38	−0.145	1.250‡	3.43	DE1,DE3,DE4
	3	118.8	117.6–120.5	2.97	0.304	0.997‡	2.04
	3	143.3	140.3–145.4	4.52†	−1.110‡	−0.070	4.70	D
	5	68.6	65.1–73.4	4.91†	−1.083‡	−0.944	6.66
	8	89.8	77.5–92.2	6.07†	1.477‡	0.359	6.17	E4,DE3
	13	57.2	46.7–63.8	3.01	−1.029‡	−0.858‡	2.51	D,E1,E3,E4
	14	119.2	105.9–121.8	2.97	−0.695‡	−0.526	2.43
	16	42.6	36.2–44.4	2.97	0.992‡	−0.417	3.10
	18	69.1	65.4–70.2	4.95†	1.225‡	0.065	4.44	D,E1,E3,E4
% Lean	1	146.8	137.7–160.3	3.10	−0.901‡	0.005	2.95
	2	69.9	64.6–73.4	3.88	−0.038	−1.416‡	3.18	DE3,DE4
	3	118.8	117.6–120.5	3.07	−0.209	−1.099‡	2.15
	3	143.3	138.8–149.0	3.83	1.114‡	0.028	4.32	D
	5	68.6	65.1–76.1	3.53	0.988‡	0.898*	5.22	D
	7	94.8	91.6–96.9	3.67	0.846‡	−0.444	2.79	DE4
	8	89.8	77.5–92.2	4.88†	−1.427‡	−0.325	5.33	E3
	18	69.1	65.4–70.2	3.85	−1.135‡	−0.134	3.71	D,E1,E3
% Change in Wt	3	140.0	131.7–148.8	2.99	1.578‡	0.858	3.07	E3,E4
	8	89.8	79.1–91.5	3.16	−1.944‡	−2.460‡	4.21
	9	112.3	111.2–113.1	3.15	0.785	1.889‡	2.87	E1,E2,E3,E4
	10	91.5	86.1–111.1	3.87	2.082‡	−1.695‡	3.69
	13	60.4	56.9–70.7	3.03	1.657‡	1.702‡	3.81
	16	5.6	4.0–6.7	3.41	−0.997*	−2.151‡	3.22
	18	69.6	65.4–79.6	4.20	−2.588‡	0.874	5.45
% Change in % fat	14	66.0	61.8–69.4	3.09	−1.681	−6.226‡	2.95
	17	31.9	30.5–37.2	3.94	4.691‡	5.914‡	5.63
% Change in % lean	5	31.9	30.5–37.2	3.94	4.691‡	5.914‡	2.98
	7	118.1	113.7–125.8	3.00	−1.701‡	−0.341	4.16
	8	87.3	77.5–91.5	3.73	2.253‡	1.835‡	4.81	E4
	11	106.1	92.5–108.0	3.77	1.754‡	0.173	4.03
	13	60.4	56.9–74.9	3.60	−1.593‡	−1.886*	4.15
Food intake/Wt	2	38.2	35.1–44.2	2.94	−0.026	0.064‡	3.02
	5	65.1	53.7–74.1	3.41	0.037*	0.080‡	4.42	DE1,E2
	6	144.5	142.8–147.2	2.94	0.033	0.092‡	3.43	S,DE2
	7	71.7	64.8–78.3	4.96†	0.065‡	−0.050*	5.29	S
	8	81.4	77.5–91.4	5.43†	−0.113‡	−0.071*	7.09
	11	24.4	17.4–36.3	2.94	0.074‡	0.059*	3.40	E2
Exercise Traits
Distance days 5–6	16	87.7	85.2–90.4	2.84	−258.497	1308.660‡	2.42
	19	19.7	17.4–22.0	3.09	150.920	1273.844‡	2.71
Time days 5–6	19	18.9	17.4–22.0	4.47†	17.422	42.256‡	3.99
Average speed days 5–6	16	87.7	86.1–90.4	3.09	−0.290	1.592‡	2.99
	19	45.0	35.2–48.9	3.20	−1.231‡	0.281	3.20
Maximum speed days 5–6	2	152.5	149.5–159.4	3.56	1.379*	−2.702‡	3.18
	11	11.0	5.9–19.1	3.04	1.512‡	−0.333	3.37
	18	62.9	58.6–63.8	3.02	−2.099‡	0.239	4.16
	19	45.0	42.1–48.7	3.41	−1.845‡	0.651	3.73
	19	53.8	48.9–55.1	3.41	−0.897*	1.883‡	4.30

Shown are all quantitative trait loci (QTLs) affecting the weight and body composition traits in the G₁₀ mice that had likelihood of odds (LOD) scores reaching the chromosome-wise or experiment-wise (†) level of significance. Locations and confidence intervals of the QTLs are given in Mb. Also shown is the percentage contribution (%) of each QTL to the total variance of each trait, and its additive (a), dominance (d) genotypic effects (*P < 0.05; ‡P < 0.01). Interactions = interactions of QTLs with sex (S), diet (D), distance run (E1) time spent running (E2), average running speed (E3) maximum running speed (E4), or diet and exercise (DE1, DE2, DE3, DE4).

Many of the QTL positions for different traits are common, and in fact only 31 of the 71 total QTLs have nonoverlapping confidence intervals. This suggests that there are at most 31 (and perhaps fewer) unique QTLs affecting the weight and body composition traits. An example of this is the QTL on chromosome 8 that affects nine of the 10 traits. This QTL is located at 81.5–95.5 Mb and may well represent a single gene (or several linked genes) with pleiotropic effects on these traits. Other examples of apparent pleiotropy are seen by the QTLs on chromosome 13 at 57.2–60.4 Mb affecting seven traits, and a QTL on chromosome 18 at 69.1–69.6 Mb affecting six traits.

The majority of the weight/body composition QTLs exhibit significant additive genotypic effects, with less than one-half showing significance dominance effects (P < 0.01 in a χ² test). In addition, the overall mean of the (absolute) additive genotypic values is 1.24 compared with 1.01 for the (absolute) dominance genotypic values. Food intake is an exception, however, with all six QTLs affecting this trait exhibiting significance dominance effects. In addition, there are several instances, especially for pre-exercise weight, of overdominance (where d exceeds the positive a value or is less than the negative a value). The numbers of positive and negative a values are nearly the same, suggesting that the HR allele at these loci tends to act about equally in increasing or decreasing these traits.

The percent of the total phenotypic variation in the weight and body composition traits contributed by the QTLs ranges from <1% (0.61%) to 11.45%, averaging 3.66%. The variation explained by the QTLs affecting both pre- and postexercise body weight tends to be the lowest (mean = 1.17%), whereas that for pre- and postexercise percent of fat is the highest (mean = 4.25%). The contribution of >11% of the variation in the percentage of fat and of lean tissue by the QTL on chromosome 8 is particularly impressive.

A total of eight of the weight/body composition QTLs show significant interactions with the dietary environment, whereas only two QTLs, both affecting food intake, show significant interactions with sex (Table 5). Four of these dietary interactions involve a single QTL on chromosome 3 (at 143.3 Mb) that exhibits pleiotropic effects on pre- and postexercise percentages of fat and lean tissue. The effects of this QTL in both dietary environments are illustrated in Fig. 1. Note that the B6 allele increases both the pre- and postexercise percentages of fat tissue while decreasing the pre- and postexercise percentages of lean tissue, but primarily in mice fed the high-fat diet, not those fed the control diet.

Fig. 1.

Genotypic effects of a quantitative trait locus (QTL) on chromosome 3 on the pre- and postexercise percentage of fat and lean tissue in the G₁₀ mice for both the low-fat and high-fat dietary environment.

A total of 18 QTLs for some of the postexercise weight/composition traits also show interactions with one or often several of the exercise traits and/or with exercise and diet combined (Table 5). Where this occurred, we calculated partial regressions of the weight/composition traits on the exercise traits for each of the three genotypes at each QTL affected. Figure 2 illustrates the interaction of exercise with four QTLs. In Fig. 2A, all three genotypes for a QTL on chromosome 13 generally produce a decrease in the percentage of postexercise fat with increases in exercise (distance run), but the effect is more pronounced for the B6 allele at this locus. A similar trend is seen for a QTL on chromosome 9 that affects the percent change in body weight after exercising (time spent running). In Fig. 2C, both homozygotes at a QTL on chromosome 16 produce increases in postexercise body weight with increasing exercise (distance) run, but heterozygotes produce a negative response in body weight. And in Fig. 2D, HR/HR homozygotes at a chromosome 8 QTL decrease the percent of change in the percentage of lean tissue with increases in maximum speed run, while the HR/B6 and B6/B6 genotypes produce an increase or no effect on lean tissue with exercise.

Fig. 2.

Genotypic effects of QTLs on postexercise percent of fat (A), percent change in the weight (B), postexercise weight (C), and percent change in the percent of lean tissue (D) in the G₁₀ mice that vary depending on the level of exercise. For each of the 3 genotypes, the regressions for each of the postexercise traits on a specific exercise trait are shown.

Figure 3 illustrates how the effect of two different QTLs on postexercise traits varies depending on both exercise and diet. In Fig. 3A, it can be seen that as mice fed the control diet increase their level of exercise (distance run), the HR/HR and HR/B6 genotypes for a QTL on chromosome 2 cause no change in postexercise fat, whereas the B6/B6 genotype tends to decrease the amount of postexercise fat. But HR/HR and HR/B6 mice on the high-fat diet tend to decrease fat with exercise, whereas B6/B6 mice actually increase the amount of fat with exercise. In Fig. 3B, it may be seen that all mice with any of the three genotypes at a chromosome 8 QTL tend to increase weight with exercise if fed the control diet but show little or no increase in weight if fed the high-fat diet.

Fig. 3.

Genotypic effects of QTLs on postexercise percent of fat (A) and postexercise weight (B) in the G₁₀ mice that vary depending on both diet and exercise. For each of the 3 genotypes, the regressions of these postexercise traits on a specific exercise trait in each dietary environment are shown.

Exercise QTLs.

A total of 10 QTLs were found to affect the exercise traits, none of which interacted with sex or diet. Only one QTL was found for time (cumulative 1 min intervals run), although it also was the only QTL to reach the experiment-wise threshold level of significance. Five chromosomes are involved, with chromosome 19 harboring QTLs that affect all four traits. A QTL on chromosome 16 also exhibits pleiotropic effects on distance and average speed. Based on the confidence intervals for these exercise QTLs, all are in separate locations from the QTLs found for the weight and body composition traits. The QTLs for distance and time show large dominance effects, whereas those for average and maximum speed exhibit both additive and dominance effects. The percentage of the phenotypic variation in the exercise traits explained by these QTLs ranges from 2.4 to 4.3%, averaging 3.4%.

G₄ vs. G₁₀ QTLs.

The results of the QTL reanalysis of the G₄ mice produced only from crosses of B6 females × HR males are shown in Table 6. QTLs are listed for all LOD scores reaching the 0.05 or 0.1 experiment-wise level of significance (see Ref. 33). Altogether, 30 QTLs were found, one more than the 29 QTLs mapped for these traits in the full population generated from both mating types (31, 33). Tests at the sites of these 30 G₄ QTLs in the G₁₀ population showed that 12 produced significant LOD scores (boldfaced in Table 6), suggesting a QTL replication of 40%. We can also directly compare locations of the G₄ (Table 6) and G₁₀ (Table 5) QTLs, and this shows that only 11 of the 81 total G₁₀ QTLs (14%) have confidence intervals that overlap with those of the G₄ QTLs. In other words, the majority of the G₁₀ QTLs affecting the weight, composition, and activity traits we discovered were not previously found in the analysis of the G₄ mice (31, 33).

Table 6.

QTL statistics for weight, composition, and exercise traits mapped in the G₄ generation mice produced only from the B6 female × HR male crosses

Trait	Ch	Position, Mb	Confidence Interval, Mb	LOD	a	d	%
Distance	none
Time	7	108.9	93–133	4.9	36.8*	−0.9	4.9
Average speed	none
Maximum speed	2	91.8	83–103	6.1	−1.5	0.6	5.5
	11	56.5	57–66	4.5	1.3	0.8	4.8
Pre-exercise WT	1	122.5	93–134	9.7	1.1	−0.1	4.0
	2	76.4	69–83	6.6	1.3	1.2	3.0
	4	149.3	147–	7.1	−1.9*	0.9	6.9
	5	10.6	−15	10.4	−0.8	−0.7	2.2
	6	36.3	30–45	17.1	1.3	−0.9	5.7
	7	82.6	73–86	7.3	0.4	1.9	4.8
Pre-exercise % fat	1	82.6	73–99	6.9	1.4*	0.9	6.4
	8	86.0	78–93	11.9	2.1	−0.1	10.7
Pre-exercise % lean	8	89.3	78–93	9.8	−2.0*	0.2	10.0
Postexercise WT	1	36.8	34–40	14.1	1.2	−0.7	5.3
	2	172.5	159–	7.5	1.0	−0.1	7.0
	4	149.3	142–	7.5	−1.5*	0.8	6.2
	5	10.6	−15	12.2	−0.6	−0.6	2.1
	6	36.3	30–45	14.6	0.9	−0.6	3.7
Postexercise % fat	1	171.8	154–179	8.8	1.0*	0.3	8.3
	5	90.1	68–93	6.8	−1.0*	0.4	5.3
	8	89.3	78–99	7.5	1.0*	−0.4	6.7
Postexercise % lean	1	171.8	154–179	7.0	−1.4*	0.3	7.6
	5	90.1	53–93	7.9	1.3*	−0.3	8.2
	8	89.3	86–99	7.7	−1.3*	0.3	7.9
% Change in WT	6	67.4	54–76	4.6	−1.8*	−1.0	5.3
	17	83.3	79–86	4.5	1.0	1.9	3.4
% Change in % fat	5	76.0	53–97	6.7	−6.8*	1.9	6.9
% Change in % lean	5	99.5	77–113	5.4	2.0*	1.3	6.0
	17	83.3	79–86	5.3	−1.3	−3.0*	5.6
Food intake/WT	6	67.4	54–76	6.1	−0.1	−0.1	6.0
	7	82.6	73–86	5.5	0.02	−0.1	5.0

Boldfaced LOD scores indicate QTLs that have replicated in the G₁₀ population. HR, high voluntary wheel runnning.

^*P < 0.05.

The mapping resolution in the G₁₀ population was enhanced because of a nearly eightfold expansion of the genome (from 1,606 cM F₂ equivalents to 12,685 cM) compared with the threefold expansion achieved in the G₄ generation (33). This expansion resulted in QTLs with reduced confidence intervals, those for the weight, composition, and exercise traits in the G₁₀ population averaging 11.3 Mb (Table 5) compared with that of 23 Mb previously calculated by Kelly et al. (31, 33) for these same traits in the G₄ population. And the confidence interval average of 11.3 Mb actually would have dropped further to about 8 Mb if we had used a 1-LOD (rather than 1.5-LOD) drop criterion as did Kelly et al. (31). For the nine replicated QTLs (Table 6), the confidence intervals were reduced from an average of 19 Mb in the G₄ to 13 Mb (or ∼8 Mb using a 1-LOD drop) in the G₁₀ mice.

DISCUSSION

The basic goals of this study were to potentially fine map the QTLs previously found to affect the 14 body weight, composition, and exercise traits in the G₄ mice, to identify any new QTLs for these traits and to test the QTLs for interactions with diet and/or exercise. We were successful in uncovering a number of QTLs for these traits, some of which replicated G₄ results, but many of which were new. In all cases it was clear that the additional generations of outbreeding in this population since the G₄ were beneficial in allowing us to map the QTL positions with greater precision. We also found that a number of these QTLs significantly interacted with diet, exercise, or both. These kinds of interactions may well be the key to understanding the perplexing amount of variation in weight or weight loss among individuals on basically similar diets and/or exercise regimes.

QTL replication.

We found a 40% replication rate for the G₄ QTLs for the weight/composition and exercise traits in our G₁₀ mouse population (Table 6). By comparison, Vaughn et al. (61) found 70% replication of QTLs for a number of growth traits between two separate F₂ intercross mouse populations. However, only 96 markers were used in that study (61), suggesting broader and inflated overlap between confidence intervals for the QTLs in the replicates. In the F₉-F₁₀ generations produced from an intercross of the LG/J × SM/J strains, Norgard et al. (43) replicated 36 of the original 70 QTLs (51%) affecting long bone lengths found in the F₂-F₃ generations. However, these investigators used a much larger number of mice (1,455) than were available in our study, and this should have increased the chance of replication. In addition, neither study cited above used mice that were subjected to different diets or a voluntary exercise period, so in general it is difficult to attach much significance to the fact that their QTL replication rates are greater than those for our G₄ mice.

There are several possible reasons why many of the G₄ QTLs may not have replicated in the G₁₀ population. One common explanation is that QTLs with marginally significant effects in early generations randomly fail to reach significance in later generations (4). This is certainly possible, although if so, we might have expected to see an association between replication and the magnitude of the LOD scores that was not apparent. A few of the QTLs in the G₄ generation also may be false positives, although those with high LOD scores are unlikely candidates. Another possibility is that some of the G₄ QTLs may represent linked genes that have separately mapped in the G₁₀ population but with too small an effect to reach significance. It also is possible that some of the replication failure may have resulted from the different methods that were used to calculate LOD thresholds in the G₄ (GRAIP procedure; see Ref. 33) compared with the G₁₀ population. But some preliminary QTL analyses done on the G₁₀ mice using the methods previously used for the G₄ population generally showed quite different patterns of LOD peaks for mice in the separate generations. A perhaps more likely possibility for some of the QTL replication failure is that one-half of the G₁₀ mice were fed a high-fat diet, whereas all G₄ mice were fed a standard diet. Although we could have tested this using the G₁₀ mice fed only the control diet, the sample size would have been reduced to 234 mice as would the power to detect QTLs, and thus this test would be of limited usefulness.

Genotypic frequency variation.

It is possible that reduced sample sizes for some genotypes produced from drift in genotypic frequencies in the G₁₀ mice (see Supplemental Fig. A1) may have resulted in sampling error that decreased the chance of detecting some G₄ QTLs. To test this possibility, we calculated the frequencies of the three genotypes at the closest markers for all 12 QTLs that replicated, as well as the 18 QTLs that did not replicate (Table 6). The (pooled) frequencies of the HR/HR, HR/B6, and B6/B6 genotypes, respectively, were 0.349, 0.465, and 0.186 for the replicate QTLs and 0.263, 0.485, and 0.252 for the nonreplicates. This suggests that the replicate QTLs tended to be associated with higher frequencies of the HR/HR genotype (and lower B6/B6 frequencies), and in fact a conventional contingency table χ² analysis showed that the genotype frequencies significantly differed between the two groups of QTLs (P < 0.0001). Variation in genotypic frequencies may also have affected the locations of the QTLs we discovered, and we tested this by comparing the genotype frequencies at the QTL sites vs. non-QTL sites. Over all chromosomes affected, the HR/HR, HR/B6, and B6/B6 frequencies were 0.274, 0.480, and 0.246 at the QTL sites and 0.278, 0.481, and 0.241 at the non-QTL sites. These frequencies clearly are very similar, and a nonsignificant (P = 0.118) χ² value from a contingency table analysis confirmed this. On the other hand, there is clearly heterogeneity in the genotype frequencies among the chromosomes, and χ² tests performed as above for each of the 16 chromosomes with QTLs showed that eight were statistically significant whereas eight did not reach significance. So it is possible that sampling variation brought about by variation especially in the frequencies of the homozygote genotypes may have impacted our QTL mapping to some extent.

QTLs for body weight and composition.

We detected eight QTLs affecting pre-exercise body weight in the G₁₀ mice, adding to the many QTLs previously mapped for this trait at various ages in different mouse populations (11, 16, 36, 50). This number is considerably less than the 14 QTLs found by Cheverud et al. (11) for 8 wk body weight and the 17 or 18 found by Rocha et al. (50) for 6 and 10 wk body weight, but both of these studies used many more mice than the 473 we had available. In addition, the B6 and HR lines were chosen on the basis of their difference in voluntary exercise, not body weight. Nehrenberg et al. (42) found that 8 wk body weight in the HR strain averaged 27.0 g compared with that of 20.5 g in the B6 strain. This difference of just 6.5 g is much less than that of 20.4 g (63-day weight) between the LG/J and SM/J strains used by Cheverud et al. (11) and that of 27 g (6 wk weight) between the M16i and L6 strains used by Rocha et al. (50). As might be expected, the QTLs discovered by these investigators also contributed greater amounts to the total variance.

It is interesting to compare the pre-exercise vs. postexercise body weight QTLs (Table 5) to discover to what extent the genetic regulation of this trait might differ before and after the short, but intensive, exercise period. Genetic correlations of body weight at contiguous weekly ages generally have been quite high, suggesting a mostly common, pleiotropic genetic basis (35). And in fact Cheverud et al. (11) found that QTLs for 8 wk and 9 wk body weight in mice colocalized in 13 of 17 instances. In our population of mice, however, only one QTL on chromosome 8 shared the same confidence intervals at both ages (8 and 9 wk), implying a largely different genetic basis for body weight at these two ages that most likely is due to the effects of wheel running. It also is suggestive that the only weight QTL common at both ages also was the only one that reached significance at the experiment-wise level (Table 5). In their analysis of the G₄ mice, Kelly et al. (31) reported only QTLs with LOD scores reaching the (0.05 or 0.1) experiment-wise significance level, and three of the four QTLs they discovered affecting pre-exercise weight also affected postexercise weight.

We discovered more QTLs for pre- and postexercise percentages of fat and lean tissue than for body weight. And unlike the QTLs for body weight, there was some apparent commonality of QTLs for pre- vs. postexercise fat and lean percentages as well as those affecting both pre-exercise traits and both postexercise traits. Particularly noticeable were QTLs on chromosomes 3 (143.3 Mb), 8 (81.5–89.8 Mb), and 18 (69.1–69.6 Mb) affecting both traits at both (8 and 9 wk) ages and a QTL on chromosome 13 (57.2–59.0 Mb) affecting the percentage of pre-exercise fat and lean tissue and postexercise fat. Of these four QTLs, that on chromosome 8 was most impressive, reaching 5% experiment-wise significance in all four instances, and contributing >11% to the total phenotypic variance of pre-exercise fat and lean tissue. Kelly et al. (31) also found a QTL on chromosome 8 at a similar position (93 Mb) affecting postexercise (although not pre-exercise) fat and lean tissue in the G₄ mice, but none that colocalized with the QTLs we found for these traits on chromosomes 3, 13, and 18. Again, however, this may be a consequence of the fact that the G₄ mice all were fed a normal-fat diet.

The majority (7 of 11) of the QTLs affecting pre-exercise fat tissue in the F₁₀ mice exhibited negative additive genotypic values, indicating that fat levels generally were decreased by the HR alleles but increased by the B6 alleles at these loci. This suggests that we might expect mice in the B6 strain to have a higher percentage of fat than those in the HR strain, and Nehrenberg et al. (42) found this to be the case (mean pre-exercise fat = 8.85% for HR, 9.40% for B6). These investigators also discovered that fat in the F₁ progeny from the HR × B6 cross averaged 10.16%, greater than that for either inbred strain parent. Ordinarily this trend results from overdominance, but only one of the 11 QTLs affecting pre-exercise fat showed overdominance in the correct direction (Table 5). Dominance, including overdominance, was actually more prominent among the QTLs affecting pre-exercise body weight in the G₁₀ mice, although body weight in the F₁ mice from the B6 × HR cross was intermediate between the two strains (42).

QTLs and diet.

We found that three body weight QTLs showed significant interactions with the dietary environment (Table 5). In other studies with mice fed either of two different diets, diet-specific QTLs for body weight have been a prevalent finding (13, 18, 24). For example, in an extensive study using 1,002 mice in an F₁₆ advanced intercross mouse population, Cheverud et al. (13) found interactions of additive, dominance, or imprinting effects with diet for 13 of 17 QTLs affecting body weight. And Gordon et al. (24) found five QTLs for 9 wk body weight in an F₂ mouse population, all of which interacted with diet. On the other hand, none of the six QTLs we found for 9 wk (postexercise) weight colocalized with those found by Gordon et al. (24) or by Cheverud et al. (13), suggesting an entirely different genetic basis for body weight in our population of mice.

Dietary interactions were more prevalent for the QTLs affecting pre-exercise and postexercise fat (and lean) tissue. The QTL on chromosome 3 was most influenced by diet, showing the highest LOD scores associated with the interactions in all four instances (pre- and postexercise fat and lean tissue percentage). Presumably this QTL and other similar QTLs provide the genetic basis for the predisposition to dietary-induced obesity of mice in the B6 strain. Thus, several previous studies have shown that B6 mice on a high-fat diet tend to increase in weight and adipose tissue more so than mice in several other strains, and typically without consuming more calories (2, 55). Besides their higher feeding efficiency (greater weight gain per calories consumed), B6 mice also show higher insulin and leptin levels and greater glucose intolerance than 129/J mice (2). Leptin is an adipose-derived hormone that plays a key role in energy homeostasis, signaling the brain about the level of fatty tissue (20). It is interesting in this regard that Harper et al. (27) discovered a QTL for increased serum leptin levels (Chlq3) that maps to an almost identical position (142.8 Mb) as the chromosome 3 QTL for increased fat we uncovered in the G₁₀ population. And the B6 allele of a QTL (Fpli) discovered by Suto and Sekikawa (56) on chromosome 3 (at 143.4 Mb) increased the level of insulin in their F₂ mouse population.

QTLs and exercise.

It was surprising to see such a low level of correlations between the postexercise weight/composition traits and the exercise traits, especially compared with the generally moderate to high correlations for these two sets of traits in the G₄ mice (31). Genetic correlations between these two sets of traits also would be expected to be quite low because none of the QTLs affecting postexercise weight and composition traits colocalized with those for the exercise traits (Table 5). This result, however, is consistent with the general independence of the direct effects of QTLs for body weight/composition and those for the four exercise traits found for the G₄ mice (Refs. 31, 33; also see Table 6). Leamy et al. (36) also reported similar results for body weight gain and each of three exercise traits, in an F₂ mouse population, and in fact the lack of common QTLs with pleiotropic effects on these traits prevented the calculation of single-locus genetic correlations.

Unlike direct effects of QTLs, we did find significant indirect effects (interactions) of exercise with a number of the QTLs affecting the postexercise weight and composition traits, including QTLs on chromosomes 3, 8, and 18 that mapped to similar areas as those affecting pre-exercise fat. These interactions indicate the presence of relationship QTLs (12, 37, 44), or QTLs that affect a trait variably depending on the level of another trait. We previously showed that the patterns of genotypic effects of these QTLs on the postexercise traits varied considerably, with either the HR or B6 allele capable of increasing or decreasing a trait with increases in exercise (Fig. 2). Leamy et al. (37) discovered a number of relationship QTLs throughout the mouse genome that affected exercise traits in mice differently depending on their individual body weights. To the extent that these sorts of QTLs exist in human populations, they help explain why the effects of exercise on weight or weight change would be expected to vary considerably among individuals each with a unique genotype.

Exercise effects on the QTLs were not always straightforward, since some QTL by exercise interactions differed between the two diets. This sort of result might be expected given the results of the study conducted by Meek et al. (41), who analyzed the simultaneous effects of diet and exercise on HR and control mice over a 2 mo period. Basically, these investigators showed that the high-fat diet produced an increase in fat in both HR and control mice when adjusted for exercise and caloric intake, but that HR mouse had significantly less fat probably because of their greater running distance. Thus, the relationship between body weight or fatness and both diet and exercise is a complicated one but is at least partially attributable to the effect of the QTLs we found that interacted with both of these environmental factors.

Identity of the QTLs.

Although the confidence intervals of QTL mapped in this study still harbor dozens of genes, some inferences are possible regarding the identity of the two key QTLs on chromosomes 3 and 8 affecting body weight and/or the proportion of fat and lean tissue in the G₁₀ mice. For the QTL we discovered on chromosome 3 (143.3 Mb), one potential candidate is Mttp (at 138 Mb), a gene that affects levels of plasma aplipoB100 levels in mice that differ depending on their diet (47) and that may be involved in signaling of appetite and satiety (62). Another possible candidate gene is Unc5c (141 Mb), which affects overall size in mice (34). For the QTL on chromosome 8, a potential candidate is Ucp1 (85.8 Mb), mutations in which induce obesity in B6 mice fed a control diet, but especially those fed a high-fat diet (21). Quite recently, Boström et al. (7) showed that PGC1-α expression in mouse muscle stimulates an increase in expression of FNDC5, a membrane protein that in turn produces a hormone, irisin. This hormone acts on white adipose tissue to stimulate UCP1 expression and brown-fat-like development. Exercise induces irisin, and even mild levels in the blood can increase energy expenditure without exercise or food intake (7). Another possibility for the chromosome 8 QTL is Rln3 (86.6 Mb), relaxin-3, a gene that modulates feeding and metabolism (52). Sutton et al. (57) found that relaxin-3 knockout mice on a 129S5:B6 background weighed less and had less fat than their congenic controls.

Conclusions

We have been successful in discovering a number of QTLs affecting body weight and composition in the G₁₀ mice. Four of these in particular on chromosomes 3, 8, 13, and 18 showed strong effects on the percentage of fat and lean tissue both before and after the exercise period. These and several other QTLs also often showed interactions with diet, exercise, or both, suggesting an underlying genetic basis for the considerable variability in weight of fat loss typically found among individuals on the same diet and/or exercise regimen. Because of the increased mapping resolution in our intercross population of mice, we were able to suggest several genes that represent candidates for QTLs on chromosomes 3 and 8. These included two genes, Ucp1 and Rln3, for the QTL on chromosome 8 that had the greatest impact on body weight and fat in the G₁₀ mouse population. Ucp1 is particularly intriguing because of its potential link with exercise (7), and the role of this gene in human obesity currently is under investigation (39).

GRANTS

This work was partially supported by National Institute of Diabetes and Digestive and Kidney Diseases (NIDDKD) Grant DK-076050 to D. Pomp. Phenotypes were collected with the Animal Metabolism Phenotyping core facility within UNC's Nutrition and Obesity Research Center (funded by NIDDKD Grant P30DK-056350).

DISCLOSURES

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

AUTHOR CONTRIBUTIONS

Author contributions: L.J.L. and S.A.K. analyzed data; L.J.L., S.A.K., and D.P. interpreted results of experiments; L.J.L. prepared figures; L.J.L. drafted manuscript; L.J.L. and D.P. edited and revised manuscript; L.J.L., S.A.K., K.H., and D.P. approved final version of manuscript; K.H. performed experiments; D.P. conception and design of research.